DOI: 10.1002/cl2.

1411

SYSTEMATIC REVIEWS
International development

The effects of agricultural output market access

interventions on agricultural, socio‐economic, food security,
and nutrition outcomes in low‐ and middle‐income
countries: A systematic review

Pierre Marion1,2 | Etienne Lwamba1 | Andrea Floridi1,3 | Suvarna Pande1,4 |

1,4 5 1,4
Megha Bhattacharyya | Sarah Young | Paul Fenton Villar | Shannon Shisler1

1
International Initiative for Impact Evaluation
(3ie), London, UK Abstract
2
Economics Department, Business School, Background: An estimated two billion people do not have sufficient access to
University of Sussex, Brighton, UK
nutritious food, and nearly half are dependent on small‐scale and subsistence
3
International Institute of Social Studies,
Erasmus University of Rotterdam, Rotterdam, farming. Projections show that the global population is not on track to reach the
The Netherlands Sustainable Development Goals. With this in mind, development actors are
4
School of Global Development, University of increasingly seeking to better integrate rural farmers into agricultural markets.
East Anglia, Norwich, UK
5
This synthesis of the literature can help to inform policy decisions to improve
CMU Libraries, Carnegie Mellon University,
Pittsburgh, Pennsylvania, USA outcomes for smallholder farmers in low‐ and middle‐income countries, and to
enable the realization of the Sustainable Development Goals. This work is the
Correspondence
Pierre Marion, International Initiative for
most comprehensive and up‐to‐date review synthesizing evidence from 262
Impact Evaluation (3ie), London, UK. interventions.
Email: [email protected]
Objectives: The purpose of this systematic review is to appraise and synthesize
evidence of the effects of five types of interventions facilitating farmers' access to
output markets in low‐ and middle‐income countries. We examine how these effects
vary across contexts and subgroups. We also identify evidence on program costs and
evidence gaps in the literature.
Search Methods: The search of included studies was based on nine major databases/
search engines and 25 institutional websites, using a set of English search terms. We
also conducted forward and backward citation tracking of literature, published a
public call for papers, and contacted key experts.
Selection Criteria: We included studies on the effects of five types of output
market access interventions, focusing on participants residing in low‐ and middle‐
income countries: (1) Farm‐to‐market transport infrastructure interventions; (2)

This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium,
provided the original work is properly cited.
© 2024 International Initiative for Impact Evaluation. Campbell Systematic Reviews published by John Wiley & Sons Ltd on behalf of The Campbell Collaboration.

Campbell Systematic Reviews. 2024;20:e1411. wileyonlinelibrary.com/journal/cl2 | 1 of 191

https://doi.org/10.1002/cl2.1411
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
2 of 191 | MARION ET AL.

Access to output market information interventions; (3) New marketplaces or

alternative marketing opportunities interventions; (4) Contract farming interventions;
(5) Improved storage infrastructure and technologies interventions. We included
published and unpublished studies from 2000 onwards, with experimental and
quasi‐experimental study designs focusing on relevant outcomes.
Data Collection and Analysis: We screened 52,366 studies, identifying a total of
439 papers representing 289 unique studies on the effects of 262 interventions in
53 countries. Data extraction and risk of bias assessments were completed by two
independent reviewers.
Main Results: All five types of output market access interventions resulted in small‐
to‐moderate positive effects on almost all measures of market participation,
agricultural production, and welfare outcomes. These improvements occurred
through a reduction in transaction costs, adoption of improved practices, greater
farm investment, access to higher prices for farmers, greater volume sold, and
increased farm income. Effects vary by intervention type. However, the body of
evidence is comprised of a large share of included studies with a high risk of bias.
Few studies have information on the cost of interventions, and there are gaps in
the available evidence.
Authors' Conclusions: Output market access interventions are effective in reducing
transaction costs and addressing market failures, thereby leading to higher income
levels. Evidence of effects on food security and nutrition is sparse and has not
provided conclusive findings. We also found that multi‐component interventions are
not necessarily more effective than single component interventions. The specific
needs and dynamics of each context should inform the choice of the intervention
and approach. Investments in additional research with low risk of selection bias and
confounding effects will improve the evidence base, especially for outcomes that
constitute gaps in the literature. For example, the effects on quality of agricultural
products and group participation are under‐researched, and we did not find any
included studies focusing on North Africa or the Middle East. Collecting and
reporting information on interventions' cost‐effectiveness will help decision‐makers
to prioritize limited resources.

KEYWORDS
agriculture, contract farming, farm‐to‐market transport infrastructure, international
development, meta‐analysis, new marketplaces and alternative marketing opportunities,
output market access interventions, output market information, storage infrastructure,
systematic reviews

1 | P L A IN LA N G U A G E S U M M A R Y outcomes in low‐ and middle‐income countries (LMICs): A

systematic review.
1.1 | Output market access interventions result in
small to moderate improvements in agricultural and
welfare outcomes for farmers 1.1.1 | The review in brief

The effects of agricultural output market access interventions Output market access interventions improve the welfare of poor and
on agricultural, socio‐economic, food security and nutrition remote farmers in LMICs through a small reduction in the cost of
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 3 of 191

accessing markets; greater use of improved techniques, inputs and farm 1.1.5 | What do the findings of this review mean?
investment; greater agricultural production; greater sales; and higher
returns from production. Output market access interventions can be an effective way to
integrate poor and remote farmers into markets and to contribute to
long‐term rural development and poverty reduction. The specific
1.1.2 | What is this review about? needs and dynamics of each context should inform the choice of
intervention and approach. In order to improve decision‐making and
An estimated two billion people globally do not have sufficient access work toward the achievement of the Sustainable Development Goals,
to nutritious food and nearly half are dependent on small‐scale and it is important to invest in expanding the evidence base and
subsistence farming. This review examines whether interventions improving its quality. The inclusion of cost data could also provide
assisting rural farmers to sell their products in agricultural markets crucial insight into the variability of expenditure needed across
can improve agricultural, socio‐economic, food security, and nutrition different contexts and enhance understanding of interventions'
outcomes. economic implications.

1.1.3 | What is the aim of this review? 1.1.6 | How up to date is this review?

This Campbell Systematic Review examines the effects of five The review authors searched studies up to November 2023. This
different types of market access interventions. It appraised and Campbell Systematic Review was published in March 2024.
synthesized evidence from 289 unique studies and 262 unique
interventions. We included rigorous quantitative impact evaluations
with a control group. 2 | B A C K GRO U N D

2.1 | The problem, condition or issue

1.1.4 | What are the main findings of this review?
Despite progress in recent decades, hunger and rural poverty remain
What studies are included? chronic global development challenges. In 2021, more than 698 million
This review includes impact evaluations on the effects of output people were living in poverty and 828 million people were chronically
market access interventions on agricultural, socio‐economic, food food insecure (FAO, IFAD, UNICEF, WFP and WHO, 2022; Suckling
security and nutrition outcomes. These interventions are: (1) Farm‐to‐ et al., 2021). Estimates indicate that approximately 80% of the world's
market transport infrastructure interventions; (2) Access to output poor live in rural areas and that the livelihoods of almost half of the
market information interventions; (3) New marketplaces or alternative world's undernourished people and 63% of those in poverty are
marketing opportunities interventions; (4) Contract farming interven- dependent on farming, particularly small‐scale and subsistence farming
tions; (5) Improved storage infrastructure and technologies interventions. (Fan & Rue, 2020; World Bank, 2020, 2022). Projections show that the
We searched 52,366 studies recorded and we identified a total of global community is not on track to reach the Sustainable Development
439 papers representing 289 unique impact evaluation studies Goals, SDGs (FAO, 2022b; World Bank, 2023).
covering the effects of five market access intervention types in Today, a key aspect of development agencies and international
53 LMICs. This work is the most comprehensive and up‐to‐date donors' efforts to address rural poverty and hunger includes better
systematic review on output market access interventions. integrating rural farmers into agricultural markets and promoting
a shift from subsistence to market‐oriented agriculture (IFAD,
What are the effects of output market access interventions? 2022; World Bank, 2022). For example, following the 2008 World
All five types of output market access interventions result in small to Development Report on Agriculture (World Bank, 2007), the World
moderate improvements in agricultural and welfare outcomes for Bank considered linking farmers to markets as one of five priority
farmers. We also found that multi‐component interventions are not areas necessary to support the development of the agricultural sector
necessarily more effective than single component interventions. Few and alleviate poverty. In 2015, it also identified improving farmer
studies have information on the cost of these interventions, and gaps market access as one of three key actions in its Agenda for the Global
in the available evidence exist. Food System, which was prepared as a call for action to provide
support for the Sustainable Development Goals on ending hunger
How has this intervention worked? and poverty (World Bank Group, 2015). Similarly, the Food Coalition
The effects on farmers' outcomes vary across the five intervention types. program1 describes improving access and participation in markets as
The evidence shows that these interventions mostly led to a reduction in
costs of accessing markets and greater market participation, thereby 1
A multistakeholder umbrella program managed by Food and Agriculture Organization of the
driving improvements in agricultural production and welfare. United Nations.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
4 of 191 | MARION ET AL.

an important long‐term solution to transforming global agrifood infrastructure. These interventions were selected as they are the
systems (FAO, 2022a). most prevalent output market access interventions in the LMICs and
The underlying premise supporting the promotion of increased the most researched in the literature.
market participation traces its origins back to, at least, Adam Smith
and David Ricardo, who highlighted the welfare gains that can be
achieved from choosing market‐oriented production and exchange 2.2.1 | Farm‐to‐market transport infrastructure
compared to operating under autarkic conditions (Barrett, 2008).
More modern arguments focusing specifically on the agricultural Transporting produce to markets can be costly if farmers must rely on
sector further highlight that it is anticipated improving market slow, poor quality transport networks and if trade, therefore, involves
participation will incentivize farmers to cultivate higher‐value crops the risk of the loss of product quality and spoilage. Chamberlin and
(including cash crops), adopt improved inputs, and increase their Jayne (2013) show rural marketing costs in many LMICs are often
agricultural production. In turn, it is expected that increased market dominated by transportation costs and de Brauw and Bulte (2021)
sales and agricultural production will also help to increase farmers' explain that such transaction costs reduce the benefits and incentives
income and support their food security by improving food access and to trade. Farm‐to‐market transport infrastructure interventions include
affordability (Barrett et al., 2010; Barrett, 2008; Chamberlin & initiatives that improve infrastructure, such as roads and bridges,
Jayne, 2013; de Janvry & Sadoulet, 2020; Gómez et al., 2011). used for delivering agricultural produce to domestic and international
However, data indicates farmers' participation in markets markets. This includes initiatives constructing, rehabilitating, or
remains limited. Indeed, approximately 35% of farmers worldwide maintaining infrastructure. They are intended to reduce the risks
produce food as part of formal or informal supply chains (Farmer and costs associated with transporting goods and produce. This may
Income Lab, 2018). Numerous factors inhibit farmers' access and make market opportunities more lucrative and improve access to new
participation in output markets. Constraints caused by lack of or distant markets (Aggarwal et al., 2022; Ludwig et al., 2016).
physical access to markets (including time and distance) have been
a principal focus of the literature on farmers' market access
(Baltenweck & Staal, 2007). Other highlighted factors related to 2.2.2 | Output market information
farmers' limited market access concern farmers' lack of connections
to established market actors, land rights, market risk, market Output market information interventions include two different types
structure, low bargaining power, low economies of scale, barriers to of information interventions. One class of interventions includes
entry, anti‐competitive behavior by market actors, economic and initiatives that directly provide farmers with information about the
trade policies, and a lack of capital (human, social, financial, market, usually about prices, completed trades, or current market
natural) (Aquino, 1998; Chamberlin & Jayne, 2013; Maitre et al., 2011; demand (including information on location, quality, variety, and
Markelova et al., 2009; Platteau, 1996; Maziku and Mashenene, 2020; quantity demanded). This can help inform farmers about markets
de Brauw & Bulte, 2021). where they might sell their produce, as well as help them to decide
Several discussions have also focused on the costs of accessing what to grow and how much to expect to sell it for. A second class of
markets (e.g., including information and coordination costs). These interventions increase access to technologies required to access
costs can drive an effective wedge between input and output prices, market information. This includes investments in mobile phone
rendering market participation non‐remunerative. Similarly, receipt services, internet or broadband access, and the basic provision of
of low farm gate prices can render market participation non‐ electricity.
remunerative (Alene et al., 2008; Barrett, 2008; de Brauw & Since Stigler's seminal work on the “Economics of Information”
Bulte, 2021; de Janvry et al., 1991). Price volatility across area, time, (Stigler, 1961), a large body of literature has shown asymmetric
and unpredictability also lead to uncertainty and risks to farmers information can affect market equilibria, creating an inefficient
(Cardell & Michelson, 2023). allocation of goods across markets, increasing price dispersion and
instability, and decreasing trade and competition (Aker, 2008;
Goyal, 2010). For example, one way information asymmetries can
2.2 | The interventions distort agriculture markets is that more informed traders may use
their knowledge to exploit farmers and pay lower farm gate prices.
In this review, we compiled evidence on the effects of a selection of Farmers who receive lower prices for their produce may be less
five key types of interventions intended to improve farmers' market inclined to trade and may limit their market operations accordingly
access and participation in LMICs. Informed by expert suggestions, (Magesa et al., 2020; Nakasone, 2013).2 Overall, it is anticipated that
we examined interventions that address farmers' access and market information may facilitate deeper agricultural markets by both
participation in output markets by (i) improving farm‐to‐market
transport infrastructure; (ii) increasing access to output market
information; (iii) creating new marketplaces or alternative marketing 2
However, there is also significant debate about whether traders are exploitative or whether
opportunities; (iv) facilitating contract farming; (v) improving storage we simply are not pricing in their services or costs correctly (see Dillon & Dambro, 2017).
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 5 of 191

encouraging and enabling farmers to participate more effectively in 2.2.5 | Storage infrastructure
them (Giovannucci & Shepherd, 2007).
Improved storage infrastructure interventions provide access to storage
facilities, such as sheds with off‐the‐ground storage, and storage
2.2.3 | New marketplaces and alternative marketing deposit systems. These interventions are designed to reduce the risks
opportunities associated with storing agricultural produce, which may help to
increase the cultivation of alternative high‐yielding crops that do not
Interventions creating new marketplaces and alternative marketing otherwise store well. This can increase access to new or alternative
opportunities aim to provide new ways for market participants to markets (Katunze et al., 2017; Omotilewa et al., 2018) and it may also
trade and/or reduce search costs by simplifying or improving improve market access by enabling farmers to perform a more
connections between market participants. It is expected that these extensive market search for buyers or enabling them to wait for more
types of interventions will increase market competition, improve favorable market conditions (i.e., perform intertemporal arbitrage)
the prices farmers receive for their produce, and, to the extent this (Lothoré & Delmas, 2009; Pingali et al., 2019). However, an increasing
creates market incentives, increase production and possibly alter the degree of controversy now surrounds storage interventions. For
production of the types of crops grown (e.g., increasing the adoption example, Cardell and Michelson (2023) show that the price for maize
of cash crops; Levi et al., 2020; Goyal, 2010). Examples of this type of often declines after harvest (~30% of the time), so storing after harvest
intervention include creating new marketplaces by connecting may also be detrimental to farmers outcomes (especially when
geographically distributed agri‐markets through online platforms programs are combined with credit facilities).
(e.g., Levi et al., 2020) or mobile trading platforms (e.g., Bergquist &
McIntosh, 2021). Other examples of alternative marketing outlets
include setting up trading hubs or internet kiosks (e.g., Goyal, 2010) 2.3 | How the interventions might work
or curating commitments or stable arrangements where traders offer
to buy farmers' produce (e.g., Bold et al., 2022; Maertens et al., 2020). The theoretical framework for this review is based on a broader
literature that indicates that interventions can strengthen market
access by affecting the way markets function and minimizing the
2.2.4 | Contract farming market inefficiencies that inhibit farmer's market access and
participation (de Castro, 2021). This includes, for instance, improving
Contract farming and outgrower schemes create arrangements connectivity or decreasing farmers' search and transaction costs,
where traders and farmers enter forward‐looking agreements for lowering farmers' market risk, increasing the prices farmers receive
3
the delivery of agricultural produce (Eaton & Shepherd, 2001). As for their produce, and so forth (see the implied conceptual model in
well as advanced knowledge of price, these arrangements can Figure 1).
provide farmers with more certainty about the quantity and quality Both theory and evidence suggest that by improving factors
of produce in demand. Contracts can reduce some of the risks which increase the prices farmers receive for their produce, decrease
associated with farmers' land allocation and investment decisions and market risks, and reducing search and transaction costs, output
they can limit the need to use spot markets to sell produce after they market access interventions can affect production outcomes. This
have already invested in the inputs required to farm their crops occurs as the returns to selling produce in the market (commerciali-
(Bellemare & Bloem, 2018; Bellemare & Lim, 2018). Increased zation) increases relative to the welfare benefits of growing staple
certainty regarding price and demand can also facilitate farmers' crops for consumption. Alternatively, this may also incentivize
investment into new markets, as well as access to finance (e.g., via farmers to expand their cultivation of crops and other agricultural
credit facilities or sometimes the conditions of a contract can provide products (Bold et al., 2022; de Brauw & Bulte, 2021; Goyal, 2010).
inputs in‐kind). For example, in the contracts studied by Bellemare To the extent that this causes farmers to adjust their decisions
(2010), seeds, pesticides, and fertilizer are provided to farmers with regarding their land allocation, this may also cause an increase in the
the contracted crop being used as collateral. It may also help farmers adoption of alternative high‐value crops and increase the production
organize supply under a common contractual agreement, thereby of higher quality crops (Bold et al., 2022). For example, work by
accessing new markets that require or benefit from some degree of Fafchamps (1992) indicates that price risk affects farmers' crop
aggregation/group participation. Where group participation improves choice. Similarly, to the extent that these factors enable greater
farmers' negotiation or bargaining power, this could also result in access to credit and market incentives encourage the use of
improved terms of trade and increase the prices they receive for their improved inputs, it is anticipated this may increase agriculture yields
produce (Barrett et al., 2012). (George, 2014; Ragasa et al., 2018).
Increased commercialization, higher returns to trade, and larger
yields are expected to increase farmers' incomes (Zeller et al., 1998;
3 Barrett, 2008). Haggblade et al. (1989), Haggblade and Hazell (1989)
Outgrower schemes refer to arrangements involving public enterprises, parastatals,
government agencies, or NGOs rather than private entities (Bellemare & Bloem, 2018). and de Janvry and Sadoulet (2002) also explain it is possible that
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
6 of 191 | MARION ET AL.

FIGURE 1 Conceptual model: potential channels of effects.

agricultural growth can generate income and employment multipliers infrastructure may also bring more goods to local markets to compete
in the rural non‐farm economy by increasing demand for production with local farmers, thereby possibly decreasing prices and local
inputs and consumer goods. This may further increase farmers' off‐ farmers' incomes (Dumas & Jativa, 2020). A growing literature also
farm income, and where production outcomes, yields, and incomes concerns competition and market structure (e.g., see Bergquist &
increase, it is anticipated interventions will help to support farmers' Dinerstein, 2020; Casaburi & Reed, 2021; Chatterjee, 2019). If
food security and nutrition by improving food access or whether they intermediaries are operating in a competitive environment, then
can afford to purchase food (Barrett et al., 2010; Barrett, 2008; market access interventions that reduce transaction costs would also
Mishra et al., 2018; Rugumamu, 2014). yield savings that traders pass on to farmers (in the form of higher
However, there are also several competing predictions about the prices). However, if traders are exercising a high degree of market
anticipated effects of market access interventions. For instance, farmers' power, cost savings may almost exclusively benefit (be captured by)
incomes can be diversified on‐farm but also off‐farm (Reardon, 1997). intermediaries, not farmers (see Bergquist & Dinerstein, 2020).
Brooks and Donovan (2020) highlight that households manage their Finally, market participation can introduce market and commercial
choice of agricultural production and labor market risk simultaneously, risks that can be difficult to manage (particularly when farmers are
and that farm investment is risky. This nuance creates a degree of highly leveraged) and increase farmers' costs. For example, market price
theoretical ambiguity about the potential effects of these interventions risks are already well documented above. Furthermore, contract
on farmers' agricultural outcomes. If market access interventions, such farming can introduce risks, such as risk of contractual default, while
as those improving roads and other forms of transport, simultaneously also creating additional costs (e.g., due to the inputs requirements for
alter farmers' access to other income‐generating opportunities, then special varieties and to manage the processes necessary to fulfill a
these interventions may not increase farm incomes as implied above. contract) (Narayanan, 2014). This emphasizes the risk‐return trade‐off
The anticipated effects on farm incomes become ambiguous because farmers may face when participating in markets.
households may substitute some of their farm activities with new Across all these interventions, the extent of the positive and
opportunities that arise off‐farm (e.g., in the labor market). unintended negative effects described above are likely to vary
Furthermore, where market access interventions cause a shift considerably across farmers' characteristics and over time. A
away from staple crops and toward higher‐value cash crops, reduction in transaction costs will affect farmers differently based
they may have an ambiguous effect on food security. Theoretically, on the type and level of sales. Also, it is reasonable to assume that
if the price of the cash crop were to fall, the farmers' income from farmers face non‐separable production and consumption decisions
cash crops may not be capable of fulfilling their dietary needs. An due to market failures and missing markets, meaning that consump-
extensive and longstanding debate concerns such issues (e.g., see tion decisions would influence production decisions and vice‐versa
Alarcon, 1993; Barbier, 1989; Bouis & Haddad, 1990; Dewey, 1981; (Key et al., 2000; Taylor & Adelman, 2003). Farmers can be
Fafchamps, 1992; Govereh et al., 2003; Timmer, 1997; etc.). distinguished into two groups: net‐buying farmers (including non‐
Increasing market access may also lead to higher market and agricultural households and subsistence farmers) and net‐selling
price competition (Minot & Hill, 2007). For example, transport farmers. Farmers' production and consumption choices vary from
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 7 of 191

farmer to farmer. Similarly, effects in the short term are not Bellemare and Bloem's [2018] review on contract farming restricts
necessarily the same as those in the long‐term (Lane et al., 2022). the search for the literature to published academic journal articles),
making a more comprehensive and up to date review necessary. Also,
many reviews do not use transparent and systematic methods for
2.4 | Why it is important to do this review synthesis or formal approaches for critical appraisal, which means
they cannot be considered systematic reviews as defined by
It is widely argued that market access may be a necessary (if not Campbell Collaboration (e.g., Aker and colleagues, 2016).
sufficient) condition required for agricultural transformation and We intended to provide a new and updated systematic review,
poverty reduction at the country level (Barrett et al., 2010; summarizing the evidence for the five focal agricultural market access
Barrett, 2008; Gómez et al., 2011), at the household level (Chamberlin intervention types indicated above. In doing so, we provide a new
& Jayne, 2013) and farm level (de Janvry & Sadoulet, 2020). Reflecting synthesis in areas where there have been relatively limited efforts to
these arguments, development agencies and international donors' synthesize evidence using systematic approaches (such as output
efforts to address poverty and rural hunger increasingly focus on better market information and storage infrastructure interventions and
integrating poor and remote farmers into markets. For example, the initiatives creating new or alternative marketplaces) and update and
International Fund for Agricultural Development (IFAD) (2022) states expand reviews where there are some existing efforts (e.g.,
increasing poor rural people's access to markets is now one of its to’ concerning infrastructure and contract farming).4
priorities, and the proportion of IFAD‐supported projects that include
work on market access has increased dramatically (increasing from 3%
to more than 75% over the last two decades). 3 | OBJEC TI VES
Correspondingly, evidence of the effects of interventions
addressing farmers' access to output markets has expanded in recent The purpose of this review is to identify, assess and synthesize
years. Bold and colleagues (2022) present one example of an evidence of the effects of interventions addressing farmers' access to
experiment examining the effects of an intervention in Uganda output markets. To achieve this objective, we answer the following
coupling agricultural extension services with a market access questions:
component (where an agro‐trading company committed to offering
to buy farmers' quality maize at a premium). Other examples have (1) How effective are output market access interventions in improving
also examined the effects of the construction of roads, increasing the agricultural, socio‐economic, food security and nutrition outcomes?
provision of market information, and using technology to create more (2) How do effects vary across different contexts, interventions, and
unified agricultural markets (e.g., see Khandker & Koolwal, 2010; Levi outcomes, and do the effects of interventions differ between
et al., 2020; Svensson & Yanagizawa, 2009). sub‐groups of the population (e.g., according to sex, race,
A handful of reviews have synthesized evidence on the effects of ethnicity, age, socio‐economic status, type of produce, farm size,
agricultural market access interventions. For example, systematic etc.)?
reviews by Hine and colleagues (2016, 2019) and Ludwig and (3) What is the risk of bias of studies included in the review?
colleagues (2016) find rural road investment prizes farmers via (4) What evidence is available in studies included in the review on
improved income, poverty reduction, agricultural employment, agricul- program costs and their cost‐effectiveness?
tural output, and sales. Aker and colleagues (2016) review the literature (5) Where do gaps exist in the literature, and how can future
on agricultural information technology initiatives (such as information research enrich the evidence on the effects of interventions
interventions using mobile phones) and conclude that evidence of the designed to improve access to agricultural markets in LMICs?
effects of such interventions on farmers' behavior and welfare
outcomes remains mixed. A systematic review on contract farming by
Ton and colleagues (2017) also suggests that such schemes may 4 | METHODS
increase farmer income but that the effects may be concentrated
among wealthier farmers who can compensate for the transaction costs 4.1 | Criteria for considering studies for this review
to take advantage of these types of arrangements.
However, while the results and implications derived from existing Next, we outline the inclusion and exclusion criteria, which define the
reviews tend to be narrow and deep, they are broadly less accessible factors determining whether a particular study was included in the
to researchers, practitioners, and funders of evidence than might be review.
ideal. They are distributed widely across the literature, often focusing
on only one particular intervention type, and the approaches for
4
We initially considered combining a synthesis of intervention types without systematic
evidence selection, appraisal, and analysis are rarely comparable. review evidence with a review of reviews for interventions with existing systematic review
Furthermore, extant reviews are also fast becoming dated (e.g., the evidence. However, we determined that existing systematic reviews warranted updating to
present the most up‐to‐date evidence and, in the process of updating the reviews,
systematic search by Ton and colleagues [2017] was last completed
standardizing the approaches of study selection, analysis, critical appraisal and effect size
in 2015). Some reviews include a limited sample of studies (e.g., calculations would increase the comparability of results across different intervention types.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
8 of 191 | MARION ET AL.

4.1.1 | Types of studies studies included evidence of the effects of interventions implemen-
ted in more than one country. A study including interventions from
We included studies using experimental and quasi‐experimental multiple countries was included if results were provided for LMIC and
designs (QEDs) to measure a change in outcomes attributable to an high‐income countries separately in the results.
intervention. These methods rely on the identification of a counter-
factual, which provides evidence about what would have happened in
the absence of the intervention. The counterfactual may be inferred 4.1.3 | Types of interventions
from a specific group of observations that do not receive the
intervention (as in a randomized controlled trial [RCT]), or that receive We included interventions that aim to strengthen farmers' output
an alternative intervention. A counterfactual may also be constructed market access using one or more of the five types of interventions we
by researchers (using propensity score matching or interrupted time describe above (Table 1). We considered both farmers already with
series). This means that included studies may apply a wide range of access to markets (commercialized farmers) and subsistence farmers.
potential designs, such as RCTs, regression discontinuity designs, This list of included interventions was not an exhaustive list of
natural experiments, instrumental variables (IVs), fixed effects (FE) potentially relevant output market access interventions but provided a
regressions, interrupted time series models, matching methods, and the selection of key strategies. The selection of interventions included
synthetic control methods (see Supporting Information: Appendix 1 for was informed by expert advice.
background and methods sections). These study designs are widely As also noted in our discussion above, there were variations in
recognized as credibly establishing causal relationships between the way interventions may have appeared in practice. Table 1
interventions and outcomes when carefully and credibly executed characterizes further the different ways these intervention types
(Aloe et al., 2017; Reeves et al., 2017; Gertler et al., 2011; Villar & were sub‐categorized, and our analysis was designed to consider
Waddington, 2019; Waddington et al., 2023). variation within intervention types, as well as between them, where
We did not exclude studies based on the comparison condition there was sufficient evidence to do so. We adopted an iterative
of a control group. A study's control group may consist of approach, where we updated these subcategories of interventions
observations subject to no intervention, on a waitlist, or a member again once we had identified the full population of studies included in
of an alternative intervention or condition. However, we excluded the review. The subcategories of interventions we presented in
studies that only use simulation or forecast models, ex‐ante impact Table 1 are based on expert advice and knowledge of existing studies
assessments or scenario analyses, as well as evaluations and case from our initial searches to develop our search strategy (discussed
studies that do not satisfy the methodological conditions described below). We performed a further mapping exercise analyzing
above. We also excluded efficacy studies, feasibility studies, interventions for any other notable affinities that can be gleaned
acceptability studies, and literature reviews, and systematic reviews from the available information about the interventions, as well as the
were not included as primary studies. differences between them.
The nature of many of the interventions studied, particularly Interventions may include activities addressing these issues in
large infrastructure interventions and widespread information in- isolation or in combination with activities. For example, a farming
itiatives, may create spillover effects and market equilibrium contract may include both a fixed price agreement and financing for
outcomes. For example, Hildebrant and colleagues (2020) provided inputs, or alternatively, an intervention may include multiple compo-
evidence of spillovers from a price alert system in Ghana. We nents consisting of both the provision of market information and a
documented where studies presented both partial and general market‐making activity using information technology. They may also
equilibrium effects of interventions (e.g., Yanagizawa‐Drott & include another intervention that is not in itself included in this review,
Svensson, 2012) or controlled for this issue via a study design (e.g., such as agriculture extension services with a market access intervention
Bergquist & McIntosh, 2021). (as in Bold et al., 2022). Our review endeavored to record the nuances
of intervention arrangements in its description and analysis.
We included interventions targeting farmers of seasonal (such as
4.1.2 | Types of participants grains, pulses, or root and tuber crops, etc.) and permanent crops
(such as cocoa, coffee, tea, etc.). This includes horticulture products,
We included studies on the effects of interventions on participants and we also included interventions concerning livestock and animal
residing in LMICs. We used the World Bank income status husbandry, which is the branch of agriculture concerned with animals
classification for defining LMICs (Supporting Information: Appendix 2 that are raised for meat, milk, skins, hides, and so forth. However, to
for background and methods sections), and studies were classified limit the overall breadth of the scope of the review, we did not
according to their status in the year the intervention began.5 Some include studies on market access interventions from related sectors
(i.e., the fishery or forestry sectors). We added this clarification
because some organizations, such as the Food and Agriculture
5
We also included studies from countries that have had a high‐income status for only one
Organization of the United Nations (FAO) and USAID, aggregate
year before reverting to LMIC status. At the time of writing this protocol, this exception only
applies to Argentina (2014, 2017), Venezuela (2014), Mauritius (2019), and Romania (2019). these economic activities into a single agriculture sector.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 9 of 191

TABLE 1 List of included interventions and subcategories of interventions.

Intervention type Intervention subcategories Definition

Farm‐to‐market transport Domestic transport infrastructure Initiatives that improve infrastructure primarily used for domestic
infrastructure interventions transport, such as rural roads, bridges, waterways, and river
transport.

Export transport infrastructure Initiatives that improve infrastructure used primarily to export
produce to foreign markets, such as at ports, border crossings or
international distribution facilities.

Output market information Provision of output market Interventions that provide farmers with output market information.
interventions information (mobile and internet) The mode of information dissemination must include mobile (e.g.,
via SMS, phone) and internet (e.g., via apps, website, etc.)
technologies.

Provision of output market Interventions that provide farmers with output market information.
information(other forms of The mode of information dissemination can include other more
communication) traditional forms of communication and technology (e.g., radio,
peer networks, etc.).

Investments in information and Interventions that increase access to technologies required to

communication technologies access market information. This may include investments in mobile
infrastructure phone services, internet or broadband access, and the basic
provision of electricity.

Initiatives creating new marketplaces Online commodity exchanges and Interventions improving connections between market participants
& alternative marketing opportunities mobile‐based marketplaces by creating new or improved marketplaces using information
technologies, for example, via online commodity exchanges, apps,
or mobile services.

Alternative physical marketplaces Interventions improving connections between market participants

by creating physical alternative marketplaces, such as physical hubs
linking farmers directly with private companies.

Arranged or curated offers from Interventions improving connections between market participants
buyers by curating commitments or stable arrangements where traders
offer to buy farmers' produce (e.g., Bold et al., 2022; Maertens
et al., 2020).

Contract farming initiatives(inc. out‐ Fixed‐price and price‐guarantee Fixed‐price contracts are forward‐looking agreements in which the
grower schemes) contracts buyer provides farmers with a guaranteed price for the delivery of
agricultural produce.

Production‐management contracts Production‐management contracts whereby the buyer provides

extension services or other technical support for the delivery of
agricultural produce.

Input or credit‐supply contracts Input or credit‐supply contracts where the buyer provides inputs or
credit upfront and deducts the cost at harvest.

Other contract types Other types of contractual arrangements between farmers and
buyers for the future delivery of agricultural produce.

Undefined contract farming Contract farming interventions where the characteristics of

interventions are not known.

Improved storage infrastructure Improved on‐farm storage Interventions that provide technical, logistical, or financial support
interventions infrastructure for improved on‐farm farm storage infrastructure, such as the use
of sheds with off‐the‐ground storage.

Storage deposit systems Interventions that enable farmers to deposit produce in off‐farm
stores, such as warehouse receipt systems.

Improved storage technologies for Interventions that support the use of improved technologies to
transit maintain produce quality or prevent it from spoiling during transit.

Note: We included studies examining the effects of households' participation in contract framing schemes, as well as field experiments providing or
offering contract farming as part of an intervention.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
10 of 191 | MARION ET AL.

With respect to interventions creating alternative marketing development channel, we included welfare, non‐ and off‐farm
opportunities or new markets, we excluded public marketing (or income, and labor market outcomes, but only from studies that also
commodity) boards, which are set up by a government to regulate the report agricultural production outcomes, intermediate outcomes, or
buying and selling of a certain commodity within an area (e.g., see off‐farm (agriculture only) outcomes. In our analysis, we examined
Fung et al., 2020). We also excluded minimum support price policies the statistical correspondence between the estimated effects on
(MSPs), which are also recognized as a type of output subsidy and are agricultural outcomes and these other outcomes (e.g., confirming that
regularly used by governments to safeguard farmers' income against evidence indicates infrastructure interventions improve both agricul-
crop price falls and ensure sufficient and balanced production of tural outcomes and welfare outcomes).
crops. Furthermore, we excluded studies examining the effects of
establishing traditional agricultural commodity exchanges or pro-
grams promoting post‐harvest practices and technologies, such as 4.1.5 | Duration of follow‐up
drying tarps (e.g., see Magnan et al., 2021). We also excluded studies
that concern certification only, such as UTZ Certified, Fair Trade, and We included studies published in 2000 or later. This restriction is
Rainforest Alliance, unless a non–transferable fixed‐term forward necessary to keep the review's search feasible within project
sales contract with a specific firm is offered. Sharecropping resources. Furthermore, analyses indicate that although there were
arrangements where a tenant farmer is provided with inputs in IEs published in international development before 2000, the number
exchange for an agreed share of the harvest were also excluded. is relatively small, especially in the agriculture sector (see Sabet &
Brown, 2018). Studies published in any language were included,
although the search terms used were in English only.
4.1.4 | Types of outcome measures

We included studies that contain data on outcomes related to the 4.1.6 | Types of settings
conceptual framework above. This includes immediate outcomes
(such as transactions costs, group participation, use/adoption of Based on the selected criteria described in this section, included studies
improved inputs, technologies, and practices, farm investment, credit in this review focused on rural settings with agricultural activities where
and prices, crop losses), agricultural production outcomes (including farmers are either reliant on subsistence or commercial farming in
measure of yields, volume, quality, produce type), intermediate LMICs. These settings are likely to face poverty and remoteness from
outcomes (farm sales and income), non‐ and off‐farm income and urban areas and major cities. Therefore, included studies did not
labor outcomes, and welfare outcomes (total household income and analyze the effects of interventions in urban areas or of large‐scale
food security). producers.
Table 2 provides further details of each of the outcome types
included. This may not be an entirely exhaustive list of potential
outcomes for these interventions. There are many channels through 4.2 | Search methods for identification of studies
which these interventions may operate, and some outcomes, such as
farmers' market risk, are also difficult to define or measure in the To identify relevant literature, we conducted a comprehensive search
context of an impact evaluation. Informed by expert advice, this for eligible published and unpublished studies. Our search strategy
review focused on a selection of key outcomes which provide good was developed in collaboration with an information specialist and
coverage of a broad range of relevant outcomes related to the with reference to guidance in Kugley and colleagues (2017). We have
interventions' theory of change and the primary welfare outcomes of developed a set of English search terms, which we used in a wide
interest. All outcomes here were relevant for the five intervention array of electronic academic and institutional databases. We also
types considered in this review. However, the prevalence of conducted citation tracking, published a blog presenting a public call
outcomes featured in the primary research evaluations dictates the for papers, and we contacted key experts and organizations to
final outcomes included for each intervention within our framework. identify additional studies.
We reflect on the prevalence of these particular outcomes in the
systematic review for each intervention type.
Interventions investing in public goods (such as roads and bridges 4.2.1 | Electronic searches
concerning farm‐to‐market interventions or internet and electricity
infrastructure under information interventions) may also have several To identify relevant studies for our review, we have developed a set
possible channels explaining their effects on welfare outcomes. For of English search terms and a search strategy in collaboration with an
example, other possible channels beyond the hypothesized effect on information specialist. Our search terms combine Boolean terms
the agricultural sector may include improved access to health care, with a list of keywords related to the review's inclusion criteria. We
education, other labor markets, and so forth. Since in this instance we used and adapted these terms to search electronic databases and
are interested in understanding their effects through the agricultural institutional websites with sufficient search functionality.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 11 of 191

TABLE 2 List of included outcomes.

Outcome type Outcome Description

Immediate outcomes Transaction costs This includes measures of costs associated with trading produce and that are
accrued by farmers (such as transport costs).

Group participation This includes outcomes measuring the incidence of farmer participation or
membership of cooperatives and farmer groups.

Use/adoption of improved inputs, This includes outcomes measuring the use or adoption of improved inputs,
technologies & practices technologies, and practices (such as pesticides, specialized seeds, etc.). This
also include measures of whether farmers have produce in storage, or the
amount of produce in storage.

Farm investment This includes measures of the monetary value of farmer investment
(spending) on improved farm infrastructure, technologies, inputs, and so
forth.

Credit This includes outcomes measuring farmers access or use of credit (such as
the total amount borrowed).

Prices This includes farm‐gate prices, that is, those effectively received by farmers
for their produce. Prices may be measured through farmers' reports of prices,
market surveys, and from records of market transactions.

Crop losses This includes measures of post‐harvest losses.

Agricultural Production Yields Yields are a measure of agricultural productivity.

outcomes
Volume This includes measures of the quantity or volume of agricultural production
or output and share or amount of land cultivated

Quality This includes measures of the quality of produce (e.g., Aflatoxin and moisture
in stored crops).

Produce type This includes outcomes where studies are specifically intending to measure
the adoption or increased volume of an alternative variety of crop or produce
(e.g., production of specialized crop varieties, production of cash crops, etc.).

Intermediate outcomes Sales This includes measures of the proportion, number and the volume of produce
sold by farmers.

Farm income This includes measures of income generated from farm activities (net and
gross income)

Off‐farm outcomes Off‐farm income This includes measures of income generated from off‐farm agriculture‐
(Agriculture only) related activities that occur beyond the farm owned by the household (e.g.,
off‐farm wage work in agriculture).

Labor market outcomes This includes measures of employment outcomes and number of hours
worked for off‐farm agriculture‐related activities that occur beyond the farm
owned by the household (e.g., off‐farm wage work in agriculture).

Non‐farm outcomes Non‐farm income This includes measures of income generated from non‐farm activities.

Labor market outcomes This includes measures of employment outcomes and number of hours
worked (not on a farm owned by the household).

Welfare outcomes Total income and wealth This includes measures of total household income and other measures of
socio‐economic status (such as total household expenditure and asset or
wealth indices).

Food security and nutrition Food security and nutrition concerns the state of having reliable access to a
sufficient quantity of affordable, nutritious food. We included indices of food
security and nutrition, composite scores of the extent to which households
have food to meet basic dietary needs, measures of nutritional intake and
food consumption and outcomes based on whether households report they
have sufficient food.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
12 of 191 | MARION ET AL.

As suggested by Sandieson (2006), we utilized pearl‐harvesting 4.3 | Data collection and analysis
techniques to develop our list of key search terms. This involves
reviewing and identifying keywords from our inclusion criteria and 4.3.1 | Description of methods used in primary
known studies relevant to the review (Supporting Information: research
Appendix 3 for background and methods sections). We identified an
initial sample of relevant studies by consulting with members of the Based on the selected criteria described in “Types of studies,” primary
project's expert advisory group and searching the results of a selection studies included in this review will use either experimental study
of existing reviews (Supporting Information: Appendix 4 for background designs or QEDs. These analysis methods can address selection and
and methods sections). We also performed targeted searches of 3ie's confounding bias and, therefore, are best to attribute specific
Development Evidence Portal (DEP), which is currently the world's most observed changes in outcomes to the intervention under study.
expansive repository of rigorous evidence on what works in interna-
tional development. We used the DEP's intervention search filters,
which index studies using a standardized vocabulary for classifying 4.3.2 | Selection of studies
interventions (Kozakiewicz et al., 2021), to help to identify a sample of
studies listed in the portal that are most relevant to our review. After collating and de‐duplicating records from our literature search,
Supporting Information: Appendix 5 (for background and methods we performed a two‐stage selection process where trained reviewers
sections) provides a list of the initial list of studies identified for pearl‐ assessed studies against the review's inclusion and exclusion criteria.
harvesting and developing our search strategy.
We also compiled a list of databases and websites we searched for De‐duplication of records in the literature search
relevant evaluations and studies (Table 3). To reduce the risk of First, the search results from databases and search engines went through
publication bias, these sources have been selected to cover a range of an initial round of de‐duplication in R using the synthesisr package.
publication types, including journal articles, working and discussion Results were then imported to the EPPI‐Reviewer software (Thomas
papers, conference proceedings, thesis and dissertations, and institutional et al., 2022), where EPPI's de‐duplication functionality was also applied
reports. We have identified relevant sources by consulting an information as a second check for duplicate records. For studies identified through
specialist, the project's expert advisory group, and systematic reviews on forward citation tracking, we also de‐duplicated records where it was
included or related interventions recording databases including impact possible to export the bibliographic information of a records reference
evaluation evidence (including Hine et al., 2016, 2019; Ludwig et al., 2016; list and uploaded this to EPPI. Any additional relevant studies identified
Oya et al., 2017, 2018; Ton et al., 2017; Waddington et al., 2014). through suggestions from our expert advisory group, study authors, and
While some websites and databases have reasonably well‐ backward citation tracking were manually captured in EPPI Reviewer,
developed search functions, some do not support complex search where they cannot be identified in our search results.
strings or allow for the direct export of materials, and others must be
browsed by keywords or even browsed in their entirety. We customized Stage 1 – Title and abstract screening
our general search strategy according to the functionality of each After de‐duplicating records from our literature search, the first stage of
database and website we searched (using the website's thesaurus or the selection process consisted of screening the information available in
keyword index if necessary to identify the most appropriate vocabulary). study titles and abstracts against the inclusion and exclusion criteria.
We consulted an information specialist who helped to troubleshoot Trained reviewers independently screened studies, and those that were
problematic sources, as well as advised on the best ways of conducting relevant and met the inclusion criteria were flagged for a full‐text review
targeted searches. We documented the literature search process and in the second stage of the selection process. More information on this
any necessary changes to the search strategy for each source. screening is available in the Protocol (Villar et al., 2023).
Supporting Information: Appendix 6 for background and methods
sections provides the search strategy and results for databases and Stage 2 – Full text screening and study selection
search engines and for institutional websites and repositories. In the second stage of the study selection process, using two
independent reviewers we double screened all studies flagged for a
review using each manuscript's full text.
4.2.2 | Searching other resources

We supplemented these searches by contacting key researchers and 4.3.3 | Data extraction and management
organizations working on issues related to this review, and we
engaged our expert advisory group for suggestions concerning other We extracted the following descriptive, methodological, quantitative,
relevant studies. Finally, we searched the included studies of other and cost and other qualitative data from each study included in the
related evidence maps and reviews. Supporting Information: Appen- review using standardized data extraction forms (forms are provided
dix 5 (for background and methods sections) provides the list of in Supporting Information: Appendix 7 for background and methods
relevant maps and reviews. sections):
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 13 of 191

TABLE 3 Table of electronic databases and websites to be searched.

Type Sources

Databases and search engines Scopus

CAB Abstracts
Web of Science1
Agricola
Econlit
EBSCO Discovery Service2
International Bibliography of the Social Sciences
Sociological Abstracts
ProQuest dissertations and theses database

Institutional websites and 3ie Development Evidence Portal

repositories Action Against Hunger
ActionAid
Abdul Latif Jameel Poverty Action Lab (J‐PAL)
African Development Bank (AfDB)
Agricultural Technology Adoption Initiative (ATAI)
Asian Development Bank (ADB)
Centre for Agricultural Bioscience International (CABI)
Danish International Development Agency (DANIDA)
Australian Government Department of Foreign Affairs and
Trade (DFAT)
Development Experience Clearinghouse (USAID)
Food and Agriculture Organization (FAO)
Farm Africa
Feed the Future Innovation Lab for Markets, Risk &
Resilience
Deutsche Gesellschaft für Internationale
Zusammenarbeit (GIZ)
International Fund for Agricultural Development (IFAD)
International Food Policy Research Institute (IFPRI)
Innovations for Poverty Action (IPA)
Inter‐American Development Bank (IADB)
Netherlands Ministry of Foreign Affairs
Norwegian Agency for Development Cooperation
(NORAD)
Oxfam Policy and Practice
Research for Development: FCDO
The Expert Group for Aid Studies (EBA)
World Food Program (WFP)

Note: Details of the search pathway and any filters applied to websites were recorded and published in supplementary materials with the final report.
1
Including Web of Science Core Collection (Social Sciences Citation Index (SSCI), Science Citation Index Expanded (SCI‐EXPANDED), Conference
Proceedings Citation Index – Science (CPCI‐S), Conference Proceedings Citation Index – Social Science & Humanities. (CPCI‐SSH), Emerging Sources
Citation Index (ESCI).
2
Including GreenFILE, Academic Search Complete, Science Direct, AGRIS, RePEc, World Bank
e‐Library.

Descriptive data, methodological information and cost data was methods sections). This examined both the internal validity and
single coded by a trained reviewer and checked for agreement by statistical conclusion validity of experimental and quasi‐experimental
another one. Two trained reviewers independently coded the impact evaluation designs (see Waddington et al., 2012). Two
quantitative data, and any disagreement was resolved through reviewers undertook the risk of bias assessment independently. If
discussion with a third reviewer (who must be a core team member). there were disagreements, we resolved them by discussion and the
involvement of a third reviewer (who must be a member of the core
team). We compiled a risk of bias assessment for each estimate we
4.3.4 | Assessment of risk of bias in included studies extracted. This reflects that reflecting estimates on different
outcomes in the same study may score differently in the assessment
We assessed the risk of bias in included studies using 3ie's risk of bias (available in Supporting Information: Appendix 12 for background
tool (see Supporting Information: Appendix 10, for background and and methods sections).
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
14 of 191 | MARION ET AL.

4.3.5 | Measures of treatment effect meta‐analytic methods assume effect size estimates were indepen-
dent and failure to qualitatively recognize estimates were derived
In this review, we examined the effects of output market access from the same intervention or study can distort (inflate) our
interventions. An effect size (or treatment effect) expresses the perceptions of the availability of evidence.
direction and magnitude of the difference in outcomes between Dependent effect sizes can arise in several circumstances. For
groups of observations, such as the difference in outcomes between example, co‐dependencies between estimates can arise when several
observations in the intervention and comparison groups (Borenstein publications stem from one study, or several studies are based on the
et al., 2009a; Valentine et al., 2015). same data set. Some studies might have multiple treatment arms that
However, effect sizes presented in empirical studies are rarely are all compared to a single control group. Other studies may report
independent of the scale or unit of the outcome in the study, and the outcome measurements from several time points or use multiple
scale or unit of the outcome is rarely directly comparable across studies. outcome measures to assess related outcome constructs. All such
For these reasons, to facilitate cross‐study comparisons of the magni- cases yield a set of statistically dependent effect size estimates
tudes of studies effects in our analysis, we extracted data from each study (Borenstein et al., 2009b).
to calculate standardized effects sizes. We chose the appropriate We assessed the extent to which relationships exist across the
formulae for standardized effect size calculations in reference to, and studies included in the review. We avoided double counting of
dependent upon, the data provided in the included studies and the identical evidence by linking papers prior to data analysis. We used
outcome type (see Supporting Information: Appendix 8 for background information provided in the studies included to help support these
and methods sections, for details of the effect size formulae sheet). assessments, such as sample sizes, program characteristics and key
If different outcome types existed under the same outcome implementing and/or funding partners. Where we have several
construct (e.g., binary measure of employment and a measure of the publications reporting on the exact same effect, one main study was
number of hours worked off‐farm), for comparability of estimated used for data extraction, and the linked studies were stored to help
effect sizes, we converted estimates to the most common standard- any required search for further or missing information. To identify the
ized metric. We used common transformations outlined in Borenstein main study, priority was given to journal articles and, in the case of
and colleagues (2009a) for converting between different measures of multiple reports/working papers, the most recent one was selected.
standardized effects. We extracted effects reported across different interventions,
outcomes, and subgroups within a study. We addressed dependent
effect sizes using data processing and selection techniques. We used
4.3.6 | Unit of analysis issues several criteria to select one effect estimate per outcome per study
(further details of the criteria determining effect estimate selection
Unit of analysis errors can arise when the unit of allocation of a are available in Supporting Information: Appendix 9 for background
treatment is different to the unit of analysis of effect size estimate, and methods sections).
and this is not accounted for in the analysis (e.g., by clustering SEs at
the level of allocation). We assessed included studies for the
prevalence of these issues and, where they exist, account for them 4.3.8 | Dealing with missing data
by adjusting the reported SEs according to the following formula
(Higgins et al., 2020; Hedges, 2009): In instances where there is missing or incomplete data, we made
every effort to contact study authors to obtain the required
(d )′ = (d ) ∙ 1 + (m − 1) c, information. If we were unable to obtain the necessary data, we
reported the characteristics of the study but stated that it could not
where d is the effect size, m is the average number of observations be included in the meta‐analysis or reporting of effect sizes due to
per cluster, and c is the intra‐cluster correlation coefficient. If the missing data. In line with recommendations on collating data in
included studies used robust Huber‐White SEs to correct for systematic reviews from study authors (see Mullan et al., 2009), we
clustering, we calculated the SE of d by dividing d by the t‐statistic reported the number of studies for which authors were contacted,
on the coefficient of interest. We searched for an appropriate ICC in the information requested, any important details of the method of
the literature and if this was not available, we assumed the ICC to be eliciting information, and the response of authors to the request.
0.05, as also described in Waddington and colleagues (2014).

4.3.9 | Assessment of heterogeneity

4.3.7 | Criteria for determination of independent
findings In our analysis, we examined and discussed the distribution of
estimated effects across intervention and outcome types. We also
It is important our analysis accurately captures and reflects on statistically assessed heterogeneity by calculating the Q statistic, I2,
co‐dependencies between study estimates. This is because standard and τ2 to provide an estimate of the amount of variability in the
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 15 of 191

distribution of study effects sizes (Borenstein et al., 2009a). We methodological and substantive characteristics. Examples of these
complemented this assessment with a graphical analysis using forest categories include:
plots to illustrate the range of the standardized effects by
intervention. This approach provides a summary effect estimate • Extrinsic characteristics: funder of the study (e.g., NGO vs. private
(demoted by a diamond) with studies weighted by the precision of sector vs. government investments), publication type, publica-
the estimate using the inverse of the variance. Whenever feasible, we tion date.
conducted moderator analyses using meta‐regression to investigate • Methodological characteristics: study design, risk of bias, length of
potential sources of heterogeneity. follow‐up, types of outcome measures.
• Substantive characteristics: participant characteristics (gender,
age, socio‐economic status, education, land ownership, farm size),
4.3.10 | Assessment of reporting biases context (geographical setting, crop type), intervention type,
intervention features, type of implementing agency.
If meta‐analysis was feasible, we assessed reporting biases in the
literature using a rank correlation test (see Begg & Mazumdar, 1994). We used random effects meta‐regression to investigate the
We also used a regression test (Sterne & Egger, 2005), using the association between moderator variables and heterogeneity of
standard error of the observed outcomes as predictor, to check for treatment effects (Borenstein et al., 2009a), and subgroup analyses
funnel plot asymmetry. to investigate heterogeneity by treatment subgroups (e.g., men and
women, poor and non‐poor, and so on). If these strategies were not
possible (e.g., if we do not have sufficient number of studies or data),
4.4 | Data synthesis we discussed and explored the factors which may be driving the
heterogeneity of results narratively by conducting cross‐case
To synthesize the effects of market access interventions, we comparisons (Miles & Huberman, 1994).
combined a narrative synthesis of study findings with a meta‐
analysis of interventions effects.
We included studies in the same meta‐analysis when we 4.4.2 | Sensitivity analysis
identified two or more effect sizes using a similar outcome construct,
the same intervention type, and where the type of comparison group We conducted sensitivity analysis to assess whether the results of
was judged to be similar across the studies. This is similar to the the meta‐analysis were sensitive to the removal of any single study.
approach taken by Wilson and colleagues (2011). Where there are We did this by removing studies from the meta‐analysis one‐by‐one
too few studies, or the included studies were considered too and assessing changes in results. We also assessed the sensitivity of
heterogeneous in terms of interventions or outcomes, we presented our results to the inclusion of studies with a high risk of bias studies
a narrative discussion of individual effect sizes alone. by removing these studies from the meta‐analysis and comparing
Because heterogeneity exists in theory due to the variety of results to the main meta‐analysis results. We examined the sensitivity
interventions and contexts that could be included in the review, we of results to the inclusion of specific outcome measures (e.g., limiting
used inverse‐variance weighted, random effects meta‐analytic meta‐analysis to the preferred measure of an outcome). We also
models (Higgins et al., 2020). We used the metafor package assessed the sensitivity of our result to the inclusion of data obtained
(Viechtbauer, 2010) in R software to conduct the meta‐analyses (R directly from study authors (as discussed above).
Core Team, 2020). Furthermore, we assessed the sensitivity of our results to outliers.
The narrative synthesis of study findings provided information on We used studentized residuals to examine whether studies estimated
the relevant information on the associated factors facilitating and effects may be outliers (Viechtbauer & Cheung, 2010) and studies with
moderating the effects. This information was synthesized in order to a studentized residual larger than the 100 × (1 − 0.05/(2 × k ))th
possibly draw conclusions beyond the study level and design percentile of a standard normal distribution were considered as
learnings on the necessary conditions for positive effects to occur. potential outliers.
This analysis closely followed the meta‐analysis results and interpre- We also used both bivariate and multivariate (or partial) effects
tations and the subgroup and heterogeneity analysis described for calculating standardized effect sizes. A partial effect size is based
above. on a regression coefficient measuring the treatment effect “holding
all other variables constant” and is, therefore, measuring a different
quantity to a bivariate relationship. Our standardized effect sizes are
4.4.1 | Subgroup analysis and investigation of only strictly comparable in studies using a common model (Keef &
heterogeneity Roberts, 2004). However, only using bivariate effect sizes to calculate
standardized effects would not be suitable in this context due to the
Following the PROGRESS‐PLUS approach (Oliver et al., 2017), we likely high risk of bias. This may cause quasi‐experimental study
assessed moderators falling into three broad categories of extrinsic, designs that control selection bias (Waddington et al., 2014). We
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
16 of 191 | MARION ET AL.

used sensitivity analysis to examine systematic differences in partial After removing duplicate records in databases and registers, we
and bivariate effects (omitting them from the analysis or controlling identified 13,035 citations for screening at the title and abstract
for these characteristics in a meta‐regression). stage. We retained 12,605 records for full text screening. The EPPI
Finally, not all multivariate models control for the same classifier allowed us to directly exclude 6578 records, and 5087
covariates and nor should models estimated for different study records were excluded by human coders. Among the 940 included
designs using data collected in different contexts necessarily do so. studies, 50 records could not be retrieved, and within the retrieved
The risk of bias assessment evaluates likely specification errors and records, 771 were excluded.
the sensitivity analysis omitting high‐risk studies (discussed above) With other methods, 39,761 records were screened after
should capture most of these issues. Otherwise, we assume the identifying 10,834 duplicates. At the title and abstract screening
possible resulting multicollinearity issues are inconsequential (see stage, 38,237 records were excluded (Supporting Information:
Waddington et al., 2014). Appendix 11 for background and methods sections), leaving 1528
records for retrieval. We found 1458 eligible studies and excluded
1138 studies.
4.4.3 | Treatment of qualitative research

We do not include qualitative research. 5.1.2 | Included studies

We included 439 papers reporting on impact evaluations meeting our

4.4.4 | Summary of findings and assessment of the PICOS (Population, Intervention, Comparison, Outcomes, and Study
certainty of the evidence design) criteria. This report examines 289 studies and 262 unique
interventions across all intervention areas. The number of interven-
We do not include a summary of findings and assessment of the tions is higher than that of unique studies, as some studies included
certainty of the evidence. several interventions or linked to multiple publications.

Population
5 | RESULTS The study interventions were implemented in 53 LMICs (Figure 3).
They include 155 studies from Sub‐Saharan Africa, 41 studies from
We first describe the search results, providing information on the East Asia and the Pacific, 26 from Latin America and the Caribbean,
number of included studies at all stages of the search and screening 59 from South Asia, and seven from Europe and Central Asia. The
of the literature. We also provide an overview of the characteristics individual countries with the largest number of included interventions
and distribution of the included evidence base across all interventions are India (n = 36), Ethiopia (n = 23), Ghana (n = 22), China (n = 18), and
covered in the review, including the risk of bias of included impact Kenya (n = 17).
evaluations.
We then report the results of our appraisal and synthesis of the Interventions
studies' estimates of the effect of interventions in each outcome Figure 4 shows the distribution of included studies across interven-
type. To fully understand these results, we discuss the expected tion subcategories. Table 1 provides information on the five
effects according to the literature in each intervention type, then considered intervention types and subcategories. Across the five
present the number of included studies and estimates with important types of market access interventions, the intervention level with
summary statistics, followed by information on the estimates' risk of the fewest included studies is storage (n = 23), while contract farming
bias. For each outcome, we synthesize the included estimates. We is the largest intervention category (n = 130). When examining the
finish with a discussion section comprising a summary of results and subcategories, we also found an uneven spread of studies, with 50
implications for policymakers, implementers, and researchers. exclusively focused on domestic transport infrastructure, and only one
exclusively focused on improved storage technologies for transit or
export transport infrastructure. However, we counted a high number
5.1 | Description of studies of studies focusing on multi‐component interventions (n = 53).

5.1.1 | Results of the search Outcomes

We also found an uneven distribution of outcome categories across
Figure 2 provides the precise number of included and excluded our body of evidence (Figure 5). At the outcome category level, we
studies in different stages of the search and screening process observed clusters of evidence focusing on intermediate (n = 175),
(PRISMA Figure, Page et al., 2021). We searched 52,366 studies immediate (n = 158), and agricultural production (n = 127) outcomes.
across nine databases and 25 institutional websites, as well as the However, the body of evidence is smaller for categories related to
cited and referenced literature of included studies. non‐farm and off‐farm outcomes (n = 47 and n = 5, respectively). At the
18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
|17 of 191

Distribution of interventions by country.

PRISMA diagram.
ET AL.

FIGURE 2

FIGURE 3
MARION
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
18 of 191 | MARION ET AL.

FIGURE 4 Distribution of studies by intervention type. The figure accounts for individual components of multi‐component interventions
(n = 53).

FIGURE 5 Distribution of studies by outcome type.

outcome type level, we included a large number of studies focusing We contacted authors of 16 studies to ask for additional
on income‐related outcomes: 143 studies analyzed the effects of information on important details of the method used and the
interventions on farm income and 102 studies focused on total income reported effects. For four of these studies, authors provided an
and wealth. adequate response.

Study design
A large majority of the included studies used QEDs (n = 255) compared 5.1.3 | Excluded studies
to experimental designs (n = 28); the remainder (n = 6) were ongoing
studies. The methods most frequently employed by QEDs were statistical As presented in our flow chart (Figure 2), the main reasons for
matching (n = 102), followed by difference in differences (DID) (n = 84), exclusion at the title and abstract stage are related to the absence or
and endogenous treatment‐effect models (n = 35) (Figure 6). relevance of the intervention. At this stage, we excluded 7086
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 19 of 191

FIGURE 6 Distribution of studies by design.

records due to their absence of an intervention, and 4381 records parallel time trend assumption. Similarly, the chosen IVs were often
due to their focus on non‐relevant interventions. not exogenous and authors used endline time‐variant covariates in
At the full text screening stage, the main reason for exclusion was statistical matching estimations, which both contributed to a high risk
non‐suitability of the study design according to our framework; 748 of confounding effects of estimates in included studies. These studies
studies were excluded due to this criterion. The discrepancy between also have a high risk of selection bias. Included studies often lacked
the title and abstract stage and the full text stage on this criterion can precise information on the location of targeted beneficiaries,
be explained by the limited information provided in a title or abstract on especially for contract farming interventions. This made the identifi-
the specificities of the study design (e.g., experimental or quasi‐ cation of an adequate counterfactual more complicated and
experimental, use of counterfactual). Once again, the absence or significantly increased the risk of selection bias. In RCT studies, the
relevance of the interventions were the primary reasons for exclusion, percentage of studies with a risk of bias is much lower. Spill‐overs
with 262 and 390 records excluded, respectively. and contamination are a main source of bias, primarily due to the risk
Across the screening stages, 11,390 records were excluded as of the control group living in the same village as the beneficiary group
they were effectively duplicates of other included papers (e.g., a and thus possibly benefitting from the intervention which might
working paper version of a journal article, which did not contain result in an underestimation of the treatment effect.
additional information). We failed to retrieve 120 full texts as they
were either unavailable or inaccessible online.
5.3 | Effects of interventions

5.2 | Risk of bias in included studies 5.3.1 | Chapter 1: Farm‐to‐market transport

infrastructure interventions
Figure 7 presents a summary of the results of our risk of bias
assessments for all 1059 estimates. We found that 60% of included In this chapter, we systematically identify and synthesize evidence of
estimates were rated as having a high risk of bias; 31% were the effects of farm‐to‐market transport infrastructure interventions on
considered to have some concerns related to their risk of bias, and agricultural, economic, and food security and nutrition outcomes in
the remaining 9% were assessed as having a low risk of bias. LMICs. We also describe characteristics of the available evidence,
Figure 7 shows a discrepancy between the risk of bias of QED assess studies' risk of bias, determine the effects of these interven-
studies compared to RCT studies, as 61% of the former are rated as tions on our outcomes of interest, and provide information on the
high risk of bias, compared to 41% for the latter. Looking at individual reported costs of included interventions. Supporting Information:
risk of bias domains, the most common source of bias is related to Appendices 1 and 2 for Chapter 1 provide references for included
selection bias (26% of estimates were rated as high risk of bias) and studies and additional references, respectively.
confounding bias (36% of estimates). Additional concerns included
spillovers, crossover, and contamination (19% of estimates were How can farm‐to‐market transport infrastructure interventions
rated as high risk of bias), reporting bias (8% of estimates were rated affect agricultural, socio‐economic, food security, and nutrition
as high risk of bias), and outcome measurement bias (2% of estimates outcomes?
were rated as high risk of bias). Fewer than one percent of estimates The theory of change illustrates the logic of how transport
were assessed as high risk for performance bias. infrastructure may lead to observed changes at the output, outcome,
In QED studies, the risk of confounding bias is prevalent in and impact levels (Figure 8). Farm‐to‐market transport infrastructure
studies with a DiD design as authors often do not report testing for interventions include initiatives that improve infrastructure used to
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
20 of 191 | MARION ET AL.

FIGURE 7 Risk of bias of included estimates between quasi‐experimental designs (QEDs) (top) and randomized controlled trials (RCTs)
(bottom).

deliver agricultural produce to domestic and international markets, transporting agricultural goods from farms to markets, and providing
such as roads and bridges. Such initiatives can include constructing, access to more and better markets (Key et al., 2000; Jacoby &
rehabilitating, or maintaining infrastructure. Minten, 2009; Mu & van de Walle, 2007; Barrett, 2008; van de Walle,
Transport infrastructure may affect access to markets in several 2009). These transaction cost reductions can lead to direct effects,
ways. The main causal channel through which the construction or including improvements in beneficiaries' incomes and agricultural
rehabilitation of roads and bridges affects beneficiaries is by reducing production, as well as indirect effects such as changes to output
transaction costs (both time and financial resources) associated with market prices in the area and greater farmer access to key services.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 21 of 191

F I G U R E 8 Theory of change of farm‐to‐market infrastructure interventions. The outcome types included in this systematic review are
immediate outcomes (green), agricultural production outcomes (purple), intermediate outcomes (teal), off‐farm and non‐farm outcomes and
welfare outcomes (light blue).

Transporting produce to markets can be costly if farmers must elasticity of product demand and supply; moreover, prices can
rely on slow, poor‐quality transport networks that involve risk of change over time due to fluctuations associated with the interactions
spoilage and loss of product quality. Chamberlin and Jayne (2013) of supply and demand.
showed that transportation costs often comprise the largest share of Transport infrastructure interventions might increase the availa-
rural marketing costs in many LMICs. High transportation costs can in bility of agricultural products in markets, leading to a resultant
turn increase production costs, affecting the price of produce, sales, increase in their demand. Prices are expected to increase in the short
and ultimately, farm income (Kilkenny, 1998). run as demand exceeds supply. Farmers may then choose to switch
de Brauw and Bulte (2021) explained that such transaction costs from food crops to cash crops in order to exploit these price rises
and risks reduce benefits and incentives for farmers to trade. (Asfaw et al., 2019; Carletto et al., 2017; Cazzuffi et al., 2020).
Therefore, initiatives to lower transaction costs can render extant This demand effect is accompanied by a supply effect, which has a
market opportunities more lucrative, and can improve access to new competing effect on prices. Increases in the quantity of commercialized
or distant markets (Aggarwal et al., 2022; Ludwig et al., 2016). This in produce and the number of traders in local markets lead to greater
turn can translate into greater household income and improved competition and potential falls in prices. Such price reductions could be
welfare (Besley & Burgess, 2000; Fafchamps & Hill, 2005; Chamberlin detrimental to farmers, particularly cash crop producers. However,
& Jayne, 2013). At the impact level, increased household income may changes in farmers' products associated with transport infrastructure
allow beneficiary households to improve their dietary diversity, invest may also have positive implications on producers' access to credit
a greater number of resources into healthcare and education, and (Khandker et al., 2009) and the value of farm land (Gonzalez‐Navarro &
increase asset accumulation. Quintana‐Domeque, 2016). Namely, paved roads increase the value of
Greater access to markets, resulting from reduced transaction surrounding houses and land, leading to improvements in housing
and production costs, may also result in greater adoption of improved quality and size; this improved house and land value facilitates access to
inputs and better farm practices, with effects on agricultural credit due to the conversion of these assets into capital.
productivity. This higher production can lead to further participation In addition, a reduction in transaction costs would decrease the
in markets and greater income and welfare effects. In addition, price of imported or non‐local agricultural products, leading to lower
greater market participation can contribute to knowledge spillovers prices. For example, the price of imported fertilizer and seeds may
among farmers (de Janvry et al., 1991). decline as road quantity and quality increases, resulting in greater
Further, transport infrastructure that links markets can have adoption of improved practices (Asher & Novosad, 2020).
demand and supply effects on agricultural products, with a subse- Another indirect effect of farm‐to‐market transport infrastructure is
quent, indirect effect on agricultural prices. The supply and demand improvement in access to other key institutions and benefits such as
effects on prices are however variegated depending on the specific health care, education, and both agricultural and non‐agricultural labor
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
22 of 191 | MARION ET AL.

opportunities. Transport infrastructure interventions may increase the total of 61 of these 83 studies informed our synthesis‐of‐findings
number of consumed food types and improve nutrition by facilitating analysis on the effects of transport infrastructure interventions. The
access to more diverse food products in the local context (Tontisirin other 22 studies were either ongoing, previous versions, or unpublished
et al., 2002; Kennedy et al., 2007). In the long‐term, this improved versions of already‐included studies, and did not provide any additional
access to health and education may translate into productivity and results.
income effects (Bouis & Haddad, 1990; Strauss & Thomas, 1998). Of the 61 studies informing our analysis, 10 are included as
The effects of infrastructure interventions vary by type of linked studies reporting additional outcome data about the effects of
producer and type of farm‐to‐market transport infrastructure. For an intervention (i.e., data on outcomes or prioritized outcome
example, large road infrastructure interventions are likely to affect indicators that are not reported by the main identified study).7
producers in very different ways than the construction of rural feeder
roads. Major roads tend to connect remote areas to large urban areas, Interventions, inputs, and activities
and therefore do not link smaller remote communities together, with We identified studies on the effects of 41 farm‐to‐market transport
implications for access to markets, as well as pricing (Casaburi et al., infrastructure interventions implemented in 30 countries. The most
2013; Burgess et al., 2015; Gachassin et al., 2010). common location of included interventions is Africa (16, Figure 9),
Large‐scale infrastructure is likely to be less impactful for smaller followed by Latin America and the Caribbean (9), East Asia and the
or subsistence farmers (Narayana et al., 1988). Although smaller‐scale Pacific (7), South Asia (7), and Europe and Central Asia (2). Individual
infrastructure could improve access to markets and increase prices, countries with the largest number of included interventions are
subsistence farmers may remain unaffected, as they often lack Bangladesh, India, Kenya, and Vietnam (each with 3) and China,
sufficient assets such as vehicles to exploit the reduction in Nicaragua, and Nigeria (each with 2).
transaction costs and to reach markets. It is equally important to The most common types of infrastructure are roads (28
recall that the type of infrastructure matters, since infrastructure interventions) and bridges (five interventions); a further eight
such as ports could have higher effects on the prices of imported and interventions combined investments in both road and bridge
exported commodities, whereas other types of domestic infra- infrastructure, and one built roads and waterway infrastructure
structure could be more effective in boosting domestic trade. (Appui au Développement du Menabe et du Melaky [AD2M] in
Finally, we briefly discuss unintended negative consequences of Madagascar). The included interventions also vary according to the
farm‐to‐market transport infrastructure. Price changes caused by these purpose of their infrastructure investment. Supporting Information:
interventions are likely to affect production decisions. Farmers are Appendix 3 for Chapter 1 provides additional information on the 41
likely to allocate more land for non‐food cash crops production that, interventions including country, type of infrastructure, type of
if not accompanied by improved income, might put them at risk of improvement, and intervention year. This information can also be
neglecting food crops, which are often essential for household food found in Table 4, below.
security and nutrition. A total of 19 interventions involve the construction of new
This is particularly relevant for non‐food, export‐oriented cash infrastructure (46%); 14 interventions consist of rehabilitated
crop producers, who are particularly exposed to price fluctuations in infrastructure (34%); 3 interventions combine infrastructure con-
international markets. For instance, Wood and colleagues (2013) struction and maintenance; 2 combine rehabilitation and mainte-
found that greater access to markets led farmers in Malawi to nance; and 3 combine infrastructure construction and rehabilitation
allocate most of their resources to the cultivation of tobacco, which (Table 4). We identified two studies (Hayakawa et al., 2020, 2022)
caused a decline in child anthropometric measures in the presence of that had a specific cross‐country focus; they analyzed the effects of
a domestic staple price shock. Miller and Urdinola (2010) found that the TLFB, which aimed to facilitate cross‐border links between
child mortality in Colombia was procyclical with coffee prices. Thailand and Laos.
Similarly, greater market participation may improve children's Several of the included transport infrastructure interventions
educational opportunities, as these interventions may offer greater were multi‐component, combining infrastructure investments with
labor market opportunities for children and encourage parents to other types of interventions (n = 11). This reflects a long‐standing
stop their schooling temporarily or prematurely (Atkin, 2016). practice in agricultural development policy, whereby projects
combine mutually supporting agricultural and non‐agricultural activi-
Description of included studies ties to address multiple constraints to growth. For example, AD2M
Overall, we included 83 studies examining the effects of 41 intervention in Madagascar combined the construction of road and
infrastructure interventions. This included one ongoing study consisting waterway infrastructure with the promotion of technologies and
of a pre‐analysis plan for a mixed methods impact evaluation of a practices, land certification, functional literacy, and hydro‐agricultural
Bridges to Prosperity (B2P) project in Rwanda (Macharia, 2022).6 A development.

6 7
Interventions were considered at the country level. We obtained 37 unique transport Where more than one published study reported outcomes for the same intervention, the
infrastructure interventions with B2P and the Millennium Village Project in two countries study with the longest follow up is selected as the “main” study, and the others are reported
each (the former in Nicaragua and Rwanda and the latter in Ghana and Kenya). as “linked studies.”
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 23 of 191

FIGURE 9 Distribution of interventions by region.

Many of these transport infrastructure investments (e.g., AD2M, (including 3 linked studies and 15 evaluated interventions). Evidence
FAD‐2, FAD‐3) are also accompanied by production value addition, on farm investment was provided by a total of 8 main studies and 3
and enterprise or marketing trainings for capacity development. linked studies focused on 10 interventions. A total of 7 studies,
These activities are jointly intended to help participants exploit new including 1 linked study, analyzed the effect of 6 interventions on
market opportunities (interventions, Table 5). In the analysis below, credit outcomes; 9 studies, including 2 linked studies, focused on the
we conduct a moderator analysis to determine whether the estimates price outcomes of 7 interventions.
from multi‐component interventions are on average statistically In terms of agricultural production outcomes, none of the included
significantly different from those of single‐component interventions. studies focused on crop loss and quality. Sixteen studies (including 1
linked study), 17 studies (including 2 linked studies), and 15 studies
Comparison (including 1 linked study) provided estimates on yields, volume, and
The comparison groups used for these estimations focused nearly produce type, respectively. The number of interventions with yields,
exclusively on comparing the outcomes of beneficiaries of transport volume, and produce type outcomes was 15, 16, and 15, respectively.
infrastructure investments to those of non‐beneficiaries located in a Estimates of intervention effects on farm income are reported for
different area. Six percent of studies (n = 4) exploited a phased‐in (or 18 interventions from 19 studies. For sales outcomes, 10 main studies
pipeline) intervention approach to identify a comparison group. One (including 1 linked study) provided estimates for 10 interventions. For
study (Aggarwal, 2018) compared districts with varying treatment off‐farm agricultural outcomes, estimates were available on off‐farm
intensities to estimate the effects of the PMGSY intervention. income (e.g., wage agricultural labor) from 3 studies (including 1 linked
Treatment intensity was defined as the percentage of the district's study) for 2 interventions; estimates on off‐farm labor were available
village populations that received a PMGSY road. Supporting from 16 studies (including 6 linked studies) for 11 interventions. For
Information: Appendix 4 for Chapter 1 provides information on non‐farm outcomes, we found 15 studies (including 2 linked studies)
these studies, including the intervention evaluated, its subcategory, with evidence on non‐farm labor outcomes for 15 interventions, and
the length of follow‐up period (in years), study design, and type of 13 studies (including 2 linked studies) with evidence on non‐farm
comparison. labor outcomes for 12 interventions. Figure 10 provides a summary of
the outcomes across the included studies.
Outcomes
The included studies provide evidence of the effects of farm‐to‐ Study design
market transport infrastructure interventions on welfare outcomes None of the 61 studies included in our analysis were RCTs. All studies
related to total income and wealth (25 main studies and 3 linked used a natural experiment or a quasi‐experimental study design to
studies) for 26 interventions, and related to food security and nutrition evaluate the effects of transport infrastructure interventions. This
(10 main studies and 2 linked studies) for 12 interventions. reflects the practical difficulty in randomizing the placement of roads
For intermediate outcomes, the effects on transaction costs are and other infrastructure construction. Furthermore, much like
available for 9 interventions from 5 main studies and 4 linked studies. infrastructure construction investments, maintenance and rehabilita-
One single main study provided an estimate on group participation tion efforts are generally strategically targeted at areas with the
outcomes. A total of 16 studies provided evidence on outcomes of greatest need and value (e.g., because of deprivation, degradation of
the use or adoption of improved inputs, technologies, and practices the infrastructure, or its local economic importance and potential).
 TABLE 4 Included farm‐to‐market transport infrastructure interventions.

First year of
24 of 191

Intervention name Acronym Country (ISO3 code) Type Improvement type intervention
|

Transport Sector Investment Program TSIP Tanzania (TZA) Road Rehabilitation 2010

Access Roads Program (Program for Rehabilitation of Access ARP Trinidad and Tobago (TTO) Road Rehabilitation 1994
Roads and Reconstruction of Bridges)

Appui au Développement du Menabe et du Melaky AD2M Madagascar (MDG) Road; waterway Construction 2007
infrastructure

Armando Emilio Guebuza (Zambezi) Bridge Investment AEGBI Mozambique (MOZ) Bridge Construction 2009

Babeldaob Road BR Palau (PLW) Road Construction 2001

Beni‐Jomsom‐Sadak Road BJS Nepal (NPL) Road Construction 2008

Bridges to Prosperity B2P Nicaragua (NIC) Bridge Construction 2015

Bridges to Prosperity B2P Rwanda (RWA) Bridge Construction 2019

Community‐level infrastructure rehabilitation projects CLIRP Georgia (GEO) Road; bridge Rehabilitation 1998

Corredores de la Paz Program CPP Colombia (COL) Road Construction 2000

Chimborazo Rural Investment Project CRIP Ecuador (ECU) Road Rehabilitation 2015

E‐384 y E‐585 E‐384 y E‐585 Ecuador (ECU) Road Rehabilitation 2017

EU Rural Feeder Roads Rehabilitation Program RFRRP Sierra Leone (SLE) Road Rehabilitation 2009

Eastern Province Feeder Road Project EPFRP Zambia (ZMB) Road Rehabilitation 1996

Fadama II FAD‐2 Nigeria (NGA) Road Construction; maintenance 2005

Fadama III FAD‐3 Nigeria (NGA) Road Construction; maintenance 2013

Farmer Support Program FSP South Africa (ZAF) Road Construction 1987

Food Production, Processing, and Marketing Project FPPM Democratic Republic of the Road, bridge Rehabilitation 2011
Congo (COD)

Golden Quadrilateral Highway Investment GQ India (IND) Road Construction 2001

Guangxi Integrated Agricultural Development Project GIADP China (CHN) Road Construction 2012

Irish Aid Infrastructure Project in Vietnam IAIV Vietnam (VNM) Road Rehabilitation 2011

Jamuna Bridge Investment JBI Bangladesh (BGD) Bridge Construction 1998

Millennium Challenge Corporation Connectivity Project MCCCP El Salvador (SLV) Road Rehabilitation 2009

Millennium Village Project MVP Ghana (GHA) Road Construction; 2012

rehabilitation
MARION
ET AL.

18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
 TABLE 4 (Continued)
MARION

First year of
Intervention name Acronym Country (ISO3 code) Type Improvement type intervention
ET AL.

Millennium Village Project MVP Kenya (KEN) Road Rehabilitation 2005

North‐South‐East‐West Highway Project NSEW India (IND) Road Construction 2001

Pradhan Mantri Gram Sadak Yojana (Prime Minister's Rural Road PMGSY India (IND) Road Construction 2000
Scheme)

Programa de Servicios Agropecuarios Provinciales PROSAP Argentina (ARG) Road Construction 2006

Promoting Food Security in South East Liberia through LIBA‐72 Liberia (LBR) Road; bridge Rehabilitation 2010
Commercial Rice Value‐Chain Development

Puding County Road Investments PCRI China (CHN) Road Construction 2004

Rural Development Project RDP Bangladesh (BGD) Road; bridge Construction; 1995
rehabilitation

Rural Road Rehabilitation Project RRRP Armenia (ARM) Road Rehabilitation 2009

Rural Road Rehabilitation and Maintenance Program PCR Peru (PER) Road Maintenance; 1996
rehabilitation

Rural Roads and Markets Improvement and Maintenance Project RRMIMP Bangladesh (BGD) Road; bridge Construction; 1995
rehabilitation

Smallholder Dairy Commercialization Program SDCP Kenya (KEN) Road Construction 2006

Smallholder Horticulture Marketing Program ShoMaP Kenya (KEN) Road; bridge Construction 2008

Thai Lao Friendship Bridge Construction TLFB Thailand (THA) Bridge Construction 2007

Third Rural Transport Project RTP‐3 Vietnam (VNM) Road Maintenance; 2008
rehabilitation

Transport Sector Support Program PAST Nicaragua (NIC) Road; bridge Construction; maintenance 1999

Universal Rural Road Access Program URRAP Ethiopia (ETH) Road Construction 2010

Vietnam Rural Transport Project I RTPI Vietnam (VNM) Road Rehabilitation 1997

Note: ISO3 country code provided in parentheses (). If the introduction of an intervention is staggered, the year of intervention refers to the earliest year that observations were undertaken for the intervention.
| 25 of 191

18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
26 of 191 | MARION ET AL.

TABLE 5 Descriptions of other interventions combined with transport infrastructure interventions.

Intervention (country Description of other intervention components combined with transport infrastructure
code)a Type interventions

AD2M (MDG) Road; waterway Other infrastructure, promotion of technologies and practices, land certificates,
infrastructure functional literacy, and hydro‐agricultural development

FAD‐2 (NGA) Road Other infrastructure (culverts, market infrastructure, post‐harvest infrastructure),
capacity development and advisory services

FAD‐3 (NGA) Road Capacity development and advisory services

FPPM (COD) Road; bridge Access to inputs, services, and markets

GIADP (CHN) Road Irrigation infrastructure, agricultural production and marketing, support, and rural
environmental improvement

LIBA‐72 (LBR) Road; bridge Training, inputs, and technical support

MVP (GHA) Road Interventions in health, education, agriculture, infrastructure, energy, water, and
sanitation sectors

MVP (KEN) Road Training in improved agricultural techniques and received seeds and fertilizers, input
subsidy, microfinance loans, rehabilitation of school facilities, development and
rehabilitation of health centers, development of water sources (such as protected
springs), and connection of local institutions to the electric grid

PROSAP (ARG) Road Electrification

SDCP (KEN) Road Training on organizational, managerial and enterprise skills; milk‐handling practices; and
value‐added opportunities

ShoMaP (KEN) Road; bridge Domestic market systems analysis, institutional strengthening, and grants to farmers
a
ISO3 country code provided in parentheses ().

Most studies rely on at least two rounds of outcome data, and their included in the study's analysis. The average follow‐up period of studies
identification strategies are primarily determined using DID and FE included in the analysis is 6.5 years. The longest reported follow‐up
regressions (36 studies, 71%). The second most prevalent design period is 24 years (Sikwela, 2013), while the shortest is seven months
employed in the included studies is cross‐sectional statistical (Thomas et al., 2021). Most studies include a follow‐up period of less
matching approaches (10 studies, 20%) (Supporting Information: than eight years (n = 33), with the largest concentration of studies
Appendix 4 for Chapter 1). examining outcomes between four to six years after the intervention
Other less common types of design in this sample of studies commenced (n = 15). However, a relatively large cluster of included
include IVs (3 studies, 5%), regression discontinuity design (4 studies, studies (n = 12) are also focused on interventions' longer‐term effects—
7%) and interrupted time series (2 studies) (Jones & Salazar, 2021; namely, 10 or more years after their inception.
Salazar‐Espinoza & Jones, 2017).8 However, designs—such as IV
design, regression discontinuity design and interrupted time series Synthesis of findings
design—are also generally less common in the broader evaluation The data extraction of studies under the farm‐to‐market transport
literature due to the data requirements often necessary to create a infrastructure intervention category identified a total of 2129
convincing application of the approach. estimates—232 of which were selected for their analysis of effects
For example, regression discontinuity design is based on a based on criteria described in Supporting Information: Appendix 9 for
discontinuity or threshold determining access to transport infra- the background and methods sections guiding the selection of
structure. Transport infrastructure investments are generally public independent estimates. In the following section, we first provide an
goods; therefore, their access is not restrictive. Interrupted time overview of the risk of bias assessment conducted on the included
series analysis requires that data are available for numerous time estimates and provide an analysis of the effects of farm‐to‐market
periods both before and after an intervention is implemented, which transport infrastructure on covered outcomes. We report additional
can be a challenging data requirement to satisfy. information on the individual meta‐analyses (sensitivity analysis,
Figure 11 shows the distribution of the length of studies' different outliers' analysis, publication bias analysis, moderator analysis, meta‐
follow‐up periods (study periods). This is the difference between the analysis without high risk of bias estimates) in Supporting Informa-
start date of the intervention and the most recent outcome data tion: Appendix 5 for Chapter 1.

8 Risk of bias. Our assessments of studies' risk of bias identify whether

Seven studies used multiple designs across the different estimated effects of transport
infrastructure. aspects of their design, implementation, analysis, or reporting may
18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
|27 of 191

Distribution of outcome types.

F I G U R E 10
ET AL.
MARION
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
28 of 191 | MARION ET AL.

F I G U R E 11 Distribution of the length of studies' follow‐up periods.

F I G U R E 12 Risk of bias of included estimates.

potentially bias their estimates of the intervention's effects (Support- This risk of bias of included estimates was based on six domains
ing Information: Appendix 10 for background and methods sections). (selection bias; confounding; performance bias; spillovers, crossovers,
Figure 12 presents a summary of the results of our risk of bias and contamination; outcome measurement bias; and reporting bias).
assessments. We found that 49% of included estimates were rated as Looking at individual risk of bias domains, the most common source
having a high risk of bias, 26% were considered to have some of bias is related to confounding bias (30% of estimates were rated as
concerns related to risk of bias, and the remaining 25% were assessed high risk of bias), which often had to do with either lack of, or
as having a low risk of bias. unreported, balance at baseline. Additional concerns included
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 29 of 191

spillovers, crossovers, and contamination (19% of estimates were resulting in a large reduction in transaction costs (Casaburi et al.,
rated as high risk of bias), and reporting bias (3% of estimates were 2013; ĝ = 0.88; 95% CI: 0.18 to 1.59; p = 0.01).
rated as high risk of bias). Estimates were not likely to be at high risk Torero and colleagues (2018) did not find that MCCCP led to a
for outcome measurement bias and performance bias. reduction in self‐reported travel costs to the nearest town market
(ĝ = 0.05; 95% CI: −0.04 to 0.14; p = 0.26). They argued that the
Summary of the meta‐analyses results of farm‐to‐market transport surveyed households most likely did not use the constructed roads to
infrastructure interventions. Table 6 displays the meta‐analyses access small town markets as the intervention focused on a highway
results across outcome types covered in subsequent sections. road connecting the Southern part of El Salvador to the rest of the
country. Households likely continued to use minor roads or would
Immediate outcomes. The immediate outcomes category covers the have accessed the town market by walking. In Bangladesh, Khandker
first outcomes that may affect farmers across the causal chain. They and Koolwal (2011) estimated that the RRMIMP (improvement of
cover financial resources (e.g., transaction costs, and credit), market feeder roads to bitumen‐surfaced standard) led to a decrease in self‐
linkages and bargaining power (e.g., group participation, and prices), reported household‐level transport costs per trip in the rainy season
and the adoption of new practices and agricultural approaches (e.g., (ĝ = 0.10 ; 95% CI: 0.01 to 0.21; p = 0.08) but not in the dry season
use/adoption of improved inputs, technologies and practices, and farm (ĝ = −0.02; 95% CI: −0.12 to 0.10; p = 0.83). In the meta‐analysis
investments). above, we calculated the synthetic effect size ( ĝ = 0.04 ; 95% CI:
−0.05 to 0.14; p = 0.38).
Transaction costs. Within nine studies, we identified nine independent None of the moderator variables significantly contributed to the
estimates of the effect of transport infrastructure on transaction costs variation in estimates. Of the nine included effect sizes, seven effect
outcomes. Studies of evaluated interventions took place in several sizes remain after excluding high risk of bias estimates. The effect of
regions: Africa (n = 4), East Asia and Pacific (n = 1), Latin America and farm‐to‐market transport infrastructure interventions on transaction
the Caribbean (n = 1) and South Asia (n = 3). Of the nine studies, three ˆ = 0.11;
costs outcomes is still small and statistically significant (SMD
evaluated multi‐component interventions (FPPM, ShoMaP and FAD‐2). 95% CI: 0.02 to 0.20; p = 0.02).
Two interventions‐built bridges (AEGBI and TLFB), while three others
constructed both roads and bridges (RDP, RRMIMP and ShoMaP). The Use/adoption of fertilizers, pesticides, herbicides and other improved
remaining interventions rehabilitated roads (MCCCP and RFRRP) or inputs. Fourteen studies, providing 15 estimates, examined the
provided new roads (PMGSY and FAD‐2). effects of 14 transport infrastructure interventions on use/adoption of
All studies here use different measures of self‐reported transport fertilizers, pesticides, herbicides and other improved inputs. Eight studies
costs to the market at the household‐level (k = 7) and objective were focused on Africa (n = 8), two in East Asia and Pacific (n = 2),
transport costs at the market‐level (k = 2 from Casaburi et al., 2013; two in Latin America and the Caribbean (n = 2), and two in South Asia
Zant, 2021). In Zant (2021), transport costs were determined (n = 2). We found two independent estimates in Ragasa and
between markets based on road quality and road distance in South colleagues (2016) as the analysis is conducted in two provinces of
Mozambique. Casaburi and colleagues (2013) used the one‐way Democratic Republic of Congo (DRC) (Bandundu and Kinshasa) and
motorbike transport fares per kilometer along all 47 rehabilitated each analysis includes an independent farmer control.
RFRRP roads in Sierra Leone. Of these studies, seven evaluated multi‐component interven-
Farm‐to‐market transport infrastructure interventions reduced tions (AD2M, FAD‐2, FPPM, GIADP, LIBA‐72, MVP, and SDCP) and
transaction costs of program beneficiaries compared to non‐ the remaining seven were single component (ARP, JBI, MCCCP, PCRI,
beneficiaries. The overall weighted average effect is very small, but PMGSY, TSIP and URRAP). Multi‐component interventions combined
positive and statistically significant (standardized mean difference road construction and/or rehabilitation with additional activities
ˆ = 0.09; 95% CI: 0.02 to 0.16; p = 0.02) (Figure 13). The effect
[SMD] benefiting households engaged in agriculture such as GIADP which
sizes here are all reversed meaning that a higher effect represents also built irrigation infrastructure, agricultural production and market-
lower transaction costs. Effect sizes vary between 0.01 and 0.88 ing, support, and rural environmental improvement.
standard deviations. The heterogeneity among the studies is Nine estimates measured the impact of transport infrastructure
substantial (Q(8) = 37.73, p < 0.001, τ2 = 0.008, I2 = 78.80). interventions on use/adoption of inorganic fertilizers. For example,
The estimates from Madu and Phoa (2012) and Casaburi and Aggarwal (2018) used the area under fertilizer in hectare as an
colleagues (2013) are statistically significant and led to a large outcome variable. The outcome variable in Qin and Zhang (2016) was
reduction effect on transactions costs. In Nigeria, households' fertilizer use (yuan per mu). Barnett and colleagues (2018) provided
reported cost of transporting products on a truck declined with an estimate of the impact of MVP on adoption of fertilizer. Three
FAD‐2 (Madu & Phoa, 2012; ĝ = 0.66; 95% CI: 0.43 to 0.89; other estimates focused on the impact on pesticides use and
9
p < 0.001). RFRRP improved the quality of roads in Sierra Leone adoption (e.g., Pesticide use in kilogram per hectare in Garbero &
Songsermsawas, 2018). The remaining three binary outcome vari-
9 ables related to adoption of livestock vaccination service (Bonilla
Information on interventions is provided in Supporting Information: Appendix 3 for
Chapter 1. et al., 2018), adoption of weeding (Fuller, 2014), and adoption of soil
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
30 of 191 | MARION ET AL.

TABLE 6 Summary of the meta‐analyses results of farm‐to‐market transport infrastructure interventions.

Outcome types Outcome subcategory SMD 95% CI k p Value

Immediate outcomes Transaction costs 0.09 0.02, 0.16 9 0.02

Group participation N/A

Use/adoption of fertilizers, 0.08 0.03, 0.14 15 0.003

pesticides, herbicides, and other
improved inputs

Use/adoption of improved 0.05 0.02, 0.08 6 0.004

varieties and quality seeds

Use/adoption of labor and 0.001 −0.08, 0.08 6 0.95

machinery

Farm investment (total) N/A

Farm investment (seeds) 0.03 −0.03, 0.08 3 0.33

Farm investment (fertilizers) 0.03 −0.02, 0.07 5 0.22

Farm investment (labor and 0.30 0.05, 0.55 7 0.02

machinery)

Credit 0.04 −0.03, 0.11 8 0.30

Prices a
0.04 −0.05, 0.14 8 0.34

Agricultural production Yields 0.13 0.04, 0.22 15 0.003

outcomes
Volume 0.04 −0.02, 0.10 16 0.18

Produce type (area) −0.01 −0.10, 0.08 10 0.84

Produce type (livestock) 0.09 0.04, 0.14 5 <0.001

Produce type (diversity) N/A

Intermediate outcomes Sales 0.001 −0.05, 0.06 10 0.92

Farm income 0.08 0.04, 0.13 19 <0.001

Off‐farm outcomes Off‐farm income −0.07 −0.27, 0.13 3 0.49

(agriculture only)
Labor market outcomes −0.01 −0.12, 0.10 12 0.85

Non‐farm outcomes Non‐farm income 0.08 0.02, 0.14 16 0.01

Labor market outcomes 0.07 0.02, 0.11 13 0.003

Welfare outcomes Total income and wealth 0.05 0.02, 0.08 27 0.004

Food security 0.02 −0.04, 0.09 11 0.46

Nutrition 0.03 −0.08, 0.14 3 0.63

Note: Outcome subcategories in bold have a statistically significant SMD at a 5% significance level. N/A indicates that we did not conduct a meta‐analysis
for these outcome types. In these cases, k = 1 or k = 2 and we chose not to conduct a meta‐analysis in those cases due to differences (e.g., intervention
types, regions).
Abbreviations: k, number of estimates included in the analysis; SMD, standardized mean difference.
a
For prices received by the farmers, yields, volume and sales outcomes, within included studies, we found distinct effects of the considered interventions on
several crops. Results in this table are based on the selection of the effects of the interventions on the targeted crops, or when targeted crops were not
specified, the effects on the most prevalent crops in the local context. For these outcomes, we conduct additional analysis focusing on the effects of
the interventions on other crops not selected here. Statistical significance of the overall weighted average effects do not change (except for sales) and the
magnitude of these effects do not substantially change. Results are available upon request. For the other chapters, only within a few included studies, we
found distinct effects of the interventions on multiple crops. Therefore, we only conduct this type of additional analysis for chapter 1 on outcomes prices,
yields, volume and sales.

fertility management (Nkonya et al., 2008). All outcome variables Farm‐to‐market transport infrastructure interventions had a very
across these 15 estimates are at the household level, except the small positive and statistically significant effect on use and adoption
estimate from Blankespoor and colleagues (2018) which was at the of fertilizers, pesticides, herbicides and other improved inputs. The
sub‐district or upazila in Bangladesh (outcome variable: share of land overall weighted average effect is 0.08 standard deviations (95% CI:
under chemical fertilizer). 0.03 to 0.14; p = 0.001) (Figure 14), suggesting a very small benefit to
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 31 of 191

F I G U R E 13 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of farm‐to‐market transport
infrastructure interventions on transaction costs.

F I G U R E 14 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of farm‐to‐market transport
infrastructure interventions on use/adoption of fertilizers, pesticides, herbicides and other improved inputs.

the intervention beneficiaries compared to the control participants. significant. Blankespoor and colleagues (2018) reported the largest
The effect sizes vary considerably from −0.06 standard deviations effect, finding that the Jamuna Bridge led to increased share of land
(95% CI: −0.28 to 0.16; p = 0.55; Ragasa et al., 2016) to 0.82 standard under chemical fertilizer in the beneficiary upazilas following
deviations (95% CI: 0.55 to 1.10; p < 0.001; Blankespoor et al., 2018) significant reduction in transport costs.
and there is substantial heterogeneity across included estimates (Q The moderator analysis indicated that effects were greater in
(14) = 106.55, p < 0.001, τ̂ 2 = 0.01, I2 = 86.86%). The effects from South Asia than in Africa and that more recent interventions seem to
Ragasa and colleagues (2016) are both negative but not statistically be realizing smaller effects. We did not identify any funnel plot
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
32 of 191 | MARION ET AL.

asymmetry meaning that publication bias is not likely present. Eight of high‐yield seeds of root crops and tubers, the results are the same
of the included estimates are assessed as high risk of bias. After for other crops.
removing estimates with a high risk of bias, seven estimates remain in None of the moderator variables significantly contributed to the
the analysis. The overall weighted effect is still small positive and variation in estimates. As five of the six included estimates are
ˆ = 0.12; 95% CI: 0.02 to 0.22; p = 0.02).
statistically significant (SMD assessed as high risk of bias, we could not conduct sensitivity
analyses by study quality. The estimate from Charles (2010) was
Use/adoption of improved varieties and quality seeds. We found six considered low risk of bias and the magnitude of the estimate was
independent effect sizes from four studies on the effect of transport discussed above.
infrastructure on use/adoption of improved varieties and quality seeds.
These studies are from Africa (n = 2), Latin America and the Use/adoption of labor and machinery. In terms of estimates on use/
Caribbean (n = 1) and South Asia (n = 1). We found three indepen- adoption of labor and machinery, we found six independent estimates
dent estimates in Ragasa and colleagues (2016) as the analysis is from four studies (Jirgi et al., 2019, Nigeria, FAD‐3; Dumas &
conducted in three provinces of DRC (Bandundu, Bas‐Congo and Játiva, 2020, Tanzania, TSIP; Charles, 2010, Trinidad and Tobago,
Kinshasa) and each analysis includes an independent farmer control. ARP; Ragasa et al., 2016, DRC, FPPM). Three estimates are from
These four studies examined the effect of transport infrastructure on Ragasa and colleagues as in the previous outcome. The outcome
household‐level measures of use (area under hybrid seeds in hectare variable in Jirgi and colleagues (2019) was number of man days used
in Aggarwal, 2018) or adoption of improved seed varieties (Charles, on their own farm. Another outcome variable here is the use of jeeps,
2010; Nkonya et al., 2008; Ragasa et al., 2016). trucks and vans (Charles, 2010). Dumas and Játiva (2020) used the
Farm‐to‐market transport infrastructure interventions increased number of household members that worked on farms over the last
the use and adoption of improved varieties and quality seeds. Overall, year as the outcome variable.
transport infrastructure had a small positive but statistically Farm‐to‐market transport infrastructure interventions had no effect
insignificant effect on use/adoption of improved varieties and quality on use/adoption of labor and machinery. Figure 16 presents the forest
seeds (SMDˆ = 0.05; 95% CI: 0.02 to 0.08; p = 0.004). Figure 15 shows plot with these six estimates and the overall weighted average effect
the forest plot of these six effect sizes. Effect sizes vary from 0.01 to (SMD = −0.00; 95% CI: −0.08 to 0.08; p = 0.95). Effect sizes vary from
0.13 standard deviations. There was no significant amount of −0.59 to 0.09 standard deviations. The amount of heterogeneity
heterogeneity among the studies (Q(5) = 1.93, p = 0.859, among the estimates was moderate (Q (5) = 7.70, p = 0.17, τ̂ 2 =
τ̂ 2 < 0.001, I2 < 0.001%). The estimate from Aggarwal (2018) is 0.003, I2 = 35.09%). Dumas and Játiva (2020) found that TSIP
statistically significant ( ĝ = 0.05; 95% CI: 0.01 to 0.09; p = 0.01). intervention (roads were upgraded to paved standards) reduced the
Aggarwal (2018) found that the road construction in India allocation of household labor on their own plots ( ĝ = −0.07; 95% CI:
(PMGSY) lead to a two percent increase in the area under cultivation −0.13 to −0.001; p = 0.04).
with hybrid seeds ( ĝ = 0.05; 95% CI: 0.01 to 0.09; p = 0.01). Nkonya None of the moderator variables significantly contributed to the
and colleagues (2008) and Charles (2010) did not distinguish positive variation in estimates. If we focus our analysis exclusively on
impacts from transport infrastructure on adoption of improved estimates without a high risk of bias, two estimates remain. These
varieties and quality seeds ( ĝ = 0.01; 95% CI: −0.08 to 0.10; estimates are from Charles (2010) and Jirgi and colleagues (2019).
p = 0.84 and ĝ = 0.13; 95% CI: −0.02 to 0.29; p = 0.14, respectively). The overall weighted average effect becomes statistically
Although the estimate here from Charles (2010) relates to adoption insignificant (SMD = −0.14; 95% CI: −0.77 to 0.49; p = 0.66), but this

F I G U R E 15 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of farm‐to‐market transport
infrastructure interventions on use/adoption of improved varieties and quality seeds.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 33 of 191

result should be interpreted with caution as it is based on a very small We found moderate heterogeneity among the estimates
number of effects. (Q(2) = 2.88, p = 0.24 , τ̂ 2 = 0.001, I2 = 30.47%), which had effect
sizes ranging from −0.09 to 0.06 standard deviations. Shamdasani
Farm investment (seeds). Three studies provided four independent (2021) found that PMGSY in India resulted in a very small positive
estimates of the impact of transport infrastructure on farm investments and statistically significant effect on households' purchase of high‐
in seeds. These three studies originate from interventions in different yielding varieties seeds through greater access to input markets
regions (Europe and Central Asia, n = 1; South Asia, n = 1; Latin (ĝ = 0.06; 95% CI: 0.002 to 0.12 p = 0.04). The estimate from Fortson
America and the Caribbean, n = 1). All included interventions here are and colleagues (2015, ĝ = 0.004; 95% CI: −0.05 to 0.06; p = 0.89) was
single component interventions meaning that they only focused on statistically insignificant.
building, improving and rehabilitating roads and bridges. None of the moderator variables significantly contributed to the
Outcome variables were continuous and binary variables linked variation in estimates. If we focus our analysis exclusively on
to farm investments in seeds. For example, Fortson and colleagues estimates without a high risk of bias, two estimates remain. These
(2015) had a variable for seeds expenditure as an outcome variable. estimates are from Shamdasani (2021) and Corral and Zane (2021).
In Shamdasani (2021), the outcome variable is a binary taking the The overall weighted average effect remains statistically insignificant
value one if households purchased high‐yielding variety seeds and (SMD = 0.03, 95% CI: −0.08 to 0.15; p = 0.57).
zero otherwise. Corral and Zane (2021) used a binary variable taking
the value one if households paid for seeds and zero otherwise. Farm investment (fertilizers). Five studies provided four independent
Farm‐to‐market transport infrastructure interventions did not affect estimates of the impact of transport infrastructure on farm investments
farm investments in seeds. in fertilizers. These four studies originate from interventions in
Figure 17 displays these three effect sizes. These interventions different regions (Africa, n = 1; Europe and Central Asia, n = 1; East
had a very small positive but statistically insignificant effect on farm Asia and Pacific, n = 1; Latin America and the Caribbean, n = 2). All
ˆ = 0.03; 95% CI: −0.03 to 0.08; p = 0.33).
investments in seeds (SMD included interventions here are single component interventions

F I G U R E 16 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of farm‐to‐market transport
infrastructure interventions on use/adoption of labor and machinery.

F I G U R E 17 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of farm‐to‐market transport
infrastructure interventions on farm investment (seeds).
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
34 of 191 | MARION ET AL.

meaning that they only focused on building, improving and interventions occurring in four regions (Africa, n = 1; East Asia and
rehabilitating roads and bridges. Brooks and Donovan (2020) and Pacific, n = 3; Europe and Central Asia, n = 1; South Asia, n = 1;
Thomas and colleagues (2021) evaluated the impact of B2P in Latin America and the Caribbean, n = 1). All included interventions
Nicaragua and Rwanda respectively. Outcome variables were binary here are single component interventions meaning that they only
or continuous variables linked to farm investments on fertilizers and focused on building, improving and rehabilitating roads and bridges.
pesticides. For example, Brooks and Donovan (2020) used house- Outcome variables here are binary or continuous measures of farm
hold's expenditure fertilizer and pesticides and Fortson and col- investments on hired labor. Van de Walle and Cratty (2002) used
leagues (2015) used a measure of fertilizer expenditure as an the number of person days hired in by household to help with
outcome variable. In Shamdasani (2021), the outcome variable is a cultivation. In Dumas and Játiva (2020), the outcome variable was a
binary taking the value of one if households purchased chemical binary variable indicating if the household hired labor over the last
fertilizer and zero otherwise. year. Qin and Zhang (2016): measured the cost of hired labor, yuan
Farm‐to‐market transport infrastructure interventions did not affect per mu. Other outcome variables related to other farm investments
farm investments in fertilizers. Figure 18 displays these four effect on inputs (such as Fortson et al., 2015, expenditure on farm
sizes. These interventions had a very small positive but statistically irrigation; Shamdasani 2021, binary variable indicating if the
ˆ = 0.03; 95%
insignificant effect on farm investments in fertilizers (SMD household invested in irrigation; Hayakawa et al., 2020, value of
CI: −0.02 to 0.07; p = 0.22). We found moderate heterogeneity households' tractors).
among the estimates (Q(4) = 5.98, p = 0.20, τ̂ 2 = 0.001, I2 = 33.10%), Transport infrastructure interventions increased farm investments in
which had effect sizes ranging from −0.02 to 0.11 standard labor and machinery. We found that transport infrastructure has a
deviations. Brooks and Donovan (2020) found a positive, small and moderate positive and statistically significant effect on farm invest-
statistically significant effect (ĝ = 0.11; 95% CI: 0.01 to 0.21; ment in labor and machinery for intervention beneficiaries
p = 0.04). Authors argued that B2P in Nicaragua facilitated market ˆ = 0.30 ; 95% CI: 0.05 to 0.55; p = 0.02) (Figure 19). Effect
(SMD
access and led to greater purchase of fertilizer inputs. The estimate in sizes of reported estimates vary from −0.03 to 0.97 standard
Shamdasani (2021) was very small and statistically insignificant deviations and there is considerable heterogeneity across included
(ĝ = 0.06; 95% CI: −0.005 to 0.12; p = 0.07). Authors found that estimates (Q (6) = 281.23, p < 0.001, τ̂ 2 = 0.11, I2 = 97.87%).
PMGSY did not reduce the price of farm inputs such as fertilizers. Qin and Zhang (2016) found that PCRI intervention in China
None of the moderator variables significantly contributed to the increased household expenditure on hired labor by 75.1%. Similarly,
variation in estimates. If we focus our analysis exclusively on estimates Van de Walle and Cratty (2002) reported that RTPI provided better
without a high risk of bias, three estimates remain. These estimates are access to roads and encouraged households to invest in agricultural
from Shamdasani (2021), Brooks and Donovan (2020) and Corral and production involving greater employment of laborers.
Zane (2021). The overall weighted average effect becomes statistically None of the moderator variables significantly contributed to the
ˆ = 0.07, 95% CI: 0.02 to 0.12; p = 0. 01), but this
significant (SMD variation in estimates. Excluding high risk of bias estimates, the
result should be interpreted with caution as it is based on a very small number of included estimates decreases to four estimates. This
number of effects. average effect is still statistically significant and increases in
ˆ = 0.42; 95% CI: 0.02 to 0.82; p = 0.04).
magnitude (SMD
Farm investment (labor and machinery). We found seven independent
estimates from seven studies on the effect of transport infra- Credit. We found eight independent estimates across six studies of
structure on farm investments linked to labor and machinery from the impact of transport infrastructure on credit. These studies are

F I G U R E 18 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of farm‐to‐market transport
infrastructure interventions on farm investment (fertilizers).
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 35 of 191

F I G U R E 19 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of farm‐to‐market transport
infrastructure interventions on farm investment (labor and machinery).

based on interventions which took place in Africa (n = 4 ), East Asia Gebresilasse, 2018; Barnett et al., 2018). The overall weighted
and Pacific (n = 1) and Latin America and the Caribbean (n = 1). Two average effect increases in magnitude but is still statistically
studies evaluated multi‐component interventions meaning that it ˆ = 0.12; 95% CI: −0.02 to 0.26; p = 0.09).
insignificant (SMD
combined construction of roads and other activities. These were the
MVP intervention in Ghana presented by Barrett and colleagues Prices. We identified eight estimates evaluating the effect of
(2018) and FPPM studied by Ragasa and colleagues (2016).10 The transport infrastructure interventions on prices. These estimates are
other interventions focused on road rehabilitation (TSIP in Tanzania, from eight studies focusing on interventions in Africa (n = 3), Europe
ARP in Trinidad and Tobago and RTPI in Vietnam) and road and Central Asia (n = 1) and South Asia (n = 4 ). There is one multi‐
construction (URRAP in Ethiopia). Across these six studies, the component intervention in Kenya (SDCP) which combined road
outcome variable is a binary variable taking the value one if a infrastructure development and establishing market linkages for
beneficiary household had access credit in the last 12 months and smallholders. Six studies evaluated the impact of road construction
zero otherwise. Van de Walle and Cratty (2002) used village‐level and rehabilitation whereas Khandker and colleagues (2009) studied
access to credit (binary variable). the impacts of both RRMIMP and RDP which combined road and
Farm‐to‐market transport infrastructure interventions did not affect bridge constructions.
access and use of credit of program beneficiaries compared to non‐ Outcome variables here measure the sales price for specific
beneficiaries. The overall weighted average effect is positive, small crops, or an index of sales prices received across all crops by
ˆ = 0.04; 95% CI: −0.03 to
and statistically significantly (SMD households (n = 5). Two studies (Healy, 2018; Dumas & Játiva, 2020)
0.10; p = 0.30) (Figure 20). Effect sizes of included estimates vary used average sale prices data at the market level and Skelley (2018)
from −0.04 to 0.57 standard deviations and there is substantial analyzed average sale prices at the district level. We selected the
heterogeneity among the estimates (Q (7) = 22.34 , p = 0.002, effects of the interventions on sales prices of crops targeted by the
τ̂ 2 = 0.005, I2 = 68.69%). Dumas and Játiva (2020) found that TSIP intervention such as maize in Tanzania (Dumas & Játiva, 2020), or
intervention in Tanzania did not affect households' access to credit as when no targeted crops are specified, we selected effects of the
the effect size is statistically insignificant (ĝ = −0.02; 95% CI: −0.12 interventions on prices of the most prevalent crops in the local
to 0.07; p = 0.65). In Van de Walle and Cratty (2002), RTPI facilitated context.
access to credit. Authors reported that credit was available to Transport infrastructure did not significantly affect prices
households in 18% more communes two years after the start of the ˆ = 0.04 ; 95% CI: −0.05 to 0.14; p = 0.34) (Figure 21). The
(SMD
road rehabilitation. The effect was positive, large and statistically effect estimates ranged from −0.52 to 0.60 and there was substantial
significant ( ĝ = 0.57; 95% CI: 0.21 to 0.93; p = 0.002). heterogeneity among effects (Q (7) = 24.65, p < 0.001, τ̂ 2 = 0.01,
The moderator analysis indicated that recent implemented I2 = 71.61%).
interventions are associated with slightly smaller effects. When we As described above, researchers have discussed that the
only focus on estimates with low risk of bias and some concerns, four effects of farm‐to‐market transport infrastructure on agricultural
estimates remain (Charles, 2010; Van de Walle & Cratty, 2002; prices cannot be easily predicted (Asfaw et al., 2019; Carletto et al.,
2017; Cazzuffi et al., 2020; Asher & Novosad, 2020). These
10 interventions can have potentially conflicting demand and supply
Three estimates are from Ragasa and colleagues (2016) as in use/adoption of labor and
machinery outcome. effects on prices received by farmers. Also, these effects are likely
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
36 of 191 | MARION ET AL.

F I G U R E 20 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of farm‐to‐market transport
infrastructure interventions on credit.

F I G U R E 21 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of farm‐to‐market transport
infrastructure interventions on prices.

to vary depending on the crops. It is therefore important to in the meta‐analysis). In Casaburi and colleagues (2013), improve-
examine local, contextual results. ment in rural road quality led to a large reduction in the prices of
Focusing on the effects of these interventions on cash crops, in cassava and rice (the two main staples produced domestically,
Bonilla and colleagues (2018), the intervention increased milk estimate for rice is not shown in the forest plot) in rural markets along
production ( ĝ = 0.13; 95% CI: 0.06 to 0.21; p = 0.001) but did not the rehabilitated roads (cassava price: ĝ = −0.52; 95% CI: −1.02 to
result in a rise in milk selling price received by farmers (ĝ = 0.07; 95% 0.02; p = 0.04). In Fortson and colleagues (2015; not shown in the
CI: −0.01 to 0.15; p = 0.08). Authors argued that that training forest plot above), the estimate of RRRP on grains price is statistically
provided within the SDCP intervention was expected to increase insignificant as for the cash crop discussed above.
quality of milk production and therefore prices. They do not discuss Estimates are also statistically insignificant when examining the
the effect of the road infrastructure on milk prices. In another study, effect of farm‐to‐market transport infrastructure intervention on the
Forston and colleagues (2015) expected that RRRP could potentially combined prices of all crops, except for the estimate looking at the
reduce agricultural prices through facilitating farmers' market access effect of RDP. Khandker and colleagues (2009) in Bangladesh, found
resulting in greater competition. However, this is not found in their a moderate and positive effect on prices ( ĝ = 0.21; 95% CI: 0.09 to
quantitative analysis where they found no significant impact on 0.33; p < 0.001). Authors used a Laspeyres price index as their
prices ( ĝ = −0.03; 95% CI: −0.08 to 0.03; p = 0.32). outcome variable.
In terms of food crops, in Tanzania (Dumas & Játiva, 2020), TSIP The moderator analysis indicated that recent interventions are
did not affect sales price of maize and sales price of rice (not included finding slightly smaller effects. Four of the included estimates are
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 37 of 191

assessed as high risk of bias. Excluding high risk of bias estimates, (Figure 22). Effect estimates ranged from −0.17 to 0.21 and there
four estimates from three studies (Khandker et al., 2009; Casaburi is substantial heterogeneity among the effects (Q (9) = 31.77,
et al., 2013; Healy, 2018) remain. The overall weighted average effect p < 0.001, τ̂ 2 = 0.01, I2 = 71.67%). As discussed above, it is not
ˆ = 0.001; 95% CI: −0.18
is still statistically insignificant (SMD possible to include all estimates of the effects of transport
to 0.18; p = 0.98). infrastructure on all considered crops. This type of analysis cannot
highlight whether households decided to change the main marketed
Intermediate outcomes. The intermediate outcomes category covers harvested crop following the intervention.
the outcomes that may affect farmers following the harvest period None of the moderator variables significantly contributed to the
through the sales of their agricultural products. They cover two variation in estimates. We did not identify any funnel plot asymmetry
outcome types: sales and farm income. meaning that publication bias is not likely present. There are three
estimates with low risk of bias or with some concerns for risk of bias
Sales. We found ten studies focusing on the effects of farm‐to‐ (Lokshin & Yemtsov, 2005; Charles, 2010; Nakamura et al., 2020).
market interventions on sales providing each an independent estimate When we run the analysis with just these three estimates, the
(n = 10 ). These studies cover Africa (n = 5), East and Central Asia overall weighted average effect is still statistically insignificant
(n = 2) and Latin America and the Caribbean (n = 3). Of these ten ˆ = −0.01; 95% CI: −0.17 to 0.14; p = 0.85), but this effect
(SMD
studies, three evaluated multi‐component interventions (i.e., FAD‐2, should be interpreted with caution as it is based on very few
LIBA‐72 and SDCP. Most interventions included rehabilitation of estimates.
transport infrastructure and four interventions constructed infra-
structure (B2P, FAD‐2 SDCP and URRAP). Farm income. From 18 studies, we have 19 independent estimates
Outcome variables were included measures of proportions of (k = 19) focusing on the effect of farm‐to‐market transport infra-
produce sold (Nakamura et al., 2020, all crops; Charles, 2010, root structure on farm income outcomes. Two estimates are from Escobal
crops and tubers; Nkonya et al., 2008, grains; Thomas et al., 2021, and Ponce (2008) as authors provided two independent estimates
maize), or the volume of produce sold by rural households. Lokshin (effect of non‐motorized and motorized rural roads) based on two
and Yemtsov (2005) used the quantity sold at the village level as distinct control groups in Peru (intervention: PCR). These studies are
the outcome variable. One study analyzed the effect of these mainly on interventions in Africa (n = 7), followed by East Asia and
transport infrastructure on sales combined across all crops (Naka- Pacific (n = 4 ), Latin America and the Caribbean (n = 3), South Asia
mura et al., 2020, outcome: share of harvested crops sold by the (n = 3) and East and Central Asia (n = 1). There are seven multi‐
household). Several authors evaluated the effect of the interven- component interventions (FAD‐3, FSP, GIADP, LIBA‐72, MVP and
tion on several crops. In those studies, we chose the effect on crops ShoMaP). Six interventions include construction of a bridge (B2P,
targeted by the intervention or when no targeted crops were LIBA‐72, RDP, RRMIMP, ShoMaP, and TLFB).
specified, we selected the effect on the most prevalent crops in the Asher and Novosad (2020) conducted this analysis at the village
local environment (e.g., Dumas & Játiva, 2020, outcome variable: level, all others are conducted at the household level. All estimates
rice sold in kilograms). measure the effect of transport infrastructure on income generated
Effect sizes from these ten studies vary from −0.17 to 0.21 from farm activities except two estimates which were farm activities
standard deviations. Zhangallimbay and Ordoñez (2022) found a profits (Fuller, 2014; Brooks & Donovan, 2020). All outcomes were
moderate and positive effect of the E‐384 and E‐585 roads in farm income or profit on all combined crops in the local contexts,
Ecuador on sales of permanent crops in tons for households within a except Garbero and Songsermsawas (2018) which focused on income
zone of 3 km (ˆg = 0.21; 95% CI: 0.08 to 0.34; p = 0.001). These from livestock in China from GIADP and Jirgi and colleagues (2019)
road construction projects increased market access and volume sold which measured the effect of FAD‐3 on income from sorghum.
by households. On the other hand, Charles (2010) estimated that the Transport infrastructure interventions resulted in very small
ARP resulted in a statistically significant decline in the percentage of improvements in farm income. The weighted average effect was very
root and grains harvested sold (ĝ = −0.17; 95% CI: −0.33 to ˆ = 0.08; 95% CI:
small positive and statistically significant effect (SMD
−0.02; p = 0.03). Authors report an increase proportion of sale 0.04 to 0.13; p < 0.001) (Figure 23). The heterogeneity of effect
levels for other minor crops such as pumpkin reflecting a possible sizes is substantial (Q (18) = 101.31, p < 0.001, τ̂ 2 = 0.01,
change in marketing decisions. Finally, the analysis of the effect of I2 = 82.23%) and effect sizes vary from −0.08 to 1.58 standard
TSIP on rice sale volume is negative and statistically significant deviations. In Charlery and colleagues (2016), the effect is negative
(Dumas & Játiva, 2020, ĝ = −0.13; 95% CI: −0.24 to −0.03; p = 0.01) but statistically insignificant (ĝ = −0.08; 95% CI: −0.53 to
which the authors explain as a decline in the price of rice caused by 0.06; p = 0.59). Jirgi and colleagues (2019) found that FAD‐3
greater competition from lower transaction costs. The return to rice increased beneficiaries' households' farm income by 1.58 standard
production and sales declined limiting the incentive to sell rice. deviations (95% CI: 0.74 to 2.41; p < 0.001). Sikwela (2013) reported
Overall, transport infrastructure did not increase sales. The average an effect size of similar magnitude (ĝ = 1.54 ; 95% CI: 1.11 to
weighted effect size is very small, positive but statistically 1.97; p < 0.001). In both studies, these estimates are based on a
ˆ = 0.01; 95% CI: −0.05 to 0.06; p = 0.02)
insignificant (SMD small sample size of beneficiaries (156 and 89 households in Jirgi
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
38 of 191 | MARION ET AL.

F I G U R E 22 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of farm‐to‐market transport
infrastructure interventions on sales.

F I G U R E 23 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of farm‐to‐market transport
infrastructure interventions on farm income.

et al., 2019 and Sikwela, 2013, respectively) which is a contributing purchases of agricultural inputs leading to greater production and
factor to the very large effect sizes computed from these studies. gains.
TLFB resulted in a rise in farm income in Thailand (Hayakawa The moderator analysis shows that effects were greater In multi‐
et al., 2020, ĝ = 0.23; 95% CI: 0.13 to 0.34; p < 0.001). Kebede component interventions relative to single component interventions.
(2020) found that URRAP, a large road infrastructure in Ethiopia, It also found that more recent interventions seem to be realizing
significantly decreased trading costs between urban and rural smaller effect, greater length of exposure increased effect sizes and
areas and increased specialization in comparative advantage crops, the use of control variables in estimating the effect in the included
though the effect was very small (ĝ = 0.03; 95% CI: 0.01 to studies was associated with significantly lower effect sizes. We
0.05; p = 0.003). B2P in Nicaragua led to a small rise in total farm identified potential funnel plot asymmetry suggesting there may be
profit (Brooks & Donovan, 2020, ĝ = 0.11; 95% CI: 0.01 to publication bias. Nine of the included estimates are assessed as high
0.21; p = 0.03). Authors claim that the bridge facilitated the risk of bias. The overall weighted average effect remains statistically
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 39 of 191

significant when we only include the 10 estimates with low of risk of E‐384 and E‐585 roads in Ecuador resulted in a significant rise of
ˆ = 0.06; 95% CI: 0.01
bias or with some concerns for risk of bias (SMD beneficiary households' production of banana per hectare for house-
to 0.11; p = 0.02). holds within a zone of 3 km from the road (Zhangallimbay & Ordoñez,
2022; ĝ = 0.44 ; 95% CI: 0.20 to 0.67; p < 0.001). Similarly, Brooks
Agricultural production outcomes. The agricultural production out- and Donovan (2020) found that a similar result of similar magnitude on
comes are yields, volume, produce type, quality and crop loss. The maize yields evaluating the impact of B2P in Nicaragua (ĝ = 0.35; 95%
included studies provided evidence on all of these outcomes, except CI: 0.14 to 0.55; p = 0.001). Also, the Jamuna Bridge led to substantial
crop loss and quality. rise in yields in the beneficiary upazilas/subdistricts (Blankespoor et al.,
2022, ĝ = 0.72; 95% CI: 0.20 to 1.12; p < 0.001). In Wanjala and
Yields. Fifteen independent estimates from 15 studies evaluated the Muradia (2013), MVP in Kenya resulted in an increase in beneficiaries'
effect of farm‐to‐market transport infrastructure on yields. Six maize yield (ĝ = 0.33; 95% CI: 0.10 to 0.55; p = 0.01). Authors in
estimates are based on interventions in Africa, two from East Asia these studies argued that the infrastructure construction facilitated
and Pacific, five from Latin America and the Caribbean and two from access to inputs and increased the incentive for agricultural
South Asia. Six of the 14 studies evaluated were multi‐component investments.
interventions (AD2M, GIADP, MVP in Kenya and MVP in Ghana, Two studies reported that the transport infrastructure resulted in a
PROSAP and ShoMaP). Eleven focused on roads and four included negative and statistically significant effect on yields (Kingombe & Di
bridges (AD2M, B2P, JBI and ShoMaP). Falco, 2012; Garbero & Songsermsawas, 2018). Garbero and Song-
Yields are a measure of agricultural productivity. All yields sermsawas (2018) farmers in project areas reallocate their arable land
outcomes considered here were at the household level, except the from growing grain and root crops to higher valued crops including
one from Blankespoor and colleagues (2022) which provided vegetables and fruits (ĝ = −0.14; 95% CI: −0.24 to −0.05; p = 0.002).
the annual yield per hectare at the subdistrict level. We selected Kingombe and Di Falco (2012) focused on the effect of EPFRP on
the effects of the interventions on crops targeted by the intervention, cotton productivity and found a small, negative, statistically significant
or when no targeted crops are specified, we selected effects of the (ĝ = −0.13; 95% CI: −0.24 to −0.01; p = 0.03).
interventions on the most prevalent crops in the local context. This analysis highlights differences in effects across crop
Farm‐to‐market transport infrastructure interventions led to a small types (food, cash and all crops). We found positive effects and
rise in yields. The overall weighted average effect is small positive and negative effects on food crops and also on cash crops as shown in
statistically significant effect (SMD ˆ = 0.13; 95% CI: 0.04 to Figure 24.
0.22; p = 0.003). There is substantial heterogeneity in the magnitude The moderator analysis shows that the magnitude of the effect
of the effect sizes (Q (14) = 88.11, p < 0.001, τ̂ 2 = 0.02, I2 = 84.11%) sizes was smaller from road infrastructure relative to bridge infra-
which vary from −0.15 to 0.72 standard deviations (Figure 24). structure. We identified potential funnel plot asymmetry suggesting
Several studies reported large positive and statistically significant there may be publication bias. Five estimates are assessed as high risk
effects of transport infrastructure on yields. The construction of the of bias. Excluding high risk of bias estimates, ten estimates remain.

F I G U R E 24 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of farm‐to‐market transport
infrastructure interventions on yields.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
40 of 191 | MARION ET AL.

The overall weighted average effect is still statistically significant and of distinguished several positive and statistically significant effects on
ˆ = 0.12; 95% CI: 0.01 to 0.22; p = 0.03).
similar magnitude (SMD production. Bonilla and colleagues (2018) and Khandker and colleagues
(2009) found a small effect in different contexts (maize production in
Volume. Sixteen estimates based on 16 studies provided information Kenya: ĝ = 0.13; 95% CI: 0.06 to 0.21; p = 0.001; all crops production
on the effect of farm‐to‐market transport infrastructure interventions in Bangladesh: ĝ = 0.18; 95% CI: 0.06 to 0.30; p = .004 , respec-
on volume outcomes. These studies focused on interventions in Africa tively). Ring and colleagues (2018) reported a moderate effect on rice in
(n = 7), Europe and Central Asia (n = 1), Latin America and the Madagascar (ĝ = 0.21; 95% CI: 0.11 to 0.32; p < 0.001). The largest
Caribbean (n = 5), and South Asia (n = 3). effects are from Lema and colleagues (2017, ĝ = 0.49; 95% CI: 0.01 to
Five studies evaluated multi‐component interventions (AD2M, 0.97; p = 0.04) and Jirgi and colleagues (2019, ĝ = 1.01; 95% CI: 0.65
FAD‐3, LIBA‐72, PROSAP and SDCP). Six focused on rehabilitation of to 1.29; p < 0.001). In Jirgi and colleagues (2019), the effect is very
farm‐to‐market transport infrastructure (ARP, EPFRP, LIBA‐72, MCCCP, large and is based on the effect of FAD‐3 which combined many
RRRP and TSIP) whereas the others constructed new infrastructure. Six different types of interventions, including the construction of roads in
interventions included bridge development (AD2M, B2P, LIBA‐72, RDP Nigeria.
and RRMIMP). The moderator analysis Indicated that effects were greater In
Outcome variables include measures of the total quantity or multi‐component interventions relative to single component inter-
volume of agricultural production. The analysis is at the household ventions. We identified potential funnel plot asymmetry suggesting
level for 15 of the included estimates, one at the individual level there may be possible publication bias. Nine of the 16 estimates are
(Thomas et al., 2021) and one at the district level data (Skelley, 2018). assessed as high risk of bias. When we only focus on estimates with
As for the other outcomes, we selected the effects of the low risk of bias and some concerns, seven estimates remain
interventions on crops targeted by the intervention, or when no (Khandker et al., 2009, Charles, 2010; Khandker & Koolwal, 2011;
targeted crops are specified, we selected effects of the interventions Ring et al., 2018; Jirgi et al., 2019; Brooks & Donovan, 2020; Corral &
on the most prevalent crops in the local context. Zane, 2021). The overall weighted average effect remains statistically
Farm‐to‐market transport infrastructure interventions did not ˆ = 0.08; 95% CI: −0.05 to 0.21; p = 0.24).
insignificant (SMD
improve volume outcomes. The overall weighted average effect is
very small positive and statistically insignificant effect on volume Non‐farm outcomes. The non‐farm outcomes category covers the
ˆ = 0.04 ; 95% CI: −0.02 to 0.10; p = 0.18) (Figure 25).
outcome (SMD outcomes that may affect farmers outside the farm activities and
Effect sizes vary from −0.21 to 1.01 standard deviations and there is agricultural production. They cover two outcome types: non‐farm
substantial heterogeneity in the magnitude of the effect sizes income and labor market outcomes.
(Q (15) = 59.55, p < 0.001, τ̂ 2 = 0.009, I2 = 74.81%).
Charles (2010) found that ARP intervention in Trinidad and Tobago Non‐farm income. We found 16 estimates from 15 studies looking at
had a moderate, negative and statistically significant effect on the the effect of farm‐to‐market transport infrastructure on non‐farm
quantity of roots crops and tubers produced (ĝ = −0.21; 95% CI: −0.36 income outcomes. Escobal and Ponce (2008) provided two estimates
to −0.05; p = 0.01) but the effect was not statistically significant on (effect of non‐motorized and motorized rural roads) as these
pumpkin production (not shown in the forest plot). We also estimates are based on two distinct control groups in Peru and are

F I G U R E 25 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of farm‐to‐market transport
infrastructure interventions on volume.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 41 of 191

therefore independent. African and South Asia are the two most 0.17 to 0.36; p < 0.001). Authors highlighted, as we discussed
represented regions with four estimates each. Three estimates were above, that this effect was at the expense of the agricultural labor
from interventions in East Asia and Pacific. One estimate was from market. In Nigeria, FAD‐2 led to a greater impact in magnitude
South Asia (India: Asher & Novosad, 2020) and four estimates from (ĝ = 0.45; 95% CI: 0.23 to 0.68; p < 0.001). Finally, Charlery and
Latin America and the Caribbean (Peru: Escobal & Ponce, 2008; colleagues (2016) described the effects in Nepal from BJS ( ĝ = 0.53;
Nicaragua: Brooks & Donovan, 2020; Ecuador: Corral & Zane, 2021). 95% CI: 0.24 to 0.83; p < 0.001). Transport costs decreased
There were four multi‐component interventions: FAD‐2, GIADP and following the intervention and facilitated access and increased
MVP in Ghana and Kenya. Four interventions included bridge demand for labor in services activities. As shown in the outcome
constructions: B2P in Nicaragua and Rwanda, JBI and RDP. off‐farm labor market outcomes, these results suggest that workers
Considered outcomes here are measures of income generated moved from the agricultural sector to non‐agricultural sectors.
from non‐farm activities. All are at the household level, except the The moderator analysis indicated that the use of control variables
estimate from Asher and Novosad (2020) in India which focused on in estimating the effect in the included studies was associated with
household income level from manual labor at the village level. Other significantly lower effect sizes. We identified potential funnel plot
outcome variables at the household level are wage from non‐ asymmetry suggesting there may be publication bias. Seven included
agricultural activities (Khandker et al., 2009; Garbero & Songsermsa- estimates are assessed as high risk of bias. When only focusing on
was, 2018; Charlery et al., 2016; Escobal & Ponce, 2008; Barnett the nine estimates with low and with some concerns for risk of bias, the
et al., 2018; Thomas et al., 2021; Wanjala & Muradia, 2013) or total overall weighted average effect becomes statistically significant
non‐farm income (Mannava et al., 2020; Madu & Phoa, 2012; Qin & ˆ = 0.07 ; 95% CI: −0.01 to 0.15; p = 0.08).
(SMD
Zhang, 2016).
Farm‐to‐market transport infrastructure interventions increased Labor market outcomes. Twelve studies evaluate the effect of farm‐to‐
non‐farm income. The overall average weighted effect is very small market transport infrastructure interventions on non‐farm labor market
ˆ = 0.08; 95% CI: 0.02 to
positive and statistically significant (SMD outcomes. Within those 12 studies, we found 13 independent effects.
0.14; p = 0.01) (Figure 26). The effect sizes vary from −0.07 to 0.53 Escobal and Ponce (2008) provided two estimates (effect of non‐
standard deviations indicating that there is substantial heterogeneity motorized and motorized rural roads) as these estimates are based on
in effects (Q (15) = 68.75, p < 0.001, τ̂ 2 = 0.009, I2 = 78.18%). two distinct control groups in Peru and are therefore independent. These
Thomas and colleagues (2021) found that B2P in Rwanda studies are on interventions across different regions: Africa (n = 2), East
successfully linked households to labor markets (ĝ = 0.12; 95% CI: Asia and Pacific (n = 3), Latin America and the Caribbean (n = 3) and
0.01 to 0.24; p = 0.04). Escobal and Ponce (2008) reported an South Asia (n = 4 ). IAIV was a multi‐component intervention. Three
impact of larger magnitude and similar mechanism ( ĝ = 0.27; 95% CI: interventions included bridge constructions: JBI, PAST and RRMIMP and

F I G U R E 26 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of farm‐to‐market transport
infrastructure interventions on non‐farm income.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
42 of 191 | MARION ET AL.

four interventions only focused on transport infrastructure rehabilitation: 0.09 to 0.49; p = 0.01). The likelihood that the main household
ARP, IAIV RTPI and TSIP. provider was working in the service sector significantly increased.
Outcome variables are measures of employment outcomes and None of the moderator variables significantly contributed to the
number of hours worked (not on farm owned by the household). Two variation in estimates. We did not identify any funnel plot asymmetry
studies (Escobal & Ponce, 2008; Charles, 2010) used a binary variable meaning that publication bias is not likely present. Four included
(employed in non‐agriculture work). Number of days last year in non‐ estimates are assessed as high risk of bias. When we only focus on
agricultural self‐employment at the household level was the estimates with low risk of bias and some concerns, nine estimates
household level outcome variable in Khandker and Koolwal (2011). remain. The overall weighted average effect increases and is still
Other outcome variables included the number of household members ˆ = 0.09; 95% CI: 0.03
small, positive and statistically significant (SMD
or share of household members in non‐agricultural wage sector to 0.15; p = 0.002).
(Akee, 2006; Nakamura et al., 2020; Dumas & Játiva, 2020; Nguyen
et al., 2017). Welfare outcomes. The welfare outcomes category covers measures
Farm‐to‐market transport infrastructure interventions improved of improvements in the farmers' well‐being and quality of life. They
non‐farm labor outcomes. This overall weighted average effect is very encompass a range of social, economic, and environmental factors,
ˆ = 0.07 ; 95% CI: 0.02
small, positive and statistically significant (SMD but in this review we focus on total income and food security and
to 0.11; p = 0.003) (Figure 27). Effect sizes vary from −0.04 to 0.29 nutrition.
standard deviations This overall weighted average effect is very small
ˆ = 0.07 ; 95% CI: 0.02 to
positive and statistically significant (SMD Total income and wealth. We identified 27 estimates examining the
0.11; p = 0.003, and there is substantial heterogeneity in effects effect of farm‐to‐market transport infrastructure interventions on total
(Q (12) = 30.81, p = 0.002, τ̂ 2 = 0.004, I2 = 61.05%). income and wealth from 23 studies and three linked papers (Khandker
Households were more likely to be engaged in non‐farm labor & Koolwal, 2011; Khandker et al., 2009; Hayakawa et al., 2020).
following the intervention in Peru (Escobal & Ponce, 2008; ĝ = 0.10 ; These studies analyzed interventions across all regions (Africa, n = 9;
95% CI: 0.01 to 0.20; p = 0.03). This shift from the agricultural to South Asia, n = 5; East Asia and the Pacific, n = 6; Europe and Central
non‐agricultural sectors is also visible in Akee (2006; ĝ = 0.16; 95% Asia, n = 1; Latin America and the Caribbean, n = 5). Escobal and
CI: 0.10 to 0.23; p < 0.001). The BR intervention caused a higher Ponce (2008) provided two estimates (effect of non‐motorized and
number of household members working in wage labor. The large‐ motorized rural roads), which are based on two distinct control
scale bridge JBI increased the share of employment in services groups in Peru and therefore independent.
industries at the subdistrict level (Blankespoor et al., 2022; ĝ = 0.22; Most estimates were based on interventions focusing only on
95% CI: 0.08 to 0.37; p = 0.003). The study from the Ministry of farm‐to‐market transport infrastructure (k = 19). The remaining eight
Foreign Affairs of Denmark (2010) found that PAST in Nicaragua had estimates were based on impact evaluations studying the effect of this
the largest impact on non‐farm labor outcomes ( ĝ = 0.29; 95% CI: type of intervention alongside at least one other intervention. For

F I G U R E 27 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of farm‐to‐market transport
infrastructure interventions on non‐farm labor market outcomes.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 43 of 191

example, Garcia and colleagues (2018) quantified the effect of ShoMaP (ĝ = −0.08; 95% CI: −0.22 to 0.07; p = 0.30), but it is not statistically
in Kenya, which combined investments in domestic horticultural value significant. Brooks and Donovan (2020) found that B2P in Nicaragua
chains, improvements in market infrastructure, and improvements in led to a large statistically significant effect ( ĝ = 0.56; 95% CI: 0.28 to
rural access roads. Similarly, in addition to roads, MVP in Kenya 0.83; p < 0.001).
invested in natural assets (soil nutrients), human capital (skills and None of the moderator variables significantly contributed to
health), power and telecommunication infrastructure, household assets, the variation in estimates. We identified potential funnel plot
and microfinance. Of the 26 interventions considered here, all included asymmetry, suggesting possible publication bias. Eleven included
construction, rehabilitation, and/or maintenance of roads, except for estimates are assessed as high risk of bias. Of the 27 included
two (B2P and TLFB) that constructed bridges. effect sizes, 16 remain after excluding high risk of bias estimates.
Outcomes primarily related to household‐level total or per capita When we restrict the analysis to those estimates, the overall
income (k = 16). Nguyen and colleagues (2017) employed a wealth weighted average effect slightly increases and remains very small
index based on household assets as an outcome variable. For seven but statistically significant (SMD ˆ = 0.08; 95% CI: 0.04
studies, the outcome variables are household‐level per capita or total to 0.13; p < 0.001).
expenditure. The remaining three estimates measured impact on the
number of household‐owned assets (Ministry of Foreign Affairs of Food security. Ten studies reported a total of eleven estimated
Denmark, 2010; Garcia et al., 2018; Ring et al., 2018). effects of transport infrastructure interventions on food security. Most
Farm‐to‐market transport infrastructure interventions led to an of these studies focused on interventions taking place in Africa
increase in income and wealth outcomes. The overall weighted average (n = 5), with the remaining studies in Latin America and the Caribbean
ˆ = 0.05;
effect is very small, positive, and statistically significant (SMD (n = 3), East Asia and the Pacific (n = 1), and South Asia (n = 1). Escobal
95% CI: 0.02 to 0.08; p = 0.004 ) (Figure 28). Effect sizes of reported and Ponce (2008) provided two estimates (effect of non‐motorized
estimates vary from −0.08 to 0.56 standard deviations and there is and motorized rural roads), as these estimates are based on two
substantial heterogeneity across included estimates (Q (26) = 70.72, distinct control groups in Peru and are therefore independent. These
p < 0.001, τ̂ 2 = 0.004, I2 = 63.24%). The smallest effect size is based eleven estimates originated from four multi‐component interventions
on Fuller (2014), which studied the impact of LIBA‐72 in Liberia (AD2M in Madagascar, LIBA‐72 in Liberia, MVP in Ghana, and

F I G U R E 28 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of farm‐to‐market transport
infrastructure interventions on income and wealth.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
44 of 191 | MARION ET AL.

F I G U R E 29 Forest plot showing the observed outcomes and estimates of the random‐effects model for the effect of farm‐to‐market
transport infrastructure interventions on food security.

SHoMAP in Kenya) and seven interventions focusing on transport 0.22; p = 0.02)—a measure of households' perceptions of food
infrastructure development. security—though the magnitude of the effect is small.
Their outcomes here are all at the household level and include The RRMIMP intervention in Bangladesh resulted in a small
measurement of the value of food consumption expenditure level increase ( ĝ = 0.14 ; 95% CI: 0.03 to 0.25; p = 0.01) (Khandker &
(k = 5), amount of rice consumption in kilograms (Dumas & Koolwal, 2011) in food security, using household food expenditure as
Játiva, 2020), two estimates on the household food insecurity access a measure, 12 years after the start of the intervention. Finally,
scale (HFIAS, Garcia et al., 2018; Fuller 2014), and two binary Barnett and colleagues (2018) found that MVP in Ghana made
outcomes on households' lack of food in the past year (Khandker & households less likely to report not having enough food in the last 12
Koolwal 2011; Barnett et al., 2018). months (ĝ = 0.19; 95% CI: 0.10 to 0.28; p < 0.001).11
Farm‐to‐market transport infrastructure interventions did not None of the moderator variables significantly contributed to the
improve food security. On average, the effect was very small and variation in estimates. Of the eleven estimated effect sizes, six are
ˆ = 0.02 95% CI: −0.04 to
positive, but not statistically significant (SMD assessed as having a high risk of bias, four are assessed as having a
0.09; p = 0.46) (Figure 29). Once again, we found substantial low risk of bias, and one has some concerns for risk of bias. When we
heterogeneity among the effects (Q (10) = 40.49, p < 0.001, remove high risk of bias estimates from the model, the overall
τ̂ 2 = 0.008, I2 = 75.30 %) which ranged from −0.14 to 0.19 standard weighted average effect remains very small and statistically
deviations. ˆ = 0.05; 95% CI: −0.08 to 0.17; p = 0.49).
insignificant (SMD
Hayakawa and colleagues (2020) found that TLFB led to a small
reduction in household food expenditure (ĝ = −0.14 ; 95% CI: −0.23 Nutrition. Three estimates provide information on the effect of farm‐
to −0.04; p = 0.01). Authors reported that the international bridge to‐market transport infrastructure interventions on nutrition outcomes.
linking Thailand and Laos negatively affected food expenditure in a Two studies evaluated interventions in Kenya (SDCP and ShoMaP)
very significant way, mainly for female and young heads of house- and one in China (GIADP). All three interventions combined the
holds (ĝ = −1.10 ; 95% CI: −1.28 to −0.91; p < 0.001) but not for male development of transport infrastructure with other activities aiming
or elder heads of households ( ĝ = −0.09; 95% CI: −0.20 to to develop agricultural activities. For example, the SDCP provided
0.03; p = 0.15). The bridge benefitted households, which were able training and constructed roads for greater smallholder access to
to expand their agricultural production and farm income, thereby markets. All outcomes here are measures of the household dietary
affecting food expenditure; however, it disadvantaged other house- diversity score (HDDS), which is the number of food groups
holds such as female‐headed households. consumed by a household over a specified period.
Garcia and colleagues (2018), Khandker and Koolwal (2011) and Farm‐to‐market transport infrastructure interventions did not
Barnett and colleagues (2018) all found small, positive, and improve nutritional outcomes. We found that transport infrastructure
statistically significant effects on food security outcomes. According interventions did not affect HDDS of program beneficiaries
to Garcia and colleagues (2018), SHoMAP increased beneficiary
households' scores on the household food insecurity access scale 11
The effect size here is reversed, meaning that a higher effect represents lower food
HFIAS for beneficiary households ( ĝ = 0.12; 95% CI: 0.02 to insecurity.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 45 of 191

ˆ = 0.03; 95% CI: −0.08 to

compared to non‐beneficiaries (SMD Gender. Table 7 shows effect sizes by gender subgroups. There were
0.14; p = 0.63) (Figure 30), but there is substantial heterogeneity not enough effects within any outcome category to examine
across included estimates (Q (2) = 9.28, p = 0.01, τ̂ 2 = 0.008, weighted average effects by gender. Brooks and Donovan (2020)
I2 = 78%). However, this result should be interpreted with caution, showed that the effects across various outcomes of the B2P
given the small number of estimates contributing to this analysis. intervention were similar for men and women. Mannava and
Garcia and colleagues (2018) found a small, positive, and colleagues (2020) reported that the effect of the intervention on
statistically significant effect (ĝ = 0.12; 95% CI: 0.02 to agricultural income and total income became negative and statisti-
0.22; p = 0.02). This result is in line with the food security outcome cally significant for female‐headed households. In this context,
effect described above. female‐headed households were less likely to take advantage of
None of the moderator variables significantly contributed to the newly created economic opportunities. Nkonya and colleagues
variation in estimates. Two of the three included estimates were (2012) do not find any difference in total income by gender.
assessed as high risk of bias (Bonilla et al., 2018; Garcia et al., 2018);
thus, we could not conduct a sensitivity analysis by study quality. The Wealth. Table 8 show effect sizes by wealth subgroups. Charlery and
estimate from Garbero and Songsermsawas (2018) was assessed as colleagues (2016) found that the effects on total income were
low risk of bias ( ĝ = −0.08; 95% CI: −0.18 to 0.01; p = 0.07). greatest for the poorest households following the BJS intervention.
The effect sizes were statistically insignificant in the whole sample
Subgroup analysis. We now examine how the effect sizes of farm‐to‐ and for the least poor households. The effects reported by Garbero
market transport infrastructure interventions may vary by subgroup. and Songsermsawas (2018) and Nkonya and colleagues (2012) do not
We here compare estimates within studies from the same interven- seem to vary by wealth levels.
tion across subgroups. We do not report here the effect sizes across
subgroups that are consistently statistically insignificant. We divide Analysis of effects over time. We aimed to reproduce this same type
this analysis by type of subgroup: gender of beneficiaries and wealth of analysis but for different periods of exposure to the interventions.
of beneficiaries. Here we compare estimates within studies from the same

F I G U R E 30 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of farm‐to‐market transport
infrastructure interventions on nutrition.

TABLE 7 Effect sizes across gender subgroups.

Study Outcome subcategory Outcome description Subgroup information ĝ p Value

Brooks and Non‐farm labor market Days worked outside Men 0.35 <0.001
Donovan outcomes the village
Women 0.28 <0.001
(2020)

Mannava Farm income Agricultural income Whole sample 0.15 0.09

et al. (2020)
Female‐headed household −0.22 0.01

Total income and wealth Total income Whole sample −0.06 0.51

Female‐headed household −0.18 0.04

Nkonya Total income and wealth Household income Men only 0.14 <0.001
et al. (2012)
Women only 0.22 0.02

Note: Subgroup estimates in bold have a statistically significant effect size (ĝ ) at a 5% significance level.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
46 of 191 | MARION ET AL.

TABLE 8 Effect sizes across wealth subgroups.

Outcome Outcome
Study subcategory description Subgroup information ĝ p Value

Charlery Total income and Total income Whole sample 0.29 0.05
et al. (2016) wealth
Least poor households −0.03 0.83

Poorest households 0.32 0.03

Garbero and Food security and Dietary diversity Whole sample −0.08 0.07
Songsermsawas nutrition score
Poor county 0.08 0.26
(2018)
Vulnerable county −0.17 0.01

Yields Grain yield (kg/ Whole sample −0.14 <0.001

ha, log)
Poor county −0.03 0.74

Vulnerable county −0.20 <0.001

Use/adoption of Pesticide use Whole sample 0.03 0.56

improved inputs, per ha (kg/
Poor county 0.16 0.03
technologies, and ha, log)
practices Vulnerable county −0.10 0.10

Volume Value of grain Whole sample −0.16 <0.001

production
Poor county −0.08 0.30
(Chinese
yuan, log) Vulnerable county −0.21 <0.001

Nkonya et al. (2012) Total income and Household Tercile 1 (the poorest) 0.03 0.61
wealth income
Tercile 2 0.27 0.01

Tercile 3 0.06 0.62

Note: Subgroup estimates in bold have a statistically significant effect size (ĝ ) at a 5% significance level.
Abbreviations: ha, hectare; kg, kilogram.

intervention across time. As above, we only focus on the effect sizes intervention was successful (or not). A thorough analysis of barriers
where the period of exposure to the intervention is different, with all and facilitators would require a dedicated search for different types
other variables held constant. We do not report here the effect sizes of evidence. Of the 83 papers in this body of evidence, our analysis
across the period of exposure, which are consistently statistically of barriers and facilitators identified 17 papers with relevant
insignificant. information.
From Table 9 (Khandker & Koolwal, 2011) (RRMIMP) and
Table 10 (Lema et al., 2017) (PROSAP), we notice that effect sizes Topographic characteristics and geographic distance to the road may
do not always increase as the exposure period increases. For impact access to intervention benefits. The construction of a road or
example, in Khandker and Koolwal (2011), the effect of the its maintenance is implemented in a specific geography whose
intervention on transaction costs is no longer statistically significant. characteristics may affect the road's use. For example, several
However, we do notice that the intervention sustainably led to a shift authors highlight that post‐intervention distance to the road may
from agricultural to non‐agricultural labor markets. The effects here have a smaller impact on those located farther from the road.
became statistically significant after 96 months from the start of the Nakamura and colleagues (2020) analyzed the effect of URRAP,
intervention. which connected rural communities through new roads. He observed
that villages with no pre‐existing roads had significantly higher
Barriers and facilitators analysis welfare gains after the intervention.
We now discuss the reported facilitators and barriers of the Khandker and Koolwal (2011) analyzed the effects of rural
interventions' effects in their local contexts. This presentation is infrastructure expansion in Bangladesh and highlighted that house-
subject to an important caveat: the included studies are quantitative holds farther away from project roads benefitted less than their
impact evaluations. Such studies tend to focus narrowly on counterparts who were closer. Similarly, Dumas and Játiva (2020)
identifying the magnitude of the intervention's effects, without found that the effects (e.g., area of rice cultivated by households,
discussing implementation details or rigorously explaining why an prices, adoption of organic fertilizer) seem to dissipate as distance
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 47 of 191

T A B L E 9 Effect sizes across periods of exposure (from Khandker terms of a decline in the share of services employment. Similarly,
& Koolwal, 2011, RRMIMP). Zhangallimbay and Ordoñez (2022) found that effects on yields
Period of and sales were greater for households further away (6 km as
Outcome Outcome exposure in opposed to three) from the intervention.
subcategory description months ĝ p Value
In addition to beneficiaries' distance from the road, the distance
Total income Household per 48 0.12 0.03 from the road to the closest market is another geographical
and wealth capita expenditure
96 0.11 0.05 characteristic to consider. Mu and Van de Walle (2011) focused on
RTPI, which aimed to link commune centers to markets through the
Food security Household per 48 0.10 0.09
and nutrition capita food rehabilitation of 5000 km of rural roads. They also observed that the
96 0.14 0.01
expenditure intervention's impact on market‐related outcomes was lower for

Transaction Average transport 48 0.15 0.01 communes that were farther away from existing markets. Similarly,
costs costs/trip: rainy Mu and Van de Walle (2011) observed that communes with pre‐
96 0.10 0.08
season existing markets saw greater intervention impacts on market‐related
Labor market Number of days 48 0.02 0.71 outcomes.
outcomes (ag) past month Finally, the authors also highlighted the importance of the road's
96 −0.18 0.001
inagricultural
land topography in maximizing intervention impact. For example,
wage work
Dappe and colleagues (2021) commented that PMGSY road
Labor market Number of days 48 0.08 0.13 interventions in India increased the travel distance from habitation
outcomes past month innon‐
96 0.16 0.003 where the terrain was flatter.
agricultural
wage work
Socio‐economic discrepancies may affect the benefits of transport
Note: Estimates in bold have a statistically significant effect size (ĝ ) at a
5% significance level. infrastructure. Both Mu and Van de Walle (2011) in Vietnam and
Melecky and colleagues (2018) in India observed that communes and
households with better access to credit saw comparatively fewer
T A B L E 10 Effect sizes across periods of exposure (from Lema benefits from the interventions. Similarly, both Mu (2011) and Mu
et al., 2017, PROSAP). and Van de Walle (2011) observed that Vietnamese households with
Period of higher literacy levels benefitted less from the interventions.
Outcome Outcome exposure in However, other authors observed higher effects among more
subcategory description months ĝ p Value educated households (Nguyen et al., 2017; Melecky et al., 2018;
Volume Milk 72 0.41 0.09 Ring et al., 2018).
production (L) Socio‐economic impacts may also vary by gender; several
108 0.49 0.04
studies reported smaller effects among female‐headed house-
Note: Estimates in bold have a statistically significant effect size (ĝ ) at a
holds compared to male‐headed households (Hayakawa et al.,
5% significance level.
2020). These qualitative findings are in line with those presented
in the subgroup analysis. We might observe a smaller effect of
road interventions among the poorest and most vulnerable
from the road increased. The effect size on the price of rice is no households due to additional costs incurred by access to and
longer statistically significant for villages located 30–50 km away. use of the roads. Similarly, the richest households might use
Conversely, some authors did observe a larger effect for alternative access to markets and not use the new roads.
households located farther from the road, especially following the However, the hypothesis on the smaller effects observed in
construction of new infrastructure (as opposed to rehabilitation), those groups would need to be corroborated by additional
and more immediate outcomes. For example, Blankespoor and quantitative research.
colleagues (2022) highlighted that villages situated farther away
from the Jamuna Bridge benefitted most from the new transport Intervention costs
link, especially in terms of the growth rate for agricultural Information about costs is available for 15 of the 41 included
productivity. interventions based on information from 15 studies. Of the 83 included
These interventions can have negative effects for some studies, 18.5% reported cost data, which is in line with the impact
subpopulations if, for example, proximity to an infrastructure evaluation literature. This statistic is consistent with the broader
project reduces the comparative advantage that villagers previ- international development field, in which Brown and Tanner (2019)
ously had over those without access to this infrastructure. For reported that fewer than one in five impact evaluations provide
example, Blankespoor and colleagues (2022) observed that information about intervention costs. As expected, overall, the reported
villagers situated closest to the Jamuna Bridge suffered most in costs of these infrastructure interventions are large (Table 11).
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
48 of 191 | MARION ET AL.

T A B L E 11 Summary of available intervention costs.

Acronym
Study (country code) Cost information description

Charles (2010) ARP (TTO) Cost: USD92.9 million

Ring et al. (2018) AD2M (MDG) Cost: USD56.7 million

Jones and Salazar (2021) AEGBI (MOZ) Cost: EUR66 million

Brooks and Donovan (2020) B2P (NIC) Cost: USD40,000

Moreno (2019) CPP (COL) Average cost per kilometer: USD156.657

Casaburi et al. (2013) RFRRP (SLE) Cost: EUR9.5 million (approximately EUR16,000 per kilometer of rehabilitated road)

Sulaiman et al. (2021) FAD‐3 (NGA) Cost: USD450 million

Nguyen et al. (2017) IAIV (VNM) Cost: EUR42 million

Blankespoor et al. (2018) JBI (BGD) Cost: USD753.7 million

Torero et al. (2016) MCCCP (SLV) The economic rate of return measures the effectiveness of a project by contrasting the
discounted flows of costs and benefits of a specific intervention. The updated, revised rate of
return for MCCCP was 21.6% over twenty years, as of 2014.

Wanjala and MVP (KEN) MVP has an estimated budget of USD110 per person. Implementing the intervention in a
Muradia (2013) village with about 60,000 people costs approximately USD7 million (MVP 2006).

Barnett et al. (2018) MVP (GHA) Cost: USD15.3 million. Health and infrastructure were the largest sectors in terms of project
spend, with management and overheads accounting for approximately a third of the total.
The cost per capita was USD360 in 2012 (in present value terms), or USD88 per capita per
annum.

Shamdasani (2021) PMGSY (IND) The average cost per kilometer of road construction under the program is estimated to be
USD43,300. (Source: PMGSY Online Management and Monitoring System). PMGSY roads
extended for 480,000 km.

Lema et al. (2017) PROSAP (ARG) Dairy farms located fewer than 15 km from the roads increased their production, on average,
between 198,000 and 208,000 (depending on the period considered). Therefore, the
additional 198,000 L per year produced by each farm represents an additional net income of
USD7227 per year and per farm—which, for the 322 farms in the target population, is
equivalent to USD2,327,094 per year.An investment of USD24,424,877 implies an
investment recovery period of about 10.5 years, which, although quite long, is a conservative
estimate, since it does not consider other benefits derived from these works.

Bonilla et al. (2018) SDCP (KEN) Cost: Kenyan shillings (KSh)1,515,614,910. Estimated total cost per beneficiary: KSh97,561.
Estimated per farmer benefit: KSh20,586 in 2016. It would take approximately 4.74 years to
cover costs.

Despite the heterogeneity in interventions (e.g., JBI in Bangla- rate of return for the MCCCP intervention (which consisted of road
desh cost more than USD750 million for the construction of a 4.5‐km rehabilitation in El Salvador) at 21.6% over 20 years, as of 2014.
bridge, whereas B2P's costs in Nicaragua did not exceed USD50,000 However, considering the large outlays and variable costs associated
per bridge), transport infrastructure represents one of the costliest with the interventions, it is likely that these estimates would vary
types of development intervention. However, they are often long‐ quite considerably across different contexts (even for relatively
term initiatives that can affect large numbers of beneficiaries and similar projects).
radically transform the local economic and social environment. For Given the heterogeneity of interventions, the available informa-
example, Blankespoor and colleagues (2018) reported that JBI tion provides limited insight into the likely variability of costs across
reduced the cost of transport from the poor North West hinterland different contexts. Similarly, scant information accompanies most of
in Bangladesh to the capital city by more than 50%. the cost information reported, meaning that comparability of these
Bonilla and colleagues (2018) and Lema and colleagues (2017) figures is not possible. Of these 15 interventions, six are multi‐
reported that the estimated cost recovery time for these types of component interventions (AD2M, FAD‐3, MVP Kenya, MVP Ghana,
interventions can be approximately 5–10 years, or more. Meanwhile, PROSAP, and SDCP) and the information below reflects costs across
Torero and colleagues (2016) provided an estimate of the economic all implemented activities.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 49 of 191

Discussion non‐agricultural labor markets. We found that farm‐to‐market

transport infrastructure interventions increased non‐farm income
Summary of findings. We identified a relatively large body of studies ˆ = 0.08; 95% CI: 0.02, 0.14; k = 16) and non‐farm labor out-
(SMD
that rigorously and quantitatively examined the effects of transport ˆ = 0.07; 95% CI: 0.02, 0.14; k = 13).
comes (SMD
infrastructure interventions, including a total of 83 papers reporting on We found substantial variation between the studies in each of
41 interventions implemented in 30 countries. Of these interven- these outcome categories. For example, we found that the B2P
tions, 11 were multi‐component interventions that combined intervention in Nicaragua (Brooks & Donovan, 2020) and Rwanda
infrastructure interventions with components such as promoting (Thomas et al., 2021) had different effects on farm investment, non‐
the uptake of new technologies or practices or providing marketing farm income, and volume outcomes (although the length of exposure
training or capacity development. The most studied infrastructure to the intervention was shorter in Thomas et al., 2021). Interventions
intervention was investment in roads (28 interventions), and all within the same country (Bangladesh) also had heterogeneous effects
studies employed a QED. Included studies covered follow‐up periods (Khandker et al., 2009; Khandker & Koolwal, 2011; Blankespoor
between seven months and 24 years, and 15 of the 41 interventions et al., 2022).
provided information on costs. When we examined potential sources of this heterogeneity, we
Farm‐to‐market transport infrastructure interventions had posi- found very little effect of intervention year, type of evaluation
tive effects along all stages of the theory of change, though these method used, length of follow‐up period, length of exposure to the
improvements were generally small. We found an overall very intervention, or whether it was a single‐ or multi‐component
small, positive, and statistically significant effect on transaction intervention. Additional analyses, such as subgroup analysis and
costs equivalent to a reduction in transaction costs findings related to intervention barriers and facilitators, further
ˆ = 0.09; 95% CI: 0.02, 0.16; k = 9). Transport infrastructure
(SMD indicate that effects are highly variable according to context and
interventions also affected access to markets, resulting in a very intervention type. We do not identify consistent patterns of higher or
small, positive, and statistically significant effect on farm income lower effects either across specific socio‐economic groups (wealth
ˆ = 0.08; 95% CI: 0.04, 0.13; k = 19). The overall effect on
(SMD level and gender) or across time.
income and wealth was also very small, positive, and statistically This heterogeneity can be explained by many different factors.
ˆ = 0.05; 95% CI: 0.02, 0.08; k = 27). We did not
significant (SMD First, farm‐to‐market transport infrastructure interventions themselves
find that these gains were realized further down the causal chain, are heterogeneous and are likely to have different effects. These
as we found no improvements in food security and nutrition. interventions improved roads or bridges but were diverse in their
However, the number of studies focused on these more distal activities (construction or rehabilitation) and scale (e.g., feeder rural
outcomes was low. roads, highways). Interventions including maintenance of developed
Farm‐to‐market transport infrastructure interventions also resulted infrastructure may have protected the infrastructure from damage.
in very small improvements in the adoption of fertilizers, pesticides, and Second, across countries, effects are expected to be different across
herbicides (SMDˆ = 0.08; 95% CI: 0.03, 0.14; k = 15), improved vari- targeted beneficiaries' baseline market access (commercialized vs.
ˆ = 0.05; 95% CI: 0.02, 0.08; k = 6), and
eties and quality seeds (SMD subsistence farmers). Third, the local context plays a significant role in
moderate improvements in farm investments in machinery and labor the magnitude of effects. However, we lacked sufficient studies or
ˆ = 0.30; 95% CI: 0.05, 0.55; k = 7). The evidence also
inputs (SMD accompanying data to test for many of the possible sources of
highlights some small gains in yields for food and cash crops heterogeneity.
ˆ = 0.13; 95% CI: 0.04, 0.22; k = 15). Across the theory of
(SMD
change, we found that the largest effects occurred in farm investments Implications for policymakers and implementers.
and yields outcomes. • Policymakers and implementers should be mindful of the hetero-
The evidence shows the competing effects of farm‐to‐market geneous effects of farm‐to‐market transport infrastructure interven-
transport infrastructure interventions on prices received by farmers. tions on households. Evidence suggested that female‐headed
These interventions did not affect agricultural prices in a uniform way households or vulnerable households do not benefit from such
across intervention types and crop types. These differences in interventions in the same way that less‐vulnerable households or
magnitude of effects across outcomes (e.g., very small for transaction male‐headed households do, thereby contributing to greater local
costs, farm income, and income vs. moderate effects on farm inequities. Decision‐makers and implementers should carefully
investments) may be due to barriers that limit access to and benefits consider how the intervention's characteristics (large‐scale trans-
from markets. port infrastructure construction or simply a rural road improve-
Beyond these effects, farm‐to‐market transport infrastructure ment) will have varying effects and may neglect different groups
interventions changed farmers' behavior and decision‐making. The of people.
evidence highlights that market access resulted, in some contexts, • Farm‐to‐market transport infrastructure interventions offer farm-
in shifting production and sales from food crops to cash crops. ers greater market access for their products but also enable
Similarly, farm‐to‐market transport infrastructure interventions led farmers to also take advantage of new income generating
laborers to move away from agricultural labor markets towards opportunities beyond the agricultural sector. In the long‐term,
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
50 of 191 | MARION ET AL.

F I G U R E 31 Theory of change for outcome market information interventions. The outcome types included in this systematic review are
immediate outcomes (green), agricultural production outcomes (purple), intermediate outcomes (teal), and welfare outcomes (light blue).
Outcomes in orange are other relevant immediate outcomes.

these interventions may lead to structural transformation of the review). This is particularly relevant as ports might have higher
local economy (i.e., labor is permanently reallocated to more effects on prices of imported and exported products.
productive non‐agricultural sectors). However, it is important to
consider the short‐term effects on the local agricultural sector of
these changes. For example, vast and rapid rural exodus towards 5.3.2 | Chapter 2: Output market information
urban areas may have negative implications on the local interventions
agricultural production and prices.
How can output market information affect agricultural, socio‐
Implications for researchers. economic, food security, and nutrition outcomes?
• Further research could focus on single‐component interventions Output market information interventions provide information about
on farm‐to‐market transport infrastructure interventions in order to market conditions, price inputs, and prices of produce to be sold. This
more clearly attribute the effects of these initiatives. We found chapter covers a wide array of interventions including, but not limited
that almost 30% of interventions from the included studies are to, information and communication technologies (ICT), informational
multi‐component interventions. In these included studies, authors campaigns, and supporting access to information technologies.
quantified the overall effect of the intervention; therefore, in most Improved access to information can affect farmers' decisions
cases, they cannot disentangle how much of the effect is concerning the inputs, techniques, and technologies to be employed
attributable to each activity. in the production stage. Information can also help farmers to
• Effects may be greater when farm‐to‐market transport infra- determine output prices, identify new markets, and access new
structure interventions are associated with additional marketing technologies.
support and capacity development. The available evidence does This chapter focuses on two types of output market information
not permit strong conclusions on this topic; therefore, more interventions: (i) provision of output market information via mobile
research comparing the effects with and without these comple- phones, internet, and other forms of communication; and (ii) and
mentary interventions would be beneficial. investments in information and communication technologies. The flow
• Research should further analyze the role of farm‐to‐market diagram in Figure 31 presents the theory of change: how output
transport infrastructure interventions on households' decisions market information might affect agricultural, socio‐economic, food
to engage in non‐farm sectors. The limited available evidence security and nutrition outcomes.
here highlighted that in some contexts, these interventions
increased households' non‐farm income and labor participation. Provision of output market information. The first group of interven-
• Further research is needed on the effectiveness of ports and tions includes initiatives providing farmers with information about
waterways infrastructure. The analyzed body of evidence high- the market. Market information can be delivered in several different
lighted a gap represented by a scarcity of studies exploring these ways. For instance, information can be sent via phone (e.g., SMS, and
specific types of interventions (only one study was included in this phone call), accessed via the internet (e.g., apps and websites) or via
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 51 of 191

other traditional modes of information dissemination such as radios above from the provision of market information. For example, farmers
and individual (peer) networks (Munyua, 2007).12 improved their productivity by accessing information via the internet
Households can benefit from access to output market informa- (Jung & López‐Bazo, 2020; LoPiccalo, 2021).
tion at different stages of cultivation and marketing. Crop prices are Despite the promise of output market information interventions,
highly difficult to predict at the local level in the presence of many factors hinder their effectiveness. Notably, this includes lack of
considerable unexpected (both non‐seasonal and seasonal) price training and practical exposure towards ICTs, and low ICT literacy
changes (Kaminski et al., 2014; Cardell & Michelson, 2023; Hill & could prevent small farmers from effectively using ICT tools and
Fuje, 2020). In addition, producers' livelihoods are most often maximizing choices (Rajoria et al., 2022). These limitations might
dependent on a few crops and therefore, households' income and undermine the effects of interventions on productivity and profit-
welfare are tied to crop price levels (Dehn, 2000; Collier & Gunning, ability, and ultimately lead to null impact on income and revenues.
1999; Bussolo et al., 2007; Mekasha et al., 2022). Limited credit access can similarly inhibit investments and
At the time of planting, output market information can facilitate adoption of inputs and new technologies. Access to credit is
farmers' decisions about the type of product and the quantity to be considered a key factor for boosting small farmers growth and
produced in order to exploit temporary price changes based on local improving household income and food security (Feder et al., 1990;
or international changes in supply and demand (Nakasone, 2013; Diagne and Zeller 2001; Awotide et al., 2015).
Ndimbwa et al., 2021; Collier & Goderis, 2012; Beck et al., 2018). A Lastly, unstable power supply, low network connectivity, lack of
study of global commodity price rises between 1963 and 2008 repairing stations and facilities in rural areas (Mishra et al., 2020;
showed that these price increases were followed by an increase in Rajoria et al., 2022) also determine the effectiveness of these
output (Collier & Goderis, 2012). Conversely, Beck and colleagues interventions in the medium and long‐run. Other relevant contextual
(2019) found that, when coffee prices were low in Vietnam, factors include limited access to transport infrastructure, equipment,
households were more likely engaged in wage employment and less and environmental or economic shocks. For example, farmers in
in agricultural production. Tanzania did not benefit from access to agricultural information from
After the harvest, farmers can exploit market information to the internet as they did not have adequate equipment and resources
determine the best time, quantity and location of sales to obtain the (Ndimbwa et al., 2021).
highest price for their output at a lower search cost. Prices often
change massively across markets in LMICs meaning that search costs Description of included studies
for farmers are high. In Niger, for example, across pairs of markets, We included 81 studies examining the effects of 47 output market
grain price varied on average by 19% of the mean national grain price information interventions. One of the analyzed studies is an ongoing
(Aker, 2008). Search costs from information asymmetries are evaluation of the FPPM project, aimed at strengthening market
therefore a key component of farmers' transaction costs. Based on efficiency and agricultural productivity in DRC (Ragasa et al., 2016).
data in Sri Lanka, De Silva and Ratnadiwakara (2010) found that The synthesis‐of‐findings analysis relies on effects from 57 papers
search costs represented 69.8% of farmers' total transaction costs. (49 main studies and eight linked papers), as 24 papers do not provide
With output market information, farmers can receive higher prices any additional results. These 24 papers are ongoing studies, previous
for their produce and face lower transaction costs increasing market versions, or unpublished versions of already included studies. These
access and participation (Okello et al., 2007; Magesa et al., 2020; eight linked papers report additional outcome data about the effects
Nakasone, 2013; Giovannucci & Shepherd, 2007). of an intervention (i.e., data on outcomes or prioritized outcome
Over time, these effects from access output market information indicators that are not reported by the main study).
are likely to encourage farmers to expand cultivation and possibly
adopt and invest in improved agricultural practices (Ndimbwa et al., Intervention, inputs, and activities
2021; Mittal & Tripathi, 2009; Goyal, 2010). In turn, farmers' also We identified 47 output market information interventions implemen-
increase income, profitability and ultimately improve farmers' food ted in 19 countries across 5 regions (Figure 32). Interventions in
security and wealth. Africa represent more than half of the included studies (28), followed
by South Asia (11), and Latin American and the Caribbean (4). The
Investments in ICTs infrastructure. A second class of interventions remaining four interventions were implemented in East Asia and the
increases access to technologies required to access market information Pacific (3), and one was implemented in Europe and Central Asia. The
rather than directly disseminating the information. These interventions individual countries with the largest number of included interventions
include investments in mobile phone services, internet or broadband are India (8), followed by Ghana (5), Kenya (4), Ethiopia (3), and Peru
access, and the basic provision of electricity. Strengthening ICT (3). One study analyzed the effects of a multi‐country intervention
infrastructure is considered a key element for supporting farmers' implemented in DRC, Rwanda, and Uganda (Nkonya et al., 2013), and
participation in markets and can then unlock the effects mentioned one examined an intervention in Kenya and Rwanda (Kamande &
Nafula, 2016).
12 Supporting Information: Appendix 3 for Chapter 2 provides
Supporting Information: Appendices 1 and 2 for Chapter 2 provide the references of
included studies and additional references, respectively. information on the 47 interventions including country, type of
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
52 of 191 | MARION ET AL.

F I G U R E 32 Distribution of interventions by region.

F I G U R E 33 Distribution of interventions by subcategory.

intervention, and year the intervention began. The most frequent the members were women. Cooperative members were kept informed
type of intervention evaluated is the provision of output market about market prices via SMS, helping them to make informed decisions
information (31), via mobile and/or internet (25), and via other forms of about when and where to sell their products. It also constructed four
communication (6), followed by the groups of interventions belonging community greenhouses, established three cold‐storage units for
to investments in ICT infrastructure (16) (Figure 33). agricultural crops, and promoted innovative agro‐technology practices
Table 12 provides an overview of the interventions included in for water management. Finally, it strengthened the participants'
this chapter in terms of country, type of intervention, and year the marketing capacities and linked them with markets.
intervention began. Another example is the FPPM in DRC, which took an active role
Nine of the included interventions used a multi‐component in the rehabilitation of feeder roads and rivers links to areas of
approach (Table 13). This reflects a long‐standing practice in concentrated agricultural production; it also provided information to
agricultural development policy, whereby projects combine mutually farmers through radio broadcasts, as well as providing storage
supporting agricultural and non‐agricultural activities addressing facilities. In the analysis below, we conducted a moderator analysis to
several issues that create a binding constraint on growth. determine whether the estimates from multi‐component interven-
For example, the ARMA66 project in Armenia established four tions are on average statistically significantly different from estimates
cooperatives: in three, all members were women; in the other, half of from single‐component interventions.
 T A B L E 12 Table of included output market information interventions.
MARION

First year of
Intervention name Acronym Country (ISO3 code) Type intervention
ET AL.

ABC Program ABC Niger (NER) Investments in ICT infrastructure 2009

Women's Economic Empowerment in Rural Communities of Vayots Dzor ARMA66 Armenia (ARM) Provision of output market information 2010
Region Project (mobile and internet)

Barangay Line Enhancement Program BLEP Philippines (PHL) Investments in ICT infrastructure 2004

Community Cellular Network CCN Philippines (PHL) Investments in ICT infrastructure 2017

Community Knowledge Workers CKW Uganda (UGA) Provision of output market information 2010
(mobile and internet)

Weather Information CPWI Colombia (COL) Provision of output market information 2009
(mobile and internet)

China Rural Electrification Program CREP China (CHN) Investments in ICT infrastructure 1998

Community‐owned Micro Hydropower CWM Nepal (NPL) Investments in ICT infrastructure 1996

E‐CHOUPAL E‐CHOUPAL India (IND) Provision of output market information 2000

(kiosk) (mobile and internet)

Ethiopian Commodity Exchange ECX Ethiopia (ETH) Provision of output market information 2008
(mobile and internet)

Ethiopia Market Information System EMIS Ethiopia (ETH) Provision of output market information 2008
(mobile and internet)

Electrification Through Microhydro Plants ETMHP Nepal (NPL) Investments in ICT infrastructure 2004

Esoko Esoko Ghana (GHA) Provision of output market information 2009

(mobile and internet)

Esoko SIM Esoko SIM Ghana (GHA) Provision of output market information 2008
(mobile and internet)

Esoko and Kenya Agricultural Commodity Exchange Esoko and KACE Rwanda (RWA) Provision of output market information 1997
(mobile and internet)

EsokoSEND EsokoSEND Ghana (GHA) Provision of output market information 2006

(mobile and internet)

Food for All Project in Kenya F4APK Kenya (KEN) Provision of output market information 2019
(mobile and internet)

FARMEX FARMEX India (IND) Provision of output market information 2009 and 2017
(mobile and internet)
|

Fund for Investments in Telecommunications FITEL Peru (PER) Investments in ICT infrastructure 2002

(Continues)
53 of 191

18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
 TABLE 12 (Continued)
54 of 191

First year of
Intervention name Acronym Country (ISO3 code) Type intervention
|

Food Production, Processing, and Marketing FPPM Democratic Republic of the Provision of output market information 2011
Congo (COD) (other forms of communication)

Ghana Market Information Service GMIS Ghana (GHA) Provision of output market information 2011
(mobile and internet)

Integrated Agricultural Research for Development IAR4D Democratic Republic of the Provision of output market information 2008
Congo (COD) (mobile and internet)

Jyotigram Yojana JGY India (IND) Investments in ICT infrastructure 2003

Kenya Maize Development Program KMDP Kenya (KEN) Provision of output market information 2004
(other forms of communication)

DrumNet DrumNet Kenya (KEN) Provision of output market information 2002

(mobile and internet)

Kitabu cha biashara, or “business book” Kichabi Tanzania (TZA) Provision of output market information 2014
(other forms of communication)

Malawi Market Information Services MMIS Malawi (MWI) Provision of output market information 2004
(mobile and internet)

Mobile phone rollout MPR Mozambique (MOZ) Investments in ICT infrastructure 1997

Mobile Phone Tower MPT Ethiopia (ETH) Investments in ICT infrastructure 2000

Maize and Soybean Prices via Mobile Phone MSPMP Malawi (MWI) Provision of output market information 2019
(mobile and internet)

National Agricultural Extension and Research Liaison Services NAERLS Nigeria (NGA) Provision of output market information 2009
(other forms of communication)

National Electrification Program NEP Ghana (GHA) Investments in ICT infrastructure 1989

Nutrient Management Interventions and Information NUI (Treatment 2) Nigeria (NGA) Provision of output market information 2016
(mobile and internet)

Purchase for Progress P4P Democratic Republic of the Provision of output market information 2012
Congo (COD) (other forms of communication)

Peru Cell Phone Crop Information PCPCI Peru (PER) Provision of output market information 2010
(mobile and internet)

Public Phones in Rural Areas PPRA Peru (PER) Investments in ICT infrastructure 1994

Sustainable Rural Electrification Project RE Bhutan (BTN) Investments in ICT infrastructure 2000

Rajiv Gandhi Grameen Vidyutikaran Yojana RGGVY India (IND) Investments in ICT infrastructure 2005
MARION
ET AL.

18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
 TABLE 12 (Continued)
MARION

First year of
Intervention name Acronym Country (ISO3 code) Type intervention
ET AL.

Reuters Market Light RML India (IND) Provision of output market information 2010
(mobile and internet)

Rollout of mobile phone coverage ROCPC Niger (NER) Investments in ICT infrastructure 2001

Smallholder Horticulture Marketing Program SHoMaP Kenya (KEN) Provision of output market information 2008
(mobile and internet)

Three formal information systems: Agricultural Market Information System SIF Mozambique (MOZ) Provision of output market information 1991
(SIMA), INFOCOM, and the Famine Early Warning System Network (mobile and internet)
(FEWS NET)

Trading System TS India (IND) Provision of output market information 2008

(mobile and internet)

Uganda Market Information Service UMIS Uganda (UGA) Provision of output market information 2000
(other forms of communication)

Village Phone VP Rwanda (RWA) Investments in ICT infrastructure 2006

Women Farmers with Global Potential WFGP India (IND) Provision of output market information 2008
(mobile and internet)

National Agriculture Market eNAM India (IND) Provision of output market information 2016
(mobile and internet)

Note: ISO3 country code provided in parentheses (). If the introduction of an intervention is staggered, the year of intervention refers to the earliest year in which observations were undertaken for the
intervention.
| 55 of 191

18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
56 of 191 | MARION ET AL.

T A B L E 13 Descriptions of other interventions combined with output market information interventions.

Intervention (country Description of other intervention components combined with transport
code)a Type infrastructure interventions

ARMA66 (ARM) Provision of output market Offers technical support throughout the agricultural cycle, connects cooperatives
information (mobile and internet) to microfinance institutions, and promotes products through market days, SMS
updates, and packaging support. Cooperative members are kept informed about
market prices via SMS, helping them to make informed decisions about when and
where to sell their products.

ECX (ETH) Provision of output market The Ethiopian Commodity Exchange (ECX) is a modern and transparent commodity
information (mobile and internet) spot trading platform, operating on a membership basis, wherein buyers and sellers
acquire transferable membership seats. The platform quickly shares real‐time
market info via electronic tickers, the internet, voice response, and media. It runs
94 tickers across the country, displaying commodity details, real‐time prices, and
price changes.

EMIS (ETH) Provision of output market EMIS disseminated information on wholesale prices to all market actors, including
information (mobile and internet) smallholder farmers, through various channels such as a website, electronic tickers
in 250 rural markets, SMS, interactive voice response services, radio broadcasts
(three times a day), television broadcasts (twice a day), newspapers, and
newsletters (daily, monthly, and half‐yearly).

FPPM (COD) Provision of output market FPPM facilitated access of smallholder groups to inputs, services, and markets,
information (other forms of enabling smallholder farms to become income‐generating businesses through the
communication) identification of potential partners in the processing sector and storage facilities. It
developed a draft strategy for implementing a market information system in the
FPPM project zone, focusing on radio broadcasts to producers to maximize
coverage of rural households, and rehabilitation of feeder roads or river links.

IAR4D (COD) Provision of output market IAR4D established innovation platforms in the East and Central African region.
information (mobile and internet) These platforms are forums that bring together different agricultural stakeholders
(e.g., farmers; researchers; extension/advisory services; private sector businesses
or agribusinesses such as processors, traders and transporters; policymakers; civil
society organizations) to handle critical development challenges and/or
opportunities. The project provided and facilitated access to research and
extension services, access to market information, horizontal linkages (farmer
groups for production and marketing), and vertical linkages (agricultural traders,
credit providers).

P4P (COD) Provision of output market P4P updated farmers on the prices of commodities from time to time via mobile
information (other forms of phones. The farmers could therefore take advantage of this market information to
communication) sell their produce when prices were favorable. The program provided support for
farmers by providing: assistance with linkages to markets; assistance with
cultivation, harvesting, and processing tools; technical assistance on farming
techniques; market information; and access to credit. Additionally, transport
facilities such as bicycles and wheelbarrows were provided to aid in the
distribution of produce.

SHoMaP (KEN) Provision of output market SHoMaP disseminated market updates and trends to farmers and traders through
information (mobile and internet) SMS and radio broadcasts to make informed decisions about their horticultural
activities. This intervention also combined spot improvement of rural access roads
and development or improvement of downstream physical market infrastructure.

SIF (MOZ) Provision of output market SIF provided farmers with access to market information through SIMA, INFOCOM,
information (mobile and internet) and FEWS NET. These provide agricultural market information to farmers to help
them negotiate better prices and to assist traders and processors in identifying
opportunities. The information covers various commodities (e.g., maize, beans,
rice), providing data on prices, transportation costs, quantities available, and
market analysis. The information is disseminated through email, radio, television,
websites, and newspapers.

eNAM (IND) Provision of output market eNAM is an electronic trading portal designed to connect existing agricultural
information (mobile and internet) produce market committee markets. It provides various services related to the
committee, including information on commodity arrivals, prices, buy‐and‐sell trade
offers, response mechanisms, payment settlements, and grievance redressal.
a
ISO3 country code provided in parentheses ().
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 57 of 191

Comparison Another significant share of studies were carried out 2–4 years
In the included 49 studies, the comparison groups used for these after the intervention began (n = 11), whilst the remaining studies
estimations nearly exclusively focused on comparing the outcomes of were conducted over four years after the interventions, with eight
beneficiaries of output market information interventions to non‐ studies lasting 10 years or more. The longest covered period is 264
beneficiaries located in a different area (88%). Six percent of studies months (22 years; Akpandjar & Kitchens, 2017), whilst the shortest is
(n = 3) exploited a phased‐in (or pipeline) approach of the interven- two months (Parker et al., 2016).
tions to identify a comparison group. Two studies (Chindarkar et al.,
2020; Belay & Ayalew, 2020) took advantage of the varying levels of Synthesis of findings
beneficiaries' treatment exposure to estimate the effects of these We identified a total of 990 estimates from the data extraction of
interventions. studies under the output market information intervention category. Of
Chindarkar and colleagues (2020) relied on a measure of village these, 172 were selected for the analysis of effects. We report
exposure to the intervention (JGY in India) and Belay and Ayalew additional information on the individual meta‐analyses (sensitivity
(2020) identified the control group based on the variation in timing analysis, outlier analysis, publication bias analysis, moderator analysis,
and spatial distance to the ECX. The comparison in Parker and meta‐analyses without high risk of bias estimates) in Supporting
colleagues (2016) is price information before the unexpected ban of Information: Appendix 5 for Chapter 2.
the RML intervention in India. Supporting Information: Appendix 4
for Chapter 2 provides information on these studies including the Risk of bias. The risk of bias of each estimate across the 49 included
evaluated intervention, its subcategory, length of follow‐up in years, studies was assessed using the Risk of Bias Assessment Tool
study design, and the type of comparison. (Supporting Information: Appendix 10 for background and methods
sections). Figure 36 presents a summary of risk of bias assessment
Outcomes results. We found that 51% of included estimates were rated as
The analyzed studies cover 17 outcomes included in our framework, having a high risk of bias, 44% were considered to have some
as specified in the protocol (Villar et al., 2023). The most frequently concerns related to risk of bias, and the remaining 5% of estimates
analyzed outcomes are prices (n = 27 studies), followed by farm were assessed as having a low risk of bias.
income (n = 24), sales (n = 17), use or adoption of improved inputs, Figure 36 shows a discrepancy between the risk of bias of QED
technologies and practices (n = 12), and total income and wealth studies compared to RCT studies, as 54% of the former is rated as
(n = 11) (Figure 34). high risk of bias, compared to 38% for the latter. This risk of bias of
included estimates was based on six domains (selection bias;
Study design confounding; performance bias; spillover, crossover, and contamina-
Of the 49 main studies, most used QEDs (n = 41) to evaluate tion; outcome measurement bias; and reporting bias).
intervention effects. The majority of these studies relied on DID When examining individual risk of bias domains, the most
and FE approaches (n = 24) and statistical matching methods common source relates to confounding bias (28% of estimates are
(n = 15). One study exploited a natural experiment (Parker et al., rated as high risk of bias), which was often due to either a lack of
2016), and another utilized an interrupted time series balance, or unreported balance, at baseline. Additional concerns
(Goyal, 2010). Burlig and Preonas (2021) conducted a regression included spillover, crossover and contamination (25% of estimates
discontinuity estimation using a population‐based eligibility were rated as high risk of bias), and reporting bias (15% of estimates
threshold (electrification under RGGVY only occurred in villages were rated as high risk of bias). Estimates were not likely to be at high
if they contained at least one neighborhood with more than risk for outcome measurement bias and performance bias.
300 people).
Two studies conducted an IV estimation (Chakravorty et al., Summary of the meta‐analyses results of output market interventions. -
2016; Meeks et al., 2022). The remaining eight studies used Table 14 displays the meta‐analyses results across outcome types
experimental designs. Four of these studies exploited village‐level covered in the subsequent sections.
clustered RCTs. Three studies estimated the effects using two study
designs: regression discontinuity, DID, and FE approach (Burlig & Immediate outcomes. The immediate outcomes category covers the
Preonas, 2021); and statistical matching, DID, and FE approach (Lee first outcomes that may affect farmers across the causal chain. They
et al., 2020; Rajkhowa, 2021). cover financial resources (e.g., transaction costs and credit), market
Figure 35 also shows the distribution of the studies' different linkages and bargaining power (e.g., group participation and prices),
lengths of follow up (study periods). This is the difference between and the adoption of new practices and agricultural approaches (e.g.,
the start date of the intervention and the most recent outcome data use/adoption of improved inputs, technologies and practices, and farm
included in the study's analysis. These 47 interventions began investments).
between 1989 and 2019 and had an average length of roughly 5.1
years. Fourteen studies were undertaken within two years after the Transaction costs. We found six studies reporting six estimates of the
intervention took place. effect of output market information on transaction costs. Regionally,
18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ET AL.
MARION

Distribution of outcome types.

F I G U R E 34
|58 of 191
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 59 of 191

F I G U R E 35 Distribution of study follow‐up period length.

these studies are from Africa (n = 4), Latin America and the Caribbean output market infrastructure interventions on transaction cost
(n = 1) and South Asia (n = 1). Of the six studies, two examined the outcomes is even smaller and still not statistically significant
effects of investments in ICT infrastructure—for example, the ˆ = 0.01, 95% CI: −0.12 to 0.13; p = 0.94 ).
(SMD
provision of at least one public satellite payphone from the FITEL
program13 in Peru (Beuermann, 2015), and accessibility of mobile Use/adoption of improved varieties and quality seeds. We included
phones through mobile phone rollout in Mozambique (Zant, 2019). seven estimates that examined the effect of output market informa-
Two studies evaluated the provision of agricultural price information tion interventions on use/adoption of improved varieties and quality
through mobile and internet through the RML program in India seeds in the production processes from five studies. All the studies
(Fafchamps & Minten, 2012), and ICT‐based market interventions on focused on Africa (n = 5). We found three estimates in Ragasa and
intervention costs for smallholder farmers in Malawi (Katengeza colleagues (2016), which were focused on three regions of DRC with
et al., 2014). Additionally, two evaluated multi‐component interven- independent comparative groups.
tions (SHoMaP and P4P). Three studies evaluated multi‐component interventions (P4P,
Most studies analyzed the effects at the household level (k = 5), apart FPPM, and ECX), which combined providing output market informa-
from the mobile phone rollout in Mozambique (Zant, 2019), which tion through mobile, internet, or other forms of ICT infrastructure
evaluated impacts at the market level. Zant (2019) identified markets with creating new marketplace, or farm‐to‐market transport infra-
within 35 km of the mobile phone tower to control for differences in structure. Two studies analyzed the effects of single interventions
access and affordability, as a means of analyzing transport costs of maize such as provision of market information using mobile phones in Ghana
per kilogram for markets with and without mobile phone facilities. (EsokoSEND) (Al‐Hassan et al., 2013) and membership durations
Overall, output market information interventions did not affect (F4APK) (Schipper et al., 2021).
transaction costs. The overall weighted average effect is very small Output market information interventions increased use/adoption of
ˆ = 0.03; 95% CI: −0.06 to
and not statistically significant (SMD improved varieties and quality seeds. The overall weighted average
0.14; p = 0.42) (Figure 37).The outcomes were reversed for analysis, ˆ = 0.14 ; 95%
effect is small, positive, and statistically significant (SMD
implying that a higher effect meant lower transaction costs. Effects CI: 0.04 to 0.25; p = 0.01) (Figure 38). Effect sizes ranged from 0.04
ranged from −0.08 to 1.33 standard deviations. There is moderate to 0.39 with a substantial heterogeneity (Q (6) = 19.58, p < 0.0003,
heterogeneity among studies (Q(5) = 9.62, p = 0.09, τ̂ 2 = 0.006, τ̂ 2 = 0.01, I2 = 69.36%).
I2 = 48.03%). None of the included estimates are statistically The estimates from EsokoSEND (Al‐Hassan et al., 2013) and
significant. P4P (Anguko, 2018) found significant effects of the allocation of
None of the moderator variables significantly contributed to improved seeds on coffee, sesame, maize, wheat, haricot, and
the variation in estimates. Of the six included effect sizes, three beans ( ĝ = 0.30 ; 95% CI: 0.09 to 0.51; p = 0.005), and the use of
remain after excluding high risk of bias estimates. The effect of modern agricultural practices ( ĝ = 0.39; 95% CI: 0.23 to
0.54; p < 0.001), respectively.
13 None of the moderator variables significantly contributed to the
Information on interventions is provided in Supporting Information: Appendix 3 for
Chapter 2. variation in estimates. Of the seven included effect sizes, three remain
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
60 of 191 | MARION ET AL.

F I G U R E 36 Risk of bias of included estimates between quasi‐experimental design (QED) (top) and randomized controlled trial (RCT)
(bottom).

after excluding high risk of bias estimates. The effect of output market herbicides. All studies focused on Africa. We included two estimates
infrastructure interventions on use/adoption of improved varieties and from Ragasa and colleagues (2016), as the evaluation focused on two
quality seeds outcomes slightly increases and is still statistically regions of DRC with independent comparative groups.
ˆ = 0.18, 95% CI: 0.004 to 0.36; p = 0.044 ).
significant (SMD Two studies evaluated multi‐component interventions (FPPM
and ECX). The multi‐component interventions combined providing
Use/adoption of fertilizers, pesticides, and herbicides. We found seven output market information through mobile, internet, or other forms of
estimates from six studies that examined the effect of output market ICT infrastructure with creating new marketplaces or farm‐market
information interventions on use/adoption of fertilizers, pesticides, and infrastructure.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 61 of 191

T A B L E 14 Summary of meta‐analyses results of output market information interventions.

Outcome types Outcome subcategory SMD 95% CI k p Value

Immediate outcomes Transaction costs 0.04 −0.06, 0.14 6 0.42

Group participation N/A

Use/adoption of output market 0.14 −0.08, 0.37 3 0.20

information

Use/adoption of fertilizers, 0.04 −0.04, 0.12 7 0.34

pesticides, and herbicides

Use/adoption of improved 0.14 0.04, 0.25 7 0.01

varieties and quality seeds

Use/adoption of improved 0.21 0.05, 0.38 9 0.01

inputs (machinery and irrigation)

Farm investment (total) 0.12 −0.01, 0.25 5 0.08

Farm investment (seeds) N/A

Farm investment (fertilizers) N/A

Farm investment (labor) N/A

Farm investment (livestock) N/A

Credit 0.03 −0.12, 0.17 6 0.72

Prices 0.06 0.001, 0.12 18 0.047

Agricultural production Crop loss 0.16 −0.14, 0.46 3 0.30

outcomes
Yields 0.07 −0.09, 0.23 6 0.37

Volume 0.05 −0.02, 0.11 10 0.17

Produce type (area) 0.08 −0.03, 0.19 4 0.15

Produce type (diversity) 0.14 0.07, 0.22 6 0.001

Intermediate outcomes Sales 0.03 −0.03, 0.1 17 0.28

Farm income 0.09 0.04, 0.13 24 <0.001

Off‐farm outcomes Labor market outcomes N/A

(agriculture only)

Non‐farm outcomes Non‐farm income 0.07 0.01, 0.13 6 0.03

Labor market outcomes 0.13 0.06, 0.20 7 <0.001

Welfare outcomes Total income and wealth 0.07 0.01, 0.13 11 0.03

Food security 0.04 −0.04, 0.12 7 0.36

Note: Outcome subcategories in bold have a statistically significant SMD at a 5% significance level. k is the number of estimates. N/A indicates that we did
not conduct a meta‐analysis for these outcome types. In these cases, k = 1 or k = 2 and we chose not to conduct a meta‐analysis in those cases due to
various differences (e.g., intervention types, regions).

Four studies analyzed the effects of single interventions such as Output market information interventions did not affect the use/
provision of market information using mobile phones in Ghana adoption of fertilizers, pesticides, and herbicides. The overall weighted
(EsokoSEND) (Al‐Hassan et al., 2013) and membership durations average effect is very small and not statistically significant
(F4APK) (Schipper et al., 2021) in Kenya, site‐specific nutrient ˆ = 0.04 ; 95% CI: −0.04 to 0.12; p = 0.34) (Figure 39). Effect
(SMD
management information in Nigeria (NUI, Treatment 2) (Oyinbo sizes ranged from −0.06 to 0.25 with substantial heterogeneity
et al., 2022), and provision of “Yellow Pages” phone directories in (Q (6) = 14.19, p = 0.03, τ̂ 2 = 0.007, I2 = 57.73%).
Tanzania (Kichabi) (Dillon et al., 2020). All seven estimates used the Oyinbo and colleagues (2022) implemented an RCT for smallholders
household as the unit of analysis. Outcomes variables here are binary in Northern Nigeria, which provided site‐specific nutrient management
variables indicating whether households used fertilizer and continu- interventions and information on expected returns before the planting
ous variables on the quantity of fertilizer used. season. They found a positive and statistically significant but small effect
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
62 of 191 | MARION ET AL.

F I G U R E 37 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of output market information
interventions on transaction costs.

F I G U R E 38 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of output market information
interventions on the use/adoption of improved varieties and quality seeds.

F I G U R E 39 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of output market information
interventions on the use/adoption of fertilizers, pesticides, and herbicides.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 63 of 191

on the use of improved seeds (ĝ = 0.26; 95% CI: 0.10 to use per capita, while the outcome variable in Chindarkar and
0.41; p = 0.001), but this estimate was identified as a clear outlier in colleagues (2020), Lombardini (2017), and Schipper and colleagues
the context of this model. Belay and Ayalew (2020) found that treated (2021) was a binary variable indicating whether households used
households—those with access to price information through the ECX in irrigation facilities (tubewells, drip irrigation systems, and irrigation
Ethiopia—were more likely to use fertilizers on coffee‐ and sesame‐ levels, respectively).
cultivated land (ĝ = 0.12; 95% CI: 0.02 to 0.23; p = 0.02). Output market information interventions increased the use/adop-
None of the moderator variables significantly contributed to the tion of improved inputs (machinery and irrigation). The overall weighted
variation in estimates. Of the seven included effect sizes, two remain average effect is positive, moderate, and statistically significant
after excluding high risk of bias estimates. The effect of output market ˆ = 0.21; 95% CI: 0.05 to 0.38; p = 0.01) (Figure 40). Effect sizes
(SMD
information interventions on use/adoption of fertilizers, pesticides, and ranged from 0.01 to 0.90 and there is considerable heterogeneity
herbicides outcomes is slightly smaller and still not statistically (Q (8) = 77.9, p < 0.001, τ̂ 2 = 0.05, I2 = 89.69%).
ˆ = 0.06, 95% CI: −0.06 to 0.18; p = 0.33).
significant (SMD The estimates from Lombardini (2017, ARMA66) and Chindarkar
and colleagues (2020, JGY) are statistically significant (ĝ = 0.89; 95%
Use/adoption of improved inputs (machinery and irrigation). Nine CI: 0.70 to 1.09; p < 0.001) and (ĝ = 0.34 ; 95% CI: 0.16 to
estimates from seven studies were included to examine the effect 0.52; p < 0.001, respectively). Belay and Ayalew (2020) found that
of output market information interventions on adopting improved inputs treated households—those with access to price information through
(machinery and irrigation) in production processes. Most of the studies the ECX in Ethiopia—were more likely to irrigate coffee‐ and sesame‐
focused on Africa (n = 5), followed by Europe and Central Asia (n = 1) cultivated land (ĝ = 0.23; 95% CI: 0.13 to 0.34; p < 0.001).
and South Asia (n = 1). We found three estimates in Ragasa and None of the moderator variables significantly contributed to the
colleagues (2016), as the evaluation focused on three regions of DRC variation in estimates. Of the nine included effect sizes, five remain
with independent comparative groups. after excluding high risk of bias estimates. The effect of output market
Three studies evaluated multi‐component interventions infrastructure interventions on use/adoption of improved inputs
(ARMA66, FPPM, and ECX), which combined providing output market (machinery and irrigation) outcomes increases and is still statistically
information through mobile, internet, or other forms of ICT ˆ = 0.34, 95% CI: 0.06 to 0.63; p = 0.02).
significant (SMD
infrastructure with creating new marketplaces, storage facilities, or
farm‐market infrastructure. The other four studies analyzed effects of Use/adoption of output market information. Three studies examined
single interventions, such as provision of market information using the effects of three output market information interventions on the
mobile phones in Ghana (EsokoSEND, Al‐Hassan et al., 2013) and use/adoption of output market information outcomes. Two studies
membership durations (F4APK, Schipper et al., 2021) in Kenya, were from Africa, and the third was from Latin America. All three
provision of phone directories in Tanzania (Kichabi, Dillon et al., examined the effects of different interventions (there were no multi‐
2020), and the JGY program in India (Chindarkar et al., 2020). component interventions). While Aker and Ksoll (2016) evaluated the
The nine estimates used the household as their unit of analysis. basic adult literacy technology infrastructure program through the
Okello and colleagues (2020) looked at changes in household input mobile phone module on price information in Niger (ABC), Dillon and

F I G U R E 40 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of output market information
interventions on the use/adoption of improved inputs (machinery and irrigation).
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
64 of 191 | MARION ET AL.

colleagues (2020) looked at the effects of the provision of local from interventions providing information, mostly via mobile phone or
enterprise phone directories in Tanzania (Kichabi). Camacho and internet (k = 4), and in one case with other forms of communication.
Conover (2015) analyzed the effects of SMS price information on the Rajkhowa (2021) analyzed the FARMEX program, an intervention
uptake of improved practices in Colombia (CPWI). supporting farmers in India via real‐time agricultural extension
The three studies used the household as the unit of analysis. Aker services and a marketplace for seeds, fertilizers, and pesticides.
and Ksoll (2016) evaluated the effect using follow‐up on price Ogutu and colleagues (2020) evaluated the effectiveness of the
information as the outcome variable. Camacho and Conover (2015) DrumNet program in Kenya, which was introduced to improve
evaluated the effect according to whether the treatment group read the agricultural extension, access to credit, and marketing services for
informational mobile messages. Similarly, Dillon and colleagues (2020) smallholder farmers via an ICT platform for mobile phones; its
evaluated whether the dissemination of a phone directory of agricultural ultimate aim was to increase productivity. Schipper and colleagues
businesses and producers (Kichabi intervention) increased households' (2021) carried out an impact evaluation of F4APK, a platform which
likelihood of using a phone for crop‐related activities. provides agroeconomic tips and disseminates information about
Output market information led to a statistically significant increase weather and prices to farmers in Kenya.
in the use/adoption of output market information. The overall weighted The analyzed outcomes are expressed as measures of production
average effect is small, positive, and not statistically significant costs (Schipper et al., 2021; Oyinbo et al., 2022), total expenditures
ˆ = 0.14, 95% CI: −0.08 to 0.37, p = 0.20 ) (Figure 41). The effect
(SMD (Dillon et al., 2020), or as expenditure per acre in inputs (Ogutu et al.,
sizes range from −0.05 to 0.32. There is a substantial, statistically 2014; Rajkhowa, 2021). All five reported estimates are derived from
significant heterogeneity (Q (2) = 15.89, p < 0.001, τ̂ 2 = 0.03, household‐level analyses.
I2 = 87.42%) and considerable variation in effect sizes. Overall, output market information interventions did not affect total
Camacho and Conover (2019) found moderate, positive, and farm investment. The overall weighted average effect is small and
statistically significant effects (ĝ = 0.32; 95% CI: 0.14 to ˆ = 0.12; 95% CI: −0.01
positive but not statistically significant (SMD
0.50; p < 0.001). Beneficiary farmers were more likely to always to 0.25; p = 0.42) (Figure 42). Effects ranged from −0.06 to 0.27
read their text messages following the intervention (CPWI), which standard deviations. There is substantial heterogeneity among
sent daily texts with price and weather information. Authors studies (Q(4) = 9.90 , p = 0.04 , τ̂ 2 = 0.013, I2 = 59.56%). Oyinbo and
therefore considered treated farmers to be more engaged in ICT colleagues (2022) reported positive effects on farm investments in
when making production and marketing decisions. Aker and Ksoll production ( ĝ = 0.26; 95% CI: 0.11 to 0.44; p = 0.04) from an RCT
(2016) did not find that the provision of a mobile phone and training providing site‐specific recommendations alongside information of
on its use in Niger (ABC) increased households' likelihood of price variability on the costs of production.
following price information to inform their marketing decisions None of the moderator variables significantly contributed to the
( ĝ = −0.05; 95% CI: −0.15 to 0.04; p = 0.32). variation in estimates. Of the four included effect sizes, two remain
None of the moderator variables significantly contributed to the after excluding high risk of bias estimates. The effect of output market
variation in estimates. Of the three included effect sizes, one information interventions on total farm investment outcomes is still
remained after excluding high risk of bias estimates (Camacho & ˆ = 0.10; 95% CI: −0.03
small and statistically insignificant (SMD
Conover, 2019, discussed above). to 0.24; p = 0.12).

Farm investment (total). Five studies, each with an independent Prices. Eighteen estimates from 18 studies evaluated the effect of
estimate, examined the effects of output market information on total output market information interventions on prices received by the
farm investment. Four of these studies focused on interventions producers. Three studies evaluated multi‐component interventions
implemented in Africa, with one in South Asia. None of the studies (ECX, EMIS, and SIF). The other 15 evaluations looked at single‐
analyzed multi‐component interventions. Four effects were derived component interventions, ranging from providing output market

F I G U R E 41 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of output market information
interventions on the use/adoption of output market information.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 65 of 191

F I G U R E 42 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of output market information
interventions on farm investment (total).

F I G U R E 43 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of output market information
interventions on prices.

information through mobile, internet, or other forms of investment in and statistically significant ˆ = 0.06;
(SMD 95% CI: 0.001 to
ICT infrastructure. 0.12; p = 0.047) (Figure 43). There is substantial heterogeneity among
Most estimates (k = 15) evaluated the interventions at house- the estimates (Q (17) = 50.95, p < 0.001, τ̂ 2 = 0.009, I2 = 66.64), and
hold level. However, Goyal (2010) analyzed the effect at market effect sizes vary from −1.14 to 0.60 standard deviations.
level, and Douglas and Thomas (2009) used the community as their Campenhout (2017) found that community knowledge workers
unit of analysis. Finally, Rajhowa (2021) analyzed the market statistically significantly increased access to agricultural information
auction center‐level price outcome. The majority of included and extension by 0.16 standard deviations (95% CI: 0.07 to
estimates consider multiple crops (k = 10). Schipper and colleagues 0.24; p < 0.001). Similarly, Goyal (2010) reported small but statisti-
(2021) focused on French beans, bananas, butternuts, onions, cally significant positive effects of the E‐CHOUPAL program on the
tomatoes, capsicums, and milk, while Kamande and Nafula (2016) wholesale market price of soybeans (ĝ = 0.60 ; 95% CI: 0.25 to
focused on beans, maize, pigeon peas, potatoes, bananas, sweet 0.96; p < 0.001).
potatoes, sugarcane, cassavas, sorghums, and other fruits and None of the moderator variables significantly contributed to the
vegetables. The other estimates (k = 8) tracked single crops. For variation in estimates. We did not identify any funnel plot asymmetry
example, Rajkhowa (2021) and Lee and colleagues (2020) analyzed meaning that publication bias is not likely present. Of the seventeen
the effect of these interventions on the price of tea and coffee, included effect sizes, ten remain after excluding high risk of bias
respectively. estimates. The effect of output market information interventions on
Output market information interventions increased prices received prices outcomes slightly increases but becomes statistically insignificant
by the producers. The weighted average effect was very small, positive, ˆ = 0.08, 95% CI: −0.0002 to 0.16; p = 0.05).
(SMD
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
66 of 191 | MARION ET AL.

Agricultural production outcomes. The agricultural production outcomes Volume. A total of 10 estimates from 10 studies were included in the
are yields, volume, produce type, quality, and crop loss. The included analysis on the impact of output market information interventions on
studies provided evidence on all outcomes, except quality. volume outcomes. These studies focused on interventions in Africa
(n = 3), Europe and Central Asia (n = 1), Latin America and the
Yields. Six studies, each providing one estimate, measured the Caribbean (n = 2), and South Asia (n = 4).
effect of output market information interventions on yields. These Four studies evaluated multi‐component interventions (FPPM,
interventions focused on providing output market information ARMA66, KMDP, and SEWA). Four studies analyzed provision of
through mobile, internet, or other forms of communication in output market information through information dissemination through
Africa (n = 5) and South Asia (n = 1). For example, Goyal (2010) mobile phones (i.e., SMS, phone calls), and internet (e.g., apps,
focused on the E‐CHOUPAL internet kiosk program to provide websites) technology (ESOKO, E‐CHOUPAL, RML, mobile phone
information about local (mandi) prices on total yields. One study provision), while others evaluated interventions that sought to
(Garcia et al., 2018) examined a multi‐component intervention increase access to technologies that improved access to market
(SHoMaP) that sought alternative physical marketplaces and information—such as mobile phone services, internet, or broadband,
increased domestic infrastructure, in addition to output market and the essential provision of electricity (e.g., rural energy develop-
information interventions. Apart from the district‐level study on ment program, CWM in Nepal, ABC mobile phone module in Niger,
India (Goyal, 2010), all other studies analyzed the effects at the FITEL in Peru).
household level. Two of the volume outcomes are measured at the district
Output market information interventions did not affect yields. (Goyal, 2010) and crop level (Desai & Joshi, 2014), while the rest
Overall, there is a very small, positive, albeit not statistically (k = 8) are measured at the household level. All studies had some
ˆ = 0.07; 95% CI: −0.09 to
significant weighted average (SMD variation of total quantity produced as their outcome variable.
0.23; p = 0.37) (Figure 44). Effect sizes vary from −0.34 to 0.39. Output market information interventions did not have a statisti-
There is substantial heterogeneity among estimates (Q (5) = 23.89, cally significant effect on volume outcomes. The overall weighted
p < 0.001, τ̂ 2 = 0.03, I2 = 79.07%). average effect is very small and not statistically significant
In Nigeria, Oyinbo and colleagues (2022) reported a moderate, ˆ = 0.05; 95% CI: −0.02 to 0.11; p = 0.17) (Figure 45). The
(SMD
positive, and statistically significant effect of information on expected observed outcomes were from −0.10 to 0.35. There is moderate
returns for maize yield (ĝ = 0.38, 95% CI: 0.23 to 0.54, p = 0.00 ). heterogeneity among the estimates (Q (9) = 14.14 , p = 0.11,
Similarly, in Kenya, Garcia and colleagues (2018) reported a very small, τ̂ 2 = 0.004, I2 = 36.36%). This evaluation found a statistically signifi-
positively significant effect of SHoMaP on yields of crops such as cant moderate effect of community‐owned micro hydropower on
amaranth, bananas, cabbages, carrots, chilies, garden peas, Irish potatoes, total agricultural output in Nepal ( ĝ = 0.34 ; 95% CI: 0.11 to
managu, mangoes, melons, onions, passion fruit, pineapples, sagas, sweet 0.57; p = 0.003) (Koirala, 2019).
potatoes, and tomatoes (ĝ = 0.10, 95% CI: 0.005 to 0.20, p = 0.03). None of the moderator variables significantly contributed to the
None of the moderator variables significantly contributed variation in estimates. We did not identify any funnel plot asymmetry,
to the variation in estimates. Of the six included effect sizes, meaning that publication bias is not likely to be present. Five of the
two remain after excluding high risk of bias estimates. included estimates are assessed as high risk of bias; after removing
The effect of output market information interventions on yields these, five estimates remain in the analysis. The overall weighted
outcomes decreases and is still not statistically signifi- effect slightly decreases but is still statistically insignificant
^ D = ‐0. 15, 95% CI : ‐0. 69 to 0. 39; p = 0. 59).
cant (SM ˆ = −0.004, 95% CI: −0.08 to 0.07; p = 0.91).
(SMD

F I G U R E 44 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of output market information
interventions on yields.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 67 of 191

F I G U R E 45 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of output market information
interventions on volume.

F I G U R E 46 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of output market information
interventions on produce type (diversity).

Produce type (diversity). Next, we turn to produce type (diversity) All three studies evaluated interventions implemented in Sub‐
outcomes. We included six independent estimates from six studies Saharan Africa. Belay and Ayalew (2020) investigated the effects of
evaluating interventions that began between 2008 and 2017. access to price information on crop choice among Ethiopian farmers
Half of the studies (n = 3) focus on Africa, followed by South Asia and found a small, positive, and significant effect ( ĝ = 0.14 ; 95% CI:
(n = 2) and Europe and Central Asia (n = 1). India is the most 0.03 to 0.24; p = 0.01). Aker and Ksoll (2016) assessed the
represented country, with two studies included in the analysis. effectiveness of a business literacy program delivered via mobile
Two of the six analyzed interventions have a multi‐component phone in DRC, and reported a small, positive, and statistically
design (ARMA66 and P4P). Namely, the study by Belay and significant effect on the number of crops grown (ĝ = 0.15; 95% CI:
Ayalew (2020) evaluated the impact of access to price informa- 0.05 to 0.25; p = 0.01). Anguko (2018) evaluated the impact of P4P
tion via internet, or more traditional information channels, for in Niger, and found moderate, positive, and statistically significant
Ethiopian farmers. effects on the number of crops grown ( ĝ = 0.32; 95% CI: 0.17 to
The outcome variable of interest was measured as number of 0.48; p < .001).
crops grown, change of crop variety compared to the previous year, None of the moderator variables significantly contributed to
and ratio of grown crops among those in the price index. All six the variation in estimates. Of the six included effect sizes, five
estimates are measured at household level. remain after excluding high risk of bias estimates. The effect of
Overall, output market information interventions did not affect output market information interventions on produce type (diversity)
product type (diversity). The overall weighted average effect is small, outcomes remains the same and stays statistically significant
ˆ = 0.14 ; 95% CI: 0.07 to
positive, but not statistically significant (SMD ˆ = 0.14, 95% CI: 0.05 to 0.24; p = 0.003).
(SMD
0.22; p = 0.001) (Figure 46). Effects ranged from 0.04 to 0.32
standard deviations. There is modest heterogeneity among studies Intermediate outcomes. The intermediate outcomes category covers
(Q(5) = 8.37, p = 0.16, τ̂ 2 =0.003, I2 = 37.68%). the outcomes that may affect farmers following the harvest period
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
68 of 191 | MARION ET AL.

through the sale of their agricultural products. They cover two to farmers and traders through mobile phone text messages
outcome types: sales and farm income. containing maize and soybean prices in Malawi.
On the contrary, the other two studies reported positive effects
Sales. The analysis includes 17 studies reporting the effects of output in Armenia and in Ethiopia, respectively. The study by Caeyers and
market information on sales. Most studies evaluated interventions Nairi Quinn (2015) reported a positive impact ( ĝ = 0.31; 95% CI: 0.14
providing output market information via mobile and internet to 0.48; p = 0.003) from the ARMA66 project. Lee and colleagues
technologies (n = 11), via other forms of communication alone (2020) found positive effects of the price index among Ethiopian
(n = 2), or combined with other interventions (n = 3); while the small farmers' coop eratives (ĝ = 0.24 ; 95% CI: 0.05 to
remaining two studies focused on alternative physical marketplaces 0.42; p = 0.01). Ochieng and colleagues (2020), however, found a
(Desai and Joshi 2014) or investment in ICT (Aker & Ksoll, 2016). reduction in farmers' sales ( ĝ = −0.31; 95% CI: −0.51 to
Most of the studies were based on interventions in Africa (n = 11), −0.12; p = 0.001). According to the authors, the maize and soybean
with additional studies from South Asia (n = 4), Europe and Central prices via mobile phone intervention in Malawi may have triggered a
Asia (n = 1), and Latin America and the Caribbean (n = 1). The included “wait‐and‐see” attitude among farmers in hope of better prices in the
interventions were implemented over the period 2008–2019. Two of future.
the 17 studies implemented a multi‐component intervention The moderator analysis indicated that estimates based on models
(ARMA66 and EMIS). adjusting for covariates found slightly smaller effects than estimates
Sixteen studies conducted the analysis at household level, while computed from models not controlling for other variables. We did
one study conducted a market‐level analysis (Goyal, 2010). Outcome not identify any funnel plot asymmetry, meaning that publication bias
variables were expressed as measures of sales volume, proportion of is not likely to be present. When we only focus on estimates with low
production sold, and commercialization index. risk of bias and some concerns, nine estimates remain. The overall
Output market information interventions did not affect sales average weighted effect stays statistically insignificant and does not
outcomes. The overall weighted average effect is very small, positive, ˆ = 0.03, 95% CI: −0.09 to 0.15; p = 0.63).
vary (SMD
ˆ = 0.03; 95% CI: −0.03 to
and not statistically significant (SMD
0.10; p = 0.28) (Figure 47). Effects ranged from −0.31 to 0.31 Farm income. We identified 24 estimates of the effects of output
standard deviations. There is substantial heterogeneity among market information interventions on farm income from 24 studies. The
studies (Q(16) = 45.90, p = 0.001, τ̂ 2 = 0.008, I2 = 65.14 %). included studies focused on interventions commencing between
The study by Campenhout (2017) found small negative effects of 1994 and 2019 and implemented in Africa (n = 12), South Asia (n = 7),
the CKW Initiative in Uganda (ĝ = −0.10 ; 95% CI: −0.19 to East Asia and the Pacific (n = 2), Latin America and the Caribbean
−0.01; p = 0.03). Likewise, Ochieng and colleagues (2020) found a (n = 2), and Europe and Central Asia (n = 1). Three of the 24
negative impact on sales ( ĝ = −0.31; 95% CI: −0.51 to interventions used multi‐component designs combining different
−0.12; p < 0.001) from an intervention providing better information interventions (EMIS, IAR4D, and SHoMaP). Most estimates were

F I G U R E 47 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of output market information
interventions on sales.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 69 of 191

derived from interventions based on the provision of output market 0.56; p < 0.001). In a similar vein, Oyinbo and colleagues (2022)
information via mobile or internet (k = 16) and investments in ICT conducted an experiment providing site‐specific nutrient manage-
infrastructure (k = 8), followed by interventions combining market ment interventions and information on expected returns to Nigerian
information with either access to alternative marketplaces (k = 3), farmers and found positive effects ( ĝ = 0.31; 95% CI: 0.16 to
arranged or curated offers (k = 2), or with domestic transport 0.46; p < 0.001).
infrastructure (k = 1) (Garcia et al., 2018). The remaining estimate Similarly, Ding et al. (2018) evaluated the impact of a massive
refers to the provision of information via other traditional forms of electrification program in China, which improved farmers' income
communication. (ĝ = 0.14 ; 95% CI: 0.06 to 0.23; p < 0.001). Garcia and colleagues
Most studies conducted the analysis at household level (k = 22), (2018) reported a small positive and significant effect of SHoMaP, a
with the other two studies focused on the market and village level, program supporting small farmers with training and marketing
respectively. Outcome variables were captured using measures of information in Kenya (ĝ = 0.12; 95% CI: 0.02 to 0.22; p = 0.02).
crop revenues, income from farming, horticultural income per capita, Nkonya and colleagues (2013) evaluated the IAR4D project in DRC,
and farm household crop income. and found small positive effects ( ĝ = 0.18; 95% CI: 0.03 to
Output market information interventions increased farm income. 0.33; p = 0.02). Chindarkar and colleagues (2020) exploited a panel
The overall weighted average effect is very small, positive, and dataset for assessing the effectiveness of investments in fixed and
statistically significant ˆ = 0.09; 95%
(SMD CI: 0.04 to variable farm inputs, and net farm income per acre. The authors
0.13; p < 0.001) (Figure 48). Effects ranged from −0.18 to 0.38 reported negative effects (ĝ = −0.18; 95% CI: −0.35 to −0.003;
standard deviations. There is substantial heterogeneity among the p = 0.04).
studies (Q(23) = 67.23, p < 0.001, τ̂ 2 = 0.007, I2 = 65.79%). None of the moderator variables significantly contributed to the
The study by Chong and colleagues (2009) found a positive variation in estimates. We did not identify any funnel plot asymmetry,
effect of the privatization of the national telecommunication systems meaning that publication bias is not likely to be present. Of the 24
in Peru (ĝ = 0.14 ; 95% CI: 0.02 to 0.27; p = 0.02). Beuermann (2015) included effect sizes, eleven remain after excluding high risk of bias
reported positive effects of the FITEL, a program which made use of estimates. The effect of output market information interventions on
the revenues collected via a taxation on telecommunication compa- farm income outcomes decreases and becomes statistically
nies in order to expand agricultural infrastructure in Peru ( ĝ = 0.11; ˆ = 0.05, 95% CI: −0.001 to 0.10; p = 0.05).
insignificant (SMD
95% CI: 0.001 to 0.21; p = 0.045). Lee and colleagues (2020) found a
large and statistically significant effect of access to price information Non‐farm outcomes. The non‐farm outcomes category covers the
on the farm income of beneficiaries (ĝ = 0.38; 95% CI: 0.20 to outcomes that may affect farmers outside farm activities and

F I G U R E 48 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of output market information
interventions on farm income.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
70 of 191 | MARION ET AL.

agricultural production. They cover two outcome types: non‐farm public phones in every town in Peru (PPRA) had a small, positive, and
income and labor market outcomes. statistically significant effect on households' annual non‐farm per
capita income.
Non‐farm income. We identified six estimates from six studies The moderator analysis indicated that recent interventions are
assessing the effects of output market information interventions on yielding slightly smaller effects. When we only focus on estimates with
non‐farm income. This category of outcomes includes income and low risk of bias and some concerns, two estimates remain (Blumen-
revenues from non‐agricultural activities (outside of farm activities stock et al., 2020; Akpandkar & Kitchens, 2017). The overall weighted
and agricultural production). Four studies evaluated the effects of average effect decreases in magnitude and becomes statistically
investments in ICT infrastructure. The remaining two studies (Smale ˆ = 0.01, 95% CI: −0.07 to 0.09; p = 0.76).
insignificant (SMD
et al., 2012; Desai & Joshi, 2014) analyzed the effects of interventions
providing output market information. Five of the six studies focused Labor market outcomes. We included seven estimates based on seven
the analysis at household level, whereas one study focused on the studies assessing the effects of output market information interven-
firm level (Akpandjar and Kitchens 2017). One estimate reflects the tions on non‐farm labor market outcomes. This outcome category
effect of multiple interventions (KMDP, KDDP, and KHDP). includes measures of the amount of time worked in their household
Two studies focused on South Asia, two studies focused on business (Douglas & Thomas, 2009; Blumenstock et al., 2020), the
Africa, and two were based on interventions in East Asia and the number (Ding et al., 2018) or the ratio (Burlig & Preonas, 2021) of
Pacific and Latin American and the Caribbean, respectively. Outcome workers in wage employment (with an indicator for wage or salaried
variables are expressed in terms of measures of non‐farm income, workers) (Meeks et al., 2022), and binary indicators for running a
monthly wages, and other (non‐agricultural) income. Five of the six small business (Dillon et al., 2020; Akpandjar & Kitchens, 2017).
studies focused on household level, while one study focused on firm Included studies focused on interventions implemented across
level (Akpandjar & Kitchens, 2017). countries in Africa (n = 3), followed by East Asia and the Pacific
Output market information interventions increased non‐farm (n = 2), and South Asia (n = 2). None of the evaluated interventions
income. The overall weighted average effect is very small, positive, used a multi‐component design.
ˆ = 0.07; 95% CI: 0.01 to
and statistically significant (SMD Six out of seven studies evaluated the effects of investments in
0.13; p = 0.03) (Figure 49). Effects ranged from 0.01 to 0.19 ICT infrastructure, while one intervention provided output market
standard deviations. We found moderate heterogeneity among the information (Dillon et al., 2020). The latter study is an experiment
estimates (Q(5) = 6.61, p = 0.25, τ̂ 2 = 0.001, I2 = 24.37%). that provided phone directories and information to enterprises in
The first study examines an experiment in the Philippines, which Tanzania. The studies had a heterogeneous focus in terms of unit of
provided equipment to install cellular network towers, free SIM cards, analysis: two studies focused on the individual level, two on
and a tariff reduction to facilitate mobile phone connection (Blumen- the household level, two on the village level, and one on the
stock et al., 2020); it found null effects of the intervention ( ĝ = 0.09; district level.
95% CI: −0.07 to 0.09; p = 0.85). The second study by Akpandjar Output market information interventions increased non‐farm labor
and Kitchens (2017) evaluated the effects of an electrification project market outcomes. The overall weighted average effect is small,
that improved connections in rural areas of Ghana; it reported small, ˆ = 0.13; 95% CI: 0.06 to
positive, and statistically significant (SMD
non‐significant effects ( ĝ = 0.12; 95% CI: −0.31 to 0.55; p = 0.59). 0.20; p < 0.001) (Figure 50). Effects ranged from 0.01 to 0.97
The overall effect is largely driven by statistically significant estimates standard deviations. We found considerable heterogeneity among
from Chong and colleagues (2009), who found that the installation of the estimates (Q(6) = 87.22, p < 0.001, τ̂ 2 = 0.006, I2 = 93.12%).

F I G U R E 49 Forest plot of observed outcomes and estimates of the random‐effects model for the effect of output market information
interventions on non‐farm income.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 71 of 191

F I G U R E 50 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of output market information
interventions on non‐farm labor market outcomes.

Dillon and colleagues (2020) conducted an experiment distribut- physical marketplace interventions providing output market information,
ing phone directories among Tanzanian farmers and reported positive arranged or curated offers from buyers, and domestic transport, each with
effects (ĝ = 0.28; 95% CI: 0.14 to 0.41; p < 0.001). In a similar vein, one estimate. The remaining four estimates were derived from
Blumenstock and colleagues (2020) carried out an experiment in the evaluations of multi‐component initiatives combining the provision of
Philippines, providing free SIM cards and multi‐band mobile phones output market information interventions with other types of interventions.
to farmers, and found a very small increase in time spent working For example, Lombardini (2017) assessed the effects of the
away ( ĝ = 0.05; 95% CI: 0.003 to 0; p = 0.04 ). Akpandjar and ARMA66 program, an intervention in Armenia combining arranged
Kitchens (2017) evaluated NEP which extends universal services in offers from buyers with the supply of storage deposit systems and
Ghana, and found large effects on the probability that the head of the provision of market information via mobile phone. The project
household runs a non‐agricultural business (ĝ = 0.82; 95% CI: 0.13 to aims to create new economic opportunities for small‐scale farmers
1.47; p < 0.001); whilst Meeks and colleagues (2022) reported a via the establishment of cooperatives and the delivery of training
significant effect from a program promoting electrification in Nepal activities and productivity‐enhancement technologies.
( ĝ = 0.01; 95% CI: 0.008 to 0.01; p < 0.001)—however, the very Ten estimates were derived from studies focused on the
small magnitude of the effect is of little practical significance. household level, except one estimate which referred to a village‐
None of the moderator variables significantly contributed to level analysis (Burlig & Preonas, 2021). The latter study evaluated
the variation in estimates. Of the seven included effect sizes, five India's national rural electrification program (RGGVY) aimed at
remain after excluding high risk of bias estimates. The effect of improving both domestic and commercial electricity access in over
output market information interventions on non‐farm labor market 400,000 rural villages across 27 states.
outcomes slightly increases and remains statistically signifi- Output market information interventions increased income and
ˆ = 0.19, 95% CI: 0.07 to 0.32; p = 0.002).
cant (SMD wealth outcomes. The overall weighted average effect is very small,
ˆ = 0.07; 95% CI: 0.01 to
positive, and statistically significant (SMD
Welfare outcomes. The welfare outcomes category covers measures 0.13; p = 0.03) (Figure 51). Effects ranged from −0.03 to 0.25
of improvements in the well‐being and quality‐of‐life of farmers. standard deviations. There is substantial heterogeneity among
They encompass a range of social, economic, and environmental studies (Q(10) = 31.85, p < 0.001, τ̂ 2 = 0.006, I2 = 68.60 %).
factors, but in this systematic review we focus on two: total income Chong and colleagues (2009) evaluated the impact of privatiza-
and food security and nutrition. tion of the national telecommunication system in Peru and found
small positive effects (ĝ = 0.22; 95% CI: 0.10 to 0.35; p < 0.001).
Total income and wealth. The analysis included 11 estimates of the Anguko (2018) reported small positive effects of the P4P, a program
effects of output market information on total income and wealth providing information and harvesting and processing tools to small
(including measures of socio‐economic status, such as household farmers in DRC (ĝ = 0.25; 95% CI: 0.10 to 0.40; p < 0.001).
expenditure and assets). The estimates stem from 11 studies focusing Blumenstock and colleagues (2020) investigated the effectiveness
on interventions implemented in five regions: Africa (n = 5), South of the CCN tower installation in improving mobile coverage in rural
Asia (n = 2), East Asia and the Pacific (n = 2), Europe and Central Asia areas in the Philippines; the authors reported small effects on total
(n = 1), Latin America and the Caribbean (n = 1). household income ( ĝ = 0.13; 95% CI: 0.04 to 0.21; p = 0.002).
Five of the 11 estimates were based on interventions providing None of the moderator variables significantly contributed to the
investments in ICT infrastructure. Other interventions included alternative variation in estimates. We did not identify any funnel plot asymmetry,
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
72 of 191 | MARION ET AL.

F I G U R E 51 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of output market information
interventions on income and wealth.

F I G U R E 52 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of output market information
interventions on food security.

meaning that publication bias is not likely to be present. Of the 11 colleagues (2018) used the food consumption expenditure, includ-
included effect sizes, five remain after excluding high risk of bias ing both the money spent on purchased food and the value of
estimates. The effect of output market information interventions on consumption of own production as proxies for food security
income and wealth outcomes increases and remains statistically measures. For one study, the effect size had to be reversed for
ˆ = 0.12, 95% CI: 0.02 to 0.21; p = 0.02).
significant (SMD the analysis (Schipper et al., 2021), as it used a binary variable
indicating whether a household faced insufficient current food
Food security. Seven estimates were included in the analysis of the consumption as the outcome measure.
effect of output market information interventions on food and nutrition Providing output market information did not affect food security
outcomes. Of these seven, three were multi‐component interventions outcomes. The overall weighted average effect is positive, very small,
(P4P, ARMA66, and SHoMaP), while the rest analyzed effects of an ˆ = 0.03, 95% CI: −0.04 to 0.12,
and not statistically significant (SMD
ICT‐based agricultural market information service in Ghana (Esoko- p = 0.36) (Figure 52). Effect sizes ranged from −0.11 to 0.21. There is
SEND) (Al‐Hassan et al., 2013), membership durations in Kenya moderate heterogeneity across the included estimates (Q (6) = 14.18,
(Schipper et al., 2021), CCN tower installation (Blumenstock et al., p = 0.03, τ̂ 2 = 0.007, I2 = 57.67%).
2020) in the Philippines, and the VP program in Rwanda (Douglas & Blumenstock and colleagues (2020) found a small,
Thomas, 2009). Effects of all seven interventions were analyzed at statistically significant positive effect of the locally owned CCN
the household level. program in the Philippines on the availability of adequate food
The seven studies used different food consumption measures (ĝ = 0.14, 95% CI: 0.06 to 0.22, p = 0.001).
to define the outcome measure. For example, Lombardini (2017) None of the moderator variables significantly contributed to
used the value of daily per capita food consumption in the previous the variation in estimates. Of the seven included effect sizes, five
seven days. Similarly, Anguko (2018) used the value of food remain after excluding high risk of bias estimates. The
consumption per adult equivalent per day, and Garcia and effect of output market information interventions on food security
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 73 of 191

outcomes decreases and remains statistically insignificant intervention across time. As above, we only focus on effect sizes
ˆ = 0.01, 95% CI: −0.11 to 0.12; p = 0.89).
(SMD where the period of exposure to the intervention is different, with
everything else held constant. We do not report here the effect sizes
Subgroup analysis. We now examine how the effect sizes of output across the period of exposure, which are consistently statistically
market information interventions may vary by subgroup. We here insignificant (Table 16).
compare estimates within studies from the same intervention across We noticed that effect sizes are not always maintained or
subgroups, and do not report here the effect sizes across subgroups, increased as the period of exposure increases. For example, KMDP
which are consistently statistically insignificant. We display this price and trade information appears to have a moderate positive
analysis by the wealth of beneficiaries. influence on the value of livestock assets after 70 months (in 2008) of
project exposure (Smale et al., 2012) but not after 92 months (in
Wealth. Table 15 shows the available effect sizes by wealth 2010), as the effect becomes very small and statistically insignificant.
subgroups. The findings are from Dillon and colleagues (2020) across Similarly, through GMIS (Hildebrandt et al., 2020), weekly SMS
four outcomes. Authors evaluated the effects of the Kichabi alerts delivered to subscribers with the most current price informa-
intervention in Tanzania, which distributed telephone directories for tion led to an increase in yam price for farmers after 10 months of
both businesses and households in selected villages, with an aim to project exposure (short‐term effects, within the first year, according
improve communication and trade. After nine months of project to authors) but not after 22 months (long term effects, in the second
exposure, beneficiary households were more likely to have used a year). Authors primarily argued that interventions may have produced
phone for crop production activities, more likely to have a non‐farm temporary price increases, highlighting that longer‐term estimates
enterprise, and less likely to have faced a maize crop failure in the may also be biased due to spillover.
most recent agricultural season. Oyinbo and colleagues (2022) found that effects from NUI
The results are different across households below and above (Treatment 2)—a tablet‐ or smartphone‐mediated decision‐support‐
median wealth. Households below median wealth levels were more tool intervention including price information—emerged and were
likely to have used a phone for crop production activities and to have sustained after two years of household project exposure. After one
a non‐farm enterprise, whereas households above median wealth year (in 2017) and two years (in 2018) of access to this tool,
levels were more likely to have a non‐farm enterprise and less likely households were more likely to apply inorganic fertilizer at sowing
to have faced a maize crop failure in the most recent agricultural and experienced a rise in maize yields. The magnitude of the effect
season. Only households below the mean wealth did not benefit from increased from one to two years.
an increase in crop sales price.
Barriers and facilitators analysis
Analysis of effects over time. We aimed to reproduce this same type This section discusses the reported facilitators and barriers to the
of analysis but for different periods of exposure to the interventions. interventions' effects in local contexts. This presentation is subject to
Here we compare estimates within studies from the same an important caveat: the included studies are quantitative impact

T A B L E 15 Effect sizes across wealth subgroups (from Dillon et al., 2020, Kichabi).
Outcome subcategory Outcome description Subgroup information ĝ p Value

Use/adoption of improved inputs, Binary variable on whether the household Whole sample 0.19 0.01
technologies & practices used a phone for crops in the most recent
Households below median wealth 0.17 0.02
agricultural season
Households above median wealth 0.12 0.07

Non‐farm labor market outcomes Binary variable on whether the household has Whole sample 0.28 <0.001
a non‐farm enterprise in the most recent
Households below median wealth 0.17 0.02
agricultural season
Households above median wealth 0.19 0.01

Prices Crop sales price (Tanzanian shillings, log) Whole sample 0.21 0.08

Households below median wealth −0.06 0.63

Households above median wealth 0.29 0.02

Crop loss Binary variable on whether the household did Whole sample 0.17 0.02
not face a maize crop failure in the most
Households below median wealth 0.04 0.63
recent agricultural season
Households above median wealth 0.18 0.02

Note: Subgroup estimates in bold have a statistically significant effect size (ĝ ) at a 5% significance level.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
74 of 191 | MARION ET AL.

T A B L E 16 Effect sizes across periods of exposure (from Smale et al., 2012, KMDP; Hildebrandt et al., 2020, GMIS; Oyinbo et al., 2022, NUI
Treatment 2).
Period of exposure
Study Outcome subcategory Outcome description in months ĝ p Value

Smale et al. (2012) Farm investment (livestock) Value of livestock assets (Kenyan 45 0.05 0.61
shillings)
70 0.29 0.01

92 0.02 0.82

Hildebrandt et al. (2020) Prices Price level (cassava, processed) 10 −0.01 0.94

22 0.77 <0.001

Price level (yam) 10 0.21 0.02

22 0.00 0.99

Oyinbo et al. (2022) Use/adoption of improved Binary variable on whether 12 0.20 0.01
inputs, technologies, & inorganic fertilizer was applied at
24 0.29 <0.001
practices sowing

Farm income Net revenue in naira per hectare 12 0.14 0.06

24 0.31 <0.001

Farm investment Production costs in naira per 12 0.13 0.08

hectare
24 0.26 0.001

Yields Maize yield (ton/ha) 12 0.23 0.002

24 0.39 <0.001

Note: Estimates in bold have a statistically significant effect size (ĝ ) at a 5% significance level.

evaluations. Such studies tend to focus narrowly on identifying the number changes or migration. Finally, in a similar intervention in
magnitude of the intervention's effects, without discussing imple- Ghana, Hildebrandt and colleagues (2014) realized that many farmers
mentation details or rigorously explaining why an intervention was broke or lost their phones or changed their SIM cards, creating
successful (or not). A thorough analysis of barriers and facilitators interruptions in service provision.
would require a dedicated search for different types of evidence. Of
the 81 papers in this body of evidence, our analysis of barriers and The impact may be facilitated by the level of ICT awareness and
facilitators identified 44 papers with relevant information. education of the targeted population. Output market information
interventions that rely on mobile or other technologies do not only
ICT programs are reliant on the quality and reach of existing or new need to provide the tool itself but must also ensure that the
infrastructure. ICT infrastructure is a prerequisite to the implementa- beneficiary population can use it optimally. Several authors com-
tion and success of the interventions, and the authors highlight the mented on the importance of ICT awareness and ICT education as
barrier it may raise if not fully considered. In India, Reddy and drivers for increasing program participation and maximizing program
Mehjabeen (2019) noted that the lack of infrastructure led to failure effects.
of eNAM program implementation in some states. Similarly, in DRC, In areas where the population already has knowledge and
Rwanda, and Uganda, Nkonya and colleagues (2013) reported that interest in using ICT solutions, participation may increase. For
the surveyed population mentioned issues such as slow broadband example, Ogutu and colleagues (2014) analyzed the effect of
and a shutdown that resulted in a partial implementation of the DrumNet in Kenya, which provided marketing services to
information platform. However, Nagpal and Sovera (2021) observed smallholder farmers by linking them to buyers and providing
a larger effect of electrification on agricultural outputs in Indian them with information through an ICT platform. The authors
states with reliable electricity supply. observed higher proportions of participation in regions where
The success of ICT infrastructure programs is also highly reliant other ICT‐based management‐information system providers were
on the quality of the infrastructure management and logistics. In implanted.
Colombia, Camacho and Conover (2011) highlighted that their SMS‐ Beyond participation, authors emphasized the potential of
based information provision had to be discontinued for 10 days due ICT awareness and knowledge on agricultural outcomes. In
to the incapacity of their information source to send the daily prices. Colombia, Camacho and Conover (2011) exploited the high level
In a similar type of intervention in India, Fafchamps and Minten of connectivity and high use of SMS technology to analyze the
(2012) noted that some phone numbers could not be used due to effect of an SMS‐based intervention; they based its design on the
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 75 of 191

results of their baseline survey, which showed that over 78% of Although the provision of price information may strengthen farmers'
farmers had knowledge on the use of text messages that could bargaining power, misalignment of information, strategies, or relation-
then be used as a medium for the intervention. ships can act as an additional barrier. Providing information about
In Niger, Aker and Ksoll (2016) analyzed the impact of the ABC outputs and prices to farmers can help to strengthen their bargaining
Program, which provided basic literacy and numeracy training and power in negotiations with traders. However, this is not the only
introduced a mobile phone module. They observed a larger effect barrier to be addressed to reduce price dispersion. Authors highlight
among those who owned a mobile phone at baseline compared to that information must be aligned, and accessed by both parties, and
those who did not. In India, Desai and Joshi (2014) observed that might not be enough to break vertical relationships. In India, Mitra
educated women were more capable of leveraging market informa- and colleagues (2018) observed that farmers did not make use of the
tion services to increase their income. Finally, in Mozambique, Zant price information they received because they did not have the option
(2017) observed a larger effect of the mobile phone rollout on price of taking their potatoes to another buyer and, therefore, continued
dispersion in source markets with a higher number of people with with the local buyer. They also commented that improving farmers'
mobile phones. knowledge and access to price information is unlikely to have positive
On the other hand, authors pointed to potential barriers raised outcomes in farmgate prices in the context of vertical supply chains.
by lower levels of literacy and ICT education. In Ghana, Hildebrandt Other authors highlighted the importance of relationships in the
and colleagues (2014) observed that 10 farmers did not have phones selling behavior of farmers and traders. In Ghana, Hildebrandt and
to carry out the SMS‐based price information intervention. They also colleagues (2020) commented that the prevalence of long‐standing
noted that sets of farmers struggled to interpret the information due relationships could hinder the effectiveness of information provision,
to low literacy levels. especially in cases of credit or agricultural provisions that reduce the
Several authors also noted the relatively higher impact of ICT bargaining power of farmers in negotiations. Similarly, in India, Reddy
interventions among the younger generation and their higher levels and Mehjabeen (2019) observed the unbalanced levels of information
of participation. Okello and colleagues (2020) analyzed the effect of between farmers and traders: farmers' ICT and pricing knowledge
ICT‐mediated market information in Kenya and explained that a limited their bargaining power against traders with experience in the
higher level of participation among younger farmers was due to their management of online banking systems.
higher ICT literacy levels. Similarly, Fafchamps and Minten (2012) In Nigeria, Sanusi (2018) also commented on unbalanced
observed that young farmers were more likely to improve the quality information levels and observed that when markets in two states
of their crops as a consequence of using the RML service compared have access to the same price information, price dispersion reduces
to other age groups. further than in situations in which only one market is accessing the
information. Finally, the authors highlighted the influence of trust
For the introduction of ICT tools to be successful, the targeted between market stakeholders, which, when lacking, can limit their
population must be interested in using them. Both the electrification ability to collaborate. In Kenya, Garcia and colleagues (2018)
of rural areas and the introduction of communication technologies observed that a lack of trust among group members reduced the
require a certain level of buy‐in from the local population who are to impact of an output market information intervention.
benefit from them. In the absence of this buy‐in, the local population Risk aversion and selling strategies can also play an important role
may not use them optimally or may raise resistance to their in farmers' pricing and selling behavior. Ochieng and colleagues (2020)
introduction. In India, Fafchamps and Minten (2012) analyzed the analyzed the effect of a mobile phone text message information
effect of the RML intervention which provided price, weather, and provision in Malawi. They observed that despite farmers' having
crop advisory information via SMS. They highlighted that, initially, the information on better pricing, their risk aversion resulted in low sale
farmers who were offered the intervention did not express interest in volume for maize. A similar observation was raised by Nakasone (2013)
it, and dismissed its usefulness despite its potential use for them. on a price information intervention in Peru. Risk‐averse farmers acted as
Similarly, some actors might see the introduction of a new ICT tool as a barrier to market participation. Finally, in India, Mitra and colleagues
a threat to their bargaining power. (2018) observed that farmers were likely to report tracked prices based
In India, Reddy and Mehjabeen (2019) focused on the eNAM on information posted daily in a public location rather than the
program, providing both platform‐based information and an e‐ information they received from private phone calls.
auction platform. The main hurdle for program implementation was
aligning the interests of all stakeholders while facing resistance from Despite the provision of information, agroecological conditions may
traders and commission agents, as they felt they were losing benefits affect prices and agricultural outputs. Outputs and price information
compared to physical transactions. are a resource for farmers but might not prevent unexpected or
To address this barrier and boost participation, some programs seasonal agroecological conditions. In Ghana, Hildebrandt and
made use of incentives or obligatory contributions. For example, in colleagues (2020) received anecdotal evidence that most farmers in
Colombia, Camacho and Conover (2011) offered a pesticide spraying the area have no other option but to sell their crops during the
machine as a prize in a raffle to encourage participation in an SMS‐ harvest season, regardless of the higher prices they could obtain
based information provision initiative. during the off‐season. Similarly, in India, Nagpal and Sovera (2021)
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
76 of 191 | MARION ET AL.

noted that the effect is driven by the season, especially cropping This is higher than the general trend in impact evaluation literature,
season, which is highly dependent on monsoon rains. where fewer than 20% of impact evaluations report any cost
Other authors highlighted the impact of agroecological events that information (Brown & Tanner, 2019). Studies have primarily provided
negatively affected prices and agricultural outputs and reduced the information on mobile phone rates and the cost of rolling out mobile
impact of information provision. In DRC, Ragasa and colleagues (2016) networks (especially in remote areas).
commented that the decrease in processing could be due to an increased Four authors provided information on the cost induced by phone
import of stable crops during a food crisis. In Kenya, Schipper and subscriptions. Aker (2015) indicated that in Niger, mobile phone rates
colleagues (2021) suggested that the COVID pandemic might explain the ranged from USD0.35–0.43 per minute and cost USD0.07 per text
reduction in production of French beans and bananas. message. In Mozambique, Zant (2019) reported that the price of a
simple mobile phone was around USD5–6 and a local phone call cost
Intervention costs approximately USD0.10 per minute, making it a relatively affordable
Thirteen of the 49 studies included in this body of evidence provided resource. Nagpal (2021) reported that households were required to
some information about the cost of interventions (27%) (Table 17). pay USD7 for their connection and would pay in 10 installments.

T A B L E 17 Summary of available intervention costs.

Study Acronym Cost information description

Aker (2015) ROCPC Mobile phone rates: USD0.35–0.43 per minute for calls and USD0.07 per text message

Cost per mobile phone: USD1

Zant (2019) MPR Cost of mobile phone: USD5–6

Cost of local phone call: USD0.10 per minute.

Meeks et al. (2022) ETMHP Communities identified through the carpet study were eligible for a subsidy, coordinated by the
Alternative Energy Promotion Centre and funded by bilateral and multilateral international donor
organizations. The subsidy increased over time. In 2006, it was NPR10,000 per household, not to exceed
NPR85,000 per kW (NPR1 = USD0.0104) for the system. In 2009, the subsidy rose to NPR15,000 per
household (NPR125,000/kW maximum). In 2013, the subsidy changed for some sites, with remote and
very remote locations receiving NPR15,000 per household (NRP100,000/kW) and NPR25,000 per
household (NPR130,000/kW).

Kizito et al. (2012) SIF Cost of running the management information system: USD130,000 for the whole country and
USD30,000 for Nampula province.

Average population gain in income: USD723,000 (approximately six times more than the entire
operating costs of the management information system in Mozambique in 2002).

Chong et al. (2009) PPRA The cost of installation in the Andean region was quite high as many poor, rural towns are at high
altitudes and difficult to access.

Douglas and Thomas (2009) VP Collaborations with microfinance institutions provide financing to operators, so that they can purchase
the phones and pay off the loan with the profit stream from airtime usage. During the study period, the
term of the loan offered by microfinance institutions was six months, and the total cost of the loan plus
financing was USD261.

Goyal (2010) E‐CHOUPA The intervention saved INR12.9 million in the first year of operation.

Parker et al. (2016) RML Phone subscription fee: USD2 per month

Hildebrandt (2020) GMIS Comments on the low cost of information delivery via mobile phone, which makes it highly cost‐
effective (returns around 200% of total costs)

Cost of training: USD37 per farmer

Chakravorty (2016) BLEP Electricity connection cost: USD325 per household

The cost of extending electricity grid to a village is recovered after a single year of realized welfare gain.

Burlig (2015) RGGVY Total cost: USD17.2 billion

Cost per village: USD36,000

Nagpal (2021) RGGVY Cost per connection for households: USD7

Garcia et al. (2018) SHoMaP Total cost: USD32.15 million

 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 77 of 191

Finally, Park and colleagues (2016) reported that in India the due to the improved profitability attributable to improved access to
subscription fee cost USD2 per month. market information.
The reported costs of implementation varied based on the type The analysis detected very small and statistically significant
and location of the intervention. Kizito and colleagues (2012) positive effects of output market information interventions on
described a total cost of USD130,000 for the implementation of ˆ = 0.07 ; 95% CI: 0.01 to
farmers' total income and wealth (SMD
the management information system intervention in Mozambique for 0.13; p = 0.03; k = 11). The modest effects challenge the view that
an average population gain in income of USD723,121 per year. information interventions can unconditionally unlock the potential of
Chong and colleagues (2009) did not provide full cost information of smallholder farmers (Sligo et al., 2005).
the intervention in Peru but commented that the cost was quite high We found positive, although very small, effects on
due to rural towns' high altitudes and difficult access. Meeks and non‐farm outcomes. Notably, we found that interventions positively
colleagues (2022) reported that the subsidy in Nepal for investment affected non‐farm income (SMD ˆ = 0.07 ; 95% CI: 0.01 to
in ICT infrastructure increased from NPR10,000 per household in ˆ = 0.13;
0.13; p = 0.03; k = 6) and non‐farm labor outcomes (SMD
2006 to NPR15,000 and NPR25,000 per household in 2013 for 95% CI: 0.06 to 0.20; p < 0.001; k = 7).
remote and very remote areas, respectively. We found no systematic improvements of agricultural production
outcomes, off‐farm income outcomes, and other immediate or
Discussion intermediate outcomes such as credit, transaction costs, and sales.
However, it is important to stress that generally, the number of
Summary of findings. We identified 81 quantitative studies carrying out included studies for these outcomes is low and that we could not
quasi‐experimental or experimental evaluations of 47 output market perform the meta‐analysis for seven outcome categories with fewer
information interventions implemented across 19 countries. Of these than three estimates.
interventions, nine have a multi‐component design combining provision We conducted exploratory moderator analyses for outcomes
of output market information interventions or investments in ICT with more than five estimates. We found little effect of intervention
infrastructure with other intervention components. The most evaluated year, type of evaluation method used, length of follow‐up period,
types of interventions are those providing output market information via length of exposure to the intervention, or whether it was a single or
the internet or mobile phone (25 studies) followed by investments in ICT multi‐component intervention.
infrastructure (16 studies). Most of the included studies used a QED (41 The effects are heterogeneous across outcomes and across
studies) and covered an average follow‐up period of 5.1 years. interventions, implying that contextual factors may be relevant. We
We then performed a meta‐analysis of the effects using an observed an overall small improvement of total income and wealth
inverse variance weighted random‐effects model. This quantitative outcomes, which seems to be driven by improvements in both farm
synthesis indicates that output market information interventions had and off‐farm outcomes (i.e., farm income, off‐farm income, and labor).
positive effects along all the stages of the theory of change, although The effect of these interventions with regard to the adoption of
the detected effects are small and often not statistically significant. ˆ = 0.14; 95%
improved varieties and quality seeds were small (SMD
Thus, the overall evidence suggests that smallholder farmers gained CI: 0.04 to 0.25; p = 0.01; k = 7). We found a similar effect on
limited advantages. However, on average there were no significant ˆ = 0.21;
adoption of improved machinery and irrigation inputs (SMD
negative effects of these programs. 95% CI: 0.05 to 0.38; p = 0.01; k = 9). Farmers cultivated a greater
Based on the theory of change, these interventions are expected ˆ = 0.14 ; 95% CI:
variety of crops because of the interventions (SMD
to permit farmers to exploit knowledge of output market information 0.07 to 0.22; p = 0.001; k = 6).
to obtain higher prices for their products and face lower transaction Reducing the information asymmetry by itself does not systemati-
costs due lower search costs. These effects in turn could influence cally improve farmers' agricultural, economic, and social outcomes. The
the quantity produced, market participation, income welfare but also small effects might indicate that other factors contribute to the
adoption and investments of agricultural improved practices. performances of smallholder farmers. One argument could be that they
We detected a very small and significant impact on three are too small to receive the actual benefits of information access; low
outcome categories: farm investment, prices received by farmers, and capital and assets might be important limitations hindering farmers'
total income and wealth. The remaining outcomes displayed small, adoption of improved inputs and access to markets. Low assets cannot
non‐significant effects. easily be used as collateral for bank loans and credit, and ultimately
The analysis indicates that output market information interven- prevent farmers from gaining potential market benefits (e.g., sales,
tions have small, positive effects on the prices per unit farmers income). Furthermore, small farmers are particularly vulnerable to
ˆ = 0.06; 95% CI: 0.001 to
receive for their produce (SMD weather and climate fluctuations (Harvey et al., 2014), which can be
0.12; p = 0.047; k = 18). Thus, although the detected effects are only partially addressed with better access to information.
positive, their magnitude is modest. Other relevant barriers could include smallholder farmers' limited
We found a very small, positive impact on farm income access to networks or low technical and management skills.
ˆ = 0.09; 95% CI: 0.04 to 0.13; p < 0.001; k = 24 ) Specifically,
(SMD Moreover, misalignment of information, strategies, and relationships
the theory of change suggests that this income boost might occur might undermine the effectiveness of output market information.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
78 of 191 | MARION ET AL.

Effectiveness of ICT programs depends on the quality of existing is less than ten meaning that the body of evidence is small.
infrastructure, and on the beneficiary interest in using ICT tools. Similarly, we did not observe any evidence on the effects of these
However, these hypotheses would need to be further corroborated interventions on off‐farm outcomes and nutrition. Future research
with a qualitative review of program and project documents. is needed on these outcomes, especially immediate outcomes
Few studies provide information concerning the costs of the which are relevant in understanding these interventions.
intervention, which is slightly over the general trend of impact • Further research should focus on the effects of output market
evaluations. In most cases, authors provided information on the costs information interventions on non‐farm income and labor out-
of individual intervention components, such as mobile phone cost comes. The limited evidence on these outcomes indicates that
and rates, whereas in a few cases, the total cost of the intervention these interventions may have a strong effect on farmers'
was reported. involvement in non‐farm sectors; however, the number of
studies covering these issues is very low and sparsely
Implications for policymakers and implementers. distributed across regions.
• Output market information interventions are more effective in • Further research is needed on the reasons for the limited increase
improving prices, farm‐income, total income, non‐farm income and in reported adoption and investments in practices exploiting the
non‐farm employment compared to other outcomes. Nevertheless, output market information provided.
the small average effects should be kept in mind when
implementing such policies.
• The design of new projects should be mindful of the quality and 5.3.3 | Chapter 3: Initiatives creating new
reach of existing ICT infrastructure, and the ICT awareness and marketplaces and alternative marketing opportunities
education of the targeted population. Existing marketing strategies
and relationships of farmers should also be considered when How can creating new marketplaces and alternative marketing
establishing new output market interventions. opportunities affect agricultural, socio‐economic, food security, and
• Although the evidence indicates that farmers experienced small nutrition outcomes?
improvements in farm and total income levels from access to better New marketplaces and alternative marketing opportunities interventions
crop prices, no evidence supports changes in adoption and establish market platforms which offer novel avenues for the
investments in improved agricultural practices based on the exchange of goods and services, or which lessen the expenses
provided output market information. The design of new projects related to finding such opportunities. Figure 53 provides an overview
should, therefore, further identify and address local barriers (e.g., of potential effects of these interventions on agricultural, socio‐
lack of infrastructure, security, credit, risks, price volatility) that economic, food security, and nutrition outcomes. This chapter
prevent farmers from taking full advantage of the output market focuses on three types of new marketplaces and alternative marketing
information provided in these interventions. opportunities interventions:

Implications for researchers. • Online commodity exchanges and mobile‐based marketplaces inter-
• For all immediate outcomes such as transaction costs, adoption of ventions aim to improve connections between market participants
improved practices and investments, the number of included studies by creating new or improved marketplaces using information

F I G U R E 53 Theory of change of market interventions. The outcome types included in this systematic review are immediate outcomes
(green), agricultural production outcomes (purple), intermediate outcomes (teal), and welfare outcomes (grey).
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 79 of 191

technologies, for example, via online commodity exchanges, apps, With lower transactions costs, higher prices and more informa-
or mobile services. tion, farmers are expected to experience higher sales and farm
income. In turn, these short‐term effects from these interventions
• Alternative physical marketplaces interventions aim to improve may also encourage farmers in the long‐term to adopt improved
connections between market participants by creating physical production methods, diversify their produce type, leading to a further
marketplaces, such as physical hubs linking farmers directly with increase in overall farm and total income which may affect food
private companies. security and nutritional outcomes.
• Arranged or curated offers from buyers interventions establish
connections between market participants by curating commit- Description of included studies
ments where traders offer to buy farmers' produce.14 We included 94 studies examining the effects of 61 initiatives
creating new marketplaces and alternative marketing opportunities
By directly connecting farmers with buyers, new marketplaces interventions. Four of the included studies are ongoing (Falcao &
and alternative marketing opportunities interventions can first decrease McIntosh, 2021; Robinson et al., 2021; Cefala & Swanson, 2021;
farmers' transactions costs by reducing their search costs. Arranged or Robinson, 2023). The quantitative synthesis of findings relies on
curated offers from buyers interventions establish arrangements effects from 67 papers (65 main studies and two linked papers), as
between farmers and buyers meaning that farmers sells to a single 27 papers do not provide any additional results. These 27 papers
buyer instead of searching for multiple buyers on the physical market. were ongoing studies, previous versions, or unpublished versions of
Online commodity exchanges and mobile‐based marketplaces interven- already included studies. Two linked papers reported additional
tions eliminate farmers' traveling costs associated with identifying a outcome data about the effects of an intervention (i.e., data on
buyer (Li et al., 2021).15 outcomes or prioritized outcome indicators that are not reported by
With greater linkages to buyers, these interventions also offer the main study).
farmers greater opportunities to access new markets and a wider
base of buyers. Farmers mostly sell at farm gate or travel to markets Intervention, inputs, and activities
selling to other households or small traders meaning that they often We identified 61 initiatives creating new marketplaces and alternative
lack connections to established market actors (Chamberlin & marketing opportunities interventions implemented in 28 countries
Jayne, 2013; de Brauw & Bulte, 2021; Farmer Income Lab, 2018). (Figure 54). Interventions in Sub‐Saharan Africa represent almost half
These interventions can link farmers with private sector buyers and of the included studies (28), followed by East Asia and the Pacific
actors which do not normally engage in traditional physical markets (12), South Asia (10), and Latin American and the Caribbean (7). The
(Iacovone & McKenzie, 2019). remaining interventions were implemented in Europe and Central
Third, with these interventions, farmers may receive better prices Asia (4). Individual countries with the largest number of included
for their products as a result of increased communication with buyers interventions are China (9), Kenya (6), India (6), Tanzania (6),
and decreased information asymmetries in price (Zeng et al., 2017). Colombia (4), and Ethiopia (4). Two studies analyzed the effects of
This is particularly true for arranged or curated offers from buyers a multi‐country intervention implemented in DRC, Rwanda, and
interventions where farmers and buyers reach an arrangement on Uganda (Nkonya et al., 2013), and one in Kenya and Uganda (Rao
price which requires information exchange and negotiations. How- et al., 2016).
ever, lower transaction costs and greater opportunities from these Supporting Information: Appendix 3 for Chapter 3 provides
interventions, as discussed above, may increase competition and information on the 61 interventions including country, intervention
reduce the prices received by farmers. type, and year of commencement. The most frequent type of
These interventions may also have positive long‐term effects on intervention evaluated is arranged or curated offers from buyers
farmers. First, these interventions may increase farmers' bargaining (n = 32), followed by the groups of interventions belonging to
power in the market as they have greater opportunities and are alternative physical marketplaces (n = 17) and online commodity
dealing with various stakeholders in the agricultural value chain exchanges and mobile‐based marketplaces (n = 13) (Figure 55). Lom-
(Desai & Joshi, 2013). Second, these arrangements may also facilitate bardini and Bowman (2015) included activities spanning two
the exchange of information accessed by farmers on production intervention categories: arranged or curated offers from buyers and
practices. These interventions may therefore reduce learning costs, alternative physical marketplaces.
thus enhancing production processes, including use of improved Table 18 provides an overview of the interventions included in
inputs, greater quality of products and cultivating different crops. this review in terms of country, intervention type, and commence-
ment year.
Thirty‐seven of the included interventions use a multi‐
14
Contrary to contract farming interventions, these connections occur after harvest and
component approach (Table 19). In the analysis below, we conduct
therefore are not forward‐looking arrangements whereas contract farming is an arrangement a moderator analysis to determine whether the estimates from multi‐
in advance of or at the time of planting.
15 component interventions are on average significantly statistically
Supporting Information: Appendices 1 and 2 for Chapter 3 provide additional information
about included studies and additional references, respectively. different than estimates from single‐component interventions.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
80 of 191 | MARION ET AL.

F I G U R E 54 Distribution of interventions by region.

F I G U R E 55 Distribution of interventions by subcategory.

Comparison Outcomes
In the included 65 studies, the comparison groups used for these The analyzed studies cover all 18 outcomes included in our
estimations nearly exclusively focused on comparing outcomes for framework specified in the protocol. The most frequently analyzed
beneficiaries of initiatives creating new marketplaces and alternative outcomes are farm income (n = 29), followed by total income and
marketing opportunities interventions to those of non‐beneficiaries wealth (n = 28), prices (n = 25), sales (n = 23), use or adoption of
located in a different area (97%). One study exploited a phased‐in (or improved inputs, technologies and practices (n = 18), and produce
pipeline) approach to the interventions to identify a comparison type (n = 16) (Figure 56).
group (Chaudhary & Suri, 2022). One study (Belay & Ayalew, 2020)
identified a control group based on the variation in timing and spatial Study design
distance to the ECX. Supporting Information: Appendix 4 for Chapter The majority of the 65 main studies used QEDs (n = 57) to evaluate
3 provides information on these studies, such as the evaluated the effects of the interventions, primarily relying on DID and FE
intervention, its subcategory, length of follow‐up periods in years, approaches (n = 21) and statistical matching methods (n = 28). One
study design, and type of comparison. study was based on an interrupted time series study design
 T A B L E 18 Table of included initiatives creating new marketplaces and alternative marketing opportunities interventions.
MARION

First year of
Intervention name Acronym Country (ISO3 code) Type intervention
ET AL.

Agriculture for Children's Empowerment ACE Liberia (LBR) Arranged or curated offers from buyers 2008

Anchor Farm Project AFP Malawi (MWI) Improved on‐farm storage infrastructure and 2014
technologies; Arranged or curated offers from buyers

Agruppa AGP Colombia (COL) Online commodity exchanges and mobile‐based 2016
marketplaces

AHL Auction Floors AHLAF Malawi (MWI) Alternative physical marketplaces 2004

AidMatch Program AMP Bangladesh (BGD) Alternative physical marketplaces 2014

AidMatch Program AMP Tajikistan (TJK) Alternative physical marketplaces 2014

Access to a market for quality maize AMQM Uganda (UGA) Arranged or curated offers from buyers 2017 and 2018

New Economic Opportunities for Small‐Scale Farmers in ARMA66 Armenia (ARM) Arranged or curated offers from buyers; Storage deposit 2010
Tavush and Vayots Dzor Regions systems; Provision of output market information
(mobile and internet)

Agricultural Value Chain Project, Oromia AVCPO Ethiopia (ETH) Storage deposit systems; Other contract types; 2011
Alternative physical marketplaces

Barangay Bagsakan System BBS Philippines (PHL) Arranged or curated offers from buyers 2006

Contract arrangement with tea farmers CATF Vietnam (VNM) Arranged or curated offers from buyers; Input or credit‐ 1999
supply contracts

Contract farming in Hanoi CFH Vietnam (VNM) Arranged or curated offers from buyers –

Co‐operative membership CM China (CHN) Production‐management contracts; Arranged or curated 1980

offers from buyers

Dairy market hub DMH Tanzania (TZA) Alternative physical marketplaces 2014

DrumNet, a horticultural export and cashless micro‐credit DMHECMCP Kenya (KEN) Alternative physical marketplaces 2004
program

E‐CHOUPAL (Hub) E‐CHOUPAL (hub) India (IND) Provision of output market information (mobile and 2000
internet); Alternative physical marketplaces

East African Dairy Development EADD Uganda (UGA) and Kenya (KEN) Alternative physical marketplaces 2008

e‐Commerce adoption ECA China (CHN) Online commodity exchanges and mobile‐based 2014
marketplaces

e‐Commerce in rural villages ECRV China (CHN) Online commodity exchanges and mobile‐based 2017
|

marketplaces

(Continues)
81 of 191

18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
 TABLE 18 (Continued)
82 of 191

First year of
Intervention name Acronym Country (ISO3 code) Type intervention
|

Ethiopian Commodity Exchange ECX Ethiopia (ETH) Online commodity exchanges and mobile‐based 2008
marketplaces; Provision of output market information
(mobile and internet)

Egiye Jai and Nigera GoriEnhancing food production and food EJNG Bangladesh (BGD) Arranged or curated offers from buyers 2013
security through homestead farming of vegetables, livestock,
and fisheries

e‐Mandi Initiative EMI India (IND) Online commodity exchanges and mobile‐based 2006
marketplaces

Empowering Small Producers, Especially Women, in the Dairy ESPEWDS Pakistan (PAK) Arranged or curated offers from buyers; Alternative 2011
Sector Project physical marketplaces

Improving living conditions in Fadama areas FAD‐2 Nigeria (NGA) Domestic transport infrastructure; Alternative physical 2005
marketplaces

Fostering Agricultural Market Activity FARMA Bosnia and Herzegovina (BIH) Arranged or curated offers from buyers 2009

FARMEX FARMEX India (IND) Provision of output market information (mobile and 2009 and 2017
internet); Online commodity exchanges and mobile‐
based marketplaces

Farmers' markets FM Colombia (COL) Alternative physical marketplaces 2009

Farmers' organizations FO Ethiopia (ETH) Arranged or curated offers from buyers 2009

Farmers' professional cooperatives FPC China (CHN) Input or credit‐supply contracts; Arranged or curated 2008
offers from buyers

Guangxi Integrated Agricultural Development Project GIADP China (CHN) Domestic transport infrastructure; Alternative physical 2012
marketplaces

Guiding Opinions on Promoting Accurate Poverty Alleviation GOPAPAEC China (CHN) Online commodity exchanges and mobile‐based 2016
for E‐Commerce Project marketplaces

Guatemala Highlands Value Chain Development Alliance GUAB38 Guatemala (GTM) Arranged or curated offers from buyers; Storage deposit 2010
systems

High‐Value Agriculture Project HVAP Nepal (NPL) Arranged or curated offers from buyers 2011

The Innovation for Agribusiness (InovAgro) project IAP Mozambique (MOZ) Arranged or curated offers from buyers 2014

Integrated Agricultural Research for Development IAR4D Democratic Republic of the Congo Alternative physical marketplaces–‐ Provision of output 2008
(COD), Rwanda (RWA) and market information (mobile and internet)
Uganda (UGA)
MARION
ET AL.

18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
 TABLE 18 (Continued)
MARION

First year of
Intervention name Acronym Country (ISO3 code) Type intervention
ET AL.

ICT infrastructure in rural regions ICTIRR China (CHN) Online commodity exchanges and mobile‐based 2009
marketplaces

Integrated Project to Increase Agricultural Productivity IPIAP Tanzania (TZA) Arranged or curated offers from buyers 2014

Kenya Horticulture Development Project KHDP Kenya (KEN) Alternative physical marketplaces; Provision of output 2004
market information (other forms of communication)

Kims Poultry Care Center Contract Farming KPCCCF Kenya (KEN) Input or credit‐supply contracts; Arranged or curated 1996
offers from buyers

Livestock Commercialization for Pastoralist Communities LCPC Niger (NER) Arranged or curated offers from buyers 2008

Mercados Campesinos MC Colombia (COL) Alternative physical marketplaces 2004

Mobile phone‐based marketplace for agricultural commodities MPBMAC Uganda (UGA) Online commodity exchanges and mobile‐based 2015
marketplaces

Optimizing Income of Local Chicken and Sunflower OILCSS Tanzania (TZA) Alternative physical marketplaces; Improved on‐farm 2012
Smallholder Producers Project storage infrastructure and technologies

Purchase for Progress P4P Democratic Republic of the Arranged or curated offers from buyers; Provision of 2012
Congo (COD) output market information (other forms of
communication)

Purchase for Progress P4P Ghana (GHA) Arranged or curated offers from buyers 2011

Purchase for Progress P4P Tanzania (TZA) Arranged or curated offers from buyers; Production‐ 2009
management contracts; Storage deposit systems

Rural Productive Partnerships Project PAAP Colombia (COL) Arranged or curated offers from buyers 2002

Pre‐existing vertical and horizontal linkages PEVHL Tanzania (TZA) Other contract types; Arranged or curated offers from ‐
buyers

Plataformas Program PP Ecuador (ECU) Arranged or curated offers from buyers 2003

Placement of supermarket supply chains PSSC Nicaragua (NIC) Arranged or curated offers from buyers 2000

Rural e‐commerce in the apple sector RECAS China (CHN) Online commodity exchanges and mobile‐based 2015
marketplaces

Rural E‐Commerce Demonstration Counties Project REDC China (CHN) Online commodity exchanges and mobile‐based 2014
marketplaces

Supply chain regulations in the market for fresh fruits and SCRMFFV Turkey (TUR) Arranged or curated offers from buyers 2012
vegetables
|

Smallholder Dairy Commercialization Program SDCP Kenya (KEN) Arranged or curated offers from buyers 2006

(Continues)
83 of 191

18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
84 of 191 | MARION ET AL.

Note: ISO3 country code provided in parentheses (). If the introduction of an intervention is staggered, the year of intervention refers to the earliest year observations were undertaken for the intervention.
(Goyal, 2010). Aysoy and colleagues (2015) conducted a regression
discontinuity estimation. The authors used data before and after
Turkish policy changes on supply chain regulations for fresh fruits
First year of
intervention
and vegetables. One study conducted an IV estimation (Biggeri et al.,
2022). Three studies estimated the effects using two study designs:
2016

2003

2014

2008

2013

2016
Wang (2020) using a DID and FE and IV approach, and Zant (2020)
and Rajkhowa (2021) both combining a statistical matching and DID

Alternative physical marketplaces; Provision of output

and FE approach. The remaining eight studies used experimental
designs.
Online commodity exchanges and mobile‐based

Online commodity exchanges and mobile‐based

Figure 57 shows the distribution of the length of the studies'

market information (mobile and internet)

follow‐up periods (study periods), which is the difference between
Arranged or curated offers from buyers

Arranged or curated offers from buyers

Arranged or curated offers from buyers

Arranged or curated offers from buyers

the intervention's start date and the most recent outcome data
included in the study's analysis. The 61 evaluated interventions
began between 1980 and 2019 with an average length of roughly 5.2
years. Twenty studies covered a period within two to four years after
the intervention took place. Another large share of studies evaluated
the effects less than two years after the intervention started (n = 15),
marketplaces

marketplaces

whilst the remaining 27 studies evaluated the effect four or more

years after the interventions. The longest covered period is 34 years
(Liang et al., 2020) whilst the shortest is one month (Lentz & Upton,
Type

2016; Bayiyana et al., 2018; Levi et al., 2020).

Synthesis of findings
The data extraction of studies under the initiatives creating new
marketplaces and alternative marketing opportunities intervention
category identified a total of 1450 estimates, of which 244 have
Country (ISO3 code)

been selected for the analysis of effects. We report additional

Tanzania (TZA)

Ethiopia (ETH)

information on the individual meta‐analyses (sensitivity analysis,

Kenya (KEN)

Kenya (KEN)

India (IND)

India (IND)

India (IND)

outlier analysis, publication bias analysis, moderator analysis, meta‐

analyses without high risk of bias estimates) in Supporting Informa-
tion: Appendix 5 for Chapter 3.
SSGACPMTSM

Risk of bias. The risk of bias for each estimate across the 65 included
studies was assessed following the Risk of Bias Assessment Tool
Acronym

(Supporting Information: Appendix 10 for background and methods

WFGP

eNAM
WIISP
VML2
SHEP

UMP

sections). Figure 58 presents a summary of the results of our risk of

bias assessments. We found that 63% of included estimates were
rated as having a high risk of bias, 33% were considered to have some
Self‐sustaining groups to access clean planting material,

National Agriculture Market (eNAM) e‐trading platform

Smallholder Horticulture Empowerment and Promotion

concerns related to their risk of bias, and the other 4% were assessed
Wheat Initiative to Increase Smallholder Productivity

as having a low risk of bias.

Despite the overall trends, Figure 58 shows a discrepancy between
the risk of bias of QED studies compared to RCT studies, since 66% of
the former are rated as high risk of bias compared to 41% for the latter.
Women Farmers with Global Potential

This risk of bias of included estimates was based on six domains

(selection bias; confounding; performance bias; spillovers, crossovers,
technical support, and markets

and contamination; outcome measurement bias; and reporting bias).

(Continued)

Vertical Market Linkages 2

Unified Market Platform

Looking at individual risk of bias domains, the most common source of

bias is related to selection (34% of estimates were rated as high risk of
Intervention name

selection bias) and confounding bias (26%).

For example, Berg and Levely (2019, p. 23) reported a high risk of
TABLE 18

selection bias for their analysis of IPIAP in Tanzania. Due to the

limited number of treatment wards, they could not select all the FOs
and observed major differences at baseline between treatment and
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 85 of 191

T A B L E 19 Descriptions of other interventions combined with interventions creating new marketplaces and alternative marketing
opportunities.
Intervention Description of other intervention components combined with transport infrastructure
(country code)a Type interventions

ACE (LBR) Arranged or curated offers from buyers The Agriculture for Children's Empowerment (ACE) project operated in rural Liberia
from 2008 to 2013. It aimed to improve household economic security by increasing
farm production and linkages with buyers. ACE used a value‐chain facilitation approach
to develop agricultural networks and stimulate entrepreneurial skills among farmers. It
provided training on modern farming methods, food, and seed preservation, and
encouraged the formation of farmer clusters.ACE also introduced input suppliers and
buyers to farmers. The project's goal was to increase farm income, leading to
investments in child well‐being including education, health, and nutrition. The project
revised its approach in 2009 to focus on direct technical assistance to smallholder
farmers.

AGP (COL) Online commodity exchanges and Agruppa (AGP), a social enterprise founded in 2014, utilizes mobile phone technology
mobile‐based marketplaces to aggregate demand for produce from small vendors, creating daily collective orders.
These orders are purchased in bulk directly from farms at a discount and then delivered
to small retail shops, thereby shortening the supply chain.Its aim is to reduce travel time
and costs for shop owners, offer fresh produce at a 5% discount compared to
Corabastos (a large market), and potentially lower food costs for poor consumers. The
project started in Bogotá in 2014 and expanded in 2016. AGP initially intended to
develop an app but instead offered other ordering methods due to technology issues.

ARMA66 (ARM) Arranged or curated offers from buyers; The ARMA66 project, New Economic Opportunities for Small‐Scale Farmers in Tavush
Storage deposit systems; Provision of and the Vayots Dzor Region in Armenia, established farmers' cooperatives as a platform
output market information (mobile and for various project activities. These activities included providing training in high‐value
internet) and climate‐resilient crops, productivity‐enhancing technologies, and financial
management to enhance incomes and livelihood security.Inputs such as improved
seeds, fertilizers, and pesticides were supplied to boost productivity. Cold storage
facilities and sun dryers were set up to improve the quality and timing of produce sales.
Harvest festivals and links with national processors aimed to increase farmers' incomes.
SMS messages on market prices, agro‐tourism development, and access to interest‐free
microfinance loans aimed to support various objectives related to income, resilience,
and productivity.

AVCPO (ETH) Storage deposit systems; Other The Agricultural Value Chain Project, Oromia (AVCPO) was created as a comprehensive
contract types; Alternative physical initiative from 2011 to 2013, targeting 6520 smallholder farmers in 15 cooperatives. It
marketplaces encompassed various interventions including training, seed distribution, storage,
processing, and marketing through cooperatives. The project aimed to introduce and
expand durum wheat cultivation, implement technological enhancements by providing
high‐quality seeds and improved agricultural extension services, and strengthen the
role of cooperatives in the agricultural value chain.

CATF (VNM) Arranged or curated offers from buyers; Contract arrangement with tea farmers (CATF) involves two types of contract farming
Input or credit‐supply contracts arrangements: contracts with state‐owned enterprises, including land leases, and
contracts with private firms or cooperatives. These contracts typically offer initial
support for new tea production, such as free seeds, input subsidies, and extension
services. The intervention likely started around 1999 in Vietnam, following the Tea
Development Plan (outlined in the Prime Minister's Decision 43/1999/QD‐TTg). This
initiative aimed to boost tea production, productivity, exports, and employment
opportunities, leading to a significant expansion of tea cultivation in the country
after 1999.

CFH (VNM) Arranged or curated offers from buyers This study of contract farming in Hanoi (CFH) explores the impact on the annual income
of peri‐urban vegetable farmers in Hanoi, Vietnam in 2008, based on their practices of
either: (1) selling produce directly to consumers in shops or markets (T1); or (2) selling
through shops or supermarkets under contract (T2). Although there is no external
intervention, the research leverages existing practices to analyze income outcomes.

DMH (TZA) Alternative physical marketplaces The dairy market hub (DMH) approach introduced by the International Livestock
Research Institute in Tanzania clusters dairy services around a milk buyer through
contractual agreements, providing farmers with access to milk markets, inputs,
information, training, and credit services.Two mechanisms enhance farmers' incomes
through hubs. First, information and training improve their knowledge and skills, leading

(Continues)
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
86 of 191 | MARION ET AL.

TABLE 19 (Continued)

Intervention Description of other intervention components combined with transport infrastructure

(country code)a Type interventions

to increased adoption, productivity, and marketed output, and benefiting all farmers
regardless of hub membership. Second, credit enhances farmers' assets, enabling
technology adoption to address production and marketing constraints, and resulting in
higher dairy productivity and market output (available only to hub‐participating
farmers).

DMHECMCP (KEN) Alternative physical marketplaces DrumNet aimed to support smallholder farmers with their horticultural exports and
access to microcredit without cash transactions. It established partnerships with
commercial banks, input providers, transportation services, and exporters to overcome
barriers faced by these farmers.Farmers were organized into groups of five and were
jointly responsible for loans. During harvest, DrumNet negotiated prices with
exporters, and collection points were established. Transaction agents at these points
facilitated grading and packaging. In the credit treatment group, DrumNet collaborated
with local agricultural retailers for in‐kind loans, with retailers trained in DrumNet's
recordkeeping. Farmers received orientation, opened savings accounts, and contributed
to a transaction insurance fund serving as collateral for their credit line.

E‐CHOUPAL Provision of output market information In October 2000, ITC Limited initiated the e‐Choupal program to purchase soybeans
(hub) (IND) (mobile and internet); Alternative directly from farmers, thereby eliminating intermediaries and ensuring better quality
physical marketplaces and cost control. They established 1700 internet kiosks and 45 hubs over 4.25 years in
soy‐growing districts of Madhya Pradesh. Warehouses (hubs) facilitated quality testing
and direct sales to ITC. This intervention empowered farmers with price information
and an outside option, while ITC improved its soybean quality and saved costs.

EADD (UGA) Alternative physical marketplaces The East African Dairy Development (EADD) project, initiated in 2008 and active until
2018 in Kenya, Uganda, and Rwanda, introduced dairy hubs aimed at improving the
dairy value chain. These hubs focused on collective action and involved aggregating and
selling milk in bulk, giving farmers a bargaining advantage with buyers and potentially
raising milk prices.Additionally, the project facilitated bulk input sourcing and
negotiated rates with service providers, reducing production costs and increasing dairy
and household incomes. The hubs pursued various milk marketing strategies, either
selling exclusively to processors (Strategy 1) or to diverse outlets (Strategy 2), including
large processors. Initial evaluations showed improved household welfare, income, and
dietary diversity.

ECX (ETH) Online commodity exchanges and The ECX was launched in April 2008 to address food distribution issues contributing to
mobile‐based marketplace; Provision of drought‐related famines. Initially trading only coffee and sesame, it later included
output market information (mobile and maize, wheat, and haricot beans. The ECX operates as a transparent membership‐based
internet) commodity trading platform in Addis Ababa. It conducts market surveillance to prevent
manipulation and assumes central counter party risk.Buyers must deposit funds, and
sellers store produce at ECX warehouses for next‐day settlement. The exchange offers
guarantees, loans, and hedging options, disseminating real‐time market information
across the country via price tickers and various media channels.

EJNG (BGD) Arranged or curated offers from buyers Between January 2013 and December 2016, two projects— Egiye Jai (EJ) in Barisal
district and Nigera Gori(NG) in Dinajpur district—aimed to enhance food production
and food security through homestead farming of vegetables, livestock, and fisheries.
These projects utilized cluster‐level training to encourage the participation of women
farmers in agricultural programs, addressing their mobility limitations. They organized
households into clusters within the same village to promote collaborative learning and
overcome physical barriers to training, fostering sustainable and diverse food
production at the household level.

ESPEWDS (PAK) Arranged or curated offers from buyers; The Empowering Small Producers, Especially Women, in the Dairy Sector Project
Alternative physical marketplaces (ESPEWDS) aimed to improve livelihoods, increase income and employment, and
enhance women's economic leadership in the dairy sector. It was implemented by
Oxfam GB in partnership with the Doaba Foundation in Muzaffargarh district,
Pakistan.Starting in 2011, the project supported four cooperatives across four tehsils
but faced interruptions due to restructuring until it resumed in 2013 and focused on
one of the original cooperatives. Activities involved forming community groups,
establishing milk collection centers, and creating an enterprise with the aim of buying
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 87 of 191

TABLE 19 (Continued)

Intervention Description of other intervention components combined with transport infrastructure

(country code)a Type interventions

milk from local farmers at competitive prices and generating profits while empowering
women.

FAD‐2 (NGA) Domestic transport infrastructure; The project aimed to alleviate rural poverty and enhance food security by improving
Alternative physical marketplaces living conditions in Fadama areas. Its three main components were capacity building
and advisory services, rural infrastructure investment, and project management and
coordination. These components collectively aimed to boost agricultural production,
productivity, and value addition for smallholder farmers and rural entrepreneurs,
fostering sustainable development in the region.

FARMA (BIH) Arranged or curated offers from buyers Fostering Agricultural Market Activity (FARMA) was designed by USAID/BiH and Sida
to support Bosnia and Herzegovina in achieving several critical objectives: increasing
agricultural competitiveness, meeting EU accession standards, reducing poverty
through sustainable production, and boosting sales, exports, and employment. Building
on lessons from the previous Linking Agricultural Markets to Producers project
(2003–2008), FARMA aimed to enhance market linkages, improve access to credit, and
contribute to agricultural policy reform. It ran from August 2009 to May 2015, focusing
on sustainable market linkages, financial access, capacity building, and policy
enhancements to promote the competitiveness of agricultural products in the country.

FO (ETH) Arranged or curated offers from buyers Farmers' organizations (FOs), like cooperatives, can significantly reduce transaction
costs for farmers in agricultural markets. They achieve this by integrating vertically
down the marketing chain, decreasing information asymmetry, and enabling better
prices for members. FOs also reduce initial fixed costs for market access, including
storage and transport costs.Donors or government entities often finance the necessary
investments to strengthen FOs, as seen in initiatives like the World Food Programme's
P4P. This program aims to enhance FOs' capacities, providing skills training, resources,
and creating additional demand for their products through food procurement programs.
This approach positively impacts farmers' welfare and investments in agriculture.

FPC (CHN) Input or credit‐supply contracts; The Farmers' Professional Cooperatives (FPC) intervention, implemented in Jiangsu
Arranged or curated offers from buyers province, China, since 2008, requires cooperative members to enter written contracts
specifying rice cultivation and procurement procedures. Members follow cooperative
instructions, record input usage, and receive discounted seeds, fertilizers, and pesticide
supplies. The cooperative provides technical guidance and advance payments to some
farmers. By complying with regulations, participants can earn a 10% premium over
market prices, though the cooperative doesn't set a minimum price.

GIADP (CHN) Domestic transport infrastructure; The Guangxi Integrated Agricultural Development Project (GIADP) aimed to improve
Alternative physical marketplaces rural infrastructure and market access for smallholder farmers in the Guangxi Zhuang
Autonomous Region of China. Approved by the International Fund for Agricultural
Development in December 2011 and implemented from January 2012 to March 2017,
the project focused on rural development and poverty reduction.It consisted of three
components: community infrastructure (rural roads and irrigation), agricultural
production and marketing support, and rural environmental improvement. The impact
assessment's primary goal was to determine the project's effectiveness in increasing
beneficiaries' agricultural revenue through innovative approaches, considering potential
impact variations based on different project interventions.

GOPAPAEC (CHN) Online commodity exchanges and The Guiding Opinions on Promoting Accurate Poverty Alleviation for E‐commerce
mobile‐based marketplaces Project (GOPAPAEC) was introduced in China in November 2016. It outlines principles,
objectives, and tasks for rural e‐commerce poverty alleviation. Since then, intensive
policies have supported rural e‐commerce projects aimed at alleviating poverty,
particularly in poverty‐stricken areas. These policies represent a top‐level design for
rural e‐commerce poverty alleviation, and various ministries and commissions have
issued supporting documents to assist poor rural areas in creating plans for
implementing e‐commerce poverty alleviation projects tailored to their specific
circumstances.

GUAB38 (GTM) Arranged or curated offers from buyers; The Guatemala Highlands Value Chain Development Alliance (GUAB38) consists of
Storage deposit systems three projects aimed at supporting small‐scale producers in diversifying and increasing
their returns from agricultural production in the Department of Sololá. It supported

(Continues)
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
88 of 191 | MARION ET AL.

TABLE 19 (Continued)

Intervention Description of other intervention components combined with transport infrastructure

(country code)a Type interventions

selected association members in establishing kitchen gardens, boosting maize

production with technical support and fertilizers, and providing maize silos for selling
during favorable market conditions.The projects began in response to the devastation
caused by Tropical Storm Agatha in May 2010, and continued with infrastructure
development and expansion to promote agricultural and economic development. The
projects also prioritize the participation and empowerment of women within producer
associations.

HVAP (NPL) Arranged or curated offers from buyers; The High‐Value Agriculture Project (HVAP) in Nepal aimed to reduce poverty and food
Improved on‐farm storage insecurity by connecting smallholder producers with markets in hilly and mountainous
infrastructure regions. HVAP worked with producer organizations across seven districts, focusing on
seven value chains including apples, ginger, and goats.The project's components
included establishing agreements between producer organizations and agribusinesses,
strengthening producer organization capacity, providing market information, offering
skills development training, and building market infrastructure. These activities aimed
to enhance the adoption of high‐value commodities, improve linkages between
producer organizations and agribusinesses, provide access to technical and price
information, reduce post‐harvest losses, and enhance management capabilities for
smallholder producers.

IAR4D (COD) Alternative physical marketplaces; The IAR4D project employs a conceptual framework to map its impact pathway. This
Provision of output market information begins with IAR4D activities, leading to intermediate impacts through farmer learning
(mobile and internet) from on‐farm experiments and advisory services offered by various partners and
stakeholders. This process also involves the formation of innovation platforms, which
enable horizontal connections among producers and vertical linkages between
producers and agricultural traders and credit service providers. Ultimately, the goal is to
achieve changes in agricultural income for participants.

KHDP (KEN) Alternative physical marketplaces; The Kenya Horticulture Development Project (KHDP) provided marketing, postharvest
Provision of output market information handling, processing, and agronomic assistance to smallholders and allied
(other forms of communication) agribusinesses. In its initial work program, the KHDP targeted six product categories:
passion fruit (fresh and processed), chili products (fresh, processed, and dried), vanilla
and spices, smallholder flowers, tree crops for processing (cashew and mango), and
local market vegetables (onions, carrots, cabbages, tomatoes, and indigenous
vegetables).

KPCCCF (KEN) Input or credit‐supply contracts; Kim's Poultry Care Center Contract Farming (KPCCCF) program involved Kim's Poultry
Arranged or curated offers from buyers Care Center, a major company, and small‐scale farmers in Nakuru County, Kenya, in
2011. It aimed to measure its impact based on the income of participating smallholder
farmers.

LCPC (NER) Arranged or curated offers from buyers This project (Livestock Commercialisation for Pastoralist Communities project, LCPC)
was implemented in partnership with the Association pour la Redynamisation de
l'Élevage au Niger (AREN) in the commune of Bermo, located in the Maradi Region of
Niger. In 2008, AREN and Oxfam encouraged representatives of pastoralist
communities in the commune of Bermo to form a groupement des interêts
économiques (GIE, economic interests group) to take a more active role in organizing
support and representing the interests of pastoralists in the area. While the GIE
provides most of its support to its own members (who numbered around 420 at the
time of the effectiveness review), its activities are also intended to benefit the
pastoralist community more widely. Oxfam and AREN supported the GIE in drawing up
a business plan and provided various forms of capacity building for the institution and
its members. The GIE's primary objective is to improve the position of pastoralists when
they buy or sell livestock. Traditionally pastoralists have conducted transactions in the
market through an intermediary (locally known as a dillali). To maximize the money they
earn from the sale of livestock, the GIE trains its members to avoid using intermediaries
and provides them with information about market prices. In early 2012, the GIE was
due to take over responsibility for managing the livestock market in the town of Bermo
from the municipal authority.

MC (COL) Alternative physical marketplaces The Mercados Campesino (MC) project, which took place in the central region of
Colombia from 2004 to 2015, aimed to address challenges faced by smallholders. It
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 89 of 191

TABLE 19 (Continued)

Intervention Description of other intervention components combined with transport infrastructure

(country code)a Type interventions

established a direct market channel in Bogotá, organizing farmers' markets every two
weeks in 16 city squares. Smallholders from rural areas around Bogotá could sell their
produce directly to consumers. The project also provided workshops and training to
help farmers understand market dynamics, improve product quality, and adopt new
production methods. It aimed to transform smallholders' access to CIT's food supply
chain, foster cooperation among local actors, and enhance the political recognition of
family farmers in food supply and policymaking processes.

OILCSS (TZA) Alternative physical marketplaces; The Optimizing Income of Local Chicken and Sunflower Smallholder Producers Project
Improved on‐farm storage (OILCSS) aimed to enhance the income of smallholder farmers in Tanzania by improving
infrastructure and technologies sunflower and chicken production and sales. The project provided inputs, technical
guidance (including planting, harvesting, storage and processing), and formed marketing
groups to boost agricultural practices, strengthen market links, and establish market
information and credit facilities. It operated in Kilosa district (Morogoro region) and
Kongwa and Chamwino districts (Dodoma region) from October 2012 to August 2015.

P4P (COD) Arranged or curated offers from buyers; The Purchase for Progress (P4P) Program was implemented in the Equateur province of
Provision of output market information DRC, specifically in the municipalities of Bikoro and Ingende, from April 2012 to
(other forms of communication) October 2014. It aimed to link farmers to markets, provide them with tools and
technical assistance, offer market information and credit facilities, and supply transport
equipment such as bicycles and wheelbarrows.The program established associations or
groups for maize, rice, groundnuts, and beans farmers. Produce was sold to
wholesalers, retailers, and the World Food Programme, which distributed it as relief
food both within the DRC and beyond. Coordination and funding were provided by
Oxfam, the Belgian government, and the French government.

P4P (GHA) Arranged or curated offers from buyers The World Food Programme's P4P Program, initiated in 2008, aims to facilitate
smallholder farmers' access to high‐value grain markets through the Programme's
robust procurement system. Traditionally, the World Food Programme sourced grains
from large commercial traders, but P4P was established to enable direct purchases
from trained smallholder farmers.Prior to engagement, the World Food Programme
provides training and connects farmers with various partners to support them
throughout the entire value chain, which encompasses production, post‐harvest
processes, business skills, and financial access. After training, farmers may choose to
prepare their grains according to World Food Programme standards for potential
supply. The program underwent pilot phases in 15 African countries before scaling up.

P4P (TZA) Arranged or curated offers from buyers, The Tanzanian P4P pilot, which took place from September 2009 to December 2013,
Production‐management contracts; aimed to address post‐harvest losses and improve smallholder farmer livelihoods. It
Storage deposit systems involved Tanzanian savings and credit cooperatives, which received both demand‐ and
supply‐side interventions. Supply‐side efforts included rehabilitating warehouses,
providing equipment and training, and enhancing marketing infrastructure.Demand‐
side interventions involved the World Food Programme purchasing maize and beans
from smallholders through competitive tenders and forward contracts. However,
purchase quantities were irregular, and some cooperatives faced partial or complete
contract defaults due to slow payments and changing market prices. Average contract
sizes varied, impacting the demand stimulus experienced by cooperative members.

SDCP (KEN) Arranged or curated offers from buyers; The Smallholder Dairy Commercialization Programme (SDCP) was a Kenyan
Farm‐to‐market transport government initiative from 2005 to 2015, focused on improving the livelihoods of rural
infrastructure households dependent on dairy farming. The SDCP included several components: (1)
training farmers in organizational and managerial skills; (2) enhancing dairy farming
productivity and reducing costs through grants, training, and demonstrations; and (3)
improving market access by enhancing road infrastructure and providing milk handling
and value‐addition training. These efforts aimed to increase the income of
impoverished rural households engaged in dairy production across nine milk‐producing
counties in Kenya.

SHEP (KEN) Arranged or curated offers from buyers The Smallholder Horticulture Empowerment and Promotion (SHEP) Program,
originating from a partnership between Kenya and Japan, has been introduced to
21,000 beneficiaries across 33 Kenyan counties. It involves a comprehensive training
package for farmers and extension officers. Initially, stakeholders are briefed on SHEP

(Continues)
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
90 of 191 | MARION ET AL.

TABLE 19 (Continued)

Intervention Description of other intervention components combined with transport infrastructure

(country code)a Type interventions

principles, followed by baseline surveys to assess farming practices and to set income
goals. Farmers engage in business talks, facilitated by extension officers, connecting
them with market participants. Gender mainstreaming, budgeting, and empowerment
training follow.The core components include market surveys to shift from “grow AND
sell” to “grow TO sell” practices, motivating farmers to meet market demands. Crop
selection and planting calendars are determined collectively based on market insights.
Popular crops include tomatoes, kale, and onions. This approach empowers small‐scale
horticultural farmers in Kenya.

SSGACPMTSM Arranged or curated offers from buyers The In response to low banana yields and limited market access for farmers,
(KEN) international NGOs Africa Harvest and TechnoServe initiated a collaborative effort in
2003—self‐sustaining groups to access clean planting material, technical support, and
markets (SSGACPMTSM). They encouraged small‐scale banana farmers in Kenya's
central highlands to form self‐sustaining groups to access clean planting material,
technical support, and markets. This initiative takes a comprehensive approach,
addressing the entire value chain, from technology adoption and production to
marketing. Over the years, thousands of banana growers in the region have organized
into farmer groups as part of this initiative.

VML2 (TZA) Arranged or curated offers from buyers Vertical Market Linkages 2 (VML2) was a framework for agricultural commercialization
that addresses various market failures. It included contract farming and outgrower
schemes (often involving contractual arrangements between farmers and buyers), or
full vertical integration (where a nucleus firm controls most stages of the supply chain).
The intervention's aim was to improve market access, reduce transaction costs, lower
crop production risks, decrease information asymmetries, and provide better access to
inputs, credit, and new technologies. The intervention comprised business contacts,
private businesspersons, groceries and other local merchants, factories, and
slaughterhouses, creating direct connections between agri‐processing (AP) firms and
sales to local businesses with less traditional market channels.

WFGP (IND) Alternative physical marketplaces; The Women Farmers with Global Potential (WFGP) initiative, led by SEWA in India,
Provision of output market information established over 100 rural producer associations or “farmer development centers” in
(mobile and internet) 300 villages across four districts in Gujarat. This initiative, which ran from 2008 to
2010, aimed to empower female farmers by providing them with training, input
procurement, financial access, price dissemination, and market linkages.Over 2000
female farmers participated in this program, with the primary goal of creating a women‐
managed agricultural extension system. SEWA also facilitated credit access through
group‐based banking, and provided market information to help women access global
agricultural markets.

WIISP (ETH) Arranged or curated offers from buyers In 2013, the Ethiopian Agricultural Transformation Agency introduced the Wheat
Initiative to Increase Smallholder Productivity (WIISP). This targeted 2000 “benchmark”
farmers in four major wheat‐producing regions, offering them a comprehensive
package that included training, inputs such as certified seeds and fertilizer, and a market
guarantee from the Ethiopian Grain Trade Enterprise. Benchmark farmers were
required to have a minimum 0.5‐hectare wheat plot and be open to adopting new
practices. The enterprise assured them of purchasing their wheat at or above market
prices. This initiative aimed to address various constraints on wheat productivity by
providing improved inputs, farming techniques, and a secure market for farmers.

eNAM (IND) Online commodity exchanges and The eNAM platform, initiated in April 2016 and operating in over 1000 regulated
mobile‐based marketplaces; Provision agricultural produce market committee markets across India, aims to unify the country's
of output market information (mobile agricultural commodity trading. By connecting 2477 of these markets, eNAM aims to
and internet) streamline agricultural trade. It has registered 17.3 million farmers, 0.22 million traders,
and 2140 farmers' producer organizations across 21 states, facilitating transactions
worth USD26.82 billion. Compared to open auctions, which are susceptible to
collusion, and manual tendering, which is prone to manipulation, e‐trading is
transparent and quick, benefiting farmers with real‐time price information.
a
ISO3 country code provided in parentheses ().
18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
|91 of 191

Distribution of outcome types.

F I G U R E 56
ET AL.
MARION
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
92 of 191 | MARION ET AL.

F I G U R E 57 Distribution of the length of studies' follow‐up periods.

comparison groups, thereby making them suspect selection bias. Estimates here use different measures of transport costs (k = 2)
Regarding the risk of confounding, several authors did not report or cost of marketing (k = 3) at the household level (k = 4) and the
testing for parallel time trend assumption, despite their use of DID product level (k = 1). For example, Ashraf and colleagues (2009)
design (Rutherford et al., 2016; Aysoy et al., 2015; Lentz & Upton, measured the total spent on marketing (in KSh1000) at the household
2016; Bayiyana et al., 2018; Salvaña et al., 2011; Warsanga & Evans, level, while Liu and colleagues (2021) measured marketing costs at
2018). Additional concerns included spillover, crossover, and con- the product level.
tamination (28% of estimates were rated as high risk of bias), Market interventions did not reduce transaction costs. The overall
reporting bias (7%), and outcome measurement bias (8%). Estimates weighted average effect is very small, negative, and not statistically
were not likely to be at high risk for performance bias. ˆ = −0.01; 95% CI: −0.09 to 0.07; p = 0.79)
significant (SMD
(Figure 59). The effect sizes here are all reversed meaning that a
Summary of the meta‐analyses results of market interventions. Table 20 higher effect represents lower transaction costs. Effect sizes vary
displays the meta‐analyses results across outcome types covered in between −0.12 and 0.09 standard deviations. The heterogeneity
the subsequent sections. among the studies is substantial (Q (4) = 11.94 , p = 0.02, τ̂ 2 = 0.005,
I2 = 66.50%).
Immediate outcomes. The immediate outcomes category covers the first The estimates from Ashraf and colleagues (2009) and Liu and
outcomes that may affect farmers across the causal chain. They cover colleagues (2021) are statistically significant but do not have the
financial resources (e.g., transaction costs and credit), market linkages and same direction. In Kenya, households reported the total cost of
bargaining power (e.g., group participation and prices), and the adoption marketing in KSh1000 (Ashraf et al., 2009; ĝ = 0.10 ; 95% CI:
of new practices and agricultural approaches (e.g., use/adoption of <0.001 to 0.19; p = 0.049). In China, RECAS increased the market-
improved inputs, technologies, and practices and farm investments). ing costs for apples (Liu et al., 2021; ĝ = −0.12; 95% CI: −0.21 to
−0.04; p = 0.01). In Ethiopia, Gelo and colleagues (2020) did not
Transaction costs. Within five studies, we identified five independent find that FOs led to a reduction in total cost of selling maize:
estimates of the effect of initiatives creating new marketplaces and despite being very small and positive, the effect of the intervention
alternative marketing opportunities (hereafter market interventions) on is not statistically significant (ĝ = 0.05; 95% CI: −0.10 to
transaction cost outcomes. Studies of evaluated interventions took 0.20; p = 0.55).
place in two regions: Africa (n = 4), and East Asia and the Pacific The exploratory moderator analysis indicated that more recent
(n = 1). Of the five studies, four evaluated multi‐component interven- interventions and longer exposure to the intervention appear to
tions (DMHECMCP, FO, P4P, SHEP).16 Three interventions sup- have smaller effects. Of the five included effect sizes, three remain
ported arranged or curated offers from buyers (FO, P4P, SHEP), one after excluding high risk of bias estimates. The effect of market
intervention focused on alternative marketplaces (DMHECMCP), and interventions on transaction cost outcomes is still negative, very
one focused on online commodity exchanges and mobile‐based ˆ = −0.01; 95% CI: −0.07
small, and not statistically significant (SMD
marketplaces (RECAS). to 0.05; p = 0.73).

16 Use/adoption of fertilizers, pesticides, herbicides, and other improved

Information on interventions is provided in Supporting Information: Appendix 3 for
Chapter 3. inputs. Five studies with five independent estimates analyzed the
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 93 of 191

F I G U R E 58 Risk of bias of included estimates between quasi‐experimental design (QED) (top) and randomized controlled trial (RCT) (bottom).

effect of market interventions on the use/adoption of fertilizers, Market interventions did not increase the use/adoption of fertilizers,
pesticides, herbicides, and other improved inputs. Studies of eval- pesticides, herbicides, and other improved inputs. The overall weighted
uated interventions took place in three regions: Africa (n = 3), Latin average effect is moderate‐sized, positive, and not statistically
America and the Caribbean (n = 1), and East Asia and the Pacific ˆ = 0.31; 95% CI: −0.20 to 0.82; p = 0.23)
significant (SMD
(n = 1). Of the five studies, three evaluated multi‐component (Figure 60). The observed outcomes ranged from −0.12 to 1.34
interventions (ECX, WIISP, GIADP). Three interventions supported and there is considerable heterogeneity among the estimates
arranged or curated offers from buyers (WIISP, PP, IPIAP), one (Q (4) = 346.211, p < 0.001, τ̂ 2 = 0.33, I2 = 98.85%).
intervention focused on alternative marketplaces (GIADP), and one In Ecuador, Cavatassi and colleagues (2011) observed a positive,
focused on online commodity exchanges and mobile‐based market- small, and statistically significant effect of PP on the quantity of
places (ECX). insecticides applied (ĝ = 0.16; 95% CI: 0.01 to 0.31;p = 0.03). In
Estimates here use different measures of farmers' use of Tanzania, Berg and Levely (2019), observed a positive, large, and
fertilizers or pesticides at the household level (k = 5). For example, statistically significant effect of IPIAP on the likelihood of using
Belay and Ayalew (2020) measured the allocation of fertilizer as a organic fertilizers (ĝ = 1.34 ; 95% CI: 1.22 to 1.46; p < 0.001).
share of ECX‐traded commodities, while Garbero and Songserm- None of the exploratory moderator variables significantly
sawas (2018) measured fertilizer use in kilograms per hectare. contributed to the variation in estimates. Of the five included effect
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
94 of 191 | MARION ET AL.

T A B L E 20 Summary of meta‐analyses results.

Outcome types Outcome subcategory SMD 95% CI k p Value

Immediate outcomes Transaction costs −0.01 −0.09, 0.07 5 0.79

Group participation 0.92 0.43, 1.41 4 <0.001

Use/adoption of fertilizers, 0.31 −0.20, 0.82 5 0.23

pesticides, and herbicides

Use/adoption of improved 0.93 −0.23, 2.08 5 0.12

varieties and quality seeds

Use/adoption of other 0.22 0.10, 0.35 12 0.001

improved inputs

Farm investment (total) 0.24 0.05, 0.42 6 0.01

Farm investment (fertilizers) 0.13 −0.04, 0.30 4 0.13

Farm investment (seeds) 0.09 −0.04, 0.22 3 0.19

Farm investment (labor) 0.17 0.04, 0.31 4 0.01

Farm investment (livestock) 0.34 0.08, 0.60 3 0.01

Credit 0.12 −0.01, 0.25 4 0.08

Prices 0.13 0.05, 0.20 24 <0.001

Agricultural Production Crop loss N/A

outcomes
Yields 0.16 0.04, 0.27 16 0.01

Volume 0.02 −0.08, 0.12 14 0.67

Quality N/A

Produce type (area) 0.001 −0.06, 0.06 3 0.92

Produce type (livestock) N/A

Produce type (diversity) 0.07 −0.03, 0.16 13 0.16

Intermediate outcomes Sales 0.12 0.03, 0.20 23 0.01

Farm income 0.15 0.10, 0.20 27 <0.001

Off‐farm outcomes Labor market outcomes N/A

(agriculture only)

Non‐farm outcomes Non‐farm income 0.02 −0.06, 0.1 8 0.60

Labor market outcomes N/A

Welfare outcomes Total income and wealth 0.10 0.05, 0.15 29 <0.001

Food security 0.10 0.02, 0.17 17 0.01

Nutrition 0.03 −0.05, 0.10 5 0.52

Note: Outcome subcategories in bold have a statistically significant SMD at a 5% significance level. K is the number of estimates. N/A indicates that we did
not conduct a meta‐analysis for these outcome types. In these cases, k = 1 or k = 2 and we chose not to conduct a meta‐analysis in those cases due to
various differences (e.g., intervention types, regions).

sizes, two remain after excluding high risk of bias estimates. The interventions on the use/adoption of improved varieties and quality
effect of market interventions on the use/adoption of fertilizers, seeds. All studies of evaluated interventions took place in Africa:
pesticides, herbicides, and other improved input outcomes becomes Ethiopia (n = 2), Mozambique, Kenya, and Tanzania (respectively
ˆ = 0.08; 95% CI:
positive, very small, and statistically significant (SMD n = 1). Of the five studies, two evaluated multi‐component interven-
0.002 to 0.16; p = 0.04). tions (WIISP, SSGACPMTSM]. Four interventions supported arranged
or curated offers from buyers (WIISP, IAP, IPIAP, SSGACPMTSM), and
Use/adoption of improved varieties and quality seeds. Five studies with one focused on online commodity exchanges and mobile‐based
five independent estimates analyzed the effect of market marketplaces (ECX).
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 95 of 191

F I G U R E 59 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of market interventions on
transaction costs.

F I G U R E 60 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of market interventions on the
use/adoption of fertilizers, pesticides, herbicides, and other improved inputs.

F I G U R E 61 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of market interventions on the
use/adoption of improved varieties and quality seeds.

Estimates here use different measures of farmers' use of −0.23 to 2.08; p = 0.12) (Figure 61). The observed outcomes ranged
certified or improved varieties and quality seeds at the household from −0.08 to 2.95 and there is considerable heterogeneity among
level (k = 5). For example, Abate and colleagues (2018) measured estimates (Q (4) = 1284.06, p < 0.001, τ̂ 2 = 1.71, I2 = 99.69%).
the quantity of certified seeds applied, while Belay and Ayalew In Kenya, Fischer and Qaim (2012) observed a positive, large, and
(2020) measured the share of improved seeds used for ECX statistically significant effect of SSGACPMTSM on the adoption of
commodities (coffee, sesame, maize, wheat, and haricot beans) in tissue culture plantlets (ĝ = 1.57; 95% CI: 0.98 to 2.16; p < 0.001). In
the enumeration area. Tanzania, Berg and Levely found a similar size, direction, and
Market interventions did not affect the use/adoption of improved significant effects of IPIAP on the likelihood of planting improved
varieties and quality seeds. The overall weighted average effect is large seeds (ĝ = 2.95; 95% CI: 2.81 to 3.10; p < 0.001). In Ethiopia, Belay
ˆ = 0.93; 95% CI:
and positive, but not statistically significant (SMD and Ayalew (2020) observed a positive, small, and statistically
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
96 of 191 | MARION ET AL.

significant effect of the ECX on the allocation of improved seeds as a (Figure 62). The observed outcomes ranged from −0.23 to
share of ECX‐traded commodities (ĝ = 0.15; 95% CI: 0.04 to 0.90 and there is considerable heterogeneity among estimates
0.25; p = 0.006). (Q (11) = 133.59, p < 0.001, τ̂ 2 = 0.04 , I2 = 91.77%).
None of the exploratory moderator variables significantly Lombardini (2017) in Armenia observed a positive, large, and
contributed to the variation in estimates. Of the five included effect statistically significant effect of ARMA66 on the adoption of drip
sizes, two remain after excluding high risk of bias estimates. The irrigation systems (ĝ = 0.90 ; 95% CI: 0.70 to 1.09; p < 0.001). In China,
effect of market interventions on the use/adoption of improved seed Liang and colleagues (2020) observed an effect of similar direction, size,
outcomes becomes positive, small, and statistically significant and significance from CM on the number of producers with a pollution‐
ˆ = 0.11; 95% CI: 0.04 to 0.18; p = 0.001).
(SMD free, green, or organic vegetable certification (ĝ = 0.63; 95% CI: 0.18 to
1.07; p = 0.01). In Ethiopia, Belay and Ayalew (2020) observed a
Use/adoption of other improved inputs. Twelve studies with twelve positive, moderate, and statistically significant effect of the ECX on the
independent estimates analyzed the effect of market interventions allocation of irrigation inputs as a share of ECX‐traded commodities
on the use/adoption of other improved inputs. The studies of (ĝ = 0.23; 95% CI: 0.13 to 0.34; p < 0.001).
evaluated interventions took place in regions: Africa (n = 7), East The moderator analysis indicated that studies with a longer
Asia and the Pacific (n = 2), Europe and Central Asia (n = 1), and follow‐up period showed larger effects. We identified funnel plot
Latin America and the Caribbean (n = 2). Of the twelve studies, ten asymmetry, meaning that publication bias is likely to be present. Of
evaluated multi‐component interventions (ARMA66, ECX, SDCP, the 12 included effect sizes, four remain after excluding high risk of
DMHECMCP, ARMA66, GIADP, MC, ACE, GUAB38, P4P). Seven bias estimates. The effect of market interventions on the use/adoption
interventions supported arranged or curated offers from buyers of other improved inputs outcomes is still positive, and moderate‐sized,
(SDCP, AFP, P4P, ARMA66, ACE, GUAB38, CM), four focused on ˆ = 0.34; 95% CI: −0.02
but is no longer statistically significant (SMD
alternative marketplaces (DMHECMCP, OILCSS, GIADP, MC), and to 0.71; p = 0.06).
one focused on online commodity exchanges and mobile‐based
marketplaces (ECX). Farm investment (total). Six studies with six independent estimates
Estimates here use different measures of farmers' use of new analyzed the effect of market interventions on the total farm
agricultural practices or the adoption of improved cattle feed at the investment outcomes. The studies of evaluated interventions took
household level. For example, Bonilla and colleagues (2018) place in four regions: Africa (n = 3), South East Asia (n = 1), Europe and
measured whether the intervention affected giving cattle concen- Central Asia (n = 1), and Latin America and the Caribbean (n = 1). Of
trate feeds and mineral supplements (binary variable), while Belay and the six studies, two evaluated multi‐component interventions
Ayalew (2020) measured the impact of the intervention on the share (SSGACPMTSM, SHEP). Four interventions supported arranged or
of irrigation used for ECX commodities (coffee, sesame, maize, wheat, curated offers from buyers (AMQM, PP, SSGACPMTSM, SHEP), one
and haricot beans) in the enumeration area. focused on alternative marketplaces (AMP), and one focused on online
Market interventions increased the use/adoption of other improved commodity exchanges and mobile‐based marketplaces (FARMEX).
inputs. The overall weighted average effect is moderate, positive, and Estimates here use different measures of farmers' total expenses
ˆ = 0.22; 95% CI: 0.10 to 0.35; p = 0.001)
statistically significant (SMD and investment in agricultural lands or resources at the household

F I G U R E 62 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of market interventions on the
use/adoption of other improved inputs.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 97 of 191

level (k = 6). For example, Bold and colleagues (2022) analyzed the Farm investment (fertilizers). Four studies with four independent
effect of AMQM in Uganda on the expenses of seeds, fertilizers, estimates analyzed the effect of market interventions on farm
boosters, and chemicals. In Tajikistan, Morgan and Stewart (2019) investment (fertilizer) outcomes. The studies of evaluated inter-
measured the impact of AMP on the total invested in farming land ventions took place in three regions: Africa (n = 2), South East
and off‐farm business in the past 12 months. Asia (n = 1), and Latin America and the Caribbean (n = 1). Of the
Market interventions increased the total farm investment. The four studies, two evaluated multi‐component interventions
overall weighted average effect is moderate‐sized, positive, and (SSGACPMTSM, SHEP). Three interventions supported arranged
ˆ = 0.24; 95% CI: 0.05 to 0.42; p = 0.01)
statistically significant (SMD or curated offers from buyers (PP, SSGACPMTSM, SHEP), and one
(Figure 63). The observed outcomes ranged from 0.01 to 0.86 and focused on online commodity exchanges and mobile‐based market-
there is substantial heterogeneity among estimates (Q (5) = 30.91, places (FARMEX).
p < 0.0001, τ̂ 2 = 0.03, I2 = 83.82%). Estimates here use different measures of farmers' total expenses
In Tajikistan, Morgan and Stewart (2019) observed a positive, and investment in fertilizers and pesticides at the household level
large, and statistically significant effect of AMP on the total invested (k = 4). For example, Fischer and Qaim (2012) analyzed the effect of
in farming land and off‐farm business in the past 12 months SSGACPMTSM in Kenya on the cost of fertilizers and pesticides. In
(ĝ = 0.38; 95% CI: 0.24 to 0.52; p < 0.001). In Kenya, Fischer and India, Rajkhowa (2021) measured the impact of FARMEX on the
Qaim (2012) observed an effect of similar direction, size, and fertilizer expenditure per acre.
significance for SSGACPMTSM on the total cost incurred by farmers Market interventions did not increase farm investment on fertilizer.
(ĝ = 0.86; 95% CI: 0.30 to 1.42; p = 0.003). The overall weighted average effect is small, positive, and not
None of the exploratory moderator variables significantly ˆ = 0.13; 95% CI: −0.04 to 0.30; p = 0.13)
statistically significant (SMD
contributed to the variation in estimates. Of the six included effect (Figure 64). The observed outcomes ranged from −0.01 to 0.54 and
sizes, two remain after excluding high risk of bias estimates. The there is considerable heterogeneity among estimates (Q (3) = 9.77,
effect of market interventions on the total farm investment outcomes is p = 0.02, τ̂ 2 = 0.02, I2 = 69.31%).
then positive, small, and no longer statistically significant Cavatassi and colleagues (2011) found a positive, small, and
ˆ = 0.10; 95% CI: −0.16 to 0.35; p = 0.44).
(SMD statistically significant effect of PP in Ecuador on the total cost of

F I G U R E 63 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of market interventions on
total farm investments.

F I G U R E 64 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of market interventions on
farm investment (fertilizer).
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
98 of 191 | MARION ET AL.

chemical fertilizers (in USD/ha) observed by (ĝ = 0.19; 95% CI: 0.04 Farm investment (labor). Four studies with four independent esti-
to 0.35; p = 0.01). mates analyzed the effect of market interventions on farm investment
The exploratory moderator analysis indicated that interventions (labor) outcomes. The studies of evaluated interventions took place in
with a longer exposure period seem to show larger effects. Of the two regions: Africa (n = 3) and Latin America and the Caribbean
four included effect sizes, one remains after excluding high risk of (n = 1). Of the four studies, two evaluated multi‐component
bias estimates. This effect is reported by Shimizutani and colleagues interventions (SSGACPMTSM, ACE). All four interventions supported
(2021): they observed a negative, very small, and not statistically arranged or curated offers from buyers (IAP, PP, SSGACPMTSM, ACE).
significant effect of SHEP on the total cost of fertilizers (ĝ = −0.01; Estimates here use different measures of farmers' total expenses
95% CI: −0.08 to 0.06; p = 0.74). and the prevalence of hired labor for agriculture at the household
level (k = 4). For example, Ghebru and colleagues (2019) analyzed the
Farm investment (seeds). Three studies with three independent effect of the IAP in Mozambique on the prevalence of household‐
estimates analyzed the effect of market interventions on farm hired labor for any farming activity. In Ecuador, Cavatassi and
investment (seeds) outcomes. The studies of evaluated interventions colleagues (2011) measured the impact of PP on the cost of paid
took place in three regions: Africa (n = 1), South East Asia (n = 1), Latin labor in (USD/ha).
America, and the Caribbean (n = 1). Of the three studies, one Market interventions increased farm investment in labor. The
evaluated a multi‐component intervention (SHEP). Two interventions overall weighted average effect is small, positive, and statistically
supported arranged or curated offers from buyers (PP, SHEP), and one ˆ = 0.17; 95% CI: 0.04 to 0.31; p = 0.01) (Figure 66).
significant (SMD
focused on online commodity exchanges and mobile‐based market- The observed outcomes ranged from 0.044 to 0.83 and there is
places (FARMEX). substantial heterogeneity among estimates (Q (3) = 7.34, p = 0.06,
Estimates here use different measures of farmers' total expenses τ̂ 2 = 0.01, I2 = 59.14 %).
and investment in seeds at the household level (k = 3). For example, Fischer and Qaim (2012) found a large positive and statistically
Cavatassi and colleagues (2011) analyzed the effect of PP in Ecuador significant effect of SSGACPMTSM in Kenya on households' amount
on the cost of seeds purchased (in USD/ha). In India, Rajkhowa (2021) of hired labor in hours per acre (ĝ = 0.83; 95% CI: 0.26 to 1.39;
measured the impact of FARMEX on seed expenditure per acre. p = 0.004) In Ecuador, Cavatassi and colleagues (2011) observed a
Market interventions did not increase farm investment in seeds. The positive, small and statistically significant effect of PP on the cost of
overall weighted average effect is very small, positive, and not paid labor in USD/ha: (ĝ = 0.19; 95% CI: 0.04 to 0.34; p = 0.01).
ˆ = 0.09; 95% CI: −0.04 to 0.22; p = 0.19)
statistically significant (SMD None of the exploratory moderator variables significantly
(Figure 65). The observed outcomes ranged from 0.01 to 0.19 and contributed to the variation in estimates. Of the four included effect
there is substantial heterogeneity among estimates (Q (2) = 4.80, sizes, one remains after excluding high risk of bias estimates. Ghebru
p = 0.09, τ̂ 2 = 0.008, I2 = 58.37%). Cavatassi and colleagues (2011) and colleagues (2019) analyzed the effect of the IAP in Mozambique
reported a positive, small, and statistically significant effect of PP in and observed a positive, small, and statistically significant effect on
Ecuador on the total cost of seeds (in USD/ha) (ĝ = 0.19; 95% CI: whether a household hired labor for any farming activity (ĝ = 0.16;
0.04 to 0.34; p = 0.01). 95% CI: 0.07 to 0.25; p < 0.001).
The exploratory moderator analysis indicated that more recent
interventions show smaller effects. Of the three included effect sizes, Farm investment (livestock). Two studies with three independent
two remain after excluding high risk of bias estimates. The effect of estimates analyzed the effect of market interventions on farm
market interventions on farm investment (seeds) outcomes is still investment (livestock) outcomes. The studies of evaluated interven-
ˆ = 0.02; 95%
positive, very small, and not statistically significant (SMD tions took place in two regions: Africa (n = 1) and South Asia (n = 1).
CI: −0.11 to 0.08; p = 0.64 ). Of the two studies, one evaluated a multi‐component intervention

F I G U R E 65 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of market interventions on
farm investment (seeds).
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 99 of 191

F I G U R E 66 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of market interventions on
farm investment (labor).

F I G U R E 67 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of market interventions on
farm investment (livestock).

(FAD‐2). All interventions supported the establishment of alternative estimates (the study is from Savani and Stewart [2019], which was
marketplaces (FAD‐2, AMP). We found two estimates in Tijani and previously discussed).
colleagues (2014), as the evaluation focused on two different states
of Nigeria with independent comparative groups. Prices. We identified 24 studies with 24 independent estimates
Estimates here use different measures of farmers' livestock analyzing the effect of market interventions on price outcomes. The
expenses and livestock assets at the household level (k studies of evaluated interventions took place in five regions: Africa
= 3). For example, Savani and Stewart (2019) analyzed the effect of (n = 10), East Asia and the Pacific (n = 3), South Asia (n = 8), Latin
AMP in Bangladesh on purchased inputs for dairy farming in the last America and the Caribbean (n = 2), and Europe and Central Asia
12 months. In Nigeria, Tijani and colleagues (2014) measured the (n = 1). Eleven studies evaluated multi‐component interventions
impact of FAD‐2 on livestock assets. (SDCP, DMHECMCP, E‐CHOUPAL (Hub), ESPEWDS, eNAM,
Market interventions increased farm investment in livestock. The SSGACPMTSM, P4P, LCPC, OILCSS, ECX, FPC). Five interventions
overall weighted average effect is moderate‐sized, positive, and supported the establishment of alternative marketplaces (e.g.,
ˆ = 0.34 ; 95% CI: 0.08 to 0.60; p = 0.01)
statistically significant (SMD DMHECMCP, WFGP), 10 interventions focused on online commodity
(Figure 67). The observed outcomes ranged from 0.10 to 0.54 and exchanges, and 11 interventions focused on arranged and curated
there is substantial heterogeneity among estimates (Q (2) = 14.24 , offers from buyers (e.g., ARMA66, FARMA).
p = 0.0008, τ̂ 2 = 0.05, I2 = 85.96%). Estimates here use different measures of prices of agricultural
Savani and Stewart (2019) found a large, positive, and outputs for the household (k = 15); group (k = 1); market (k = 2); or
statistically significant effect of AMP in Bangladesh on whether group, product, and individual level (k = 6). For example, Bold and
households purchased inputs for dairy farming in the last 12 months colleagues (2022) analyzed the effect of AMQM in Uganda on the
(ĝ = 0.54 ; 95% CI: 0.40 to 0.69; p < 0.001). In Kogi State, Nigeria, price of maize per kilogram at the household level.
Tijani and colleagues (2014) observed a positive, large, and Market interventions improved the prices received by farmers. The
statistically significant effect of FAD‐2 on the number of livestock overall weighted average effect is small, positive, and statistically
assets (ĝ = 0.37; 95% CI: 0.19 to 0.54; p < 0.001). ˆ = 0.13; 95% CI: 0.05 to 0.20; p < 0.001) (Figure 68).
significant (SMD
None of the exploratory moderator variables significantly The observed outcomes ranged from −4.61 to 0.65 and there is
contributed to the variation in estimates. Of the three included considerable heterogeneity among estimates (Q (23) = 239.29,
effect sizes, one remains after excluding the high risk of bias p < 0.001, τ̂ 2 = 0.02, I2 = 90.39%).
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
100 of 191 | MARION ET AL.

F I G U R E 68 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of market interventions on
prices.

An examination of the studentized residuals revealed that Aysoy Yields. We identified 16 studies with 16 independent estimates
and colleagues (2015) may be a potential outlier in the context of this analyzing the effect of market interventions on yield outcomes. The
model. This might be due to the very small sample size of this study studies of evaluated interventions took place in four regions: Africa
(37 agricultural items). According to Cook's distances, none of the (n = 7), Latin America and the Caribbean (n = 2), South Asia (n = 3), and
studies could be considered overly influential. If we drop Aysoy and East Asia and the Pacific (n = 4). Eight studies evaluated multi‐
colleagues (2015) from the analysis, the overall effect stays positive, component interventions (VML2, FPC, WIISP, E‐CHOUPAL, GIADP,
ˆ = 0.14 ; 95% CI: 0.08 to
small, and statistically significant: (SMD SSGACPMTSM, GUAB38, HVAP). Three interventions supported the
0.21; p < 0.001). establishment of alternative marketplaces (E‐CHOUPAL, GIADP,
Chaudhary and Suri (2022) observed a positive, large, and AHLAF), two interventions focused on online commodity exchanges
statistically significant effect of eNAM on prices in India (ĝ = 0.65; (ICTIRR, FARMEX), and 11 interventions focused on arranged and
95% CI: 0.40 to 0.89; p = 0.006). In Niger, Fuller (2012) observed a curated offers from buyers (VML2, AMQM, WIISP, AFP, CM, FPC, PP,
positive, large, and statistically significant effect of LCPC on prices SSGACPMTSM, IPIAP, GUAB38, HVAP).
(ĝ = 0.64 ; 95% CI: 0.27 to 1.01; p < 0.001). In Pakistan, Lombardini Estimates here use different measures of yields and technical
and Bowman (2015) observed a negative, moderate, and statistically efficiency at the household level (k = 13), district level (k = 1), plot level
significant effect of ESPEWDS on the price of milk (ĝ = −0.29; 95% (k = 1), and extension planning‐area level (k = 1). For example, Abate and
CI: −0.48 to −0.10; p = 0.002). colleagues (2018) analyzed the effect of WIISP in Ethiopia on wheat
None of the moderator variables significantly contributed to the yields at the household level. In India, Goyel (2010) analyzed the effect
variation in estimates. We identified funnel plot asymmetry, meaning of E‐CHOUPAL on the total yield at the district level.
that publication bias is likely to be present. Of the 24 studies, eight effect Market interventions increased yields. The overall weighted
sizes remain after excluding high risk of bias estimates. The effect of average effect is small, positive, and statistically significant
market interventions on price outcomes is then positive, very small, and ˆ = 0.16; 95% CI: 0.04 to 0.27; p = 0.01) (Figure 69).
(SMD
ˆ = 0.10 ; 95% CI: 0.01 to 0.19; p = 0.03).
statistically significant (SMD The observed outcomes ranged from −0.94 to 0.45 and there
is substantial heterogeneity among estimates (Q (15) = 95.20 ,
Agricultural production outcomes. The agricultural production outcomes p < 0.0001, τ̂ 2 = 0.04 , I2 = 84.24 %).
are yields, volume, produce type, quality, and crop loss. The included Bold and colleagues (2022) observed a positive, moderate‐sized,
studies provided evidence on all these outcomes. and statistically significant effect of AMQM in Uganda on the harvest
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 101 of 191

F I G U R E 69 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of market interventions on
yields.

in kilogram per acre of cultivated land (ĝ = 0.32; 95% CI: 0.01 to 0.63; ˆ = 0.02; 95% CI: −0.08 to 0.12; p = 0.67)
statistically significant (SMD
p = 0.04 ). In Tanzania, Berg and Levely (2019) observed a large, (Figure 70). The observed outcomes ranged from −0.82 to 0.39 and
positive, and statistically significant effect of IPIAP on maize yields there is substantial heterogeneity among estimates (Q (13) = 77.83,
(ĝ = 0.44 ; 95% CI: 0.32 to 0.56; p < 0.001). In Kenya, Fischer and p < 0.0001, τ̂ 2 = 0.03, I2 = 83.30 %).
Qaim (2012) observed a negative, large, and statistically significant Bonilla and colleagues (2018) in Kenya analyzed the effect of
effect of SSGACPMTSM on banana yields (in kilogram/hectare) SDCP on total milk production for the previous day. They observed a
(ĝ = −0.94 ; 95% CI: −1.51 to −0.37; p = 0.001). positive, small, and statistically significant effect of the intervention
The moderator analysis indicated that multi‐component inter- (ĝ = 0.13; 95% CI: 0.06 to 0.21; p < 0.001). In Bangladesh, Savani and
ventions show smaller effects. We did not identify any funnel plot Stewart (2019) analyzed the effect of AMP on dairy production in the
asymmetry, meaning that publication bias is not likely to be present. past week and observed a positive, large, and statistically significant
Of the 16 included effect sizes, four remain after excluding high risk effect (ĝ = 0.39; 95% CI: 0.25 to 0.54; p < 0.001).
of bias estimates. The effect of market interventions on yields is then In Tanzania, Anguko and Rana (2017) observed a positive,
ˆ = 0.17 ;
positive, small, and no longer statistically significant (SMD moderate‐sized, and statistically significant effect of OILCSS on the
95% CI: −0.15 to 0.48; p = 0.31). total quantity of crops produced in the previous 12 months (ĝ = 0.22;
95% CI: 0.05 to 0.39; p = 0.01). In Malawi, Zant (2020) observed a
Volume. We identified 14 studies with 14 independent estimates large, negative, and statistically significant effect of AHLAF on total
analyzing the effect of market interventions on volume outcomes. The tobacco production (ĝ = −0.82 95% CI: −1.15 to −0.49; p < 0.001).
studies of evaluated interventions took place in four regions: Africa The moderator analysis indicated that non‐staggered interven-
(n = 7), Latin America and the Caribbean (n = 1), South Asia (n = 5), and tions show larger effects. We did not identify any funnel plot
Europe and Central Asia (n = 1). Ten studies evaluated multi‐component asymmetry, meaning that publication bias is not likely to be present.
interventions (SDCP, DMHECMCP, E‐CHOUPAL, WFGP, ESPEWDS, Of the 14 included effect sizes, five remain after excluding high risk
EJNG, GUAB38, KHDP, ARMA66, OILCSS). Eight interventions of bias estimates. The effect of market interventions on volume
supported the establishment of alternative marketplaces (DMHECMCP, ˆ = 0.16; 95% CI:
increases but is still not statistically significant (SMD
E‐CHOUPAL, WFGP, AMP, OILCSS, ESPEWDS, AHLAF, KHDP), and −0.03 to 0.35; p = 0.11).
seven interventions focused on arranged and curated offers from buyers
(AMQM, SDCP, ESPEWDS, EJNG, IPIAP, GUAB38, ARMA66). Intermediate outcomes. The intermediate category covers outcomes
Estimates here use different measures for volume of agricultural that may affect farmers following the harvest period through the sale
outputs at the household level (k = 12), district level (k = 1), and of their agricultural products. This includes two outcome types: sales
extension planning‐area level (k = 1). For example, Ashraf and and farm income.
colleagues (2009) analyzed the effect of DMHECMCP in Kenya on
the total production of French beans at the household level. In India, Sales. We identified 23 studies with 23 independent estimates
Goyal and colleagues (2010) focused on the impact of E‐CHOUPAL analyzing the effect of market interventions on sales outcomes. The
on total crop production at the district level. studies of evaluated interventions took place in four regions: Africa
Market interventions did not increase the volume of outcomes. The (n = 11), Latin America and the Caribbean (n = 4), South Asia (n = 6),
overall weighted average effect is very small, positive, and not and Europe and Central Asia (n = 2). Fifteen studies evaluated multi‐
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
102 of 191 | MARION ET AL.

F I G U R E 70 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of market interventions on
volume.

component interventions (SDCP, E‐CHOUPAL, P4P, WFGP, AVCPO, meaning that publication bias is not likely to be present. Of the 23
SSGACPMTSM, EJNG, SHEP, GUAB38, LCPC, KHDP, ARMA66, included effect sizes, eight remain after excluding high risk of bias
FARMA, OILCSS, HVAP). Seven interventions supported the estab- estimates. The effect of market interventions on sales is then positive,
lishment of alternative marketplaces (E‐CHOUPAL, WFGP, AVCPO, ˆ = 0.20; 95% CI:
moderate‐sized, and not statistically significant (SMD
AMP, FM, OILCSS, KHDP), two interventions focused on online 0.03 to 0.38; p = 0.02).
commodity exchanges (MPBMAC, FARMEX), and 14 interventions
focused on arranged and curated offers from buyers (SDCP, P4P, IAP, Farm income. We identified 29 studies with 29 independent estimates
HVAP, PP, SSGACPMTSM, EJNG, PAAP, SHEP, IPIAP, GUAB38, analyzing the effect of market interventions on farm income outcomes.
LCPC, ARMA66, FARMA). The studies of evaluated interventions took place in five regions: Africa
Estimates here use different measures of volume for the sale of (n = 13), Latin America and the Caribbean (n = 3), South Asia (n = 4), East
agricultural outputs at the household level (k = 21), market level Asia and the Pacific (n = 8), and Europe and Central Asia (n = 1).
(k = 1), and producer organization level (k = 1). For example, Lentz and Seventeen studies evaluated multi‐component interventions (KPCCCF,
Upton (2016) analyzed the effect of P4P in Tanzania on the CFH, DMH, AGP, WFGP, ARMA66, GIADP, SHEP, GOPAPAEC, EADD,
percentage of crops sold at the household level. In India, Goyal and LCPC, KHDP, IAR4D, FPC, HVAP, OILCSS, HVAP).
colleagues (2010) focused on the impact of E‐CHOUPAL on the Nine interventions supported the establishment of alternative
volume of sales at the market level. marketplaces (DMH, WFGP, AMP, FM, OILCSS, GIADP, EADD,
Market interventions increased sales. The overall weighted average KHDP, IAR4D), six interventions focused on online commodity
ˆ = 0.12; 95%
effect is small, positive, and statistically significant (SMD exchanges (MPBMAC, RECAS, AGP, ECA, FARMEX, GOPAPAEC),
CI: 0.03 to 0.20; p = 0.01) (Figure 71). The observed outcomes ranged and 15 interventions focused on arranged and curated offers from
from −0.31 to 0.82 and there is substantial heterogeneity among buyers (KPCCCF, CFH, AMQM, AFP, CM, HVAP, FPC, P4P, ARMA66,
estimates (Q (22) = 180.87, p < 0.0001, τ̂ 2 = 0.03, I2 = 87.84 %). PP, BBS, SHEP, PEVHL, LCPC, IAR4D).
Biggeri and colleagues (2022) in Ethiopia observed a positive, Estimates here use different measures of farm income for the
large, and statistically significant effect of AVCPO on the share of household level (k = 24), individual level (k = 3), product level (k = 1),
cereal production sold through cooperatives (ĝ = 0.82; 95% CI: 0.66 and group level (k = 1). For example, Wainaina and colleagues (2014)
to 0.98; p < 0.001). In Bangladesh, Savani and Stewart (2019) analyzed the effect of KPCCCF in Kenya on farmers' net income from
observed a positive, large, and statistically significant effect of AMP poultry at the individual level. In Vietnam, Wang and colleagues
on quantity of milk sold (in liters in the past week, ĝ = 0.40 ; 95% CI: (2014) analyzed the effect of CFH on the yearly household income.
0.25 to 0.54; p < 0.001). In Tanzania, Berg and Levely (2019) Market interventions increased the farm income. The overall
observed a negative, moderate, and statistically significant effect of weighted average effect is small, positive, and statistically significant
IPIAP on the percentage of maize crops sold (ĝ = −0.27; 95% CI: ˆ = 0.15; 95% CI: 0.10 to 0.20; p < 0.001) (Figure 72).
(SMD
−0.39 to −0.15; p < 0.001). The observed outcomes ranged from −0.15 to 0.60 and there
None of the moderator variables significantly contributed to the is substantial heterogeneity among estimates (Q (28) = 79.26,
variation in estimates. We did not identify any funnel plot asymmetry, p < 0.0001, τ̂ 2 = 0.009, I2 = 64.67%).
18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
103 of 191

Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of market interventions on

Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of market interventions on
|

F I G U R E 71

F I G U R E 72
ET AL.

farm income.
MARION

sales.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
104 of 191 | MARION ET AL.

Li and colleagues (2021) in China observed a positive, large, and income and wealth outcomes. The studies of evaluated interventions
statistically significant effect of ECA on the income of the household took place in five regions: Africa (n = 16), Latin America and the
from production and marketing in 10,000 yuan/capita (ĝ = 0.49; 95% Caribbean (n = 4), South Asia (n = 4), East Asia and the Pacific (n = 4),
CI: 0.16 to 0.81; p = 0.003). In Tanzania, Anguko and Rana (2017) and Europe and Central Asia (n = 1). Twenty‐two studies evaluated
observed a positive, large, and statistically significant effect of multi‐component interventions (e.g., CATF, DMHECMCP P4P).
OILCSS on the total revenue from sales of all crops in the previous 12 Eleven interventions supported the establishment of alternative
months (ĝ = 0.37; 95% CI: 0.14 to 0.60; p = 0.002). marketplaces (DMHECMCP, WFGP, AMP, FM, OILCSS, ESPEWDS,
The moderator analysis indicated that estimates from studies using GIADP, FAD‐2, EADD, KHDP), two interventions focused on online
QEDs and estimates based on average treatment effects on the treated commodity exchanges (ECRV, ECA), and 17 interventions focused on
(relative to those based on intention to treat) show larger effects. We arranged and curated offers from buyers (placement of supermarket
identified funnel plot asymmetry, meaning that publication bias is likely supply chains [PSSC], CATF, P4P, FO, IAP, HVAP, P4P Ghana, P2P
to be present. Of the 29 included effect sizes, 10 remain after excluding Tanzania, VML2, ARMA66, ESPEWDS, SSGACPMTSM, PAAP, SHEP,
high risk of bias estimates. The effect of market interventions on farm ACE, GUAB38, LCPC). We found two independent estimates in Tijani
ˆ = 0.12;
income is still positive, small, and statistically significant (SMD and colleagues (2014), as the evaluation focused on two different
95% CI: 0.04 to 0.19; p = 0.002). states of Nigeria with independent comparative groups.
Estimates here use different measures of total income at the
Welfare outcomes. The welfare outcomes category covers measures of household level (k = 29). For example, Michelson (2013) analyzed the
improvements in farmers' well‐being and quality of life. They effect of PSSC in Nicaragua on the welfare effect of supermarket
encompass a range of social, economic, and environmental factors; supply chain participation. In Vietnam, Saigenji (2012) analyzed the
however, in this review we focus on two: total income and food security. effect of CATF on the difference in daily mean income per capita.
Market interventions increased total income and wealth. The overall
Total income and wealth. We identified 29 studies with 29 indepen- weighted average effect is very small, positive, and statistically
dent estimates analyzing the effect of market interventions on total ˆ = 0.10 ; 95% CI: 0.05 to 0.15; p < 0.001) (Figure 73).
significant (SMD

F I G U R E 73 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of market interventions on
total income.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 105 of 191

F I G U R E 74 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of market interventions on
food security.

The observed outcomes ranged from −0.19 to 0.45 and there is (P4P Ghana, P4P Tanzania, VML2, FO, HVAP, P4P, ARMA66,
substantial heterogeneity among estimates (Q (28) = 108.62, ESPEWDS, EJNG, PAAP, ACE, GUAB38, LCPC).
p < .0001, τ̂ 2 = 0.01, I2 = 74.22%). Estimates here use different measures of food consumption and
Lentz and Upton (2016) in Tanzania observed a negative, small, food security indices at the household level (k = 17). For example, Lentz
and statistically significant effect of P4P (ĝ = −0.19; 95% CI: −0.37 and Upton (2016) analyzed the effect of P4P in Tanzania on the food
to −0.01; p = 0.04). In Nigeria, Tijani and colleagues (2014) consumption scope of households. In Ethiopia, Gelo and colleagues
observed a positive, large, and statistically significant effect of (2020) analyzed the effect of FOs on food expenditure share.
FAD‐2 on household income (ĝ = 0.36; 95% CI: 0.18 to 0.54; Market interventions improved food security. The overall weighted
p < 0.001). In Tanzania, de Castro (2021) found that VML2 had a average effect is small, positive, and statistically significant
positive, large and statistically significant effect on household ˆ = 0.10 ; 95% CI: 0.02 to 0.17; p = 0.01) (Figure 74). The
(SMD
income (ĝ = 0.41; 95% CI: 0.28 to 0.53; p < 0.001). However, observed outcomes ranged from −0.12 to 0.46 and there is
according to Cook's distance, this estimate is overly influential and substantial heterogeneity among estimates (Q (16) = 60.50 ,
could be considered an outlier. p < 0.001, τ̂ 2 = 0.02, I2 = 75.55%).
The moderator analysis indicated that interventions with longer Biggeri and colleagues (2022) in Ethiopia observed a positive, large,
exposure periods show larger effects. We did not identify funnel plot and statistically significant effect of AVCPO on household food insecurity
asymmetry, meaning that publication bias is unlikely to be present. Of access scale scores17 (ĝ = 0.37 ; 95% CI: 0.21 to 0.53; p < 0.001). In
the 29 included effect sizes, 11 remain after excluding high risk of Tanzania, Anguko and Rana (2017) observed a positive, moderate‐sized,
bias estimates. The effect of market interventions on total income and statistically significant effect of OILCSS on food consumption per
ˆ = 0.09;
stays positive, very small, and statistically significant (SMD adult equivalent per day (ĝ = 0.33; 95% CI: 0.18 to 0.48; p < 0.001).
95% CI: 0.01 to 0.17; p = 0.02). Effects are lower, on average, if they are based on arranged or
curated offers from buyers interventions, as compared to other types of
Food security. We identified 17 studies with 17 independent market interventions. We did not identify any funnel plot asymmetry,
estimates analyzing the effect of market interventions on food security meaning that publication bias is not likely to be present. Of the 17
outcomes. The studies of evaluated interventions took place in four studies, six remain after excluding high risk of bias estimates. The effect
regions: Africa (n = 9), Latin America and the Caribbean (n = 3), South of market interventions on food security outcomes is still positive and
Asia (n = 3), and Europe and Central Asia (n = 2). Fourteen studies ˆ = 0.13; 95% CI: −0.05
small, but no longer statistically significant (SMD
evaluated multi‐component interventions (e.g., P4P Ghana, P4P to 0.32; p = 0.15).
Tanzania, VML2). Five interventions supported the establishment of
alternative marketplaces (AVCPO, FM, AMP, OILCSS, ESPEWDS), and 17
The effect was reversed in the analysis to account for an improvement in food security as a
13 interventions focused on arranged and curated offers from buyers positive effect size.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
106 of 191 | MARION ET AL.

F I G U R E 75 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of market interventions on
nutrition.

Nutrition. We identified five studies with five independent estimates Land size. Table 22 shows effect sizes by land size subgroup from
analyzing the effect of market interventions on nutritional outcomes. Cavatassi and colleagues (2011), Fischer and Qaim (2012), and Li and
The studies of evaluated interventions took place in three regions: colleagues (2021). We do not show here all the relevant estimates
Africa (n = 2), South Asia (n = 2), and East Asia and the Pacific (n = 1). from Cavatassi and colleagues (2011) and Fischer and Qaim (2021).
Five studies evaluated multi‐component interventions (SDCP, AVC- Cavatassi and colleagues (2011) evaluated the effect of PP in
PO, GIADP, EJNG, HVAP). Two interventions supported the Ecuador, which created partnerships between small‐scale farmers
establishment of alternative marketplaces (AVCPO, GIADP), and three and various agricultural support providers, aiming to enhance potato
interventions focused on arranged and curated offers from buyers production and marketing. Effects by land size subgroup are
(SDCP, HVAP, EJNG). displayed across several outcomes in the table.
Estimates here use different measures of nutrition and nutrition Fischer and Qaim (2012) found that SSGACPMTSM reduced
indices at the household level (k = 5). For example, Bonilla and households' yields. Across land size subgroups, the effect is large,
colleagues (2018) analyzed the effect of SDCP in Kenya on the negative, and statistically significant for households with 1–3 acres of
HDDS. In Nepal, Kafle and colleagues (2022) used a similar index to landholding. This intervention encouraged small‐scale banana farm-
measure the dietary score of HVAP beneficiaries. ers in Kenya's central highlands to form self‐sustaining groups to
Market interventions did not improve nutritional outcomes. The access clean planting material, technical support, and markets.
overall weighted average effect is very small, positive, and not In Li and colleagues (2021), the effect of e‐commerce adoption in
ˆ = 0.03; 95% CI: −0.05 to 0.10; p = 0.52)
statistically significant (SMD China (ECA) on household net income is positive and statistically
(Figure 75). The observed outcomes ranged from −0.08 to 0.18 and significant for both types of households. The effect is slightly larger
there is substantial heterogeneity among estimates (Q (4) = 11.61, for households that cultivated more than 4.35 mu.18
p = 0.02, τ̂ 2 = 0.005, I2 = 65.56%). In Bangladesh, Lee and colleagues
(2022) observed a positive, small, and statistically significant effect of Education. Table 23 shows effect sizes by education subgroup. The
EJNG on the HDDS (ĝ = 0.18; 95% CI: 0.05 to 0.31; p = 0.01). effects, reported by Li and colleagues (2021), are statistically significant
None of the exploratory moderator variables significantly across lower‐ and higher‐educated household heads. The magnitude of
contributed to the variation in estimates. Of the five studies, two the effect is greater for households in which the head‐of‐household
effect sizes remain after excluding high risk of bias estimates. The had less than or equivalent to nine years of formal education.
effect of market interventions on nutrition outcomes becomes negative
ˆ = −0.07 ;
but is still very small and not statistically significant (SMD Gender. Table 24 shows effect sizes by gender subgroups from
95% CI: −0.15 to 0.009; p = 0.08). Shimizutani and colleagues (2021) (SHEP) and Wang (2021) (REDC).
REDC supported the expansion of e‐commerce in rural areas with the
Subgroup analysis. We now examine how the effect sizes of market establishment of county town‐village delivery systems and e‐
interventions may vary by subgroup. We here compare estimates within commerce service terminals in China. It reduced the likelihood of
studies from the same intervention across subgroups. We do not report working in the agricultural sector and increased the likelihood of
here the effect sizes across subgroups, which are consistently working in the wage sector for households with male and female
statistically insignificant. We divide this analysis into types of heads; the magnitude of the effects was similar across these two
subgroups: wealth, land size, education, and gender of beneficiaries. outcomes and the gender of household heads.

Wealth. Table 21 shows effect sizes by wealth subgroup. The effects

reported by Garbero and Songsermsawas (2018) do not consistently 18
A mu is a Chinese unit of land measurement. One mu equates to approximately 666.5
vary by wealth levels across outcomes here. square meters.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 107 of 191

T A B L E 21 Effect sizes across wealth subgroups (from Garbero & Songsermsawas, 2018).
Outcome subcategory Outcome description Subgroup information ĝ p Value

Food security and nutrition Dietary diversity score Whole sample −0.08 0.07

Poor county 0.08 0.26

Vulnerable county −0.17 0.01

Yields Grain yield (kg/ha, log) Whole sample −0.14 <0.001

Poor county −0.03 0.74

Vulnerable county −0.20 <0.001

Use/adoption of improved inputs, Pesticide use per ha (kg/ha, log) Whole sample 0.03 0.56
technologies, and practices
Poor county 0.16 0.03

Vulnerable county −0.10 0.10

Volume Value of grain production (Chinese Whole sample −0.16 <0.001

yuan, log)
Poor county −0.08 0.30

Vulnerable county −0.21 <0.001

Note: Subgroup estimates in bold have a statistically significant effect size (ĝ ) at a 5% significance level.
Abbreviations: ha, hectare; kg, kilogram.

T A B L E 22 Effect sizes across land‐size subgroups (from Cavatassi et al., 2011; Fischer & Qaim, 2012; Li et al., 2021).
Study Outcome subcategory Outcome description Subgroup information ĝ p Value

Cavatassi Farm income Gross margin ($/ha) Whole sample 0.21 0.01
et al. (2011)
Households with large farms (>5 ha) 0.39 0.07

Households with medium farms 0.36 0.003

(between 1–5 ha)

Households with small farms (<1 ha) −0.02 0.85

Farm investment Input cost ($/ha) Whole sample 0.15 0.06

Households with large farms (>5 ha) −0.27 0.20

Households with medium farms 0.13 0.30

(between 1–5 ha)

Households with small farms (<1 ha) 0.36 0.002

Yields Log of total harvest (kg/ha) Whole sample 0.27 0.01

Households with large farms (>5 ha) 0.05 0.81

Households with medium farms 0.35 0.01

(between 1–5 ha)

Households with small farms (<1 ha) 0.33 0.004

Fischer and Yields Yield (kg/acre) Whole sample −0.94 0.001

Qaim (2012)
Households with 0.1–1‐acre landholding −0.16 0.63

Households with 1–3‐acre landholding −0.63 0.01

Households with more than 3 acres of −0.16 0.63

landholding

Li et al. (2021) Total income and Net income Households which cultivated land 0.24 <0.001
wealth (10,000 yuan/capita) size ≤ 4.34 mu

Households which cultivated land 0.31 <0.001

size > 4.34 mu

Note: Estimates in bold have a statistically significant effect size (ĝ ) at a 5% significance level.
Abbreviations: ha, hectare; kg, kilogram.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
108 of 191 | MARION ET AL.

T A B L E 23 Effect sizes across education subgroups (from Li et al., 2021, ECA).

Outcome
subcategory Outcome description Subgroup information ĝ p Value

Total income and Net income Whole sample of households 0.41 <0.001
wealth (Yuan10,000/capita)
Household heads with less than or 0.36 <0.001
equal to nine years of education

Household heads with more than nine 0.18 <0.001

years of education

Note: Subgroup estimates in bold have a statistically significant effect size (ĝ ) at a 5% significance level.

T A B L E 24 Effect sizes across gender subgroups (from Shimizutani et al., 2021; Wang, 2021).

Study Outcome subcategory Outcome description Subgroup information ĝ p Value

Shimizutani Sales Sales of horticultural crops (KSh) Female‐headed 0.23 0.04

et al. (2021) households

Male headed 0.08 0.14

households

Wang (2021) Labor market outcome Binary variable indicating whether the individual Female −0.09 0.01
(off‐farm) was working in the agriculture sector
Male −0.12 0.001

Labor market outcome Binary variable indicating whether the individual Female 0.12 0.001
(non‐farm) was working in the wage sector
Male 0.08 0.02

Note: Estimates in bold have a statistically significant effect size (ĝ ) at a 5% significance level.

T A B L E 25 Effect sizes across periods of exposure (from Smale et al., 2012).

Outcome Period of exposure
subcategory Outcome description in months ĝ p Value

Total income and Income 35 0.18 0.09

wealth
60 0.12 0.25

82 0.23 0.03

Farm income Value of horticultural sales from main and short 35 −0.08 0.46
seasons valued at the district median price of
60 −0.004 0.97
each crop for the respective district
82 0.27 0.01

Note: Estimates in bold have a statistically significant effect size (ĝ ) at a 5% significance level.

Shimizutani and colleagues (2021) evaluated the effect of SHEP There is available data from Smale and colleagues (2012) on the
on sales of horticultural crops; only female‐headed households effect of KHDP on households' total income and the value of
experienced a statistically significant effect. This intervention horticultural sales, both in Kenyan Shillings. We notice that effect
combined training and market connections for farmers in Kenya. sizes increase and become statistically significant as the period of
exposure to the intervention increases. The project reported its
Analysis of effects over time. We aimed to reproduce this same type largest effects in 2010 after 82 months of project exposure.
of analysis but for different periods of exposure to the interventions.
We here compare estimates within studies from the same interven- Barriers and facilitators' analysis
tion across time periods. As above, we only focus on the effect sizes This section discusses the reported facilitators and barriers to the
where the period of exposure to the intervention changes, with interventions' effects within local contexts. This presentation is
everything else held constant. We do not report here the effect sizes subject to an important caveat: the included studies are quantitative
across the period of exposure, which are consistently statistically impact evaluations. Such studies tend to focus narrowly on
insignificant (Table 25). identifying the magnitude of the intervention's effects, without
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 109 of 191

discussing implementation details or rigorously explaining why an travel time from the villages to the market were barriers to
intervention was successful (or not). A thorough analysis of barriers participation. Zant (2018) also highlighted the incentives driven by
and facilitators would require a dedicated search for different types distance to the market in Malawi: after the establishment of auction
of evidence. Of the 94 papers in this body of evidence, our analysis of floors for tobacco, the author observed that a 10‐km reduction in
barriers and facilitators identified 34 papers with relevant distance to the market led to an increase in production.
information. On the other hand, poor infrastructure and networks can hinder
access to and participation in the market or may require farmers to
Access to information may be an important driver of farmers' decisions adopt alternative market strategies. For example, in the REDC project
and actions around entering markets. Access to information on market in China, Wang (2021) observed that remote areas were more prone
prices or agricultural outputs may play a key role in farmers' decisions to use e‐commerce options than the physical market. However, this is
to participate in the market and in their selling approaches. In that only made possible by good internet access. This outcome was also
sense, authors highlighted the benefits of phone ownership or access observed by Nkonya and colleagues (2013) in DRC, Rwanda, and
to market information as drivers of success for their new market- Uganda, where interventions implemented innovation platforms to
places or alternative marketing interventions. In India, Rajkhowa strengthen market linkages; however, a slow broadband network
(2021) analyzed the effect of FARMEX, a digital agri‐tech platform, affected farmers' participation.
and observed a positive effect of phone ownership on off‐farm In Malawi, Savani and Stewart (2019) identified weak and
employment allowing farmers to access digital information and inadequate infrastructure on the river Islands as barriers to accessing
cultivate additional crops. Similarly, in Kenya, Fischer and Qaim the market on the mainland. In a similar intervention in Tajikistan,
(2012) commented that farmers with mobile phones received more Morgan and Stewart (2019, AMP Tajikistan) commented that poor
efficient communication, as they were more easily contacted and market infrastructure in local areas hindered producers from
notified to attend market days and other activities in their self‐ exporting their outputs to the market.
sustaining groups. In Malawi, Maertens and colleagues (2020)
highlighted that despite low uptake of the output marketing program, Incentives to participate and sell quality products in an enabling
some farmers were aware of price evolution and managed to sell environment to sustain new markets. A new or alternative market is
their products at higher prices. only as good as its use by farmers and the quality of the traded
On the other hand, the lack of information, or perception that the products. It requires take‐up and buy‐in from local stakeholders (both
information is not reliable, may hinder intervention effectiveness. In farmers and traders) to drive business and generate wealth. In India,
the ECX, a membership‐based commodity trading platform, farmers Reddy (2018) analyzed eNam (a pan‐India electronic trading
who did not receive reference price information about ECX‐traded platform), and observed barriers including fear among farmers
crops produced less of these commodities and benefitted less from regarding the recovery of their loans and the number of farmers
this initiative (Belay & Ayelew, 2020). The authors noted that without e‐market bank accounts.
reducing the distance to the nearest plasma display screen where Similarly, the Anchor Farm Model in Malawi, which supported
information was provided increased program benefits. Similarly, in access to structure output markets, suffered from a low uptake that
the Mozambique IAP, which promoted sustainable market systems, minimized the benefits of the intervention (Maertens et al., 2020). In
some farmers did not make any changes to their selling approaches Colombia, AGP used mobile phone technology to aggregate demand
even after receiving information, as they perceived it as wrong or of and promote market linkages. However, take‐up was low, and several
low quality (Ghebru et al., 2019). firms were not interested and thus did not market the service
(Iacovone & McKenzie, 2019).
Geographical and infrastructural accessibility are important considera- For stakeholders participating in the market, the quality of the
tions for farmers' participation in markets. Geographical proximity to a products they trade is a key criterion to making the market attractive
market can be a facilitator for farmers' participation in new markets and sustainable. In India, the Unified Market Platform integrated all
or alternative marketing approaches, thereby allowing them to derive wholesale transactions in the state of Karnataka. Farmers who
more benefits from the intervention. In China, Couture and produced higher‐quality products benefitted more from the inter-
colleagues (2021) analyzed the effect of an e‐commerce intervention vention (Levi et al., 2020). In Colombia, Fuller (2015) observed a
(ECRV) and observed that the population living in closer proximity to similar phenomenon in an intervention (FM) establishing new
the program's terminal was more likely to use the new e‐commerce farmers' markets in Bogotá, hypothesizing that higher prices among
terminal shopping option. They also highlighted that villages with pre‐ participants might be a result of increased product quality.
existing parcel delivery service connections had a higher intervention Finally, market conditions can turn into an enabling or challeng-
uptake. ing factor for intervention implementation. The market must be well‐
In a similar intervention in China, Liu and colleagues (2021) managed and logistically sustainable for farmers and traders to
observed the same phenomenon. In Bangladesh, the gendered maximize their benefits. In India, Reddy (2018) highlighted the
enterprise and markets approach supported market linkages: Savani challenges faced by eNAM and its electronic trading platform, as
and Stewart (2019, AMP Bangladesh) commented that distance and market arrivals were not increasing and the program faced resistance
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
110 of 191 | MARION ET AL.

from traders and agents. Similarly in Armenia, Caeyers and Nairi market channels in Bogotá and organizing bi‐weekly farmer markets
Quinn (2015) observed that in the ARMA66 project (which facilitated led to larger benefits for richer farmers due to their increased market
access to microfinance and organized harvest festivals), farmers faced abilities (Romagnoli et al., 2018). Finally, in Malawi, Maertens et al.
difficulties in sourcing inputs for the local market, such as quality (2020) analyzed the Anchor Farm Model, which provided access to
seeds and seedlings, fertilizers, and pesticides. structured output markets. They observed that farmers' income and
productivity increased among households with higher baseline soil
Knowledge and experience with market tools and resources can be a fertility compared to households with lower baseline soil fertility.
driver of intervention participation and success. Farmers come to the
market with different backgrounds and experiences in trading and Intervention costs
markets. This discrepancy in experience can hinder intervention Five of the 67 studies included in this body of evidence provide
impact for those with limited knowledge of the practices and information on intervention cost (8%) (Table 26). This is in line with the
resources they can use, or it can be a catalyst of change for those general trend in impact evaluation literature, where less than 20% of
who know how to navigate the process. In Azerbaijan, Lombardini impact evaluations report any cost information—although the propor-
(2017) observed that women in the intervention group had less tion here is much lower (Brown & Tanner, 2019). Little information is
knowledge of loan markets than men, which reduced the likelihood of provided in the studies about the cost of interventions or specific costs
their access to credit and diminished the benefits they received from incurred by the establishment of new markets or alternative markets.
established cooperatives. Most available information reported the cost per beneficiary.
Similarly, in India, the pan‐India electronic trading platform
revealed a discrepancy of knowledge between farmers and traders; Discussion
several farmers did not have knowledge of the bidding process and
suffered from unbalanced awareness of the market processes Summary of findings. The body of evidence of studies that quantita-
compared to traders (Reddy & Mehjabeen, 2019). The same tively analyze the effect of initiatives creating new marketplaces and
observations are true for e‐commerce: in China, the REDC project alternative marketing opportunities on agricultural outputs and market
showed a discrepancy between older and younger participants outcomes is relatively large compared to the other intervention
regarding their likelihood of using e‐commerce due to their categories of our systematic review. Evidence on the topic includes
technological capabilities (Wang, 2021). In Tanzania, Bayiyana and 94 studies reporting on 61 unique interventions in 28 countries (with
colleagues (2018) saw farmers' higher education levels as a driver of 27 studies focusing on Sub‐Saharan Africa). The analysis relied on
income, as they could use their knowledge and resources efficiently quantitative data from 67 studies, as 27 papers were excluded since
to maximize the benefits from the dairy market hub. they are ongoing studies, or unpublished versions of already included
This knowledge and experience might explain the additional studies.
benefits accessed by larger landowners and richer farmers. Ashraf The body of evidence includes 37 studies focusing on Interven-
and colleagues (2009) analyzed the effect of DrumNet in Kenya and tions combining market interventions with other market components
observed that prior growers of export crops benefitted from reduced or with other intervention categories. Most studies focused on the
marketing costs through the intervention compared to first‐time effect of arranged or curated offers from buyers (n = 32) with exposure
growers. Similarly, in the MC project in Colombia, establishing direct periods from three months to 34 years. Included studies mostly used

T A B L E 26 Summary of available intervention costs.

Study Acronym Cost information description

Falcao and MPBMAC Total cost to run the platform for three years: USD927,190 including USD560,112 for management,
McIntosh (2021) USD168,105 for targeting and promotion, USD39,784 for training, USD53,648 for material, and
USD46,757 for monitoring

Ashraf et al. (2009) DMH The farmer's commitment fee to participate in the transaction insurance fund was USD10 (equivalent to
a laborer's weekly wage)
Monthly cost for DrumNet: USD1200

Goyal (2010) E‐CHOUPAL The program saved Rs12.9 million in the first year of operation through better quality oil and de‐oiled
cake obtained from processing soybeans procured through the e‐Choupal intervention.

Maertens et al. (2020) AFP Cost of hub farmer's time: USD120/club or USD6/farmer
Cost of government extension agent: USD187–250/monthCost of farmer field day: USD650 or
USD3.25/farmer
Cost of demonstration plot: USD280 or USD14/farmer

Kafle et al. (2022) HVAP Total cost: USD17,083,470

Cost per producer: USD158/capita
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 111 of 191

QEDs (n = 57) over follow‐up periods of up to five years. A small interventions or run additional moderator analyses (e.g., crop losses,
minority of studies provided information on costs (n = 5). livestock type, off‐farm and non‐farm labor market outcomes). These
Overall, we found positive, very small to moderate, and often outcomes represent a clear gap in the evidence base for market
statistically significant effects of market interventions on outcomes interventions, and a critical mass of primary research in these areas
included in the theory of change of this intervention. None of the will be necessary before a robust synthesis can be undertaken.
meta‐analyses revealed significant negative effects of the interven- When we explored potential sources of heterogeneity, we
tions, while 11 of the 27 included outcomes showed positive and found some effects of the types of analysis and design of the
statistically significant results. intervention: multi‐component interventions showed smaller ef-
Market interventions produce positive and statistically significant fects on yields; non‐staggered interventions showed larger effects
effects across the causal chain of agricultural, social, and economic on volume but smaller effects on non‐farm income; and studies with
outcomes, particularly in terms of immediate outcomes. The meta‐ QEDs reported larger effects on farm income. However, time‐
analysis showed that the interventions led to positive, large, and related effects are more nuanced: although longer exposure
statistically significant effects on group participation outcomes periods showed larger effects on farm investments (fertilizers) and
ˆ = 0.92; 95% CI: 0.43, 1.41; k = 4 ). Despite the smaller body
(SMD total income and wealth, studies with longer follow‐up periods
of evidence available on this outcome type, the findings are in line reported larger effects on the use/adoption of improved inputs, but
with the theory of change for this intervention, and corroborate the smaller effects on credit.
potential effect of market interventions on the connections between Similarly, more recent interventions showed smaller effects on
market partners (both between farmers and buyers and between transaction costs and farm investment (seeds) and larger effects on
farmers themselves). group participation. Additional analyses, such as subgroup analysis
We also observe a positive, small to moderate, and statistically and findings related to intervention facilitators, emphasize the
significant effect of the interventions on use and adoption of other potential influence of knowledge and education levels of farmers
improved inputs (SMD ˆ = 0.22; 95% CI: 0.10, 0.35; k = 12), total accessing the markets, but also their physical and economic access
farm investment (SMD ˆ = 0.24; 95% CI : 0.05, 0.42; k = 6), farm to the market (e.g., land ownership, distance to the market, access
ˆ = 0.34; 95% CI : 0.08, 0.60; k = 3),
investments in livestock (SMD to information).
ˆ = 0.17; 95% CI : 0.04, 0.31; k = 4 ),
farm investment in labor (SMD The body of evidence reveals a gap in the reporting of cost
ˆ = 0.13; 95% CI: 0.05, 0.20; k = 24).
and prices (SMD information, with less than 10% of studies including any cost
With regard to agricultural production outcomes, we found a information. Among studies providing information on cost, most
positive, moderate, and statistically significant effect of market focused on the reporting of the total or component‐specific budget
ˆ = 0.16; 95% CI: 0.04, 0.27; k = 16).
interventions on yields (SMD of the interventions.
Further down the causal chain, we also observed a positive, moderate,
and statistically significant effect of market interventions on intermediate Implications for policymakers, implementers.
outcomes: sales (SMD ˆ = 0.12; 95% CI: 0.03, 0.20; k = 23) and farm • Market interventions are effective in improving agricultural,
ˆ = 0.15; 95% CI: 0.10, 0.20; k = 27).
income (SMD socioeconomic, food security, and nutrition outcomes. Although
These interventions potentially decreased search costs, barri- most of the effects are small to moderate, market interventions
ers to market entry, and improved market competition. Those have statistically significant positive effects under all outcome
benefits do not only include financial aspects (e.g., income) but also categories. In cases where effects are very small, consider
resources (e.g., group participation, investments, or adoption of new whether the increases will be practically significant relative to
resources and practices) and agricultural opportunities (e.g., volume, program costs.
sales, prices). • To maximize the effectiveness of market interventions, policy-
We also found very small, positive, but statistically makers should consider both the technical accessibility (knowl-
significant effect of the interventions on total income and edge and education of targeted farmers) and the physical
ˆ = 0.10; 95% CI: 0.05, 0.15; k = 29) and food security
wealth (SMD accessibility (transport, infrastructure, and distance) of the newly
ˆ
(SMD = 0.10; 95% CI: 0.02, 0.17; k = 17). This smaller effect, also established market.
observed on food security outcomes, might be explained by their • To maximize the value for money of implemented interventions,
position at the end of the pathway to change, as they are a policymakers should invest in collecting evidence on the cost of
consequence of the diverse range of outcomes included in the market interventions and ensuring such cost are reported.
market interventions theory of change (e.g., transaction costs, yields,
volume, sales, prices, farm income, etc.). However, the significant Implications for researchers.
and positive effect corroborates the potential of market interven- • More evidence is needed on the establishment of alternative market-
tions in affecting the full causal chain of agricultural, social, and places or online commodity exchanges, especially considering the
economic outcomes increasing number of phone and internet users in LMICs. Similarly,
Despite the overall promising results of the meta‐analysis, research should consider expanding the body of evidence outside of
several outcomes lack data to robustly measure the effect of Africa, as this region is currently covered by half of included studies.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
112 of 191 | MARION ET AL.

• The body of evidence of market interventions includes the second (Dorward, 2001).19 First, contracts reduce the search costs for a
highest prevalence of high‐risk bias studies among the interven- buyer as they set agreements between farmers and a single trader
tion types in this systematic review, mostly due to confounding, whereas farmers often sell to multiple actors. Second, contracts can
spillovers, and risk of selection bias. To maximize the accuracy of also facilitate marketing aspects such as the delivery of the
future studies, researchers should consider these risks in their commodity and obtaining payment (Gray & Boehlje, 2005). These
evaluation design. lower transaction costs can permit farmers to access new markets
• Expanding follow‐up periods and/or including a cost‐benefit (Masakure & Henson, 2005; Guo & Jolly, 2009).
analysis of market interventions are both important. Unlike in Contract farming can also increase the price farmers receive for
other chapters, more than half of included studies had a follow‐up their produce (Minot, 2007). These interventions improve farmers'
period or less than four years. Information on intervention costs is knowledge of price and reduce information asymmetries across
only available for five of the 61 included interventions. contracting parties. Contract farming entails a negotiation process
well in advance of the sale between farmers and buyer which
should permit farmers to receive better prices than on the spot
5.3.4 | Chapter 4: Contract farming interventions market. This effect can apply to all contract farming types but
mostly in the case of fixed‐price and price guarantee contracts
How can contract farming interventions affect agricultural, socio‐ where farmers and buyers agree in advance the price of the
economic, food security, and nutrition outcomes? contracted produce.
Contract farming can be defined as forward‐looking, vertical market The other pathways of short‐term effects on quantity
arrangements (linked to volume, value, quality, and price, in produced, sales quantity, and product quality depend on the
advance of or at the time of planting) where traders/buyers may attributes of the contract. Production‐management contracts and
provide inputs, credit, and/or extension services to farmers as a input or credit‐supply contracts provide agricultural extensions
part of an agreement to purchase farmers' agricultural produce leading to new skills and additional inputs and production credit,
(Little & Watts, 1994; Eaton & Shepherd, 2001; Prowse, 2012). respectively. Production‐management contracts often set require-
Contract farming also includes out‐grower schemes, which refer to ments in terms of quality of the produce. Contract farming
similar arrangements involving public enterprises, parastatals, arrangements may also allow farmers to diversify into new crops
government agencies, or NGOs, rather than private entities and income sources. These arrangements may also introduce
(Bellemare & Bloem, 2018). contracted farmers to new and better agricultural practices leading
This chapter focuses on five types of contract farming: to improvements in productivity, and quality, and therefore higher
revenues and income (Glover, 1984).
• Fixed‐price and price guarantee contracts wherein the buyer In the long‐term, across all types of contract farming,
provides farmers with a guaranteed price (or a range of prices) in beneficiary farmers may face lower uncertainty and risks as
advance of the transaction for the delivery of agricultural produce. contracts provide farmers with advanced knowledge on prices,
• Production‐management contracts wherein the buyer provides quantity, and quality of expected agricultural production (Bellemare
extension services or other technical support for the delivery of & Bloem, 2018; Oya, 2012; Minot, 2007; Bellemare & Novak,
agricultural produce. 2017). In the absence of contract farming, faced with many
• Input or credit‐supply contracts wherein the buyer provides inputs different risks, farmers employ coping mechanisms to reduce their
or credit upfront and deducts the cost at harvest. exposure to risk which limit investments and maintain them in
• Other contract types are contracts which cannot be categorized in poverty (Rosenzweig & Binswanger, 1993; Jalan & Ravaillon, 2001;
the three types of contracts above. These contracts may comprise Elbers et al., 2007). Farmers in contracts arrangements were more
a range of requirements on the production (processes, crop likely to invest and adopt new labor and modern inputs practices
varieties, timing of harvest, etc.) and also on the marketing side than farmers without access to these arrangements (Bellemare &
(time of transaction, quantity, quality standards and sorting, price Bloem, 2018; Bellemare & Lim, 2018).
or price band). Contract farming can therefore improve market access, lower
• Undefined contract farming where the characteristics of contract transaction costs, increase farmers' and output prices in the short‐
farming interventions are not known. term. Longer term effects are potential greater agricultural
production and, ultimately, households' welfare including greater
Figure 76 provides an overview of the Theory of Change (TOC) food security and improved nutrition. Contract farming can also
of contract farming on agricultural, socio‐economic, food security and improve food security and nutrition as it can provide more stable
nutrition outcomes. Approaches for contract farming interventions income across the year because of reduced uncertainty and risk
may vary and have distinct effects. We describe here the expected (Bellemare & Novak, 2017; Michelson et al., 2012). Reduction in
channel of effects and impacts of contract farming.
Transaction costs for farmers in contract farming are likely to be 19
Supporting Information: Appendices 1 and 2 for Chapter 4 provide additional information
lower relative to those trading with partners on spot markets about included studies and additional references, respectively.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 113 of 191

F I G U R E 76 Theory of change for contract farming interventions. The outcome types included in this systematic review are immediate
outcomes (green), agricultural production outcomes (purple), intermediate outcomes (teal), and welfare outcomes (light blue).

income fluctuations can limit seasonal fluctuations in food Smaller scale or less literate farmers may be less likely to engage
consumption affecting food security and nutrition (Christian & in contract farming and may be disadvantaged (Minot, 2007;
Dillon, 2016). Meemken & Bellemare, 2020). A systematic review on contract
Contract farming can, however, introduce new risks for farmers. farming by Ton and colleagues (2017) suggested that contract
They may face contractual default and possible indebtedness schemes may increase farmers' income but that the effects may be
(Narayanan, 2014). Contract farming engaging farmers in new crops concentrated among already wealthier farmers who are better placed
may also cause crop failures and income losses. New contracted at the outset to take advantage of contract farming.
crops may be unsuitable for the local production and marketing
context. Alternatively, farmers may not have the necessary skills to Description of included studies
cultivate a new crop. Similarly, contract farming may disrupt the We included 194 papers (including three ongoing studies: Arouna
existing farming system with the introduction of new inputs and et al., 2019; Casaburi & Wills, 2019; Udry & Karlan, 2021) examining
requirements. the effects of 129 contract farming interventions. The quantitative
Also, farmers may face difficulties when dealing with the traders. synthesis of findings relies on effects from 144 papers (125 main
Traders may also fail to satisfy contractual arrangements, may exploit studies and 19 linked papers), as 50 papers do not provide any
their monopolistic position, or may set unrealistic targets for the additional results. These 50 papers were ongoing studies, previous
farmers (Little & Watts 1994; Glover, 1990; Sivramkrishna and versions, or unpublished versions of already included studies.
Jyotishi 2008). Also, traders can exploit imperfect information and Nineteen linked papers report additional outcome data about the
farmers' limited resources. Disputes between traders and farmers in effects of an intervention (i.e., data on outcomes or prioritized
contractual relationships are rarely settled through the legal system outcome indicators that are not reported by the main study).
due to the high costs of bringing a dispute to court. These contracts
are therefore rarely legally enforced (Minot, 2007; Guo & Interventions, inputs, and activities
Jolly, 2009). We identified 129 contract farming interventions implemented in 36
Contract farming is based on negotiation between traders and countries (Figure 77). Interventions in Sub‐Saharan Africa represent
farmers and therefore entails additional costs linked to coordination, more than half of the included interventions (n = 73), followed by
aggregation and governance. The distribution of these costs across South Asia (n = 26, including one intervention that also took place in
farmers and the buyer vary across arrangements but may limit the Africa), East Asia and the Pacific (n = 23), Latin America and the
positive effects on farmers. On the other hand, contract farming can Caribbean (n = 6), and Europe and Central Asia (n = 1). Individual
facilitate the creation of horizontal coordination between farmers; countries with the largest number of included interventions are India
leading farmers to create institutionalized groups, associations, and (n = 15), Ghana (n = 13), Ethiopia (n = 11), China (n = 8), Tanzania
cooperatives (Rehber, 2007). (n = 9), Vietnam (n = 6), and Senegal (n = 5). Two studies analyzed the
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
114 of 191 | MARION ET AL.

F I G U R E 77 Distribution of interventions by region.

effects of a multi‐country intervention, the first implemented in Comparison

Bangladesh, Côte d'Ivoire, Mozambique, Nigeria, Tanzania, and In all 125 included studies, the comparison groups used for the
Uganda (Meemken & Bellemare, 2020), and the second in Ghana, estimations nearly exclusively focused on comparing outcomes for
Kenya, and Zambia (Dihel et al., 2018). beneficiaries of contract farming interventions to those of non‐
Supporting Information: Appendix 3 for Chapter 4 provides contracted farmers who had access to spot transaction markets
information on the 129 interventions: country, type of intervention, (96%). One study exploited a phased‐in (or pipeline) intervention
and year the intervention began. The most frequent type of approach to identify a comparison group (Wollni et al., 2013). Four
intervention evaluated is production‐management contracts (n = 47), studies used beneficiaries of other interventions as the control group.
followed by groups of interventions belonging to input or credit‐ For example, in Soullier and Moustier (2018), the control group was
supply contracts (n = 40), fixed‐price and price‐guarantee contracts comprised of producers who used the national agricultural bank loan
(n = 23), undefined contract farming (n = 22), and other contract types and sold through spot transactions. Supporting Information: Appen-
(n = 21) (Figure 78). dix 4 for Chapter 4 provides information on these studies such as the
Other contract types include similar forward‐looking market evaluated intervention, the intervention subcategory, the length of
contracts between traders and farmers which contain a combination follow‐up periods in years, the study design, and the type of
of fixed prices and price guarantees, production‐management comparison.
aspects, and/or input or credit‐supply elements. Undefined contract
farming covers other contracts for which the included studies Outcomes
provided insufficient information to classify them into one of these The analyzed studies cover all 17 outcomes included in our
four intervention subcategories. Ten interventions have activities framework specified in the protocol (Villar et al., 2023). The most
across two intervention subcategories, and seven interventions have frequently analyzed outcomes are farm income (n = 86), followed by
activities across three subcategories. yields (n = 46), total income and wealth (n = 43), use or adoption of
Table 27 provides an overview of the interventions included in improved inputs, technologies and practices (n = 20), farm investment
this review in terms of country, type of intervention, and year the (n = 20), and prices (n = 20) (Figure 79).
intervention began. Information on the intervention's first year is
often not specified (n = 50). Study design
Seven of the included interventions used a multi‐component For all 125 main studies, most of those in our analysis were based on
approach (Table 28). In the analysis below, we conduct a moderator QEDs (n = 119); the majority relied on statistical matching methods
analysis to determine whether the estimates from multi‐component (n = 55) and endogenous treatment effect methods (n = 30). Seven-
interventions are on average statistically significantly different than teen studies conducted an IV estimation. One study exploited a
estimates from single‐component interventions. natural experiment (Väth, 2014). Six studies used DID and FE
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 115 of 191

F I G U R E 78 Distribution of interventions by subcategory.

approaches. Ten studies estimated the effects using two study bias assessments. We found that 75% of included estimates were
designs. The remaining six studies used experimental designs. rated as having a high risk of bias, 23% were considered to have some
Figure 80 also shows the distribution of the length of studies' concerns related to the risk of bias, and the remaining 2% were
different follow‐up periods (study periods). This is the difference assessed as having a low risk of bias.
between the start date of the intervention and the most recent Figure 81 shows a consistency in the risk of bias in both QED
outcome data included in the study's analysis. It is not specified in 50 studies and RCT studies, since 75% of the former, and 73% of the
of the included studies. The remaining interventions began between latter, are rated as high risk. This risk of bias of included estimates
1950 and 2020 and had an average length of roughly 13.2 years. was based on six domains (selection bias; confounding; performance
Most studies with this information covered a period 10 years or bias; spillovers, crossovers, and contamination; outcome measure-
more after the intervention took place (n = 33). Thirteen studies ment bias; and reporting bias). Looking at individual risk of bias
evaluated the effects for less than two years; the remaining studies domains, the most common source of bias is related to selection (52%
evaluated the effects in a period of two to ten years after the of estimates were rated as high risk of bias) and confounding (64%).
intervention started (n = 28). The longest covered period was 63 years; Additional concerns included spillovers, crossovers, and contamina-
the authors indicated that this contract farming intervention in Tanzania tion (12% of estimates were rated as high risk of bias), reporting bias
(CFT) started in 1950 (Herrmann et al., 2018). The shortest covered (14%), and outcome measurement bias (2%). Estimates were not
period was one month (Adebisi et al., 2019; Lentz & Upton, 2016). likely to be at high risk for performance bias.
Relative to the other four chapters in this systematic review,
Synthesis of findings contract farming intervention studies had a higher percentage of
The data extraction of studies under the contract farming intervention estimates that were assessed as having a high risk of bias (75%). This
category identified a total of 1296 estimates, of which 289 have been percentage was 49%, 51%, 63%, and 46% for Chapters 1, 2, 3, and 5,
selected for the analysis of effects. We report additional information respectively. Included estimates for contract farming interventions
on the individual meta‐analyses (sensitivity analysis, outlier analysis, were more likely to have a high risk of selection bias and confounding
publication bias analysis, moderator analysis, and meta‐analyses effects than in the other chapters.
without high risk of bias estimates) in Supporting Information: Included studies often lacked precise information on the location
Appendix 5 for Chapter 4. of targeted beneficiaries in contract farming interventions. Contract
farming is an older and more widespread intervention in LMICs than
Risk of bias. The risk of bias of each estimate across the 125 included the other intervention types in this review. This made the
studies was assessed following the Risk of Bias Assessment Tool identification of an adequate counterfactual more complicated, and
(Supporting Information: Appendix 10 for background and methods significantly increased the risk of selection bias. This situation also
sections). Figure 81 presents a summary of the results of our risk of explains the greater use of methods such as statistical matching,
 T A B L E 27 Table of included contract farming interventions.

Intervention name Acronym Country (ISO3 code) Type First year of intervention
116 of 191

Adoption of a bio‐control product: Aflasafe ABCPA Senegal (SEN) Other contract types 2019
|

Aterna Baby Corn Production and Marketing Cooperative ABCPMCSL India (IND) Production‐management contracts 2009
Society Limited

Agribusiness contract farming in the Senegal River Delta ACFSRD Senegal (SEN) Input or credit‐supply contracts −

Agricultural commodity supply ACOS Ethiopia (ETH) Production‐management contracts 2014

Avocado Growers Association of Kenya AGA Kenya (KEN) Production‐management contracts 2009

Angkor Kasekam Roongroeung Co. Ltd. AKR Cambodia (KHM) Fixed‐price and price‐guarantee contracts; Production‐management 2001
contracts; Input or credit‐supply contracts

Akate, masara, and other contract farming schemes AMOCF Ghana (GHA) Other contract types; Production‐management contracts; Input or 2009 and 2011
credit‐supply contracts

Annapurna Organics Ltd. AOL Nepal (NPL) Other contract types 2001

A‐rachide de Bouche Program ARBP Senegal (SEN) Input or credit‐supply contracts 1990

Agricultural Value Chain Project, Oromia AVCPO Ethiopia (ETH) Storage deposit systems; Other contract types; Alternative physical 2011
marketplaces

Broiler contract farming in China BCF China (CHN) Other contract types −

Brown rice contract farming in India BRCF India (IND) Undefined contract farming −

The Bundibugyo Scheme in Uganda BS Uganda (UGA) Fixed‐price and price‐ guarantee contracts 2002

Brinjal contract farming in Bangladesh BrinjalCF Bangladesh (BGD) Production‐management contracts −

Contractual arrangements in the oil palm industry in CAOPI Indonesia (IDN) Input or credit‐supply contracts 1989
Indonesia

Contractual arrangements for potato in Peru CAP Peru (PER) Production‐management contracts −

Contract arrangement with tea farmers CATF Vietnam (VNM) Arranged or curated offers from buyers; Input or credit‐supply 1999
contracts

Coffee Certification in Ethiopia CC Ethiopia (ETH) Other contract types 2002

Cashew contract farming in Ghana CCF Ghana (GHA) Fixed‐price and price‐guarantee contracts; Production‐management −
contracts

Contract Farming in Peru CF Peru (PER) Other contract types 2001

Contract farming for apple and onion in China CFAO China (CHN) Production‐management contracts −

Contract farming arrangements in Punjab CFAP India (IND) Undefined contract farming −

Contract farming in China CFC China (CHN) Undefined contract farming −

MARION
ET AL.

18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
 TABLE 27 (Continued)
MARION

Intervention name Acronym Country (ISO3 code) Type First year of intervention
ET AL.

Contract farming in developing countries CFDC Bangladesh (BGD) Undefined contract farming −

Dairy contract farming in Ethiopia CFE Ethiopia (ETH) Undefined contract farming −

Contract farming in Ghana CFG Ghana (GHA) Undefined contract farming −

Contract farming in Madagascar CFM Madagascar (MDG) Fixed‐price and price‐ guarantee contracts −

Contract farming for malt barley in Ethiopia (production) CFMBEP Ethiopia (ETH) Production‐management contracts 2013

Contract farming for malt barley in Ethiopia (price and CFMBEPI Ethiopia (ETH) Fixed‐price and price‐guarantee contracts; Input or credit‐supply −
inputs) contracts

Contract farming in the Mekong River Delta CFMRD Vietnam (VNM) Production‐management contracts 2002

Contract farming in rice farming for environmentally CFRFESSP China (CHN) Undefined contract farming −
sustainable production in China

Contract farming in rice value chain CFRVC Ghana (GHA) Undefined contract farming −

Contract farming schemes in southern India CFS India (IND) Production‐management contracts 1990

Contract farming for small‐scale farmers CFSS South Africa (ZAF) Production‐management contracts −

Contract farming in Tanzania CFT Tanzania (TZA) Undefined contract farming 1950

Contract farming in Zimbabwe CFZ Zimbabwe (ZWE) Production‐management contracts; Input or credit‐supply contracts 2010

Contractual honey production in Ethiopia CHP Ethiopia (ETH) Production‐management contracts 2007

Cotton outgrower programs in Zambia COP Zambia (ZMB) Input or credit‐supply contracts 1997

Contracts for oil palm farming in Ghana COPF Ghana (GHA) Fixed‐price and price‐ guarantee contracts; Input or credit‐supply 1995 and 2007
contracts

Contracts with supermarkets in Kenya CS Kenya (KEN) Other contract types −

Cotton and tobacco contract farming in Mozambique CTCF Mozambique (MOZ) Production‐management contracts −

Cattle contract farming in Turkey CattleCF Turkey (TUR) Fixed‐price and price‐ guarantee contracts 2011

Coffee contract farming in Haiti CoffeeCF Haiti (HTI) Fixed‐price and price‐guarantee contracts; Production‐management 2001
contracts

Dairy contract farming in Bangladesh DCF Bangladesh (BGD) Production‐management contracts 1974

Dairy contract farming in Ethiopia DCFE Ethiopia (ETH) Production‐management contracts 1980

Dimma contract farming for Honey DCFH Ethiopia (ETH) Input or credit‐supply contracts −
|

Dairy contract farming in India DCFI India (IND) Production‐management contracts −

Dwangwa Cane Growers Limited DCGL Malawi (MWI) Production‐management contracts 2000

(Continues)
117 of 191

18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
 TABLE 27 (Continued)

Intervention name Acronym Country (ISO3 code) Type First year of intervention
| 118 of 191

Export supply chain in Senegal ESC Senegal (SEN) Production‐management contracts 1994

Entreprises de Services et Organisations de Producteur in ESOPB Benin (BEN) Production‐management contracts; Input or credit‐supply contracts; 2008
Benin Fixed‐price and price‐ guarantee contracts

Enterprise de Service et Organization des Producteur ESOPT Togo (TGO) Production‐management contracts 2004
in Togo

French bean contract farming in Tanzania FBCF Tanzania (TZA) Fixed‐price and price‐ guarantee contracts −

Field experiment in Vietnam FECF Vietnam (VNM) Production‐management contracts 2010

Farmers' professional cooperatives in China FPC China (CHN) Input or credit‐supply contracts; Arranged or curated offers from 2008
buyers

Ginger contract farming in Nepal GCF Nepal (NPL) Input or credit‐supply contracts; Production‐management contracts; 2010
Contract farming – Undefined contract

Gold Delta RSC contract farming scheme GDRCF Myanmar (MMR) Input or credit‐supply contracts 2009

GOPDC contract farming GOPDC Ghana (GHA) Production‐management contracts 1986

High‐yielding variety paddy seed and ginger in Nepal HYVPG Nepal (NPL) Production‐management contracts 1999

Interlocked contractual arrangements in Zambian dairy ICAZDS Zambia (ZMB) Production‐management contracts −
sector

Informal models of contract farming IMCF Ghana (GHA) Input or credit‐supply contracts −

IMPROVE contract farming IMPROVE Uganda (UGA) Production‐management contracts 2003

Kaleya Smallholders Company Ltd. Outgrower Scheme KASCOLOS Zambia (ZMB) Input or credit‐supply contracts; Production‐management contracts 1980

Kims Poultry Care Center contract farming KPCCCF Kenya (KEN) Input or credit‐supply contracts; Arranged or curated offers from 1996
buyers

Kilombero Sugar Company Limited KSCL Tanzania (TZA) Fixed‐price and price‐ guarantee contracts 1960 and 2008

Lentils contract farming in Nepal LCF Nepal (NPL) Undefined contract farming −

LT Firm in Vietnam LTF Vietnam (VNM) Production‐management contracts 2012

Malt barley contract farming in Ethiopia MBCF Ethiopia (ETH) Fixed‐price and price‐ guarantee contracts −

Malt barley contract farming in Arsi Zone MBCFA Ethiopia (ETH) Production‐management contracts −

Malt barley and sugarcane contract farming in Ethiopia MBSCF Ethiopia (ETH) Undefined contract 1975

Maize contract farming in Ghana MCF Ghana (GHA) Production‐management contracts −

Maize contract farming in Nigeria MCFN Nigeria (NGA) Input or credit‐supply contracts −
MARION

Nucleus Farmer‐Outgrower Scheme in Ghana NFOS Ghana (GHA) Input or credit‐supply contracts −
ET AL.

18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
 TABLE 27 (Continued)
MARION

Intervention name Acronym Country (ISO3 code) Type First year of intervention
ET AL.

Nestlé India Limited NIL India (IND) Production‐management contracts 1961

Onion contract farming in India OCF India (IND) Undefined contract farming −

Onion, okra, and pomegranate contract farming in India OOPCF India (IND) Undefined contract farming 1989, 1992 and 2001

Purchase for Progress in Tanzania P4P Tanzania (TZA) Arranged or curated offers from buyers, Production‐management 2009
contracts; Storage deposit systems

Poultry contracts in Bangladesh PCB Bangladesh (BGD) Input or credit‐supply contracts −

Poultry contracts in Ghana PCG Ghana (GHA) Other contract types −

Poultry contracts in India PCI India (IND) Production‐management contracts 1990

Poultry contracts in Nigeria PCN Nigeria (NGA) Undefined contract farming −

Contract arrangements in tea sector in Phu Tho PHCATS Vietnam (VNM) Other contract types 2002

Pioneer seed contracts in Indonesia PSC Indonesia (IDN) Fixed‐price and price‐ guarantee contracts; Input or credit‐supply 1986
contracts; Production‐management contracts

PT Pertani seed rice contracts PT seed rice (IDN) Indonesia (IDN) Production‐management contracts 1988

Pakistan Tobacco Company PTC Pakistan (PAK) Production‐management contracts 1959

Paprika contract farming in Tanzania PaprikaCF Tanzania (TZA) Input or credit‐supply contracts 2004

PepsiCo India PepsiCI India (IND) Fixed‐price and price‐ guarantee contracts 2001, 2006 and 2012

PepsiCo PepsiPotato Pakistan (PAK) Other contract types 1970

Pineapple contract farming in Costa Rica PineappleCF Costa Rica (CRI) Other contract types −

Paras Spices Pvt. Ltd. PSPL India (IND) Fixed‐price and price‐ guarantee contracts 2006

Rice contract farming in Nigeria RCF Nigeria (NGA) Other contract types 2013

Rice contract farming in Benin and other RCFBO Benin (BEN) Other contract types −

Rice contract farming in Benin price RCFBP Benin (BEN) Fixed‐price and price‐ guarantee contracts −

Rice contract farming in Côte d'Ivoire RCFCI Cote d'Ivoire (CIV) Production‐management contracts; Input or credit‐supply contracts 2013

Rice contract farming in Indonesia RCFI Indonesia (IDN) Undefined contract farming −

Rice contract farming with Lao Arrowny Corporation RCFLAC Laos (LAO) Fixed‐price and price‐ guarantee contracts; Input or credit‐supply 2002
contracts

Rice contract farming in Vietnam RCFV Vietnam (VNM) Input or credit‐supply contracts 2002
|

Maize Contract Farming by Rafhan RMPLPC maize Pakistan (PAK) Other contract types

(Continues)
119 of 191

18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
 TABLE 27 (Continued)

Intervention name Acronym Country (ISO3 code) Type First year of intervention
| 120 of 191

Rice Outgrower Scheme in Ghana ROS Ghana (GHA) Input or credit‐supply contracts −

Rice seed contract farming in India RSCF India (IND) Other contract types 2017

Rice seed contract farming in Punjab RSCFP India (IND) Input or credit‐supply contracts −

Rana Sugars Limited Ltd RSL India (IND) Fixed‐price and price‐ guarantee contracts 2012

Reasmey Stung Sen Agricultural Development Cooperative RSSADC Cambodia (KHM) Input or credit‐supply contracts 2003

Rice value chains in Senegal RVC Senegal (SEN) Input or credit‐supply contracts 2010 and 2011

Sunflower contract farming in Tanzania SCF Tanzania (TZA) Input or credit‐supply contracts 2007

Sustainable Quality Program SQP Colombia (COL) Production‐management contracts; Input or credit‐supply contracts 2007

Sipi Scheme by Kawacom (U) Ltd. SSKL Uganda (UGA) Production‐management contracts 2000

Sesame contract farming in Ethiopia SesaCF Ethiopia (ETH) Undefined contract farming −

Soybean contract farming in Ghana SoyabCF Ghana (GHA) Other contract types −

Tomato contract farming in Nepal TCFN Nepal (NPL) Fixed‐price and price‐ guarantee contracts; Input or credit‐supply 1992
contracts; Undefined contract farming

Tomato contract farming in Nigeria TCFNi Nigeria (NGA) Undefined contract farming −

Tobacco contract farming in the Philippines TCFP Philippines (PHL) Input or credit‐supply contracts 2011

The Technological Guarantee Program TGP Mexico (MEX) Fixed‐price and price‐ guarantee contracts 2013

Soybean contract farming model in Nigeria SoyabCFM Nigeria (NGA) Other contract types 2004 and 2008

Olam Nigeria Limited ONL Nigeria (NGA) Other contract types

Tobacco contract farming in Malawi TobCFMa Malawi (MWI) Input or credit‐supply contracts 2013

Tobacco contract farming in Malawi (wives on contracts) TobCFMaw Malawi (MWI) Production‐management contracts −

Tobacco contract farming in Mozambique TobCFMo Mozambique (MOZ) Input or credit‐supply contracts −

Tobacco contract farming in Zimbabwe TobCFZ Zimbabwe (ZWE) Undefined contract farming 2003

Union Nationale des Producteurs de Soja du Bénin UNPS Benin (BEN) Input or credit‐supply contracts 2020

Vegetable contract farming in Ethiopia VCF Ethiopia (ETH) Production‐management contracts −

Vegetable contract farming in China VCFC China (CHN) Production‐management contracts; Arranged or curated offers from 1990
buyers

Value chain participation in Ghana, Kenya, and Zambia VCP Ghana (GHA) Input or credit‐supply contracts −
MARION
ET AL.

18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 121 of 191

endogenous treatment effect methods, or IV methods than in other

First year of intervention

chapters.20
These methods require extensive data and knowledge of the

1975 and 2008

local context to confidently address selection bias and confounding
effects. The share of estimates based on an RCT is much lower in this
chapter (4%)21 than the others (20%, 14%, and 22% in Chapters 2, 3,

2007
and 5, respectively). RCTs can randomly assign the intervention,
Other contract types; Production‐management contracts; Fixed‐price −

−
thereby limiting selection bias and confounding effects. The chosen
IVs were often not exogenous, and authors used endline time‐variant

Fixed‐price and price‐ guarantee contracts; Input or credit‐supply

covariates in the statistical matching estimation; these both
Other contract types; Arranged or curated offers from buyers

contributed to a high risk of confounding effects of estimates in

included studies.

Summary of the meta‐analyses results of contract farming

interventions. Table 29 displays the meta‐analysis results across
Fixed‐price and price‐ guarantee contracts

outcome types covered in the subsequent sections.

Input or credit‐supply contracts

Immediate outcomes. The immediate outcomes category covers the

and price‐guarantee contracts

Undefined contract farming

first outcomes that may affect farmers across the causal chain. They
cover financial resources (e.g., transaction costs, credit), market
linkages and bargaining power (e.g., group participation, prices), and
Undefined contract

the adoption of new practices and agricultural approaches (e.g., use/

adoption of improved inputs, technologies, and practices, farm
contracts

investments).
Type

Use/adoption of fertilizers, pesticides, and herbicides. Within seven

Country (ISO3 code)

studies, we identified seven independent estimates of the effect of

contract farming on the use/adoption of fertilizers, pesticides, and
Tanzania (TZA)

Tanzania (TZA)

Tanzania (TZA)

Ethiopia (ETH)
Ghana (GHA)
China (CHN)

China (CHN)

herbicides outcomes. Studies of evaluated interventions took place in

two regions: Africa (n = 6), and Latin America and the Caribbean
(n = 1). No studies evaluated multi‐component interventions. Two
interventions provided fixed‐price and price‐guarantee contracts
(COPF and TGP), while two others provided production‐
management contracts (IMPROVE and RCFCI). Three interventions
also provided input or credit‐supply contracts (RVC, COPF, and
Acronym

WSSOS
WACF

WRCF
VML1
VESC

RCFCI). The remaining interventions provided other contract types

WPC
WCF

(AMOCF, ABCPA). RCFCI and COPF provided contracts that

combined both production‐management and input or credit‐supply
elements.
All studies here used measures of use/adoption of fertilizers,
pesticides, and herbicide outcomes at the household level (k = 3) and
Watermelon‐producing cooperative in China

the plot level (k = 4). For example, Deutschmann and colleagues

Vegetables export supply chain in Tanzania

Wonji‐Shoa Sugarcane Outgrower Scheme

Written apple contract farming in China

(2023) used a binary variable of whether households treated their

Written rice contract farming in Ghana
Wheat contract farming in Tanzania

fields with Alflasafe, which prevents aflatoxin‐causing fungi from

developing on crops. Corral and colleagues (2016) and Ruml and
Qaim (2020a) also conducted this analysis, using a binary variable
(Continued)

Vertical Market Linkages 1

indicating whether households used fertilizers during sowing, and

Intervention name
TABLE 27

20
A total of 86% of the studies with QEDs in this chapter used these methods, whereas in in
the other chapters, this percentage was much lower (e.g., only 42% and 51% in Chapters 2
and 3, respectively).
21
Eleven estimates out of 289.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
122 of 191 | MARION ET AL.

T A B L E 28 Descriptions of other interventions combined with contract farming interventions.

Intervention (country Description of other intervention components combined with contract farming
code)* Type interventions

AVCPO (ETH) Storage deposit systems; Other The AVCPO was developed as a comprehensive intervention program. In its initial
contract types; Alternative physical phase (2011–2013), it targeted 6520 smallholder farmers organized into 15
marketplaces cooperatives. The project included a range of initiatives such as training, providing
seeds, storage, processing, and marketing support through these cooperatives.The
interventions aimed to introduce or expand durum wheat cultivation and incorporated
technological innovations like high‐quality seed provision and improved agricultural
extension services. Additionally, the project sought to strengthen the role of
cooperatives in the agricultural value chain.

FPC (CHN) Input or credit‐supply contracts; FPCs in Jiangsu province, China, were established in 2008 and functioned through a
Arranged or curated offers from structured system. Members were required to sign written contracts at the beginning
buyers of the cultivation season, detailing rice cultivation and procurement procedures.
These contracts were binding for both the cooperative and its members.The
cooperatives provided a prescribed instruction handbook, dictating fertilizer and
pesticide use. The cooperative collectively purchased essential inputs, offered
technical advice, and provided advance payments to facilitate rice production.
Members adhering to the regulations could receive a 10% premium over the market
price, thereby fostering cooperative farming practices with economic incentives.

MCATS (VNM) Arranged or curated offers from In Moc Chau, Vietnam, there were two primary types of contract arrangements for tea
buyers; Input or credit‐supply farming: those between tea farmers and a state‐owned enterprise, which may have
contracts included a land lease component; and contracts between tea farmers and private firms
or cooperatives. Notably, most of these contractual arrangements in Moc Chau
included initial support for commencing tea production, including free seeds,
subsidization of inputs, and free extension services.

P4P (TZA) Arranged or curated offers from The Tanzanian P4P pilot (2009–2013) aimed to enhance the marketing and credit
buyers, Production‐management services provided to Tanzanian Savings and Credit Cooperatives (referred to as FOs).
contracts; Storage deposit systems This initiative focused on supply‐ and demand‐side interventions. Supply‐side actions
involved the rehabilitation of 23 warehouses, equipment provision, and various
training programs.Demand‐side interventions encompassed the procurement of maize
and beans by the World Food Programme. However, these purchases varied in
quantity and consistency across Fos, with some facing default issues due to payment
delays and market price fluctuations. On average, Fos with contracts received around
185 metric tons, but demand stimulus per FO member fluctuated from 0.18–0.77
metric tons. Not all members directly participated.

SQP (COL) Production‐management contracts; The SQP was implemented in several countries; its main distinguishing factor was the
Input or credit‐supply contracts buyer's commitment to purchase coffee from program farmers if it complied with
stringent quality standards. Additionally, the program emphasized quality and
environmental sustainability in coffee production. Its contractual arrangements
included provisions such as export‐gate price premiums, farm‐gate price premiums,
and a lump‐sum contribution to cover program implementation costs. These costs
encompassed training, extension services, plot renewal support, and farm inspections
to ensure compliance with program requirements.

VCFC (CHN) Production‐management contracts; In China, vegetable farmers who opted for marketing contracts used various contract
Arranged or curated offers from types, such as agreements with cooperatives, firms, middlemen, and wholesale
buyers markets. This included producers signing written or oral contracts with a single buyer
(often a marketing or processing company). These companies may have also provided
input to ensure product quality or reduce production costs. This form of coordination
was widespread in the Chinese vegetable sector.

VML1 (TZA) Undefined contract VML1 was a framework for agricultural commercialization that addresses various
market failures. It included contract farming and outgrower schemes—often involving
contractual arrangements between farmers and buyers, or full vertical integration
where a nucleus firm controls most supply chain stages.The intervention's aim was to
improve market access, reduce transaction costs, lower crop production risks,
decrease information asymmetries, and provide better access to inputs, credit, and
new technologies. It included cooperatives, Fos, and contract farming with employers,
focusing on traditional cooperative market channels that enable farmers' groups to
leverage market power and profit margins. This group also included interactions with
agri‐processing firms, either informally or under contract.

Note: ISO3 country code provided in parentheses (). If the introduction of an intervention is staggered, the year of intervention refers to the earliest year
that observations were undertaken for the intervention.
18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
|123 of 191

Distribution of outcome types.

F I G U R E 79
ET AL.
MARION
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
124 of 191 | MARION ET AL.

F I G U R E 80 Distribution in the length of study follow‐up periods.

whether households used chemical fertilizer on their plots, farming interventions on the use/adoption of fertilizers, pesticides, and
respectively. herbicides outcomes increases but remains statistically insignificant
Other outcome variables related to the quantity of nitrogen ˆ = 0.57 ; 95% CI: −0.91 to 2.05; p = 0.45).
(SMD
fertilizer used (Ragasa et al., 2018) and the quantity of chemical
fertilizer in kilogram per hectare (Soullier and Moustier 2018; Mano Use/adoption of improved varieties and quality seeds. Within five
et al., 2017). The last measure is the adoption of “good agricultural studies, we identified five independent estimates of the effect of
practices,” which include the use of fertilizers, pesticides, and contract farming on the use/adoption of improved varieties and quality
herbicides (Hansen and Rosenthal 2014). seeds outcomes. Studies of evaluated interventions took place in two
Contract‐farming interventions did not affect the use/adoption of regions: Africa (n = 3) and Latin America and the Caribbean (n = 2). No
fertilizers, pesticides, and herbicides. The overall weighted average studies evaluated multi‐component interventions. One intervention
ˆ = 0.48;
effect is large, positive, and statistically insignificant (SMD provided fixed‐price and price‐guarantee contracts (TGP), while
95% CI: −0.14 to 1.11; p = 0.13) (Figure 82). Effect sizes vary another provided production‐management contracts (RCFCI). Three
between −0.42 and 1.80 standard deviations. The heterogeneity interventions also provided input or credit‐supply contracts (RVC
among the studies is considerable (Q(6) = 275.54, p < 0.001, Production contr and RCFCI). The remaining interventions provided
τ̂ 2 = 0.69, I2 = 97.82%). other contract types (AMOCF, CF). RCFCI provided contracts that
Corral and colleagues (2016) found that TGP in Mexico produced combined both production‐management and input or credit‐supply
a large and statistically significant reduction in farmers' likelihood of elements.
use of fertilizers during sowing (ĝ = −0.42; 95% CI: −0.79 to Included studies used measures of use/adoption of improved
−0.05; p = 0.02). Authors reported that farmers targeted by TGP varieties and quality seeds outcomes at the household level (k = 2)
experienced a delay in the delivery of fertilizer, which explained this and at the plot level (k = 3). For example, Corral and colleagues
negative effect. Deutschmann and colleagues (2023) observed a (2016) and Ragasa and colleagues (2018) used a binary variable
positive, large, and statistically significant effect of ABCPA in Senegal indicating whether households used hybrid seeds on specific
on the adoption of Aflasafe (ĝ = 1.32; 95% CI: 1.11 to 1 plots.
.53; p < 0.001). In Ghana, Ragasa and colleagues (2018) observed a Contract‐farming interventions did not affect the use/adoption of
positive, large, and statistically significant effect of AMOCF on the improved varieties and quality seeds. The overall weighted average
quantity of nitrogen fertilizer used (ĝ = 1.80 ; 95% CI: 1.58 to ˆ = 1.00;
effect is large and positive, but statistically insignificant (SMD
2.02; p < 0.001). 95% CI: −0.37 to 2.38; p = 0.15) (Figure 83). Effect sizes vary
None of the moderator variables significantly contributed to the between −0.23 and 4.29 standard deviations. The heterogeneity
variation in estimates. Of the seven included effect sizes, two remain among the studies is considerable (Q(4) = 818.95, p < .001,
after excluding high risk of bias estimates. The effect of contract τ̂ 2 = 2.46, I2 = 99.51%).
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 125 of 191

F I G U R E 81 Risk of bias of included estimates between quasi‐experimental design (QED) (top) and randomized controlled trial (RCT)
(bottom).

Corral and colleagues (2016) found that TGP in Mexico resulted from the analysis, the overall effect is substantially reduced but
in a large and statistically significant increase in farmers' likelihood of ˆ = 0.15; 95% CI: −0.13 to
remains statistically insignificant (SMD
using hybrid seeds (ĝ = 0.82; 95% CI: 0.45 to 1.19; p < 0.001). In 0.44; p = 0.30).
Ghana, Ragasa and colleagues (2018) observed a large, positive, and None of the moderator variables significantly contributed to
statistically significant effect from AMOCF on the adoption of hybrid the variation in estimates. Of the five included effect sizes, one
seeds (ĝ = 4.30 ; 95% CI: 4.00 to 4.59; p < 0.001). We identify that remains after excluding high risk of bias estimates. This estimate is
this effect size is considered as an outlier according to the provided by Mano and colleagues (2017), which raised some
studentized residuals of the meta‐analysis. According to the model's concerns about the risk of bias. This estimate is very small,
Cook's distances, this estimate is overly influential in terms of its positive, and statistically insignificant (ĝ = 0.09; 95% CI: −0.17
weight. If the estimate from Ragasa and colleagues (2018) is removed to 0.35; p = 0.52).
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
126 of 191 | MARION ET AL.

T A B L E 29 Summary of meta‐analyses results.

Outcome types Outcome subcategory SMD 95% CI k p Value

Immediate outcomes Transaction costs N/A

Group participation N/A

Use/adoption of fertilizers, pesticides, and herbicides 0.48 −0.14, 1.11 7 0.13

Use/adoption of improved varieties and quality seeds 1.00 −0.38, 2.38 5 0.15

Use/adoption of labor and machinery N/A

Use/adoption of other improved inputs 0.27 0.07, 0.47 14 0.01

Farm investment (total) 0.25 −0.04, 0.54 12 0.09

Farm investment (fertilizers and pesticides) 0.11 −0.24, 0.47 3 0.53

Farm investment (labor) 0.03 −0.04, 0.1 8 0.40

Credit N/A

Prices 0.27 0.10, 0.45 19 <0.001

Agricultural production outcomes Yields 0.20 0.10, 0.29 43 <0.001

Volume 0.27 0.12, 0.41 5 0.001

Quality 0.42 0.13, 0.72 7 0.01

Produce type (diversity) 0.02 −0.06, 0.10 7 0.69

Intermediate outcomes Sales 0.34 0.18, 0.50 8 <0.001

Farm income 0.33 0.26, 0.40 84 <0.001

Off‐farm outcomes (agriculture only) Off‐farm income N/A

Non‐farm outcomes Non‐farm income 0.12 −0.06, 0.30 3 0.20

Labor market outcomes N/A

Welfare outcomes Total income and wealth 0.23 0.17, 0.30 39 <0.001

Food security 0.16 −0.01, 0.33 13 0.06

Food nutrition −0.07 −0.71, 0.57 3 0.82

Note: Outcome subcategories in bold have a statistically significant SMD at a 5% significance level. K is the number of estimates. N/A indicates that we did
not conduct a meta‐analysis for these outcome types. In these cases, k = 1 or k = 2 and we chose not to conduct a meta‐analysis in those cases due to
various differences (e.g., intervention types, regions).

F I G U R E 82 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of contract farming
interventions on the use/adoption of fertilizers, pesticides, and herbicides.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 127 of 191

F I G U R E 83 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of contract farming
interventions on the use/adoption of improved varieties and quality seeds outcomes.

F I G U R E 84 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of contract farming
interventions on the use/adoption of other improved inputs.

Use/adoption of other improved inputs. Within 14 studies, we Contract‐farming interventions increased the use/adoption of other
identified 14 independent estimates of the effect of contract farming improved inputs. The overall weighted average effect is moderate,
on the use/adoption of other improved input outcomes. Studies of ˆ = 0.27; 95% CI: 0.07 to
positive, and statistically significant (SMD
evaluated interventions took place in several regions: Africa (n = 7), 0.47; p = 0.01) (Figure 84). Effect sizes vary between −0.20 and 1.23
East Asia and the Pacific (n = 3), and South Asia (n = 4). One study standard deviations. The heterogeneity among the studies is
evaluated a multi‐component intervention (VCFC). One intervention considerable (Q(13) = 140.47, p < 0.001, τ̂ 2 = 0.127, I2 = 90.74 %).
provided fixed‐price and price‐guarantee contracts (CCF), while eight Dubbert and colleagues (2021) observed a negative, moderate,
others provided production‐management contracts (e.g., ABCPMCSL, and statistically significant effect of CCF on the adoption of
VCFC, FECF). Five interventions were based on credit‐supply sustainable farm practices in Ghana (ĝ = −0.20 ; 95% CI: 0.40 to
contracts (RVC, PCB, VCP, IMCF, and RCFCI). The remaining −0.003; p = 0.05). In the same country, Ragasa and colleagues (2018)
interventions provided other contract types (AMOCF, BCF). observed a positive, large, and statistically significant effect of
The outcome variables in these studies were at the household AMOCF on the adoption of improved practices (ĝ = 1.23; 95% CI:
level (k = 12) and the plot level (k = 2). They are all related to the use/ 1.02 to 1.44; p < 0.001). Although both interventions took place in
adoption of improved agricultural practices. For example, Abdul‐ Ghana, they were implemented in different regions and the authors
Rahman and colleagues (2019) used a discrete variable ranging from used different measures of improved inputs. In the study by Ragasa
0–4 illustrating the number of improved farm technologies adopted and colleagues (2018), the improved practices were plant spacing,
by the household. In Liang and colleagues (2020), the outcome seeding rate, planting method, and row planting, whereas in Dubbert
variable is a binary variable indicating whether the household has a and colleagues (2021), sustainable farm practices included leveling
certificate demonstrating that the producer has pollution‐free, green, terrain, mulching, catch pits, and reverse terraces.
or organic vegetable production. Roy and colleagues (2021) exploited Effects based on average treatment effect estimates are on
data on the Food Safety Adoption Index measuring households' average 0.45 standard deviation units smaller than average treatment
adoption of food safety measures. effects on the treated estimates. We did not identify any funnel plot
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
128 of 191 | MARION ET AL.

asymmetry, meaning that publication bias is unlikely to be present. end of the spectrum, in China, Ren and colleagues (2021) found a
Of the 14 included effect sizes, five remain after excluding high positive, large, and statistically significant increase in total cost of rice
risk of bias estimates. The effect of contract farming interventions production per mu (ĝ = 1.00 ; 95% CI: 0.83 to 1.16; p < 0.001).
on improved seeds outcomesbecomes smaller and is no Authors reported that farmers invested more in sustainable produc-
ˆ = 0.12; 95% CI: −0.14 to
longer statistically significant (SMD tion practices. In Ethiopia, Tefera and Bijman (2021) observed a
0.39; p = 0.36). positive, large, and statistically significant effect of CFMBEP on the
total cost of malt barley production in Ethiopian birr/ha (ĝ = 1.41;
Farm investment (total). Within 12 studies, we identified 12 indepen- 95% CI: 1.16 to 1.67; p < 0.001).
dent estimates of the effect of contract farming interventions on total None of the moderator variables significantly contributed to the
farm investment outcomes. Studies of evaluated interventions took variation in estimates. We did not identify any funnel plot asymmetry,
place in several regions: Africa (n = 5), East Asia and the Pacific (n = 3), meaning that publication bias is unlikely to be present. Of the twelve
and South Asia (n = 4). There were no multi‐component interventions. included effect sizes, two remain after excluding high risk of bias
Three interventions provided fixed‐price and price‐guarantee estimates. The effect of contract farming interventions on total farm
contracts (RCFLAC, TCFN, MBCF), while three others were based investment outcomes increases but remains statistically insignificant
on production‐management contracts (HYVPG, CFMBEP, ACOS). ˆ = 0.69; 95% CI: −0.72 to 2.11; p = 0.34 ). However, given this
(SMD
The remaining interventions were input or credit‐supply contracts was based on a very small number of studies, this result should be
(RCFLAC, RVC, RCFV), other contract types (PCG, RMPLPC maize), interpreted with caution.
and undefined contract farming (CFRFESSP, and LCF). RCFLAC
combined elements from several contract types. Farm investment (fertilizers and pesticides). Within three studies, we
All studies here used measures of total farm investment at the identified three independent estimates of the effect of contract
household level (k = 12). For example, Mishra and colleagues (2018c) farming interventions on farm investments in fertilizers and pesticides
used households' total investments in agriculture per quintal in Nepal. outcomes. Evaluated interventions took place in three regions: Africa
In Mishra and colleagues (2018d), the outcome variable was a similar (n = 1), East Asia and the Pacific (n = 1), and South Asia (n = 1). There
measure per hectare, while Bannor and Gyekye (2022) relied on total were no multi‐component interventions. One intervention provided
farm expenditure in Ghana. production‐management contracts (ABCPMCSL) while CFRFESSP was
Contract farming interventions did not affect total farm investment. based on undefined contract farming in China. ESOPB combined
The overall weighted average effect is moderate and positive elements from several contract types.
ˆ = 0.25; 95% CI: −0.04 to
but statistically insignificant (SMD ESOPB was a contract farming intervention in Benin for rice
0.54; p = 0.09) (Figure 85). Effect sizes vary between −0.24 and producers that combined elements from fixed‐price and price‐
1.41 standard deviations. There is considerable heterogeneity among guarantee contracts, input or credit‐supply contracts, and other contract
the studies (Q(11) = 262.03, p < 0.001, τ̂ 2 = 0.25, I2 = 95.80 %). types. Farmers committed to written contracts specifying delivery
Mishra and colleagues (2016) observed a negative, moderate, quantity, time, fixed producer price, quality criteria, and payment
and statistically significant effect of HYVPG on the total cost for terms. They received credit for inputs, training, technical assistance,
farmers (ĝ = −0.24 ; 95% CI: −0.38 to −0.10; p < 0.001). On the other and transport, processing, and branding the rice as a premium

F I G U R E 85 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of contract farming
interventions on total farm investment.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 129 of 191

product sold in urban markets (Mishra et al., 2022b). In Haryana, provided contracts that combined both production management and
India, baby corn farmers had formal marketing contracts with input or credit‐supply elements. We included two estimates from
ABCPMCSL. In this contract farming agreement, farmers received Simmons and colleagues (2005), as the evaluation focused on two
training on cultivation practices and had to supply good‐quality different contract farming interventions (PT seed rice and PSC broiler)
produce as per contract specifications (Maertens & Vande, 2017). with independent comparative groups.
All studies here used household‐level measures of investments in All studies here used outcome variables relating to farm
fertilizer and pesticides. Mishra and colleagues (2022b) relied on fertilizer investment in labor at the household level (k = 7) and the plot level
investments in rupees per acre, Maertens and Vande (2017) used (k = 1) (Mano et al., 2017). For example, the outcome variable here in
investments in pesticides per hectare, and Ren and colleagues (2021) Mishra and colleagues (2018b) was the total hired‐worker days per
relied on data on investments in organic fertilizer per mu in China. acre. In Simmons and colleagues (2005), the analysis focused on the
Contract farming interventions did not affect farm investment effect of two interventions on non‐family labor use in days‐per‐year.
(fertilizers and pesticides). The overall weighted average effect is small Contract farming interventions did not affect farm investments in
ˆ = 0.11; 95% CI: −0.24
and positive but statistically insignificant (SMD labor. The overall weighted average effect is very small and positive
to 0.47; p = 0.53) (Figure 86). Effect sizes vary between −0.44 and ˆ = 0.03; 95% CI: −0.04 to
but statistically insignificant (SMD
0.41 standard deviations. There is substantial heterogeneity among 0.10; p = 0.40) (Figure 87) Effect sizes vary between −0.45 and
the studies (Q(2) = 17.74 , p < 0.01, τ̂ 2 = 0.09, I2 = 88.73%). 0.26 standard deviations. The heterogeneity among the studies is
Mishra and colleagues (2022b) observed a negative, large, and substantial (Q(7) = 47.55, p < 0.001, τ̂ 2 = 0.004 , I2 = 85.28%).
statistically significant effect of ABCPMCSL on fertilizer expenditures Mano and colleagues (2017) observed a negative, large, and
(ĝ = −0.44 ; 95% CI: −0.79 to −0.08; p = 0.02). On the other end of statistically significant effect of RCFCI on the machine cost per
the spectrum, in China, Ren and colleagues (2021) observed a hectare in thousand Central African francs/ha (ĝ = −0.45; 95% CI:
positive, large, and statistically significant effect of CFRFESSP on −0.71 to −0.18; p < 0.001). On the other end of the spectrum, in
investments in organic fertilizers (ĝ = 0.44 ; 95% CI: 0253 India, Mishra and colleagues (2018b) observed a positive, moderate,
to 0.56; p < 0.001). and statistically significant effect of OCF on total hired‐worker days
Exploratory moderator analyses were limited, but none of the per acre (ĝ = 0.26; 95% CI: 0.10 to 0.41; p = 0.002).
tested variables significantly contributed to the variation in estimates. Effects are on average lower if they are based on input or credit‐
As all included estimates are assessed as high risk of bias, we did not supply contract interventions relative to other types of contract
conduct a sensitivity analysis by risk of bias. farming interventions. Of the eight included effect sizes, one remains
after excluding high‐risk bias estimates (Mano et al., 2017). The
Farm investment (labor). Within seven studies, we identified eight magnitude of this effect was discussed above.
independent estimates of the effect of contract farming interventions
on farm investments in labor outcomes. One study evaluated an Prices. Within 19 studies, we identified 19 independent estimates of
intervention that took place in Africa. Two interventions occurred in the effect of contract farming interventions on price outcomes. Studies of
East Asia and the Pacific, and three took place in South Asia. CFDC evaluated interventions took place in several regions: Africa (n = 9), East
intervention spanned across Africa and South Asia. There were no Asia and the Pacific (n = 5), Latin America and the Caribbean (n = 1), and
multi‐component interventions. One intervention provided fixed‐ South Asia (n = 4). Two studies evaluated multi‐component interven-
price and price‐guarantee contracts (TCFN), while another was based tions (FPC, SQP). Five interventions provided fixed‐price and price‐
on production‐management contracts (PT seed rice). PSC broiler was guarantee contracts (e.g., MBCF, CCF Marketing contract, RCFLAC),
an input or credit‐supply contracts intervention. while seven others were based on production‐management contracts
The remaining interventions were other contract types (CF, (e.g., IMPROVE, SQP, RCFCI). The remaining interventions were input
RMPLPC Maize) and undefined contract farming (CFDC, OCF). RCFCI or credit‐supply contracts (eight interventions including FPC and RCFV),

F I G U R E 86 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of contract farming
interventions on farm investment (fertilizers and pesticides).
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
130 of 191 | MARION ET AL.

F I G U R E 87 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of contract farming
interventions on farm investments in labor.22

F I G U R E 88 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of contract farming
interventions on prices.

other contract types (AOL, RMPLPC Maize), and undefined contract et al., 2022b). Only one outcome variable here used the average price
farming (CFT, BRCF, CFRFESSP). RCFCI, WRCF, SQP, RCFLAC, and received by household across all crops (Herrmann et al., 2018).
ESOPB combined elements from several contract types. Contract farming interventions increased prices. The overall
Eighteen estimates were at the household level and one estimate weighted average effect is moderate, positive, and statistically
was at the plot level (Mano et al., 2017). Outcome variables here ˆ = 0.27 ; 95% CI: 0.10 to 0.45; p < 0.001)
significant (SMD
measure the sales price for specific crops, or an index of sales prices (Figure 88). Effect sizes vary between −1.35 and 1.15 standard
received. We selected the estimates of sales prices for prevalent crops deviations. The heterogeneity among the studies is considerable
in the local contexts or crops targeted by the intervention, such as (Q(18) = 412.12, p < 0.001, τ̂ 2 = 0.14 , I2 = 95.63%).
coffee in Uganda in the IMPROVE intervention (Hansen & Rosenthal, In Senegal, Soullier and Moustier (2018) observed a negative,
2014) or baby corn in India in the ABCPMCSL intervention (Mishra large, and statistically significant effect of RVC on rice prices
(ĝ = −1.35; 95% CI: −1.64 to −1.06; p < 0.001). The control group
22 comprised producers who used the national agricultural bank loan
For the estimate from Meemken and Bellemare (2020), these several countries are
Bangladesh, Côte d'Ivoire, Mozambique, Nigeria, Tanzania, and Uganda. and sold through spot transactions. On the other end of the
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 131 of 191

spectrum, in Ethiopia, Tefera and Bijman (2021) observed a positive, estimates with a high risk of bias, eleven remain in the analysis. The
large, and statistically significant effect of CFMBEP on the selling overall weighted effect is reduced and becomes statistically
price of malt barley (ĝ = 1.15; 95% CI: 0.90 to 1.40; p < 0.001). ˆ = 0.11; 95% CI: −0.07 to 0.30; p = 0.24).
insignificant (SMD
None of the moderator variables significantly contributed to the
variation in estimates. We did not identify any funnel plot asymmetry, Volume. Within five studies, we identified five independent estimates
meaning that publication bias is not likely to be present. Twelve of of the effect of contract farming interventions on volume outcomes.
the included estimates are assessed as high risk of bias. After Studies of evaluated interventions took place in Africa (n = 4) and
removing these, seven estimates remain in the analysis. The overall Latin America and the Caribbean (n = 1). There were no multi‐
weighted effect decreases and becomes statistically insignificant component interventions. One intervention provided fixed‐price and
ˆ = 0.09; 95% CI: −0.27 to 0.45; p = 0.63).
(SMD price‐guarantee contracts (MBCF), while another intervention was
based on production‐management contracts (CoffeeCF). The remain-
Agricultural production outcomes. The agricultural production out- ing interventions were input or credit‐supply contracts (DCFH) and
comes are yields, volume, produce type, quality, and crop loss. The undefined contract farming (CFE). ESOPB is a contract farming
included studies provided evidence on all these outcomes, except intervention in Benin for rice producers that combined fixed‐price
crop loss. and price‐guarantee contracts, input or credit‐supply contracts, and
other contract types.
Yields. Within 43 studies, we identified 43 independent estimates of Outcome variables include measures of the total quantity or
the effect of contract farming interventions on yield outcomes. Studies of volume of agricultural production. The analysis is at the household
evaluated interventions took place in several regions: Africa (n = 25), level for all five included estimates. As for the other outcomes, we
East Asia and the Pacific (n = 7), Latin America and the Caribbean selected estimates focusing on crops targeted by the intervention
(n = 1), and South Asia (n = 10). Three studies evaluated multi‐ (such as coffee in CoffeeCF in Haiti, Fuller, 2013) or the most
component interventions (VML1, FPC, VCFC). Nine interventions prevalent crops in the local context.
provided fixed‐price and price‐guarantee contracts, while 12 others Contract farming interventions increased volume. The overall
were based on production‐management contracts. The remaining weighted average effect is moderate, positive, and statistically
interventions were input or credit‐supply contracts (13 interventions), ˆ = 0.27 ; 95% CI: 0.12 to 0.41; p = 0.001)
significant (SMD
other contract types (five interventions), and undefined contract farming (Figure 90). Effect sizes vary between 0.09 and 0.41 standard
(9 interventions). RCFCI, FPC, COPF marketing contracts, RCFLAC, deviations. The heterogeneity among the studies is substantial
and ESOPB combined elements from several contract types. (Q(4) = 9.48, p = 0.05, τ̂ 2 = 0.02, I2 = 57.79%).
There were 37 estimates at the household level, one estimate at In Ethiopia, Alemu and colleagues (2016) observed a positive,
the individual level (Yegbemey et al., 2021), and five estimates at the large, and statistically significant effect of DCFH on honey produc-
plot level. All outcome variables relate to measures of agricultural tion in kilograms (ĝ = 0.40 ; 95% CI: 0.22 to −0.57; p < 0.001). In
yields. We first selected estimates that examined the interventions' Benin, Maertens and Vande (2017) observed a similar direction, size,
effects on targeted‐crop yields. Alternatively, we selected estimates and significant effect of ESOPB on the total quantity of rice produced
focusing on the most prevalent crops in the local context. (ĝ = 0.41; 95% CI: 0.24 to 0.58; p < 0.001).
Contract farming interventions increased yields. The overall None of the moderator variables significantly contributed to the
weighted average effect is moderate, positive, and statistically variation in estimates. Of the five included effect sizes, one remains
ˆ = 0.20; 95% CI: 0.10 to 0.29; p < 0.001)
significant (SMD after excluding the high risk of bias estimates (Fuller, 2013). This
(Figure 89). Effect sizes vary between −0.93 and 1.15 standard estimate is small, positive, and statistically insignificant (ĝ = 0.12;
deviations. The heterogeneity among the studies is substantial 95% CI: −0.05 to 0.30; p = 0.19).
(Q(42) = 356.29, p < 0.001, τ̂ 2 = 0.08, I2 = 88.21%).
In India, Mishra and colleagues (2018a) observed a negative, Quality. Within seven studies, we identified seven independent
large, and statistically significant effect of BRCF on rice yields estimates of the effect of contract farming interventions on quality
(ĝ = −0.93; 95% CI: −1.19 to −0.67; p < 0.001). On the other end of outcomes. Studies of evaluated interventions took place in several
the spectrum, in Ghana, Ragasa and colleagues (2018) observed a regions: Africa (n = 4), East Asia and the Pacific (n = 1), and Latin
positive, large, and statistically significant effect of AMOCF on maize America and the Caribbean (n = 2). SQP evaluated multi‐component
yields (ĝ = 1.15; 95% CI: 0.94 to 1.36; p < 0.001). SSKL had a interventions, MBCF provided fixed‐price and price‐guarantee con-
positive, large, and statistically significant increase in coffee yields tracts, while ESOPT, IMPROVE, SQP, FECF, and CFMBEP provided
(production per tree) in Uganda (Bolwig et al., 2009, ĝ = 0.73; 95% CI: production‐management contracts. The remaining interventions were
0.40 to 1.05; p < 0.001). based on other contract types (PineappleCF). SQT contracts also
None of the moderator variables significantly contributed to the combined elements linked to input or credit‐supply contracts.
variation in estimates. We did not identify any funnel plot asymmetry, Outcome variables relate to measures of produce quality. Wollini
meaning that publication bias is not likely to be present. Thirty‐one of and colleagues (2013) used households' percentage graded as high‐
the included estimates are assessed as high risk of bias. After removing quality produced pineapples to evaluate the effect of Pineapple CF in
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
132 of 191 | MARION ET AL.

F I G U R E 89 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of contract farming
interventions on yields.

F I G U R E 90 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of contract farming
interventions on volume.

Costa Rica. Similarly, Saenger and colleagues (2016) relied on vary between 0.01 and 1.05 standard deviations. The heterogeneity
measures of absolute milk fat in kilograms produced by households among the studies is considerable (Q(6) = 82.13, p < 0.001,
during the last 12 months in Vietnam (FECF). τ̂ 2 = 0.137, I2 = 92.70 %).
Contract farming interventions improved quality. The overall Abade and colleagues (2019) in Togo observed a positive, large,
weighted average effect is large, positive, and statistically significant and statistically significant effect of ESOPT on the degree of paddy
ˆ = 0.42; 95% CI: 0.13 to 0.72; p = 0.01) (Figure 91). Effect sizes
(SMD purity (ĝ = 0.71; 95% CI: 0.51 to 0.91; p < 0.001). In Ethiopia, Tefera
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 133 of 191

F I G U R E 91 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of contract farming
interventions on quality.

F I G U R E 92 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of contract farming
interventions on sales.

and Bijman (2021) observed a positive, large, and statistically contracts. ESOPB is a contract farming intervention in Benin for rice
significant effect of CFMBEP on barley quality (ĝ = 1.05; 95% CI: producers that combined fixed‐price and price‐guarantee contracts,
0.80 to 1.30; p < 0.001). input or credit‐supply contracts, and other contract types.
None of the moderator variables significantly contributed to the All studies here (k = 8) used household‐level measures of
variation in estimates. Of the seven included effect sizes, two remain proportion or quantity of produce sold. In Lentz and Upton (2016),
after excluding high risk of bias estimates. The effect of contract farming the outcome variable was households' percentage of all cultivated
interventions on quality outcomes increases but becomes statistically crops sold (the ratio of the value of crops sold divided by the value of
ˆ = 0.59; 95% CI: −0.29 to 1.48; p = 0.19).
insignificant (SMD crops produced). Deutschmann and colleagues (2023) used the
quantity of produced groundnut sold. In those studies, we chose
Intermediate outcomes. The intermediate outcomes category includes measures combining all cultivated crops (e.g., Lentz & Upton, 2016;
outcomes that may affect farmers following the harvest period Herrmann et al., 2018), the most prevalent crops in the local
through the sale of their products. They cover two outcome types: environment, or crops targeted by the intervention (e.g., Maertens &
sales and farm income. Vande, 2017; Fuller, 2013; Deutschmann et al., 2023).
Contract farming interventions increased sales. The overall
Sales. Within eight studies, we identified eight independent esti- weighted average effect is moderate, positive, and statistically
mates of the effect of contract farming interventions on sales ˆ = 0.34; 95% CI: 0.18 to 0.50; p < 0.001)
significant (SMD
outcomes. Studies of evaluated interventions took place in Africa (Figure 92). Effect sizes vary between 0.07 and 0.82 standard
(n = 7) and Latin America and the Caribbean (n = 1). Two studies deviations. There is substantial heterogeneity among the studies
evaluated multi‐component interventions (AVCPO and P4P Tanza- (Q(7) = 43.03, p < 0.001, τ̂ 2 = 0.04 , I2 = 83.73%).
nia). One intervention provided fixed‐price and price‐guarantee Tefera and Bijman (2021) in Ethiopia observed a positive, large,
contracts (MBCF), while two others were based on production‐ and statistically significant effect of CFMBEP on the malt barley sold
management contracts (P4P Tanzania and CFMBEP). The remaining ratio (ĝ = 0.38; 95% CI: 0.13 to 0.63; p = 0.002). In the same country,
interventions were other contract types (AVCPO and ABCPA) and Biggeri and colleagues (2022) also observed a large positive and
undefined contract farming (CFT). CoffeeCF combined elements of significant effect from AVCPO on the share of cereal sold through
fixed‐price and price‐guarantee contracts and production‐management cooperatives (ĝ = 0.82; 95% CI: 0.66 to 0.98; p < 0.001).
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
134 of 191 | MARION ET AL.

None of the moderator variables significantly contributed to the statistically significant effect on farmers' net returns (yuan/mu) on
variation in estimates. Of the eight included effect sizes, four remain apples (ĝ = 0.95; 95% CI: 0.73 to 1.16; p < 0.001). Pavneet (2017,
after excluding high risk of bias estimates. The effect of contract farming PSPL, Chicory) observed a large, positive, and statistically significant
interventions on sales outcomes slightly increases and is still statistically effect of PSPL in India on gross income from chicory (ĝ = 3.73; 95%
ˆ = 0.39; 95% CI: 0.03 to 0.74; p = 0.03).
significant (SMD CI: 3.19 to 4.26; p < 0.001).
We identify that the effect size provided by Abdulai and Al‐
Farm income. Within 79 studies, we identified 84 independent Hassan (2016) and Pavneet (2017, PSPL, Chicory) are considered
estimates of the effect of contract farming interventions on farm outliers according to the studentized residuals of the meta‐analysis.
income outcomes. Studies of evaluated interventions took place in According to the model's Cook's distances, the estimates from
several regions: Africa (n = 41), East Asia and the Pacific (n = 17), Ragasa et al. (2018), Abdulai and Al‐Hassan (2016), Adabe (2017),
Europe and Central Asia (n = 1), Latin America and the Caribbean and Pavneet (2017, PSPL, Chicory) are overly influential in terms of
(n = 3), and South Asia (n = 17). Two studies evaluated multi‐ their weight. The effect size from Pavneet (2017, PSPL, Chicory) is
component interventions (FPC and VCFC). Nineteen interventions particularly large, but this may be due to the very small sample size
provided fixed‐price and price‐guarantee contracts (e.g., MBCF, of this study (100 farmers). If we remove those four estimates from
CattleCF, CCF marketing contracts), while 32 others were based on the analysis, the overall effect slightly reduces and is still
production‐management contracts (e.g., NIL, DCFI, SSKL, AKR). The ˆ = 0.31; 95% CI: 0.25 to
statistically significant (SMD
remaining interventions were input or credit‐supply contracts (21 0.38; p < 0.001).
interventions, e.g., NFOS, DCFH, IMCF), other contract types (11 Effects are on average higher if they are based on fixed‐price and
interventions), and undefined contract farming (12 interventions). Ten price‐guarantee contract interventions relative to other types of
interventions combined elements of several contract farming types contract farming interventions. We identified funnel plot asymmetry,
(ESOPB, CoffeeCF, WRCF, RCFCI, RCFLAC, AKR, COPF Marketing meaning that publication bias is likely to be present. Of the 84
contr, PSC, CFZ).23 included effect sizes, 23 remain after excluding high risk of bias
We included two estimates from Pavneet (2017), as the evaluation estimates. The effect of contract farming interventions on farm income
focused on three different contract farming interventions (PSPL, PepsiCI, outcomes remains the same and is still statistically significant
and RSL) with independent comparative groups. Similarly, this analysis ˆ = 0.34 ; 95% CI: 0.22 to 0.47; p < 0.001).
(SMD
contains four estimates from Narayanan (2014), as the author
conducted four distinct analyses evaluating the effects of four different Welfare outcomes. The welfare outcomes category covers measures of
contract farming schemes in India as part of the CFS intervention—each improvements in farmers' well‐being and quality of life. These
centered on a crop with an independent comparative group (marigolds, encompass a range of social, economic, and environmental factors;
papayas, broilers, and gerkins). however, in this review, we focus on two: total income and food
All studies here used measures of income generated from farm security and nutrition.
activities including profits at the household level (k = 78), individual
level (n = 3), and plot level (n = 3). All outcomes were farm income or Total income and wealth. Within 38 studies, we identified 39
profit on all combined crops in the local contexts (e.g., Ito et al., 2012; independent estimates of the effect of contract farming interventions
Zhu, 2007; Sharma, 2016) or crops targeted by the intervention (e.g., on income and wealth outcomes. Studies of evaluated interventions
Ramaswami et al., 2009, poultry; Alemu et al., 2016, honey; San, primarily took place in Africa (n = 29), with additional studies
2019, rice). Outcome variables included net income (e.g., Benfica implemented in East Asia and the Pacific (n = 4), Latin America and
et al., 2006), gross agricultural income (e.g., Warning & Key, 2002; the Caribbean (n = 1), South Asia (n = 3), and one intervention in
Pavneet, 2017), net profit (e.g., Narayanan, 2014), or gross profit Africa and South Asia (CFDC).
(e.g., Simmons et al., 2005). Three studies focused on multi‐component interventions (P4P
Contract farming interventions increased farm income. The overall Tanzania, CATF, and VML1). Seven interventions provided fixed‐price
weighted average effect is moderate, positive, and statistically and price‐guarantee contracts (e.g., CFMBEPI and KSCL), while 13
ˆ = 0.33; 95% CI: 0.26 to 0.40; p < 0.001)
significant (SMD others were based on production‐management contracts (e.g., VCF,
(Figure 93). Effect sizes vary between −0.79 and 3.73 standard CHP, DCF). The remaining interventions were input or credit‐supply
deviations. There is considerable heterogeneity among the studies contracts (12 interventions, e.g., RCF and VCF), other contract types
(Q(84) = 1358.90, p < 0.001, τ̂ 2 = 0.10 , I2 = 93.89%). (seven interventions), and undefined contract farming (five
In Ghana, Abdulai and Al‐Hassan (2016) observed a negative, interventions).
large, and statistically significant effect of SoyabCF on crop income Five interventions combined elements of several contract farm-
(ĝ = −0.79; 95% CI: −1.00 to −0.57; p < 0.001). Ma and Abdulai ing types (ESOPB, CFMBEPI, WRCF, COPF Marketing contr24). We
(2016) found that WACF in China had a positive, large, and found two estimates in Olomola (2010), as the evaluation focused on

23 24
Information is in Supporting Information: Appendix 3 for Chapter 4. Information is in Supporting Information: Appendix 3 for Chapter 4.
18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
135 of 191

F I G U R E 93 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of contract farming
|

interventions on farm income.

ET AL.
MARION
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
136 of 191 | MARION ET AL.

two different contract farming interventions (SoyabCFM and ONL) positive, and statistically significant effect of VESC on yearly income
with independent comparative groups. per capita (ĝ = 1.97 ; 95% CI: 1.59 to 2.34; p < 0.001).
All studies here used measures of total household gross income None of the moderator variables significantly contributed to the
(e.g., Bezabeh et al., 2020), net income (Miyata et variation in estimates. We identified funnel plot asymmetry, meaning
al., 2009), and other measures of socio‐economic status such as that publication bias is likely to be present. Of the 39 included effect
total household expenditure (e.g., Saenger et al., 2016), assets (e.g., sizes, four remain after excluding high risk of bias estimates. The
Fort & Vargas, 2015), or wealth indices (e.g., Schüpbach, 2014) at effect of contract farming interventions on income and wealth slightly
the household‐level. ˆ = 0.20 ; 95% CI: 0.08
reduces and is still statistically significant (SMD
Contract farming interventions increased income and wealth. The to 0.32; p = 0.01).
overall weighted average effect is moderate, positive, and statistically
ˆ = 0.23; 95% CI: 0.17 to 0.3; p < 0.001) (Figure 94).
significant (SMD Food security. Within 13 studies, we identified 13 independent
Effect sizes vary between = −0.68 and 1.97 standard deviations. The estimates of the effect of contract farming interventions on food security
heterogeneity among the studies is considerable (Q(38) = 383.95, outcomes. Studies of evaluated interventions took place in several
p < 0.001, τ̂ 2 = 0.03, I2 = 90.10 %). regions: Africa (n = 11), Latin America and the Caribbean (n = 1), and
Megersa and Assefa (2019) in Ethiopia observed a negative, South Asia (n = 1). Three studies evaluated multi‐component interven-
large, and statistically significant effect of MBSCF on net income tions (VML1, AVCPO, and P4P Tanzania). Five interventions provided
(ĝ = −0.68; 95% CI: −1.26 to −0.11; p = 0.02). Wendimu and fixed‐price and price‐guarantee contracts (VML1, CFM, CCF marketing
colleagues (2016) found that WSSOS in Ethiopia similarly had a contracts, RCFBP, and CoffeeCF), while three others were based on
negative, large, and statistically significant effect on total net income production‐management contracts (P4P Tanzania, CTCF, and CoffeeCF).
(ĝ = −0.55; 95% CI: −0.84 to −0.26; p < 0.001). Farmers faced very The remaining interventions were input or credit‐supply contracts (RVC
low prices for sugar produced in this contract farming scheme and Production contr), other contract types (AVCPO), and undefined contract
were not allowed to cultivate other profitable crops. On the other farming (PCN, CFG, CFT, and OCF). CoffeeCF combined elements
end of the spectrum, in Tanzania, Benali (2017) observed a large, across different contract farming types.

F I G U R E 94 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of contract farming
interventions on income and wealth.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 137 of 191

F I G U R E 95 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of contract farming
interventions on food security.

Outcomes here are all at the household level and measure the share of food expenditure (ĝ = 1.01; 95% CI: 0.84 to
value of daily calorie intake (Adebisi et al., 2019), food consumption 1.17; p < 0.001). In Nigeria, PCN led to a positive, large, and
expenditure levels (k = 2, Herrmann et al., 2018; Fuller, 2013), and statistically significant effect on adult equivalent household calorie
household food insecurity access scale scores (HFIAS, Biggeri et al., intake (Adebisi et al., 2019) (ĝ = 0.50 ; 95% CI: 0.21 to
2022; Binpori et al., 2021; Soullier & Moustier, 2018). Two studies 0.80; p = 0.001)
used food consumption score measures (FCS, Olounlade et al., Effects are on average lower if they are based on fixed‐price and
2020; Lentz & Upton, 2016). Other outcome variables were price‐guarantee contract interventions. We did not identify any funnel
duration of the hunger period (Navarra, 2017; Bellemare & Novak, plot asymmetry, meaning that publication bias is not likely to be
2016), a binary variable on whether a household faced a period of present. Of the thirteen included effect sizes, six remain after
hunger (Dubbert & Abdulai, 2021), and the share of food excluding the high risk of bias estimates. The effect of contract
expenditure (Mishra et al., 2018b). Measures on the household farming interventions on food security outcomes is slightly greater and
food insecurity access scale, HFIAS, relating to period and still statistically insignificant (SMDˆ = 0.17 ; 95% CI: −0.03
occurrence of hunger and share of food expenditure were reversed, to 0.38; p = 0.10 ).
meaning that a higher effect represents greater food security.
Contract farming interventions did not affect food security. The Nutrition. Within three studies, we identified three independent
overall weighted average effect is small, positive, and statistically estimates of the effect of contract farming interventions on nutritional
ˆ = 0.16; 95% CI: −0.01 to 0.33; p = 0.06)
insignificant (SMD outcomes. All studies of evaluated interventions took place in Africa
(Figure 95). Effect sizes vary between −0.43 and 1.01 standard (n = 3). One evaluated a multi‐component intervention (AVCPO). One
deviations. There is considerable heterogeneity among the studies intervention provided input or credit‐supply contracts (MCFN), and
(Q(12) = 191.88, p < 0.001, τ̂ 2 = 0.09, I2 = 93.75%). another was based on other contract types (AVCPO) and undefined
Olounlade and colleagues (2020) in Benin observed a negative, contract farming (CFT). All outcomes here are measures of the HDDS.
large, and statistically significant effect of RCFBP on the food Contract farming interventions did not affect nutrition. The overall
consumption score (ĝ = −0.43; 95% CI: −0.71 to −0.16; p = 0.002). weighted average effect is very small, negative, and statistically
Similarly, Dubbert and Abdulai (2021) found that CCF Marketing ˆ = −0.07; 95% CI: −0.71 to 0.57; p = 0.82)
insignificant (SMD
contract in Ghana had a negative, moderate, and statistically (Figure 96). Effect sizes vary between −0.62 and 0.44 standard
significant effect on food security (ĝ = −0.24 ; 95% CI: −0.44 to deviations. There is considerable heterogeneity among the studies
−0.05; p = 0.02). The outcome variable was the length of time that (Q(2) = 127.14 , p < 0.001, τ̂ 2 = 0.31, I2 = 98.43%).
members of the farm household went to bed without three meals In Tanzania, Hermann and colleagues (2018) observed a negative,
per day.25 In India, Mishra and colleagues (2018b) observed a large, and statistically significant effect of CFT on the HDDS
positive, large, and statistically significant effect of OCF on the (ĝ = −0.62; 95% CI: −0.70 to −0.53; p < 0.001). Conversely, in
Nigeria, Nazifi and colleagues (2021) observed a positive, large, and
25 statistically significant effect of MCFN (ĝ = 0.44 ; 95% CI: 0.26 to
The effect size here is reversed, meaning that a higher effect represents lower food
security. 0.63; p < 0.001).
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
138 of 191 | MARION ET AL.

F I G U R E 96 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of contract farming
interventions on nutrition.

Exploratory moderator analyses were limited, but none of the but the magnitude of the effect increases alongside households'
tested moderator variables significantly contributed to the variation farm size.
in estimates. Of the three included effect sizes, one remains after
excluding the high risk of bias estimates ( ĝ = −0.03; 95% CI: −0.19 to Gender. Table 31 shows effect sizes by gender subgroups from
0.13; p = 0.69) (Biggeri et al., 2022). Yegbemey and colleagues (2021). REDC supported the expansion of
e‐commerce in rural areas with the establishment of county town‐
Subgroup analysis. We now examine how the effect sizes of contract village delivery systems and e‐commerce service terminals in China. It
farming interventions may vary by subgroup. We here compare reduced the likelihood of working in the agricultural sector and
estimates within studies from the same intervention across subgroups. increased the likelihood of working in the wage sector for households
We do not report here the effect sizes across subgroups, which are with male and female heads; the effect magnitude was similar across
consistently statistically insignificant. We divide this analysis into two these two outcomes and the gender of household heads.
types of subgroups: land size and gender of beneficiaries. Across the included studies, we did not find estimates within
studies from the same intervention across time periods.
Land size. The following included studies provided relevant estimates
of effects by land size: Mishra and colleagues (2016, 2018c, 2018d, Barriers and facilitators' analysis
2022a), Dubbert and Abdulai (2021), Dubbert (2019), Ruml and Qaim This section discusses the reported facilitators and barriers to the
(2020a), Benali (2017), and Kumar and colleagues (2019). Across interventions' effects within local contexts. This presentation is subject
these studies, we do not find that farmers with smaller land holdings to an important caveat: the included studies are quantitative impact
consistently experience lower benefits from contract farming inter- evaluations. Such studies tend to focus narrowly on identifying the
ventions across outcomes. Table 30 focuses on effect sizes by land magnitude of the intervention's effects, without discussing implemen-
size subgroups from Mishra and colleagues (2016, 2018d). We do not tation details or rigorously explaining why an intervention was
present all relevant estimates due to the large number of them. The successful (or not). A thorough analysis of barriers and facilitators
selected estimates provide some insight into the variation of effects would require a dedicated search for different types of evidence. Of
according to farmers' land size. the 144 papers in this body of evidence, our analysis of barriers and
Mishra and colleagues (2016, a linked study of Kumar et al., facilitators identified 56 included papers with relevant information.
2019) evaluated the effect of HYVPG (production‐management
contracts) in Nepal. Effects by land size subgroups are displayed in The effectiveness of contract farming interventions can be hindered by
Table 30 across two outcomes: total revenues and rice yields. the power imbalance between the buyer and the farmer: Group
Positive, moderate, and statistically significant effects in revenues participation and education can mitigate this risk. The collaboration
and yields from this intervention are shared across farmers of all between the buyer, or firm, and the farmer can be negatively
sizes, except for revenues of large‐sized farms. impacted by the bargaining power imbalance between them. Firms
Mishra and colleagues (2018d) found that TCFN in Nepal and buyers might have access to additional information and
benefited most farmers with larger farms (in terms of income and resources, putting them at an advantage in their negotiation with
tomato yields). This contract farming intervention offered benefits farmers and hindering farmers' benefits from the intervention.
such as subsidized seeds, fertilizers, and pesticides, low‐interest In Kenya, Mwambi and colleagues (2016) analyzed the effect of
credit, training, and market guarantees to farmers. The effect on farm the AGAK agreement with farmers through the facilitation of training
income is large, positive, and statistically significant for households and contract farming with farming groups. They observed that the
with more than 0.51 hectares. This effect was positive, small, but not intermediary‐based contract had a negative impact on producers, as
statistically significant for households with small‐sized farms. they were unaware of all the contract terms and were not involved in
Similarly, the effect on yields is positive and statistically significant their negotiation, thereby leading to contract breaches and
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 139 of 191

T A B L E 30 Effect sizes across land‐size subgroups (from Mishra et al., 2016, 2018b).
Outcome Outcome
Study subcategory description Subgroup information ĝ p Value

Mishra Farm income Total revenues Large (more than 0.73 ha) 0.37 0.01
et al. (2016)
Medium (more than 0.43 ha, less than 0.42 0.003
or equal to 0.73 ha)

Small (less than or equal to 0.43 ha) 0.34 0.01

Yields Rice yields (kg/ha) Large (more than 0.73 ha) 0.27 0.06

Medium (more than 0.43 ha, less than 0.29 0.04

or equal to 0.73 ha)

Small (less than or equal to 0.43 ha) 0.30 0.03

Mishra et al. Farm income Total revenues Large farms (less than 0.85 ha) 0.79 <0.001
(2018d) (NPR/ha)
Medium farms (more than 0.51 ha and 0.48 0.01
less than or equal to 0.85 ha)

Small farms (less than or equal to 0.17 0.17

0.51 ha)

Yields Tomato yields Large farms (less than 0.85 ha) 0.55 <0.001
(kg/ha)
Medium farms (more than 0.51 ha and 0.39 0.03
less than or equal to 0.85 ha)

Small farms (less than or equal to 0.29 0.02

0.51 ha)

Note: Estimates in bold have a statistically significant effect size (ĝ ) at a 5% significance level.
Abbreviations: ha, hectare; kg, kilogram.

T A B L E 31 Effect sizes across gender subgroups (from Yegbemey reported a reduction in transaction costs among FPC members
et al., 2021). entering contract farming in China.
Outcome Outcome Subgroup The authors also identified education and access to information
subcategory description information ĝ p Value as mitigating factors for power imbalances. In Ghana, Kenya, and
Yields Soy yields Men 0.49 <0.001 Zambia, Dihel and colleagues (2018) noted that a lack of information
(kg/ha) about potential contract buyers hindered the realization of agree-
Women 1.06 <0.001
ments as part of VCP. In Benin, Maertens and Vanden (2017)
Note: Estimates in bold have a statistically significant effect size (ĝ ) at a
commented that higher levels of education were associated with
5% significance level.
increased benefits from ESOPB contract farming. Similarly, in Ghana,
Abbreviations: ha, hectare; kg, kilogram.
Dubbert (2019) reported that contract farmers with phones had
higher cashew yields compared to those without, due to better
access to information.
disadvantages in accessing benefits. Liang and colleagues (2020)
reported a similar issue in China, wherein producers under VCFC Farm characteristics and types of crops produced can increase the
found themselves in a position of limited influence over the prices benefits of entering contact farming. Contract farming may require a
they received. In other instances, such as RSCFP in India (Swain, certain volume or quality of production. To reach these thresholds,
2016), farmers lost decision‐making power in selecting crops outside producers with larger farms, labor forces, or specific crops might
the contract, resulting in a decrease in productivity. realize more benefits from contract farming. Several authors observed
Some authors highlighted that group participation of farmers that farmers with larger‐sized farm secured greater participation and
could mitigate this risk of power imbalance with companies entering benefits from contract farming (Herrmann [2017] in Tanzania; Kumar
contract farming. By bringing resources together, members of farmer and colleagues [2016] in Nepal; Maertens and Vande [2017] in Benin;
groups may increase their influence in negotiation and obtain more Abdul‐Rahaman and Abdulai [2020] in Ghana; Mishra et al., 2018] in
favorable terms. Maertens and Vanden (2017) observed this Nepal; Seba and colleagues [2018] in Ethiopia; Zhu [2007] in China;
phenomenon in ESOPB in Benin, where members of FOs were able Feyiso [2021] in Ethiopia). The size of the farm is often associated
to secure higher rice incomes due to improved management practices with farmers' revenues and access to financial resources, which may
and access to extension services. Similarly, Hoken and Su (2018) facilitate their participation in contract farming: in Sierra Leone,
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
140 of 191 | MARION ET AL.

Casaburi and colleagues (2014) observed that farmers with larger markets: when spot market prices are higher than prices offered in the
farms were less likely to face limitations in credit access or storage, contracts, farmers miss out on potential earnings.
and could therefore maximize their benefits from entering contract Finally—and relevant to all other intervention categories of our
farming. systematic review—natural characteristics and events can affect
Several authors also reported the benefits of contract farming for participation in and benefits from contract farming. In India,
small landowners (Ito et al., 2012 in China; Mishra et al., 2018 in Narayanan (2014) witnessed the impact of a mealybug infestation
Nepal; Dubbert & Abdulai, 2021 in Ghana). They also noted, on CFS contract farming. Papaya contract farmers could have gained
however, that the benefit of contract farming is related to their more from the intervention, but the infestation impacted the quantity
transition from subsistence farming to commercialization. For and quality of agricultural products. In Ghana, Abdul‐Rahman and
example, in Kenya, Rao and Qaim (2011) highlighted gains among colleagues (2019) made similar observations about the impact of
poor households and small landowners as they exploited opportuni- drought on agricultural production and benefits. Finally, in Senegal,
ties to sell their products to supermarkets at more stable prices. This Deutschmann and colleagues (2023) commented that the COVID‐19
effect among small and large farmers alike might be explained by crisis had a notable, negative impact on cooperatives and the
requirements in contract farming that may repel middle‐scale groundnut industry.
farmers, who have less incentive than small‐scale farmers to enter
contract farming, and less capacity to meet the requirements than Intervention costs
large‐scale farmers. Seven of the 144 papers included in this body of evidence provided
Authors also identified the labor force as a core driver of information on intervention cost (6%) (Table 32). This is lower than
participation in and success of contract farming. For example, in the general trend in impact evaluation literature, wherein fewer than
Senegal, Maertens and Swinnen (2009) found that households with a 20% of impact evaluations report any cost information (Brown &
greater labor force and higher education levels benefited more from Tanner, 2019). In other intervention categories in our systematic
ESC contract farming. Several authors used the family size as a proxy review, there is evidence on costs for at least five percent of their
for labor force and observed that larger households tend to benefit respective bodies of evidence. Moreover, none of the studies that
more from contract farming. For example, in Ethiopia, Meshesha report costs provide any information on the full intervention cost;
(2011) highlighted that larger households with more family members rather, they primarily provide the cost of one component or activity
participating in honey production had increased incomes and capacity of the intervention. Corral (2016) provides a full cost‐benefit analysis
to respond to production demand from CHP contract farming. This of TGP in Mexico and shows that the average cost per hectare is
observation should, however, be tempered by the risk of unintended lower among intervention groups.
consequences, such as children potentially participating in agricul-
tural labor. Discussion
Finally, the authors highlighted that farmers selling specific
crops or higher‐quality products benefited from a comparative Summary of findings. A large body number studies quantitatively
advantage in contract farming. In Nepal, Kumar and colleagues analyze the effect of contract farming interventions on agricultural
(2016) analyzed the effect of GCF contract farming and observed outputs and market outcomes: 194 papers reporting the effect of
that farmers using a superior variety of seeds (i.e. producing ginger 129 unique interventions in 36 countries (with 72 studies focusing on
with low fiber contract increased their net profit from contract Africa), and 125 studies (and 19 linked papers). Our synthesis‐of‐
farming. Similarly, farmers with access to storage facilities that findings analysis focused on 144 studies, as 50 papers did not provide
maintained their product quality experienced more benefits. For any additional results. These 50 papers were unpublished versions of
instance, in Kenya, Mwami and colleagues (2016) found that already included studies.
farmers with suitable storage and transportation resources could The included studies cover five intervention types combining
ensure that they supply fresh, high‐quality products, which multiple subcategories of contract farming interventions, or contract
allowed them to obtain better prices and thereby enhanced their farming interventions with other intervention categories. Most
earnings. studies focused on the effect of production‐management contracts
(n = 47) with exposure periods from one month to 63 years
Market and local characteristics may both facilitate an enabling (although 50 studies did not provide information on exposure and
environment for contract farming or hinder the participation of farmers the follow‐up period). Included studies mostly used QEDs (n = 119
and companies. The authors highlighted that farmers' decisions to out of 125), and all studies that reported on exposure and follow‐up
participate in contract farming are informed by market conditions. Prices periods measured the effect on outcomes immediately after the
are an important decision criterion. In Senegal, Soullier and Moustier end of the intervention. A very small minority of studies provided
(2018) observed that price fluctuations over different seasons cost information (n = 7).
contributed to farmers' decisions to enter RVC contract farming. In Overall, we found positive, moderate to large, and sometimes
Vietnam, Hoang (2021) commented that the reason for the limited, statistically significant effects of contract farming interventions on
short‐term impact of contract farming was the price fluctuation of spot outcomes included in the theory of change of this intervention. None
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 141 of 191

T A B L E 32 Summary of available intervention costs.

Study Acronym Cost information description

Yegbemey et al. (2021) UNPS Cost of 0.25‐hectare delivery of soy‐specific inputs: USD15

Corral (2016) TGP Average cost per hectare:

• Control: USD5330
• Soil analysis (receiving free soil analysis and input recommendations): USD4871
• Soil analysis and agricultural extension worker (receiving a free soil analysis plus frequent plot visits from
an agricultural extension worker): USD4653

Deutschmann et al. (2023) ABCPA Provision of bio‐control technology against aflatoxins: treating one hectare of cropland costs about USD17
at market price. The mean production on one hectare (about 900 kg.) is worth about USD350 at 2018
market prices.

Mwambi (2015) AGA Cost of registration to farmer group: USD12.50

Bellemare (2021) CFM Each respondent was asked whether they would participate in a contract farming agreement that would
raise their income by 10% in exchange for a one‐time monetary investment. The amount of the monetary
investment was randomly selected from six investment amounts of USD12.50, 25, 37.50, 50, 62.50, or 75.

Ruml (2020) COPF Cost for plantation setup and maintenance: USD1600/ha

Mishra (2018) TCFN Average irrigation cost reduction: INR323/acre

Abbreviations: ha, hectare; kg, kilogram.

of the meta‐analyses revealed any statistically significant negative categories we reviewed. Indeed, 76% of included estimates were
effects of the interventions. assessed as having a high risk of bias, primarily because of selection
In terms of immediate outcomes, the meta‐analyses showed that bias, bias from confounding effects, and reporting bias. Included
contract farming led to positive, moderate, and statistically significant estimates were more likely to have a high risk of selection bias and
ˆ = 0.27; 95% CI: 0.10, 0.45; k = 19) and the
effects on prices (SMD confounding bias than in the other chapters.
use and adoption of improved inputs (SMD ˆ = 0.27; 95% CI: 0.07, Included studies often lacked precise information on the location
0.47; k = 14 ). Contract farming interventions also led to improvements of targeted beneficiaries of contract farming interventions. Contract
in agricultural production outcomes such a positive, large, and statistically farming is an older and more widespread intervention in LMICs than
ˆ = 0.42; 95% CI: 0.13, 0.72; k = 7)
significant effect on quality (SMD the other intervention types in this review. This made the
and a positive, moderate, and statistically significant effect on yields identification of an adequate counterfactual more complicated, and
ˆ = 0.20; 95% CI: 0.10, 0.29; k = 43) and volume (SMD
(SMD ˆ = 0.27; significantly increased the risk of selection bias. Included studies
95% CI: 0.12, 0.41; k = 5). The evidence also shows positive, moderate, were more likely to use statistical matching, endogenous treatment
ˆ = 0.34; 95% CI:
and statistically significant effects on sales (SMD effect methods, or IV methods than in other chapters. The chosen IVs
0.18, 0.50; k = 8), ˆ = 0.33; 95% CI:0.26, 0.40;
farm income (SMD were often not exogenous, and authors used endline time‐variant
k = 84 ), and ˆ = 0.23; 95% CI:
total income and wealth (SMD covariates in the statistical matching estimation, which contributed to
0.17, 0.30; k = 39). In the body of evidence, we however do not a high risk of confounding effects of estimates in included studies.
observe statistically significant effects on farm investments, production We found substantial variation between the studies in each of
diversity, and food security. these outcome categories. For example, interventions within the
Despite the large body of evidence available on the effect of same country had heterogeneous effects (Bezabeh et al., 2020;
contract farming interventions, we identified zero, or very few, Wendimu et al., 2016; Megersa and Assefa 2019 in Ethiopia on
estimates for several outcomes of our framework, revealing some income and wealth outcomes). When we explored potential sources of
evidence gaps. The evidence on transaction costs is extremely thin, heterogeneity, we did not find crosscutting patterns that expect
although one aim of contract farming is to reduce transaction costs. context‐specific effects of interventions. We did, however, find that
Similarly, there is currently no or very limited evidence available on for two outcomes (food security and farm income), effects varied by
the effect of contract farming on group participation, use, and adoption contract type.
of labor and machinery, credit, off‐farm income, and non‐farm labor For food security outcomes, effects were lower on average if
market outcomes. Limited evidence is available on the effect of these they were based on fixed‐price and price‐guarantee contract interven-
interventions on farm investment (fertilizers and pesticides), non‐farm tions and higher if they were based on undefined contract farming
income, and nutrition (respectively, k = 3). On the other hand, some interventions, relative to other types of contract farming interventions.
outcomes are widely covered by the literature, such as farm income For farm income, we observed a different result: effects were higher
(k = 84), yields (k = 42), and total income and wealth (k = 38). on average if they were based on fixed‐price and price‐guarantee
This body of evidence on contract farming is characterized by a contract interventions, relative to other types of contract farming
relatively higher risk of bias compared to the other four intervention interventions. The lack of data on follow‐up and exposure periods
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
142 of 191 | MARION ET AL.

limits the ability to analyze the effects' heterogeneity across follow‐ • Contract farming interventions often lack critical information to
up and exposure periods. allow precise attribution of the effects to specific contract
This heterogeneity can be explained by many different factors. characteristics, or understanding of their effects over time. Future
Contract farming interventions themselves are heterogeneous in their studies should aim to provide more information on the terms of
design and objectives and likely to have different effects on the contract being implemented (e.g., fixed‐price, management‐
participant farmers. Across countries, effects are expected to vary production), the exposure to the intervention (including farmers'
across targeted beneficiaries' baseline market access (commercial vs. movements in and out of contracts), and the follow‐up period.
subsistence farmers). Local context can also play a significant role in
the magnitude of the effects.
Additional analyses, such as subgroup analysis and findings 5.3.5 | Chapter 5: Improved storage infrastructure
related to intervention facilitators and barriers, emphasize the risk of and technologies interventions
power imbalances in contract farming between farmers and compa-
nies, with disadvantages for the former. This analysis also reveals the How can storage infrastructure affect agricultural, socio‐economic,
potential influence of farmers' resources (e.g., education, farm size, food security, and nutrition outcomes?
labor force, quality products) as a facilitator of participation and Storage infrastructure and technologies refer to the establishment,
benefits. Finally, prices and environmental characteristics can enhancement, and utilization of physical facilities and systems to
influence the willingness of farmers to enter contract farming. efficiently store, protect, and manage agricultural products and
commodities in terms of their quality and quantity.
Implications for policymakers, implementers. This chapter focuses on the three types of improved storage
• Policymakers can rely on the potential of contract farming infrastructure and technologies interventions:
interventions to improve the quality, sales, and income from
agricultural products. Although these interventions do not affect • on‐farm storage infrastructure and technologies provide technical,
all expected outcomes across the theory of change, farmers logistical, or financial support to improve on‐farm storage
realized moderate‐to‐large positive effects on a specific set of practices including storage bags, containers for oils, and materials
outcomes that policymakers may want to prioritize. to construct on‐farm sheds to farmers.
• To maximize the effectiveness of contract farming interventions, • storage deposit systems often take the form of warehouse receipt
policymakers should ensure symmetry of information between the systems and enable the common storage of agricultural outputs in
different parties of the contract. This will enable farmers to comply off‐farm stores (Maertens et al., 2020). Storage deposit systems also
with the contract terms while ensuring that they receive fair and provide inventory credit where farmers deposit a portion of their
profitable offers from buyers. They should also specifically target harvested crops in designated warehouses after harvesting, which
groups that face additional challenges when entering contract in turn secures a six‐month loan.
farming (e.g., small landowners or those with low labor forces) to • storage technologies for transit provide safe and weatherproof
provide them with additional incentives and support. storage technologies for transit, such as refrigerated trucks, to
• Despite coverage by a large body of evidence, very little facilitate both collective or individual transportation of agricultural
information is available on the cost of contract farming interven- outputs.
tions. Policymakers should consider funding more research
analyzing the value for money and cost‐effectiveness of contract Figure 97 provides an overview of the Theory of Change (TOC) of
farming interventions. storage infrastructure on agricultural, socio‐economic, food security and
nutrition outcomes. Approaches for storage of agricultural outputs and
Implications for researchers. types of infrastructure vary and have different effects. We describe the
• Future studies should aim to mitigate the risks of bias related to specific effects and impacts by each intervention type.
selection, confounding, and reporting to continue improving the
reliability of findings. As discussed above, included studies often Improved on‐farm storage infrastructure and technologies. On‐farm
lacked precise information on the location of targeted beneficia- storage infrastructure and technologies aim to reduce post‐harvest loss
ries meaning that it was not possible to identify an adequate and maintain the quality of the harvest over time. Producers can
counterfactual. The chosen IVs were often not exogenous, and therefore sell a greater share of their agricultural output at a better
authors used endline time‐variant covariates in the statistical prices leading to greater farm income and welfare impacts (Katunze
matching estimation. et al., 2017; Omotilewa et al., 2018).26 These effects can encourage
• Despite the large body of evidence available on contract farming, producers to invest in their own production with long‐term
some specific contexts, outcomes, and interventions show implications on productivity, income, and welfare. Reducing the
evidence gaps. Future research could therefore aim to fill evidence
gaps in non‐African regions, and on transaction costs, non‐farm 26
Supporting Information: Appendices 1 and 2 for Chapter 5 provide the references of
outcomes, and fixed‐price contract farming. included studies and additional references, respectively.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 143 of 191

F I G U R E 97 Theory of change for storage infrastructure interventions. The outcome types included in this systematic review are immediate
outcomes (green), agricultural production outcomes (purple), intermediate outcomes (teal), off‐farm and non‐farm outcomes and welfare
outcomes (light blue).

post‐harvest losses can strengthen producers' expectations that storage can help increase household food stocks for smallholder
returns from cultivation are secure, creating an incentive to invest in farmers (Biggeri et al., 2022). In the lean season, several months after
their production capacities. harvest, households tend to have worse food security and nutrition
Storage interventions also give farmers the option to sell their harvest outcomes (Branca et al., 1993; Dercon & Krishnan, 2000; Savy et al.,
at the right time in the year to maximize farm income. In the absence of 2006; Basu & Wong, 2015; Hirvonen et al., 2016). Authors have shown
adequate storage infrastructure, farmers often sell most of their that diet seasonality in childhood can have long‐lasting effects on
production immediately after the harvest to avoid post‐harvest losses height and educational attainment (Christian & Dillon, 2016). Access to
(Affognon et al., 2015; Burke et al., 2019). First, farmers can conduct stored food in the lean season can reduce diet seasonality.
more extensive market search for buyers to find better prices (Lothoré &
Delmas, 2009; Pingali et al., 2019). Second, storage interventions help Storage deposit systems. Storage deposit systems interventions are
farmers manage output price fluctuations. After the harvest, in the warehouse receipt systems. This approach grants farmers the flexibility
season when grain is abundant in local markets, food prices tend to to store crops in proximity to the market where they intend to sell.
significantly decline. In the lean season, prices increases, but at this point Alternatively, these interventions permit farmers to access storage
as most of their harvest was already sold, households can only sell little facilities when establishing on‐farm storage is not feasible either
(Stephens & Barrett, 2011; Casaburi & Reed, 2017; Brander et al., 2021). financially or logistically. The effects from these interventions are the
Farmers can therefore exploit this seasonal price arbitrage. same as from on‐farm storage infrastructure and technologies interventions.
However, prices vary extensively in markets across time and Another effective related strategy of storage deposit systems is
locations, based on the context's dynamics of supply and demand inventory credit, a mechanism that not only provides credit access
(Kaminski et al., 2014; Cardell & Michelson, 2023; Hill & Fuje, 2020). but also encourages farmers to capitalize on favorable seasonal price
The benefits of storage in terms of temporal arbitrage are not variations. Farmers deposit a portion of their harvested crops in
guaranteed, however, given the presence of considerable unexpected designated warehouses after harvesting, which in turn secures a six‐
(non‐seasonal) price volatility during the year and difficulties predicting month loan. The access to credit means that it can initiate different
prices in these contexts. For example, Cardell and Michelson (2023) effects than the ones described above.
show that the price for maize often declines after harvest (30% of the As some of storage deposit systems interventions are associated with
time), so storing after harvest may also be detrimental to farmers access to credit, these interventions can address post‐harvest liquidity
outcomes (especially if programs are combined with credit). constraints and contribute to improvements in agricultural production
Storage interventions also can have large effects on households' and food outcomes. Agricultural production requires use of costly
food security and nutrition. During the post‐harvest phase, on‐farm inputs which farmers often cannot afford. Access to credit can reduce
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
144 of 191 | MARION ET AL.

these liquidity constraints and facilitate access to inputs (Stampini & technologies for transit interventions. These interventions can reduce
Davis, 2008; Oseni & Winters, 2009; Stephens & Barrett, 2011). harvest‐losses and maintain the quality of the products having effect on
Access to credit and commitment to store outputs for a specific time agricultural production, intermediate outcomes and welfare outcomes.
can help limit behaviors which are detrimental to the households. First, As for the other storage interventions, these changes may encourage
households have less of an incentive to sell immediately post‐harvest farmers to invest and adopt improved practices and technologies with
when prices tend to be low and therefore, they can exploit temporal potentially long‐term effects on productivity, income and welfare.
price arbitrage (Le Cotty et al., 2019). Second, households do not need to As for storage deposit systems, these interventions were led by
sell household productive assets to obtain cash for food purchases agricultural cooperatives or producer organizations to efficiently
(“distress selling”), particularly in the lean season (Ceballos et al., 2020). handle beneficiaries' harvest to store for transit. Intervention
Beneficiary farmers of storage deposit systems interventions who implementation costs were most often covered by both the
successfully fulfill their loan obligations have their collateral returned agricultural cooperatives or producer organizations and farmers.
during the lean season, a period when grain prices typically increase. Issues linked to management of cooperatives or producer organiza-
Other farmers may however face difficulties in repaying the credit tions' operations are also relevant here.
after the period of storage has ended. As indicated above, the
presence of considerable unexpected (non‐seasonal) price volatility Description of included studies
during the year means that the value of the stored output may Overall, we included 27 studies examining the effects of 20 storage
decrease and limit their ability to pay off their debt (Cardell & infrastructure interventions. This includes two ongoing studies (Ragasa
Michelson, 2023; Channa et al., 2022). et al., 2016; McIntosh, 2023). The quantitative synthesis of findings
These interventions are often conducted at the community level, relies on effects from 21 papers (20 main studies and one linked
through a cooperative or producer organization which requires paper), as six papers do not provide any additional results. These six
coordination between all parties and management. However ade- papers were ongoing studies, previous versions, or unpublished
quate and fair management of storage can contribute to improved versions of already included studies. Caeyers and Nairi Quinn (2015)
social capital in the community (Biggeri et al., 2022). However, if reported additional outcome data about the effects of ARMA66 (i.e.,
farmers face high storage costs or difficulties in accessing their stored data on outcomes or prioritized outcome indicators that are not
products, the interventions are less likely to produce lasting impacts. reported by the main study identified).
Intervention implementation costs were most often covered by both
the agricultural cooperatives or producer organizations and farmers. Intervention, inputs, and activities
We identified studies on the effects of storage infrastructure
Improved storage technologies for transit. Improved storage technologies interventions implemented in 15 countries (Figure 98). The most
for transit interventions provide safe and weatherproof storage common location of the included interventions is Africa (14), followed
technologies for transit, such as refrigerated trucks, to facilitate both by South Asia (3), and Latin America and the Caribbean, East Asia and
collective or individual transportation of agricultural outputs. Only a the Pacific, and Europe and Central Asia, with one study each,
subset of the effects described above are relevant for improved storage respectively. Meanwhile, the individual countries with the largest

F I G U R E 98 Distribution of interventions by region.

 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 145 of 191

number of included interventions are Tanzania (3) and Burkina Faso, Information: Appendix 4 for Chapter 5 provides information on these
Kenya, and India, with 2 each, respectively. studies such as the evaluated intervention, its subcategory, length of
Supporting Information: Appendix 3 for Chapter 5 provides follow‐up period in years, study design, and type of comparison.
information on the 20 interventions regarding country, type of
infrastructure, and year the intervention began. The most common Outcomes
types of storage infrastructure interventions included storage deposit Our body of evidence covers 15 of the 19 outcomes included in our
systems (11 interventions), followed by on‐farm storage (8) and storage framework. Most studies analyze the effect of improved storage
for transit (1) (Figure 99). Eight of the included interventions use a infrastructure and technologies interventions on the use and adoption of
multi‐component approach, primarily through the combination of improved inputs, technologies, and practices (n = 16), and, more
storage deposit systems and arranged or curated offers from buyers (3). specifically, the adoption of new storage infrastructure and practices
Table 33 provides information about the 20 interventions regarding (n = 10). Several studies also focus on the effect of interventions on
country, type of storage technology, and year the intervention began. sales (n = 12), food security (n = 12), farm income (n = 10), total income
Eight studies focus on multi‐component interventions combining (n = 8) and produce types (n = 8) (Figure 100).
the storage technology with another type of storage practice and/or
another type of intervention (Table 34). For example, the ARMA66 Study design
project in Armenia established four cooperatives: in three, all The 20 studies included in our analysis mostly used QEDs to evaluate
members were women; in one, half of the members were women. the effects of storage infrastructure interventions (n = 15). These quasi‐
It also constructed four community greenhouses, established three experimental approaches primarily used DID and FE approaches
cold‐storage units for agricultural crops, and promoted innovative (n = 6) and statistical matching (n = 6). A subset of studies exploited an
agro‐technology practices for farmers' water management. The experimental approach (n = 5), but we did not observe any trend
project also worked to strengthen participant marketing capacities between the type of interventions, regions, and the study design or
and to link them with markets. method (Supporting Information: Appendix 4 for Chapter 5).
Another example is the FPPM in DRC, which provided important Figure 101 shows the distribution of studies' different follow‐up
marketing support in addition to storage facility construction. The periods (study periods). This is the difference between the start date of
project took an active role in the rehabilitation of feeder roads and the intervention and the most recent outcome data included in the
waterways to areas with concentrated agricultural production, and study's analysis. The 20 included interventions were initiated between
provided price and product information to farmers through radio 2001 and 2020 for an average duration of three years. The average
broadcasts. In the analysis below, we conduct a moderator analysis to follow‐up period of the studies is 3.7 years; the longest reported follow‐
determine whether the estimates from multi‐component interven- up is 10 years (Chatterjee, 2018), while the shortest is two months
tions are on average statistically significantly different than estimates (Takeshima et al., 2021). Most studies included a follow‐up period of less
from single‐component interventions. than eight years, with the largest concentration of studies examining
outcomes less than two years after the intervention began (n = 6).
Comparison
The comparison groups used for these estimations exclusively focused Synthesis of findings
on comparing the outcomes of beneficiaries of storage infrastructure The data extraction of studies under the improved storage infra-
investments to non‐beneficiaries located in a different area. Supporting structure and technologies intervention category identified a total of

F I G U R E 99 Distribution of interventions by subcategory.

 T A B L E 33 Included improved storage infrastructure and technologies interventions.
146 of 191

First year of
|

Intervention name Acronym Country (ISO3 code) Type intervention

The Anchor Farm Model AFP Malawi (MWI) Improved on‐farm storage infrastructure and technologies; Arranged or curated 2014
offers from buyers

Women's Economic Empowerment in Rural ARMA66 Armenia (ARM) Arranged or curated offers from buyers; Storage deposit systems; Provision of 2010
Communities of Vayots Dzor Region output market information (mobile and internet)

Agricultural Value Chain Project, Oromia AVCPO Ethiopia (ETH) Storage deposit systems; Other contract types; Alternative physical marketplaces 2011

Bulk Milk Freezers BMC India (IND) Improved storage technologies for transit 2007

Cooperative de Prestation de Services Agricoles ‐ COPSA‐C Burkina Faso (BFA) Storage deposit systems 2015
Coobsa

Farm Risk Management for Africa FARMAF Burkina Faso (BFA) Storage deposit systems 2013

Food Production, Processing, and Marketing Project FPPM Democratic Republic of Domestic transport infrastructure; Improved on‐farm storage infrastructure and 2011
Congo (COD) technologies; Provision of output market information (other forms of
communication)

Grameen Bhandaran Yojana GBY India (IND) Improved on‐farm storage infrastructure 2001

Groups Savings and Reinvestment Account GSRA Kenya (KEN) Improved storage infrastructure 2015

Guatemala Highlands Value Chain Development GUAB38 Guatemala (GTM) Arranged or curated offers from buyers; Storage deposit systems 2010
Alliance

High‐Value Agriculture Project in Hill and HVAP Nepal (NPL) Arranged or curated offers from buyers; Improved on‐farm storage infrastructure 2011
Mountainous Areas

Millenium Village Project MVP Ghana (GHA) Domestic transport infrastructure; Storage deposit systems 2012

One Acre Fund OAF Kenya (KEN) Improved on‐farm storage infrastructure 2012

Optimizing Income of Local Chicken and Sunflower OILCSS Tanzania (TZA) Alternative physical marketplaces; Improved on‐farm storage infrastructure and 2012
Smallholder Producers' Project technologies

Purchase for Progress P4P Tanzania (TZA) Arranged or curated offers from buyers; Production‐management contracts; 2009
Storage deposit systems

Purchase for Progress P4P Uganda (UGA) Storage deposit systems 2008

Program d'Appui au Développement Rural dans le PADER‐G Chad (TCD) Domestic transport infrastructure; storage deposit systems 2011
Guéra

Purdue Improved Crop Storage PICS Tanzania (TZA) Improved on‐farm storage infrastructure 2017

Scientific Grain Storage Project SGSP China (CHN) Storage deposit systems N/A

Solar‐Powered Cold‐Storage SPCP Nigeria (NGA) Storage deposit systems 2020

MARION
ET AL.

18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 147 of 191

T A B L E 34 Descriptions of other interventions combined with storage infrastructure and technology interventions.
Intervention (country Description of other intervention components combined
code)a Type with storage infrastructure interventions

AFP (MWI) Improved on‐farm storage infrastructure and technologies; Support to improve farmers' access to input markets—
Arranged or curated offers from buyers particularly credit and seed markets

ARMA66 (ARM) Arranged or curated offers from buyers; Storage deposit Establish cooperatives, provide technical support during the
systems; Provision of output market information (mobile and vegetable production cycle, link cooperatives with a non‐
internet) profit micro‐finance institute providing zero‐interest loans,
increase awareness of beneficiary farmers on market prices
via mobile SMS

FPPM (COD) Domestic transport infrastructure; Improved on‐farm storage Radio broadcasts to producers, rehabilitation of feeder
infrastructure and technologies; Provision of output market roads or river links
information (other forms of communication)

GUAB38 (GTM) Arranged or curated offers from buyers; Storage deposit Partnership with a private‐sector exporter to support
systems producers to produce on a commercial scale

HVAP (NPL) Arranged or curated offers from buyers; Improved on‐farm Link producers with input suppliers and traders along the
storage infrastructure value chain, form cooperatives (collectively called POs),
provide production and marketing training

MVP (GHA) Domestic transport infrastructure; Storage deposit systems Construction and rehabilitation of roads

OILCSS (TZA) Alternative physical marketplaces; Improved on‐farm storage Strengthen market links, provide tools and training, and
infrastructure and technologies establish market information and credit facilities

P4P (TZA) Arranged or curated offers from buyers; Production‐ Develop viable warehouse receipt systems, support
management contracts; Storage deposit systems rehabilitation of warehouses, purchase of 26,182 metric
tons of maize and pulses from P4P smallholders
a
ISO3 country code provided in parentheses (). If the introduction of an intervention is staggered, the year of intervention refers to the earliest year
observations were undertaken for the intervention.

Summary of the meta‐analyses results of improved storage infrastructure

715 estimates, of which 125 have been selected for the analysis of
and technologies interventions. Table 35 displays the meta‐analyses
effects. We report additional information on the individual meta‐
results across outcome types covered in the subsequent sections.
analyses (sensitivity analysis, outlier analysis, publication bias analy-
sis, moderator analysis, meta‐analyses without high risk of bias
Agricultural production outcomes. The agricultural production outcomes
estimates) in Supporting Information: Appendix 5 for Chapter 5.
are yields, volume, produce type, quality, and crop loss. The included
studies provided evidence on all these outcomes.
Risk of bias. Overall, we identified 58 estimates with a high risk of
bias (46%), 48 estimates raising some concerns (38%), and 19
Crop loss. The crop loss outcome type measures post‐harvest losses.
estimates with a low risk of bias (15%) (Supporting Information:
The four estimates under this outcome type are sourced from four
Appendix 10 for background and methods sections).
studies focusing on interventions initiated between 2011 and 2020.
Despite the overall trends, Figure 102 presents a summary of
These studies focused on storage interventions in Africa (n = 2),
results from our risk of bias assessments of all included 1059
Europe and Central Asia (n = 1), and South Asia (n = 1). Examples
estimates. We found that 59% of included estimates were rated as
include Luo and colleagues (2022), who analyzed the effect of SGSP
having a high risk of bias, 31% were considered to have some
in China. Crops under this outcome type include potatoes, tomatoes,
concerns related to the risk of bias, and the remaining estimates (9%)
maize, grains, sorghum, groundnuts, and cassava. One of the four
were assessed as having a low risk of bias.
evaluated interventions was a multi‐component intervention
Figure 102 shows a discrepancy between the risk of bias of QED
(ARMA66), which included many other project activities such as the
studies compared to RCT studies, since 51% of the former are rated
establishment of cooperatives and the provision of technical,
as high risk of bias, compared to 32% for the latter. In addition, no
marketing, and business support. All interventions focused on the
QED estimates were assessed as having a low risk of bias. In both
establishment and implementation of storage deposit systems.
study designs, spillovers and contamination are a main potential
source of bias, primarily due to the risk of the control group living in Outcome variables used here are quantity of post‐harvest loss
the same village as the beneficiary group and thus possibly (Cavatassi et al., 2018; Luo et al., 2022). Takeshima et al. (2021) and
benefitting from the intervention; such intra‐village spillovers might Lombardini (2017) used the value of post‐harvest losses. All analysis
result in an underestimation of the treatment effect. is at the household level. The effect sizes presented are all reversed
18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
ET AL.
MARION

Distribution of outcome types.

F I G U R E 100
|148 of 191
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 149 of 191

F I G U R E 101 Distribution of the length of studies' follow‐up periods.

such that positive numbers indicate a benefit to the treatment group systems, except GBY and PADER‐G, which were improved on‐farm
compared to the control group, meaning that a higher effect storage infrastructure interventions.
represents lower crop loss. Outcome variables measure agricultural productivity of the most
Improved storage infrastructure and technologies interventions prevalent crops, or crops targeted by the project (soy in Maertens
reduced crop loss of program beneficiaries compared to non‐ et al., 2020; maize in Fuller, 2013). Barnett and colleagues (2018)
beneficiaries. The overall weighted average effect is positive, created a measure of agricultural productivity combining all
ˆ = 0.27 ; 95% CI: 0.04 to
moderate, and statistically significant (SMD harvested crops. All analyses here are at the household level except
0.50; p = 0.02) (Figure 103). Effect sizes vary from 0.01 to 0.45 and Chatterjee (2018), who conducted an analysis at the district level in
there is substantial heterogeneity among the studies (Q (3) = 21.09, India.
p < 0.001, τ̂ 2 = 0.04 , I2 = 85.78%). In Nigeria, Takeshima and Improved storage infrastructure and technologies interventions
colleagues (2021), reported a positive, moderate, and statistically increased yields. The overall weighted average effect is positive,
significant effect of SPCP on the share of loss value to total gross ˆ = 0.15; 95% CI: 0.04 to
small, and statistically significant (SMD
revenue (ĝ = 0.29, 95% CI: 0.14 to 0.44, p < 0.001). 0.26; p = 0.01) (Figure 104). Effect sizes vary from 0.01 to 0.45
In Armenia, Lombardini (2017) found a positive, large, and standard deviations and there is substantial heterogeneity
statistically significant effect of ARMA66 on the value of crops lost among effects (Q (4) = 13.10 , p = 0.01, τ̂ 2 = 0.01, I2 = 69.45%).
(ĝ = 0.39, 95% CI: 0.05 to 0.73, p < 0.001). Finally, in China, Luo In Chad, Cavatassi and colleagues (2018) reported a positive,
and colleagues (2022) observed a positive, large, and statistically small, and statistically significant effect of PADER‐G on the harvest
significant effect of SGSP on storage losses in kilograms of grains per hectare (ĝ = 0.19, 95% CI: 0.09 to 0.30, p < 0.001).
(ĝ = 0.45, 95% CI: 0.26 to 0.63, p < 0.001). None of the estimates In India, Chatterjee (2018) reported a positive, large, and
was assessed as high risk of bias. statistically significant effect of GBY on rice yield (ĝ = 0.45,
95% CI: 0.15 to 0.75, p = 0.004 ). Finally, in Ghana, Barnett and
Yields. The five estimates measuring the effect of storage interven- colleagues (2018) identified a positive, small, and statistically
tions on yields are extracted from five studies focusing on interven- significant effect of MVP on agricultural productivity of all crops
tions initiated between 2010 and 2014. These studies focused on (ĝ = 0.18, 95% CI: 0.09 to 0.27, p < 0.001).
storage interventions in Africa (n = 3), Latin America and the The moderator analysis indicated that more recent interventions
Caribbean (n = 1), and South Asia (n = 1). For example, Maertens seem to be realizing smaller effects. Three out of the five estimates
and colleagues (2020) analyzed the effect of the Anchor Farm Model are assessed as high risk of bias. If we focus our analysis exclusively
of the Clinton Development Initiative and its provision of on‐farm on estimates without a high risk of bias, two remain (Cavatassi et al.,
storage technologies. In India, Chatterjee (2018) focused on the 2018; Barnett et al., 2018). The overall weighted average
effect of the GBY program providing scientific storage facilities for effect decreases and is no longer statistically significant
good grain production. Three studies evaluated multi‐component ˆ = 0.14, 95% CI: −0.21 to 0.50; p = 0.43), but this result
(SMD
interventions (AFP, GUAB38, and MVP Ghana). All interventions should be interpreted with caution given the small number of
focused on the establishment and implementation of storage deposit estimates contributing to this analysis.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
150 of 191 | MARION ET AL.

F I G U R E 102 Risk of bias of included estimates between quasi‐experimental design (QED) (top) and randomized controlled trial (RCT)
(bottom).

Volume. The volume outcome type measures the total quantity or Outcome variables measure the agricultural production volume
volume of agricultural production. The six estimates under this of the most prevalent crops or those targeted by the project (cereals
outcome type are sourced from seven studies focusing on interven- in Biggeri et al., 2022; milk in Breza et al., 2021; maize in Fuller, 2013
tions initiated between 2007 and 2013. These studies focused on and Le Cotty et al., 2019; sunflower in Anguko & Rana, 2017). The
storage interventions in Africa (n = 3), Europe and Central Asia (n = 1), other studies created a measure of agricultural production volume
South Asia (n = 1), and Latin America and the Caribbean (n = 1). combining all harvested crops. All analyses here are at the household
Three studies evaluated multi‐component interventions level, except Breza and colleagues (2012), who conducted their
(ARMA66, GUAB38, and OILCSS). Three focused on the establish- analysis at the village level in India.
ment and implementation of storage deposit systems (AVCPO, Improved storage infrastructure and technologies interventions did
GUAB38, FARMAF, and ARMA66), and one was an improved on‐ not affect volume outcomes. The overall weighted average effect is
farm storage infrastructure intervention (OILCSS). Another implemen- ˆ = 0.10 ; 95% CI:
positive, small, and statistically insignificant (SMD
ted an improved storage technologies for transit intervention (BMC). −0.12 to 0.31; p = 0.37) (Figure 105). Of the six estimates, three are
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 151 of 191

T A B L E 35 Summary of meta‐analyses results.

Outcome types Outcome subcategory SMD 95% CI k p Value

Immediate outcomes Transaction costs N/A

Use/adoption of storage practices 0.19 0.06, 0.31 10 0.03

Use/adoption of fertilizers, pesticides, herbicides, and other improved 0.05 −0.01, 0.12 8 0.12
inputs

Use/adoption of improved varieties and quality seeds 0.05 −0.06, 0.15 3 0.39

Use/adoption of labor and machinery 0.22 −0.01, 0.45 7 0.06

Farm investment (total) N/A

Farm investment (seeds) N/A

Farm investment (fertilizers) N/A

Farm investment (labor and machinery) N/A

Credit 0.03 −0.05, 0.11 7 0.48

Prices 0.11 −0.04, 0.26 8 0.14

Agricultural Production outcomes Crop loss 0.27 0.04, 0.50 4 0.01

Yields 0.15 0.04, 0.26 5 0.01

Volume 0.10 −0.12, 0.31 6 0.37

Quality N/A

Produce type (area) 0.00- −0.11, 0.10 5 0.93

1

Produce type (livestock) N/A

Produce type (diversity) N/A

Intermediate outcomes Sales 0.21 0.11, 0.32 13 <0.001

Farm income 0.10 0.02, 0.18 10 0.02

Non‐farm outcomes Non‐farm income N/A

Welfare outcomes Total income and wealth 0.12 0.03, 0.20 8 0.01

Food security 0.09 0.02, 0.16 12 0.01

Nutrition 0.05 −0.01, 0.10 4 0.09

Note: Outcome subcategories in bold have a statistically significant SMD at a 5% significance level. K is the number of estimates. N/A indicates that we did
not conduct a meta‐analysis for these outcome types. In these cases, k = 1 or k = 2 and we chose not to conduct a meta‐analysis in those cases due to
differences (e.g., intervention types, regions).

F I G U R E 103 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of improved storage
infrastructure and technologies on crop loss.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
152 of 191 | MARION ET AL.

F I G U R E 104 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of improved storage
infrastructure and technologies on yields.

F I G U R E 105 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of improved storage
infrastructure and technologies on volume.

assessed as high risk of bias. Effect sizes vary from −0.28 to 0.51 They include financial resources (e.g., transaction costs, credit),
standard deviations. The heterogeneity among the studies is market linkages and bargaining power (group participation, prices),
substantial (Q (5) = 47.18, p < 0.001, τ̂ 2 = 0.06, I2 = 89.40%). and the adoption of new practices and agricultural approaches (use/
In Ethiopia, Biggeri and colleagues (2022) found a positive, large, adoption of improved inputs, technologies and practices, farm
and statistically significant effect of AVCPO on the growth of investments). There is no evidence on group participation outcomes.
cereal production values between 2013 and 2010 (ĝ = 0.51,
95% CI: 0.35 to 0.67, p < 0.001). In India, Breza and colleagues Use/adoption of storage practices. The ten estimates measuring the
(2012) observed a positive, moderate, and statistically significant effect effect of storage interventions on the use/adoption of storage practices
of BMC on the volume of milk per month (ĝ = 0.20, 95% CI: 0.07 to are extracted from ten unique studies. These studies examined
0.34, p = 0.003). In Guatemala, Fuller (2013) reported a negative, storage interventions in Africa (n = 8), Europe and Central Asia (n = 1),
moderate, and statistically significant effect of GUAB38 on the and East Asia and the Pacific (n = 1); they all focus on the introduction
quantity of maize produced in 2011 (ĝ = −0.28, 95% CI: −0.49 to of storage deposit systems. For example, Casaburi and colleagues
−0.08, p = 0.006). (2014) focused on community storage rehabilitation with extra palm
The moderator analysis indicated that additional months of oil containers and marketing support in Sierra Leone, while Kizito and
exposure to the intervention was associated with greater effects. Of Kato (2018) analyzed a component of the P4P27 initiative in Uganda,
the six estimates, three are assessed as high risk of bias. If we focus including the construction of stores in rural areas with an increased
our analysis exclusively on estimates without a high risk of bias, storage capacity of 100 to 300 metric tons. Two studies evaluated
three remain (Lombardini, 2017; Le Cotty and colleagues, multi‐component interventions (ARMA66, P4P in Uganda). Eight
2019; Biggeri and colleagues, 2022). The overall weighted interventions focused on the establishment and implementation of a
average effect increases but is still statistically insignificant storage deposit system, and two interventions entailed improved
ˆ = 0.14, 95% CI: −0.21 to 0.50; p = 0.43).
(SMD on‐farm storage infrastructure and technologies.

Immediate outcomes. The immediate outcomes category covers the 27

Information on interventions is provided in Supporting Information: Appendix 3 for
first outcomes that may affect farmers across the causal chain. Chapter 1.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 153 of 191

F I G U R E 106 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of improved storage
infrastructure and technologies on use/adoption of storage practices.

Outcome variables here relate to the use or adoption of storage interventions on the adoption of fertilizers, pesticides, herbicides, and
practices measures. Delavallade and Godlonton (2020) and Le Cotty other improved inputs are based on seven unique studies focusing on
and colleagues (2019) used a binary variable indicating whether interventions initiated between 2011 and 2020. These studies
respondents stored their harvest. In Luo and colleagues (2022) and focused on storage interventions in Africa (n = 5), South Asia (n = 1),
Cavatassi and colleagues (2018), the outcome variation was the and East Asia and the Pacific (n = 1).
duration of storage among respondents. The remaining studies Examples of studies under this outcome include work from
measured the quantity stored by respondents. All analyses here are Chatterjee (2018) in India analyzing the effect of the GBY program
at the household level. providing storage facilities to avoid deterioration and wastage of food
Improved storage infrastructure and technologies interventions grain. We found two independent estimates in Ragasa and colleagues
increased the use and adoption of storage practices. The overall (2016), as the analysis is conducted in two provinces of DRC
weighted average effect is positive, small, and statistically significant (Bandundu and Kinshasa), and each analysis includes an independent
ˆ = 0.19, 95% CI: 0.06 to 0.31; p = 0.003) (Figure 106). Effect
(SMD farmer control group. Three studies evaluated multi‐component
sizes vary from −0.06 to 0.65 and the heterogeneity among the interventions (AFP, FPPM, and MVP Ghana). All interventions
estimates is substantial (Q (9) = 68.37, p < 0.001, τ̂ 2 = 0.033, focused on the establishment and implementation of a storage
I2 = 86.84 %). In Uganda, Kizito and Kato (2018) observed a positive, deposit system, except GBY and FPPM, which were improved on‐farm
large, and statistically significant effect of the P4P initiative on the storage infrastructure interventions.
quantity stored ( gˆ = 0.56, 95% CI: 0.34 to 0.77; p < 0.001). Outcome variables measured the use of fertilizer (binary variable)
Similarly, in Chad, Cavatassi and colleagues (2018) in all studies except that of Le Cotty and colleagues (2019), which
reported a small, positive, and statistically significant effect of measured the quantity of fertilizer used. All analyses here are at the
PADER‐G on the storage duration of grains in months household level, except Chatterjee (2018), who conducted an
( gˆ = 0.14, 95% CI: 0.04 to 0.25; p < 0.001). Finally, in Armenia, analysis at the district level in India.
Lombardini (2017) reported a large, positive, and statistically Improved storage infrastructure and technologies interventions did
significant effect of ARMA66 on the prevalence of cold storage not affect the use/adoption of fertilizers, pesticides, herbicides, and other
( gˆ = 0.65, 95% CI: 0.46 to 0.84; p < 0.001). improved inputs. The overall weighted average effect is statistically
We identified potential funnel plot asymmetry, suggesting that ˆ = 0.05; 95% CI: −0.01 to 0.12; p = 0.12)
insignificant (SMD
there may be publication bias. The moderator analysis indicated (Figure 107). Effect sizes vary from −0.15 to 0.29 standard deviations
that effects were greater in multi‐component interventions relative and the heterogeneity is moderate (Q (7) = 13.80 , p = 0.06,
to single‐component interventions. Two of the ten estimates are τ̂ 2 = 0.004, I2 = 49.29%). In Nigeria, Takeshima and colleagues
assessed as having a high risk of bias. If we focus our (2021) reported a positive, small, and statistically significant effect
analysis exclusively on estimates without a high risk of bias, eight of SPCP on the application of pesticides and chemicals before sales
estimates remain, and the overall weighted average effect (gˆ = 0.09, 95% CI: 0.02 to 0.17; p = 0.02). In Burkina Faso, Le Cotty
decreases slightly and remains statistically significant and colleagues (2019) observed a positive, small, and statistically
ˆ
(SMD = 0.15, 95% CI: 0.02 to 0.28; p = 0.02). significant effect of FARMAF on the quantity of fertilizers used
(in kilograms) ( gˆ = 0.15, 95% CI: 0.01 to 0.29; p = 0.04 ).
Use/adoption of fertilizers, pesticides, herbicides, and other improved In India, Chatterjee (2018) reported a positive, moderate,
inputs. The eight estimates measuring the effect of storage and statistically significant effect of the GBY program on
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
154 of 191 | MARION ET AL.

the total use of nitrogen, potash, and phosphate fertilizers Use/adoption of machinery and labor. The seven estimates measuring
( gˆ = 0.29, 95% CI: 0.02 to 0.58; p = 0.04 ). Finally, in Ghana, Barnett the effect of interventions on the adoption of inputs are extracted
and colleagues (2018) reported a positive, small, and statistically from five unique studies focusing on interventions initiated between
significant effect of MVP on the use of protective chemicals during 2010 and 2012. These studies focused on improved storage
storage (gˆ = 0.10, 95% CI: 0.007 to 0.19; p = 0.04 ). interventions in Africa (n = 3), Europe and Central Asia (n = 1), and
The moderator analysis indicated that the use of control variables Latin America and the Caribbean (n = 1). In addition to Casaburi and
in estimating the effect in included studies was associated with colleagues (2014) and Ragasa and colleagues (2016), Fuller (2013)
significantly lower effect sizes. Four of the estimates are assessed as focused on GUAB38, which constructed storage centers and seedling
having a high risk of bias. If we focus our analysis exclusively on banks. We selected these estimates from Ragasa and colleagues'
estimates without a high risk of bias, four estimates remain (Barnett (2016) volume outcome analysis.
and colleagues, 2018; Le Cotty and colleagues, 2019; Luo and Anguko and colleagues (2017) analyzed the effect of OILCSS,
colleagues, 2022; Takeshima and colleagues, 2021). The overall which improved on‐farm storage infrastructure. Finally, Lombardini
weighted average effect increases slightly and becomes statistically (2017) evaluated the effect of ARMA66, which constructed four
ˆ = 0.09, 95% CI: 0.02 to 0.16; p = 0.02).
significant (SMD community greenhouses and established three cold‐storage units
for crops.
Use/adoption of improved varieties and quality seeds. We selected Four studies evaluated multi‐component interventions
three estimates measuring the effect of interventions on the (ARMA66, FPPM, GUAB38, and OILCSS). All interventions focused
adoption of improved varieties and quality seeds. We selected these on the establishment and implementation of storage deposit systems,
estimates from one study: Ragasa and colleagues (2016) volume except OILCSS and FPPM, which were improved on‐farm storage
outcome analysis of the FPPM initiative in Bandundu, Bas‐Congo, and infrastructure interventions.
Kinshasa. FPPM was a multi‐component intervention focused on the Outcome variables measured the adoption of labor and machin-
establishment and implementation of improved on‐farm storage ery inputs. For example, Ragasa and colleagues (2016) employed a
infrastructure. The authors used a binary variable indicating binary variable indicating whether households had access to
whether households adopted improved cassava varieties as their mechanization for land preparation. In Lombardini (2017), the
outcome variable. outcome variable was a binary variable indicating whether they used
Improved storage infrastructure and technologies interventions did a drip irrigation system. All analyses here are at the household level.
not affect the use or adoption of improved varieties and quality seeds. Improved storage infrastructure and technologies interventions did
The overall weighted average effect is positive, very small, and not affect the use or adoption of labor and machinery. The overall
ˆ = 0.05; 95% CI: −0.06 to 0.15;
statistically insignificant (SMD weighted average effect is positive and moderate but statistically
p = 0.39) (Figure 108). Effect sizes vary from 0.04 to 0.09 standard ˆ = 0.22; 95% CI: −0.01 to 0.45; p = 0.06)
insignificant (SMD
deviations and there was no significant amount of heterogeneity (Figure 109). Effect sizes vary from −0.06 to 0.90 standard deviations
among the studies (Q (2) = 0.09, p = 0.96, τ̂ 2 = 0.000 , I2 = 0.00 %), and there is considerable heterogeneity across included estimates
which is unsurprising given that all estimates are evaluations of the (Q (6) = 84.18, p < 0.001, τ̂ 2 = 0.09, I2 = 92.87%).
same intervention in the same country. As this outcome type only In Tanzania, Anguko and Rana (2017) reported a small, positive,
includes high risk of bias estimates, we did not run sensitivity and statistically significant effect of OILCSS on the average number
analyses by study quality. of agricultural tools accessed and used by households

F I G U R E 107 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of improved storage
infrastructure and technologies on use/adoption of fertilizers, pesticides, herbicides, and other improved inputs.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 155 of 191

F I G U R E 108 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of improved storage
infrastructure and technologies on use/adoption of improved varieties and quality seeds.

F I G U R E 109 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of improved storage
infrastructure and technologies on use/adoption of labor and machinery.

( gˆ = 0.17, 95% CI: 0.01 to 0.32; p = 0.03). In Armenia, Lombardini All interventions focused on the establishment and implementation
(2017) found a large, positive, and statistically significant effect of of a storage deposit system, except FPPM, which was an improved on‐
the ARMA66 initiative on the prevalence of drip irrigation systems farm storage infrastructure intervention. Outcome variables measure
( gˆ = 0.90, 95% CI: 0.70 to 1.09; p < 0.001). Finally, in Guatemala, access to and use of credit. All studies used a binary variable except
Fuller (2013) reported a large, positive, and statistically significant Cavatassi and colleagues (2018) and Lombardini (2017), who
effect of GUAB38 on the use of improved agricultural techniques employed the number of loans and the value of credit, respectively.
( gˆ = 0.37, 95% CI: 0.16 to 0.57; p < 0.001). All analyses here are at the household level.
None of the moderator variables significantly contributed Improved storage infrastructure and technologies interventions did not
to the variation in estimates. All estimates but two (Casaburi et al., affect access to and use of credit. The overall weighted average effect is
2014; Lombardini, 2017) are assessed as high risk of bias. ˆ = 0.03; 95%
very small and positive but statistically insignificant (SMD
If we remove high risk of bias estimates, the overall weighted CI: −0.05 to 0.11; p = 0.48) (Figure 110). Effect sizes vary from −0.15
average effect is large, but remains statistically insignificant to 0.17 standard deviations and there is substantial heterogeneity
ˆ = 0.41, 95% CI: −0.51 to 1.35; p = 0.38). However, this result
(SMD among the estimates (Q (6) = 13.32, p = 0.04 , τ̂ 2 = 0.006,
should be viewed with caution due to the small number of estimates I2 = 54.96%). Through COPSA‐C's warrantage system in Burkina Faso,
contributing to this analysis. smallholder farmers can deposit their grains in a village warehouse as
collateral for a loan of up to 80% of the grain value (Delavallade &
Credit. We identified seven estimates from five studies measuring Godlonton, 2020). However, the effect here is statistically insignificant
farmers' access to or use of credit, including Ragasa and colleagues (ĝ = 0.14, 95% CI: −0.03 to 0.32, p = 0.10 ). Other interventions did
(2016) on the FPPM project in DRC, Cavatassi and colleagues (2018) not entail similar access to credit.
on PADER‐G in Chad, and Lombardini (2017) on ARMA66 in None of the moderator variables significantly contributed
Armenia. These studies focused on storage interventions in Africa to the variation in estimates. If we focus our analysis exclusively
(n = 4) (including three estimates from Ragasa and colleagues [2016] on estimates without a high risk of bias, four remain. The overall
in DRC), and Europe and Central Asia (n = 1). Three studies evaluated weighted average effect is still small and statistically
multi‐component interventions (ARMA66, FPPM, and MVP Ghana). ˆ = 0.05, 95% CI: −0.07 to 0.18; p = 0.40 ).
insignificant (SMD
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
156 of 191 | MARION ET AL.

Prices. We identified eight estimates from eight studies measuring However, in Nigeria, Takeshima and colleagues (2021) observed a
farm‐gate prices (those effectively received by farmers for their large, positive, and statistically significant effect of SPCP on the sales
products). Prices are measured several a few different ways: through price per kilogram (ĝ = 0.39, 95% CI: 0.31 to 0.47, p < 0.001). Simi-
farmers' reports of prices, market surveys, and records of market larly, in Kenya, Aggarwal and colleagues (2018) report a moderate,
transactions. One study from Kizito and Kato (2018) analyzed the positive, and statistically significant effect of GSRA on the log average
effect of the P4P initiative in Uganda building stores with storage price of crops (ĝ = 0.20, 95% CI: 0.04 to 0.36, p = 0.01). Finally, in
capacity of 100 to 300 metric tons. Another study by Maertens and Uganda, Kizito and Kato (2018) found a positive, moderate, and
colleagues (2020) in Malawi evaluated the effects of the Anchor Farm statistically significant effect of P4P on the prices received for maize
Model of the Clinton Development Initiative, which provided on‐farm (ĝ = 0.30, 95% CI: 0.09 to 0.51, p = 0.005).
storage technologies. The moderator analysis indicated that additional months of exposure
All estimates focus on Africa except Kafle and colleagues (2022), to the intervention was associated with greater effects. Five of the
who examined an initiative providing cold storage facilities in major estimates are assessed as high risk of bias. If we focus our analysis
market outlets in Nepal. Four studies evaluated multi‐component exclusively on estimates without a high risk of bias, three remain. The
interventions (AFP, HVAP, OILCSS, and P4P Uganda). All interven- overall weighted average effect increases but remains statistically
tions focused on the establishment and implementation of a storage ˆ = 0.16, 95% CI: −0.11 to 0.44; p = 0.24 ). How-
insignificant (SMD
deposit system, except OAF, OILCSS, HVAP, and AFP, which were ever, this result should be interpreted with caution given the small
improved on‐farm storage infrastructure interventions. number of estimates contributing to this analysis.
Outcome variables here measure the sales price specifically for
the most prevalent crops or those targeted by the project (soy in Intermediate outcomes. The intermediate outcomes category covers
Maertens et al., 2020; sunflower in Anguko & Rana, 2017; sorghum in the outcomes that may affect farmers following the harvest period
Cavatassi et al., 2018; maize in Kizito & Kato, 2018). The other through the sale of their agricultural products. They cover two
studies used an index of sales prices received across all crops. All outcome types: sales and farm income.
analyses here are at the household level, except Takeshima and
colleagues (2021), who conducted their analysis at the market level in Sales. The sales outcome measures the proportion, number, and
Nigeria. volume of produce sold by farmers. We found 13 estimates sourced
Improved storage infrastructure and technologies interventions do from 13 studies focusing on interventions initiated between 2008
not affect prices. The overall weighted average effect is statistically and 2020. These studies focused on storage interventions in Africa
insignificant (SMDˆ = 0.11; 95% CI: −0.04 to 0.26; p = 0.14 ) (n = 11), South Asia (n = 1), and Latin America and the Caribbean
(Figure 111). Effect sizes vary from −0.11 to 0.39 standard deviations (n = 1). Among the included studies, Lentz and Upton (2016)
and there is considerable heterogeneity across included estimates evaluated P4P in Tanzania, which supported the rehabilitation of 23
(Q (7) = 94.15, p < 0.001, τ̂ 2 = 0.04 , I2 = 92.57%). warehouses. Aggarwal and colleagues (2018) focused on GSRA,
The only study reporting a statistically significant negative effect which encouraged members to set aside maize together in communal
is Kafle and colleagues (2022), who reported a very small negative bags stored at a single member's house. Six studies evaluated multi‐
effect of HVAP in Nepal on the price index for project commodities— component interventions (AVCPO, GUAB38, HVAP, OILCSS, and
this can be interpreted as a reduction in the selling price of apple, P4P Uganda). Nine interventions focused on the establishment and
ginger, turmeric, timur, off‐season vegetables, vegetable seeds, and implementation of a storage deposit system, and four interventions
meat goats (ĝ = −0.08, 95% CI: −0.15 to − 0.002, p = 0.04 ). were improved on‐farm storage infrastructure interventions.

F I G U R E 110 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of improved storage
infrastructure and technologies on credit.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 157 of 191

F I G U R E 111 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of improved storage
infrastructure and technologies on prices.

F I G U R E 112 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of improved storage
infrastructure and technologies on sales.

Outcome variables are measures of the proportion of produce sales (ĝ = 0.57, 95% CI: 0.35 to 0.78, p = 0.01). In Ethiopia, Biggeri
sold, the volume of produce sold by rural households, or binary and colleagues (2022) observed a positive effect of AVCPO
variables indicating whether households sold any harvested produc- on the share of cereal production sold through cooperatives
tion. All analyses here are at the household level. Crops under this (ĝ = 0.82, 95% CI: 0.66 to 0.96, p < 0.001).
outcome type include palm oil, maize, vegetables, grains, sunflower, We found that estimates based on IV methods were larger than
and horticultural products. those based on DID and FE. Similarly, effects were greater in multi‐
Improved storage infrastructure and technologies interventions component interventions relative to single‐component interventions.
increased sales. The overall weighted average effect is positive, We did not identify any funnel plot asymmetry, meaning that
ˆ = 0.21; 95% CI: 0.11 to
moderate, and statistically significant (SMD publication bias is not likely to be present. Of the 13 estimates, six
0.32; p < 0.001) (Figure 112). Effect sizes vary from −0.01 to 0.82 are assessed as high risk of bias. If we focus exclusively on estimates
standard deviations and there is substantial heterogeneity across without a high risk of bias, seven estimates remain. The overall
included estimates (Q (12) = 99.31, p < 0.001, τ̂ 2 = 0.03, weighted average effects are similar and still statistically signifi-
I2 = 87.92%). The meta‐analysis includes three large, positive, and ˆ = 0.20, 95% CI: 0.04 to 0.36; p = 0.02).
cant (SMD
statistically significant estimates.
In Tanzania, Anguko and Rana (2017) observed a positive Farm income. Ten estimates, sourced from ten studies, measure
effect of OILCSS on the quantity of sunflower sold in the previous income generated from farm activities. These studies focused
12 months (ĝ = 0.39, 95% CI: 0.07 to 0.70, p = 0.02). In Uganda, primarily on storage interventions in Africa (n = 8); one study was
Kizito and Kato (2018) reported a positive effect of P4P on maize implemented in Nepal, where Kafle and colleagues (2022) worked on
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
158 of 191 | MARION ET AL.

an initiative providing cold storage facilities in major market outlets, examined GUAB38, which constructed storage centers and seedling
and another was implemented in Armenia (Lombardini 2017). Other banks. Anguko and colleagues (2017) analyzed the effects of OILCSS,
examples include work from Barnett and colleagues (2018) in Ghana, which improved on‐farm storage infrastructure. Six studies evaluated
where the MVP supported farmer‐based organizations to collectively multi‐component interventions (ARMA66, GUAB38, HVAP, MVP
store their agricultural products. Six studies evaluated multi‐ Ghana, and P4P Tanzania). All interventions focused on the
component interventions (AFP, ARMA66, HVAP, MVP Ghana, establishment and implementation of a storage deposit system, except
OILCSS, and P4P Tanzania). Six interventions focused on the OILCSS and HVAP, which were improved on‐farm storage infra-
establishment and implementation of a storage deposit system. Four structure interventions.
interventions were improved on‐farm storage infrastructure The outcomes related to household‐level total income, per capita
interventions. income, or adult‐equivalent income (k = 3). Lombardini (2017) and
Outcome variables measured revenues from agricultural produc- Cavatassi and colleagues (2018) employed a wealth index based on
tion of all crops except those of Takeshima and colleagues (2021), household assets. The other outcome variables are household‐level
Maertens and colleagues (2020), Aggarwal and colleagues (2018), and total expenditure, per capita expenditure, or adult‐equivalent
Cavatassi and colleagues (2018), which focused on revenues from expenditure.
horticultural products, soy, maize, and grains, respectively. All Improved storage infrastructure and technologies interventions im-
included analyses are at the household level. proved income and wealth outcomes. The overall weighted average effect
Improved storage infrastructure and technologies interventions ˆ = 0.12; 95% CI: 0.03
is positive, small, and statistically significant (SMD
increased farm income. The overall weighted average effect is positive to 0.20 ; p = 0.01) (Figure 114). Effect sizes vary from −0.19 to 0.26
ˆ = 0.10 ; 95% CI: 0.02 to
and small but statistically insignificant (SMD standard deviations. There is substantial heterogeneity across included
0.18, p = 0.02) (Figure 113). Effect sizes vary from −0.22 to 0.37 estimates (Q (7) = 23.004, p = 0.002, τ̂ 2 = 0.01, I2 = 69.57%). In
standard deviations and there is substantial heterogeneity across Tanzania, Lentz and Upton (2016) observed a negative, small, and
included estimates (Q (9) = 39.77, p < 0.001, τ̂ 2 = 0.01, I2 = 77.37%). statistically significant effect of P4P on the log income
In Tanzania, Lentz and Upton (2016) observed a negative, (ĝ = −0.19, 95% CI: −0.37 to − 0.01, p = 0.04). In the same country,
moderate, and statistically significant effect of P4P on the log Anguko and Rana (2017) observed a positive, moderate, and statistically
of crop value (ĝ = −0.22, 95%CI: −0.41 to − 0.02, p = 0.03). In significant effect of OILCSS on the total household consumption per
Nigeria, Takeshima and colleagues (2021) reported a small, positive, adult‐equivalent per day (ĝ = 0.26, 95% CI: 0.11 to 0.42, p < 0.001).
and statistically significant effect of SPCP on the gross None of the moderator variables significantly contributed to the
revenue percentage increase for farmers in USD per week variation in estimates. Of the eight estimates, four are assessed as high
( ĝ = 0.16, 95% CI: 0.01 to 0.31, p = 0.03). In Nepal, Kafle and risk of bias. If we focus our analysis exclusively on estimates without a
colleagues (2022) reported the positive, moderate, and statistically high risk of bias, four estimates remain (Lombardini, 2017; Delavallade
significant effect of HVAP on the log of agricultural income & Godlonton, 2020; Cavatassi et al., 2018; Barnett et
( gˆ = 0.26, 95% CI: 0.19 to 0.34, p < 0.001). In Tanzania, Anguko al., 2018). The overall weighted average effect without high risk of
and Rana (2017) reported a large, positive, and statistically bias estimates is still statistically significant, and the magnitude remains
significant effect of OILCSS on farming income sources the same (SMDˆ = 0.12, 95% CI: 0.03 to 0.21; p = 0.01).
(ĝ = 0.37, 95% CI: 0.14 to 0.60, p = 0.01).
None of the moderator variables significantly contributed to the Food security. The 12 estimates of this outcome type are sourced
variation in estimates. Of the ten estimates, five are assessed as high risk from 12 studies analyzing interventions initiated between 2009 and
of bias. We did not identify any funnel plot asymmetry, meaning that 2015. Aggarwal and colleagues (2018) focused on GSRA, which
publication bias is not likely to be present. If we focus our analysis encouraged members to set aside maize together in communal bags
exclusively on estimates without a high risk of bias, five estimates remain. stored at a single member's house. Anguko and colleagues (2017)
The overall weighted average effect is slightly smaller and still statistically analyzed how OILCSS improved on‐farm storage infrastructure.
ˆ = 0.09, 95% CI: 0.03 to 0.15; p = 0.003).
significant (SMD Finally, Lombardini (2017) evaluated the effect of ARMA66, which
constructed four community greenhouses and established three cold‐
Welfare outcomes. The welfare outcomes category covers measures storage units for crops. Six studies evaluated multi‐component
of improvement in farmers' well‐being and quality of life. They interventions (ARMA66, GUAB38, HVAP, MVP Ghana, and P4P
encompass a range of social, economic, and environmental factors Tanzania). All interventions focused on the establishment and
but in this review, we focus on two: total income and food security. implementation of a storage deposit system, except OAF, OILCSS,
and HVAP, which were improved on‐farm storage infrastructure
Total income and wealth. The eight estimates of this outcome type interventions.
measure total household income and socio‐economic status (e.g., Outcomes were all at the household level and measured the
total household expenditure, asset or wealth indices). They are value of food consumption expenditure (k = 5), household food
sourced from eight studies focusing on interventions initiated insecurity access scale scores (HFIAS, k = 2), binary outcomes on
between 2009 and 2015. These studies include Fuller (2013) who households' lack of food (k = 1), food insecurity experience scale
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 159 of 191

F I G U R E 113 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of improved storage
infrastructure and technologies on farm income.

F I G U R E 114 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of improved storage
infrastructure and technologies on income and wealth.

scores (FIES, k = 1), number of days without food (k = 1), binary We found that estimates based on IV methods were larger
variables measuring whether households reduced food intake during than those based on DID and FE. We did not identify any funnel
planting to afford inputs (k = 1), and number of days with food self‐ plot asymmetry, meaning that publication bias is not likely to be
sufficiency (k = 1). present. Of the 12 estimates, five are assessed as high risk of bias.
Improved storage infrastructure and technologies interventions If we focus our analysis exclusively on estimates without a high
improved food security. The overall weighted average effect is risk of bias, seven estimates remain. When high risk of bias
ˆ = 0.09; 95%
positive, very small, and statistically significant (SMD estimates are removed from the model, the overall weighted
CI: 0.02 to 0.16; p = 0.01) (Figure 115). Effect sizes vary from average is the same size, but becomes statistically
−0.04 to 0.37 standard deviations. There is substantial heteroge- ˆ = 0.09, 95% CI: −0.002 to 0.19; p = 0.05).
insignificant (SMD
neity across included estimates (Q (11) = 34.54 , p < 0.001,
τ̂ 2 = 0.009, I2 = 68.16%). The meta‐analysis includes two estimates Nutrition. The four estimates of this outcome type are sourced from
with a positive and statistically significant effect. In Tanzania, four studies analyzing interventions initiated between 2011 and 2012.
Anguko and Rana (2017) reported a moderate‐sized effect of Cavatassi and colleagues (2018) analyzed the effect of the PADER‐G
OILCSS on food consumption per adult‐equivalent per day project in Chad, which implemented a cereal storage intervention to
( gˆ = 0.33, 95% CI: 0.18 to 0.48, p < 0.001). In Ethiopia, Biggeri support poor rural households and smallholder farmers. Kafle and
and colleagues (2022) reported a large effect of AVCPO on colleagues (2022) worked on a 2011 initiative providing cold‐storage
household food insecurity access scale scores, HFIAS facilities in major market outlets in Nepal. Biggeri and colleagues (2022)
(ĝ = 0.37, 95% CI: 0.21 to 0.53, p < 0.001).28 assessed the impact of AVCPO, which provided a package of
interventions (training, seed provision, storage, processing, and market-
28 ing) to Ethiopian smallholder farmers. Two studies evaluated multi‐
The effect size here is reversed, meaning that a higher effect represents lower food
insecurity. component interventions (HVAP and MVP Ghana).
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
160 of 191 | MARION ET AL.

F I G U R E 115 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of improved storage
infrastructure and technologies on food security.

All interventions focused on the establishment and implementa- to 0.55; p = 0.42). The effect sizes here are all reversed, meaning
tion of a storage deposit system except HVAP, which was an improved that a higher effect represents lower prevalence of underweight
on‐farm storage infrastructure intervention. Outcomes were measures children. This analysis provides limited insight into how improved
of the HDDS (k = 3) and the percentage of underweight children storage interventions affect different types of individual beneficiaries
under five in the village (k = 1) (Barnett et al., 2018). as MVP was a multi‐component intervention and it is unlikely that
Improved storage infrastructure and technologies interventions did this difference was exclusively due to its storage deposit system.
not improve nutrition outcomes. The weighted average effect is not
ˆ = 0.05, 95% CI: −0.01 to 0.11, p = 0.09)
statistically significant (SMD Analysis of effects over time. We aimed to reproduce this same type
(Figure 116). Effect sizes vary from −0.03 to 0.16 standard deviations of analysis but for different periods of intervention exposure. We
and there is no significant amount of heterogeneity across estimates here compare estimates within studies from the same intervention
(Q (3) = 2.93, p = 0.40 , τ̂ 2 = 0.001, I2 = 0.001%). In Chad, Cavatassi across time periods. As above, we only focus on effect sizes in which
and colleagues (2018) observed a positive, small, and statistically the period of exposure to the intervention differs, with everything
significant effect of PADER‐G on the HDDS over the course of a else held constant. However, we do not find effect sizes that are
week (ĝ = 0.11, 95% CI: 0.01 to 0.21, p = 0.04 ). statistically significant in all analyzed periods, or a change in statistical
Of the four estimates, one is assessed as high risk of bias. If we focus significance across the periods analyzed. Barnett and colleagues
our analysis exclusively on estimates without a high risk of bias, three (2018) provided estimates of the effect of MVP in Ghana across
remain (Biggeri and colleagues, 2022, AVCPO in Ethiopia; Cavatassi and different outcomes over different times of exposure to the
colleagues, 2018, PADER‐G in Chad; Barnett & colleagues, 2018, MVP in intervention, but the effects were consistently statistically
Ghana). The overall weighted average effect is still statistically insignificant.
ˆ = 0.07, 95% CI: −0.03 to 0.18; p = 0.20 ).
insignificant (SMD
Barriers and facilitators analysis
Subgroup analysis. We now examine how the effect sizes of improved The following section presents the barriers and facilitators reported in
storage Infrastructure and technologies interventions may vary by our body of evidence. We again remind the reader that the included
subgroup. We compare estimates within studies from the same studies are quantitative impact evaluations. Such studies tend to focus
intervention across subgroups, with only one study providing narrowly on identifying the magnitude of the intervention's effects,
estimates of effects across subgroups (Barnett et al., 2018). Here, without discussing implementation details or rigorously explaining why
we specifically focus on estimates wherein the intervention is the an intervention did or did not result in a particular effect. A thorough
same and only the subgroup is different. analysis of barriers and facilitators would require a dedicated search for
different types of evidence, and this was outside the scope of work for
Gender. Barnett and colleagues (2018) provided estimates of the this project. Of the 27 papers included in this body of evidence, we
effect of MVP in Ghana on the percentage of underweight children identified 13 with relevant information for this subsection.
under five years of age at the village level. The effects are statistically
insignificant for both females ( ĝ = −0.04 ; 95% CI: −0.42 to Success is more likely when the logistics of storage infrastructure
0.34; p = 0.84) and for the overall sample ( ĝ = 0.16; 95% CI: −0.22 construction are well‐organized. The installation of a storage facility,
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 161 of 191

F I G U R E 116 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of improved storage
infrastructure and technologies on nutrition.

or adoption of storage technologies, required a sufficient level of Intervention costs

organization to facilitate the construction, shipping of materials, or Eight of the 27 studies included in this body of evidence provided
access to the storage technology. Logistical organization of the information about the cost of interventions (30%) (Table 36). This is
storage infrastructure and technology then either catalyzed the higher than the norm than in the impact evaluation literature, where
effect of the intervention (if well‐structured and scheduled) or less than 20% of impact evaluations reported any cost information
challenged its implementation. (Brown & Tanner, 2019). The information provided is primarily limited
Casaburi and colleagues (2014) analyzed the effect of two to the total budget or a component of the intervention, and does not
interventions in Sierra Leone: rehabilitation of community storage typically report the value for money of each intervention.
and installation of extra palm oil containers, and implementation of
an inventory credit loan system that gives access to credit based on Discussion
the amount of agricultural products stored. The authors noted that
one of the villages targeted by the inventory credit project could Summary of findings. The body of evidence of studies that rigorously
not identify a suitable storage space, which meant that this and quantitatively examine the effects of storage infrastructure and
component was not implemented in the community. In India, Breza technologies is relatively small compared to the other intervention
and colleagues (2012) also highlighted the importance of logistical categories of this chapter: it includes 27 papers reporting on 20
capacity and support as important considerations for storage interventions through 20 studies covering 15 countries (including 11
interventions. They analyzed the effects of bulk milk freezers that countries in Africa). Of these interventions, eight combined storage
provided remote villages with refrigerating technologies. interventions with other storage components, or with other interven-
tion categories. The most studied storage intervention was storage
Shared storage facilities may benefit from organized and structured deposit systems (n = 11). Included studies primarily employed a QED
farmer groups. Authors also highlighted the benefits of relying on (n = 15) and covered follow‐up periods between two months and ten
structured farmer groups as facilitators of shared storage‐facility years. Eight of the 20 included studies provided information on costs.
implementation. In Uganda, Kizito and Kato (2018) focused on P4P, We found an overall positive, moderate, and often significant
which included the construction of stores in rural areas. The authors effect of storage interventions on outcomes included in the
commented that farmers from Western Uganda tended to store more theory of change of this intervention. Storage interventions
maize than their Eastern counterparts due to their group produce positive, very‐small‐to‐moderate, statistically
organization. significant effects on the adoption of storage practices
A similar comment is made by Aggarwal and colleagues (2018), ˆ = 0.19; 95% CI: 0.06, 0.31; k = 10 ), the reduction of crop
(SMD
who analyzed the effect of GSRA, which includes a product to ˆ = 0.27; 95% CI: 0.04, 0.50; k = 4 ), and the increase in
loss (SMD
facilitate maize storage after the harvest. The authors observed that, ˆ = 0.21; 95% CI: 0.11, 0.32; k = 13), farm
the volume of sales (SMD
following the intervention, GSRA farmers sold later and received ˆ = 0.10; 95% CI: 0.02, 0.18; k = 10), total income and
income (SMD
higher prices for their outputs. They speculated that this larger effect ˆ = 0.12; 95% CI: 0.03, 0.20; k = 8), and food security
wealth (SMD
compared to control groups might also be due to farmers benefitting ˆ = 0.09; 95% CI: 0.02, 0.16; k = 12).
(SMD
from collective bargaining and intermediation. In Nepal, Kafle and In the absence of overall positive effects on adoption of improved
colleagues (2022) observed the same phenomenon when analyzing practices and farm investments, we still observed a small, positive,
the effect of a program supporting producer organizations through and statistically significant effect of the intervention on yields
storage systems. The authors noted that the relatively small size of ˆ = 0.15; 95% CI: 0.04, 0.26; k = 5). This lack of statistically
(SMD
the producer organization allowed the program to engage closely significant effects on adoption and investments may simply be due to
with beneficiaries and facilitate program implementation. the dearth of available evidence on these issues. We also did not
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
162 of 191 | MARION ET AL.

T A B L E 36 Summary of available intervention costs.

Study Acronym Cost information description

Casaburi et al. (2014) SRT Rehabilitation cost per community: USD30

Maertens et al. (2020) AFP Farmer field day cost: USD650 total or USD3.25 per farmer
Marketing component cost: USD120 per club or USD6 per farmer

Aggarwal et al. (2018) GSRA Results suggest that GSRA is cost‐effective (modest cost on a per‐person basis)

Cavatassi et al. (2018) PADER‐G Investment of USD2.4 million for the construction of 66 community cereal banks. Additional USD0.6
million for capacity development of the cereal banks management committees.

Kafle et al. (2022) HVAP Cost per producer: USD158Cost benefit per beneficiary over three years: USD190

Luo et al. (2022) SGSP The high cost of advanced storage facilities is an impediment for small farmers: if a farmer purchased
four 1000‐kg metal silos, the amount per household is USD92, but the value of maize saved every year
is USD12.97. This means it takes seven years for households to recover their costs

Barnett et al. (2018) MVP Total cost 2012–2016: USD15.3 millionCost per capita 2012: USD360 or USD88 per annum

Delavallade and COPSA‐C Sensitization campaign cost: USD570 or USD2.30 per household or USD0.70 per bag storedCost of
Godlonton (2020) granting access to warrantage: USD9.20 per bag or USD30.10 per household

observe an effect on prices, indicating that farmers may not have infrastructure and technologies interventions to inform new
taken advantage of seasonal price differences. initiatives; however, they should be mindful that the available
None of the meta‐analyses show significant negative effects of evidence is highly focused on interventions in Africa (13 of the
the interventions. However, some outcomes lack sufficient data to 19 included interventions took place in 11 countries in Africa). In
robustly measure intervention effects (i.e., quality, transaction costs, other regions, the available evidence is much more limited; it
and farm investments). analyzed the effects of six interventions covering five countries.
When we examined potential sources of this heterogeneity, we It is important to consider the findings in context.
found very little effect of intervention year, type of evaluation method • Policymakers and implementers should ensure that improved
used, length of follow‐up period, length of exposure to the intervention, storage infrastructure and technologies interventions are adapted
or whether it was a single‐component or multi‐component interven- to the local environment. This requires both physical and technical
tion. Additional analyses, such as subgroup analysis and findings related accessibility of the storage facilities and technologies, but also the
to intervention barriers and facilitators, further indicate that effects are reliability and maintenance of those storage services. In this way,
highly variable according to context and intervention type. We do not these interventions are more likely to be operational and used by
identify consistent patterns of increasing or reducing effects across beneficiaries.
specific socio‐economic groups (wealth level and gender) or across • Policymakers and implementers should also engage with local
time. We were unable to test many of the possible sources of farmers' groups to increase adoption of storage practices and
heterogeneity due to either a limited number of included studies, or efficiently coordinate the use of shared storage facilities.
because relevant data was not presented.
Finally, the impact evaluations include sparse information on Implications for researchers.
barriers and facilitators that may inform the design and implementa- • There is a need for more evidence on the effect of storage for
tion of storage interventions. First, the evidence highlighted that transit systems and on‐farm storage. The current body of evidence
storage deposit systems interventions needs to be physically and mainly covers shared storage facilities, and more evidence on
technically accessible, reliable and well‐maintained in order to ensure other types of storage approaches might allow us to identify
that these systems are operational and are used by the beneficiaries. different effects between intervention types and better under-
Second, as these interventions are often conducted at the community stand the impact of storage interventions.
level, through a cooperative or producer organization, the barriers • To better understand the effect of storage infrastructure over
and facilitators analysis also emphasized the importance of function- time, researchers should consider assessing effects over a longer
ing, fair and inclusive farmers' groups to coordinate the use of shared follow‐up period. The construction of storage infrastructure, or
storage facilities. Thirty‐three percent of the included studies the introduction of storage technologies, might require mainte-
provided some information on costs. nance over time to continue bearing positive effects. However,
the current body of evidence does not include evidence with a
Implications for policymakers, implementers. follow‐up period of over ten years, which limits insights into the
• Policymakers and implementers should use the findings of this potential effects of degradation or good maintenance of storage
systematic review on the effects of improved storage facilities.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 163 of 191

5.3.6 | Chapter 6: Multi‐component output market outcomes among the other five chapters: transaction costs, volume,
access interventions yields, prices, sales, farm income, total income, and food security.

We now review and synthesize evidence of the effects of output Description of included studies
market access interventions which combine multiple components. We included 59 papers (including three ongoing studies: Ragasa et al.,
These effects correspond in a subset of the effects already 2016; Lema et al., 2017; Jirgi et al., 2019 and three linked papers)
analyzed in the previous chapters. In the analysis above, we examining the effects of 53 multi‐component output market access
conducted a moderator analysis to determine whether the interventions. The quantitative synthesis of findings relies on effects
estimates from multi‐component interventions were on average from these 59 papers (56 main studies and three linked papers).
statistically significantly different from those of single‐ Three linked papers report additional outcome data about the effects
component interventions. of an intervention (i.e., data on outcomes or prioritized outcome
Effects synthesized here are based on interventions combining indicators that are not reported by the main study).
multiple output market access interventions or on interventions with
development activities complementing an output market access Intervention, inputs, and activities
intervention. These effects therefore represent the aggregated Among all the interventions considered, we identified 53 multi‐
change to beneficiaries. We cannot disentangle how much of the component output market access interventions implemented in 26
effect is attributable to each activity. We consider effects from countries (Figure 117). Interventions in Sub‐Saharan Africa represent
interventions combining: more than half of the included interventions (n = 33), followed by East
Asia and the Pacific (n = 7), South Asia (n = 6), Latin American and
• Multiple subcategories under the same intervention cat egory (e.g., Caribbean countries (n = 5) and Europe and Central Asia (n = 2).
fixed‐price contracts and production‐management contracts under Individual countries with the largest number of included interventions
contract farming interventions), are Kenya (n = 8), Ethiopia (n = 5), China (n = 4), and Tanzania (n = 4).
• Multiple approaches under the same subcategories (e.g., different One studies analyzed the effects of an intervention in Democratic
on‐farm storage technologies), Republic of Congo, Rwanda and Uganda (Nkonya et al., 2013).
• Multiple approaches including one of subcategories of output The most frequent type of intervention in this analysis is the
market access interventions (e.g., different farm‐to‐market transport arranged or curated offers from buyers (n = 22), followed by the groups
infrastructure and capacity development and advisory services), of interventions belonging to domestic transport infrastructure
• Or multiple subcategories under different intervention categories (n = 12), alternative physical marketplaces (n = 11), provision of output
(e.g., other contract types and arranged or curated offers from buyers). market information, mobile and internet (n = 8) and storage deposit
systems (n = 6) (Figure 118). Sixteen interventions have activities
This chapter first describes the extent of available evidence, the across two intervention subcategories and four interventions have
interventions, outcomes, comparisons, and study designs. We then activities across three subcategories. Arranged or curated offers from
present these effects, specifically focusing on the most reported buyers interventions were often combined with technical assistance

F I G U R E 117 Distribution of the interventions by region.

 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
164 of 191 | MARION ET AL.

F I G U R E 118 Distribution of interventions by intervention subcategories.

and capacity development (e.g., ACE and AFP) or with contract Study design
farming interventions (e.g., CATF and WPC). Similarly, domestic Of the 56 main studies, most of the studies in our analysis were
transport infrastructure interventions were often combined with based on quasi‐experimental designs (n = 51) to evaluate the
capacity development (e.g., FAD‐2 and FAD‐3) but also other effects of the interventions. The majority of these studies relied
infrastructure development (e.g., AD2M). on statistical matching methods (n = 27) and difference‐in‐
Supporting Information: Appendix 2 for Chapter 6 provides differences and fixed‐effect approaches (n = 21). Two studies
information on the 53 interventions in terms of the country, the type conducted an IV estimation and another study utilized an
of intervention, and the year of intervention. Table 37 provides an interrupted time series (Goyal, 2010). The remaining five studies
overview of the interventions included in this review in terms of were studies with experimental designs.
country, type of intervention and the year the intervention began. Figure 120 also shows the distribution of the length of studies'
The information on the first year of the intervention is not specified different follow‐up periods (study periods). This is the difference
for three of the included interventions. between the start date of the intervention and the most recent
outcome data included in the study's analysis. This information is not
Comparison specified in two of the included studies. The remaining interventions
In all 53 included studies, the comparison groups used for the started in the period between 1980 and 2016 and with an average
estimations nearly exclusively focused on comparing the outcomes of length of roughly 5.8 years. Most studies with this information
beneficiaries of multi‐component output market access interventions to covered a period between two and four years after the intervention
farmers who did not have access to these interventions (96%). One took place (n = 13). Nine studies evaluated the effects for less than
study exploited a phased‐in (or pipeline) approach of the interven- two years. The remaining studies evaluated the effects in a period of
tions to identify a comparison group (Chaudhary & Suri, 2022). four to ten years after the intervention started (n = 32). The longest
Another study used beneficiaries of other interventions as the covered period is 63 years as authors indicated that the multi‐
control group (Belay & Ayalew, 2020). The authors identified the component output market access interventions started in 1950 in
control group based on the variation in the timing and spatial distance Tanzania (CFT, Herrmann et al., 2018) whilst the shortest is 1 month
to the ECX. Supporting Information: Appendix 3 for Chapter 6 (Adebisi et al., 2019; Lentz & Upton, 2016).
provides information on these studies such as the evaluated
intervention, the intervention sub‐category, the length of follow‐up Synthesis of findings
periods in years, the study design, and the type of comparison. The data extraction of studies under the multi‐component output
market access interventions category identified a total of 253 selected
Outcomes estimates based on criteria described in Supporting Information:
The analyzed studies cover all 19 outcomes included in our Appendix 9 for background and methods sections, guiding the
framework specified in the protocol (Villar et al., 2023). The most selection of independent estimates. In the following section, we first
frequently analyzed outcomes are total income and wealth (n = 28), provide an overview of the risk of bias assessment conducted on the
farm income (n = 27), followed by food security and nutrition (n = 20), included estimates and provide an analysis of the effects of multi‐
use or adoption of improved inputs, technologies and practices component output market access interventions on covered outcomes.
(n = 20), sales (n = 19), produce type (n = 17) and prices (n = 17) We report additional information on the individual meta‐analyses
(Figure 119). (sensitivity analysis, outlier analysis, publication bias analysis,
 T A B L E 37 Table of multi‐component output market access interventions included.
MARION

First year of
Intervention name Acronym Country (ISO3 code) Type intervention
ET AL.

Agriculture for Children's Empowerment ACE Liberia (LBR) Arranged or curated offers from buyers 2008

Appui au Développement du Menabe et du Melaky AD2M Madagascar (MDG) Domestic transport infrastructure 2007

Anchor Farm Project AFP Malawi (MWI) Improved on‐farm storage infrastructure and technologies; Arranged or 2014
curated offers from buyers

Agruppa AGP Colombia (COL) Online commodity exchanges and mobile‐based marketplaces 2016

Women's economic empowerment in rural ARMA66 Armenia (ARM) Arranged or curated offers from buyers; Storage deposit systems; 2010
communities of Vayots Dzor region Provision of output market information (mobile and internet)

Agricultural Value Chain Project Oromia AVCPO Ethiopia (ETH) Storage deposit systems; Other contract types; Alternative physical 2011
marketplaces

Contract arrangement with tea farmers CATF Viet Nam (VNM) Arranged or curated offers from buyer; Input or credit‐supply contracts 1999

Contract farming in Hannoi CFH Viet Nam (VNM) Arranged or curated offers from buyers

dairy market hub DMH Tanzania (TZA) Alternative physical marketplaces 2014

DrumNet, a horticultural export and cashless micro‐ DMHECMCP Kenya (KEN) Alternative physical marketplaces 2004
credit program

E‐CHOUPAL E‐CHOUPAL (Hub) India (IND) Provision of output market information (mobile and internet) 2000

East African Dairy Development EADD Uganda (UGA) Alternative physical marketplaces 2008

Ethiopian Commodity Exchange ECX Ethiopia (ETH) Online commodity exchanges and mobile‐based marketplaces; Provision 2008
of output market information (mobile and internet)

Egiye Jai and Nigera Gori EJNG Bangladesh (BGD) Arranged or curated offers from buyers 2013

Ethiopia Market Information System EMIS Ethiopia (ETH) Provision of output market information (mobile and internet) 2008

Empowering Small Producers, Especially Women, in ESPEWDS Pakistan (PAK) Arranged or curated offers from buyers; Alternative physical 2011
the Dairy Sector marketplaces

Fadama II FAD‐2 Nigeria (NGA) Domestic transport infrastructure 2005

Fadama III FAD‐3 Nigeria (NGA) Domestic transport infrastructure 2008

Fostering Agricultural Market Activity FARMA Bosnia and Arranged or curated offers from buyers 2009
Herzegovina (BIH)

Farmer Organizations FO Ethiopia (ETH) Arranged or curated offers from buyers 2009

Farmers' Professional Cooperatives in China FPC China (CHN) Input or credit‐supply contracts; Arranged or curated offers from buyers 2008
|

(Continues)
165 of 191

18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
 TABLE 37 (Continued)

First year of
166 of 191

Intervention name Acronym Country (ISO3 code) Type intervention

|

Food Production, Processing, and Marketing FPPM Democratic Republic of the Domestic transport infrastructure; Improved on‐farm storage 2011
Congo (COD) infrastructure and technologies; Provision of output market information
(other forms of communication)

Farmer support program FSP South Africa (ZAF) Domestic transport infrastructure 1987

Guangxi Integrated Agricultural Development Project GIADP China (CHN) Domestic transport infrastructure; Alternative physical marketplaces 2012

The Guiding Opinions on Promoting Accurate GOPAPAEC China (CHN) Online commodity exchanges and mobile‐based marketplaces 2016
Poverty Alleviation for E‐commerce Project

Guatemala Highlands Value Chain Development GUAB38 Guatemala (GTM) Arranged or curated offers from buyers; Storage deposit systems 2010
Alliance

High‐Value Agriculture Project in Hill and HVAP Nepal (NPL) Arranged or curated offers from buyers; Improved on‐farm storage 2011
Mountainous Areas infrastructure

Irish Aid infrastructure project in Vietnam IAIV Viet Nam (VNM) Domestic transport infrastructure 2011

Integrated agricultural research for development IAR4D Democratic Republic of the Alternative physical marketplaces – Provision of output market 2008
Congo (COD) information (mobile and internet)

Kenya Maize Development Program, KHDP Kenya (KEN) Alternative physical marketplaces 2004

Kims Poultry Care Center contract farming KPCCCF Kenya (KEN) Input or credit‐supply contracts; Arranged or curated offers from buyers 1996

Livestock Commercialisation for Pastoralist LCPC Niger (NER) Arranged or curated offers from buyers 2008
Communities

LIBA‐72 Liberia (LBR) Domestic transport infrastructure 2010

Mercados Campesinos MC Colombia (COL) Alternative physical marketplaces 2004

Millenium Village Project MVP Ghana (GHA) Domestic transport infrastructure; Storage deposit systems 2012

MVP Kenya (KEN) Domestic transport infrastructure 2005

The ‘optimizing income of local chicken and OILCSS Tanzania (TZA) Alternative physical marketplaces; Improved on‐farm storage 2012
sunflower smallholder producers' project infrastructure and technologies.

Purchase for Progress P4P Democratic Republic of the Arranged or curated offers from buyers; Provision of output market 2012
Congo (COD) information (other forms of communication)

Purchase for Progress P4P Ghana (GHA) Arranged or curated offers from buyers 2011

P4P Tanzania (TZA) Arranged or curated offers from buyers, Production‐management 2009
contracts and Storage deposit systems

P4P Uganda (UGA) Storage deposit systems 2008

MARION
ET AL.

18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
 TABLE 37 (Continued)
MARION

First year of
Intervention name Acronym Country (ISO3 code) Type intervention
ET AL.

Programa de Servicios Agropecuarios Provinciales PROSAP Argentina (ARG) Domestic transport infrastructure 2006

Smallholder Dairy Commercialization Programme SDCP Kenya (KEN) Arranged or curated offers from buyers 2006

Smallholder Horticulture Empowerment and SHEP Kenya (KEN) Arranged or curated offers from buyers 2016
Promotion

Smallholder Horticulture Marketing Programme ShoMaP Kenya (KEN) Domestic transport infrastructure; Provision of output market 2008
information (mobile and internet)

Three formal system of information system: SIMA, SIF Mozambique (MOZ) Provision of output market information (mobile and internet) 1991
INFOCOM and FEWS NET

Sustainable Quality Program SQP Colombia (COL) Production‐management contracts; Input or credit‐supply contracts 2007

self‐sustaining groups to access clean planting SSGACPMTSM Kenya (KEN) Arranged or curated offers from buyers 2003
material, technical support, and markets

Vegetables Contract Farming in China VCFC China (CHN) Production‐management contracts 1980

Vertical Market Linkages VML Tanzania (TZA) Arranged or curated offers from buyers; Undefined contract

Women Farmers with Global Potential WFGP India (IND) Alternative physical marketplaces; Provision of output market 2008
information (mobile and internet)

Wheat Initiative to increase smallholder productivity WIISP Ethiopia (ETH) Arranged or curated offers from buyers 2013

National Agriculture Market e‐trading platform eNAM India (IND) Online commodity exchanges and mobile‐based marketplaces 2016
| 167 of 191

18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
168 of 191 | MARION ET AL.

F I G U R E 119 Distribution of outcome types.

moderator analysis, meta‐analyses without high risk of bias estimates) (Supporting Information: Appendix 10 for background and methods
in Supporting Information: Appendix 4 for Chapter 6. sections). Our assessments of studies' risk of bias identify whether
aspects of their design, implementation, analysis, or reporting may
Risk of bias. The risk of bias of each estimate across the 56 included potentially bias their estimates of the effects of interventions. We
studies was assessed following the Risk of Bias Assessment Tool assessed the risk of bias associated with each estimate of the average
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 169 of 191

F I G U R E 120 Distribution of the length of studies' follow‐up periods.

effect of an intervention included in our analysis. Figure 121 presents creating new marketplaces and alternatives market opportunities
a summary of the results of our risk of bias assessments. We found (k = 4), contract farming initiatives (k = 0), improved storage infra-
that 61% of included estimates were rated as having a high risk of structure and technologies (k = 3). We found three estimates in Ragasa
bias, 36% of estimates were considered to have some concerns and colleagues (2016) as the evaluation focused on three different
related to the risk of bias, and the remaining 3% of estimates were regions of DRC with independent comparative groups.
assessed as having a low risk of bias. All estimates were at the household level (k = 9) and measure the
Figure 121 shows a consistency in the risk of bias in both QED cost of production of crops. For example, Ashraf and colleagues (2009)
studies and RCT studies since 61% of the former are rated as high risk analyzed the effect of DMHECMCP, combining different offers of
of bias and 64% of the latter. This risk of bias of included estimates was alternative physical marketplaces, in Kenya on the total spending in
based on six domains (selection bias; confounding; performance bias; marketing (in Ksh 1000). In Ethiopia, Gelo and colleagues (2020), analyzed
spillovers, crossovers, and contamination; outcome measurement bias; the effect of FO, combining approaches for arranged or curated offers from
and reporting bias). Looking at individual risk of bias domains, the most buyers, on the total cost of selling maize. The effect size here is reversed
common source of bias is related to non‐randomness of the assignment meaning that a higher effect represents lower transaction cost.
mechanism in RCTs (45% of estimates rated as high risk of bias, only Multi‐component output market access interventions did not decrease
relevant for RCTs), spillovers, cross‐over, and contamination (27% of transaction costs. The overall weighted average effect is very small,
estimates) and confounding bias (26% of estimates). Additional ˆ = 0.05; 95% CI: −0.04 to
positive, and not statistically significant (SMD
concerns included selection bias (22% of estimates rated as high risk 0.13; p = 0.29) (Figure 122). Effect sizes vary between −0.14 and 0.66
of bias), reporting bias (5% of estimates rated as high risk of bias), and standard deviations. The heterogeneity among the studies is substantial
outcome measurement bias (4% of estimates rated as high risk of bias). (Q(8) = 45.32, p < 0.001, τ̂ 2 = 0.01, I2 = 82.35%).
Estimates were not likely to be at high risk for performance bias. In Kinshasa, DRC, Ragasa and colleagues (2016) reported the
negative, small, and statistically significant effect of FFPM on the fare to
Summary of the meta‐analyses results of multi‐component output the nearest market in urban area by motor vehicle (ĝ = −0.14 ; 95% CI:
market access interventions. Table 38 displays the meta‐analysis −0.24 to −0.04; p = 0.01). In Kenya, Ashraf and colleagues (2009) found
results across outcome types covered in the subsequent sections. the positive, small and statistically significant effect of DMHECMCP on
the total spending in marketing (in Ksh 1000; ĝ = 0.10 ; 95% CI: 0.001 to
Immediate outcomes. 0.19; p = 0.049). In Nigeria, Madu and Phoa (2012) observed the
positive, large and statistically significant effect of FAD‐2 on the change
Transaction costs. Within seven studies, we identified nine indepen- in mean cost of transporting load on truck (ĝ = 0.66; 95% CI: 0.43 to
dent estimates of the effect of multi‐component output market access 0.89; p < 0.001). However, this estimate is considered as an outlier. If
interventions on transaction costs outcomes. Studies of evaluated this estimate is removed from the analysis, the overall effect remains
interventions all took place in Africa (n = 7). The estimates of multi‐ positive, very small, and not statistically significant.
component interventions cover four of the five intervention The exploratory moderator analysis revealed that effects are on
categories of our body of evidence: farm‐to‐market transport average lower by −0.03 standard deviations units for more recently
infrastructure (k = 2), output market information (k = 2), initiative started interventions. Effects are on average 0.31 standard deviations
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
170 of 191 | MARION ET AL.

F I G U R E 121 Risk of bias of included estimates between QED (top) and RCT (bottom).

higher if they are based on farm‐to‐market transport infrastructure implemented in East Asia and Pacific (n = 1), Latin America and
interventions. We identified funnel plot asymmetry meaning that the Caribbean (n = 1), and South Asia (n = 4). The estimates of
publication bias is likely present. Of the nine included effect sizes, four multi‐component interventions cover the five intervention cate-
remain after excluding high risk of bias estimates. The effect of multi‐ gories of our body of evidence: farm‐to‐market transport infra-
component output market access interventions on transaction costs structure (k = 1), output market information (k = 2), initiative
ˆ = 0.14 ;
remains positive, small, and not statistically significant (SMD creating new marketplaces and alternatives market opportunities
95% CI: −0.09 to 0.37; p = 0.22). (k = 12), contract farming initiatives (k = 2), improved storage
infrastructure and technologies (k = 4).
Prices. Within 17 studies, we identified 17 independent estimates Most estimates were at the household level (k = 15) and two
of the effect of multi‐component output market access interven- estimates were at the market level (Goyal, 2010; Chaudhary & Suri,
tions on price outcomes. Studies of evaluated interventions 2022) all measuring the impact of interventions on prices received by
primarily took place in Africa (n = 11), with additional studies farmers. For example, Bonilla and colleagues (2018) analyzed the effect
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 171 of 191

T A B L E 38 Summary of meta‐analyses results.

Outcome types Outcome sub‐category SMD 95% CI k p Value

Immediate outcomes Transaction costs 0.05 −0.04, 0.14 9 0.29

Group participation 0.83 0.43, 1.24 5 <0.001

Use/adoption of fertilizers, pesticides, 0.04 −0.02, 0.10 12 0.16

herbicides

Use/adoption of improved varieties and quality 0.11 −0.03, 0.25 7 0.13

seeds

Use/adoption of labor and machinery 0.21 0.09, 0.33 13 <0.001

Use/adoption of other improved inputs 0.02 −0.17, 0.21 3 0.87

Farm investment (fertilizers and pesticides) N/A

Farm investment (labor and machinery) N/A

Farm investment (livestock) N/A

Farm investment (seeds) N/A

Farm investment (total) N/A

Credit 0.06 −0.03, 0.15 8 0.22

Prices 0.13 0.04, 0.22 17 0.01

Agricultural Quality N/A

production outcomes
Crop loss N/A

Produce type (area) −0.08 −0.19, 0.03 6 0.14

Produce type (diversity) 0.09 0.01, 0.17 11 0.02

Produce type (livestock) 0.07 0.001, 0.15 3 0.049

Volume 0.07 −0.001, 0.14 15 0.05

Yields 0.10 0.02, 0.18 16 0.02

Intermediate Farm income 0.17 0.10, 0.24 27 <0.001

outcomes
Sales 0.15 0.06, 0.25 19 0.001

Off‐farm outcomes Off‐farm income N/A

(Agriculture only)
Labor market outcomes N/A

Non‐farm outcomes Non‐farm income 0.08 −0.01, 0.16 10 0.07

Labor market outcomes N/A

Welfare outcomes Total income and wealth 0.12 0.07, 0.16 30 <0.001

Food security 0.07 0.002, 0.13 19 0.04

Food Nutrition 0.05 −0.02, 0.11 7 0.19

Note: Outcome subcategories in bold have a statistically significant standardized mean difference (SMD) at a 5% significance level. K is the number of
estimates. N/A indicates that we did not conduct a meta‐analysis for these outcome types. In these cases, k = 1 or k = 2 and we chose not do to a meta‐
analysis in those cases due to differences in intervention type, in the regions, and so forth.

of SDCP, an intervention combining approaches of arranged or curated (Figure 123). Effect sizes vary between −0.29 and 0.65 standard
offers from buyers, in Kenya on the selling price of milk per liter. In deviations. The heterogeneity among the studies is substantial
Ethiopia, Belay and Ayalew (2020) analyzed the effect of ECX, and (Q(16) = 144.02, p < 0.001, τ̂ 2 = 0.03, I2 = 88.89%).
intervention combining approaches of online commodity exchanges and In Pakistan, Lombardini and Bowman (2015) analyzed the effect of
mobile‐based marketplaces and the provision of output market information ESPEWDS, an intervention combining arranged or curated offers from
(mobile and internet), on the average price received by farmers. buyers and alternative physical marketplaces, and found the negative,
Multi‐component output market access interventions increased moderate, and statistically significant effect of the intervention on milk
prices. The overall weighted average effect is small, positive, and price (ĝ = −0.29; 95% CI: −0.48 to −0.10; p = 0.002). In India,
ˆ = 0.13; 95% CI: 0.04 to 0.22; p = 0.01)
statistically significant (SMD Chaudary and Suri (2022) reported the positive, large and statistically
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
172 of 191 | MARION ET AL.

F I G U R E 122 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of multi‐component output
market access interventions on transaction costs.

F I G U R E 123 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of multi‐component output
market access interventions on prices.

significant effect of eNAM, combining approaches of online commodity South Asia (n = 2). The estimates of multi‐component interventions
exchanges and mobile‐based marketplaces, on wholesale prices cover the five intervention categories of our body of evidence: farm‐
(ĝ = 0.65; 95% CI: 0.40 to 0.89; p < 0.001). to‐market transport infrastructure (k = 6), output market information
The moderator analysis revealed that effects are on average (k = 1), initiative creating new marketplaces and alternatives market
lower by −0.02 standard deviations units for studies with longer opportunities (k = 8), contract farming initiatives (k = 3), improved
follow‐up period. We did not identify funnel plot asymmetry meaning storage infrastructure and technologies (k = 3). We found two estimates
that publication bias is unlikely present. Of the 17 included effect in de Castro (2021) as the evaluation focused on two different
sizes, four remain after excluding high risk of bias estimates. The interventions with independent comparative groups.
effect of multi‐component output market access interventions on prices Most estimates were at the household level (k = 13), completed
remains positive, but very small, and not statistically significant by one estimate at the district level (Goyal, 2010), and two estimates
ˆ = 0.04; 95% CI: −0.18 to 0.27; p = 0.71).
(SMD at the plot level. All outcome variables relate to measures of
agricultural yields. We first selected estimates that looked at the
Agricultural production outcomes. effect of these interventions on the yields of the crops targeted by
the intervention. Alternatively, we selected estimates focusing on the
Yields. Within 15 studies, we identified 16 independent estimates of most prevalent crops in the local context. Examples include the work
the effect of multi‐component output market access interventions on from Abate and colleagues (2018) in Ethiopia on WIISP, an
yield outcomes. Studies of evaluated interventions primarily took intervention combining approaches for arranged or curated offers
place in Africa (n = 9), with additional studies implemented in East from buyers, on wheat yields. In India, Goyal (2010) focused on the
Asia and Pacific (n = 3), Latin America and the Caribbean (n = 2), and effect of E‐CHOUPAL, an intervention combining the provision of
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 173 of 191

F I G U R E 124 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of multi‐component output
market access interventions on yields.

output market information (mobile and internet) and alternative physical cover four of the five intervention categories of our body of
marketplaces, on the total yields. evidence: farm‐to‐market transport infrastructure (k = 5), output market
Multi‐component output market access interventions increased information (k = 1), initiative creating new marketplaces and alternatives
yields. The overall weighted average effect is small, positive, and market opportunities (k = 9), contract farming initiatives (k = 0), im-
ˆ = 0.10 ; 95% CI: 0.02 to 0.18; p = 0.02)
statistically significant (SMD proved storage infrastructure and technologies (k = 3).
(Figure 124). Effect sizes vary between −0.94 and 0.39 standard Outcome variables include measures of the quantity or volume
deviations. The heterogeneity among the studies is substantial of agricultural production. The analysis is mostly at the household
(Q(15) = 70.00, p < 0.001, τ̂ 2 = 0.02, I2 = 78.57%). level (k = 15) completed by one estimate at the district level. As for
In Kenya, Fischer and Qaim (2012) analyzed the effect of the other outcomes, we selected estimates focusing on crops
SSGACPMTSM, an intervention combining arranged or curated offers targeted by the intervention (such as coffee in the CoffeeCF in Haiti,
from approaches, and found the negative, large, and statistically Fuller, 2013) or the most prevalent crops in the local context.
significant effect of the intervention on banana yields (ĝ = −0.94 ; Examples include the work from Bonilla and colleagues in Kenya on
95% CI: −1.51 to ‐0. 37; p = 0. 001). In Tanzania, de Castro (2021) the effect of SDCP, combining arranged or curated offers from buyers
reported the positive, moderate and statistically significant effect of approaches, on the total milk production. In Kenya, Ashraf and
VML2, comparing approaches of arrange or curated offers from buyers colleagues (2009) analyzed the effect of DMHECMCP, combining
and fixed‐price contract farming, on production outcomes (ĝ = 0.33; alternative physical marketplaces approaches, on the total production
95% CI: 0.10 to 0.55; p < 0.001). of French beans.
The moderator analysis revealed that effects are on average Multi‐component output market access interventions increased
higher by 0.09 standard deviations units for studies with longer volume. The overall weighted average effect is very small, positive,
follow‐up period. We did not identify funnel plot asymmetry meaning ˆ = 0.07; 95% CI: −0.001 to
and statistically insignificant (SMD
that publication bias is unlikely present. Of the 16 included effect 0.14; p = 0.05) (Figure 125). Effect sizes vary between −0.28 and
sizes, six remain after excluding high risk of bias estimates. The effect 1.01 standard deviations. The heterogeneity among the studies is
of multi‐component output market access interventions on yields is substantial (Q(14) = 38.55, p < 0.001, τ̂ 2 = 0.01, I2 = 63.68%).
ˆ = 0.11; 95% CI:
positive, small, but not statistically significant (SMD In Guatemala, Fuller (2013) analyzed the effect of GUAB38, an
−0.04 to 0.26; p = 0.16). intervention combining arranged or curated offers from buyers and
storage deposit systems, and found the negative, moderate, and
Volume. Within 15 studies, we identified 15 independent estimates statistically significant effect of the intervention on the quantity of
of the effect of multi‐component output market access interventions on maize produced in 2011 (ĝ = −0.28; 95% CI: −0.49 to
volume outcomes. Studies of evaluated interventions primarily took ‐0.08; p < 0.001). In Nigeria, Jirgi and colleagues (2019) observed
place in Africa (n = 7), with additional studies implemented in Europe the positive, large and statistically significant effect of FAD‐3,
and Central Asia (n = 2), Latin America and the Caribbean (n = 2), and comparing approaches of domestic transport infrastructure, on the
South Asia (n = 4). The estimates of multi‐component interventions volume of sorghum (ĝ = 1.01; 95% CI: 0.21 to 1.80; p = 0.01).
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
174 of 191 | MARION ET AL.

F I G U R E 125 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of multi‐component output
market access interventions on volume.

The moderator analysis revealed that effects are on average cultivated crops sold (the ratio of the value of crops sold divided
lower by −0.27 standard deviations units for studies in Latin‐America by the value of crops produced). In Uganda, Kizito and Kato (2018)
compared to Africa. Similarly, effects are on average lower by −0.12 analyzed the effect of P4P, an intervention combining approaches to
standard deviations units for studies in South Asia compared to storage deposit systems, on maize sales.
Africa. We also found that effects are on average larger by 0.002 Multi‐component output market access interventions increased
standard deviations units for interventions with longer exposure sales. The overall weighted average effect is small, positive, and
period. Effects are on average 0.16 standard deviations higher if they ˆ = 0.15; 95% CI: 0.06 to 0.25; p = 0.001)
statistically significant (SMD
are based on farm‐to‐market transport infrastructure interventions We (Figure 126). Effect sizes vary between −0.31 and 0.82 standard
did not identify funnel plot asymmetry meaning that publication bias deviations. The heterogeneity among the studies is substantial
is unlikely present. Of the 15 included effect sizes, six remain after (Q(18) = 137.52, p < 0.001, τ̂ 2 = 0.03, I2 = 86.91%).
excluding high risk of bias estimates. The effect of multi‐component In Ethiopia, Biggeri and colleagues (2022) found the positive,
output market access interventions on volume is still positive, very large and statistically significant effect of AVCPO, combining storage
ˆ = 0.09; 95% CI: −0.07
small, and not statistically significant (SMD deposit systems, other contract farming types, and alternative physical
to 0.25; p = 0.27). marketplaces, on the share of cereal production sold through
cooperatives (ĝ = 0.82; 95% CI: 0.66 to 0.98; p < 0.001). However,
Intermediate outcomes. this estimate is considered as an outlier. If this estimate is removed
from the analysis, the overall effect remains positive, small, and
Sales. Within 19 studies, we identified 19 independent estimates of statistically significant. In Tanzania, Anguko and Rana (2017) analyzed
the effect of multi‐component output market access interventions on the effect of OILCSS, combining alternative physical marketplaces and
sales outcomes. Studies of evaluated interventions primarily took improved on‐farm storage infrastructure and technologies, on the
place in Africa (n = 12), with additional studies implemented in Europe quantity of sunflower sold in the previous 12 months (ĝ = 0.39;
and Asia (n = 2), Latin America and the Caribbean (n = 1), and South 95% CI: 0.07 to 0.70; p = 0.02).
Asia (n = 4). The estimates of multi‐component interventions cover The moderator analysis revealed that effects are on average
the five intervention categories of our body of evidence: farm‐to‐ larger by 0.002 standard deviations units for interventions with
market transport infrastructure (k = 3), output market information longer exposure period. Effects are on average 0.31 standard
(k = 2), initiative creating new marketplaces and alternatives market deviations higher if they are based on improved storage infrastructure
opportunities (k = 14), contract farming initiatives (k = 2), improved and technologies We did not identify funnel plot asymmetry meaning
storage infrastructure and technologies (k = 5). that publication bias is unlikely present. Of the 19 included effect
Most studies here (k = 17) used household‐level measures of sizes, five remain after excluding high risk of bias estimates.
proportion or quantity of produce sold. The remaining two studies The effect of multi‐component output market access interventions on
used producer organization or market levels. In Lentz and Upton sales is positive, moderate, but becomes statistically insignificant
(2016), the outcome variable was households' percent of all ˆ = 0.22; 95% CI: −0.09 to 0.53; p = 0.16).
(SMD
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 175 of 191

F I G U R E 126 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of multi‐component output
market access interventions on sales.

Farm income. Within 27 studies, we identified 27 independent offers from buyers, production‐management contracts, and Storage
estimates of the effect of multi‐component output market access deposit systems, on the crop value (ĝ = −0.22; 95% CI: −0.41 to
interventions on farm income outcomes. Studies of evaluated −0.02; p = 0.03). In Nigeria, Jirgi and colleagues (2019) observed the
interventions primarily took place in Africa (n = 18), with additional positive, large and statistically significant effect of FAD‐3, combining
studies implemented in East Asia and Pacific (n = 1), Europe and domestic transport infrastructure and other approaches, on the income
Central Asia (n = 1), Latin America and the Caribbean (n = 1), and from sorghum production (ĝ = 1.58; 95% CI: 0.74 to 2.41; p < 0.001).
South Asia (n = 2). The estimates of multi‐component interventions The moderator analysis revealed that effects are on average
cover the five intervention categories of our body of evidence: farm‐ larger by 0.002 standard deviations units for interventions with
to‐market transport infrastructure (k = 7), output market information longer exposure period. It also revealed that effects are on average
(k = 5), initiative creating new marketplaces and alternatives market smaller by −0.02 standard deviations units for more recently started
opportunities (k = 16), contract farming initiatives (k = 2), improved interventions. We did identify funnel plot asymmetry meaning that
storage infrastructure and technologies (k = 6). publication bias is likely present. Of the 27 included effect sizes, nine
All studies here used measures of income generated from farm remain after excluding high risk of bias estimates. The effect of multi‐
activities including profits at the household level (k = 25), individual component output market access interventions on farm income remains
level (n = 1), and group level (n = 1). Examples include the work from ˆ = 0.11; 95% CI: 0.02
positive, small, and statistically significant (SMD
Iacovone and McKenzie (2019) in Colombia on AGP, combining online to 0.19; p = 0.01).
commodity exchanges and mobile‐based marketplaces approaches,
analyzing the effect of the intervention on the profit in the past Welfare outcomes.
week. In Tanzania, Bayiyana and colleagues (2018) analyzed the
effect of DMH, combining alternative physical marketplaces ap- Total income and wealth. Within 28 studies, we identified 30
proaches, on dairy income measured as the net value of households' independent estimates of the effect of multi‐component output
annual fresh milk production. market access interventions on income and wealth outcomes. Studies
Multi‐component output market access interventions increased of evaluated interventions primarily took place in Africa (n = 22), with
farm income. The overall weighted average effect is small, positive, additional studies implemented in East Asia and Pacific (n = 3), Europe
ˆ = 0.17 ; 95% CI: 0.10 to
and statistically significant (SMD and Central Asia (n = 1) Latin America and the Caribbean (n = 1), and
0.24; p < 0.001) (Figure 127). Effect sizes vary between −0.22 South Asia (n = 3). The estimates of multi‐component interventions
and 1.58 standard deviations. The heterogeneity among the studies is cover the five intervention categories of our body of evidence: farm‐
substantial (Q(26) = 131.01, p < 0.001, τ̂ 2 = 0.02, I2 = 80.13%). to‐market transport infrastructure (k = 8), output market information
In Tanzania, Lentz and Upton (2016) reported negative, moderate (k = 3), initiative creating new marketplaces and alternatives market
and statistically significant effect of P4P, combining arranged or curated opportunities (k = 21), contract farming initiatives (k = 3), improved
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
176 of 191 | MARION ET AL.

F I G U R E 127 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of multi‐component output
market access interventions on farm income.

storage infrastructure and technologies (k = 6). We found two estimates deposit systems, on the total household income (ĝ = −0.19; 95% CI:
in Tijani and colleagues (2014) as the evaluation focused on two −0.37 to −0.01; p = 0.04 ). In Vietnam, Sagenji (2012) observed the
different states of Nigeria with independent comparative groups. positive, large and statistically significant effect of CATF, combining
There are also two estimates in de Castro (2021) as authors evaluated arranged or curated offers from buyers and input or credit‐supply
the effects of two different interventions with independent control contracts, on the daily income per capita (ĝ = 0.45; 95% CI: 0.10 to
groups. −0.80; p = 0.01).
All studies here used measures of total household gross income, The moderator analysis revealed that effects are on average
net income, and other measures of socio‐economic status such as smaller by −0.02 standard deviations unit for more recently started
total household expenditure, asset or wealth indices at the interventions. We also found a larger average effect by 0.002
household‐level (k = 30 ). For example, de Castro (2021) analyzed standard deviation units for interventions with longer exposure
the effect of VML in Tanzania, an intervention combining fixed‐price period. We did not identify funnel plot asymmetry meaning that
and price guarantee contracts and arranged or curated offers from publication bias is unlikely present. Of the 30 included effect sizes,
buyers, on household welfare levels. In Nepal, Kafle and colleagues 13 remain after excluding high risk of bias estimates. The effect of
(2022) analyzed the effect of HVAP, combining arranged or curated multi‐component output market access interventions on income
offers from buyers and improved on‐farm storage infrastructure, on the and wealth is still positive, small and statistically significant
total household income. ˆ = 0.13; 95% CI: 0.06 to 0.20; p < 0.001).
(SMD
Multi‐component output market access interventions increased income
and wealth. The overall weighted average effect is small, positive, and Food security. Within 18 studies, we identified 19 independent
ˆ = 0.12; 95% CI: 0.07 to 0.16; p < 0.001)
statistically significant (SMD estimates of the effect of multi‐component output market access
(Figure 128). Effect sizes vary between −0.19 and 0.45 standard interventions on food security outcomes. Studies of evaluated
deviations. The heterogeneity among the studies is substantial interventions took place in several regions: Africa (n = 14), Europe
(Q(29) = 105.27, p < 0.001, τ̂ 2 = 0.01, I2 = 72.45%). and Central Asia (n = 1), Latin America and the Caribbean (n = 1), and
In Tanzania, Lentz and Upton (2016) found negative, small and South Asia (n = 3). The estimates of multi‐component interventions
statistically significant effect of P4P, combining arranged or curated cover the five intervention categories of our body of evidence: farm‐
offers from buyers and production‐management contracts and Storage to‐market transport infrastructure (k = 2), output market information
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 177 of 191

F I G U R E 128 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of multi‐component output
market access interventions on income and wealth.

(k = 3), initiative creating new marketplaces and alternatives market ˆ = 0.07; 95% CI: 0.002 to
and statistically significant (SMD
opportunities (k = 14), contract farming initiatives (k = 3), improved 0.13; p = 0.04) (Figure 129). Effect sizes vary between −0.12 and
storage infrastructure and technologies (k = 5). There are two estimates 0.46 standard deviations. The heterogeneity among the studies is
in de Castro (2021) as the evaluation focused on two different substantial (Q(18) = 69.91, p < 0.001, τ̂ 2 = 0.01, I2 = 74.25%).
interventions with independent comparative groups. Gelo and colleagues (2020) on FO in Ethiopia reported a positive,
Outcomes here are all at the household level and measure value large and statistically significant effect of the intervention on food
of daily calorie intake, food consumption expenditure level, and expenditure shares (ĝ = 0.46; 95% CI: 0.24 to 0.68; p < 0.001). In
Household Food Insecurity Access Scale. Other outcome variables Ethiopia, Biggeri and colleagues (2022) observed the positive, large
were the duration of the hunger period, a binary variable on whether and statistically significant effect of AVCPO, combining storage
a household faced a period of hunger, and the share of food deposit systems, other contract types and alternative physical market-
expenditure. For example, Lentz and Upton (2016) analyzed the places, on HFIAS (ĝ = 0.37; 95% CI: 0.21 to 0.53; p < 0.001).
effect of P4P in Tanzania, combining arranged or curated offers from None of the moderator variables significantly contributed to the
buyers and production‐management contracts and storage deposit variation in estimates. We did not identify funnel plot asymmetry
systems, on the food consumption score (covering the frequency of meaning that publication bias is unlikely present. Of the 19 included
consumption of different food groups in the past seven days, effect sizes, nine remain after excluding high risk of bias estimates.
weighted by nutritional importance). In Ethiopia, Gelo and colleagues The effect of multi‐component output market access interventions on
(2020) anlayzed the effect of FO, combining approaches of arranged food security remains positive, small, but becomes statistically
or curated offers from buyers, on the food expenditure share. ˆ = 0.09; 95% CI: −0.03 to 0.20; p = 0.13).
insignificant (SMD
Measures on HFIAS, relating to period and occurrence of hunger
and share of food expenditure were reversed meaning that a higher Discussion
effect represents greater food security.
Multi‐component output market access interventions improved food Summary of findings. There is a large body of evidence of studies
security. The overall weighted average effect is very small, positive, quantitatively analyzing the effect of multi‐component output market
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
178 of 191 | MARION ET AL.

F I G U R E 129 Forest plot showing observed outcomes and estimates of the random‐effects model for the effect of multi‐component output
market access interventions on food security.

access interventions: it includes 53 studies reporting the effect of 53 subgroups, barriers and facilitators, or cost analysis here as they are
unique interventions in 26 countries (with 33 studies focusing on the same as in the previous chapters. Overall, we found positive,
Africa). very small to small, and mostly statistically significant effects of
The studies here cover effects based on interventions combining: multi‐component output market access interventions on the main
outcomes included in our analysis. None of the meta‐analyses
• Multiple subcategories under the same intervention category (e.g., revealed any statistically significant negative effects of the
fixed‐price contracts and production‐management contracts under interventions.
contract farming interventions), In terms of immediate outcomes, the meta‐analyses showed
• Multiple approaches under the same subcategories (e.g., different that multi‐component output market access interventions led to
on‐farm storage technologies), positive, small, and statistically significant effects on prices
• Multiple approaches including one of subcategories of output ˆ = 0.13; 95% CI: 0.04 to 0.22; k = 17). However, we did not
(SMD
market access interventions (e.g., different farm‐to‐market transport find a statistically significant effect of the interventions on
infrastructure and capacity development and advisory services), ˆ = 0.05; 95% CI: −0.04 to 0.13; k = 9).
transaction costs (SMD
• Or multiple subcategories under different intervention categories Regarding agricultural production outcomes, the meta‐analysis
(e.g., other contract types and arranged or curated offers from buyers). revealed positive, very small to small, and statistically significant
ˆ = 0.10; 95% CI: 0.02 to
effect of the interventions on yields (SMD
At the sub‐category level, many studies focused on the effect 0.18; k = 16). The intermediate outcomes show the largest average
of arranged or curated offers from buyers (n = 22). At the category effects for this intervention category. We estimate a small, positive
level, initiatives creating new marketplaces and alternative marketing and statistically significant effect of the interventions on sales
opportunities were the most prevalent (n = 37). Exposure periods ˆ = 0.15; 95% CI: 0.06 to 0.25; k = 19) and farm income
(SMD
span from 5 month to 34 years (although 2 studies did not provide ˆ = 0.16; 95% CI: 0.10 to 0.24; k = 27). Finally, the analysis
(SMD
information on exposure and follow‐up period). Follow‐up periods of the effect of interventions on welfare outcomes revealed their
are up to three years, although 45 studies analyzed the effect positive, very small to small, and statistically significant average
immediately after the end of the intervention implementation. ˆ = 0.12; 95% CI: 0.07 to
effect on the total income and wealth (SMD
Included studies mostly used quasi‐experimental designs (n = 48 ˆ = 0.07; 95% CI: 0.002
0.16; k = 30), and on food security (SMD
out of 53). to 0.13; k = 19).
We analyzed the effect of interventions against the full When comparing with the rest of the chapters, we found that the
spectrum of outcomes of our framework. However, we focused effects of multi‐component intervention have similar direction, size,
here on the most reported outcomes among the other five and significance of effect to the ones observed in their individual
chapters: transaction costs, volume, yields, prices, sales, farm income, components. In line with the observations from other chapters, the
total income, and food security. We did not conduct additional body of evidence on multi‐component interventions is characterized by
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 179 of 191

high risk of bias. Similarly, to trends observed in other chapters, the 6 | D IS CU SS IO N

main sources of risk of bias were selection, confounding effects and
reporting. This systematic review is the most comprehensive and up‐to‐date
We found substantial variation between the studies in each of review of output market access interventions, analyzing their effects
these outcome categories. When we explore potential sources of on agricultural, socio‐economic, food security, and nutrition outcomes in
heterogeneity, we did observe time‐related but no context‐specific LMICs. The included studies examined the effects of 262 interven-
or approach‐related patterns. For yields, farm income and income and tions in 53 unique LMICs in all regions of the world, including: 156
wealth outcomes, we reported a slightly larger average effect of studies from Sub‐Saharan Africa, 41 from East Asia and the Pacific,
intervention with a larger exposure period. However, we also 26 from Latin America and the Caribbean, 59 from South Asia, and
distinguished a slightly smaller average effect of more recent seven from Europe and Central Asia.
interventions (from −0.03 to −0.02 standard deviation units) for
sales, farm income and income and wealth outcomes. The magnitude
of the effects from multi‐component output market access interven- 6.1 | Summary of main results
tions are not different by intervention type. We found that effects
were slightly higher for multi‐component domestic transport infra- We first summarize the results by intervention type. Figure 130
structure interventions for transactions costs and volume outcomes reports the overall weighted effects found across chapters by
relative to the other types of interventions. outcome.

Implications for policymakers and implementers.

• Policymakers might rely on multi‐component interventions to 6.1.1 | Farm‐to‐market transport infrastructure
complement single component failing to achieve effects on the interventions
targeted outcome. For example, while farm‐to‐market transport
infrastructure or initiative creating new marketplaces and alternative Farm‐to‐market transport infrastructure interventions include initiatives
marketing opportunities individually fail at impact volume outcome, that improve infrastructure (e.g., road, bridges) used for delivering
they bear positive and statistically significant effect when agricultural produce to domestic and international markets. This
combined in multi‐component interventions. includes initiatives constructing, rehabilitating, or maintaining
• However, when a single component is already bearing positive infrastructure.
and statistically significant effects on a targeted outcome, In our review of the evidence, we found 83 papers reporting the
combining it with an additional component might not lead to a effects of 41 separate interventions implemented across 30 countries
larger effect. Indeed, for all positive and statistically significant (16 interventions in Africa, 9 in Latin America and the Caribbean, 7 in
effects of multi‐component intervention on any given outcome, a East Asia and the Pacific, 7 in South Asia, and 2 in Europe and Central
single component intervention bears a larger effect: for example, Asia). The most common types of infrastructure were roads (28
farm‐to‐market transport infrastructure single component inter- interventions) and bridges (five interventions), with eight additional
ventions show a larger effect on transaction costs; market interventions combining investments in both road and bridge
interventions show a larger effect on prices; contract farming infrastructure, and one intervention that built roads and waterway
interventions show a larger effect on volume, yields, sales, or farm infrastructure.
income; storage interventions show a larger effect on total We found that 49% of included estimates were rated as having a
income and food security. high risk of bias, 26% were considered to have some concerns related
to risk of bias, and the remaining 25% were assessed as having a low
Implications for researchers. risk of bias. Information on intervention costs is available for 15 of
• Further research could focus on filling the evidence gaps: this the 41 included interventions, based on data from 15 studies.
might include a focus on less covered regions (e.g., East Asia and For farm‐to‐market transport infrastructure interventions, we
Pacific, Europe and Central Asia, Latin America and the Caribbean, detected very small and statistically significant effects on reducing
South Asia), experimental designs, or longer follow‐up periods transaction costs, increasing farm income, and total income and wealth.
allowing to analyze the effect of interventions over time. We found no statistically significant effects on food security and
• Future synthesis research could further analyze the effect of nutrition outcomes. These interventions also resulted in small
interventions over a larger list of outcomes and/or incorporate improvements in the adoption of fertilizers, pesticides, and herbicides
cost, qualitative, or barriers and facilitators analysis to expand the and the adoption of improved varieties and quality seeds; moderate
body of synthesis evidence on this type of intervention. improvements in farm investments; and small gains in yields in food
• Finally, future research should seek to address the main sources of and cash crops. The evidence also indicates that in some contexts
bias identified in the body of evidence. This will increase the these interventions shifted household production and sales from
confidence in primary findings and overall accuracy of the analysis food crops to cash crops, and facilitated participation and income
of effect of the interventions. generation in non‐farming sectors.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
180 of 191 | MARION ET AL.

F I G U R E 130 Summary of overall weighted average effects.

We did not find improvements on the use/adoption of improved 6.1.2 | Output market information interventions
practices (besides adoption of fertilizers, pesticides, and herbicides and
the adoption of improved varieties and quality seeds) or in farm Output market information interventions consist of interventions that
investment, credit, prices, crop loss, volume, quality, sales, off‐farm labor provide information about market conditions, prices of inputs, and
market outcomes, off‐farm income, food security and nutrition. We did prices of produce to be sold. This category covers a wide array of
not observe any evidence of the effects of these interventions on interventions including but not limited to ICT, informational campaigns,
production diversity. and supporting access to information technologies. Improved access to
The barriers and facilitators analysis of the included studies information can affect the decisions of farmers concerning the inputs
suggests that female‐headed households or vulnerable households and procedures to be employed in the production stage, as well as in
did not benefit from the interventions in the same way that less‐ determining prices, accessing new technologies and new markets, and
vulnerable households or men did, which created greater inequality in adapting to climate variations and other natural factors.
the local context. Decision‐makers and implementers are advised to This chapter focuses on three types of output market information
carefully consider how the characteristics of the intervention (large‐ interventions:
scale transport infrastructure construction or simply a rural road
improvement) will have varying effects and may neglect different 1. Provision of output market information via mobile and internet;
groups of people. 2. Provision of output market information via other forms of
The long‐term effects of farm‐to‐market transport infrastructure communication; and
interventions and their role in household decision‐making around 3. Investments in ICT.
engagement in non‐farm sectors—and the implications for the local
agricultural sector—both remain under explored. The limited available We identified 81 quantitative studies carrying out quasi‐
evidence highlighted that in some contexts, interventions increase experimental or experimental evaluations for 47 output market
households' non‐farm income and labor participation. In the long‐ information interventions implemented across 19 countries. Of these
term, these interventions may lead to structural transformation of the interventions, nine have a multi‐component design combining
local economy (i.e., labor is permanently reallocated to more provision of output market information interventions or investments in
productive non‐agricultural sectors). However, it is important to ICT infrastructure alongside other intervention components. The most
consider the short‐term effects on the local agricultural sector of evaluated types of interventions are those providing output market
these changes. For example, vast and rapid rural exodus towards information via the internet or mobile phones (25 studies), followed
urban areas may have negative implications on the local agricultural by investments in ICT infrastructure (16 studies). We found that 51%
production and prices. of included estimates were rated as having a high risk of bias, 44%
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 181 of 191

were considered to have some concerns related to risk of bias, and commitments or stable arrangements in which traders offer to buy
the remaining 5% were assessed as having a low risk of bias. farmers' produce.
Information about interventions' costs is available for 13 of the 47
included interventions based on information from 13 studies. Our review of the evidence identified 94 studies reporting on 61
In the review of the effects of output market information unique interventions in 28 countries (with 28 studies focusing on
interventions, we found positive, very small, and statistically signifi- Sub‐Saharan Africa). The body of evidence includes 37 studies
cant effects on the prices received by farmers, on farm income, and on focusing on interventions combining market interventions with other
total income and wealth outcomes. We detected a small, positive market components or with other intervention categories. Most
effect on the adoption of improved varieties and quality seeds and a studies focused on the effect of arranged or curated offers from buyers
moderate, positive effect on the adoption of improved inputs (32 studies). We found that 63% of included estimates were rated as
(machinery and irrigation). We did not find a statistically significant having a high risk of bias, 33% were considered to have some
effect of these interventions on transaction costs, credit, adoption of concerns related to risk of bias, and 4% were assessed as having a
output market information, adoption of fertilizers, pesticides, and low risk of bias. Information on intervention costs is available for five
herbicides, farm investments, yields or other agricultural production of the 61 included interventions, based on information from five
outcomes, sales, or food security. Neither did we observe any studies.
evidence of the effects of these interventions on off‐farm outcomes Initiatives creating new marketplaces and alternative marketing
and nutrition. opportunities on average led to positive, small, and statistically
Based on the theory of change, these interventions are expected significant effects on sales and farm income outcomes, as well as
to permit farmers to exploit knowledge of output market information positive, very small, and statistically significant effects on total
to obtain higher prices for their products and face lower transaction income and wealth and food security outcomes. We found positive
costs due lower search costs. These effects in turn could influence and statistically significant effects on immediate outcomes (group
the quantity produced, market participation, income welfare but also participation, price received for produce, farm investment, and
adoption and investments of agricultural improved practices. adoption of improved inputs) and on yields.
However, we only found very small improvements among These benefits do not only include financial aspects (e.g., income),
beneficiaries. This may be due to farmers having insufficient access but also resources (e.g., group participation, investments, or adoption of
to credit, capital and assets to exploit market information, to adopt new resources and practices) and agricultural opportunities (e.g.,
improved inputs, or to access markets. Other elements undermining volume, sales, prices). The available evidence, however, does not
the effects may include limited networks or low technical and indicate a statistically significant reduction in farmers' transaction
management skills among smallholder farmers, as well as mis- costs.
alignment of information, strategies, entrenched relationships, low‐ Despite the overall promising results of the meta‐analysis,
quality infrastructure, and minimal beneficiary interest in using ICT several outcomes lack data to robustly measure intervention
tools. effects (e.g., crop losses, livestock type, off‐farm and non‐farm labor
market outcomes). These outcomes represent a clear gap in the
evidence base for market interventions. Also, we found very
6.1.3 | New marketplaces and alternative marketing limited evidence on the effects of the establishment of alternative
opportunities marketplaces or online commodity exchanges, especially consider-
ing the increasing number of mobile phone and internet users in
New marketplaces and alternative marketing opportunities interventions LMICs.
establish market platforms that offer novel avenues to exchange These interventions most likely contributed to decreased
goods and services or lessen the expenses related to finding such geographical search costs, reduced barriers to market entry, and
opportunities. This chapter focuses on three types of interventions: improved market competition, with implications for total income and
food security. This body of evidence emphasizes the importance of
1. Online commodity exchanges and mobile‐based marketplaces, which farmers' knowledge and education levels in accessing markets, as well
aim to improve connections between market participants by as their physical and economic access to them (e.g., land ownership,
creating new or improved marketplaces using information distance to a market, access to information).
technologies (e.g., via online commodity exchanges, apps, or
mobile services);
2. Alternative physical marketplaces interventions, which aim to 6.1.4 | Contract farming interventions
improve connections between market participants by creating
physical marketplaces (e.g., physical hubs linking farmers directly Contract farming can be defined as forward‐looking and vertical
with private companies); and market arrangements (linked to volume, value, quality, and price) in
3. Arranged or curated offers from buyers interventions, which which traders provide inputs, credit, and/or extension services to
establish connections between market participants by curating farmers as a part of an agreement to purchase their agricultural
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
182 of 191 | MARION ET AL.

produce (Little & Watts, 1994; Eaton & Shepherd, 2001; of the former, and the influence of farmers' resources (e.g., education,
Prowse, 2012). farm size, labor force, quality products) as a facilitator of participation
This chapter focuses on five types of contract farming: and benefits.

• Fixed‐price and price‐guarantee contracts, wherein the buyer

provides farmers with a guaranteed price for the delivery of 6.1.5 | Improved storage infrastructure and
agricultural produce; technologies
• Production‐management contracts, wherein the buyer provides
extension services or other technical support for the delivery of Storage infrastructure and technologies refer to the establishment,
agricultural produce; enhancement, and utilization of physical facilities and systems to
• Input or credit‐supply contracts, wherein the buyer provides inputs efficiently store, protect, and manage agricultural products and
or credit upfront and deducts the cost at harvest; commodities in terms of quality and quantity of agricultural harvest.
• Other contract types are contracts between farmers and buyers for This chapter focuses on three types of interventions:
the future delivery of agricultural produce which cannot be
categorized in the three types of contracts above. These contracts 1. On‐farm storage infrastructure and technologies interventions, which
may combine comprised a range of requirements on the provide technical, logistical, or financial support to improve farm
production (processes, crop varieties, timing of harvest, etc.) and storage practices (e.g., the use of sheds with storage off the
also on the marketing side (time of transaction, quantity, quality ground);
standards and sorting, price or price band); and 2. Storage deposit systems interventions, which enable farmers to
• Undefined contract farming where the characteristics of interven- deposit produce in off‐farm stores (e.g., warehouse receipt
tions are not known. systems); and
3. Storage technologies for transit interventions, which support the use
The review of this evidence identified 194 papers reporting the of improved technologies to maintain produce quality or prevent
effect of 129 unique interventions in 36 countries (with 73 studies it from spoiling during transit.
focusing on Africa) and through 125 studies (and 19 linked papers).
We found that 75% of included estimates were rated as having a high We found 27 studies on 20 interventions implemented in 15
risk of bias, 23% were considered to have some concerns related to countries (14 interventions in Africa, 3 in South Asia, 1 each in Latin
risk of bias, and 2% were assessed as having a low risk of bias. America and the Caribbean, East Asia and the Pacific, and Europe and
Information about intervention costs is available for seven of the 129 Central Asia, respectively). The most common types of storage
included interventions, based on information from seven studies. infrastructure interventions included storage deposit systems (11
The evidence on the effects of contract farming interventions is interventions) followed by on‐farm storage (8) and storage for transit
relatively large compared to other intervention categories of our (1). We found that 46% of included estimates were rated as having a
systematic review. These interventions firstly had positive, moderate, high risk of bias, 38% were considered to have some concerns related
and statistically significant effects on prices received by farmers for to risk of bias, and 15% were assessed as having a low risk of bias.
their produce, as well as on adoption of improved inputs, quality Information about intervention costs is available for eight of the 20
outcomes, yields, and the volume of quantity produced. included interventions, based on information from eight studies.
In terms of market‐related outcomes, farmers benefited from Improved storage infrastructure and technologies interventions led to
increased sales and farm income from contract farming interventions. moderate improvements in reducing crop loss, increasing volume of sales,
We found positive, moderate, and statistically significant effects on and also resulted in small effects in incomes and food security. We did
total income and wealth, but no statistically significant effect on food not find improvements in group participation, use/adoption of improved
security and nutrition. We also did not observe improvements in farm practices, credit, price, produce type, off‐farm income, or nutrition.
investment and non‐farm income. We found evidence gaps regarding We found evidence gaps regarding the effects of these
intervention effects on transaction costs, which is a very important interventions on important and relevant outcomes (e.g., quality of
outcome, as contract farming interventions aim to reduce transaction agricultural production). We did not identify any research on effects
costs. We also did not find any evidence on the effects of contract on transaction costs, farm investment, quality, off‐farm outcomes, or
farming on group participation, farm investment, credit, off‐farm non‐farm outcomes. The available evidence is highly focused on
outcomes, non‐farm outcomes, and nutrition. interventions in Africa and on storage deposit systems interventions.
The evidence indicates that contract farming reduced farmers' Greater evidence on the effects of storage for transit systems and on‐
uncertainty and risks associated with marketing their agricultural farm storage is needed. Similarly, the body of evidence only focused
produce, with the provision of advanced knowledge on prices, on the immediate effects of storage and ignores the effect of storage
quantity, and quality of expected agricultural production. The infrastructure over time.
evidence also highlighted the risk of power imbalances between The evidence highlights that storage infrastructure must be adapted
farmers and companies entering contract farming at the disadvantage to the local environment (physical and technical accessibility, reliability
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 183 of 191

and maintenance of those storage services) to ensure that it is output market access multi‐component interventions on the main
operational and used by beneficiaries. The barriers and facilitators outcomes included in our analysis.
analysis also emphasizes the importance of functioning, fair and inclusive We found substantial variation between the studies in each of
farmers' groups to coordinate the use of shared storage facilities. these outcome categories. When we explore potential sources of
heterogeneity, we did observe time‐related but no context‐specific
or approach‐related patterns. For yields, farm income and income and
6.1.6 | Multi‐component output market access wealth outcomes, we reported a slightly larger average effect of
interventions intervention with a larger exposure period. However, we also
distinguished a slightly smaller average effect of more recent
Effects from multi‐component output market access interventions interventions (from −0.03 to −0.02 standard deviation units) for
correspond in a subset of the effects analyzed in the previous sales, farm income and income and wealth outcomes. The magnitude
chapters. In the previous chapters, we conducted a moderator of the effects from output market access multi‐component interven-
analysis to determine whether the estimates from multi‐component tions are not different by intervention type. We found that effects
interventions were on average statistically significantly different from were slightly higher for multi‐component domestic transport infra-
those of single‐component interventions. structure interventions for transactions costs and volume outcomes
For these effects, multiple output market access interventions relative to the other types of interventions.
were combined or other activities complemented output market When comparing with the rest of the chapters, we found that the
access interventions. These effects therefore represent the aggre- results of multi‐component intervention show similar direction, size,
gated change to beneficiaries and we cannot disentangle how much and significance of effect to the ones observed in their individual
of the effect is attributable to each activity. We consider effects from components.
interventions combining:

• Multiple subcategories under the same intervention category (e.g., 6.2 | Overall completeness and applicability of
fixed‐price contracts and production‐management contracts under evidence
contract farming interventions),
• Multiple approaches under the same subcategories (e.g., different The mandate and the scope of this review is broad; it synthesizes the
on‐farm storage technologies), effects of five different intervention types (complemented by an
• Multiple approaches including one of subcategories of output analysis of multi‐component interventions) with more than sixteen
market access interventions (e.g., different farm‐to‐market transport subcategories between them, on six outcome types. While we had a
infrastructure and capacity development and advisory services), large evidence base, the spread of evidence is quite uneven, with
• Or multiple subcategories under different intervention categories some important gaps. In this section, we describe the overall extent,
(e.g., other contract types and arranged or curated offers from applicability and key gaps in the evidence base for each intervention
buyers). group.
First, we identified a relatively large body of evidence for
We found 59 studies on 53 interventions implemented in 26 interventions facilitating farm‐to‐market transport infrastructure. Of
countries (33 interventions in Africa). At the sub‐category level, the total studies, the most common infrastructure intervention was
most studies focused on the effect of arranged or curated offers domestic transport infrastructure through the construction or mainte-
from buyers (n = 22). At the category level, initiatives creating new nance of roads. More evidence on export transport infrastructure and
marketplaces and alternative marketing opportunities were the most other types of infrastructure (e.g., bridges or waterways) would
prevalent (n = 37). Arranged or curated offers from buyers interven- facilitate the generalization of findings. Moreover, although 30% of
tions were often combined with technical assistance and capacity the studies analyzed the effect of multi‐component interventions,
development (e.g., ACE and AFP) or with contract farming authors rarely disentangled how much of the effect was attributable
interventions (e.g., CATF and WPC). Similarly, domestic transport to each activity. Regarding geographies, the prevalence of Africa‐
infrastructure interventions were often combined with capacity focused studies (representing almost half of the body of evidence)
development (e.g., FAD‐2 and FAD‐3) but also other infrastructure limits the ability to generalize findings to other regions such as
development (e.g., AD2M). We found that 61% of included Europe and Central Asia (only including two studies) or MENA (not
estimates were rated as having a high risk of bias, 36% were represented in the body of evidence).
considered to have some concerns related to risk of bias, and 3% For interventions aimed at providing output market information,
were assessed as having a low risk of bias. the most evaluated types of interventions are those disseminating
We reported on the most reported outcomes among the other information via the internet or mobile phones, followed by invest-
five chapters: transaction costs, volume, yields, prices, sales, farm ments in ICT infrastructure. While our findings might inform decisions
income, total income, and food security. Overall, we found positive, in a variety of contexts for those two types of interventions, findings
very small to small, and mostly statistically significant effects of are more limited on provision of output market information through
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
184 of 191 | MARION ET AL.

other forms of communication. Once again, the prevalence of evidence 6.3 | Quality of the evidence
from Africa provides thorough insights into what works in this
context but more limited findings from other regions such as East Across all intervention types, the available evidence is subject to
Asia and Pacific (covered by three studies) or Europe and Central Asia significant risks of bias. For all chapters, the most common source of
(covered by one study). bias is related to confounding and selection bias. In this section, we
The body of evidence analyzing the effects of initiatives creating discuss the quality of studies and the evidence for each intervention
new marketplaces and alternative marketing opportunities on agricul- group in more detail.
tural outputs and market outcomes is relatively large compared to the Looking at studies on farm‐to‐market transport infrastructure
other categories in our systematic review. Our findings draw on a interventions, the most common source of bias is related to confounding
large body of evidence on arranged and curated offers from buyers but bias, which often had to do with either a lack of balance at baseline, or
fewer contexts are covered in the body of evidence on online balance at baseline being unreported. Additional concerns included
commodity exchanges and mobile‐based marketplaces. Similarly to spillovers, crossovers, and contamination and reporting bias.
previous chapter, the vast majority of studies focus on African Of the included studies on output market information interventions,
contexts while evidence from other regions is limited and hinders the confounding is the most common source of bias, which often had to do
generalizability of findings. with either a lack of balance at baseline, or balance at baseline being
Despite the large body of evidence available on the effect of unreported. Like farm‐to‐market transport infrastructure interventions,
contract farming interventions, the generalizability of findings remains additional concerns included spillovers, crossovers, and contamination
limited. This is particularly due to the lack of definition of types of and reporting bias. Similarly, for improved storage infrastructure and
contract farming interventions in studies: 22 studies are classified as technologies interventions, spillovers and contamination are the main
undefined contract farming and 20 as other contract types. While the rest sources of bias: these primarily occur due to the risk of the control
of the types of interventions are well covered, there is an opportunity group living in the same village as the beneficiary group, and thus
to bring additional contexts and approaches to expand the coverage of possibly benefitting from the intervention.
findings. Despite a larger sample of studies in East Asia and Pacific and For initiatives creating new marketplaces and alternative marketing
South Asia, more than half of the studies remain focused on the African opportunities, the majority of studies were rated as having a high risk
context which might affect the overall findings. of bias, and most QED studies carried a high risk of bias. The most
Finally, storage deposit systems have a thinner evidence base. common source of bias is related to selection bias and confounding
The most common interventions focused on storage deposit systems bias. Regarding the risk of confounding, several authors do not report
and improved on‐farm storage infrastructure and technologies. testing for parallel time trend assumption, despite their use of DID
However, our findings do not allow us to draw conclusions on design. Additional concerns included spillovers, crossovers, and
improved storage technologies for transit as it only includes one contamination, reporting bias, and outcome measurement bias.
study. The available evidence is highly focused on interventions in The body of evidence on contract farming is characterized by a
Africa (13 of the 19 included interventions took place in 11 relatively higher risk of bias compared to the other four intervention
countries in Africa) which limits the ability to draw conclusions on categories we reviewed. Indeed, 76% of included contract farming
effects in other contexts. estimates were assessed as having a high risk of bias, as compared to
Across chapters we observe crosscutting gaps of evidence that approximately 50%–60% for other intervention groups. This is primarily
may limit the generalizability of findings as the evidence is unevenly because of selection bias, bias from confounding effects, and reporting
distributed across contexts and settings. For example, Africa bias; contract farming estimates were more likely to have a high risk of
represents over half of the studies in our analysis (153 of 289 selection bias and confounding bias than other interventions.
studies), but we did not find any evidence of evaluated interventions Included studies often lacked precise information on the location
in MENA, and it was primarily concentrated in Anglophone‐speaking of targeted beneficiaries of contract farming interventions. Contract
countries in Eastern and Western Africa. Similarly, in South Asia, the farming is an older and more widespread intervention in LMICs than
body of evidence is heavily focused on India (36 of 59 studies). the other intervention types. This made the identification of an
Therefore, the effects of these interventions have not been adequate counterfactual more complicated, and significantly
researched in many contexts, which severely limits the utility of the increased the risk of selection bias.
evidence for informing policymaking decisions in these countries. In Included studies were more likely to use statistical matching,
addition, the low number of studies assessed as low risk of bias might endogenous treatment effect methods, or IV methods than in other
affect the accuracy of findings in some contexts. Moreover, some chapters. The chosen IVs were often not exogenous, and authors
outcomes suffered gaps of evidence to run meta‐analyses across used endline time‐variant covariates in the statistical matching
contexts: this is the case for off‐farm income, group participation, crop estimation—both of which contributed to a high risk of confounding
loss, quality, access to credit or transaction costs. Finally, in the absence effects for estimates in included studies.
of an additional search for qualitative and cost evidence, more Information was missing on context, intervention design, and
evidence may be available to expand the findings on value for money, implementation. This was especially the case when synthesizing the
barriers and facilitators across geographies and interventions. effects of contract farming interventions (Chapter 4). In 21 studies,
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 185 of 191

authors did not provide sufficient information on the evaluated of policy initiatives and a greater number of outcomes than previous
intervention to categorize it as a contract farming intervention. Similarly, reviews. Overall, the findings are partially consistent with the
for 49 interventions, information on the first year of the intervention was evidence provided by previous systematic reviews.
not specified in the included studies. Both omissions limited the analysis. The results of our review are aligned with those of Aker and
Although most of the manuscripts across all five chapters colleagues (2016), who reported mixed evidence on the effects of
reported sample characteristics, standard errors, and standard output‐market information‐technology initiatives (such as information
deviations, details were missing in 16 studies on the exact sample interventions using mobile phones) on farmers' behavior and welfare
sizes or precision estimates. We contacted authors to collect missing outcomes. However, whereas Aker and colleagues (2016) conducted
data, but in the absence of responses for 12 studies, the effects with a non‐systematic literature review, we conducted a rigorous
missing or unclear data could not be included in the analysis. systematic review to take stock of the existing experimental and
quasi‐experimental evidence.
In line with Ton and colleagues (2017), we found that contract
6.4 | Potential biases in the review process farming can improve farm income, although we detected smaller
effects. This incongruence might be explained by the substantially
There are several limitations to the review process that should be different size of the analyzed body of evidence: the study by Ton
borne in mind when interpreting the results. and colleagues (2017) relied on 75 studies, 22 of which were
First, many of the meta‐analyses presented in this review are based meta‐analyzed; our review includes 193 total studies, 143 of
on a small number of estimates. This implies that bias might arise due which were utilized for meta‐analyzing 127 interventions across
to variance between studies, although the random‐effects model 36 countries.
employed allows us to partially account for heterogeneity between Our results also corroborate findings reported by Hine and
studies. Moreover, in numerous cases the small number of estimates colleagues (2016, 2019) and Ludwig et al. (2016), who found
precluded us from testing the presence of publication bias and that rural road investment is associated with increased income
performing the moderator analysis to detect potential sources of (and poverty reduction), employment, and agricultural output.
variation. Nevertheless, we opted to perform small k meta‐analyses, Again, however, the magnitude of our effects is generally smaller,
since the latter are preferable to the unknown bias of ad hoc and we did not detect the same statistically significant improve-
summaries (Borenstein et al., 2021). ments in sales. Thus, our findings are more aligned with de Abreu
Another limitation is the large share of included studies judged to et al. (2022), who conducted a systematic review with narrative
have high risk of bias, which raises concerns about the internal validity of analysis of rural transportation interventions and reported mixed
the analyzed evidence. Namely, roughly 90% of the meta‐analyzed effects.
estimates either do not provide sufficient information to assess the Unlike these studies, our systematic review covers a substantially
risk of bias, or have been assessed as having high risk of bias. The large body of evidence; moreover, our review contributes rigor,
main sources of risk of bias are confounding factors, spillovers and particularly regarding the reporting of evidence and risk of bias
crossovers, and selection bias. assessment.
Another factor is the proportion of included estimates from multi‐ In some instances, the findings of our review diverge from those
component interventions (19%). In many cases, authors provided informa- in existing reviews. For instance, Nandi and colleagues (2021)
tion on the effects of the overall interventions without isolating the conducted a systematic review with descriptive analysis of 28
effects of each component. This limited our ability to attribute the effects studies, reporting on the associations between market access and
of multi‐component interventions and determine the role of each activity. food security. In contrast, our study conducted a quantitative meta‐
Finally, data extraction and risk of bias assessments are based on analysis covering a higher number of studies and found either very
data reported in the manuscripts (and in supplementary materials). small or non‐significant average effects on food security and nutrition
Although authors were contacted about missing quantitative data, across the five intervention groups. Besides the methodological
they were not contacted for missing descriptive data nor for data factors, this difference can be explained by the larger set of
related to risk of bias assessments. In cases where authors omitted databases searched in our review, which resulted in a considerably
critical quantitative data necessary for estimating the effects, we higher number of included studies.
estimated the relevant information with conservative assumptions Other factors accounting for the discussed differences could be
that might underestimate the intervention impact. the difference in sources and databases searched, the adopted
inclusion criteria, and differences in how outcomes are coded or
grouped. An equally influential factor could be related to the type of
6.5 | Agreements and disagreements with other studies included in the reviews. Namely, the papers by Aker and
studies or reviews colleagues (2016), Nandi and colleagues (2022), de Abreu et al.
(2022), and Kaiser and Barstow (2022) reviewed studies with no
This systematic review is the most comprehensive and up‐to‐date restrictions on study design, whereas our review focused exclusively
review on these topics, as it covers current evidence on a wider range on studies using experimental designs and QEDs.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
186 of 191 | MARION ET AL.

7 | A U TH O R S ' C O NC LUS I O NS and practitioners can ensure information access or education on the
technologies or services provided (e.g., the purpose of the new road,
7.1 | Implications for practice and policy cost of the phone service, distance to the new market, terms and
conditions of the farming contract, user guidelines for a storage
Policymakers and practitioners can rely on output market access technology). Without this prerequisite, farmers might not realize the
interventions to improve agricultural and socio‐economic outcomes of full benefits of the interventions or may not have optimal use of the
farmers in LMICs. The channel of effects varies across types of output newly provided resources or services. Similarly, policymakers and
market access interventions. For example, some interventions, such as practitioners should ensure that these resources and services are
creating new marketplaces or alternative marketing opportunities, will physically accessible for farmers to enjoy their benefits (e.g., road or
primarily impact immediate outcomes and bear effects further along storage infrastructure needs maintenance over time, phone connec-
the causal chain. In contrast, other interventions, such as improved tions need to be reliable, and the market needs to be connected and
storage, have positive and statistically significant effects on agricul- geographically close enough to farmers).
tural production and intermediate outcomes. Policymakers can still invest in expanding the body of evidence.
In terms of food security and nutrition outcomes, conclusions from Despite the relatively large body of evidence on the effect of
the available evidence are much less clear for policymakers and agricultural output market access interventions, the scarcity of
practitioners. First, evidence on the effects of output market access evidence on costs might limit the ability of policymakers and
on these outcomes, especially nutrition, is thinner and more sparsely practitioners to know the value‐for‐money of interventions. Funding
distributed across contexts than for other outcomes. Second, the of cost‐benefit analyses could therefore contribute to a more
available evidence is only relevant for a specific time of the year. We accurate analysis of those interventions' aspects and effects.
did not see any effects of output market access interventions on food
security or nutrition seasonality. Finally, the causal pathway between
market access interventions and improvements in food security and 7.2 | Implications for research
nutrition are more complex than for other outcomes, meaning that
local social and intra‐household dynamics may play a greater role and Researchers can still fill many contextual evidence gaps. Research should
should be further understood. work on filling evidence gaps, as this will allow a better understanding
The specific needs and dynamics of each context should inform the of the effects of interventions across contexts and outcomes. For
choice of the intervention and approach. Despite interventions' overall example, while Africa represents over half of the studies in our analysis,
positive and statistically significant effects, policymakers and practi- none focuses on countries in North Africa or the Middle East, which
tioners should consider the variations according to context, population, might be an area of interest for future research. Similarly, while
or exposure to the intervention. This is particularly true for vulnerable outcomes such as farm income or total income and wealth gathered over
groups (e.g., small landowners, those with lower levels of education, 50 estimates for some interventions, other outcomes such as farm
women). Tailoring the intervention design and approach to the local investments, group participation or non‐farm outcomes still lack enough
context's specificities will encourage the local population's participation evidence to analyze the effect of interventions through meta‐analyses
and maximize the sustainability of effects over time. and would benefit from a prioritization for future research.
Across all interventions, access to information appears as a critical Future research should mitigate the risks of confounding and
facilitator of participation and impact. Across chapters, authors have selection biases. Researchers can contribute to improving confidence
also highlighted the risk of information gaps as a potential barrier to in findings by considering those risks of bias in the design of their
impact. Reducing the information gap between stakeholders, policy- research products. Developing approaches to mitigate those risks will
makers, and practitioners can enable higher levels of trust in the improve the accuracy of findings and increase the overall quality and
intervention—leading to greater participation, but also catalyzing the confidence in the evidence.
adoption of resources or practices, reducing vulnerability to price Studies should report their research in a way that allows the
volatility or natural phenomena, and increasing farmers' bargaining identification of precise effect estimates and provides sufficient information
power in contract negotiations. According to the type of intervention, to assess study quality. Suggestions include registering a pre‐analysis plan,
the type of information to be provided varies. While information on establishing a relevant counterfactual (using either random assignment or
prices supports market interventions, access to information on quasi‐experimental methods), and collecting data from all participants at
contract terms and crops is a resource for farmers entering contract baseline, where feasible, and at equivalent intervals for both control and
farming. Authors also comment on the diversity of approaches treatment groups during post‐intervention follow‐ups.
information sharing can take: development of communication When reporting findings, authors should report on all outcomes
infrastructure, participation in farmers groups, provision of mobile without regard for the significance of the results, account for multiple
phone, or internet‐based information. This will allow beneficiaries to hypothesis testing, and report precise data (including exact p‐values,
maximize their benefits and potentially increase intervention impact. standard errors of regression coefficients, and standard deviations
Policymakers and practitioners should be mindful of both the when reporting means). To help assess study quality, authors should
technical and physical accessibility of their interventions. Policymakers also be transparent about the assignment mechanism, the baseline
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 187 of 191

balance between control and treatment groups, overall attrition and Please note that this is the recommended optimal review team
differential attrition, potential concerns related to spillovers or composition.
performance bias, and any deviations from the intended intervention.
Including cost data will provide crucial insights into the variability of • Content: PM, EL, AF, and SP developed the content of the review
costs across different contexts and will enhance understanding of with inputs and quality assurance from SS and Anil Thota (AT).
interventions' economic implications. This can be achieved by including • Systematic review methods: PFV, Tomasz Kozakiewicz (TK) and SS
cost‐effectiveness or cost‐benefit analyses, or a more systematic drafted the review methods.
reporting of the costs of interventions. Similarly, including qualitative • Statistical analysis: SS oversaw the statistical analysis. PM, EL, AF,
analysis and using mixed methods can provide additional insights into SP and a pool of consultants extracted effects data, with quality
implementation considerations and inform future design and invest- assurance from PM, EL, AF and SP. PM, EL, AF and SP also led
ment in agricultural output market access interventions. outcome classification and grouping with quality assurance and
inputs from SS, PM and EL.
A C KN O W L E D G M E N T S • Information retrieval: PFV, TK and Vinitha Bachina (VB) developed
We are grateful to 3ie colleagues who provided input at various the initial search strings and quality assured by Sarah Young
stages of the review and supported on the administrative, (Information specialist). PFV and TK led screening, gray literature
communication and finance management. We are grateful to the review, snowballing and references checks with the support of a
consultants and colleagues who contributed to the screening, pool of consultants and under the oversight of VB.
retrieval and extractions processes: Tomasz Kozakiewicz, Sarah
Young, and Vinitha Bachina. We would also like to thank Hope DECLARATIONS OF INTE REST
Michelson, the subject expert for this systematic review, for her There are no reported conflicts of interest in this review. One of
comments and the members of our advisory group who provided the review authors is involved with the International Development
comments on the systematic review content: David Spielman, Kyle Coordination Group of the Campbell Collaboration. However, the
Emerich, Marcella McClatchey, Melissa Schweisguth and Loren- IDCG editor for this review is not involved in the review. The
zo Casaburi. We are also grateful to the other members of the IDCG's independent co‐chair will also independently assure this
Advisory group: Boran Altincicek, Jonathan Robinson, Kevin review.
Donovan, Rocco Macchiavello, Shoumitro Chatterjee, Wyatt
Brooks and Yaw Nyarko. The authors of this systematic review PRELIMINARY TIMEFRAME
would like to specifically thank the following people for their The time frame for this review was as follows:
contributions which made this systematic review possible: Kishore
Basak, William Fernandez, Sangyoung Jung, María Camila Arias • Draft and Final Protocols: May 2023–August 2023.
Álvarez, Vinitha Bachina, Clément Dumont, Juan Pascual Torres, • Draft and Final Report: October–December 2023.
Zahin Shafiq, Danish Us‐Salam and Zhaocheng Gu. Finally, thanks
to all the authors who shared unpublished and forthcoming papers PLANS FOR UPDATING THIS REVIEW
and any additional data that helped improve the review. The search of this study has been completed in November 2022
based on a search strategy developed in May 2022. The core team
C O NT R I B U T I O N S OF AU TH O R S will explore opportunities for funding an update of this review upon
Pierre Marion (PM), Etienne Lwamba (EL), Andrea Floridi (AF), submission of the final report in 2024.
Suvarna Pande (SP), Megha Bhattacharyya (MB), Paul Fenton Villar
(PFV) and Shannon Shisler (SS) are the core team for this review. DI F F ER EN CE S B ET WE EN P R OT OCOL AND RE VI EW
PM is a research associate with experience in agriculture, rural The title of this systematic review was slightly changed from “The
development and education sector. EL is an evaluation specialist effects of agricultural output market access interventions on
with experience in the conflict and development sector. AF is an agricultural, socio‐economic and food and nutrition security out-
evaluation specialist with experience in development economics, comes in low‐ and middle‐income countries: A systematic review” to
private sector development, and informal economy. SP is a “The effects of agricultural output market access interventions on
research associate specializing in development economics particu- agricultural, socio‐economic, food security and nutrition outcomes in
larly on behavioral interventions, gender, violence, economic and low‐ and middle‐income countries: A systematic review.”
social preferences and decision‐making. MB is a research assistant The first three research questions were slightly changed from the
with experience in development economics, health, and climate following:
change. PFV is a former evaluation specialist at 3ie. SS is a
systematic review and quantitative methods expert, with a PhD in (1) What does evidence indicate about the direction and magnitude
Quantitative Methods and over a decade of experience in of the effects of output market access interventions on
designing, managing and analyzing quantitative research, including agricultural, socio‐economic and food and nutrition security
meta‐analyses. outcomes?
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
188 of 191 | MARION ET AL.

(2) How does the distribution of effects differ across different Guatemala. Economic Development and Cultural Change, 41(2),
contexts, interventions, and outcomes and do the effects of 319–342.
Alene, A. D., Manyong, V. M., Omanya, G., Mignouna, H. D., Bokanga, M.,
interventions differ between sub‐groups of the population (e.g.,
& Odhiambo, G. (2008). Smallholder market participation under
according to sex, race, ethnicity, age, socio‐economic status, type transactions costs: Maize supply and fertilizer demand in Kenya.
of produce, farm size, etc.)? Food Policy, 33(4), 318–328.
(3) What is the risk of bias of studies on the effects of output market Aloe, A. M., Becker, B. J., Duvendack, M., Valentine, J. C., Shemilt, I., &
access interventions on farmers' on farmers' agricultural, socio‐ Waddington, H. (2017). Quasi‐Experimental Study Designs Series—
Paper 9: Collecting Data from Quasi‐Experimental Studies. Journal of
economic and food and nutrition security outcomes in LMICs?
Clinical Epidemiology, 89, 77–83.
Aquino, C. (1998). When tariffs rule: Philippine smallholder agriculture under
We also adjusted the name of three of the included outcomes in the GATT/WTO tariff and trade liberalization regime. Philippine
this review. The outcome name “non/off farm outcomes” was Peasant Institute.
Baltenweck, I., & Staal, S. (2007). Beyond one‐size‐fits‐all: Differentiating
changed to “non‐farm outcomes” to eliminate any confusion with
market access measures for commodity systems in the Kenyan
other outcome names (notably “off‐farm outcomes”). We also highlands. Journal of Agricultural Economics, 58(3), 536–548.
changed the name of “aggregation” outcome to “Group participation” Barbier, E. B. (1989). Cash crops, food crops, and sustainability: The case
and we changed “access to credit” to “credit” as it also measures use of Indonesia. World Development, 17(6), 879–895.
Barrett, C. B. (2008). Smallholder market participation: Concepts and
of credit, not just access to credit.
evidence from eastern and southern Africa. Food Policy, 33(4),
As specified in the protocol, in the Dealing with missing data section, 299–317.
we did not report the impact that information obtained from authors had Barrett, C. B., Bachke, M. E., Bellemare, M. F., Michelson, H. C.,
on the results (i.e., using sensitivity analyses) as we only obtained Narayanan, S., & Walker, T. F. (2012). Smallholder participation in
information on four studies after contacting authors of 16 studies. contract farming: Comparative evidence from five countries. World
Development, 40(4), 715–730.
Also, as specified in the Data synthesis section of the protocol,
Barrett, C. B., Carter, M. R., & Timmer, C. P. (2010). A Century‐Long
we did not examine the statistical correspondence between welfare
Perspective on Agricultural Development. American Journal of
outcomes and other more proximate outcomes in the conceptual Agricultural Economics, 92(2), 447–468.
model by examining the correspondence in effects between Begg, C. B., & Mazumdar, M. (1994). Operating characteristics of a rank
immediate and agricultural outcomes and welfare outcomes. We correlation test for publication bias. Biometrics, 50(4), 1088.
Bellemare, M. F. (2010). Agricultural extension and imperfect supervision
had planned to conduct this through descriptive analysis and, where
in contract farming: Evidence from Madagascar. Agricultural
possible, regression analysis and fuzzy Qualitative Comparative Economics, 41(6), 507–517.
Analysis (QCA) (Thomas et al., 2014; Ton et al., 2017). We also Bellemare, M. F., & Bloem, J. R. (2018). Does contract farming improve
planned to document the results and correspondence of estimated welfare? A review. World Development, 112, 259–271.
Bellemare, M. F., & Lim, S. (2018). In all shapes and colors: Varieties of contract
effects for studies that provided insights through general equilibrium
farming. Applied Economic Perspectives and Policy, 40(3), 379–401.
analysis, as well as partial effects.
Bergquist, L. F., & Dinerstein, M. (2020). Competition and entry in
agricultural markets: Experimental evidence from Kenya. American
S O U RC E S OF S U P P O RT Economic Review, 110(12), 3705–3747.
Bergquist, L. F., & McIntosh, C. (2021). Search cost, intermediation, and
trade: Experimental evidence from Ugandan agricultural markets.
I n te rn a l s o u r c e s
Working paper. https://www.atai-research.org/wp-content/
None. uploads/2021/08/BML_paper_v8-copy.pdf
Bold, T., Ghisolfi, S., Nsonzi, F., & Svensson, J. (2022). Market access and
External source s quality upgrading: Evidence from four field experiments. American
Economic Review, 112(8), 2518–2552.
Borenstein, M., Hedges, L. V., Higgins, J. P. T., & Rothstein, H. R. (2009a).
Introduction to meta‐analysis. John Wiley & Sons Ltd.
REFERENCES Borenstein, M., Hedges, L. V., Higgins, J. P. T., & Rothstein, H. R. (2009b).
de Abreu, V. H. S., Santos, A. S., & Monteiro, T. G. M. (2022). Climate Multiple comparisons within a study. In M. Borenstein, L. Hedges, J.
change impacts on the road transport infrastructure: A systematic Higgins, & H. R. V. Rothstein (Eds.), Introduction to meta‐analysis (Ch.
review on adaptation measures. Sustainability, 14, 8864. https://doi. 25). (pp. 239–242). John Wiley & Sons Ltd.
org/10.3390/su14148864 Bouis, H. E., & Haddad, L. J. (1990). Effects of agricultural commercialization
Aggarwal, S., Giera, B., Jeong, D., Robinson, J., & Spearot, A. (2022). on land tenure, household resource allocation, and nutrition in the
Market access, trade costs, and technology adoption: Evidence from Philippines. International Food Policy Research Institute in collabo-
Northern Tanzania. Unpublished working paper. ration with the Research Institute for Mindanao Culture. https://
Aker, J. C. (2008). Does digital divide or provide? The impact of cell books.google.co.in/books?id=A3wumdoEWHUC
phones on grain markets in Niger. Center for Global Development de Brauw, A., & Bulte, E. (2021). African farmers, value chains and
Working Paper No.154. agricultural development. An economic and institutional perspective.
Aker, J. C., Ghosh, I., & Burrell, J. (2016). The promise (and pitfalls) of ICT Palgrave Macmillan Cham.
for agriculture initiatives. Agricultural Economics, 47(S1), 35–48. Brooks, W., & Donovan, K. (2020). Eliminating uncertainty in market
Alarcon, J. A. (1993). Household income, food availability, and commercial access: The impact of new bridges in Rural Nicaragua. Econometrica,
crop production by smallholder farmers in the western highlands of 88(5), 1965–1997.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 189 of 191

Brown, E. D., & Tanner, J. (2019). Integrating value for money and impact Gertler, P., Martinez, S., Premand, P., Rawlings, L. B., & Vermeersch, C. M.
evaluations: Issues, institutions, and opportunities, World Bank Policy (2011). Impact evaluation in practice. World Bank Publications.
Research Working Paper 9041. Independent Evaluation Group, the Giovannucci, D., & Shepherd, A. (2007). Market information services,
World Bank. https://openknowledge.worldbank.org/handle/10986/ 996770. Social Science Research Network. https://papers.ssrn.com/
32586 abstract=996770
Cardell, L., & Michelson, H. (2023). Price risk and small farmer maize Gómez, M. I., Barrett, C. B., Buck, L. E., De Groote, H., Ferris, S.,
storage in Sub‐Saharan Africa: New insights into a long‐standing Gao, H. O., McCullough, E., Miller, D. D., Outhred, H., Pell, A. N.,
puzzle. American Journal of Agricultural Economics, 105(3), 737–759. Reardon, T., Retnanestri, M., Ruben, R., Struebi, P., Swinnen, J.,
https://doi.org/10.1111/ajae.12343 Touesnard, M. A., Weinberger, K., Keatinge, J. D. H., Milstein, M. B.,
Casaburi, L., & Reed, T. (2021). Using individual‐level randomized treatment & Yang, R. Y. (2011). Research principles for developing country
to learn about market structure. Working Paper. https://www.econ. food value chains. Science, 332(6034), 1154–1155.
uzh.ch/dam/jcr:5a734efb-afb6-4552-92f9-282af8930811/cocoa_ Govereh, J., Govereh, J., & Jayne, T. S. (2003). Cash cropping and food
paper_AEJ2021.pdf crop productivity: Synergies or trade‐offs? Agricultural Economics,
de Castro, A. B. R. (2021). Impact of Agricultural Market Linkages on 28(1), 39–50.
Small‐Scale Farmers' Welfare: Evidence from Tanzania. Paper Goyal, A. (2010). Information, direct access to farmers, and rural market
presented at the 31st International Conference of Agricultural performance in central India. American Economic Journal: Applied
Economists, held on the 17–31 August 2021. https://agris.fao.org/ Economics, 2(3), 22–45.
agris-search/search.do?recordID=US2022245917 Haggblade, S., & Hazell, P. (1989). Agricultural technology and farm‐
Chamberlin, J., & Jayne, T. S. (2013). Unpacking the meaning of ‘Market nonfarm growth linkages. Agricultural Economics, 3(4), 345–364.
Access’: Evidence from rural Kenya. World development, 41, 245–264. Haggblade, S., Hazell, P., & Brown, J. (1989). Farm‐nonfarm linkages in
Chatterjee, S. (2019). Market power and spatial competition in rural India. rural sub‐Saharan Africa. World Development, 17(8), 1173–1201.
Faculty of Economics, University of Cambridge Working Papers in Hedges, L. V. (2009). Stochastically dependent effect sizes. In H. Cooper,
Economics, No. 1921. L. V. Hedges & J. C. Valentine (Eds.), The handbook of research
Dewey, K. G. (1981). Nutritional consequences of the transformation from synthesis and meta‐analysis (Ch.13) (pp. 281–298). Russell Sage
subsistence to commercial agriculture in Tabasco, Mexico. Human Foundation.
Ecology, 9(2), 151–187. Higgins, J., Thomas, J., Chandler, J., Cumpston, M., Li, T., Page, M., &
Dillon, B., & Dambro, C. (2017). How competitive are crop markets in Sub‐ Welch, V. (Eds.). (2020). Cochrane handbook for systematic reviews of
Saharan Africa? American Journal of Agricultural Economics, 99(5), interventions, version 6.1 (updated September 2020). Cochrane.
1344–1361. Hildebrandt, N., Nyarko, Y., Romagnoli, G., & Soldani, E. (2020). Price
Dumas, C., & Játiva, X. (2020). Better roads, better off? Evidence on information, inter‐village networks, and ‘Bargaining Spillovers’: Experimental
improving roads in Tanzania. FSES Working Papers 518. Faculty of evidence from Ghana, 3694558. Social Science Research Network.
Economics and Social Sciences, University of Freiburg. https://ideas. Hine, J., Abedin, M., Stevens, R. J., Airey, T., & Anderson, T. (2016). Does
repec.org/p/fri/fribow/fribow00518.html the extension of the rural road network have a positive impact on
Eaton, C., & Shepherd, A. (2001). Contract farming: Partnerships for growth, poverty reduction and resilience for the rural areas served? If so how,
FAO Agricultural Services Bulletin 145. Food & Agriculture Organiza- and if not why not? A systematic review. EPPI‐Centre, Social Science
tion of the United Nations. https://pacificfarmers.com/wp-content/ Research Unit, UCL Institute of Education, University College
uploads/2014/07/Contract-farming-by-CharlesEatonA. London.
Shepherd-.pdf Hine, J., Sasidharan, M., Torbaghan, M. E., Burrow, M., & Usman, K.
Fafchamps, M. (1992). Cash crop production, food price volatility, and (2019). Evidence of the impact of rural road investment on poverty
rural market integration in the third world. American Journal of reduction and economic development, Knowledge, Evidence & Learning
Agricultural Economics, 74(1), 90–99. for Development (K4D) helpdesk report. Institute of Development
Fan, S., & Rue, C. (2020). The role of smallholder farms in a changing Studies.
world. In S. Gomez y Paloma, L. Riesgo & K. Louhichi (Eds.), The role International Fund for Agricultural Development (IFAD). (2022). Poverty
of smallholder farms in food and nutrition security (pp. 13–28). reduction through market participation. Market access and value
Springer International Publishing. chains. Available at https://www.ifad.org/en/market-access
FAO. (2022a). Food coalition: Act, accelerate, transform. Food and de Janvry, A., Fafchamps, M., & Sadoulet, E. (1991). Peasant household
Agriculture Organization of the United Nations. https://www.fao. behaviour with missing markets: Some paradoxes explained. The
org/food-coalition/en/ Economic Journal, 101(409), 1400–1417.
FAO. (2022b). Tracking progress on food and agriculture‐related SDG de Janvry, A., & Sadoulet, E. (2002). World poverty and the role of
indicators 2022. Food and Agriculture Organisation. https://doi.org/ agricultural technology: Direct and indirect effects. Journal of
10.4060/cc1403en Development Studies, 38(4), 1–26.
FAO, IFAD, UNICEF, WFP and WHO. (2022). The State of Food Security de Janvry, A., & Sadoulet, E. (2020). Using agriculture for development:
and Nutrition in the World 2022. Repurposing food and agricultural Supply‐ and demand‐side approaches. World Development, 133,
policies to make healthy diets more affordable. Food and Agriculture 105003.
Organization of the United Nations. Kaiser, N., & Barstow, C. K. (2022). Rural transportation infrastructure in
Farmer Income Lab. (2018). What works to increase smallholder farmers' low‐ and middle‐income countries: A review of impacts, implications,
income? A landscape review. Farmer Income Lab Working paper. and interventions. Sustainability, 14, 2149. https://doi.org/10.3390/
https://www.farmerincomelab.com/sites/g/files/jydpyr621/files/ su14042149
2019-09/What%20Works_FINAL_9.19.pdf Katunze, M., Kuteesa, A., Mijumbi, T., & Mahebe, D. (2017). Uganda
Fung, W., Liverpool‐Tasie, L. S. O., Mason, N. M., & Oyelere, R. U. (2020). Do warehousing receipt system: Improving market performance and
crop purchase programs improve smallholder welfare? The case of productivity. African Development Review, 29(2), 135–146.
Zambia's Food Reserve Agency. Agricultural Economics, 51(4), 519–533. Keef, S. P., & Roberts, L. A. (2004). The meta‐analysis of partial effect
George, T. (2014). Why crop yields in developing countries have not kept sizes. British Journal of Mathematical and Statistical Psychology, 57(1),
pace with advances in agronomy. Global Food Security, 3(1), 49–58. 97–129.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
190 of 191 | MARION ET AL.

Key, N., Sadoulet, E., & Janvry, A. D. (2000). Transactions costs and Nakasone, E. (2013). The role of price information in agricultural markets:
agricultural household supply response. American Journal of Experimental evidence from rural Peru. Paper presented at the
Agricultural Economics, 82(2), 245–259. Agricultural & Applied Economics Association's 2013 AAEA & CAES
Khandker, S. R., & Koolwal, G. B. (2010). How infrastructure and financial Joint Annual Meeting, held on the 4–6 August 2013, Washington,
institutions affect rural income and poverty: Evidence from DC. https://ageconsearch.umn.edu/record/150418
Bangladesh. Journal of Development Studies, 46(6), 1109–1137. Nandi, R., Nedumaran, S., & Ravula, P. (2021). The interplay between food
Kozakiewicz, T., Snilstveit, B., & Engelbert, M. (2021). A classification of market access and farm household dietary diversity in low and middle
interventions and outcomes for international development evidence. income countries: A systematic review of literature. Global Food
International Initiative for Impact Evaluation (3ie). https://www. Security, 28, 100484. https://doi.org/10.1016/j.gfs.2020.100484
3ieimpact.org/blogs/classification-interventions-and-outcomes- Narayanan, S. (2014). Profits from participation in high value agriculture:
international-development-evidence Evidence of heterogeneous benefits in contract farming schemes in
Kugley, S., Wade, A., Thomas, J., Mahood, Q., Jørgensen, A. M. K., Southern India. Food Policy, 44, 142–157.
Hammerstrøm, K., & Sathe, N. (2017). Searching for studies: A guide Oliver, S., Dickson, K., Bangpan, M., & Newman, M. (2017). Getting started
to information retrieval for Campbell systematic reviews. Campbell with a review. An Introduction to Systematic Reviews (pp. 71–92).
Systematic Reviews, 13(1), 1–73. SAGE Publications.
Lane, C., Marion, P., Pande, S., Storhaug, I., CordovaArauz, D. B., & Omotilewa, O. J., Ricker‐Gilbert, J., Ainembabazi, J. H., & Shively, G. E. (2018).
Fenton Villar, P. (2022). Rapid evidence assessment: What facilitates Does improved storage technology promote modern input use and food
success and failure of food systems interventions in the long‐term. security? Evidence from a randomized trial in Uganda. Journal of
International Initiative for Impact Evaluation (3ie). Development Economics, 135, 176–198.
Levi, R., Rajan, M., Singhvi, S., & Zheng, Y. (2020). The impact of unifying Oya, C., Schaefer, F., & Skalidou, D. (2018). The effectiveness of
agricultural wholesale markets on prices and farmers' profitability. agricultural certification in developing countries: A systematic
Proceedings of the National Academy of Sciences, 117(5), review. World Development, 112, 282–312.
2366–2371. Oya, C., Schaefer, F., Skalidou, D., McCosker, C., & Langer, L. (2017).
Lothoré, A., & Delmas, P. (2009). Market access and agricultural product Effects of certification schemes for agricultural production on socio‐
marketing: Promoting farmer initiatives. Afd, CTA and Inter‐Réseaux economic outcomes in low‐ and middle‐income countries: A
Développement Rural. systematic review. Campbell Systematic Reviews, 13(1), 1–346.
Ludwig, C., Nagarajan, G., & Zaman, L. (2016). Systematic review of the Page, M. J., McKenzie, J. E., Bossuyt, P. M., Boutron, I., Hoffmann, T. C.,
effects of rural roads on expanding agricultural markets in developing Mulrow, C. D., Shamseer, L., Tetzlaff, J. M., Akl, E. A., Brennan, S. E.,
countries. Social Impact Inc. Chou, R., Glanville, J., Grimshaw, J. M., Hróbjartsson, A., Lalu, M. M.,
Maertens, A., Mhango, W., & Michelson, H. (2020). The effect of Li, T., Loder, E. W., Mayo‐Wilson, E., McDonald, S., … Moher, D.
demonstration plots and the warehouse receipt system on integrated (2021). The PRISMA 2020 statement: An updated guideline for
soil fertility management adoption, yield and income of smallholder reporting systematic reviews. Systematic Reviews, 10(1), 89. https://
farmers: A study from Malawi's Anchor Farms, 3ie Impact Evaluation doi.org/10.1186/s13643-021-01626-4
Report 122. International Initiative for Impact Evaluation (3ie). Pingali, P., Aiyar, A., Abraham, M., & Rahman, A. (2019). Linking farms to
Magesa, M. M., Michael, K., & Ko, J. (2020). Access and use of agricultural markets: Reducing transaction costs and enhancing bargaining
market information by smallholder farmers: Measuring informational power. In Transforming food systems for a rising India (Ch.8) (pp.
capabilities. The Electronic Journal of Information Systems in 193–214). Palgrave Studies in Agricultural Economics and Food
Developing Countries, 86(6), e12134. Policy. Palgrave Macmillan.
Magnan, N., Hoffmann, V., Opoku, N., Gajate‐Garrido, G., & Kanyam, D. A. Platteau, J. P. (1996). The evolutionary theory of land rights as applied to
(2021). Information, technology, and market rewards: Incentivizing Sub‐Saharan Africa: A critical assessment. Development and Change,
aflatoxin control in Ghana. Journal of Development Economics, 151, 27(1), 29–86.
102620. R Core Team. (2020). R: A language and environment for statistical
Maitre, H. E., Lemeilleur, S., & Bienabé, E. (2011). Linking smallholders to computing. R Foundation for Statistical Computing. https://www.r-
efficient markets. Paper presented at the 3rd European Forum on project.org
Rural Development, Breakout Session no. 6, held on the 29 March–1 Ragasa, C., Lambrecht, I., & Kufoalor, D. S. (2018). Limitations of contract
April 2011, Palencia, Spain. https://agritrop.cirad.fr/560830/3/ farming as a pro‐poor strategy: The case of maize outgrower
document_560830.pdf schemes in upper west Ghana. World Development, 102, 30–56.
Markelova, H., Meinzen‐Dick, R., Hellin, J., & Dohrn, S. (2009). Collective Reardon, T. (1997). Using evidence of household income diversification to
action for smallholder market access. Food Policy, 34(1), 1–7. inform study of the rural nonfarm labor market in Africa. World
Maziku, P., & Mashenene, R. (2020). Effect of non‐tariff barriers on maize Development, 25(5), 735–747.
production and marketing by smallholder farmers in Tanzania. Global Reeves, B. C., Wells, G. A., & Waddington, H. (2017). Quasi‐experimental
Business Review, 25, 150–161. study designs series—Paper 5: A checklist for classifying studies
Miles, M. B., & Huberman, A. M. (1994). Qualitative data analysis: An evaluating the effects on health interventions—A taxonomy without
expanded sourcebook (2nd ed.). Sage Publications Inc. labels. Journal of Clinical Epidemiology, 89, 30–42.
Minot, N., & Hill, R. V. (2007). Developing and connecting markets poor Rugumamu, C. P. (2014). Empowering smallholder rice farmers in Tanzania
farmers. 2020 focus brief on the World's Poor and Hungry People. to increase productivity for promoting food security in Eastern and
Washington: International Food Policy Research Institute Southern Africa. Agriculture & Food Security, 3(1), 7.
(IFPRI). Sabet, S. M., & Brown, A. N. (2018). Is impact evaluation still on the rise?
Mishra, A. K., Kumar, A., Joshi, P. K., & D'Souza, A. (2018). Production The new trends in 2010–2015. Journal of Development Effectiveness,
risks, risk preference and contract farming: Impact on food security 10(3), 291–304.
in India. Applied Economic Perspectives and Policy, 40(3), 353–378. Sandieson, R. (2006). Pathfinding in the research forest: The pearl
Mullan, R. J., Flynn, D. N., Carlberg, B., Tleyjeh, I. M., Kamath, C. C., harvesting method for effective information retrieval. Education
LaBella, M. L., Erwin, P. J., Guyatt, G. H., & Montori, V. M. (2009). and Training in Developmental Disabilities, 41(4), 401–409.
Systematic reviewers commonly contact study authors but do so Sterne, J. A. C., & Egger, M. (2005). Regression methods to detect
with limited rigor. Journal of Clinical Epidemiology, 62(2), 138–142. publication and other bias in meta‐analysis. In H. R. Rothstein, A. J.
 18911803, 2024, 2, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/cl2.1411 by Kungliga Tekniska Hogskolan Kth Royal Institute Of Technology, Wiley Online Library on [25/09/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
MARION ET AL. | 191 of 191

Sutton & M. Borenstein (Eds.), Publication bias in meta‐analysis Waddington, H. S., Villar, P. F., & Valentine, J. C. (2023). Can non‐
(pp. 99–110). John Wiley & Sons Ltd. randomised studies of interventions provide unbiased effect
Stigler, G. J. (1961). The economics of information. Journal of Political estimates? A systematic review of internal replication studies.
Economy, 69(3), 213–225. Evaluation Review, 47(3), 563–593.
Suckling, E., Christensen, Z., & Walton, D. (2021). Poverty trends: Global, Waddington, H., White, H., Snilstveit, B., Hombrados, J. G., Vojtkova, M.,
regional and national. Development initiatives: Factsheet. Develop- Davies, P., Bhavsar, A., Eyers, J., Koehlmoos, T. P., Petticrew, M.,
ment Initiatives Poverty Research. Valentine, J. C., & Tugwell, P. (2012). How to do a good systematic
Svensson, J., & Yanagizawa, D. (2009). Getting prices right: The impact of review of effects in international development: A tool kit. Journal of
the market information service in Uganda. Journal of the European Development Effectiveness, 4(3), 359–387.
Economic Association, 7(2/3), 435–445. Wilson, D. B., Weisburd, D., & McClure, D. (2011). Use of DNA testing in
Taylor, J. E., & Adelman, I. (2003). Agricultural household models: Genesis, police investigative work for increasing offender identification,
evolution, and extensions. Review of Economics of the Household, 1, arrest, conviction and case clearance. Campbell Systematic Reviews,
33–58. 7(1), 1–53.
Thomas, J., Graziosi, S., Brunton, J., Ghouze, Z., O'Driscoll, P., Bond, M., & World Bank. (2007). World development report 2008: Agriculture for
Koryakina, A. (2022). EPPI‐Reviewer: Advanced software for system- development. World Bank.
atic reviews, maps and evidence synthesis. EPPI‐Centre, UCL Social World Bank. (2020). Reversals of fortune: Poverty and shared prosperity.
Research Institute, University College London. World Bank.
Thomas, J., McNaught, J., & Ananiadou, S. (2011). Applications of text World Bank. (2022). Agribusiness and value chains. https://www.
mining within systematic reviews. Research Synthesis Methods, 2(1), worldbank.org/en/topic/agribusiness
1–14. World Bank. (2023). Food security | Rising food insecurity in 2023.
Thomas, J., O'Mara‐Eves, A., & Brunton, G. (2014). Using qualitative Agribusiness and value chains. https://www.worldbank.org/en/
comparative analysis (QCA) in systematic reviews of complex topic/agriculture/brief/food-security-update
interventions: A worked example. Systematic Reviews, 3, 67. World Bank Group. (2015). Ending poverty and hunger by 2030: An agenda
Timmer, C. P. (1997). Farmers and markets: The political economy of new for the global food system (2nd ed.). World Bank Group.
paradigms. American Journal of Agricultural Economics, 79(2), 621–627. Yanagizawa‐Drott, D., & Svensson, J. (2012). Estimating Impact in Partial vs.
Ton, G., Desiere, S., Vellema, W., Weituschat, S., & D'Haese, M. (2017). General Equilibrium: A Cautionary Tale from a Natural Experiment in
The effectiveness of contract farming for raising income of Uganda. John F. Kennedy School of Government at Harvard University.
smallholder farmers in low‐ and middle‐income countries: A Zeller, M., Diagne, A., & Mataya, C. (1998). Market access by smallholder
systematic review. Campbell Systematic Reviews, 13(1), 1–131. farmers in Malawi: implications for technology adoption, agricultural
Valentine, J. C., Aloe, A. M., & Lau, T. S. (2015). Life after NHST: How to productivity and crop income. Agricultural Economics, 19(1–2),
describe your data without “p‐ing” everywhere. Basic and Applied 219–229.
Social Psychology, 37(5), 260–273.
Viechtbauer, W. (2010). Conducting meta‐analyses in R with the metafor
package. Journal of Statistical Software, 36, 1–48.
SUPP ORTING INFO RM ATION
Viechtbauer, W., & Cheung, M. W.‐L. (2010). Outlier and influence
diagnostics for meta‐analysis. Research Synthesis Methods, 1(2), Additional supporting information can be found online in the
112–125. Supporting Information section at the end of this article.
Villar, P. F., Kozakiewicz, T., Bachina, V., Young, S., & Shisler, S. (2023).
PROTOCOL: The effects of agricultural output market access
interventions on agricultural, socio‐economic and food and nutrition
security outcomes in low‐ and middle‐income countries: A system-
How to cite this article: Marion, P., Lwamba, E., Floridi, A.,
atic review. Campbell Systematic Reviews, 19(3), e1348. https://doi.
org/10.1002/cl2.1348
Pande, S., Bhattacharyya, M., Young, S., Villar, P. F., & Shisler, S.
Villar, P. F., & Waddington, H. (2019). Within study comparisons and risk (2024). The effects of agricultural output market access
of bias in international development: Systematic review and critical interventions on agricultural, socio‐economic, food security, and
appraisal. Campbell Systematic Reviews, 15(1–2), e1027. nutrition outcomes in low‐ and middle‐income countries: A
Waddington, H., Snilstveit, B., Hombrados, J., Vojtkova, M., Phillips, D.,
systematic review. Campbell Systematic Reviews, 20, e1411.
Davies, P., & White, H. (2014). Farmer field schools for improving
farming practices and farmer outcomes: A systematic review. https://doi.org/10.1002/cl2.1411
Campbell Systematic Reviews, 10(1), i–335.