World Development Perspectives 18 (2020) 100203

Contents lists available at ScienceDirect

World Development Perspectives

journal homepage: www.elsevier.com/locate/wdp

Climate-smart agriculture, household income and asset accumulation among

smallholder farmers in the Nyando basin of Kenya
Maurice Juma Ogadaa, Elizaphan J.O. Raob, , Maren Radenyc, John W. Rechac, Dawit Solomonc
⁎

a
School of Business and Economics, Taita Taveta University, Kenya
b
Policy, Institutions and Livelihoods, International Livestock Research Institute (ILRI), P.O. Box 30709 - 00100, Nairobi, Kenya
c
CGIAR Research Program on Climate Change, Agriculture and Food Security (CCAFS), East Africa, International Livestock Research Institute (ILRI), P.O. Box 30709 -
00100, Nairobi, Kenya

ARTICLE INFO ABSTRACT

Keywords: Climate change, characterized by rising average temperatures and changes in rainfall amounts and patterns,
Climate smart agriculture remains a major threat to rural livelihoods. For smallholder households that heavily depend on agriculture,
Household income adaptation and mitigation measures, including coping strategies are therefore needed to secure household li-
Selection bias velihoods and incomes. In this article, we assess the impacts on livelihood outcomes, of climate smart agri-
Propensity score matching
cultural (CSA) practices that are promoted as adaptative strategies against effects of climate change. Using data
Africa
Kenya
from a survey of households in Nyando basin in Kenya, we combine statistical matching and simultaneous
equation econometric modelling, to evaluate pathways through which CSA practices impact on household asset
accumulation and income. We find that uptake of multiple stress-tolerant crops improves household income by
83%, which in turn improves household asset accumulation. Impact pathway modelling also show that adoption
of improved livestock breeds significantly reduces household income by 76%. For improved livestock, however,
household income does not impact on asset accumulation, possibly because income is invested in form of li-
vestock rather than household assets. These findings show that adoption of multiple stress-tolerant crops im-
proves household assets accumulation via the income pathway. However, given an option, households would
invest increased incomes in livestock assets, confirming that livestock are a form of savings, which can be
liquidated to bridge household income gaps. This is a better resilience measure as compared to investment in
domestic household assets.

1. Introduction option, although on its own is not a solution to climate change, and may
come with such residual costs as reduced productivity (Adger, Huq,
Emission of greenhouse gases has been blamed for the changing Brown, Conway, & Hulme, 2003). However, adaptation enables the
climate (Zenghelis, 2006). The change has been characterized by rise in smallholders to secure livelihoods, and improve incomes and asset base
average temperatures, and changes in rainfall amounts and distribution (Niang et al., 2014).
(Field, 2014; I.P.C.C., 2007)). In Sub-Saharan Africa, for example, the In East Africa, about 80% of the population live in the rural areas
rise in temperature is projected to exceed the global average while and are intricately dependent on rainfed agriculture. Therefore, the
rainfall amount, in many parts, is projected to decline (Niang et al., emerging climatic challenges and their impacts on household wealth,
2014). So far, the changing climate has manifested in increased fre- income streams and the general livelihood sources, coupled with ex-
quency and intensity of extreme weather events such as droughts and isting challenges in smallholder agriculture could be devastating. Local
floods (Field, 2014). Smallholders, who are heavily dependent on rain- communities have previously used diverse coping mechanisms and
fed agriculture, are some of the greatest casualties of these changes adaptation strategies to avoid or minimize risks from the recurrent
(Pachauri, Allen, Barros, Broome, Cramer, Christ, Church, Clarke, Dahe, weather extremes such as droughts and floods in the region. These
& Dasgupta, 2014). Thus, cushioning them through climate change approaches include migration, income diversification and use of ap-
mitigation and adaptation is important (McCarthy, Lipper, & Branca, propriate technology (Babatunde & Qaim, 2010; Burney & Naylor,
2011). For the smallholders, adaptation is apparently the most viable 2012; Karamba, Quiñones, & Winters, 2011). More often, however,

Corresponding author.
⁎

E-mail addresses: [email protected] (M.J. Ogada), [email protected] (E.J.O. Rao), [email protected] (M. Radeny), [email protected] (J.W. Recha),
[email protected] (D. Solomon).

https://doi.org/10.1016/j.wdp.2020.100203
Received 25 June 2019; Received in revised form 22 March 2020; Accepted 16 April 2020
Available online 28 April 2020
2452-2929/ © 2020 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/BY-NC-ND/4.0/).
M.J. Ogada, et al. World Development Perspectives 18 (2020) 100203

households rely on indigenous knowledge and some degree of trial and appropriate agricultural technologies is an important step towards
error to cushion themselves from climate extremes and change. Often, achieving food and nutritional security (Langyintuo & Mungoma, 2008)
they lack complete information on the nature of the emerging climate and general household welfare (Mendola, 2007). For the CSVs, it is
extremes and the associated challenges, and the required climate-smart envisioned that adoption of CSA technologies and innovations such as
technologies to respond effectively. Even when they have the right in- multiple stress-tolerant crops, coupled with good agronomic practices,
formation, scarcity of resources may limit their responses. Conse- would boost production of the food staples even with the changing
quently, their responses may be inadequate, less effective and un- climate and bridge the hunger gap. Surplus output could also be mar-
sustainable. Effective adaptation requires multi-stakeholder approach keted to enhance household income, providing opportunities for further
involving policy makers, extension agents, researchers, communities as investments. The same applies to mitigation interventions such agro-
well as relevant non-state actors. Against this back drop, the CGIAR forestry, composting and other soil and land management options that
Research Program on Climate Change, Agriculture and Food Security build soil carbon with positive consequences for crop yields. Improved
(CCAFS), in partnership with other stakeholders, is working with livestock breeds with faster growth, coupled with improved husbandry
farmers in East Africa to test and promote a portfolio of climate-smart or management practices realized through farm advisory services is
agriculture (CSA) technological and institution options for dealing with expected to improve availability of animal source foods, thus enhancing
climate change in agriculture through the Climate-Smart Villages household food and nutrition security. When the animals and/or their
(CSVs) research for development approach, with the aim of scaling the products are sold, household incomes are also improved. Moreover,
appropriate options and drawing lessons for policy makers across scales household participation in group-based savings and loans schemes en-
(Aggarwal et al., 2018). CSVs are clusters of villages found in climate hances their access to credit, which can be re-invested in agriculture
change hotspots across a wide range of agro-ecological zones with and other income generating ventures with positive implications for
different farmer typologies, climate risks and vulnerabilities allowing household food security, income and asset accumulation.
comparison, learning, extrapolation and climate analogue analysis. Previous studies have shown positive impacts of agricultural tech-
In East Africa, CCAFS started piloting CSVs approach in 2012 with nologies on yields and gross farm returns, often leading to improved
six clusters of villages: Lushoto (Tanzania), Wote and Nyando (Kenya), consumption expenditure and reduced poverty (see Amare, Asfaw, &
Hoima and Rakai (Uganda), and Borana (Ethiopia) (Recha, Kimeli, Shiferaw, 2012; Asfaw, Shiferaw, Simtowe, & Lipper, 2012; Kassie,
Atakos, Radeny, & Mungai, 2017). In these villages, the smallholders Shiferaw, & Muricho, 2011; Kiiza & Pederson, 2012). However, impacts
face increased frequency and intensity of extreme weather events, and vary, with wealthier households sometimes gaining more while the
pests and diseases. Every CSV has a wide range of activities: Weather- poor experience negative impacts (Bourdillon, Hebinck, & Hoddinott,
Smart (seasonal weather forecast, agro-advisory services, and climate 2007; Gabre-Madhin & Haggblade, 2004; Hossain, Lewis, Bose, &
analogues), Water-Smart (rain water harvesting and soil erosion control), Chowdhury, 2003). Wider impacts can also be limited by adoption
Carbon-Smart (agroforestry, trees for fodder and forage, fuel wood and rates, which is determined by farmer or farm-specific factors, access to
fruit trees, and composting manure), Crop-Smart (improved beans, im- information, access to the technology, human capital and institutional
proved Irish and sweet potatoes, multiple stress-tolerant cereals, seed and factors (Amare et al., 2012; Asfaw et al., 2012). Moreover, information
fodder bulking), Livestock-Smart (improved sheep and goats, improved asymmetry may limit proper use of the adopted technology, leading to
poultry, community para-veterinary workers), and Knowledge-Smart reduced impacts of technologies. Given the mixed results from previous
(smart farms, collective action groups and farmer-to-farmer learning). The studies, it is necessary to evaluate the climate-smart agricultural tech-
portfolio of technologies and innovations, however, differ across CSVs nologies that have been promoted under the CSVs program. This will
based on farming systems, agro-ecosystems, livelihoods, climate and inform selection of technologies to scale up and help in the review of
environment-related risk factors. The overarching goal is to stimulate implementation approaches if necessary. Using household data col-
action that would enable communities and households respond to cli- lected from the Nyando basin in Kenya and controlling for possible
mate variability so as to reduce hunger, ensure food security and en- selection bias, this study therefore estimates the impact of the different
hance household incomes (Kinyangi, Recha, Kimeli, & Atakos, 2015). CSA technologies on household income and asset accumulation.
To enhance their capacity to cope with or adapt to climate change, The rest of the paper is organized as follows. In section 2, we discuss
farmers within the CSVs work in groups. These groups have become the data collection process and present some descriptive statistics. We
important although their initial aims might have had nothing to do with then present the evaluation framework and the estimation approaches
agriculture and climate change. Some were formed for the purposes of followed by presentation and discussion of results in section 3. Finally,
pooling financial resources through Rotating Savings and Credit Asso- we make some concluding remarks and draw policy implications.
ciation (ROSCA) schemes, and pooling of farm labor. Others were
mainly for such social reasons as support to orphans, widows and other 2. Data and descriptive statistics
vulnerable groups. The groups provide platforms for innovative part-
nerships for new knowledge and skills, and for building capacity of the 2.1. Sampling strategy and data collection
local farmers to change farming practices and to adopt new crops and
livestock. They also provide avenues for members to pool financial Data for this study comes from a project evaluation survey that was
resources for savings, provide farm labor and enhance efficient delivery conducted to assess the impacts of CSA technologies and practices that
of extension services and high-quality farm inputs at affordable prices. had been introduced in the CSVs in the Nyando basin in Western Kenya.
Research for Development (R4D) is active in the CSVs, led by several The portfolio of CSA technologies in Nyando CSVs include crop, live-
CGIAR and national research centers. stock, integrated soil and water conservation (SWC), and agroforestry.
Adoption of CSA technologies and innovations can improve house- Crop related technologies include drought, pest and disease tolerant,
hold livelihood outcomes, either directly or indirectly (Becerril & and early maturing crop varieties and integrated with improved agro-
Abdulai, 2010; Moyo, Norton, Alwang, Rhinehart, & Deom, 2007). nomic practices, integrated pest management, crop disease surveillance
Direct effects could include crop and livestock productivity gains and and control, crop rotation, use of intercrop innovations and appropriate
reduced cost of production, leading to improved food and nutrition use of fertilizers. Among the new and/or improved crop varieties in
security, and household income and wealth indicators. Indirect effects Nyando are pigeon peas, and sorghum varieties such as KARI Mtama 2,
could include increased supply of food staples, leading to improved KARI Mtama 4, Seredo, and Serena which are fast maturing and tolerant
availability of food through the markets. Further, increased pro- to water logging. There are also improved cassava variety, which is
ductivity may enhance demand for farm labour, resulting into improved resistant to mosaic virus, improved potato varieties which are fast
returns to agriculture wage labour. Overall, adoption of improved and maturing and tolerant to drought, and tissue culture bananas which are

2
M.J. Ogada, et al. World Development Perspectives 18 (2020) 100203

resistant to bacterial wilt. Most households in Nyando are diversifying CSV villages for a cross-sectional survey.
into new crops to increase productivity and reduce chances of complete
crop failure. New crops such as pigeon pea and green grams are being 2.2. Descriptive analyses
cultivated alongside the traditional legumes such as beans and cowpeas.
Pigeon pea has the advantage of withstanding drought as well as wa- Before delving into impact evaluation, we explore the data to reveal
terlogging, while the leaves can be harvested and used as livestock any observable relationship between adaptive and mitigation measures
fodder. The livestock interventions focus on crossbreeding of im- and key variables of interest – domestic asset index and household income.
proved/exotic small ruminants (Red Masai sheep and Galla goats) and Household domestic asset index is calculated for all movable assets with
indigenous breeds (Small East African Sheep and Small East African each type of asset or groups of assets assigned weights which are then
Goats), coupled with training of the farmers on improved management adjusted for age as follows (Bill and Melinda Gates Foundation, 2010):
practices. Crossbreeding combines the high yield and early maturity G N
traits of improved breeds with the hardiness, disease resistance and HHdomesticassetindex = ( gi × a) i = 1, 2, .;g = 1, 2, ,G
adaptability traits of indigenous local breeds. The Galla goats are well g=1 i=1
adapted to drylands, with faster growth and mature about six months
earlier than the indigenous goats. The Red Maasai sheep are popular for where, gi = weight of the ith item of asset g, N = number of asset g
fast growth and maturity, are resistant to internal parasites, and tol- owned by household, a = age adjustment to weight, G = number of
erant to trypanosomes, drought and heat stress, and suitable for meat assets owned by household. We derive this index for domestic, trans-
production. In addition, the cross-breeds attract higher market prices as port and productive assets. Household income was obtained via ex-
compared to the indigenous breeds. Integrated SWC includes water- penditure approach that is known to be less sensitive to measurement
harvesting pans, ploughing across contours, and use of terraces and than the direct income estimation (Deaton, 1997). The expenditure
stone bunds. These measures are aimed at preventing soil erosion and approach approximates total household income from total household
improving soil fertility. Agroforestry integrates trees into crop and li- expenditure on food items, non-food items and contributions ex-
vestock production and across agricultural landscapes. In Nyando, the penditure. The total expenditure in local currency was then converted
focus is on trees which grow faster under water stress and have multiple to the purchasing power parity (PPP) equivalent using PPP rates that
uses. Agroforestry contributes to building ecological resilience through are referenced to 2011 (IBRD. International Bank for Reconstruction
providing carbon sinks thereby removing greenhouse gases from the and Development, 2018).
atmosphere. Other benefits include minimising the risk of soil erosion Descriptive analyses show that adopters1 of multiple stress-tolerant
by wind and surface runoff, removes nutrients from deeper soil layers to crop (DTC) varieties have superior asset index on average, both prior to
surface layers, and provides soil organic matter through decomposition (pre-CSV) and after (post-CSV) the introduction of CSV interventions
of leaf biomass and roots. Farmers can accrue additional benefits by (See Fig. 1). This is also true for all categories of assests (See Fig. 5), and
using leguminous tree species that will fix nitrogen in the soil and across the entire range of distribution as can be seen from Fig. 2.
provide fodder to farm animals. To cope with the rising demand for tree However, the difference in asset index is more pronounced with/after
seedlings farmers have been encouraged and supported to establish tree CSV interventions; cumulative distribution function (CDF) of adopters
nurseries. Institutional innovations include smart farms as adaptation significantly dominates the CDF of non-adopters.
learning hubs, innovation funds through the CBOs and improved access While asset index increased for both adopters and non-adopters
to inputs of high quality at affordable prices through a community input after the CSV interventions, increases were significantly larger (p-
supply shop. value = 0.03) for adopters of drought-tolerant crops (DTC), a possible
CCAFS has been collecting monitoring data among participating indication of the impacts of DTC on asset accumulation. We also notice
households in Nyando. From these data, adoption of CSA technologies a close association between increases in asset index and increases in
and innovations has improved overtime. Uptake of improved climate income. It possible that drought-tolerant crops could be impacting on
resilient crop varieties has been on an upward trajectory, with the asset accumulation via an income pathway. We explore this in the
number of farmers introducing improved crop varieties gradually in- econometric analyses in the next section.
creasing from 2012. Similarly, uptake of more resilient small ruminant We also describe the relationship between uptake of improved small
livestock breeds, particularly Galla goats and Red Maasai Sheep has also ruminant livestock (sheep and goats) on the one hand and changes in
been on the rise. Other changes include increased uptake of agrofor- household income and asset index on the other. Again, as is illustrated
estry, especially fruit trees; diversification into beekeeping by some in Fig. 3, asset indices increased for both adopters and non-adopters.
households; early planting, crop rotation and intercropping; and soil However, we find significantly larger increases (p-value = 0.09) for
and water conservation practices. For example, in 2011, only 32% of adopters of improved livestock leading to a wider gap in asset index
the farm households had adopted improved crop varieties. This had after the CSV interventions. Indeed, Fig. 4 illustrates a more pro-
risen to 93% in the intervention villages by 2016. The uptake of ferti- nounced difference in the distribution of asset indices after the in-
lizer had also risen from about 8% to about 94% in the same period. troduction of CSV interventions. The picture is quite mixed when we
However, non-participating households were completely ignored in this consider individual asset groups as ilustrated in Fig. 6. Adopter of im-
process. This renders the data inappropriate for impact evaluation due proved livestock exhibit slightly lower increase in transport asset index
to lack of a counterfactual. To create a counterfactual, we identified than non-adopters, after introduction of CSV interventions.
villages which were very similar to the CSVs in terms of observable Additionally, while asset index increased for adopters of improved
biophysical (i.e., temperature, precipitation, soil type, landscape posi- livestock (IL) breeds, income per adult equivalent declined, an indica-
tion) and socio-economic (i.e., most prevalent farming system, main tion that households could be converting their incomes into livestock
agricultural crop, livestock ownership and husbandry practices, market assets. Given the possible impact of DTC on household income, it is
behaviors) characteristics. These villages were far enough from the possible that DTC is impacting on livestock via the income pathway.
CSVs to minimize “contamination”. Households from these villages were Again, we explore this argument in the next section via econometric
then listed with the help of the local administration to provide a sam- analysis.
pling frame for the non-treated villages. A random sample of 216
households was then made from this sampling frame. For the CSVs, a 1
Farm households which have consistently used a given climate-smart
list of all participating and non-participating households provided a technology since uptake, irrespective of the proportion of land under the
sampling frame from which a random sample of 217 households was technology. Although the land area covered did not matter in the classification,
drawn. This yielded a sample of 433 households across CSV and non- such households had completely switched to the new practices.

3
M.J. Ogada, et al. World Development Perspectives 18 (2020) 100203

Fig. 1. Average asset index and household income per adult equivalent by Fig. 3. Average asset index and household income per adult equivalent by
adoption of DTC. adoption of IL.

Fig. 4. Cumulative distribution of asset index by adoption of IL both before

(panel a) and after (panel b) CSV interventions.Note: Kolmogorov-Smirnov test
Fig. 2. Cumulative distribution of asset index by adoption of DTC both before statistic of 0.174 (p-value = 0.008) for panel (a) and 0.202 (p-value = 0.001)
(panel a) and after (panel b) CSV interventions.Note: Kolmogorov-Smirnov test for panel (b) show that the distributions are significantly different.
statistic of 0.201 (p-value = 0.000) for panel (a) and 0.250 (p-value = 0.000)
for panel (b) shows that the distributions are significantly different.
With respect to adoption of improved livestock breeds, we also
observe growth in asset index for both adopters and non-adopters
Breaking down assets into categories—domestic, transport and across all asset categories. Similar to the case of DTC, domestic assets
production assets—we also observe differences in growth of asset for recorded the largest growth. However, unlike in the case of DTC, this
different categories of assets. While both adopters and non-adopters of was followed by the growth in production asset. Again, similar to the
DTC experienced growth across all categories of assets, domestic assets case of DTC, domestic and production assets recorded significant
experienced the largest growth followed by transport assets. However, growth.
growth in transport asset over the intervention period was insignificant
(p-value = 0.26).

4
M.J. Ogada, et al. World Development Perspectives 18 (2020) 100203

However, households may self-select into uptake of interventions

and this may bias estimates of treatment effects of CSA interventions. It
is possible that some factors determining uptake of CSA intervention
may also affect household income. If such factors are not included ex-
plicitly in equation (1), especially when such variables are unobserved,
then the indicator for uptake of CSA interventions in equation (1) will
be correlated with the error term ( 1), leading to a biased estimation of .
One way to address this problem is to monitor households who have
applied the CSA technologies and those who are not using these tech-
nologies over time and then apply difference-in-difference (DiD) ana-
lytical techniques to isolate the impact of CSA technologies. However,
this was not possible in this study, since monitoring of project inter-
ventions was limited to the treatment group (in the CSVs).

Fig. 5. Average asset index by asset category and adoption of drought-tolerant 3.2. Matching approach
crops.
To address the potential selection bias, we used two approaches:
Propensity Score Matching (PSM) and selection models. PSM assumes
that conditioning on observable variables eliminates sample selection
bias (Heckman & Navarro-Lozano, 2004). Matching essentially creates
an experimental condition in which uptake of CSA interventions is
randomly assigned, allowing for identification of causal link between
respective interventions and livelihood outcomes. Instead of directly
comparing outcome and impact variables between households who
have applied CSA practices and those applying conventional (non-CSA)
practices, PSM restricts comparison to households that are similar in
terms of observable characteristics and therefore reduces the bias that
would otherwise occur if the two groups were systematically different
(Dehejia & Wahba, 2002).
PSM is a two-stage process: first, we use the entire sample to esti-
mate a probit model that generates propensity scoresP (z ) —estimates of
the probability that a household with a vector of characteristicsz , will
apply CSA practices on their farms. The vector z are assumed to be
Fig. 6. Average asset index by asset category and adoption of improved live-
those observable variables that determine whether a household applies
stock breeds.
CSA technologies. In this estimation, households with similar ob-
servable characteristics are likely to have similar propensity scoresP (z ) ,
3. Impacts of climate smart agriculture on household income and even if some of them may not have implemented CSA interventions.
assets Using similarity in propensity scores, we can, therefore, construct
comparable groups - groups of households with similar propensity
The descriptive analysis in the previous section show positive cor- scoresP (z ) but where one group has applied CSA practices while the
relation between asset index and adoption of both drought tolerant other group of households has not implemented CSA interventions.
crops and improved livestock breeds. We also see positive correlation In the second stage, we calculate the average outcome for the two
between adoption of DTC and household income while there is a ne- groups and then estimate impacts of CSA interventions as the difference
gative correlation between adoption of improved livestock and house- in average outcomes between these groups. This difference is known as
hold income. However, we cannot attribute these changes to adoption the PSM estimator of average treatment effect on the treated house-
of the respective CSA technologies. We therefore undertake statistical holds (ATT), which is expressed as follows:
and econometric modelling that can help us establish causality.
PSM
ATT = EP (z | I = 1) [E {R1 |I = 1, P (z )} E {R 0 |I = 0, P (z )}] (2)
3.1. Evaluation framework
Where R1and R 0 are outcomes for households that have applied CSA
practices and the comparison group of households, respectively;
Program evaluation often follows approaches suggested by Maddala
I = 1indicates that households practice CSA and I = 0 refer to compar-
(1983):
ison group of households that do not practice CSA. The PSM procedure
y = X1 + I + 1 (1) therefore allows us to compare outcomes for comparable groups (apples
Where y can be considered as one of the livelihood outcomes such as with apples rather than apples with oranges) since the groups are similar in
food security, household income or wealth (asset ownership); X1is a observable characteristics and hence in propensity scores.
vector of farm, household and contextual characteristics that could
influence livelihood outcomes and; I is a dummy indicating whether or 3.2.1. Estimation results
not a household uses any one of the CSA interventions. We hypothesize The matching process begins with the estimation of propensity
that these CSA interventions could influence livelihood outcomes such scores, P(z). We avoid the use of potentially endogenous variables in
as household income and assets. Improved yields engendered by ap- the probit model, as this could cause problems in result interpretation
propriate crop varieties can possibly lead to surplus production for sale (Caliendo & Kopeinig, 2008). The probit results are shown in Tables A1
with positive impacts on household income. Some of this income can and A2 in the Appendix. Farmer age, farming experience, residence in
then be invested by household in accumulating assets. Holding other CSVs, provision of weather forecast information to farmers and social
factors constant, therefore, the coefficient ( ) captures partial effects of capital (group membership) influence adoption of DTC. Conversely,
household uptake of CSA interventions. adoption of improved livestock is significantly influenced by gender of

5
M.J. Ogada, et al. World Development Perspectives 18 (2020) 100203

Table 1
Average treatment effects and results of sensitivity analyses.
Matching algorithm Welfare Outcome Non-adopters Adopters

Average values for matched samples of adopters and non-adopters of DTC

Nearest neighbour Domestic asset index 23.29 39.92
Household income/AE 888.74 2,000.26
Kernel Domestic asset index 25.92 39.92
Household income/AE 1,293.18 2,000.26
Average values for matched samples of adopters and non-adopters of improved livestock
Nearest neighbour Domestic asset index 24.20 36.50
Household income/AE 3,396.29 1,284.39
Kernel Domestic asset index 29.55 36.50
Household income/AE 1,692.36 1,384.39
Matching algorithm Welfare Outcome ATT T-statistic Critical level of hidden bias ( ) Treated (n) Control (n)
Treatment effect of multiple stress-tolerant crops
Nearest neighbour Domestic asset index 16.63 3.10*** 1.05–1.10 202 214
Household income/AE 1,110.5 2.48*** 1.70–1.75 201 214
Kernel Domestic asset index 13.99 2.82*** 1.05–1.10 202 214
Household income/AE 707.1 1.73** – 201 214
Treatment effect of improved livestock breeds
Nearest neighbour Domestic asset index 12.31 2.34*** 1.60–1.65 131 296
Household income/AE −2,011.9 −1.70* – 131 295
Kernel Domestic asset index 6.96 1.32 – 131 296
Household income/AE −308.0 −0.65 – 131 295

Note: ∗, ∗∗, and ∗∗∗ Significant at the 10%, 5%, and 1% levels, respectively. The t-values for the ATTs are based on bootstrapped standard errors with 500
replications.

household head, ethnicity of households, residence in CS village, off- 3.2.2. Validity of the matching assumption
farm livelihood activities and proximity to market centers. Despite the general ability of PSM to control for selection bias, the
There are various matching techniques; the most common ones in- estimates are only valid subject to two conditions: (i) balancing in
clude nearest neighbor matching (NNM), kernel-based matching covariates is achieved, and (ii) there is no systematic household het-
(KBM), stratified radius matching, and Mahalanobis matching erogeneity due to unobservables (Caliendo & Kopeinig, 2008; Dehejia &
(Caliendo & Kopeinig, 2008). In this study, we apply the KBM and the Wahba, 2002). The objective of estimating the propensity scores is to
NNM methods. NNM involves pairing households that use CSA tech- balance the distribution of variables relevant to the matching process.
nologies with their non-user counterparts that are closest in terms of P Balancing tests are therefore necessary after matching to determine if
(z) as matching partners. KBM, on the other hand, uses a weighted the matching process has reduced the bias by eliminating differences in
average of the outcome variable for all individuals in the control group covariates. We evaluate the balancing condition and bias reduction
(non-users of CSA technologies) to construct a counterfactual outcome. following Rosenbaum and Rubin (1983). Results of these tests are
Observations that provide better matches are given more weight. The shown in Table 2. Results of balancing test reveal substantial reduction
weighted average is compared to the outcome for users of CSA tech- in bias due to balancing achieved via statistical matching (see column 5
nologies, and the difference provides an estimate of the treatment effect of Table 2), underlining the fact that systematic difference that are due
of CSA technologies. A sample average over all households that have to observable factors are properly eliminated.
adopted CSA technologies then provides an estimate of ATT. To test for potential hidden bias due to unobservable factors, we use
We apply NNM and KBM methods and impose the common support Rosenbaum bounds test (Rosenbaum, 2002). Assuming two individuals
condition to ensure proper matching. The matching procedure was have the same observed covariates z (as implied by the matching proce-
conducted with STATA 15 software, following steps described by dure), the two matched observations would differ in their odds of up-
Leuven and Sianesi (2003). The upper panel of Table 1 shows the take of CSA technologies and practices only by the difference in un-
average values of asset index and household income for matched samples observed covariates, which is measured by the parameter . The test
of adopter and non-adopters of DTC and improved livestock. The lower procedure involves changing the level of and deriving the bounds on
panel of the same Table shows treatment effects for the treated (ATT) the significance levels of the ATT under the assumption of endogenous
for adopters of drought tolerant crops and improved livestock. Both self-selection into adoption. This allows for identification of the critical
matching methods reveal significant impacts of DTC adoption on asset levels of at which the estimated ATT would become insignificant.
index and household income. Compared to average asset index of 23.3 Results of this test are shown in column 5 of Table 1. Using the example
for the controls/non-adopters of drought-tolerant crops, the ATT of of asset index, the critical values for hidden bias ( ) are 1.05–1.10 for
16.63 dislayed in Table 1 represents an increase of 71% in asset index both KBM and NNM. The lowest value of = 1.05 implies that if
due to adoption of drought-tolerant crops. Using the kernel matching households that have the same z-vector would differ in their odds of
procedure, and applying the same principle, adoption of drought-tol- adopting DTC by only a factor of 1.05 (5%), the ATT for domestic asset
erant crops leads to 54% increase in asset index. We therefore conclude index would be rendered insignificant. This implies that any small
that adoption of DTC increases asset index by 54–71%. Impacts on systematic differences between adopters and non-adopters due to un-
household income are even bigger. Compared to average household observed factors would invalidate our conclusion. While the critical
income per adult equivalent for non-adopters (USD 1,293.2), adoption values for ATT of household income are higher (1.70–1.75), they are
of DTC increases household income by 55–124%. As for adoption of only valid for NNM method. These findings show that odds of adoption
improved livestock, only the NNM method reveal significant impacts are likely to be influenced by unobserved factors. We address this
with effects on income being negative and only significant at 10%. limitation via econometric approach in the next section.
Compared to household income for non-adopters (USD 1,692.35),
adoption of improved livestock leads to reduction in income by
18–60%.

6
M.J. Ogada, et al. World Development Perspectives 18 (2020) 100203

3.3. Analysis of impact pathways

p-value of LR (matched)

Results in the previous section showed that adoption of CSA tech-

nologies (DTC and improved livestock breeds) have varied effects on
household income and household asset index. However, the pathways
through which these impacts are realized is not revealed in these re-
0.635
0.317
1.000
1.000

0.825
0.355
0.999
1.000
sults. We hypothesize that adoption of multiple stress-tolerant crops
would boost production of the staple food crops even with the changing
p-value of LR (unmatched)

climate, resulting not only in household food security, but yielding

surplus that could be marketed to enhance household income. We also
hypothesize that the resultant income increases will enhance household
asset accumulation. As for adoption of improved livestock breeds, we
hypothesize that sale of such livestock could boost household income
that again can be invested in household asset. However, it is also pos-
0.000
0.000
0.000
0.000

0.000
0.000
0.000
0.000

sible that investment in improved livestock draws on household earn-

ings, thus implying negative consequences for household income. We
Pseudo R (matched)

explore these pathways econometrically in this section.

3.3.1. Estimation strategy

To test the stated hypotheses, we specify a simultaneous equation
0.03
0.05
0.01
0.01

0.04
0.06
0.02
0.02

model as follows:
(2)
Pseudo R (unmatched)

A= 0 + 1Y + 2 X2 + 2

Y= 0 + 1U + 2 X3 + 3 (3)

U= 0 + 1Y + 2 X4 + 4 (4)
0.17
0.18
0.17
0.18

0.21
0.21
0.21
0.21

Where A represents household asset index, Y represents household

income and U represents uptake of CSA technologies – multiple stress-
% bias reduction

tolerant crop (DTC) and improved livestock (IL). Variables X2 X 4 are

additional covariates affecting the respective equations. The system of
simultaneous equation comprises dependent variables with mixed
specifications – some are binary while some are continuous. As dis-
77.2
41.0
89.2
83.7

72.6
62.9
80.6
82.8

cussed previously, it is possible that farmers could self-select into

adoption of CSA technologies, thus making equation (4) endogenous.
Equation 4 therefore needs to be modelled with appropriate instru-
Median absolute bias (after

ments to address the potential selection bias. The simultaneous equa-

tion is estimated via mixed process maximum likelihood procedure,
also known as conditional mixed process (cmp) (Roodman, 2011). We
apply this method and estimate separate models for impact pathways
for adoption of drought tolerant crop and improved livestock. For each
matching)

of these estimations, we use the matched sample of adopters and non-

adopters emerging from the propensity score matching undertaken in
5.7
8.0
2.7
2.2

5.8
7.8
4.9
4.3

section 3.
Median absolute bias (before

3.3.2. Results and discussion

Results of this estimation are shown in Tables 3 and 4. We find that
household income has a positive and significant effect on asset index
Indicators of covariate balancing before and after matching.

while household income is also positively and significantly influenced

by adoption of drought tolerant crops. A 1% change in household in-
come results into an increase in asset index by 0.77 units. On the other
matching)

hand, adoption of drought tolerant crops increases household income

25.0
13.5
25.0
13.5

21.2
21.0
25.3
25.0

by 83%. This implies that crop production constitutes a major source of

income for households in the Nyando basin and hence use of crops that
Domestic asset index

Domestic asset index

Domestic asset index

Domestic asset index

are adapted to climate risks enhances the resilience of these households.

Household income

Household income

Household income

Household income

These results are largely consistent with the PSM results that
showed positive and significant impacts of DTC on household income
Outcome

and asset index. Indeed, the matching technique showed that adoption
of DTC had an impact of 55–124% (an average of 89.5%) on household
Multiple stress-tolerant crops

income depending on the matching method used. Similarities in find-

Matching algorithm

ings based on the two different methods is probably indicative of lim-

Improved livestock
Nearest neighbor

Nearest neighbor

ited selection on unobservables.

With regards to livestock, adoption of improved breeds reduces
household income by 76%. This can be interpreted as household in-
Table 2

Kernel

Kernel

come that is saved in the form of livestock. It is possible that households

that adopt improved livestock view these as assets and therefore tend to

7
M.J. Ogada, et al. World Development Perspectives 18 (2020) 100203

Table 3
Impact pathways for adoption of multiple stress- tolerant crops.
Independent variables Asset index Household income DTC adoption
Log income per adult equivalent 76.588*** (21.033)
Age of household head −0.387 (0.299) 0.001 (0.004)
Educational status a
Primary education −21.141 (15.443) 0.206 (0.175) −0.265 (0.230)
Secondary education −31.012* (18.085) 0.314 (0.194) −0.404 (0.257)
Tertiary/college/university −16.408 (22.453) 0.287 (0.254) −0.131 (0.337)
Occupation of household head b
Farm wage employment −26.423 (27.06) 0.400 (0.300)
Non-farm employment −20.843 (13.505) 0.175 (0.152)
Micro-enterprise −26.709* (14.28) 0.323** (0.142)
Other employment −4.766 (22.315) 0.051 (0.261)
Gender of household head (dummy) 8.908 (10.249) −0.034 (0.121) 0.022 (0.157)
Value of credit accessed −4.56E-05 (7.47E-05) 7.93E-07 (8.23E-07) 4.69E-06*** (1.78E-06)
Farm wage income −6.81E-07 (1.01E-06)
Micro-enterprise income 1.13E-07 (1.10E-07)
Formal employment income −2.33E-07 (1.69E-07)
Land rent income −3.28E-09 (9.14E-07)
Remittance income 4.67E-06** (1.93E-06)
Household size 5.068** (2.033) −0.052** (0.021)
DTC adoption 0.830*** (0.206)
Total land owned (acres) −0.134 (1.346) 0.005 (0.016) 0.008 (0.023)
Residence in CS village −7.873 (8.784) −0.095 (0.114) 0.730*** (0.179)
Group membership 20.572** (9.67) −0.111 (0.111)
Lagged domestic asset index 1.142*** (0.155) 0.028*** (0.007)
Farming experience 0.007 (0.007)
Household received forecast:
On extreme weather occurrence 0.251 (0.157)
On onset of rains −0.115 (0.161)
Three months in advance 0.199* (0.110)
Kalenjin ethnic group c −0.507*** (0.191)
Lagged group membership −0.034 (0.127)
Distance to closest river 0.012 (0.037)
Distance to closest road 0.008 (0.045)
Distance to closest market −0.029 (0.044)
Livestock rearing experience −0.009 (0.007)
Constant −770.018*** (217.03) 10.118*** (0.304) 0.129 (0.417)
Number of observations 416

Note: *, **, *** Significant at the 10%, 5%, and 1% level, respectively
a
Base level of education is no education
b
Reference occupation is farming (crop/livestock)
c
Reference ethnic group is Luo

prioritize these over other household assets. This could possibly explain such calamities (Mogues, 2011; Ngigi & Birner, 2013). These effects are
the positive but non-significant impact of household income on expected to be larger for poorer households.
household assets in this model. This is also a fundamental departure
from the multiple stress-tolerant crop model where adoption of DTC 4. Conclusion
significantly improved household income, subsequently leading to
household asset accumulation. Since a large proportion of this savings This paper evaluated the impacts of climate change adaptation and
goes towards acquisition of improved livestock little is left for invest- mitigation strategies, piloted in East Africa, on household income and
ment in accumulation of other household assets. Indeed, with the in- asset accumulation. Using household data from the Nyando basin of
vestment of household income in improved livestock, a 1% increase in Western Kenya and controlling for possible endogeneity, the paper
household income only improves household asset index by 0.17 units. shows that CSA technologies and practices have diverse consequences
Similar to impact pathways for adoption of DTC, results of the im- for household income and asset accumulation.
pact pathways analysis for adoption of improved livestock are con- First, adoption of DTCs significantly enhances household income
sistent with the PSM results. Adoption of improved livestock does re- which, in turn, improves household asset accumulation. Thus, such
duce household income. The magnitude of effect is, however, higher in crop interventions would have even greater impact if input and output
the impact pathway model; 76% as compared to 18–60% (an average of markets are made more efficient through appropriate policy instru-
39%), signaling possible existence of selection on observables under the ments. Second, adoption of improved small ruminant livestock breeds
matching model. These findings are consistent with Mogues (2011), and has a negative impact on household income, indicating that households
confirm that rural households tend to store their wealth in form of li- draw on their incomes to invest in livestock assets. These findings are
vestock, especially in the absence of or malfunction of credit or in- consistent with Pica-Ciamarra, Tasciotti, Otte, and Zezza (2015), who
surance markets. Livestock, especially small ruminants can easily be show that majority of households in less developed countries keep li-
sold off to smooth income and consumption fluctuations occasioned by vestock for various reasons. For example, the small animals are often
shocks. Indeed, livestock and especially small livestock are usually the kept as an insurance against emergencies or a form of investment,
first to be sold off during weather shocks, reducing their stock during especially because they pose less financial risk due to minimal initial

8
M.J. Ogada, et al. World Development Perspectives 18 (2020) 100203

Table 4
Impact pathways for adoption of improved livestock breeds.
Independent variables Asset index Household income IL adoption
Log income per adult equivalent 16.738 (10.85)
Age of household head −0.052 (0.125) 0.002 (0.004) 0.005 (0.007)
Educational status a
Primary education −8.323 (6.716) 0.252 (0.181) 0.296 (0.269)
Secondary education −6.218 (8.095) 0.333* (0.202) 0.231 (0.306)
Tertiary/college/university 5.689 (10.256) 0.489* (0.261) 0.331 (0.384)
Occupation of household head b
Farm wage employment 10.41 (11.373) 0.304 (0.309)
Non-farm employment −4.205 (5.513) 0.114 (0.155)
Micro-enterprise 0.983 (6.652) 0.387*** (0.150)
Other employment 10.037 (9.274) 0.043 (0.271)
Gender of household head (dummy) 2.005 (4.193) 0.071 (0.128) 0.346* (0.180)
Value of credit accessed −6.60E-05 (7.73E-05) 8.60e-07* (8.24E-07)
Farm wage income −7.55E-07 (9.44E-07)
Micro-enterprise income 1.19E-07 (1.09E-07)
Formal employment income −2.96E-07 (1.73E-07)
Land rent income 7.13E-08 (9.08E-07)
Remittance income 4.58E-06*** (1.95E-06)
Household size 2.815*** (0.870) −0.045** (0.021)
Adoption of improved livestock −0.760*** (0.244) −
Total land owned (acres) 0.086 (0.551) 0.009 (0.017) −0.006 (0.028)
Residence in CS village −3.289 (3.716) 0.385*** (0.128) 1.274*** (0.204)
Group membership 11.088*** (3.940) −0.039 (0.111)
Lagged domestic asset index 1.288*** (0.093) 0.001 (0.004)
Lagged group membership 0.320** (0.157)
Livestock rearing experience -0.007 (0.009)
Kalenjin ethnic group c 0.486** (0.246)
Drought perceived as cause of crop failure −0.921* (0.496)
Distance to closest road 0.136** (0.058)
Distance to closest market −0.120*** (0.046)
Household received forecast:
On extreme weather occurrence 0.251 (0.157)
On onset of rains −0.115 (0.161)
Farming experience 0.013 (0.010)
Lagged value of TLU 0.027 (0.018)
Constant −166.019 (110.531) 10.123*** (0.316) −1.651** (0.774)
Number of observations 427

Note: *, **, *** Significant at the 10%, 5%, and 1% level, respectively
a
Base level of education is no education
b
Reference occupation is farming (crop/livestock)
c
Reference ethnic group is Luo

investment (McDermott, Staal, Freeman, Herrero, & Van de Steeg, original draft, Writing - review & editing. Maren Radeny:
2010; Peacock & Hastings, 2011). Conceptualization, Writing - original draft, Writing - review & editing,
Overall, these findings are consistent with previous studies (e.g. Project administration, Funding acquisition. John W. Recha: Project
Hellin and Fisher, 2018), which demonstrate that CSA technologies and administration, Writing - review & editing. Dawit Solomon:
practices are viable pathways out of poverty, especially when targeted Supervision, Writing - review & editing.
properly. A key question is how to scale these CSA technologies and
practices. While collaboration among stakeholders is a clear pathway
through which adoption may be accelerated and sustained, there is Declaration of Competing Interest
need to stimulate demand for the increased output through food and
nutrition policies which integrate the products into food systems so as The authors declare that they have no known competing financial
to drive commercialization. For improved livestock breeds, well de- interests or personal relationships that could have appeared to influ-
signed, accessible and affordable breeding programs is critical for up- ence the work reported in this paper.
take while linkages to output markets are necessary for commerciali-
zation and sustainability. Complementary services such as climate
information, agro-advisories and financing also need to be strength- Acknowledgements
ened.
This work was implemented as part of the CGIAR Research Program
CRediT authorship contribution statement on Climate Change, Agriculture and Food Security (CCAFS), which is
carried out with support from CGIAR Trust Fund Donors and through
Maurice Juma Ogada: Conceptualization, Investigation, Data bilateral funding agreements. For details please visit https://ccafs.c-
curation, Writing - original draft, Writing - review & editing. Elizaphan giar.org/donors. The views expressed in this document cannot be taken
J.O. Rao: Conceptualization, Methodology, Formal analysis, Writing - to reflect the official opinions of these organizations.

9
M.J. Ogada, et al. World Development Perspectives 18 (2020) 100203

Appendix

Table A1
Probit model for adoption of drought-tolerant-crops (DTC).
Independent variables Coefficient SE

Residence in CS village 1.658*** −0.456

Age of household head −0.013* −0.007
Farming experience 0.013* −0.007
Lagged group membership 0.322* −0.165
Household received forecast:
On extreme weather occurrence 0.428** −0.187
Total land owned (acres) 0.033 −0.026
Educational status a
Primary education −0.098 −0.25
Secondary education −0.167 −0.281
Tertiary/college/university 0.025 −0.365
Lagged domestic asset index 0.012 −0.008
Drought perceived as cause of crop failure −1.200** −0.512
Gender of household head (dummy) 0.194 −0.17
Occupation of household head b
Farm wage employment −0.521 −0.445
Non-farm employment −0.136 −0.218
Micro-enterprise 0.06 −0.203
Other employment −0.351 −0.368
Distance to health facility 0.057 −0.086
Constant −0.249 −0.727
Observations 432

Note: *, **, *** Significant at the 10%, 5%, and 1% level, respectively
a
Base level of education is no education
b
Reference occupation is farming (crop/livestock)

Table A2
Probit model for adoption of improved livestock.
Independent variables Coefficient SE

Residence in CS village 1.340*** −0.254

Age of household head 0.006 −0.008
Gender of household head (dummy) 0.317* −0.185
Lagged group membership 0.256 −0.166
Livestock rearing experience −0.012 −0.009
Kalenjin ethnic group a 0.707*** −0.199
Educational status b
Primary education 0.289 −0.273
Secondary education 0.338 −0.308
Tertiary/college/university 0.416 −0.386
Crop farming is household primary livelihood −0.08 −0.054
Off-farm activities is household primary livelihood −0.132*** −0.042
Drought perceived as cause of crop failure −0.782* −0.429
Distance to health facility 0.059 −0.099
Distance to closest market −0.135** −0.057
Farming experience 0.015 −0.011
Distance to closest river 0.059 −0.052
Household received forecast:
On likely start of rains −0.25 −0.196
Household has a child 0.008 −0.191
Total land owned (acres) 0.008 −0.027
Constant −1.496* −0.860
Observations 432

Note: *, **, *** Significant at the 10%, 5%, and 1% level, respectively:
a
Reference ethnic group is Luo
b
Base level of education is no education;

10
M.J. Ogada, et al. World Development Perspectives 18 (2020) 100203

References Agricultural Seed Technologies: Insights from Uganda. Telecommunications Policy,

36(4), 253–259.
Kinyangi, J., Recha, J., Kimeli, P., & Atakos, V. (2015). Climate-Smart Villages and the Hope
Adger, W. N., Huq, S., Brown, K., Conway, D., & Hulme, M. (2003). Adaptation to Climate of Food Secure Households. Agriculture and Food Security (CCAFS): Cgiar Research
Change in the Developing World. Progress in development studies, 3(3), 179–195. Program on Climate Change.
Aggarwal, P. K., Jarvis, A., Campbell, B. M., Zougmoré, R., Khatri-Chhetri, A., Vermeulen, Langyintuo, A. S., & Mungoma, C. (2008). The Effect of Household Wealth on the
S., ... Bonilla Findji, O. (2018). The Climate-Smart Village Approach: Framework of Adoption of Improved Maize Varieties in Zambia. Food policy, 33(6), 550–559.
an Integrative Strategy for Scaling up Adaptation Options in Agriculture. Ecology and Leuven, E., & Sianesi, B. (2003). Psmatch2. STATA module to perform full Mahalanobis and
Society, 23(1, 14). propensity score matching, common support graphing, and covariate imbalance testing,
Amare, M., Asfaw, S., & Shiferaw, B. (2012). Welfare Impacts of Maize-Pigeonpea version, 1(3).
Intensification in Tanzania. Agricultural Economics, 43(1), 27–43. Maddala, G. S. (1983). Limited-Dependent and Qualitative Variables in Econometrics.
Asfaw, S., Shiferaw, B., Simtowe, F., & Lipper, L. (2012). Impact of Modern Agricultural Cambridge: Cambridge University Press.
Technologies on Smallholder Welfare: Evidence from Tanzania and Ethiopia. Food McCarthy, Nancy, Lipper, Leslie, & Branca, Giacomo (2011). Climate Smart Agriculture:
policy, 37(3), 283–295. Smallholder Adoption and Implications for Climate Change Adaptation and
Babatunde, R. O., & Qaim, M. (2010). Impact of Off-Farm Income on Food Security and Mitigation. Mitigation of Climate Change in Agriculture Series, 4.
Nutrition in Nigeria. Food policy, 35(4), 303–311. McDermott, J. J., Staal, S. J., Freeman, H., Herrero, M., & Van de Steeg, J. (2010).
Becerril, J., & Abdulai, A. (2010). The Impact of Improved Maize Varieties on Poverty in Sustaining Intensification of Smallholder Livestock Systems in the Tropics. Livestock
Mexico: A Propensity Score-Matching Approach. World development, 38(7), science, 130(1–3), 95–109.
1024–1035. Mendola, M. (2007). Agricultural Technology Adoption and Poverty Reduction: A
Bill and Melinda Gates Foundation (2010). Agricultural Development Outcome Indicators: Propensity-Score Matching Analysis for Rural Bangladesh. Food policy, 32(3),
Initiative and Sub-Initiative Progress Indicators & Pyramid of. Outcome Indicators, 372–393.
BMGF. Mogues, T. (2011). Shocks and Asset Dynamics in Ethiopia. Economic Development and
Bourdillon, M., Hebinck, P., & Hoddinott, J. (2007). Assessing the Impact of Hyv Maize in Cultural Change, 60(1), 91–120.
Resettlement Areas. of Zimbabwe. Moyo, S., Norton, G. W., Alwang, J., Rhinehart, I., & Deom, C. M. (2007). Peanut
Burney, J. A., & Naylor, R. L. (2012). Smallholder Irrigation as a Poverty Alleviation Tool Research and Poverty Reduction: Impacts of Variety Improvement to Control Peanut
in Sub-Saharan Africa. World Development, 40(1), 110–123. Viruses in Uganda. American journal of agricultural economics, 89(2), 448–460.
Caliendo, M., & Kopeinig, S. (2008). Some Practical Guidance for the Implementation of Ngigi, M. W., Birner, R., 2013. Shocks, Livestock Assets and Climate Change Adaptation
Propensity Score Matching. Journal of economic surveys, 22(1), 31–72. in Kenya.
Deaton, A. (1997). The Analysis of Household Surveys: A Microeconometric Approach to Niang, I., Ruppel, O. C., Abdrabo, M. A., Essel, A., Lennard, C., Padgham, J. 2014. Africa.
Development Policy. World Bank Publications. In Climate Change 2014: Impacts, Adaptation, and Vulnerability. Part B: Regional
Dehejia, R. H., & Wahba, S. (2002). Propensity Score-Matching Methods for Aspects. Contribution of Working Group Ii to the Fifth Assessment Report of the
Nonexperimental Causal Studies. Review of Economics and statistics, 84(1), 151–161. Intergovernmental Panel on Climate Change., V. R. Barroset al (eds). Cambridge
Field, C. B. (2014). Climate Change 2014–Impacts, Adaptation and Vulnerability: Regional University Press: Cambridge, UK.
Aspects. Cambridge University Press. Pachauri, R. K., Allen, M. R., Barros, V. R., Broome, J., Cramer, W., Christ, R., Church, J.
Gabre-Madhin, E. Z., & Haggblade, S. (2004). Successes in African Agriculture: Results of A., Clarke, L., Dahe, Q., Dasgupta, P., 2014. Climate Change 2014: Synthesis Report.
an Expert Survey. World development, 32(5), 745–766. Contribution of Working Groups I, Ii and Iii to the Fifth Assessment Report of the
Heckman, J., & Navarro-Lozano, S. (2004). Using Matching, Instrumental Variables, and Intergovernmental Panel on Climate Change: IPCC.
Control Functions to Estimate Economic Choice Models. Review of Economics and Peacock, C., & Hastings, T. (2011). Meru Dairy Goat and Animal Healthcare Project.
statistics, 86(1), 30–57. International journal of agricultural sustainability, 9(1), 203–211.
Hellin, J., & Fisher, E. (2018). Building Pathways out of Poverty through Climate Smart Pica-Ciamarra, U., Tasciotti, L., Otte, J., & Zezza, A. (2015). Livestock in the Household
Agriculture and Effective Targeting. Development in Practice, 28(7), 974–979. Economy: Cross-Country Evidence from Microeconomic Data. Development policy re-
Hossain, M., Lewis, D., Bose, M. L., & Chowdhury, A. (2003). Rice Research, Technological view, 33(1), 61–81.
Progress, and Impacts on the Poor: The Bangladesh Case (Summary Report): Intl Food. Recha, J. W., Kimeli, P., Atakos, V., Radeny, M. A., & Mungai, C. (2017). Stories of
Policy Res Inst. Success: Climate-Smart Villages in East Africa, CGIAR Research Program on Climate
I.P.C.C., 2007. Climate Change 2007: The Physical Science Basis: Summary for Change, Agriculture and Food Security (CCAFS). Available online at www.ccafs.c-
Policymakers. Geneva: IPCC. giar.org.
IBRD. International Bank for Reconstruction and Development, 2018. Global Purchasing Roodman, D. (2011). Fitting Fully Observed Recursive Mixed-Process Models with Cmp.
Power Parities and Real Expenditure: 2011 International Comparison Program. The The Stata Journal, 11(2), 159–206.
World Bank. Rosenbaum, P. R., 2002. Observational Studies: Springer.
Karamba, W. R., Quiñones, E. J., & Winters, P. (2011). Migration and Food Consumption Rosenbaum, P. R., & Rubin, D. B. (1983). The Central Role of the Propensity Score in
Patterns in Ghana. Food policy, 36(1), 41–53. Observational Studies for Causal Effects. Biometrika, 70(1), 41–55.
Kassie, M., Shiferaw, B., & Muricho, G. (2011). Agricultural Technology, Crop Income, Zenghelis, D. (2006). Stern Review: The Economics of Climate Change. London, England: HM
and Poverty Alleviation in Uganda. World Development, 39(10), 1784–1795. Treasury.
Kiiza, B., & Pederson, G. (2012). Ict-Based Market Information and Adoption of

11