Environ Syst Decis

DOI 10.1007/s10669-017-9652-7

Food security in Africa: a cross-scale, empirical investigation

using structural equation modeling
Riva C. H. Denny1 • Sandra T. Marquart-Pyatt1 • Arika Ligmann-Zielinska2 •
Laura Schmitt Olabisi3 • Louie Rivers4 • Jing Du5 • Lenis Saweda O. Liverpool-Tasie6

Ó Springer Science+Business Media, LLC 2017

Abstract Despite consistent gains in global agricultural the critical importance of availability and accessibility for
productivity in the last 50 years, lack of food security mitigating food insecurity.
persists in many regions of the world. Addressing this issue
is especially pertinent in Africa where 39 of the nearly five Keywords Food security
Undernutrition
Africa

dozen nations most at risk of food insecurity are located. Structural equation modeling
We draw from interdisciplinary research to develop an
empirical model that outlines the four interconnected
aspects of food security—availability, access, utilization 1 Introduction
and stability. Given the complexity of this issue, we
develop a model that considers agricultural, socio-political, Persistent food insecurity is a challenge for the global
and economic factors as drivers of food security and its community. In the face of global wealth gains, economic
manifestations, related in a complex system of relations growth, and increased food production since at least the
that includes both direct and indirect paths. We use struc- 1950s, populations remain vulnerable to shifts in food
tural equation modeling with latent variables to specify a availability and access. Food insecurity is the norm rather
model that seeks to determine the primary drivers of food than the exception in the developing world and is a par-
security over 55 years in Africa, West Africa as a region, ticularly critical challenge in Africa where recent work
and for a group of 5 West African countries: Burkina Faso, estimates 1 in 4 people go hungry (UNDP 2017). Since the
Ghana, Mali, Niger, and Nigeria. Empirical results reveal factors shaping food security are complicated and interre-
lated, encompassing social, economic, political, and envi-
ronmental issues from poverty and inequality to health,
temperature, and rainfall, developing an empirical model to
& Riva C. H. Denny test these proposed paths requires careful and rigorous
[email protected]
specification, assessment, and validation. In this study, we
1
Department of Sociology, Michigan State University, build a model of food security that specifies multiple paths
East Lansing, MI, USA from social, economic, political, and agricultural factors
2
Department of Geography, Environment, and Spatial and tests these proposed direct and indirect relations for
Sciences, Michigan State University, East Lansing, MI, USA African countries, with a West African regional emphasis,
3
Department of Community Sustainability, Michigan State and for a five-country West African sub-sample, over 55
University, East Lansing, MI, USA years using structural equation modeling with latent vari-
4
Department of Forestry and Environmental Resources, North ables (SEMLV).
Carolina State University, Raleigh, NC, USA For decades scholars have investigated food security as
5
Department of Construction Science, Texas A&M an important component of well-being, since answering the
University, College Station, TX, USA question of who goes hungry and why is a topic of great
6
Department of Agricultural, Food, and Resource Economics, importance to populations, governments, and policymakers
Michigan State University, East Lansing, MI, USA worldwide. Food security is recognized as a fundamental

123
 Environ Syst Decis

right and is a central goal of development efforts and originated from efficient, effective, [equitable] and
international organizations including the Food and Agri- low-cost food systems that are compatible with sus-
culture Organization (FAO) of the United Nations, the tainable use of natural resources. (Quisumbing et al.
World Bank, and the International Food Policy Research 1995, p. 50)
Institute (IFPRI). Famines and disasters draw episodic
The four key aspects of food security—availability, access,
international media attention to this recalcitrant problem,
stability, and utilization—are described by the FAO (2008)
such as in Niger in 2004/2005 (BBC 2005), and, as of this
and encompass the cross-scale relevance of daily struggles
writing in 2017, in Somalia (Gettleman 2017), yet food
for subsistence related to adequate food along with system-
insecurity reflects deeper structural challenges related to
level production and distribution challenges. Availability
inequality and distributional issues related to resource
and access describe (1) the presence or existence of food,
access. Ecological factors contributed to both crises, yet
that is whether food in the area or particular place is
played out via a slightly differing confluence of events.
physically available, and (2) physical and economic access
Although the poor harvest the preceding fall was the
to that food. Utilization refers to food preparation and
proximate cause of the 2005–2006 famine in Niger, the
feeding practices, food diversity and distribution in the
lack of rain during the rainy season in the month of August
household, and the biological ability to utilize the food
the prior year—typically when the heaviest rains occur—
consumed, which determines nutrition status and thus
and subsequent pest damage where locusts destroyed crops
represents where physical health becomes a crucial com-
were key precipitating events. Drought, given the pro-
ponent of food security. Although present in varying
longed absence of rain for nearly 2 years, is one factor in
degrees depending on any number of social, economic, and
what is being labeled as the 2017 famine in Somalia. Crop
political factors, the configuration and stability over time of
failure and livestock deaths are occurring, and more than
the other three pillars in relation to one another can define a
six million Somalians are facing increased prospects of
population as food insecure and/or be applied to the system
malnutrition and related diseases. Thus, food security is a
in its entirety.
challenging and critically important topic given its com-
Research on food security is evident in numerous dis-
plexity with regard to conceptual dimensions, its cross-
ciplinary fields in the social sciences including sociology,
scale attributes, and variability across spatial, temporal,
economics, geography, and political science as well as in
and social gradients.
the physical sciences including agronomy, ecology, and
The paper proceeds as follows. The next section
biogeochemistry. Increasingly, much of the food security
describes prior research that informs this study, in which a
work is transdisciplinary, incorporating insights from
path diagram of the drivers of food insecurity across the
relevant fields as needed to address this critical issue with
four pillars of availability, access, stability, and utilization
a shared objective: to improve human well-being. Given
is developed. Next, the data and empirical models are
this broad disciplinary breadth, the unit of analysis of
discussed, followed by results and discussion of the anal-
studies of food security ranges from individuals and
yses. The paper concludes with a discussion of avenues for
households to communities, regions, countries, and groups
future academic work.
of countries as defined by geographic boundaries (e.g.,
Sub-Saharan Africa) and/or geo-political/economic cate-
gories (e.g., less developed countries or LDCs). Extant
2 Defining and explaining food insecurity
work on food security chronicles not just the particular
constellation of the pillars but also how these intercon-
Consensus has emerged among scholars in recent decades
nected aspects play out over time, over space (e.g., across
regarding how to define food security in relation to the
communities and/or countries), and in relation to defining
household. At the household level, food security is the
vulnerable groups. Many programs run by international
ability of a household to secure enough food to ensure
organizations aimed at reducing hunger and food insecu-
adequate dietary intake of socially acceptable foods for all
rity are targeted at the national level to allow for tailored
its members. Efforts to explain food insecurity focus on
approaches with regard to how policies are constructed
how availability, access, utilization, and stability take
and implemented (Stamoulis and Zezza 2003). As a result,
shape across different scales ranging from the global to the
studies of food security are wide-ranging and varied,
individual and how this affects human populations and
including regionally focused investigations, country-
decision-making, especially at the household level.
specific work, within-country research, examinations of
Food security exists when every person has access to households specifically, as well as multi and cross-scale
sufficient food to sustain a healthy and productive approaches. Throughout, the crucial roles of contexts are
life, where malnutrition is absent, and where food highlighted.

123
Environ Syst Decis

In explaining food security, scholars have developed population, thus curtailing the potential for economic and
conceptual models that to varying degrees incorporate the social development and thus institutional transformation.
pillars of food security. Building models is challenging This includes work on food security, particularly at the
given the uneven spatial distribution of these concepts and national or macro-comparative scale as incorporated in
the identification of—to varying degrees—both proximate cross-national studies (Firebaugh and Beck 1994; Jenkins
and underlying causes of hunger, undernutrition, malnu- and Scanlan 2001).
trition, and food insecurity. The seminal work of Sen Studies also examine economic and political determi-
(1981), for instance, outlined a clear shift in the puzzle of nants of food availability and access using a macro-com-
food security. How and why in the context of an increasing parative, cross-national approach to investigate whether or
global food supply is food insecurity possible? Moving not an available food supply and by extension the broader
from the availability or supply-side argument, Sen posited food system holds the key to enhancing food security
that more attention should be paid to questions of access through reducing food insecurity (Jenkins and Scanlan
that illuminate demand-side dynamics and bring social, 2001; Scanlan and Jenkins 2001; Brady et al. 2007; Jenkins
political, and economic factors squarely into focus as key et al. 2007). This work reveals that food availability does
drivers (Barrett 2010). Specifically, questions of access are not necessarily result in food security in less developed
demand-side topics that maintain a focus on adequacy of countries (LDCs) (Jenkins and Scanlan 2001) and that
economic production and growth. More recent work builds economic gains do not consistently result in improved
on this, introducing notions of equity and justice with food, nutrition, and health (Brady et al. 2007). It is also the
regard to distributional issues that also have cross-scale case that increases in food security, as measured by
relevance. Briefly, harkening back to defining food inse- reductions in child hunger, is not uniformly found in LDCs
curity, differences can be shown within a nation as well as but rather is confined to a subset of well-to-do LDCs
within a household in terms of whether all societal or (Scanlan and Jenkins 2001). This work also demonstrates
household members have access to sufficient food to meet the relevance of more than just food supply, such as key
their needs. socio-political variables including gender stratification and
A central element in work on food security is the role of military conflicts that increase the prevalence of child
environmental conditions in facilitating agricultural pro- hunger in the developing world (Jenkins et al. 2007).
duction and economic growth. Ecological factors linked Expanding social and political rights and improving
with temperature, rainfall, and soil conditions have clear gender equality has been advanced as important for
roles in agricultural production, particularly in Africa reducing child hunger and by extension food insecurity of
where most agriculture is rain-fed. Given this fundamental vulnerable populations (Jenkins et al. 2007; FAO 2015a;
reliance on natural system irregularity, populations are UNDP 2017). Reductions in health and educational dis-
especially vulnerable to shifts in ecological conditions like parities between income groups, women and men, children
climate variability. Food production as evidenced by and adults, rural and urban residents, and other social
agriculture and increased food supply is posited as crucial categories have the potential to reduce barriers to food
for development. Measured at the national scale and pro- accessibility as well as the stability and utilization of
moted as vital to improving societal well-being, increases available resources. Strategies for poverty reduction that
in agricultural production through increased crop yields are emphasize access to basic needs and increasing opportunity
touted as positive, and expected to reduce food insecurity for livelihood gains have long been argued to result in
with respect to food availability. promoting increased overall well-being such as education
A long line of research in development argues that gains for girls that can with time result in gains in gender
economic growth promotes societal and human well-being, equality for women (Bouis and Hunt 1999; Brady et al.
at least since the inception of the growth consensus model 2007).
(Rostow 1959; Kerr 1960). Within the field of develop- Given this broad disciplinary breadth, the unit of anal-
ment, promotion of the ‘benefits to growth’ model spills ysis of studies of food security ranges from individuals and
over into topics linked with economic growth like indus- households to communities, regions, countries, and groups
trial and agricultural production. In effect, wealth accu- of countries as defined by particular discipline-relevant
mulation reaches all social and economic categories in geographic boundaries (e.g., Sub-Saharan Africa) and/or
society—from rich to poor—and thus improves human geo-political/economic categories (e.g., less developed
welfare, although likely via different mechanisms related countries or LDCs). Some conceptual models include
to the international stratification system including capital, multiple paths of influence (Hammond and Dube 2012;
aid, and investment as well as the intra-national income Stamoulis and Zezza 2003). Given our focus on the nation-
inequality distribution. The development potential of a state as the unit of analysis and use of SEMLV, we draw
nation can be hampered by poverty and poor health of the from related substantive work investigating life expectancy

123
 Environ Syst Decis

and the likelihood of experiencing a disaster for the com- boxes depict observed (manifest) variables and ovals show
parative, cross-national approach. latent (unobserved) variables. Conceptual groupings that
Substantively related cross-national, comparative work map onto the four pillars of food security are shown by
using SEMLV investigates life expectancy and disaster boxes enclosed with dashed lines. Our model includes
devastation (Austin and McKinney 2012, 2016). In the latent variables, created from the conceptual underpinnings
former, hunger is included in a model of health, political, in previous work, which are abstract constructs best mod-
and economic factors affecting life expectancy in devel- eled using multiple indicators that are conceptually rele-
oping countries, specifically LDCs, and in Sub-Saharan vant and intercorrelated. We describe each latent variable
Africa (SSA) given the low life expectancy across the and the specific individual measures of which they are
region and also in comparison globally with the rest of the comprised in the data and methods section. This model
world. In the latter, disaster vulnerability is examined in shows that environmental, socio-political, and economic
poor nations as the result of social, economic, political, and variables have direct and indirect effects on measures of
environmental factors associated with underdevelopment. availability, accessibility, utilization, and the final outcome
Examining a path model of disaster vulnerability with 85 variable: undernutrition.
nations, Austin and McKinney (2016) demonstrate direct Moving from left to right in the path diagram in Fig. 1,
effects of health resources and gender equality and indirect our hypothesized model shows that we use three variables
effects of economic and political variables working to incorporate social and environmental stability through-
through these two measures. out the model. At the top of the model diagram, agricul-
Extant work demonstrates that the factors shaping food tural inputs are posited to affect agricultural production
security are complex and interrelated, encompassing social, which then affects variability in the food supply and food
economic, political, and environmental issues from poverty aid, all of which affect the adequacy of dietary energy
and inequality to social rights protections, health, and supply. These effects comprise the food supply or avail-
temperature and rainfall. Many empirical models tend to ability part of the model. At the bottom of the model are
emphasize agricultural variables as central to understand- the utilization variables related to disease burden, preven-
ing food security, often as stand-alone models without tion, and treatment. Also of importance is accessibility,
specific ties to social systems. We seek to integrate these measured with poverty, which is in the center of our path
disparate threads into an empirical model that accounts for diagram and serves an important link between the stability,
complexities related to identifying primary and secondary availability, and utilization components and their ultimate
factors affecting food security (Hammond and Dube 2012; effect on our final outcome variable: undernutrition. We
Drimie and McLachlan 2013; Stamoulis and Zezza 2003). use a latent variable of undernutrition that encapsulates
Prior work outlines the factors shaping food security to be multiple measures of the nutritional status of the population
direct and indirect but do not necessarily construct models as described in detail below. The nutritional status of the
that reflects the complexity. In addition, although work population, especially as it applies to vulnerable groups
includes aspects of accessibility and availability, stability, like children or the elderly, status has long been viewed as
and utilization, the latter is explicitly examined with less a measurable outcome of hunger and food insecurity
frequency in empirical models. In this study, we build a (Anderson 1990; UNICEF 1990; Campbell 1991; Sta-
model of food security that specifies multiple paths from moulis and Zezza 2003; Bhattacharya et al. 2004; FAO
social, economic, political, and agricultural factors and 2015b).
tests these proposed relations in African countries, the To develop this model, we drew from past studies
West Africa region, and a group of five dryland West regarding what concepts were essential to include and how
African countries over 55 years using structural equation they were related to one another. Our model does not
modeling with latent variables (SEMLV). replicate a model from previous work but does share
emphasis on the pillars of availability and stability and
relies on prior conceptual work to also bring access and
3 Modeling food security utilization into the framework (Hammond and Dube 2012;
Stamoulis and Zezza 2003). Data coverage and availability,
Drawing from prior research described in an earlier section, both spatially and temporally, were significant factors
here we outline an empirical model that encompasses the informing our decisions about how availability, access,
four main pillars of food security: stability, availability, stability, and utilization were operationalized and
access, and utilization (FAO 2008, 2010; Hammond and interconnected.
Dube 2012; Stamoulis and Zezza 2003). Figure 1 shows
our hypothesized model of food security and undernutrition
using conventions from path analysis and SEMLV where

123
Environ Syst Decis

Fig. 1 Conceptual path diagram. Boxes indicate observed variables, ovals indicate latent variables, dashed boxes indicate variables that
represent the four pillars of food security

4 Data and methods specification as latent (unobserved) or manifest (observed)

variables.
Structural equation modeling (SEM) has several benefits Figure 2 shows a pictorial representation of the latent
over the linear regression model because it can accom- variable water and sanitation infrastructure using con-
modate multiple relationships between multiple variables ventions from SEMLV where boxes depict observed or
by simultaneously estimating multiple regression models. indicator variables and ovals show latent variables. Error
This allows one to model indirect, mediating and covarying terms, here measurement errors, are represented by the
relationships in addition to the direct relationships possible arrows pointing to the boxes. The latent level of water and
with simple regression models. Furthermore, the ability to sanitation infrastructure in a nation consists of four critical
include latent variables in an SEM [i.e., structural equation measures: how much of the population has access to
modeling with latent variables (SEMLV)] provides a way improved sanitation facilities and to an improved water
to include more abstract concepts, as well as additional source, and the percentages of the population in rural and
error terms. Latent variables are constructed as measure- urban areas not practicing open defecation. This measure
ment models and tested using confirmatory factor analysis builds on and extends previous work by specifying multiple
(CFA). We use SEMLV to specify the relations between measures linked together as depicted in the figure and
variables in our model of food insecurity in Africa, West including measurement error in each variable (Bollen
Africa, and five dryland West African nations, according to 1989) as well as correlation between the errors of the open
the main conceptual categories or pillars as well as their defecation measures. Further, this latent variable can be
thoroughly examined by way of component and overall

123
 Environ Syst Decis

Fig. 2 Confirmatory factor

analysis (CFA) of measurement
model for water and sanitation
infrastructure latent variable,
with component and overall fit
measures

model fit that examine whether this model adequately 4.1 Outcome variables
captures the relations shown.
As an initial step in model building and testing, CFA ‘Undernutrition’ is a latent construct (or measurement
analyses or measurement models were completed for each model evaluated using confirmatory factor analysis (CFA),
of the latent variables. That is, we completed CFA analyses a technique in SEM) that includes three variables capturing
for each latent construct on its own, using the full sample critical dimensions of undernourishment: prevalence of
of countries and years, to check the reliability of each undernourishment as percent of the population experienc-
observed variable in the measurement model, and overall ing it, prevalence of undernourishment as a 3-year average
fit. Detailed results for component and overall fit for each of the percent of the population experiencing it, and share
latent variable in the model in Fig. 1 are available upon of dietary energy supply from starchy staples (i.e., cereals,
request. Numerous measurement models were tested while roots and tubers) as a 3-year average of the percent of the
selecting the specific measurement models used for the diet as a measure of lack of dietary diversity (Hoddinott
four latent variables in this model. Many latent variables et al. 2002). Empirical checks provide information
were considered that we ultimately reduced to summed regarding the validity and reliability of the individual
indexes (such as the poverty index, natural disaster index, measures (e.g., standardized factor loadings ranging from
and education index variables) or single observed variables 0.995 to 1 and unstandardized loadings from 0.892 to 1, all
(such as the tuberculosis rate variable), due to lack of data significant).
and/or poor measurement model fit. The Poverty Index variable comes from combining two
Variables in the full empirical model include measures measures of poverty: the percentage of the population
from social, economic, political, health, and agricultural living on less than $2 a day, and the poverty gap at $2 a
domains. All variables were gathered from secondary data day—the mean shortfall from $2 a day expressed as a
sources including the World Bank’s World DataBank percentage.
(http://databank.worldbank.org/data/home.aspx), and the We include four measures of food availability: three
FAO’s FAOSTAT (http://www.fao.org/faostat/en/#home), measures of food from different sources and one of the
unless otherwise noted. Variable transformations were adequacy of the food supply. Food production is the
completed as appropriate and as described in the following. average value of food produced in standardized dollars per
We have 8 outcome (endogenous) variables in our full person expressed as a 3-year average. Food imports are
empirical model, of which 6 are observed and 2 are latent, shown as a percentage of merchandise imports. Food aid is
and 12 predictor (exogenous) variables of which 10 are in grams per capita per day and includes both cereal and
observed and 2 are latent. non-cereal foods. Diet energy adequacy is the 3-year
average of dietary energy supply adequacy expressed as a
percentage (e.g., 100% = ‘adequate’).
We include two measures related to food utilization.
Tuberculosis rate is the incidence of tuberculosis per

123
Environ Syst Decis

100,000 people; we use this variable as a proxy for the CPIA gender equality rating (rescaled to look like a per-
health of the population. Disease prevention and treatment centage), and the percentage of teenagers (ages 15–19) that
is a latent variable with six indicators. Three indicators are do not give birth (a transformed version of the adolescent
immunization rates for: diphtheria, pertussis and tetanus fertility rate). The CFA results and fit statistics indicate a
(DPT) immunization (percentage of children ages reasonable fit (Chi-square is significant,3 RMSEA =
12–23 months), Hepatitis B (percentage of children 1-year 0.069, IFI & CFI = 0.973).
old), and measles (percentage of children ages Water and Sanitation Infrastructure is a latent variable
12–23 months).1 The additional four indicators are: the that includes 4 predictors: the percentage of the population
percentage of newborns protected from tetanus by virtue of with access to improved sanitation facilities, the percentage
being born to mothers who had been vaccinated, the per- of the population with access to an improved water source,
centage of children under 6 months old who are exclusively the percentage of the rural population not practicing open
breastfed, and the percentage of children under age 5 with defecation, and the percentage of the urban population not
diarrhea who are treated with oral rehydration salts. Fit practicing open defecation. Fit statistics of this measure-
statistics of this measurement model indicate very good to ment model indicate very good to excellent fit of this latent
excellent fit of this latent construct (Bollen 1989; West construct (Bollen 1989; West et al. 2012). The fit of this
et al. 2012). Overall model fit statistics are excellent—the latent variable is very good (Chi-square is significant,
Chi-square value is non-significant and values for the RMSEA = 0.052, IFI & CFI = 0.996).
Incremental Fit Index (IFI), and the Comparative Fit Index The Gross domestic product (GDP) per capita in con-
(CFI), are 0.999. Values above 0.95 suggest very good to stant 2005 US dollars is included to measure economic
excellent fit for these measures (West et al. 2012). The development. The age dependency ratio (ADR) is the ratio
root-mean-square error of approximation (RMSEA) is of individuals younger than 15 and older than 64 (depen-
0.015 (CI 0.000, 0.031). For the RMSEA, values closer to dents) to those between the ages of 15 and 64 years
zero are desirable and below 0.05 suggests very good fit. expressed as a percentage. The remaining three exogenous
observed variables are related to agricultural productive
4.2 Predictor variables capacity: the amount of agricultural land in 1000 sq. km,
the tractor density, as the number of tractors per 100 sq. km
We include three measures of stability, one environmental of arable land, and the amount, in kilograms, of fertilizer
and two social. The natural disaster index is a sum of nine used per hectare of arable land. Education is measured as
variables from the International Disaster Database from the the summed index of three variables: the gross enrolment
Centre for Research on the Epidemiology of Disasters ratio in primary, secondary and tertiary schools of both
(CRED) (http://www.emdat.be/). These were nine dummy sexes combined expressed as a percentage. The gross
variables on whether or not there was a: drought, earth- enrolment ratio is the total enrollment of all ages compared
quake, extreme storm, temperature extreme, flood, insect to total population of expected age for the education level.
infestation, landslide or dry mass movement, volcanic Public health expenditures are the healthcare portion paid
activity or a wildfire in the year. Social rights is a summed by government entities expressed as a percentage of total
index of two variables: the scores of the degree of civil health expenditures.
liberties and the degree of political rights from Freedom
House (https://freedomhouse.org/reports), reverse coded so 4.3 Sample background
that high values indicate high degrees of rights and liber-
ties. The Refugee population variable is the percentage of Using the above-mentioned model and variables, we focus
the country’s population that are refugees. on food insecurity in Africa where roughly one-fourth of
The latent construct ‘Gender Equality’ includes six the population goes hungry. Fourteen African countries
variables. Four variables are related to education: the have undernourishment rates of 25% or more of the pop-
gender parity index (GPI) for gross enrollment in primary, ulation, another 7 have undernourishment rates of
secondary and tertiary schools,2 which is the ratio of girls 15–24.9%, and 20 countries have failed to meet their
to boys enrolled at each level of school expressed as a Millennium Development Goal hunger target—which in
percentage, and the ratio of female to male literacy rates the aggregate are higher rates than any other continent
expressed as a percentage. The other two variables are the (FAO 2015a, b). Paarlberg (1999) suggests that a combi-
nation of historical, political, economic, ethnic and envi-
1 ronmental factors contribute to Africa’s distinctiveness
We include theoretically based correlated errors between the three
immunization rate variables.
2 3
We include theoretically based correlated errors between the three Give the large size of our sample a significant Chi-square test is not
education GPI variables. surprising.

123
 Environ Syst Decis

Table 1 Background on region sub-samples and African continent: 2005–2014 averages per country
Africa West Africa 5 Countries

Years of independence 88.3a 58.2 53.6

Number of former British colonies 20 4 2
Number of landlocked countries 16 3 3
Land area (sq. km) 550,183.5 378,979.4 779,760.0
Percent of arable land 13.6 17.8 19.0
Percent of agricultural land 47.3 48.4 51.5
Total population 19,217,412.8 19,045,638.6 45,670,998.8
Rural population (% of total population) 58.8 57.4 65.6
Urban population (% of total) 41.2 42.6 34.4
Population density (people per sq. km of land area) 88.0 75.1 72.1
Age dependency ratio 80.8 87.0 93.4
Life expectancy at birth (years) 57.7 56.4 55.0
GDP per capita (constant 2005 US$) 1944.0 659.7 552.0
GINI index 43.7 39.7 38.7
Access to improved water source (% of population) 71.6 70.6 68.3
Access to improved sanitation facilities (% of population) 38.8 25.5 18.6
Total ecological footprint per capita 1.3 1.4 1.5
Cereal yield (kg per hectare) 1494.4 1217.6 1186.3
Prevalence of undernourishment (% of population) 20.9 16.2 11.2
Share of dietary energy supply derived from starchy staples (%) (3-year average) 23.6 17.4 12.5
Prevalence of food inadequacy (%) (3-year average) 11.3 3.1 1.7
Prevalence of stunting (% of children under 5) 35.2 33.9 37.3
Depth of the food deficit (kcal/person/day) 151.5 116.5 79.4
a
This number includes Ethiopia, which was never colonized. If Ethiopia is left out, the average years of independence is 52.5 years

compared to developing countries in other parts of the nations,5 and a five-country sample of nations within West
world when it comes to efforts to improve food security. Africa.6 Our approach creates multiple spatial comparisons
Not only does Africa seem to be unique from other global to broaden our understanding of how food security takes
regions, but there is also likely to be substantial variation shape across the African continent, across a region (West
within global regions including within Africa (Barrett Africa), and then further zooms into five countries to pro-
2010). vide a comprehensive look at how these dynamics play out
In our analyses, we seek to further our understanding of at different spatial scales. Comparing across these three
the spatial distribution of vulnerable populations and the groupings, there is considerable variation in many indica-
social, economic, political, and ecological contexts within tors (see Table 1).
which these individuals are embedded. Thus, we consider We highlight a few values from the table. First, the
Africa as a continent,4 a regional sample of West African cereal yields are highest when a continental lens is used
(close to 1500 kg per hectare averaged over countries and
years) and lower across West Africa as a region and the
group of five West African countries at 1218 and 1186 kg
4
Southern Region: Botswana, Lesotho, Namibia, South Africa, per hectare averaged as appropriate. Second, it should be
Swaziland; Eastern Region: Burundi, Comoros, Djibouti, Eritrea,
Ethiopia, Kenya, Madagascar, Malawi, Mauritius, Mozambique,
noted that the prevalence of undernourishment as a percent
Rwanda, Seychelles, Somalia, South Sudan, Tanzania, Uganda, of the population is highest across the African continent
Zambia, Zimbabwe; Middle Region: Angola, Cameroon, Central (approximately 21%) and lower in both West Africa as a
African Republic, Chad, Congo, Democratic Republic of the Congo,
Equatorial Guinea, Gabon, Sao Tome and Principe; Northern Region: 5
Algeria, Egypt, Libya, Morocco, Sudan, Tunisia; Western Region: Benin, Burkina Faso, Cape Verde, Cote d’Ivoire, Gambia, Ghana,
Benin, Burkina Faso, Cape Verde, Cote d’Ivoire, Gambia, Ghana, Guinea, Guinea-Bissau, Liberia, Mali, Mauritania, Niger, Nigeria,
Guinea, Guinea-Bissau, Liberia, Mali, Mauritania, Niger, Nigeria, Senegal, Sierra Leone, and Togo.
6
Senegal, Sierra Leone, Togo. Burkina Faso, Ghana, Mali, Niger, and Nigeria.

123
Environ Syst Decis

Table 2 Descriptive statistics for model variables

Mean SD Range min Range max N

Exogenous observed variables

Fertilizer use 37.289 99.759 0.000 696.593 413
Tractor density 39.282 83.694 0.004 1170.000 1568
Agricultural land (1000 sq. km) 207.256 248.465 0.030 1366.980 2725
Natural disasters index 0.440 0.682 0.000 4.000 2885
Refugee population 0.738 1.437 0.000 15.744 1092
Age dependency ratio (ADR) 88.847 12.904 40.796 114.371 2885
GDP per capita 1368.800 2090.210 68.567 15,592.170 2429
Education index 116.340 53.236 11.432 276.293 1067
Public heath expenditures 46.835 16.965 3.091 94.567 1035
Social rights 49.290 9.729 16.670 75.000 387
West Africa 0.305 0.460 0.000 1.000 2885
5 Countries 0.095 0.294 0.000 1.000 2885
Exogenous latent variables
Water and sanitation infrastructure
Improved sanitation facilities 36.275 26.011 2.600 98.400 1284
Improved water source 66.451 18.868 13.200 99.900 1281
Rural people not practicing open defecation 59.868 28.334 0.000 100.000 1259
Urban people not practicing open defecation 90.347 12.010 32.400 100.000 1259
Gender equality
Female literacy ratio 69.877 20.972 21.907 129.672 185
Gender equality rating 53.630 8.757 33.330 75.000 388
Primary school enrollment GPI 82.454 18.258 31.913 150.735 1881
Secondary school enrollment GPI 71.307 28.103 7.857 160.186 1381
Tertiary school enrollment GPI 51.363 39.366 0.000 329.519 1052
Adolescent non-fertility rate 87.326 5.283 76.468 99.396 2885
Endogenous observed variables
Food production 151.171 60.032 58.000 346.000 1009
Food imports 19.020 8.327 0.474 62.416 1501
Total food aid 23.028 52.596 0.000 513.125 1429
Diet energy adequacy 108.000 15.640 69.000 152.000 1032
Poverty index 96.522 43.061 0.000 176.180 170
Tuberculosis rate 291.530 259.504 15.000 1354.000 1325
Endogenous latent variables
Undernutrition
Prevalence of undernourishment 25.216 15.443 5.000 76.800 1012
Share of dietary energy supply from cereals 27.294 14.678 5.000 76.800 941
Prevalence of undernourishment (3-year average) 33.613 16.756 5.000 83.500 1006
Disease prevention and treatment
DPT immunization 65.666 25.954 1.000 99.000 1702
Measles immunization 65.528 23.000 1.000 99.000 1694
Hep B immunization 81.193 17.224 10.000 99.000 576
Tetanus immunization 62.000 24.973 1.000 99.000 1574
Diarrhea treatment 29.414 20.408 1.000 93.600 354
Exclusive breastfeeding 31.612 19.304 1.200 88.400 164

123
 Environ Syst Decis

Fig. 3 Path diagram showing simplified model results for Africa model

region and the five West African countries at 16.2 and outlined in Fig. 1, specifying that environmental, socio-
11.2%, respectively. Third, the prevalence of food inade- political, and economic variables have direct and indirect
quacy is largest at the continent at 11.3% of the population, effects on measures of availability, accessibility, utiliza-
lower in the West Africa region at 3.1% and lower still for tion, and food insecurity or undernutrition. Our analyses
the group of five West African countries at 1.7% of the use the four pillars of food security—accessibility, avail-
population. Fourth, the depth of the food deficit is the ability, utilization, and stability—as specified in the model
largest at the continental scale at 151 kcal per day com- shown in Fig. 1 and explained in detail in this section. Our
pared with the West Africa region and group of five West sample is all 54 African countries over 55 years
African countries at 116 and 79 kcal per day, respectively. (1960–2014), which provides 2885 observations. We use
Although important, these values do not show the full structural equation modeling with latent variables
range of variability across the continent nor do they show (SEMLV) to estimate the model using AMOS software.
the within-region differences for individual nations. We run three models: one unrestricted model that covers
the entire African continent, and two that use a dummy
4.4 Analysis variable to control for the West African region, and a five-
country sub-sample of West African countries, respectively
Table 2 presents the descriptive statistics for the variables (see footnotes 4, 5, and 6).
used in the analysis. We investigate the empirical model

123
Environ Syst Decis

Table 3 Standardized and unstandardized maximum likelihood coefficients, and standard errors from SEMLV of undernutrition, predictors of
dietary energy adequacy in Africa, with controls for West Africa region, and five-country sub-sample, 1960–2014 (n = 2885)
Africa West Africaa 5 Countrya
Unstd coeffs (SE) Std coeffs Unstd coeffs (SE) Std coeffs Unstd coeffs (SE) Std coeffs

Food production 0.164*** 0.611 0.163*** 0.610 0.161*** 0.603

(0.007) (0.007) (0.007)
Food imports 0.384*** 0.200 0.254*** 0.131 0.365*** 0.190
(0.055) (0.056) (0.054)
Food aid -0.007 -0.022 -0.007 -0.022 -0.004 -0.013
(0.008) (0.008) (0.008)
Social rights 0.331** 0.063 0.260* 0.050 0.269* 0.051
(0.127) (0.124) (0.126)
Natural disasters index 0.580 0.025 0.821 0.035 0.517 0.022
(0.571) (0.558) (0.567)
* p \ 0.05; ** p \ 0.01; *** p \ 0.001 (two-tailed)
a
Regional dummy variable not shown

5 Results and discussion standardized coefficients, followed by food imports and

social rights. Interestingly, food aid does not have a sig-
Figure 3 shows the generalized results of the Africa model nificant effect on diet energy adequacy in any of our
from Fig. 1, while in Tables 3, 4, 5, 6 and 7 we present the models. The lack of significance, and small effect size, of
main results7 of our models, focusing on the drivers of four food aid could be due to the nature of food aid as an
key variables: diet energy adequacy, poverty index, tuber- emergency measure—it is not a primary source of food at
culosis rate, and undernutrition. In viewing Fig. 3, it is clear the national level and may be responding to the condition
that undernutrition is driven by a complex set of factors. On of low diet energy adequacy, suggesting a reciprocal
the whole, we find many of the significant relationships relationship not fully captured in this model. The effect
between variables in our model to be as we would expect. sizes are quite similar across our three models indicating
Broadly speaking, agricultural resources increase food that the drivers of food availability are consistent in West
production, which in turn increases diet energy adequacy, Africa and in our five-country sub-sample compared to the
which reduces undernutrition, while poverty increases entire continent.
undernutrition and better healthcare reduces it.
5.2 Accessibility
5.1 Availability
While the first prerequisite for food security is that there is
Diet energy adequacy, the average dietary energy supply food present in the country, the second requirement is that
adequacy expressed as a percent, is the variable through people can physically get to the food and are then able to
which we hypothesize that the other food availability acquire some of it. The accessibility of food is perhaps the
variables work. Food production, food imports and food aid most difficult aspect of food security to measure at the
are all food sources that would be expected to directly national level because many of its properties rely on indi-
contribute to overall food availability. In addition, we vidual-level capacity. We approximate accessibility in our
expect that more socially and institutionally stable coun- model with an index measure of the poverty rate. As shown
tries will have a more consistent and sufficient food supply in Table 4, the poverty index in Africa is most significantly
and that natural disasters are likely to disrupt the food reduced by higher GDP per capita, and by greater social
supply. While having enough food present in a country is rights, and higher levels of the education index to a lesser
not sufficient to ensuring food security for the population, it degree. The ADR has the most influential positive effect on
is a necessary component of food security. Table 3 shows the poverty index, consistent with the expectation that
the model results for diet energy adequacy. children and the elderly are most vulnerable to poverty,
As expected, food production is the most important followed by gender equality. The refugee population is not
driver of diet energy adequacy, as indicated by the found to significantly influence the poverty rate.
Of the significant effects, GDP, social rights, ADR and
7
the education index are in the expected direction, while
Detailed results are available upon request.

123
 Environ Syst Decis

Table 4 Standardized and unstandardized maximum likelihood coefficients, and standard errors from SEMLV of undernutrition, predictors of
poverty index in Africa, with controls for West Africa region, and five-country sub-sample, 1960–2014 (n = 2885)
Africa West Africaa 5 Countrya
Unstd coeffs (SE) Std coeffs Unstd coeffs (SE) Std coeffs Unstd coeffs (SE) Std coeffs

Social rights -1.934** -0.125 0.417 0.026 -1.619** -0.105

(0.616) (0.622) (0.609)
Refugee population 2.089 0.067 2.028 0.061 1.771 0.057
(1.296) (1.201) (1.298)
ADR 1.175*** 0.319 0.658*** 0.170 1.165*** 0.319
(0.195) (0.195) (0.191)
GDP per capita -0.011*** -0.510 -0.014*** -0.586 -0.012*** -0.529
(0.001) (0.001) (0.001)
Gender equality 0.424* 0.181 -0.013 -0.005 0.435* 0.184
(0.199) (0.205) (0.193)
Education index -0.164* -0.178 -0.206** -0.211 -0.133 -0.144
(0.081) (0.074) (0.078)
* p \ 0.05; ** p \ 0.01; *** p \ 0.001 (two-tailed)
a
Regional dummy variable not shown

gender equality is not. A potential explanation for the size of ADR is about half the size as in the Africa model,
effect direction of gender equality is that the measures we and the effect of the education index on the poverty index is
use in the latent variable, such as education participation, modestly larger. The five-country sub-sample model is
literacy, and adolescent non-fertility rate, simply do not very similar to the Africa model, with the exception of the
function to improve economic status due to the presence of effect of the education index on the poverty index, which is
other forms of gender discrimination. not significant.
There are some notable cross-group differences for the
drivers of the poverty index, particularly when controlling 5.3 Utilization
for West Africa, that show important spatial differences. In
the West Africa model, social rights, and gender equality The ability to absorb and thereby utilize the nutrients that
change sign and become non-significant, while the effect are consumed is an important element of nutritional

Table 5 Standardized and unstandardized maximum likelihood coefficients, and standard errors from SEMLV of undernutrition, predictors of
tuberculosis rate in Africa, with controls for West Africa region, and five-country sub-sample, 1960–2014 (n = 2885)
Africa West Africaa 5 Countrya
Unstd coeffs (SE) Std coeffs Unstd coeffs (SE) Std coeffs Unstd coeffs (SE) Std coeffs

Social rights 11.440*** 0.136 18.840*** 0.224 14.631*** 0.174

(2.651) (2.541) (2.605)
Refugee population 19.357*** 0.114 23.725*** 0.139 9.001 0.053
(5.408) (5.044) (5.389)
ADR -2.328** -0.116 -3.798*** -0.190 -2.394** -0.120
(0.780) (0.747) (0.764)
Water and sanitation infrastructure -3.274*** -0.265 -5.489*** -0.444 -3.884*** -0.314
(0.554) (0.553) (0.559)
Disease prevention and treatment -0.758 -0.063 -0.660 -0.056 -1.405*** -0.118
(0.389) (0.347) (0.393)
Public health expenditures 2.834*** 0.186 1.690*** 0.111 2.865*** 0.188
(0.512) (0.488) (0.499)
* p \ 0.05; ** p \ 0.01; *** p \ 0.001 (two-tailed)
a
Regional dummy variable not shown

123
Environ Syst Decis

Table 6 Standardized and unstandardized maximum likelihood coefficients, and standard errors from SEMLV of undernutrition, predictors of
undernutrition in Africa, with controls for West Africa region, and five-country sub-sample, 1960–2014 (n = 2885)
Africa West Africaa 5 Countrya
Unstd coeffs (SE) Std coeffs Unstd coeffs (SE) Std coeffs Unstd coeffs (SE) Std coeffs

Diet energy adequacy -1.013*** -0.925 -0.958*** -0.898 -1.001*** -0.923

(0.012) (0.011) (0.012)
ADR -0.192*** -0.141 -0.154*** -0.116 -0.186*** -0.137
(0.025) (0.023) (0.025)
Poverty index 0.092*** 0.246 0.098*** 0.286 0.095*** 0.256
(0.008) (0.007) (0.008)
Disease prevention and treatment -0.050*** -0.061 -0.069*** -0.087 -0.067*** -0.083
(0.010) (0.009) (0.010)
Tuberculosis rate -0.008*** -0.117 -0.010*** -0.151 -0.009*** -0.127
(0.001) (0.001) (0.001)
* p \ 0.05; ** p \ 0.01; *** p \ 0.001 (two-tailed)
a
Regional dummy variable not shown

security. Poor health can cause as well as be caused by of public health expenditures on tuberculosis rate may be
poor nutrition. Table 5 shows the results of the drivers of due to having unmeasured reciprocal effects—the govern-
tuberculosis rate, a proxy for overall population disease ment might be spending a larger proportional share of
burden. In Africa water and sanitation infrastructure is the health care costs because there are higher tuberculosis
most important driver, as indicated by the size of the rates, not because publicly funded healthcare somehow is
standardized coefficient, and it has an expected negative less effective or contributes to the tuberculosis rate.
effect on tuberculosis rate. The ADR is the other predictor The West Africa results for tuberculosis rate are
that has a negative and significant effect on the tuberculosis somewhat different in magnitude than for all of Africa,
rate, while social rights, refugee population, and public with greater effects of social rights, refugee population,
health expenditures have a positive effect on the tubercu- ADR, and water and sanitation infrastructure, and a
losis rate. There is also a significant and positive non- smaller effect of public health expenditures. The five-
causal relationship between the tuberculosis rate and the country sub-sample model is quite similar to all of Africa
poverty index.8 with the notable exceptions of refugee population, which is
Three variables have counterintuitive effects. The neg- not significant, likely due to there being very low rates of
ative effect of ADR on tuberculosis rate is not as expected, refugees in these countries,9 and disease prevention and
but may be a result of the use of tuberculosis rate as a treatment having a significant negative effect on tubercu-
proxy for disease burden. We would expect having a larger losis rate. These counterintuitive effects reinforce the
proportion of young and old people in the population importance of considering multiple levels of granularity
would increase the disease burden of the population as the and especially smaller-scale context effects.
young and old are more vulnerable to disease. However, if
young and old people are less likely to contract tubercu- 5.4 Undernutrition
losis specifically, then the negative effect of ADR is rea-
sonable. The positive effects on tuberculosis rate by social Table 6 shows the direct effects on our key outcome
rights and public health expenditures are also not as variable, undernutrition, and Table 7 shows the total
expected, but may be due to false assumptions of what effects of all the model variables on undernutrition. Un-
effect these variables should have. For instance, we include dernutrition in Africa is directly reduced (Table 6) by
social rights as a social stability measure, but it may be greater diet energy adequacy, most substantially, the ADR,
embedded within other neoliberal concepts and policies disease prevention and treatment, and tuberculosis rate. As
that ultimately decrease peoples’ health. The positive effect expected, poverty is found to increase undernutrition. Of
these effects, the ADR and tuberculosis rate are not in the
8
This non-causal relationship was included because of the suspected
9
reciprocal relationship between health status and poverty that we were The mean for these five countries ranges from 0.006 to 0.090%,
not able to otherwise include in this model. while the Africa average is 0.738%.

123
 Environ Syst Decis

expected direction. The negative effect of ADR may be due of negative. Gender equality has a small but negative total
to the working age population shielding the more vulner- effect in the West Africa model, in contrast to the other two
able members of society through household food distribu- models, and the effect of water and sanitation infrastruc-
tion strategies (Bhattacharya et al. 2004). The more ture is much larger. We also see the total effect of social
surprising negative effect of tuberculosis rate on under- rights range from 0.003 to 0.131 across the models, simi-
nutrition could be the result of an unmodeled reciprocal larly to GDP per capita, which ranges from -0.677 to
relationship between undernutrition and tuberculosis rate -0.906. The total effect size of education and disease
as a proxy for disease burden, where undernutrition prevention and treatment also varies across the models.
increases disease rate more than disease rate increases
undernutrition. 5.5 Model discussion
The total effects on undernutrition in Table 7 compli-
ment the results of Tables 3, 4, 5 and 6 and show the In addition to finding important scale and context effects in
indirect effects that some variables have on undernutrition. the differences between our models, we make some
Of particular interest are the relatively large indirect effects observations on modeling food security and undernutrition
of fertilizer use, tractor density, and GDP per capita on that may prove useful to future modeling efforts. One
undernutrition that were not visible in the direct effects observation is that the composition of the latent variables
presented in Table 6. Additionally, the total effect of ADR matters for their role in the overall model. By definition,
on undernutrition is smaller than the direct effect shown in the observed or indicator variables must have particular
Table 6 (approximately -0.04 vs. -0.1). attributes regarding their substantive content and their
While the five direct drivers of undernutrition are quite empirical properties (e.g., directionality, number, and
consistent across spatial groups, several of the total effects measurement scale), which requires careful selection and
on undernutrition vary more substantially, as shown in attention to detail. However, even if the predictors all
Table 7, although they are generally in a direction that contribute to the same latent concept, the latent variable
confirms theoretical expectations. Looking across the may still not be measuring all the aspects of the concept
models by variable we see that tractor density and refugee that is desired. We potentially see this happening in the
population have a larger effect in the five-country model positive effect of gender equality on poverty in our model,
than in the Africa or West Africa models, while the effect where the type of gender equality we are capturing in the
of agricultural land is positive, though very small, instead latent variable, which is largely focused on education, may

Table 7 Standardized
Africa West Africa 5 Country
maximum likelihood
coefficients from SEMLV of Fertilizer use -0.511 -0.507 -0.613
undernutrition, effects on
undernutrition in Africa, with Tractor density 0.642 0.623 0.852
controls for West Africa region, Agricultural land -0.062 -0.062 0.007
and five-country sub-sample, Food production -0.570 -0.552 -0.559
1960–2014 (n = 2885)
Food imports -0.181 -0.115 -0.173
Food aid 0.020 0.019 0.012
Diet energy adequacy -0.925 -0.898 -0.923
Social rights 0.003 0.057 0.131
Natural disasters index 0.041 0.003 0.018
Refugee population 0.016 0.013 0.025
ADR -0.048 -0.038 -0.040
GDP per capita -0.677 -0.759 -0.906
Gender equality 0.045 -0.001 0.047
Education index -0.044 -0.061 -0.037
Poverty index 0.247 0.287 0.256
Water and sanitation infrastructure 0.031 0.067 0.040
Disease prevention and treatment -0.053 -0.079 -0.068
Tuberculosis rate -0.117 -0.151 -0.127
Public health expenditures -0.022 -0.017 -0.024
Region (West Africa) -0.317
Region (5 Countries) -0.241

123
Environ Syst Decis

not be the aspects of gender equality that are important to the observed complexity of food security is the reason we
reducing poverty. chose to use SEMLV rather than a classic regression model
A second observation is that we need to pay more where we would be assured of getting model fit measures.
attention to how we incorporate the stability pillar into the Our analysis examines food insecurity using country-
model. The positive effect of social rights on TB rates level data, which represent one scale—the aggregate
suggests that we may need to include more stability mea- scale—that is important to chronicle in seeking to improve
sures that are more specifically selected for each outcome well-being worldwide. However, national-level processes
variable of interest, rather than trying to establish national can mask different patterns at the sub-national scale as our
stability with a few universal variables. If there are enough analysis shows. While studies using countries or regions as
appropriate variables available, this will also be useful the unit of analysis are useful for identifying inadequacies
empirically for model identification and estimation. in supply, studies also need to delve more deeply into
A third observation is that we need to consider disen- within-country or intra-national challenges related to
tangling the potentially reciprocal effects between poverty, access and utilization (Barrett 2010).
utilization/health, and undernutrition. There is an estab-
lished conceptual connection between poverty, which
contributes to household food insecurity and undernutrition 6 Conclusion
in individuals, which can cause physical and cognitive
limitations, which then makes it more difficult to break the The results of our model suggest that using undernutrition
cycle of poverty (FAO 2008). There are also predictors that as a single, combined outcome of food insecurity, as rep-
have conceptually ambiguous causal directions at least at resented by the four pillar concepts, is a useful practice. By
the national scale. For example, does disease prevention using SEMLV, we are able to not only assess the direct
and treatment reduce the amount of disease or does the predictors of undernutrition but also the drivers of the
amount of disease drive the amount of disease prevention predictors of undernutrition. Based on standardized total
and treatment happening? effects (Table 7), we find that the most important overall
The SEMLV method can be used to test reciprocal or drivers of national-level undernutrition in all three models
feedback relationships between two or more variables, but are fertilizer use, food production, diet energy adequacy,
requires very careful model specification and sufficient GDP per capita, tractor density, and poverty. Looking
data, which we could not do in the context of our current across regional scales, we see that while the direct drivers
model beyond including the error correlation between the of undernutrition and diet energy adequacy are quite con-
poverty index and the tuberculosis rate. A model designed sistent, the drivers of poverty and tuberculosis rate are not.
just to test the empirical reciprocity between poverty, This finding indicates important inter- and intra-regional
health status/utilization, and undernutrition would be an heterogeneity that should be considered and accounted for
exciting extension of this research. Other avenues for in future research, as well as the potential need accom-
future research using SEMLV include determining relevant modate reciprocal effects among accessibility, utilization
ecological factors that can be added to the existing model, and undernutrition measures.
incorporating data on rural versus urban populations, Economic growth, political stability and social rights
comparing models for different African regions, and non- and protections are vital to ensuring continued progress is
African regions, and integrating multiple spatial scales and made toward reducing food insecurity worldwide but
time-lagged effects. especially in regions, countries, communities, and house-
Modifications to the models are possible, including holds where vulnerable populations are at high risk of food
additional paths not shown here or additional variables that insecurity. Although some notable gains have been made in
may be of growing importance in this region for food recent decades, food insecurity in Africa persists and
security. We need to simultaneously grapple with goals of remains a pressing challenge for scholars and practitioners
model parsimony and data quality, as is the case in any (FAO 2015b). Moving forward, it is clear that addressing
country-level analysis. Building multiple models across this intractable challenge requires transdisciplinary
varying gradients remains an important goal of this approaches and thinking across disciplinary lines. Criti-
research and model parsimony remains instrumental to cally, even while economic development efforts chronicle
achieve this. However, given the limitations of our data we progress in well-being, these gains are not equally dis-
decided to use a pooled sample for this analysis and will tributed across social categories and nontrivial populations
examine models for individual countries using an alterna- remain vulnerable to shifts in food supply and access.
tive but related strategy of model building, testing, and Academics and stakeholders must continue to engage in
evaluation. Reducing the complexity of the model too meaningful dialogues in order to ensure that the emphasis
much threatens its usefulness; the ability to model more of

123
 Environ Syst Decis

is on both chronic undernutrition and disaster-related Austin KF, McKinney LA (2012) Disease, war, hunger, and
hunger. deprivation: a cross-national investigation of the determinants
of life expectancy in less-developed and Sub-Saharan African
Scholars are faced with many decisions related to model Nations. Soc Perspect 55:421–447. doi:10.1525/sop.2012.55.3.
building that are important for analytical purposes and for 421
subsequent work to use as a baseline for expansion, mod- Austin KF, McKinney LA (2016) Disaster devastation in poor
ification, and possible replication in future empirical work. nations: the direct and indirect effects of gender equality,
ecological losses, and development. Soc Forces 95:355–380.
Our analyses using SEMLV include latent variables and doi:10.1093/sf/sow056
direct, indirect and total effects in a path model. There are Barrett CB (2010) Measuring food insecurity. Science
likely reciprocal relations in the model, such as between 80(327):825–828
TB rate, poverty and undernutrition, that, with more data BBC (2005) Niger food crisis timeline. BBC News, London
Bhattacharya J, Currie J, Haider S (2004) Poverty, food insecurity,
and an alternate model specification, should be assessed in and nutritional outcomes in children and adults. J Health Econ
future studies using the same sample or perhaps another 23:839–862. doi:10.1016/j.jhealeco.2003.12.008
region or a different observational unit like the household Bollen K (1989) Structural equations with latent variables. Wiley,
instead of the country level. New York
Bouis H, Hunt J (1999) Linking food and nutrition security: past
Issues related to measurement are also of critical lessons and future opportunities. Asian Dev Rev 17:168–213
importance. This includes trade-offs related to data Brady D, Kaya Y, Beckfield J (2007) Reassessing the effect of
reduction strategies, issues of data quality, and the always economic growth on well-being in less-developed countries,
relevant question of scale. Data reduction and variable 1980-2003. Stud Comp Int Dev 42:1–35. doi:10.1007/s12116-
007-9003-7
selection criteria are necessary and incredibly difficult Campbell CC (1991) Food insecurity: A nutritional outcome or a
when the research team takes into account all possible predictor variable? J Nutr 121:408–415
variables that could be used to measure each of the food Drimie S, McLachlan M (2013) Food security in South Africa-first
security pillars one by one—determining which indicator is steps toward a transdisciplinary approach. Food Secur
5:217–226. doi:10.1007/s12571-013-0241-4
the best requires developing protocols and best practices in FAO (2008) An introduction to the basic concepts of food security.
terms of data coverage temporally and spatially. Issues of The Food and Agriculture Organization of the United Nations
data quality go hand in hand with the aforementioned (FAO), Rome
challenges pertaining specifically to data availability and FAO (2015a) The state of food insecurity in the world 2015. Meeting
the 2015 international hunger targets: taking stock of uneven
data sources. Finally, issues of scale are of vital importance progress. Food and Agriculture Organization of the United
with regard to data distillation and in terms of model design Nation, Rome, Italy
and specification and model building and evaluation. Our FAO (2015b) Regional overview of food insecurity Africa, Rome,
results are based on national-level data which are best able Italy
Firebaugh G, Beck FD (1994) Does economic growth benefit the
to point to broad trends in the relations among stability, masses? Growth, dependence, and welfare in the third world.
availability, access, and utilization in explaining food Am Soc Rev 59:631–653
security. As an example, the effect of GDP on food inse- Gettleman J (2017) Drought and war heighten threat of not just 1
curity is at an aggregate scale that makes it an imperfect famine, but 4. New York Times, New York
Hammond RA, Dube L (2012) A systems science perspective and
proxy for measuring the income of individuals within a transdisciplinary models for food and nutrition security. Proc
nation, region, or household. Macro-level analyses are not Natl Acad Sci 109:12356–12363. doi:10.1073/pnas.0913003109
able to answer questions about how these relations play out Hoddinott J, Yohannes Y, Food and Nutrition Technical Assistance
within nations and, crucially, within households. Com- (FANTA) Project (2002) Dietary diversity as a household food
security indicator, Washington D.C.
panion analyses that ask similar questions yet at different Jenkins JC, Scanlan SJ (2001) Food security in less developed
scales complement this work by zooming in on these countries, 1970–1990. Am Soc Rev 66:718–744
interrelations on a finer level of granularity. Future schol- Jenkins JC, Scanlan SJ, Peterson L (2007) Military famine, human
arly work on food insecurity will continue to grapple with rights, and child hunger: a cross-national analysis, 1990–2000.
J Confl Resolut 51:823–847. doi:10.1177/0022002707308215
these challenges through combining analysis strategies and Kerr C (1960) Industrialism and industrial man: the problems of labor
via collaboration across disciplinary boundaries. and management in economic growth. Harvard University Press,
Cambridge
Acknowledgements Funding was provided by National Science Paarlberg RL (1999) Politics and food insecurity in Africa. Rev Agric
Foundation (Grant No. SMA-1416730). Econ 21:499–511. doi:10.2307/1349894
Quisumbing AR, Brown LR, Feldstein HS et al (1995) Women: the
key to food security. International Food Policy Research
Institute, Washington, D.C.
References Rostow WW (1959) Stages of economic growth. Econ Hist Rev
12:1–16. doi:10.1111/j.1468-0289.1959.tb01829.x
Anderson SA (1990) Core indicators of nutritional state for difficult- Scanlan SJ, Jenkins JC (2001) Military power and food security: a
to-sample populations. J Nutr 120:1555–1600 cross-national analysis of less-developed countries, 1970–1990.
Int Stud Q 45:159–187

123
Environ Syst Decis

Sen A (1981) Poverty and famines: an essay on entitlement and UNICEF (2017) Sustainable development goals. Goal 2: Zero
deprivation. Oxford University Press, Oxford Hunger, New York
Stamoulis K, Zezza A (2003) A conceptual framework for national West SG, Taylor AB, Wu W (2012) Model fit and model selection in
agricultural, rural development, and food security strategies and structural equation modeling. In: Hoyle RH (ed) Handbook of
policies. The Food and Agriculture Organization of the United structural equation modeling. Guilford Press, New York,
Nations, Agricultural and Development Economics Division, pp 209–231
Rome
UNICEF (1990) A UNICEF policy review: strategy for improved
nutrition of children and women in developing countries. United
Nations Children’s Fund (UNICEF), New York

123