THE MULTIDIMENSIONAL NATURE
OF HOUSEHOLD FOOD INSECURITY IN RURAL ZAMBIA:
A PANEL ANALYSIS
By
Ayala Wineman

A THESIS
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
Agricultural, Food, and Resource Economics – Master of Science
2013

ABSTRACT
THE MULTIDIMENSIONAL NATURE
OF HOUSEHOLD FOOD INSECURITY IN RURAL ZAMBIA:
A PANEL ANALYSIS
By
Ayala Wineman

Food security is recognized as a multifaceted condition of complex causality that is related to,
yet distinct from, poverty and hunger. Given its broad definition, it is no surprise that food
security eludes precise measurement. This study considers there to be three components of
household food security (quantity, quality, and stability), and attempts to address the "conceptto-measurement" gap in food security by building an index that spans these three dimensions. A
panel data set is used for descriptive analysis of food security indicators in rural Zambia in
2000/01, 2003/04, and 2007/08. A multidimensional index of food security for rural Zambia is
then developed using principal component analysis. This composite index is used to explore the
spatial patterns of food security in Zambia over time, to assess correlates of food insecurity, and
to measure the impacts of climate shocks on food security for different types of households,
including female-headed households. Results indicate that both seasonal rainfall and temperature
have a significant impact on a household's food security score, although not for all individual
components of the food security index. The paper concludes with a consideration of the merits
and shortcomings of developing a composite food security index.

ACKNOWLEDGEMENTS
I would like to thank my major adviser, Dr. Eric Crawford, for his guidance on this thesis and for the
support he has offered throughout my graduate studies at Michigan State University. I am also grateful
for the feedback provided by the other members of my thesis committee, Dr. David Mather and Dr.
Robert Richardson. The data set used in this thesis was made available to me by the Food Security
Research Project. Finally, I was privileged to be included in the data collection activities of the 2012
Rural Agricultural Livelihoods Survey in Zambia, an experience which provided me with the necessary
insight to develop this thesis. Thank you to the Food Security Group for giving me that opportunity.

iii

TABLE OF CONTENTS
LIST OF TABLES ................................................................................................................................... vi
LIST OF FIGURES ............................................................................................................................... viii
KEY TO ABBREVIATIONS ................................................................................................................. ix
1.

INTRODUCTION ............................................................................................................................ 1

2.

AGROECOLOGY AND RURAL LIVELIHOODS IN ZAMBIA ............................................... 3

3.

FOOD SECURITY MEASUREMENT .......................................................................................... 6
3.1 Definition and drivers ..................................................................................................................... 6
3.2 The three components: quantity, quality, and stability .............................................................. 7
3.3 Indicators of household food security ......................................................................................... 12
3.4 Construction of a composite index of food security ................................................................... 14

4.

DATA SOURCES ........................................................................................................................... 17

5.

FOOD SECURITY INDICATORS IN ZAMBIA, 2001-2008 .................................................... 19
5.1 Income-based measures of welfare .............................................................................................. 19
5.2 Non-income food security indicators ........................................................................................... 23
5.3 Descriptive statistics...................................................................................................................... 26
5.4 Correlation among indicators ...................................................................................................... 38
5.5 Food security profiles ................................................................................................................... 40

6.

THE MULTIDIMENSIONAL FOOD SECURITY INDICES .................................................. 42
6.1 Construction of the indices........................................................................................................... 42
6.2 Composition of the indices ........................................................................................................... 44
6.3 Suitability and factor loadings ..................................................................................................... 46
6.4 Descriptive statistics...................................................................................................................... 49

7.

APPLICATIONS OF THE FOOD SECURITY INDICES ........................................................ 61
7.1 Correlates of food sufficiency and food security (cross-sectional methods) ............................ 61
7.2 Correlates of food sufficiency and food security (panel methods) ........................................... 69
7.3 Avenue of climate impacts on food security ............................................................................... 77
iv

8.

DISCUSSION & EVALUATION OF THE FOOD SECURITY INDICES.............................. 80

APPENDIX .............................................................................................................................................. 91
REFERENCES...................................................................................................................................... 102

v

LIST OF TABLES

Table 5-A: Poverty indicators from 2001 to 2008 ................................................................................ 20
Table 5-B: Poverty indicators across subgroups in 2008 .................................................................... 20
Table 5-C: Transition matrix of household income per capita for 2004 and 2008 ........................... 23
Table 5-D: Descriptive statistics of various food security indicators over time ................................ 26
Table 5-E: Food security indicators by gender of household head in 2008 ....................................... 31
Table 5-F: Food security indicators in 2008, by agro-ecological zone ............................................... 32
Table 5-G: Transition matrix of calories/adult equivalent/day for 2004 and 2008 .......................... 34
Table 5-H: Location of calorie-deficient households in 2008 .............................................................. 36
Table 5-I: Correlation matrix of food security indicators in 2008 ..................................................... 39
Table 5-J: Food security profiles in 2008.............................................................................................. 41
Table 6-A: Elements of the food sufficiency and food security indices.............................................. 45
Table 6-B: Factor loadings for multi-year indices: FSU and FSI ...................................................... 47
Table 6-C: Factor loadings for 2008: FSU and FSI ............................................................................. 48
Table 6-D: FSU and FSI in 2008............................................................................................................ 50
Table 6-E: FSU and FSI across years ................................................................................................... 53
Table 6-F: Transition matrix of food sufficiency score from 2004 to 2008 ....................................... 55
Table 6-G: Transition matrix of food security score from 2004 to 2008 ........................................... 55
Table 6-H: Income and food security indicators, 2004 and 2008 ....................................................... 56
Table 6-I: Food security indicators in 2004 and 2008 ......................................................................... 56
Table 6-J: Type I and II errors when income and FSI are used as targeting criteria ..................... 57
Table 6-K: Persistence of food security from 2001 to 2008 ................................................................. 57
Table 6-L: Proportion of households by FSU and FSI trajectory ...................................................... 58
vi

Table 6-M: OLS regression of change in FSU and previous FSU ...................................................... 60
Table 7-A: Logit and OLS regressions of household food sufficiency/ FSU score in 2008 .............. 63
Table 7-B: Determinants of being always food secure or insecure, 2001 .......................................... 67
Table 7-C: Determinants of FSU and FSI with HH fixed effects: 2001, 2004, and 2008 ................. 72
Table 7-D: Effect of lagged FSU and FSI with HH fixed effects: 2001, 2004, and 2008 .................. 76
Table 7-E: Impact of climate shocks by poverty level: 2001, 2004, and 2008 ................................... 77
Table 7-F: Effect of climate on individual elements of the FSU: 2001, 2004, and 2008 ................... 79
Table 9-A: Cost of monthly food basket and basic needs basket per adult equivalent by province 91
Table 9-B: Income indicator definitions ............................................................................................... 92
Table 9-C: Food security indicator definitions .................................................................................... 92
Table 9-D: Other independent variable definitions ............................................................................. 95
Table 9-E: Rainfall and temperature variables ................................................................................... 97
Table 9-F: Calorie content of common foods consumed in Zambia .................................................. 97
Table 9-G: Demographic and agricultural variables for 2001, 2004, and 2008 ................................ 98
Table 9-H: Demographic and agricultural variables across subgroups in 2008............................. 100
Table 9-I: Example of food sufficiency score calculation .................................................................. 101

vii

LIST OF FIGURES

Figure 3-1: The four pillars of food security ........................................................................................ 10
Figure 3-2: An alternate framework: four domains of food security ................................................ 11
Figure 3-3: Household food security framework in this thesis ........................................................... 12
Figure 5-1: Distribution of per capita income in 2008......................................................................... 21
Figure 5-2: Distribution of per capita income in 2008 among MHHs and FHH widows................. 21
Figure 5-3: Distribution of income per capita over the panel years................................................... 22
Figure 5-4: Distribution of calories/ae/day for 2001, 2004, and 2008 ................................................ 28
Figure 5-5: Distribution of calories by gender of household head ..................................................... 32
Figure 5-6: Distribution of months without food stocks in 2008 by agro-ecological region ............ 35
Figure 5-7: CDF of calories/adult equivalent/day by agro-ecological zone ....................................... 37
Figure 5-8: CDF of proportion of calorie requirements met, by agro-ecological zone .................... 37
Figure 6-1: Radar graph of food security indicators for 2008 ............................................................ 51
Figure 6-2: Distribution of FSU and FSI across years ........................................................................ 52
Figure 6-3: Geographic pattern of average district FSU for 2001, 2004, and 2008 .......................... 54
Figure 6-4: FSU and change in FSU, 2001 to 2004 and 2004 to 2008 ................................................ 59
Figure 7-1: Seasonal weather and food sufficiency .............................................................................. 74
Figure 9-1: Agro-ecological zones in Zambia ....................................................................................... 91
Figure 9-2: Timeline of food security indicators .................................................................................. 95

viii

KEY TO ABBREVIATIONS

AE

Adult equivalent

AEZ

Agro-ecological zone

CSO

Central Statistical Office

FAO

Food and Agriculture Organization

FEWS NET

Famine Early Warning Systems Network

FFV

Fresh fruits and vegetables

FHH

Female-headed household

FSU

Food sufficiency index

FSI

Food security index

FSRP

Food Security Research Project

GDP

Gross domestic product

GRZ

Government of the Republic of Zambia

ha

Hectare

HH

Household

kcal

Kilocalorie

KMO

Kaiser-Mayer-Olkin measure of sampling adequacy

LCMS

Living Conditions Monitoring Survey

MACO

Ministry of Agriculture and Cooperatives

MHH

Male-headed household

TLU

Tropical livestock unit

WFP

World Food Program

ZMK

Zambian kwacha

ix

1. INTRODUCTION

This thesis will analyze a longitudinal data set of households in rural Zambia in order to measure
household food security, inclusive of vulnerability to future food shortfalls. Accurate measurement
drives the diagnosis of food insecurity, the exploration of its determinants, and the design of effective
policies to bolster household welfare. Measurement is necessary to understand whether a situation is
getting better or worse and whether food insecurity is chronic or transitory. It reveals information about
who is food insecure in a population and where insecurity is concentrated. Measurement, if done well,
can also bring to light the nature of food insecurity, detailing whether it is a problem of food availability
or economic access, of diet quantity or quality. It is a prerequisite for any analysis of the causes of food
insecurity in a population, which in turn allows for appropriate interventions to address insecurity.
Barrett (2010) further stresses the importance of developing measures that account for the multiple
dimensions of food security and exhibit predictive power.

This study aims to broaden our understanding of food security measurement and dynamics, first by
analyzing the relationships among various food security indicators and tracking these individual
measures over time. It then develops a relatively simple measure of food security in the form of a single
composite index that incorporates indicators of its multiple dimensions. This composite index is applied
to household survey data from Zambia to address the following questions:
•

Where are food insecure households found within Zambia?

•

What is the nature of the food insecurity problem?

•

How has food insecurity changed over time?

•

To what extent does this food security index correlate with money-metric indicators of welfare?

1

•

What are the determinants of food insecurity?

•

Through what avenues do rainfall shocks affect food security?

Few other existing studies use a multidimensional measure of food security to address these kinds of
questions, and to the author's knowledge, no other study thoughtfully reflects on the merits and
shortcomings of using such a composite index. This thesis seeks to fill that gap in the literature.

Chapters 2 and 3 will present background information on agro-ecology and livelihoods in Zambia and a
literature review on food security concepts and measurement. Chapter 4 will describe the data sources
used in this study. Chapter 5 will provide descriptive statistics of various poverty measures and food
security indicators in Zambia, disaggregated by region and gender of household head. Chapter 6 will
detail the construction of a food sufficiency index and food security index using principal component
analysis and will include descriptive statistics of the indices. Chapter 7 will present several applications
of the indices, including econometric analyses of the determinants of food security and of the persistence
of food insecurity. A separate analysis will examine the impact of rainfall shocks on food security.
Chapter 8 will offer conclusions and a summary of lessons learned from the application of the composite
indices of food security developed in this paper.

2

2. AGROECOLOGY AND RURAL LIVELIHOODS IN ZAMBIA

Zambia is a landlocked country characterized by low population density, and it is divided into 9
1

provinces and 72 districts. Although it is among the most urbanized of sub-Saharan African countries,
roughly 45 percent of the population live in rural areas and depend on agriculture for their livelihoods
(Jain 2006).

Zambia's approximately one million farmers can be grouped into three categories: small-scale farmers
(with up to five hectares of farming land), emergent farmers (with 5 to 20 hectares), and large-scale
commercial farmers (with over 20 hectares). As of 2003, approximately 94 percent were small-scale
farmers, 5.5 percent were emergent farmers, and commercial farmers accounted for less than 0.5 percent
(Siegel and Alwang 2005). While most large-scale commercial farmers are located near urban centers,
most small-scale farmers are rural, and about 95 percent of rural households are engaged in crop
production. There is one main cropping season, and among small-scale farmers, crop production
depends almost entirely on rainfall (Jain 2006).

Zambia is divided into four agro-ecological zones distinguished by divergent rainfall patterns (Figure 91). Zone I, located in the south, is relatively dry with unpredictable and poorly distributed rainfall and
limited potential for crop production. This zone also experiences high temperatures during the growing
season, which limits the range of potential crops (Kambikambi 2006). Zone IIa covers the centraleastern part of the country and has the highest agricultural potential, with fertile soil and rain that is
evenly distributed throughout the growing season. Zone IIb is characterized by lower rainfall, sandier

1

In 2011, this was changed to 10 provinces.
3

soils, and high risk of drought. Finally, zone III in the north experiences high rainfall, although this
pattern has produced leached, acidic, and unfertile soils (Jain 2006).

The farming systems and crops vary by agro-ecological region: Zone I is dominated by subsistence crop
production, largely using family labor. The region is suitable for production of drought tolerant crops
(e.g. cotton, sesame, sorghum, millet), though poor soils and unreliable rainfall make farming risky.
Alternative livelihood activities include goat rearing, mat and basket making, and fishing (Siegel 2008).

Zone IIa is characterized by a maize regime and is the most mechanized and commercialized region.
Cash crops, including cotton, irrigated wheat and soybeans are also grown here. Much of this zone is
located along the north-south "line-of-rail", and it is the most populous zone with over four million
inhabitants. This area has better access to infrastructure, higher use of inputs, and a higher share of
medium- and large-sized farms using improved technologies (Siegel and Alwang 2005). The existence
of medium- and larger-scale farms in the area is beneficial to smallholders through labor linkages and
the existence of outgrower schemes, and also through demonstration effects. In contrast, Zones I and III
are characterized by a virtual absence of medium- and larger-scale commercial farms (Siegel 2008). In
sharp contrast to zone IIa, zone IIb exhibits substantially lower capacity for crop production, with
Kalahari sands covering the degraded soils.

Zone III is planted with a cassava and maize regime and is dominated by subsistence production. The
rural areas of this region have the lowest population density in Zambia, and farmers use very low-input
shifting and semi-permanent cultivation techniques (Saasa et al. 1999). Chitemene (slash-and-burn) and
fundakila (using decomposing plant matter to improve soil fertility) are two widely used, traditional

4

methods of cultivation. In zone III, farmers rely on hand hoes rather than oxen, and crops have been
grown mainly from local varieties rather than improved cultivars (Kambikambi 2006).

The year-to-year variability in rainfall and other climatic conditions is important in determining crop
output in most parts of Zambia. Drought has been the biggest shock to food security in the country
during the last two decades (Muchinda 2001, cited in Jain 2006), with large shortfalls in maize yield
consistently occurring in seasons with below normal rainfall. In some years the yield has been only 40
percent of the long-term average, owing to long dry spells within the growing season and the shorter
rainy seasons which have become more common over the past several decades. Droughts in the 1990s
disastrously reduced crop yields, particularly in the south, and extreme climate events have also resulted
in substantial losses of livestock and fertile soil (Jain 2006). At the same time, Zambia sometimes
experiences heavy localized floods that also threaten agricultural production. The general climate
outlook for southern Africa will be characterized by a decrease in precipitation, a rise in temperatures,
and a higher frequency and severity of extreme weather events (Kotir 2011). Thus, the general
consensus among climatologists is that climate change will act as a multiplier of existing threats to food
security in southern Africa.

5

3. FOOD SECURITY MEASUREMENT

3.1 Definition and drivers
The definition of food security is generally understood as a situation whereby "all people, at all times,
have physical, social and economic access to sufficient, safe and nutritious food to meet their dietary
needs and food preferences for an active and healthy life" (FAO 2002). Conversely, food insecurity
exists when people do not have such access, and households experience food insecurity whenever they
are unable to absorb, reduce or mitigate the impact of a negative food shock (Misselhorn 2005; Webb et
al. 2006). With this understanding, food insecurity is not the result of an agricultural failure to produce
sufficient food; rather, it is a failure of local livelihoods to guarantee food access (Devereux and
Maxwell 2001). Food insecurity is related to, yet distinct from, concepts such as poverty and
malnutrition (Webb et al. 2006), and is experienced at a range of spatial scales from households to
regions, as well as a range of time scales.

The failure of early attempts to alleviate food insecurity has arguably been due to their overwhelming
emphasis on food availability and the primacy of agricultural solutions (Ziervogel et al. 2006). While
food security clearly depends on agro-climatic conditions and aggregate food production, it also depends
on socio-economic conditions, including the distribution, access, and affordability of food. People in
rural areas depend on agriculture and natural resources to generate cash income as well as food for home
consumption, and food production affects both supply-side and demand-side indicators of food security.
A meta-analysis of food insecurity in southern Africa suggests that future determinants of food security
lie primarily beyond agricultural production (Misselhorn 2005). An exclusive focus on improving crop
yields necessarily neglects the interrelated economic (e.g. poverty, unemployment, and market failures)
and socio-political factors (e.g. conflict, poor governance, and a high disease burden) that create

6

increasingly vulnerable and unstable communities in the region (Kotir 2011; Misselhorn 2005). Thus,
measuring any change in food security and identifying its cause is a complex task (Maredia 2009).

It is important to note that aggregate food availability is a poor predictor of other food insecurity
indicators, and food insecurity "does not arise exclusively –or even predominantly– because of covariate
shocks to an entire population" (Barrett 2002). Furthermore, research that calculates average food
availability from national food production and imports cannot shed light on how this food is distributed
among communities, households, and individuals (Misselhorn 2005). Rather, the correlates and causes
of food insecurity are likely to be found at the level of households and individual livelihoods.

3.2 The three components: quantity, quality, and stability
This paper considers there to be 3 components of household food security: quantity, quality, and the
stability of the first two components. However, this conceptual framework deviates a bit from some
more conventional frameworks. This section will introduce the more common conceptualizations of
food security and will explain the rationale for this divergence.

Most works consider there to be three interlinked components embedded in the definition of food
security: (1) the availability of food in terms of its physical presence in a given country/household; (2)
the access to food as reflected by people's ability to obtain food from own stock/home production, or
through market purchases, gifts or borrowing; and (3) the utilization of food, in terms of the ability to
derive full biological benefits from food, based on food safety and nutritional/ socio-cultural value
(Thompson et al. 2010).

7

Food availability is a function of the combination of domestic food production and food stocks,
commercial food imports, and food aid, as well as the determinants of these factors. The term
'availability' is usually used in reference to food supplies at the regional or national level (Riely et al.
1999). Access depends on the range of food choices open to people, given their income, market prices,
market accessibility, employment, distribution of wealth, and formal or informal safety net
arrangements. Access rests on having adequate resources, or entitlements, to acquire a sufficient
quantity of food, and while consumers' purchasing power in the form of real incomes and food prices is
important, entitlements are not necessarily monetary. Instead, they include all commodity bundles
within a person's control, given the legal, political, economic, and social arrangements of the
community. Thus, they may also include traditional rights, such as a share of common resources
(Morton 2007). Whereas availability reflects the supply side of food security, access reflects the
demand side and in practice typically results in inequality of inter- and intra-household food distribution.
Access is directly compromised by adverse shocks such as unemployment spells, price spikes, or the
loss of livelihood-producing assets. The access lens is therefore able to highlight the close relationship
between food security, poverty, and social, economic, and political marginalization (Barrett 2010).
However, the multidimensional nature of individual or household access makes it complicated to
measure. Access rests on a wide variety of activities, and the lack of a 'typical' set of activities makes it
difficult to define a universal set of indicators (Swindale and Bilinsky 2006).

Utilization is related to health and reflects concerns about whether individuals make good use of the
food to which they have access (Barrett 2010). There are two forms of food utilization: physical and
biological utilization. Physical utilization reflects the level of a household's physical means to safely use
the food available, and depends on the sanitary conditions along the entire food chain (Morton 2007).

8

This may include adequate housing, access to potable water, cooking utensils, cultural feeding
hierarchies, family structure, and caretaker behavior, knowledge, and workload. Biological utilization is
a measure of the body's ability to effectively use the nutrients consumed, and it depends on hygiene,
infestation (e.g., hookworms), increased nutrient demand resulting from infection, and dietary quality
(Renzaho and Mellor 2009). Particular attention is given to micronutrient deficiencies, and utilization
encompasses both food consumption in the short term (e.g., breast-feeding, food intake, food habits and
practices) and nutritional status parameters in the long run.

Availability, access, and utilization, also referred to as the three 'pillars' of food security, are inherently
hierarchical, with availability as necessary but not sufficient to ensure access, and access as necessary
but not sufficient for effective utilization. While local, regional, and national food production are
extremely important determinants of food security, the inability of poor households to secure food
through markets and non-market channels may limit their food security, even when food is abundant
(Barrett 2010). Of course, this conceptualization includes several feedback loops, with appropriate
utilization necessary for achieving adequate access (via health and other human capital effects), and
access necessary for maintaining food availability (through enhanced labor productivity and the
avoidance of resource depletion). These feedback loops mean that there are both upstream and
downstream linkages associated with food security interventions (Webb and Rogers 2003).

Researchers and development agencies increasingly identify a fourth conceptual pillar of food security
that is alternately labeled as stability or vulnerability (Figure 3-1; adapted from Gregory et al (2005) and
expanded)). At the household level, vulnerability can be thought of as the likelihood that at a given time
in the future, a household will have a level of welfare (often expressed in terms of income or

9

consumption) below some benchmark, such as the expenditure required to meet the minimum caloric
requirement per capita per day (Hoddinott and Quisumbing 2003). This concept is forward-looking and
related to expectations and uncertainty. The definition of food security includes a consideration of its
temporal dynamics by noting that security must exist 'at all times', and the term 'food security' connotes
certainty of future well-being. Along these lines, food insecurity reflects uncertain access to enough and
appropriate foods, and is therefore a dynamic problem that can be best conceptualized as an ex ante
status rather than a purely ex post outcome (Barrett 2002). The definition of food security means that
either a currently inadequate diet or a high probability of an inadequate diet in the future both render a
person food insecure (Christiaensen and Boisvert 2000). According to Barrett (2002), a useful
conception of food security must consider changes over time and people’s perceptions of and responses
to these changes (e.g. consumption smoothing), and must reflect uncertainty and risk. Without viable
expectations of the other three dimensions of food security, a household is subject to uncertainty that
affects all of its investment and disinvestment decisions.

Figure 3-1: The four pillars of food security

Note: For interpretation of the references to color in this and all other figures, the reader is referred to
the electronic version of this thesis.

10

It should be noted that this widely used framework is not the only way that food security has been
conceptualized. A similar food security conceptual framework includes four slightly different aspects,
as outlined in Figure 3-2 (adapted from Campbell (1991)). This framework is often used to describe how
food insecurity is experienced and managed at the household or individual level, where availability and
access cannot readily be distinguished. The first two domains capture the physiological adequacy of
food, while the next two are psychological and social in nature. Note that the health and sanitation
components of utilization are absent here.

Figure 3-2: An alternate framework: four domains of food security
Due to the nature of the household-level data available for this paper’s application, this paper will adopt
a third framework that combines aspects of Figures 3-1 and 3-2. It will regard food security as
comprised of 3 components: food quantity available in the household, food quality as captured by diet
diversity and the presence of important nutrients, and the stability of adequate food supplies (Figure 33). It should be noted that this is not intended to be the most comprehensive framework, but rather the
one most appropriate for the present analysis.

11

Figure 3-3: Household food security framework in this thesis
3.3 Indicators of household food security
The accurate measurement of food security is necessary to recognize crises as they occur, to target
resources toward those most in need or at risk of sliding into hunger, to track the impact of food
security-focused interventions, and to quantitatively analyze the dynamics of food security.
Policymakers need to know how many people are at risk of food insecurity, who and where they are, and
how best to reach them. While accurate measurement of household food security is essential for
effective research and well-targeted policies and programs, there is no standard methodology for
measuring food security, and despite an improved theoretical understanding of food security, the FAO
notes that there exists no "perfect single measure that captures all aspects of food insecurity" (FAO
2002). Food security is an unobservable, cross-cutting, multifaceted process of complex causality, and
there are considerable conceptual and measurement problems associated with estimating the incidence
and intensity of food insecurity (Barrett 2002). For this reason, it is rare for impact assessments of food
security interventions or agricultural development programs to include an indicator explicitly called
'food security' (Maredia 2009).
12

Commonly used indicators of household-level food availability may include variables such as household
crop or food production (e.g., average cereal yields per hectare or food production per capita) or
livestock ownership (Renzaho and Mellor 2009). Although there are no exact indicators of access
failure, commonly used proxy measures include household income relative to food prices, food
expenditures, and per capita food, calorie and nutritional intake. However, each of these proxies is a
partial measure of the multifaceted phenomenon of food access (Webb et al.2006), and the pathways or
mechanisms through which these indicators affect food security are implied rather than empirically
determined and calculated (Maredia 2009; Webb et al. 2006). A common indicator of utilization is the
growth/ nutritional status of children under five years of age as measured by height and weight (Webb et
al. 2006), or ‘hygiene’ measures, such as type of latrine or source of water (Coates et al. 2003). Readers
should note that in this study, indicators will be classified a bit differently, as indicated in Figure 3-3.

While most measures reflect past or current levels of food security (e.g. past or current food production
and income; current expenditures and dietary diversity; or current health and nutritional status), ideal
food security indicators would also reflect stability/vulnerability, or the probability of adequate
consumption in the future. Vulnerability is essentially forward-looking, and is determined by the risks a
household or individual faces in making a living, the livelihood options available, and the ability to
handle this risk (Alinovi et al. 2009). Indicators of vulnerability include a household's asset stock as a
measure of wealth, land owned as a measure of production security, and an index of income
diversification (Coates et al. 2003). A simple access-based indicator of vulnerability is how much
'buffer' there is in current income for accommodating higher food prices. A food security indicator
concerned with caloric sufficiency but not the proportion of income spent on food may misclassify a

13

household as food secure (Maxwell et al. 1999; Løvendal and Knowles 2007). Other indicators include
access to assets and asset liquidity, and crop and income diversification. Measures of stability should
account for the choices and tradeoffs households face when allocating their resources over time to
balance current access and jeopardizing future food consumption (Wolfe and Frongillo 2000).

Though many of these indicators are similar, they are also somewhat unique in what they are measuring
(Barrett 2010), and different indicators capture different dimensions of food security. The FAO
concludes that "no individual measure suffices to capture all aspects of food insecurity" (FAO/FIVIMS
2003), and the absence of such a 'gold standard' measure of food insecurity makes it unreasonable to use
a single benchmark to proxy a (non-existent) gold standard. In light of the multidimensional nature of
food security, it is generally agreed that a suite of indicators and methods are needed for its assessment
(CFS 2011). These different measures are needed for complementary analysis and the triangulation or
cross-referencing of indicators (a convergence of evidence approach) (Coates 2003; Maxwell et al.
1999; Migotto et al. 2005). In particular, conventional measures based only on household expenditures
are considered to miss many aspects of food security, and a non-monetary approach illuminates some of
the deeper mechanisms of food insecurity (e.g. vulnerability, persistence) that are often overlooked by
conventional, money-metric analysis (Dasgupta and Baschieri 2010).

3.4 Construction of a composite index of food security
Several papers have combined various food security indicators into a single composite index. This
seems to build on the literature of multidimensional poverty indices (e.g. Alkire and Foster 2007), and it
allows for the ranking of different countries, regions, populations, or households in terms of the severity
of food insecurity. Such an exercise may be useful in order to target resources and policies toward those

14

most in need and to track changes over time. An index may be constructed in five steps: (1)
identification of the dimensions of food security, (2) selection of indicators for each dimension, (3)
standardization of indicators, (4) weighting of indicators, and (5) ranking of units (e.g. households or
countries) according to summed scores. When building such an index, the main challenges are to select
the most appropriate indicators and to calculate the weights attached to each component in order to best
represent their values in practice.

The WFP's Vulnerability Analysis and Mapping Unit applies Principal Component Analysis (PCA) to
generate a food security and vulnerability index for household profiling (WFP 2009). PCA is a
statistical technique that can reduce the dimensions found among multiple variables into a single
measure that captures as much as possible of the variation in the original data set (Abeyasekara 2005).
Thus, from a handful of food security indicators, PCA can identify a latent food security variable and
can assign a food security score to each household. The Famine Early Warning System Network
(FEWSNET) outlines a method to construct a household index of chronic vulnerability. Elements in the
index may include drought risk, the diversity of income portfolios, and reliance on export crops. These
variables are each converted into comparable units (z-scores), weighted, and summed to generate a
simple ranking of households (FEWS 1999). Demeke et al. (2011) use PCA to create a food security
index for Ethiopia that includes several household-level variables related to food availability/access and
vulnerability. The authors find the index to be highly correlated with factors such as per capita
consumption expenditures and value of food consumption per month, which seems to validate the index
as a reliable measure of household food security. Alinovi et al. (2009) also use PCA to build an index of
resilience to food insecurity, considering resilience to be a latent variable comprised of income and food

15

access, assets, access to public services, and social safety nets, as well as two cross-cutting dimensions:
stability and adaptive capacity.

Although these studies each build an index that measures either food security or one of its aspects, none
includes a discussion of the merits and limitations of doing so. While the argument for creating an index
is compelling, there may be drawbacks to collecting diverse factors into one score, particularly when
trying to understand the determinants of food security and the channels of impact. For example,
Ravallion (2011) notes that while he does not deny that poverty is multidimensional, he is skeptical
about the value of a single (uni-dimensional) index for sound development policy making in practice.
His reasoning is that, in practice, policymakers can already refer to multiple indicators to measure
different aspects of poverty, but the use of a single index measure necessarily relies on assumptions
regarding which aspects of poverty should have been included in the measure and how they should be
weighted. The same argument can be levied at the construction of a multidimensional food security
index. While an acknowledgement of food security's multiple dimensions merits consideration, it is not
clear that collapsing these dimensions into a single composite index can be useful. This thesis will
explore this very question.

16

4. DATA SOURCES

This analysis uses nationally representative panel data on rural farm households in Zambia. Households
were surveyed in 2000/01, 2004, and 2008 and asked about their activities and income over the previous
12 months, as well as household changes over the previous 4 years. Hence the surveys refer to the
1999/2000, 2002/03, and 2006/07 agricultural years, and the 2000/01, 2003/04, and 2007/08 marketing
years. The first wave is comprised of both the 1999/2000 Post-Harvest Survey conducted by the
Zambian Central Statistical Office (CSO) and Ministry of Agriculture and Cooperatives (MACO), and
the CSO/MACO/Michigan State University Food Security Research Project (FSRP) Supplemental
Survey. The second and third waves are Supplemental Surveys. 6,922 households were interviewed in
2001; 5,419 of these households were re-interviewed in 2004; and 8,094 households were included in
the 2008 survey, including both panel and non-panel households. In total, 4,286 households were reinterviewed in all three waves of the panel survey.

Data on the calorie content of food items are taken from the Food Consumption Table for Use in Africa
(FAO 1968) and the Tanzania food composition table (Lukmanji et al. 2008). Information on the
contents and provincial prices of a basic needs basket in Zambia is drawn from the Zambia Living
Conditions Monitoring Survey (LCMS) (CSO 2004). Data on historical prices of food items at the
provincial level are taken from the FAO Global Information and Early Warning System (FAO 2013).
Historical rainfall and average temperature data are obtained from records collected by 35
meteorological stations run by the Zambian Meteorological Department.

In this paper, panel analyses include the balanced panel of households interviewed in all three waves,
while cross-sectional analyses include all households interviewed in a given year. Population or panel
weights are included in all relevant analyses. In panel regressions where a test for attrition bias rejects
17

the null hypothesis of no bias, attrition-adjusted weights created with the inverse probability weighting
procedure (Woodridge 2002) are used. Monetary values are inflated to 2007/08 values using the
consumer price index, and the exchange rate for this year was 3,829 ZMK = 1 U.S. dollar. No effort
was made to impute missing data points.

18

5. FOOD SECURITY INDICATORS IN ZAMBIA, 2001-2008

5.1 Income-based measures of welfare
In many studies, income is used as a general welfare measure or as an indicator of the 'access' dimension
of food security. Here, income is calculated as the sum of the value of agricultural production (retained
or sold), income from the sale of live or slaughtered animals, any salary or wages earned, whether in
cash or in kind, any gifts or remittances received by the household, and the net revenue from household
own-business activities. Given data limitations, income here is calculated differently in longitudinal
analyses, where only information captured in all three years is used, and in cross-sectional analyses for
2008, where all information captured in this last year is used.

In this paper, the poverty rate is conceived in two ways: (1) The dollar-per-day poverty rate is defined
as the proportion of households whose annual per capita income is below one U.S. dollar per day. This
threshold is used because it was the World Bank international poverty line until 2008 (Chen et al. 2008),
while most of the data were collected before this date. (2) The moderate poverty rate is defined as the
proportion of households whose average monthly income is below the cost of a basic needs basket (in
per adult equivalent terms), or the collection of food and non-food items required to purchase or acquire
a minimum standard of caloric intake and essential non-food items. The extreme poverty rate is defined
as the proportion of households unable to afford even the minimum monthly per adult equivalent food
basket. The cost of these baskets in each province was obtained from the 2002-03 Living Conditions
Monitoring Survey (CSO 2004; Table 9-A), and the definition of extreme poverty is similarly borrowed
from this report.

Several definitions are also employed for female-headed households (FHHs), where "FHH widow"
captures households that self-identify as both female-headed and widowed. "FHH nm" stands for FHH
19

in which there is no prime-age (15 – 59 years) male member of the household, and the rationale here is
that these households would be most vulnerable to gender discrimination, whereas FHHs that contain
adult men might be better placed to access the same resources as male-headed households (MHHs).
Note that these definitions of FHH are not mutually exclusive. Definitions of all variables included in
this study are provided in the appendix (Tables 9-B through 9-E), and population sampling weights are
used for all descriptive statistics in this chapter. Throughout this report, binary variables will be denoted
with 1=, where a value of 1 indicates “yes” or “true”, and a value of 0 indicates otherwise. A proportion
of 0.75 means that 75 percent of observations have a value of 1.

Table 5-A: Poverty indicators from 2001 to 2008
Indicator
1=HH below $1-per-day poverty line
1=HH unable to afford basic needs basket
1=HH unable to afford food basket
Median net income per capita (ZMK)

2001
0.91
0.93
0.88
289,030

2004
0.89
0.89
0.83
373,682

2008
0.88
0.89
0.82
366,667

Table 5-B: Poverty indicators across subgroups in 2008
Agro-ecological zones
Indicator
1=HH below $1per-day poverty
line
1=HH unable to
afford basic needs
basket
1=HH unable to
afford food basket
Median net income
per capita (ZMK)

All
households

FHH
widow

FHH
nm

|t|

0.87

0.91

0.89

0.88

0.91

0.81
384,000

I

IIa

IIb

III

3.78

0.91

0.86

0.91

0.87

0.90

3.82

0.93

0.87

0.91

0.88

0.86

0.83

3.93

0.87

0.79

0.84

0.82

321,000

349,027

3.29

288,419

433,483

337,894

364,518

These figures illustrate that the poverty rate is sensitive to the definition of the poverty cutoff. The tstatistics in the fourth column of Table 5-B compare the mean values for widowed FHHs and all other
20

households. However, using any of these poverty cutoffs, FHHs exhibit a higher rate of poverty. The
actual distributions of per capita income are shown in Figures 5-1 and 5-2. Figure 5-3 shows that this

0

5.0e-07

Density

1.0e-06

1.5e-06

2.0e-06

distribution has changed over 2001, 2004, and 2008, although there is not a consistent trend over time.

0

1000000
1500000
500000
Income per capita (ZMK)

2000000

0

Density
5.0e-07
1.5e-06
1.0e-06
2.0e-06

Figure 5-1: Distribution of per capita income in 2008

0

1000000
1500000
500000
Income per capita (ZMK)
MHH

2000000

FHH

Figure 5-2: Distribution of per capita income in 2008 among MHHs and FHH widows
Note: In almost all figures, outliers are omitted for clarity of visual presentation. An outlier can
therefore be defined as an observation that lies beyond the range displayed in each figure.

21

Density
5.0e-07
1.5e-06
1.0e-06
2.0e-06
0

0

1000000
500000
1500000
Income per capita (ZMK)
2008
2001

2000000

2004

Figure 5-3: Distribution of income per capita over the panel years
In terms of average per capita income, Zambia experienced a significant improvement from 2001 to
2004 (t-stat = 6.73), but no significant difference between 2004 and 2008 (t-stat = 0.61).

It should be noted that the appropriateness of an income flow measure of household wealth has been
called into question. This is because a flow-based measure of poverty can be extremely stochastic due to
random price and yield fluctuations and irregular earnings from remittances and gifts. Another critique
of survey measures of household income is that measurement error is expected to be relatively high for a
single-visit survey that elicits recall data on 12 months of agricultural production and non-farm income.
In this study, the value of assets (usually considered a more stable indicator of household welfare) will
be considered an indicator of food security vulnerability, rather than a proxy for wealth. Keep in mind
that this does not capture the household’s stock of human capital, and is therefore still an imperfect
measure of vulnerability.

22

A transition matrix of income for 2004 and 2008 (Table 5-C) shows that households do not readily
experience large changes in income status. The income brackets were formed as quintiles in 2004, so
that category 1 represents the poorest 20 percent of households in that year, and the same cutoffs were
maintained in 2008. Of households that started out in the poorest category, 35 percent remained in that
income bracket four years later, while 10 percent moved up to the highest bracket.

Table 5-C: Transition matrix of household income per capita for 2004 and 2008
Income status 04
1
2
3
4
5

1
0.35
0.28
0.21
0.17
0.09

Income status 08
2
3
0.21
0.20
0.25
0.20
0.21
0.21
0.17
0.20
0.13
0.16

4
0.14
0.16
0.21
0.23
0.22

5
0.10
0.12
0.15
0.23
0.40

5.2 Non-income food security indicators
The food security indicators included in this thesis are listed in Table 5-D, and definitions are available
in Table 9-C. Panel values are used in all panel analyses, and non-panel values are calculated for
2007/08 and used only in cross-sectional analyses from that year. Where these diverge, the panel
variables consider only the information that is consistently recorded across all years of the survey, while
the non-panel variables often include additional information added in 2008.

Calories are calculated by converting the volume of agricultural products harvested or food items
procured/ received into calorie content, and the conversion values are presented in Table 9-F. It should
be emphasized that these values are a rough estimate, though it is not immediately obvious whether they
are an over- or under-estimate of calories available to a household. It may be an underestimate because
the survey does not include information on food eaten away from home, the collection of wild foods, the
purchase of non-staple food items (such as fruits, vegetables, oils, or animal products), or food obtained
23

in socially unacceptable ways, such as by stealing. At the same time, it may be an overestimate because
the calculation of retained calories does not account for losses in storage, nor does it consider food
provided to guests.

The household status of food-energy deficiency (light) is calculated using the total calories produced or
acquired by the household, relative to age and sex-specific caloric requirements and assuming a light
level of activity (Smith and Subandoro 2007). Accordingly, the statuses of food-energy deficiency
(moderate) and (intense) are calculated using the calorie requirements for moderate and intense levels of
activity. Each household’s total energy requirement is calculated by multiplying the number of
household members in each age-sex category by the corresponding energy requirement, weighting the
individuals by time spent at the homestead, and summing across household members. An additional 500
calories is added for each child younger than one year old at the end of the survey period in order to
account for the higher energy needs of breastfeeding mothers.

The household endowment of livestock reflects what each household owned at the beginning of the
reporting period, 12 months prior to the survey. The calculation of edible tropical livestock units (TLU)
in 2008 excludes work oxen, as this variable is intended to capture the availability of meat products to
the household. The number of crops is a simple sum of the different crops or fruits/vegetables produced
and retained by a household, excluding any cash crops that are not edible. This is a very coarse
indicator of dietary diversity, as it does not account for the inclusion of different food groups in the
household's diet. Because it is a simple sum, it also potentially includes crops that are grown in a very
small amount. The binary variable indicating whether a household participates in a transfer network (i.e.
whether it received or provided any remittances or gifts) is intended as a proxy for social capital,

24

assuming that a household that engages in such exchange has access to a network that can serve as a
social safety net in the event of an idiosyncratic food shortfall. It should be noted that when this variable
is used in a regression (as in Chapter 7), it is highly susceptible to reverse causality.

The proportion of crop value sold is intended to be an indicator of surplus production. This assumption
is not perfectly straightforward, and a household may sell crops and still be a net buyer (Barrett 2008).
The value of productive assets is estimated for each year to reflect what each household owned at the
beginning of the reporting period. This is considered to be an indicator of vulnerability because assets
serve as a safety net during negative shocks and also determine a household’s ability to improve its
situation (Carter and Barrett 2006). For 2008 this includes both the value of farm equipment and work
oxen, while for multi-year analyses it includes only farm equipment (see Table 9-C for details of
variable construction).

As mentioned earlier, it is important to be aware of the limitations of using this survey to measure food
security: The survey only asks about the purchase of staple foods. Yet according to the 2002-03 LCMS,
fruits and vegetables comprise, on average, 20 percent of food purchases among households in Zambia
(CSO 2004). The purchase of cooking oil or meat could be important, yet overlooked, sources of
calories or protein. Although the survey asks about the collection of forest products and wild fish as an
income source, it does not account for a household's consumption of these goods. It also accounts for
food produced/acquired but not food consumed, although losses in storage could drive a sizable wedge
between these two values (World Bank 2011). In terms of the stability dimension of food security, the
survey does not ask about access to credit or degree of debt in a household, although these could have
important implications for a household's vulnerability.

25

Particularly with regard to food purchases and food earned as in-kind wages, the survey responses are
subject to recall error over the course of a 12-month recall period. Most food security measures that try
to capture food consumption do so with much shorter time-frames, such as one day or two weeks, and
this is due to concern over recall errors that could arise with a longer recall period (Beegle et al. 2012).
This study develops a much coarser estimate of food availability over a 12-month time period, with
reference only to large purchases.

5.3 Descriptive statistics
Table 5-D presents descriptive statistics of these food security indicators for each year.

Table 5-D: Descriptive statistics of various food security indicators over time
2001

2004

2008

|t| difference
01/08
04/08

QUANTITY
Total calories acquired/ ae/ day (median)
1=Food energy deficient (light)
1=Food energy deficient (moderate)
1=Food energy deficient (intense)
QUALITY
Tropical livestock units
Number of different field crops retained
1=HH produces eggs or milk

2

2,413.47 2,537.08 2,336.31
0.53
(0.50)
0.61
(0.49)
0.68
(0.47)

0.50
(0.50)
0.58
(0.49)
0.65
(0.48)

1.82
(6.90)
1.91
(1.20)
0.31
(0.46)

2.58
(12.87)
2.32
(1.33)
0.14
(0.35)

Test for difference in mean calories/ae/day
26

0.54
(0.50)
0.61
(0.49)
0.68
(0.47)

0.44

1.93

2

1.07

4.39

0.12

4.16

0.17

4.50

2.35 4.12
(8.42)
1.89 1.13
(1.17)
0.47 20.34
(0.50)

1.25
20.23
43.01

Table 5-D (cont’d)
1=HH produces some vegetables/fruits

0.16
(0.37)

---

0.38 29.08
(0.49)

---

1.82
(4.62)
1.91
(0.84)

2.45
(5.21)
1.85
(0.84)

2.19 4.69
(4.94)
2.18 19.33
(0.85)

2.89

---

1.31

1.84

(2.36)

(2.47)

---

0.60

0.40

0.50
(0.50)
0.15
(0.23)
0.12
0.37

(0.49)
0.64
(0.48)
0.20
(0.25)
0.15
0.40

0.09

0.08

0.12

(0.19)
(0.18)
Hectares cultivated
1.44
1.56
(1.34)
(1.47)
No. observations
6,922
5,419
Values are averages with standard deviations in brackets.

(0.21)
1.42
(1.63)
8,094

STABILITY
Value of productive assets (ln)
Number of income sources
Number of months without any food
stocks (of cropping HHs)
1=Maizecass (HH had food in stock at
end of survey period)
1=HH participates in transfer network
Proportion of crop value sold
No. skilled off-farm workers
Proportion food expenditure relative to
income

22.45

---

12.19

---

23.17

(0.49)
0.83 45.13
(0.38)
0.29 27.58
(0.33)
0.23 14.55
0.48

24.66
18.09
15.26

8.41

10.38

0.79

5.3

3

These results reflect stagnancy in the number of calories per adult equivalent available to households
each day. The first t-statistic is a test of the difference in mean (not median) calories between 2001 and
2008, and it does not indicate a significant trend. The actual distribution of calories across panel years is
presented in Figure 5-4, which illustrate that a higher proportion of households in 2001 and 2008 report
a relatively lower number of calories accessed – though the difference is small. This nonlinear trend in
calories over the panel years mirrors the trend in income (Table 5-A and Figure 5-3), and it is clear that

3

Adult equivalent is a measure that adjusts the size of a household to reflect its caloric consumption
needs based on the age and gender or each individual in the household (WHO 1985).
27

although 2004 was a relatively good year for households in Zambia, by 2008 several food security
indicators had returned to their 2001 levels. In other words, although many of the t-statistics for a
difference in mean values between years are significant, the trend often changes direction between
2001/04 and 2004/08. However, it does seem that households are marketing an increasing proportion of
their crop production, producing more skilled off-farm workers, and collecting income from a greater

0

Density
.0001
.0002

.0003

number of sources.

0

2000
4000
6000
Calories per adult equivalent per day
2001
2008

8000

2004

Figure 5-4: Distribution of calories/ae/day for 2001, 2004, and 2008
The cutoff point for a household being considered food energy deficient is at the age- and genderspecific calorie requirements suggested by Smith and Subandoro (2007) for three different activity
levels. These cutoffs produce markedly divergent rates of food energy deficiency, although with any
cutoff, the proportion of households that exhibit food energy deficiency is lower than those that cannot
afford a basic needs food basket based on income calculations (Table 5-A). This is a noteworthy
finding, as it suggests there may be a problem with how the food basket is priced in the LCMS (CSO
2004), or alternatively a problem with how income is calculated. In this thesis crop income is measured

28

at farmgate prices, whereas the LCMS probably prices its food basket at retail prices. This may suggest
that we are underestimating the real incomes earned by households, resulting in a deceptively high
measure of extreme poverty.

The inclusion of participation in a transfer network as an indicator of stability may appear questionable
as such participation in a given year may be the direct cause or consequence of current food security
status. However, a cursory glance at transfer network patterns suggests that participation alone does not
unambiguously relate to either food sufficiency or insufficiency. Fifteen percent of households
participate only as receivers, 25 percent participate as givers, 43 percent of households both give and
receive within their transfer network, and the remaining 17 percent neither give nor receive in 2008.
Only the last category receives a value of zero for “transfer network participation”, with the intention of
capturing a household’s access to a network rather than specifically whether they are recipients or
benefactors of assistance.

Tables 5-E and 5-F present descriptive statistics for 2008, disaggregated by gender of household head
and agro-ecological zone. Where applicable, the non-panel values are used. In this table, FHHs are
defined alternately as widowed FHHs or FHHs without any prime-age male residing in the household.
Note again that these categories are not mutually exclusive; rather they are both included to explore the
implications of defining FHHs differently. It is quite surprising to observe that these definitions of
FHHs, which produced somewhat poorer households according to income measures, actually display
significantly higher calories per capita and correspondingly lower rates of food energy deficiency. One
possible explanation to reconcile these results with an a priori expectation of lower food security is that
only FHHs that are sufficiently empowered remain as independent households, whereas FHHs that are

29

unable to achieve relative food security in Zambia are absorbed into the households of extended family
(Geisler 1993). This hypothesis seems to be borne out by the 2008 data, in which 10.12 percent of
households contain a woman who is not the head of household and is widowed, separated, or divorced.
Meanwhile, just 2.6 percent of households contain a man in this position. The discrepancy suggests that
the FHHs captured in this household survey may represent the “survivors”, and this pattern underscores
how the categorization of households as male- or female-headed does not truly capture the gender
dynamics of food security. This point will be revisited in Chapter 8.

FHHs do report a higher average number of months without any food stocks and lower measures of
dietary diversity, making their lower rate of calorie deprivation all the more surprising. These FHHs
exhibit lower average values of several other stability indicators: They cultivate smaller areas of land,
have fewer productive assets, and are less likely to participate in a transfer network. Figure 5-5 further
indicates that the significantly higher calories reported by FHHs may be attributed to those households
at the high end of the spectrum, where both categories of FHHs cross the density curve of MHHs. In
other words, a greater proportion of FHHs are doing relatively well, as compared with the general
population.

In the appendix, Tables 9-G and 9-H display the average values, for each year and for sub-groups in
2008, of other demographic and agricultural variables that will be included in later analyses related to
the correlates of food security.

30

Table 5-E: Food security indicators by gender of household head in 2008

QUANTITY
Total calories acquired/ ae/ day
(median)
1=Food energy deficient (light)
1=Food energy deficient (moderate)
1=Food energy deficient (intense)
QUALITY
Tropical livestock units
Number of different crops or
vegetables/fruits retained
1=HH produces eggs or milk
1=HH produces some vegetables/fruits
STABILITY
Value of productive assets (ln)
Number of income sources
Number of months without any food
stocks (of cropping HHs)
1=Maizecass (HH had food in stock at
end of survey period)
1=HH participates in transfer network
Proportion of crop value sold
No. skilled off-farm workers

All HHs

FHH widow

FHH nm

|t| FHH
widow/
others
difference

2,521.40

2,581.34

3,186.06

4.7

0.52
(0.50)
0.59
(0.49)
0.67
(0.47)

0.50
(0.50)
0.56
(0.50)
0.63
(0.48)

0.42
(0.49)
0.47
(0.50)
0.53
(0.50)

1.53

1.88
(6.92)

1.09
(3.64)

0.55
(2.07)

4.15

2.72

2.55

2.32

3.13

(1.97)
0.47
(0.50)
0.38
(0.49)

(1.80)
0.40
(0.49)
0.31
(0.46)

(1.83)
0.33
(0.47)
0.29
(0.45)

8.31
(6.49)
2.18
(0.85)

4.85
(6.42)
1.93
(0.82)

3.14
(5.60)
1.86
(0.76)

19.85

1.84

2.19

2.10

4.97

(2.47)

(2.62)

(2.72)

0.40

0.44

0.42

(0.49)
0.83
(0.38)
0.31
(0.33)
0.23
(0.48)

(0.50)
0.70
(0.46)
0.24
(0.32)
0.09
(0.32)

(0.49)
0.68
(0.47)
0.26
(0.35)
0.08
(0.28)

31

2.23
3.1

4.94
5.1

4.5

4.34
1.83
7.06
10.31

Table 5-E (cont’d)
Proportion food expenditure relative to
income

0.13

0.15

0.15

2.75

0

Density
.0001
.0002

.0003

(0.22)
(0.24)
(0.25)
Land access (ha)
2.65
1.83
1.43
3.22
(9.27)
(4.12)
(3.23)
Proportion of population
1.00
0.14
0.10
No. observations
8,094
1,095
821
The last column refers to the difference in means between all households and widowed FHHs.

0

2000
4000
6000
Calories per adult equivalent per day
MHH
FHH nm

8000

FHH widow

Figure 5-5: Distribution of calories by gender of household head
Table 5-F: Food security indicators in 2008, by agro-ecological zone
Agro-ecological zones
I
IIa
IIb

III

QUANTITY
Total calories acquired/ ae/ day (median)
1=Food energy deficient (light)

2,185.99 2,823.79 1,765.98 2,245.90
0.57
0.44
0.64
0.56
(0.50)
(0.50)
0.48
0.50
0.67
0.52
0.70
0.63
(0.47)
(0.50)
0.46
0.48
0.73
0.62
0.74
0.69
(0.44)
(0.49)
0.44
0.46

1=Food energy deficient (moderate)
1=Food energy deficient (intense)

32

Table 5-F (cont’d)
QUALITY
Tropical livestock units
Number of different field crops retained
1=HH produces eggs or milk
1=HH produces some vegetables/fruits
STABILITY
Value of productive assets (ln)
Number of income sources
Number of months without any food stocks
(of cropping HHs)
1=Maizecass (HH had food in stock at end of
survey period)
1=HH participates in transfer network
Proportion of crop value sold (out of
cropping HHs)
No. skilled off-farm workers
Proportion food expenditure relative to
income
Land access (ha)
Proportion of population
No. observations

3.30
(7.44)
2.25
(1.57)
0.54
(0.50)
0.29
(0.45)

3.25
(9.64)
2.73
(1.86)
0.57
(0.49)
0.45
(0.50)

1.55
(6.30)
1.54
(1.10)
0.30
(0.46)
0.14
(0.35)

0.55
(2.24)
2.99
(2.13)
0.40
(0.49)
0.37
(0.48)

7.93
(7.01)
2.11
(0.84)

9.95
(6.22)
2.19
(0.86)

3.97
(6.22)
2.14
(0.76)

7.71
(6.21)
2.18
(0.85)

2.75

2.42

2.36

1.10

(2.79)

(2.51)

(2.88)

(2.07)

0.22

0.21

0.36

0.60

(0.42)
0.80
(0.40)

(0.41)
0.84
(0.37)

(0.48)
0.78
(0.42)

(0.49)
0.83
(0.38)

0.18

0.33

0.26

0.31

(0.28)
0.18
(0.43)

(0.32)
0.27
(0.51)

(0.38)
0.15
(0.39)

(0.34)
0.22
(0.47)

0.16

0.14

0.16

0.12

(0.26)
2.27
(4.17)
0.06
517

(0.21)
2.96
(11.80)
0.4
3,108

(0.25)
1.46
(1.97)
0.08
655

(0.21)
2.64
(7.90)
0.45
3,798

A transition matrix of household calorie status (Table 5-G) indicates that households move around more
readily with regard to calorie consumption as compared with income (Table 5-A). Again, the categories
of calorie status were formed as quintiles of calories/adult equivalent/day in the year 2004, with these
cutoffs maintained in 2008. Thus, of households in the lowest calorie bracket in 2004, only 26 percent
33

remained in that bracket four years later. This is somewhat surprising, as it is often found that
consumption (a "realized" welfare achievement) is smoother than income (a "potential" welfare
opportunity), whereas these results indicate greater movement in household consumption.

Table 5-G: Transition matrix of calories/adult equivalent/day for 2004 and 2008
Calorie status 04
1
2
3
4
5

1
0.26
0.24
0.19
0.19
0.10

Calorie status 08
2
3
4
0.26
0.18
0.16
0.25
0.17
0.19
0.22
0.20
0.20
0.19
0.19
0.22
0.14
0.17
0.27

5
0.14
0.15
0.18
0.21
0.32

An assessment of the number of months without food stocks, among cropping households, shows that
the distribution differs according to agro-ecological region (Figure 5-6). Specifically, zone III in the
north exhibits a much higher proportion of households in 2008 that report that they always had food
stocks in the previous 12 months. This is probably due to the dominance of cassava in the region, as
cassava can often be left in the field until it is ready to be consumed, rather than requiring a harvest
during a specific season. Alternatively, this may be due to localized flooding during the 2006/07
agricultural season that negatively, and disproportionately, affected the other agro-ecological zones
(UNICEF 2007).

34

Proportion of households

0.7
0.6
0.5

All of Zambia

0.4

AEZ I

0.3

AEZ IIa

0.2

AEZ IIb

0.1

AEZ III

0
0

1

2

3 4 5 6 7 8 9
No. months without stocks

10 11

Figure 5-6: Distribution of months without food stocks in 2008 by agro-ecological region
Food energy deficient households are not equally distributed across Zambia's provinces. Table 5-H
presents estimated numbers of households that do not meet their calorie requirements for a moderate
level of activity. While the highest density of calorie deficient households is found in Luapula, the
greatest number of calorie-deficient households can be found in Northern and Eastern provinces. This
information should be of interest to development agencies interested in evaluating the cost-effectiveness
of geographically-targeted food security programs.

35

Table 5-H: Location of calorie-deficient households in 2008

Province

Central
Copperbelt
Eastern
Luapula
Lusaka
Northern
North West
Southern
Western
Total calorie-deficient
Total Zambia

Number
energydeficient
HHs
107,100
63,972
154,643
145,694
32,732
184,825
71,460
100,953
131,933
993,311
1,669,861

Number
energydeficient
individuals
740,827
394,184
992,772
858,346
209,158
1,102,541
456,770
692,357
791,477
6,238,432
9,627,873

Energydeficient
HHs / km²
1.14
2.04
2.24
2.88
1.50
1.47
0.48
1.19
1.04

Proportion of
province
rural
population
(HHs)
0.55
0.63
0.51
0.81
0.67
0.58
0.54
0.49
0.71

Proportion
of national
rural
population
(HHs)
0.06
0.04
0.09
0.09
0.02
0.11
0.04
0.06
0.08
0.59

Figure 5-7 presents the cumulative distribution of calories (per adult equivalent, per day) in each agroecological zone. As zone IIa first-order stochastically dominates the other zones, it is clear that this
region fared better in 2008 in terms of this indicator. A similar trend is seen when evaluating the
proportion of household calorie requirements that are met (Figure 5-8), where zone IIa sees the highest
proportion of households exceeding their requirements.

36

1
0

Cumulative density
.2
.4
.6
.8
0

5000
4000
3000
1000
2000
Household calories per adult equivalent per day
AEZ I
AEZ IIb

AEZ IIa
AEZ III

0

Cumulative density
.2
.4
.6
.8

1

Figure 5-7: CDF of calories/adult equivalent/day by agro-ecological zone

0

1
2
3
4
Proportion of calorie requirements met
AEZ I
AEZ IIb

5

AEZ IIa
AEZ III

Figure 5-8: CDF of proportion of calorie requirements met, by agro-ecological zone

37

5.4 Correlation among indicators
Table 5-I shows the correlations between the various food security and income indicators in 2008.
Although many of the variables are significantly correlated in the expected directions, the magnitude of
the coefficients seems small. The number of months without food stocks and the proportion of income
spent on food purchases are generally negatively correlated with variables associated with food security.
The number of income sources claimed by a household has a very weak correlation with the other
indicators.

Although not presented here, a similar correlation table was constructed for each agro-ecological zone
separately. The correlation coefficients often suggest that food security indicators are similarly related
in the different regions. However in zones IIb and III, the indicators for whether a household produces
fruits or vegetables is negatively correlated with total calories, while in zone IIb a similar negative
relationship is seen between calories and whether a household produces milk/eggs. These patterns are
unexpected and merit further consideration. As will be discussed in Chapter 6, this is taken into
consideration in the construction of a country-wide food security index.

38

Table 5-I: Correlation matrix of food security indicators in 2008

Calories
Income
Productive assets
Hectares cultivated
Land access
No. crops
TLU
Milk/eggs
Garden
Prop. crops sold
Maize/cassava in
storage
Months w/out food
Income sources
No. skilled off-farm
workers
Prop. food
expenditures
Transfer network

Milk/eggs
Garden
Prop. crops sold
Maize/cassava
Months
Income sources
No. off-farm
workers
Prop. food
expenditures
Transfer
network

Calories
1
0.34***
0.14***
0.29***
0.11***
0.16***
0.15***
0.10***
-0.02**
-0.21***

Income

Productive
assets

1
0.36***
0.28***
0.11***
0.18***
0.21***
0.15***
0.13***
0.15***

1
0.30***
0.11***
0.14***
0.25***
0.24***
0.12***
0.17***

0.19***
-0.22***
0.03***

0.11*** 0
-0.19*** -0.13***
0.30*** 0.06***

0.12***
-0.17***
0

0.05*** 0.31*** 0
-0.06*** -0.26*** -0.06***
0
0.16*** 0

-0.05***

0.35***

-0.08***

0.03**

-0.11***
0.05***

-0.35*** -0.12***
0.17*** 0.07***

-0.17***
0.06***

-0.04*** -0.23*** -0.08***
0.04*** 0.09*** 0.04***

0.17***

Ha
Land
cultivated access

No.
crops

TLU

1
0.26***
0.25***
0.31***
0.23***
0.09***
0.08***

1
0.08***
0.30***
0.07***
0.05***
0.02***

1
0.05***
0.23***
0.60***
-0.17***

1
0.18***
0.06***
0

-0.11*** 0

Milk/eggs
1
0.16***
-0.02**
0.05***
-0.07***
0.11***

Garden

Prop.
crops
sold

1
0.03***
0.12***
-0.08***
0.15***

1
-0.17***
-0.11***
-0.10***

1
-0.57***
0.06***

1
0

1

-0.04***

-0.04***

0.14***

-0.06***

0.02***

0.16***

1

-0.08***

-0.11***

0.038***

-0.25***

0

-0.11***

-0.05***

0.10***

0.10***

0.04***

0.03***

-0.04***

0.37***

0.07***

Prop. food expenditures
Transfer network

Maize/
cassava

Months
without
stocks

Income
sources

Prop. food exp.
1
-0.05***
39

Transfer
network
1

Off-farm
workers

5.5 Food security profiles
The last descriptive exercise will use the household-level information from 2008 to create a series of
“food security profiles” using cluster analysis. Clustering is used to group households into categories
such that households within a category are similar to one another but dissimilar to households in other
categories (Babu and Sanyal 2009). The households within a cluster should resemble one another along
a range of indicators or food security dimensions, and this technique can be therefore used to recognize
patterns among the food security indicators. Similarity is measured with the error sum of squares
between households in a cluster and the cluster mean, and once the number of clusters is specified, a “kmeans” cluster procedure will classify households to minimize this error sum of squares. Non-panel
variables are used in this cluster analysis, and results are provided in Table 5-J. Households with any
missing observations are dropped from the analysis, and a fourth category that includes just several
dozen very wealthy households is not presented here. Please note that this clustering process was done
without consideration of household weights, and the number of clusters was determined arbitrarily to
elicit a set of food security profiles that seem to “tell a story”.

The cluster analysis readily categorizes households into groups of low, medium, and high food security,
in which all the indicators align as one would expect. Approximately one third of the Zambian rural
population is categorized into the “low” group: These households do not own any productive assets and
have low average values of all other food security indicators, and about one quarter of these households
are FHH widows. At the other end of the spectrum, the “high” food security group experiences (on
average) the shortest length of time without food stocks and the highest level of crop diversity. Most of
these households come from agro-ecological zone IIa.

40

Table 5-J: Food security profiles in 2008
Average values
Calories/AE/day
Calories/AE/day (median)
Crops retained
1=HH produced milk or eggs
1=HH has garden
Months without food stocks
Hectares cultivated
1=Participation in transfer network
TLU
Value of productive assets (ZMK)
1= FHH widows
1=AEZ I
1=AEZ IIa
1=AEZ IIb
1=AEZ III
No. observations
Proportion of population

41

Low
3,367.40
2,191.12
2.41
0.33
0.31
3.10
0.91
0.79
0.31
0
0.24
0.07
0.30
0.15
0.48
2,702
0.33

Category
Medium
High
3,544.79 5,938.91
2,524.63 4,111.29
2.94
3.26
0.52
0.86
0.42
0.49
2.43
1.17
1.54
3.89
0.84
0.92
1.15
16.18
1,454,971 9,166,691
0.08
0.08
0.05
0.12
0.44
0.71
0.04
0.07
0.47
0.10
4,576
566
0.57
0.07

6. THE MULTIDIMENSIONAL FOOD SECURITY INDICES

6.1 Construction of the indices
From the array of food security indicators calculated in Chapter 5, it can be difficult to extract a
household’s overall status of food security. To do so, it may be useful to construct a
multidimensional index that incorporates the most important indicators from each dimension of
food security. However, the construction of such an index can be highly subjective, particularly
with regard to the weights assigned to each element of the index. This thesis uses principal
component analysis (PCA) to construct a composite food security index, such that the weights
are derived objectively from the data.

PCA is a type of factor analysis that can reduce dimensions, or uncover latent variables, by
extracting a linear combination that best describes the co-variance among all elements and
transforms them into one index. The first principal component is the linear combination
capturing the greatest variation among the set of variables. This is converted into factor scores,
which serve as weights for the creation of an index for each household. In other words, from an
initial set of n correlated variables (X1, X2… Xn), PCA creates m uncorrelated principal
components where each is a linear weighted combination of the initial variables as follows:
PCm = am1X1 + am2X2 + am3X3 + … + amnXn
Here, 𝑎𝑚𝑛 represents the weight for the mth principal component and the nth variable. The

components are ordered so that the first component explains the largest amount of variation in

the data subject to the constraint that the sum of the squared weights is equal to one. Each
subsequent component explains some additional but smaller proportion of variation of the

42

variables. As data reduction is the primary objective of this exercise, only the first component
will be used.

Once the first component is identified, the food security index for each household is derived as
follows:

𝐹𝑆𝐼𝑗 = ∑ 𝐹𝑖 [

𝑋𝑗𝑖 − 𝑋𝑖
𝑆𝑖

]

Where FSIj is the Food Security Index, which follows a normal distribution with a mean of 0. Fi
th

th

is the weight for the i variable in the PCA model (the square of ami), Xji is the j household’s
th

th

value for the i variable, Xi and Si are the mean an d standard deviations of the i variable for
overall households. Essentially, the FSI is the sum of the weighted z-scores for each variable.

Each PCA index is based on a scale which is relevant only to that estimation, such that a set of
indices from different estimations cannot be meaningfully compared. As this study uses three
rounds of household panel data, it is necessary to generate an index that is comparable over time.
To this end, following the approach of Cavatassi et al. (2004), data for the three rounds are
pooled (including all households interviewed in each year) and principal components are
estimated over the combined data. The resulting weights are then applied to the variable values
for each round of the survey. As discussed in section 6.2, the principal components are carefully
inspected to discern whether the variables used to construct the index and their respective
weights remain similar in all the three rounds, as this condition enables the comparison of
changes over time (Vyas and Kumaranayake 2006). The pooled estimate is used in panel
analyses, while a separate index is computed for 2008 and used in cross-sectional analyses.
43

6.2 Composition of the indices
The selection of indicators included in the indices is driven by both the food security literature
and the goal of maximizing the variation explained by the first principal component. For
example, the use of hectares cultivated, rather than land access (i.e. potential hectares cultivated)
results in a greater amount of variation explained. Two versions of the index are created: The
first is the Food Sufficiency Index (FSU), based mostly on the quantity and quality measures of
food security. The second index is called the Food Security Index (FSI) and is based on the
quantity, quality, and stability measures. Thus, the FSI explores the merit of including indicators
of future vulnerability in the index. The variables included as elements of each index are listed
in Table 6-A.

44

Table 6-A: Elements of the food sufficiency and food security indices

Each index for 2008 includes indicators of the various dimensions of food security: quantity
(total calories), quality (crop count, TLU, milk/eggs), and stability (months without food stocks,
proportion of crop value sold, access to a social safety net, value of assets, and size of land
cultivated). Traditional multivariate methods are based on continuous variables. In this analysis,
one or two categorical variables are included in a given index, but because they are outnumbered
by continuous variables, it should not harm the analysis (Abeyasekara 2005).

Unfortunately, there is no indicator of stability in the FSU for multi-year analysis. This is
because the 2001 survey did not capture information on cassava stocks, and a variable that only
covers the stability of other crop stocks would misrepresent the food security situation in zone III
45

where cassava is widely consumed. Another point to note is that non-cropping households
(which comprise about 5 percent of households in rural Zambia) are included in this index
construction exercise, and are given a value of 0 for crops retained if they did not produce any
crops and a value of 12 for months without food stocks.

It is also unfortunate that no measure of income diversification is included in the FSI. Several
variables were considered, including the number of income sources from which a household
draws income and the number of household members that participate in skilled work or work
that requires a capital investment. However, each variable is often associated with distress rather
than stability. It seems that non-farm income options are so limited in rural Zambia that they
cannot be meaningfully regarded as a component of a household’s underlying food security
status.

6.3 Suitability and factor loadings
All of the original variables are loaded onto one factor that explains the most variance in the
original components. For the multi-year indices, Table 6-B provides the factor loadings of the
first principal component, as well as the eigenvalue (proportion of variance explained) and
Kaiser-Meyer-Olkin (KMO) statistic. The factor loadings can also be thought of as the square
root of the weight assigned to each variable. To construct the multi-year indices, all households
interviewed in each year are included, using population weights. The PCA is not rotated as this
does not seem to enhance the results.

46

Table 6-B: Factor loadings for multi-year indices: FSU and FSI
Factor loadings
Pooled years 2001 2004

Food sufficiency index (FSU)
Component
Calories (ln)
No. different crops retained
Tropical livestock units
Household produces milk or eggs
KMO
Proportion variation explained
Number observations

2008

0.64 0.65 0.65 0.55
0.6 0.62 0.56 0.54
0.33 0.27 0.38 0.56
0.35 0.34 0.35 0.60
0.52
0.35
0.4 0.33 0.39
20,435 6,922 5,419 8,094
Factor loadings
Pooled years 2001 2004

Food security index (FSI)
Component
Calories (ln)
No. different crops retained
Tropical livestock units
Household produces milk or eggs
Hectares cultivated
Participates in transfer network
Value of productive assets (ln)
KMO
Proportion variation explained
Number observations

2008

0.39 0.41 0.39 0.36
0.34 0.38 0.29 0.33
0.39 0.40 0.37 0.42
0.31 0.24 0.37 0.34
0.52 0.53 0.51 0.50
0.12 0.05 0.18 0.09
0.45 0.44 0.44 0.46
0.65
0.29 0.29 0.27 0.31
20,435 6,922 5,419 8,094

In the multi-year indices, the data reduction of the PCA explains 35 percent of the original
variation of the data for the FSU. For the FSI, this value is 29 percent. These values are similar
to those seen in other studies that use PCA to develop a food security or vulnerability index:
Filmer and Pritchett (1998) use the first principal component of PCA to construct a household
wealth index, where the first factor explains 25.6 percent of the variation in the data. Dasgupta
and Baschieri (2010) construct an index of vulnerability to climate change, where the first
principal component explains 25 percent of the variation, and Demeke et al. (2011) similarly use
PCA to build an index of food security in Ethiopia, where the first factor explains 32.5 percent of
the variation.
47

The factors that load most heavily on the FSU are calories and crop diversity, and the value of
the Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy is 0.52. A large KMO value
(greater than 0.5) indicates that correlations between pairs of components can be explained by
the other variables in the index (Kaiser 1974). It should be noted, however, that this value is
close to the cutoff of acceptability. In deciding to pool the data from all years of the panel, it is
important to verify that the pattern of factor loadings do not change fundamentally from year to
year, when the exercise is repeated for each year independently. An inspection of the yearly
values in Table 6-B confirms that it is acceptable to create a pooled index. An example of how
these factor loadings are used to create the food sufficiency score for a specific household is
available in the appendix (Table 9-I).

Table 6-C provides the same information for the indices derived when using only the year 2008.
When applicable, these variables are non-panel values. TLU now includes all animals besides
oxen, which are considered productive assets, and the FSU now includes an indicator of stability
(months without food stocks).
Table 6-C: Factor loadings for 2008: FSU and FSI

Food sufficiency index (FSU)
Component
Calories (ln)
Number of different crops retained
Edible tropical livestock units (TLU)
Household produces milk or eggs
Number of months with no crop
stocks
KMO
Proportion variation explained
Number observations

Agro-ecological zones
All of
Zambia
0.41
0.52
0.28
0.39

I

IIa

IIb

III

0.38
0.343
0.49
0.46

0.48
0.46
0.33
0.36

0.43
0.57
0.25
0.29

0.31
0.59
0.16
0.42

-0.57

-0.48

-0.56

-0.58

-0.60

0.60
0.36
7,909

0.36
489

0.38
3,074

0.39
612

0.34
3,734

48

Table 6-C (cont’d)
Food security index (FSI)
Component
Total calories (ln)
Number of different crops retained
Household produces milk or eggs
Number of months without food stocks
Hectares cultivated (ha)
Participates in transfer network
Edible tropical livestock units
Value of productive assets, including oxen
(ln)
KMO
Proportion variation explained
Number observations

Agro-ecological zones
All of
Zambia
0.34
0.38
0.35
-0.41
0.46
0.13
0.32

I

IIa

IIb

III

0.30
0.33
0.36
-0.37
0.43
0.05
0.42

0.39
0.34
0.31
-0.43
0.44
0.14
0.32

0.34
0.48
0.26
-0.47
0.44
0.02
0.28

0.28
0.48
0.35
-0.48
0.48
0.13
0.23

0.34

0.41

0.36

0.31

0.20

0.71
0.29
7,909

0.30
489

0.32
3,074

0.31
612

0.26
3,734

The principal component tends to load positively on variables which contribute to food security,
and negatively on variables which contribute to food insecurity. For 2008, Table 6-C displays
the results when these indices are created for each agro-ecological zone separately. It would be
problematic if the pattern of factor loadings differed between regions, but a visual inspection
shows that the importance of different variables to food security is relatively consistent across
zones. The number of observations for which each index can be calculated is slightly lower than
the number of households interviewed in 2008 due to missing observations.

6.4 Descriptive statistics
The average and standard deviation of these indices in 2008 are shown in Table 6-D. As
expected, the average value for all households is zero, as the PCA always produces a relative
index centered at zero. It is clear that female-headed households have lower average scores, as
compared with the general population, and this is true whether FHHs are defined as widowed
households or FHHs without a prime-age male (FHH nm). However, it seems that FHHs

49

without a prime-age male experience lower food security than the alternative definition of FHHs.
Among agro-ecological zones, the drier regions (zone I and IIb) experience lower average food
sufficiency, and only zone IIa has a positive average value for both the FSU and FSI.
Interestingly, zone I fares much better with regard to its average FSI, while zone III fares much
worse. This is surely because of the high disease burden for livestock in zone III, compared with
the important cultural role of livestock in zone I.

Table 6-D: FSU and FSI in 2008
Agro-ecological zones

Food sufficiency
score
Food security score

All HHs

FHH
widow

FHH
nm

I

IIa

IIb

III

0.00

-0.18

-0.34

-0.04

0.17

-0.68

0.00

(1.41)
0.00
(1.51)

(1.30)
-0.42
(1.35)

(1.36)
-0.68
(1.29)

(1.32)
-0.05
(1.44)

(1.50)
0.33
(1.67)

(1.36)
-0.89
(1.38)

(1.31)
-0.13
(1.31)

To understand what is driving the average value in each agro-ecological zone, it is necessary to
analyze the average values of each element of the index. The radar graph in Figure 6-1 displays
the average values for the FSI elements for each zone. Note that for this figure, one of the food
security indicators (months with food stocks) was adjusted so that higher values represent greater
security. These values are normalized z-scores, such that a value of zero represents the median
level in Zambia. A higher score farther from the center of the graph is relatively good, while a
value closer to the center represents a lower level of welfare. Zones I, IIa, and III exhibit
relatively "smooth" circles, though zone I stands out in terms of the low number of months with
food stocks, and zone III stands out on terms of high crop diversity and time spent with food
stocks. Zone IIb consistently scores low for indicators of food security, and the low value for

50

number of crops/FFV retained seems to stem from the much lower prevalence of gardens in the
region (Table 5-F).

Calories
0.4
0.2
0
-0.2
-0.4
-0.6
-0.8

Value of
productive assets

Participates in
transfer program

Number crops/
FFV retained
AEZ I
Edible TLU

AEZ IIa
AEZ IIb
AEZ III

Hectares
cultivated

Months with food
stocks
Produces milk or
eggs

Figure 6-1: Radar graph of food security indicators for 2008
The distribution of the FSU scores and FSI scores across years is presented in Figure 6-2. The
FSU seems to change little from year to year, while there is more variation in the FSI, where the
year 2008 exhibits a leftward shift.

51

.4
Density
.3
.2
.1
0
-5

0
Food sufficiency score (FSU)
2004

0

.1

Density
.2
.3

.4

2001
2008

5

-5

0

5

Food security score (FSI)
2001
2008

2004

Figure 6-2: Distribution of FSU and FSI across years
The average score and standard deviation of the FSI and FSU for each year are displayed in
Table 6-E. T-tests show that from 2001 to 2004, Zambian households experienced a significant
increase in food sufficiency and security. From 2004 to 2008, the average values of food
sufficiency dropped while food security improved ever so slightly. Although statistically

52

significant, it seems the difference from 2004 to 2008 may be so small as to be practically
insignificant – reflecting once again a situation of stagnancy in Zambia over this time period.

Table 6-E: FSU and FSI across years
Years

FSU
FSI

All HHs

2001

2004

2008

0.00
(1.18)
0.00
(1.42)

-0.10
(1.14)
-0.15
(1.29)

0.05
(1.16)
0.04
(1.41)

0.03
(1.23)
0.07
(1.50)

|t| 2001
& 2004
3.9

|t| 2004
& 2008
2.88

2.87

6.08

The spatial pattern of how food security has changed over time can be seen in the maps of Figure
6-3, which shows the average value of the FSU for each district. While the relative degree of
food insecurity does change, there are clearly pockets of insecurity, namely in zones IIb and III.
The pattern in 2008 closely reflects the location of flooding in Zambia in the 2006/07
agricultural season (UNICEF 2007). It seems that this index can be effectively used to gauge the
household impact of short-term shocks. At the same time, a pattern observed in a single year
should not be conflated with regional food security over the long-term, and this fluctuation
demonstrates the value of precisely this sort of longitudinal analysis.

53

Figure 6-3: Geographic pattern of average district FSU for 2001, 2004, and 2008
A transition matrix of households' food sufficiency scores (Table 6-F) from 2004 to 2008 shows
that households do not tend to move very far along the FSU scale. The status categories are
derived from the FSU quintiles of 2004. Of the households that fall into the first quintile in 2004,
31 percent retain the same status in 2008. For the food security scores (Table 6-G), this pattern
is even more pronounced: Of the households that fall into the top quintile in 2004, 52 percent
retain the same status in 2008.

54

Table 6-F: Transition matrix of food sufficiency score from 2004 to 2008

FSU status 04
1
2
3
4
5

FSU status 08
1
2
0.31
0.26
0.26
0.24
0.18
0.21
0.16
0.14
0.10
0.14

3
0.20
0.22
0.22
0.19
0.17

4
0.19
0.18
0.20
0.24
0.19

5
0.08
0.11
0.20
0.25
0.36

Table 6-G: Transition matrix of food security score from 2004 to 2008

FSI status 04
1
2
3
4
5

FSI status 08
1
2
0.32
0.27
0.32
0.32
0.19
0.18
0.12
0.14
0.06
0.09

3
0.18
0.26
0.24
0.22
0.10

4
0.13
0.20
0.24
0.25
0.17

5
0.04
0.09
0.14
0.21
0.52

Given this trend, it is useful to think about the predictive power of various indicators in terms of
how well they would predict a household's status 4 years in the future. Table 6-H presents the
correlation coefficients between household income (in both 2004 and 2008) and a series of food
security outcomes in 2008. All correlations are positive and strongly significant, though the
magnitude of the relationship between income and food security drops precipitously when
income is related to food security four years in the future. Interestingly, income in 2008 is most
strongly related to calories in 2008, while income in 2004 is most strongly correlated with the
FSI in 2008.

One might also want to assess the predictive power of food security indicators in 2004 (Table 6I). The correlation between FSI in 2004 and 2008 is 0.55, which makes sense as many of the
components of this index might be consistent within a household from one year to the next. In
terms of predicting a household's food sufficiency experience in 2008, FSI outperforms FSU.
55

This is as expected, as the FSI specifically includes indicators of vulnerability to future food
shortfalls. In terms of predicting the proportion of calorie requirements met in 2008, none of the
2004 indicators exceed a correlation of 0.17.

Table 6-H: Income and food security indicators, 2004 and 2008
2008

Income (ln) 2008
Income (ln) 2004

FSI
FSU
Calories (ln)
0.40*** 0.39*** 0.43***
0.19*** 0.13*** 0.13***

Proportion of calorie
requirements met
0.41***
0.11***

Table 6-I: Food security indicators in 2004 and 2008
2008
FSI
FSU
0.55*** 0.40***
FSI
0.35*** 0.33***
FSU
0.21*** 0.19***
Calories (ln)
2004
Proportion of calorie
0.18*** 0.14***
requirements met*
*assuming moderate level of activity

Calories (ln)
0.19***
0.15***
0.16***

Proportion of calorie
requirements met
0.17***
0.13***
0.14***

0.12***

0.17***

A final exercise in assessing the relative predictive power of the FSI is a thought experiment, in
which the goal is to identify in 2004 the households that will be food insecure in 2008. A
household is targeted if it is in the lowest one-third of either the income or FSI distribution in
2004. The targeting is considered accurate if the household would be targeted in 2004 and is in
the lowest one-third of the FSU distribution in 2008. A type I error is a ‘false positive’, where
the household is targeted when it should not have been, and a type II error is a ‘false negative’,
where the household is not targeted when it should have been.

Table 6-J displays the proportion of 2008 households that fall into each category, given the
income and FSI targeting criteria. This outcome suggests that the FSI from 2004 does a better
56

job of accurately targeting households that will be poor in 2008. If the FSI is accepted as a
practical measure of food security, then a household's FSI score in 2004 is a better predictor of
food sufficiency four years later, as compared with a money-metric indicator, and incomepoverty status alone seems insufficient to identify households that are vulnerable to food
insecurity. However, the two targeting criteria have different patterns of type I and II errors, and
using the FSI results in 12 percent of 2008 households being overlooked when they should have
been targeted.

Table 6-J: Type I and II errors when income and FSI are used as targeting criteria
Accurate targeting
Type I error
Type II error

Income 2004
0.34
0.63
0.03

FSI 2004
0.65
0.23
0.12

Are households merely "bouncing around" the food security indices without a pattern? Table 6K presents the percent of households that are always, sometimes, and never secure, in terms of
food sufficiency or food security. Here, a household is categorized as "always food insecure" if
its FSU or FSI score is in the lowest one-third of the distribution for all three waves of the panel.
A household is considered "sometimes" food insecure if its score is in the lowest one-third for
one or two years, and never food insecure if its score is never in the lowest one-third.

Table 6-K: Persistence of food security from 2001 to 2008
Always food
insecure
Sometimes
Never

FSU

FSI

0.08
0.54
0.38

0.09
0.50
0.41

Table 6-K breaks down the proportion of households that experience the following trajectories in
terms of their FSU score and FSI score: A positive change from 2001 to 2004 and another
57

positive change from 2004 to 2008 (+ +); A positive followed by a negative change (+ -); a
negative followed by a positive change (- +); and two negative moves down the index (- -). For
food sufficiency, 22 percent of households consistently moved higher on this scale, while 34
percent consistently fell lower, and the remaining households (~43 percent) experience a change
in their direction of movement. This pattern of movement is mirrored in the food security index,
which suggests strong consistency between the two indices.

Table 6-L: Proportion of households by FSU and FSI trajectory
Trajectory
+ +
+ - +
- -

FSU
0.22
0.14
0.30
0.34

FSI
0.25
0.14
0.29
0.33

Figure 6-4 shows the relationship between the FSU in years 2001 or 2004 and the subsequent
change in FSU from that year to the next panel year. Although there is a good deal of variation,
the trend is generally negative: A higher FSU is correlated with a negative (or smaller positive)
change in the FSU. If a higher FSU were correlated with greater security, one might expect to
see a quadratic relationship between FSU in a given year and the subsequent change in FSU.
However, the regression results presented in Table 6-M indicate that this is not the case. This
pattern does not necessarily indicate that households are haphazardly "bouncing around", as a
higher food security score leaves less potential for additional improvement. As well, it may be
overreaching to conclude that the variability in food security in this short time frame represents
the longer-term trends of household movement. (See Naschold and Barrett (2011) for an
analogous discussion of how to interpret stochastic income over varying time frames.) However,
this does illustrate that household experiences regularly fluctuate in the short term.

58

5
Change in FSU 2001 to 2004
0
-5

0
Food sufficiency score (FSU) 2001

5

-5

0
Food sufficiency score (FSU) 2004

5

-5

Change in FSU 2004 to 2008
0

5

-5

Figure 6-4: FSU and change in FSU, 2001 to 2004 and 2004 to 2008

59

Table 6-M: OLS regression of change in FSU and previous FSU
∆FSU 2004-08
OLS

Explanatory variables
FSU 2004

-0.658***
(0.024)
0.003

[FSU 2004]²

(0.003)
0.124***
(0.024)

Constant

Observations
4,340
0.302
R²
*** p<0.01, ** p<0.05, * p<0.1

60

7. APPLICATIONS OF THE FOOD SECURITY INDICES

7.1 Correlates of food sufficiency and food security (cross-sectional methods)
7.1.1. Probability of being food insufficient
In this chapter, the food sufficiency and food security indices are applied to various exercises
aimed at identifying the correlates and determinants of food security. The selection of variables
included in the models is guided by the food security literature and the significant determinants
that have been identified in similar contexts (Feleke et al. 2005; Garrett and Ruel 1999;
Misselhorn 2005; Tschirley and Weber 1994), the variables that seem like they may plausibly
affect food security, and data availability. Section 7.1.1 is limited to cross-sectional analyses of
the 2008 data in order to take advantage of the larger number, and more nuanced definitions, of
food security indicators and regressors that are available only in this last wave of the panel
survey. Although attempts are made to ensure the regressors are exogenous, readers should be
cautious when interpreting the coefficients in section 7.1. Because time-invariant household
fixed effects cannot be captured in cross-sectional regressions, the coefficients are best
interpreted as measures of partial correlation rather than causation. In addition, it should be noted
that because the dependent variable is a normalized food security score unique to this population,
the coefficients are a bit difficult to interpret. Readers are encouraged to focus on the sign,
significance, and relative magnitude of the coefficients rather than their specific value.

A first attempt to identify the correlates of food insufficiency is done with a logit regression, in
which the dependent variable is an indicator of being in the bottom one-third of the FSU
distribution in 2008. This is modeled as
wi = α + βiXi + εi

61

th

where wi = ln(pi/(1-pi)), pi = prob (yi = 1| xi), yi = 1 if the i household is food insecure and 0
otherwise, and Xi = a vector of socioeconomic and geographic characteristics of the i

th

household. Each district is matched with a meteorological station in Zambia, where rainfall and
temperature records reference the 2006/07 agricultural season. Average partial effects are
presented in the first column of Table 7-A. In column 1, a positive value means that an increase
in the explanatory variable leads to an increase in the predicted probability of being food secure.
Conversely, a negative coefficient means that an increase in the explanatory variable results in a
decrease in the probability of being food secure.

Among the notable results is the lack of significance of being a female-headed widowed
household. The insignificance here does not negate the fact that female-headed widowed
households have significantly lower FSU than the general population (Table 6-D), but it
indicates that once all of these other factors are controlled for, there is no residual causal
relationship between widowhood and food insufficiency. Rather, their status is explained by the
other household characteristics included as controls in this regression. Ceteris paribus, a higher
dependency ratio is actually associated with sufficiency; this is surprising, given the importance
attributed to the burden of a high dependency ratio in other studies (Kennedy and Peters 1992).

Several explanatory variables relate to a household’s endowment of social capital: Relation to
the village chief is a significant determinant of food sufficiency, while the number of years a
household has resided in the village is not significant. Participation in a transfer network (the
receipt or provision of cash or goods) is associated with being food sufficient, though this
potentially exhibits reverse causality as it may have been directly determined by a household’s
62

food security experience. For example, if households receive remittances during times of
distress, this would attenuate the detected relationship between social capital and avoidance of
insufficiency. Alternatively, if households provide assistance to others during a good year, their
food sufficiency status makes it possible for them to participate in a transfer network. One
option would have been to use a household’s transfer network status from 2004, as this could be
exogenous with 2008 food sufficiency. However, this would have sharply limited the sample
size, so the household’s 2008 transfer network status is included with the caveat that it is most
certainly endogenous.

The variables related to a household’s asset endowment tell a consistent story: A higher value of
productive assets (including work oxen) and larger area of land cultivated are significant
determinants of sufficiency. The variables related to information access (ownership of a radio
and receipt of agricultural information services) are also positively correlated with sufficiency. It
is interesting to see that a higher proportion of crop value marketed is negatively associated with
food sufficiency, though the relationship may be driven by inclusion of a dummy variable for
production of a cash crop. Finally, rainfall is a significant determinant of food sufficiency in the
expected quadratic relationship, and a joint F-test of significance of average season temperature
(𝐹 = 8.41, prob > 𝐹 = 0.000) reveals that temperature is also a determinant of food sufficiency in
a quadratic manner.

Table 7-A: Logit and OLS regressions of household food sufficiency/ FSU score in 2008

Explanatory variables

(1)
1=Food sufficient
Logit

HH size

-0.013***
(0.002)
63

(2)
FSU 2008
OLS
-0.020**
(0.008)

Table 7-A (cont’d)
Dependency ratio
Maximum education
Maximum women’s educ.
1=FHH widow
1=HH head recently died
Age of HH head
Age² of HH head
1=Owns radio
1=Agricultural services
Km to main road (ln)
1=HH related to chief
Years in village (ln)
1=Transfer network
No. skilled workers
1=Woman works off-farm
Proportion income on food
1=Cash crop
Kgs nitrogen/ ha maize (ln)
1=Grows maize
1=Owns water pump
Proportion crop value sold
Value of productive assets (ln)
Hectares cultivated
64

0.076**
(0.030)
0.000
(0.002)
0.002
(0.002)
0.009
(0.018)
-0.019
(0.030)
0.003
(0.002)
-0.000

0.163*
(0.084)
0.004
(0.007)
0.009
(0.006)
-0.048
(0.046)
-0.073
(0.076)
0.013**
(0.006)
-0.000*

(0.000)
0.040***
(0.012)
0.075***
(0.011)
-0.006*
(0.003)
0.060***
(0.011)
0.003
(0.007)
0.061***
(0.014)
-0.041***
(0.015)
-0.001
(0.029)
-0.407***
(0.032)
0.032
(0.020)
0.013***
(0.004)
0.023
(0.015)
0.098
(0.073)
-0.254***
(0.020)
0.006***
(0.001)
0.161***

(0.000)
0.079***
(0.031)
0.303***
(0.030)
-0.002
(0.009)
0.237***
(0.029)
0.045***
(0.015)
0.232***
(0.037)
-0.099**
(0.041)
0.030
(0.095)
-1.248***
(0.073)
0.296***
(0.048)
0.063***
(0.009)
0.412***
(0.039)
1.392***
(0.457)
-1.262***
(0.061)
0.027***
(0.003)
0.262***

Table 7-A (cont’d)
1=Zone IIa
1-Zone IIb
1=Zone III
mm important rain (ln)
Important rain²
Season temperature (°C)
Season temperature²

(0.010)
-0.000
(0.027)
-0.066**
(0.033)
0.129***
(0.032)
3.510***
(0.867)
-0.262***

(0.026)
-0.106
(0.070)
-0.379***
(0.083)
0.327***
(0.085)
11.575***
(2.124)
-0.862***

(0.066)
0.062
(0.079)
-0.001

(0.162)
0.558***
(0.208)
-0.011**

(0.002)

(0.004)
-47.399***
(7.469)
7,234
0.445

Constant

Observations
7,234
R²
Average partial effects (column 1) and OLS coefficients (column 2)
with robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
7.1.2 Correlates of food sufficiency score

Similar results are seen when food sufficiency is maintained as a continuous variable. An
ordinary least squares regression of households in 2008 is modeled as
FSUi = α + βiXi + εi
where FSUi = the Food Sufficiency score of household i in 2008, and Xi = a vector of
th

socioeconomic and geographic characteristics of the i household. The results are presented in
column 2 of Table 7-A. The unit of the dependent variable is 1 standard deviation of the FSU
distribution. However, because this is a population-specific index, the sign and significance of
the coefficients are probably more important than their magnitude.

65

An older household head is generally associated with higher food sufficiency, although the size
of the coefficient is small. Consistent with the results of column 1, the coefficient on FHH
widows is not significant, although the sign is in the expected direction. A higher proportion of
income spent on market food purchases is negatively related to food sufficiency, which suggests
that households that are more reliant on the market for their food needs experience more
insecurity. It should be acknowledged that this variable certainly suffers from simultaneity bias,
as it is measured during the same time period as the dependent variable. In other words, even if
a household does not alter its food purchase behavior in a bad year, the proportion of household
income devoted to food purchases would necessarily rise as income falls. The coefficient is
probably biased downward.

Among the geographic variables, zone I (in the south) is used as the reference zone, and a
household located in zone IIa or zone III is predicted to have a similar or significantly higher
food sufficiency score. This is consistent with the descriptive results shown in Figure 6-3, where
only zone IIb stands out as being less food secure than zone I. It is interesting to note that
remoteness, as measured by distance to a main road, is not found to be associated with food
sufficiency.

7.1.3 Correlates of the persistence of food security
The correlates of the persistence of food security are explored by first classifying households into
3 mutually-exclusive and exhaustive categories as always food insecure, always food secure, or
sometimes food secure. A household is considered food insecure in a given year if its FSU score
is in the lowest one-third of the multi-year FSU index, and it is classified as always food insecure

66

if it is food insecure in all 3 years of the panel, always food secure if it is never in the lowest onethird of the distribution, or sometimes food secure if its classification changes at least once over
the study period. This follows the approach developed by Demeke et al. (2011), and also
responds to the appeal of Hulme (2003) regarding the importance of studying the causes of
chronic poverty relative to transient poverty.

A multinomial logistic model (ML) is used to identify the determinants of being either always
food secure or insecure. The ML model provides partial estimates of the effects of each
explanatory variable on the probability that a household experiences one of the discrete
outcomes listed above, relative to a base category. In this case, the base category is the status of
being sometimes food secure, and the average partial effects are presented in Table 7-B. The
explanatory variables are restricted to initial household conditions in 2001, and readers should be
cautious about interpreting these values in a causal manner: Rather the purpose of this exercise is
to capture the partial correlations between each explanatory variable and a household’s long-term
food security experience.
Table 7-B: Determinants of being always food secure or insecure, 2001
(1)
1=Never
food secure
Logit

Explanatory variables (2001)
HH size

0.005***
(0.002)
0.027
(0.029)
-0.004*
(0.002)
0.001
(0.002)
0.012
(0.016)

Dependency ratio
Maximum education
Maximum women’s educ.
1=FHH widow
67

(2)
1=Always
food secure
Logit
-0.012***
(0.003)
0.103**
(0.043)
0.007**
(0.003)
0.007**
(0.003)
0.031
(0.026)

Table 7-B (cont’d)
1=HH head recently died
Age of HH head
Age² of HH head
1=Owns radio
Km to main road (ln)
1=Transfer network
No. skilled workers
1=Woman works off-farm
Proportion income on food
1=Cash crop
Kgs nitrogen/ ha maize (ln)
1=Grows maize
1=Owns water pump
Proportion crop value sold
Value of productive assets (ln)
Hectares cultivated
1=Zone IIa
1=Zone IIb
1=Zone III
1=Zero tillage
1=Intercrop
1=Crop rotation

68

-0.014
(0.060)
-0.002
(0.002)
0.000

-0.035
(0.094)
0.005
(0.003)
-0.000*

(0.000)
-0.020*
(0.012)
-0.001
(0.003)
-0.012
(0.011)
0.029**
(0.014)
0.025
(0.028)
0.031
(0.019)
0.013
(0.030)
0.001
(0.004)
-0.014
(0.012)
-0.128**
(0.057)
-0.016
(0.029)
-0.008***
(0.002)
-0.083***
(0.012)
-0.003
(0.020)
0.079***
(0.022)
0.001
(0.021)
0.022
(0.015)
-0.023**
(0.011)
-0.027**
(0.011)

(0.000)
-0.016
(0.016)
0.011***
(0.004)
0.048***
(0.015)
-0.029
(0.024)
-0.020
(0.055)
-0.131***
(0.047)
0.058*
(0.030)
0.020***
(0.005)
0.085***
(0.021)
-0.074
(0.080)
-0.101***
(0.036)
0.011***
(0.002)
0.106***
(0.007)
0.072**
(0.033)
-0.096**
(0.043)
0.067**
(0.034)
-0.073**
(0.029)
0.032**
(0.016)
0.096***
(0.015)

Table 7-B (cont’d)
Observations
4,023
4,023
Average partial effects with robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
The results are not particularly surprising, as the same variables significantly contribute to the
probability of being persistently secure as had affected the food sufficiency score (Table 7-A).
The results presented are all relative to the base outcome of being sometimes food secure. For
example, in column 2 a one-year increase in the maximum education level results in an increase
in the probability of being always food secure, as opposed to being sometimes food secure. The
results indicate that, even holding the household maximum education level constant, a higher
maximum education level for women is a positive determinant of long-term food sufficiency.
Ownership of a radio is correlated negatively with the probability of being always insecure rather
than sometimes secure, but this is not a significant determinant of being located further along on
the security spectrum. Several farm management variables fall under the heading of
conservation farming: Practicing crop rotation from year to year is associated with long-term
food sufficiency, as is intercropping. However the correlation between practicing zero tillage
and being food secure is ambiguous.

7.2 Correlates of food sufficiency and food security (panel methods)
7.2.1 Correlates of food sufficiency and security scores, controlling for household fixed
effects
All previous regressions have not controlled for unobservable household characteristics that may
determine food security. For example, a farmer that applies more effort may experience a higher
level of food security, but such traits are not captured in the data. If an explanatory variable is
correlated with an omitted variable, the coefficient from an OLS regression will be biased. A
69

fixed effects model controls for these unobservables and should therefore more likely provide
unbiased estimates of the coefficients. The appropriateness of a fixed-effects model is confirmed
2

with Hausman tests, which in each case reject the null hypothesis (Prob > 𝜒 = 0.000) that there
is no systematic difference in coefficients between a random effects and fixed effects model.

The next several models exploit the variation found within multiple observations on a single
household to understand the correlates of food security. The model is
Yit = γ + βXit + αi + εit
where Yit = the dependent variable of household i at time t, Xit = a vector of socioeconomic
characteristics and agricultural practices of household i at time t, and αi = unobserved household
fixed effects that are time-invariant. All explanatory variables that are time-invariant, such as
geographic location or relationship to the village chief, are necessarily dropped from this analysis.
As a regression-based test for attrition bias rejects the null of no bias (p-value = 0.000), attritionadjusted weights are used. Column 1 of Table 7-C presents the results of this model when FSIit
is the dependent variable. Many of the coefficients have maintained their sign, even after
accounting for household fixed effects. The indicator for FHH widow is still included here,
although now the coefficient represents the marginal effect of becoming a widow, rather than the
effect of a long-term status as FHH widow. However, it is noteworthy that becoming a FHH
widow has a significantly negative impact on food security. This may be driven by the choice of
dependent variable as the FSI index includes both productive assets and land cultivated, which
are held in smaller amounts by FHHs (Table 5-E). Household size is no longer significant, and

70

neither is the maximum education level. It now appears that having a woman work off-farm is
associated with a higher FSI score.

Column 2 of Table 7-C presents the results of this model when FSUit is the dependent variable.
Attrition bias (p-value = 0.000) is again addressed with attrition-adjusted weights. An F-test of
joint significance of the temperature variables indicates that they are significant in the typical
quadratic manner (𝐹 = 4.78, prob > 𝐹 = 0.009). Intercropping in the previous agricultural season,
having rotated the main crop on a field, and using a no-tillage method of land preparation all
positively impact food sufficiency after harvest. This short-term effect is interesting, as the
impacts of conservation farming techniques are expected to be seen over a longer time horizon
(Haggblade and Tembo 2003).

It might be worrisome that these multi-year indices do not incorporate a variable for the stability
of current food supply or the maintenance of food stocks. To explore how this omission affects
the results, the analysis is repeated for just years 2004 and 2008 when data are available for the
indicator of having maize or cassava in storage from the previous harvest. This indicator is used
in the construction of a brand new FSU index, and the results are presented in column 3 of Table
7-C. This model does not exhibit attrition bias, and it seems clear that the pattern does not
change markedly when this stability variable is included in the FSU index.

71

Table 7-C: Determinants of FSU and FSI with HH fixed effects: 2001, 2004, and 2008

Explanatory variables
HH size
Dependency ratio
Maximum education
Maximum women’s educ.
1=FHH widow
1=HH head recently died
Age of HH head
Age² of HH head
1=Owns radio

(1)
FSI

(2)
FSU

OLS

OLS

0.001
(0.007)
-0.082
(0.081)
0.004
(0.007)
0.008
(0.006)
-0.199***
(0.052)
-0.060
(0.073)
0.042***
(0.011)
-0.000***

-0.056***
(0.006)
0.033
(0.070)
0.001
(0.006)
0.004
(0.006)
-0.081
(0.051)
-0.056
(0.067)
0.022**
(0.010)
-0.000**

-0.067***
(0.009)
0.064
(0.113)
-0.001
(0.009)
0.001
(0.008)
-0.057
(0.077)
-0.081
(0.082)
-0.008
(0.015)
0.000

(0.000)
0.138***
(0.028)

(0.000)
0.087***
(0.024)
0.127***
(0.023)
0.006
(0.032)
0.079
(0.077)
-0.341***
(0.061)
0.228***
(0.040)
0.043***
(0.007)
0.556***
(0.044)
0.229*
(0.125)
-0.704***
(0.057)
0.023***

(0.000)
0.083**
(0.035)
0.091**
(0.035)
-0.038
(0.046)
0.156
(0.105)
-0.734***
(0.086)
0.291***
(0.057)
0.036***
(0.010)
0.536***
(0.070)
0.113
(0.224)
-1.031***
(0.080)
0.009*

(0.003)

(0.005)

1=Transfer network
No. skilled workers
1=Woman works off-farm
Proportion income on food
1=Cash crop
Kgs nitrogen/ ha maize (ln)
1=Grows maize
1=Owns water pump
Proportion crop value sold

(3)
FSU
(2004 and 2008 only)
OLS

0.070*
(0.037)
0.172*
(0.089)
-0.528***
(0.062)
0.465***
(0.051)
0.037***
(0.009)
0.487***
(0.041)
0.608***
(0.226)
0.056
(0.060)

Value of productive assets
(ln)

72

Table 7-C (cont’d)
Hectares cultivated
1=Zero tillage
1=Intercrop
1=Crop rotation
mm important rain (ln)
Important rain²
Season temperature (°C)
Season temperature²
1=2004
1=2008
Constant

Observations

0.030
(0.040)
0.209***
(0.031)
0.216***
(0.028)
1.342*
(0.775)
-0.117*

0.212***
(0.014)
0.075**
(0.036)
0.231***
(0.030)
0.212***
(0.026)
1.217**
(0.615)
-0.094*

0.201***
(0.023)
0.212***
(0.062)
0.382***
(0.051)
0.319***
(0.038)
1.984***
(0.712)
-0.138**

(0.063)
0.570***
(0.204)
-0.012***

(0.050)
0.258
(0.199)
-0.006

(0.058)
0.274
(0.306)
-0.007

(0.004)
0.056**
(0.028)
0.299***
(0.042)
-12.689***
(3.462)

(0.004)
0.047*
(0.026)
0.175***
(0.037)
-7.946**
(3.113)

(0.006)

11,869
11,869
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1

-0.252***
(0.035)
-9.496**
(4.212)
7,846

Because both rainfall and temperature enter the model in a quadratic manner, the coefficients can
be used to determine the “optimal” levels for food sufficiency or food security. In column 2, the
optimal rainfall (from mid-December through February) occurs at 647.68 mm, and the optimal
average season temperature occurs at 21.5°C (Figure 7-1). The household average for seasonal
temperature over the panel years is 23.06°C, which indicates that any additional warming will be
harmful to food sufficiency. This is potentially an important finding in light of the expected
higher average temperatures in southern Africa associated with climate change (Kotir 2011). At
the same time, the optimal temperature in terms of the FSI score (column 1) is 23.75°C, which
suggests that additional warming may be beneficial for food security until a critical point, after
73

which it will be deleterious. Similarly, the household average for rainfall during this interval is
543.82 mm, which indicates that any decrease in rainfall would be harmful to food sufficiency.
Attempts to break apart this impact by agro-ecological zone were not successful, though this will
be discussed as a direction for future research in Chapter 8.

Food sufficiency score

Temperature and FSU
3
2.8
2.6
2.4
2.2
2
15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
Temperature (°C)

3.95
3.90
3.85
3.80
250
300
350
400
450
500
550
600
650
700
750
800
850
900
950
1000

Food sufficiency score

Rainfall and FSU

Rainfall (mm), mid-December - February

Figure 7-1: Seasonal weather and food sufficiency

7.2.2. Persistent effects of food security and food sufficiency
In order to explore the persistent effects of a household’s FSU or FSI score, it is useful to
implement a dynamic panel method that includes as a regressor the lagged FSU or FSI score. In

74

a cross-sectional regression, a lagged dependent food security score will be biased upward in the
presence of serial correlation, which is found to be present in regressions of the FSU and FSI
(𝑢�2008 regressed on 𝑢�2004; p-value = 0.000 in both cases). A fixed effect model can address the

household effects that influence both past and current food security experiences of a household.
However the inclusion of a lagged dependent variable within a fixed effects model results in a
coefficient that is biased downward, a phenomenon known as the Nickell bias (Nickell 1981).
This is especially true in a “small T, large N” context. To overcome this problem, the ArellanoBond Generalized Method of Moments estimator (Arellano and Bond 1991) is used here, in
which a first-differenced model is used in combination with an instrumental variable method to
address the endogeneity of the lagged dependent variable. The instrumental variables used for
the lagged variables include all level terms of the regressors from the previous period, and for
this reason, only one lag can be included with this three-wave panel. The model is
∆Yit = 𝛼 + β1 ∆Yi,t-1+𝛽𝑖 ∆Xit + ∆ εit

where ∆Yit is the differenced dependent variable (e.g. FSU2008 – FSU2004) and ∆Yi,t-1 is the

differenced lagged dependent variable (e.g.FSU2004 – FSU2001). The instrumental variables for
∆Yi,t-1 include all Xi,t-2 (e.g. control variables from 2001). Results are presented in Table 7-D,
and although only the coefficients on the lagged variables are reported, all other HH
characteristics used in Chapter 7 are included here as controls.

In column 1, the FSI is regressed on the household’s lagged FSI score, in addition to the standard
controls from the current period. The lagged value is strongly significant, indicating that the FSI
indeed exhibits persistence and is not determined only by current shocks to the household. In
75

column 2, the FSU score is regressed on the lagged FSU score, and again the coefficient is
significant. In column 3, the FSU is regressed on lagged FSI with the aim of discerning whether
the FSU or the FSI is the stronger determinant of future food sufficiency. The slightly larger
coefficient on the lagged FSI suggests that the extra variables included in the FSI, with the aim
of capturing a household’s vulnerability to future food shortfalls, do a better job of predicting
future food sufficiency. However, the difference seems rather small.
Table 7-D: Effect of lagged FSU and FSI with HH fixed effects: 2001, 2004, and 2008

Explanatory
variables
Lagged FSI

(1)
FSI
GMM

(2)
FSU
GMM

0.248***
(0.031)

0.138***
(0.031)
Observations
4,155
4,155
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1

(3)
FSU
GMM

0.160***
(0.033)

Lagged FSU

4,155

7.2.2 Impact of climate shocks on food sufficiency
In Table 7-E, the impact of rainfall on food sufficiency is studied with the samples restricted to
those households that are poor or not poor (according to the $1-per-day cutoff) in any year.
Although only the coefficients on climate variables are reported, all other HH characteristics
used in Chapter 7 are included here as controls. Among poor households (column 2), the
standard relationship is found in which there is an optimal amount of rainfall that maximizes
food sufficiency. However, among non-poor households (column 1), the coefficients on rainfall
variables are not significant, and F-tests of joint significance confirm that neither rainfall nor
temperature is a significant determinant of food sufficiency. Several other climate variables are
also determinants of FSU for poor households: A longer rainy season in associated with a higher
76

food sufficiency score, and this is probably because a longer rainy season is more likely to
provide adequate rain early in the season to feed a rainfall-sensitive maize crop. In addition, a
longer rainy season provides the opportunity to add mixed beans or other fast-maturing crops to
the household’s harvest. Average season temperature is also significant. These findings
highlight the vulnerability of poor households to climate outcomes, while wealthier households
are buffered from such climate risks.

Table 7-E: Impact of climate shocks by poverty level: 2001, 2004, and 2008
(1)
FSU (Not poor)
OLS

Explanatory variables
mm important rain (ln)
Important rain²
Deviation from mean rainfall (ln)
Start of rainy season
Length of rainy season (days)
Season temperature (°C)
Season temperature²

0.481
(4.078)
-0.018

1.844***
(0.668)
-0.145***

(0.344)
0.043
(0.078)
-0.083
(0.180)
-0.734
(1.673)
-0.774
(1.743)
0.020

(0.054)
-0.008
(0.013)
0.014
(0.022)
0.559***
(0.203)
0.413*
(0.239)
-0.010**

(0.036)
-0.157
(0.101)
F-test rainfall (Prob > F)
0.428
F-test temperature (Prob > F)
0.736
Observations
1,326
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
No. hot dekads

(2)
FSU (Poor)
OLS

(0.005)
0.011
(0.010)

10,732

7.3 Avenue of climate impacts on food security
The final application of the food security indices developed in this thesis is to explore the avenue
of impact of rainfall and temperature shocks on food security. In all regressions in this section,
77

control variables related to household characteristics are included in the econometric analysis,
although the coefficients on these controls are not presented here. The model is
Yit = γ + βWt + βXit + αi + εit
where Wt = a vector of rainfall and temperature variables of the household’s district during the
agricultural season relevant to time t. For all but one regression in this section, a test of attrition
bias fails to reject the null of no bias at a 5 percent level of significance, and attrition-adjusted
weights are therefore used.

In column 1 of Table 7-F, the FSU is regressed on climate and control variables. As we have
seen before, the relationship between rainfall/ temperature and food sufficiency is quadratic, such
that they both have a positive but diminishing impact of food sufficiency until the optimal level
is reached, after which additional rainfall or additional warming decreases food sufficiency. In
columns 2 – 5, this regression is repeated for each element of the FSU index separately, with the
aim of understanding the causal path through which climate affects this index.

The relationship between rainfall and calories is similar to that found with the total index, and
similar coefficients are also seen when the number of crops retained is the dependent variable.
However rainfall seems to have a different relationship with the production of milk/eggs, and is
not a significant determinant of TLU. At the same time, temperature seems to be important for
all elements of the FSU, though there does not seem to be an “optimal” temperature when the
dependent variable is calories. The lesson seems to be that climate affects the various elements
of this food sufficiency index differently. In terms of food security policy development, this
disaggregated analysis is important to identify the interventions that might bolster household
78

resilience to climate shocks. For example, certain livestock management practices may diminish
the relationship between climate shocks and a household’s likelihood of producing milk or eggs.
Table 7-F: Effect of climate on individual elements of the FSU: 2001, 2004, and 2008

Explanatory variables
mm important rain (ln)
Important rain²
Season temperature (°C)
Season temperature²

(1)
FSU

(2)
Calories (ln)

(4)
Milk/
eggs
Probit

(5)
TLU

OLS

(3)
Crops
retained
OLS

OLS
1.373**
(0.653)
-0.107**

1.269**
(0.620)
-0.099**

0.832
(0.582)
-0.051

-0.052
(0.825)
-0.004

-1.571
(8.249)
0.083

(0.053)
0.417*
(0.219)
-0.010**

(0.051)
-0.088
(0.232)
0.000

(0.048)
(0.852)
0.925*** 0.184***
(0.230)
(0.001)
-0.020*** -0.004***

(0.652)
1.068
(1.068)
-0.018

(0.004)

(0.005)

F-test rainfall (Prob > F)
F-test temperature (Prob > F)

0.000

OLS

(0.005)

(0.002)

(0.022)

0.000

0.000

0.443
0.084

Observations: 12,058
OLS coefficients or average partial effects (column 4) with robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1

79

8. DISCUSSION & EVALUATION OF THE FOOD SECURITY INDICES

8.1 Summary of findings
This thesis first offers information on a range of household-level food security indicators in
Zambia. The proportion of households below the money-metric poverty lines (Table 5-A) is far
higher than the proportion of households that are found to be food-energy deficient (Table 5-D).
As the extreme poverty line is based on the cost of a food basket that meets calorie needs without
attention to macronutrients (Table 9-A), it seems these two calculations should be aligned. This
suggests there may be a problem with the food basket prices listed in the 2002-03 LCMS, or
alternatively with the way household income is calculated in the panel data set. Other food
security indicators related to diet quality and stability fail to reveal a clear, linear trend between
2001 and 2008, as the average values fluctuate and seem to closely reflect varying conditions
during the previous agricultural season. For example, the number of months spent without any
stocks from own-production increased markedly between 2004 and 2008. However, the 2006/07
agricultural season was characterized by localized floods (UNICEF 2007), and the
accompanying drop in food security measures should not be interpreted as a general trend in the
well-being of Zambian households. Rather, most explorations of the time trends reveal
stagnancy in the food security situation of Zambian households.

In 2008, it seems that female-headed households (whether defined as FHH widows or FHH nm)
generally experience higher levels of food quantity, although lower average values of other
indicators suggest that they do not unambiguously experience higher levels of food security. For
example, FHHs are less likely to produce milk or eggs, have a garden, or retain a diverse set of
crops from own-production. They tend to gather their income from a less diverse livelihood base,
and are less oriented toward the market as sellers (Table 5-E). It is therefore difficult to
80

determine whether FHHs experience an overall level of food security that differs from other
households in Zambia, and this underscores the potential value of generating a composite index
of food security.

Geographic patterns of food security are immediately evident, as zone IIa consistently exhibits
favorable values of food security indicators, often followed by zone III (Table 5-F). It is not
clear that these variables share the same relationship in each agro-ecological zone, and this will
present some complications when generating a food security index that is comparable over space.
For example, zone I has a relatively high average value of TLU, although other food security
indicators are low, while the reverse is true for zone III. Another interesting pattern is seen in the
distribution of months without food stocks, in which the cassava belt (zone III) exhibits much
higher levels of food stability than other regions. Although the government of Zambia has
recently focused much of its agricultural policy on promoting the production of maize, this
finding highlights the importance of cassava to Zambia’s food security.

The mobility in calorie status evident in the transition matrix from 2004 to 2008 (Tables 5-G and
5-H) suggests that households may be less able to smooth consumption, relative to income, than
is commonly thought. For example, a household in the fourth calorie bracket in 2004 was
equally likely to find itself in the first bracket as to remain in the same bracket. This represents a
wide spectrum of potential welfare for the household, and it calls attention to the importance of
consumption-smoothing mechanisms to the maintenance of household welfare.

81

The FSU and FSI are constructed from a subsample of food security indicators that seem to best
span multiple dimensions of food security, and also seem to have similar meanings across the
four agro-ecological zones. While the FSU is designed to reflect only food sufficiency and
consistency within the previous 12 months, the FSI includes variables that plausibly reflect a
household’s vulnerability to future food shortfalls. When the two indices are calculated for each
household, it becomes clear that FHHs have lower food sufficiency and food security scores, on
average, than the general population (Table 6-D). Figure 6-3 demonstrates that western Zambia
is persistently worse off than other regions. A transition matrix for the FSU (Table 6-F) shows a
lower mobility of households, as compared with the transition matrix based on calories alone.
This is as expected, as the livestock included in the index can take considerable time to
accumulate, and can also serve as a buffer against crop failure.

In general, it is surprising that the FSU and FSI scores in 2004 are not more tightly correlated
with food security indicators in 2008 (Table 6-I). For example, the correlation coefficient
between the 2004 FSI and 2008 calories is just 0.19. This, among other results, casts doubt on
the quality of the FSI or perhaps on the merit of creating a FSI that explicitly accounts for
vulnerability. The relationship between a household’s FSU and its subsequent change in FSU
further complicates the story, as it is unclear whether or how households are moving along a
“path” of food sufficiency. Higher food sufficiency scores are associated with negative
movements along the FSU spectrum (Figure 6-4), and this might imply that a higher FSU score
is not associated with lower vulnerability.

82

The final portion of this thesis employs the food sufficiency and security indices in econometric
analyses to discern the correlates and determinants of food security. Many of the results are as
expected: Assets associated with agricultural production and information access are positively
associated with food sufficiency. Several variables relate to women: The maximum level of
women’s education in the household is sometimes (though not always) associated with higher
FSU and FSI scores, which draws attention to the importance of girls’ education in Zambia. The
coefficient on FHH widow is not significant in cross-sectional regressions, which suggests that
these households do not experience lower FSU scores due specifically to their widowed status.
(Keep in mind that in panel regressions, the variable denoting FHH widowed status references
the impact of a household becoming a FHH widow rather than the characteristic of being a FHH
widow.)

In panel regressions that control for HH fixed effects –such as knowledge, motivation, and
ability– several variables related to farm management may be interpreted causally: Planting a
cash crop is associated with higher food security, even though households may now be
somewhat more dependent on the market to acquire their food. The use of crop rotation,
intercropping, and minimum tillage land preparation are all positively associated with food
sufficiency (Table 7-C). However the impact of improved fallow techniques in the previous
agricultural season seems more ambiguous. In the panel regressions, both rainfall and
temperature continue to be significant determinants of food security in a quadratic manner.

The final exercise relates to the impact of climate shocks on FSU. Rainfall and temperature are
understood to be exogenous, which allows for a causal interpretation of the coefficients (Tables

83

7-E and 7-F). Specifically for poor households, the early-growing season rainfall positively
affects food sufficiency until a maximum point, after which it reduces the FSU score. Given the
occurrence of both droughts and floods in Zambia, this pattern makes sense. At the same time,
the climate variables seem to have little or no explanatory power for households that are not poor.
This affirms the vulnerability of poor households to climate shocks, and points toward the
possible value of weather-indexed insurance or other measures that reduce this vulnerability.
Section 7.3 shows that climate affects each element of the FSU index differently (Table 7-H).
Although informative, this complicates the challenge of developing interventions to improve
household resilience to climate shocks.

8.2 Does using a food security index shed light on the story of food security?
As noted in Chapter 3, the idea of using a multidimensional index of welfare, whether in
reference to poverty or food security, is regarded with some skepticism (Ravallion 2011). The
benefit would seem to be that the food security score encompasses the many dimensions of food
security and reveals a household’s latent food security status. This allows for a straightforward
description of the state of food security in Zambia and enables a statistical analysis that links
explanatory variables with a household’s overall experience of food security.

However, the results of section 7.3 do reveal the shortcomings of such an index: The initial result
shows that seasonal rainfall is a significant determinant of food security in a quadratic
relationship. This is an important finding, as southern Africa is expected to experience drier
growing seasons and more erratic rainfall in the coming decades (Kotir 2011). At the same time,
the policy implication is not entirely clear until the index is decomposed and a regression is run

84

on each of its components separately. The realization that temperature is a determinant of the
number of crops retained suggests that agricultural extension officers may encourage crop mixes
that include less heat-sensitive crops, even if these are less profitable or produce fewer calories
per hectare. The realization that climate shocks in the preceding agricultural season do not seem
to affect TLU indicates that including TLU in a household’s livelihood portfolio may reduce the
household’s vulnerability to climate shocks. In addition, following a season of low rainfall,
households may require assistance with other food sufficiency indicators but not necessarily with
TLU ownership. These avenues of impact and potential intervention would not be readily
deduced when using the index in its composite form.

A final note on the information lost when constructing a food security score is that it is possible
for households with very different characteristics to be given the same score. For example, a
household in zone I may have plenty of livestock along with an extended season without food
stocks. A household in zone III may have few livestock but still experience consistency in its
food supply. The score for each household will be similar, even though the underlying
experience of food security is quite different and merits a unique intervention or policy response.

The inclusion of variables associated with vulnerability (i.e. the extra variables added to the FSI)
seems useful but does not add a great deal to the analysis. For example, the correlation
coefficient between the FSU in 2004 and the FSU in 2008 is 0.33, while the value linking the FSI
in 2004 and the FSU in 2008 was 0.40 (Table 6-I). A priori, it might have been expected that
FSU would be more variable and susceptible to transitory shocks, while the asset and social
network variables included in the FSI render it a better representation of the household’s overall

85

capacity to mitigate negative shocks to food security. However it is not clear that this thesis
would have suffered had it excluded consideration of these indicators of vulnerability, despite
their conceptual appeal.

8.3 Limitations of the study
Several caveats are warranted that have not been adequately emphasized: This data set was not
collected with the intention of measuring consumption or food security, and it is therefore
necessary to identify proxies for even the three dimensions of food security considered here:
quantity, quality, and stability. For example, rather than capturing whether households consume
specific nutrients (e.g. protein) in their diet, the homestead production of milk or eggs, or
ownership of livestock, is used as a proxy for the likely presence of animal protein in the diet.
Yet there may be a weak relationship between household production and consumption of these
products. When measuring food quantity in the household, this study does not account for the
possible draw-down of food stocks from the start of the study period, although whenever
possible it does note whether the household possessed such stocks. Some other possible errors
involved in conflating household production and acquisition of certain products with food
sufficiency have already been outlined in Chapter 5.

This study considers all food security measures at the household level, but is unable to address
the intra-household distribution of resources (i.e. the food security of individual members within
the household). Essentially, it assumes a unitary household model in which food is allocated
according to relative needs rather than bargaining power. However, it is well known that intrahousehold distribution of food often does not follow this pattern (Haddad et al. 1997). Some

86

individuals may be residing in food insecure households but still experience food security, and
vice versa. It is also unclear that the associational or causal relationships identified in this thesis
would apply at the individual level. Take the matter of gender: although the results often
indicate that being a FHH widow does not explain a household’s food sufficiency status, it may
be that at the individual level, gender does affect one’s food sufficiency experience. Indeed, the
data set reveals that widowed or separated women are often absorbed into other households,
suggesting that a survey that is only able to account for the female-headed households that have
“survived” is a poor instrument for gauging the food security impact of being a woman or of
being widowed.

Given that there is no "gold standard” indicator of food security available in the data set, it is
difficult to validate the two indices constructed in this study. Had a module of household
consumption been included in the surveys, the indices’ validity could have been explored
through their correlations with these more accurately measured consumption outcomes (see
Headey and Ecker 2012). Given the data limitations, the merit of the indices is largely based on
their conceptual soundness rather than statistical validation.

8.4 Directions for future research
Future research might build on this exercise by attempting to validate the food sufficiency and
food security indices using another data set that includes additional information on household
experiences, including their consumption and anthropometric outcomes. Even if such
information were available only in a cross-section, it could still be used to test the relationship
between the indices and realized food security experiences. It would also be useful to further

87

explore the causal relationship between certain right-hand-side variables included in the
regression analysis (e.g. social capital, women’s economic opportunities) and a household’s food
security experience. This can only be done when exogeneity of the explanatory variables has
been established. In terms of the impact of different agricultural practices on a household’s
future foods security, a dynamic analysis would have been better suited to discern the longerterm impact of conservation farming techniques.

In order to understand the determinants of food security, it may be more useful to disaggregate
the entire exercise by agro-ecological zone. This would allow the weights on the elements of the
FSU or FSI indices to differ by zone. Although it would no longer allow for the country-wide
comparison of results, it may reveal stronger determinants of food security within a given region.
Another direction for research relates to the differences between households that are categorized
into different trajectories of the FSU and FSI (Table 6-L). Are households on an upward track
systematically different from those on a downward path or those whose scores fluctuate back and
forth? What life events or household decisions seem to characterize households in these
different categories?

A final direction for research is to explore the long-term effects of climate shocks, as in
Hoddinot (2006). This thesis only studies the effect of climate in a given agricultural season on
the immediate food sufficiency outcome in the following year. The results generally indicate
that a level of rainfall or temperature that is too high or too low will negatively affect food
sufficiency. This regression analysis holds constant the household’s ownership of productive
assets at the start of the reporting period during which food sufficiency is measured. However,

88

the significance of these short-term impacts may be overshadowed by their long-term effects on
food security, particularly if households are forced to liquidate their assets in response to a
negative shock.

89

APPENDIX

90

APPENDIX

Figure 9-1: Agro-ecological zones in Zambia
Table 9-A: Cost of monthly food basket and basic needs basket per adult equivalent by
province
Province
Central
Copperbelt
Eastern
Luapula
Lusaka
Northern
Northwestern
Southern
Western

Food index
(Paasche)
0.86
0.87
0.83
0.87
1
0.89
0.9
0.99
0.9

Food basket
(ZMK)
107,995
109,250
104,228
109,250
125,575
111,763
113,018
124,320
113,018

Basic needs basket
(ZMK)
154,277
156,071
148,896
156,071
179,392
159,660
161,454
177,598
161,454

Note: Values are taken from the 2002-03 Living Conditions Monitoring Survey (CSO 2004),
with prices expressed in 2007/08 kwacha. The food basket meets a recommended minimum
calorie requirement of 2,094 kcal/ person/ day. However, the food basket is not constructed to
reflect protein and micronutrient needs. Furthermore, this calorie-per-day recommendation is far
lower than that suggested by Smith and Subandoro (2007), where an individual given a weight of
one adult equivalent is recommended to consume 2,900 calories, assuming a moderate level of
activity.

91

Table 9-B: Income indicator definitions
Construction
Indicator

Gross income

Net income

Income poverty
Food basket
poverty
Basic needs
poverty

Panel
Includes gross value of harvested
crops & vegetable sales; total
income from live & slaughtered
animals; value of eggs and milk
produced; gross income from offfarm wages and business, including
in-kind income; and remittances
Gross income minus total fertilizer
cost and business expenditures
1 if HH has net income/ adult
equivalent/ day less than $1, 0
otherwise
1 if HH cannot afford the cost of a
monthly food basket with net
income
1 if HH cannot afford the cost of a
basic needs basket

Non-panel

Gross income (panel) plus value
fruits/vegetables retained and
value of fish consumed and sold
Gross income (panel) minus total
fertilizer and transportation costs
and business expenditures
1 if HH has net income (nonpanel)/ adult equivalent/ day less
than $1, 0 otherwise

uses non-panel measure of income
uses non-panel measure of income

Table 9-C: Food security indicator definitions
Construction
Indicator
QUANTITY

Total calories acquired

Panel

Non-panel

Sum of: calories produced
from field crops minus
sales; calories from milk
and eggs produced minus
sales; and calories from
staple food purchases, of
food items common to all
years

92

Sum of: calories produced from
field crops and vegetables/ fruits
minus sales; calories from milk
and eggs produced minus sales;
calories from staple food
purchases, of all food items
included in SS 2008; calories from
food items received as in-kind
wages; and calories from food
items received as remittances

Table 9-C (cont’d)

Calories per adult
equivalent per day

Adult equivalents per
household calculated from
formula of Smith and
Subandoro (2007), with
males aged 30 – 60 years
considered to be 1 adult
equivalent

Food energy deficient
(light)

1 if HH meets its calorie
requirements (assuming
light activity level), 0
otherwise

Food energy deficient
(moderate)

1 if HH meets its calorie
requirements (assuming
moderate activity level), 0
otherwise

Food energy deficient
(intense)

1 if HH meets its calorie
requirements (assuming
intense activity level), 0
otherwise

1 if HH meets its calorie
requirements (using non-panel
calculation of calories and
assuming light activity level), 0
otherwise
1 if HH meets its calorie
requirements (using non-panel
calculation of calories and
assuming moderate activity level),
0 otherwise
1 if HH meets its calorie
requirements (using non-panel
calculation of calories and
assuming intense activity level), 0
otherwise

QUALITY
Tropical Livestock Units
(TLU)
Number of different field
crops or vegetables
retained
Household produces eggs
or milk
Household produces some
vegetables/fruits

Units of panel livestock
Units of non-panel livestock
owned at start of the
owned at start of the reporting
reporting period (12 months
period, excluding oxen
prior to interview)
Number of non-panel field crops
Number of panel field crops
and vegetables/fruits retained by
retained by HH after sales
HH after sales
1 if HH produces eggs or
milk, 0 otherwise
1 if HH produces
vegetables or fruits, 0
otherwise (only available
for 2001 and 2008)

STABILITY
Number of months
without any food stocks

(not used in panel analyses)

HH had maize or cassava
in stock at end of
reporting period

1 if HH had either maize or
mature cassava in stock/ in
the field at end of reporting
period, 0 otherwise
93

No. months since the HH last
reported running out of a food
stock

Table 9-C (cont’d)

Hectares cultivated

Value of panel equipment
owned at start of the
reporting period (ploughs,
harrows, and ox-carts)
Number of income sources
from the following groups:
on-farm income,
agricultural wage income,
non-agricultural wage or
salary income, business
income, and remittances
1 if HH gave or received
any cash or commodities
(remittances or gifts), 0
otherwise
Proportion of value of crops
produced that were sold
Number of individuals in
HH that engage in nonagricultural/ skilled income
generating activities
Proportion of HH net
income (panel) spent of
food purchases
Hectares cultivated

Land access (ha)

(not used in panel analyses)

Value of productive assets

Number of income
sources

Transfer network
Proportion of crop value
sold
No. skilled off-farm
workers
Proportion food
expenditure

94

Value of all non-panel equipment,
machinery, and oxen owned at
start of reporting period

Proportion of value of non-panel
crops produced that were sold

Proportion of HH net income (nonpanel) spent of food purchases
Landholding size plus land rented
or borrowed in

Figure 9-2: Timeline of food security indicators
Table 9-D: Other independent variable definitions
Indicator
HH size

Dependency ratio

Maximum education in
HH

Maximum women’s
education
Death of head
Radio

Construction
Panel
Non-panel
Number of HH members,
weighted by time spent at the
homestead in past 12 months
Number of household members
not between ages 15 - 59/ total
household size (weighted by time
spent at home)
Maximum level of education
among prime-age adults. Through
high school, this corresponds to
years of school completed, and
thereafter corresponds to further
academic achievements
Highest level of education
reported by a female in the HH
1 if HH head had died in previous
4 years, 0 otherwise
1 if HH owns radio at time of
interview, 0 otherwise
95

Table 9-D (cont’d)
Women off-farm
Relation to chief

1 if HH had a prime-age woman
working off-farm, 0 otherwise
N/A

Number years in village N/A
Female-headed
household definitions
FHH widow
FHH no adult male

Farming practices
Intercrop
Zero tillage

Crop rotation

Cash crop
Fertilizer
Water pump
Services

Nitrogen per ha maize

Maize HH

1 if HH is related to village
chief, 0 otherwise
No. years HH has resided in
current village

1 if HH identifies as FHH and
head is widowed, 0 otherwise
1 if HH identifies as FHH and
there is no prime-age (15-59)
male member of the HH, 0
otherwise
1 if HH had any fields
intercropped, 0 otherwise
1 if HH had any fields that had
been prepared using zero tillage, 0
otherwise
1 if HH had any fields used for a
different crop from the previous
agricultural season, 0 otherwise
1 if HH produced tobacco or
cotton, 0 otherwise
1 if HH acquired any fertilizer, 0
otherwise
1 if HH owns water pump, 0
otherwise
(not used in panel analysis)

Kgs nitrogen acquired per hectare
maize planted (0 for households
that do not plant maize). This
assumes that households have
applied compound D as basal
fertilizer and urea as top dressing.
1 if household planted any maize,
0 otherwise

96

1 if HH receive advice on amount
of seed or fertilizer to use;
minimum tillage techniques; and
crop residues; optimal planting
date; and nitrogen-fixing crop
rotation, 0 otherwise

Table 9-E: Rainfall and temperature variables
Variable
Season rainfall (mm)
Deviation from long-term mean

Construction
Total precipitation recorded from November through March
Absolute deviation from average growing season rainfall
(1990/91 – 2009/10)
Length of growing season (days) Number of days from first dekad with > 20 mm rainfall to last
dekad with >20 mm rainfall
Rain start
Number of dekads after November 1 until > 20 mm rainfall
within a dekad
Season variance
Variance in dekadal rainfall over the season (November
through March)
Rain stress
Number of 20-day periods in growing season with < 40 mm
rainfall
Important rainfall (mm)
Total precipitation from mid-December through February
Average temperature (Celsius)
Average dekadal temperature over the growing season
No. hot dekads
Number of dekads during growing season with average
temperature over 25°C
Note: Missing observations were imputed using an average of the observations of nearby
meteorological stations of similar altitude. In total, 13.75 percent (245/1,782) of dekadal rainfall
observations were imputed in this manner, and 34.23 percent (616/1,782) of temperature
observations were imputed. Though the rate of missing temperature observations is high, it was
determined that because temperature is less localized than rainfall, imputation would likely
produce suitable temperature estimates.
Table 9-F: Calorie content of common foods consumed in Zambia
Food items
purchased/received
roller meal
breakfast meal
maize meal from
grinding mill
cassava chips
cassava flour
wheat flour
wheat bread
buns/ fritters

Miscellaneous
milk (per liter)

kcal/kg
3680
3530
3530
3570
3510
3640
2330
3300

kcal
790

Field crops kcal/kg
maize
3570
sorghum
3500
rice
millet
sunflower
groundnuts
soya beans
irish
potatoes
mixed
beans
bambara
nuts
cowpeas

97

3440
3410
4860
5490
4050
820

Vegetables and
fruits
oranges
bananas
pineapple
guavas
avocado
watermelon
mangoes

kcal/kg
430
880
470
640
1210
220
600

tangerines

490

3670

lemons

290

3670
3380

grapefruit
fresh groundnuts

340
5490

Table 9-F (cont’d)
egg (per unit)

140

fish (per kg)

3340

oil (per kg)

8840

velvet
beans
sweet
potatoes
raw
cassava

3600

cabbage

260

1210

rape

220

1490

spinach
tomato
onion
okra
eggplant
pumpkin
chilies
chomoli
cauliflower
carrots
lettuce
green beans
green maize
impwa
pumpkin leaves
sweet potato
leaves
cassava leaves
bean leaves
chinese cabbage
sugarcane

260
210
410
360
320
5110
3120
220
250
400
200
340
1520
400
270
490
910
360
250
620

Source: Lukmanji et al. 2008; Wu Leung et al. 1968
Table 9-G: Demographic and agricultural variables for 2001, 2004, and 2008
Demographic variables
HH size
Dependency ratio
Age of HH head
Maximum education (level)
Maximum woman's education (level)
1=HH includes a skilled off-farm
woman worker
1=Death of head in previous 4 years
1=HH self-identifies as femaleheaded
1=FHH widow
1=FHH with no adult male

2001
5.84
0.52
45.53
7.03
5.00

(2.96)
(0.21)
(15.26)
(3.36)
(3.38)

2004
5.91
0.45
48.50
7.18
5.31

(2.91)
(0.22)
(15.01)
(3.41)
(3.40)

2008
5.77
0.55
48.33
7.54
5.61

0.02
0.01

0.03
0.05

0.05
0.04

0.23
0.10
0.10

0.24
0.13
0.07

0.24
0.14
0.11

98

(2.80)
(0.22)
(15.30)
(3.44)
(3.42)

Table 9-G (cont’d)
Agricultural practices & access to information
(proportions of cropping HHs)
1=Fertilizer
0.23
1=Cash crop
0.09
1=Water pump
0.01
1=Services
--1=HH owns radio (of all HHs)
0.34
1=Sell crop
0.65
1=Plant maize
0.78
1=Zero tillage
0.08
1=Intercrop
0.35
1=Crop rotation
0.53

99

0.29
0.18
0.01
0.47
0.47
0.64
0.81
0.10
0.20
0.68

0.31
0.16
0.01
0.50
0.58
0.65
0.77
0.03
0.12
0.71

Table 9-H: Demographic and agricultural variables across subgroups in 2008
Demographic variables
All HHs
FHH widow
FHH nm
HH size
5.77 (2.80) 4.42 (2.69) 3.13 (2.06)
Dependency ratio
0.55 (0.22) 0.61 (0.26) 0.69 (0.28)
Age of HH head
48.33 (15.30) 59.24 (12.73) 53.55 (16.64)
Maximum education (level)
7.54 (3.44) 7.01 (3.39) 5.27 (3.61)
Maximum woman's education
(level)
5.61 (3.42) 5.04 (3.62) 4.43 (3.66)
1=HH includes a skilled off-farm
woman worker
0.05
0.06
0.11
1=Death of head in previous 4
years
0.04
0.20
0.12
1=HH self-identifies as femaleheaded
0.24
1=FHH widow
0.14
1=FHH with no adult male
0.11
Agricultural practices & access to information
Proportions of cropping
households
1=Fertilizer
0.31
0.25
0.17
1=Cash crop
0.16
0.09
0.06
1=Water pump
0.01
0.01
0.00
1=Services
0.53
0.52
0.43
1= HH owns radio (of all HHs)
0.58
0.33
0.26
1=Sell crop
0.65
0.55
0.49
1=Plant maize
0.77
0.75
0.66
1=Zero tillage
0.03
0.03
0.03
1=Intercrop
0.12
0.11
0.12
1=Crop rotation
0.71
0.70
0.65
Table 9-H (cont’d)
Demographic variables
HH size
Dependency ratio
Age of HH head
Maximum education
(level)
Maximum woman's
education (level)
1=HH includes a skilled
off-farm woman worker
1=Death of head in
previous 4 years
1=HH self-identifies as
female-headed
1=FHH widow

Zone I
Zone IIa
Zone IIb
Zone III
5.88 (2.62) 6.06 (3.01) 5.46
(2.61) 5.56 (2.63)
0.57 (0.21) 0.56 (0.21) 0.55
(0.22) 0.54 (0.22)
49.23 (15.88) 48.93 (15.43) 49.41 (16.52) 47.49 (14.84)
6.78

(3.44)

7.80

(3.56)

6.59

(3.42)

7.58

(3.29)

5.23

(3.13)

5.87

(3.56)

4.99

(3.39)

5.55

(3.32)

0.03

0.06

0.03

0.06

0.05

0.05

0.03

0.04

0.24
0.13

0.24
0.03

0.31
0.04

0.23
0.03

100

Table 9-H (cont’d)
1=FHH with no adult
male
0.11
0.08
Agricultural practices & access to information
Proportions of cropping
households
1=Fertilizer
0.06
0.43
1=Cash crop
0.08
0.35
1=Water pump
0.01
0.01
1=Services
0.50
0.69
1= HH owns radio (of all
HHs)
0.44
0.64
1=Sell crop
0.45
0.70
1=Plant maize
0.95
0.93
1=Zero tillage
0.00
0.06
1=Intercrop
0.13
0.02
1=Crop rotation
0.55
0.78

0.15

0.12

0.06
0.01
0.00
0.27

0.27
0.03
0.00
0.50

0.37
0.38
0.77
0.00
0.08
0.39

0.58
0.68
0.62
0.01
0.21
0.74

Table 9-I: Example of food sufficiency score calculation
Multi-year food sufficiency index (FSU)
Component
Factor loading
Calories (ln)
0.64
No. different crops retained
0.6
Tropical livestock units
0.33
Household produces milk or
0.35
eggs

z-scores

HH
1
1
1

Year
2001
2004
2008

Calories
0.76
2.35
3.73

Crop
count
0.78
1.58
-0.02

TLU
1.72
-0.07
-0.23

Milk/
eggs
1.45
1.45
-0.69

Food
Formula for calculation sufficiency
of food sufficiency score
score
2.04
0.64²·(calorie z-score) +
2.94
0.60²·(crop count z2.06
score) + 0.33²·(TLU zscore) + 0.35²·(milk/eggs
z-score)

101

REFERENCES

102

REFERENCES
Abeyasekara, S. 2005. Statistical analysis of survey data. In Household Sample Surveys in
Developing and Transition Countries. New York, NY: United Nations.
Alinovi, L., E. Mane, and D. Romano. 2009. Measuring household resilience to food insecurity:
Application to Palestinian households. FAO‐ESA Working Paper. FAO, Agricultural and
Development Economics Division, Rome.
Alkire, S., and J. Foster. 2007. Counting and multidimensional poverty measures. OPHI
Working Paper Series 7. University of Oxford Poverty and Human Development Initiative,
Oxford.
Arellano, M., and S. Bond. 1991. Some tests of specification for panel data: Monte Carlo
evidence and an application to employment equations. The Review of Economic Studies, 58:
277 – 297.
Babu, S., and P. Sanyal. 2009. Food Security, Poverty, and Nutrition Policy Analysis.
Amsterdam: Academic Press.
Barrett C. 2002. Food security and food assistance programs. In B. L. Garner and G. C. Rausser,
eds. Handbook of Agricultural Economics. Amsterdam: Elsevier Science.
Barrett, C. 2008. Smallholder market participation: Concepts and evidence from eastern and
southern Africa. Food Policy 33(4):299-317.
Barrett, C. 2010. Measuring food insecurity. Science 327: 825–828.
Beegle, K., J. de Weerdt, J. Friedman, and J. Gibson. 2012. Methods of household consumption
measurement through surveys: Experimental results from Tanzania. Journal of Development
Economics, 98 (1): 3-18.
Campbell, C. 1991. Food insecurity: a nutritional outcome or a predictor variable? The Journal
of Nutrition 121 (3): 408–415.
Carter, M., and C. Barrett. 2006. The economics of poverty traps and persistent poverty: An
asset-based approach. Journal of Development Studies. 42 (2): 178-199.
Cavatassi R., B. Davis, and L. Lipper. 2004. Estimating Poverty Over Time and Space:
Construction of a Time-Variant Poverty Index for Costa Rica. ESA Working Paper n. 04-21,
FAO, Rome, Italy.
Christiaensen, L., and R. Boisvert. 2000. On Measuring Household Food Vulnerability: Case
Evidence from Northern Mali. Working Paper. Department of Agricultural, Resource, and
Managerial Economics, Ithaca, New York: Cornell University.

103

Coates, J., P. Webb, and R. Houser. 2003. Measuring Food Insecurity: Going Beyond Indicators
of Income and Anthropometry. Washington, D.C.: Food and Nutrition Technical Assistance
Project, Academy for Educational Development.
Committee on World Food Security (CFS). 2011. Measuring food insecurity: Meaningful
concepts and indicators for evidence-based policy-making. Round Table on Monitoring Food
Security, Technical background paper. September 12-13. Rome.
Central Statistics Office (CSO). 2004. Living Conditions Monitoring Survey Report 2002-03.
Government of Zambia, Lusaka.
Chen, S., M. Ravallion and P. Sangruala. 2008. Dollar a day revisited. World Bank Policy
Working Paper No.4620, World Bank, Washington, DC.
Dasgupta, A., and A. Baschieri. 2010. Vulnerability to climate change in rural Ghana:
mainstreaming climate change in poverty-reduction strategies. Journal of International
Development 22: 803-820.
Demeke, A., A. Kiel, and M. Zeller. 2011. Using panel data to estimate the effect of rainfall
shocks on smallholders’ food security and vulnerability in rural Ethiopia. Climatic Change
108 (1): 185–206.
Devereux S., and S. Maxwell. 2001. Food Security in Sub-Saharan Africa. London: Intermediate
Technology Development Group Publishing.
Feleke, S., R. Kilmer, and C. Gladwin. 2005. Determinants of food security in Southern Ethiopia
at the household level. Agricultural Economics 33: 351–363.
Filmer D., and L. Pritchett. 2001. Estimating wealth effects without expenditure data-or tears: An
application to educational enrollments in states of India. Demography 38(1): 115-132.
Food and Agriculture Organization of the United Nations (FAO). 2002. International Scientific
Symposium on Measurement and Assessment of Food Deprivation and Undernutrition.
Executive Summary. June 26-28. Rome: FAO/Food Insecurity and Vulnerability
Information and Mapping System.
FAO Food Insecurity and Vulnerability Information and Mapping Systems (FIVIMS). 2013.
Website accessed online at http://www.fivims.org/.
FAO Global Information and Early Warning System. 2013. Website accessed online at
www.fao.org/giews/english/index.htm.
Garrett, J., and M. Ruel. 1999. Are determinants of rural and urban food security and nutritional
status different? Some insights from Mozambique. World Development 27(11): 1955-1975.

104

Haddad, L., J. Hoddinott, and H. Alderman. 1997. Intrahousehold Resource Allocation in
Developing Countries: Models, Methods, and Policy. Baltimore, MD: Johns Hopkins
University.
Haggblade, S., and G. Tembo. 2003. Conservation farming in Zambia. Discussion Paper No.
108. International Food Policy Research Institute, Environment and Production Technology
Division.
Headey, D., and O. Ecker. 2012. Improving the measurement of food security. Discussion Paper
No. 01225. Washington, D.C.: International Food Policy Research Institute.
Hoddinott, J. 2006. Shocks and their consequences across and within households in rural
Zimbabwe. Journal of Development Studies 42(2): 301-321.
Hoddinott, J. and A. Quisumbing. 2003. Methods for microeconometric risk and vulnerability
assessments. Social Protection Discussion Paper Series 29138, The World Bank.
Hulme, D., and A Shepherd. 2003. Conceptualizing chronic poverty. World Development 31(3):
403-423.
Jain, S. 2006. An empirical economic assessment of impacts of climate change on agriculture in
Zambia. Discussion Paper No. 27. University of Pretoria, South Africa: Center for
Environmental Economics and Policy in Africa (CEEPA).
Kambikambi, T. 2006. CROPWAT exercise report for Zambia. Discussion Paper No. 39.
University of Pretoria, South Africa: Center for Environmental Economics and Policy in
Africa (CEEPA).
Kaiser, H. 1974. An index of factor simplicity. Psychometrika 39: 31–36.
Kennedy, E., and P. Peters. 1992. Household food security and child nutrition: The interaction of
income and gender of household head. World Development 20(8): 1077–85.
Kotir, J. H. 2011. Climate change and variability in Sub-Saharan Africa: a review of current and
future trends and impacts on agriculture and food security. Environmental Development and
Sustainability 13: 587-605.
Løvendal, C., and M. Knowles. 2005. Tomorrow’s Hunger: A Framework for Analyzing
Vulnerability to Food Insecurity. Agricultural and Development Economics Division, FAO,
ESA Working Paper No. 05-07.
Lukmanji, Z., E. Hertzmark, N. Mlingi, V. Assey, G. Ndossi, and W. Fawzi. 2008. Tanzania
Food Composition Tables, First Edition. Dar es Salaam, Tanzania, and Boston, MA:
Muhimbili University of Health and Allied Sciences, Tanzania Food and Nutrition Centre,
and Harvard School of Public Health.

105

Maredia, M. 2009. Improving the proof: evolution of and emerging trends in impact assessment
methods and approaches in agricultural development. Discussion Paper No. 929.
Washington, D.C.: International Food Policy Research Institute (IFPRI).
Maxwell, D., C.Ahiadeke, M. Armar-Klemesud, C. Levin, S. Lamptey, and G. Mary. 1999.
Alternative food-security indicators: revisiting the frequency and severity of ‘coping
strategies’. Food Policy 24: 411–429.
Migotto M., B. Davis, C. Carletto, and K. Beegle. 2005. Measuring food security using
respondents' perception of food consumption adequacy. FAO-ESA Working Paper No. 0510. FAO, Agricultural and Development Economics Division, Rome.
Misselhorn, A. 2005. What drives food insecurity in southern Africa? A meta-analysis of
household economy studies. Global Environmental Change 15: 33–43.
Morton, J. 2007. The impact of climate change on smallholder and subsistence agriculture.
Proceedings of the National Academy of Sciences 104 (50): 19680–19685.
Naschold, F. and C. Barrett. 2011. Do short-term observed income changes overstate structural
economic mobility? Oxford Bulletin of Economics and Statistics 73(5): 705-717.
Nickell, S. 1981. Biases in dynamic model with fixed effects. Econometrica 49: 1417–26.
Ravallion, M. 2011. On multidimensional indices of poverty. The World Bank Development
Research Group. World Bank Policy Research Working Paper 5580.
Renzaho, A., and D. Mellor. 2009. Food security measurement in cultural pluralism: Missing the
point or conceptual misunderstanding? Nutrition 26 (1): 1–9.
Riely, F., N. Mock, B. Cogill, L. Bailey, and E. Kenefick. 1999. Food security indicators and
framework for use in the monitoring and evaluation of food aid programs. FANTA
Background Paper. Washington, D.C.: Food and Nutrition Technical Assistance Project
(FANTA), Academy for Educational Development.
Saasa, O., D. Chiwele, F. Mwape, and J. Keyser. 1999. Comparative economic advantage of
alternative agricultural production activities in Zambia. USAID Technical Paper No. 104.
Siegel, P. 2008. Profile of Zambia's smallholders: Where and who are the potential beneficiaries
of agricultural commercialization? World Bank Africa Region Working Paper Series No.
113.
Siegel, P., and J. Alwang. 2005. Poverty reducing potential of smallholder agriculture in Zambia:
Opportunities and constraints. World Bank Africa Region Working Paper Series No. 85.

106

Smith, L., and A. Subandoro. 2007. Measuring food security using household expenditure
surveys. Food Security in Practice series. Washington, D.C.: International Food Policy
Research Institute (IFPRI).
Swindale, A., and P. Bilinsky. 2006. Development of a universally applicable household food
insecurity measurement tool: process, current status, and outstanding issues. The Journal of
Nutrition 136 (5): 1449–1452.
Thompson, H., L. Berrang-Ford, and J. Ford. 2010. Climate change and food security in subSaharan Africa: a systematic literature review. Sustainability 2 (8): 2719–2733.
Tschirley, D., and M. Weber. 1994. Food security strategies under extremely adverse conditions:
The determinants of household income and consumption in rural Mozambique. World
Development 22:2 (159-173).
The United Nations Children's Fund (UNICEF). 2007. Zambia: Immediate Needs for Children
and Women Affected by the Floods. Accessed online at:
www.unicef.org/infobycountry/files/Zambia_Immediate_Needs_6Mar07.pdf.
Vyas, S., and L. Kumaramayake. 2006. Constructing socio-economic status indices: how to use
principal components analysis. Health Policy and Planning 21 (6): 459-468.
Webb, P., and B. Rogers. 2003. Putting the ‘‘in’’ back into food insecurity. Occasional Paper
No. 1. Washington, D.C.: Office of Food for Peace/USAID.
Webb, P., J. Coates, E. Frongillo, B. Rogers, A. Swindale, and P. Bilinsky. 2006. Measuring
household food insecurity: why it’s so important and yet so difficult to do. The Journal of
Nutrition 136: 1404–1408.
Wolfe, W., and E. Frongillo, Jr. 2000. Building household food security measurement tools from
the ground up. FANTA Background Paper, Washington, D.C.: Food and Nutrition Technical
Assistance Project (FANTA), Academy for Educational Development.
Wooldridge, J. 2002. Econometric Analysis of Cross Section and Panel Data. Cambridge, MA:
Massachusetts Institute of Technology Press.
World Bank. 2011. Missing Food: The case of Postharvest Grain Losses in sub-Saharan Africa.
Report No. 60371.
World Food Program (WFP). 2009. Comprehensive Food Security & Vulnerability Analysis
Guidelines. Rome.
World Health Organization (WHO). 1985. WHO Technical Report Series No. 724. Geneva:
WHO.

107

Wu Leung, W., F. Busson, and C. Jardin. 1968. Food composition table for use in Africa. US
Dept. of Health, Education, and Welfare and the FAO. Bethesda, MD and Rome.
Ziervogel, G., A. Nyong, B. Osman, C. Conde, S. Cortés, and T. Downing. 2006. Climate
variability and change: implications for household food security. AIACC Working Paper No.
20. Assessments of Impacts and Adaptations to Climate Change (AIACC).

108