Vous êtes sur la page 1sur 25

Agrekon

Agricultural Economics Research, Policy and Practice in Southern Africa

ISSN: 0303-1853 (Print) 2078-0400 (Online) Journal homepage: http://www.tandfonline.com/loi/ragr20

Multidimensional Household Food Security


Measurement in Rural Zambia

Ayala Wineman

To cite this article: Ayala Wineman (2016) Multidimensional Household Food Security
Measurement in Rural Zambia, Agrekon, 55:3, 278-301, DOI: 10.1080/03031853.2016.1211019

To link to this article: http://dx.doi.org/10.1080/03031853.2016.1211019

Published online: 13 Sep 2016.

Submit your article to this journal

Article views: 28

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at


http://www.tandfonline.com/action/journalInformation?journalCode=ragr20

Download by: [Michigan State University] Date: 27 February 2017, At: 14:25
MULTIDIMENSIONAL HOUSEHOLD FOOD
SECURITY MEASUREMENT IN RURAL
ZAMBIA

Ayala Wineman
International Development in the Department of Agricultural,
Food, and Resource Economics,
Michigan State University, USA
Email: wineman1@msu.edu

ABSTRACT
Food security is recognised as a multifaceted condition of complex causality,
and 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 “concept-
to-measurement” gap in food security by building an index that spans these three
dimensions. A panel data set from rural Zambia is used for descriptive analysis of
food security indicators in 2001, 2004 and 2008. A multidimensional index of food
security for rural Zambia is then developed using principal component analysis. We
use this index to explore the spatial patterns of food security over time and to assess
correlates of food security and impacts of climate shocks. Results indicate that both
rainfall and temperature have a significant impact on a household’s food security
score, though not for each individual component of the index. The paper concludes
with a discussion of the merits and shortcomings of developing a composite food
security index.

Keywords: climate, food security measurement, principal component analysis, Zambia

JEL Classification: C38; Q12; Q54; I32

university
of south africa

Agrekon DOI: 10.1080/03031853.2016.1211019


Volume 55 | Number 3 | 2016 Print ISSN 0303-1853 | Online 2078-0400
pp.278-301 ©Agricultural Economics Association of South Africa

278
Wineman Multidimensional household food security

1. INTRODUCTION
This paper draws from 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.
Along these lines, measurement is necessary to understand where food insecurity is
concentrated and whether a situation is growing better or worse. Measurement can also
bring to light the nature of food insecurity, detailing whether it tends to be chronic or
transitory and whether it is a problem of food availability or economic access, of diet
quantity or quality.
This paper aims to broaden our understanding of food security measurement and
dynamics, first by analysing the relationships among various food security indicators and
tracking these indicators over time. It then develops two relatively simple measures of
food security in the form of composite indices that incorporate indicators of its multiple
dimensions. The food sufficiency index (FSU) reflects a household’s realised welfare
over the previous year, while the food security index (FSI) incorporates forward-looking
indicators of future vulnerability. These indices are applied to household data from rural
Zambia to address the following questions: Where are food insecure households found
within Zambia? What are the correlates of food insecurity, and through what avenues do
climate shocks affect food security? Few existing studies construct a multidimensional
measure of household food security to address these questions, and to our knowledge,
no other study draws from an empirical exercise to thoughtfully reflect on the merits
and shortcomings of using a single composite index. This paper seeks to fill this gap in
the literature.
The paper is organised as follows: Section 2 offers background information on food
security measurement and the agro-ecology of Zambia. Section 3 describes the data
sources, and section 4 provides descriptive statistics of food security indicators. Section
5 details the construction of the food sufficiency and food security indices. Section 6
presents several econometric applications of the food sufficiency index, and section 7
offers a summary of lessons learned.

2. BACKGROUND

2.1 Food security measurement


The definition of food security is generally understood as a situation whereby “all people,
at all times, have physical and economic access to sufficient, safe and nutritious food
to meet their dietary needs and food preferences for an active and healthy life” (FAO,
1996). By 2001, this definition was expanded to also include social access (FAO, 2002).
Conversely, food insecurity exists when people do not have such access, and households

279
Wineman Multidimensional household food security

experience insecurity whenever they are unable to absorb or mitigate the impact of a
negative food shock. Food insecurity is related to, yet is distinct from, concepts such
as poverty and malnutrition (Webb et al., 2006), and the failure of early attempts to
alleviate food insecurity has arguably been due to their overwhelming emphasis on
food availability. While food security clearly depends on agricultural conditions and
aggregate food production, it also depends on socio-economic conditions, including
the distribution, access and affordability of food. In fact, aggregate food availability
is a poor predictor of other indicators of food insecurity, and such insecurity “does not
arise exclusively – or even predominantly – because of covariate shocks to an entire
population” (Barrett, 2002). Thus, the correlates of food insecurity are likely to be found
at the level of households and individuals.
Most works now consider there to be four interlinked components embedded in the
definition of food security: (1) availability of food in terms of its physical presence in a
given country or region; (2) access to food as reflected by the ability to obtain food from
own-production or through purchases, gifts or borrowing; (3) utilisation, or the ability
to derive full biological benefits from food, based on food safety and personal health;
and (4) stability of access, independent of shocks or cyclical patterns (Carletto et al.,
2012). However, the availability and access components cannot be readily distinguished
at the household level, where data do not reveal the extent to which food is available in
local markets, and household behaviours regarding food acquisition do not necessarily
reflect economic access to food (Jones et al., 2013). At the same time, in locales without
active food markets or strong safety nets, availability and access are hardly two separate
dimensions. Breaking with the four-pillar convention, Coates (2013) identifies 5 slightly
different dimensions of food security, including food sufficiency, nutrient adequacy,
cultural acceptability, safety, and certainty/stability. In this way, the experience of food
security can be isolated from its potential causes and consequences.
This paper presents a slightly different framework to describe how food insecurity
is experienced and managed at the household level. We consider there to be three
components of household food security: (1) food quantity available in the household;
(2) food quality as captured by dietary diversity and the presence of important nutrients;
and (3) the stability of adequate food supplies, encompassing both realised regularity of
food supply and the likelihood of future food shortfalls.1 This is not intended to be the
most comprehensive framework, as it overlooks, inter alia, themes of food safety and
cultural suitability. Rather, this framework is most appropriate for the present analysis,
subject to data limitations typical of many rural household surveys.
While accurate measurement of food security is essential for effective research
and well-targeted policies and programmes, there is no standard methodology for
measurement, and despite an improved theoretical understanding of food security, the

1 According to the framework presented by Leroy et al. (2015), quantity and quality are among the “components” of
the availability, access and stability dimensions of food security. In this paper, we use the terms “component” and
“dimension” interchangeably, in reference to the quantity, quality and certainty/stability of food supply.

280
Wineman Multidimensional household food security

FAO notes that there exists no “perfect single measure that captures all aspects of food
insecurity” (FAO, 2002). The absence of such a gold standard makes it unreasonable
to use a single benchmark as a proxy for food security. In light of its multidimensional
nature, it is generally agreed that a suite of indicators is needed for the assessment of
food security (Carletto et al., 2012), and different measures are sometimes used for the
triangulation or cross-referencing of results (Maxwell et al., 2014). At the same time,
several papers have combined various food security indicators into a single composite
index, an approach that seems to build on the literature of multidimensional poverty
indices (Alkire and Foster, 2011). The development of a composite index allows for the
ranking of different countries, regions or households according to the severity of food
insecurity. The World Food Program generates a food security and vulnerability index
for household profiling (WFP, 2009), and in Ethiopia, Demeke et al. (2011) also create
a food security index composed of household-level variables related to food availability,
access and vulnerability.2,3
Though the argument for creating such an index is compelling, there may be
drawbacks to combining diverse factors into one score. Referring to multidimensional
poverty indices, Ravallion (2011) expresses scepticism of reliance on a single poverty
index for policy making. The author argues that policy makers can already refer to
multiple indicators that capture different aspects of poverty, while the use of a single
index necessitates assumptions regarding which aspects of poverty should be included
and how they should be weighted. The same argument can be made regarding the
construction of a multidimensional food security index. Coates (2013) observes that
lumping together various elements of food security risks obscuring the relationships
between these individual elements and their respective determinants. The term “food
security”, the author notes, is “potent as a whole for advocacy, but perhaps more usefully
deconstructed into its constituent parts for the purposes of systematic assessment, action,
and evaluation”. This paper uses quantitative methods to explore whether collapsing the
components of food security into a single index can, in fact, be useful for monitoring
and analysis.

2.2 Agro-ecology and rural livelihoods in Zambia


Zambia is a landlocked country in southern Africa characterised by relatively low
population densities, where roughly 45% of the population live in rural areas and depend
on agriculture for their livelihoods (Jain, 2007). Zambia’s farmers can be grouped into
three categories, including small-scale (with up to 5ha of farm land), emergent (5–20ha),

2 More commonly, multidimensional food security indices have been applied to country-level analyses. A review
and assessment of such indices is provided by Masset (2011).
3 The widely used Household Food Insecurity Access Scale collects, in a simple survey module, information on
quantity, quality and recent stability of food access (Coates et al., 2007). Although composed differently from our
food security index, the lessons of this paper should be useful in evaluating these more experiential/subjective
composite measures of food security.

281
Wineman Multidimensional household food security

and large-scale (≥ 20ha) farmers. As of 2003, approximately 94% were small-scale


farmers (Siegel and Alwang, 2005) dependent almost entirely on rain-fed production.
Zambia is divided into four agro-ecological regions distinguished by distinct rainfall
patterns (Figure 1). Region I, located in the south, is relatively dry with unpredictable
and poorly distributed rainfall and limited potential for crop production. Region IIa
covers the central-eastern part of the country and has the highest agricultural potential,
with fertile soil and rains that are evenly distributed throughout the growing season.
Region IIb is characterised by sandy soils and a high risk of drought. Finally, region III
in the north experiences ample rainfall, though this pattern has produced leached and
acidic soils (Jain, 2007). The farming systems in Zambia also vary by agro-ecological
region: Region I is suitable for drought-tolerant crops (e.g., cotton or sorghum), though
poor soils and unreliable rainfall make farming risky. Region IIa, the most densely
populated region, is characterised by a maize regime and is also the most mechanised
and commercialised region, with better access to infrastructure and higher use of inputs.
Region IIb exhibits substantially lower capacity for crop production, while region III is
characterised by a dual cassava and maize regime, with farmers tending to use low-input
shifting and semi-permanent cultivation techniques (Siegel and Alwang, 2005).

Figure 1: Agro-ecological regions in Zambia

Year-to-year variability in rainfall and other climatic conditions are important


determinants of crop output in most parts of Zambia. Drought has been the biggest
shock to food security during the last two decades (Jain, 2007), with large shortfalls
in maize yield consistently occurring in seasons with below normal rainfall. At the
same time, Zambia sometimes experiences heavy localised floods that also threaten
agricultural production. The climate outlook for southern Africa is characterised by a
rise in temperatures and an increasing frequency and severity of extreme rainfall events
(Kotir, 2011). Thus, the general consensus is that climate change will act as a multiplier
of existing threats to food security in the region.

282
Wineman Multidimensional household food security

3. DATA SOURCES
This analysis draws from a nationally representative panel data set of 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 four 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 comprises both the 1999/2000 Post-Harvest Survey conducted by the
Zambian Central Statistical Office (CSO) and Ministry of Agriculture and Cooperatives
(MACO), and the 2001 CSO/MACO/Michigan State University Food Security Research
Project (FSRP) Supplemental Survey. The second and third waves are Supplemental
Surveys. The number of households interviewed was 6 922 (2001), 5 419 (2004) and
8 094 (2008). In total, 4 286 households were re-interviewed in all three waves of the
panel.
The calorie content of food items is taken from Wu Leung et al. (1968) and Jumanji
et al. (2008). Historical rainfall and temperature data are obtained from the records of
meteorological stations run by the Zambia Meteorological Department, with each district
matched to a nearby meteorological station to estimate seasonal climate variables. In
this article, panel analyses include the balanced panel of households interviewed in all
three waves, and population weights are included in all relevant analyses. Regression-
based tests for attrition bias (Wooldridge, 2002: 577) fail to reject the null hypothesis
of no bias for our panel regressions, indicating that patterns of attrition do not bias the
results. Monetary values are inflated to 2007/08 Zambian kwacha (ZMK) using the
consumer price index.

4. DESCRIPTIVE STATISTICS OF FOOD SECURITY


INDICATORS
The available food security indicators are defined in Table 1, grouped according to
dimension (quantity, quality or stability). Measures of quantity include the number
of calories available to the household per adult equivalent (AE) per day, and binary
indicators of whether this number falls below the household’s minimum requirements.
The cut-offs are determined with reference to age- and gender-specific calorie
requirements (UNU/ WHO/ FAO, 2004, as summarised in Smith and Subandoro, 2007)
for three different activity levels. Thus, the calorie requirement for a household depends
on its demographic composition. For a light, medium, or intense level of activity, one
adult equivalent requires 2 500, 3 000 or 3 550 calories per day, respectively.4
Indicators of quality aim to capture the diversity of field crops and the likely presence
of meat, other livestock products, and vegetables or fruits in the household diet. Several

4 Mothers of children under one year, assumed to be breastfeeding, require approximately 567.5 additional calories
per day (Smith and Subandoro, 2007).

283
Wineman Multidimensional household food security

authors have documented a link between livestock ownership and increased dietary
diversity (Jodlowski et al., 2016). A correlation between home gardens and improved
micronutrient consumption has similarly been observed (Galhen et al., 2013), along
with a correlation between on-farm crop diversity and household diet quality (Mazunda
et al., 2015).
Stability, the third dimension of food security, first covers the regularity of food
supply over the past year with an indicator for whether a household retains maize or
cassava in stock at the time of interview, and the number of months the household has
been without any food stocks from own-production. In addition, stability encompasses
the certainty of future well-being, a state of being also referred to as “vulnerability”.
Indicators of forward-looking stability include measures of asset ownership (land and
equipment) and income diversity, and as social capital may reduce vulnerability to future
food insecurity (Misselhorn, 2009), access to a support network is also regarded as an
indicator of stability. In 2008, 15% of households participated in an inter-household
transfer network only as receivers, 25% participated as givers, 43% both gave and
received within their network, and the remaining 17% neither gave nor received. Only
the last category receives a value of zero for “transfer network participation”.
Table 2 provides descriptive statistics of these indicators over the panel years. The
results reflect a startling stagnancy over a range of indicators: Although 2004 was a
relatively good year for households in rural Zambia, by 2008 several food security
indicators had returned to their 2001 levels. Thus, while average calories per day
increased by 175 calories between 2001 and 2004, the average values of 2001 and 2008
do not significantly differ. The three activity level cut-offs used to identify households
as “food energy deficient” produce markedly divergent rates of energy deficiency
among the rural population, ranging from 54% to 68% in 2008. Between 2001 and
2008, these rates also do not change significantly. However, over the study period, some
improvement is evident among indicators of the quality and stability dimensions of
food security. Thus, by 2008, households possessed more livestock (on average) and
collected income from a greater number of sources.

284
Wineman Multidimensional household food security

Table 1: Food security indicator definitions


INDICATOR CONSTRUCTION
Quantity Calories/ AE/ day Sum of crop calories produced minus sales
(from previous growing season); calories from
milk and eggs produced minus sales (over past
year); and calories from staple food purchases
(over past year). This is scaled per adult
equivalent per day.
Food energy deficient (light/ 1= Household (HH) does not meet its calorie
moderate/ intense) requirements (for light, moderate, or intense
activity levels), 0 otherwise
Quality No. field crops retained Number of different field crops retained by HH
after sales
Tropical Livestock Units (TLU) Livestock units owned as of one year before
interview
HH produces eggs or milk 1= HH produces eggs or milk, 0 otherwise
HH produces vegetables/ 1= HH produces vegetables or fruits, 0
fruits otherwise (available for 2001 and 2008)
Stability No. months without food Number of months since HH ran out of food
stocks stocks from own-production (available for 2008).
Estimated as 12 for non-cropping HHs.
HH has maize or cassava in 1= HH has maize or mature cassava in stock/
stock in the field, 0 otherwise (available for 2004 and
2008)
Proportion food expenditure Proportion of HH income spent on food
purchases over past year
Land cultivated (ha) Land area (ha) cultivated in previous growing
season
Value of productive assets Value of farm equipment owned as of one year
(ln, 2007/08 ZMK) before interview
No. income sources Number of income sources from the following
categories: on-farm income, agricultural wage
income, non-agricultural wage or salary income,
business income, and remittances
HH participates in transfer 1= HH gave or received cash or gifts in the past
network year, 0 otherwise

285
2001 2004 2008 t-tests
‘01 = ‘04 =
Mean SD Mean SD Mean SD
Wineman

Table 2:
‘08 ‘08
Quantity Calories/ AE/ day 3,422.51 (4,399.95) 3,597.12 (4,169.83) 3,454.28 (4,342.77) *
  1= Food energy deficient (light) 0.53 (0.50) 0.50 (0.50) 0.54 (0.50) ***
  1= Food energy deficient 0.61 (0.49) 0.58 (0.49) 0.61 (0.49) ***
(moderate)
  1= Food energy deficient (intense) 0.68 (0.47) 0.65 (0.48) 0.68 (0.47) ***
Quality No. field crops retained 1.91 (1.20) 2.32 (1.33) 1.89 (1.17) ***
TLU 1.82 (6.90) 2.58 (12.87) 2.35 (8.42) ***
  1= HH produces eggs or milk 0.31 (0.46) 0.14 (0.35) 0.47 (0.50) *** ***
  1= HH produces vegetables/ fruits 0.16 (0.37) --- a 0.38 (0.49) *** ---
Stability 1= HH has maize or cassava in --- 0.60 (0.49) 0.40 (0.49) --- ***

286
stock
No. months without food stocks --- --- 2.82 (3.81) --- ---
  Proportion food expenditure 0.11 (0.22) 0.09 (0.19) 0.12 (0.22) *** ***
Land cultivated (ha) 1.44 (1.34) 1.56 (1.47) 1.42 (1.63) ***
Value of productive assets (ln) 1.82 (4.62) 2.45 (5.21) 2.19 (4.94) *** ***
  No. income sources 1.91 (0.84) 1.85 (0.84) 2.18 (0.85) *** ***
  1= HH participates in transfer 0.50 (0.50) 0.64 (0.48) 0.83 (0.38) *** ***
network
  Observations 6,922 5,419 8,094
Summary statistics of food security indicators, 2001–2008

Note: Asterisks denote significance levels of t-tests for the equality of means. *** p<0.01, ** p<0.05, * p<0.1
a Variable not captured for some years, as indicated by ---.
Multidimensional household food security
FIGURES

Wineman Multidimensional household food security

Patterns of food security in Zambia also differ by agro-ecological region, as illustrated


with a radar graph of the average normalised z-score5 of specific indicators for each
region in 2008 (Figure 2).6 A z-score is positive when the value exceeds the population
average, and in this graph, a higher score farther from the centre is relatively good. It
is clear that households in region IIb tend to experience diminished levels of most food
security indicators, with fewer field crops retained, a lower likelihood of producing
milk or eggs, and fewer months spent with food stocks. In contrast, region III stands out
in terms of high dietary diversity according to both the number of field crops retained
and theFigure
likelihood of producing vegetables/fruits.
1. Agro-ecological regions in Zambia Households in region III in the north
also report fewer months spent without food stocks from own-production. This likely
reflects the dominance of cassava in the region, as cassava does not require a seasonal
harvest and can often be left in the field until it is consumed. Households in region IIa
fare relatively well (on average) for most food security indicators.

z‐score
Calories
0,4
Participates in
0,2 TLU
transfer program
0,0
‐0,2
Value of productive No. field crops I
‐0,4
assets retained
‐0,6 IIa

IIb
Produces milk or
Land cultivated (ha) III
eggs

Months with food Produces
stocks vegetables/ fruits

Figure 2: Food security indicator z-scores by agro-ecological region, 2008


Figure 2. Food security indicator z-scores by agro-ecological region, 2008

5 A z-score is a measure of a value’s relationship to the mean value in a population. z-scorei = (xin- µn)/ σn where
xin is the value of indicator n for household i, µn is the average value of n in the population, and σn is the standard
deviation.
6 Summary statistics across regions are available from the author upon request.

287 29
Wineman Multidimensional household food security

Figure 3. Elements of the food sufficiency and food security indices


Figure 3: Elements of the food sufficiency and food security indices

5. THE MULTIDIMENSIONAL FOOD SECURITY INDICES

5.1 Construction of the indices


From the array of food security indicators presented in section 4, 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. This
study uses principal component analysis (PCA) to construct a set of two composite
food security indices, such that the weights are derived objectively from the data.
PCA is a type of factor analysis that can reduce dimensions within data by extracting
linear combinations that best describe the co-variance among all elements (Vyas and
Kumaramayake, 2005). Although food security is a latent variable that cannot be directly
observed, PCA can be used to infer a household’s level of food security from other
variables that are themselves observable. The first principal component captures the
greatest variation, and because data reduction is the primary objective of this exercise,
only the first component is used in this paper. This component is converted into factor
scores that serve as weights for the creation of an index.

288
Wineman Multidimensional household food security

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

� ���
�� = ∑ �� � �� ��
= (1) (1)
��

where
where��Iis isthe
theindex
indexvalue
valuefor for household
household �,i, which
which follows
follows aa normal
normal distribution
distribution with with aa mean
i
ofmean
0, ��ofis0,the is the weight
Fnweight for thefor�ththe n variable of theofindex
the index
(the (the squared factor score of � is
th
variable squared factor score of �), ��
n), xin is the value of variable n for household i, and µn and σn are the mean and standard
the value ofofvariable
deviations variable� n. for household �, and �� and �� are the mean and standard deviations
Essentially,
of variable �. the index is the sum of the weighted z-scores for each element of the
index. Each PCA-generated 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. the
Essentially, As index
this study
is theuses
sumthree
of the rounds of household
weighted z-scores forpanel data,
each it is necessary
element of the index.to Each
generate an index that is comparable over time. Following the approach of Cavatassi et
PCA-generated
al. (2004), dataindex for theis three
basedrounds
on a scale, whichand
are pooled is relevant
principalonly to that estimation,
components are estimatedsuch that a
over
set the combined
of indices data. estimations cannot be meaningfully compared. As this study uses
from different
The selection of elements included in the indices is driven by the food security
three rounds of household panel data, it is necessary to generate an index that is comparable
literature (see references in section 4), an imperative to span multiple dimensions of
over
foodtime. Following
security, and thethe goal approach of Cavatassi
of maximising et al. (2004),
the variation explained data
byforthe the
firstthree rounds are
principal
component.
pooled In othercomponents
and principal words, the variables (as well
are estimated as the
over anycombined
variable transformations)
data. were
selected to produce an index with the greatest ability to capture the variation among
The selection
these elements. of elements
Two indices included
are created in the
(Figure indices
3). The foodissufficiency
driven by index the food
(FSU)security
literature (see references in section 4), an imperative to span multiple dimensions aof food
is based mostly on the quantity and quality indicators of food security, reflecting
household’s realised welfare over the previous year. Unfortunately, the FSU does not
security,
include and the goal of
an indicator ofmaximizsing
food supply the variationasexplained
regularity, by the first
this information wasprincipal component.
not captured
Ininother
all survey
words,waves. The food(as
the variables security
well as index
any(FSI) is based
variable on the quantity,
transformations) werequality
selected to
and stability measures, now incorporating forward-looking indicators of vulnerability.
produce
Thus, two an index
measures with of
theasset
greatest
stocksability
(landtoarea
capture the variation
cultivated and theamong
value these elements. Two
of equipment
owned)are
indices capture
created future stability,
(Figure as they
3). The foodmay be used toindex
sufficiency continually
(FSU) produce
is basedfood andon the
mostly
can be liquidated in times of distress. A third indicator of vulnerability, participation in
quantity
a transfer and qualityaims
network, indicators
to captureof food security,potential
a household’s reflecting a household’s
access to assistance. realiszed
The FSIwelfare
is used
over the to exploreyear.
previous the merits of including
Unfortunately, thethese
FSUadditional indicators.
does not include
7
an indicator of food supply
The factor loadings of the indices are given in Table 3. The data reduction of
regularity, as this information
the PCA explains 35% of thewas not captured
original variationinofallthesurvey waves. The
data included food
in the security
FSU, and index
this value
(FSI) is based is 29%
on the for quantity,
the FSI. These
quality,values are similar
and stability to thosenow
measures, seenincorporating
in other studies forward-
that use PCA to develop a wealth or food security index (Demeke et al., 2011; Filmer
looking indicators of vulnerability. Thus, two measures of asset stocks (land area cultivated
and Pritchett, 1998). The factors that load most heavily on the FSU are calories and
cropthe
and diversity,
value of andequipment
the value ofowned)the Kaiser-Meyer-Olkin (KMO) measure
capture future stability, as they ofmay sampling
be used to
continually
7 produce
Our measure food
of income andis not
diversity canincluded
be liquidated ininclusion
in the FSI, as its times reduced
of distress. A explained
the variance third indicator
by the of
index. This seems to indicate that, in the rural Zambian context, the combination of food security indicators into a
vulnerability,
single indexparticipation in without
was more appropriate a transfer network,
considering aims to capture a household’s potential
income diversification.

access to assistance. The FSI is used to explore the merits of including these additional
289
indicators.7
The factor loadings of the indices are given in Table 3. The data reduction of the PCA
Wineman Multidimensional household food security

adequacy is 0.52. A large KMO value (greater than 0.5) points to an adequate proportion
of variance among the variables that may be indicative of a latent factor (Kaiser, 1974).
It should be noted, however, that our value is close to this cut-off. For the FSI, the
largest factor score is held by land area cultivated, and the KMO measure is 0.65. In
deciding to pool data from all panel years, it is important to verify that the pattern of
factor loadings does not change fundamentally from year to year when the exercise is
repeated for each year independently (Vyas and Kumaranayake, 2006). An inspection
of the values for individual years and agro-ecological regions confirms that the factor
loadings are generally consistent across time and space, and it is therefore acceptable to
create a pooled index.

Table 3: Factor loadings for the food sufficiency (FSU) and food security (FSI)
indices
Factor loadings
Element FSU FSI
Calories (ln) 0.64 0.39
No. field crops retained 0.60 0.34
TLU 0.33 0.39
HH produces milk or eggs 0.35 0.31
Land cultivated (ha) 0.52
HH participates in transfer network 0.12
Value of productive assets (ln) 0.45
KMO 0.52 0.65
Proportion variation explained 0.35 0.29
Observations 20,435 20,435

5.2 Descriptive statistics of the indices


The average values of these indices across years and regions are given in Table 4. Among
agro-ecological regions in 2008, region IIa has the highest average value for both the
FSU and FSI. Region I fares well with regard to average FSI, while region III fares
much worse, and this is surely due to the high disease burden for livestock in region
III (Siegel and Alwang, 2005), coupled with the important cultural role of livestock
in region I. Figure 4 illustrates how these indices can be used to explore the spatial
patterns of food security over time, showing the average FSU score across districts
and years. While the relative degree of food sufficiency in each district does change,
there are pockets of persistent insufficiency, particularly in the west (region IIb). This

290
Wineman Multidimensional household food security

information should be useful for the design of policies and the targeting of interventions
to improve food security in Zambia.

Figure 4. Geographic pattern of food sufficiency over time


Figure 4: Geographic pattern of food sufficiency over time

We next explore the predictive power of our indices by assessing the correlation between
index values in 2004 and 2008 (Table 5). The correlation between the FSI in 2004 and
the FSU in 2008 is 0.39, which is larger than that between the FSU in 2004 and 2008.
It therefore seems the FSI outperforms the FSU in predicting a household’s experience
of food sufficiency four years in the future, and this is as expected, given that the FSI
specifically includes indicators of vulnerability to future food shortfalls.

291
Wineman Multidimensional household food security

Table 4: Food sufficiency and food security index values


  FSU FSI
  Mean SD Mean SD
Pooled years 0.00 (1.18) 0.00 (1.42)
2001 -0.10 (1.14) -0.15 (1.29)
2004 0.05 (1.16) 0.04 (1.41)
2008 0.03 (1.23) 0.07 (1.50)
Agro-ecological regions (2008)
I 0.02 (1.03) 0.42 (1.50)
IIa 0.25 (1.15) 0.58 (1.75)
IIb -0.60 (1.35) -0.55 (1.33)
III -0.05 (1.24) -0.30 (1.09)

Table 5: Correlation of food security indices, 2004 and 2008


2008
FSI FSU
FSI 0.56*** 0.39***
2004

FSU 0.34*** 0.34***

*** p<0.01, ** p<0.05, * p<0.1

6. ECONOMETRIC APPLICATIONS OF THE FOOD


SUFFICIENCY INDEX
In this section, the food sufficiency index is used within econometric analyses of food
security dynamics. Following the approach of Demeke et al. (2011), we first use the FSU
to explore determinants of the persistence of food sufficiency. Households are classified
into three categories as never, always, or sometimes food sufficient, and a household is
considered food insufficient in a given year if its FSU score is in the lowest one-third
of the index distribution. For this exercise, a household is never food sufficient if it is
food insufficient in all three waves of the panel, always food sufficient if it is always in
the top two-thirds of the distribution, or sometimes food sufficient if its classification
changes at least once over the study period. With this categorisation, 39% of households
are always food sufficient, while a greater proportion (54%) have indeed moved in and
out of sufficiency, and just 7% are never food sufficient.

292
a greater proportion (54%) have indeed moved in and out of sufficiency, and just 7% are
never food sufficient.
WinemanA multinomial logit (MNL) model is used to identify the determinants
Multidimensional of being
household food either
security
always food sufficient or insufficient. With the MNL model, we can estimate the partial
A multinomial
effects logit (MNL)
of each explanatory variablemodel
on theisprobability
used to identify the determinants
that a household experiencesofone
being
of
either always food sufficient or insufficient. With the MNL model, we can estimate
these discrete outcomes, relative to the status of being sometimes food sufficient
the partial effects of each explanatory variable on the probability that a household
(Wooldridge,one
experiences 2002: 497). Explanatory
of these variables relative
discrete outcomes, are restricted
to theto status
initial household conditions
of being sometimes
food sufficient
in 2001, (Wooldridge,
with the selection of 2002: 497).
variables Explanatory
guided by the foodvariables are restricted
security literature to et
(Feleke initial
al., Formatte
household conditions in 2001, with the
2005; Misselhorn, 2005). The equation is:
selection of variables guided by the food security
literature (Feleke et al., 2005; Misselhorn, 2005). The equation is:

ܻ௜ ൌ ߛ ൅ ࢄ࢏ǡ૛૙૙૚ ࢼ ൅ ൅ࡳ࢏ ࢾ ൅ ߝ௜ (2) (2)

where Yi is the status of being always or never food sufficient for household i, Xi,2001 is
where ܻ is the status of being always or never food sufficient for household ݅, ࢄ࢏ǡ૛૙૙૚ is a
a vector ௜of socioeconomic characteristics and agricultural practices of household i in
vectorGofis socioeconomic
2001, a vector of characteristics
geographic and agricultural
variables, and ε is a practices
stochasticof household
error term. ݅ in
The 2001, ࡳ࢏
results
i i
(Table 6) indicate that greater educational attainment of the household
is a vector of geographic variables, and ߝ௜ is a stochastic error term. The results (Table 6) head is positively
correlated with being always food sufficient (column 2), as is a larger farm size, greater
indicate that greater educational attainment of the household head is positively correlated
asset holdings, and household participation in a transfer network. Residence along the
withrail
main being always
line, a proxyfoodforsufficient
market(column
access, 2), as isassociated
is also a larger farm size,
with greater
being asset holdings,
consistently food
sufficient
Africa over
and household the following
(Muyangaparticipation 7 years,
in a transfer
and Jayne, 2014) while
network.
are likely greater
Residence
to pose population
a threat along density
the main
to long-term is associated
rail line, food
household a proxy
with being never
for market asaccess,
sufficient. This suggests that rising population densities observed in
sufficiency, throughis heightened
also associated with being
land scarcity. consistently
Relative to region food sufficient
IIa (the over the
base region),
some regions of rural sub-Saharan Africa (Muyanga and Jayne, 2014) are likely to pose
following
threat to 7long-term
ahouseholds inyears,
regionswhile orgreater
Ihousehold population
IIb are density
significantly
food lessislikely
sufficiency, associated
as with
to experience
through being
heightened never
landsufficient.
consistent food
scarcity.
Relative to and
sufficiency,
This suggests region
that IIacase
in the (the
rising ofbase
regionregion),
population IIb, theyhouseholds
densitiesareobserved
more likely in some
regions
to never I or IIb
attain
regions are significantly
ofsufficiency.
rural sub-Saharan
less likely to experience consistent food sufficiency, and in the case
We next explore the short-term determinants of a household’s food sufficiency score. of region IIb, they
are more likely to never attain sufficiency.
Because the dependent variable is a normaliszed 11 score unique to this population, the
We next explore the short-term determinants of a household’s food sufficiency
coefficients
score. Because are difficult to interpret,
the dependent and readers
variable are encouraged
is a normalised to focus
score uniqueon their signs,
to this levels
population,
the coefficients and
of significance, are relative
difficultmagnitudes.
to interpret, and readers
A household fixed are encouraged
effects to focus on
(HHFE) regression
their signs, levels of significance, and relative magnitudes. A household
exploits the variation found within multiple observations of a single household to understand fixed effects
(HHFE) regression exploits the variation found within multiple observations of a single
the correlates of food sufficiency. The equation is:
household to understand the correlates of food sufficiency. The equation is:

‫ܷܵܨ‬௜௧ ൌ ߛ ൅ ࢄ࢏࢚ ࢼ ൅ ࢃ࢏࢚ ࣂ ൅  ࢅ࢚ ࣐ ൅ ߙ௜ ൅ ߝ௜௧


(3) (3)

where ‫ܷܵܨ‬௜௧ is the food sufficiency score for household ݅ at time ‫ ݐ‬, ࢄ࢏࢚ is a vector of
socioeconomic characteristics and agricultural practices of household ݅ at time ‫ݐ‬, ࢃ࢏࢚ is a Formatted: Fo

vector of climate outcomes of household ݅ at time ‫( ݐ‬referring to the previous growing


season), ࢅ࢚ is a vector of year dummy variables intended to capture time-variant shocks
common to all households, and ߙ௜ is the unobserved, time-invariant fixed effect for
household݅. Explanatory variables that are time-invariant, such as geographic location, are
necessarily omitted.
293
Results are given in Table 7. Ceteris paribus, a greater household size is associated
with a lower level of food sufficiency. The variables related to a household’s asset
endowment tell a consistent story: A higher value of productive assets and larger area of land
Wineman Multidimensional household food security

Table 6: Determinants of being never or always food sufficient (MNL regression)


  (1) (2)
1= Never food 1= Always food
sufficient sufficient
Explanatory variables (2001) Mean APE SE APE SE
values
 
HH size 5.84 0.00** (0.00) -0.01*** (0.00)

Proportion dependents 0.53 0.04* (0.02) 0.06 (0.04)


1= Female-headed HH (widowed) 0.10 0.02 (0.01) 0.02 (0.03)
Head’s age 45.56 -0.00 (0.00) 0.00 (0.00)
Years education of head 5.23 -0.00 (0.00) 0.01*** (0.00)
1= Owns radio 0.34 -0.01 (0.01) -0.01 (0.02)
1= HH grew cash crops 0.08 -0.02 (0.02) 0.06** (0.03)
1= HH grew maize 0.79 -0.02 (0.01) 0.09*** (0.02)
1= HH acquired fertilizer 0.22 -0.01 (0.02) 0.05** (0.02)
1= HH owns irrigation equipment 0.01 -0.07*** (0.01) -0.05 (0.08)
Land cultivated (ha) 1.44 -0.08*** (0.01) 0.11*** (0.01)
Value of assets (ln) 1.85 -0.01*** (0.00) 0.01*** (0.00)
1= HH participated in transfer 0.50 -0.01 (0.01) 0.06*** (0.02)
network
1= HH is related to village chiefa 0.50 -0.03*** (0.01) 0.06*** (0.01)
1= District is along rail line 0.31 -0.05*** (0.01) 0.11*** (0.03)
District population density 21.16 0.00*** (0.00) -0.00** (0.00)
(persons/ km2)
1= Region I 0.06 -0.01 (0.01) -0.11*** (0.03)
1= Region IIb 0.10 0.08*** (0.03) -0.21*** (0.03)
1= Region III 0.45 -0.02 (0.01) -0.01 (0.03)

Observations 4,286 4,286


APE = Average partial effects; standard errors clustered at enumeration area; *** p<0.01, **
p<0.05, * p<0.1
a
Information collected in 2008

294
Wineman Multidimensional household food security

where FSUut is the food sufficiency score for household i at time t, Xit is a vector of
socioeconomic characteristics and agricultural practices of household i at time t, Wit is
a vector of climate outcomes of household i at time t (referring to the previous growing
season), Yt is a vector of year dummy variables intended to capture time-variant shocks
common to all households, and αi is the unobserved, time-invariant fixed effect for
household i. Explanatory variables that are time-invariant, such as geographic location,
are necessarily omitted.
Results are given in Table 7. Ceteris paribus, a greater household size is associated
with a lower level of food sufficiency. The variables related to a household’s asset
endowment tell a consistent story: A higher value of productive assets and larger area
of land cultivated are positive and significant determinants of sufficiency. Participation
in a transfer network (the receipt or provision of cash or goods) is also associated with
being food secure.8 As rainfall and temperature enter the model in a quadratic manner, it
is possible to roughly estimate the “optimal” levels of rainfall and temperature for food
sufficiency. The optimal early season rainfall occurs at 809.62mm, while the household
average for early season rainfall over the panel years is 570.74mm. This indicates that
any decrease in rainfall from the current average would be harmful to food sufficiency,
though there is room for improvement with an increase. The optimal average growing
season temperature occurs at 22.58°C, while the household average for temperature
over the panel years is 23.13°C. This indicates that any warming above the present-
day average will be harmful to food sufficiency, an important finding in light of the
expected increase in temperatures in southern Africa associated with climate change
(Kotir, 2011).
Our final exercise is to explore the avenues through which rainfall and temperature
affect food sufficiency. Using equation (3), we disaggregate the FSU, treating each
element, in turn, as a dependent variable. The decision to include a quadratic term for
temperature and rainfall is based on the sign on the coefficient of the linear term and a
test for the joint significance of both terms. Results, given in Table 8, show that when the
FSU is regressed on seasonal climate variables (column 1), the relationships between
rainfall/temperature and food sufficiency are quadratic. A similar relationship is evident
when calories or the number of crops are used as the dependent variable (columns 2–3).
However, rainfall has only a negative relationship with the production of milk/eggs,
and is not a significant determinant of TLU (columns 4–5). And while temperature is
significant for most elements of the FSU, this is not true for the production of milk/eggs.

8 Transfer network participation may be susceptible to reverse causality if participation is a response to a household’s
food security experience. For example, if households receive remittances during times of distress, this would
attenuate the detected relationship between network participation and avoidance of food insufficiency. On the
other hand, households experiencing a good year may be more likely to provide assistance to others. Results
should be interpreted as correlations only.

295
Wineman Multidimensional household food security

Table 7: Determinants of food sufficiency score (HHFE regression)


FSU
Coef SE
     
HH size -0.05*** (0.01)
Proportion dependents -0.01 (0.06)
1= Female-headed HH (widowed) -0.07 (0.05)
Head’s age 0.03*** (0.01)
Head’s age 2
-0.00** (0.00)
Years education of head 0.01* (0.00)
1= Owns radio 0.11*** (0.02)
1= HH grew cash crops 0.08** (0.04)
1= HH grew maize 0.77*** (0.05)
1= HH acquired fertilizer 0.08*** (0.03)
1= HH owns irrigation equipment 0.29** (0.13)
Land cultivated (ha) 0.21*** (0.02)
Value of assets (ln) 0.02*** (0.00)
1= HH participated in transfer network 0.14*** (0.02)
Early season rain (100s mm) a
0.06** (0.03)
Rain² -0.00** (0.00)
Avg. season temperature (°C) a
0.74*** (0.24)
Temperature² -0.02*** (0.00)
1= 2004 -0.03 (0.02)
1= 2008 0.02 (0.04)
Constant -10.21*** (2.87)

Observations 12,858
Within R 2
0.19  
Standard errors clustered at household; *** p<0.01, ** p<0.05, * p<0.1
a
Growing season temperature (averaged over the daytime high and nighttime low temperatures) refers
to November-March, while early season rainfall refers to mid-December-February. The latter interval is
understood as most important for maize development.

296
Wineman Multidimensional household food security

These results suggest that livestock ownership may improve household resilience to
rainfall shocks if diet quality can be maintained in the face of uncertain rainfall. The
realisation that temperature is a determinant of the number of crops retained suggests
that agricultural extension services may promote food sufficiency by encouraging
crop mixes that include more heat-tolerant crops. Table 8 reveals that seasonal climate
outcomes affect the various elements of this food sufficiency index differently, and
this disaggregated analysis is appropriate to identify interventions to bolster household
resilience to climate shocks.

Table 8: Effects of climate shocks on individual FSU elements (HHFE


regressions)
  (1) (2) (3) (4) (5)
Calories Crops
FSU Milk/ eggs TLU
(ln) retained
Early season rain (100s mm) 0.06** 0.10*** 0.05** -0.01*** -0.06
(0.03) (0.03) (0.02) (0.00)a (0.08)b
Rain² -0.00** -0.01*** -0.001
(0.00) (0.00) (0.00)
Avg. season temperature (°C) 0.74*** 0.18 1.25*** -0.002 1.59
(0.24) (0.26) (0.24) (0.01) b
(1.22)
Temperature² -0.02*** -0.01 -0.03*** -0.03
(0.00) (0.01) (0. 0) (0.03)

HH controls of Table 7 Y Y Y Y Y
P > F(Rain= 0 & Rain2= 0) 0.05 0.00 0.00 --- ---
P > F(Temp= 0 & Temp2= 0) 0.00 0.00 0.00 --- 0.05
Within R2 0.19 0.13 0.20 0.19 0.05
Observations 12,858 12,858 12,858 12,858 12,858
Standard errors in parentheses, clustered at household; *** p<0.01, ** p<0.05, * p<0.1
a
When a quadratic term is included, the squared term coefficient is also negative, and the two are jointly
significant (p=0.01).
b
When a quadratic term is included, the squared term coefficient is positive, and the two are not jointly
significant (p > 0.1).

297
Wineman Multidimensional household food security

7. DISCUSSION AND CONCLUSIONS


This study first provides information on a range of household-level food security
indicators in Zambia from 2001 to 2008, with variables categorised by the quantity,
quality and stability components of food security. Particularly with regard to indicators
of quantity, the findings reveal stagnancy in the food security situation over the study
period. Our descriptive analysis also reveals geographic patterns of food security, with
region IIa consistently exhibiting favourable values of various indicators. In addition,
the “cassava belt” (region III) exhibits higher levels of food stability (in terms of
regularity of food supply) than other regions. Although the government of Zambia has
recently focused much of its agricultural policy on promoting maize production, this
finding highlights the importance of cassava to Zambia’s food security.
In an attempt to reflect the wide-ranging definition of food security (FAO, 2002), we
construct the food sufficiency index (FSU) and food security index (FSI), incorporating
a subsample of indicators that span multiple dimensions. The FSI specifically includes
indicators of future well-being, including land size, asset base, and access to a social
support network. We demonstrate how these indices can be used to map the geography
of food insecurity in Zambia, with pockets of persistent insufficiency evident in the west.
In terms of the predictive power of the indices, the FSI exhibits a larger correlation with
food sufficiency four years in the future, indicating that it better encapsulates forward-
looking food “security”, inclusive of a household’s capacity to mitigate negative shocks.
The final portion of this article employs the food sufficiency index in econometric
analyses of the dynamics of food security. Coefficients are used to estimate the “optimal”
values of seasonal climate outcomes, showing that any increase in temperature from the
current average would reduce food sufficiency. To assess the avenues through which
climate shocks affect a household’s latent food sufficiency status, we disaggregate the
index into its individual components. The results suggest that livestock ownership and
production of livestock products are less sensitive to different weather shocks, such that
their inclusion in a household’s livelihood portfolio may comprise an important part of
a strategy to improve food security in the face of climate uncertainty. What does this tell
us about the usefulness of a composite food security index in analysis? On one hand, we
have seen that by measuring a household’s latent food security status, the food security
score allows for a straightforward description of the state of food security in Zambia (see
Figure 4). On the other hand, the policy implications of our econometric analysis are not
clear until the index is decomposed. The avenues of impact and potential interventions
would not be readily deduced when using the index in its composite form, a finding that
reflects the conclusion reached by Coates (2013).
Several caveats are warranted that have not been adequately emphasised: First,
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

298
Wineman Multidimensional household food security

directly measuring household consumption of specific macronutrients or micronutrients


(e.g., protein), livestock ownership and homestead production of milk or eggs are used
as proxies for the likely presence of animal protein in the diet. Second, the use of PCA
to construct our food security indices involves a linear aggregation of the weighted
elements of each index. Though the weights reflect the trade-offs between the index
elements, the implication of this linear form is that these elements are perfect substitutes.
However, each household’s food sufficiency and security scores may be sensitive to this
methodological choice (Santeramo, 2015). Third, with a household-level analysis, we
were unable to consider the intra-household allocation of food, which is highlighted by
Headey and Ecker (2013) as a “best practice” in food security measurement.
Nevertheless, this study sheds light on the food security of households across rural
Zambia, showing that a measure of quantity alone cannot capture the breadth of the
food security experience. Rather, a combination of indicators is needed to reflect its
multiple dimensions. We demonstrate how composite food security indices can be used
to monitor (and predict) food security and study its correlates and determinants. At
the same time, this paper shows that some information is lost when aiming to broadly
measure food security, inclusive of its multiple dimensions. For certain purposes, we
find that a composite food security index is well suited to track progress across the
population. But for other purposes, such as the analysis of sensitivity to climate shocks,
or the design of interventions that would address different types of food insecurity, we
concur with Coates (2013) that evaluation is most usefully levelled at the concept’s
individual components.

ACKNOWLEDGEMENTS
Financial support for this study was provided by the USAID Bureau for Food
Security (Associate Award AIDOAA-LA-11-00010 under Food Security III,
CDG-A-00-02-00021-00). The Food Security Research Project of Michigan State
University generously provided access to the household data used in this paper, while
the Zambia Meteorological Department provided access to historical climate records.
The author wishes to thank Eric W. Crawford, David Mather and Robert Richardson
of Michigan State University for their review of the original version of this work. The
views expressed are those of the author only.

REFERENCES
Alkire, S. & Foster, J. 2011. Counting and multidimensional poverty measurement. Journal of Public
Economics 95(7–8): 476–487.
Barrett, C. 2002. Food security and food assistance programs. In Garner, B.L. & Rausser, G.C. (eds.),
Handbook of agricultural economics, 2103–2190. Amsterdam: Elsevier Science.

299
Wineman Multidimensional household food security

Carletto, C., Zezza, A. & Banerjee, R. 2013. Towards better measurement of household food security:
Harmonizing indicators and the role of household surveys. Global Food Security 2(1): 30–40.
Cavatassi, R., Davis, B. & Lipper, L. 2004. Estimating poverty over time and space: Construction of
a time-variant poverty index for Costa Rica. Agricultural and Development Economics Division
Working Paper No. 04-21. Rome: Food and Agriculture Organization.
Coates J. 2013. Build it back better: Deconstructing food security for improved measurement and
action. Global Food Security 2(1): 188–194.
Coates, J., Swindale, A. & Bilinsky, P. 2007. Household Food Insecurity Access Scale (HFIAS) for
measurement of household food access: Indicator guide. Washington, D. C.: Food and Nutrition
Technical Assistance Project (FANTA), Academy for Educational Development.
Demeke, A., Kiel, A. & Zeller, M. 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.
Feleke, S., Kilmer, R. & Gladwin, C. 2005. Determinants of food security in Southern Ethiopia at the
household level. Agricultural Economics 33(3): 351–363.
Filmer D. & Pritchett, L. 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). 1996. Rome Declaration on World
Food Security and World Food Summit Plan of Action, November 13–17. Rome: World Food
Summit.
FAO. 2002. The state of food insecurity in the world 2001. Rome: FAO.
Galhena, D.H., Freed., R. & Maredia, K.M. 2013. Home gardens: A promising approach to enhance
household food security and wellbeing. Agriculture and Food Security 2(8). doi: 10.1186/2048-
7010-2-8
Headey, D. & Ecker, O. 2013. Rethinking the measurement of food security: From first principles to
best practice. Food Security 5: 327–343.
Jain, S. 2007. An empirical economic assessment of impacts of climate change on agriculture in
Zambia. World Bank Policy Research Working Paper No. 4291. New York: World Bank.
Jodlowski, M., Winter-Nelson, A., Baylis, K. & Goldsmith, P.D. 2016. Milk in the data: Food security
impacts from a livestock field experiment in Zambia. World Development 77: 99–114.
Jones, A.D., Ngure, F.M., Pelto, G. & Young, S.L. 2013. What are we assessing when we measure food
security? A compendium and review of current metrics. Advances in Nutrition 4(5): 481–505.
Jumanji, Z., Hertzmark, E., Mlingi, N., Assey, V., Ndossi, G. & Fawzi, W. 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.
Kaiser, H. 1974. An index of factorial simplicity. Psychometrika 39(1): 31–36.
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(3): 587–605.
Leroy, J.L., Ruel, M., Frongillo, E. A., Harris, J. & Ballard, T.J. 2015. Measuring the food access
dimension of food security: A critical review and mapping of indicators. Food and Nutrition
Bulletin 36(2): 167–195.

300
Wineman Multidimensional household food security

Masset, E. 2011. A review of hunger indices and methods to monitory country commitment to fighting
hunger. Food Policy 36: S102–S108.
Maxwell, D., Vaitla, B. & Coates, J. 2014. How do indicators of household food insecurity measure
up? An empirical comparison from Ethiopia. Food Policy 47: 107–116.
Mazunda, J., Kankwamba, H. & Pauw, K. 2015. Food and nutrition security implications of crop
diversification in Malawi’s farm households. In Aberman, N., Meerman, J. & Benson, T. (eds.),
Mapping the linkages between agriculture, food security, and nutrition in Malawi, 44–49).
Washington, D. C.: International Food Policy Research Institute.
Misselhorn, A. 2005. What drives food insecurity in southern Africa? A meta-analysis of household
economy studies. Global Environmental Change 15(1): 33–43.
Misselhorn, A. 2009. Is a focus on social capital useful in considering food security interventions?
Insights from KwaZulu-Natal. Development Southern Africa 26(2): 189–208.
Muyanga, M. & Jayne, T.S. 2014. Effects of rising rural population density on smallholder agriculture
in Kenya. Food Policy 48: 98–113.
Ravallion, M. 2011. On multidimensional indices of poverty. World Bank Policy Research Working
Paper No. 5580. New York: World Bank.
Santeramo, F.G. 2015. Food security composite indices: Implications for policy and practice.
Development in Practice 25(4): 594–600.
Siegel, P. & Alwang, J. 2005. Poverty reducing potential of smallholder agriculture in Zambia:
Opportunities and constraints. World Bank Africa Region Working Paper No. 85. New York:
World Bank.
Smith, L. & Subandoro, A. 2007. Measuring food security using household expenditure surveys.
Food security in practice technical guide series. Washington, D.C.: International Food Policy
Research Institute (IFPRI).
United Nations University, World Health Organization and Food and Agriculture Organization of the
United Nations (UNU/ WHO/ FAO). 2004. Human energy requirements: Report of a joint FAO/
WHO/UNU expert consultation, October 17–24, 2001. Food and Nutrition Technical Report
Series 1. Rome: FAO.
Vyas, S. & Kumaramayake, L. 2006. Constructing socio-economic status indices: How to use principal
components analysis. Health Policy and Planning 21(6): 459–468.
Webb, P., Coates, J., Frongillo, E., Rogers, B., Swindale, A. & Bilinsky, P. 2006. Measuring household
food insecurity: Why it’s so important and yet so difficult to do. The Journal of Nutrition 136(5):
1404–1408.
Wooldridge, J. 2002. Econometric analysis of cross section and panel data. Cambridge: Massachusetts
Institute of Technology Press.
World Food Program (WFP). 2009. Comprehensive food security and vulnerability analysis guidelines
(First edition). Rome: WFP.
Wu Leung, W., Busson, F. & Jardin, C. 1968. Food composition table for use in Africa. Bethesda,
MD and Rome: U.S. Dept. of Health, Education, and Welfare and the Food and Agriculture
Organization of the United Nations.

301

Vous aimerez peut-être aussi