94092

The Impact of the Global Food Crisis
on Self-Assessed Food Security

Derek D. Headey1

We provide the ﬁrst large-scale survey-based evidence on the impact of the global
food crisis of 2007– 08 using an indicator of self-assessed food security from the
Gallup World Poll. For the sampled countries as a whole, this subjective indicator of
food security remained the same or even improved, seemingly owing to a combination
of strong economic growth and limited food inﬂation in some of the most populous
countries, particularly India. However, these favorable global trends mask divergent
trends at the national and regional levels, with a number of countries reporting sub-
stantial deterioration in food security. The impacts of the global crisis therefore
appear to be highly context speciﬁc. JEL codes: I32, O11

The global food crisis of 2007–08 involved approximately a doubling of inter-
national wheat and maize prices in the space of two years and a tripling of in-
ternational rice prices in the space of just a few months. Understandably, such
rapid increases in the international prices of staple foods have raised concerns
about the impact on the world’s poor. Household surveys suggest that most
poor people earn signiﬁcant shares of their incomes from agriculture but are
nevertheless often net food consumers (World Bank 2008b). Consistent with
this stylized fact, several multicountry World Bank simulation studies ﬁnd that
poverty typically increases when food prices increase (holding all else equal),
with much of the increase in poverty taking place in poorer rural areas (Ivanic
and Martin 2008; de Hoyos and Medvedev 2009; Ivanic, Martin, and Zaman

1. Derek Headey, Research Fellow, International Food Policy Research Institute, PO Box 5689,
Addis Ababa, Ethiopia. D.Headey@cgiar.org. A supplemental appendix to this article is available at
http://wber.oxfordjournals.org/. The author particularly wishes to thank Angus Deaton for the
introduction to the GWP data as well as very detailed comments on an early draft. Thanks also to
Gallup staff for answering a number of questions and to Shahla Shapouri of the USDA for providing
comments and answering questions regarding the USDA model. John Hoddinott, Olivier Ecker, Paul
Dorosh, Bart Minten, Maggie McMillan, Maximo Torero, and Shenggen Fan contributed useful
comments and suggestions. Participants at various seminars at the FAO and IFPRI provided insightful
comments. The author also thanks USAID for ﬁnancial support and Yetnayet Begashaw, Teferi
Mequaninte, and Sangeetha Malaiyandi for excellent research assistance. Any errors are the author’s
own.

THE WORLD BANK ECONOMIC REVIEW, VOL. 27, NO. 1, pp. 1 – 27 doi:10.1093/wber/lhs033
Advance Access Publication January 7, 2013
# The Author 2013. Published by Oxford University Press on behalf of the International Bank
for Reconstruction and Development / THE WORLD BANK. All rights reserved. For permissions,
please e-mail: journals.permissions@oup.com

1
2 THE WORLD BANK ECONOMIC REVIEW

2011). Likewise, the U.S. Department of Agriculture’s (USDA 2009) simulation
found that approximately 75 –80 million people went hungry during the 2008
food crisis, a number that the Food and Agriculture Organization (FAO) of the
UN (FAO 2009) applied to its precrisis baseline numbers in the absence of an
FAO model that could simulate a food price shock.2 Subsequent USDA simula-
tions were used by the FAO to estimate that over one billion people went
hungry in 2009, up from 873 million in 2005–06.3
These studies have led some observers to conclude that global poverty or
hunger increased during the 2008 food crisis. Fundamentally, however, most of
the simulation studies cited above aim to predict and understand the impacts of
higher relative food prices, holding all else equal. The use of this kind of partial
simulation approach is justiﬁable on several grounds. First, partial simulations
have an advantage in being able to produce very timely ex ante estimates of
what might happen if food prices increase. Second, more sophisticated ap-
proaches (Ivanic and Martin 2008; de Hoyos and Medvedev 2009; Ivanic,
Martin, and Zaman 2011) are useful for identifying the mechanisms by
which higher food prices could inﬂuence poverty and the distributional conse-
quences of food price changes. In that sense, they are certainly policy rele-
vant. Third, these approaches provide the scope to explore the sensitivity of
results to alternative assumptions.
However, the use of partial approaches to infer actual changes in global
poverty is inappropriate because there are many ways their predictions might
not eventuate. For example, several simulation studies assumed rates of interna-
tional price transmission to domestic markets rather than using observed price
changes (e.g., Ivanic and Martin 2008). There is also the poorly informed ques-
tion of whether wages (rural and urban) might adjust to higher food prices,
with some evidence suggesting that agricultural wages might adjust even in the
short run (Lasco et al. 2008). More generally, strong income or wage growth
(even without “adjustment”) may have buffered any negative impacts of higher
prices in the 2000s, as Mason et al. (2011) observed in urban Kenya and
Zambia. More ambiguously, households could mitigate the worst forms of
hunger or poverty through any number of coping mechanisms, such as

2. Some basic problems with the FAO model are reviewed in Headey (2011a) and FAO (2002). In
the 2008 crisis, the FAO had an underlying model that only incorporated quantities, not prices, so the
FAO’s capacity to simulate the effects of food price increases was very limited. Therefore, the FAO
relied on a USDA trade model (USDA 2009). A major shortcoming of the USDA model was that it did
not include middle-income countries, including large ones such as China, Mexico, and Brazil. Headey
(2011a) also shows that the USDA (2009) estimates are contradicted by the USDA’s own historical
production and import estimates for 2007–08 (USDA 2011).
3. In addition to the two basic approaches described above (the World Bank poverty simulations
and the FAO/USDA hunger simulations), several authors have taken mixed approaches to estimate
´quez et al. (2010) and Tiwari and Zaman (2010). Dessus et al.
calorie availability trends, including Anrı
(2008) adopt the net beneﬁt ratio approach, but only for urban areas. There are also many
country-speciﬁc simulation exercises; a particularly good one is Arndt et al. (2008). See Headey (2011a)
for a more extensive overview and critique.
Headey 3

reducing dietary quality, selling assets, working longer hours, or reducing
nonfood expenditures.4
Because of these complexities, this article takes a different route by provid-
ing the ﬁrst ex post analysis of survey data collected before, during and shortly
after the 2008 food crisis across a large number of countries. Speciﬁcally, we
examine the results from an indicator of self-assessed problems affording sufﬁ-
cient amounts of food, which was collected as part of the Gallup World Poll
(GWP). Although subjective data certainly have shortcomings (an issue we discuss
in detail below), their advantage in this context is that they are substantially
cheaper to collect relative to the more objective monetary or anthropometric indi-
cators found in standard household welfare surveys. Hence, the country and time
coverage of the GWP surveys is their primary advantage. Speciﬁcally, the GWP
surveys allow us to examine self-assessed food insecurity trends in 69 low- and
middle-income countries, of which China is the most prominent exclusion. This
substantial cross-country coverage also allows us to test whether changes in this
indicator are explained by variations in food inﬂation and economic growth.
The basic conclusion from the Gallup data is that at the peak of the crisis
(2008), global food insecurity was either not higher or even substantially lower
than it was before the crisis. The raw results for the 69 countries for which we
have precrisis (2005–06) and mid-crisis (2008) data suggest that 132 million
people became more food secure. If 2007 is used as the “precrisis” benchmark,
the picture is more neutral because self-assessed food insecurity was essentially
unchanged between 2007 and 2008. However, these surprisingly optimistic
global trends mask large regional variations. Global trends are clearly driven
by declining food insecurity in India and several other large developing coun-
tries. However, on average, self-assessed food insecurity increased in many
African countries and most Latin American countries. It decreased somewhat in
Eastern Europe and Central Asia, but it probably rose in the Middle East (for
which the GWP sample is very small). In the average Asian country, there was
basically no change, although we again observe variations around the mean.
Because this article introduces a new method for gauging trends in global
food security, it is especially important to investigate the reliability of the
Gallup indicator and to understand the factors that might explain these some-
what surprising results. In the analysis below, we note some of the general
shortcomings of subjective indicators, which are now widely used in the con-
texts of general well-being (e.g., Headey et al. 2010; Kahneman and Deaton
2010; Deaton 2010; 2011), poverty (Ravallion 2012), and food security
(Deitchler et al. 2010), as well as some speciﬁc problems with the Gallup indi-
cator. We also conduct econometric tests to determine whether the observed
trends in self-assessed food security are plausibly explained by changes in per

4. Inevitably, measurement and estimation issues constrain these studies. Headey and Fan (2010)
and Headey (2011a) provide an overview of some measurement and estimation issues (see also footnote
2). Of course, measurement issues also apply to the data used in this study (see section 2).
4 THE WORLD BANK ECONOMIC REVIEW

capita GDP and various food price indices. As expected, we ﬁnd that real eco-
nomic growth improves self-assessed food security. Real GDP growth already
controls for aggregate price changes. We also ﬁnd some additional effects of
aggregate inﬂation, but we ﬁnd no signiﬁcant additional effect of relative food
price changes (i.e., changes in the food terms of trade). We also show that in
many of the largest developing economies (i.e., those with the largest poor pop-
ulations), nominal economic growth generally outpaced food inﬂation, even in
2008. Hence, it appears that strong real income growth has largely offset the
adverse impacts of food inﬂation in many developing countries, including those
with the largest poor populations.

II . AN OVE RVI EW OF THE GALL UP WORLD POL L FOOD
I N S E C U R I T Y I N D I C AT O R

In this section, we provide an overview of the GWP and the speciﬁc food secur-
ity indicator used in this study. Our goal is limited to answering three questions.
First, what is the general quality of the GWP surveys? Second, what limitations
might the GWP indicator of self-assessed food insecurity have? Third, do basic
cross-country patterns in this indicator align with expectations? Because the
GWP is conducted by a private organization and its collaborators, much of the
description of the formal survey characteristics relies on Gallup materials. We
explore correlations between the GWP indicator and non-GWP welfare indica-
tors by conducting a correlation analysis of a cross-section of countries and, in
the next section, a multivariate analysis of the full panel dataset.5
General Characteristics of the Gallup World Poll
Since 2005–06, the GWP has interviewed households in approximately 150
countries, although not always annually. Most questions are constructed to
have yes or no answers to minimize translation errors. In developing countries,
all but one of the GWP surveys are conducted face to face (China 2009 is the
exception), and most take approximately one hour to complete. The surveys
follow a complex design and employ probability-based samples intended to be
nationally representative of the entire resident civilian noninstitutionalized pop-
ulation aged 15 years and older. In the ﬁrst stage of sampling, primary sam-
pling units consisting of clusters of households are stratiﬁed by population size,
geography, or both, with clustering achieved through one or more stages of
sampling. When population information is available, sample selection is based
on probabilities proportional to population size; otherwise, simple random
sampling is used. Gallup typically surveys 1,000 individuals in each country,
except in larger countries such as India (roughly 6,000), China (4,000), and
Russia (3,000). In the second stage, random route procedures are used to select

5. Much of what follows is drawn directly from the Gallup Worldwide Research Methodology
(Gallup 2010a). The present author purchased country-level data directly from Gallup and
corresponded with senior Gallup staff about speciﬁc questions.
Headey 5

sampled households within a primary sampling unit, and Kish grids are used to
select respondents within households. Finally, the data are internally assessed
for consistency and validity and then centrally aggregated and cleaned. Data
weighting is used to ensure a nationally representative sample for each country,
with oversampling corrected accordingly.6
This approach generates margins of error that are generally in the 3 –4
percent range at the 95 percent conﬁdence level, with a mean error margin of
3.3 percent.7 Note, however, that because these surveys have a clustered
sample design, the margin of error varies by question. It is therefore possible
that the margin of error is greater for certain questions. We also note that the
margins of error in China and India tend to be lower than the average (by 1.6
to 2.6 percentage points). However, in China in 2005–06, the food insecurity
question followed some fairly detailed questions on income and welfare, which
may have primed respondents to be more likely to answer “yes” to the food in-
security question. Although we were aware of this problem in China, there may
be similar problems in other countries. It is certainly possible that the ﬁrst
wave of the GWP (2005–06) contains greater measurement error than subse-
quent waves because Gallup faced a steep learning curve in conducting such an
ambitious global survey (we address this issue below in a sensitivity analysis).

The Gallup World Poll Question on Food Security
Although these general characteristics of the GWP surveys are pertinent, we
now turn to the speciﬁc question of interest, which is phrased as follows:
“Have there been times in the past 12 months when you did not have enough money to buy the
food that you or your family needed?”

A simple yes or no answer is recorded. For simpliﬁcation, we refer to this as
the “food insecurity” indicator rather than a more cumbersome term such as
“unaffordability of food.”8 What are some of the strengths and weaknesses of

6. In a handful of cases, certain sections of the population are oversampled (see appendix S3). For
example, urban areas were oversampled in Pakistan, Russia, and Ukraine in at least one year, and in the
August–September 2009 survey in China, the provinces of Beijing, Shanghai, and Guangzhou were
oversampled, possibly because of the unusual switch to telephone surveying. In other contexts, it appears that
Gallup oversampled more educated groups (Senegal, Zambia), and in some developing countries, certain parts
of the country were not sampled at all because of ongoing political instability or other accessibility problems.
7. Thus, if the survey were conducted 100 times using the same procedures, the “true value” around an
assessed percentage of 50 would fall within the range of 46.7 percent to 53.3 percent in 95 out of 100 cases.
8. We note that there are other welfare indicators measured by Gallup, including a question
pertaining to hunger rather than food affordability as well as a general life satisfaction question (scaled
from one to ten). In earlier versions of this paper, we considered the hunger variable, but the sample
size for that indicator was much smaller, and trends in that variable could not be signiﬁcantly explained
by economic growth or food inﬂation. The life satisfaction question was not explored because it is not
obvious that changes in this indicator over 2006–08 would be substantially related to food inﬂation.
Even so, that indicator generally suggests sizeable improvements in well-being in developing countries,
with only a handful of exceptions (Pakistan, Sierra Leone, Egypt, and Afghanistan). Hence, we
concentrate on the more relevant food insecurity question.
6 THE WORLD BANK ECONOMIC REVIEW

this question? The strengths include a focus on access rather than availabili-
ty, a recall period (12 months) capable of capturing seasonality and other
short-run food price movements, and large cross-country and multiyear cov-
erage. This last strength is a signiﬁcant advantage in the absence of more
regular economic or nutritional surveys, but there are also limitations with
subjective data. Unlike simulation approaches, for example, subjective data
do not provide much information about the mechanisms or magnitudes of
welfare impacts. However, there are some indications that the simple yes/no
indicator used here may not lead to much loss of information in practice.
The GWP has data for Africa in which a similar question is asked that
allows for ﬁve different answers based on the frequency of deprivation.
Those data show a similar trend to the dichotomous indicator (see ﬁg. S.1 in
the supplemental online appendix, available at http://wber.oxfordjournals.
org/).
A more signiﬁcant problem is that the deﬁnition of food needs is not univer-
sal. For a well-off or well-educated family accustomed to a high-quality diet,
“food” may mean a food bundle of sufﬁciently high quality (e.g., meat, eggs,
dairy). For a very poor family, however, “food” may just mean enough cereals
or other staple foods. Hence, it is possible that the food insecurity measure is
biased upward by education or income or downward by overly low standards
of food intake. There is some indication of such biases in the data, although
formal tests of the presence of biases proved to be inconclusive (Headey
2011a). For example, there is surprisingly high self-assessed food insecurity in
developing countries with relatively high levels of education/literacy, such as
the former Soviet Bloc countries and Sri Lanka (see the online supplemental
appendix S2 for individual country-year observations). At the other extreme,
food insecurity appears too low in several countries where we know that un-
dernutrition is quite prevalent. In Ethiopia, for example, where diets are very
monotonous and undernutrition is very high, self-assessed food insecurity
was just 14 percent in 2006 (although it subsequently rose rapidly). However,
in cross-country regressions, we did not ﬁnd an impact of education on food
insecurity after controlling for income (see Headey 2011a). There are no indi-
cations that large numbers of poor countries systematically underreport food
insecurity.
To illustrate this issue, table 1 reports regional means (the full Gallup data
are presented in appendix S2). At the bottom of table 1, we observe that the
mean “global” prevalence of households reporting problems with affording
food is almost 32 percent. As expected, however, there are large variations
around the world, with some countries reporting almost no food insecurity and
others reporting that 80 percent of households had problems affording food.
For the most part, the pattern across continents is plausible. Food insecurity is
Headey 7

T A B L E 1 . Regional Unweighted Means for the Two GWP Measures, Circa
2005, for Developing Countries Only (Percent)
Food insecurity

Mean No. of obs.

sub-Saharan Africa 58.3 27
South Asia* 31.2 5
East Asia* 24.0 6
Middle East & North Africa 26.5 2
Central America & Caribbean 34.7 9
South America 36.0 10
Transitiona countries 29.1 23
OECDb 8.3 22
Low incomec 48.6 49
Middle incomec 29.6 28
Upper incomec 11.0 34
Mean, total sample 31.7 433

Note: *Note that two outliers are excluded. Nepal is excluded from the South Asia results,
and Cambodia is excluded from the East Asia results. In the case of Nepal, its food insecurity
score is much lower than that of the other South Asian countries, whereas Cambodia’s is much
higher. With the inclusion of these two outliers, the food insecurity scores for South Asia
and East Asia are roughly equal at 31 percent. a Transition refers to former Communist countries.
b
Members of Organization for Economic Co-operation and Development. c Low income is
deﬁned as a 2005 GDP per capita of less than USD 5,000 purchasing power parity; middle
income, as USD 5,000– 13,000; and upper income, as greater than USD 13,000.
Source: Data are from the GWP (Gallup 2010b).

highest in sub-Saharan Africa, which is by far the poorest region in the world
in monetary terms. Food insecurity in South Asia is higher than in East Asia,
as expected, but only when two large outliers, Nepal and Cambodia, are ex-
cluded.9 In Latin America, food insecurity is surprisingly high (34 percent).
This may relate to the greater prevalence of urban poverty and of relatively
poor net food consumers, although this is only a speculation.
The data also suggest a strong income gradient for food insecurity.
Low-income countries have food insecurity rates that are 17 percentage points
higher than middle-income countries, and the same difference is observed
between middle- and upper-income countries. In terms of correlations with
other welfare indicators (table 2), there is some support that cross-country pat-
terns impart meaningful information. Of course, extremely high correlations
are not necessarily expected given the well-known problems associated with
measuring hunger and poverty10 and the fact that anthropometric indicators

9. Self-assessed food insecurity in Cambodia is unusually high (67 percent), but in Nepal, it is
extremely low (9 percent). Including these two countries leaves the South and East Asian means roughly
equal, at 31 percent.
10. Indeed, in the context of critiquing standard poverty measures, Deaton (2010) suggested that
the Gallup indicators used in this study might be more reliable than the World Bank poverty estimates.
As a rough demonstration of their suitability, Deaton showed that the food security variable is highly
correlated with GDP per capita.
8 THE WORLD BANK ECONOMIC REVIEW

T A B L E 2 . Correlations between the Self-Reported Food Security Indicator and
Other Indicators of Income, Poverty, Hunger, and Malnutrition, Circa 2005
Alternative poverty/hunger indicator (source) Self-reported hunger

GDP per capita, purchasing power parity, log Correlation 2 0.71***
(World Bank) No. of obs. 44
Household income per capita, USD, log Correlation 2 0.68***
(World Bank Povcal) No. of obs. 59
Prevalence of hunger Correlation 0.58***
(FAO) No. of obs. 62
Prevalence of poverty, USD 1/day Correlation 0.77***
(World Bank Povcal) No. of obs. 58
Prevalence of poverty, USD 2/day Correlation 0.67***
(World Bank Povcal) No. of obs. 49
Prevalence of low-BMI women, excluding outliers Correlation 0.73***
(DHS & WHO) No. of obs. 17
Prevalence of underweight preschoolers, log Correlation 0.55***
(DHS & WHO) No. of obs. 45
Prevalence of stunted preschoolers, log Correlation 0.48***
(DHS & WHO) No. of obs. 45

Note: *, **, and *** indicate signiﬁcance at the 10 percent, 5 percent, and 1 percent levels, re-
spectively. All variables are measured in 2005 or the nearest available year. Log indicates that
variable is expressed in logarithms to account for a nonlinear relationship. Excluding outliers
refers to the exclusion of six countries with the highest prevalence of low-BMI women in the
sample, all above 20 percent: India, Bangladesh, Ethiopia, Cambodia, Nepal, and Madagascar.
Without this exclusion, the correlation is statistically insigniﬁcant. Samples vary in size because of
the paucity of some of the poverty and malnutrition indicators.
Source: Dependent variable is from the GWP (Gallup 2010b). The sources of the independent
variables are as follows: World Bank, World Bank (2010b) WDI; World Bank Povcal, World
Bank (2010a); FAO; Food and Agriculture Organization (2011); DHS; Demographic Health
Surveys (2010); WHO, World Health Organization (2010).

are heavily inﬂuenced by nonfood factors, such as health, education, family
planning, and cultural norms. Bearing this in mind, we ﬁnd that GDP per
capita, mean household income, poverty rates, hunger rates, and anthropomet-
ric indicators are signiﬁcantly correlated with the two GWP indicators, almost
invariably at the one-percent level (table 2). The correlations are particularly
strong for the (logarithmic) income and poverty indicators. In a very small
sample—which excludes six important outliers—the correlation between the
GWP indicators and the body mass index (BMI) of adult women is also very
high (0.68). Table S1.1 in the appendix presents the full correlation matrix
among the variables. It shows that the correlations between the GWP measure
and the various benchmarks are at least as strong as the benchmark correla-
tions for the FAO hunger measure and the World Bank poverty measure, if not
stronger.
In table 3, we also show that the GWP food insecurity indicator is signiﬁ-
cantly explained by “relative food prices,” which is measured as the ratio of
Headey 9

T A B L E 3 . Whether Self-Assessed is Food Security Explained by Relative Food
Prices
Regression No. 1 2 3

Dependent variable Food price level Food insecurity Food insecurity
No. of observations 99 91 91
Constant 61.74*** 17.0** 31.1**
GDP per capita ($1,000s) 2.80*** 2 3.1*** 2 2.3***
GDP per capita, squared 0.04*** 0.03***
Food price ratio 63.8*** 48.7
Food price ratio, squared 2 19.4*** 2 9.2
Africa dummy 30.4 18.6
Latin America dummy 2 12.3 10.5
Asia dummy 5.0 4.6
Europe-plus dummy 2 12.5 5.9

R-squared 0.65 0.73 0.76
Adjusted R-squared 0.63 0.72 0.75

Note: *, **, and *** indicate signiﬁcant at the 10 percent, 5 percent, and 1 percent levels, re-
spectively. “Europe-plus” includes Eastern European countries plus North America and
Australasia. Note that self-reported food insecurity data are measured in 2005 or 2006, whereas
the food price ratio is measured in 2005.
Source: “Food insecurity” is from the GWP (Gallup 2010b) and is described in the text. GDP
per capita is from the World Bank (2010) and is measured in constant purchasing power parity
dollars. “Relative food prices” are measured as the purchasing power parity of food and nonalco-
holic beverages relative to the nominal exchange rate for the year 2005. Information is from the
World Bank (2008b).

the purchasing power parity for food items to the exchange rate (both mea-
sured in 2005). This index can be interpreted as the extent to which a coun-
try’s food basket is expensive or cheap relative to the costs of importing food
(values of more than 100 imply that food is relatively expensive, whereas
values of less than 100 imply that food is relatively cheap). However, because
of Balassa-Samuelson effects, this indicator is likely to be higher in richer
countries than in poorer countries. Hence, we use multivariate regressions to
control for GDP per capita. However, even after controlling for GDP per
capita, there are still substantial variations in food prices across developing
countries (as the continent dummies in regression 1 suggest), which could be
explained by transport costs, variations in agricultural productivity, the
limited tradability of food ( partly due to tastes), or even exchange rate distor-
tions. Indeed, regression 2 suggests that variation in “relative food prices”
across countries signiﬁcantly explains variations in self-assessed food security
after controlling for GDP per capita. However, the relationship is nonlinear:
at low levels of food prices, the marginal effects of higher prices are quite
large, but at the highest observed levels of relative food prices, the marginal
effects are insigniﬁcantly different from zero. A caveat is that the result of
10 THE WORLD BANK ECONOMIC REVIEW

regression 2 in table 3 is not very robust to the inclusion of continental
dummies (introduced in regression 3), particularly the dummy for sub-Saharan
Africa. This lack of robustness appears to be because relative food prices and
self-assessed food insecurity are both very high in Africa.11 Speciﬁcally, the in-
clusion of continent dummies results in the food price coefﬁcients no longer
being signiﬁcant at the 10 percent level, although this insigniﬁcance also
applies to the continent dummy coefﬁcients, suggesting that multicollinearity is
an issue.
Overall, the results reported above present a mixed picture of the validity of
cross-country patterns in the Gallup data. On the one hand, there are certainly
some worrying outliers in the GWP indicator ( particularly in the 2005–06
round). On the other hand, the data as a whole are plausibly patterned across
countries and strongly correlated with other welfare indicators and relative
food prices. However, we acknowledge that many social scientists are wary of
subjective indicators of welfare, even if this skepticism has been moderated in
recent decades. There is, of course, an immense body of economic literature
that uses indicators of self-assessed well-being and health (e.g., Headey et al.
2010), including indicators collected by Gallup (Kahneman and Deaton 2010;
Deaton 2010; 2011). On the positive front, comparisons of self-assessed
poverty and objectively measured indicators of poverty have uncovered close
relationships between the two (Ravallion 2012). A recent assessment of food
insecurity questions in six developing countries also found that questions per-
taining to more severe forms of deprivation were highly comparable across
countries, although concepts related to anxiety and dietary quality were not
(Deitchler et al. 2010). In addition, there are longstanding concerns that such
measures are sensitive to framing, question ordering, and other response biases.
In terms of the third item, there is an extensive body of literature that examines
biases in self-reported indicators (see, e.g., Benitez-Silva et al. 2004; Krueger
and Schkade 2008; Ravallion 2012). A speciﬁc concern in the context of food
security is that respondents may believe that more negative answers increase
their chances of accessing food or cash transfers. Many such biases may only
exist at certain levels but disappear when trends in the data are observed.
However, any changes in question ordering could bias results, as a recent
paper by Deaton (2011) shows. Substantial measurement errors could also
mean that subjective indicators perform adequately in the cross-section but
poorly in ﬁrst differences (Bertrand and Mullainathan 2001). Clearly, there are

11. An issue here is that food prices may be higher in Africa because of the way in which the 2005
round of the International Comparison Program was conducted on a continental basis. Speciﬁcally, it is
possible that food prices in Africa are biased upward by methodological issues, although it is difﬁcult to
substantiate such a claim. A more general problem with purchasing power parities is the challenge of
ﬁnding common items to compare across countries. Exchange rate distortions may be problematic for
this index, although data on black market premia on exchange rates suggest that most exchange rate
distortions have declined markedly over time.
Headey 11

important reasons to explore the validity of trends in the Gallup indicator, not
just at certain levels.
III. EXPLAINING CHANGES IN SELF-ASSESSED FOOD
INSECURITY

In this section, we explore the validity of changes in self-assessed food insecuri-
ty at the national level by gauging whether trends in the GWP indicators are
explained by changes in disposable income. The underlying model for these
regressions is that the prevalence of food insecurity (F) at time t in country i is
a function of disposable income per capita, or nominal income per capita (Y),
deﬂated by a relevant set of prices (P):

Fi;t ¼ f ðYi;t =Pi;t Þ: ð1Þ

Although intuitive in principle, in practice, disposable income at the national
level is measured with considerable error for several reasons. First, income
inequality means that GDP per capita may be a ﬂawed indicator of the pur-
chasing power of a poor or vulnerable household in a country (the same is true
of GDP growth as an indicator of changes in welfare). Second, the price index
(P) used to deﬂate GDP per capita (the GDP deﬂator) may not represent the
consumption patterns of the food-insecure population because the budget share
they allocate to food expenditures will typically be higher than the share
employed in calculating the (consumer price index) CPI.
Because of these complications, it is not obvious that changes in real GDP
per capita adequately capture trends in the purchasing power of the poor.
Hence, in the regressions below, we estimate several different speciﬁcations.
First, we vary the choice of price index used to deﬂate growth in per capita
GDP (the GDP deﬂator, the total CPI, and the food CPI). Second, we test
whether changes in the total CPI or changes in relative food prices (i.e., the
food CPI over the nonfood CPI) provide some additional explanatory power.
Third, we test whether these relationships vary over income levels, in accor-
dance with Engel effects and the fact that welfare programs may play a larger
role in determining food security in wealthier countries than economic growth.
Finally, we add ﬁxed effects to the speciﬁcations to partially control for unob-
servable factors, such as income inequality and social safety net.12
In addition to these issues of speciﬁcation, there are some measurement con-
siderations. First, in our preferred regression models, we specify the dependent
variables as the change in the prevalence of food insecurity across two successive
periods. This approach is in contrast to most of the analogous growth-poverty lit-
erature, in which it is common to measure the dependent variable as a percentage

12. Although adding ﬁxed effects would seem desirable in principle, the valid addition of ﬁxed
effects rests on the assumption that both right-hand side variables are strictly exogenous at all leads and
lags, which is unlikely. Hence, we do not solely rely on the ﬁxed effects estimator.
12 THE WORLD BANK ECONOMIC REVIEW

change. However, taking the percentage changes of a prevalence rate can cause
scaling problems and create outliers (Deaton 2006; Headey 2011b).13 The only
signiﬁcant advantage of using a percentage change is that it allows for the deriva-
tion of elasticities that can be directly compared to the literature that examines the
impact of economic growth on poverty. Therefore, in some of our results, we also
report these elasticities, although our preferred estimates focus on ﬁrst differences.
A second issue pertains to measurement error in the Gallup data. Some ap-
parent outliers are indicative of this measurement error. In ﬁgure 1, we consid-
er potential outliers more systematically with scatter plots between changes in
food security and various indicators of economic growth and price changes.14
In all of the scatter plots, there are some potentially inﬂuential outliers, includ-
ing Azerbaijan, Angola, and Venezuela, which are three oil producers, several
Eastern European countries (Armenia, Latvia, Estonia, and Ukraine) and
several African countries (Tanzania, Mali, and Malawi). Note that these outli-
ers are sometimes driven by large changes in the dependent variable as well as
by unusual economic growth or inﬂation rates. Measurement error is therefore
a problem in both the left- and right-hand side variables.
To gauge the inﬂuence of outlying observations, we calculated dfbetas (an
indicator of the inﬂuence of outliers) and earmarked observations with dfbetas
greater than 0.2.15 One option is to run regressions that exclude outliers,
which we do in the case of ﬁxed effects regressions. Another option is to use a
robust regressor that downweights outlying observations without completely
discounting them. Hence, we use both robust regressors and ﬁxed effects
estimates that exclude these outlying observations. Furthermore, we report

13. The problem with taking percent changes in prevalence rates can be illustrated with an example
of a country with high food insecurity and a country with low food insecurity. In the food-insecure
country, suppose that food insecurity decreases from 42 percent at time t 2 1 to 40 percent at time t.
This yields a ﬁrst difference of two percentage points and a percent change of approximately 2 4.7
percent (that is, 2/40 Â 100). However, an equally large reduction in malnutrition prevalence in the
food-secure country from 4 to 2 percent yields a percent change of 50 percent. Not only is a 50 percent
change likely to be an outlier, but it is also 10 times the value of the equally large reduction in
malnutrition in the high-malnutrition country. Of course, one could argue that this may not matter if
percent differences are applied to the right-hand-side variables. In the case of per capita income,
however, this is not true because the denominator (initial income) is invariably large enough to produce
more meaningful estimates of percent change. Moreover, percent changes in income make sense if there
is a diminishing marginal impact of income on food insecurity.
14. Note that in all our regressions, we exclude observations for Zimbabwe because of its
hyperinﬂationary episode, which leaves the country as an enormous outlier on the food inﬂation-food
insecurity relationship.
15. This cut-off is fairly conservative. The usual cut-off for this sample size, 2/sqrt(N), is equal to
0.12. We calculate these dfbetas for various models and exclude a common set of outlying observations:
Algeria, 2009; Angola, 2008; Armenia, 2007 and 2009; Azerbaijan, 2007; Botswana, 2008; China,
2008; Denmark 2007 and 2008, and 2009; Djibouti, 2009; Iraq, 2008 and 2009; Kenya, 2007; Kuwait,
2009 and 2010; Romania, 2007; Rwanda, 2009; Tanzania, 2008; Trinidad and Tobago, 2008;
Vietnam, 2009; and Zimbabwe, all observations. A good explanation of dfbetas can be found in Stata
Web Books: Regressions with Stata, Chapter 2 – Regression Diagnostics: http://128.97.141.26/stat/
stata/webbooks/reg/chapter2/statareg2.htm
Headey 13

F I G U R E 1. Scatter plots of self-reported food insecurity, economic growth, and
various inﬂation indicators.

Sources: The Y-axis variable is from the GWP (Gallup 2010b). Economic growth data are
from the IMF (2011), and food inﬂation data are from the ILO (Headey 2011b).

ordinary least squares regressions in appendix S1, in which all outliers are
included.
Turning to some results, we begin with descriptive statistics and correlations
for our dependent and independent variables (tables 4 and 5). Over the entire
period, the mean change in the ﬁrst difference of the food insecurity measure
was close to zero (0.2), although the standard deviation and range of this vari-
able is quite large. The statistics for the percentage change in food insecurity
show a similar pattern and indicate the presence of some of the previously
mentioned problems with the use of percentages of a prevalence variable.
There is a tendency to inﬂate small changes at lower levels of food insecurity
due to the small base. Next, the three economic growth indicators show similar
variation around the mean, but the relatively rapid rate of food inﬂation over
this period means that the GDP growth deﬂated by the food CPI has a mean of
only 0.4, whereas deﬂating by the GDP or CPI deﬂators results in means of 2.7
percent and 2.9 percent, respectively. Thus, food inﬂation typically exceeded
nonfood inﬂation. Turning to table 5, it is noteworthy that the correlations
among different price indices are quite large, as high as 0.82 in the case of the
14 THE WORLD BANK ECONOMIC REVIEW

relationship between food inﬂation and total inﬂation. Table 5 also presents
some bivariate evidence that changes in food security are signiﬁcantly related
to both economic growth and overall inﬂation but not to our estimates of rela-
tive food inﬂation.
Table 6 reports the results for the full sample of countries with ﬁrst differ-
ences in food insecurity as the dependent variable and various indicators of
economic growth as the sole explanatory variable. Regressions 1 through 3
report results from the analysis using the robust regressor, and regressions 4
through 6 report results from using the ﬁxed effects estimator. The main
ﬁnding from table 6 is that the economic growth coefﬁcient is always highly
negative, signiﬁcant, and quite large in magnitude. In terms of the size of the
coefﬁcients, the point estimates suggest that doubling the GDP per capita
would reduce the rate of food insecurity by 12 to 24 percentage points, de-
pending on the estimator and the indicator of economic growth. In general, the
ﬁxed effects estimators produce larger estimates. When ﬁxed effects are used
and outliers are removed, the choice of deﬂator makes virtually no difference.
In table 6, we report elasticities in addition to the ﬁrst difference coefﬁcients.
The elasticities are quite large, varying from 0.47 to 1.25, and are commensu-
rate in size to growth-poverty elasticities (for example, those reported in
Christiaensen et al. 2011).
In table 7, we run the same regressions with the addition of separate price
change indicators to determine whether certain types of inﬂation have addition-
al explanatory power over real economic growth rates. Speciﬁcally, we add in-
ﬂation in the total CPI and food CPI relative to the nonfood CPI. The ﬁrst
represents an aggregate price effect, and the second represents a relative food
price effect. Table 7 shows that overall inﬂation has a signiﬁcant positive effect

T A B L E 4 . Descriptive Statistics for Dependent and Independent Variables
Count Mean Std. De. Min. Max.

Change in food insecurity 296 0.2 7.2 2 31.0 24.0
Percent change in food insecurity 290 4.9 31.6 2 83.0 200.0
Economic growth (GDP deﬂator) 291 2.9 5.7 2 17.6 32.1
Economic growth (CPI deﬂator) 276 2.7 8.4 2 27.1 41.1
Economic growth (food CPI deﬂator) 271 0.4 8.4 2 34.2 34.0
Total CPI inﬂation 276 8.6 7.5 2 8.9 51.8
Food CPI inﬂation 276 10.9 9.6 2 11.5 67.6
Nonfood CPI inﬂation 276 6.4 6.2 2 9.5 34.4
Relative food inﬂation 276 4.4 7.7 2 20.8 41.8

Note: All data are in percent or percentage points. Economic growth is reported with three dif-
ferent means of deﬂation: the GDP deﬂator, the CPI deﬂator, and the food CPI deﬂator. Relative
food inﬂation is the change in the ratio of the food CPI to the nonfood CPI.
Source: Food insecurity is from the GWP (Gallup 2010b). Economic growth data are from the
IMF (2011), and all inﬂation data are from the ILO (Headey 2011b).
Headey 15

T A B L E 5 . Correlations between Changes in Food Insecurity and Various
Explanatory Variables
Change in food Economic Total Food Nonfood
insecurity growtha inﬂation inﬂation inﬂation

Economic growtha 2 0.10**
Total inﬂation 0.18*** 0.20***
Food inﬂation 0.15*** 0.19*** 0.82***
Nonfood inﬂation 0.19*** 0.19*** 0.71*** 0.51***
Relative food inﬂationb 0.04 0.06 0.41*** 0.76*** 2 0.17***

Note: *, **, and *** indicate signiﬁcance at the 10 percent, 5 percent, and 1 percent levels, re-
spectively. a Growth in GDP per capita deﬂated by the GDP deﬂator. b Changes in the ratio of
the food CPI to the nonfood CPI.

on the prevalence of food insecurity. Again, the coefﬁcient point estimates are
larger in the ﬁxed effects regressions (0.22 versus 0.11), but these marginal
effects are relatively large for both estimators. Doubling the CPI, for example,
is expected to increase the prevalence of food insecurity by 11 to 22 percentage
points, holding real economic growth constant.
Somewhat surprisingly, the relative food inﬂation coefﬁcients in table 7 are
insigniﬁcant at the 10 percent level, but they are still positive (in one regression,
the relative food inﬂation coefﬁcient is signiﬁcant at the 13 percent level). One
explanation may be greater measurement error in relative food inﬂation
because we were required to estimate nonfood inﬂation rates for approximately
half of our sample.16 Nevertheless, the fact that food inﬂation was the main
driver of overall inﬂation over the period in question (food inﬂation explained
almost 80 percent of variation in total inﬂation from 2006 to 2008 in develop-
ing countries) indirectly points to the generally adverse role of higher food
prices on self-assessed food insecurity. Moreover, a signiﬁcant additional effect
of overall price inﬂation on food insecurity could be consistent with microeco-
nomic theories of labor markets. Speciﬁcally, most poor people engaged in
wage labor (i.e., those who are not self-employed, such as farmers) tend to
work in labor markets that are characterized by substantial slack (unemploy-
ment or underemployment). If various food and nonfood prices increase, then
the nominal wages of workers in such markets would not be expected to in-
crease commensurately, leading to a fall in real incomes (Headey et al. 2012).

16. The reason for the larger error in the relative food inﬂation measure is that the ILO only reports
the total CPI and the food CPI. Because relative food inﬂation is measured as changes in the ratio of the
food CPI to the nonfood CPI, we had to derive the nonfood CPI from the total CPI, the food CPI, and
the share of food in the total CPI. However, only approximately 50 percent of countries reported the
food weight to the ILO, so we were required to estimate food CPI weights for the remaining countries
using regressions against GDP per capita (i.e., Engel effects). This interpolation is the best we could do,
but it may mean that relative food inﬂation is measured with sizeable error. That said, alternative
indicators of relative food prices, such as the change in the food CPI minus the change in the total CPI,
essentially yield the same insigniﬁcant results.
16
THE WORLD BANK ECONOMIC REVIEW
T A B L E 6 . Regressions of Changes in Self-Reported Food Insecurity against Economic Growth
Regression No. 1 2 3 4 5 6

Means of deﬂating economic growth GDP deﬂator Total CPI Food CPI GDP deﬂator Total CPI Food CPI
Outliers removed? No No No Yes Yes Yes
Regressor Robust regressor Robust regressor Robust regressor Fixed effects Fixed effects Fixed effects
Economic growth Coefﬁcients 2 0.24*** 2 0.14*** 2 0.12*** 2 0.21*** 2 0.22*** 2 0.23***
(0.06) (0.04) (0.04) (0.08) (0.06) (0.07)
Elasticities 2 0.56** 2 0.55*** 2 0.47** 2 1.25** 2 0.93*** 2 0.82***
(0.27) (0.20) (0.19) (0.48) (0.29) (0.30)
No. of observations 291 275 271 271 256 252
No. of countries 120 112 111 113 106 105
R-squared 0.05 0.04 0.03 0.06 0.05 0.05

Note: *, **, and *** indicate signiﬁcance at the 10 percent, 5 percent, and 1 percent levels, respectively. Standard errors are reported in parentheses.
The robust regressions are estimated using the rreg command in stata, with default settings. For ﬁxed effects regressions, standard errors are adjusted for
country clusters. Outliers are identiﬁed based on dfbetas greater than 0.20. Economic growth is the percent change in GDP per capita between the two
years in which the GWP surveys were conducted. Note that the robust regressor does calculate a pseudo R-squared, but it is generally regarded as inap-
propriate to report this value. Hence, the R-squared reported in this table is derived from an ordinary least squares regression that excludes outlying
values.
Source: Dependent variables are from the GWP (Gallup 2010b). Economic growth data are from the IMF (2011), and food and total CPI data are
from the ILO (Headey 2011b).
T A B L E 7 . Augmenting the Regressions with Measures of Inﬂation
Regression No. 1 2 3 4

Means of deﬂating economic growth Total CPI Total CPI Total CPI Total CPI
Regressor Robust regressor Robust regressor Fixed effects Fixed effects
Economic growth (CPI) 2 0.14*** 2 0.13*** 2 0.22*** 2 0.19***
(0.04) (0.04) (0.06) (0.07)
Relative food inﬂation 0.04 0.12
(0.05) (0.08)
Total inﬂation 0.11*** 0.22**
(0.05) (0.11)
Number of countries 105 105 105 105
Number of observations 252 252 252 252
R-squared: overall 0.05 0.04 0.05 0.06

Note: *, **, and *** indicate signiﬁcance at the 10 percent, 5 percent, and 1 percent levels, respectively. Standard errors are reported in parentheses.
Note that outliers are removed for all regressions. Outliers are identiﬁed based on dfbetas greater than 0.20. The robust regressions are estimated using
the rreg command in stata with default settings. For ﬁxed effects regressions, standard errors are adjusted for country clusters. Economic growth is the
percent change in GDP per capita between the two years in which the GWP surveys were conducted deﬂated by the total CPI. Total inﬂation is the
percent change in the food CPI between the month of the GWP survey and the month of the previous GWP survey, where the food CPI in any given
month is actually the average food CPI in the previous 12 months. Relative food inﬂation is the percentage change in the ratio of the food CPI to the
nonfood CPI, where the both CPIs in any given month are actually the average CPIs in the previous 12 months. Note that the robust regressor does calcu-
late a pseudo R-squared, but it is generally regarded as inappropriate to report this value. Hence, the R-squared reported in this table is derived from an
ordinary least squares regression that excludes outlying values.
Source: Dependent variables are from the GWP (Gallup 2010b). Economic growth data are from the IMF (2011), and food and total CPI data are
from the ILO (Headey 2011b).

Headey
17
18 THE WORLD BANK ECONOMIC REVIEW

Hence, it is possible for nominal price increases to induce real wage declines,
and there is signiﬁcant evidence pointing to the adverse impact of inﬂation on
poverty reduction (see Ferreira, Prennushi, and Ravallion 2000).17
Finally, we ran a number of additional speciﬁcation tests related to income-
level effects and alternative inﬂation effects. Speciﬁcally, we ran interaction
terms with GDP per capita (in linear and log form) and with income dummy
variables (low, middle, upper). Although we strongly expected that changes in
food insecurity would be more sensitive to changes in disposable income at
lower levels of income, there were no signiﬁcant interaction terms (results avail-
able upon request). We suspect that this result may be driven by the fact that
growth rates, inﬂation rates, and changes in food insecurity were all much
lower in upper-income countries, which would have the effect of making the
relationships approximately linear.
From the perspective of providing validation that changes in self-assessed
food insecurity impart useful information, the results in tables 6 and 7 are en-
couraging. It is particularly encouraging that changes in real GDP per capita
signiﬁcantly explain changes in self-assessed food insecurity, suggesting that the
latter is sensitive to changes in disposable income.
Despite signiﬁcant and robust marginal effects, there are some caveats to
these results. First, there is the inﬂuence of outliers. In the online appendix
(table S1.2), we report the results of reestimating the regressions in table 6 and
including outliers. Although all of the economic growth coefﬁcients are still sig-
niﬁcant at the 10 percent level or higher, the standard errors are signiﬁcantly
larger, and the point estimates are sometimes larger and sometimes smaller in
magnitude than those in table 6. Our treatment of outliers therefore does not
lead to qualitatively different results.
Nevertheless, the presence of outliers and the low explanatory power of the
regressions reemphasize our concerns about measurement error. These con-
cerns must be tempered, however, because the analogous literature on the
impact of economic growth on poverty reduction reports regression models
with similarly low explanatory power (see Christiaensen et al. 2011, for
example), suggesting that these types of short-run poverty/food insecurity epi-
sodes suffer from the measurement errors and misspeciﬁcation problems noted
above. Although the presence of large marginal effects of economic growth and
inﬂation rates on self-assessed food insecurity are encouraging, we must inter-
pret trends in the latter quite cautiously.

17. Ferreira et al. (2000) write, “While changes in the relative short-term returns to holding bonds
versus stocks may redistribute income only among the non-poor, there is one major asset-type impact
which affects the poor: inﬂation. The rate of inﬂation is a tax on money holdings. Because there are
barriers to entry in most markets for non-money ﬁnancial assets, the poor are constrained in their
ability to adjust their portfolio to rises in inﬂation. Typically, they will hold a greater proportion of
their wealth in cash during inﬂationary episodes than do the non-poor. The non-poor are generally
better able to protect their living standards from inﬂationary shocks than the poor.”
They go on to cite evidence from India, Brazil, the Philippines, and a larger cross-country review.
Headey 19

I V. M E A S U R I N G A N D I N T E R P R E T I N G K E Y T R E N D S I N
T H E G A L L U P D AT A

In the introduction to this paper, we noted our basic result at the global level:
132 million fewer people were food insecure in 2008 relative to 2005–06. In
this section, we examine Gallup trends in more detail by observing regional
variations within this global trend, considering important exclusions from the
sample, engaging in an important sensitivity analysis, and exploring the factors
that might explain the surprisingly positive global trend.
In table 8, we report simple averages of the GWP food insecurity indicator
by various regions of the developing world for 2005–06, 2008, and 2009.
These years quite neatly correspond to a precrisis survey round, a food crisis
round, and an early ﬁnancial crisis round. Starting at the top of table 8, we
observe what superﬁcially explains the very positive global trend: in the eight
most populous developing countries (excluding China), food insecurity de-
creased by 4.7 percentage points between 2005–06 and 2008. However, in
many other regions of the world, food insecurity increased, including coastal
West Africa (but not the Sahel), Eastern and Southern Africa, and Latin
America. In other developing regions, there was either no change or some im-
provement. We also note that the deterioration of food insecurity in much of
Africa and Latin America is consistent with a number of simulation studies
(see Headey and Fan 2010 for a review).
Although the results in table 8 cover the majority of the developing world’s
population, there are still sizeable omissions. Although the GWP surveys cover
China, we excluded the 2005–06 rounds due to speciﬁc concerns about biases
in the responses to the food insecurity question. However, a number of other
countries are lacking the requisite data for 2005–06 or 2008. China, of course,
has a population of over a billion people, but 16 other omitted developing
countries represent close to 600 million people. Hence, one way to explore the
sensitivity of our “global” estimate to the omission of these countries is to
posit some plausible trends for these omitted countries and then recalculate the
global ﬁgures.
With regard to China, the assessed GWP observations for 2006 and 2008
suggest an unrealistically large drop in food insecurity over that time (20 per-
centage points), which is probably related to the aforementioned problems
with the ordering of questions in the 2006 round. It is therefore pertinent to
consider a more plausible scenario for China and what this scenario would
suggest about global trends in food insecurity. Given China’s phenomenal eco-
nomic growth and rather limited level of food inﬂation (nominal mean incomes
increased by 65 percent over 2006–08, whereas the food CPI increased by ap-
proximately 30 percent), it is plausible that food insecurity fell several percent-
age points in China. We thus consider a 3-percentage-point reduction from
2006 to 2008 to be relatively conservative. However, the countries omitted
from one of more rounds of the GWP include many that could be suspected to
20 THE WORLD BANK ECONOMIC REVIEW

T A B L E 8 . Regional Trends in Self-Reported Food Insecurity (Percent
Prevalence)
No. of 2005– 06 surveys 2008 surveys 2009 surveys
Developing region obs. (precrisis) (food crisis) (ﬁnancial crisis)

Eight most populous 8 32.7 28.0 30.6
developing countries*
sub-Saharan Africa 14 55.8 54.6 57.2
West Africa, coastal 4 48.5 51.3 58.0
West Africa, Sahel 5 59.6 49.2 55.2
Eastern & Southern 5 57.8 62.8 58.6
Africa
Latin America & 15 33.2 36.4 35.7
Caribbean
Central America, 7 38.4 41.4 40.3
Caribbean
South America 8 28.6 32.0 31.6
Middle East (including 3 19.7 26.0 21.3
Turkey)
Transition countries 13 31.9 30.2 34.6
Eastern Europe 6 21.8 19.7 25.8
Central Asia 7 40.6 39.1 42.1
Asia 12 28.8 29.0 30.8
East Asia 7 30.1 30.6 32.7
South Asia 5 26.8 26.8 28.6

Note: * “Large and fast growing” includes India, Indonesia, Brazil, Pakistan, Bangladesh,
Nigeria, Mexico, and Vietnam but excludes China.
Source: Author’s calculations from GWP (Gallup 2010b) self-reported food insecurity preva-
lence rates.

have experienced rapid food inﬂation, including the Philippines (the largest rice
importer in the world), a number of Middle Eastern and North African coun-
tries (some of the largest wheat importers in the world), and Ethiopia (the
second largest country in Africa, one of the poorest countries in the world, and
a country that experienced one of the fastest inﬂation rates in the world over
2007–08). In table S1.3 in the appendix, we make rather pessimistic assump-
tions about trends in food insecurity in these 16 countries (based largely on ob-
served food inﬂation data) and adjust the raw GWP estimates by adding the
assumed changes in food insecurity from the omitted countries. The results of
this exercise are assessed in table 9. The inclusion of assumed changes for these
16 countries adds 62 million people falling into food insecurity rather than
coming out it, but the assumed trend in China would result in close to 40
million people coming out of poverty. In short, the core results reported in the
introduction are not highly sensitive to the omission of these admittedly impor-
tant countries.
Another objection may be that the 2005–06 GWP results are less reliable
than subsequent rounds because the ﬁrst round of the GWP may be regarded
Headey 21

T A B L E 9 . Alternative Estimates of Global Self-Reported Food Insecurity
Trends after Allowing for Omitted Countries (Millions of People)
Estimated change in global food
Estimation scenarios insecurity, 2005– 06 to 2007– 08

Raw results, 69 countries (excluding China), covering 2 132
57% of developing world population
As above, plus pessimistic assumptions for 16 omissions, 2 60
covering 67% of developing world population
As above, plus a 3-percentage-point reduction in China, 2 100
covering 87% of world population

Note: See text in this section for more details regarding the assumptions and data as well as
table A3.
Source: Author’s calculations from GWP data (Gallup 2010b), FAO Global Information and
Early Warning System data (2010), and ILO food inﬂation data (Headey 2011b).

as a trial run for Gallup. We have the option of using the second round of the
GWP in 2007 as a base year instead of the 2005–06 round, but the 2007
round contains fewer countries and does not include China. Nevertheless, the
2007 GWP round includes India and other large countries and therefore covers
approximately 43 percent of the population in the developing world. A second
potential problem with using the 2007 round as a base year is that maize and
wheat prices were already increasing in 2007, so it is difﬁcult to regard 2007
as a pure precrisis period. Thus, we might underestimate the food insecurity
impacts of the crisis if the 2005–06 round is shown to be unreliable. However,
we note that there is no analogous problem with the 2008 data. The vast
majority of the GWP surveys in 2008 were conducted in the last three quarters
of the year after international food prices peaked. Therefore, they cover the
period of peak international prices. Some lag in domestic food inﬂation may
still be a problem, although we have already assessed results for surveys con-
ducted in 2009, which may capture the twin effects of slower growth (due to
the ﬁnancial crisis) and higher food prices.
Bearing these caveats in mind, table 10 reports the results of calculating the
population-weighted averages of food insecurity prevalence and population
numbers for 2007 and 2008. The results thus suggest that there was basically
no change in the “global” prevalence of food insecurity between 2007 and
2008. However, table 8 also shows that this result is heavily driven by trends
in India, where food insecurity fell 4 percentage points from 2007 to 2008.
The bottom half of table 8 calculates trends excluding India (which, admitted-
ly, represents approximately one-quarter of the developing world’s population)
and ﬁnds that population-weighted food insecurity in the rest of the sample
went up by 2.53 percentage points, representing approximately 43 million
people. Therefore, using the 2007 round as a base suggests that many develop-
ing countries were somewhat worse off in the peak food crisis year relative to
22 THE WORLD BANK ECONOMIC REVIEW

T A B L E 1 0 . Changes in Self-Reported Food Insecurity from 2007 to 2008
Prevalence of food insecurity (%) Population of food insecure (millions)

48 developing countries (43.3% of developing world population)
2007 29.33% 821.4
2008 29.28% 820.1
2 0.05 percentage points 2 1.3 million
47 developing countries excluding India (23.3% of developing world population)
2007 31.51% 532.8
2008 34.04% 575.9
2.53 percentage points 43.1 million

Source: Author’s calculations from GWP data (Gallup 2010b).

the previous year. The largest increases in self-assessed food insecurity occur in
Tanzania (23 points), Turkey (21 points), Burkina Faso (14 points), Uganda
(14 points), Mozambique (12 points), Kenya (11 points), Ecuador (10 points),
Cameroon (9 points), Sri Lanka (9 points), Armenia (7 points), and Honduras
(7 points). Although we cannot ignore measurement error and the role of other
factors in explaining these trends (the result in Turkey stands out as somewhat
implausible), it is notable that many of the countries listed above did experi-
ence quite rapid food inﬂation. Indeed, the average rate of food inﬂation in
these countries was approximately 4 points higher than the rest of the sample.
Although we have explored validity issues in previous sections, another rele-
vant question is whether the GWP results are supported by any other survey ev-
idence. One other reasonably large survey of developing countries that was
conducted before and during the crisis is the Afrobarometer survey. A recent
working paper by Verpoorten, Arora and Swinnen (2011) explores trends in an
Afrobarometer indicator that pertains to a very similar question to the one
asked in the GWP and ﬁnds a 3-percentage-point increase in food insecurity in
urban Africa from 2005 to 2008 and a 2-percentage-point increase in rural
Africa. Thus, the overall picture of some deterioration in food insecurity in
Africa is common across both the GWP and Afrobarometer surveys. Second,
and perhaps most important, the most recent World Bank estimates of poverty
trends also suggest that global poverty fell between 2005 and 2008 on every
continent (World Bank 2012). Third, the FAO (2012) has revised its estimates
of large increases in global hunger in 2009. The most recent estimates show a
relatively steady decline in global undernourishment, consistent with both the
Gallup and World Bank poverty estimates.
Finally, it is worth exploring why the GWP results (and the new World
Bank and FAO numbers) tell a positive story at the global level. One clear
pattern is that events in the largest developing countries heavily inﬂuence any
appraisal of global trends, not only because of the obvious inﬂuence of their
sheer size on global trends, but also because many large countries are charac-
terized by limited food inﬂation, rapid economic growth, or both. The ﬁrst of
Headey 23

these is not surprising. Large countries are very reluctant to rely heavily on sig-
niﬁcant food imports and are more likely to impose export restrictions and set
aside signiﬁcant food reserves. For example, China, India, Indonesia, and
Vietnam all imposed some restrictions on grain exports in 2007 or 2008, and
Nigeria abolished a 100 percent tariff on rice imports (Headey and Fan 2010).
Of course, the effect of these attempts to insulate domestic markets on global
poverty is ambiguous given that effective trade restrictions by large countries
may protect their own poor but may hurt the rest of the world’s poor by spur-
ring further international food inﬂation. Another domestic policy factor that
may explain the apparent reduction of food insecurity in some of the larger de-
veloping countries is the spread of major social safety net programs in these
countries, particularly India’s National Rural Employment Guarantee Scheme.
However, in addition to these factors, strong economic growth in most of the
world’s largest developing countries clearly provides a plausible explanation
for the largely favorable trends in self-assessed data in these countries.
To examine disposable income issues more explicitly, we deﬂate nominal
economic growth in recent years by changes in the food CPI rather than by an
overall price index (as we did in some of the regressions in tables 6 and 7).
This indicator is clearly an imperfect indicator of food security because poor
people also spend money on nonfood items (but not much on fuel, which is the
major source of nonfood inﬂation) and because mean GDP growth is often not
representative of the income growth of lower income groups. Nevertheless, this
crude indicator of “food-disposable mean income” at least indicates whether
mean nominal income growth outpaced food inﬂation.
Figure 2 plots this indicator for the nine largest developing countries from
2005 to 2008, with Brazilian and Mexican incomes measured on a separate
axis because of scaling issues. The results are quite striking. “Mean food-
disposable income” rose by over USD 700 per capita in China, over USD
1,800 in Brazil, and over USD 400 in Indonesia. In India, the increase was sur-
prisingly modest (USD 80), but the increase was notably large in Nigeria (USD
240). In Mexico, however, the results are completely reversed, with food-
disposable incomes declining sharply in 2007 and especially in 2008, during
the so-called “tortilla crisis.” In the other countries, the data show much more
modest trends and some general declines in 2008 as food prices rose substan-
tially in Bangladesh, Vietnam, and Pakistan. Although there are variations
among these nine countries, it is clear that nominal income growth generally
outpaced food inﬂation in most country-year observations by a large margin in
the four most populous countries (China, India, Indonesia, and Brazil).
These results apply to the largest countries, but we can also experiment with
predicting changes in self-assessed food security for the entire sample of develop-
ing countries based on our regression results. Speciﬁcally, we use the growth and
inﬂation coefﬁcients derived in regressions 2 and 4 in table 7 in conjunction
with actual growth and inﬂation rates to predict changes in self-assessed food in-
security over 2006–08. The results of this simulation are reported in Table 11.
24 THE WORLD BANK ECONOMIC REVIEW

F I G U R E 2. Nominal average per capita GDP deﬂated by the food CPI, 2005
to 2008.

Source: The indicator above is nominal GDP per capita between 2005– 06 and 2007– 08 from
the IMF (2011) deﬂated by food CPI data from the ILO (Headey 2011b).

From these regressions, it is clear that economic growth reduces self-assessed
food insecurity, whereas the total inﬂation rate increases it. For the most part,
the results in table 7 suggest that these two effects cancel each other out over
2006–08. In one estimate, global food insecurity decreases by 25 million
people, and in the other, it increases by 6 million. In other words, the economet-
ric predictions lead to a conclusion that is qualitatively similar to the raw de-
scriptive statistics: a decrease in global self-assessed food insecurity or little or
no change overall.

V. C O N C L U S I O N S

The innovation of this study is its use of survey-based evidence, rather than
simulations, to assess the impact of the food crisis on global and regional food
insecurity. We ﬁnd no evidence that global food insecurity was higher in 2008
than it was in previous years, although it affected many regions, particularly a
number of countries in Africa and Latin America. Though qualiﬁed by mea-
surement issues, these results are broadly corroborated by recent World Bank
estimates of a declining global poverty trend over 2005–08. Our results also
cast doubt on the usefulness of simulation approaches in predicting global
poverty trends, although the more sophisticated of these approaches are still
useful for exploring the mechanisms and distributional impacts of food price
impacts in an experimental setting.
Finally, our results raise the question of whether self-assessed indicators
might be a useful addition to existing food security metrics. Further research is
needed in this regard. As we noted in section 2, self-assessed indicators are sus-
ceptible to a number of biases. Nevertheless, these weaknesses must be traded
off against the fact that such indicators are easily, quickly, and cheaply mea-
sured relative to household expenditure or consumption data. There are also
Headey 25

T A B L E 1 1 . Econometrically Predicted Changes in Self-Assessed Food
Insecurity over 2006–2008 (Millions of People)
Predicted change in food Predicted change in food
insecurity using regression 2 in insecurity using regression 4 in
table 7 table 7

Total predicted change in 2 104.5 2 152.8
self-reported food
insecurity
Change due to economic 79.4 158.8
growth
Change due to total inﬂation 2 25.1 6.0

Note: Changes in self-reported food insecurity are estimated by multiplying the coefﬁcients
from table 7 by changes in GDP per capita and changes in the total CPI from 2006 to 2008.
Source: Author’s calculations based on the GWP (Gallup 2010a).

some potentially signiﬁcant ways to improve self-reported data. For example, a
more disaggregated ordering of self-assessed food security (such as using scales
of 0 to 5) might reduce measurement errors. King et al. (2004) also argue for
anchoring vignettes to make measurements more comparable across different
socioeconomic groups. Another approach might be to ask households to report
the frequency of consumption across different groups rather than asking about
more subjective feelings of deprivation. Such dietary diversity or food con-
sumption scores have been shown to be strong predictors of household calorie
consumption and individual anthropometric outcomes (Wiesmann et al. 2006;
Arimond and Ruel 2006) and might capture the fact that reducing dietary diver-
sity is a common means of coping with higher staple food prices (Block et al.
2004). Others have argued for the use of sentinel sites to collect higher frequency
measurements of food security and nutrition outcomes (Barrett 2010).
Regardless of the path that is pursued, there are strong grounds for making
a large push to improve the measurement of food security. The global food
crisis of 2007–08 revealed some signiﬁcant deﬁciencies in our capacity to
monitor coping strategies and welfare impacts in an acceptable timeframe.
Moreover, if strong economic growth had not been prevalent in substantial
parts of the developing world prior to and during the food crisis, the impacts
of higher food prices might have been far more disastrous. Indeed, predictions
of higher food prices and continued price volatility in the next decade or
beyond (Headey and Fan 2010) would seem to justify greater investment and
experimentation in food security measurement in the near future.

REFERENCES

´quez, G., S. Daidone, and E. Mane. 2010. “A Cross-country Inquiry, Gustavo Anrı
Anrı ´quez, and Silvio
Daidone. Rising Food Prices and Undernourishment A Cross-Country Inquiry.” ESA Working Paper
No. 10-01. Food and Agriculture Organization, Rome.
26 THE WORLD BANK ECONOMIC REVIEW

Arimond, M., and M. Ruel. 2006. “Dietary Diversity Is Associated with Child Nutritional Status:
Evidence from 11 Demographic and Health Surveys.” The Journal of Nutrition 134: 2579–85.
Arndt, C., R. Benﬁca, N. Maximiano, A.M.D. Nucifora, and J. T. Thurlow. 2008. “Higher Fuel and
Food Prices: Impacts and Responses for Mozambique.” Agricultural Economics 39(2008 supple-
ment):497–511.
Barrett, S. C. 2010. “Measuring Food Insecurity.” Science 327: 825– 8.
Benitez-Silva, H., M. Buchinsky, H.M. Chan, S. Cheidvasser, and J. Rust. 2004. “How Large Is the
Bias in Self-Assessed Disability?” Journal of Applied Econometrics 19: 649–70.
Bertrand, M., and S. Mullainathan. 2001. “Do People Mean What They Say? Implications for
Subjective Survey Data.” The American Economic Review 91(2): 67 –72.
Block, S.A., L. Kiess, P. Webb, S. Kosen, R. Moench-Pfanner, M.W. Bloem, and P.C. Timmer. 2004.
“Macro Shocks and Micro Outcomes: Child Nutrition during Indonesia’s Crisis.” Economics and
Human Biology 2(1): 21 –44.
Christiaensen, L., L. Demery, and J. Ku¨ hl. 2011. “The Role of Agriculture in Poverty Reduction: pAn
Empirical Perspective.” Journal of Development Economics 96(2), 239 –254.
de Hoyos, R., and D. Medvedev. 2009. “Poverty Effects of Higher Food Prices: A Global Perspective.”
Policy Research Working Paper No. 4887. The World Bank, Washington DC.
Deaton, A. 2006. “Global Patterns of Income and Health: Facts, Interpretations, and Policies.” NBER
Working Paper No.12735. National Bureau of Economic Research, Cambridge, MA, 2006.
———. 2010. “Price Indexes, Inequality, and the Measurement of World Poverty.” Presidential Address
to the American Economic Association. Atlanta, GA.
———. 2011. “The Financial Crisis and the Well-Being of America.” NBER Working Paper No.
17128. National Bureau of Economic Research, Cambridge, MA.
Deitchler, M., T. Ballard, A. Swindale, and J. Coates. 2010. “Validation of a Measure of Household
Hunger for Cross-Cultural Use.” Food and Nutrition Technical Assistance II Project (FANTA-2)
Technical Report. USAID, Washington, DC.
Dessus, S., S. Herrera, and R. de Hoyos. 2008. “The Impact of Food Inﬂation on Urban Poverty and Its
Monetary Cost: Some Back-of-the-Envelope Calculations.” Policy Research Working Paper 4666.
The World Bank, Washington DC.
FAO. 2002. “Measurement and Assessment of Food Depreviation and Undernutrition,” Food and
Agriculture Organisation (FAO), Rome. ftp://ftp.fao.org/docrep/fao/005/y4250e/y4250e00.pdf.
———. 2009. “The State of Food Insecurity in the World 2008.” Food and Agriculture Organisation
(FAO), Rome.
———. 2012. “The State of Food Insecurity in the World 2012.” Food and Agriculture Organisation
(FAO), Rome.
Ferreira, F., G. Prennushi, and M. Ravallion. 2000. “Protecting the Poor from Macroeconomic Shocks:
an Agenda for Action in a Crisis and Beyond.” World Bank, Washington, DC.
Gallup. 2010a. “World Wide Research Methodology.” Gallup, Inc., Washington, DC.
———. 2010b. “Worldwide Research – Country Data Set Details.” Gallup, Inc., Washington, DC.
Haddad, L., J. Hoddinott, and H. Alderman. 1996. Intrahousehold Resource Allocation in Developing
Countries: Models, Methods and Policy. Baltimore: Johns Hopkins University Press.
Headey, B., R. Muffels, and G.G. Wagner. 2010. “Long-Running German Panel Survey Shows That
Personal and Economic Choices, Not Just Genes, Matter for Happiness.” Proceedings of the
National Academy of Sciences Pre-print Online Version, October 4.
Headey, D. 2011a. “Was the Global Food Crisis Really a Crisis? Simulations versus Self-Reporting.”
IFPRI Discussion Paper 01087. International Food Policy Research Institute, Washington, DC.
———. 2011b. “Turning Economic Growth into Nutrition-Sensitive Growth.” Background paper for
the “2020 Conference: Leveraging Agriculture for Improving Nutrition and Health,” New Delhi, India.
Headey 27

Headey, D., and S. Fan. 2010. “Reﬂections on the Global Food Crisis: How Did It Happen? How Has
It Hurt? And How Can We Prevent the Next One?” International Food Policy Research Institute
(IFPRI), Washington, DC.
Headey, D., F. Nisrane, I. Worku, M. Dessa, and A. Seyoum Taffesse. 2012. “Urban Wage Behavior
during Food Price Hikes: The Case of Ethiopia.” Paper presented at “CSAE Conference 2012:
Economic Development in Africa,” March 18-21, 2012, Centre for the Study of African Economies,
University of Oxford.
IMF. 2011.World Economic Outlook - April 2011. Washington, DC: International Monetary Fund (IMF).
Ivanic, M., and W. Martin. 2008. “Implications of Higher Global Food Prices for Poverty in
Low-Income Countries.” Agricultural Economics 39(s1): 405–16.
Ivanic, M., W. Martin, and H. Zaman. 2011. Estimating the Short-Run Poverty Impacts of the 2010–
11 Surge in Food Prices. Policy Research Working Paper 5633. The World Bank, Washington, DC.
Kahneman, D., and A. Deaton. 2010. “High Income Improves Evaluation of Life but Not Emotional
Well-Being.” Proceedings of the National Academy of Sciences 107(38): 16489– 93.
King, G., C.J.L. Murray, J. A. Salomon, and A. Tandon. 2004. Enhancing the Validity and
Cross-Cultural Comparability of Measurement in Survey Research. American Political Science
Review 98(1): 191–207.
Krueger, A.B., and D.A. Schkade. 2008. “The Reliability of Subjective Well-Being Measures.” Journal
of Public Economics 92(8-9): 1833–45.
Lasco, C.D., R.J. Myers, and R.H. Bernsten. 2008. Dynamics of Rice Prices and Agricultural Wages in
the Philippines. Agricultural Economics 38(3): 339–48.
Mason, N.M., T.S. Jayne, A. Chapoto, and C. Donovan. 2011. “Putting the 2007/2008 Global Food
Crisis in Longer-Term Perspective: Trends in Staple Food Affordability in Urban Zambia and
Kenya.” Food Policy 36(3): 350–67.
Ravallion, M. 2012. “Poor, or Just Feeling Poor? On Using Subjective Data in Measuring Poverty.”
Policy Research Working Paper Series with number 5698. World Bank, Washington, DC, 2010.
Swinnen, J., and P. Squicciarini. 2012. Mixed Messages on Prices and Food Security. Science
335(January 27): 405–6.
Tiwari, S., and H. Zaman. 2010. “The Impact of Economic Shocks on Global Undernourishment.”
Policy Research Working Paper Series with number 5215. World Bank, Washington, DC
USDA. 2008. “Food Security Assessment, 2007.” United States Department of Agriculture (USDA)
Economic Research Service (ERS), Washington, DC.
———. 2009. “Food Security Assessment, 2008-09.” United States Department of Agriculture (USDA)
Economic Research Service (ERS), Washington, DC.
———. 2011. “Ps&D Online Database,” United States Department of Agriculture (USDA) Economic
Research Service (ERS), Washington, DC.
Verpoorten, M., A. Arora, and J. Swinnen. 2011. Self-Reported Food Security in Africa During the
Food Crisis. LICOS Working Paper. LICOS Centre for Institutions and Economic Performance &
Department of Economics, University of Leuven (KUL).
Wiesmann, D., L. Bassett, T. Benson, and J. Hoddinott. 2006. Validation of the World Food
Programme’s Food Consumption Score and Alternative Indicators of Household Food Security.
IFPRI Discussion Paper 00870. International Food Policy Research Institute (IFPRI), Washington,
DC.
World Bank. 2010. World Development Indicators Online. Washington, DC: The World Bank.
———. 2008a. International Comparison Program Tables of Final Results. Washington, DC: World Bank.
———. 2008b. World Development Report 2008: Agriculture for Development. Washington, DC:
World Bank.
———. 2012. An update to the World Bank’s estimates of consumption poverty in the developing
world. Washington, DC: The World Bank.