World Development Vol. 27, No. 11, pp.

1955±1975, 1999
Ó 1999 Elsevier Science Ltd. All rights reserved
Printed in Great Britain
www.elsevier.com/locate/worlddev 0305-750X/99/$ - see front matter
PII: S0305-750X(99)00091-1

Are Determinants of Rural and Urban Food Security

and Nutritional Status Di€erent? Some Insights from
Mozambique
JAMES L. GARRETT and MARIE T. RUEL *
International Food Policy Research Institute, Washington, DC, USA
Summary. Ð Undernutrition of children 0±60 months old in Mozambique is much higher in rural
than in urban areas. Food security is about the same, although substantial regional di€erences
exist. Given these outcomes, we hypothesized that the determinants of food security and nutritional
status in rural and urban areas of Mozambique would di€er as well. Yet we ®nd that the
determinants of food insecurity and malnutrition, and the magnitudes of their e€ects, are very
nearly the same, although some di€erentiation in determinants of undernutrition does begin to
appear among children 24±60 months old. The di€erence in observed outcomes appears primarily
due to di€erences in the levels of critical determinants rather than in the nature of the determinants
themselves. Ó 1999 Elsevier Science Ltd. All rights reserved.

Key words Ð Africa, Mozambique, food security, nutrition, rural, urban

1. INTRODUCTION national household survey of living conditions in

Mozambique. The paper concludes by high-
National and municipal governments in lighting implications of the ®ndings for improv-
developing countries have long struggled to ing food and nutrition security in Mozambique
conquer urban poverty, food insecurity, and and, more generally, for policy and program
malnutrition. In an increasingly urbanized design in both rural and urban areas.
world, that challenge will not go away soon.
International aid agencies, whose programs
have traditionally concentrated on rural areas, 2. FOOD AND NUTRITION SECURITY IN
are now moving systematically to develop MOZAMBIQUE
strategies to improve urban livelihoods.
As they detail their strategies for urban areas, Mozambique faces severe challenges in
governments and assistance organizations ask eliminating poverty and food and nutrition
whether they can simply transfer their concep- insecurity. The country is only now emerging
tual frameworks and programs from rural areas from decades of civil strife and is still coping
to the cities. A number of studies have looked with a transition to a more liberalized market
at food security or nutritional status in rural economy. Despite Mozambique's largely rural
and urban areas (Alderman, 1990; Alderman population, urban food insecurity and malnu-
and Higgins, 1992; Blau, Guilkey, and Popkin, trition are signi®cant problems. In Mozam-
1996; Ricci and Becker, 1996; Sahn, 1994;
Sahn, 1988; Thomas, Strauss, and Henriques,
1991; Thomas and Strauss, 1992), but none * The authors are grateful to Gaurav Datt, Carlo del
have explored in depth the question of whether Ninno, Lawrence Haddad, Lourdes Hinayon, Haydee
the factors that determine food and nutrition Lemus, Jan Low, Margaret McEwan, John Maluccio,
security are di€erent between rural and urban Stephan Meershoek, Ellen Payongayong, and Prem
areas, and what the implications of these Sangraula for their comments and assistance. We are
di€erences are for the design and operation of also thankful to the entire team at the Mozambican
food and nutrition programs. Ministry of Planning and Finance, the National Statis-
The present study answers some of these tics Institute, and at IFPRI who assisted in the collection
questions using new data from a 1996±97 and preparation of the data on which this study is based.
1955
1956 WORLD DEVELOPMENT

bique, 62% of the urban population is poor, both areas, in ways that can assist policy
and 18% of the total poor live in urban areas makers and program administrators to act
(MPF/UEM/IFPRI, 1998). Given the UNÕs most e€ectively to reduce poverty and food and
(1998) estimate of an urbanization level of 35% nutrition insecurity in rural and urban areas of
and a population of 15.7 million people (INE, Mozambique.
1999), two million poor people now live in the
cities and towns of Mozambique. This is more
than the number of urban poor in highly 3. MODELS AND ESTIMATION
urbanized Colombia and over half the number PROCEDURES
of urban poor in Indonesia, with a population
more than 12 times that of Mozambique Following well-known expositions (Behrman
(Haddad, Ruel, Garrett, this issue). and Deolalikar, 1988; Strauss and Thomas,
Recent data also suggest that food insecurity 1995), we use a standard household utility
is slightly higher in urban areas of Mozambique model to examine the determinants of food
than in the countryside. Sixty-seven percent of security 3 and nutritional status by specifying,
the urban population is food insecure, respectively, a demand function for calories
compared to 64% of rural residents. 1 These and a production function for child nutritional
aggregate igures, however, mask important status. Conceiving of demand for calories as
regional di€erences. While levels of food inse- similar to demand for any other good, demand
curity in the rural areas of the central and for calories will be in¯uenced by income (Yh ),
southern regions, and in Maputo and other prices (Ph ), and demographic characteristics
cities, are roughly similar, at about 65±75%, and other exogenous factors (Zh ): The vector of
food insecurity is much lower in the northern prices includes not only food but also prices of
region of Mozambique, which is the most other purchased and home-produced ``goods,''
productive agricultural area, at only 48% such as nutrition and health.
(MPF/UEM/IFPRI, 1998). Maximizing utility subject to income
Prevalence of childhood malnutrition, on the constraints and the nutrition production func-
other hand, is clearly worse in rural than in tion, we can derive a reduced-form household-
urban areas, regardless of region (MPF/UEM/ level demand function for calories (Kh ). A
IFPRI, 1998), as is typical of most developing reduced-form equation includes only exoge-
countries (Ruel et al., 1998). In Mozambique nous variables. We consider income to be
46% of rural preschoolers are stunted (height- endogenous, so in the reduced form assets (Ah ),
for-age z-scores <ÿ2), compared to 26% in a predetermined variable, replaces income:
urban areas. 2 Despite its seemingly low preva-
Kh ˆ f Ph ; Ah ; Zh †:
lence in urban areas, stunting a€ects approxi-
mately 225,000 urban preschoolers. Considering Nutrition for individual i can be conceived of
the well-documented, long-term negative conse- as the output of a production function in which
quences of stunting on adult stature, body a speci®c technology translates inputs into
composition, work capacity, and women's nutritional outcomes, which are represented by
reproductive performance (Martorell, 1993), as some standardized anthropometric measure
well as new evidence of an association with such as height-for-age. Guided by the under-
increased risks of chronic disease and obesity lying biological and economic determinants of
(Barker, 1994), these numbers are alarming. nutritional status (UNICEF, 1990), we can
As one of the poorest countries in the world, generalize to say that nutrition is produced by a
with an annual per capita expenditure of set of inputs, including caring behaviors direc-
$US170 (MPF/UEM/IFPRI, 1998), Mozambi- ted toward the individual (Ci ), health status
can policy makers must make hard choices and the household environment (Hi ), and
about how best to use their limited domestic dietary intake (Ki ), to which calorie availability
resources. The above ®gures highlight the fact at the household level, Kh , contributes:
that they cannot focus their attention only on
Ni ˆ N Ci ; Hi ; Ki †:
rural areas. Yet information on factors a€ect-
ing food insecurity and malnutrition in urban
areas is scarce and most programs are designed (a) Econometric considerations
to alleviate poverty and food insecurity in rural
areas. This paper provides insights into the To estimate the independent e€ect of, say,
factors that a€ect food and nutrition security in income on calories or nutritional status, ordi-
 FOOD SECURITY AND NUTRITIONAL STATUS 1957

nary-least-squares (OLS) estimates will only be compensate for any shortfall. Factors other
unbiased if we can rule out correlation between than income and prices can also a€ect house-
the error term and all explanatory variables. hold calorie availability, mostly by in¯uencing
For income in the calorie availability regression preferences. These factors include household
or in the nutritional status regression, such a demographic structure (such as the presence of
correlation is likely to exist. For instance, some small children or the elderly and gender of the
unobserved in¯uence, such as entrepreneurial household head), educational levels of house-
talent, could in¯uence both household calorie hold members, and location (including di€er-
availability and household income, and so the ences among regions as well as between urban
explanatory variable (income) would be corre- and rural areas). Household income and the
lated with the error term. kinds of food available can also vary by season.
Use of OLS in the presence of such endo- Of special interest in Mozambique is whether
geneity gives biased and inconsistent estimates. those displaced during the war have unique
In principle, use of instrumental-variables (IV) characteristics that a€ect their food security.
estimation can resolve this problem. For IV The calorie availability model includes variables
estimation to eliminate the correlation between to re¯ect the in¯uence of each of these factors.
the explanatory variable suspected of endo-
geneity and the error term, the instrumental (ii) Nutritional status
variable must be contemporaneously uncorre- In addition to factors that a€ect the house-
lated with the error and correlated with the hold's access to food, which can a€ect an
potentially endogenous variable it is instru- individual's own dietary intake, a child's
menting. But, if the correlation between the nutritional status will also be a€ected by the
instrumental variable and its corresponding hygienic condition of the household, ease of
endogenous variable is weak, the variance-co- access to and quality of health care, and
variance matrix of the IV estimator will maternal caregiving practices.
increase. Consequently, the IV estimator may Because the survey was designed primarily to
not be very precise, and, using a criterion of collect expenditure and demographic informa-
mean-squared error, we might prefer the OLS tion, it did not collect detailed information on
estimator (Kennedy, 1992). the proximal determinants of nutritional status,
In addition, if even a weak relationship exists such as individual dietary intake, nor on genetic
between the instrumental variable and the error or biological factors that are associated with
in the equation with the potentially endogenous growth such as mother's height. As a result, in
variable, IV estimates may themselves be some cases we use household-level variables to
inconsistent. In ®nite samples (even large ones) represent the e€ects of these potential determi-
if this association exists, IV estimates will be nants at the individual level, while understand-
biased in the same direction as OLS estimates, ing that intrahousehold mechanisms mediate
and the magnitude of the bias will approach the e€ect of these factors on child nutrition.
that of OLS as the F-statistic on the excluded Factors a€ecting the household's access to
instruments in the ®rst-stage regression goes to food, which may also a€ect the child's access to
zero (Bound, Jaeger, and Baker, 1995). food, are detailed above in the discussion of the
calorie availability model. Maternal educa-
(b) Conceptual and empirical models tional attainment is used as a proxy for caring
practices. 4 Use of prenatal care is included as a
Conceptual models and empirical evidence proxy for preventive health care use, and
can help place this theoretical discussion in availability of a latrine, source of water, and in-
context and suggest which variables belong in house crowding (denoted by number of rooms
the estimated equations and what estimating per capita) are used to capture the environ-
procedures are appropriate. mental conditions of the household. Because
Mozambique is home to over 40 di€erent tribal
(i) Calorie availability languages, the mother's ability to speak
The access that a household has to food Portuguese, the country's lingua franca, is
depends on whether the household has enough included as a potential indicator of the
income to purchase food at prevailing prices or mother's ability to access or understand health
has sucient land and other resources to grow or nutrition education messages. As local clin-
its own food. It can also receive assistance from ics may in fact use local languages, maternal
formal programs or informal networks to ability to speak Portuguese may also capture
1958 WORLD DEVELOPMENT

further enrichment of a mother's own human possession, use of prenatal care, and latrine
capital or a wealth e€ect not otherwise captured availability), although this was generally less
in the model. than 10% of the observations. These variables
Because the determinants of malnutrition were recoded so that a missing or 0 value was
may di€er according to the age of the child, as represented by 0. A separate dummy variable,
suggested by others (Grosse, 1996; Sahn and coded 1 when the value for the corresponding
Alderman, 1997), 5 we separated the sample variable was missing and 0 when it was not, was
into two subsamples of observations on chil- also included in the equation. This method
dren less than 24 months of age and those 24± allows us to retain other nonmissing observa-
60 months old and ran age-speci®c models. tions from the household for which the infor-
mation is missing in the sample, but it removes
(iii) General the in¯uence of the missing value in estimating
In the calorie availability and the nutritional the parameter for where the household has no
status models, instead of reported income, we observation (Ward, 1982). The coecient on
use the value of total household consumption each of these variables is conditional on the
(also referred to by many, including us, as observation being present. Tests for levels of
household consumption expenditure). Expen- signi®cance of the variable that denotes a
ditures are a better representation than income missing observation indicate whether those
of total resources available to the household households with missing observations di€er in
because households typically try to smooth some way from those with observations. Tables
consumption over time in the face of ¯uctua- 1±3 indicate the number of observations that
tions in income. Consumption expenditure were available for each variable.
includes values for all current consumption, We test for equality of coecients in urban
imputing values where necessary for items such and rural areas using a dummy variable tech-
as rent or home-grown food. nique. Although an alternative method would
Because the variables for calories/AEU and be to estimate and test coecients from similar
consumption expenditure share a basis of models using separate urban and rural samples,
calculation (the quantities of food consumed), the dummy variable approach allows for easier
any unobserved factor that a€ects quantities of manipulation of the variables and permits
food consumed will a€ect them both and create information from both samples to be used
a source of endogeneity. Two-stage least when there is no signi®cant di€erence between
squares (2SLS) procedures were used to control them (Kennedy, 1992). For our models, we ®rst
for this endogeneity, with an index of house- interacted all variables (except regional and
hold assets (the total number of di€erent community-level dummies) with an urban-rural
household items and vehicles possessed by the dummy and tested the signi®cance of the
household) as the identifying instrument in interaction terms. Interactions that were insig-
combination with the other (exogenous) vari- ni®cant at the 0.1 level or higher were dropped
ables in the ®rst-stage equation. 6 STATA 5.0, from the model. The results for each model,
which provides corrected standard errors in then, are based on only one equation using a
2SLS estimation, was used to estimate the combined sample.
regression equations.
Factors speci®c to each community, such as
prices, could also a€ect calorie availability and 4. DATA
nutritional status. Unfortunately, the commu-
nity survey was not carried out in both urban The data are from a national cross-sectional
and rural areas, so comparable data are not household demographic and expenditure
available for these factors in both areas. To survey carried out by the Government of
control for unobserved community-level heter- Mozambique from February 1996 to March
ogeneity, we employed a community ®xed-ef- 1997. Data from 8,274 households were
fects model using a dummy variable for each of collected to be representative at the level of
the communities in the sample (an administra- each of the 10 provinces and the capital,
tively-de®ned neighborhood in an urban area Maputo. Although nonfood expenditures made
and a locality in a rural area). 7 monthly or quarterly were collected in the
Some variables had a number of missing principal questionnaire, food expenditures were
observations (adult and mother's education, collected from a daily consumption module
mother's ability to speak Portuguese, land that covered a seven-day recall period. House-
 FOOD SECURITY AND NUTRITIONAL STATUS 1959

Table 1. Variable description, means, and standard errors: Calorie availability modela
Variables Rural Urban
Mean Standard n Mean Standard n
error error
Continuous variables
Calories/day/AEU 3033.34 58.28 4454 2834.52 92.85 2009
Expenditure/day/capita (meticais) 4796.72 135.08 4454 10476.83 981.74 2009
Land per capita (hectares)b 0.59 0.04 4238 0.37 0.03 835
Percentage of household <5 years old 0.14 0.01 4454 0.15 0.01 2008
Percentage of household P 5 and 6 17 0.32 0.01 4454 0.34 0.01 2009
Percentage of household P 60 0.06 0.01 4454 0.05 0.01 2009
Household size (no. of members) 4.70 0.06 4454 5.51 0.15 2009
Dummy variables
Highest level of education by adult
male in household
None/illiterate 0.71 0.01 2566 0.27 0.04 386
Some but less than primary 0.20 0.01 791 0.29 0.02 499
Completed primary or more 0.09 0.01 302 0.44 0.04 851
Highest level of education by adult
female in household
None/illiterate 0.91 0.01 3812 0.54 0.04 885
Some but less than primary 0.06 0.01 333 0.25 0.02 473
Completed primary or more 0.02 0.00 114 0.21 0.03 505
War migrant 0.06 0.01 4454 0.03 0.01 2009
Female household head 0.21 0.01 4454 0.22 0.01 2009
Seasonsc
Early rains (Dec±Feb) 0.28 0.04 4454 0.29 0.07 2009
Rains (Mar±May) 0.24 0.04 4454 0.39 0.07 2009
Harvest (Jun±Aug) 0.19 0.03 4454 0.25 0.08 2009
Post-harvest (Sep±Nov) 0.29 0.04 4454 0.07 0.02 2009
Regions
Maputo ÿ ÿ ÿ 0.07 0.02 6463
Secondary citiesd ÿ ÿ ÿ 0.06 0.02 6463
Other urban areas ÿ ÿ ÿ 0.08 0.02 6463
Rural provinces, north 0.26 0.04 6463 ÿ ÿ ÿ
Rural provinces, central 0.37 0.04 6463 ÿ ÿ ÿ
Rural provinces, south 0.16 0.02 6463 ÿ ÿ ÿ
a
Weighted to re¯ect number of households in the sample. Standard errors adjusted for survey design.
b
Conditional on having an observation.
c
For central and northern provinces. Periods begin and end one month earlier for southern provinces.
d
Beira, Matola, and Nampula.

hold-level calorie availability was calculated of age. When households with values for calo-
from this daily consumption module. After ries/AEU outside the acceptable range were
excluding extraordinarily unreasonable values then omitted, 1,201 and 1,514 observations,
of calorie availability per Adult Equivalent respectively, remained.
Unit (<800 calories/AEU and >7000 calories/
AEU), the sample used for estimation included
6,463 households. 5. RESULTS
Anthropometric measures of height, along
with age in months, were also collected for all (a) Descriptive analyses
children 60 months old and younger in the
household. Z-scores were derived using the The descriptive data suggest some striking
WHO/CDC reference values (WHO, 1979). di€erences between rural and urban areas as
When outlying values of ‹5 Z-scores for well as some interesting similarities. Table 1
height-for-age were excluded, a total of 1,474 uses data from the calorie availability sample
observations remained on children <24 months and shows that, with a higher cost of living in
old and 1835 for those P 24 and 6 60 months the cities, the total mean consumption
1960 WORLD DEVELOPMENT

Table 2. Means and standard errors, nutritional status model, 0±23 monthsa
Variables Rural Urban
Mean Standard n Mean Standard n
error error
Continuous variables
Height-for-age Z score ÿ1.35 0.08 806 ÿ0.72 0.15 395
Child age in months 10.30 0.29 806 11.16 0.44 395
Expenditure/day/ capita (meticais) 4080.47 200.65 806 7323.31 825.87 395
Land per capita (hectares)b 0.39 0.02 764 0.25 0.03 188
Percentage of household <5 years old 0.30 0.01 806 0.28 0.007 395
Percentage of household P 5 and 6 17 0.30 0.01 806 0.34 0.01 395
Percentage of household P 60 0.01 0.00 806 0.01 0.002 395
Household size (no. of members) 6.65 0.18 806 6.96 0.24 395
Rooms in dwelling per capita 0.49 0.02 802 0.51 0.03 395
Dummy variables
Female child 0.55 0.02 806 0.56 0.03 395
Highest level of education by adult male
in householdb
None/illiterate 0.61 0.04 435 0.27 0.06 81
Some but less than primary 0.27 0.03 198 0.32 0.03 114
Completed primary or more 0.12 0.02 95 0.41 0.07 159
Mother's level of educationb
None/illiterate 0.88 0.03 607 0.55 0.07 192
Literate or any schooling 0.12 0.03 81 0.45 0.07 177
War migrant 0.09 0.02 806 0.05 0.02 395
Female household head 0.10 0.01 806 0.11 0.02 395
Seasonsc
Early rains (Dec±Feb) 0.23 0.05 806 0.30 0.08 395
Rains (Mar±May) 0.32 0.05 806 0.37 0.09 395
Harvest (Jun±Aug) 0.18 0.04 806 0.28 0.12 395
Post-harvest (Sep±Nov) 0.27 0.05 806 0.05 0.02 395
Regions
Maputo ÿ ÿ ÿ 0.05 0.01 1201
Secondary citiesd ÿ ÿ ÿ 0.05 0.02 1201
Other urban areas ÿ ÿ ÿ 0.13 0.04 1201
Rural provinces, north 0.23 0.04 1201 ÿ ÿ ÿ
Rural provinces, central 0.39 0.05 1201 ÿ ÿ ÿ
Rural provinces, south 0.15 0.03 1201 ÿ ÿ ÿ
Mother speaks Portugueseb 0.32 0.03 707 0.78 0.05 380
Mother had prenatal careb 0.62 0.04 699 0.92 0.05 377
Have latrineb 0.36 0.04 779 0.68 0.06 347
Use well waterb 0.49 0.03 777 0.44 0.11 341
Use piped water or public tapb 0.14 0.03 777 0.52 0.11 341
Use river waterb 0.36 0.03 777 0.04 0.03 341
a
Weighted to re¯ect number of individuals in the sample. Standard errors adjusted for survey design.
b
Conditional on having an observation.
c
For central and northern provinces. Periods begin and end one month earlier for southern provinces.
d
Beira, Matola, and Nampula.

expenditure/day/capita is more than twice as as with the comparison of levels of food

high in urban as in rural areas. Re¯ecting security, the number must be treated with
higher food prices in the cities, the average some caution.
value of food consumption /day/capita is also From the poverty assessment report, we
higher in urban than in rural areas, 4,655 know that city-dwellers purchase 83% of their
Mts. versus 2,840 Mts. (MPF/UEM/IFPRI, food, while those who live in rural areas
1998). Despite substantially higher expendi- purchase only 30%, either growing or gathering
tures, the mean calories/day/AEU is slightly the rest or receiving it as transfers from social
lower in urban than in rural areas, although, assistance programs or other households.
 FOOD SECURITY AND NUTRITIONAL STATUS 1961

Table 3. Means and standard errors, nutritional status model, 24±60 monthsa
Variables Rural Urban
Mean Standard n Mean Standard n
error error
Continuous variables
Height-for-age Z score ÿ1.91 0.09 957 ÿ1.19 0.10 557
Child age in months 41.25 0.53 957 41.35 0.53 557
Expenditure/day/capita (meticais) 4061.35 206.64 957 7606.66 677.27 557
b
Land per capita (has.) 0.40 0.02 899 0.31 0.05 253
Percentage of household <5 years old 0.29 0.01 957 0.27 0.01 555
Percentage of household P 5 and 6 17 0.34 0.01 957 0.37 0.01 557
Percentage of household P 60 0.01 0.00 957 0.01 0.00 557
Household size (no. of members) 7.07 0.14 957 7.63 0.20 557
Rooms in dwelling per capita 0.46 0.01 952 0.47 0.01 557
Dummy variables
Female child 0.51 0.02 957 0.44 0.04 557
Highest level of education by adult male in
b
household
None/illiterate 0.58 0.04 480 0.19 0.03 92
Some but less than primary 0.28 0.03 234 0.34 0.04 164
Completed primary or more 0.14 0.02 124 0.47 0.04 240
b
Mother's level of education
None/illiterate 0.89 0.02 720 0.56 0.04 280
Literate or any schooling 0.11 0.02 101 0.44 0.04 243
War migrant 0.10 0.03 957 0.04 0.01 557
Female household head 0.10 0.01 957 0.15 0.02 557
Seasonsc
Early rains (Dec±Feb) 0.27 0.06 957 0.36 0.08 557
Rains (Mar±May) 0.31 0.05 957 0.38 0.08 557
Harvest (Jun±Aug) 0.16 0.03 957 0.20 0.07 557
Post-harvest (Sep±Nov) 0.26 0.04 957 0.06 0.03 557
Regions
Maputo ÿ ÿ ÿ 0.05 0.01 1514
Secondary citiesd ÿ ÿ ÿ 0.05 0.02 1514
Other urban areas ÿ ÿ ÿ 0.13 0.04 1514
Rural provinces, north 0.23 0.04 1514 ÿ ÿ ÿ
Rural provinces, central 0.39 0.05 1514 ÿ ÿ ÿ
Rural provinces, south 0.15 0.03 1514 ÿ ÿ ÿ
Mother speaks Portugueseb 0.30 0.03 826 0.84 0.03 528
Mother had prenatal careb 0.62 0.04 801 0.95 0.02 517
Have latrineb 0.37 0.04 922 0.71 0.05 490
b
Use well water 0.48 0.03 908 0.36 0.06 491
Use piped water or public tapb 0.13 0.03 908 0.62 0.07 491
Use river waterb 0.39 0.03 908 0.02 0.01 491
a
Weighted to re¯ect number of individuals in the sample. Standard errors adjusted for survey design.
b
Conditional on having an observation.
c
For central and northern provinces. Periods begin and end one month earlier for souther provinces.
d
Beira, Matola, and Nampula.

Educational levels are much higher in urban almost all rural households in the sample have
areas. No male adult is literate or has any land, only 42% of urban households do.
education in 71% of rural households, while at Demographic structure and household size
least one adult male in 73% of urban house- are about the same in rural and urban areas.
holds is literate or has some education. In an On average, about half the household members
astounding 91% of rural households, no female are younger than 18 years old, and the average
adult is literate or has any education, while this household has about ®ve members. Women
is true of 54% of urban households. Although head about 20% of households in both areas.
1962 WORLD DEVELOPMENT

In summary, expenditure levels are higher in (b) Multivariate analyses

urban areas but calorie availability is not,
which, in addition to greater expenditure on Table 4 summarizes the results of our tests
nonfood needs, may also re¯ect higher prices exploring the potential endogeneity of the
and lower energy requirements due to lower household expenditure variable in the calorie
physical activity. Education, which is assumed availability and the nutritional status models.
to a€ect income levels, is higher for both men The Durbin-Wu-Hausman test determined
and women in urban areas, though women still whether OLS and 2SLS give signi®cantly
lag far behind. di€erent coecients on household expenditure.
Tables 2 and 3 describe the data used in the If the two procedures provide signi®cantly
nutritional status models. The mean height-for- di€erent results, and we have a reasonably good
age Z-score is much lower in rural than in prediction of household expenditure in the ®rst-
urban areas. For children 0±23 months old, the stage regression, then we can be reasonably
mean Z-score is ÿ1.35 in rural areas vs. ÿ0.72 con®dent that we should use the 2SLS esti-
in urban areas. Corresponding ®gures for the mates. For the calorie availability model and
24±60 month old age group are ÿ1.91 and the nutritional status model for 24±60 month
ÿ1.19. Although not reported in the tables, old children, the Durbin-Wu-Hausman statistic
prevalence of stunting among children 0±23 rejects the hypothesis that the OLS and 2SLS
months old is 39% in rural areas and 23% in estimates are identical. This hypothesis cannot
urban areas, and 51% and 28% in rural and be rejected for the nutritional status model for
urban areas, respectively, for children 24±60 0±23 month old children.
months old. Following the suggestion of Bound, Jaeger,
Educational levels of adults in households and Baker (1995), the F-statistic on the identi-
with preschoolers are much lower in rural than fying instruments in the ®rst-stage estimation of
in urban areas, mirroring the levels in the each model is reported as an indicator of its
general population, as shown by comparison quality. Table 4 shows that the instrumental
with Table 1. Whereas 80% of mothers with variables perform well, and are highly signi®-
preschoolers in urban areas speak Portuguese, cant in the ®rst-stage regressions of all three
only about 30% of rural mothers do. Well over models. The results in Table 4 suggest that the
90% of urban mothers, but only 62% of rural 2SLS estimates are preferred for the calorie
mothers, receive some prenatal care. availability model and the nutritional status
Urban households have greater access to model for 24±60 month olds. OLS is preferred
improved sanitation and piped water or public for the nutritional status model for 0±23 month
taps. About 70% of urban households with olds.
preschoolers have latrines and 50±60% have For reported results for all models, when
piped water or get water from public taps. Only rural and urban coecients are the same, the
35% of rural households have latrines. Over interaction term was insigni®cant and was
one-third of rural households get water from dropped before the ®nal model was estimated.
rivers and lakes, compared to fewer than 5% of Therefore, there is just one coecient (the same
urban-dwellers. Almost half of rural house- in both areas) to report. Where rural and urban
holds get water from public or private wells. coecients di€er, the interaction term was
In-house crowding is similar in rural and signi®cant and parameters were calculated
urban areas, with about two household separately. The coecient in the table repre-
members per room. While about 10% of these sents the main e€ect of the variable in either the
households with preschoolers are headed by rural or the urban area, as noted.
women in rural and urban areas, half that of
the general population as shown in Table 1, (i) Calorie availability model
means of other variables are about the same. Table 5 presents the OLS and 2SLS estimates
Neither are any substantial di€erences apparent for the calorie availability model. We focus on
between Tables 2 and 3. the 2SLS estimates for reasons given above.
In summary, preschoolers in urban areas Only household expenditure, household size,
seem to have higher levels of the inputs household composition, seasonality, and loca-
required for good child nutrition: higher levels tion have any signi®cant e€ect on calorie
of expenditure and maternal and adult male availability. Of these, only household expendi-
education as well as better access to sanitation ture and household size have a di€erent e€ect
and safe water. on calorie availability in urban and rural areas.
 FOOD SECURITY AND NUTRITIONAL STATUS 1963

Table 4. Instrumental variable estimation test statisticsa

Potentially endogenous variable F-statistic on instruments: Durbin Wu-Hausman F-statistic:
®rst-stage regressions second-stage regressions
Total expenditure per capita (log)
Calorie Availability Model 46.32 (Pr > 0.00)
Assets 1029.64 (2, 6169)
Assets ´ (urban-rural dummy) 2065.41 (2, 6169)
Nutritional Status Model, 0±23 0.40 (Pr > 0.53)
months
Assets 277.51 (1, 912)
Nutritional Status Model, 24±60 7.75 (Pr > 0.01)
months
Assets 383.57 (1, 1220)
a
Total no. of di€erent household assets and vehicles is used as an instrument. Since the expenditure variable appears
alone and in interaction with the urban-rural dummy in the calorie availability model, that model has two instru-
ments.

The expenditure elasticity for calorie avail- The positive e€ect on calorie availability of
ability is slightly higher in urban than in rural having higher percentages of children or of
areas (0.14 compared to 0.12) 8, which suggests elderly people in the household probably
that a city-dweller is more likely to spend an re¯ects the construction of the dependent
increase in income on food than a rural- variable: When household size is the same,
dweller. Although low, the estimates of the households with larger percentages of children
expenditure-calorie elasticity are consistent or the elderly will, because these individuals
with other 2SLS ®ndings (Behrman and have lower calorie requirements, have a lower
Deolalikar, 1987; Bouis and Haddad, 1992). On number of ``Adult Equivalents'' than those
the other hand, this contrasts with analysts who with young adults. The same per capita sums,
have found that expenditure-calorie elasticities then, could be used to provide more food for a
are typically higher for poor families and those smaller number of ``Adult Equivalents,'' creat-
living in rural areas (Alderman and Higgins, ing a positive relationship.
1992; Alderman, 1986; Sahn, 1988). 9 Surprisingly, seasonality does not have a
Greater household size has a large negative di€erential e€ect between urban and rural
impact on calorie availability, with the e€ect areas. Rather, calorie availability declines
initially greater in rural than in urban areas. substantially in both urban and rural areas
The quadratic term is signi®cant and positive, during the period of early rains and the harvest
indicating that the negative impact is increasing season, as compared to the rainy season (the
at a declining rate. This relationship re¯ects the omitted dummy variable). Calorie availability
ability of larger households to begin to mitigate in the postharvest period does not di€er
the negative e€ects of an additional household signi®cantly from the rainy season. This result
member through exploiting economies of scale is dicult to interpret. Although calorie avail-
in consumption. The negative e€ect is larger in ability might understandably decrease during
rural areas than in urban areas at small the early rains (a postharvest period of intense
household sizes, although larger household size agricultural activity), it is unclear why calorie
begins to have a positive e€ect at 10 members in availability would increase in the rainy season
rural areas and 13 in urban areas. The reason only to decline again during the harvest period.
for this di€erential relationship is an important Correlation of these seasons with a determinant
topic for future research. Perhaps additional of calorie availability could help explain this
members in rural areas have limited opportu- result, but such a correlation is not readily
nities to improve household income and food apparent. Although only individually the
availability but stronger social networks in parameter on ``other urban areas'' (smaller
rural areas eventually begin to compensate for urban areas outside Maputo and the secondary
this e€ect. 10 In any case, fewer than 10% of cities) is signi®cant, the parameters on the
households are composed of more than 10 regional variables as a whole are highly signi-
individuals, so the negative e€ect of household ®cant (Pr > 0.0001) and, in many cases, are
size on calorie availability holds for almost all quite large. Residence in smaller urban centers
households in the sample. has a substantial negative e€ect on calories/
1964 WORLD DEVELOPMENT

Table 5. Calorie availability modela

Dependent variable: calories per day per adult Rural Urban
equivalent unit
Independent variables OLS 2SLS OLS 2SLS
Expenditure per capita (log)b; c 1077.5 341.6 751.6 396.1
(40.7) (102.8) (44.4) (72.9)
Adult male, literate or some school ÿ112.7 ÿ34.3 ÿ112.7 ÿ34.3
(44.0) (46.2) (44.0) (46.2)
Adult male, completed primary or more ÿ191.2 ÿ47.8 ÿ191.2 ÿ47.8
(53.4) (57.2) (53.4) (57.2)
Adult female, literate or some school ÿ53.4 21.6 ÿ53.4 21.6
(52.7) (54.9) (52.7) (54.9)
Adult female, completed primary or more ÿ105.2 67.9 ÿ105.2 67.9
(68.3) (73.5) (68.3) (73.5)
War migrant ÿ16.2 ÿ74.3 ÿ16.2 ÿ74.3
(74.1) (76.7) (74.1) (76.7)
Land (has.) per capita (log) 3.8 3.7 3.8 3.7
(3.4) (3.5) (3.4) (3.5)
Percentage of household <5 years old 921.0 650.3 921.0 650.3
(133.0) (140.2) (133.0) (140.2)
Percentage of household P 60 306.6 201.9 306.6 201.9
(96.7) (100.3) (96.7) (100.3)
Percentage of household P 5 and 6 17 231.0 145.9 231.0 145.9
(107.8) (111.4) (107.8) (111.4)
Household sizec ÿ241.7 ÿ412.9 ÿ233.3 ÿ299.5
(32.4) (40.2) (34.2) (36.7)
Household size (squared)c 12.1 20.1 8.7 10.9
(2.1) (2.4) (2.0) (2.1)
Female, household head 56.9 ÿ19.6 56.9 ÿ19.6
(54.1) (56.3) (54.1) (56.3)
Early rains ÿ408.1 ÿ477.8 ÿ408.1 ÿ477.8
(233.7) (240.8) (233.7) (240.8)
Harvest ÿ472.6 ÿ525.6 ÿ472.6 ÿ525.6
(217.2) (223.8) (217.2) (223.8)
Post-harvest ÿ339.2 ÿ398.0 ÿ339.2 ÿ398.0
(267.6) (275.7) (267.6) (275.7)
Secondary cities ÿ64.5 ÿ7.5 ÿ64.5 ÿ7.5
(448.0) (461.5) (448.0) (461.5)
Other urban areas ÿ778.7 ÿ1067.8 ÿ778.7 ÿ1067.8
(456.3) (471.6) (456.3) (471.6)
Rural provinces, north ÿ1443.5 1491.9 ÿ1443.5 1491.9
(784.1) (1217.6) (784.1) (1217.6)
Rural provinces, central ÿ3372.0 ÿ297.1 ÿ3372.0 ÿ297.1
(737.0) (1196.4) (737.0) (1196.4)
Rural provinces, south ÿ4346.1 ÿ648.0 ÿ4346.1 ÿ 648.0
(756.2) (1261.8) (756.2) (1261.8)
Missing observations
Adult male education 31.4 56.7 31.4 56.7
(67.2) (69.3) (67.2) (69.3)
Adult female education ÿ275.6 ÿ233.5 ÿ275.6 ÿ233.5
(76.8) (79.2) (76.8) (79.2)
Land per capita ÿ430.1 ÿ426.1 ÿ430.1 ÿ426.1
(157.1) (161.8) (157.1) (161.8)
Community dummy variablesd
R2 0.38 0.34 0.38 0.34
F 14.75 10.95 14.75 10.95
N 6462 6461 6462 6461
a
Standard errors are reported in parentheses. 2SLS estimates are preferred.
b
Endogenous variable, predicted by assets.
c
Coecient di€ers between rural and urban models.
d
Not reported. Prob > F, 0.00 (OLS), 0.00 (2SLS).
*
Signi®cantly di€erent from zero at 10% or higher level.
 FOOD SECURITY AND NUTRITIONAL STATUS 1965

AEU/day and residence in rural areas of the child stunting is lower than in other areas.
northern provinces has a substantial positive Aside from this seasonal, potentially regional,
e€ect. The high signi®cance level of the dummy e€ect, then, we ®nd no di€erences between
variables included for each community, when urban and rural areas in the relative e€ects of
considered as a group, con®rm that communi- the determinants of nutritional status of chil-
ty-level factors, such as prices, also exert strong dren 0±23 months old.
in¯uences on household-level calorie availabil- Statistically signi®cant determinants of child
ity. height-for-age Z-scores in both urban and rural
Just as interesting as what is signi®cant, is areas in this sample include child characteristics
what is not. Level of education (as represented (sex and age), maternal characteristics (educa-
by the highest level of education attained by an tion level) and household characteristics
adult member of the household) did not a€ect (expenditure per capita and percentage of
household calorie availabilityÐat least beyond household less than ®ve years of age).
the e€ect of income, which is already included Girls' nutritional status is better than boys'
in the model. In fact, to test whether the income by 0.36 Z-scores. This re¯ects the documented,
e€ect was working through education (and yet not well understood, greater vulnerability of
therefore reducing the signi®cance of the edu- boys at this age (Svedberg, 1990). Nutritional
cation variable), we dropped the expenditure status deteriorates rapidly with age, at the rate
variable and reestimated the model. Education of 0.13 Z-scores per month, re¯ecting deterio-
remained insigni®cant. ration in the rapid growth that typically occurs
Finally, we might expect to ®nd some asso- in young children.
ciation between calorie availability and vari- Mother's education has a signi®cant e€ect on
ables such as land availability per capita, the children's nutritional status at this age. Even
gender of the head of the household, and just being literate or having some education
whether the household had members who had improves the child's nutritional status by more
migrated during the war. The results show that than one-third of a Z-score. 11 Because the
these characteristics do not a€ect calorie model controls for income (through the
availability independently, once we control for expenditure variable), the e€ect of maternal
other factors. education documented here is independent of
In general the results from this estimation income and probably re¯ects better maternal
conform to expectations: income and prices (as caring practices such as child feeding, use of
represented by the community dummy vari- health services and hygiene. The highest level of
ables) matter to household-level calorie avail- education attained by an adult male in the
ability. Demographic structure and regional household is not signi®cant.
location of the household are also important, In contrast to ®ndings by Sahn and Alder-
and probably a€ect availability through their man in Mozambique (1997), we ®nd that
in¯uence on food consumption patterns. Only expenditure matters even for this young age
household size and household expenditure group, with a 10% rise in expenditure contrib-
levels have a small di€erential e€ect on calorie uting to a 2.4% improvement in Z-score. Still,
availability in urban and rural areas. for the e€ect of income to equal that of
maternal literacy or only some maternal
(ii) Nutritional status, 0±23 months of age schooling, income would have to more than
Table 6 presents the preferred OLS estimates double.
of the factors a€ecting nutritional status of To investigate whether the e€ect of maternal
children 0±23 months old. There are no signi- education varied by income level, we added an
®cant di€erences in the determinants of nutri- interaction term of expenditure and maternal
tional status of children 0±23 months of age in education to the model. We do not present the
urban and rural areas with the exception of results here, but a positive sign on the interac-
households surveyed in the postharvest period, tion term, which was signi®cant, indicated that
where there is a negative impact on child education enhances the positive e€ect of
stunting relative to the rainy season. This increases in expenditure. Other researchers
di€erence is most likely due to the fact that no have documented similar results showing that
households in urban areas outside Maputo City maternal schooling is associated with improved
were surveyed during the postharvest period. child nutrition only among households that
This variable is therefore acting, in part, as a have access to a minimum level of resources
dummy variable representing Maputo, where without being among the wealthiest group
1966 WORLD DEVELOPMENT

Table 6. Nutritional status model, 0±23 monthsa

Dependent variable: Height-for-age Z-score Rural Urban
Independent variables OLS 2SLS OLS 2SLS
Female child 0.36 0.36 0.36 0.36
(0.10) (0.10) (0.10) (0.10)
Age in months ÿ0.13 ÿ0.13 ÿ0.13 ÿ0.13
(0.03) (0.03) (0.03) (0.03)
Age squared 0.002 0.002 0.002 0.002
(0.001) (0.001) (0.001) (0.001)
Expenditure per capita (log)b 0.26 0.38 0.26 0.38
(0.10) (0.21) (0.10) (0.21)
Adult male education, literate or some school 0.01 ÿ0.001 0.01 ÿ0.001
(0.13) (0.13) (0.13) (0.13)
Adult male education, completed primary ÿ0.003 ÿ0.03 ÿ0.003 ÿ0.03
or more (0.17) (0.17) (0.17) (0.17)
Mother's education, literate or any schooling 0.37 0.35 0.37 0.35
(0.16) (0.16) (0.16) (0.16)
War migrant ÿ0.006 0.01 ÿ0.006 0.01
(0.22) (0.22) (0.22) (0.22)
Land (has.) per capita (log) 0.01 0.01 0.01 0.01
(0.01) (0.01) (0.01) (0.01)
Percentage of household <5 years old ÿ1.35 ÿ1.31 ÿ1.35 ÿ1.31
(0.67) (0.68) (0.67) (0.68)
Percentage of household P 60 ÿ1.29 ÿ1.25 ÿ1.29 ÿ1.25
(1.38) (1.38) (1.38) (1.38)
Percentage of household P 5 and 6 17 ÿ0.05 ÿ0.002 ÿ0.05 ÿ0.002
(0.52) (0.52) (0.52) (0.52)
Household size ÿ0.02 0.001 ÿ0.02 0.001
(0.10) (0.11) (0.10) (0.11)
Household size (squared) 0.008 0.007 0.008 0.007
(0.006) (0.006) (0.006) (0.006)
Female, household head ÿ0.03 ÿ0.02 ÿ0.03 ÿ0.02
(0.23) (0.23) (0.23) (0.23)
Early rains ÿ0.53 ÿ0.51 ÿ0.53 ÿ0.51
(0.61) (0.61) (0.61) (0.61)
Harvest 0.21 0.18 0.21 0.18
(0.67) (0.68) (0.67) (0.68)
Post-harvestc ÿ3.71 ÿ3.70 ÿ1.04 ÿ1.03
(1.07) (1.07) (1.05) (1.05)
Mother speaks Portuguese ÿ0.02 ÿ0.03 ÿ0.02 ÿ0.03
(0.14) (0.14) (0.14) (0.14)
Rooms per capita 0.23 0.21 0.23 0.21
(0.21) (0.21) (0.21) (0.21)
Prenatal care 0.17 0.18 0.17 0.18
(0.16) (0.16) (0.16) (0.16)
Latrine ÿ0.14 ÿ0.13 ÿ0.14 ÿ0.13
(0.13) (0.13) (0.13) (0.13)
Well water 0.02 0.01 0.02 0.01
(0.18) (0.18) (0.18) (0.18)
River/lake water 0.25 0.25 0.25 0.25
(0.21) (0.21) (0.21) (0.21)
Missing observations
Adult male education 0.10 0.10 0.10 0.10
(0.26) (0.26) (0.26) (0.26)
c
Mother's education 0.59 0.55 ÿ0.55 ÿ0.57
(0.40) (0.40) (0.48) (0.48)
Land (has.) per capita (log) 0.13 0.12 0.13 0.12
(0.42) (0.42) (0.42) (0.42)
Mother speaks Portuguesec ÿ1.23 ÿ1.22 0.28 0.26
(0.55) (0.55) (0.72) (0.72)
Prenatal care ÿ0.10 ÿ0.09 ÿ0.10 ÿ0.09
(0.40) (0.40) (0.40) (0.40)
Continued opposite
 FOOD SECURITY AND NUTRITIONAL STATUS 1967

Table 6Ðcontinued
Dependent variable: Height-for-age Z-score Rural Urban
Independent variables OLS 2SLS OLS 2SLS
Latrine ÿ0.45 ÿ0.45 0.41 0.37
(0.37) (0.37) (0.36) (0.36)
Source of water 0.42 0.42 0.42 0.42
(0.25) (0.25) (0.25) (0.25)
Community dummy variablesd
R2 0.23 0.23 0.23 0.23
F 2.27 2.26 2.27 2.26
N 1197 1197 1197 1197
a
Standard errors are reported in parentheses. OLS estimates are preferred.
b
Endogenous variables, predicted by assets.
c
Coecient di€ers between rural and urban models.
d
Not reported. Prob > F, 0.01 (OLS), 0.01 (2SLS).
*
Signi®cantly di€erent from zero at 10% or higher level.

(Bairagi, 1980; Reed, Habicht and Niamego, dren (those 0±23 months old), the primary
1996). determinants of nutritional status are child
Although household size does not a€ect biological characteristics (sex, age), and
nutritional status, larger percentages of chil- maternal schooling which probably acts
dren less than ®ve years old in the household through good child care feeding, health and
negatively a€ect height-for-age, with a 10% hygiene practices (Ricci and Becker, 1996; Ruel
increase in this percentage causing a 3.7% et al., 1999). Household income is also a
deterioration in height-for-age Z-score. This determinant of child nutritional status in our
could re¯ect the increased demands on mater- sample, a ®nding that contrasts with other
nal time that a larger number of small children studies in urban Mozambique (Sahn and
exert; the shorter periods between births, which Alderman, 1997) and in Accra (Ruel et al.,
can result in lower birth weights and poorer 1999).
postnatal growth; or, in light of the previous The R2 is similar to those found in other
®nding that households with small children estimations of nutritional status (Alderman and
have higher levels of calorie availability, Garcia, 1994; Sahn, 1994), but its relatively low
resource allocation patterns within the house- value indicates that the explanatory power of
holds that do not favor younger children. With the model remains to be improved, perhaps by
more than half of these households reporting including more speci®c descriptions of the more
more than one child under ®ve, further research proximal determinants of nutritional status.
into the causes of the decline would be well
justi®ed. (iii) Nutritional status, 24±60 months of age
Access to land and gender-based character- Table 7 shows that the determinants of
istics, such as gender of the head of the nutritional status in urban and rural areas only
household or the mother's ability to speak begin to di€erentiate once children are older.
Portuguese, do not a€ect nutritional status. Among the older age group, the e€ects of
Variables that re¯ect physical environment, land holdings per capita, source of water, and
including source of water, are not signi®cant, seasonality di€er between urban and rural
nor is whether the mother received prenatal areas. In urban areas, nutritional status
care. If regional e€ects exist, they appear to be declines if the household has land. The reason
captured by other variables in the model. Due for this negative e€ect may be that, in urban
to the low level of signi®cance of regional areas, landholding is associated with poor
variables (Pr > 0.62), they were dropped from environmental conditions and poor access to
the ®nal model. On the other hand, the dummy health care. 12 For example, urban households
variables for the communities were jointly with land tend to live in cities outside Maputo
signi®cant, indicating the importance of local- where public services are less available and
level variation in determining nutritional status. nutritional status is lower (MPF/UEM/IFPRI,
In summary, our results concur with those of 1998). Landholdings by urban households may,
other researchers who ®nd that in young chil- then, in some sense be capturing regional
1968 WORLD DEVELOPMENT

Table 7. Nutritional status model, 24±60 monthsa

Dependent variable: Height-for-age Z-score Rural Urban
Independent variables OLS 2SLS OLS 2SLS
Female child ÿ0.03 ÿ0.03 ÿ0.03 ÿ0.03
(0.08) (0.08) (0.08) (0.08)
Age in months ÿ0.02 ÿ0.03 ÿ0.02 ÿ0.03
(0.03) (0.03) (0.03) (0.03)
Age squared 0.0002 0.0003 0.0002 0.0003
(0.0004) (0.0004) (0.0004) (0.0004)
Expenditure per capita (log)b 0.27 0.70 0.27 0.70
(0.09) (0.18) (0.09) (0.18)
Adult male education, literate or some school 0.06 0.02 0.06 0.02
(0.12) (0.12) (0.12) (0.12)
Adult male education, completed primary or more 0.16 0.08 0.16 0.08
(0.14) (0.14) (0.14) (0.14)
Mother's education, literate or any schooling 0.17 0.10 0.17 0.10
(0.13) (0.13) (0.13) (0.13)
War migrant 0.03 0.05 0.03 0.05
(0.19) (0.19) (0.19) (0.19)
Land (has.) per capita (log)c 0.03 0.03 ÿ0.02 ÿ0.02
(0.02) (0.02) (0.01) (0.01)
Percentage of household <5 years old ÿ0.68 ÿ0.62 ÿ0.68 ÿ0.62
(0.58) (0.59) (0.58) (0.59)
Percentage of household P 60 ÿ0.99 ÿ0.85 ÿ0.99 ÿ0.85
(0.99) (1.0) (0.99) (1.0)
Percentage of household P 5 and 6 17 ÿ0.12 ÿ0.04 ÿ0.12 ÿ0.04
(0.46) (0.46) (0.46) (0.46)
Household size 0.13 0.20 0.13 0.20
(0.09) (0.10) (0.09) (0.10)
Household size (squared) ÿ0.006 ÿ0.009 ÿ0.006 ÿ0.009
(0.005) (0.005) (0.005) (0.005)
Female, household head ÿ0.19 ÿ0.13 ÿ0.19 ÿ0.13
(0.19) (0.19) (0.19) (0.19)
Early rainsc ÿ2.63 ÿ2.46 ÿ0.08 0.03
(0.82) (0.84) (0.53) (0.54)
Harvest ÿ0.03 ÿ0.10 ÿ0.03 ÿ0.10
(0.47) (0.48) (0.47) (0.48)
Post-harvest ÿ0.38 ÿ0.24 ÿ0.38 ÿ0.24
(0.80) (0.80) (0.80) (0.80)
Mother speaks Portuguese 0.14 0.12 0.14 0.12
(0.12) (0.12) (0.12) (0.12)
Rooms per capita 0.53 0.35 0.53 0.35
(0.20) (0.21) (0.20) (0.21)
Prenatal care 0.02 0.02 0.02 0.02
(0.14) (0.14) (0.14) (0.14)
Latrine ÿ0.10 ÿ0.07 ÿ0.10 ÿ0.07
(0.12) (0.12) (0.12) (0.12)
Well waterc ÿ0.09 ÿ0.07 ÿ0.72 ÿ0.73
(0.20) (0.20) (0.23) (0.24)
River/lake water ÿ0.22 ÿ0.18 ÿ0.22 ÿ0.18
(0.21) (0.21) (0.21) (0.21)
Missing observations
Adult male education 0.24 0.23 0.24 0.23
(0.21) (0.21) (0.21) (0.21)
Mother's education ÿ0.64 ÿ0.70 ÿ0.64 ÿ0.70
(0.40) (0.40) (0.40) (0.40)
Land (has.) per capita (log)c 0.35 0.38 ÿ0.35 ÿ0.42
(0.56) (0.57) (0.59) (0.60)
Mother speaks Portuguese 0.65 0.63 0.65 0.63
(0.43) (0.44) (0.43) (0.44)
Prenatal care ÿ0.21 ÿ0.21 ÿ0.21 ÿ0.21
(0.25) (0.26) (0.25) (0.26)
Continued opposite
 FOOD SECURITY AND NUTRITIONAL STATUS 1969

Table 7Ðcontinued
Dependent variable: Height-for-age Z-score Rural Urban
Independent variables OLS 2SLS OLS 2SLS
Latrine 0.14 0.03 0.14 0.03
(0.21) (0.22) (0.21) (0.22)
Source of waterc ÿ0.22 ÿ0.22 0.06 0.13
(0.34) (0.34) (0.24) (0.25)
Community dummy variablesd
R2 0.14 0.12 0.14 0.12
F 1.86 1.85 1.86 1.85
N 1509 1509 1509 1509
a
Standard errors are reported in parentheses. 2SLS estimates are preferred.
b
Endogenous variables, predicted by assets.
c
Coecient di€ers between rural and urban models.
d
Not reported. Prob > F, 0.00 (OLS), 0.00 (2SLS).
*
Signi®cantly di€erent from zero at 10% or higher level.

e€ects. Households with land may also be longer signi®cant. Surprisingly, maternal edu-
raising animals in close proximity to the house, cation did not have any e€ect on children's
which in a crowded area may lead to a poorer height-for-age Z-scores at this age. Still, this
quality environment inside and outside the concurs with ®ndings by Sahn and Alderman
house. Little work has been done on the (1997) for the same age group in Maputo. As
impacts of urban agriculture and urban land with the model for 0±23 month old children, we
holdings on child nutritional status (Maxwell, tested an interaction term of expenditure and
Levin, Csete, 1998); additional research is maternal education to see if the e€ect of edu-
needed to shed light on the nature of the cation varied with income level. The interaction
association found here. term was not signi®cant.
The importance of the physical environment The impact of expenditure on nutritional
for this age group is also highlighted by the status is stronger among the group of older
negative e€ect of using well water in urban children as compared to the 0±24 months old
areas, as compared to piped water or public group. Although still provoking less than a
taps. Using well water in urban areas is asso- one-to-one response, the expenditure elasticity
ciated with decline of 0.73 in height-for-age Z- for nutritional status of 0.43 is almost twice
score. This probably re¯ects contamination or that of the younger age group and is fairly high.
lack of access to sucient water. 13 Surpris- As distinct from its negative e€ect on house-
ingly, the source of water does not a€ect the hold calorie availability, larger household size
nutritional status of children in rural areas. has a positive e€ect on older children's nutri-
The period of early rains has a signi®cant tional status. This may re¯ect some economies
negative e€ect on children in rural areas, rela- of scale in providing for the needs, in addition
tive to the other seasons of the year. The to food, of older children. The positive in¯u-
reasons for this decline are uncertain and ence of less in-house crowding further argues
should be explored further, but it could re¯ect a for the importance of the physical environment
slowing of growth in a season where morbidity for this age group. Each additional room per
may be increasing along with moisture and person in the household improves height-for-
temperatures, where calorie availability is fall- age Z-scores by 0.35.
ing, and demand for agricultural labor is As in the model for children 0±23 months
increasing, perhaps drawing attention away old, once other factors are controlled for, the
from child care. potential determinants of gender of head of
Household expenditure, household size and household, use of prenatal care, existence of a
the number of rooms per capita were the only latrine, or maternal ability to speak Portuguese
other factors that a€ected the nutritional status do not exert independent in¯uences on nutri-
of children 24±60 months old. The e€ects of tional status. The signi®cance of the commu-
these determinants did not di€er between urban nity dummy variables support the analysis that
and rural areas. At this age, the more biological community-level factors are also important
in¯uences of age and sex of the child are no to nutritional status. As with the model for
1970 WORLD DEVELOPMENT

children 0±23 months of age, better measure- than grow, their own food. Household size, for
ment of more proximal determinants of nutri- reasons to be explored, initially exerts a larger
tional status would probably go far to negative e€ect in rural than in urban areas.
improving the explanatory power of the model. Regional factors are also in¯uencing calorie
availability in ways beyond the e€ects speci®ed
in the model. Future research should focus on
6. DISCUSSION elucidating the causes for these regional di€er-
ences.
(a) Main ®ndings For children 0±23 months old, the same
factors explain nutritional status in urban and
Because we know that urban and rural live- rural areas with no di€erence in magnitude. For
lihoods and lifestyles di€er, we had hypothe- these children, biological factors and maternal
sized that the levels and determinants of food education have the greatest positive in¯uence
security and nutritional status in rural and on growth. To a large extent, mothers are
urban areas of Mozambique would be di€erent, responsible for feeding these younger children
too. Instead, we found that while levels of key and they largely control the interaction they
determinants of food insecurity and malnutri- have with their physical environment. Thus, it
tion, such as expenditure or education, may is logical that little urban-rural di€erences are
di€er between rural and urban areas, the nature found in this age group.
of the determinants and the magnitudes of their The analysis of nutritional status of the older
e€ects are very nearly the same. This study, of children points out that the determinants not
course, evaluates the e€ects of important, but only di€er between urban and rural areas but
mostly distal and mostly household-level, also di€er for older and younger children. In
determinants. Future research might ®nd comparison with the younger children, children
di€erences in urban and rural areas in the e€ect 24±60 months old are more mobile, they are
of more proximal determinants of food inse- weaned from the breast, they are starting to eat
curity and malnutrition, such as intrahousehold a variety of family foods, and they are increas-
resource allocation or maternal behaviors, or in ingly exposed to environmental contamination,
the mechanisms by which these in¯uential which results in high rates of infectious diseases
factors have their e€ects on food and nutrition and poor growth. Factors related to environ-
security. mental hygiene and food safety thus become
Nevertheless, our results largely reject the critical for their health and nutritional status.
hypothesis that the determinants of household Our results con®rm that environmental factors
calorie availability and nutritional status for do play a large role in determining nutritional
children 0±23 months old are di€erent between status of older children, but also point out that
urban and rural areas. For children 24±60 the nature of the threat is di€erent in urban and
months old, the determinants still have signi®- rural areas.
cant overlap but some di€erentiation between
urban and rural areas begins to appear, espe- (b) Explaining urban-rural di€erences
cially in factors having to do with the child's
surrounding physical environment. For nutritional status of both younger and
In urban and rural areas, much the same older children, most of the urban-rural di€er-
factors determine calorie availability. Only ence appears due to di€erences in the levels of
expenditure and household size exert signi®- critical determinants, such as in income or
cantly di€erent e€ects, and even then the mother's educational level. For example, stun-
magnitude of the di€erence is not large. The ting of children 0±23 months old is much more
most important factors are, as one would prevalent in rural than in urban areas (39% vs.
expect from a demand equation, income 23%, respectively), but there were no signi®cant
(proxied by expenditure), prices (as re¯ected in di€erences in the determinants of their nutri-
the community dummy variables), and demo- tional status. Within the context of the signi®-
graphics, such as household size. Urban- cant variables in our model, we explain this
dwellers do seem to be slightly more sensitive to di€erence in urban-rural levels of stunting by
changes in incomes than rural-dwellers. This noting that expenditure levels and maternal
may re¯ect urban residents' lack of a natural- education are much lower in rural areas.
resource ``cushion'' to absorb income or price Expenditure levels in rural areas are only about
shocks and also their need to purchase, rather half those in urban areas, and only 12% of rural
 FOOD SECURITY AND NUTRITIONAL STATUS 1971

mothers are literate or have any education at in both urban and rural areas is undoubtedly
all, while a much higher 45% of urban mothers important for achieving food and nutrition
do. security in Mozambique. The recent poverty
On the other hand, levels of food insecurity, assessment of the Ministry of Planning and
as measured by calories/AEU/day, are much Finance (MPF/UEM/IFPRI, 1998) emphasized
the same despite large di€erences in the levels of the importance of investing in education, in
some critical determinants of calorie availabil- agricultural productivity, and in rural infra-
ity between rural and urban areas. Failure to structure, as key elements of a poverty-reduc-
take potentially higher activity levels in rural tion strategy there. Continued support of social
areas into account could potentially underesti- assistance programs in Mozambique, such as
mate the level of food insecurity there; perhaps the urban cash transfer program, will also be
the levels of food insecurity, then, are not that necessary to help those who cannot participate
similar. But in the end, even if adjusted for in the labor market and receive the bene®ts of
energy requirements or household composition, overall increasing economic growth.
rural households actually do, on average, Women's education is also important to
receive more calories than urban ones. improving children's nutritional status. In the
The observed higher level of calorie avail- long term, improving girls' formal education
ability in rural areas is more likely due to and women's literacy and job skills will raise
regional, agroecological in¯uences not speci®- household incomes. In Mozambique, not only
cally captured in the model. These regional did maternal education have a positive e€ect on
di€erences can be large. For example, the young children's nutritional status above and
regression analysis shows that residing in the beyond the income e€ect, but it actually
rural northern provinces has a large positive enhanced the positive e€ect of income among
e€ect on calorie availability, beyond that of the the 0±24 months old group. In the long run,
other variables included in the model. Such a increases in both income and maternal educa-
result is reasonable, given the northern zone's tion could have large pay-o€s in terms of
higher levels of agricultural production, but the reducing childhood malnutrition in Mozam-
reasons for it are unclear. Perhaps cultural bique. Although our study does not elucidate
preferences of households in the region for low- the mechanisms by which maternal education
cost, energy-rich foods are at work, as well as a€ects child nutrition, others have shown that
other nonincome factors. Perhaps the data do education acts largely through greater know-
not adequately capture the value of consump- ledge and improved caregiving practices (Cebu
tion for rural households that depend to a large Study Team, 1991; Ruel et al., 1992; Ruel et al.,
extent on natural resources that they use with- 1999). Thus, well-targeted nutrition education
out ®nancial cost. In that case, levels of programs to improve speci®c caregiving prac-
consumption expenditure, a signi®cant deter- tices, such as child feeding, hygiene and use of
minant of calorie availability, would in fact be health services, could in the short term help
more similar than the data indicate. More mothers make better use of their scarce
research along the lines of Sharma et al. (1996), resources and protect their children's health
who look at the ecoregional dimensions of and nutrition.
malnutrition, and Huang and Bouis (1996), Given our results indicating the negative
who consider the impact of nonincome factors e€ect of larger percentages of preschoolers on
on consumption patterns, could shed light on nutritional status of 0±23 months old and of
the reasons for these regional, not strictly rural- larger household size on 24±60 month olds,
urban, di€erences. attention should also be directed at attenuat-
ing these conditions. Higher incomes and
(c) Policy recommendations higher levels of girls' and women's education
will over time probably lead to reductions in
Our analysis demonstrates that income is an fertility and lengthen time between births,
essential determinant of calorie availability and resulting in smaller household sizes and lower
nutritional outcomes in both rural and urban proportions of children under ®ve in the
areas. Despite some debate about the e€ec- household over time, although there may be
tiveness of increasing income in reducing food scope for direct actions that can, in the shorter
insecurity and malnutrition (Subramanian and term, also assist families in exercising their
Deaton, 1996; Bouis and Haddad, 1992; Behr- preferences in this area. In the meantime,
man and Deolalikar, 1987), income-generation social assistance programs should be sure to
1972 WORLD DEVELOPMENT

take into account the additional needs of between rural and urban areas. But policy
larger households. makers and program administrators cannot
For older children, the physical environment simply transfer their programs from rural to
emerges as a key di€erence between urban and urban areas, just as they cannot simply
rural determinants of nutritional status. In- transfer rural programs from one area to
house crowding exerts a negative in¯uence on another. Because the levels, if not the magni-
children's nutritional status in both urban and tude of the e€ects, of the determinants are
rural areas, but the urban environment seems di€erent in each location, policy makers and
to exert a greater negative impact. Programs in program administrators must understand
urban areas, then, should concentrate on speci®c community-level conditions so that
providing sanitation, garbage disposal, and they can identify which of the key variables
clean water to households, especially those with programs must address.
children under ®ve. Just providing the 37% of We recognize that creating programs and
households that use well water with adequate making policies that are ¯exible and adequately
amounts of safe water could improve nutri- re¯ect the needs, conditions, and resources in
tional status signi®cantly. Further investigation each community is quite a challenge. On a more
is warranted to con®rm the possible link positive note, this study does note some key
between land holdings and a poorer household areas for intervention to reduce food insecurity
environment, but targeting these programs to and malnutrition in both rural and urban areas,
urban households that have land may be a way and it suggests that policy makers and admin-
to identify and reach the most a€ected groups. istrators need not abandon the conceptual
In conclusion, our analysis indicates that, frameworks and toolkits they have developed
conceptually, the determinants of food security for rural areas but can bring them along as they
and nutritional status are not very di€erent move to work in the city.

NOTES

1. In both this paper and the Mozambique poverty urban and rural areas), this study chooses to focus only
assessment report (MPF/UEM/IFPRI, 1998), food secu- on the determinants of chronic malnutrition.
rity is de®ned as whether the household had enough In addition these estimates should be interpreted as a
calories available to meet caloric requirements of lower bound for malnutrition in Mozambique as
household members, using an adult equivalent unit approximately one-third of the anthropometric data
measure. The number of Adult Equivalent Units (AEU) collected for the survey were not usable. These unusable
in the household was determined by scaling the require- observations tended to be on children from rural areas
ments of each individual in the household to those of a and from poorer households (which tend to have higher
reference adult, based on age, sex, and an assumption of prevalence of malnutrition), perhaps leading to an
a moderate activity level. The 3,000 kcals/day require- underestimate of actual prevalence of malnutrition
ment of the reference adult was based on the estimated (MPF/UEM/IFPRI, 1998).
requirements of an adult male, 18 to 30 years old, with
moderate activity levels (FAO/WHO /UNU, 1985). 3. Although calorie availability is a widely accepted
Expressing food security in these units makes it indicator of food security, we recognize it measures only
dicult to tell whether the di€erence between levels of quantity of food. It does not incorporate other possible
food security in rural and urban areas is signi®cant. dimensions of food security, including nutritional ade-
Because the proportion of moderately active individuals quacy, safety, or cultural acceptability (Oshaug, 1994).
may be higher in rural than in urban areas, adjustment De®nitions of food security now re¯ect, according to
for higher activity levels in rural areas could reduce the Maxwell (1996), a ``cornucopia of ideas.'' Consequently,
magnitude of the observed di€erence in levels of food this paper will, for the most part, use the term that more
security between urban and rural areas. accurately re¯ects the focus of the studyÐcalorie avail-
abilityÐrather than the potentially more confusing term
2. Although the level of wasting, or acute malnutrition, ``food security.''
in Mozambique is indicative of poor country conditions
(WHO, 1995), it is low in Mozambique relative to 4. Maternal education and adult female education
stunting, or chronic malnutrition. Given the relatively were highly correlated. Logically maternal education is
low prevalence of acute malnutrition (6±7% in both the more appropriate variable to include in the nutri-
 FOOD SECURITY AND NUTRITIONAL STATUS 1973

tional status models, and so adult female education was other hand, average daily calories/AEU are about 10%
omitted. lower in urban areas (2,835 vs. 3,033), so the elasticity
may be re¯ecting a logical greater marginal propensity
5. Additional tests for our sample, not reported here, to spend on food.
support the hypothesis of di€erences in determinants of
the nutritional status of children 0±23 and 24±60 months 10. See MPF/UEM/IFPRI (1998) for a more complete
old. discussion of social networks in Mozambique.

6. The asset index comprised mainly household items 11. In the nutritional status models, mother's educa-
and appliances, such as radios and sewing machines, and tion has only two categories: none/illiterate and literate
vehicles. Although these assets seem more appropriate or at least some formal education. Insucient numbers
to urban than rural areas, this index performed better of women in the sample for the nutritional status models
than other asset indexes composed of other, more rural- had completed primary level or higher to create a third
based assets, including livestock and other animals, even and separate category.
for this highly rural sample.
12. Although in this sample urban households with
7. In some cases, observations were not available land do tend to be poorer than those without land, and
within a particular community or inclusion of the may be using land as a coping strategy (Maxwell, 1995),
dummy variable for community caused problems of it is less likely that this negative association with land
perfect collinearity. The total number of community holding and nutritional status is associated with low
dummy variables for each model is 264 for the calorie incomes. Including expenditures in the model should
availability model, and 268 and 252 for the 0±23 month have controlled for this e€ect. This conclusion is
age group and the 24±60 month age group, respectively, strengthened by noting that land holding did not show
for the nutritional status models. up as a key determinant in the calorie availability model.

8. Because the data used in the regression were not 13. The bene®ts of using piped water to the nutritional
weighted, elasticities are calculated using the unweighted status of children are well known, but quantity as well as
mean of 2950 calories/AEU/day for the sample. quality of water are important to reducing morbidity
and improving child nutrition (Burger and Esrey, 1995).
9. In Mozambique mean consumption expenditure per The data do not permit further exploration of which is
capita in rural areas is only about half that of urban the key factor in urban areas of Mozambique.
areas (4,797 Mts. vs. 10,476 Mts, respectively). On the

REFERENCES

Alderman, H. (1986) The e€ect of food price and income Barker, D. J. P. (1994) Mothers, babies and disease in
changes on the acquisition of food by low-income later life. British Medical Journal Publishing Group,
households. International Food Policy Research London.
Institute, Washington, DC. Behrman, J. and Deolalikar, A. (1987) Will developing
Alderman, H. (1990) Nutritional status in Ghana and its country nutrition improve with income? A case study
determinants. Social Dimensions of Adjustment in for rural south India. Journal of Political Economy 95
Sub-Saharan Africa Working Paper No. 3, The (31), 492±507.
World Bank, Washington, DC. Behrman, J. and Deolalikar, A. (1988) Health and
Alderman, H. and Garcia, M. (1994) Food security and nutrition. In Handbook of Development Economics,
health security: Explaining the levels of nutritional Vol. 1, ed. H. Chenery and T. N. Srinivasan, pp.
status in Pakistan. Economic Development and 630±711. North-Holland, NY.
Cultural Change 42 (1), 485±507. Blau, D., Guilkey, D., and Popkin, B. (1996) Infant
Alderman, H. and Higgins, P. (1992) Food and nutri- health and the labor supply of mothers. Journal of
tional adequacy in Ghana. Cornell Food and Nutri- Human Resources 31 (1), 90±139.
tion Policy Program Working Paper No. 27, Cornell Bouis, H. and Haddad, L. (1992) Are estimates of
University Food and Nutrition Policy Program, calorie-income elasticities too high? Journal of Devel-
Ithaca, NY. opment Economics 39 (2), 333±364.
Bairagi, R. (1980) Is income the only constraint on child Bound, J., Jaeger, D. and Baker, R. (1995) Problems
nutrition in rural Bangladesh? Bulletin of the World with instrumental variables estimation when the
Health Organization 58, 767±772. correlation between the instruments and the endog-
1974 WORLD DEVELOPMENT

enous explanatory variable is weak. Journal of the Philippines. American Journal of Clinical Nutrition
American Statistical Association 90 (430), 443±450. 63, 966±975.
Burger, S. and Esrey, S. (1995) Water and sanitation: Ruel, M.T., Habicht, J.P., Pinstrup-Andersen, P. and
Health and nutrition bene®ts to children. In Child Grohn, Y. (1992) The mediating e€ect of maternal
Growth and Nutrition in Developing Countries, ed. P. nutrition knowledge on the association between
Pinstrup-Andersen, D. Pelletier, and H. Alderman. maternal schooling and child nutritional status in
Cornell University Press, Ithaca, NY. Lesotho. American Journal of Epidemiology 135,
Cebu Study Team (1991) Underlying and Proximate 904±914.
Determinants of Child Health: The Cebu Longitu- Ruel, M., Levin, C., Armar-Klemesu, M., Maxwell, D.
dinal Health and Nutrition Study. American Journal and Morris, S. (1999) Good care practices mitigate
of Epidemiology 133 (2), 185±201. the negative e€ects of poverty and low maternal
FAO/WHO/UNU (Food and Agriculture Organization/ schooling on children's nutritional status: Evidence
World Health Organization/ United Nations Univer- from Accra. World Development, 27 (11), this issue.
sity) Expert Consultation (1985) Energy and protein Ruel, M., Garrett, J., Morris, S., Maxwell, D., Oshaug,
requirements. WHO Technical Report Series No. A., Engle, P., Menon, P., Slack, A. and Haddad, L.
724, World Health Organization, Geneva. (1998) Urban challenges to food and nutrition
Grosse, S. (1996) Economic and demographic determi- security: A review of food security, health, and
nants of child growth in Rwanda. Ph.D. dissertation, caregiving in the cities. Food Consumption and
University of Michigan, Ann Arbor, MI. Nutrition Division Discussion Paper No. 51, Inter-
Haddad, L., Ruel, M. and Garrett, J. (1999) Are urban national Food Policy Research Institute, Washing-
poverty and undernutrition growing? Some newly ton, DC.
assembled evidence. World Development 27 (11), this Sahn, D. (1994) The contribution of income to improved
issue. nutrition in C^ote d'Ivoire. Journal of African Econ-
Huang, J. and Bouis, H. (1996) Structural changes in the omies 3 (1), 29±61.
demand for food in Asia. 2020 Vision Discussion Sahn, D. (1988) The e€ect of price and income changes
Paper No. 11, International Food Policy Research on food-energy intake in Sri Lanka. Economic
Institute, Washington, DC. Development and Cultural Change 36 (2), 315±340.
Instituto Nacional de Estatõstica (INE) (1999) II Sahn, D. and Alderman, H. (1997) On the determinants
Recenseamento Geral da Populacß~ ao e Habitacß~ao. of nutrition in Mozambique: The importance of age-
Mocßambique. http://www.ine.gov.mz/Censo97/RE- speci®c e€ects. World Development 25 (4), 577±588.
CENS.HTM. Sharma, M., Garcia, M., Qureshi, A. and Brown, L.
Kennedy, P. (1992) A Guide to Econometrics, Third (1996) Overcoming malnutrition: Is there an ecore-
Edition. The MIT Press, Cambridge, MA. gional dimension? 2020 Vision Discussion Paper No.
Martorell, R. (1993) Enhancing human potential in 10, International Food Policy Research Institute,
Guatemalan adults through improved nutrition in Washington, DC.
early childhood. Nutrition Today 16 (4), 349±355. Strauss, J. and Thomas, D. (1995) Human resources:
Maxwell, D. (1995) Alternative food security strategy: A Empirical modeling of household and family deci-
household analysis of urban agriculture in Kampala. sions. In Handbook of Development Economics Vol.
World Development 23 (10), 1669±1681. 3A, ed. J. Behrman and T. N. Srinivasan, pp. 1883±
Maxwell, D., Levin, C. and Csete, J. (1998) Does urban 2023. North-Holland, NY.
agriculture help prevent malnutrition? Evidence from Subramanian, S. and Deaton, A. (1996) The demand for
Kampala. Food Consumption and Nutrition Divi- food and calories. Journal of Political Economy 104
sion Discussion Paper 45. International Food Policy (1), 133±162.
Research Institute, Washington, DC. Svedberg, P. (1990) Undernutrition in Sub-Saharan
Maxwell, S. (1996) Food security: A post-modern Africa: Is there a gender bias? Journal of Develop-
perspective. Food Policy 21 (2), 155±170. ment Studies 26 (3), 469±486.
Ministry of Planning and Finance, Government of Thomas, D., Strauss, J. and Henriques, M. (1991) How
Mozambique / Universidade Eduardo Mondlane / does mother's education a€ect child height? Journal
International Food Policy Research Institute (MPF/ of Human Resources 26 (2), 184±211.
UEM/IFPRI) (1998) Understanding Poverty and Thomas, D. and Strauss, J. (1992) Prices, infrastructure,
Well-Being in Mozambique: The First National household characteristics and child height. Journal of
Assessment (1996±97). Ministry of Planning and Development Economics 39 (2), 301±331.
Finance, Maputo and International Food Policy UNICEF (United Nations Children's Fund) (1990)
Research Institute, Washington, DC. Strategy for improved nutrition of children and women
Oshaug, O. (1994) Nutrition security in Norway? A in developing countries. UNICEF Policy Review
situation analysis. Scandanavian Journal of Nutrition Paper, UNICEF, New York.
38 (Supplement 28), 1±68. United Nations (1998) World Urbanization Prospects.
Reed, B. A., Habicht, J.-P., and Niamego, C. (1996) The The 1996 Revision. Department of Economic and
e€ects of maternal education on child nutritional Social A€airs, Population Division, United Nations,
status depend on socio-environmental conditions. New York.
International Journal of Epidemiology 25, 585±592. Ward, R. (1982) Asymmetry in retail, wholesale, and
Ricci, J. and Becker, S. (1996) Risk factors for wasting shipping point pricing for fresh vegetables. American
and stunting among children in Metro Cebu, Journal of Agricultural Economics 64 (2), 205±212.
 FOOD SECURITY AND NUTRITIONAL STATUS 1975

WHO (World Health Organization) (1979) Measurement WHO (World Health Organization) (1995) Physical
of Nutritional Impact. World Health Organization, Status: The Use and Interpretation of Anthropometry.
Geneva. WHO Technical Report Series 854, World Health
Organization, Geneva.