Community and International Nutrition

Dietary Diversity Is Associated with Child Nutritional Status:

Evidence from 11 Demographic and Health Surveys1,2
Mary Arimond3 and Marie T. Ruel
Food Consumption and Nutrition Division, International Food Policy Research Institute (IFPRI),
Washington, DC 20006

ABSTRACT Simple indicators reflecting diet quality for young children are needed both for programs and in some
research contexts. Measures of dietary diversity are relatively simple and were shown to be associated with
nutrient adequacy and nutritional status. However, dietary diversity also tends to increase with income and wealth;
thus, the association between dietary diversity and child nutrition may be confounded by socioeconomic factors.
We used data from 11 recent Demographic and Health Surveys (DHS) to examine the association between dietary
diversity and height-for-age Z-scores (HAZ) for children 6 23 mo old, while controlling for household wealth/
welfare and several other potentially confounding factors. Bivariate associations between dietary diversity and
HAZ were observed in 9 of the 11 countries. Dietary diversity remained significant as a main effect in 7 countries in Downlo
from
multivariate models, and interacted significantly with other factors (e.g., child age, breast-feeding status, urban/
jn.nutrit
rural location) in 3 of the 4 remaining countries. Thus, dietary diversity was significantly associated with HAZ, g by gu
either as a main effect or in an interaction, in all but one of the countries analyzed. These findings suggest that Octobe
there is an association between child dietary diversity and nutritional status that is independent of socioeconomic 2016
factors, and that dietary diversity may indeed reflect diet quality. Before dietary diversity can be recommended for
widespread use as an indicator of diet quality, additional research is required to confirm and clarify relations
between various dietary diversity indicators and nutrient intake, adequacy, and density, for children with differing
dietary patterns. J. Nutr. 134: 2579 2585, 2004.

KEY WORDS: Demographic and Health Surveys (DHS) dietary diversity child nutritional status
diet quality socioeconomic factors

All people need a variety of foods to meet requirements for
essential nutrients, and the value of a diverse diet has long
been recognized. Lack of diversity is a particularly severe
problem among poor populations in the developing world,
where diets are based predominantly on starchy staples and
often include few or no animal products and only seasonal
fruits and vegetables. For vulnerable infants and young chil-
dren, the problem is particularly critical because they need
energy- and nutrient-dense foods to grow and develop both
physically and mentally and to live a healthy life. For these
reasons, dietary diversity is now included as a specific recom-
mendation in the recently updated guidance for complemen-
tary feeding of the breast-fed child aged 6 to 23 mo (1).
Because of the perceived importance of dietary diversity
for health and nutrition, indicators of dietary diversity have
be- come increasingly popular in recent years. These types
of
1
Preliminary results were reported in Proceedings of the 2nd International
Workshop, Ouagadougou, November 2328, 2003 [Ruel, M. T. & Arimond, M.
(2004) Dietary diversity and growth: an analysis of recent demographic and
health surveys. In: Food Based Approaches for a Healthy Nutrition in West Africa
(Brower, E. D., Traore, A. S. & Treche, S., eds.). University Press, Ouagadougou
(in press)].
2
Funded in part by the Food and Nutrition Technical Assistance Project
(FANTA) managed by the Academy for Educational Development for USAID.
3
To whom correspondence should be addressed.
E-mail: m.arimond@cgiar.org.
indicators are particularly attractive because they are relatively
simple to measure and they are thought to reflect nutrient
adequacy, i.e., individuals consuming more diverse diets are
thought to be more likely to meet their nutrient needs. Simple
yet valid indicators are of particular importance for large
household surveys and for program management.
In developed countries, there are a number of studies link-
ing dietary diversity to nutrient intake, particularly among
adults; these studies are reviewed by Kant (2). Although there
is some indication from the literature that dietary diversity is
positively associated with a greater intake of energy and sev-
eral other nutrients among young children in developing
countries (3 6), additional research is warranted to character-
ize the exact nature of the relation between dietary diversity
and nutrient intake and adequacy. In young children, dietary
diversity has also been associated with improved nutritional
status (4,79), suggesting that diversity may indeed reflect
higher dietary quality and greater likelihood of meeting daily
energy and nutrient requirements.
However, dietary diversity was also shown to be strongly
associated with household socioeconomic status (8,10), and
links between socioeconomic status and child nutrition and
health outcomes have long been established. Interpretation of
associations between dietary diversity and nutritional status is
therefore complicated by the fact that both are strongly linked

0022-3166/04 $8.00 2004 American Society for Nutritional Sciences.

Manuscript received 2 June 2004. Initial review completed 1 July 2004. Revision accepted 2 August 2004.

2579
2580 ARIMOND AND RUEL
to household socioeconomic factors. Families with greater
incomes and resources tend to have more diverse diets, but
they are also likely to have better access to health care, and
better environmental conditions. Clearly, children in wealth-
ier households are better off and grow better for a number of
reasons, but improved nutrient adequacy may be one impor-
tant way in which household wealth and resources translate
into better outcomes for children. Thus, a key question is
whether dietary diversity is independently associated with
better child nutritional status because it accurately reflects
nutrient adequacy, or whether the association is found primar-
ily because dietary diversity is a particularly good proxy for
household socioeconomic status.
The present study addresses this question using data from
4
recent Demographic and Health Surveys (DHS) from 11
countries. Our focus is on infants and young children 6 23
mo of age, during the vulnerable period of transition from
breast- feeding to the family diet. The overall goal of the
research was to determine whether an association between
child dietary diversity and nutritional status among 6- to 23-
mo-old chil- dren was found across countries and regions with
varying dietary patterns, and whether this association
remained once socioeconomic factors were controlled for by
multivariate analyses. Answers to these questions are key to
understanding the nature of the links between dietary diversity
and child outcomes, and to fostering progress in developing
simple in- dicators of dietary quality.
SUBJECTS AND METHODS
Data
Data from recent DHS surveys from 11 countries were used. The
DHS are a series of standardized, nationally representative surveys
that have been implemented in ~70 countries since 1984. The
selection criteria for the countries included in the analyses were the
following: 1) Data set was available in mid-2002 and used the general
format of the most recent MEASURE DHS+ questionnaire; 2)
The country was from the African, South or Southeast Asian, or the
Latin America/Caribbean (LAC) region; 3) At least 6 of 7 broad food
groups needed to create the diversity indicator (see description be-
low) were represented in the questionnaire. Eleven data sets met
these criteria: Benin (2001), Cambodia (2000), Colombia (2000),
Ethiopia (2000), Haiti (2000), Malawi (2000), Mali (2001), Nepal
5
(2001), Peru (2000), Rwanda (2000), and Zimbabwe (1999).
All of the DHS that follow a standard protocol were given blanket
approval by the ORC Macro Institutional Review Board. Every
survey that deviated substantially from the standard protocol was
reviewed and approved separately. Each survey also received
approval from an in-country ethical review board, if such an
organization existed (personal communication, Altrena Mukuria,
ORC Macro, International).
Samples
After excluding children for whom age information was
missing, we randomly selected 1 child < 2 y of age in each
household. The proportion of children with missing values for age
ranged from 0% in 3 countries (Ethiopia, Colombia, and Peru) to
8% in Zimbabwe. Sample sizes for children aged 6 23 mo ranged
from 958 in Zimbabwe to 3662 in Peru. A number of children
were missing anthropometric
4
Abbreviations used: DHS, Demographic and Health Surveys [The DHS
program is funded by the U.S. Agency for International Development (USAID)
and administered by ORC Macro. ORC Macro provides technical assistance to
part- ner institutions in each country.]; HAZ, height-for-age Z-score(s); LAC, Latin
America/Caribbean; VIF, variance inflation factor; WHZ, weight-for height Z-
score(s).
5 7
At the time data were accessed, data from Ethiopia, Haiti, Mali, Nepal,
and Peru were indicated to be preliminary data on ORC Macro website.
measurements or had unacceptably extreme values. The proportion of
children with missing or extreme anthropometric values ranged from
2% in Nepal to 20% in Zimbabwe; these children were excluded from
bivariate and multivariate analyses.
Variable creation
Dietary diversity. The dietary diversity indicator used in the
analysis was created using data from the 7-d recall of foods/food
groups available in the MEASURE DHS+ surveys.6 Our general
approach was to develop a score that included a point for each of the
major nutritionally important types of food the child may have eaten,
while providing some balance between plant foods and animal-source
foods. Therefore, for the purpose of our analysis, foods/food groups
were regrouped and summed into a 7-point dietary diversity score, as
follows:7 1) starchy staples (foods made from grain, roots, or tubers);
2) legumes; 3) dairy (milk other than breast milk, cheese, or yogurt);
4) meat, poultry, fish, or eggs; 5) vitamin A-rich fruits and vegetables
(pumpkin; red or yellow yams or squash; carrots or red sweet
potatoes; green leafy vegetables; fruits such as mango, papaya, or
other local vitamin A-rich fruits); 6) other fruits and vegetables (or
fruit juices); and 7) foods made with oil, fat, or butter. Foods/food
groups that the child had consumed on 3 d in the previous week
received a score of
1 and those that8 the child had consumed <3 times in the past week Downlo
were scored 0. The choice of 3 d was arbitrary but was meant from
to capture foods eaten regularly. jn.nutrit
Terciles of dietary diversity were used to classify children into g by gu
low, average, and high diversity. The terciles were derived separately Octobe
for each country, and were made age-specific within the following 2016
age ranges: 6 8 mo, 9 11 mo, 1217 mo, and 18 23 mo. The
rationale for using age-specific terciles is that diversity increases
rapidly with age; by using age-specific terciles, children were ranked
as having low, average, or high diversity compared with other
children in their age range. For example, in Malawi, the high
diversity tercile included infants 6 8 mo who ate 2 or more food
groups, those 9 11 mo who ate 3 or more, and those 1223 mo who
ate 4 or more. Country- specific terciles were used because there are
currently no international guidelines or recommendations on which to
base cutoffs for low or high diversity. Tercile cutoffs were lowest
in Mali and Ethiopia, and highest in the LAC region across all age
groups.
Maternal and child nutritional status. Height-for-age Z-
scores (HAZ) were used as an indicator of nutritional status, and
maternal height and BMI were used for maternal nutritional status;
extreme values were excluded (11).
Proxies for household wealth and welfare. A variety of ap-
proaches have been used to characterize household wealth, welfare,
and socioeconomic status, including measurement of income and
expenditures and approaches incorporating information about house-
hold assets and access to services (12). Recently, authors analyzing
DHS and other similar surveys developed indices using information
on household assets, water and sanitation, and services (13,14). We
used a similar approach, with factor analysis as a data reduction tool,
to combine a large number of household-level variables into several
factors, with the objective of constructing a proxy for household
wealth and welfare. Categories of variables included in the factor
analysis (when available) were as follows: ownership of household
assets (radios, telephones, television, refrigerator), productive assets
(agricultural implements, land, sewing machines, bicycles, boats),
animals; main source of drinking water; type of sanitation facility;
main material of the floor and of the roof; and crowding (number of
household members per sleeping room).
Factor analysis was done separately for each country; some cate-
gories of variables were not available in all countries. Variables were
entered into the factor analysis either as summed scores or as ordered
variables with increasing scores reflecting increasing quality. For
6
In Haiti, the 24-h food group recall was used to construct the dietary
diversity variable, because the Haitian questionnaire did not include a 7-d recall.
Ten of the 11 countries included all 7 food groups. In Zimbabwe, the food
group list did not include foods made with fats and oils.
8
In Haiti, children received a score of 1 if they had the food yesterday, and
0 if not.
 DIETARY DIVERSITY AND CHILD NUTRITIONAL STATUS 2581

example, household assets, productive assets, and animals were each

summed, with items scored 1 if present, 0 if not. Water source,
sanitation facilities, and floor and roof materials were scored from
lowest to highest quality.
The factors were derived separately for urban and rural areas,
because the assets and household characteristics that differentiate
better off from worse off households in urban and rural areas are
likely to differ. After initial exploration, all models were restricted to
2 factors that, taken together, explained from 47 to 68% of the shared
variability in urban areas, and from 33 to 62% of the shared variabil-
ity in rural areas. In most cases, retaining 2 factors was equivalent to
retaining all factors with initial eigenvalues > 1. Scores for the 2
factors were used as continuous variables in the models.
Analytical methods
Sample weights were used for all analyses, and statistical testing
was performed in Stata (version 7) (15). Stata allows specification of
the sample design (stratification and clustering) of the surveys.
Descriptive analyses are presented first, to provide general infor-
mation on the characteristics of the study populations. They are
followed by results of the bivariate analyses of the association be-
tween childrens dietary diversity terciles and mean HAZ. The sig-
nificance of differences between means was tested using an adjusted
Wald test for joint hypothesis testing. Associations were considered
significant at P-values < 0.05.
Multivariate ordinary least-squares methods were then used to test
whether associations between dietary diversity and HAZ remained
significant after controlling for several potentially confounding fac-
tors at the child (age, age squared, sex, breast-feeding status), mater-
nal [height, BMI, education and number of prenatal care visits (a
proxy for access to health care)], and household level (wealth/welfare
factors 1 and 2, urban/rural location, and number of children < 5 y
old). Least-square means (adjusted for continuous variables in the
model) were computed to assess the difference in HAZ by dietary
diversity terciles. Multicollinearity was assessed in the models using
the variance inflation factor (VIF) (15); only age and age squared
showed evidence of multicollinearity (VIF > 10). Removing age
squared from the models did not change results for dietary diversity;
thus, age squared was retained in the models for theoretical reasons.
Two-way interactions between dietary diversity and several factors
were also tested in the multivariate analyses because we hypothesized
that the association between diversity and child nutritional status
might vary depending on certain child, maternal, or household char-
acteristics, i.e., we tested the two-way interactions between dietary
diversity and the following plausible factors: child age, whether child
was still breast-fed, mothers education, urban/rural location, and
wealth/welfare factors. Main effects and interactions were considered
significant at P-values < 0.05. For categorical variables, statistical
significance was assessed with joint tests of main effects.
RESULTS
Characteristics of survey households, mothers, Downlo
and children from
jn.nutrit
Key descriptive statistics for the survey households,
g by gu
mothers and children highlight some of the major differences Octobe
between countries (Table 1). In most countries, more than 2016
two-thirds of the households lived in rural areas; in
Colombia and Peru,

TABLE 1
Selected household, maternal, and child characteristics, by country
Africa Asia Latin America/Caribbean

Characteristic Benin Ethiopia Malawi Mali Rwanda Zimbabwe Cambodia Nepal Colombia Haiti Peru

Households, n 1312 2697 3228 1136 2110 958 2049 1809 1346 1758 3662

Rural 67 90 86 71 84 67 87 93 31 67 44
Female-headed 13 12 19 22 17 32 18 12 21 37 13
Piped water 40 13 22 27 35 43 3 34 83 51 66
No sanitary facility 71 86 20 19 3 29 85 77 12 46 30
Electricity 19 7 4 12 7 34 12 18 93 32 60
Maternal
Height,1 cm 158.1 157.2 155.7 161.6 158.0 159.4 152.5 150.3 154.5 158.1 150.2
BMI,1 kg/m2 21.9 20.0 21.9 21.8 22.4 22.9 20.4 20.1 24.4 22.3 24.8
<18.5, % 11 26 6 9 6 5 20 24 3 10 1
>25.0, % 13 3 10 13 15 20 4 3 38 19 42
Education
None, % 71 81 32 81 33 6 32 73 3 7 38

Child (623 mo)

Stunted (HAZ < 2) 28 47 47 35 40 31 36 44 16 20 22
Wasted (WHZ < 2) 16 18 9 18 10 9 19 18 1 8 1
Still breast-fed 88 92 93 90 91 78 81 96 47 69 75
Fed complementary foods
at least the minimum
recommended number
of times (if breast-fed)2 39 43 50 25 15 42 59 68 66 22 68
No solid food groups in last
7 d (68 mo old) 28 59 5 57 19 5 22 35 8 NA3 21

1 Values are means.

2 Breast-fed children 6 8 mo old should be fed meals of complementary foods at least 2 times/d, with additional snacks as desired, whereas
breast-fed children 9 23 mo old should be fed at least 3 times/d, with additional snacks (1).
3 NA, not available.
2582
ARIMOND AND RUEL

TABLE 2
Dietary diversity for children aged 6 23 mo (food groups eaten 3 d in the last week), by country

Mean diversity score % with low diversity 0 % with middle diversity 3 % with high diversity
(range 07) 2 food groups 4 food groups 57 food groups

Africa
Benin 3.2 38 31 33
Ethiopia 2.2 61 33 6
Malawi 2.4 57 37 6
Mali 1.7 70 21 8
Rwanda 2.9 42 43 16
Zimbabwe1 3.1 38 44 18
Asia
Cambodia 2.8 44 40 15
Nepal 2.8 43 45 12
Latin America/Caribbean
Colombia 4.8 11 25 65
Haiti (24-h)2 (3.8) (19) (47) (34)
Peru 4.5 13 29 58

1 The mean diversity score for Zimbabwe is on a scale of 0 6 because one food group was missing from the questionnaire.
2 The scores for Haiti are based on a 24-h recall, because the 7-d recall was not available.
the proportion was much lower. In general, Colombia and
Peru had more favorable household characteristics, whereas
Ethiopia consistently ranked low.
Mean maternal height was lowest in the Asian countries
and in Peru and highest in Haiti, Benin, Mali, and Zimbabwe.
The proportion of women with low BMI (<18.5) ranged from
1 to 11% in most countries, but was markedly higher in
Ethiopia and in both Asian countries. At the other end of
the spectrum, the highest rates of overweight and obesity
(BMI 25) were in the 2 Latin American countries.
Maternal education and literacy varied widely among coun-
tries. In 4 countries, more than two thirds of the women
reported that they had never attended school (Benin, Ethio-
pia, Mali and Nepal), whereas this was reported by approxi-
mately one third of the mothers in another set of 4 countries
(Malawi, Rwanda, Cambodia, and Haiti). In the remaining
countries (Zimbabwe, Colombia, and Peru) the proportion of
women who had no schooling was <10% and in these same
countries, >50% of the women reported having at least some
secondary education.
Among children aged 6 23 mo, the prevalence of stunting
(HAZ less than 2 SD) was highest in Ethiopia and Malawi,
and was notably lower in all 3 countries in the LAC region.
The prevalence of wasting (WHZ less than 2 SD) was
highest in Ethiopia, the West African countries (Benin and
Mali), and in both Asian countries (Cambodia and Nepal),
and very low in the 2 Latin American countries.
Feeding practices
Feeding practices for children aged 6 23 mo also differed
by country (Table 1). Breast-feeding was maintained through y
2 of life for most children in these countries. Over 85% of the
children were still breast-fed in 5 of the 6 African countries,
and in Nepal. Rates were lowest in Colombia and Haiti. Low
frequency of feeding appeared to be a problem in most coun-
tries, and particularly in Mali, Rwanda, and Haiti. In these 3
countries, the mean frequency of feeding was <2 on the day
before the survey. Late introduction of solids/semisolids was a
problem in a number of countries, and is particularly extreme
in Ethiopia and Mali, where more than half of the 6- to 8-mo-
old children received none of the food groups in the previous
week.
Downlo
from
jn.nutrit
Mean dietary diversity was lowest in Mali, followed by g by gu
Ethiopia and Malawi (Table 2). Note that in Mali and Ethi- Octobe
opia, the low mean reflects a large proportion of children who 2016
received none of the food groups (Table 1); in Malawi, very
few children ate none of the groups in the previous week, yet
diversity was very low. Mean dietary diversity was observed
to be highest in Peru and Colombia.
Similar patterns were observed when examining the per-
centages of children with low, average, or high dietary diver-
sity in each country, using fixed cutoff points to define these 3
categories. A very high percentage of children from Mali,
Ethiopia, and Malawi scored in the lowest diversity group
(having consumed only 0 2 food groups on 3 or more days in
the previous week), whereas more than half of the children in
the 2 Latin American countries (Colombia and Peru) scored
in the highest diversity group (having consumed 57 food
groups on 3+ d in the previous week). Mean dietary
diversity
was consistently higher in urban than in rural areas in every
country studied (not shown); this is consistent with findings
from previous analyses of other DHS data sets (16,17).
Associations between dietary diversity and height-for-age
Bivariate associations. Significant associations between
HAZ and dietary diversity terciles were found in bivariate
analyses in 9 of the 11 countries, but not in Benin or Cam-
bodia. Differences between extreme terciles in the 9 countries
ranged from 0.26 in Haiti to 0.56 in Peru. The differences
were generally in the expected direction, but in some cases
were not consistent in direction. For example, in Malawi and
Mali, children in the middle diversity tercile had the lowest
mean HAZ.
Multivariate analyses. Associations between dietary di-
versity and HAZ were significant as a main effect in 7 of the
countries studied: 4 in Africa (Ethiopia, Mali, Rwanda, and
Zimbabwe), the 2 Asian countries (Cambodia and Nepal) and
Colombia (Table 3).9 In these countries, the size of the
adjusted Z-score differences between low and high diversity
9
In 2 countries (Mali and Rwanda), P-values for individual contrasts between
high and low diversity tended to be significant (P = 0.06 and 0.07, respectively),
but the joint test for significance of all contrasts was significant, P < 0.05.
 DIETARY DIVERSITY AND CHILD NUTRITIONAL STATUS 2583

TABLE 3
Summary of regression results with dietary diversity terciles as one determinant of HAZ:
coefficients and significant main effects, by country1
Africa Asia Latin America/Caribbean

Benin Ethiopia Malawi Mali Rwanda Zimbabwe Cambodia Nepal Colombia Haiti Peru

Child age 0.25* 0.24* 0.24* 0.19* 0.28* 0.19* 0.06 0.22* 0.11* 0.10* 0.16*
Child age squared 0.00* 0.01* 0.00* 0.00 0.01* 0.00 0.00 0.00* 0.00 0.00 0.00*
Gender2 0.10 0.26* 0.17* 0.10 0.14* 0.11 0.07 0.01 0.18* 0.26* 0.03
Maternal height 0.05* 0.04* 0.04* 0.03* 0.04* 0.03* 0.05* 0.05* 0.06* 0.05* 0.05*
Maternal BMI 0.01 0.01 0.02 0.05* 0.02 0.03 0.00 0.04* 0.02* 0.05* 0.00
Maternal education
Primary vs. none 0.03 0.18 0.01* 0.14 0.12 0.12 0.15 0.17* 0.09 0.09 0.11*
Secondary vs. none 0.03 0.37 0.39* 0.12 0.15 0.16 0.45 0.24* 0.07 0.04 0.24*
Prenatal care visits
13 vs. none 0.21 0.03 0.15 0.14 0.03 0.38 0.12 0.10 0.02 0.30* 0.12
4+ vs. none 0.12 0.10 0.21 0.00 0.13 0.50 0.27 0.15 0.14 0.21* 0.09
Urban or rural3 0.13 0.10 0.36* 0.36* 0.47* 0.06 0.43* 0.38* 0.10 0.17 0.55*
Wealth/welfare factors
1st factor 0.03 0.05 0.12* 0.07 0.11* 0.01 0.01 0.14* 0.08* 0.10* 0.14*
2nd factor 0.06 0.01 0.10* 0.04 0.03 0.14* 0.11 0.09* 0.04 0.02 0.03
Still breast-fed4 0.20 0.60* 0.35* 0.01 0.00 0.11 0.28 0.09 0.06 0.00 0.03 Downlo
Number of children from
<5 y old 0.10* 0.14* 0.02 0.07 0.05 0.15 0.23* 0.06 0.17* 0.18* 0.16* jn.nutrit
Dietary diversity tercile g by gu
Middle vs. low 0.06 0.19* 0.08 0.11* 0.08* 0.47* 0.23* 0.07* 0.06* 0.08 0.03 Octobe
High vs. low 0.01 0.35* 0.06 0.23* 0.17* 0.68* 0.37* 0.23* 0.19* 0.10 0.09 2016
R2 0.24 0.19 0.19 0.28 0.20 0.15 0.17 0.24 0.26 0.23 0.29
F 26.62 20.29 33.22 14.58 29.20 6.09 7.56 25.98 23.79 27.02 40.98
n 1072 2372 2651 849 1802 617 771 1632 1179 1574 2874

1* Significant main effect, P < 0.05. For continuous variables and dichotomous variables, each coefficient with an asterisk was significant. For
categorical variablesmaternal education, number of prenatal care visits, and dietary diversity terciles coefficients are shown to be significant if
joint tests of contrasts were significant.
2 A indicates a negative coefficient for boys compared with girls.
3 A indicates a negative coefficient for rural areas compared with urban areas.
4 A indicates a negative coefficient for continued breast-feeding compared with no breast-feeding.

groups ranged from 0.24 in Colombia, to 0.59 in Zimbabwe
(Fig. 1). The bivariate associations between dietary diversity
and HAZ in Malawi, Haiti, and Peru were no longer signifi-
cant as main effects in multivariate analyses that controlled for
child, maternal, and household factors. In contrast, a signifi-
cant association was observed in Cambodia in the multivari-
ate, but not in the bivariate results.
In examining two-way interactions (Table 4), dietary di-
versity interacted with selected characteristics in a number of
countries, including 3 of the 4 countries in which the main
effect of dietary diversity was not significant in the multivar-
iate analysis. Only Benin had no association between
dietary diversity and HAZ (no main effect and no
interaction with other factors).
The most frequently observed interactions were between
dietary diversity and the age of the child, and between diver-
sity and current breast-feeding status (still breast-fed or not).
In 2 of the 3 countries in which dietary diversity interacted
with the childs breast-feeding status [Cambodia (Fig. 2) and
Nepal], the findings showed that dietary diversity was more
strongly associated with HAZ among children who were no
longer breast-fed.

FIGURE 1 Adjusted mean HAZ

by diet diversity tercile in 11 countries.
Values are means SEM, n = 617
2874. Means were adjusted for child
age and age squared, maternal height
and BMI, number of children < 5 y old
in household, and the 2 wealth/welfare
factor scores. Differences in HAZ by
diet diversity tercile were significant as
main effects in 7 countries (Ethiopia,
Mali, Rwanda, Zimbabwe, Cambodia,
Nepal, and Colombia), P < 0.05 (joint
test of significance of categories).
 2584
ARIMOND AND RUEL

TABLE 4
Adjusted differences in mean HAZ between highest and lowest dietary diversity terciles, by country for subgroups
of each variable that interacted significantly with dietary diversity in multivariate analyses1
Latin
America/Caribbean
Africa Asia

Malawi Mali Rwanda Cambodia Nepal Haiti Peru

Age category
611 mo 0.22 0.67 0.34 0.07
1217 mo 0.26 0.08 0.14 0.54
1823 mo 0.16 0.42 0.49 0.41
Location
Urban 0.07 0.44
Rural 0.45 0.07
1st wealth/welfare factor
Low 0.15
Middle 0.11
High 0.22
2nd wealth/welfare
factor
Low 0.12 0.58
Middle 0.61 0.39 Downlo
High 0.83 0.06 from
Still breast-fed jn.nutrit
Yes 0.15 0.21 0.34 g by gu
No 0.01 1.15 1.53 Octobe
2016
1 Interactions were considered to be significant when P < 0.05. There were no significant interactions in Benin, Ethiopia, Zimbabwe, or Colombia.
Positive values for differences indicate that HAZ was highest in the high diversity tercile.

The direction of the interaction between dietary diversity DISCUSSION

and other factors was not consistent across countries. For
example, the interaction between diversity and child age
group showed that diversity was most strongly associated with
HAZ among older children in some countries (e.g., Peru),
whereas the opposite was true in Rwanda, where the strongest
association was among children 6 11 mo old. Urban/rural
differences in the association between dietary diversity and
HAZ were also observed in 2 countries, with stronger associ-
ations in urban areas in Haiti; the opposite was true in Mali
with stronger associations in rural areas.
FIGURE 2 Interaction between breast-feeding status and dietary
diversity terciles in Cambodia DHS+ 2000. Values (adjusted mean
HAZ) are means SEM, n = 771. Means were adjusted for child age
and age squared, maternal height and BMI, number of children < 5 y
old in household, and the 1st wealth/welfare factor score. The interac-
tion was significant (P < 0.05).
This analysis of DHS data confirms that dietary diversity is
generally associated with child nutritional status, and that the
associations remain when household wealth and welfare fac-
tors are controlled for by multivariate analyses. This was
observed for a range of countries and populations with widely
different dietary patterns. Dietary diversity was significant as a
main effect in 7 countries in multivariate models, and inter-
acted significantly with other factors (e.g., child age, breast-
feeding status, urban/rural location) in 3 of the 4 remaining
countries. Thus, dietary diversity was significantly associated
with HAZ, either as a main effect or in an interaction, in all
but 1 of the countries analyzed. The existence of significant
interactions in some countries means that dietary diversity was
more strongly associated with child HAZ among some sub-
groups of the population.
Positive associations between dietary diversity and child
nutritional status were documented previously in China (7),
Kenya (4), Mali (8), and Haiti (18). Two additional studies in
Niger (5) and Guatemala (6) showed positive but not signif-
icant associations; however, sample sizes in both studies were
relatively small (reducing the statistical power to detect dif-
ferences) and in one of these (Guatemala), the children were
younger than in the other studies (9 11 mo). In addition to
variations in age groups, a variety of dietary methods were
used, and diversity indicators and cutoffs were defined differ-
ently in each study. The fact that a positive association be-
tween dietary diversity and child nutritional status was ob-
served in most studies, in spite of the lack of uniformity in
methodological approaches and populations studied, suggests
that the association is robust.
Two previous studies also documented an interaction be-
tween dietary diversity and breast-feeding status. Our results
for Cambodia and Nepal confirm their findings in showing a
 DIETARY DIVERSITY AND CHILD NUTRITIONAL STATUS
stronger association between dietary diversity and HAZ for
nonbreast-fed children (4,19). Dietary diversity may be more
important for nonbreast-fed children because they rely on
complementary food to meet all of their energy and nutrient
needs.
Other observed interactions are less consistent, and some
are difficult to interpret. There may be a variety of reasons
why diversity appears to be more strongly associated with
HAZ in subgroups. Depending on local diet patterns, high
diversity scores may be more or less nutritionally meaningful.
For ex- ample, if many food groups are given, but in
extremely small quantities, diversity scores are less
nutritionally meaningful. In some subgroups, there may be a
lack of nutritionally important variation; for example, low,
middle, and high terciles may all in fact reflect quite low
diversity (among the youngest children in Mali and Ethiopia
the age- and sample-specific terciles defined high diversity as
2 or more food groups, and very few children consumed >2).
There may also be 3-way interac-
tions; this was not assessed because subgroups become too
small. Interactions do indicate that more complex relation-
ships were present, and that coefficients for main effects
were misleading.
Some previous studies reporting associations between diet
diversity and child nutritional status did not control for likely
confounders. The purpose of controlling for wealth and wel-
fare factors in the analysis presented here was to try to disen-
tangle the association between dietary diversity and nutri-
tional status from household socioeconomic status. Although
our results are a step forward in determining that the associ-
ation is, at least in part, independent of socioeconomic factors,
it is important to recognize the limitations of this type of
cross-sectional analysis.
First, a childs nutritional status as reflected in HAZ repre-
sents a long-term cumulative process, whereas the dietary
information available in the DHS reflects only the previous
week. One major and unproven assumption underlying our use
of a 1-wk recall for dietary diversity is that recent diversity is
a good proxy for longer-term dietary diversity. Note that a
failure of this assumption would likely result in a lack of
association between dietary diversity and nutritional status, a
finding obtained for only 1 country in our analysis (Benin). A
second potential limitation is in our measurement of wealth
and welfare factors. Although measurement approaches
similar to ours are increasingly popular, like other measures of
socio- economic factors, they are imperfect, and we cannot
rule out the possibility that our control for socioeconomic
status was not complete.
The motivation for focusing on simple dietary diversity
indicators, measurable in cross-sectional surveys, is to move
forward in meeting the needs of programs and research
seeking simple measurement tools. In the context of
programs, or of research with multiple objectives (e.g., large
household sur- veys), detailed dietary assessment is usually
impossible. Dietary diversity indicators could be particularly
useful in these con- texts. Before they can be recommended
for widespread use, however, additional research is essential,
to confirm that di- versity is meaningfully associated with
nutrient adequacy in different population groups and in
countries with varying dietary patterns. In particular, relations
between dietary diver- sity and nutrient intake, adequacy, and
density, must be clar- ified. Additional research will also be
required to address a number of methodological issues related
to the construction of dietary diversity indicators (for example,
choosing food groups
2585
and diversity score cutoffs) and to explore the potential to
harmonize measurement tools and indicators for universal use.
If research does establish that indicators of dietary diversity
are good and consistent predictors of nutrient adequacy, these
indicators could become invaluable tools with which to assess
dietary quality as it relates to nutrient deficiencies, and to
monitor and evaluate progress aimed at improving diet quality
for young children.
ACKNOWLEDGMENTS
The authors thank Altrena Mukuria, Casey Aboulafia, and Noah
Bartlett of ORC Macro, International for sharing information on the
DHS data sets and for discussions of preliminary results. We thank
Eunyong Chung of USAID, and Anne Swindale and Paige Harrigan
of the Food and Nutrition Technical Assistance Project (FANTA)
managed by the Academy for Educational Development for USAID
for their helpful comments on a preliminary report. We are also
grateful to Wahid Quabili of the International Food Policy Research
Institute (IFPRI) for assistance with data analysis.
LITERATURE CITED
Downlo
1. PAHO/WHO (2003) Guiding Principles for Complementary Feeding of from
the Breastfed Child. PAHO/WHO, Washington, DC.
jn.nutrit
2. Kant, A. K. (1996) Indexes of overall diet quality: a review. J. Am. Diet.
Assoc. 96: 785791. g by gu
3. Hatloy, A., Torheim, L. E. & Oshaug, A. (1998) (1998) Food vari- etya Octobe
good indicator of nutritional adequacy of the diet? A case study from an urban 2016
area in Mali, West Africa. Eur. J. Clin. Nutr. 52: 891 898.
4. Onyango, A., Koski, K. G. & Tucker, K. L. (1998) Food diversity versus
breastfeeding choice in determining anthropometric status in rural Kenyan tod-
dlers. Int. J. Epidemiol. 27: 484 489.
5. Tarini, A., Bakari, S. & Delisle, H. (1999) The overall nutritional quality
of the diet is reflected in the growth of Nigerian children. Sante 9: 2331.
6. Brown, K. H., Peerson, J. M., Kimmons, J. E. & Hotz, C. (2002) Options
for achieving adequate intake from home-prepared complementary foods in low
income countries. In: Public Health Issues in Infant and Child Nutrition (Black,
R.E. & Michaelsen, K. F, eds.), pp. 239 256. Nestec, Ltd., Vevey/Lippincott
Williams & Wilkins, Philadelphia, PA.
7. Taren, D. & Chen, J. (1993) A positive association between extended
breast-feeding and nutritional status in rural Hubei Province, Peoples Republic of
China. Am. J. Clin. Nutr. 58: 862 867.
8. Hatloy, A., Halland, J., Diarra, M. & Oshaug, A. (2000) Food variety,
socioeconomic status and nutritional status in urban and rural areas in Koutiala
(Mali). Public Health Nutr. 3: 57 65.
9. Arimond, M. & Ruel, M. T. (2002) Progress in Developing an Infant and
Child Feeding Index: an Example Using the Ethiopia Demographic and Health
Survey 2000. Food Consumption and Nutrition Division Discussion Paper 143.
International Food Policy Research Institute, Washington, DC.
10. Hoddinott, J. & Yohannes, Y. (2002) Dietary Diversity as a Food
Security Indicator. Food Consumption and Nutrition Division Discussion Paper
136. International Food Policy Research Institute, Washington, DC.
11. World Health Organization (1995) Physical Status: The Use and Inter-
pretation of Anthropometry. WHO Technical Report Series. WHO, Geneva, Swit-
zerland.
12. Falkingham, J. & Namazie, C. (2002) Measuring Health and Poverty: A
Review of Approaches to Identifying the Poor. Health Systems Resource Centre,
London, UK.
13. Filmer, D. & Pritchett, L. (1998) Estimating Wealth Effects Without
Expenditure Data or Tears: An Application to Educational Enrollments in States
of India. World Bank Policy Research Working Paper. The World Bank, Washing-
ton, DC.
14. Gwatkin, D. (2000) Health inequalities and the health of the poor: What
do we know? What can we do? Bull. WHO 78: 318.
15. Stata Corporation (2001) Stata Statistical Software: Release 7.0 Ref-
erence. Stata Corporation, College Station, TX.
16. Ruel, M. T. (2000) Urbanization in Latin America: constraints and
opportunities for child feeding and care. Food Nutr. Bull. 21:1224.
17. Ruel, M. T. & Garrett, J. (2004) Features of urban food and nutrition
security and considerations for successful urban programming. e-JADE (in
press).
18. Ruel, M. T., Menon, P., Arimond, M. & Frongillo, E. (2004) Food
insecurity: an overwhelming constraint for child dietary diversity and growth in
Haiti. FASEB J. 18: A106 (abs.).
19. Marquis, G. S., Habicht J.-P., Lanata, C. F., Black, R. E. & Rasmussen,
K. M. (1997) Breast milk or animal-product foods improve linear growth of
Peruvian toddlers consuming marginal diets. Am. J. Clin. Nutr. 66: 11021109.