HHS Public Access

Author manuscript
Ethiop J Health Dev. Author manuscript; available in PMC 2018 January 01.
Author Manuscript

Published in final edited form as:

Ethiop J Health Dev. 2017 ; 31(2): 75–84.

The effect of early childhood stunting on children’s cognitive

achievements: Evidence from young lives Ethiopia
Tassew Woldehanna1, Jere R. Behrman2, and Mesele W. Araya2
1Department of Economics, Addis Ababa University, Ethiopia
2University of Pennsylvania, Philadelphia, PA, USA
Author Manuscript

Abstract
Background—There is little empirical evidence on the effect of childhood malnutrition on
children’s cognitive achievements in low income countries like Ethiopia. A longitudinal data is
thus vital to understand the factors that influence cognitive development of children over time,
particularly how early childhood stunting affects cognitive achievement of children up to the age
of 8 years.

Objective—To examine the effect of early childhood stunting on cognitive achievements of

children using longitudinal data that incorporate anthropometric measurements and results of
cognitive achievement tests such as Peabody Picture Vocabulary Test and Cognitive Development
Assessment quantitative tests.

Method—Defining stunted children as those having a standardized height for age z-score less
Author Manuscript

than −2; we used a Propensity Score Matching (PSM) to examine the effect of early childhood
stunting on measures of cognitive performance of children. The balance of the propensity score
matching techniques was checked and found to be satisfied (P<0.01)

Results—Early childhood stunting is significantly negatively associated with cognitive

performance of children. Controlled for confounding variables such as length of breastfeeding,
relative size of the child at birth, health problems of early childhood such as acute respiratory
illness and malaria, baseline household wealth, child gender, household size and parental
education, estimates from PSM show that stunted children scored 16.1% less in the Peabody
Picture Vocabulary Test and 48.8% less in the Quantitative Assessment test at the age of eight,
both statistically significant at P<0.01.

Conclusions—It is important to realize the importance of early investment in terms of child

Author Manuscript

health and nutrition until five years for the cognitive performance of children. As household
wealth and parental education are particularly found to play an important role in children’s
nutritional achievements, policy measures that are directed in improving household’s livelihood
may have a spill-over impact in improving child nutritional status, and consequently cognitive
development and schooling.

Corresponding Author: [email protected]. P.O. Box: 24570/1000, Addis Ababa, Ethiopia.

 Woldehanna et al. Page 2

Keywords
Author Manuscript

Early childhood; stunting; cognitive achievements; Ethiopia

Introduction
For children to be successful at later stages in life, cognitive skills are deemed crucial (1).
Evidence also suggests that such cognitive skills are highly dependent on the home
environment and parents’ nutritional investment throughout the life time of the children
although early investments have a more positive impact than do later investments (1–3). This
assertion stems from the fact that the years from conception through birth to eight years of
age are critical for healthy physical and mental development of children (4).

However, notwithstanding the importance of early nutrition on children’s health status as

Author Manuscript

well as their ability to learn, think analytically and socialize with others and their capacity to
adapt to changes, child malnutrition rates are very high in many parts of the developing
world (1–10), where chronic malnutrition as measured by stunting affects 37% of all under-5
children in Sub-Saharan Africa and robs at least 2–3% of its GDP growth due to losses from
increased health care costs in addition to losses from poor cognitive function and the deficits
it causes in schooling and learning ability (5). Malnutrition in Ethiopia specifically
contributes to an estimated 270, 000 deaths of under-five children each year (11).

A growing number of both experimental and non-experimental empirical studies have been
done in developing countries. Many of those studies indicate that early nutritional
supplements have positive effects on cognitive achievement among teenagers, and later on
their adulthood development (12–18). As example, a randomized nutrition intervention in
Author Manuscript

rural Guatemala that captures the effect of exposure to an intervention from birth to 36
months indicated significantly positive and fairly substantial effects a quarter century after it
ended – increased attainment and speedier grade progression by women, higher scores on
reading comprehension tests as well as on non-verbal cognitive tests for both women and
men were observed (12). A recent experimental study in slum areas of Delhi also showed
that iron supplements administered two to six year olds reduced pre-school absenteeism by
one-fifth and had significant positive effects on child weight within the first six months of
the program, particularly among poorer communities (18–19). A study from Vietnam that
used cross-sectional data collected during a baseline survey of a randomized trial to examine
the association between results of educational tests and the anthropometric status of school
children also concluded that better nutrition during early childhood would significantly
increase educational achievement in adulthood under the condition that the individuals
Author Manuscript

completed the transition to complete primary education (20).

Similar to the randomized experimental studies, longitudinal studies also demonstrate that
children who experience early malnutrition are more likely to have lower scores in tests
assessing cognitive function, psychomotor development, fine motor skills, activity levels and
attention span (1–2, 21–24). A study on more than 2000 Filipino children showed a negative
consequence of malnutrition on cognition and schooling, where severe or moderate stunting
resulted in lower scores in cognitive tests at the ages of 8 and 11, as compared to non-

Ethiop J Health Dev. Author manuscript; available in PMC 2018 January 01.
 Woldehanna et al. Page 3

stunted children (25). A similarly crafted study also examined the effects of early childhood
Author Manuscript

nutrition and subsequent academic achievement using a longitudinal data set that follows a
large sample of children from birth until the end of their primary education (21). The
findings demonstrate that malnourished children perform more poorly in school, even after
correcting for the effects of unobserved heterogeneity both across and within households.
The advantage (at least part of it) that well-nourished children enjoy arises from the fact that
they enter school earlier and thus have more time to learn (26).

There are few studies on this issue in Africa and the existing results are somehow mixed for
two main reasons (22, 27). One, it is very difficult to get data on all variables of interest,
which might result in serious problems of omitted variable bias; and two, the variables that
have data are frequently measured with error, which can lead to problems of attenuation bias
(21). Yet, in spite of the difficulties, a longitudinal data from rural Zimbabwe showed that
early childhood nutritional deficiencies in the form of low height-forage can be linked to
Author Manuscript

poorer cognitive attainment later in life (22). Another study from Nigeria also compared
school academic performances of primary school pupils after a retrospective assessment of
their health and nutritional history dating back to their prenatal gestation periods. However,
there was no significant association between early nutritional status and later school
performance (26–27). To sum up, it has long been known that good nutrition is essential to
children’s physical and cognitive development (1–3, 18, 20), but recent evidence sheds new
light on the optimal timing of interventions to improve child nutrition and the long-term
effects of such interventions. Authors like Dobbing particularly argue that the period from
birth to six month is a critical period for brain development so that poor nutrition during this
period has a long lasting effect on cognitive development of children (28–29), whereas some
others like Glewwe and King, who assessed the impact of timing of malnutrition on
cognitive development by employing longitudinal data, suggest that the period from 18–24
Author Manuscript

months since birth may be critical for cognitive development (30). Given such differences in
the estimated impacts of the timing of malnutrition on cognitive development and learning
outcomes, it is worthwhile to provide evidence on whether children’s stunning up to the age
of five years has an effect on cognitive development of children at the age of eight.
Moreover, it is particularly useful to provide such evidence for Ethiopians where
malnutrition of children is highly prevalent and has become a major contributor to child
mortality as underlying cause for nearly 50% of under-five deaths (31), but there have been
relatively few studies. By estimating the effect of the z-score of height-for-age on cognitive
development of eight year old children using longitudinal data from Young Lives project,
this study hence provides fresh evidence on the relation between early childhood stunting
and cognitive development of children for a typical developing country, Ethiopia.
Author Manuscript

Method
Sample children
The data used in this study comes from Young Lives, a 15-year survey investigating the
changing nature of childhood poverty in Ethiopia, India, Peru and Vietnam
(www.younglives.org.uk). In Ethiopia, the project is overseen by the Ethiopian Development
Research Institute (EDRI) and University of Oxford, funded by UK aid from the Department

Ethiop J Health Dev. Author manuscript; available in PMC 2018 January 01.
 Woldehanna et al. Page 4

for International Development (DFID). In each country, the Project follows 2000 children
Author Manuscript

aged 1 and other 1000 children aged 8 years in 2002. While the first group is referred as a
Younger Cohort, the latter is termed as an Older Cohort. The children were selected from 20
sentinel sites of Ethiopia’s five major regions (Tigray, Amhara, Oromia, SNNP and Addis
Ababa) that represent for more than 90% of the total population.

The methodology in the first stage was purposive because the sentinel1 sites in each region
were chosen such that the cost of tracking children in the future was manageable, to reduce
the probability of attrition in remote pastoralist areas (32). In each region three to five
sentinel sites were selected, with a balanced representation of poor and less-poor
households, and urban and rural areas. Finally, from each sentinel site 100 children who
were born between April 2001 and June 2002 (the Younger Cohort) and 50 children born
between April 1994 and June 1995 (the Older Cohort) were selected using simple random
sampling.
Author Manuscript

As the aim of this study is to examine the impact of early childhood stunting on cognitive
performance of children aged 5 and 8, we, however, made use of data only from the Younger
cohort who were first appraised in 2002 when they were around age one. The children were
then re-interviewed in 2006 and 2009. At the beginning, there was no refusal from parents of
the kids to be part of the long-term study, but once started some children were untraceable
(n=33) and some others (n=10) refused to give responses up to round three. Additionally, 72
children died up to the third round. Excluding deaths, the total attrition rate over eight years
is 2.15%, reducing the sample of the children to 1884 by the third round (Table 1).

The Young Lives longitudinal data is of high quality, where the validity and reliability of the
tools used in each survey were approved by the Ethics Committee of University of Oxford
Author Manuscript

and the data collection process was managed by the Ethiopian Development Research
Institute (EDRI). Prior data collection in each survey an extensive training focusing on the
child, household and community questionnaires, tracking schedule, preliminary interview,
consent form, household roster and contact details was given for all field supervisors and
enumerators

Immediately following the trainings, pilot surveys were conducted so that necessary
amendments could be made to the questionnaires and organization of the field work.

The field enumerators were considered well trained if no error of administrating and scoring
was reported by their field supervisors during the piloting. Many of the field supervisors
were educated to university level and were with a good track of field work supervision, and
more interestingly, working with the same field supervisors and enumerators since the
Author Manuscript

baseline survey in 2002 has enabled us to build stable relations with the families of the
children and to minimize errors of administering the interviews and cognitive tests (33–34).

The data collected in all the three rounds include a great deal of information on children’s
and their families’ circumstances (including, but not limited to, prenatal health care, self-

1The concept of a sentinel site comes from health surveillance studies and is a form of purposive sampling where the site is deemed to
represent a certain type of population, and is expected to show typical trends affecting those particular people or areas (41).

Ethiop J Health Dev. Author manuscript; available in PMC 2018 January 01.
 Woldehanna et al. Page 5

reported five point (Likert) birth weight ranging from −2 to 2, child illness in the last 24
Author Manuscript

hours as reported by parents, family background and wealth, household composition, family
structure, per capita consumption expenditure, work patterns and social relationships).

Measures
The data from the Young Lives also contains information on children’s nutritional,
educational and cognitive outcomes. Therefore, childhood stunting in this study is measured
by the z-scores of the child’s height-for-age2 at the ages of 1 and 5 years. The construction
of height-for-age z-score is based on comparisons with a “healthy” reference population
based on the WHO (2007) criteria. The WHO (2007) has developed a table to measure
nutritional achievements based on cross-national data. We have used this WHO table to
compute height-forage z-scores (35). Based on the tables’ cut-off, a child whose height-for-
age z-score is less than −2 is termed as stunted, and if it is less than −3, the child is severely
Author Manuscript

stunted. Those with less than −6 and above 6 height-for-ages are considered as biologically
implausible scores and hence were removed from the analysis.

The validity of this reference standard stems from the empirical observation that well-
nourished and healthy children have a very similar distribution of height and weight to the
reference population, regardless of their ethnic background or where they live (36–37).

Regarding the outcome variables, child’s cognitive achievements were measured by The
Peabody Picture Vocabulary Test (PPVT), the Cognitive Developmental Assessment
Quantitative (CDA-Q) and Mathematics Tests. The PPVT is a widely-used test of receptive
vocabulary. Its main objective is to measure vocabulary acquisition in persons from 2.5 years
old to adulthood (38). The test is individually-administered, untimed, norm-referenced and
orally-administered. The maximum limit of the score is 204, but to avoid any bias the raw
Author Manuscript

scores were standardized for poor statistical behaviors. As the same set of questions was
given to the children at both five and eight years of ages, the PPVT results are
straightforwardly comparable between the second and third rounds of the survey.

The quantitative assessment used to measure numerical abilities of the children was also
derived from the International Evaluation Association (IEA). At the age of 5, the Cognitive
Developmental Assessment quantitative (CDA-Q) had 15 items that were administered
orally. But, at the age of eight the assessment test has two parts. While the first part was
administered orally and had 9 items, scoring “1” for correct and “0” for blank or incorrect,
the second part was a written test administered by the child and had 20 items, making the
overall questions 29. This means that, unlike in the PPVT test, the results from Mathematics
test between the two rounds are not straightforwardly comparable (39–40).
Author Manuscript

All cognitive achievement tests were also adapted to the Ethiopian context and translated
into local languages during the test administration. To make the results straightforwardly
interpretable in the analysis, we also transformed the PPVT and Mathematics scores into

2Height-for-age reflects cumulative linear growth. Height-for-age deficits indicate past or chronic inadequacies in nutrition and/or
chronic or frequent illness, but cannot measure short-term changes in malnutrition. Low height-forage relative to a child of the same
sex and age in the reference population is referred to as ‘shortness’. Extreme cases of low height-for-age, where shortness is
interpreted as pathological, are referred to as ‘stunting’.

Ethiop J Health Dev. Author manuscript; available in PMC 2018 January 01.
 Woldehanna et al. Page 6

logarithm form. As additional evidence, we also looked at the school readiness and progress
Author Manuscript

of the children at the ages of 7 and 8, respectively.

Analysis
In addition to a simple descriptive statistics, we used Propensity Score Matching (PSM)
model to estimate the effect of childhood stunting on cognitive performance of children aged
5 and 8 years. The goal of such estimation strategy is to re-establish experimental conditions
in a non-experimental setting (41). The process of estimation proceeds in two steps.

First, a propensity score for each child as a conditional probability of being stunted given a
full set of covariates is estimated from a logit model. That is, the probability that a child to
be stunted is given by

(1)
Author Manuscript

where d is dummy variable for a child being stunted, while x are a vector of variables that
affect both the stunting and cognitive development outcomes, ε is an error term, α are vector
of parameters to be estimated and F(.) is a logistic function.

Second, the propensity scores generated from relation 1 are used to create a matched control
group of children who were not stunted. That is, the Average Treatment effect on the Treated
(ATT) is provided by

(2)
Author Manuscript

where di is dummy variable if the child is stunted and 0 otherwise, CDi1 and CDi0 are
cognitive outcomes, with CDi1 the score of outcome that would be observed if the child was
stunted, CDi0 the outcome score that would be observed on the same age if the child was not
stunted.

Selecting variables to be included in the logit model of propensity score needs a deeper
knowledge of the reasons for stunting, if not omitting relevant variables can seriously
increase bias in resulting estimates. There is no guideline on how to choose conditioning
variables, x, but selection of x variables intuitively is very important such that the covariates
must include variables that affect both the outcome and being stunted (42).

The advantage of our longitudinal data is that we can capture much of the differences among
Author Manuscript

the children by including variables what happened to mothers and the households before and
after the birth of the children and match the stunted with the non-stunted in a better way than
many studies relying on cross-section data (20–23). If the matching process proceeds
appropriately, the stunted and non-stunted will have similar measured characteristics and the
effects of stunting can be estimated by comparing the mean scores of the matched children.
To do this, we examined the level of matching scores using kernel density graphs and limit
the sample children for whom there is sufficient overlap in propensity scores (in the area of

Ethiop J Health Dev. Author manuscript; available in PMC 2018 January 01.
 Woldehanna et al. Page 7

common support). To check the robustness of the results we estimated the PSM using
Author Manuscript

alternative techniques such as Kernel Matching Method, Radius Matching and One-to-One
Matching techniques.

Results
Descriptive statistics
Table 2 reports the variables of interest of this study. It appears that 52% of the children are
male and 60% of the sample households reside in rural areas. On average, at age 1 the
children had a height-for-age z-score that is 1.53 standard deviations below the WHO
standard. This makes more than one-third (42%) of them in the sample are to be stunted
(height-for-age z-scores less than –2), while more than one-fourth (22%) of them are
severely stunted (height-for-age z-scores less than –3). There seems, however, a sort of
recovery from early stunting by the age of 5 years, where the height-for-age z-score was
Author Manuscript

1.45 standard deviations below the WHO standard and the percentages of early stunted and
severely stunted children declined to about 31% and 8%, respectively.

Also, as reported by the primary caregivers of the children about one-half (47%) of the
children have experienced some kind of health problem by the age of 1 year, with 13% of
them severely ill or injured such as acute respiratory illness and malaria in the last 24 hours
of the survey time. Self-reported five point (Likert) scale for the relative size of the children
at birth (ranging from −2 to 2) also show that the sample children have an average birth size
of 0 on the 5 point scale, indicating that the average birth weight of the children was
extremely small that might be one of the factors for the high level of stunting (−1.53 of
height-for-age z-score at age 1).
Author Manuscript

Regarding maternal health care services, many of the children’s mothers were young adults
with average age of 27 years during giving birth. Also, only about 62% of them reported that
the pregnancy was wanted, while 38% of them were unaware of their pregnancy till the first
few months. As the result, the average number of antenatal visits during pregnancy was very
limited in number, only 2 times during the whole pregnancy period. It was also reported that
about 6% of the mothers were disabled or had long-term health problems. Again, at the age
of 1 year, about 6% of the children’s parents were not living together as the result of divorce,
separation or death.

Although the wealth indices3 of the sample households over the three rounds of the survey
consistently improved–from 0.21 in 2002 to 0.23 in 2006 and further to 0.35 in 2009, many
of the households where the children were drawn from were still poor as the wealth index
criterion of Young Lives states that a household whose wealth index is below 0.45 is
Author Manuscript

considered poor (32). The average household size during the baseline survey was 6, with per

3Wealth index variable captures indicators that are broader than production assets, such as home ownership and the durability of that
home, plus access to infrastructure such as water and sanitation. It is constructed by summing three components: measures of housing
quality, consumer durables and services. The components are calculated as scaled values (0 to 1). The measure of housing quality is
based on the type of material the floor, roof and walls were made of, and the number of rooms relative to household size. The service
component is the average of the dummy variables on the availability of electricity, piped water, fuel for cooking and toilet facilities.
The consumer durables measure is the sum of the dummy variables related to households’ ownership of radio, TV, refrigerator,
bicycle, motorcycle, car, mobile phone, landline phone and fan.

Ethiop J Health Dev. Author manuscript; available in PMC 2018 January 01.
 Woldehanna et al. Page 8

capita annual consumption expenditure 1480 Ethiopian Birr. Also, the average years of
Author Manuscript

schooling for the mothers and fathers were found to be low at 2.8 and 4.4, respectively. A
quarter (25%) of the children also attended preschool during their early childhoods.

Preliminary difference of outcomes

one of the most important issues in this study is to explore whether early childhood stunting
does really limit cognitive performance of children. To examine those preliminary
differences between the means of different children, we applied both one-sample-t test and
multivariate statistics. If we begin with the multivariate statistics as seen at the bottom of
Table 3, the T2 value is highly significant (Hotelling T2=35074.27 & P<0.01) indicating that
we don’t capitalize on chance when computing multiple t-tests. This statistical test indicates
that stunted children had significantly lower cognitive scores than their non-stunted
counterparts both at the ages of 5 and 8, where the differences in standard PPVT scores
Author Manuscript

between the two groups at the ages of 5 and 8 were respectively 6.19 and 12.37. The same is
true with the percentage of correctly answered CDA-Q and Maths tests, where stunted
children scored respectively 4.58 and 8.60 percentage points lower than the non-stunted
children. Furthermore, stunted children were less likely to enroll in formal schooling at the
age of 7, with slower early grade progression at the age of eight, than those none-stunted
children.

Propensity Score Matching (PSM)

to infer causality, it may be appropriate to see if stunting has an effect on PPVT and
Mathematics test scores. Accordingly, we employed propensity score matching techniques
where we run first stage logit model estimation of stunting at age 5 so as to predict
propensity scores of stunting that are used for matching to estimate average treatment effects
Author Manuscript

of the treatment (stunting) on the treated (stunted). Variables potentially associated with the
probability of a child being stunted (treatment) and cognitive developments (outcomes) are
included in the first stage logit regression model. Those mainly include dummy for a child
being stunted at the age of one, which is expected to capture the carry-over effect that might
exist starting from birth, length of breastfeeding, relative size of the child at birth, dummy
for child had health problems such as acute respiratory illness and malaria by the age of one
year, baseline family wealth index, child gender, household size, and parental education
levels in years. The coefficients from logit model indicated that being stunted at age 5 is
highly associated with being stunted at age 1. The inclusion of being stunted at age one year
in the first logit model is particularly very important to capture the carry-over effect that
might arise from time lag at the ages of five and eight. As expected, early household
circumstances such as baseline household wealth index, education and residential areas (in
Author Manuscript

favour of urban area) were found to be negatively associated with being stunted at the age of
5.

Having identified the contributing factors for early childhood stunning in this way and
driving predicted values from the logit model, we computed the distribution of the
propensity score for each child in the treated and control groups to identify the existence of a
common support and the level of matching in terms of covariates.

Ethiop J Health Dev. Author manuscript; available in PMC 2018 January 01.
 Woldehanna et al. Page 9

Figurer 4 exhibits the distribution of the children with respect to the estimated propensity
Author Manuscript

scores. Most of the stunted children are found in the right side of the distribution, while
those stunted children are found in the left side of the distribution graph. As seen in the
second panel of the figure, the match between the stunted and non-stunted appears to be
satisfactory. In other words, the matching result indicates that the differences between the
two group of children can be captured by the observables and the only remaining relevant
difference between the two groups of children is being stunted or not at early years of time
and there is no individual effect in relation to the cognitive outcomes. To ensure that the
stunted and non-stunted children look identical in terms of their covariates, we further ran a
balancing test, which helps to check whether the propensity score model is adequately
specified. Results of this test also indicated that the balancing condition is satisfied. The
conditional probabilities estimated from the propensity score model are then used to
compute the matching analysis, where as the result we find 1813 of children in the region of
Author Manuscript

common support, with 569 stunted and 1244 non-stunted. Only 70 children (18 stunted and
52 non-stunted) were found to be off the common support, which as the result we excluded
them from the propensity score matching analysis.

Average Treatment effect of the Treatment (ATT) estimates

As discussed earlier in the methodology part, we used various matching techniques
including kernel, radius and one-to-one matching techniques with the application of stata
“psmatch2” command (42) to estimate the Average Treatment effect of the Treatment (ATT)
on cognitive development of stunted children. The results indicate that effect of estimates
from the three matching techniques are generally in the same direction, but for economic of
space we only reported estimation results from the kernel and radius matching analysis in
Table 4.
Author Manuscript

The results of the ATT confirm that there are cognitive achievement advantages from being
non-stunted in early years of time. The differential effects are consistently significant and
higher in magnitude in both matching models. From the kernel analysis it is apparent to
observe that stunted five-year-old children scored 16.1% lower in the standard PPVT test
and 7% lower in the Cognitive Developmental Assessment quantitative test (CDA-Q) than
their none-stunted counterparts. The differences in scores also are visible at the age of eight,
where the difference in standard PPVT score remained flat at 16.1% and the difference in
Mathematics score was much higher (44.8%) at age 8 than at the age of 5 (7.2%). This may
signify that the negative effect of childhood stunting magnifies as the children get older.

Furthermore, although it might not be an indicator of cognitive development, we also tried to

look at whether there is an association between childhood stunting and schooling at the ages
Author Manuscript

of 7 and 8. The results show similar effects to those of cognitive outcomes, where stunted
children were found with 17.8% less likely to be enrolled in school at the age of 7 than
none-stunted children. Among those who got enrolled at the seven years of age, stunted
children also showed a slow grade progression at the age of 8, implying that none-stunted
children showed higher primary school enrolment and grade progression rates than stunted
children of the same age.

Ethiop J Health Dev. Author manuscript; available in PMC 2018 January 01.
 Woldehanna et al. Page 10

All in all, the result from the PSM analysis sheds light on the causal effect of malnutrition on
Author Manuscript

cognitive achievement of children at the ages of 5 and 8 years, which is in support of our
initial hypothesis that early childhood nutrition is one of the key intervention points for the
development of children’s brain and cognitive performance.

Discussion
This study was conducted to find empirical evidence in support of the claim that nutrition in
the early years of life can have a profound effect on children’s health status as well as their
ability to learn, think analytically and socialize with others and their capacity to adapt to
changes (2–10). To do this, we used a very rich dataset, gathered over three time periods
(infancy, early childhood and elementary age) with very low attrition rates from the Young
Lives project in Ethiopia that is a 15-year Longitudinal survey investigating the changing
nature of childhood poverty in the country. We specifically looked at whether stunting at age
Author Manuscript

five (and at age one) really limits children’s cognitive development at later ages. The data
obtained from the surveys revealed that about one-third of sample children were stunted by
the age of 5 and to look at the potential effect of this childhood stunting on cognitive
performance of the children, we used both simple statistics (multivariate mean test) and
Propensity Score Matching (PSM) model.

The multivariate mean test revealed a marked difference in scores between the stunted and
non-stunted children at both the ages of five and eight years. For example, at the age of 8 the
average PPVT score was 72.88 for stunted children and 81.29 for their non-stunted
counterparts, which is more than 12 points of difference. As in the PPVT test scores, the
quantitative assessment score was also much higher for non-stunted children with 8.60
percentage points of difference (in favor of non-stunted children) at the age of 8, all
Author Manuscript

differences statistically significant at P<0.01. These preliminary results show that stunting
has noticeable negative effect on cognitive development of children at the age of five and
eight years old in Ethiopia.

However, to infer causality we employed different propensity score matching techniques:

kernel analysis, radius and one to one matching controlling for confounders. We found
consistent results across all these estimating techniques (with some difference in magnitude
of parameter estimates) that reinforced our hypothesis that childhood nutrition has
significantly and positively impact on children’s cognitive development as measured by two
separate achievement test scores, namely Picture Peabody Vocabulary test (PPVT) score and
cognitive development assessment-quantitative test (CDA-Q) or Mathematics test. Our
propensity score matching results from the Kernel analysis particularly shows that stunted
children scored 16.1% less in PPVT test and 48.8% less in Mathematics test at the age of
Author Manuscript

eight, confirming these preliminary results from the simple statistics test. The estimation
results also are robust to alternative estimates of matching techniques, providing further
evidence, consistent with the literature (2–10), that well-nourished children are found to
perform better while malnourished children are in cognitive disadvantage.

Ethiop J Health Dev. Author manuscript; available in PMC 2018 January 01.
 Woldehanna et al. Page 11

Strength and Limitation

Author Manuscript

There is one caveat on the representativeness of the sample children, however. As the aim of
the Young Lives longitudinal survey has been mainly to follow the lives of children and to
track the impact of poverty on their overall development over the 15 years of time and
accordingly to inform policy makers better about the reality of childhood poverty, there was
some selection bias towards poor households from the beginning of the survey in 2002. If
this assumption of oversampling of poor households holds true, the data does not need to be
representative for the whole Ethiopia and our interpretation of the results is under this
caution. Nevertheless, in spite of this selection bias towards poor family of the children, the
longitudinal survey design is an appropriate and valuable instrument in analyzing causal
relations and modeling child welfare in the country, where it is hard to find any kind of such
longitudinal dataset (40).
Author Manuscript

In conclusion, the result of the study show those children that were nutritionally
disadvantaged at age five (and at age one) are found to perform relatively poorly on
cognitive performance and schooling at a later age. The policy implications that we can draw
from our empirical exercise are the importance of early investment in terms of child health
and nutrition at around the age of one and five years on cognitive performance of children.
Since providing balanced diet and health services are the key for children’s nutritional
achievements, efforts (and perhaps increased resources) should be devoted to improving
prenatal and postnatal care, mothers’ (parental) education, and other related environmental
factors. As baseline household wealth and parental education also are found to play an
important role in children’s nutritional achievements, policy measures that are directed in
improving household’s livelihood may have a spill-over impact in improving child health,
and consequently cognitive development and reducing school drop-out.
Author Manuscript

Acknowledgments
This study is based on research funded by the Bill & Melinda Gates Foundation (Global Health Grant
OPP1032713) and Eunice Shriver Kennedy National Institute of Child Health and Development (Grant R01
HD070993). The data used in this study comes from Young Lives, a 15-year survey investigating the changing
nature of childhood poverty in Ethiopia, India (Andhra Pradesh), Peru and Vietnam (www.younglives.org.uk).
Young Lives is core-funded by UK aid from the Department for International Development (DFID) and co-funded
from 2010 to 2014 by the Netherlands Ministry of Foreign Affairs. The findings and conclusions in this study are of
the authors and do not necessarily reflect positions or policies of the Bill & Melinda Gates Foundation, the Eunice
Shriver Kennedy National Institute of Child Health and Development, Young Lives, DFID or other funders.

References
1. Alderman H, Behrman JR, Lavy V, Menon R. Child health and school enrollment: A longitudinal
analysis. Journal of Human resources. 2001; 1:185–205.
Author Manuscript

2. Glewwe P, Jacoby HG, King EM. Early childhood nutrition and academic achievement: a
longitudinal analysis. Journal of Public Economics. 2001; 81(3):345–68.
3. Victora CG, Adair L, Fall C, Hallal PC, Martorell R, Richter L, Sachdev HS. Maternal and Child
Under nutrition Study Group. Maternal and child under nutrition: consequences for adult health and
human capital. The lancet. 2008; 371(9609):340–57.
4. UNESCO. The World Conference on Early Childhood Care and Education (ECCE): Building the
Wealth of Nations”; 2010/ED/BAS/ECCE/ME/1 REV.2
5. Heckman JJ, Masterov DV. The productivity argument for investing in young children. National
Bureau of Economic Research. 2007

Ethiop J Health Dev. Author manuscript; available in PMC 2018 January 01.
 Woldehanna et al. Page 12

6. Garcia, MH.Pence, A., Evans, J., editors. Africa’s future, Africa’s challenge: early childhood care
and development in Sub-Saharan Africa. World Bank Publications; 2008. p. 18
Author Manuscript

7. Shore, R. Families and Work Institute. New York: 1997. Re-thinking the Brain: New Insights into
Early Development.
8. Sternberg RJ. Beyond IQ: A triarchic theory of human intelligence. CUP Archive. 1985
9. Pollitt E, Gorman KS, Engle PL, Martorell R, Rivera J, Wachs TD, Scrimshaw NS. Early
supplementary feeding and cognition: effects over two decades. Monographs of the Society for
Research in Child Development. 1993; 1:115–8.
10. Maluccio JA, Hoddinott J, Behrman JR, Martorell R, Quisumbing AR, Stein AD. The impact of
improving nutrition during early childhood on education among Guatemalan adults. The Economic
Journal. 2009; 119(537):734–63.
11. AgRIDIET. Improving Nutrition through Agriculture in Ethiopia: An AgriDiet Policy Analysis.
University College of Coke; Ireland: 2015.
12. Li H, Barnhart HX, Stein AD, Martorell R. Effects of early childhood supplementation on the
educational achievement of women. Pediatrics. 2003; 112(5):1156–62. [PubMed: 14595062]
13. Hoddinott J, Maluccio JA, Behrman JR, Flores R, Martorell R. Effect of a nutrition intervention
Author Manuscript

during early childhood on economic productivity in Guatemalan adults. The lancet. 2008;
371(9610):411–6.
14. Behrman JR, Calderon MC, Preston SH, Hoddinott J, Martorell R, Stein AD. Nutritional
supplementation in girls influences the growth of their children: prospective study in Guatemala.
The American journal of clinical nutrition. 2009; 90(5):1372–9. [PubMed: 19793851]
15. Grantham-McGregor SM, Walker SP, Chang SM, Powell CA. Effects of early childhood
supplementation with and without stimulation on later development in stunted Jamaican children.
The American journal of clinical nutrition. 1997; 66(2):247–53. [PubMed: 9250101]
16. Grantham-McGregor SM, Powell CA, Walker SP, Himes JH. Nutritional supplementation,
psychosocial stimulation, and mental development of stunted children: the Jamaican Study. The
Lancet. 1991; 338(8758):1–5.
17. Gertler, P., Susan Walker, U. Labor market returns to very early childhood stimulation. Santiago,
Chile: 2011.
18. Bobonis GJ, Miguel E, Sharma CP. Iron deficiency anemia and school participation. Poverty action
Author Manuscript

lab paper. 2004; 7:1118–20.

19. Jukes M. The long-term impact of preschool health and nutrition on education. Food and Nutrition
Bulletin. 2005; 26(2S)
20. Hall A, Khanh LN, Son TH, Dung NQ, Lansdown RG, Dar DT, Hanh NT, Moestue H, Khoi HH,
Bundy DA. An association between chronic undernutrition and educational test scores in
Vietnamese children. European journal of clinical nutrition. 2001; 55(9):801–4. [PubMed:
11528497]
21. Glewwe P, Miguel EA. The impact of child health and nutrition on education in less developed
countries. Handbook of development Economics. 2007; 4:3561–606.
22. Alderman H, Hoddinott J, Kinsey B. Long term consequences of early childhood malnutrition.
Oxford economic papers. 2006; 58(3):450–74.
23. Alderman H, Hoogeveen H, Rossi M. Preschool nutrition and subsequent schooling attainment:
longitudinal evidence from Tanzania. Economic Development and Cultural Change. 2009; 57(2):
239–60.
24. Behrman JR, Lavy V. Child health and schooling achievement: Causality, association and
Author Manuscript

household allocations.
25. Mendez MA, Adair LS. Severity and timing of stunting in the first two years of life affect
performance on cognitive tests in late childhood. The Journal of nutrition. 1999; 129(8):1555–62.
[PubMed: 10419990]
26. Maluccio JA, Hoddinott J, Behrman JR, Martorell R, Quisumbing AR, Stein AD. The impact of
improving nutrition during early childhood on education among Guatemalan adults. The Economic
Journal. 2009; 119(537):734–63.
27. Abidoye RO, Eze DI. Comparative school performance through better health and nutrition in
Nsukka, Enugu, Nigeria. Nutrition Research. 2000; 20(5):609–20.

Ethiop J Health Dev. Author manuscript; available in PMC 2018 January 01.
 Woldehanna et al. Page 13

28. Dobbing J. Infant nutrition and later achievement. Nutrition Reviews. 1984; 42(1):1–7. [PubMed:
6199699]
Author Manuscript

29. Dobbing, J. The biology of human fetal growth. London: Taylor and Francis; 1976. Vulnerable
periods in brain growth and somatic growth; p. 137-47.
30. Glewwe P, King EM. The impact of early childhood nutritional status on cognitive development:
does the timing of malnutrition matter? The World Bank Economic Review. 2001; 15(1):81–113.
31. Central Statistics Agency (CSA). Ethiopia Demographic and Health Survey 2011. Calverton, MA:
ICF International and Central Statistical Agency of Ethiopia; 2012.
32. Young Lives. Young Lives Survey Design and Sampling in Ethiopia. Preliminary Findings from the
2013 Young Lives Survey (Round 4) Oxford: Young Lives. 2014
33. Woldehanna T. Inequality, preschool education and cognitive development in Ethiopia: Implication
for public investment in pre-primary education. International Journal of Behavioral Development.
2016; 40(6):509–16.
34. Woldehanna, T., Araya, M. Educational Inequalities among Children and Young People in
Ethiopia. Country Report, Oxford: The Young Lives; 2016.
35. De Onis M. Assessment of differences in linear growth among populations in the WHO
Author Manuscript

Multicentre Growth Reference Study. Acta Paediatrica. 2006; 95:56–65. [PubMed: 16373297]
36. De Onis M, Blössner M, Borghi E. Prevalence and trends of stunting among pre-school children,
1990–2020. Public health nutrition. 2012; 15(1):142–8. [PubMed: 21752311]
37. De Onis M, Onyango AW, Borghi E, Siyam A, Nishida C, Siekmann J. Development of a WHO
growth reference for school-aged children and adolescents. Bulletin of the World health
Organization. 2007; 85(9):660–7. [PubMed: 18026621]
38. Dunn LM, Dunn LM. Examiner’s manual for the PPVT-III Peabody picture vocabulary test: Form
IIIA and Form IIIB. AGS. 1997
39. Cueto S, Leon J, Guerrero G, Muñoz I. Psychometric characteristics of cognitive development and
achievement instruments in Round 2 of Young Lives. Young Lives. 2009
40. Outes-Leon I, Sanchez A. An assessment of the Young Lives sampling approach in Ethiopia.
Young Lives Technical Note. 2008; 1:1–37.
41. Blundell R, Dias MC. Alternative approaches to evaluation in empirical microeconomics. Journal
of Human Resources. 2009; 44(3):565–640.
Author Manuscript

42. Leuven E, Sianesi B. PSMATCH2: Stata module to perform full Mahalanobis and propensity score
matching, common support graphing, and covariate imbalance testing. Statistical Software
Components. 2015
Author Manuscript

Ethiop J Health Dev. Author manuscript; available in PMC 2018 January 01.
 Woldehanna et al. Page 14
Author Manuscript
Author Manuscript

Figure 1.
Kernel density distribution of propensity score before and after matching of the stunted and
non-stunted children
Author Manuscript
Author Manuscript

Ethiop J Health Dev. Author manuscript; available in PMC 2018 January 01.
 Woldehanna et al. Page 15

Table 1

Sample size by Round and preschool experience of urban and rural sample
Author Manuscript

Residence Round 1 Round 2 Round 3 Preschool

area (2002) (2006) (2009) Attendance
Age 1 Age 5 Age 8
Urban area 700 762(+) 745 56.91%

Rural Area 1299 1150 1139 3.33%

Total 1999 1912 1884 24.53 %

Source: Woldehanna (33), 2016

Note: (+) shows an increase in the urban sample in Round 2 as the result of mobility from rural to urban areas
Author Manuscript
Author Manuscript
Author Manuscript

Ethiop J Health Dev. Author manuscript; available in PMC 2018 January 01.
 Woldehanna et al. Page 16

Table 2

Basic characteristics of the sample children and their families (1883)

Author Manuscript

Std.
Mean Dev. Min Max
−
Z-score of height-for-age at age one 1.53 1.86 −5.98 5.91
Dummy for stunted child at age 1 (zhfa<-2) 0.42 0.49 0 1
Dummy for severely stunted child at age 1 (zhfa<-3) 0.22 0.41 0 1
Z-score of height-for-age at age 5 −1.45 1.12 − 4.59
Dummy for stunted child at age 5 (zhfa<-2) 0.31 0.46 0 5.57 1
Dummy for severely stunted child at age 5 (zhfa<-3) 0.08 0.27 0 1
# of antenatal visits by mother during pregnancy 2.05 2.62 0 16
Mother’s age at birth (years) 27.48 6.39 15 55
Dummy for wanted pregnancy 0.62 0.48 0 1
Author Manuscript

Dummy if child was born before expected 0.10 0.30 0 1

Mother has a permanent health problem 0.06 0.24 0 1
Dummy for child had health problems at age 1 0.47 0.50 0 1
Likert scale for the relative size of child at birth (from −2 to 2) 0.00 1.02 −2 2
# of months mother breastfeed the child 32.32 10.35 0 36
Dummy for severe illness or injury 0.13 0.34 0 1
Age of the child in months 2006 62.32 3.83 52.7 75.36
Household wealth index at age 1 ( 2002) 0.21 0.17 0.00 0.74
Household wealth index at age 5 ( 2006) 0.28 0.18 0.01 0.87
Household wealth index at age 8 ( 2009) 0.33 0.18 0.01 0.86
Household size at age 1 5.75 2.15 2 16
Per capita annual consumption expenditure (Birr in 100) 14.79 11.54 97.08 106.62
Author Manuscript

Highest grade completed by primary caregiver (in years) 2.76 3.66 0 14

Highest grade completed by father ( in years) 4.40 4.19 0 14
Dummy for parental divorce or separation by age 1 0.05 0.21 0 1
Dummy for male 0.53 0.50 0 1
Dummy for urban site 0.40 0.49 0 1
Dummy if household received support from NGOs or 0.66 0.47 0 1
Dummy variable for child being enrolled in preschool 0.25 0.43 0 1

Source: Own computation

Author Manuscript

Ethiop J Health Dev. Author manuscript; available in PMC 2018 January 01.
 Woldehanna et al. Page 17

Table 3

Cognitive achievement and school enrolment and progression

Author Manuscript

All stunted non-stunted

children at age 5 at age 5
(N=1883) (n=587) (n=1296)

Mean
Mean Mean Mean Difference(a)
Standardized core of PPVT test at age 5 (out of 204) 67.54 63.25 69.44 −6.19***
Standardized core of PPVT test at age 8 ( out of 204) 78.51 69.99 82.36 −12.37***
% Standardized core of PPVT test at age 5 33.11 31.00 34.04 −3.04***
% Standardized core of PPVT test at age 8 38.48 34.31 40.37 −6.06***
# of correctly answered Q-CDA test at age 5 (out of 15) 8.21 7.74 8.43 −0.69***
# of correctly answered Q-CDA test at age 8 (out of 29) 6.34 4.62 7.12 −2.49***
Author Manuscript

% of math questions correctly answered at age 5 54.76 51.60 56.18 −4.58***

% of math questions correctly answered at age 8 21.87 15.94 24.55 −8.60***
Dummy variable for a child begun formal school at 7 0.77 0.64 0.84 −0.20***
Grade completed at the age 8 0.79 0.59 0.88 −0.30***

Note:
*
p < 0.1;
**
p < 0.05;
***
p < 0.01
a
Test that all means are the same: Hotelling T2 = 35074.2; Prob > F = 0.0000
Author Manuscript
Author Manuscript

Ethiop J Health Dev. Author manuscript; available in PMC 2018 January 01.
 Author Manuscript Author Manuscript Author Manuscript Author Manuscript

Table 4

ATT estimates of Propensity Score Matching (PSM)

Kernel matching Radius matching

Variables a ATTk S.E. T-stat ATT S.E. T-stat

Woldehanna et al.

Logarithm of standardized PPVT score at age 5 −0.161 0.027 −5.9*** −0.165 0.026 −6.38***
Logarithm of standardized PPVT score at age 8 −0.161 0.021 −7.51*** −0.166 0.019 −8.78***
Logarithm of % of CDA-A score at age 5 −0.072 0.023 −3.16*** −0.077 0.022 −3.42***
Logarithm of % of math score at age 8 −0.448 0.046 −9.71*** −0.452 0.041 −10.93***
Probability of being enrolled in school at 7 −0.178 0.023 −7.7*** −0.177 0.023 −7.62***
Level of grade achieved by a child at 8 −0.281 0.035 −8.12*** −0.280 0.035 −7.96***

note:
***
p<0.01,
**
p<0.05,
*
p<0.1

(T>=1.65=*; T>1.96=***; T>2.6=***); N=1,833; ATT=Average treatment effect on the treated. first stage logit model used to predict propensity scores of stunting at age 5 that are used for matching the
children and to estimate Average Treatment effects of the Treatment (stunting) on the treated (stunted) was controlled for being stunted at age one year ( that might capture any carry-over effect), number of
months breastfeeding, relative size at birth, health problems at early years such as respiratory problems and malaria incident, sex of the child, baseline household wealth index, household size and
composition and parental/primary caregivers’ educational level in years ( readers interested in this issue can contact the authors at the corresponding email address).

Ethiop J Health Dev. Author manuscript; available in PMC 2018 January 01.
Page 18