1 Introduction
India contains a bulk of global infant
undernutrition with around 46 per cent of
children under three ‘stunted’ in 1998/99 (the
latest year for which nationally representative
data is available, Gragnolati et al. 2008).
1
The
adverse impact of early age stunting on various
outcomes later on in terms of health,
productivity, cognitive development and
schooling achievement have been well
established in the literature (Grantham-
McGregor et al. 1997, 1999; Dercon and Sanchez
2008). The focus of most papers on stunting,
however, is the under five age group. This is not
surprising, as stunting is argued to occur within
the first few years of life (Martorell et al. 1994).
Children enter middle childhood (roughly
between the ages of 7–12), with nutritional
deficits accrued earlier on and are likely to follow
the same growth path, continuing to be ‘stunted’.
However, although middle childhood is a period
where children experience a slower phase of
growth (Tanner 1990), there is evidence that an
already stunted child may ‘catch up’ given a
suitable environment, adoption, emigration or
treatment of disease that may have retarded
growth (Golden 1993; Tanner 1986). Quite in
contrast, children may also falter in terms of
linear growth, increasing the prevalence of
stunting with age. This is emerging as a key issue
in recent data available for sub-Saharan Africa
(Friedman et al. 2005; Lwambo et al. 2000;
Monyeki et al. 2000, Stoltzfus et al. 1997).
This article attempts to add to this recent strand
in the literature by looking at how
undernutrition seems to have persisted in middle
childhood in Andhra Pradesh, India, using
longitudinal data. It looks particularly at what
observable factors contributed towards the
persistence of stunting (i.e. following a ‘stable’
growth path of continuing to be stunted) and in
contrast what helped children move out of or
into being stunted between the ages of 7 and 12.
The article offers a unique contribution to the
literature in this area, as there is no other paper
in the empirical literature, as far as the author is
aware, that looks at this issue for India using
longitudinal data. The article is also unique in
that it looks at the lives of persistently stunted
children using a rich dataset that provides
information on several conventional and
unconventional aspects: what types of households
they are from, what their typical diet contains,
30
Persistent Stunting in Middle
Childhood: The Case of Andhra
Pradesh Using Longitudinal Data
Rozana Himaz
*
Abstract This article looks at what observable characteristics influence a child being persistently stunted,
moving from being stunted or moving into being stunted in middle childhood, between 7 and 12, using
longitudinal data for Andhra Pradesh. It finds the key factors that help a child move out of being stunted are
mother’s education and coming from the more prosperous region of Coastal Andhra. In contrast, the key
factors that pushed a child into being stunted were the child being a girl and being a younger sibling. We
also find that children who moved out of being stunted consume a diet higher in protein and micronutrients
than others. The article suggests that even if a child starts middle childhood with significant shortfalls in
height accrued from earlier on in life, nutritional interventions and adult female education may have a
positive impact on linear growth and perhaps mitigate consequences of early age stunting.
IDS Bulletin Volume 40 Number 4 July 2009 © 2009 The Author. Journal compilation © Institute of Development Studies
Published by Blackwell Publishing Ltd, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA
how they perceive their own lives (subjective
wellbeing) and their psychosocial outcomes (such
as self-esteem, self-efficacy and sense of
inclusion).
2 Data
The data for this article comes from a unique
panel survey of 1,000 children (one child per
household, referred to as the ‘index child’ in this
article), their communities and households
conducted in Andhra Pradesh, India in 2002 and
2006, drawn from a pro-poor sample managed by
the Young Lives project (University of Oxford).
The survey was carried out across 20 sentinel
sites in three regions.
2
The children were aged
seven to eight years of age in 2002 (Round 1) and
11–12 at Round 2.
The article looks at three groups of children:
(1) those who were stunted (i.e. have height for
age z-scores of below 2 standard deviations of the
mean) in Round 1 and continued to be stunted
four years later in Round 2. In Round 1, 31 per
cent of the children were stunted and over 70 per
IDS Bulletin Volume 40 Number 4 July 2009
31
Table 1 Child and household characteristics
Not stunted in Stunted in Stunted in Stunted in
either round both rounds Round 1 but Round 2 but not
(‘persistent’) not Round 2 Round 1 (‘moved
(‘moved out’) in’)
Child characteristics
Height for age z-score (Round 1) –0.95 –2.78** –2.42** –1.50*
Height for age z-score (Round 2) –0.96 –2.61** –1.49* –2.35**
Body mass index (Round 2) –1.25 –1.87** –1.65* –1.44*
Boys (%) 0.49 0.52 0.56 0.31**
Caregiver characteristics
Main caregiver is not mother or father (%) 3 5* 1 4
Father’s education (years) 4.98 3.70** 5.25 3.66**
Mother’s education (years) 3.18 1.54** 3.28 1.92**
Caste (%)
SC 20 21 14 17
ST 8 12** 09 15**
BC 46 51 54 50
OC 23 13** 20 15*
Wealth, expenditure, vulnerability
Wealth index (Round 1) 0.39 0.28*** 0.35* 0.33**
Wealth index (Round 2) 0.32 0.34* 0.32 0.30
Description of household compared to About average Poorer than About average Poorer than
others (Round 2) average average
Adjusted per capita expenditure (Rupees) 720 723 675 727
Households that had a food shortage in the 8 15** 9 7
12 months before Round 2 (%)
Suffered a serious environmental hazard 56 79* 61 66*
such as a flood, drought, pest invasion, fire,
collapse of building, crop failure between
rounds (%)
In serious debt at Round 2 (%) 49 56* 47 52
Budget shares of expenditure
Alcohol and tobacco share (%) 1 2* 1 1.5
Number of observations 610 224 82 92
Source Young Lives, India (Andhra Pradesh) Older cohort data, Round 1 and Round 2 for 2002 and 2006.
cent of them continued to be stunted in Round 2
(i.e. 224 children). While a majority of children
remained ‘stunted’, a sizeable proportion (30 per
cent) had moved out of being stunted by Round 2.
These 82 children form the second group we
examine. The final group is one of 92 children
(nearly 10 per cent of the sample) who were not
stunted in Round 1 but were stunted by Round 2.
3
The analysis uses descriptive statistics,
multivariate regressions and results from other
papers using Young Lives data to glean insights as
to what observables may explain such different
growth trajectories for stunted children in
middle childhood.
3 Analyses
3.1 Descriptive statistics
Table 1 shows child and household
characteristics for the three groups of children
compared to those children who were never
below 2 standard deviations of the mean in terms
of height for age z-scores. One of the first things
to notice is that the height for age of all three
groups (columns 2, 3 and 4) are significantly
lower than that of the comparison group of
children (column 1) who were never stunted in
our data. This reflects that even if the children
moved in/out of being stunted in middle
childhood, they remain a group that are
significantly worse off in terms of their linear
growth compared to those never stunted by the
time they entered middle childhood.
The parents of persistently stunted children and
those who move into being stunted are
significantly less educated. Quite interestingly,
the education levels of parents of children who
moved out of being stunted are not significantly
different to that of the never stunted group.
Similar results are found for statistics regarding
caste: persistent stunting and moving into being
stunted in middle childhood is significantly more
prevalent among Scheduled Tribes (ST) and less
so in other castes (OC). However, the caste
distribution of children who move out of being
stunted is not significantly different to that of
the comparison group in column 1 of children
never stunted.
Similarly, households with a child who is
persistently stunted or moved into stunting by
Round 2 consider themselves poorer than
average in Round 2. But households with a never
stunted child or those with a child that moved
out of being stunted consider themselves ‘about
average’. This perception is not reflected in the
wealth index figure for Round 2 or the per capita
expenditure figures.
4
Although the per capita expenditure figures are
not significantly different among the groups,
households with a persistently stunted child or a
child that moved into being stunted by Round 2
seem to have been subject to significantly more
adverse shocks between rounds. For example,
more households with a stunted child report to
have had a food shortage within the past 12
months (15 per cent vs. 8 per cent); a
significantly higher proportion are in serious
debt compared with households with a child who
is not stunted (56 per cent vs. 48 per cent) and a
significant proportion (79 per cent vs. 56 per
cent) had suffered from a serious environmental
hazard such as drought, flood or crop failure.
Moreover, although per capita expenditure figures
for Round 2 do not show a significant difference
between the households, per capita incomes are
significantly different between these households
in Round 2 to the detriment of households with a
stunted child at Rs. 439.70 vs. Rs. 613.50. The
gap in expenditure and income may be explained
by the fact that a higher proportion of
households with a stunted child claim to benefit
from charity, food aid, debts
5
and access to
government programmes,
6
that probably help
them consume much more than they claim to
earn.
7
Households with a persistently stunted child also
spend significantly more on alcohol and tobacco
consumption (2 per cent vs. 1 per cent). Neufeld
et al. (2005) show, using India’s National Sample
Survey for 1995–6, that higher alcohol and
tobacco consumption is correlated strongly with
poorer households from rural areas as well as
those from scheduled castes and tribes. Prasad
(2009) argues that alcohol consumption is rising
in India with more than half of all drinkers
falling into the criteria for hazardous drinking,
which is characterised by bingeing and solitary
consumption to the point of intoxication. It
remains to be explored further how much
household alcohol consumption has an impact on
child health outcomes and whether this issue has
to be a direct focus of policy.
Himaz Persistent Stunting in Middle Childhood: The Case of Andhra Pradesh Using Longitudinal Data
32
Table 2 shows that the mothers of stunted
children are also significantly shorter and lighter
than the mother’s of non-stunted children. While
this is partly indicative that genetic factors may
be driving child height, it is also suggestive of the
fact that stunting and poor nourishment is likely
to be a problem transmitted through generations
with poorly fed anaemic mothers giving birth to
underweight children who are also likely to be
malnourished. Stunted children are also more
likely to have long-term health problems than
non-stunted children (10 per cent vs. 5 per cent)
and consider their health to be significantly
worse than that of their peers (unreported).
The stunted children eat a less diverse diet than
the non-stunted children, with a significantly
lower consumption of fruits, roots, tubers and
sugary food. Quite notably, a significantly larger
proportion of children who moved out of being
stunted had a diet containing legumes (such as
lentils and beans that are high in protein) and
fish or seafood. A significantly lower proportion
had cereals and sugary food. The data on diet
pertains to consumption in the 24 hours before
the survey, but indicates that children who
moved out of stunting may have consumed diets
higher in protein than the other groups.
The children who were stunted in Round 1 were
reported to be more often unhappy, tearful and
downhearted than non-stunted children. They
also felt they were less liked by other children
than did their non-stunted counterparts. These
poorer outcomes have persisted over the four
years up to the time of the second round of data
collection, with these children scoring lower in
terms of subjective wellbeing (perception of how
good life is, based on a ladder of 1–9, with 1
denoting the worst possible outcome and 9 the
best), sense of optimism about the future
(reflected in a lower ladder rung four years from
now), low self-esteem (an opinion about oneself),
and sense of inclusion. Even the caregivers of the
persistently stunted children seem to be suffering
from a low sense of inclusion compared with the
caregivers of children who were never stunted.
8
Dercon and Sanchez (2008) find clear evidence
that stunting early on in childhood has serious
adverse effects on a child’s cognitive and
psychosocial abilities later on at age 12, in all
four Young Lives countries: Ethiopia, Peru,
Vietnam and India.
3.2 Regression analysis
Table 3, column 1 shows that the key observable
characteristics at Round 1 that exerted a positive
significant influence on a child’s height for age
z-score in Round 2 was the child’s gender being
male, the mother’s height (capturing partly
genetic factors as well as the mother’s
accumulated investments in health that have an
impact on child health, especially prenatally),
the mother’s education and household wealth.
The significant negative influences come from
belonging to a scheduled tribe or backward caste
compared with coming from ‘other’ castes and
rural residence. Column 2 looks at round two
heights as explained by Round 2 characteristics.
The results are the same as in the previous
column except that household wealth (now
proxied by expenditure per capita) ceases to be
significant. This makes sense in the context that
short-term household expenditure is unlikely to
have an impact on a child’s height which reflects
long term investments in health.
Column 3 looks at what might explain a child
remaining stunted through middle childhood. The
significant negative influences are the mother’s
height and education. The significant positive
influences are coming from a scheduled caste or
backward caste (compared to coming from other
castes) and coming from Rayalaseema, a
relatively poor area compared to Telangana.
Column 4 looks at what factors helped a child
move out of being stunted compared with the
group of children who remain stunted in both
rounds. The key contributing factors are the
mother’s education and coming from Coastal
Andhra – a state relatively more affluent than
the omitted group Telangana. The main factors
that push a child to being stunted between
rounds compared with the group that were never
stunted are being a girl and having a higher
birth order (column 5).
Note that the regression results are not different
in terms of sign and significance if we re-run it
separately for boys and girls, apart from for
column 1, when household size and birth order
become important for determining a girl’s height
for age.
The regressions here do not account for the fact
that within the different groups of children
linear growth between rounds may have been
IDS Bulletin Volume 40 Number 4 July 2009
33
driven due to different reasons and that growth
may have occurred at different speeds. To
account for this, similar regressions are run with
the difference in height for age between rounds
as the dependent variable explained by the same
Round 1 variables. In this case, the only variables
Himaz Persistent Stunting in Middle Childhood: The Case of Andhra Pradesh Using Longitudinal Data
34
Table 2 Health and subjective wellbeing
Not stunted Stunted in Stunted in Stunted in
in either both rounds Round 1 but Round 2 but not
round (‘persistent’) not Round 2 Round 1 (‘moved
(‘moved out’) in’)
Health
Mother’s weight (kg) 50.43 46.75* 47.3* 46.02
Mother’s height (cm) 151.22 148.92* 150.64 149.92
Child has long-term health problem (%) 5 10** 7 5
Dietary diversity (consumption from 5.8 5.3** 5.8 5.6
different food groups, range: 0–11)
Proportion that consume the following
food groups in the last 24 hours before
the survey:
Cereals (maize, rice, bread) 0.99 0.99 0.97* 1
Roots/tubers (potato, cassava) 0.37 0.27** 0.43 0.35
Legumes (lentils, beans, peas) 0.42 0.42 0.56** 0.38
Milk, milk products 0.66 0.62 0.60 0.67
Eggs 0.16 0.15 0.19 0.10
Meat/offal 0.08 0.07 0.13 0.10
Fish/seafood 0.04 0.04 0.12** 0.05
Oil/fat 0.95 0.95 0.96 0.97
Sugar/honey 0.77 0.70** 0.66** 0.76
Fruits 0.39 0.28** 0.36 0.37
Vegetables 0.95 0.96 0.95 0.91*
Subjective wellbeing and psychosocial outcomes
Round 1:
Often unhappy, downhearted or 22 34** 28 30
tearful (%)
Generally liked by other children (%) 95 91** 95 93
Easily distracted, concentration 30 37 32 29
wanders (%)
Round 2:
Ladder (scale 1–9: 1, worst; 9, best) 3.73 3.01** 3.58 3.39
Ladder four years from now 5.12 4.01* 4.85 4.68*
Child self-esteem (based on a 0.01 –0.04* –0.01 0.03
standardised index with mean zero and
variance 1)
Child self-efficacy –0.04 0.01 –0.05 0.06
Child sense of inclusion –0.00 –0.07** 0.06 0.05
Caregiver’s sense of inclusion 0.02 –0.05* 0.04 0.00
Number of observations 600 224 82 92
*Significant at the 10 per cent level **Significant at the 5 per cent level or higher
Note The PPVT score is based on the Peabody picture vocabulary test administered to children in the Young Lives
sample, along with a maths test. See Ceuto
et al.
(2009) for more details on the tests.
Source Young Lives, India (Andhra Pradesh) Older cohort data, Round 1 and Round 2 for 2002 and 2006.