Health Policy and Planning, 31, 2016, 582–591
doi: 10.1093/heapol/czv105
Advance Access Publication Date: 13 November 2015
Original article
Overweight and obesity in India: policy issues
from an exploratory multi-level analysis
Md Zakaria Siddiqui1 and Ronald Donato2,*
1
Crawford School of Public Policy, Australian National University, Canberra, Australia; and Institute of
Development Studies Kolkata, Kolkata, India, 2UniSA Business School, University of South Australia, Way Lee
Building, City West Campus, 37-44 North Terrace, Adelaide, 5001, South Australia
*Corresponding author. UniSA Business School, University of South Australia, Way Lee Building, City West Campus,
37-44 North Terrace, Adelaide, 5001, South Australia. E-mail: [email protected]
Accepted on 2 October 2015
Abstract
This article analyses a nationally representative household dataset—the National Family Health
Survey (NFHS-3) conducted in 2005 to 2006—to examine factors influencing the prevalence of overweight/obesity in India. The dataset was disaggregated into four sub-population groups—urban and
rural females and males—and multi-level logit regression models were used to estimate the impact
of particular covariates on the likelihood of overweight/obesity. The multi-level modelling approach
aimed to identify individual and macro-level contextual factors influencing this health outcome. In
contrast to most studies on low-income developing countries, the findings reveal that education for
females beyond a particular level of educational attainment exhibits a negative relationship with the
likelihood of overweight/obesity. This relationship was not observed for males. Muslim females and
all Sikh sub-populations have a higher likelihood of overweight/obesity suggesting the importance of
socio-cultural influences. The results also show that the relationship between wealth and the probability of overweight/obesity is stronger for males than females highlighting the differential impact of
increasing socio-economic status on gender. Multi-level analysis reveals that states exerted an independent influence on the likelihood of overweight/obesity beyond individual-level covariates, reflecting the importance of spatially related contextual factors on overweight/obesity. While this study
does not disentangle macro-level ‘obesogenic’ environmental factors from socio-cultural network influences, the results highlight the need to refrain from adopting a ‘one size fits all’ policy approach in
addressing the overweight/obesity epidemic facing India. Instead, policy implementation requires a
more nuanced and targeted approach to incorporate the growing recognition of socio-cultural and
spatial contextual factors impacting on healthy behaviours.
Key words: India; overweight-obesity; multi-level; policy issues
Key Messages
•
•
•
•
Increased years of education for women beyond a certain level exhibit a negative relationship with the likelihood of
overweight-obesity. This was not observed for men.
Gender and rurality matter in how covariates influence the likelihood of overweight-obesity highlighting the need to discern across sub-populations when effecting policy.
States exerted an independent influence on the likelihood of overweight-obesity reflecting the significance of sociocultural and spatially-related contextual effects.
Formulating more targeted and nuanced health interventions, rather than adopting a one-size-fits-all approach, is of key
importance if policy strategies are to be effective.
C The Author 2015. Published by Oxford University Press. All rights reserved. For permissions, please email: [email protected]
V
582
Health Policy and Planning, 2016, Vol. 31, No. 5
Introduction
The dramatic increase in the prevalence of overweight/obesity over
the past two decades and its consequent impact on the burden of
non-communicable diseases (NCD) such as diabetes and cardiovascular disease is recognized as a global pandemic (Swinburn et al.
2011; Popkin et al. 2012). India is not immune from this development; recent estimates reveal that >12% of the entire population is
overweight or obese and India, after China, has the highest number
of people with diabetes in the world. Rapid economic development
has seen dietary patterns shift towards energy-dense foods which together with increased urbanization and an associated sedentary lifestyle have contributed to the rise in overweight/obesity (Khandelwal
and Reddy 2013; Misra and Shrivastava 2013). Given India’s limited resources and the strains on its health system from having the
greatest number of malnourished people in the world, rising overweight/obesity with the causal link with NCD represents a major
public health challenge (FAO 2014).
Much of the health economics research examining overweight/
obesity has focused on individual-level health behaviours associated
with choice and changing incentives. The decline in the relative prices
of energy-dense foods and an increase in sedentary behaviours in
work and leisure have received strong empirical support as important
determinants of the increased overweight/obesity over time (Cutler
et al. 2003; Chou et al. 2004; Lakdawalla and Philipson 2007).
However, recent studies have moved beyond individual-level behaviours and have explored non-market factors, particularly socio-cultural influences in sub-populations, as important determinants of
overweight/obesity (Christakis and Fowler 2007; Trogdon et al.
2008; Mora and Gill 2012). The significance of India’s socio-cultural
diversity and spatial patterning on the spread and social clustering of
overweight/obesity represents an important area of inquiry.
This study analyses a nationally representative household dataset—the National Family Health Survey (NFHS-3) conducted in
2005 to 2006—in examining the nature and extent of individual and
macro-level contextual factors associated with the prevalence of
overweight/obesity in India. Attention is given to analysing the significance of covariates on particular sub-populations. The study
aims to contribute to the broader literature in two major ways. First,
separating the population into four sub-groups (urban and rural and
females and males) enables a deeper analysis of the factors influencing the likelihood of overweight/obesity and its diverse patterning
across the population. Second, we explore the significance of macrolevel contextual factors as an area of enquiry in itself, rather than
simply controlling for them when analysing individual-level covariates (Subramanian and Smith 2006; Subramanian et al. 2009). As
economic growth continues to transform Indian society, understanding individual-level behaviours and the influence of spatial and
socio-cultural contexts are important in developing a more nuanced
policy approach for addressing overweight/obesity.
Data and methods
Data
The NFHS-3 dataset analysed in this study is a nationwide household survey, which provides information on health-related matters,
including fertility, morbidity and mortality, family planning and nutrition. NFHS-3 covers a representative sample of 124 384 females
aged 15–49 years and 74 369 males aged 15–54 years across 29
states in India. The primary sampling unit (PSU) comprises two
main strata—urban (wards or municipal localities) and rural (villages)—which were drawn separately within each state and are
583
proportional to the size of the respective urban and rural areas. The
primary outcome measure used for assessing overweight/obesity was
the body mass index (BMI), which is calculated as weight in kilograms divided by height in metres squared (kg/m2). Using World
Health Organization classifications, BMI cut-offs were divided into
those overweight/obese (BMI 25) and those not overweight/obese
(BMI < 25). The study analyses the factors associated with the likelihood of overweight/obesity relative to the population who are not.
The dataset was separated into four sub-populations: urban–female,
rural–female, urban–male and rural–male to enable a detailed analysis of factors associated with high BMI.
Covariates
Socio-economic status is well recognized as an important determinant of individual-level BMI. To infer economic status, NFHS-3
adopts a wealth index score based on the household ownership of
particular assets. The survey uses principal component analysis to
weight and standardize the asset scores and each individual within
the household is assigned a wealth score. Our analysis treats wealth
as a continuous variable. Another important dimension of socio-economic status is education which in this study is treated as a continuous variable specified in terms of years of education.
Other covariates include age, town/city, religion/caste, occupation, frequency of TV viewing, access to clean cooking fuel and any
form of tobacco use. Age is treated as a continuous variable: 15–49
years for females and 15–54 years for males. In the case of the urban
sub-populations, large city, small city and town were included as
separate categorical variables to discern the degree of urbanization.1
Religion and caste were combined as a categorical variable, with
Hindu divided into two sub-categories: Hindu–caste (scheduled
castes, scheduled tribe and other backward castes) and
Hindu–general (those not legally recognized as lower caste). The
other religious groupings were Muslims, Sikhs, Christians and
others. Occupation was divided into three sub-categories: not working/home, white collar and manual labour. TV viewing was included
as a ‘lifestyle’ categorical variable: the sub-categories were not at all
(TV cat-0), less than once a week (TV cat-1), at least once a week
(TV cat-2) and almost every day (TV cat-3). Another covariate was
cooking fuel availability: the sub-categories were clean fuel (kerosene, gas and electricity) and biofuel (smoke-related solid fuels such
as animal dung, wood, coal/charcoal and biomass).2 Tobacco use
(either smoking or chewing) was included as a ‘lifestyle’ categorical
variable.3 The descriptive statistics for predictor variables considered in this study (in terms of sample size and percentage of overweight/obese), tabulated across the four sub-populations are shown
in Appendix 1.
Statistical methods
A multi-level binomial logit regression model was used to estimate a
binary outcome in terms of a log likelihood ratio. The model is written as follows:
!
X
1ijk
ln
bm Xm
(1)
¼aþ
ijk þ uk þ vjk þ eijk
m
1 1ijk
•
•
•
•
1ijk —probability of overweight/obesity for individual i nested in
jth PSU nested in kth state
Xm
ijk —fixed component covariates—where superscript m represents number of covariates
uk —intercept effect of kth state
vjk —intercept effect of jth PSU nested in kth state
584
•
eijk —residual error of individual i nested in jth PSU nested in kth
state
The model shown in Equation (1) contains a fixed component which
has a separate intercept (a) and a slope parameter (bm ) estimating
the effects of a one unit change in the covariates on the log of the
odds of overweight/obesity ð1) relative to not being overweight or
obese (1 1). The remaining terms in Equation (1) capture the independent effect of geographical space as a separate explanatory
variable on the likelihood of overweight/obesity—in terms of the impact of the state, and the PSU (neighbourhood) within it, in which
the individuals reside. The advantage of the multi-level modelling
approach is that it allows intercepts to vary with respect to the state
and city/town/village to account for the unobserved macro-level
contextual factors of the state and the PSU beyond individual-level
factors. Thus, the intercept for any given state and neighbourhood
can be expressed as a þ uk and a þ uk þ vjk , respectively.
Separate regressions were generated for each sub-population
where the maximum likelihood or odds ratio (OR) of overweight/
obesity was estimated relative to the population within that group
who were not overweight or obese. Education, wealth and age were
treated as continuous variables. The other covariates were treated as
categorical variables which involved adopting a reference category
as a benchmark from which maximum likelihood or OR was estimated. The multi-level modelling approach attempts to identify the
extent to which macro-level contextual factors and individual-level
characteristics are associated with overweight/obesity. All econometric analyses were conducted using Stata 12.
Results
Table 1 presents the binomial logistic regression results for the four
sub-populations. The results are divided into a fixed component encompassing the parameter estimates of covariates and an intercept
component which incorporates spatially related multi-level effects.
The categorical parameter estimates are shown in terms of the OR
of overweight/obesity relative to a reference category within the subpopulation. All results reported below are those that are statistically
significant at either P < 0.05 or <0.001.
As expected, the results show that socio-economic status as
measured by the ‘wealth’ variable has a strong positive effect on the
likelihood of overweight/obesity across the four sub-populations.
Thus for urban females, holding all factors constant, a one unit increase in wealth increases the likelihood of overweight/obesity by a
factor of 2.09. With the exception of rural females, the ‘wealth2’
parameter is not statistically significant, suggesting that there is no
diminishing or increasing returns present with respect to incremental
changes in wealth.
Another way to interpret the wealth variable is to transform the
OR into estimated probabilities of overweight/obesity for each subpopulation. Under this transformation, all the covariates of the
model presented in Table 1 are held constant at a base level while
allowing wealth to vary in order to determine its isolated effect on
the estimated probability of overweight/obesity for a particular subpopulation. The base model chosen for each sub-population holds
constant the continuous variables of education and age at their respective mean levels and the categorical variables are fixed at the
classification which contains the largest population for that subpopulation. The category classifications of covariates held constant
for the sub-populations are presented in Table 2.
Figure 1 shows the relationship between the estimated probability of overweight/obesity and the increasing wealth levels for the
Health Policy and Planning, 2016, Vol. 31, No. 5
sub-populations. Thus, holding all factors constant at the respective
base categories specified in Table 2, urban and rural males have a
substantially higher probability of overweight/obesity than urban
and rural females for any given wealth point. For example, at the
60th wealth point,4 urban and rural males have a 48.4 and 45.1%
probability of overweight/obesity compared with a probability of
26.1 and 11.4% for urban and rural females, respectively.
Interestingly, rural males exhibit a similar probability of
overweight/obesity as urban males; this observation is not found
with females.
With regards to the ‘education’ variable, all four sub-populations
exhibit an increasing likelihood of overweight/obesity with more
years of education (Table 1). However, the results show that for
urban and rural females the ‘education2’ variable is statistically significant with a parameter estimate of <1 suggesting the likelihood
of overweight/obesity increases with respect to education at a diminishing rate.
Again, as with the wealth variable, the effect of education on
overweight/obesity can be transformed into estimated probabilities
for the sub-populations. Figure 2 shows the probability of overweight/obesity within each of the sub-populations for different years
of education. (In this case wealth and age are held constant at their
mean.) Thus, e.g. using the sub-populations based on their respective categories (Table 2), at 5 years of education, urban females have
a 26.4% probability of overweight/obesity while rural females have
a 5.6% probability. Interestingly, in the case of females there is an
inverted U-shaped relationship in the probability of overweight/
obesity with respect to education. For both urban and rural females,
increasing the initial years of education is associated with a rise in
the probability of overweight/obesity, but beyond 7–8 years of education the impact of additional years of education begins to lower
the probability of overweight/obesity. That is, beyond a certain
threshold level of educational attainment, there is a ‘negative’ association between years of education and the probability of overweight/obesity. No statistically significant negative relationship
between education and overweight/obesity was found with males.
In relation to the categorical variables in the model, the OR of
overweight/obesity is referenced against a benchmark within each
category (Table 1). Regarding the degree of urbanization, females
living in small towns have a lower likelihood of overweight/obesity
(OR ¼ 0.88) relative to those living in major cities. No statistically
significant difference was observed for males living in small towns.
There was a statistically significant positive association between the
frequency of TV viewing and the likelihood of overweight/obesity
for urban and rural females and rural males relative to those who
did not watch TV. Thus, e.g. urban females who watch TV at least
once a week (TV cat-2) or daily (TV cat-3) had 1.25 times and 1.37
times, respectively, the likelihood of overweight/obesity relative to
urban females who do not watch TV. Those who did not use tobacco had a lower likelihood of overweight/obesity relative to those
who did use tobacco; this was statistically significant for urban and
rural females (OR ¼ 0.87 and 0.77, respectively) and urban males
(OR ¼ 0.86). With regards to occupation, urban females who work
in either white collar or manual occupations have a lower likelihood
of overweight/obesity (OR ¼ 0.87 and 0.68, respectively) relative to
urban females who do not work. For rural females, only those
involved in manual work had a statistically significant lower likelihood of overweight/obesity (OR ¼ 0.67). However, in the case of
males, urban and rural white collar workers have a higher likelihood
of overweight/obesity (OR ¼ 1.36 and 1.48, respectively) relative to
urban and rural males who do not work. For urban and rural females, access to clean cooking fuel (kerosene, gas and electricity)
Health Policy and Planning, 2016, Vol. 31, No. 5
585
Table 1. Multi-level analytical model for overweight/obesity in India, 2005–06
Urban female (n ¼ 53 124)
Rural female (n ¼ 65 560)
Urban male (n ¼ 34 565)
Rural male (n ¼ 34 469)
Parameters—fixeda
OR
CIb (95%)
OR
CIb (95%)
OR
CIb (95%)
OR
CIb (95%)
Wealth
Wealth2
Education
Education2
Age
Age2
City (Ref)
Small city
Town
TV cat-0 (Ref)
TV cat-1
TV cat-2
TV cat-3
Not working (Ref)
White collar
Manual labour
Hindu–general (Ref)
Muslim
Christian
Sikh
Hindu–caste
Other
Biofuel (Ref)
Clean cooking fuel
No tobacco use (Ref)
Tobacco use
Constant
Intercept Effects
State
State/PSU
2.10**c
0.99
1.05**
0.99**
1.38**
0.99**
1.00
0.93
0.88*
1.00
1.11
1.25**
1.37**
1.00
0.87**
0.68**
1.00
1.25**
0.89
1.38*
0.84**
0.88
1.00
1.08*
1.00
0.87*
0.063**
ICCd
0.03**
0.07**
(1.72–2.57)
(0.95–1.03)
(1.03–1.07)
(0.99–0.99)
(1.35–1.41)
(0.99–0.99)
Ref
(0.84–1.03)
(0.80–0.96)
Ref
(0.97–1.27)
(1.11–1.41)
(1.24–1.53)
Ref
(0.82–0.93)
(0.63–0.74)
Ref
(1.15–1.36)
(0.79–1.02)
(1.14–1.68)
(0.78–0.89)
(0.76–1.01)
Ref
(1.00–1.16)
Ref
(0.76–0.98)
(0.043–0.093)
CIb (95%)
(0.02–0.06)
(0.05–0.09)
2.92**c
0.92**
1.06**
0.99**
1.35**
0.99**
NA
(2.45–3.48)
(0.88–0.96)
(1.04–1.08)
(0.99–0.99)
(1.32–1.39)
(0.99–0.99)
NA
(2.53–3.01)
(0.91–1.04)
(1.01–1.08)
(0.99–1.00)
(1.32–1.43)
(0.99–0.99)
NA
Ref
(1.13–1.43)
(1.22–1.53)
(1.33–1.62)
Ref
(0.93–1.16)
(0.62–0.72)
Ref
(1.29–1.66)
(0.91–1.25)
(1.16–1.77)
(0.86–1.01)
(0.81–1.24)
Ref
(1.06–1.27)
Ref
(0.66–0.88)
(0.023–0.063)
CIb (95%)
(0.02–0.07)
(0.06–0.11)
(1.93–2.33)
(0.98–1.09)
(1.01–1.06)
(0.99–1.00)
(1.29–1.36)
(0.99–0.99)
Ref
(0.91–1.17)
(0.88–1.10)
Ref
(0.73–1.11)
(0.84–1.24)
(0.97–1.39)
Ref
(1.19–1.54)
(0.97–1.29)
Ref
(0.96–1.21)
(0.77–1.05)
(1.19–2.14)
(0.77–0.91)
(0.74–1.08)
NA
2.76**c
0.98
1.05*
0.99
1.37**
0.99**
NA
1.00
1.27**
1.37**
1.47**
1.00
1.04
0.67**
1.00
1.47**
1.07
1.43
0.93
1.01
1.00
1.16*
1.00
0.77**
0.033**
ICCd
0.04**
0.08**
2.12**c
1.03
1.04*
0.99
1.33**
0.99**
1.00
1.03
0.98
1.00
0.90
1.02
1.16
1.00
1.36**
1.12
1.00
1.08
0.90
1.59*
0.84**
0.90
NA
1.00
1.11
1.18
1.32*
1.00
1.48*
1.10
1.00
0.99
0.94
1.99**
0.85*
1.06
NA
Ref
(0.92–1.33)
(0.97–1.43)
(1.11–1.58)
Ref
(1.15–1.90)
(0.86–1.41)
Ref
(0.80–1.22)
(0.75–1.18)
(1.37–2.90)
(0.75–0.97)
(0.78–1.44)
NA
1.00
0.86*
0.022**
ICCd
0.03**
0.07**
Ref
(0.78–0.94)
(0.012–0.032)
CIb (95%)
(0.02–0.07)
(0.05–0.09)
1.00
0.94
0.083**
ICCd
0.04*
0.09**
Ref
(0.83–1.07)
(0.043–0.163)
CIb (95%)
(0.01–0.07)
(0.06–0.12)
NA ¼ Not applicable
a
From equation 1, we use expb1 which is the effect of a one unit increase in the covariate on the odds of being overweight/obese.
b
CI denotes confidence interval at 95%.
c
Asterisks denote level of statistical significance: ** P < 0.001; * P < 0.05.
d
ICC refers to Intra-class correlation, which measures the proportion of the total variance that is due to the differences between states/PSUs.
has a higher likelihood of overweight/obesity (OR ¼ 1.08 and 1.16,
respectively) compared with those who use biofuel.
Table 1 shows that urban and rural Muslim females have a much
higher likelihood of overweight/obesity relative to the Hindu–
general group (OR ¼ 1.25 and 1.47, respectively). However, there is
no statistical difference in OR for urban and rural Muslim males.
Sikhs have a statistically significant higher likelihood of overweight/
obesity across the four sub-populations. Urban and rural Sikh males
had a higher likelihood of overweight/obesity (OR ¼ 1.59 and 1.99,
respectively) compared with the Hindu–general group.
Notwithstanding the significance of particular individual-level
covariates, the multi-level modelling results reveal that unobserved
factors at state and associated PSU levels significantly explain
around 7–9% of the total residual variance. In Figure 3a and b the
states are ranked in terms of the impact of state-level unobserved
factors on the likelihood of individuals’ overweight/obesity along
with confidence intervals (95%) in the form of caterpillar plots for
the urban female and male sub-populations. Due to space limitations only the urban female and male results are presented: they
have a substantially higher overweight/obesity prevalence compared
with the rural population. From Figure 3a it can be seen that states
such as Tamil Nadu, Andhra Pradesh, Kerala, Karnakata and
Punjab have a higher likelihood of overweight/obesity relative to the
Table 2. Category classifications used in estimating the probability
of overweight/obesity
Urban female
Rural female
Urban male
Rural male
Major city
Watching TV
daily
Not working
Hindu–general
Clean cooking
fuel
No tobacco
NA
No TV watching
NA
No TV watching
Manual work
Hindu–general
Biofuel
Major city
Watching TV
daily
White collar
Hindu–general
NA
Manual work
Hindu–general
NA
No tobacco
No tobacco
Tobacco
NA, not applicable.
Indian mean. (See Appendix 2 for the acronyms for each of the
states.) Conversely, other major states such as Rajasthan, Bihar,
Jharkand and Madhya Pradesh have a lower likelihood of overweight/obesity relative to the Indian mean. Ostensibly, the statelevel effects can be considered as intercept shifts of the model estimates. Results also show that some smaller north-eastern states had
a lower likelihood of overweight/obesity than the Indian mean: for
example, Nagaland, Mizoram and Meghalaya. (Although not
shown in this article, our separate analysis reveals that the
586
Health Policy and Planning, 2016, Vol. 31, No. 5
0.90
Urban Female
Rural Female
Urban Male
Rural Male
0.80
Probablity of O-O
0.70
0.60
0.50
0.40
0.30
0.20
0.10
0.00
0
20
40
60
80
100
120
Wealth Level
Figure 1. Probability of overweight/obesity (O-O) by wealth level
0.30
Urban Female
Rural Female
Urban Male
Rural Male
Probablity of O-O
0.25
0.20
0.15
0.10
0.05
0.00
0
5
10
15
20
25
30
Years of Educaon
Figure 2. Probability of overweight/obesity (O-O) with respect to the level of education (years)
North-eastern states also have a lower probability of being underweight relative to the Indian mean.)
For urban males the caterpillar plot in Figure 3b shows that
some states such as Andhra Pradesh and Tamil Nadu and, to a lesser
extent, Karnataka and Punjab have a statistically higher likelihood
of overweight/obesity relative to states such as Madhya
Pradesh, Bihar and Rajasthan. The results show that even after
taking into account individual-level covariates, state-level contextual factors exert a separate influence on the likelihood of
overweight/obesity.
Health Policy and Planning, 2016, Vol. 31, No. 5
587
1
(a)
State level random effect
-.5
0
.5
TAN
ANP
KER
PUN
UTP
GUJ
GOA
CHG SKM
ORS UTC
MHR DEL
ASM TRP
HAR JAK WBG
KAR
MNP
HMP
ARP
MAP JHK
BIH NGL
RAJ
MZR
-1
MGH
0
10
20
30
State (ranked)
1
(b)
State level random effect
-.5
0
.5
ANP
TAN
KAR
PUN
UTP MNP KER
NGL
ARP
JHK JAK RAJ
ASM
GOA
WBG
DEL ORS MZR
TRP
HAR
GUJ
CHG SKM MHR
HMP
UTC BIH
MAP
-1
MGH
0
10
20
30
State (ranked)
Figure 3. (a) State-level intercept effects: urban females. (b) State-level intercept effects: urban males
Discussion
By disaggregating the dataset into sub-populations and giving focus
to macro-level contextual factors our study yields some interesting
findings not reported elsewhere. Studies reveal that in developed
countries, overweight/obesity is more prevalent among the poorer
and less educated sections of the population whereas in low-income
developing countries the reverse is the case, where such prevalence is
greater among those with higher socio-economic status (Dinsa et al.
2012; Cohen et al. 2013). Consistent with this, a number of studies
reveal that Indians from the upper socio-economic strata have a
higher calorific and fat content diet and engage in less physical activity in comparison to those from lower socio-economic backgrounds
resulting in a higher prevalence of overweight/obesity (Agrawal,
2005; Griffiths and Bentley 2005; Misra et al. 2011; Misra and
Shrivastava 2013). Interestingly however, our findings show that for
female Indians, in particular urban women, there is initially a positive association between overweight/obesity and education, but once
educational attainment exceeds around seven to eight years a
588
negative association begins to develop and the probability of overweight/obesity declines with additional years of education. One possible interpretation for the inverted U-shaped findings, as argued by
Cohen et al. (2013), could be that illiterate and lowly educated females have different physical patterns of activity while highly educated females engage in weight-control behaviours, such as a
healthier diet and/or increased physical activity, which becomes
dominant at very higher levels of education. In this context, our
findings could suggest that highly-educated urban women in India
are beginning to exhibit weight-control behavioural traits similar to
educated women from developed countries. Interestingly, this inverted-U relationship between education and overweight/obesity, as
yet, does not extend to males. Nor, as yet, does a similar relationship
apply between wealth status and overweight/obesity. One can speculate on the reasons for this but clearly further research is required to
understand the differing dynamics governing the relationship between education and gender, the differing perceptions and attitudes
to overweight/obesity and the changing behavioural patterns
overtime.
It is also worth noting, as argued by Cohen et al. (2013), the importance of functional form and the need to consider non-linear models when modelling the relationship between education and
overweight/obesity—as we have done in this study. Also our study
has given particular consideration to isolating the independent
effects of education on the probability of overweight/obesity
(i.e. Figure 2). Attention to these aspects may explain in part why
other studies have not discerned the positive effect that education
can play in lowering the likelihood of overweight/obesity for females
in India. Education is particularly important because it enhances
knowledge of good health maintenance and offers the potential to
increase the effectiveness of public health policies, including dietary
information and healthy lifestyle campaigns, and thereby deal with
the deleterious effects of both under and over nutrition (Lecerof et al.
2011).
Our study findings reveal Muslim females and Sikh males and females have a greater likelihood of overweight/obesity relative to the
Hindu–general group after allowing for socio-economic status and
other individual-level covariates. In the case of female sub-populations, some studies have raised the spectre of cultural practises and
social restrictions on physical activity (and on workforce participation) and on dietary behaviour as ‘possible’ reasons for higher obesity prevalence rates but data limitations, as is the case in our study,
prevented further analysis (Griffiths and Bentley 2001; Chopra et al.
2013; Misra and Shrivastava 2013; Misra et al. 2013). Regarding
the Sikh sub-population, our separate analyses (not reported here)
of the 61st round (2004–05) of the National Sample Survey on consumption expenditure, reveal that notwithstanding differences in
dietary patterns, the total calorific and fat intake of Sikhs across the
income groups differs little from the equivalent income groups of
the general Indian population. Data limitations in this study prevent
us from identifying salient socio-cultural characteristics particular to
the Sikhs which may contribute to their high BMI. Although we are
not able to determine the underlying causal mechanisms accounting
for the differences reported in this study there is a growing recognition of the need to better understand the significance of socio-cultural determinants of dietary choices and physical activity which
impact on overweight/obesity (Swinburn et al. 2011). In this regard,
policy is likely to be more effective if it is targeted in a way which
recognizes socio-cultural diversity rather than being applied uniformly across all socio-cultural groups. Another important observation in this study is that the relationship between wealth and
overweight/obesity is more intense for males than for females. Thus,
Health Policy and Planning, 2016, Vol. 31, No. 5
it needs to be recognized that as the socio-economic status of the
Indian population increases over time overweight/obesity is likely to
rise more markedly for men than women.
The multi-level analysis reveals that, in addition to individuallevel covariates such as socio-economics status, state-level macro environment exerts an independent influence over the likelihood of
overweight/obesity. One can speculate as to whether some states
may have a more ‘obesogenic’ environment—e.g. a built environment limiting physical activity, a high concentration of fast-food
outlets or particular regionally based socio-cultural-attributes—
which may commonly expose individuals to overweight/obesity
types of behaviour (Cohen and Fletcher 2008; Swinburn et al.
2011). The nature of the underlying causal linkages of state-level
factors beyond the individual-level characteristics on overweight/
obesity needs to be more fully understood. Related to macro-level
contextual factors, a growing body of empirical literature analysing
the role of social influences affecting overweight/obesity has accumulated over the past two decades (Christakis and Fowler 2007;
Auld 2011; Fletcher 2011; Renna et al. 2008; Trogdon et al. 2008;
Mora and Gil 2012). Cultural norms and values, habituation and
imitative behaviour are embedded in the social network which influences the behaviour of individuals in an interdependent way. The
implication is that policy interventions which target the social network rather than the individual can create a virtuous circle of behavioural change as people respond interdependently to changes in their
social network (Smith and Christakis 2008; Corrado and Distante
2012).
Associated with social transmission mechanisms, is the need to
understand socio-cultural perceptions regarding body size and
related weight-control activity. A recent multi-city study in India
found there was a prevalent misconception that an overweight child
was considered a ‘healthy child’ leading to unhealthy dietary habits
flowing from mother to child (Gulati et al. 2013). Similarly, a positive pattern of attitudes towards obesity found in a study of parents
of obese teenagers in the Punjab raised concerns over parents not
taking measures to combat overweight/obesity (Rashmi and Jaswal
2011). Interestingly, a recent study of female adults in India reported
a discrepancy between self-perceived and actual body weight
(Agrawal et al. 2014). The study reported that 25% of overweight
females and 10% of obese females perceived themselves as being a
normal weight and a majority of these women did not engage in
physical activity to reduce their weight. A similar misconception of
understating body weight was found by Ramesh (2012) in a study of
adolescents in the Indian state of Kerala. Given the considerable diversity of the Indian sub-continent, policy strategies need to be more
nuanced, recognizing the socio-cultural and macro-level contextual
factors which influence decisions regarding dietary behaviour, physical activity and perceptions about body size.
This study has a number of caveats. A key limitation is that lifestyle factors cannot be fully explored using NFHS-3 data as it does
not collect information on energy intake or expenditure. Thus,
watching TV was included as a proxy for physical inactivity, but it
is a crude measure and does not necessarily capture differences in
sedentary behaviours across sub-populations or differences in lifestyle behaviours affecting diet.5 In addition, whilst our study reveals
that sub-populations from different socio-religious/socio-cultural
backgrounds have differing probabilities of overweight/obesity, we
are unable to identify underlying casual mechanisms by which the
imbalance between calorie intake and expenditure occurs across
such groups. Accordingly, further research (especially of a qualitative nature) is required. Similarly, without further research we are
unable to determine the nature of the transmission mechanism by
Health Policy and Planning, 2016, Vol. 31, No. 5
which education for females, but not males, exhibits an inverted
U-shaped relationship in its association with the probability of
overweight/obesity. Clearly, more needs to be understood in
terms of how gender interacts with education in this respect. Finally,
although the use of multi-level analysis reveals the significance
of state/PSU-related contextual factors influencing overweight/
obesity, we are not able to disentangle macro-level ‘obesogenic’
environmental factors from social influences in explaining such
variations. In this regard, our analysis can be considered
exploratory in revealing the importance of spatially and sociocultural related contextual factors in addition to individual-level
factors in explaining the variations in the prevalence of overweight/
obesity.
589
4
5
Wealth is divided into equally discrete units on a scale ranging
from 1 to 102 units.
We are unable to determine whether the lower likelihood of
overweight/obesity from tobacco use is associated with lower
health status (from the deleterious effects of smoking) or the
possible appetite suppressant nature of tobacco. Also, it is outside the scope of this study to determine whether the positive association between the use of cleaner cooking fuel and the
probability of overweight/obesity is associated with higher
health status (by avoiding exposure to toxic pollutants) or the
possible changes in meal preparation and dietary patterns
brought about by the use of more time-convenient cooking fuel.
References
Conclusion
Study findings reveal considerable differences in the underlying relationships influencing the likelihood of overweight/obesity across
sub-populations in India. For females beyond a particular education
level there is a negative association with overweight/obesity.
Findings show that Muslim females and all of the Sikh sub-populations have higher overweight/obesity likelihoods relative to the
Hindu–general sub-population. This suggests the importance of
socio-cultural influences. In addition, there is a stronger proclivity in
the likelihood of overweight/obesity with regards to wealth for
males raising the spectre of the differential impact of economic development on particular sub-populations. Also, multi-level analysis
reveals that spatially related factors exert an independent influence
on overweight/obesity.
Although this study is limited in identifying the precise causal
mechanisms underpinning particular macro-level contextual factors,
the results highlight the need to refrain from adopting a ‘one size fits
all’ approach to policy. Instead, policy implementation requires a
more nuanced and targeted approach to incorporate the growing
recognition of spatial and socio-cultural related factors on overweight/obesity and to exploit such influences in spreading healthy
behaviours. India is undergoing considerable economic growth,
which is seeing a massive transformation in nutrition patterns,
increasing urbanization and a growing sedentary lifestyle which all
contribute to the overweight/obesity epidemic. Moreover, for any
given BMI, Indians are much more susceptible than Caucasians to
obesity-related diseases and metabolic-related abnormalities
(Khandelwal and Reddy 2013). Such diseases accounted for more
than 50% of deaths in 2008 and NCD represents a major public
health challenge for the Indian government (Upadhyay 2012).
Understanding the determinants driving the overweight/obesity pattern, both at the individual and the macro levels, is critical to the implementation of appropriate policy strategies. This article makes an
initial contribution towards this understanding.
Conflict of interest statement. None declared.
Notes
1
2
3
Large city, small city and town populations were >1 million,
100 000 to 1 million and <100 000, respectively.
The notion of distinguishing the type of cooking fuel between clean
and biofuel was adopted from Jose and Navaneethan (2010).
Our analysis reveals that alcohol consumption was not statistically significant across any of the sub-populations and it was
therefore excluded from the base model.
Agrawal P. 2005. Role of lifestyle and diet in emerging obesity among Indian
women and its impact upon their health status. Paper for the oral presentation in the IUSSP XXV International Population Conference Tours, France,
18–23 July 2005.
Agrawal P, Gupta K, Mishra V, Agrawal S. 2014. A study on body-weight perception, future intention and weight-management behaviour among normal-weight,
overweight and obese women in India. Public Health Nutrition 17: 884–95.
Auld M. 2011. Effect of large-scale social interactions on body weight. Journal
of Health Economics 30: 303–16.
Chopra S, Misra A, Gulati S, Gupta R. 2002. Overweight, obesity and related
non-communicable diseases in Asian Indian girls and women. European
Journal of Clinical Nutrition 67: 688–96.
Chou S-Y, Grossman M, Saffer H. 2004. An economics analysis of adult obesity: results from the Behavioural Risk Factor Surveillance System. Journal of
Health Economics 23: 565–87.
Christakis N, Fowler, J. 2007. The spread of obesity in a large social network
over 32 years. The New England Journal of Medicine 357: 370–9.
Cohen A, Rai M, Rehkopf D, Abrams B. 2013. Educational attainment and
obesity: a systematic review. Obesity Reviews 14: 989–1005.
Cohen E, Fletcher J. 2008. Is obesity contagious? Social networks vs. environmental factors in the obesity epidemic. Journal of Health Economics 27: 1382–7.
Corrado L, Distante R. 2012. Obesity is Contagious! Evidence from US Data.
Review of Environment, Energy and Economic. http://dx.doi.org/10.7711/
feemre3.2012.11.002, accessed 15 November 2013.
Cutler D, Glaeser E, Shapiro M. 2003. Why have Americans become more
obese?. Journal of Economics Perspective 17: 93–118.
Dinsa G, Goryakin Y, Fumagalli E, Suhrcke M. 2012. Obesity and socioeconomic status in developing countries: a systematic review. Obesity Reviews
13: 1067–79.
FAO. 2014. The state of food security in the world. Rome: Food and
Agriculture Organization of the United Nations..
Fletcher J. 2011. Peer effects and obesity. In: Cawley J (ed). The Oxford
Handbook of the Social Science of Obesity. Published Oxford Handbook
online. www.oxfordhandbooks.com, accessed 30 September 2013.
Griffiths P, Bentley M. 2001. The nutrition transition is underway in India.
Community and International Nutrition 131: 2692–700.
Griffiths P, Bentley M. 2005. Women of higher socio-economic status are
more likely to be overweight in Karnataka, India. European Journal of
Clinical Nutrition 59: 1217–20.
Gulati S, Misra A, Colles S et al. 2013. Dietary intakes and familial correlates
of overweight/obesity: a four cities study in India. Annals of Nutrition and
Metabolism 62: 279–90.
Jose S, Navaneetham K. 2010. Social infrastructure and women’s undernutrition. Economic and Political Weekly XLV: 83–9.
Khandelwal S, Reddy S. 2013. Eliciting a policy response for the rising epidemic of overweight-obesity in India. Obesity Reviews 14: 114–25.
Lakdawalla D, Philipson T. 2007. Labor supply and weight. Journal of
Human Resources 42: 85–116.
Lecerof S, Westerling R, Moghaddassi M, Östergren P-O. 2011. Health information for migrants: The role of educational level in prevention of overweight. Scandinavian Journal of Public Health 39: 172–8.
590
Health Policy and Planning, 2016, Vol. 31, No. 5
Smith K, Christakis N. 2008. Social networks and health. Annual Review of
Sociology 34: 405–29.
Subramanian S, Smith G. 2006. Patterns, distribution, and determinants of
under- and overnutrition: a population-based study of women in India. The
American Journal of Clinical Nutrition 84: 633–40.
Subramanian S, Perkins J, Khan K. 2009. Do burdens of underweight and
overweight coexist among lower socioeconomic groups in India? The
American Journal of Clinical Nutrition 90: 369–76.
Swinburn B, Sacks G, Hall K et al. 2011. The global obesity
pandemic: shaped by global drivers and local environment. The Lancet 378:
804–14.
Trogdon J, Nonnemaker J, Pais J. 2008. Peer effects in adolescent overweight.
Journal of Health Economics 27: 1388–99.
Upadhyay R. 2012. An overview of the burden of non-communicable diseases
in India. Iranian Journal Public Health 41: 1–8.
Misra A, Singhal N, Sivakumar B et al. 2011. Nutrition transition in India:
secular trends in dietary intake and their relationship to diet-related noncommunicable diseases. Journal of Diabetes 3: 278–92.
Misra A, Shrivastava U. 2013. Obesity and dyslipidemia in South Asians.
Nutrients 5: 2708–33.
Mora T, Gil J. 2012. Peer effects in adolescent BMI: evidence from Spain.
Health Economics 22: 510–6.
Popkin B, Adair L, Ng S. 2012. Global nutrition transition and the pandemic
of obesity in developing countries. Nutrition Reviews 70: 3–21.
Ramesh K. 2012b. Body/weight perceptions and prevalence of obesity among
adolescents – Kerala, India. International Journal of Health and Allied
Sciences 1: 92–7.
Rashmi, Jaswal S. (2011). Attitudes of parents, peers and teachers towards
obese teenagers. Journal of Psychology 2: 45–51.
Renna F, Grafova I, Thakur N. 2008. The effect of friends on adolescent body
weight. Economics and Human Biology 6: 377–87.
Appendix
Appendix 1. Descriptive statistics for NFHS-3 dataset showing sample size of different predictor variables and respective overweight/obese
percentage.
Total sub-population
Wealth
Poorest
Poorer
Middle
Richer
Richest
Education
No education
Primary
Secondary
Higher
Age (years)
15–19
20–24
25–29
30–34
35–39
40–44
45–49
50–54
Type of residence
Capital city
Small city
Town
TV watching
Never
Less than once a week
At least once a week
Everyday
Occupation
Not working
White collar
Manual labour
Socio-religious group
Hindu–general
Muslim
Christian
Sikh
Hindu–caste
Urban Female
N (% overweight)
Rural Female
N (% overweight)
Urban Male
N (% overweight)
Rural Male
N (% overweight)
53174 (23.2)
65560 (7.3)
34656 (16.7)
34572 (5.9)
866 (4.2)
2361 (6.8)
6330 (11.2)
15363 (17.7)
28254 (32.4)
12694 (1.7)
14665 (3.5)
16591 (6.3)
13595 (13.3)
8015 (24.7)
497 (3.9)
1511 (3.5)
4555 (6.9)
10940 (11.1)
17153 (24.9)
6199 (1.3)
8217 (2.1)
9497 (4.6)
7088 (10.2)
3571 (23.7)
10191 (17.2)
6201 (21.0)
27372 (24.0)
9402 (30.8)
27806 (4.9)
10959 (8.4)
24275 (9.7)
2516 (15.3)
3016 (7.7)
4144 (10.9)
19906 (15.2)
7575 (28.3)
6910 (3.1)
6624 (4.2)
18202 (6.6)
2824 (14.9)
9683 (4.9)
9672 (10.9)
8921 (20.3)
7703 (29.2)
7119 (35.0)
5778 (41.8)
4298 (42.2)
NA
13131 (1.4)
11988 (3.6)
10782 (6.2)
9412 (8.7)
8362 (11.6)
6784 (13.6)
5101 (15.7)
NA
6037 (3.5)
6242 (7.7)
5159 (14.5)
4498 (20.1)
4155 (24.9)
3565 (26.0)
2933 (27.6)
2067 (28.8)
6152 (0.7)
5309 (2.9)
5080 (5.5)
4548 (7.6)
4408 (8.5)
3713 (8.7)
3185 (9.5)
2177 (9.9)
23940 (26.8)
8763 (22.4)
20471 (20.8)
NA
NA
NA
18018 (18.4)
4567 (16.3)
12071 (15.5)
NA
NA
NA
4730 (12.7)
3662 (16.4)
5748 (19.4)
39026 (26.1)
25113 (3.6)
8647 (5.9)
8704 (8.2)
23081 (13.1)
1670 (8.6)
3196 (10.2)
4590 (12.5)
25195 (19.1)
6962 (2.9)
7936 (3.5)
6618 (4.7)
13054 (10.6)
36886 (24.4)
9471 (25.4)
6817 (14.9)
33727 (9.2)
3914 (16.1)
27919 (4.2)
5904 (7.2)
14772 (24.1)
13923 (13.0)
4271 (2.3)
6010 (12.2)
24224 (5.1)
16078 (28.6)
8435 (22.7)
4706 (26.1)
763 (41.7)
21393 (19.3)
13580 (10.9)
7151 (8.2)
5925 (12.3)
1927 (27.7)
35005 (5.3)
9283 (21.1)
5472 (13.8)
2673 (20.7)
336 (35.8)
15653 (13.8)
6450 (9.6)
3079 (5.1)
3699 (9.9)
629 (21.9)
19704 (4.4)
(continued)
Health Policy and Planning, 2016, Vol. 31, No. 5
591
Appendix 1. (continued)
Other
Cooking
Biofuel
Clean fuel
Use of tobacco
No
Yes
Urban Female
N (% overweight)
Rural Female
N (% overweight)
Urban Male
N (% overweight)
Rural Male
N (% overweight)
1799 (20.9)
1972 (4.4)
1239 (22.4)
1011 (2.9)
15413 (12.7)
37757 (28.9)
57343 (5.8)
8210 (22.5)
50586 (23.4)
2557 (18.4)
60150 (7.5)
5374 (3.6)
NA ¼ Not applicable
Appendix 2. Glossary of acronyms for Indian states.
ANP
ARP
ASM
BIH
CHG
DEL
GOA
GUJ
HAR
HMP
JAK
JHK
KAR
KER
MAP
MGH
MHR
MNP
MZR
NGL
ORS
PUN
RAJ
SKM
TAM
TRP
UTC
UTP
WBG
Andhra Pradesh
Arunachal Pradesh
Assam
Bihar
Chhattisgarh
Delhi
Goa
Gujarat
Haryana
Himachal Pradesh
Jammu and Kashmir
Jharkhand
Karnataka
Kerala
Madhya Pradesh
Meghalaya
Maharashtra
Manipur
Mizoram
Nagaland
Orissa
Punjab
Rajasthan
Sikkim
Tamil Nadu
Tripura
Uttaranchal
Uttar Pradesh
West Bengal
NA
NA
28536 (17.5)
6106 (12.9)
NA
NA
25663 (6.5)
8886 (4.4)