Women’s Malnutrition in India: The Role of Economic and Social Status

Abstract

This study has used data from the India Human Development Survey (IHDS), 2005–06, to study the factors influencing the body mass index (BMI) of women between 20 and 40 years of age in India. BMI captures both undernutrition and over-nutrition, and a quantile regression model has been used to capture the differential impact of the explanatory variables across the wide range of its values.

Variables such as per capita income, per capita consumption expenditure and wealth are important in explaining variations in BMI, but the impact varies across the quantiles. The impact of per capita consumption expenditure is greater than that of per capita income, indicating the effectiveness with which resources are converted to consumption. Higher levels of wealth status affect BMI more across all the quantiles. Women’s autonomy index shows a positive impact only for higher levels of the index value, but the magnitude is very small, while caste and religion play an important role even after controlling for economic status.

JEL Classification: I12, J10, O18

Keywords

BMI Women Quantile Regression Income Consumption

1. INTRODUCTION

Nutrition forms the basis of the overall well-being of a person. It is even more important for women since an unhealthy woman gives birth to an unhealthy child. This results in a vicious circle, the results of which linger for a long period of time. The health status of women in India is one of the lowest in the world, with India having one of the highest proportions of malnourished women in developing countries (Navaneetham & Jose, 2008). A recent study reveals that in 2000 about 70 per cent of non-pregnant and 75 per cent of pregnant women aged 15–49 years in India were anaemic (Mason et al., 2005). According to a United Nations Development Programme (UNDP) report, there have been some improvements over the last 50 years; for instance, female life expectancy at birth has increased to 65.3, the maternal mortality rate has fallen to 450 (per lakh live births) and the infant mortality rate has reduced to 56 (per 1,000 live births) in 2005 (UNDP, 2008), but there is still lot of scope for improvement.

The reasons behind women’s undernutrition in India are multiple and complex. Discriminatory practices combined with high repression and low socioeconomic standards play a major role in determining the overall (low) nutrition status of women. Women in India are also expected to do heavy household work manually with limited assistance from male household members and in many instances are subjected to frequent pregnancies. Along with this, inadequate quantity and quality of their diet also takes a toll on their health and nutritional status. Undernutrition is both the cause and effect of poverty. Poor women do not get the right diet, neither are they aware of the benefits of such a diet.

Today, urban women have become more empowered and are more aware. This change is increasingly evident in their nutrition status and overall well-being. Education has played a major role in providing more autonomy to women. Studies show that education not only contributes to their own physical well-being but also to that of the whole family.

This study uses the body mass index (BMI) of women in India (between 20 and 40 years of age) as an indicator of nutrition level and tries to assess the factors affecting it. An ordinary least-square (OLS) as well as quantile regression analysis were carried out for this. It is well known that both low (chronic energy deficiency [CED]) and high (overweight and obesity) levels of BMI are not considered healthy and are caused by malnourishment but of different kinds. Hence, a quantile regression model (QRM) is used to help distinguish between different factors affecting women with different levels of BMI.

The primary focus of this study is to identify different aspects of economic prosperity, that is, income, consumption and wealth. These variables are supposed to be complements or substitutes of each other, but each may have a different and significant impact on the BMI. There are several studies which discuss the relevance of income in improving nutritional intakes, the earliest being the one conducted by Behrman and Deolalikar (1987). However, such discussions are limited in the context of nutritional outcomes, partly because fewer studies exist on nutritional outcomes in India and because the only large sample data set available has information only on wealth status, which is more a long-term indicator of economic status.

The India Human Development Survey (IHDS) data set (2005–06) for the first time allows us to analyze the impact of flow variables like income or consumption along with a stock variable like wealth. It is expected that these three variables would be endogenous in determining nutritional outcomes, but we treat them as exogenous because the nutritional outcome indicator is at the individual level while the economic status variables are at the household level, which would pose difficulty in model specification. Furthermore, BMI is measured only for some women (referred to as ‘eligible’ women) in the sampled households and is not available for all members of the household. This makes it difficult to create an index of nutritional outcome for the household and then consider estimating a household-level model, allowing for feedback between nutritional outcomes and economic status. The advantage of estimating an individual-level model is that we also consider variables like women’s empowerment or their work status that would determine individual nutritional outcome. Besides these variables, other social and infrastructural variables have also been included as control variables in the quantile regression analysis.

2. BRIEF REVIEW OF LITERATURE

The study conducted by Navaneetham and Jose (2008) based on the National Family Health Survey 2005–06 (NFHS-3) shows that around 40 per cent women in rural India are affected by CED (i.e., they have a BMI below 18.5). The incidence of CED among urban women is 15 per cent lower than among rural women. About half the women below the age of 20 years suffer from CED, but this figure falls to half for women aged between 40 and 49 years, indicating a clear improvement in CED rates with age; however, this age group also shows a higher rate of overweight women, more so in urban areas and among states like Tamil Nadu and Kerala (Seshadri, 2009). Alongside a low BMI, studies also show that compared to men, women particularly in urban areas have high BMI values; nearly one-fourth of urban women suffer from obesity (a BMI above 25), which is a new nutrition problem emerging in urban India (Navaneetham & Jose, 2008). Ackerson, Kawachi, Barbeau and Subramanian (2008) and Singh, Srinivas and Sekher (2011) are among the few studies that address the double burden of malnutrition among Indian women using the NFHS-2 (1998–99) and NFHS-3 (2005–06) data, respectively. According to NFHS-3, 50 per cent of women from the poorest quintile suffer from CED whereas women belonging to disadvantaged social groups also show a far higher rate. Ackerson et al. show at the more granular level of districts and villages that there are clearly regional patterns—a contiguity of low-BMI regions and high-BMI regions and the association with regional development is an important finding of this study. There are also a few studies that analyze the determinants of women’s BMI based on a single state as in Kumar, Ramachandran and Viswanathan (2009) for Uttar Pradesh or for a few states as in Seshadri (2009) for the states of Bihar, Gujarat, Punjab and Tamil Nadu. The study by Kumar et al. (2009) is somewhat unique for it tries to analyse the factors that affect women’s vulnerability to CED instead of focusing only on the determinants of CED.

According to NFHS-3 data, the highest incidence of CED is found in the eastern states, such as Bihar (45.1 per cent), Jharkhand (43.4 per cent), Orissa (41.7 per cent) and West Bengal (39.1 per cent), in India. On the other hand, the southern states have the lowest incidence of CED (Deaton, 2008). There is high variation in the health conditions of women belonging to different states. In this study, we have thus controlled for the regional effects by including state dummies. The IHDS shows that the highest mean BMI of women is in the state of Punjab (23.59) and Kerala (23.10) while the lowest is in Bihar (20.81) and Orissa (20.91).¹

Women’s access to basic social infrastructure facilities such as toilet facilities, clean cooking fuel and drinking water are also important variables determining their well-being. Lack of access to sanitation makes a woman vulnerable to infections and, hence, to feeble health in the longer run. Unclean fuels, on the other hand, expose her to toxic pollutants and fetching water from long distances consumes a lot of energy. There is a difference of 20 percentage points between women with access to toilet facilities and those without in terms of incidence of CED. If women have access to clean cooking fuels then the incidence of undernutrition tends to be almost halved. Access to drinking water at home also plays an important role in determining CED among women in India, but the effect is relatively smaller (Jose & Navaneetham, 2010). In this study, access to sanitation and electricity, regular availability of tapped water and use of clean cooking fuel are all considered control variables, affecting BMI of women.

Women’s empowerment has a significant effect on a woman’s health status as well as on the health status of her children. Many women in developing countries cultivate, purchase and prepare much of the food eaten by their families, but they often have limited access to information about nutrition. An educated woman is more likely to spend on food and health care and will make an effort to diversify the family’s diet. She has greater ability to control the household’s physical and financial assets, is motivated to eat a healthy diet and feeds her babies and children with foods that meet their special nutritional requirements. There is significant evidence that a mother’s educational status directly influences her nutritional and health status as well as her children’s. Many studies have used proxy measures of women’s status, for example, indicators that depict sources of power such as education or age at marriage as measures of women’s autonomy. Bhagowalia, Menon, Quisumbing and Soundararajan (2012) used direct evidence of a woman’s empowerment, including her mobility, decision-making power and attitudes towards verbal and physical abuse, which could affect her power over making decisions about what food to consume, visiting the health care centre, dietary diversity, attitudes towards domestic violence and so on. The results from this study show that empowerment variables are positively associated with a woman’s nutritional outcome as well as her children’s nutrition. In another study of malnutrition in Zimbabwe, Malawi and Zambia, evidence suggests that women who have lower levels of autonomy and status within the household are more likely to experience undernutrition (Hindin, 2005).

Studies have also shown that women’s participation in paid work enhances her well-being; for instance, Sen (1990) concludes that ‘women’s paid employment would enable them through a variety of ways to attain a higher well-being.’ In this study, dummy variables related to the occupation of women in different sectors have been used to assess their impact individually, including a reference category for women who do not work in the labour market.

Education is another form of capturing empowerment and awareness. Smith and Haddad (2000), on the basis of an OLS regression and an error components model, studied 63 developing countries from 1975 through 1996. They find that for all countries in the sample women’s education alone explains 43 per cent of the overall decline in childhood malnutrition experienced by these countries during the years of the study. They also showed that national income played an important role in reducing child malnutrition in these countries. But this relationship was not seen in sub-Saharan Africa because of the decline in the overall income of the region between 1970 and 1995.

To bring this discussion into the context of this study we need to understand that there are multiple factors that influence women’s nutrition in India. There are not many studies on adult nutritional outcomes in India. India is growing fast economically with notable sociocultural changes. Hence, a household’s economic well-being and the changing status of women in society can play a major role in improving women’s health. Subramanian and Smith (2006) used the NFHS-2 for 1998–99 and found that the standard-of-living index, which they consider directly related to the amount of disposable household income available for food, is most strongly associated with undernutrition and overnutrition. They also showed that undernutrition was most prevalent among women belonging to lowest quintile of the standard of living. As the standard of living improved, the risk of being undernourished declined systematically. People from higher-income groups consume a diet containing 32 per cent of energy from fat while people from lower-income groups consumed only 17 per cent of their energy from fat. According to the authors, this partially explained the positive relationship between socioeconomic standards and the BMI of women.

In poor households women are at greater risk of undernutrition than men. Undernutrition in mothers, especially those who are pregnant or nursing, can be very detrimental since an unhealthy mother gives birth to an unhealthy child. The child is hence more susceptible to diseases and poor school performance and contributes inadequately to the economy in the longer run. This may lead to undernutrition in women and girls and, hence, forms a vicious circle of undernutrition and poverty. Also, women require more dietary iron than men do and more protein than usual when pregnant and lactating. Poor women, especially those in female-headed households, tend to have lower access than men to income, credit and other financial services and other resources that are needed to improve food security. It is important to study, learn and analyze the severity and nature of the complex factors responsible for the low nutritional/health standard of women in India.

Based on the findings of the earlier studies and the nature of information available in the IHDS data set, this study aims to understand the relevance of variables capturing economic prosperity and food intake on one hand and women’s work status and empowerment on the other, after controlling for the effects of other social and infrastructure variables on a nutritional outcome indicator like the BMI. Furthermore, the aim is to assess how the relevance of these variables varies across different quantiles of the BMI.

3. DATA AND METHODOLOGY

The analysis has been done using data from the IHDS 2005–06 (Desai, Dubey, Joshi, Sen, Shariff, and Vanneman, 2007) conducted jointly by the University of Maryland and National Council of Applied Economic Research (NCAER). It contains information from 41,554 households in 1,503 villages and 971 urban neighbourhoods across all the states and union territories of India.

The log of the BMI (lnbmi, henceforth) of women is taken as the dependent variable. The BMI signifies the nutrition status in the shorter run while height signifies longer-run health status. Hence, here an attempt has been made to study variables affecting the health of women in the shorter run. (The outlier values have been dropped so that the results are more efficient and reliable.) There can be changes in body height due to physiological changes before menarche, after menopause and later due to old age. Height remains constant between 20 and 40 years of age indicating that the numeraire in the BMI (weight in kilograms divided by the square of the height in meters) does not vary with age. So, the changes in body weight due to various social, economic and environmental factors are effectively captured by the BMI. The analysis in this study is carried out for women in the age group of 20–40 years.

The data set provides information on variables related to health, education, employment, economic status, marriage, fertility and social capital. Interviews related to additional village, school and medical facilities are also available. Data provide regional segregated information as well as gender-centric and institutional information. The survey included two questionnaires, one for the household head and the other for eligible women, typically the wife of the household head. Questions related to household wealth, income and expenditure were addressed to the household head while questions relating to health, education and other social indicators were addressed to women.

Studies which have focused on women’s BMI similar to this study also use determinants such as women’s age, educational and marital status, number of children ever born, caste, religion, food habits representing quality of diets, place of residence, economic status captured by wealth status and access to social infrastructure (Jose & Navaneetham, 2010; Kumar et al., 2009; Singh et al., 2011). This study is somewhat different in that it emphasizes the impact of wealth, income and consumption, three different aspects of economic status, on the BMI of women in India. Although all three of these are closely related, they have differential and significant effects on the well-being of women. Income as a whole does not guarantee the rightful use of resources. Hence, per capita consumption and the wealth index are included as explanatory variables. The per capita consumption of a household reflects the use of the income available to the household. The IHDS provides information regarding the basic availability of household assets which can significantly ease the work of women and save a lot of energy. Also, income is a short-term representation of economic well-being while wealth is a long-term indicator. A household accumulates assets after saving from income earned over a period of time.

The data set collected information on expenditure on broad groups of food and non-food items. The consumption of certain food items like cereals is available; however, for many other food items the information is either not available or the food group is too broad (like for pulses and products or meat, egg and fish). The conversion of such broad groups of food intakes in quantities into their macro and micro nutrient content is not possible. Hence, the expenditure related to various food items is used to calculate dietary diversity.

To analyse the impact of the social status of women on their well-being and healthy BMI various empowerment-related variables have been used. The IHDS also provides occupational information for women who participate in the labour market and it has been used in this study as one of the explanatory variables.

In this study, an attempt has been made to study the factors influencing women’s BMI across its entire distribution, which is captured in a QRM. There may be different factors that affect women who are chronically malnourished and those who are obese; hence, the quantile regression helps to determine factors affecting women belonging to different quantiles of BMI. Seshadri (2009) was among the earliest studies in India to use a QRM to analyze the determinants of the BMI, as well as heights among women in the states of Bihar, Gujarat, Punjab and Tamil Nadu, using NFHS-3 data. Here, the analysis is carried out for five different quantiles of lnbmi, that is, the 10th, 25th, 50th, 75th and 90th quantile, where the 10th quantile represents women in the bottom 10 per cent (0.10) of the distribution of lnbmi. Similarly, the 25th quantile represents all those women whose lnbmi is such that the area under the probability density function of lnbmi lies between 10 per cent (0.1) and 25 per cent (0.25) and so on; finally, the 90th (0.90) quantile includes lnbmi in the top 10 per cent of the distribution. The lower quantiles would encompass those who are CED whereas the top quantiles would include those who are overweight and obese. Compared to the OLS method, where a single estimated coefficient is obtained for the different explanatory variables as shown in Equation (1) below, in the QRM the number of estimated coefficients would be the number of quantiles specified by researchers as shown in Equation (2) below:

OLS model: Y_{i} =β X {+ε}_{i}

(1)

QRM: Y_{i} = β(p) X {+ε}_{i} (p),p=0 .1,0 .25,0 .50 .75 and 0 .9

(2)

Here Y is the lnbmi, X is the vector of the explanatory variables, β is the vector of coefficients for the OLS model to be estimated by the specification in Equation (1) and β(p) would be the coefficient vector for any given pth quantile, so each quantile has an estimated vector of coefficients.

Other socioeconomic variables such as caste and religious affiliation, household composition and size, and social infrastructure—access to clean and safe drinking water, sanitation, electricity and clean cooking fuel—are also used as explanatory variables. State-level dummy variables have been used to control for regional variations. The analysis has been carried out for women in the age group of 20–40 years as explained above. Further details on the set of explanatory variables and related descriptive statistics are given in Section 4 below.

4. PRELIMINARY ANALYSIS OF THE DATA

According to the data, 26.3 per cent of the women in rural areas and 15.1 per cent of women in urban areas are chronically undernourished; that is, they have a BMI less than 18.5. The national average for the BMI for women is about 21.3, around 20.7 in the rural areas and 22.4 in the urban areas.

It is important to note that the mean BMI for women is higher than 18.5 in all states. It can be seen from Figure 1 that there is not much variation in the mean BMI across states and it is far lower than the variation in the standard deviation of the BMI. Some of the smaller states and union territories show high values of standard deviation, while many states in the north-eastern region show far lower values of standard deviation. The fact that the mean BMI for women across all states is above 18.5 refutes the hypothesis of an ‘ethnically determined predisposition of low BMI’ put forward by Panagariya (2012) recently. Other aspects of the descriptive statistics and results from the econometric analysis given below show that a low BMI is associated with lower economic status, access to poor quality of basic amenities and socio-demographic features associated with a lower standard of living. This aspect has been well argued in Jose (2014).

Figure 1

Source: Authors’ estimations

A brief discussion on the choice of explanatory variables used in the model is given below.

4.1 Economic Variables

4.1.1 Per Capita Income

Per capita income takes into account resources available per person and reflects the ability of the household, ex-ante, to spend on goods and services that would allow them to attain a healthy level of BMI. The higher the per capita income the better would be the standard of living and hence the coefficient is expected to be positive.

4.1.2 Per Capita Consumption

This indicates the revealed standard of living, including the availability and accessibility of goods and services. At lower levels of income, per capita income and consumption are highly correlated as income is mainly geared towards subsistence consumption; however, at higher levels of income the two variables would be less correlated, depending on the household’s preferences for current and future consumption. Thus, after controlling for per capita income we can expect per capita consumption to be positively significant for some quantiles.

Table 1
Mean and Standard Deviation of BMI (kg/m²) Across Wealth Groups

Wealth Groups Mean Standard Deviation

Poorest 19.4 2.68

Poor 20.2 2.95

Mid-wealth 20.7 3.35

Rich 21.8 3.69

Richest 23.3 4.11

Wealth Groups	Mean	Standard Deviation
Poorest	19.4	2.68
Poor	20.2	2.95
Mid-wealth	20.7	3.35
Rich	21.8	3.69
Richest	23.3	4.11

Source: Authors’ estimations.

Note: The wealth groups are formed on the basis of quintiles of wealth index.

4.1.3 Wealth Index

This is estimated using principal component analysis by taking into account basic household amenities like owned house, cycle, motor, sewing machine, wall clock, cot, chair, fan and so on. The quality of the house, such as a pucca (permanent) wall, roof and so on, is also taken into consideration. The first principle component forms the index, which is then categorized into five groups comprising the five quintiles—the poorest, poor, middle, rich, richest wealth groups (from lowest to highest quintile). The poorest quintile is taken as the reference category in the econometric models. Table 1 shows that women in the poorest quintile of the wealth index have a mean BMI of 19.4, which increases to 23.3 for the richest wealth group, but so does the standard deviation. There is a steeper increase across wealth groups in standard deviations compared to the mean values from the poorest to the richest quintiles. Given this observation and the model specification, we expect the magnitude of the coefficients for the remaining quintiles to increase, though we can expect the lower wealth groups to be not significant for some of the lower quantiles of BMI in the QRM.

4.2 Dietary Variables

4.2.1 Share of Food Expenditure in Total Household Expenditure

At the outset, the effect of this variable on BMI is ambiguous. This is because as we increase expenditure on food in absolute terms the nutrition status of a person is expected to rise. On the other hand, as food expenditure as a proportion of total expenditure falls, the household is expected to be better off as per Engel’s Law. The average propensity to consume food items for the poorest families is usually close to 1, and this ratio falls as the household becomes economically better off. This implies that nutrition status should improve with a fall in the ratio of food expenditure to total expenditure. Hence the effect on the log of BMI cannot be predicted a priori.

4.2.2 Dietary Diversity

This variable is estimated by finding the proportion of expenditure on various food groups (f_i) within total food expenditure (f), taking a square of each of these terms and adding them up. For the hth household and with i = 1,2,…n food commodities, dietary diversity is defined as

D D^{h} = \sum_{i =1}^{n} {(f_{i}^{h} / f^{h})}^{2}

(3)

The value of the above expression lies between 0 and 1. The lower the value of this index, the greater the diversity of food items consumed by the household. One would expect women to have a higher average BMI if they consume a more diversified diet and hence the dietary diversity index as defined above (lower value, higher dietary diversity) is expected to have a negative relationship with the BMI. However, high levels of BMI also lead to overweight and obesity due to inappropriate diets, and one can expect this coefficient to be smaller in magnitude for the uppermost quantile of BMI in the QRM compared to other quantiles.² Given that not many women are overweight and obese and that this is an indicator of household dietary diversity and not of individual dietary diversity, there can be some variations in the expected significance and magnitude of the estimated coefficient compared to what has been hypothesized.

4.3 Other Social Infrastructure Variables

This consists of dummy variables like access to LPG cooking fuel, regular piped drinking water, sanitation facility within the home and electricity. These variables signify some very basic needs of civilised social life and have a direct impact on an individual’s health status by preventing frequent illnesses. Access to regular piped drinking water indicates that women are less receptive to diseases and that women do not have to travel long distance to fetch water every day. Similarly, a proper sanitation facility at home is more hygienic and reduces the chances of falling ill. Use of unclean fuel exposes a woman to harmful smoke, which is not good for her health and increases her chances of having respiratory diseases. Access to electricity is another basic amenity for households, which increases access to electrical gadgets and in turn reduces a woman’s household drudgery. Electricity has reached the majority of towns and villages in India, but its absence signifies severe incompetency for a household. All these variables reflect the well-being and awareness of a family indirectly. The government also plays a major role in providing these facilities to the people and hence these variables also indicated the regional effectiveness of some public programmes.

4.4 Socio-demographic and Regional Variables

4.4.1 Religion

There are four religion dummy variables corresponding to Hindu, Muslim, Christian and Other religions, of which ‘Hindu’ is taken as the reference category. This variable is included to capture another aspect of dietary diversity, such as the consumption of plant-based food in comparison to animal-based food products, which is not adequately captured by the dietary diversity variable. Although there is information in the data on the consumption of different food items, many of these are in aggregated groups; for example, meat, eggs, chicken and fish are all grouped into a single category. This makes conversion of food items into its nutrient content difficult; also, due to this aggregation a large number of households report consumption in value terms. The majority of Hindus are vegetarians while a majority of Muslims and Christians in India consume non-vegetarian food. Animal-based protein intake is known to have a better impact on nutritional status and, therefore, one could expect a better nutrition status among Muslim and Christian women than their Hindu counterparts. Of course, the frequency and adequacy of diets is important, but one would expect lower frequency and per capita consumption to be correlated with the economic status of the household capturing affordability and also the prosperity of the region, which would improve regular accessibility. Since economic status is already controlled for through several other variables and regional (state) dummy variables are used (which would in a broad sense control for accessibility), we expect these coefficients to indicate the net effects. However, if such dietary habits are not latently captured by the religion dummy variable, one can expect the coefficients to be not significantly different from each other.

Table 2 below shows that mean and standard deviation of women’s BMI do not vary substantially across the religious groups though Hindus have the lowest mean value followed by Muslims and Christians while the other religions have an intermediate value between Hindus and Muslims. Since these are pure effects not controlling for social status, one can expect such a pattern of variation across religious groups.

Table 2
Mean and Standard Deviation of Women’s BMI (kg/m²) Across Religious Groups

Religious Group Mean Standard Deviation

Hindu 21.0 3.62

Muslim 21.6 3.77

Christian 22.6 3.77

Other religions 21.8 3.99

Religious Group	Mean	Standard Deviation
Hindu	21.0	3.62
Muslim	21.6	3.77
Christian	22.6	3.77
Other religions	21.8	3.99

Source: Authors’ estimations.

4.4.2 Caste

Indian society has been divided into various castes (the varna system of the Hindus) on the basis of their occupation since ancient times. The relegation of menial jobs to some social groups, with limited or no access to productive resources, and persistent discrimination in various other domains of social and economic status have created high socioeconomic disparity among different groups in Indian society. Persistent disparity and repression from the so-called ‘upper caste’ people have had adverse impacts on well-being, including health. This could be reflected in a low BMI among the ‘lower castes’ and a high percentage of them suffering from CED.

This is borne out by the results in Table 3, which show that the mean BMI of Brahmin women is 22.4 while women belonging to the Scheduled Tribes have a mean BMI of 20.4. The share of women belonging to the Brahmin caste with a BMI of less than 18.5 is 14.8 per cent, while the percentage of women belonging to a Scheduled Tribe with a BMI of less than 18.5 is 28.8 per cent (not reported in the Table below).

Table 3
Mean and Standard Deviation of Women’s BMI (kg/m²) Across Caste Groups

Caste Groups Mean Standard Deviation

Brahmins 22.4 4.02

Other Backward Castes (OBCs) 21.1 3.60

Scheduled Caste (SC) 20.4 3.33

Scheduled Tribe (ST) 20.0 2.89

Others 22.0 3.95

Caste Groups	Mean	Standard Deviation
Brahmins	22.4	4.02
Other Backward Castes (OBCs)	21.1	3.60
Scheduled Caste (SC)	20.4	3.33
Scheduled Tribe (ST)	20.0	2.89
Others	22.0	3.95

Source: Authors’ estimations.

4.4.3 Household Composition Variables

This relates to the proportion of people in the following age groups in a household:

0–4 years (nf1)

5–14 years (nf2)

15–60 years (nf3)

above 60 years (nf4)

Since these proportions add up to 1, the first proportion is excluded and is taken as equivalent to the reference category, as in the case of categorical variables. If a household has a larger share of people in the first category, that is, 0 to 4 years, it requires the woman to make a significantly greater effort to take care of the children. This consumes considerable energy, and hence, her BMI is expected to be comparatively lower. Also, it implies that her BMI is negatively affected by frequent pregnancies, which consequently leads to other kinds of deficiencies, such as iron deficiency causing anaemia. Hence, the three variables (nf2, nf3 and nf4) are expected to have a positive sign.

A woman who belongs to a household with a high proportion of people in the third category, that is, the adult category, her BMI is expected to be relatively better than women belonging to households in the other three categories. It is however important to control for household size while the effect of household composition is analyzed, as a four-member household can have three adults and one child or two adults and one child, which would affect the women’s BMI status differently. Thus, household size is also included in the model, but we expect it to be insignificant for most quantiles.

Alongside the variables mentioned above, regional variables like rural/urban residence and the state of current residence are also controlled for as categorical variables.

4.5 Individual Variables

4.5.1 Age

Age can have a negative or a positive impact on the BMI of women, since the index shows changes with time. Younger generations would enjoy the benefits of economic development and improvements in access to infrastructure, which would reflect in better nutritional status, and hence, coefficient of age would be positive. Or the age coefficient could be positive, showing that older women tend to improve their BMI as younger women would tend to have a lower BMI due to reproductive and child-rearing phases perhaps resulting in an adverse impact on their body weight, in addition to other household chores and drudgeries most women face. Finally, one can also expect the coefficient to be insignificant if the explanatory variables capturing these aspects are already factored into the model adequately.

Table 4
Mean and Standard Deviation of Women’s BMI (kg/m²) Across Education Levels

Completed Education Levels Mean Standard Deviation

Not literate 20.4 3.26

Primary 21.4 3.71

Secondary 22.1 3.86

Higher secondary 22.3 4.07

Graduation 23.5 4.14

Completed Education Levels	Mean	Standard Deviation
Not literate	20.4	3.26
Primary	21.4	3.71
Secondary	22.1	3.86
Higher secondary	22.3	4.07
Graduation	23.5	4.14

Source: Authors’ estimations.

4.5.2 Education Levels

This is a categorical variable with five groups: not literate, studied up to primary, or secondary, or higher secondary, or college-level education. The reference category in the model is chosen as ‘not literate’, and the mean BMI is expected to improve with education (controlling for other factors) since an educated women is more likely to be conscious about her health and well-being. She is more likely to diversify her diet to get better nutrition and does not support frequent pregnancies. Table 4 presents the mean and standard deviation of the BMI of women for different education levels, which rises consistently with a rise in the education level.

4.5.3 Women’s Occupation Status

There are five dummy variables relating to a woman’s occupation:

Self-employed only in agriculture (farm work and animal rearing).

Self-employed in agriculture and sometimes engaged as agricultural wage labour.

Working only as agricultural wage labour.

Working as non-agricultural labour.

Salaried worker or a business women, referred as ‘other’ work.

Not actively engaged in an economic activity in the labour market.

The reference category chosen in the econometric model is the last one consisting of women who do not actively participate in the labour market and are involved in domestic work. The signs of the coefficients of the categorical variables may be ambiguous and cannot be predicted a priori. It can however be expected that women who are involved in household chores alone are likely to have lower physical activity than those involved in agricultural work or non-agricultural labour work. Women working as agricultural labourers or who are self-employed in agriculture spend a lot of energy doing hard manual work on the farm and receive meagre wages which are spent on the necessities of the household. In fact, we expect women in the occupational category (ii) to have the lowest levels of BMI, followed by those in categories (i), and then (iii) and (iv), which could have near-similar BMI magnitudes. However, working women are expected to be more empowered and their access to a cash income could make a better-quality diet more affordable, which would be the case for women in the ‘other’ work— category (v). Furthermore, women involved in other work could also have lower levels of physical activity. So compared to the reference category, the coefficient for this group could be insignificant, especially in the QRM for the lower quantiles.

Table 5
Mean and Standard Deviation of BMI (kg/m²) Across Work Status

Work Status Mean Standard Deviation

Self-employed in agriculture alone (seag) 20.5 3.28

Self-employed in agriculture and agricultural wage labour (seaglab) 19.6 2.69

Agricultural wage labour (aglab) 20.2 3.03

Non-agricultural wage labour (naglab) 20.6 3.24

Other work includes salaried and business (othwk) 22.1 4.07

Not participating in the labour market 22.0 3.90

Work Status	Mean	Standard Deviation
Self-employed in agriculture alone (seag)	20.5	3.28
Self-employed in agriculture and agricultural wage labour (seaglab)	19.6	2.69
Agricultural wage labour (aglab)	20.2	3.03
Non-agricultural wage labour (naglab)	20.6	3.24
Other work includes salaried and business (othwk)	22.1	4.07
Not participating in the labour market	22.0	3.90

Source: Authors’ own estimations.

Note: Work status refers to the type of work carried out by women in the labour market.

The results of the model show that women who do not work have the highest mean BMI, that is, 22, and women who work as farm workers or agricultural wage labour have the lowest mean BMI, that is, 19.6, as shown in Table 5. Expectedly, the percentage of women suffering from CED is highest in the former category while those who stay at home have the lowest percentage of CED.

4.5.4 Women’s Short-Term Morbidity Status

We use short-term morbidity status captured by the number of days a woman is ill with fever and so on prior to the survey in the previous month, so that we expect that this could have resulted in loss of body weight, and hence, a lower average BMI even after controlling for other variables.

4.5.5 Women’s Pregnancy Status

Similarly, a pregnant woman is expected to have a somewhat higher-average BMI compared to others and has been used as a control variable.

4.6 Women’s Autonomy Index

This index captures the autonomy enjoyed by a woman within the household. It is created by considering various aspects of empowerment, such as whether she has a bank account in her name and cash in hand and whether she is consulted at the time of making important expenditures, whether small ones, like treating a child’s sickness, or large expenditures, like a child’s marriage. Other variables reflecting her freedom of thought and actions, such as permission to visit a friend, health centre or kirana shop, are also taken into account while estimating this index, using principal component analysis, with the first principal component being taken as the index. This index is further classified into five quintiles to create categories with increasing levels of autonomy. The bottom quintile corresponds to the level with the lowest autonomy level, and higher levels indicate greater autonomy in decision-making and so on. The first quintile of the autonomy index is taken as the reference category in the regression models.

A woman who is better empowered is less dependent on others in making her decisions and, hence, has more freedom and access to resources to take care of her health. Given this, coefficients of the higher quintiles of the autonomy index are expected to have a positive sign. Table 6 shows that the mean BMI of women increases marginally from the first level to the second level but there is very little variation after that, even though the magnitude increases. The standard deviation also does not show the kind of increases that were noticed for some of the other categorical variables, but it does increase in magnitude with the level of autonomy. We expect the autonomy to be correlated with socioeconomic status and also with education levels. After controlling for these variables, it could also be that this coefficient is largely insignificant across BMI quantiles in the QRM, but as can be observed from the values of the standard deviations, there could be substantial variation in women’s status even among the richer sections of the population as well as among the educated, given that socioculturally decision-making among Indian households is largely by the adult male members of the household.

Table 6
Mean and Standard Deviation of BMI (kg/m²) Across Levels of ‘Autonomy’

‘Autonomy’ Level Mean Standard Deviation

Autonomy_1 20.9 3.51

Autonomy_2 21.3 3.75

Autonomy_3 21.2 3.78

Autonomy_4 21.7 3.68

Autonomy_5 21.6 3.96

‘Autonomy’ Level	Mean	Standard Deviation
Autonomy_1	20.9	3.51
Autonomy_2	21.3	3.75
Autonomy_3	21.2	3.78
Autonomy_4	21.7	3.68
Autonomy_5	21.6	3.96

Source: Authors’ estimations.

Note: ‘Autonomy’ refers to the woman’s choice and freedom in decision-making for the family as well as for herself. This is an index and the five levels are the five quintiles of this index that capture increasingly greater levels of ‘autonomy’.

5. RESULTS: QUANTILE REGRESSION ESTIMATES

The estimates of the quantile regression as well as the OLS regression are given in Table 7 with the explanatory variables described in Table A1. The OLS regression estimates show that almost all variables in the five categories—individual, household, dietary, social infrastructure and women empowerment—are significant in determining women’s BMI. All the variables representing the economic status of the household, that is, the log of monthly consumption expenditure, log of per capita income and all the wealth quintiles are significant in the OLS estimates. The quantile estimates are also similar and for the richest wealth quintile we notice a significant increase in the impact on the estimated coefficients.

As mentioned earlier, we incorporate two different aspects of food consumption variables in the analysis. Food expenditure share coefficients are all significant and positive, indicating a quality effect, although from Engel’s Law we could observe that the estimated coefficient could also be negative, as households with better incomes could have a lower expenditure share on food. After controlling for food expenditure share (and other socioeconomic variables) we observe that the coefficient on the dietary diversity index is negative and significant, which is an expected result. Lower values of the index indicate greater diversity in the food basket, are related to higher values of lnbmi and, hence, there is an inverse relationship. The estimated coefficient has lower magnitudes for the bottommost (0.1) and topmost (0.9) quintiles, compared to the middle quintiles, clearly signifying that diets are not that diversified for both the under- and over-nourished.

Table 7
OLS and Quantile Regression Estimates with lnBMI as the Dependent Variable

Quantiles
OLS
0.1
0.25
0.5
0.75
0.9

Variable Coefficient p value Coefficient p value Coefficient p value Coefficient p value Coefficient p value Coefficient p value

LnCOPC 0.021*** 0.000 0.010*** 0.010 0.015*** 0.000 0.018*** 0.000 0.020*** 0.000 0.019*** 0.000

lnPCI 0.005** 0.012 0.006*** 0.002 0.009*** 0.000 0.006*** 0.000 0.007*** 0.000 0.005** 0.026

Poor 0.008** 0.028 0.002 0.624 0.002 0.707 0.008** 0.026 0.004 0.298 –0.010** 0.036

Midwlt 0.010* 0.060 0.002 0.627 0.002 0.730 0.008** 0.029 0.005 0.224 0.006 0.264

Rich 0.025*** 0.000 0.017*** 0.003 0.018*** 0.000 0.025*** 0.000 0.024*** 0.000 0.025*** 0.000

Richest 0.045*** 0.000 0.039*** 0.000 0.039*** 0.000 0.049*** 0.000 0.047*** 0.000 0.046*** 0.000

fdexp_sh 0.041*** 0.001 0.028** 0.024 0.033*** 0.003 0.021** 0.021 0.023** 0.033 0.009 0.492

Dietdiv –0.046** 0.048 –0.038** 0.045 –0.052*** 0.002 –0.044*** 0.001 –0.033** 0.050 –0.038* 0.065

Qwtr 0.014*** 0.000 0.018*** 0.000 0.018*** 0.000 0.012*** 0.000 0.006* 0.076 0.001 0.735

Vwtr 0.009** 0.015 0.003 0.515 0.006 0.150 0.007** 0.020 0.004 0.258 0.001 0.798

wtr_grp1 0.007** 0.024 0.003 0.403 0.006* 0.065 0.006** 0.018 0.010*** 0.002 0.011*** 0.006

wtr_grp2 0.008** 0.039 0.008* 0.061 0.007* 0.059 0.002 0.482 0.003 0.458 0.005 0.229

Ckfuel 0.011*** 0.003 0.006 0.113 0.004 0.274 0.005* 0.066 0.006* 0.082 0.012*** 0.003

san_grp2 0.028*** 0.000 0.023*** 0.000 0.028*** 0.000 0.034*** 0.000 0.035*** 0.000 0.040*** 0.000

san_grp3 0.019*** 0.000 0.013** 0.041 0.026*** 0.000 0.024*** 0.000 0.021*** 0.000 0.021*** 0.001

san_grp4 0.031*** 0.000 0.028*** 0.000 0.030*** 0.000 0.028*** 0.000 0.030*** 0.000 0.028*** 0.000

own_elect 0.004 0.300 0.004 0.357 0.010*** 0.012 0.007** 0.022 0.014*** 0.000 0.015*** 0.001

Muslim 0.009*** 0.020 –0.003 0.486 0.006 0.153 0.018*** 0.000 0.024*** 0.000 0.027*** 0.000

Christian 0.014* 0.075 0.007 0.490 0.010 0.274 0.016** 0.019 0.009 0.295 0.019* 0.053

Oth relg 0.016*** 0.005 0.014* 0.068 0.007 0.316 0.012** 0.035 0.011* 0.103 0.019** 0.020

UCH 0.026*** 0.000 0.008 0.321 0.015** 0.048 0.020*** 0.001 0.026*** 0.000 0.045*** 0.000

OBC 0.014*** 0.007 0.005 0.376 0.005 0.349 0.009** 0.033 0.010* 0.065 0.013** 0.037

SC –0.005 0.378 –0.010 0.131 –0.008 0.160 0.0004 0.929 0.002 0.771 0.004 0.497

Oth Caste 0.018*** 0.001 0.011* 0.100 0.008 0.153 0.011** 0.017 0.019*** 0.001 0.022*** 0.001

nf2 0.026*** 0.010 0.008 0.529 0.020* 0.066 0.036*** 0.000 0.046*** 0.000 0.031** 0.018

nf3 0.038*** 0.001 0.021* 0.099 0.029** 0.011 0.050*** 0.000 0.048*** 0.000 0.032** 0.020

nf4 0.053*** 0.001 0.027 0.138 0.040** 0.015 0.062*** 0.000 0.065*** 0.000 0.060*** 0.003

Npersons 0.001* 0.070 0.001 0.209 0.001*** 0.007 0.002*** 0.000 0.001** 0.025 0.0004 0.569

Age 0.004*** 0.000 0.002*** 0.000 0.003*** 0.000 0.003*** 0.000 0.004*** 0.000 0.005*** 0.000

ed_prim 0.012*** 0.001 0.001 0.777 0.006* 0.081 0.005* 0.064 0.005* 0.084 0.006* 0.084

ed_sec 0.012*** 0.007 –0.004 0.432 0.001 0.854 0.005 0.122 0.007* 0.094 0.006 0.282

ed_hsc 0.006 0.366 –0.014** 0.035 –0.008 0.171 –0.004 0.399 0.001 0.806 0.003 0.724

ed_grad 0.021*** 0.001 0.021*** 0.004 0.010 0.144 0.008 0.116 0.011* 0.087 0.010 0.201

Seag –0.028*** 0.000 –0.012*** 0.002 –0.013*** 0.000 –0.019*** 0.000 –0.015*** 0.000 –0.025*** 0.000

Seaglab –0.036*** 0.000 –0.015** 0.017 –0.023*** 0.000 –0.033*** 0.000 –0.035*** 0.000 –0.052*** 0.000

Aglab –0.013** 0.020 –0.016** 0.013 –0.018*** 0.002 –0.018*** 0.000 –0.020*** 0.000 –0.043*** 0.000

Naglab –0.025*** 0.000 –0.008 0.336 –0.023*** 0.002 –0.030*** 0.000 –0.036*** 0.000 –0.038*** 0.000

Othwrk –0.013** 0.019 –0.010* 0.075 –0.011** 0.034 –0.013*** 0.002 –0.007 0.153 –0.008 0.210

Shmorbd –0.002*** 0.000 –0.002*** 0.000 –0.002*** 0.000 –0.002*** 0.000 –0.002*** 0.000 –0.001** 0.023

Preg 0.044*** 0.000 0.052*** 0.000 0.049*** 0.000 0.039*** 0.000 0.041*** 0.000 0.032*** 0.000

autnmy_2 –0.002 0.747 –0.017* 0.063 –0.006 0.479 –0.001 0.857 0.001 0.931 0.003 0.796

autnmy_3 –0.006* 0.072 0.001 0.844 0.003 0.352 –0.00003 0.992 –0.001 0.733 –0.004 0.380

autnmy_4 0.005 0.148 0.018*** 0.000 0.015*** 0.000 0.013*** 0.000 0.004 0.260 0.003 0.455

autnmy_5 0.0002 0.978 0.005 0.295 0.013*** 0.001 0.017*** 0.000 0.017*** 0.000 0.015*** 0.001

Urban 0.022*** 0.000 0.006 0.129 0.008** 0.023 0.014*** 0.000 0.030*** 0.000 0.036*** 0.000

Intercept 2.68*** 0.000 2.63*** 0.000 2.59*** 0.000 2.66*** 0.000 2.70*** 0.000 2.84*** 0.000

Quantiles	OLS	0.1	0.25	0.5	0.75	0.9
LnCOPC	0.021***	0.000	0.010***	0.010	0.015***	0.000	0.018***	0.000	0.020***	0.000	0.019***	0.000
lnPCI	0.005**	0.012	0.006***	0.002	0.009***	0.000	0.006***	0.000	0.007***	0.000	0.005**	0.026
Poor	0.008**	0.028	0.002	0.624	0.002	0.707	0.008**	0.026	0.004	0.298	–0.010**	0.036
Midwlt	0.010*	0.060	0.002	0.627	0.002	0.730	0.008**	0.029	0.005	0.224	0.006	0.264
Rich	0.025***	0.000	0.017***	0.003	0.018***	0.000	0.025***	0.000	0.024***	0.000	0.025***	0.000
Richest	0.045***	0.000	0.039***	0.000	0.039***	0.000	0.049***	0.000	0.047***	0.000	0.046***	0.000
fdexp_sh	0.041***	0.001	0.028**	0.024	0.033***	0.003	0.021**	0.021	0.023**	0.033	0.009	0.492
Dietdiv	–0.046**	0.048	–0.038**	0.045	–0.052***	0.002	–0.044***	0.001	–0.033**	0.050	–0.038*	0.065
Qwtr	0.014***	0.000	0.018***	0.000	0.018***	0.000	0.012***	0.000	0.006*	0.076	0.001	0.735
Vwtr	0.009**	0.015	0.003	0.515	0.006	0.150	0.007**	0.020	0.004	0.258	0.001	0.798
wtr_grp1	0.007**	0.024	0.003	0.403	0.006*	0.065	0.006**	0.018	0.010***	0.002	0.011***	0.006
wtr_grp2	0.008**	0.039	0.008*	0.061	0.007*	0.059	0.002	0.482	0.003	0.458	0.005	0.229
Ckfuel	0.011***	0.003	0.006	0.113	0.004	0.274	0.005*	0.066	0.006*	0.082	0.012***	0.003
san_grp2	0.028***	0.000	0.023***	0.000	0.028***	0.000	0.034***	0.000	0.035***	0.000	0.040***	0.000
san_grp3	0.019***	0.000	0.013**	0.041	0.026***	0.000	0.024***	0.000	0.021***	0.000	0.021***	0.001
san_grp4	0.031***	0.000	0.028***	0.000	0.030***	0.000	0.028***	0.000	0.030***	0.000	0.028***	0.000
own_elect	0.004	0.300	0.004	0.357	0.010***	0.012	0.007**	0.022	0.014***	0.000	0.015***	0.001
Muslim	0.009***	0.020	–0.003	0.486	0.006	0.153	0.018***	0.000	0.024***	0.000	0.027***	0.000
Christian	0.014*	0.075	0.007	0.490	0.010	0.274	0.016**	0.019	0.009	0.295	0.019*	0.053
Oth relg	0.016***	0.005	0.014*	0.068	0.007	0.316	0.012**	0.035	0.011*	0.103	0.019**	0.020
UCH	0.026***	0.000	0.008	0.321	0.015**	0.048	0.020***	0.001	0.026***	0.000	0.045***	0.000
OBC	0.014***	0.007	0.005	0.376	0.005	0.349	0.009**	0.033	0.010*	0.065	0.013**	0.037
SC	–0.005	0.378	–0.010	0.131	–0.008	0.160	0.0004	0.929	0.002	0.771	0.004	0.497
Oth Caste	0.018***	0.001	0.011*	0.100	0.008	0.153	0.011**	0.017	0.019***	0.001	0.022***	0.001
nf2	0.026***	0.010	0.008	0.529	0.020*	0.066	0.036***	0.000	0.046***	0.000	0.031**	0.018
nf3	0.038***	0.001	0.021*	0.099	0.029**	0.011	0.050***	0.000	0.048***	0.000	0.032**	0.020
nf4	0.053***	0.001	0.027	0.138	0.040**	0.015	0.062***	0.000	0.065***	0.000	0.060***	0.003
Npersons	0.001*	0.070	0.001	0.209	0.001***	0.007	0.002***	0.000	0.001**	0.025	0.0004	0.569
Age	0.004***	0.000	0.002***	0.000	0.003***	0.000	0.003***	0.000	0.004***	0.000	0.005***	0.000
ed_prim	0.012***	0.001	0.001	0.777	0.006*	0.081	0.005*	0.064	0.005*	0.084	0.006*	0.084
ed_sec	0.012***	0.007	–0.004	0.432	0.001	0.854	0.005	0.122	0.007*	0.094	0.006	0.282
ed_hsc	0.006	0.366	–0.014**	0.035	–0.008	0.171	–0.004	0.399	0.001	0.806	0.003	0.724
ed_grad	0.021***	0.001	0.021***	0.004	0.010	0.144	0.008	0.116	0.011*	0.087	0.010	0.201
Seag	–0.028***	0.000	–0.012***	0.002	–0.013***	0.000	–0.019***	0.000	–0.015***	0.000	–0.025***	0.000
Seaglab	–0.036***	0.000	–0.015**	0.017	–0.023***	0.000	–0.033***	0.000	–0.035***	0.000	–0.052***	0.000
Aglab	–0.013**	0.020	–0.016**	0.013	–0.018***	0.002	–0.018***	0.000	–0.020***	0.000	–0.043***	0.000
Naglab	–0.025***	0.000	–0.008	0.336	–0.023***	0.002	–0.030***	0.000	–0.036***	0.000	–0.038***	0.000
Othwrk	–0.013**	0.019	–0.010*	0.075	–0.011**	0.034	–0.013***	0.002	–0.007	0.153	–0.008	0.210
Shmorbd	–0.002***	0.000	–0.002***	0.000	–0.002***	0.000	–0.002***	0.000	–0.002***	0.000	–0.001**	0.023
Preg	0.044***	0.000	0.052***	0.000	0.049***	0.000	0.039***	0.000	0.041***	0.000	0.032***	0.000
autnmy_2	–0.002	0.747	–0.017*	0.063	–0.006	0.479	–0.001	0.857	0.001	0.931	0.003	0.796
autnmy_3	–0.006*	0.072	0.001	0.844	0.003	0.352	–0.00003	0.992	–0.001	0.733	–0.004	0.380
autnmy_4	0.005	0.148	0.018***	0.000	0.015***	0.000	0.013***	0.000	0.004	0.260	0.003	0.455
autnmy_5	0.0002	0.978	0.005	0.295	0.013***	0.001	0.017***	0.000	0.017***	0.000	0.015***	0.001
Urban	0.022***	0.000	0.006	0.129	0.008**	0.023	0.014***	0.000	0.030***	0.000	0.036***	0.000
Intercept	2.68***	0.000	2.63***	0.000	2.59***	0.000	2.66***	0.000	2.70***	0.000	2.84***	0.000

Source: Authors’ estimates.

Notes: *Denotes significance at 10 per cent level.

**Denotes significance at 5 per cent level.

***Denotes significance at 1 per cent level.

Social infrastructure variables like access to LPG fuel, electricity and good-quality sanitation (toilet) facilities at home also show a significant positive impact on the BMI of women. Access to regular piped water also influences women’s health significantly if it is available inside the house or if she does not waste too much time in fetching water. Access to cleaner cooking fuels (like LPG or kerosene) in terms of reducing the indoor air pollution also seems to be an important factor in determining the BMI of women. But it is important to note that this is not the case for the lower quantiles and one observes the effect only after the 50th percentile. Perhaps most households with women having low BMI primarily use the less clean cooking fuel and a lack of variability in the variable for this quintile results in the choice of cooking fuel coefficient being insignificant in the econometric results. However, women who use more polluting cooking fuels among those in the upper quantiles of the BMI are definitely disadvantaged compared to their counterparts, even though economically they may be able to afford better cooking fuel, but perhaps regular access to it may be an issue.

Households without a proper sanitation facility are at a higher risk of suffering from communicable diseases. This reduces a woman’s resistance further, with a negative impact on her health. The quantile estimates show clearly that all types of toilet facilities are significantly better than no facility (mainly open defecation). The results once again highlight that women in the upper lnbmi quantiles are more disadvantaged with respect to having no toilet facility. There are significant improvements when moving from a household with no facility, to a traditional to a flush toilet in the lower quantiles and this is an important result to take note of.

Electricity is one basic infrastructural facility provided by the government. Access to electricity has a significant positive impact on the BMI of women in the upper quantiles and is not significant in either the OLS model or for the bottom quantile. This could be because electricity has an amenity value and access to a regular, clean source of lighting and the possibility of using cooking and cleaning-related gadgets reduces the drudgery of household chores among those who can afford such appliances. Hence, among women in the upper quantiles, the disadvantaged are those who do not have access to electricity.

Women belonging to the Muslim, Christian and Other religions have a significantly higher BMI than Hindu women. As explained earlier, dietary habits vary across religious groups, and we expect that to be reflected in variations in the average lnbmi. Even though some aspect of dietary diversity was included in the econometric model, aspects of the quality of diets in terms of vegetarian versus non-vegetarian may not have been adequately reflected in the dietary diversity index. The caste structure in India originated among the Hindus and has become entrenched in a rigid socioeconomic hierarchy, yet we observe that even after controlling for several socioeconomic and regional variables, socially disadvantaged women from the Scheduled Tribes and Scheduled Castes are worse off in terms of their BMI. More importantly, Brahmins in most part of India are predominantly vegetarian, yet we observe that even after religious affiliation is controlled for, women in such households benefit substantially health-wise from belonging to this caste, with the exception of the lowest (0.1) quantile where caste is not salient similar to religion. However, the increase in average lnbmi is also substantial for the topmost quantile (0.9), which would also signify overweight and obesity, given that the average lnbmi is higher as noted from the intercept value for this group.

The number of persons in a household as well as household composition dummies are significant. Household size impacts more in the middle quantiles, whereas in the lowest and topmost quantiles no impact is observed. Household composition coefficients show that women with younger children (0–4 years) at home spend considerable energy taking care of them, which affects their BMI adversely. This is observed across the quantiles, but it is equally important to observe that even after controlling for household size, women in the lowest lnbmi quantile (0.1) are disadvantaged even with somewhat older children (5–14 years). Age has a significant impact on the BMI of women in the OLS regression as well as the quantile regression: older women have higher BMI than younger women and this is true across all quantiles; however, once again, the possibility of older women with high levels of BMI is also a matter of concern.

The education dummies, except for higher secondary level of education, are positive and significant for the OLS estimates, but many of these coefficients are not significant across the quantiles. Wherever they are significant, women with a higher level of education have a better average lnbmi but with very small changes in the estimated coefficients. The women’s occupation dummy coefficients are all negative and significant indicating that women who are engaged only in household activities have a higher-average lnbmi than women who are involved in labour market activities as well. The reasons are that most women are likely to be engaged in manual work related to agriculture or non-agriculture activities which, alongside the daily drudgery of household activities with limited amenities, results in the use of lots of physical energy but with limited ways to support this with adequate dietary intake and health inputs. Women working in the formal and organized service sector form a very small segment and, given their lifestyle, the regular salaried or business occupation dummy is insignificant in the 75th and 90th quantile, so that they are no different from the reference group. However, for the lower quantiles we find that even these women have a lower-average lnbmi after controlling for various other factors.

Short-term morbidity captured by the number of sick days (with the average being about three) is negative and significant for all quantiles and with very similar magnitude. We would have expected this coefficient to be insignificant in the upper quantile as women are healthier and, hence, their chance of losing weight during short illnesses would be lower. Similarly, pregnant women would be expected to have a higher BMI, and the fact that the magnitude of the coefficients does not vary much across quantiles is somewhat unexpected.

The autonomy index captures aspects of a woman’s empowerment so she has greater freedom and involvement in decision-making on important issues, such as going to the hospital when ill, spending on her diet and so on. The women’s empowerment index does not show the expected results, in that only the higher levels of the autonomy index are significant. Women living in urban areas have significantly higher BMIs than women living in rural areas throughout the distribution of the BMI. The state dummy coefficients are also significant and the results are not reported here. Despite controlling for several socioeconomic variables that could explain variations in lnbmi of women, the state of residence still explains a lot of variation.

6. CONCLUSION

In this study the factors affecting the BMI of women in India have been analyzed, focusing on three different types of economic variables and other socio-demographic variables, including a women’s empowerment index. The BMI captures both undernutrition and overnutrition. A QRM and the quantile regression estimates from this study highlight variations in the impacts of different explanatory variables across the distribution of the BMI. For instance, the coefficients of dietary diversity, women’s work status and water quality have clear variation in their impacts across the quantiles. In contrast, the women’s autonomy index has an impact only at the higher levels, indicating that after other variables are controlled for, only a fairly high level of autonomy makes an impact, and there is some variation in the impact across the quantiles.

Footnotes

Appendix

Table A1

Summary of the Explanatory Variables

Variable Name	Type	Description
Economic variables
LnCOPC	Continuous	Log of monthly consumption expenditure by household
LnPCI	Continuous	Log of monthly per capita income
Poor, mid-wealth, rich and richest	Categorical	Five quintiles of wealth index created through principal component analysis; the poorest quintile is taken as reference category
Dietary variables
fdexp_sh	Continuous	Proportion of food expenditure in total household expenditure
Dietdiv	Continuous	Dietary diversity index
Social infrastructure variables
Ckfuel	Categorical	Access to proper LPG facility for cooking
wtr_grp1, wtr_grp2	Categorical	Dummy for access to regular piped water inside (wtr_grp1), outside but consumes less time (wtr_grp2). Base category is time taken to fetch water is more than 20 minutes
Qwtr	Categorical	Whether water is purified or not (reference group)
Vwtr	Categorical	Whether water covered with lid or not (reference group)
san_grp2, san_grp3, San_grp4	Categorical	Dummy for sanitation whether it is a flush toilet (san_grp4), pit latrine (san_grp3), traditional toilets (san_grp2); reference category is taken as ‘no facility’ or open defecation
own_elec	Categorical	Access to electricity or not
Socio-demographic and regional variables
Muslim, Christian, Others	Categorical	Dummies for religious groups: Muslims, Christians and Others, with Hindus as the reference category.
UCH, OBC, SC, Oth Caste	Categorical	Dummies for caste groups: upper-caste Hindus (UCH), Other Backward Castes (OBC), Scheduled Castes (SC) and Other Castes (Oth Caste) with Scheduled Tribes (ST) as the reference.
nf2, nf3, nf4	Continuous	Dummies for composition of family according to age groups as mentioned in Section 4.4.3; nf1 is taken as base category
Npersons	Continuous	Number of persons in family
Urban	Categorical	Belonging to urban area with rural as the reference category
Individual variables
Age	Continuous	Age of woman
Ed_prim, ed_sec, ed_hsc, ed_grad	Categorical	Completed level of education: primary, secondary, higher secondary and graduation and above; not literate is the base category
Seag, seaglab, aglab, naglab, othrwrk	Categorical	Farm work (seag), farm work and agricultural wage labour (seaglab), agricultural wage labour (aglab), non-agricultural wage labour (naglab), salaried work and business (othwrk); not in the labour market is the reference group.
Preg	Categorical	Dummy for pregnant woman; non-pregnant woman is reference category
Shmorbd	Continuous	Number of days sick capturing the effect of short-term morbidity
Women’s empowerment variables
autonomy_2, autonomy_3, autonomy_4, autonomy_5	Categorical	Quantiles of women autonomy index estimated through principal component analysis, with the first category taken as the reference. Captures ‘freedom’ aspects of decision-making within the household and mobility

Source: Authors’ own summary.

Acknowledgements

This study started as part of the summer internship of the first author at Madras School of Economics (MSE) supported by the Centre for Development Economics, Delhi School of Economics, Delhi, which was then extended as a master’s thesis at MSE. We thank J.V. Meenakshi, Zareena Begum and Umakant Dash for their valuable comments in improving the quality of the discussions. This article was subsequently presented at the IHDS user’s conference at New Delhi 20–21 June 2013, and we would like to thank the seminar participants, in particular Sonalde Desai, for useful comments. We thank Getsie David for assisting us in revising some of the results based on comments made at the conference. The responsibility of the errors, if any, remains with the authors.

Notes

References

Ackerson

L. K.

, Kawachi

, Barbeau

E. M.

, & Subramanian

S. V.

(2008). Geography of underweight and overweight among women in India: A multilevel analysis of 3,204 neighborhoods in 26 states. Economics and Human Biology, 6(2), 264–80.

Behrman

J. R.

, & Deolalikar

(1987). Will developing country nutrition improve with income? A case study for rural South India. Journal of Political Economy, 95(3), 108–138.

Bhagowalia

, Menon

, Quisumbing

, & Soundararajan

(2012). What dimensions of women’s empowerment matter most for child nutrition? (IFPRI Discussion Paper No. 01192). Washington, D.C.: International Food Policy Research Institute.

Deaton

(2008). Height, health, and inequality: The distribution of adult heights in India. American Economic Review Papers and Proceedings, 98(2), 468–74.

Desai

, Dubey

, Joshi

B. L.

, Sen

, Shariff

, & Vanneman

(2007). India Human Development Survey (IHDS) [Computer file]. ICPSR22626-v2. University of Maryland and National Council of Applied Economic Research, New Delhi [producers]; Inter-university Consortium for Political and Social Research, Ann Arbor, MI [distributor].

Hindin

(2005). Women’s autonomy, status, and nutrition in Zimbabwe, Zambia, and Malawi. In Kishor

(Ed.), A focus on gender—Collected papers on gender using DHS data (pp. 93–115). Calverton, MD: US Agency for International Development and ORC Macro.

Jose

(2014). Are Indian adults predisposed to low body mass index? (Mimeograph). Hyderabad, India: Tata Institute of Social Sciences.

Jose

, & Navaneetham

(2010). Social infrastructure and women’s undernutrition. Economic & Political Weekly, 45(13), 83–89.

Kumar

K. S. K.

, Ramachandran

, & Viswanathan

(2009). What characterises women vulnerable to chronic energy deficiency? Social Indicator Research, 90(3), 365–80.

10.

Mason

, Bailes

, Beda-Andourou

, Copeland

, Curtis

, Deitchler

, . (2005). Recent trends in malnutrition in developing regions: Vitamin A deficiency, anaemia, iodine deficiency, and child underweight. Food and Nutrition Bulletin, 26(1), 59–162.

11.

Navaneetham

, & Jose

(2008). A factsheet on women’s malnutrition in India. Economic & Political Weekly, 43(33), 61–67.

12.

Panagariya

(2012, 1 December). Once we do our malnutrition numbers correctly, we will find that India has no more to be ashamed of its malnutrition level. Interview with Jain

Akshai

. Tehelka Magazine, 9(48). Retrieved from http://archive.tehelka.com/story_main54.asp?filename–e011212Once.asp (accessed on 1 December 2012 ).

13.

Sen

(1990, 20 December). More than 100 million women are missing. The New York Review of Books, 37(20). Retrieved from http://www.nybooks.com/articles/archives/1990/dec/20/more-than-100-million-women-are-missing/ (accessed on 15 January 2015 ).

14.

Seshadri

(2009). Nutritional outcomes and well-being of women in India. Unpublished master’s thesis, Madras School of Economics, Chennai, India.

15.

Singh

, Srinivas

, & Sekher

T. V.

(2011). Double burden of nutritional disorder among Indian women: An assessment of differentials and determinants. Indian Journal of Maternal and Child Health, 13(4).

16.

Smith

, & Haddad

(2000). Explaining childhood malnutrition in developing countries (IFPRI Research Report No. 111). Washington, D.C.: International Food Policy Research Institute.

17.

Subramanian

S. V.

, & Smith

G. D.

(2006). Patterns, distribution, and determinants of under- and overnutrition: A population-based study of women in India. The American Journal of Clinical Nutrition, 84(3), 633–40.

18.

UNDP. (2008). Human development report 2007–2008. Retrieved February 2013 from http://hdr.undp.org/en/reports/global/hdr2007-2008/ (accessed on 10 August 2012 ).

Women’s Malnutrition in India: The Role of Economic and Social Status

Abstract

Keywords

1. INTRODUCTION

2. BRIEF REVIEW OF LITERATURE

3. DATA AND METHODOLOGY

4.1.1 Per Capita Income

4.1.2 Per Capita Consumption

Table 1 Mean and Standard Deviation of BMI (kg/m2) Across Wealth Groups Wealth Groups Mean Standard Deviation Poorest 19.4 2.68 Poor 20.2 2.95 Mid-wealth 20.7 3.35 Rich 21.8 3.69 Richest 23.3 4.11

4.2 Dietary Variables

4.2.1 Share of Food Expenditure in Total Household Expenditure

4.2.2 Dietary Diversity

4.4 Socio-demographic and Regional Variables

4.4.1 Religion

Table 2 Mean and Standard Deviation of Women’s BMI (kg/m2) Across Religious Groups Religious Group Mean Standard Deviation Hindu 21.0 3.62 Muslim 21.6 3.77 Christian 22.6 3.77 Other religions 21.8 3.99

Table 3 Mean and Standard Deviation of Women’s BMI (kg/m2) Across Caste Groups Caste Groups Mean Standard Deviation Brahmins 22.4 4.02 Other Backward Castes (OBCs) 21.1 3.60 Scheduled Caste (SC) 20.4 3.33 Scheduled Tribe (ST) 20.0 2.89 Others 22.0 3.95

4.5 Individual Variables

4.5.1 Age

Table 4 Mean and Standard Deviation of Women’s BMI (kg/m2) Across Education Levels Completed Education Levels Mean Standard Deviation Not literate 20.4 3.26 Primary 21.4 3.71 Secondary 22.1 3.86 Higher secondary 22.3 4.07 Graduation 23.5 4.14

4.5.3 Women’s Occupation Status

4.5.5 Women’s Pregnancy Status

4.6 Women’s Autonomy Index

Table 6 Mean and Standard Deviation of BMI (kg/m2) Across Levels of ‘Autonomy’ ‘Autonomy’ Level Mean Standard Deviation Autonomy_1 20.9 3.51 Autonomy_2 21.3 3.75 Autonomy_3 21.2 3.78 Autonomy_4 21.7 3.68 Autonomy_5 21.6 3.96

Footnotes

Appendix

Acknowledgements

Notes

References

Table 1
Mean and Standard Deviation of BMI (kg/m²) Across Wealth Groups

Wealth Groups Mean Standard Deviation

Poorest 19.4 2.68

Poor 20.2 2.95

Mid-wealth 20.7 3.35

Rich 21.8 3.69

Richest 23.3 4.11

Table 2
Mean and Standard Deviation of Women’s BMI (kg/m²) Across Religious Groups

Religious Group Mean Standard Deviation

Hindu 21.0 3.62

Muslim 21.6 3.77

Christian 22.6 3.77

Other religions 21.8 3.99

Table 3
Mean and Standard Deviation of Women’s BMI (kg/m²) Across Caste Groups

Caste Groups Mean Standard Deviation

Brahmins 22.4 4.02

Other Backward Castes (OBCs) 21.1 3.60

Scheduled Caste (SC) 20.4 3.33

Scheduled Tribe (ST) 20.0 2.89

Others 22.0 3.95

Table 4
Mean and Standard Deviation of Women’s BMI (kg/m²) Across Education Levels

Completed Education Levels Mean Standard Deviation

Not literate 20.4 3.26

Primary 21.4 3.71

Secondary 22.1 3.86

Higher secondary 22.3 4.07

Graduation 23.5 4.14

Table 6
Mean and Standard Deviation of BMI (kg/m²) Across Levels of ‘Autonomy’

‘Autonomy’ Level Mean Standard Deviation

Autonomy_1 20.9 3.51

Autonomy_2 21.3 3.75

Autonomy_3 21.2 3.78

Autonomy_4 21.7 3.68

Autonomy_5 21.6 3.96