Slums and Malnourishment: Evidence From Women in India

Abstract

Objectives. We examined the association between slum residence and nutritional status in women in India by using competing classifications of slum type.

Methods. We used nationally representative data from the 2005–2006 National Family Health Survey (NFHS-3) to create our citywide analysis sample. The data provided us with individual, household, and community information. We used the body mass index data to identify nutritional status, whereas the residential status variable provided slum details. We used a multinomial regression framework to model the 3 nutrition states—undernutrition, normal, and overnutrition.

Results. After we controlled for a range of attributes, we found that living in a census slum did not affect nutritional status. By contrast, living in NFHS slums decreased the odds of being overweight by 14% (95% confidence interval [CI] = 0.79, 0.95) and increased the odds of being underweight by 10% (95% CI = 1.00, 1.22).

Conclusions. The association between slum residence and nutritional outcomes is nuanced and depends on how one defines a slum. This suggests that interventions targeted at slums should look beyond official definitions and include current living conditions to effectively reach the most vulnerable.

More than 50% of the world population was classified as urban for the first time in 2009 and is expected to reach around 69% in 2050.¹ The proportion of the urban population in the developing world is expected to increase from 45% to 66% during the same period. One of the immediate consequences of population pressure in urban spaces is the growth of slums or urban communities that are characterized by poor access to civic services, inadequate housing, and overcrowding.² It has been estimated that slum populations would double before 2035 in the low- and middle-income countries.³

One of the main concerns regarding the growth of slum populations is that the living conditions of the slum dwellers could become a public health issue. The attention gained by the relation between poor health outcomes and living conditions is neither new nor restricted to the developing world. As early as the 19th century, the Public Health Acts of Britain aimed to improve water systems and sanitation facilities in slums.² This was also true of other developed countries—notably, France and the United States—which attempted to regulate residential dwellings to contain the spread of disease among other things.

Although the pace of urbanization in India historically has been slow, it is increasing rapidly. India’s urban population grew by about 230 million between 1971 and 2008, and it is estimated that 250 million more will swell the urban population within the next 2 decades.⁴ This urban growth has led to a population explosion in cities, and India boasts of 2 cities with a population of at least 10 million (Delhi and Mumbai).

Literature from the developing world suggests that both communicable and noncommunicable diseases are a major concern for urban populations, particularly the slum populations. Already malnourished slum dwellers may experience additional stress because of overcrowding and poor living conditions and are more likely to have poor health outcomes. However, India-specific research findings paint a mixed picture. A study on urban slums in Maharashtra in 1999 indicated that women living in slums were more disadvantaged with respect to antenatal care than were women not living in slums.⁵ This was reaffirmed by another study that compared the health status of poor populations in slums and in resettlement colonies in Delhi and Chennai and found that slum dwellers had worse health outcomes than those in resettlement colonies.⁶ Recent research in Chandigarh that used primary data collected in 2006 showed that immunization status of children younger than 5 years was poorer in slum areas than in the rural and urban areas.⁷ In contrast, a 2005–2006 National Family Health Survey (NFHS-3) report suggested that slum residents were not necessarily worse off than nonslum residents on several deprivation dimensions including poor health.⁸ These studies have used prevalence rates of all illnesses, morbidity rates, incidence of hospitalization, and other health indicators as various proxies of health status.

Our study examined the distribution of women’s malnutrition in 8 cities across slum and nonslum populations. Malnutrition is a significant problem among Indian women. According to several studies that used the NFHS-3, only 52% of the women were within the normal weight range for a given height.^8,9 Following the World Health Organization, we defined malnutrition to include the dual burden of undernutrition and overnutrition. Until recently, attention has been exclusively focused on undernutrition. However, recent trends indicate that Indian women are facing a double burden of malnutrition because of the increasing prevalence of overnutrition largely caused by changing lifestyle and diet patterns.¹⁰

Being underweight could affect productivity and pose health risks, particularly for women, by increasing the likelihood of negative maternal health outcomes, including low-birth-weight infants.¹¹ However, being overweight also could lead to poor health outcomes because of the increased risk of diabetes, cardiovascular diseases, hypertension, and respiratory-related mortality.¹²

Figures 1 and 2 show the prevalence of underweight and overweight women, respectively, in 8 cities in India by slum residence status. Women residing in nonslum areas were more likely to be overweight, whereas those residing in slum areas were more likely to be underweight. In cities such as Delhi, these gaps appear to be large, with 36% being overweight in nonslum areas as opposed to 26% in slum areas; in Indore, 38% were undernourished in slums, but only 28% appear to be undernourished in nonslum areas. These numbers suggest that undernutrition is a larger problem in slums, and overnutrition is mainly a nonslum problem. Therefore, slums could be used as a valid unit to study undernutrition-related policies, and nonslum areas could be used to study overweight-related problems.

FIGURE 1—

Distribution of underweight women in 8 cities in India by slum status: 2005–2006 National Family Health Survey (NFHS-3).

Note. Prevalence was calculated with 2005–2006 NFHS-3 data that were weighted with the provided weights. The prevalence ratios were calculated as follows: the numerator is the number of people who have body mass index (BMI) < 18.5 kg/m², and the denominator is those with normal weight (BMI = 18.5–24.99 kg/m²). The slum variable includes both census-defined slums and those identified as slums by NFHS field staff.

FIGURE 2—

Distribution of overweight women in 8 cities in India by slum status: 2005–2006 National Family Health Survey (NFHS-3).

Note. Prevalence was calculated with 2005–2006 NFHS-3 data that were weighted with the provided weights. The prevalence ratios were calculated as follows: the numerator is the number of people who have body mass index (BMI) ≥ 25 kg/m², and the denominator is those with normal weight (BMI = 18.5–24.99 kg/m²). The slum variable includes both census-defined slums and those identified as slums by NFHS field staff.

However, such differences in prevalence may be attributed to differences in the configuration of infrastructure, socioeconomic and other amenities that distinguish a slum from a nonslum area, or individual characteristics between those who live in slums and those who live in nonslum areas. Individual differences tend to matter more for malnutrition outcomes than do slum characteristics.

METHODS

Household data focusing on health outcomes with city-level geographic identifiers are rare in India; the nationally representative NFHS-3 data collected in 2005 to 2006 are an exception in this regard. We used NFHS-3 data to observe maternal health and family welfare of approximately 15 000 women from 8 cities: 5 state capitals (Delhi, Kolkata, Mumbai, Hyderabad, and Chennai) and 3 small cities (Meerut, Indore, and Nagpur). We investigated how nutritional outcomes of slum residents differed from those of nonslum residents in these cities. The 5 state capitals were among the 6 most populous cities in India with populations ranging from 3.8 million in Hyderabad to 12.9 million in Delhi. Populations in smaller cities range from 1.5 million (Meerut) to 2.6 million (Nagpur) and traditionally have poorer infrastructure than do the larger cities.

Measurement of Malnutrition

Body mass index (BMI, defined as weight in kilograms divided by height in meters squared) gives us 3 outcome categories and was used to identify if individuals were underweight (BMI < 18.5 kg/m²), normal (BMI = 18.5–24.99 kg/m²), or overweight (BMI ≥ 25 kg/m²). Pregnant women and women who had delivered within 3 months of their interview were excluded from the analysis because their BMI could not be directly compared with that of other women.

Measurement of Slum

In India, the census in 2001 provided the first estimate of slum and nonslum populations based on 3 criteria. First, an area could be officially notified by a government body as a slum under any act including the Slum Act. Second, an area could be recognized as a slum by a government body irrespective of its notification status. Third, “a compact area of at least 300 population or about 60 to 70 households of poorly built congested tenements, in unhygienic environment usually with inadequate infrastructure and lacking in proper sanitary and drinking water facilities” also could be notified as a slum.¹³ Once an area was legally recognized as a slum, civic authorities were obligated, at least in theory, to provide municipal services in that area.⁸ Thus, notified slums may not be the worst residential spaces in cities, at least in terms of housing and sanitation infrastructure.

Residential areas in the NFHS-3 data were designated as slum or nonslum on the basis of 2 criteria: (1) whether they were classified as slums by a census or (2) an independent judgment of slum status by the NFHS supervisor based on the Census 2001 definition of overcrowding and poor infrastructure. The rationale provided for the independent classification was 2-fold. First, the areas were classified in 2001, and with rapid urban growth, living conditions in these areas likely would have changed considerably. Second, census classifications usually followed government norms and limited themselves to those areas that were recognized as slums by some local or state authority. This implies that the census classifications often did not include their own criteria of overcrowded houses with poor or no civic facilities. In our analysis, we used both definitions of slums (hereafter referred to as “census slum” and “NFHS slum”) to enhance our understanding of the relation between slum residence and malnutrition.

Other Covariates

We included individual demographics and socioeconomic variables that affect a woman’s quality of life, level of activity, and access to nutrition (Table 1).^9,10,14 Marital status could affect nutritional status as women take on different roles in and out of marriage. In India, caste affects a woman’s access to economic and social opportunities. Thus, we controlled for scheduled castes and tribes, who are among the most marginalized; other backward castes, who are higher in the social hierarchy; and those at the top (the general category).

TABLE 1—

Summary Statistics for Women in 8 Cities in India: 2005–2006 National Family Health Survey (NFHS-3)

Variable	Mean (SD)	Minimum	Maximum
Nutritional status
Overnourished	0.3025 (0.4594)	0	1
Normal	0.5062 (0.5000)	0	1
Undernourished	0.1913 (0.3934)	0	1
Key independent variable: type of slum
Census-designated slum	0.4723 (0.4993)	0	1
NFHS-designated slum	0.3860 (0.4868)	0	1
Household variables
Has piped drinking water?	0.8027 (0.3980)	0	1
Has toilet (sewer/septic tank)?	0.7905 (0.4070)	0	1
Has electricity?	0.9748 (0.1567)	0	1
Has a BPL card?a	0.1017 (0.3023)	0	1
House details
Owns a house?	0.7458 (0.4354)	0	1
Construction type permanent?	0.9043 (0.2942)	0	1
Religion (base: Hindu)
Muslim	0.1699 (0.3756)	0	1
Other	0.0336 (0.1803)	0	1
Social group (base: general)
Scheduled caste or tribe	0.2004 (0.4003)	0	1
Other backward caste	0.3039 (0.4600)	0	1
Current age, y	31.6277 (8.8857)	18	49
Marital status (base: currently married)b
Never married	0.1992 (0.3998)	0	1
Once married	0.0567 (0.2314)	0	1
Education (base: no education)c
≤ primary	0.1199 (0.3249)	0	1
≤ secondary	0.4604 (0.4985)	0	1
> secondary	0.2221 (0.4156)	0	1
Occupation (base: not employed)
Manual work	0.1092 (0.3119)	0	1
Any nonmanual employment	0.2280 (0.4195)	0	1
No. of food items consumed	4.3402 (1.5230)	1	7
Citywide sample distribution
Meerut	0.1300 (0.3364)	0	1
Kolkata	0.1200 (0.3250)	0	1
Indore	0.1224 (0.3277)	0	1
Mumbai	0.0998 (0.2998)	0	1
Nagpur	0.1319 (0.3384)	0	1
Hyderabad	0.1540 (0.3609)	0	1
Chennai	0.1159 (0.3201)	0	1

Note. BPL = below poverty line.

^aBPL cards distributed by the government identify the poor in India.

^bOnce married women are those who are now divorced, separated, or (in the extreme case) deserted.

^cPrimary education is until class 5; secondary is until class 12. Higher education refers to college, diploma, and so forth.

NFHS does not collect information on household income or consumption, but computes a wealth index based on household amenities and the assets owned. The asset index is widely used as a proxy for household income. However, it has been criticized for not being able to account for interstate and rural–urban differences in economic status.^15,16 The wealth index is computed at the national level, which makes city wealth distributions disproportionately skewed toward higher income categories. Fewer than 12% of the women in our sample were within the lowest 3 quintiles, reducing variation in economic status. Consequently, we did not use the NFHS wealth index per se; rather, we unpacked it to separately observe the effect of specific attributes important for nutritional status—presence of a flush toilet, piped water for drinking, and electrification status. These variables also constituted a measure of the slum-level infrastructure. The ownership status of their residence and type of construction (cement and brick construction or otherwise) and the presence of a below-poverty-line card were used to proxy economic status.

Given the importance of diet for nutrition, we constructed a diversity variable that captured the number of food groups that an individual consumes on a daily or weekly basis. Additionally, we included city dummies in the model to account for city-specific differences that affect all individuals in the same city. An example of such differences could be the level of economic development that generally varies across cities and has implications for its own residents in terms of the inflow of migrants and, consequently, the number of slum dwellers. Economic development also could define the level of health infrastructure that the city is able to sustain.

Model

An individual’s nutritional status could be either underweight, normal, or overweight according to increasing values of BMI. However, these nutritional categories in themselves are not ordered. Although being normal weight is clearly better than being underweight or overweight, it is not clear if being underweight is better or worse than being overweight. Thus, we used a multinomial logit model of nutritional status to capture the categorical nature of our outcome variable. Within a multinomial logit framework, if there are J categories, then the probability of observing a particular outcome (m), given x (the full set of covariates including city dummies to estimate the city-fixed effects) is specified as

where j = 0, 1, 2. To facilitate interpretation, the multinomial also can be expressed as an odds model:

where gives the odds of outcome m versus outcome n given x.¹⁷ These odds ratios (ORs) are to be interpreted as the effect of a unit increase in the independent variable on the odds of being nutritionally underweight (or overweight) versus being nutritionally normal, holding all other variables constant. The multinomial model was estimated separately for census slum residence and for NFHS slum residence with the reference category being the normal nutritional status.

We built the regression model by beginning with a simple correlation between slum status and nutritional status. We added a range of household and individual-level covariates on the basis of a literature review. To ensure parsimony, variables that did not contribute significantly to reducing the log likelihood (and therefore pseudo R²) of the model were omitted.

RESULTS

The unadjusted differences in the prevalence of malnutrition between slum areas and nonslum areas according to the census slum definition (n = 15 258; Wald χ² = 116.757) were significant for both underweight and overweight individuals (OR = 1.303 and 0.782, respectively; both P < .01), with normal weight as the reference category. These differences were also significant according to the NFHS-3 definition (n = 15 258; Wald χ² = 102.514; OR = 1.226 and 0.759, respectively; both P < .01). The differences were large but declined once controls for socioeconomic, infrastructural, and other household variables were added to the model.

Tables 2 and 3 present the odds ratios for the risk of being underweight and overweight, respectively, compared with being normal weight, conditional on a range of household- and individual-specific variables as well as slum residence. For model 3 in Table 2 we used the census-designated slum and found no difference in the relative odds for the risk of being underweight or overweight. However, when we identified a slum on the basis of the NFHS supervisor’s categorization (Table 3), we found that after we controlled for all socioeconomic factors, slum residence increased the odds of being underweight by 10% (95% confidence interval [CI] = 1.00, 1.22) and decreased the odds of being overweight by 14% (95% CI = 0.79, 0.95). The housing infrastructure and economic status variables (with the exception of piped water in the underweight models and electricity in the overweight models) were statistically significant and showed expected effects. Individuals living in more affluent households and with better sanitation facilities were at lower risk for undernutrition and at higher risk for overnutrition. Consistent with earlier studies, the results also suggested that investing in education could yield high returns with respect to improving undernutrition, but it could increase an individual’s risk of being overweight.^10,14 The relative odds of being underweight compared with being normal weight was reduced by 6% (95% CI = 0.92, 0.98) as the number of food groups consumed weekly or daily increased.

TABLE 2—

Odds Ratios From a Multinomial Model for Malnutrition in Census Slums in 8 Cities in India: 2005–2006 National Family Health Survey (NFHS-3)

	Model 3
Variable	Underweight, OR (95% CI)	Overweight, OR (95% CI)
Census slum	1.08 (0.98, 1.18)	0.96 (0.89, 1.05)
Household variables
Has piped drinking water?	1.09 (0.96, 1.24)	1.15 (1.03, 1.28)
Has toilet (sewer/septic tank)?	0.87 (0.77, 0.98)	1.19 (1.06, 1.34)
Has electricity?	0.71 (0.56, 0.91)	1.22 (0.90, 1.65)
Has a BPL card?a	1.16 (1.00, 1.34)	0.77 (0.67, 0.90)
House details
Owns any house?	0.86 (0.78, 0.97)	1.13 (1.02, 1.25)
Construction type permanent?	0.92 (0.80, 1.07)	1.40 (1.19, 1.65)
Religion (base: Hindu)
Muslim	1.04 (0.91, 1.19)	1.19 (1.05, 1.33)
Other religion	0.96 (0.72, 1.28)	1.33 (1.09, 1.65)
Social group (base: general)
Scheduled caste or tribe	1.28 (1.13, 1.46)	0.81 (0.72, 0.92)
Other backward caste	1.11 (0.99, 1.25)	0.76 (0.69, 0.85)
Current age, y	0.91 (0.87, 0.96)	1.26 (1.21, 1.33)
Marital status (base: currently married)b
Never married	1.17 (1.02, 1.35)	0.67 (0.57, 0.79)
Once married	1.00 (0.82, 1.25)	0.84 (0.72, 1.01)
Education (base: no education)c
≤ primary	0.78 (0.66, 0.92)	1.25 (1.08, 1.46)
≤ secondary	0.76 (0.67, 0.87)	1.62 (1.45, 1.83)
> secondary	0.57 (0.48, 0.68)	1.89 (1.65, 2.19)
Occupation (base: not employed)
Manual work	1.30 (1.14, 1.49)	0.64 (0.56, 0.74)
Any nonmanual employment	0.99 (0.89, 1.12)	0.75 (0.69, 0.84)
No. of food items consumed	0.95 (0.92, 0.98)	1.01 (0.99, 1.05)
Sample size	14 927
Wald χ2	2474.581

Note. BPL = below poverty line; CI = confidence interval; OR = odds ratio. We report relative ORs when the reference category is being nutritionally of normal weight.

^aBPL cards distributed by the government identify the poor in India.

^bOnce married women are those who are now divorced, separated, or (in the extreme case) deserted.

^cPrimary education is until class 5; secondary is until class 12. Higher education refers to college, diploma, and so forth.

TABLE 3—

Odds Ratios From a Multinomial Model for Malnutrition in 2005–2006 National Family Health Survey (NFHS-3) Slums in 8 Cities in India

	Model 4
Variable	Underweight, OR (95% CI)	Overweight, OR (95% CI)
NFHS slum	1.10 (1.00, 1.22)	0.86 (0.79,0.95)
Household variables
Has piped drinking water?	1.09 (0.97, 1.24)	1.15 (1.03, 1.29)
Has toilet (sewer/septic tank)?	0.87 (0.78, 0.99)	1.17 (1.05, 1.32)
Has electricity?	0.71 (0.56, 0.92)	1.20 (0.89, 1.64)
Has a BPL card?a	1.15 (1, 1.34)	0.78 (0.67, 0.91)
House details
Owns any house?	0.86 (0.78, 0.97)	1.12 (1.02, 1.24)
Construction type permanent?	0.92 (0.81, 1.07)	1.39 (1.18, 1.64)
Religion (base: Hindu)
Muslim	1.03 (0.91, 1.18)	1.20 (1.07, 1.35)
Other religion	0.96 (0.72, 1.28)	1.33 (1.08, 1.64)
Social group (base: general)
Scheduled caste or tribe	1.29 (1.14, 1.47)	0.82 (0.73, 0.93)
Other backward caste	1.11 (0.99, 1.25)	0.76 (0.69, 0.85)
Current age, y	0.91 (0.87, 0.96)	1.26 (1.21, 1.33)
Marital status (base: currently married)b
Never married	1.17 (1.02, 1.35)	0.67 (0.58, 0.79)
Once married	1.00 (0.81, 1.24)	0.85 (0.72, 1.01)
Education (base: no education)c
≤ primary	0.78 (0.67, 0.92)	1.24 (1.08, 1.45)
≤ secondary	0.76 (0.67, 0.87)	1.59 (1.42, 1.80
> secondary	0.57 (0.49, 0.68)	1.83 (1.59, 2.12)
Occupation (base: not employed)
Manual work	1.30 (1.14, 1.49)	0.64 (0.56, 0.74)
Any nonmanual employment	0.99 (0.89, 1.12)	0.75 (0.69, 0.84)
No. of food items consumed	0.95 (0.92, 0.98)	1.01 (0.99, 1.05)
Sample size	14 927
Wald χ2	2493.915

Note. BPL = below poverty line; CI = confidence interval; OR = odds ratio. We report relative ORs when the reference category is being nutritionally of normal weight for each coefficient.

^aBPL cards distributed by the government identify the poor in India.

^bOnce married women are those who are now divorced, separated, or (in the extreme case) deserted.

^cPrimary education is until class 5; secondary is until class 12. Higher education refers to college, diploma, and so forth.

DISCUSSION

We used competing classifications for slums to examine the association between slum residence and nutritional status. The main finding was that this association is nuanced and depends on how one defines a slum. In the definition of slums used by the census, we found no relation between living in a slum and malnutrition, but the converse was true if one used the NFHS definition.

This difference was likely driven by the fact that the NFHS definition goes beyond mere notification status to consider actual living conditions that can influence an individual’s health status such as overcrowding and congestion, poorly built settlements, and inadequate infrastructure. These areas usually do not have even the basic facilities of a notified slum and are typically inhabited by extremely marginalized populations such as new migrants and temporary workers.⁸ A slum survey conducted by the National Sample Survey Organization in 2008 to 2009 found that when compared with notified slums, a greater proportion of nonnotified slums lacked access to civic facilities such as garbage disposal, latrines, and underground drainage systems.¹⁸ Better hygiene and adequate provision of water and sanitation facilities can help reduce the occurrence of communicable diseases such as diarrhea and other infectious diseases that directly affect malnutrition.¹⁹ Traditionally, notified slums are more likely to have experienced an improvement as compared with nonnotified slum areas, and no process allows a notified slum area to become a nonslum area if the facilities improve.⁸ An upgrade of water and sanitation facilities that may have occurred between the census and the NFHS survey time frames could explain the lack of association between malnutrition and a census slum.

We found that undernutrition was mainly a slum problem, whereas overnutrition was mainly a nonslum problem. Regardless of the type of slum, the risk of undernutrition was higher among individuals who lived in a below-poverty household and who did not own a house. We also found the association between education and undernutrition to be stronger than the association between slum residence and undernutrition. A disturbing finding was the positive association between higher education, better economic standard of living, and risk of being overweight. Although this suggests that lifestyle changes that set in with overall economic growth and development need to be targeted, this could also reflect social norms that favor heavier body types.²⁰

It is important to bear in mind the several limitations of this study. The NFHS data do not provide information on the caloric intake or the physical activities undertaken by individuals. Both these aspects clearly affect nutritional status. The available qualitative measures of different food groups consumed are a poor substitute. Additionally, disaggregated information about conditions within the slum could help segregate neighborhoods with poor sanitation and sewage conditions from those with only high population density. Nutritional status could be differentially affected by these conditions.

Despite these limitations, the comparison of the relation between the slum definitions and malnutrition is important from a policy perspective. Our analysis indicates that relying mainly on official definitions may not be sufficient if slums are to be a target unit for public policies to improve the well-being of the vulnerable urban populations. A location-based targeting approach must be based on current living conditions, similar to the NFHS approach. The evidence presented here suggests that such programs need to embrace the dual challenge of avoiding both false-positive outcomes (notified slum population that does not need a targeted intervention) and false-negative outcomes (overlooking settlements that may not be officially classified as slums). The findings endorse the strategy of the latest slum development program of the Indian government, the Rajiv Awas Yojana, which promises a survey of all slums, notified and nonnotified, in selected cities to achieve its goal of “slum-free” cities.²¹ It is, however, important to be vigilant during the implementation stages of the program to ensure that notified or official slums are not given preference over nonnotified areas in an attempt to show that targets have been achieved.

Dietary diversity is a neglected area among policymakers. In India, no official dietary recommendations or food guide pyramids are available; these could influence an individual’s understanding of the optimal diet requirements to prevent undernutrition and overnutrition. The need to make cities slum-free has emphasized the need for planning settlements and provision of civic infrastructure and property rights for slum dwellers. Although the importance of these initiatives cannot be denied, the provision of health and education facilities and sustainable livelihood strategies should not lag behind. The traditional sources of deprivation (lack of access to education or income-generating activities) continue to deserve urgent attention in tackling the problem of malnutrition. People are malnourished not only in slums but also elsewhere. The prevalence of malnutrition cannot be substantially diminished only by making the cities slum-free.