Slum Definitions in Urban India: Implications for the Measurement of Health Inequalities

Transcription

1 Slum Definitions in Urban India: Implications for the Measurement of Health Inequalities Laura B. Nolan More than half of the world s population lives in urban areas, and it is projected that by 2030 over half of residents in low- and middle-income countries will reside in cities (Montgomery 2008). As rural residents move to urban areas in search of jobs, and villages are overtaken by expanding urban agglomerations, many low- and middle-income countries are increasingly concerned with the urbanization of poverty (Pradhan 2012). The speed and large scale of urban growth have outpaced the provision of services in many locations (Yach et al. 1990) and have led to a proliferation of informal settlements and the development of new, smaller cities (Montgomery 2009) without access to water and sanitation, garbage collection, or security of tenure. Concentrated urban poverty and deprivation are often characterized by residential crowding, exposure to environmental hazards, and social fragmentation and exclusion (Wratten 1995) a cluster of conditions frequently referred to with the catch-all term of slum-dwelling. Indeed, policy and media discussions of urban poverty tend to focus on slums, the characterization of which is presumed to be straightforward. There are many different ways in which slums are defined, however. In this article I examine various ways of characterizing a slum and explore how choice of definition affects the relationship between slum-dwelling and child health. The word slum was first used in London at the beginning of the nineteenth century to describe a room of low repute or low, unfrequented parts of the town, but has since undergone many iterations in meaning and application (UN-HABITAT 2003b). While early definitions of slumdwelling combined physical, spatial, social, and behavioral aspects of urban poverty (UN-HABITAT 2003a), the range of associations has more recently narrowed. According to the United Nations Program on Human Settlements (UN- HABITAT), a slum is a contiguous settlement where the inhabitants are POPULATION AND DEVELOPMENT REVIEW 41(1): (MARCH 2015) The Population Council, Inc.

2 60 S l u m D e f i n i t i o n s i n U r b a n I n d i a characterized as having inadequate housing and basic services. A slum is often not recognized and addressed by the public authorities as an integral or equal part of the city (UN-HABITAT Urban Secretariat and Shelter Branch 2002). Alternative definitions of slums The definition of what constitutes a slum, like the definition of what constitutes an urban area more generally (Dorélien et al. 2013), differs by country (United Nations 2014a) and, in India, by state (Ministry of Housing and Urban Poverty Alleviation 2008), and even city (O Hare et al. 1998). The United Nations incorporated attention to slums in the Millennium Development Goals as part of Goal 7, to Ensure Environmental Sustainability, with a target to Achieve, by 2020, a significant improvement in the lives of at least 100 million slum dwellers (United Nations 2013). A revised UN definition devised to monitor progress toward this goal characterized slums as one or a group of individuals living under the same roof in an urban area, lacking in one or more of the following five amenities: 1) Durable housing (a permanent structure providing protection from extreme climatic conditions); 2) Sufficient living area (no more than three people sharing a room); 3) Access to improved water (water that is sufficient, affordable, and can be obtained without extreme effort); 4) Access to improved sanitation facilities (a private toilet, or a public one shared with a reasonable number of people); and 5) Secure tenure (de facto or de jure secure tenure status and protection against forced eviction) (UN-HABITAT 2006/7). While this definition of what constitutes a slum was used by the UN to evaluate whether its MDG target had been met, it is quite different from the definitions used by many individual countries for their own policy and planning purposes. Uganda, for example, in a document outlining a slumupgrading strategy and action plan from 2008, defines slums as combining the UN definition with more localized characteristics reflecting the Ugandan situation : 1) attracting a high density of low-income earners and/or unemployed persons with low levels of literacy; 2) an area with high rates/levels of noise, crime, drug abuse, immorality (pornography and prostitution) and alcoholism, and high HIV/AIDS prevalence; or 3) an area where houses are in environmentally fragile lands, e.g., wetlands. Applying the UN s slum definition to Ugandan cities results in nearly two-thirds of the urban population living in slums (Uganda, Ministry of Lands 2008). In India, notification, or legal designation, as a slum settlement is central to the recognition of slums by the government and is intended to afford residents rights to the government provision of potable water and sanitation. But many communities exhibiting distinctly slum-like characteristics are never notified (Subbaraman et al. 2012); Delhi, for example, has notified no new slums since 1994 (Bhan 2013). The UN definition incorporates tenure as

3 L a u r a B. N o l a n 61 just one of its components, likely leading to disagreement over the distribution and absolute number of slum residents in India. The differences between multilateral and country-level definitions of what constitutes a slum prompt the central research question: does it matter how slums are defined? In other words, do different definitions identify the same underlying circumstances of concentrated urban deprivation and categorize the same areas as slums? In this article I investigate slum dwelling in the context of India. The definition and identification of slums are of current policy and programmatic importance to the government of India, which is increasingly concerned with poverty, inequality, and poor health among its 400 million urban residents. The government has developed policy initiatives such as the Rajiv Awas Yojana, which envisages a slum-free India (Ministry of Urban Housing and Poverty Alleviation 2010), and it may benefit from empirical work on the implications of the measurement of the urban poor population. Urbanization in India, a country of over a billion people, is a massive planning and policy challenge. After economic liberalization in the early 1990s, India s urban population grew by almost 32 percent in a decade (Agarwal et al. 2007). The UN estimates that by 2030 about 583 million Indians will live in cities (United Nations 2014b). Although India s cities are seen as the engine of the country s growth and development, poor living conditions like those found in slums likely have substantial adverse consequences for productivity and human capital development. Slum residents, for example, spend significant time and resources obtaining water and waiting to use public toilets (Subbaraman et al. 2014). Lack of infrastructure and physical security in slums may reduce residents involvement in the labor force and their participation in society, both of which may exact a toll on the country s development. The National Family and Health Survey (NFHS), India s Demographic and Health Survey (DHS), from , includes separate determinations from both the Census and survey enumerator supervisors regarding whether a respondent s household is located in a slum area, allowing for the comparison of slum definitions and their association with indicators of wellbeing. Population-based sampling frames drawn from the Census for nationally representative surveys like the DHS are rarely stratified by slum status (Montana et al. forthcoming), and the inclusion of these data in the most recent NFHS allows the comparison of multiple definitions of slum-dwelling in the Indian context. A number of empirical studies have incorporated the two definitions of slum-dwelling included in the NFHS. Swaminathan and Mukherji found that the association between slum-dwelling and under- and overweight among adults in eight urban areas in India yielded different results both in terms of significance and magnitude depending on the definition used (Swaminathan and Mukherji 2012). Dev and Balk combined the two defi-

4 62 S l u m D e f i n i t i o n s i n U r b a n I n d i a nitions, identifying households as residing in slums if they met the criteria of either of the two definitions (Dev and Balk 2014). Most other researchers have simply chosen to focus on the Census (Gaur et al. 2013; Hazarika 2010) or the survey enumerator (Rooban et al. 2012) definitions exclusively, with little justification for their decision. But the slum definitions used in the NFHS are not the only manner in which to empirically characterize slum-dwelling. Günther and Harttgen used survey respondents reported characteristics of their household and its surroundings to characterize families as living in a slum or not in sub-saharan Africa (Günther and Harttgen 2012), and Fink, Günther, and Hill employed a similar methodology across 73 countries to compare the health of rural, urban, and slum residents (Fink, Günther, and Hill 2014). Recent research has indicated that slums in India may be more heterogeneous than is often assumed (Goli et al. 2011; Chandrasekhar and Montgomery 2009; Agarwal and Taneja 2005). Some slum residents are quite well established and have been able to obtain services over time either legally or illegally. Conversely, poor people like pavement-dwellers do not live in slums and are therefore not counted by the standard definitions (Agarwal 2011). Identifying and comparing definitions of what constitutes slum-dwelling is important not only for urban planning but also in assessing the extensive literature documenting the effect of area-level poverty on health, much of which is based in the context of urban India (ibid.). Urban deprivation, slums, and health Although the mechanisms including social, environmental, geographic, and institutional (Galster 2010) by which community-level poverty may be associated with poor health are still under investigation, poor health in slum areas has been documented in sub-saharan Africa (Bocquier et al. 2011; Günther and Harttgen 2012) and South Asia, particularly Bangladesh (Gruebner et al. 2011) and India (Gaur et al. 2013; Hazarika 2010; Gupta et al. 2009). Close living quarters, poor sanitation, and lack of access to potable water (Sclar et al. 2005), all characteristics of slum-like communities, appear to produce poor health over and above the effects of simply living in a poor household and other individual-level characteristics (Rice and Rice 2009). Crowding, for example, tends to promote the transmission of infectious diseases like pneumonia, diarrhea, and tuberculosis (Unger and Riley 2007), and neighbors open defecation is negatively associated with child height (Spears 2013). These health problems are exacerbated by the illegality and social exclusion that tend to characterize slum settlements (de Snyder et al. 2011; Subbaraman et al. 2012), by poorly regulated and ineffective health services

5 L a u r a B. N o l a n 63 (Agarwal et al. 2007), by exposure to environmental hazards (Unger and Riley 2007), and, in the Indian case, by uncertainty regarding which level of government is responsible for protecting and promoting the health of the poorest urban residents (Nolan et al. 2014). In short, the possibility of an urban mortality penalty, such as occurred during industrialization in European and American cities in the nineteenth century, is a distinct possibility for India (Konteh 2009). To investigate the implications of the definition of slum-dwelling, I focus on one indicator of human and economic wellbeing, namely child health (Strauss and Thomas 1998). I use child height to investigate the effects of past epidemiological and nutritional environment (Deaton 2007), and child weight to examine acute and current health issues, as well as nutritional stressors. About 50 percent of children in India are undernourished, and lower height for age in particular has been associated with reduced cognitive and educational achievement (Hoddinott et al. 2011), as well as lower wages and labor market productivity over the life course (Case and Paxson 2008). Undernutrition not only directly affects children s physical and cognitive growth, but is also implicated in deaths from infectious diseases such as malaria, pneumonia, and measles, making the underlying condition responsible for over 20 percent of the country s burden of disease (Gragnolati et al. 2005). Most studies of the health of slum-dwellers investigate this topic within a particular slum (Subbaraman et al. 2013), in one city (Fotso et al. 2013; More et al. 2013), or across many countries employing a standardized definition of a slum dwelling (Fink et al. 2014). While Montgomery and Hewett investigated the effect of neighborhood socioeconomic status on height for age (Montgomery and Hewett 2005) using the NFHS, the association between slum-dwelling and child height and weight has, to my knowledge, not been systematically investigated in India. Data and methods Data This study uses data from the third NFHS, collected in , the first and only Demographic and Health Survey to include multiple measures of slum designation at the level of the primary sampling unit (PSU). Slum designation, however, is available for only the eight cities shown in Figure 1: Chennai, Delhi, Hyderabad, Indore, Kolkata, Meerut, Mumbai, and Nagpur (International Institute for Population Sciences and Macro International 2007). In what follows, I compare four different slum definitions: two embedded in the individual-level data as dichotomous indicator variables and two constructed from the household questionnaire. The four definitions are described in Tables 1 and 2.

6 64 S l u m D e f i n i t i o n s i n U r b a n I n d i a FIGURE 1 Eight cities in India with NFHS data on slum designation By way of summary, the 2001 Census definition is an administrative one that includes legality; the NFHS definition is based on survey enumerator supervisor observations of slum characteristics; the UN definition is comprised of universally recognized components of a healthy environment (UN-HABITAT 2006/7); and the Committee definition has been tailored to the Indian context as recommended for the 2011 Census in a Report to the Committee on Slum Statistics/Census (Ministry of Housing and Urban Poverty Alleviation 2008). It is not possible to use the slum definition from the 2011 Census because it requires information that is not available in the household-level NFHS, such as whether the community is legally recognized by various local authorities. 1 All information on slum designation and household characteristics was taken from the household recode file. The Census and NFHS dummy variables (0 not slum; 1 slum) are embedded in the individual-level questionnaire at the PSU level, which suffices as a proxy for the respondent s neighborhood. The NFHS data are not geo-referenced, and there is no other man-

7 L a u r a B. N o l a n 65 TABLE 1 Origin and characteristics of four slum definitions for India Name Census NFHS UN Committee a See text. Origin 2001 Census of India Survey enumerator supervisor observation Household questionnaire a Household questionnaire a Empirical generation Legality Density Housing Water Sanitation Variable included in the NFHS-3 Variable included in the NFHS-3 Aggregated to the primary sampling unit level Aggregated to the primary sampling unit level ner in which to operationalize spatial proximity (International Institute for Population Sciences and Macro International 2007). In rural areas, PSUs are villages. In urban areas, the NFHS uses a slightly more complex procedure. Wards were first selected systematically from the 2001 Census, and then one census enumeration block of about households was selected from each ward (both selections were done with probability proportional to size). A household listing was compiled for each enumeration block and, on average, 30 households were targeted for interview, with a minimum and maximum of about 15 and 50 households. The NFHS definition was operationalized by highly trained survey enumerator supervisors, who identified an area as a slum on the basis of the characteristics listed in Table 2 (ibid.). The UN and Committee definitions are constructed from the household questionnaire and based on reports of the family s living circumstances. First, each of the four indicators for the UN definition (insufficient living area, nondurable housing, and lack of access to improved water and to improved sanitation) were coded 1 if the household displayed the slum-related deprivation and 0 if not. A household was designated as slum-like by the UN definition if the sum of the four UN indicators equaled or exceeded 1. This analytical operationalization differs slightly from the one employed by Fink, Günther, and Hill (2014), who found that defining slum-dwelling in this manner across 73 countries resulted in an implausible proportion of households living in slums. This prompted the authors to use a more stringent coding of slumdwelling, defined as a household experiencing two or more of the four UN

8 TABLE 2 Details of four slum definitions, India Characteristic Census NFHS Legality Density Housing Water Sanitation 1) All specified areas in a town or city notified as Slum by State/Local Government and Union Territory Administration under any Act, including a Slum Act ; and/or 2) All areas recognized as Slum by State/Local Government and Union Territory Administration, Housing, and Slum Boards, which may not have been formally notified as a slum under any Act Compact area of at least 300 population or about households Poorly built, congested tenements in unhygienic environment, usually with inadequate infrastructure Lacking proper drinking water facilities Lacking proper sanitary facilities UN at least one or more NA a NA b NA Compact area of at least 300 population or about households Poorly built, congested tenements in unhygienic environment, usually with inadequate infrastructure Lacking proper drinking water facilities Lacking proper sanitary facilities Insufficient living area c Non-durable housing; d lack of a permanent structure providing protection from extreme climate conditions Lack of access to improved water e that is sufficient, affordable, and attained without extreme effort Lack of access to improved sanitation facilities g Committee all four NA Predominant material of roof is anything other than concrete Drinking water source not available within the premises f Household does not have a latrine facility within the premises (e.g., members use either a public latrine or no latrine) and does not have closed drainage h a Not applicable: information on this slum characteristic is not included in this definition. b UN definition technically includes security of tenure and protection against forced eviction. But because this information is not captured in DHS surveys, it is standard procedure to omit it from empirical analyses on this topic. c More than three people sharing a room. d Kaccha houses constructed of low-quality materials like mud, thatch, or tarpaulin, and semi-pucca houses constructed using a mix of low- and high-quality materials. e Unimproved water sources include unimproved dug well, unprotected spring, cart with small tank/drum, bottled water, tanker-truck, and surface water (river, dam, lake, pond, stream, canal, irrigation channels). f Coded as not piped into dwelling. g Unimproved sanitation facilities include flush or pour-flush to elsewhere, pit latrine without slab or open pit, bucket, hanging toilet or hanging latrine, no facilities, or bush or field. h Information on drainage was not included in the NFHS; this indicator was coded as the household either sharing a toilet facility or having none at all.

9 L a u r a B. N o l a n 67 deprivations, reducing substantially the proportion living in slums. Because my objective is to operationalize as faithfully as possible the four different slum definitions, I designate households as slum-like if they exhibited only one of the four deprivations as indicated by the UN definition. I take a similar approach for the Committee definition: each of the three indicators (non-concrete roofing material, no drinking water facility on the premises, and use of public latrine or no latrine) was coded as 0/1 and summed. Households with a score of 3 were designated as slum-like; that is, the definition required the household to display all three characteristics. Finally, to make a fair unit-wise comparison across all four definitions and for consistency in the conceptualization of slum-dwelling as a community-level phenomenon, the two definitions constructed from the household survey the UN and Committee definitions were aggregated to the PSU level as the proportion of surveyed households in that PSU characterized as slum-like by each definition. This ensures that slum-dwelling reflects concentrated community-level deprivation; only households displaying slum-like characteristics that are surrounded by other such households are considered to be located in a slum. Empirically, PSUs are defined as slums by the UN and Committee definitions if over 50 percent of the households interviewed exhibited slum-like characteristics. This approach has been used previously for the UN definition (Günther and Harttgen 2012), although Fink et al. again used a more stringent cutoff (of 75 percent) given their focus on crosscountry assessment of the urban advantage in child health, rather than an accurate operationalization of slum dwelling in any particular context (Fink, Günther, and Hill 2014). The results presented here for weight for age are not sensitive to the choice of 50 percent, although at much lower cutoffs (in which PSUs with very few households displaying slum-like characteristics are designated as slums) the association between slum-dwelling and height for age, not surprisingly, is weakened (results not shown). There are 597 PSUs in the eight cities that have non-missing values for the four slum designations included in these analyses. All independent and dependent variables were taken from the child recode file. The dependent variables are height-for-age and weight-for-age z-scores of children under five years old, scaled to the World Health Organization s reference chart and excluding children with questionable scores of under 6 and over 6 as is standard practice. As detailed in the survey report, height for age and weight for age are missing for children who were not at home at the time of the survey and whose parent refused to have them weighed and measured (9 percent of the full NFHS sample), while another 8 percent of eligible children either had no known month or year of birth or had grossly improbable height or weight measurements (International Institute for Population Sciences and Macro International 2007). I do not investigate stunting or wasting dichotomous variables defined as two or

10 68 S l u m D e f i n i t i o n s i n U r b a n I n d i a more standard deviations below the median of the reference population to preserve power, as has been recommended in the literature (Spears 2013). I also do not investigate mortality, given the relatively small number of child deaths in the sample, nor child morbidities such as diarrhea, given concerns about the accuracy of self-reported health diagnoses. Independent variables included in the regressions are all known to be associated (though not invariably) with both poor child health and poverty in the Indian context. Child characteristics include child s sex, multiple births, and size at birth. Son preference remains prevalent in India (Deaton 2008) and leads to poor nutrition even among pregnant women (Coffey et al. 2014); this is associated with low birthweight (Ramalingaswami et al. 1996) and poorer later-life health (Adair 2007), and is controlled for in the analyses with an indicator of whether the mother reported the child was small, medium, or large at birth. Being one of multiple births is also controlled for in the multivariate models; birthweight is related to later height, health, and development outcomes (Currie and Vogl 2013). Mothers characteristics include her education, religion, number of children ever born, whether she works outside the home, and her height, caste, and migration status. The effect of a mother s education on the health and wellbeing of her children is one of the most robust and consistent findings in the development literature (Lutz and KC 2011). Education is related to maternal health and height attainment, as well as to increased human capital and productivity (Strauss and Thomas 1998). High levels of childbearing take a toll on women s health and are associated with poorer health and greater stress, in some cases leading to low-birthweight babies (Cleland et al. 2006; Shah 2010). Recent work in an urban demographic surveillance site in Nairobi City, Kenya found that mothers employment outside the home was associated with child morbidity (Taffa et al. 2005). Being a member of a backward or scheduled caste or tribe in India is a marker of historic experience of marginalization and deprivation that may significantly influence current and future health and development (Gragnolati et al. 2005). Being a Muslim rather than a Hindu may also indicate lack of social and/or economic privilege. Finally, a large literature on migration status and health indicates that the children of recent migrants have worse health outcomes and higher odds of dying than non-migrants (Brockerhoff 1995). These findings have been replicated in India (Stephenson et al. 2003); the analyses presented here also include a control variable for whether the family migrated from a rural area or another city or town. Other control variables include husband s/father s education, a fixed effect for the eight cities in which the data were collected, and a fixed effect for the year of the child s birth. Father s education is associated with higher standards of living and presumably a healthier nutritional and epidemiological environment, although the effect in the literature is usually found not to be

11 L a u r a B. N o l a n 69 as large as that for maternal education (Breierova and Duflo 2010). Indeed, women are thought to be more likely than their male counterparts to invest additional available resources more heavily in their household s wellbeing (Duflo 2012). City fixed effects are included to control for characteristics of each city to which all respondents living there were exposed, and to account for the uneven distribution of health status across the eight urban areas in the sample. Year-of-birth fixed effects are included to control for any changes over the five-year period during which the births took place. Finally, a household wealth index was constructed using the first principal component (Filmer and Pritchett 2001) of a factor analysis of 19 household assets indicative of socioeconomic status in urban India. These comprise a radio, television, refrigerator, bicycle, motorcycle/scooter, car, modern cooking fuel, mobile phone, watch, mattress, pressure cooker, chair, cot/ bed, table, electric fan, sewing machine, computer, and water pump. About 96 percent of the analytical sample reported having electricity, so these assets are both relevant and usable in urban India. Items like tractors, livestock, and irrigated land were excluded because they were unlikely to be relevant to socioeconomic status in an urban setting. Modern cooking fuel was designated as 1 for electricity, liquefied petroleum gas/natural gas, and biofuel, and 0 for kerosene, coal, lignite, charcoal, wood, straw/shrubs/grass, agricultural crop, and animal dung. The score produced from the principal components analysis was not broken into quintiles because its cumulative distribution is not linear, but is used as a continuous variable (ranging from 5.49 to 3.72, with a mean of 0.14) in all regressions. Table 3 presents summary statistics for all other control and outcome variables included in the models, with the exception of slum designation. Methods The final analytical sample of children with no missing data on the dependent or any of the independent variables consisted of 4,609 children under the age of five years. The four definitions of slums are first compared descriptively, and their association with child health is then assessed beginning with bivariate regressions of each slum definition on each health outcome. This is followed by a comparison of ordinary least squares models containing all independent variables (Fink et al. 2012): Y chi = α + β c + X ch γ + X chi δ + ε chi where Y chi is the height-for-age or weight-for-age z-score of child i in household h of community (i.e. PSU) c; β c is one of four indicators of slum dwelling for community c in which child i lives and is the coefficient of interest; X ch is a vector of household-level control variables; and X chi is a vector of individual-

12 70 S l u m D e f i n i t i o n s i n U r b a n I n d i a TABLE 3 Characteristics of the study population (N = 4,609) Percent or mean Percent or mean Characteristic (standard deviation) Characteristic (standard deviation) Child Mother s partner Height for age 1.47 (1.65) Education Weight for age 1.38 (1.21) None 13.4 Sex Primary 11.3 Male 52.8 Secondary 54.1 Female 47.2 Higher 21.2 Multiple birth Area Yes 1.2 City of residence No 98.8 Delhi 13.3 Size at birth Meerut 18.8 Small 14.2 Kolkata 8.5 Medium 63.5 Indore 14.0 Large 22.3 Mumbai 8.0 Mother Nagpur 12.5 Education Hyderabad 15.4 None 22.7 Chennai 9.4 Primary 10.8 Year of birth Secondary Higher Religion Hindu Muslim Other No. of children born in last 5 years 1.6 (0.64) Age (years) 26.8 (4.59) Height (cm) (5.86) Working Yes 82.6 No 17.4 Scheduled caste or tribe Yes 20.9 No 79.1 Migrant Yes, from a rural area 29.5 Yes, from city/town 43.7 No 26.8 level control variables. All models cluster the standard errors at the household level to account for the possibility of more than one child in a household. The error term, ε chi, is assumed to be normally distributed.

13 L a u r a B. N o l a n 71 The results from the ordinary least squares models lead to further investigation into the components of the UN definition of slum-dwelling that are predictive of poor child health. In this third step, the four components of the UN definition are entered separately into an ordinary least squares regression model as follows: Y chi = α + X c η + X ch γ + X chi δ + ε chi where Y chi again is the height-for-age or weight-for-age z-score of child i in household h and community c; X c is a vector of the four characteristics constituting the UN definition of slum-dwelling; and X ch and X chi are the same vectors of household- and individual-level control variables as in the model with slum designation as a dichotomous indicator variable. This model is run for both health outcomes. 2 Results Table 4 shows the percent of children in the study sample who live in slums in each city; estimates vary widely by definition. In the capital city of New Delhi, the UN definition (comprised of community-wide lack of access to improved water, sanitation, and durable housing, and crowding) indicates that 65 percent of children live in PSUs characterized as slums, whereas the Committee definition (comprised of community-wide unimproved roof material, lack of access to potable water, and poor sanitation facilities) identifies only 32 percent of children as living in slums. The variation is widest in Meerut and Hyderabad, where the UN definition finds over 60 percent of children characterized as slum-dwelling, while the Committee definition finds the proportion to be close to zero. 3 The Census definition (which includes legality) systematically produces slightly higher estimates than the NFHS TABLE 4 Percent of children in the study sample in eight Indian cities identified as living in slums, by slum definition City (N) Census NFHS UN Committee Delhi (612) Meerut (866) Kolkata (389) Indore (644) Mumbai (368) Nagpur (576) Hyderabad (719) Chennai (435) Total

14 72 S l u m D e f i n i t i o n s i n U r b a n I n d i a TABLE 5 Percent of children in the study sample in eight Indian cities identified as living in slums according to two definitions Census NFHS UN Committee Census NFHS UN Committee definition (which relies on enumerator observations of local surroundings); the UN definition generally produces the highest estimates. The Committee definition consistently produces the lowest estimates. Proportions designated as slum-dwelling by each possible combination of two definitions (Table 5) indicate significant variation in overlap between the four definitions. For example, while 13 percent of children are designated as living in a slum by the Committee and UN definitions, 33 percent of children are designated as such by the UN and NFHS definitions. Figure 2 presents a Venn Diagram of the overlap between different slum designations. While it does not display these results proportionally, the diagram provides additional support for the variability in overlap between definitions. These descriptive results should give pause to researchers, policymakers, and public health practitioners who might consider slum-dwelling to be conceptually or empirically straightforward. Table 6 presents the results of bivariate regression models of each of the four slum designations on the two health outcomes. Four regression models are presented, one for each of the four slum definitions. All slum designa- FIGURE 2 Venn Diagram detailing overlap of number of children identified as slum-dwelling, by slum definition, in eight Indian cities

15 L a u r a B. N o l a n 73 TABLE 6 Bivariate regressions of slum designation on child health, eight Indian cities Model (1) (2) (3) (4) Child health variable Census NFHS UN Committee Height for age Coefficient 0.319* 0.331* 0.568* 0.443* (0.051) (0.052) (0.052) (0.072) R Weight for age Coefficient 0.224* 0.182* 0.362* 0.333* (0.039) (0.039) (0.039) (0.054) R *Statistically significant at p<0.05. NOTE: Standard errors (in parentheses) clustered at the household level. tions are associated with statistically significantly lower height for age and weight for age of children under age five, although the indicator explains a very small proportion of the variation in the outcome in each case. Living in a slum is associated with a range of 19 percent (0.319/1.65) to 34 percent (0.568/1.65) standard deviations lower height for age according to the Census and UN definitions, respectively. The negative association ranges from 15 percent (0.182/1.21) to 30 percent (0.362/1.21) of a standard deviation in magnitude for weight for age for the NFHS and UN definitions, respectively. Children living in communities designated as slums by three or all four of the definitions also had lower height for age and weight for age in bivariate regression models (results not shown). Tables 7 and 8 present multivariate models of the relationship between slum-dwelling and the two health outcomes and include a wide variety of control variables. Four regression models are again presented, one for each of the slum definitions. When including all covariates, the only slum designation that is statistically significantly associated (at the 5 percent level) with child height for age is the UN s. Specifically, children characterized by the UN as living in slums have, on average, a height-for-age z-score that is (or about 11 percent of a standard deviation) lower than their non-slum-dwelling counterparts (Table 7). The results are similar for weight for age, shown in Table 8. When including all covariates in the model, the UN definition is the only slum indicator that is statistically significantly associated with child weight for age. A number of covariates appear to explain away the bivariate relationship between the Census, NFHS, and Committee definitions of slum-dwelling and poor heath, including children s size at birth, maternal education, being Muslim rather than Hindu, mother s height, whether the mother works outside the home, and household wealth. Father s high level

16 74 S l u m D e f i n i t i o n s i n U r b a n I n d i a TABLE 7 Ordinary least squares regression of height for age on slum-dwelling and other covariates, eight Indian cities Model (1) (2) (3) (4) Variable Census NFHS UN Committee Slum-dwelling indicator * (0.048) (0.051) (0.053) (0.073) Child characteristics Sex Male Female (0.044) (0.044) (0.044) (0.044) Multiple birth No Yes (0.257) (0.256) (0.260) (0.257) Size at birth Small Medium 0.316* 0.319* 0.320* 0.315* (0.066) (0.066) (0.066) (0.066) Large 0.444* 0.443* 0.445* 0.445* (0.077) (0.077) (0.077) (0.077) Mother s characteristics Education None Primary 0.271* 0.269* 0.263* 0.268* (0.092) (0.092) (0.092) (0.092) Secondary 0.195* 0.194* 0.177* 0.188* (0.071) (0.071) (0.072) (0.071) Higher 0.445* 0.441* 0.407* 0.443* (0.104) (0.104) (0.105) (0.103) Religion Hindu Muslim 0.195* 0.191* 0.156* 0.202* (0.062) (0.062) (0.063) (0.061) Other (0.142) (0.142) (0.141) (0.142) No. of children born in last 5 years * (0.050) (0.050) (0.050) (0.050) * 0.184* (0.089) (0.088) (0.088) (0.088) Age 0.017* 0.017* 0.017* 0.017* (0.006) (0.006) (0.006) (0.006) Height 0.052* 0.052* 0.052* 0.053* (0.004) (0.004) (0.004) (0.006) /

17 L a u r a B. N o l a n 75 TABLE 7 (continued) Model (1) (2) (3) (4) Variable Census NFHS UN Committee Working No Yes 0.159* 0.160* 0.164* 0.161* (0.060) (0.061) (0.060) (0.061) Caste Scheduled caste/tribe Not scheduled (0.061) (0.061) (0.062) (0.061) Migrant No Yes, from city/town (0.054) (0.054) (0.054) (0.055) Yes, from rural area (0.063) (0.063) (0.063) (0.063) Partner s education None Primary (0.096) (0.096) (0.096) (0.096) Secondary (0.082) (0.082) (0.082) (0.082) Higher (0.107) (0.107) (0.106) (0.107) Wealth index 0.082* 0.081* 0.078* 0.079* (0.015) (0.015) (0.015) (0.015) City fixed effects Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Constant * * * * (0.672) (0.674) (0.672) (0.674) R *Statistically significant at p<0.05. NOTE: Standard errors (in parentheses) clustered at the household level. of education is also associated with higher weight for age but not height for age. To investigate which of the four components of the UN definition might be driving its relationship with poor health, in spite of the inclusion of these additional covariates, a final ordinary least squares regression model was estimated with the four components housing, density, water, and sanitation entered separately (not shown). The most prevalent of these components in the sample, in PSUs designated as slums by any of the definitions, as well as in PSUs designated as slums by only one of the four definitions, was high-density living.

18 76 S l u m D e f i n i t i o n s i n U r b a n I n d i a TABLE 8 Ordinary least squares regression of weight for age on slum-dwelling and other covariates, eight Indian cities Model (1) (2) (3) (4) Variable Census NFHS UN Committee Slum-dwelling indicator * (0.036) (0.039) (0.041) (0.057) Child characteristics Sex Male Female (0.033) (0.033) (0.033) (0.033) Multiple birth No Yes (0.201) (0.201) (0.205) (0.200) Size at birth Small Medium 0.281* 0.283* 0.284* 0.279* (0.050) (0.050) (0.050) (0.050) Large 0.424* 0.423* 0.425* 0.425* (0.059) (0.059) (0.059) (0.059) Mother s characteristics Education None Primary 0.245* 0.243* 0.238* 0.241* (0.068) (0.068) (0.068) (0.068) Secondary 0.150* 0.148* 0.133* 0.142* (0.054) (0.054) (0.055) (0.054) Higher 0.372* 0.366* 0.366* 0.367* (0.078) (0.078) (0.080) (0.078) Religion Hindu Muslim 0.142* 0.138* 0.107* 0.148* (0.048) (0.048) (0.049) (0.048) Other (0.095) (0.095) (0.094) (0.095) No. of children born in last 5 years (0.037) (0.048) (0.037) (0.004) (0.074) (0.074) (0.074) (0.074) Age 0.014* 0.014* 0.014* 0.013* (0.004) (0.004) (0.004) (0.004) Height 0.042* 0.042* 0.042* 0.042* (0.003) (0.003) (0.003) (0.003) /

19 L a u r a B. N o l a n 77 TABLE 8 (continued) Model (1) (2) (3) (4) Variable Census NFHS UN Committee Working No Yes 0.066* 0.066* 0.070* 0.067* (0.046) (0.046) (0.046) (0.046) Caste Scheduled caste/tribe Not scheduled (0.046) (0.046) (0.046) (0.046) Migrant No Yes, from city/town (0.042) (0.042) (0.042) (0.042) Yes, from rural area (0.049) (0.049) (0.049) (0.049) Partner s education None Primary (0.073) (0.073) (0.073) (0.073) Secondary (0.063) (0.063) (0.063) (0.063) Higher 0.165* 0.163* 0.150* 0.158* (0.082) (0.082) (0.082) (0.082) Wealth index 0.058* 0.057* 0.054* 0.055* (0.011) (0.011) (0.011) (0.011) City fixed effects Yes Yes Yes Yes Year fixed effects Yes Yes Yes Yes Constant 8.722* 8.708* 8.637* 8.709* (0.519) (0.518) (0.517) (0.117) R *Statistically significant at p<0.05. NOTE: Standard errors (in parentheses) clustered at the household level. Rather than dichotomize the four components, they were left as continuous variables representing the proportion of households in the PSU lacking each of the four amenities. The models, which were run separately for heightfor-age and weight-for-age z-scores, indicate that the negative association between height for age and the UN slum definition is driven by the housing quality component. For weight for age, both housing quality and crowding are statistically significantly associated with a lower z-score, with a particularly large effect of density. Living in a neighborhood filled with crowded homes is associated with almost a third of a standard deviation lower weight for age.

20 78 S l u m D e f i n i t i o n s i n U r b a n I n d i a Discussion In this article I have described four different ways to characterize what constitutes a slum area within India and demonstrated that these definitions frequently do not identify the same households as slum-dwelling. The manner in which a slum is defined appears to determine whether a relationship is found between slum-dwelling and health; only one of the definitions of slum-dwelling presented is associated with child health when controlling for individual- and household-level characteristics, and this relationship seems to be driven by only two of its four components. These findings have implications for the empirical measurement and study of area-level deprivation and intra-urban health inequality, and, by extension, for current policy and media attention focusing on slums (Bhaumik 2012). The way in which slums are defined affects both the descriptive characterization of urban populations and the magnitude and significance of the empirical association between community disadvantage and child health. I find significant discrepancies between slum definitions in regard to which households are designated as slum-dwelling. The mismatch between the UN definition, which is used to compare slum-dwelling across countries, and definitions used to monitor an individual country s levels of concentrated urban disadvantage for policy and planning purposes can lead to divergent and even conflicting development priorities. There are a number of potential explanations for the relatively small overlap among the four slum designations. First, the Census was conducted in 2001, while the NFHS was undertaken in Thus, it is possible that slum areas changed substantially between these dates (Montana et al. forthcoming), an issue that has not been addressed in studies using these data (Swaminathan and Mukherji 2012). A second reason may be the variation in components that make up the four definitions. While the Census includes notification (i.e. recognition as a legal settlement by a governing body), the other definitions consist of a variety of characteristics associated with slumdwelling, with one definition based on enumerator observation alone and the two others differing significantly in terms of the number of slum-related indicators the households and by extension the communities in which they are located must exhibit. Distinctions such as these are particularly important in India, where legal status confers rights to provision of public services. A definition that depends solely on government recognition will underestimate the prevalence of communities with slum characteristics (Agarwal 2011) and will likely overlook communities experiencing the highest levels of exclusion and disadvantage (Subbaraman et al. 2012), while definitions that include notification along with other area characteristics will be more inclusive. The regression results namely, that child height for age and weight for age are negatively associated with only one slum definition net of individual

21 L a u r a B. N o l a n 79 and other household characteristics point to the need for a more inclusive approach to studying the relationship between area-level deprivation and child health. Given the many probable adverse health effects of slum-dwelling (Rice and Rice 2009) and the mechanisms by which disease and poverty are thought to be perpetuated in urban areas (Wratten 1995), it is surprising that more robust slum effects were not uncovered in these analyses. It may be that neighborhood effects influence adult health more significantly than they do child health, as has been found in sub-saharan Africa (Günther and Harttgen 2012), or that individual and household characteristics are much more proximal and relevant for child height for age (Fink et al. 2014). A relatively small effect (as compared to that of individual-level characteristics) of neighborhoods has also been found in cities in developed countries (Fitzpatrick and LaGory 2003). Why might the UN definition in particular be the only slum indicator associated with poor child health? One possibility is that it is based on household reporting of current living conditions, which may better capture disadvantage than the Census, which was based on administrative data collected almost five years before the survey took place. Slum areas are dynamic in nature, and frequent updates are needed to maintain the accuracy of their identification (Montana et al., forthcoming). Indeed, information collected at the same time as child health measurement may be particularly informative about health hazards in the immediate vicinity. Further, the way in which slums are characterized by the UN definition is particularly well suited to identifying the characteristics of urban disadvantage that may be most predictive of poor health. The results presented here indicate that a number of individual components included in the UN definition are particularly important, especially housing quality and crowding. While one can imagine that the use of modern floor or roof material may protect against flooding and the spread of diseases, housing type may also be indicative of other unmeasured neighborhood advantages. Similarly, high-density living arrangements may promote the spread of infectious diseases, and may be a proxy for other characteristics of the area such as large amounts of garbage or open defecation. Since there is no absolute standard against which to compare these alternative definitions, it is not possible to discern which of these explanations is most indicative of the underlying process at work. Researchers, rather than trying to devise a better definition, should employ a number of options to evaluate their robustness, and policymakers should be aware of definition sensitivities and prioritize efficient data collection. Continued investigation of intra-urban differentials in health (Montgomery 2009) is therefore essential, as is more widespread acknowledgment that slums are not homogeneous entities (Gaur et al. 2013), but complex and dynamic. Important initiatives and areas of future research include distinguishing between legal and illegal slum settlements in household surveys in order to further examine between-slum inequality and to promote comparability with the 2011 Census designations; and investigating the use

22 80 S l u m D e f i n i t i o n s i n U r b a n I n d i a of a slum scale, which may provide more information than a dichotomous slum measure. The use of a slum scale will allow insight into whether there is a cumulative nature to the negative effects of slum-dwelling and/or a nonlinear relationship between slum adversity and health outcomes, which has been found for common mental disorders in a particularly deprived slum community in Mumbai (Subbaraman et al. 2014). Indexes of urban deprivation have previously been used in the study of the urban environment more generally (Dahly and Adair 2008). While a slum index was proposed at an Expert Group Meeting on Urban Indicators at the United Nations in 2002 (UN-HABITAT Urban Secretariat and Shelter Branch 2002), further action does not yet appear to have been taken (Doig 2014). Slum growth is not inevitable (Ooi and Phua 2007). City governments can and should take responsibility for strategic planning and intervention on behalf of deprived urban populations by linking their area s economic development with urban growth, housing, and the infrastructural needs of the individuals and families who move to cities seeking a better quality of life. Notes Support for this research was provided by grants from the Eunice Kennedy Shriver National Institute of Child Health and Human Development of the National Institutes of Health (P32CHD and T32HD007163). 1 The 2011 Census defines a slum as a residential area where dwellings are unfit for human habitation for reasons of dilapidation, overcrowding, faulty arrangements and design of such buildings, narrowness or faulty arrangement of street, lack of ventilation, light, or sanitation facilities or any combination of these factors which are detrimental to the safety and health (Chandramouli, n.d.). The definition further distinguished between slums that are Notified by State or other Local authorities under any Slum or other Act; slums that are Recognized by local authorities but not formally notified under any Act; and Identified slums, which are defined as a compact area of at least 300 population or about households of poorly built congested tenements, in unhygienic environment, usually with inadequate infrastructure and lacking in proper sanitary and drinking water facilities. 2 All analyses employed STATA Statistical Software version 13 (StataCorp 2013) and R version (R Core Team 2014). 3 This wide variation in slum designation may be related to the diverse origins and evolution of slums in each of the eight cities. For example, some local governments may have a tradition of providing notification and services more quickly than others. Notified slums that receive services are not de-notified when their conditions improve; while their residents may be better off, they are still considered to live in a slum.