Who gains, who loses and how: Leveraging gender and class intersections to secure health entitlements

Social Science & Medicine 74 (2012) 1802e1811

Contents lists available at ScienceDirect

Social Science & Medicine journal homepage: www.elsevier.com/locate/socscimed

Who gains, who loses and how: Leveraging gender and class intersections to secure health entitlements Gita Sen*, Aditi Iyer Indian Institute of Management Bangalore, Bannerghatta Road, Bangalore 560 076, Karnataka, India

a r t i c l e i n f o

a b s t r a c t

Article history: Available online 7 July 2011

This paper argues that a focus on the middle groups in a multi-dimensional socioeconomic ordering can provide valuable insights into how different axes of advantage and disadvantage intersect with each other. It develops the elements of a framework to analyse the middle groups through an intersectional analysis, and uses it to explore how such groups leverage economic class or gender advantages to secure entitlements to treatment for long-term illness. The study draws upon household survey data on health-seeking for long-term ailments from 60 villages of Koppal district, Karnataka (India). The survey was designed to capture gender, economic class, caste, age and life stage-based inequalities in access to health care during pregnancy and for short and long-term illnesses. There were striking similarities between two important middle groups - non-poor women and poor men e in some key outcomes: their rates of non-treatment when ill, treatment discontinuation and treatment continuation, and the amounts they spent for treatment. These two groups are the obverse of each other in terms of gender and economic class advantage and disadvantage. Non-poor women have an economic advantage and a gender disadvantage, while poor men have the exact opposite. However, despite the similarities in outcomes, the processes by which gender and class advantage were leveraged by each of the groups varied sharply. Similar patterns held for the poorest men except that the class disadvantage they had to overcome was greater, and the results are modified by this. Ó 2011 Elsevier Ltd. All rights reserved.

Keywords: Gender Economic class Intersectionality Middle groups Leveraging Treatment Entitlements India Health-seeking Access to care Long-term illness

Introduction e an approach to intersectionality The growing social science literature on intersectionality highlights the fact that multiple axes of power and socioeconomic inequality do not only operate together but interact in complex ways that affect individual lives, social practices, institutional arrangements and cultural ideologies (Davis, 2008). Collins (2000) identified intersectionality as the micro processes that determine the social positions that individuals and groups occupy within a macro system of “interlocking oppressions”. The value of examining these intersections derives from its opening of new analytical and empirical terrain, and new ways of addressing policies to tackle inequality and disadvantage along multiple dimensions (Bredstrom, 2006; Krieger et al., 2008; Iyer, Sen, & Ostlin, 2010). A growing number of authors (Lynch & Kaplan, 2000; Ostlin, 2002) have pointed out that women’s and men’s exposure and vulnerability to ill-health, their access to health protective

* Corresponding author. Tel.: þ91 80 26993076. E-mail address: [email protected] (G. Sen). 0277-9536/$ e see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.socscimed.2011.05.035

resources, and the consequences to them of illness, violence and disability are shaped by social relations of power that manifest along dimensions such as gender, economic class, caste, race and ethnicity. By examining how these multiple dimensions interact with each other, an intersectional analysis of social inequities in health tells us, inter alia, whether the burden of inequity among the poor is borne equally by different castes or racial groups or by different household members. It allows us to examine, whether women, men, income earners or heads of poor households are equally trapped by medical poverty, and whether they are treated alike in the event of catastrophic illness or injury. It explores whether households tighten their belts equally for men and women when health costs go up overall, and whether these patterns are similar across different income quintiles. Such analysis poses a challenge for policies to ensure not only equity across, but also within, households. Theoretical and empirical work on intersections and their impact on health is necessary, therefore, for the advancement of social theory, for better empirical analysis, and for appropriate social and health policies. One challenge, however, that much intersectionality research has faced until very recently is the absence of a simple methodology for

G. Sen, A. Iyer / Social Science & Medicine 74 (2012) 1802e1811

quantitative analysis of multiple intersections. As a result, it has been very difficult (if not impossible) to compare groups placed at different points along a multi-dimensional socioeconomic spectrum. This has led much analysis to focus on differences between groups at the extremes that typically have only advantages or only disadvantages. A comparison between such groups generates large and statistically significant differences as might be expected. Similar large differences may not hold, however, when one analyses groups in the middle of the social spectrum, who manifest different mixes of social advantage and disadvantage (e.g., Griffin, Fuhrer, Stephen, & Marmot, 2002). Analysis of these middle groups can provide more nuanced understanding of the intersections among different axes of power, as well as their relative effects. This is because crucial politics of leveraging, competition, accommodation, negotiation, or cooptation often happens with groups in the middle of the social spectrum (Agrawal & Aggarwal, 1991; Kaul, 1993). Recent work by Sen, Iyer, and Mukherjee (2009) has shown one way to fill this methodological gap, and thereby to rigorously compare any socioeconomic group with any other. This paper builds on the analysis developed therein to explore how groups in the middle of the social spectrum leverage gender or economic class advantages in order to secure access to health care. Leveraging social advantages - towards a framework Leveraging entitlements It is well recognised that the relationships among different social groups are often shaped by the struggle for resources and entitlements. Such struggles and the associated competition, collaboration and negotiation happen both in the public sphere (as in arguments over quotas/reservations/affirmative action in education, employment and political representation), and in the private space within households (over allocation of household resources by gender, lifecycle status, age, ability, etc.). Ever since Amartya Sen’s (1990) well-known discussion of the family as the site of cooperative conflict, the question of how entitlements are governed by the fallback positions of different members of the household has been a subject of inquiry (e.g., Agarwal, 1994,1997; Kabeer, 1994; Katz, 1997; Pfeiffer, 2003). For example, a woman’s fallback position in terms of educational level or income earning status can affect her power to negotiate health resources or other entitlements. In this paper, we develop the concept of leveraging in order to go beyond a focus on the positionality of particular members to the processes by which positions may be translated into entitlements. Positionality itself is, nevertheless, crucial. An individual’s position depends both on the social and economic characteristics of her/his household (wealth, income, caste, ethnicity) and on her/his own characteristics (gender, age, marital/lifecycle status, assets, income earning) and therefore position within the household. Some of these characteristics confer advantages, and others clearly do not, although it is important to recognise that which ones do so is both historically shaped (and may change over time) and context specific. Thus, different groups are endowed with specific social advantages and disadvantages along different dimensions. While economic advantage (via assets and income) is well recognised, other dimensions such as gender, caste, ethnicity, also provide important and sometimes pre-eminent sources of advantage/ disadvantage. Leveraging occurs as groups use their advantages along some dimensions to compensate for disadvantages along others. For example, disabled persons in affluent households are usually better able to mitigate the consequences of their disability than similarly abled persons in poor households, by using better technological aids, having regular carers, etc. The former are thus

1803

able to leverage their economic class position to compensate for other disadvantages. Clearly, leveraging is not relevant for the groups at the extremes of a multi-dimensional social spectrum, i.e., those who are blessed with having only advantages and therefore do not need to leverage, or those who have only disadvantages and have nothing to leverage. Leveraging is important, however, for the “middle groups”, who have a mix of advantages and disadvantages. Such leveraging by one (middle) group may be independent of other (middle) groups, but it may also set groups against each other or, alternatively, lead to collaborative behaviour. Competition for resources/entitlements is the other face of leveraging, but not all middle groups compete directly with each other. The extent of direct competition depends on whether and how two groups meet each other in private or public spaces, and the social relations between them. Thus, poor men do not usually compete with non-poor women in the private sphere of the household, but their interests may clash in the public space over education or employment quotas. In addition, because each middle group has a particular combination of advantage/disadvantage, the mechanisms of leveraging will vary across groups. Who are the middle groups? The simplest definition of middle groups is that they are the groups that are not at the extremes. As the number of dimensions of difference or inequality increase, so too does the number of groups in the middle. Not all of these groups are on par with each other as we will argue. For purposes of illustration, if we treat gender and economic class as simple dichotomies, we have four groups: non-poor men, non-poor women, poor men and poor women. In this case, the extreme groups with only advantages or only disadvantages are non-poor men and poor women; the middle groups, each with a mix of advantages and disadvantages, are non-poor women and poor men. However, if there are more than two categories along one or both dimensions, the number of middle groups increases and the process of distinguishing one group from the other can become more complex. Consider, for example, the case where economic class is represented by three categories e poorest, poor and non-poor e and gender by two e women and men e (Fig. 1). The extreme groups can still be determined on the basis of apriori assumptions that are likely to be uncontroversial in most cases. There would be little argument against identifying non-poor men and the poorest women as extreme groups that are respectively doubly advantaged and doubly disadvantaged. However, the other groups in the middle vary in their capacity to leverage advantage and/or dominate other groups. By dominance, we mean being better off along a dimension, e.g., non-poor over poor over poorest, or men over women. It is B

A

Men

Women D

Poorest Dominant middle groups

C

Poor Subordinate middle groups

Non-poor x Extreme groups

Direction of dominance

Fig. 1. Types of gender and class-based groups.

1804

G. Sen, A. Iyer / Social Science & Medicine 74 (2012) 1802e1811

important to note that dominance as defined here implies hierarchy or ordering, but does not necessarily imply power or authority. In specific cases, however, dominance may also confer power/authority. A group is defined as being dominant over another if it is better off along one dimension, and at least equal along others. In Fig. 1, there are technically four “middle groups”; i.e., those which are not at the extremes e poor men (A), poorest men (B), non-poor women (C), poor women (D). Focussing on these groups, it is clear that poor women and the poorest men are not dominant groups. Poor women are dominated by both poor men and non-poor women, while the poorest men are dominated by poor men. Among the middle groups, the two categories of poor men and non-poor women constitute the dominant groups with the difference that poor men dominate both poor women and the poorest men, while non-poor women only dominate poor women. The importance of the dominant groups is that they are likely to be the ones with the strongest possibility of leveraging, i.e., using one’s advantage along one dimension to counteract disadvantage along another. Groups that are worse off than these dominant categories may also be able to leverage some advantage; e.g., the poorest men can leverage the fact of being male vis-à-vis the poorest women, but one would expect this capacity to be lower than that of the dominant groups. The likely relationship among subordinate categories cannot be hypothesised apriori. When inequalities are analysed along three dimensions, say, gender, race and class, the figure becomes three-dimensional and the process of distinguishing between the middle groups more complex. However, our methodology still works even in such cases. Analysing the middle groups e context and methodology In this section of the paper, we first describe the specific context for our research. Next, we describe the methodology that we use to analyse how the middle groups leverage their advantages to secure entitlements to health care for long-term ailments. The context There is a considerable social science literature on the multidimensional nature of socioeconomic inequality in South Asia. Economic class, gender systems, caste, and religion interact in complex ways to shape the experience of inequality of different sub-groups in the population. Gender power relations within households mean that girls are often fed less, educated less, and have poorer access to health care, especially if they have older sisters (Harriss, 1995; Pande, 2003). Restrictions on their physical mobility, sexuality, and reproductive capacity are perceived to be natural; and in many instances, accepted codes of social conduct and legal systems condone and even reward violence against them. While the above is true for women as a whole vis-à-vis men, there can be significant differences based on age or lifecycle status (Das Gupta, 1995), as well as on the basis of economic class, religion and caste (Jejeebhoy & Sathar, 2001; Mumtaz & Salway, 2007). The impact of gender power is often worse for women from better off or upper caste households than for their poorer and lower caste sisters who are less likely to have their physical mobility restricted; but this does not hold uniformly. The data for our research are drawn from a drought-prone agrarian district of Karnataka state in southern India. Koppal district is characterized by widespread poverty and social inequalities based on gender and caste. Belief systems and traditional practices are strongly gender-biased and inimical to women’s health and wellbeing. For women, adverse gender beliefs and practices mean lower value for their lives; relentless and unremunerated labour; early marriage and childbearing amidst poor

pregnancy outcomes, uncertain child survival and son preference, and lower entitlements to health care due to lack of acknowledgement of their health needs (George, Iyer, & Sen, 2005). Our field work revealed gender, economic class and caste to be the most important axes of power relations and dynamics in Koppal. Health care in the district is poor overall, and is delivered by informal providers comprising traditional healers and unqualified practitioners of allopathic medicine, together with a few formally qualified private providers practising in small towns, and a poor quality government-run public health sector. Most forms of health care have to be purchased out-of-pocket, making cost a major consideration for access. These features of the health system together with social inequalities shape the entitlements of different groups to health care. The methodology Entitlements to health care can be of different kinds. They include inter alia the entitlement to begin treatment of any kind, and to continue receiving treatment after it is begun. We focus here on how the middle groups draw upon their gender and/or class advantages to secure these entitlements. Although we began our analysis with the hypothesis that caste would play an independent role, the statistical analysis revealed this not to be the case as its influence seems to be picked up by the economic class variable. The core of the methodology is that it assigns a separate and unique identity to each sub-group. Thus, when logit regressions (both binomial and multinomial) are run with these uniquely identified groups, it becomes possible to test the significance of differences between them, pair-wise, anywhere on the social spectrum (Sen et al., 2009). This makes it possible to compare middle groups with each other as well as with those on the extremes. We used this methodology to analyse household survey data on health-seeking for long-term ailments from 60 villages of Koppal district, Karnataka. The survey, which was conducted for the Gender and Health Equity Project in 2002, was designed to capture gender, class, caste, age and life stage-based inequalities in access to health care during pregnancy and for short and long-term illnesses. Ethical approval for the research was obtained from the Institutional Ethics Committee at the Centre for Public Policy, Indian Institute of Management Bangalore. A village census preceding the survey enumerated 15,358 households in 60 villages, which formed the sample units in a unistage-stratified sampling frame. The villages affiliated to the same Primary Health Centre constituted a stratum. Within each stratum, households were first grouped by religion-caste and then by economic class (measured by monthly consumption expenditure). A stratified sample of 12.5% of all households was drawn resulting in 1920 households and 12328 individuals within them. Of these, 1364 individuals were found to have long-term ailments (extending beyond 90 days) during the recall period. Dropping those with incomplete or ambiguous data resulted in a sample of 1290 which was used. The instrument was a structured interview schedule with checkoff boxes plus additional space for comments. The survey defined treatment as all actions taken to alleviate illness symptoms, including self-care and treatment by unqualified persons. Consequently, non-treatment meant no attempt whatsoever to reduce symptoms. We ran multinomial logit regressions of the likelihood of nontreatment and discontinued treatment, and also binomial logit regressions of continued treatment. Prior to running the regressions, each of the variables were regressed individually for each type of entitlement (i.e., being able to start and continue

G. Sen, A. Iyer / Social Science & Medicine 74 (2012) 1802e1811

treatment). The variables included gender, class (measured by average per capita monthly consumption expenditure), caste, age (in completed years), illness severity (indicated by reported difficulty in doing routine work for 90 days or more prior to the date of the interview), household headship (i.e., being the household’s main decision maker) and income earning status (indicated by participation in work for income). Only significant variables were retained. Diagnostic tests were then run using SPSS for Windows (Version 14) to identify any multi-collinearity between the variables. The results, which revealed tolerance values greater than 0.1 and VIF (Variance Inflation Factor) values below 10, indicated that the independent variables were not highly correlated among themselves. The effects of gender and class inequalities were tested by using non-poor men as the reference group with separate dummies for each of the other sub-groups, while controlling for age, illness severity, household headship and income earning. The significance of differences between the groups was tested pair-wise using chisquared tests on the regressed variables. As the survey was conducted on a stratified random sample, population estimates were computed (by weighting the data for each household by the probability of its selection) and used in the cross tables and regressions. The robust standard error was used to correct for any heteroscedasticity while calculating p values. The estimates in the regressions were generated using the NewtoneRaphson method for maximum likelihood estimation in STATA (version 9). The odds ratios for gender and class-based sub-groups are presented in diagrams to facilitate easy identification of similarities and differences, especially among the middle groups (Figs. 2, 5 and 7). The groups being analysed are lined up on the X-axis in ascending order of their respective odds ratios, which are plotted, in log scale, along the Y-axis. Groups that are significantly different

1805

100 I didnot know what to do

Percentage

80

Not getting cured/incurable

60

Too expensive

40

I/family did not think it necessary

20 0 Poorest (N=189)

Poor (N=332)

Non-poor (N=223)

Women

100 I didnot know what to do

Percentage

80

Not getting cured/incurable

60

Too expensive

40

I/family did not think it necessary

20 0 Poorest (N=32)

Poor (N=124)

Non-poor (N=72)

Men

Fig. 3. Main reasons for non-treatment of long-term ailments: Differences by gender and economic class.

from the reference group are indicated using a different type of data point compared to those that are not significant. The diagrams are accompanied by matrices at the bottom, which contain the entire range of chi-squared test results, although the analysis primarily draws upon the results pertaining to the dominant middle groups.

Odds ratios

5.47 3.99 2.62

1.39 1.00 0.92

Poorest women

Poor women

p < 0.05

Non-poor women p > 0.05

Poor men

Non-poor men

Poorest men

x Reference group

Significance of differences in non-treatment Non-poor ♂ Poor ♂ Poorest ♂ Non-poor ♀ Poor ♀ Non-poor ♂ Poor ♂ Poorest ♂ Non-poor ♀ Poor ♀ Poorest ♀

-

-

* -

** ** ** -

Poorest ♀

** ** ** * -

Notations: Definition:

♂: men/boys, ♀: women/girls; - p>0.05, * p<0.05, ** p<0.01 Economic classes were defined by the consumption expenditure quintile to which they belonged: poorest (bottom quintile), poor (next 2 quintiles) and non-poor (top 2 quintiles).

Note:

Test of significance is the chi-squared test conducted on the regressed variables Fig. 2. Likelihood of non-treatment of long-term ailments: Differences by gender and economic class.

1806

G. Sen, A. Iyer / Social Science & Medicine 74 (2012) 1802e1811

50.0

100

41.1

40.0

80

26.1

Percentage

25.0

30.0 20.0 4.2

10.0

0.0

0.0

Non-poor men

Poorest men

Not cured or any better

60

Too expensive

40

I/family did not think it necessary

20

0.0 Poorest women

Poor men

Poor women

Non-poor women

0 Poorest (N=426)

Poor (N=745)

Non-poor (N=522)

Percentage of non-treatment due to lack of acknowledgement Women

Note: Non-treatment due to lack of acknowledgement = [(percentage of people who cited their illness was not serious) x (percentage of those who cited non-serious illnesses but had 100

difficulties doing work)] / 100

80 Percentage

Fig. 4. Non-treatment due to lack of acknowledgement of the need for care: Differences by gender and economic class.

Differences in the monthly expenditures incurred on treatment were tested using the non-parametric Mann Whitney U Test (for two independent samples), as the distributions were skewed and not suitable for the use of parametric tests. The distributions were log-transformed to enable better symmetry prior to running the tests. To correct for Type 1 errors due to the multitude of tests being run, the Bonferroni Correction was applied, wherein the p values obtained for each test was multiplied by the number of tests.

Odds ratios

Poor women

I/family did not think it necessary

0 Poorest (N=200)

Poor (N=320)

Non-poor (N=308)

Men

Fig. 6. Main reasons for discontinued treatment of long-term ailments: Differences by gender and economic class.

access to and control over intra-household resources, being able to mobilise resources outside the home via loans or sale of assets, and drawing on the labour of other household members including children.

Previous applications of our methodology (Iyer, 2007; Iyer, Sen, & George, 2007; Sen, Iyer, & George, 2007) showed that the ability to use economic and/or caste advantage to gain access to health care and education was mediated by gender. The analyses also showed that the consequences of gender relations on the ability to leverage entitlement varied by economic class. In this paper, our objective is to understand specifically how the “middle groups” leverage gender and/or economic class to gain access to health care for long-term ailments. In particular, we study the mechanisms that the dominant middle groups - non-poor women and poor men e use. These include drawing support from gendered norms, having

Poorest women

Too expensive

40 20

Results

1.05 1.00 0.90

Not cured or any better

60

Entitlement 1: to begin treatment Leveraging household support for the decision to begin treatment of any kind is the first step towards securing access to health care. We found that, as expected, the groups at the extremes were significantly different from each other in treatment rates (Fig. 2, Table 1): compared to non-poor men, the poorest women were five times more likely to never seek treatment (OR: 5.47, p < 0.001). By

Poorest men

Non-poor women Non-poor men

Poor men

0.62 0.55

0.44 p < 0.05

p > 0.05

x Reference group

Significance of differences in the likelihood of continued treatment Non-poor ♂ Poor ♂ Poorest ♂ Non-poor ♀ Non-poor ♂ Poor ♂ * Poorest ♂ Non-poor ♀ Poor ♀ Poorest ♀

Poor ♀

Poorest ♀

** ** **

** ** ** -

Notations:

♂: men/boys, ♀: women/girls; - p>0.05, * p<0.05, ** p<0.01

Note:

Test of significance is the chi-squared test conducted on the regressed variables

Fig. 5. Likelihood of continued treatment of long-term ailments: Differences by gender and economic class.

G. Sen, A. Iyer / Social Science & Medicine 74 (2012) 1802e1811

1807

1.40 1.34

Odds ratios

1.15

Non-poor women

Poor women

1.00

Non-poor men

Poor men

Poorest women

Poorest men

0.71

0.56 p < 0.05

p > 0.05

x Reference group

Significance of differences in loan taking or sale of assets Non-poor ♂ Poor ♂ Poorest ♂ Non-poor ♀ Poor ♀

-

Non-poor ♂ Poor ♂ Poorest ♂ Non-poor ♀ Poor ♀ Poorest ♀

-

* ** **

* * -

Notations:

♂: men/boys, ♀: women/girls; - p>0.05, * p<0.05, ** p<0.01

Note:

Test of significance is the chi-squared test conducted on the regressed variables

Poorest ♀

** *

Fig. 7. Likelihood of loan taking or sale of assets to pay for continued treatment of long-term ailments: Differences by gender and economic class.

contrast, the dominant middle groups - non-poor women and poor men - had treatment rates that were quite similar to each other. The similarity among the dominant middle groups was the result of the intersecting pressures of gender and economic class. While class made treatment affordable, gender operated through the likelihood of an illness being acknowledged as serious enough to merit treatment. Men’s illnesses tended to be recognised, acknowledged and treated to a much greater extent than women’s illnesses. As is not uncommon in South Asia, women’s needs for food, nutrition and health are not treated on par with male needs through the entire lifecycle, and women themselves often internalise these biased beliefs and norms (Das Gupta, 1995; Harriss, 1995). Our research corroborated this. Fig. 3 provides the main reasons cited by both men and women for non-treatment. The reason “I/ family did not think it necessary” was much more important for women than for men. In particular, this reason accounted for as many as 62% of non-poor women who did not start treatment of Table 1 Likelihood of non-treatment and discontinued treatment for long-term ailments e Estimates of odds ratios. Independent variables

Age Severity Income earning status Household headship

Gender-class sub-groups

Odds ratios

Income earner Non-earner Household head Non-head Non-poor men Poor men Poorest men Non-poor women Poor women Poorest women

Nontreatment

Discontinued treatment

1.00 0.78** 1.00 0.66 1.00

1.00 0.94 1.00 0.69** 1.00

0.61 1.00 1.39 0.92 2.62*

1.60* 1.00 0.86 1.74* 0.86

3.99** 5.47**

1.46 1.76*

Notations: *p < 0.05, **p < 0.01. Note: Multinomial logit regression was used to estimate the odds ratios, wherein continued treatment ¼ 1, discontinued treatment ¼ 2, non-treatment ¼ 3.

any kind. Yet, two thirds of these women had ailments that were severe enough to cause difficulty in work for a median duration of two years (results not shown, but can be obtained from the authors). We infer from this that far from being non-serious, much of the lack of treatment stemmed from the refusal to acknowledge health problems either by the women themselves or by their families. Lack of acknowledgement was most serious for non-poor women, indicating the power of the negative influence of gender relations within better off households (Fig. 4). They were unable to leverage their class position enough to bring them on par with the men of their own households. This was a much less serious problem for poor women who, by virtue of income earning (and therefore better fallback positions), tended to have greater ability to acknowledge when they were ill and needed treatment than nonpoor women, who were less likely to be income earners. This did not however counter adequately their disadvantages in terms of economic class vis-à-vis non-poor women, and ended up with both groups having similar rates of non-treatment overall. For poor men, while there was some lack of acknowledgement, this did not really prevent them from getting treatment. The consequence was that all men e poorest, poor and non-poor e had statistically similar treatment likelihoods, suggesting men’s ability to leverage their gender advantage to compensate for poverty. As will be shown later, men were also able to convert this advantage into a larger share of the household’s overall resources for health. On the other hand, the group of non-poor women were only able to leverage their class advantage to a limited extent; gendered lack of acknowledgement tended to drag them down.

Entitlement 2: to continue receiving treatment Not all those who started treatment were able to continue with it until they got well. As expected between extreme groups, the poorest women were only half as likely to continue treatment as were non-poor men (OR: 0.44, p < 0.01). By contrast, among the middle groups, poor men (and even the poorest men) were similar to non-poor women in their continuation rates (Fig. 5).

1808

G. Sen, A. Iyer / Social Science & Medicine 74 (2012) 1802e1811

To understand how such dissimilar groups could have similar outcomes, and whether these outcomes came about through similar processes, we examined four crucial variables and how treatment continuation or discontinuation was mediated by them: the individual’s headship status, income earning status, the ability to mobilise financial and other resources, and the quality of providers sought. Impact of household headship In our sample, household heads, defined as the main decision makers, were largely but not exclusively men. Women headed 12.7% of all households. Female-headed households were either single member units or nuclear families in which they had no spouse. Compared to male heads, female heads were older, mostly single, poorer and less educated. Regressions to explore the relationship between gender, class and household headship used non-poor male heads as the reference group and separate dummies for other sub-groups, while controlling for age, illness severity and income earning. The results showed that headship mattered only in poor households and only for male heads vis-à-vis female non-heads. Male heads were much less likely to discontinue treatment than female non-heads (p < 0.001). However, this was not true vis-à-vis other men in their households. By contrast, women heads did not differ significantly from either male or female non-heads. Clearly, headship does confer leveraging possibility, but only to male heads and only vis-à-vis women household members. As the main decision makers, male household heads had greater control over decisions to continue treatment and greater bargaining power over non-heads, but only if the latter were women. By contrast, women heads had no such leveraging capacity at all. Impact of income earning The relationship between gender, class and income earning was similarly explored while controlling for age, illness severity and household headship. The first point to be noted is that, contrary to expectations, an income earner was more likely to discontinue treatment than a non-earner. Similar results have been found in Mukherjee (2007). The reasons are the opportunity costs in terms of foregone earnings and in addition, for women, the time requirements of domestic work. In non-poor households, women who did not go out to earn an income had significantly better chances of continuing treatment than other household members (including male and female earners) because there was no direct opportunity cost for their time (in terms of earning), and they were more likely to have somebody to take over household work when they needed to seek treatment. This is corroborated by the types of households in which they lived. Compared to women earners, relatively more women non-earners lived in joint families (22.2% versus 11.5%) or extended nuclear families (31.3% versus 16.2%), where there were other adult women with whom housework could be shared, thereby freeing them up to seek treatment when needed. By contrast, two thirds of the non-poor women earners lived in nuclear families with no other adult women with whom they could share the burden of domestic work. Men in Koppal seldom participate in housework. Thus, with no one to substitute for them in the home, and with significant opportunity costs of opting out of housework and wage work, non-poor women earners tended to opt out of treatment instead. In poor households, there was no difference between women who earned versus those who did not earn incomes. Non-earning women did not typically have the same kind of support from other women in their households, as their counterparts in better off households. Although 64.8% lived either in joint families or in

extended nuclear families, their position was not better than anyone else’s. Many more women in poor households work to earn incomes and hence the availability of free labour to support other women may be much lower than in non-poor households. The only real difference in poor households was between male and female earners with the latter being worse off. This was due to the fact that they had to juggle the time demands of both income earning and housework, which the men did not have to do. Indeed, income earning did not differentially affect discontinuation between male earners and male non-earners in any household, whether poor or non-poor. Compared to the dominant middle groups - non-poor women and poor men - the poorest men were significantly more likely to discontinue treatment (p < 0.01). However, they were no different from the poorest women: gender differences between them were non-significant. Neither headship nor income earning status significantly affected the likelihood of discontinuation among the poorest. However, there were important differences between the poorest men and other middle groups in the reasons for discontinuation (Fig. 6) Fewer of the poorest men thought that treatment was unnecessary compared to poor men (8% versus 20% respectively), but substantially more of the poorest men cited financial barriers compared to poor men (48.5% versus 30.3% respectively) and nonpoor men (13%). In other words, the ability of the poorest men to continue treatment depended on their ability to negotiate the financial barrier through mobilisation of resources outside the home. Mobilising resources for continued treatment In terms of the amount they spent each month to continue treatment, non-poor women were surprisingly similar to poor men, and even the poorest men (Table 2 in the summary section). Clearly, the economic position of the household is inadequate to explain this. Gender plays a crucial role. Poor men seem to have managed to corner a greater share of the household resources for continued treatment: they spent significantly more each month than poor women did (Medians: Rs. 167 versus Rs. 100; p < 0.01). This appropriation both reduced the likelihood of continued treatment by poor women (Fig. 5) and probably the quality of treatment received by them. Non-poor women also spent significantly less than non-poor men did (Medians: Rs. 125 versus Rs. 208; p < 0.01). The poorest men, by contrast, spent as much each month on continued treatment as the poorest women. Unequal spending on treatment partly reflects varying capacities and compulsions to mobilise resources outside the home. Given the costs of treatment, all households e poor and non-poor e borrow or sell assets to raise money for treatment. However, financial mobilisation is largely a prerogative of men. Non-poor women did not have the same access to credit or the market, and consequently, were less likely to take loans or sell assets than nonpoor men (p < 0.05, Fig. 7). In poor households as well, men were more likely to take loans or sell assets (p < 0.05). Only in the poorest households were men and women equally likely to mobilise external resources to continue treatment, if they could. While they did not raise financial resources outside the home, non-poor women did draw upon human resources within the home to continue treatment, or diverted financial resources from other purposes. The survey considered different types of social burdens due to treatment seeking. Of these, the most commonly cited burdens included cutting back on usual spending on food, and spouses/sons working longer or harder. Children being pulled out of school or pushed into wage work, or adults being pushed into wage work were less commonly cited. More non-poor women continued treatment without loans or sale of assets but via one or

G. Sen, A. Iyer / Social Science & Medicine 74 (2012) 1802e1811

more of the above social burdens as compared to poor men and the poorest men (21.1% versus 10.8% and 5.3% respectively). In reverse, relatively more poor and the poorest men mobilised financial resources with or without accompanying social burdens than nonpoor women (42.8% and 47.3% versus 26.8% respectively). These differences reveal the unequal terms of their access to and control over resources. Compared to the sale of assets, loan taking was the more common mode of resource mobilisation. This is hardly surprising, given the paucity of material and productive assets in Koppal. Most of these loans were taken in the informal credit market, mainly from moneylenders or employers, and wherever possible, from relatives. While this type of credit did enable men to continue treatment, it nevertheless risked the economic and social wellbeing of household members, as it entailed rates of interest as high as 36% per year on average (if a moneylender had to be repaid), and a possible loss of independence (if an employer had to be repaid). In the short term, though, loan taking enabled substantially higher spending on treatment by members of the male middle group compared to what they were able to spend with their own resources. This was in sharp contrast to both non-poor and poor women. Quality of providers sought The number of health care providers consulted prior to treatment discontinuation provides a clue to the urgency with which the middle groups sought to restore health, as providers were often changed when their treatment did not make the patients feel better. In this respect, non-poor women did not differ from poor men. However, the types of providers consulted by non-poor women were qualitatively different from those used by men. Non-poor women went substantially less to private hospitals run by qualified doctors than poor men (6.3% versus 12.6%), or poorest men (23.3%). Instead, they went relatively more to private ‘doctors’ who practised outside of hospitals than both poor men and the poorest men (52.3% versus 33.9% and 30.2% respectively). Many of these private solo practitioners in Koppal are unqualified, especially those who provide services within the village. All men made more contact with qualified private practitioners, who were generally more rigid with regard to the payment of fees, while women tended to rely more on unqualified practitioners, who could be more flexible. Non-poor women also went to significantly fewer providers than non-poor men (p < 0.01) before discontinuing treatment. It was only in their use of government facilities and service providers that non-poor women were similar to poor and the poorest men. The survey did not examine the quality of care rendered by the providers who were consulted. This is a limitation, as the quality of inter-personal communication and the effectiveness of treatment would probably have influenced the decision to continue treatment. Nor is it possible from the survey to analyse differences in the strategies used to leverage treatment by type of illness.

Summary Our analysis in this paper has thrown up both similarities and differences between the dominant middle groups in terms of treatment seeking for long-term ailments. The similarities are summarised in Table 2, which also contrasts these groups with the extreme groups and other middle groups. There were striking similarities between the dominant groups in their rates of nontreatment, treatment discontinuation and treatment continuation, and the amounts they spent for treatment, despite their opposing class and gender characteristics.

1809

Table 2 Significance of differences between groups in the middle and at both extremes.

Likelihood of non-treatment Likelihood of continued treatment Likelihood of discontinued treatment Median expenditures on continued treatment per month

Poor men versus non-poor women (dominant middle groups)

Poorest men versus non-poor women

Poorest men versus poor women

Non-poor men versus poorest women (extreme groups)

e

e

**

**

e

e

e

**

e

**

e

*

e

e

**

**

Notations: - p > 0.05, *p < 0.05, **p < 0.01.

However, when we probed the reasons behind these similarities, we found that the processes by which gender and class advantage were leveraged by each of the groups varied sharply. Poor men and even the poorest men were able to leverage gender power within their households to ensure rates of non-treatment that were as low as for the best-off group e the non-poor men. By contrast, a gendered lack of acknowledgement of their health care needs worked to disadvantage non-poor women so that their rates of non-treatment were worse than the men of their economic class group. Once some form of treatment was begun, the processes at work were somewhat different, and this was reflected in the impacts of headship and income earning, in the quality of providers sought, and the way in which resources were mobilised and used to ensure continued treatment. Poor men used their position as heads of household and as income earners to ensure much greater rates of continued treatment than the women of their households. Furthermore, as men, their capacity to borrow contributed to their expenditure on treatment being much higher than for the women in their households. The quality of the providers they sought tended to be much better as well. In all of the above, poor men seem to have been at a significant advantage relative to women in their households. For poor men, treatment continuation was at the expense of poor women, who were forced to drop out. Poor men’s entitlement was secured via gender power linked to their positions as household heads and income earners and the consequent control they were able to exert over treatment seeking decisions and the use of resources. From the perspective of non-poor women, the equal discontinuation rates identified in Table 2 between them and poor men were partly because of their privileged economic class position and their capacity to use it. Thus, they were able to ensure lower rates of discontinuation, and the non-earners among them were able to draw on the time of other women in their households. Continuing treatment for them did not entail any reduction in treatment for non-poor men or any disruption of existing gendered arrangements within the home. Indeed, the possibility of their continuing treatment depended on their ability to manage the demands of household labour with that of treatment seeking. They were also able to continue treatment through reduced spending on food and drawing on the labour of spouses/sons. Nonetheless, the quality of health providers they saw was worse than for any of the men. Non-poor women clearly drew upon their class endowments to buffer their gender disadvantage. They were able to have some

1810

G. Sen, A. Iyer / Social Science & Medicine 74 (2012) 1802e1811

access to financial and human resources within the home (over which they also had some control), although they were limited in their ability to borrow or sell assets. Poor men, in contrast, leveraged gender to counter their class disadvantage. They were able to corner resources (financial and human) within the home, and borrow or sell assets to secure the resources required to continue treatment. Similar patterns held for the poorest quintile of men except that the class disadvantage they had to overcome was greater, and the results are modified by this. While their likelihood of nontreatment was as low as for all other men, treatment continuation depended on their ability to negotiate the financial barrier. Thus, they did not spend significantly more than the poorest women did or go to significantly more providers. Nonetheless, they went to more qualified private providers who were generally more rigid with regard to payment, while the women went to mainly unqualified private doctors who could be more flexible. Although poor women do constitute a middle group, they are a subordinate group, and their ability to leverage social advantage is far more ambiguous, as it is not clear what social advantages they enjoy, being both poor and women. In practically all respects, their entitlements and outcomes were similar to those of the poorest women.

Implications This paper argues that a focus on the middle groups in a multidimensional socioeconomic ordering can provide valuable insights into how different axes of advantage and disadvantage intersect with each other. In doing this, the paper extends and deepens intersectionality research in two directions. First, it makes it clear that it is necessary to go beyond focussing on the well-known and often stark differences between groups that are at the extremes of the multi-dimensional socioeconomic spectrum. Unlike the groups at the extremes, those in the middle have a mix of advantages and disadvantages and the processes by which they leverage their advantages to counter their disadvantages can be complex and varied. The fact that the middle groups quite often constitute a significant majority in any given population means that an understanding of these processes is essential. Second, by further exploring the usefulness of an innovative quantitative methodology usable in larger datasets, the paper fills some of the lacunae in existing intersectionality research by providing complementary approaches to qualitative ones. The paper supports five propositions about leveraging among the “middle groups” that provide pointers and hypotheses for analysis beyond the health field. 1. Leveraging can lead to groups that are apparently dissimilar in their characteristics along different dimensions (economic class, caste, gender, ethnicity, etc.) ending up with similar entitlements and outcomes. This is especially so for the dominant middle groups. 2. Despite similarities in some outcomes, however, there can also be critical differences in others because different advantages work differently. 3. While all dimensions of inequality result from social hierarchies and there is no apriori reason to assume that one dimension is stronger than another, different dimensions of advantage may be leveraged more strongly in particular circumstances. In this paper, we show that both class and gender are leveraged to secure access to health care, but gender is often the more powerful lever. 4. Leveraging can occur both within and outside the household.

5. Factors such as income earning, household headship, the ability to borrow or sell assets, the ability to draw on the labour of other household members including children are important to the processes through which leveraging happens within households. But their relevance varies considerably across different middle groups; this is something that only an intersectional analysis can uncover. The propositions explored in the paper also have potentially important policy implications. They provide a handle for deepening our understanding of the politics of the middle groups. Some of these groups are often the ones with enough voice and resources to demand entitlements and challenge the status quo. This has implications outside health to a number of other areas such as education, jobs, social security, to name only a few. It also has implications for how policies address the politics of power within households. As we have seen, this varies across different subgroups of households; an intersectional analysis gives us a more nuanced understanding but it also demands a corresponding nuanced approach to policy making in support of entitlements.

Acknowledgements We are grateful to Chandan Mukherjee for partnering with us on the methodology that forms the basis for the present analysis. We were also fortunate to have Asha George’s and Margaret Whitehead’s support for the household survey, Paul Jacob’s inputs into the survey’s design, Ashish Kumar’s suggestions for database management, Frances Drever’s suggestions for the diagrams, and Shon John’s help with the computations. Without each of them, the survey and its analysis would have been poorer. Thank you also to the Robert Wood Johnson Foundation Health and Society Scholars Program Working Group on Gender and Health at Columbia University for editorial feedback.

References Agarwal, B. (1994). A field of one's own: Gender and land rights in south Asia. New Delhi: Cambridge University Press and Foundation Books in association with Book Review Literary Trust. Agarwal, B. (1997). 'Bargaining' and gender relations: within and beyond the household. Feminist Economics, 3(1), 1e51. Agrawal, S. P., & Aggarwal, J. C. (1991). Eductional and social uplift of backward classes at what cost and how: Mandal Commission and after. New Delhi: Concept Publication and Co. Bredstrom, A. (2006). Intersectionality: a challenge for feminist HIV/AIDS research? European Journal of Women's Studies, 13(3), 229e243. Collins, P. H. (2000). Black feminist thought: Knowledge, consciousness and the politics of empowerment. Boston: Unwin Hyman. Das Gupta, M. (1995). Life course perspectives on women's autonomy and health outcomes. American Anthropologist, 97(3), 481e491. Davis, K. (2008). Intersectionality as buzzword: a sociology of science perspective on what makes a feminist theory successful. Feminist Theory, 9(1), 67e85. George, A., Iyer, A., & Sen, G. (2005). Gendered health systems biased against maternal survival: Preliminary findings from Koppal, Karnataka, India (IDS Working Paper 253). Brighton, Sussex: Institute of Development Studies. Griffin, J. M., Fuhrer, Rebecca, Stansfeld, Stephen, A., & Marmot, M. (2002). The importance of low control at work and home on depression and anxiety: do these effects vary by gender and social class? Social Science & Medicine, 54(5), 783e798. Harriss, B. (1995). The intrafamily distribution of hunger in South Asia. In J. Dreze, A. Sen, & A. & Hussain (Eds.), The political economy of hunger: Selected essays (pp. 224e297). Oxford: Clarendon Press. Iyer, A. (2007). Gender, caste and class in health: Compounding and competing inequalities in rural Karnataka, India. Ph.D Thesis. Liverpool, U.K.: Division of Public Health, University of Liverpool. Iyer, A., Sen, G., & George, A. (2007). The dynamics of gender and class in access to health care: evidence from rural Karnataka, India. International Journal of Health Services, 37(3), 537e554. Iyer, A., Sen, G., & Ostlin, P. (2010). Inequalities and intersections in health: a review of the evidence. In G. Sen, & P. Ostlin (Eds.), Gender equity in health: The shifting frontiers of evidence and action (pp. 70e95). New York: Routledge.

G. Sen, A. Iyer / Social Science & Medicine 74 (2012) 1802e1811 Jejeebhoy, S. J., & Sathar, Z. A. (2001). Women's autonomy in India and Pakistan: the influence of religion and region. Population and Development Review, 27(4), 687e712. Kabeer, N. (1994). Reversed realities: Gender hierarchies in development thought. London and New York: Verso. Katz, E. (1997). The intra-household economics of voice and exit. Feminist Economics, 3(3), 25e46. Kaul, R. (1993). Caste, class and education: Politics of the capitation fee phenomenon in Karnataka. New Delhi: Sage Publications. Krieger, N., Rehkopf, D. H., Chen, J. T., Waterman, P. D., Marcelli, E., & Kennedy, M. (2008). The fall and rise of US inequities in premature mortality: 1960e2002. PLoS Med, 5(2), 0227e0241, 5(2, e46). Lynch, J., & Kaplan, G. (2000). Socioeconomic position. In L. F. Berkman, & I.Kawachi. (Eds.), Social epidemiology (pp.13e35). Oxford & New York: Oxford University Press. Mukherjee, S. (2007). The importance of quality and price in choice of health care: A study of West Bengal. Ph.D Thesis. Thiruvananthapuram: Centre for Development Studies.

1811

Mumtaz, Z., & Salway, S. (2007). Gender, pregnancy and the uptake of antenatal care services in Pakistan. Sociology of Health & Illness, 29(1), 1e26. Ostlin, P. (2002). Gender perspective on socioeconomic inequalities in health. In J. P. Machenbach, & M. J. Bakker (Eds.), Reducing inequalities in health: A European perspective (pp. 315e324). London and New York: Routledge. Pande, R. P. (2003). Selective gender differences in childhood nutrition and immunization in rural India: the role of siblings. Demography, 40(3), 395e418. Pfeiffer, J. (2003). Cash income, intrahousehold cooperative conflict, and child health in Central Mozambique. Medical Anthropology, 22, 87e130. Sen, A. (1990). Gender and cooperative conflicts. In I. Tinker (Ed.), Persistent inequalities (pp. 123e149). New York and Oxford: Oxford University Press. Sen, G., Iyer, A., & George, A. (2007). Systematic hierarchies and systemic failures: gender and health inequities in Koppal district. Economic and Political Weekly, XLII(8), 682e690. Sen, G., Iyer, A., & Mukherjee, C. (2009). A methodology to analyse the intersections of social inequalities in health. Journal of Human Development and Capabilities, 10(3), 397e415.