A Methodology to Analyse the Intersections of Social Inequalities in Health

Abstract

An important issue for health policy and planning is the way in which multiple sources of disadvantage, such as class, gender, caste, race, ethnicity, and so forth, work together to influence health. Although ‘intersectionality’ is a topic for which there is growing interest and evidence, several questions as yet remain unanswered. These gaps partly reflect limitations in the quantitative methods used to study intersectionality in health, even though the techniques used to analyse health inequalities as separable processes can be sophisticated. In this paper, we discuss a method we developed to analyse the intersections between different social inequalities, including a technique to test for differences along the entire span of the social spectrum, not just between the extremes. We show how this method can be applied to the analysis of intersectionality in access to healthcare, using cross‐sectional data in Koppal, one of the poorest districts in Karnataka, India.

Key words:

• Intersectionality
• Social inequalities
• Gender
• Economic class
• Methodology
• Health
• Karnataka
• India

Acknowledgements

Much of the detailed empirical work that has led to a number of our insights on intersectionality was done for the Gender and Health Equity Project in Koppal district of northern Karnataka, India. We would like to acknowledge the work of Shon John, who assisted us in developing the methodology presented in this paper, and Asha George, who was a key member of the research team.

A number of the ideas worked through in this paper were presented at the National Conference in Honour of Prof. A. Vaidyanathan: Macroeconomic Policy, Rural Institutions, and Agricultural Development in India, held at the Institute for Social and Economic Change, Bangalore, 9–10 April 2006. We are grateful to the participants and especially to Prof. Vaidyanathan for his insightful comments and his sustained interest in understanding social inequalities.

Notes

1 There has been considerable discussion about the appropriateness of the terms ‘sex’ and ‘gender’. It has been argued that sex should be used whenever the reference is simply to male–female distinctions, for example, in data; and that ‘gender’ should be used only to refer to social processes and social relations. However, this can sometimes be cumbersome and potentially confusing for readers who are not aware of the differences in usage, especially when discussion of data and processes is intermingled. In this paper, therefore, we use the term ‘gender’ to refer to both the data and the processes.

2 This is not to suggest that qualitative approaches are not important to our understanding of intersectionality, but they need to be balanced and their evidence corroborated by quantitative results.

3 The gradients (slopes) of fitted lines across cross‐sectional data relating health outcomes to economic or social status are used in the health field as a simple measure of inequality (Anand et al., 2001).

4 A recently published study for the USA (Cummings and Jackson, 2008) attempted a similar approach to estimate the effects of sex, race, and socio‐economic status changes in perceptions of health over time. Although the study poses an interesting research question, there are grounds for reservation about the conclusions drawn from the model specification and estimation. First, the dependent variable (namely, self‐assessed health status) is an ordinal variable on a discrete scale of one to four. However, the model treats it as a continuous variable in order to apply the classical ordinary least‐square method of estimation. Although the sample size is reasonably large (n = 1500), it does not ensure that the model assumption of normal distribution of the error term is valid, which is required for the subsequent tests of significance. Second, the conclusions are drawn from simple comparisons of the estimated coefficients without testing. Standard tests of significance could be applied for the purpose, provided of course the residuals passed the diagnostics tests for the assumption as stated earlier.

5 Although we included caste as a possible explanatory variable initially, it did not have explanatory power for the health data, as its effects seemed to be overwhelmed by economic class differences.

6 Our data (Iyer, 2007) suggest that treatment being too expensive was the reason why nearly 49% of the poorest men discontinued treatment (versus 30% of poor men and 13% of non‐poor men).

7 Research using data for education provided rather different answers to such questions than the results for health (Iyer, 2007).