Year: 2013
Type: article
Abstract: We thank Dr. Talih for his commentary (1) on our paper regarding temporal comparisons of health disparities (2). Curiously, Talih agrees with our key conclusion: Changing relationships between dimensions of socioeconomic position (e.g., education and income) can result in bias when comparing disparities over time, especially when researchers rely on a single measured indicator of socioeconomic position. What are Talih's criticisms? They are that we 1) “reduce” education to income and treat education as a linear variable, disregarding the importance of credentials; 2) violate causal agnosticism via our use of causal directed acyclic graphs; and 3) aver that the slope index of inequality (SII) was not intended to be used for temporal comparisons. These erroneous concerns are all easy to address. First, we concur with Talih that education cannot be reduced to income and likely influences health through pathways above and beyond its association with material resources. Nevertheless, extensive evidence exists regarding associations between income and health and dramatic changes in both educational attainment and its relationship with income (3, 4). Thus, even if education has strong independent effects on health (e.g., via literacy), estimates of educational disparities in health (as shown by our models) will reflect both the independent mechanisms connecting education to health and the likely substantial effect of pathways involving income and material resources (refer to our Web Appendix 2 available at http://aje.oxfordjournals.org/), and these concerns hold whether education is treated as a continuous or categorical variable. Our findings of potential bias stand. Second, we find Talih's call for “causal agnosticism” to be highly problematic. Although we concur that the conditional independence approach he advocates (with only undirected edges) would yield the same simulated data as our directed acyclic graph–based approach, on a conceptual level, we argue that causally complex situations require researchers to be more explicit, rather than less explicit, about the variety of possible causal structures that can give rise to observed multivariate associations (Figures 1 and 2 of our paper) (2). Multiple competing causal graphs can sharpen thinking by requiring explicit attention to which variables are hypothesized to be confounders or mediators. Here, one is not trying to infer causality from regressions, as Talih asserts, but rather, using regressions as tools for estimating associations between nodes of a causal graph. We should not run from explicit causal propositions; rather, we should embrace and test them! Third, Talih would do well to refer back to the classic 1985 paper by Pamuk that introduced the SII (5). In her article, Pamuk explicitly stated the following: [the] change in the size of a given class is likely to reflect a selection progress that influences its relative mortality level. If, for example, upwardly mobile people are healthier, then we would expect an increase in the relative mortality of Social Class V as the proportion of the population in this class is reduced over time …. It would thus seem advisable to take the relative sizes of the classes, as well as their relative mortality, into account when measuring the degree of inequality if comparisons across time or place are to be made (5, p. 18). Thus, the SII as originally proposed was intended to enable comparisons of inequality over time or place (5, 6). Our intent is not to dispute the SII's usefulness as a descriptive measure of inequality at a single time point but, rather, to highlight that observed temporal changes in the SII must be interpreted and attributed carefully. In summary, we appreciate Talih's recognition of the important issues our paper raises. We look forward to other investigators furthering the critical work needed to improving monitoring trends in health inequities, because this information is vital for action for health equity. Author affiliations: Department of Social and Behavioral Sciences, Harvard School of Public Health, Boston, Massachusetts (Jarvis T. Chen, Pamela D. Waterman, Nancy Krieger); and Department of Sociology, Harvard University, Cambridge, Massachusetts (Jason Beckfield). Conflict of interest: none declared.
Source: American Journal of Epidemiology
DOI: 10.1093/aje/kwt040
Institution: Harvard University
Language: English
Subfield: Health
Field: Social Sciences
Domain: Social Sciences
Open Access status: bronze
OpenAlex ID: W2130724148