Causal Analysis in the Health Sciences

Abstract

The final quarter of the twentieth century witnessed a burgeoning of formal methods for the analysis of causal effects. Of the methods that appeared in the health sciences, most can be identified with approaches to causal analysis that originated much earlier in the century in other fields: counterfactual (potential outcomes) models, graphical models, and structural equations models. Connections among these approaches were elucidated during the 1990s, and the near future may bring a unified methodology for causal analysis. This vignette briefly reviews the counterfactual approach to causal analysis in the health sciences, its connections to graphical and structural equations approaches, its extension to longitudinal data analysis, and some areas needing further work. For deeper and more extensive reviews, I especially recommend Sobel's (1995) discussion of the connections among causal concepts in philosophy, statistics, and social sciences; Pearl's (2000) unified approach to counterfactual, graphical, and structural equations models; and Robins's (1997) review of causal analysis for longitudinal data.