An integrated approach to casuality: the role of casual graphs

Abstract

Causal questions are central for most biomedical and social science studies. The main frameworks that allow the analysis of causal relations are Potential Outcomes and Causal Graphs. The approaches have often been compared, contrasting their relative strengths. This paper evaluates the implications of merging the two methodologies in an integrated approach. In particular, we assess how the limits of one can be compensated by the solutions provided by the other. The outlined approach employs causal graphs to discover and formalize a causal model that is then used as a guide to implementing potential outcomes identification strategies. The integrated approach could be beneficial to both frameworks. The assumptions of potential outcome methods can be assessed directly from a causal graph even in high dimensional contexts, thus making the obtained causal estimates more reliable. On the other hand, causal graphs can benefit from the several ad hoc identification strategies that have been developed in the potential outcomes literature.