Econometric Identification of Causal Effects: Graphical Causal Models in Practice

Abstract

It is well known that causal inference relies on untestable a-priori causal assumptions. Identification refers to whether a causal relationship can be inferred from observed statistical associations; it requires an understanding of what statistical associations are induced by those causal assumptions. Since the assumptions are untestable, a transparent description of their statistical consequences helps the readers. However, the relation between causal assumptions and their induced statistical associations may not be obvious. Graphical Causal Models developed in the computer science literature in the 1980s (Pearl 2009) help trace these consequences, and are therefore a tool for both analysis and exposition. In this paper I describe the technique and illustrate its application to several research settings, including a case study in auditing.