The Limits of Causal Inference
 
How do our notions of causality connect with the tests we use to attribute causality? Drawing on a distinction proposed by Hall (2004) between causal dependence and causal production, I argue that when we expand our notion of causality to include both these aspects, the potential outcomes framework is revealed to be powerful yet incomplete. When causal production occurs without causal dependence and visa versa, counterfactual tests will lead us to incorrectly conclude that there is no causal relationship between connected phenomena, or that causal relationships exists between events connected by omissions. While relatively few cases in political science exhibit complete dependence without production or visa versa, I will argue that many causal questions are at least somewhat influenced by these phenomena.  In short, Hall (2004)’s distinction between dependence and production forces us to narrow the number of cases for which counterfactual dependence is a complete test of causality and highlights the essential role of descriptive inference in the study of causal processes.
 
This work was presented at the Visions in Methodology Conference at UC Davis in May 2016. The working paper can be found here.