Michael Sobel, University of Arizona

Causal Inference in the Social and Behavioral Sciences

Thursday, May 1 at 7:30pm in Physics-Astronomy A110

Social scientists routinely employ regression analysis and a variety of related statistical models to draw causal inferences from survey data. Typically, the parameters of the models are interpreted as effects that indicate the change in a dependent variable that would occur if one or more independent variables were set to values other than the values actually taken. I will show why this interpretation does not generally hold, even when the model is correctly specified and a causal theory is given. I will discuss some implications for the way in which social research is and should be conducted. In particular, the usual strategies for testing competing causal explanations are misdirected. Further, the emphasis on causation in contemporary social science is often misdirected.