Abstract: Recent work on mediation analysis has detailed the assumptions required for estimating causal—explanatory—mechanisms. Most troublingly for the social sciences, it is now well know that conventional mediation analysis biased in the presence of unmeasured mediator-outcome confounding. Worse, not even randomized experiments can guarantee the absence of unmeasured mediator-outcome confounders. We introduce a new—and easily computed—estimator that identifies causal mechanisms in the presence of unmeasured mediator-outcome confounding, using a simple constraint that is commonly accepted in quasi-experimental research. We demonstrate the superior performance of our method, compared to other recent mediation estimators, by re-analyzing an active labor market intervention with known mediation effects. (Joint work with Yongnam Kim.)

Bio: Felix Elwert (Ph.D. Harvard, 2007) is Romnes Professor of Sociology and Professor of Biostatistics and Medical Informatics at the University of Wisconsin-Madison.  He develops methods for causal inference in sociology and studies problems of social stratification, education, and social demography in the United States and Europe. His research has been published in the American Journal of Sociology, the American Sociological Review, Demography, and Biometrics. He is the winner of the 2018 Leo Goodman Award from the American Sociological Association and the 2013 winner of the Causality in Statistics Education Award from the American Statistical Association.