Replication of Causal Effects with Quasi-Experimental Designs


Prof. Dr. Peter M. Steiner (University of Maryland)


Dienstag, 22. Juni, 14:00–16:00 Uhr



The workshop starts with an introduction to the Causal Replication Framework that lays out all the assumptions required for a direct replication of causal effect. That is, the conditions under which the causal quantity of interest (causal estimand) is identical and thus replicable across studies. Then, we briefly review the causal estimands and assumptions of the strongest quasi-experimental designs: Regression discontinuity and randomized encouragement designs (instrumental variable estimator), matching designs and gain score (difference-in-differences) estimators for observational studies, but also interrupted time series designs for longitudinal data. Finally, we discuss the importance of systematically controlling study factors for conceptual replications using multiple sites, switching replication or stepped wedge designs. Issues with regard to assessing replication success—choice of an appropriate correspondence measure and power implications—will be addressed. We also discuss an applied replication example.