Placebo Tests for Causal Inference

Placebo tests allow researchers to probe the soundness of a research design by checking for an association that should be present if the design is flawed but not otherwise. Despite the growing popularity of placebo tests, the principles for designing and interpreting them have remained obscure. Drawing on a comprehensive survey of recent empirical work in political science, we define placebo tests, introduce a typology of tests, and analyze what makes them informative. We consider examples of each type of test and discuss how to design and evaluate tests for specific research designs. In sum, we offer a guide to understanding and using placebo tests to improve causal inference. Authors can be reached at allan.dafoe@politics.ox.ac.uk, andrew.eggers@nuffield.ox.ac.uk, tunon@princeton.edu. For helpful input, we thank Devin Caughey, Thad Dunning, Anthony Fowler, Don Green, Sophia Hatz, Luke Miratrix, Jas Sekhon, Nils Weidmann, Baobao Zhang, and the Yale and Uppsala students of Advanced Quantitative Methods. Hayley Pring, Rhys Dubin, and Jonne Kamphorst provided excellent research assistance. Audiences at ISA, Oxford, NYU-Abu Dhabi, Columbia, and the International Methods Colloquium provided useful feedback.