Placebo Tests for Causal Inference

Andrew C. Eggers, Guadalupe Tuñón, Allan Dafoe

Research output: Contribution to journalArticlepeer-review

Abstract

Placebo tests are increasingly common in applied social science research, but the methodological literature has not previously offered a comprehensive account of what we learn from them. We define placebo tests as tools for assessing the plausibility of the assumptions underlying a research design relative to some departure from those assumptions. We offer a typology of tests defined by the aspect of the research design that is altered to produce it (outcome, treatment, or population) and the type of assumption that is tested (bias assumptions or distributional assumptions). Our formal framework clarifies the extra assumptions necessary for informative placebo tests; these assumptions can be strong, and in some cases similar assumptions would justify a different procedure allowing the researcher to relax the research design's assumptions rather than test them. Properly designed and interpreted, placebo tests can be an important device for assessing the credibility of empirical research designs.

Original languageAmerican English
JournalAmerican journal of political science
DOIs
StateAccepted/In press - 2023

ASJC Scopus subject areas

  • Sociology and Political Science
  • Political Science and International Relations

Fingerprint

Dive into the research topics of 'Placebo Tests for Causal Inference'. Together they form a unique fingerprint.

Cite this