Introducing the meta-analytic event study
Event studies are among econometricians’ tools for estimating dynamic treatment effects. They rely on the general assumption that outcomes evolve stably before an event but may diverge afterwards (i.e., ‘parallel trends’). Another tool is meta-regression, which economists increasingly use to explain between-study variance in outcomes. Marrying the two methods yields the meta-analytic event study, which exploits temporal variation in study-level estimates to identify an event’s impact on treatment outcomes.
This presentation briefly introduces the meta-analytic event study approach using data from dozens of correspondence experiments on hiring discrimination. The steady growth of such experiments since the turn of the millennium permits constructing a large, unbalanced location–time–occupation panel meta-dataset with over 1,300 study-level estimates. I use these data to evaluate how the Great Recession impacted ethnic discrimination in hiring. The expectation is that slack labour market conditions temporarily intensified discrimination.
Methodologically, I extend the commonly used unrestricted weighted least squares (UWLS) estimator of Stanley and Doucouliagos (2015) to a difference-in-differences event study framework in R. The approach isolates the crisis’s impact while controlling for relevant study-level characteristics and publication-selection bias from small-sample studies with high-variance estimates. I also illustrate the compatibility of the method with Sun and Abraham’s (2021) estimator, which accounts for the differing timing of the crisis’s impact across Europe and North America.