At What Level Should One Cluster Standard Errors in Paired and Small-Strata Experiments?

Abstract

In clustered and paired experiments, to estimate treatment effects, researchers often regress their outcome on the treatment and pair fixed effects, clustering standard errors at the unit-of-randomization level. We show that even if the treatment has no effect, a 5%-level t-test based on this regression will wrongly conclude that the treatment has an effect up to 16.5% of the time, an error rate much larger than the researcher's 5% target. To achieve their targeted error rate, researchers should instead cluster standard errors at the pair level. Using simulations, we show that similar results apply to clustered experiments with small strata.

Tasks

Reproductions