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

In matched-pairs experiments in which one cluster per pair of clusters isassigned to treatment, to estimate treatment effects, researchers often regress theiroutcome on a treatment indicator and pair fixed effects, clustering standard errorsat the unit-of-randomization level. We show that even if the treatment has noeffect, a 5%-level t-test based on this regression will wrongly conclude that thetreatment has an effect up to 16.5% of the time. To fix this problem, researchersshould instead cluster standard errors at the pair level. Using simulations, weshow that similar results apply to clustered experiments with small strata.