A1619
Title: A new design-based variance estimator for finely stratified experiments
Authors: Xun Huang - The University of Chicago (United States) [presenting]
Abstract: Design-based inference for the average treatment effect in finely stratified experiments is considered. Design-based inference refers to settings where the only source of uncertainty stems from the randomness in treatment assignment itself; finely stratified experiments refer to settings where units are first stratified into groups of a fixed size according to baseline covariates and then, within each group, a fixed number of units are assigned uniformly at random to treatment and the remainder to control. A novel estimator of the variance of the difference-in-means based on pairing adjacent strata is presented. This estimator is well defined even in the challenging setting where there is exactly one treated or control unit per stratum. The estimator is proven to be upward-biased, and thus can be used for inference under mild restrictions on the finite population. Comparison with well-known estimators proposed previously in this setting demonstrates that, while these estimators are also upward-biased, the proposed estimator has smaller bias and therefore leads to more precise inferences whenever adjacent strata are sufficiently similar.
