27 juillet 2023
Yuehao Bai et al., « On the Efficiency of Finely Stratified Experiments », arXiv - économie
This paper examines finely stratified designs for the efficient estimation of treatment effect parameters in randomized experiments. In such designs, units are divided into groups of fixed size, with a proportion within each group randomly assigned to a binary treatment. We focus on parameters defined using moment conditions constructed from known functions of the observed data. We establish that the naive method of moments estimator under a finely stratified design achieves the same asymptotic variance as that obtained using ex post covariate adjustment in i.i.d. designs, and further that this variance achieves the efficiency bound in a large class of designs.