|
Abstract:
|
A basic feature of many field experiments is that investigators are only able to randomize clusters of individuals even when individuals are the unit of interest. To recoup some of the resulting efficiency loss, many studies use the ``matched-pair cluster-randomized design.'' Other studies avoid pairing, because some claim to have identified serious problems with this design. We prove that all such claims are unfounded, and show that the estimator favored in the literature is appropriate only in situations where matching is not needed. To address this problem, we propose a simple nonparametric estimator with improved statistical properties. We show that from the perspective of bias, efficiency, power, or robustness, pairing should be used whenever feasible. We develop these techniques in the context of a randomized evaluation of the Mexican Universal Health Insurance Program.
|
