Design and Analysis of Two-Stage Randomized Experiments

Kosuke Imai, PhD
Harvard University

Abstract: In many social science experiments, subjects often interact with each other and as a result, one unit's treatment can influence the outcome of another unit. Over the last decade, a significant progress has been made towards causal inference in the presence of such interference between units.  In this talk, we will discuss two-stage randomized experiments, which enable the identification of the average spillover effects as well as that of the average direct effect of one's own treatment.  In particular, we consider the setting with noncompliance, in which some units in the treatment group do not receive the treatment while others in the control group may take up one.  This implies that there may exist the spillover effect of the treatment assignment on the treatment receipt as well as the spillover effect of the treatment receipt on the outcome.  To address this complication, we generalize the instrumental variables method by allowing for interference between units and show how to identify the average complier direct effect.  We also establish the connections between our nonparametric randomization-inference approach and the two-stage least squares regression.   The proposed methodology is motivated by and applied to an ongoing randomized evaluation of the India's National Health Insurance Program (RSBY).  Joint work with Zhichao Jiang and Anup Malani