A0591
Title: Blocking, rerandomization, and regression adjustment in randomized experiments with high-dimensional covariates
Authors: Ke Zhu - Tsinghua University (China) [presenting]
Hanzhong Liu - Tsinghua University (China)
Yuehan Yang - Central University of Finance and Economics (China)
Abstract: Blocking, a special case of rerandomization, is routinely implemented in the design stage of randomized experiments to balance baseline covariates. Regression adjustment is highly encouraged in the analysis stage to adjust for the remaining covariate imbalances. Researchers have recommended combining these techniques; however, the research on this combination in a randomization-based inference framework with a large number of covariates is limited. Methods are proposed that combine blocking, rerandomization, and regression adjustment techniques in randomized experiments with high-dimensional covariates. In the design stage, we suggest the implementation of blocking or rerandomization or both techniques to balance a fixed number of covariates most relevant to the outcomes. For the analysis stage, we propose a regression adjustment method based on the Lasso to adjust for the remaining imbalances in the additional high-dimensional covariates. Moreover, we establish the asymptotic properties of the proposed estimator and outline conditions under which this estimator is more efficient than the unadjusted estimator. In addition, we provide a conservative variance estimator to facilitate valid inferences. Our analysis is randomization-based, allowing the outcome data generating models to be misspecified. Simulation studies and two real data analyses demonstrate the advantages of the proposed method.
