🤖 AI Summary
In randomized experiments, conventional rerandomization methods suffer from poor computational efficiency under structured designs—such as stratification and clustering—hindering large-scale randomization inference. To address this, we introduce Variable Neighborhood Search (VNS), a metaheuristic optimization technique, into covariate-balanced rerandomization for the first time. Our approach formulates rerandomization as a combinatorial optimization problem, yielding an efficient framework that guarantees unbiased estimation and achieves the theoretical lower bound on variance reduction. Crucially, it enables batch generation of high-quality, covariate-balanced treatment assignments. The method generates thousands of valid allocations satisfying balance constraints in seconds—over 100× faster than state-of-the-art alternatives—while natively supporting complex experimental structures including stratification and clustering. This substantially enhances the scalability and practical applicability of rerandomization in causal inference.
📝 Abstract
Rerandomization discards undesired treatment assignments to ensure covariate balance in randomized experiments. However, rerandomization based on acceptance-rejection sampling is computationally inefficient, especially when numerous independent assignments are required to perform randomization-based statistical inference. Existing acceleration methods are suboptimal and are not applicable in structured experiments, including stratified experiments and experiments with clusters. Based on metaheuristics in combinatorial optimization, we propose a novel variable neighborhood searching rerandomization(VNSRR) method to draw balanced assignments in various experiments efficiently. We derive the unbiasedness and a lower bound for the variance reduction of the treatment effect estimator under VNSRR. Simulation studies and a real data example indicate that our method maintains the appealing statistical properties of rerandomization and can sample thousands of treatment assignments within seconds, even in cases where existing methods require an hour to complete the task.
