Rerandomization Methods for Stratified Randomized Experiments

A methodological paper by Xinhe Wang of Tsinghua University proposing two rerandomization techniques for stratified randomized experiments. The methods are based on overall and stratum-specific Mahalanobis distances and aim to improve covariate balance. Their statistical properties, including asymptotic distributions and variance reduction formulas, are derived under a randomization inference framework without modeling assumptions.

Use Cases

• Designing more efficient stratified randomized experiments based on the proposed rerandomization methods.
• Estimating variance reduction for treatment effects compared to standard stratified randomization.
• Constructing conservative confidence intervals for average treatment effects without outcome modeling assumptions.
• Comparing the performance of overall versus stratum-specific Mahalanobis distance criteria in simulation studies.

Strengths

• The paper provides explicit asymptotic distributions and variance reduction formulas for the proposed methods.
• Analysis is conducted under a randomization inference framework requiring no modeling assumptions for potential outcomes.
• Method performance is validated through an extensive simulation study and a real-data example.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and sample data are unknown, which may limit suitability assessment.

Provenance

Source

Xinhe Wang, Tsinghua University

Collection Method

Theoretical derivation and simulation.