SoftwareX · 2026

FastRerandomize

Goldstein, Jerzak, Kamat & Zhu

Fast rerandomization using accelerated computing — enabling stronger experimental designs and exact inference at practical speeds.

Read (DOI) ScienceDirect arXiv GitHub CRAN

Rerandomization GPU acceleration Randomization tests Experimental design

Overview

Rerandomization improves covariate balance by repeatedly drawing random assignments and accepting only those that pass a balance criterion. The challenge is speed: naïve rerandomization can be too slow for large experiments. FastRerandomize brings accelerated computing to the bottlenecks.

Faster acceptance

Scale rerandomization to larger N without losing rigor.

Exact inference

Randomization tests that match the assignment mechanism.

Practical tooling

A concrete implementation for applied researchers.

Figures

Performance snapshots

Light/dark variants are shown automatically based on the site theme.

n = 100

CPU vs GPU timing

n = 1000

CPU vs GPU timing

Back to projects Get BibTeX