Residual randomization inference (RRI) is a method for testing and inference in regression models based on invariance assumptions on the errors; e.g., inference assuming only exchangeability of errors, or sign symmetry, or both. Compared to bootstrap, residual randomization is more flexible, it does not rely on asymptotic normality and addresses the inference problem in a unified way.

Papers

Wang, S., Lee S.K., Kolar, M., Toulis, P. (2021) Robust inference for high-dimensional linear models via residual randomization. International Conference on Machine Learning (ICML’21), oral. (forthcoming)

Toulis, P. (2019). Life after bootstrap: Residual randomization inference in regression models with complex error structure. arxiv code slides

Toulis, P. (2019). Randomization Inference in Regression Models — R Package RRI. (Technical report). pdf

Code

RRI R package:

—on CRAN: https://cran.r-project.org/package=RRI

—on GitHub: https://github.com/ptoulis/residual-randomization