2020arXiv (Cornell University)Open access

Which Wilcoxon should we use? An interactive rank test and other alternatives

Boyan Duan, Aaditya Ramdas, Larry Wasserman

Open full text 1 citations

Abstract

Classical nonparametric tests to compare multiple samples, such as the Wilcoxon test, are often based on the ranks of observations. We design an interactive rank test called i-Wilcoxon---an analyst is allowed to adaptively guide the algorithm using observed outcomes, covariates, working models and prior knowledge---that guarantees type-I error control using martingales. Numerical experiments demonstrate the advantage of (an automated version of) our algorithm under heterogeneous treatment effects. The i-Wilcoxon test is first proposed for two-sample comparison with unpaired data, and then extended to paired data, multi-sample comparison, and sequential settings, thus also extending the Kruskal-Wallis and Friedman tests. As alternatives, we numerically investigate (non-interactive) covariance-adjusted variants of the Wilcoxon test, and provide practical recommendations based on the anticipated population properties of the treatment effects.

About this research paper

What this paper is about

Classical nonparametric tests to compare multiple samples, such as the Wilcoxon test, are often based on the ranks of observations. We design an interactive rank test called i-Wilcoxon---an analyst is allowed to adaptively guide the algorithm using observed outcomes, covariates, working models and prior knowledge---that guarantees type-I error control using martingales. Numerical experiments demonstrate the advantage of (an automated version of) our algorithm under heterogeneous treatment effects. The i-Wilcoxon test is first proposed for two-sample comparison with unpaired data, and then extended to paired data, multi-sample comparison, and sequential settings, thus also extending the Kruskal-Wallis and Friedman tests. As alternatives, we numerically investigate (non-interactive) covariance-adjusted variants of the Wilcoxon test, and provide practical recommendations based on the anticipated population properties of the treatment effects.

Why it matters

OpenAlex reports 1 citations for this work. Citation counts describe recorded attention and do not establish research quality.

Key contribution

A contribution statement is not available in the OpenAlex record.

Method / approach

Method details are not available in the OpenAlex metadata.

Main findings

Findings are not separately available in the OpenAlex metadata.

Limitations

Limitations are not available in the OpenAlex metadata.

Applications

Application details are not available in the OpenAlex metadata.

Available abstract

Classical nonparametric tests to compare multiple samples, such as the Wilcoxon test, are often based on the ranks of observations. We design an interactive rank test called i-Wilcoxon---an analyst is allowed to adaptively guide the algorithm using observed outcomes, covariates, working models and prior knowledge---that guarantees type-I error control using martingales. Numerical experiments demonstrate the advantage of (an automated version of) our algorithm under heterogeneous treatment effects. The i-Wilcoxon test is first proposed for two-sample comparison with unpaired data, and then extended to paired data, multi-sample comparison, and sequential settings, thus also extending the Kruskal-Wallis and Friedman tests. As alternatives, we numerically investigate (non-interactive) covariance-adjusted variants of the Wilcoxon test, and provide practical recommendations based on the anticipated population properties of the treatment effects.

Key concepts: Wilcoxon signed-rank test, Friedman test, Nonparametric statistics, Covariate, Rank (graph theory), Test (biology), Population, Statistics

Related papers

Back to paper searchBrowse research topicsOriginal source
Which Wilcoxon should we use? An interactive rank test and other alternatives — Research Paper | ScholarLens