Causal inference for Mann-Whitney-Wilcoxon rank sum and other nonparametric statistics

Abstract

The nonparametric Mann-Whitney-Wilcoxon (MWW) rank sum test is widely used to test treatment effect by comparing the outcome distributions between two groups, especially when there are outliers in the data. However, such statistics generally yield invalid conclusions when applied to nonrandomized studies, particularly those in epidemiologic research. Although one may control for selection bias by using available approaches of covariates adjustment such as matching, regression analysis, propensity score matching, and marginal structural models, such analyses yield results that are not only subjective based on how the outliers are handled but also often difficult to interpret. A popular alternative is a conditional permutation test based on randomization inference [Rosenbaum PR. Covariance adjustment in randomized experiments and observational studies. Statistical Science 2002; 17(3):286-327]. Because it requires strong and implausible assumptions that may not be met in most applications, this approach has limited applications in practice. In this paper, we address this gap in the literature by extending MWW and other nonparametric statistics to provide causal inference for nonrandomized study data by integrating the potential outcome paradigm with the functional response models (FRM). FRM is uniquely positioned to model dynamic relationships between subjects, rather than attributes of a single subject as in most regression models, such as the MWW test within our context. The proposed approach is illustrated with data from both real and simulated studies.

Circulating mutational portrait of cancer: manifestation of aggressive clonal events in both early and late stages

“…Patients were assigned to low or high mutation load groups based on the cohort median mutation number. We applied a Mann-Whitney-Wilcoxon rank sum [34] and a Kruskal-Wallis test [35] to compare mutation load with clinical variables with two groups and multiple groups, respectively. A Fisher’s exact test was used to determine association between smoking status (defined as an ordinal variable—current/recent, former, never smokers) and DNA damage genes.…”

Population metrics for suicide events: A causal inference approach

“…3–5 History of suicide event, which is highly correlated to the outcome, is not controlled for when performing comparisons. If we control for this confounder and compare the two facilities within each category of the history variable, then results make perfect sense.…”

“…To be able to estimate Δ based on observed outcomes and information in the risk groups H k , we assume, as in the literature, a strong ignorable assumption 3–5,7 $$false(y_{i1},y_{i2}false)\perp false(F_1,F_{2}false)false|H_k,1\le k\le K$$where ⊥ denotes stochastic independence. The condition in (3.1) ensures that within each subgroup H k , assignment of F l is independent of prior suicide event.…”

Rank‐preserving regression: a more robust rank regression model against outliers

“…Although the FRM is a generalization of the GLM and its longitudinal data extensions GEE (WGEE), the similarity between the two classes of models ends when . The FRM has been applied to address a wide range of issues, including causal inference for multi‐layered intervention studies , causal inference for the MWW rank sum test , extensions of the MWW rank sum test to the regression and longitudinal data settings , models for population mixtures , and reliability coefficients . Next, we apply the FRM to develop an alternative of rank regression that not only addresses the MAR, but its estimates are functionally independent of observed values of y i .…”