91
reads

Song Xi Chen

2010-12-16
10:30:00 - 12:00:00

R440 , Astronomy and Mathematics Building

The conventional Wilcoxon/Mann-Whitney tests may be invalid due to the presence of certain pre-treatment variables which contribute to the propensity of the missingness or the allocation of treatments in observational studies. We propose an adjusted Mann-Whitney test for comparison of treatment effects between two samples that is applicable for both missing outcomes and observational studies. The adjusted Mann-Whitney test is consistent and can utilize the baseline covariate information. We also propose a semiparametrically adjusted Mann-Whitney test which leads to dimension reduction for high dimensional covariate when a parametric propensity model for the missingness or the treatment allocation is available. A novel bootstrap procedure is devised to approximate the null distribution of the test statistics that respects the conditional distributions of the treatment effects given pre-treatment variables. A simulation study and a real economic observational data application are presented.