Wald-Type Rank Tests with Improved Small-Sample Properties

• Activity Number: 659
• Type: Contributed
• Date/Time: Thursday, August 2, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Chunpeng Fan*+ and Donghui Zhang
• Companies: Sanofi U.S. Inc. and Sanofi U.S. Inc.
• Keywords: Factorial Designs ; ANOVA-Type Rank Test ; GEE ; Empirical Covariance Estimator ; Robust Covariance Estimator ; Nonparametric

Abstract

Factorial designs have been widely used in many scientific fields. Traditionally, such designs can be analyzed by the generalized linear mixed models (GLMMs). When making inference for the fixed effects in GLMM, however, even the robust generalized estimating equations (GEE) method may give biased results when the distributional assumption is violated. In this case, rank-based tests can be an option for inferential procedures. Although previous literatures investigated rank tests in various selected designs, no unified rank test has been derived. For small sample sizes, the currently most promising ANOVA-type rank test does not work well when testing one-dimensional contrasts. This work applies the GEE technique to rank transformed data and derives a unified Wald-type rank test which can be used in any factorial design. The asymptotic properties of the proposed test are derived under the