Incorporate Clustering Effects for the Wilcoxon Rank Sum Test: The Cluswilcox Command

• Activity Number: 279 - Nonparametric Tests and Estimations for Clustered Data: It Is Essential for Non-Normal Data
• Type: Topic Contributed
• Date/Time: Tuesday, August 1, 2017 : 8:30 AM to 10:20 AM
• Sponsor: International Chinese Statistical Association
• Author(s): Xin He* and Jing Zhang and Mei-Ling Ting Lee and Bernard Rosner
• Companies: University of Maryland, College Park and University of Maryland, College Park and University of Maryland and Harvard Medical School
• Keywords: clustering effects ; cluswilcox ; Wilcoxon rank sum test ; stratum

Abstract

Wilcoxon rank sum test, which assumes that individual sampling units are independent, is frequently used to compare measures of location when the underlying distributions are far from normal or not known in advance. However, in the presence of clustering effects, p-values will be underestimated and confidence interval width will be too narrow if we still assume independence. Rosner et al. (2003) proposed a correction to the variance of the Wilcoxon rank sum statistic in order to account for clustering effects. In this talk, we introduce the cluswilcox Stata command, which incorporates the clustering effects for balanced (same number of subunits per cluster) and unbalanced (different number of subunits per cluster) data, stratied (for important confounding variables) and non-stratied data, in the presence or absence of tied rankings.