Title
A bootstrap pairwise comparison for the means of skewed data
Speaker
Wanzu Tu, Division of Biostatistics, Indiana University School of Medicine
Abstract
In this presentation, I consider the problem of all pairwise comparisons of the means of skewed data with a special emphasis on the positively skewed log-normal distributions. Two bootstrap procedures are proposed to construct the simultaneous confidence sets of the pairwise differences. The first procedure provides a direct bootstrap approximation to the sampling distribution of the maximum pivotal statistic that generates the simultaneous confidence intervals. The second procedure is based on the idea of prepivoting and can be shown to produce balanced asymptotic coverage probabilities for all the pairs involved as well as the tail areas of each pair. The consistency in the coverage probability of the two proposed procedures will also be discussed. To illustrate, I consider a clinical application on the evaluation of medication compliance among patients with chronic heart failures.
