Sanduni Palliyage
Department of Statistics
Western Michigan University
Abstract:
The Cochran-Maentel-Haentzel statistic provides a stratified test for
independence by combining observed counts over
the strata.
The CMH test assumes that there is a common effect size between tables. This homogeneity
assumption is typically
tested using the Breslow-Day test. We propose an alternative test for homogeneity by
calculating each odds ratio and examining their combined variability. In particular, we
propose rejecting
homogeneity if the range of log(OR) is too large.
We also investigate testing for homogeneity by looking at pairwise comparisons between log(OR). Since these are multiple comparisons, we investigate performance of various multiple comparison adjustments for maintaining overall size without sacrificing too much power.
Bio:
Sanduni Palliyage joined the WMU statistics PhD program from Sri Lanka in 2016.
---
All statistics graduate students are expected to attend.