Testing Association in Stratified 2×2 Contingency Tables Using Multiple Comparison Methods

• Activity Number: 76 - Contributed Poster Presentations: Section on Statistics in Epidemiology
• Type: Contributed
• Date/Time: Monday, August 3, 2020 : 10:00 AM to 2:00 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Sanduni Pathum Laksara Palliyage* and Magdalena Niewiadomska-Bugaj
• Companies: Western Michigan University and Western Michigan University
• Keywords: Multiple comparisons in odds ratios; Stratified contingency tables; Dunn-Šidák procedures; Bonferroni procedure; Holm-Bonferroni procedures; Holland-Copenhaver procedure

Abstract

Multiple comparisons methods are used to identify strata that can be clustered together and strata that are significantly different from others. The Breslow-Day test is widely used for testing association across strata. However, the Breslow-Day test can only identify whether there is at least one odds ratio that is different from other odds ratios. Using Breslow-Day test, we cannot find which stratum has that different odds ratio. In this paper, we are proposing four multiple comparison procedures for comparing odds ratios. The Bonferroni procedure with the Bonferroni correction, the Dunn-Šidák procedure with the Šidák correction, the Holm-Bonferroni procedure which is a modification of the Bonferroni procedure, and the Holland-Copenhaver procedure are adopted for comparing odds ratios. These four procedures are conservative, but they protect the family wise error rate. Among the pairwise comparison methods proposed, the Holland-Copenhaver procedure and the Holm-Bonferroni procedure can be recommended for any sample size. However, Bonferroni and Dunn-Šidák procedures have slightly higher power compared to Holland-Copenhaver and Holm-Bonferroni procedures for very large sample sizes