Mindy McCann

Department of Statistics

Oklahoma State University

Multiple Comparisons for Odds Ratios

Despite the frequent use of odds ratios in the social sciences and other areas, very few researchers even attempt to correctly report multiplicity-adjusted confidence intervals for odds ratios. The construction of multiple intervals often arises, for example, when reporting odds ratios for a categorical variable with K > 2 levels. Unless suitable adjustments for multiplicity are used, the researcher cannot draw simultaneous conclusions among the different levels because the family wise error rate in the set of K - 1 comparisons (with a reference level) is not properly controlled. Perhaps surprisingly, only naïve methods currently exist for handling this ubiquitous dilemma. We investigate simultaneous confidence intervals for odds ratios in standard settings. Model-based solutions are also discussed.

Back to Colloquium Series