Resampling-based multiple comparisons for generalized linear models

Josephine Sarpong Akosa, University of Texas at El Paso


Diverse applications in medical and epidemiological research routinely utilize generalized linear modeling to explain the relationship between the incidence of disease and particular risk factors. Researchers' interest in such models are estimated quantities from the model such as the response probabilities, the relative risks or the odds ratios and not the model itself. Often, the simultaneous estimation of these quantities or a subset of the quantities are warranted. The results are usually reported via confidence intervals at a pre-specified level of significance. Utilizing the usual 95% pointwise confidence intervals for the simultaneous inference inflates the risk of making type I errors. Several procedures have been proposed to control for multiplicity among a set of inferences, but no one solution is applicable for all situations. This study develops and via simulation evaluate resampling-based multiple comparison procedures for contrasts of generalized linear model parameters. The main results of the study is a new algorithm for resampling-based multiple comparisons for quantities from generalized linear models that will in some situations outperform existing conservative procedures and in those situations be less data-wasteful.

Subject Area


Recommended Citation

Akosa, Josephine Sarpong, "Resampling-based multiple comparisons for generalized linear models" (2014). ETD Collection for University of Texas, El Paso. AAI1564656.