Reaction times (RTs) are among the most widely used measures in cognitive psychology, yet there is still no consensus on the best statistical approach for analyzing them. Researchers increasingly rely on generalized linear mixed-effects models (GLMMs), which are often assumed to provide advantages over linear mixed-effects models (LMMs) or analyses based on aggregated data. But how large are these advantages with human-generated RTs?

This seminar presents the results of a preregistered simulation study comparing three GLMMs, a Gamma GLMM, an Inverse Gaussian GLMM, and a novel Probit GLMM, with LMMs and t-tests on aggregated data. Using virtual experiments derived from five large datasets, we evaluated the methods using coverage-adjusted power, a novel metric that adjusts power so that a desired nominal coverage is achieved.

The results show that no single method dominates in all situations. However, Gamma and Probit GLMMs, on average, provide substantial gains in power, even after adjusting for coverage, over traditional methods such as t-tests on aggregated data