Are Reaction Time Transformations Really Beneficial?
We investigate whether or not the common practice of transforming response times prior to conventional analyses of central tendency yields any notable benefits. We generate data from a realistic single-bound drift diffusion model with parameters informed by several different typical experiments in cognition. We then examine the effects of log and reciprocal transformation on expected effect size, statistical power, and Type I error rates for conventional two-sample t-tests. One of the key elements of our setup is that RTs have a lower bound, called the shift, which is well above 0. We closely examine the effect that different shifts have for the analyses. We conclude that logarithm and reciprocal transformation offer no gain in power or Type I error control. In some typical cases, reciprocal transformations are detrimental as they lead to a lowering of power.