Systematic biases and Type I error accumulation in tests of the race model inequality

摘要

In simple, goo-go, and choice reaction time (RT) tasks, responses are faster to two redundant targets than to a single target. This redundancy gain has been explained in terms of a race model assuming that whichever target is processed faster determines RT (Raab, 1962). Miller (1982) presented a race model inequality to test the race model by comparing the RT distributions of single and redundant target conditions. Here, we present simulations indicating that the standard tests of this inequality (for a description of the testing algorithm, see Ulrich, Miller, & Schroeter, 2007) are afflicted with systematic biases and Type I error accumulation. Systematic biases tend to produce violations of the race model inequality, but they decrease as the numbers of observations increase. Reasonably unbiased tests of the race model inequality are obtained for sample sizes of at least 20 for each target condition. In addition, Type I error accumulates because of testing the inequality at multiple percentiles. To reduce Type I error, the race model inequality should be tested in a restricted range of percentiles, preferably in the percentile range 10% to 25%.