Extended Data Fig. 2: Power when the RT difference is 0.8s as the condition difference increases.

Only models that had controlled error rates in Fig. 3 (main text) are shown and since the RT range of this simulation is not within the range where ConsDurRTMod has controlled error rates, it is excluded. The two-alternative, forced-choice categorization task RT distribution was used in the top panels, while Stroop RT distribution was used in the bottom panels. An ISI between 2-4s was used and the inference of interest was for the 1-sample t-test of the condition difference effect with 100 subjects. Model 2, RTDur, is the true model for the left column (orange line) and model 1, ConsDurNoRT, is the true model for the right column (blue line), and the corresponding power curves indicate maximal power. When the signal duration scales with RT (left panel), ConsDurRTDur (red) has some power loss due to model misfit, while the 4 regressor interaction model (gray line) loses considerable power due to collinearity. When the duration does not scale with RT (right panels), the ConsDurRTDur model (red line) has similar power to the true model (blue), illustrating some power loss due to including an RT regressor in the time-series analysis. This power loss is not seen for shorter RT differences (.1s result in main paper Fig. 4).