(A) Schematic of the task. The task requires subjects to judge the direction (left versus right) of a random-dot-motion (RDM) movie in blocks of free response (FR) and variable stimulus duration (VSD) trials. In a second task (not shown), subjects were presented with 100% coherent motion only and were instructed to respond as fast as possible while maintaining perfect accuracy. (B) Simulation of FR trials in the RDM task (N = 2145 trials, 165 trials per stimulus strength). Reaction time (top) and the proportion of positive choices (bottom) are plotted as a function of stimulus strength. Data (black points) were generated from an extrema detection model (same parameters as in Figure 4). Positive (negative) motion coherence corresponds to rightward (leftward) motion. The arrow shows the non-decision time used to generate the simulated data. Solid curves are model fits to the mean RTs. Both models were constrained to use the data-generating non-decision time. Dashed curves are predictions using the parameters obtained from the RT fits. With the non-decision time constrained, only the data-generating model succeeds in fitting and predicting the data. (C) Simulation of VSD trials (N = 3000 trials). Data (black points) were generated with an extrema detection model, using the same κ parameter as in B. Dashed curves are model fits. Each model’s κ parameter was fixed to the value estimated from fits to the FR data. (D, E) Same as in (B) and (C) with integration as the data-generating model.