Skip to main content
. 2020 Apr 27;9:e55365. doi: 10.7554/eLife.55365

Figure 6. Failure of the extrema detection model and mixed success of the integration model in human subjects performing the RDM task.

(A) Free response (FR) trials. Reaction times (top) and proportion of rightward choices (bottom) as a function of motion strength for six human subjects (error bars are s.e.; ~1800 trials per subject). Positive (negative) motion coherence corresponds to rightward (leftward) motion. The arrows (top) indicate each subject’s estimated non-decision time from the speeded decision trials. Solid curves are model fits to the mean RTs (integration, magenta; extrema detection, cyan). Dashed curves are model predictions using the parameters obtained from the RT fits. Solid black lines (bottom) are fits of a logistic function to the choice-data alone. For visualization purposes, data from the interleaved 100% coherence trials are not shown (see Figure 6—figure supplement 1). (B) Variable stimulus duration (VSD) trials (~3000 trials per subject). Sensitivity as a function of stimulus duration for the same subjects as in (A). Sensitivity is estimated using logistic regression (sliding logarithmic time window, see Materials and methods). Dashed curves are model fits to the VSD data, using the κ parameter derived from model fits to the FR data.

Figure 6—source data 1. Data from the RDM task for each subject, separated by trial-type.

Figure 6.

Figure 6—figure supplement 1. The reaction times from the ‘speeded decision’ experiment provide a better estimate of the non-decision time than those from the interleaved, 100% coherence trials.

Figure 6—figure supplement 1.

The abscissa shows the estimated non-decision time from the speeded decisions, in which subjects were presented with 100% coherence trials only and were instructed to respond as fast as possible while maintaining perfect accuracy. The ordinate shows the mean RT on the 100% coherence trials that were interleaved in the main experiment. All subjects had perfect accuracy for both trial-types. Error bars are S.E.M.
Figure 6—figure supplement 2. Without constraints on the non-decision time and SNR, the models cannot be easily distinguished.

Figure 6—figure supplement 2.

Data are the same as those in Figure 6. (A) FR trials. Sold curves are unconstrained fits of the integration (magenta) and extrema detection (cyan) models to mean RTs (top). The non-decision time was a free parameter and trials of 100% coherence were excluded. Dashed curves (bottom) are predictions generated from the parameters obtained from the mean RT fits. The quality of the fits and predictions are similar across both models. (B) VSD trials. Solid curves are model fits to subjects’ VSD data, allowing the K parameter to take on any value. Both models capture the main features of the data.