The early median has less GO forecast error than the earlier two-response mean in 264 ds000030 participants.

In 264 participants from ds000030 version 1.0.0, a fixed median of each participant’s correct GO reaction times (RTs) at active trial positions 1–64 had lower mean participant absolute error for later GO targets than the earlier two-response mean. The paired difference was −10.095643 ms (95% participant bootstrap interval −13.454620 to −6.791150 ms) across 4,944 test windows. The early median used 30–48 responses per participant and remained fixed during testing; no forecast parameter was fitted to test data. The early mean had the lowest displayed absolute error. Against the immediate two-response mean, the early median’s mean squared error difference had an interval that included zero. These results apply to selected future correct GO sequences in one short scanner run per person, with diagnosis groups combined. The intervals describe resampling of this sample and do not establish forecast error for new people, diagnoses, schedules, or datasets. The comparison does not show that trial history has no useful information, that the median measures time change correctly, or that a biological cause explains the result. The early baseline uses more observations than the two-response means.

reaction time · GO trials · forecast error · median baseline · participant bootstrap · ds000030 · stop-signal task