Training means from other participants give less mean squared error for scanner arrow-history contrasts in 117 ds000030 participants.

In 117 paired control participants from the UCLA Consortium for Neuropsychiatric Phenomics stop-signal task, the mean training arrow-history contrast of the other participants predicted the recorded scanner contrast with lower mean squared error than each participant’s own training contrast. The participant-minus-group error difference was 2,489.9086889287964 ms² (36.64827556276706% lower group-prediction error); its participant-bootstrap 95% interval was 1,240.3952115421305 to 3,782.7175253192545 ms², and its schedule-group interval was 1,312.6801224115572 to 3,811.445670126499 ms². The pattern was consistent across three sensitivity checks. Mean reaction time showed the opposite result. These are exploratory estimates for selected observed runs: training preceded scanning, paired inclusion depended on both phases, intervals were not adjusted for multiple comparisons, and the analysis does not estimate stable traits, future-participant error, or a biological cause. The analysis used task event tables, not brain images. The source was OpenNeuro ds000030 snapshot 00016 at a recorded Git commit; its exact correspondence to cached version-specific metadata is uncertain.

stop-signal task · reaction time · individual differences · training prediction · scanner scores · bootstrap · cognitive neuroscience