Early cue calibration with weight 0.5 reduced later mean squared response time error in 141 selected ds004574 v1.0.0 records with equal cue weights

We tested a fixed early cue calibration rule in OpenNeuro ds004574 version 1.0.0. We selected 141 person records, with 93 Parkinson's disease records and 48 control records. Both rules used the same personal mean time from blocks 1 and 2. The reference also used a cue contrast fit from other people, with group, age, and gender predictors. The added rule used half of the early personal visual minus auditory contrast deviation. We scored later correct responses in blocks 3 and 4 with equal cue and person weights. The mean squared error fell from 8102.08 to 7966.98 ms squared. The mean gain was 135.11 ms squared, with a 95% refitted person bootstrap interval of 37.85 to 283.74. The relative gain was 1.67%, with an interval of 0.48% to 3.44%. Only 73 of 141 records had positive gains. The median gain was 0.94 ms squared. Both stricter trial selection checks had intervals that included zero. Selection used later outcomes, and prior findings from this same cohort guided the study. This is an exploratory test in selected records. The package preserves source bytes and hashes, the saved plan, code, trial losses, all bootstrap draws, figures, and reproduction instructions. A separate checker verified every primary bootstrap fit. A full repeat from a new download matched all scientific output.

response time · cue calibration · forecasting · mean squared error · bootstrap · Parkinson disease · OpenNeuro