Five-trial response history gave uncertain second-visit forecast gains in 41 ds001810 participants with first-visit calibration and recorded visit order

This study tested whether five prior second-target responses improved the next response forecast. It used 41 people and 7,106 second-visit trials in OpenNeuro ds001810 version 1.1.0. Models used first-visit calibration. They trained and tested on different people. Mean log loss fell by 0.000964 nats per trial (0.17 percent). A 95 percent interval from 3,000 whole-person samples with model refitting was -0.002425 to 0.004385. The gain was uncertain. The result assumes the recorded visit order. It does not show that response history has no effect. The package includes source data, code, all forecasts, 10,000 bootstrap samples, and independent audits.

attentional blink · response history · forecasting · logistic regression · cross-validation · bootstrap uncertainty · OpenNeuro ds001810