A history of five trials gave no clear gain for logistic forecasts on a later visit in 40 ds001810 participants using stored visit order

In OpenNeuro ds001810, a history of five previous trials gave no clear gain for fixed logistic forecasts on a later visit. We fit models on visit 1 and tested 5,820 trials on visit 2 for the same 40 people. We used stored visit order. The baseline used target delay, cumulative accuracy, trial number, previous delay, and time since the previous response. Baseline minus history log loss was -0.000918 per trial. The 95% paired person bootstrap interval was [-0.004277, 0.000375]. Both models were fit again in each of 2,000 samples. All six fixed checks had negative mean contrasts. This result applies to the tested models and selected people. It does not show that all recent outcome information lacks value. The package includes 188 verified source files, code, a local plan, parameters, forecasts, bootstrap samples, and reproduction instructions.

attentional blink · behavioral forecasting · logistic regression · trial history · logarithmic loss · paired bootstrap · OpenNeuro ds001810