Five previous decision times reduced later log time forecast error in 21 ds004119 people under fixed models and early personal calibration

We tested whether five previous decision times improve a forecast of later recorded time. We used OpenNeuro ds004119 version 1.0.0 and fixed ridge models with personal calibration. The models trained on blocks 1 through 5 for 21 people. The test used 2,958 eligible decisions from their later five blocks. Adding history reduced mean squared natural log error from 0.20918 to 0.17807. The reduction was 14.87%. The 95% person bootstrap interval was 7.85% to 21.90%, with both models refitted. All five planned secondary checks gave lower point error with history. The result supplies a behavioral baseline for EEG forecast tests. The scope is later eligible decisions from the same calibrated people. The package preserves source events, file hashes, the local plan, code, exclusions, models, forecasts, bootstrap draws, and reproduction commands. Independent calculations passed. A fresh-source replay gave identical results.

decision time · behavioral forecasting · OpenNeuro · personal calibration · ridge regression · person bootstrap · EEG forecast baseline