Adding relative response time spread increased observed MoCA error in 96 ds004574 v1.0.0 Parkinson records with fixed ridge models

We tested whether relative response time spread adds MoCA score prediction value in OpenNeuro ds004574 v1.0.0. The fixed ridge comparison used 96 selected Parkinson's disease records. The base predictors were age, recorded gender, source UPDRS, and log mean standard response time. The added feature was ln(IQR / median response time), where IQR is the interquartile range. Each test person was excluded from training and scaling. Adding spread changed MSE from 11.157 to 11.289 squared MoCA points. Base MSE minus added MSE was -0.132. A 10,000-draw person bootstrap fitted both models again and excluded all copies of each test identity. Its 95% interval was -0.665 to 0.423. All six planned checks had negative point differences and intervals that included zero. The interval assumes independent people and fixed observed trial records. The result extends a related 94-record study with a different spread measure and sample rule. It uses the same source cohort. The two excluded source records had MoCA scores 9 and 11. The result does not establish absence of prediction value or future cognitive prediction. The package preserves source files, hashes, code, trial decisions, predictions, bootstrap samples, figures, and instructions. Independent fits matched all person predictions. A fresh source download reproduced every result file exactly.

Parkinson disease · MoCA · response time · interquartile range · ridge regression · cross-validation · bootstrap · OpenNeuro · cognitive assessment