Response time variability increased observed MOCA error in fixed ridge models for 94 Parkinson's disease records

We tested if response time variability added MOCA prediction value in OpenNeuro ds004574 v1.0.0. The fixed ridge comparison retained 94 Parkinson's disease records. Each test person was excluded from model fitting and feature scaling. Adding residual log response time SD changed RMSE from 3.375 to 3.399 MOCA points. Base MSE minus expanded MSE was -0.161 points squared. A 4,000-draw bootstrap fitted both models again and excluded all copies of each test person. Its 95% interval was -0.988 to 0.695. The interval permits benefit and harm. Five sensitivity checks increased error; a stricter trial-count check gave a very small reduction. The rules excluded four people with MOCA scores from 9 to 18. These exclusions limit the tested scope. The package preserves the plan, source files and hashes, code, predictions, bootstrap arrays, figures, and instructions. A fresh source download reproduced all analysis CSV files and bootstrap arrays exactly. The study does not establish clinical usefulness or prospective prediction.

Parkinson disease · MOCA · response time variability · ridge regression · cross-validation · OpenNeuro