Flanker conflict gave a small and uncertain gain in linear ADHD symptom prediction for 174 ds005754 participants

We tested flanker task data from OpenNeuro ds005754 version 1.1.1. We used 174 participants with complete task, parent symptom, and demographic data. We fit each model without the tested participant. The fixed models used age, age squared, source sex, and congruent response speed. Adding response-time conflict reduced mean squared error by 1.90 CBRS points squared (1.1%). The central 95% range from 2,000 refitted participant bootstrap samples was -5.42 to 12.91 points squared. Thus, the gain had an uncertain sign. In 188 task-qualified participants, conflict gave a random trial-half coefficient of 0.823 after length correction. Fixed filter and median checks also gave uncertain gains. A later check of common-length files gave the same interpretation. The package preserves 197 source files, hashes, code, scores, predictions, bootstrap results, figures, and instructions. Independent arithmetic and model checks passed. The evidence applies to the tested score, linear models, and selected public cohort.

flanker task · response-time conflict · ADHD symptoms · linear prediction · held-out validation · trial consistency · youth · OpenNeuro ds005754