An early weighted reward contrast gave uncertain added value in second-run ridge prediction for 187 AgeRisk records

We tested if a first-run reward contrast adds value when a fixed ridge model predicts the second-run contrast. We used 187 people from OpenNeuro AgeRisk ds004711 version 1.0.0. Each score used the first seven trials per reward type and inverse burst probability weights, with an intention cap of 12. The baseline used age, gender, and the early overall pump score. Adding the early contrast reduced observed mean squared error from 6.515 to 6.255, a 4.0% reduction. The refitted person bootstrap 95% interval for relative gain was -3.3% to 13.0%. All six planned checks and a separate protocol check also had intervals that included zero. Thus added value was uncertain for this within-session target and these models. This result does not prove that stable reward response differences are absent. The package preserves source data, hashes, a local plan, code, scores, predictions, bootstrap samples, figures, and reproduction commands. Independent numerical checks and a fresh-source replay passed.

Balloon Analogue Risk Task · risk preference · prediction · ridge regression · burst censoring · AgeRisk