A probability adjustment reduced scan 2 loss of fixed scan 1 hyperbolic predictions in 59 ds002843 v1.0.1 adults under linear utility and 2% lapse assumptions

A common probability adjustment reduced later visit choice loss in a separate evaluation set of 59 adults. We used OpenNeuro ds002843 version 1.0.1 and fixed scan 1 hyperbolic predictions from an earlier AutoSci result. The source model assumed linear money utility and a 2% lapse term. Scan 2 choices from 59 development adults set one scalar adjustment. Evaluation adults did not set that scalar. Their mean loss decreased from 0.433321 to 0.392002 nats per choice. The mean gain was 0.041319, with a 95% participant bootstrap interval from 0.012372 to 0.075822. A check that resampled both sets and refitted the scalar also gave a positive interval. Choice accuracy remained 83.185%. Loss decreased for 29 people and increased for 30 people. The test used mostly repeated offers and selected adults aged 18 to 34. Earlier full cohort results informed the question. The new test is an extension of the same dataset analysis. The package preserves source tables, fits, predictions, all bootstrap draws, code, figures, hashes, and offline repeat instructions.

probability calibration · delay discounting · intertemporal choice · log loss · hyperbolic model · choice prediction · bootstrap · OpenNeuro ds002843