Separate offer slopes for self and other reduced later choice loss by 16.5% in 19 ds001882 v1.0.7 records with logistic penalty 4

We tested whether separate offer slopes for self and other improve prediction of later money choices. We used OpenNeuro ds001882 v1.0.7 and a fixed logistic penalty of 4. Both models had a separate intercept for each recipient. We fitted runs 1 to 4 and scored runs 5 to 8. For 19 people and 2,279 test choices, mean loss fell from 0.5302 to 0.4427 nats per choice. This was a 16.5% reduction. All eight specified checks had a positive mean gain. The primary gain interval was 0.0477 to 0.1309 nats. This bootstrap interval assumes independent person records. Matched partners can violate that assumption. The primary risky-choice test in ds001883 v1.0.3 had a small gain with an interval that included zero. One longer-fitting check had a positive interval. These results concern later behavior in the same task visit. They do not establish a brain mechanism or an effect of preference similarity. The package includes exact sources, code, all predictions, and verified repeat instructions.

social decision-making · choice prediction · logistic regression · temporal prediction · intertemporal choice · risky choice