Two terms for confidence response time gave no clear gain in logistic accuracy predictions on later runs for 53 people in ds004086 v1.2.0

We tested whether the time to rate confidence adds information about word recognition accuracy. We used OpenNeuro ds004086 version 1.2.0. Each person's test model used no records from that person. Models used early runs from other people and scored later runs. Both models used confidence level, the old or new answer, and decision time. The added model used two terms for log confidence response time. There were 53 people, 5,989 early trials, and 6,063 later trials. Mean log loss was 0.653399 for the base model and 0.653835 for the added model. The paired gain was -0.000436 nats per trial. A 95% person bootstrap interval with model refits was [-0.002146, 0.000259]. It included zero. All eight planned checks had negative point estimates. These results do not establish equal models or a general absence of timing information. The package preserves source files, a fixed local plan, code, model fits, test scores, all bootstrap draws, and checks. A repeat from new source downloads gave equal result files.

recognition memory · confidence · response time · prediction · logistic regression · held-person evaluation · OpenNeuro ds004086