The 95% bootstrap interval included zero for gain from match positions over counts on later 3-back non-targets from 60 ds007022 women under fixed ridge models

Letter match positions gave a log-loss gain of 0.001670 over match counts in later 3-back visits. The 95% refit bootstrap interval was -0.004266 to 0.007228 and included zero. We fitted fixed ridge models on visit 0 and tested visits 4 and 5 in all 60 women from OpenNeuro ds007022 version 1.1.0. The test used 4,338 non-target trials. The two main models each added three history terms. Both history models had lower loss than a base model. The result does not prove model equivalence or identify a neural mechanism. The package preserves source files, provenance, predictions, all bootstrap coefficients, verification results, and repeat instructions.

n-back task · working memory · letter-history interference · false alarms · log loss · ridge regression · bootstrap · longitudinal data