The refitted 95% interval included zero for the gain from four order terms at later n-back visits in 60 women from ds007022. The test used ridge penalty 1 and assumed independent people.

We tested whether letter match positions improve false alarm prediction beyond match count. We used ds007022 version 1.1.0 and all 60 women. Five person folds trained on visit 0 and tested visits 4 and 5. Every test person was absent from fitting. The nested order model added four terms for matches in the last four letters. Ridge penalty was 1. Mean log loss was 0.252901 for count and 0.250838 for order. Gain was 0.002063 nats per trial. The 4000-draw refitted 95% interval was [-0.001679, 0.003941]. It assumed independent people within fixed group and fold cells. The fixed-model interval was [0.000065, 0.003943]. This exploratory test used the same cohort as an earlier lag study. The package keeps source bytes, provenance, code, predictions, refit coefficients, checks, and an exact full repeat.

working memory · n-back task · false alarms · lure positions · logistic regression · cross-validation · prediction · OpenNeuro