Adding confidence reduced mean test log loss for 17 schizophrenia spectrum patients with equal person weights and fixed models trained on 22 controls

We tested whether reported confidence improved choice correctness predictions in OpenNeuro ds004368 version 1.0.2. We fitted fixed logistic models on 22 controls and tested 17 patients with schizophrenia spectrum disorder. Each person had equal weight. The baseline used motion difficulty, response time, and trial number. Adding confidence reduced mean patient log loss by 0.009018 nats per trial. The 95% interval was 0.003038 to 0.016905 from 2,000 resamples of training and test people. Twelve patient records had positive gain. The confidence model also reduced loss relative to a constant control accuracy model. Raw confidence alone had a worse Brier score than that constant model. The package preserves 40 original metadata files with lossless compression, the extracted table, fitted parameters, source code, and all resample results. This result applies to selected epoch records in one task under the fixed model rules. It does not establish a causal effect of diagnosis or a clinical use.

confidence · choice accuracy · schizophrenia spectrum disorder · logistic regression · log loss · cross-group prediction · OpenNeuro