Modelling Metacognition: A Joint Prediction-Confidence Model for Predictive Inference Task Data

Metacognition, the ability to reflect on and evaluate one's own cognitive processes, can be affected in various mental health conditions. However, the specific computational mechanisms driving these changes are not fully understood. In this research, the authors build upon the Hierarchical Gaussian Filter (HGF) models to simultaneously analyze both predictions and confidence ratings in a predictive inference task. This model allows for an individualized assessment of metacognitive processing.

By applying this cognitive computational model to a large dataset of subclinical participants (N=430), the researchers achieved an impressive fit of prediction responses (r = 0.91) and a moderate to good fit of confidence ratings (r = 0.4). Moreover, the model successfully captured the dynamics of confidence self-reports around specific change-points in the data.

The posterior parameter estimates from the model showed that prediction errors have a negative impact on confidence ratings, while prediction precision positively influences these ratings.

Interestingly, the study also replicated established findings related to compulsivity, a transdiagnostic factor. The results indicate that individuals with compulsive tendencies tend to exhibit inflated confidence and a disconnect between action updates (prediction errors) and confidence. These findings underscore the robustness of the proposed methodology and highlight the potential of jointly modeling predictions and confidence to reveal underlying metacognitive alterations in individuals with psychopathology.