Fig. 3: Variation of the number of factors selected for retention.

AMT Amazon’s Mechanical Turk. A-D Anxious-depression, OCD Obsessive compulsive disorder, Social anx Social anxiety, Trait anx Trait anxiety, r Pearson correlation coefficient, p p-value. A Fifty-five factors were generated from solutions retaining 1-factor, 2-factors, …, 10-factors. Heatmap indicates the loading of individual items onto each resulting factor (top panel). Weighted effect sizes (f²) from 10 linear regression analyses predicting model-based planning, averaged across 5 datasets (N = 4990) (middle panel) and metacognition in 4 datasets (N = 2575) (bottom panel). Factor loadings and effect sizes are aligned along the same x-axis. B Top-performing factors’ effect sizes within each dataset and the weighted averaged effects across datasets. The top-performing factors had a statistically significant effects on model-based planning overconfidence and underconfidence within each dataset (all p < 0.05). *p < 0.05, **p < 0.01, ***p < 0.001. The top performing factor for model-based planning and overconfidence was extracted from the 3-factor model (in purple) and the top performing factor for underconfidence came from the 2-factor model (in grey). C Correlation between scores on the first factor in a 3-factor solution (model 3.1 i.e., ‘Anxious-depression’) and the first factor from a 2-factor solution (model 2.1) in the discovery dataset. D Correlation between factor loadings of ‘Anxious-depression’ and model 2.1 in the discovery dataset.