Figure 9. Predictions of Standard SDT, Gaussian meta noise, and Lognormal meta noise models for the dependence of existing measures of metacognition on the confidence criterion.
We observed a stark difference in the models’ predictions for the relationship between confidence criteria and the measures meta-d’/d’ and meta-d’. While the Lognormal noise model successfully captured the empirically observed negative relationship between metacognition and confidence criterion for meta-d’/d’ and meta-d’, the Gaussian meta noise and Standard SDT models predicted qualitatively different relationships between metacognitive scores and confidence criteria for these measures. For the other two measures, Type-2 AUC and Phi, all the three models consistently predicted an inverted U-shaped function, suggesting that the strong parametric assumptions of these measures overshadowed the effect of metacognitive noise. The shaded areas represent the S.E.M. of the measures across subjects.