Cognitive reflection is a better predictor of metacognition than actual task performance

Theoretical models of metacognitive assessment postulate that the correctness of this type of assessment depends on the difference between actual performance and some type of “overconfidence” bias. In contrast, Stanovich’s Tripartite model postulates that an adequate performance on tasks that requires reasoning depends first on cognitive inhibition and then on cognitive reflection. The objective of this study was to test if actual performance or measures of cognitive reflection are better predictors of metacognitive assessment. Our sample consisted of 120 undergraduate students, with ages ranging from 18 to 58 (M = 27.11, SD = 9.79). To measure performance, we fitted a Signal Detection Theory (SDT) model to the assessment of validity of 32 syllogisms. To measure cognitive reflection, we fitted a Multinomial Processing Tree (MPT) model to the Cognitive Reflection Test. Metacognitive assessment was done by asking the respondents to rate, on a three-points scale, how confident they were of their responses on the syllogisms. The SDT model generated estimates of discrimination (i.e., the ability to differentiate between logically valid and invalid syllogisms) and criterion (i.e., the bias towards choosing the “valid” or the “invalid” responses). The MPT model generated estimates of inhibition and cognitive reflection. A regularized network analysis, using the atan regularization of the correlation matrix of the measures, indicated that the predictability (i.e., a measure of the variance explained by the other variables in the model) of the metacognitive assessments was higher for the MPT measures than for the SDT measures. The implications of these results are discussed.