Metamemory viewed through the judgment lens

Abstract

Metamemory research makes extensive use of judgments, such as judgments of learning (JOLs). In a JOL, people predict their chance of remembering a recently studied item in a memory test. There is a general agreement that JOLs rely on probabilistic cues that are combined in an inference process. Accuracy as measured by the gamma correlation between JOLs and actual performance is usually mediocre, suggesting limited metacognitive abilities. In judgment and decision-making research, Brunswik's lens model is often used to decompose judgmental accuracy: A matching index G measures how adequately people's cue weights match the optimal weights, two reliability indices assess the predictability of judgments and environment, respectively, and a nonlinear component measures systematic variance not captured by the cues. We employed the lens model equation for the first time to analyze four published and one new JOL data sets. There was considerable interindividual variance in metamemory monitoring. Although gamma was on average higher than the Pearson correlation, it still underestimated metacognitive ability in terms of matching (G). Also, the nonlinear component was considerably higher than in other judgment domains, pointing to substantial item-person-interactions that we interpret as idiosyncratic encoding strategies. An exploratory cluster analysis suggests different metacognitive strategies used by subgroups of participants. We suggest the lens model as a potentially promising tool in metacognition research.