Prediction accuracy and memory performance: Correlational and experimental tests of a metamemory hypothesis

Summary In the performance prediction paradigm both a prediction of subjects' own memory performance (P) and their actual performance (A) are assessed. Prediction accuracy (PA) has been regarded as a measure of memory monitoring which in turn was assumed to be related positively to memory performance. A popular approach to test this metamemory hypothesis is based on correlations between measures of PA and of A. This approach is reexamined critically with regard to two classes of problems. The first is concerned with the definitions of at least six different PA measures represented in the literature. The second refers to statistical artifacts resulting from the part-whole character of the correlations between PA and A. Proceeding from the conclusion that a valid test of the metamemory hypothesis is not possible on the basis of correlations of PA and A, we propose an experimental procedure that properly permits the testing of the hypothesis that the utilization of monitoring activities during learning yields higher levels of memory performance.