Abstract
Recent studies on judgments of learning (JOLs) suggest that beliefs play an important role in the formation of JOLs. The current article introduces a multilevel mediation model to quantify the contribution of beliefs to JOL formation when both JOLs and global belief-based predictions are measured from the same group of participants. Our examples of fitting the multilevel mediation model to hypothetical and published datasets revealed that it is feasible to use the multilevel mediation model to examine the mediation effect of beliefs on the relationship between a cue and JOLs, and quantitatively compare the effects of beliefs and processing fluency on JOLs in one model. Then we compared the current multilevel mediation model and the multilevel moderation model implemented in previous studies, and discussed their similarities and differences. Finally, a data simulation was performed to explain the inflation of Type I error for the multilevel mediation model when we regress global belief-based predictions on the cue, and suggestions about appropriate steps for conducting multilevel mediation analysis are proposed.