Judgments of Learning at Delays: Shifts in Response Patterns or Increased Metamemory Accuracy?

Nelson and Dunlosky (1991) found that judgment-of-learning (JOL) accuracy (measured using G) was nearly perfect if the JOL was made several minutes after study (the delayed-JOL effect). However, over time, the distribution of judgments changed radically: When JOLs were made immediately, subjects typically used the middle of the scale; after a delay, more than 50% of judgments were made using the ends of the scale (Dunlosky & Nelson, 1994, Experiment 1). We replicated the delayed-JOL effect and found a similar rating shift. Is the delayed-JOL effect an artifact produced by this shift, or does it reflect true metamemory improvement? Monte Carlo simulations allowed us to separate these effects. Shifting judgments to ends of the scale did inflate JOL accuracy somewhat. The bulk of the delayed-JOL effect, however, resulted from increases in calibration. We conclude that the delayed-JOL effect reflects true metamemory improvement.