A judgment of learning predicts whether an item will be recalled later. In Nelson and Dunlosky's word-pair study, undergraduates made these predictions using only the first-word cue, without seeing the target word. Judgments made after 10–33 intervening study or judgment items better distinguished pairs later recalled from pairs forgotten than immediate judgments. This improvement was in relative accuracy: ranking items by their likelihood of later recall. It does not establish improved memory performance or perfect calibration of confidence, and the evidence concerns cued word-pair recall rather than general understanding.

Source: Nelson and Dunlosky (1991)