In two general-knowledge experiments, undergraduates were required to answer multiple-choice questions even when uncertain and then rate their confidence. Correct-answer feedback improved later cued recall of initially correct, low-confidence responses compared with no feedback. The final test followed a five-minute distractor in one experiment and a two-day delay in the other. Thus, under forced initial responding, feedback benefited some answers that were correct but uncertain—not only errors. This finding concerns correct-answer feedback, not generic praise, and does not establish benefits for every learning outcome.

Source: Butler, Karpicke, and Roediger(2008)