Assessing confidence reduces the benefits of response revisions in a general knowledge test

Abstract Students change their responses during tests, and these revisions are often correct. Some studies have suggested that decisions regarding revisions are informed by metacognitive monitoring. We investigated whether assessing and reporting response confidence increases the accuracy of revisions and the final test score, and whether confidence in a response is related to the probability and accuracy of its revision. One hundred and seventy-three participants performed a general knowledge test in which they provided an initial response and then either confirmed or revised it. We compared a group in which participants rated their confidence after the initial response against two control groups: one in which participants were not required to do anything between the initial and final response, and one in which participants performed an additional (not metacognitive) task. The results provide no evidence that reporting confidence improves response revisions or general test accuracy. We observed the highest accuracy improvement in the control group that did not include any additional task. Furthermore, reporting confidence decreased the number of revisions. However, confidence level was related to the probability and accuracy of revisions: responses associated with lower confidence were more often revised, and revisions were more often correct when participants reported some degree of initial response confidence than when they were initially guessing. The results suggest that metacognitive reports, being a demanding cognitive task, could disturb the potentially beneficial process of revising one’s response. Additionally, assessing confidence in one’s response might lead to confirmation bias and decrease the tendency to revise it.