关键词: 元记忆自信偏差, 学习判断, 锚点, 低锚反馈, 高锚反馈

Abstract: Judgments of learning (JOL) refer to the prediction of whether learned items can be recalled in a future test. Previous studies have shown that JOLs are not always accurate and often exhibit systematic biases. When JOL is higher than the recall performance, it manifests as an overestimation bias; conversely, when JOL is lower than the recall performance, it manifests as an underestimation bias. These biases may affect individuals’ selection of learning strategies and allocation of learning time, thereby reducing learning efficiency. Studies on whether feedback reduces metamemory bias remain inconsistent. Some studies reported no effect, whereas others found partial calibration. Furthermore, the mechanism underlying these effects remains unclear. According to the anchoring-and-adjustment heuristic, individuals make JOLs based on a psychological anchor, which is partially derived from prior learning experiences or memory beliefs. This anchor may change during the learning of different lists. Therefore, feedback may affect JOLs for subsequent lists by updating individuals’ anchor associated with the list. Based on this heuristic, the present study designed two types of feedback in a multi-list learning context: high–anchor (performance + 10%) and low–anchor (performance – 10%) feedback. In two experiments, participants learned four lists. Before learning, they made a global judgment of learning (GJOL) to predict the number of items they could recall in a future memory test. This indicator was used to measure participants’ psychological anchor before learning each list. Subsequently, participants learned each item and made item–level JOLs. After learning, they performed a 2–minute distractor task, followed by a memory test, and received list-level feedback. Experiment 1 used 40 word pairs as materials and adopted a cued recall task to elicit overestimation bias, while Experiment 2 used 30 word pairs and employed a two-alternative force–choice recognition task to elicit underestimation bias. Experiment 1 employed a 2 (Feedback Type: high–anchor vs. low–anchor) × 4 (List Order: Lists 1–4) mixed experimental design with 69 participants (34 in high–anchor and 35 in low–anchor group). The results showed that high-anchor feedback had no calibration effect on overestimation bias across all four lists, whereas low-anchor feedback fully calibrated this bias. Meanwhile, the GJOL (i.e., the anchor) mediated the effect of feedback from the previous list on subsequent JOLs. Experiment 2 used the same design with 63 participants (33 in high–anchor and 30 in low–anchor group). The results showed that low-anchor feedback had no calibration effect on underestimation bias across all four lists, whereas high-anchor feedback fully calibrated this bias. Also, GJOL mediated the effect of feedback on subsequent JOLs. The above results indicate that low–anchor and high–anchor feedback can specifically calibrate metamemory underestimation and overestimation biases, respectively. Different types of feedback achieve the calibration by updating individuals’ psychological anchor, which in turn adjusts their JOLs. Theoretically, this study expands the application boundary of the anchoring-and-adjustment heuristic to metamemory calibration. Practically, it offers implications for educational and cognitive training programs by providing targeted methods to correct different types of metamemory confidence bias.

Key words: metamemory confidence bias, judgments of learning, psychological anchor, low–anchor feedback, high–anchor feedback