Running Head: DISSOCIATIONS BETWEEN JOLS, BETTING, AND RECALL Judgments of Learning Index Relative Confidence Not Subjective Probability

The underconfidence-with-practice (UWP) effect is a common finding in calibration studies concerned with judgments of learning (JOLs) elicited on a percentage scale. The UWP pattern is present when, in a procedure consisting of multiple study-test cycles, mean scale JOLs underestimate mean recall performance on cycle 2 and beyond. Although this pattern is present both for items recalled and unrecalled on the preceding cycle, to date research has concentrated mostly on the sources of UWP for the latter type of items. The present study aimed at bridging this gap. In three experiments, we examined calibration on the third of three cycles. The results of Experiment 1 demonstrated the typical pattern of higher recall and scale JOLs for previously recalled items compared to unrecalled ones. More important, they also revealed that even though the UWP effect was found for both items previously recalled once and twice, its magnitude was greater for the former class of items. Experiments 2 and 3, which employed a binary betting task and a binary 0/100% JOL task, respectively, demonstrated that people can accurately predict future recall for previously recalled items with binary decisions. In both experiments, the UWP effect was absent both for items recalled once and twice. We suggest that the sensitivity of scale JOLs, but not binary judgments, to the number of previous recall successes strengthens the claim of Hanczakowski, Zawadzka, Pasek, and Higham (2013) that scale JOLs reflect confidence in, rather than the subjective probability of, future recall. DISSOCIATIONS BETWEEN JOLS, BETTING, AND RECALL 3 Judgments of Learning Index Relative Confidence Not Subjective Probability Metacognitive theorists use a variety of different judgments to investigate how people assess their own memory processes. One common one is the judgment of learning (JOL) for which people assess their future memory performance. In a typical experiment employing JOLs, participants study a list of single words or word pairs. After the presentation of each item, a prompt appears instructing participants to rate the likelihood of future recall of that item on a scale from 0% to 100% the JOL. 1 After the study phase, a recall test for the whole list follows. By comparing JOLs to recall performance, two measures can be calculated. First, resolution is the degree to which JOLs distinguish between items that will and will not be recalled at test. In order for resolution to be maximized, later recalled versus later unrecalled items should be assigned high versus low JOLs, respectively. Second, calibration is the difference between mean JOLs and mean recall performance. If the two measures are equal, assessments of future recall are said to be realistic. Mean JOLs lower versus higher than recall performance indicate underconfidence versus overconfidence, respectively. Although most JOL studies reveal overconfidence (e.g., see Koriat, 2012), there are exceptions. For example, the underconfidence-with-practice (UWP) effect (e.g., Koriat, 1997; Koriat, Sheffer, & Ma瀞ayan, 2002) is a common finding in JOL research involving repeated study and recall of the same list over at least two cycles. In most UWP studies, recall performance increases with each additional study-test cycle, as 1 Although the probabilistic 0-100% scale is most commonly used for eliciting predictions of future memory performance, alternative judgment formats (e.g., Hanczakowski, Zawadzka, Pasek, & Higham, 2013; McCabe & Soderstrom, 2011; McGillivray & Castel, 2011) and framings of the JOL prompt (e.g., Finn, 2008; Serra & England, 2012) have been employed by researchers as well. DISSOCIATIONS BETWEEN JOLS, BETTING, AND RECALL 4 does resolution. However, although JOLs are typically similar to recall on the first cycle, they do not increase as much as recall on subsequent cycles, causing calibration to worsen with practice the UWP effect. One explanation of the UWP pattern is based on people瀞s memory for past test performance (e.g., Finn & Metcalfe, 2007, 2008; Tauber & Rhodes, 2012; for alternative accounts of UWP see, e.g., England & Serra, 2012; Koriat, 1997; Scheck & Nelson, 2005). According to the memory-for-past-test (MPT) account of the UWP effect, after cycle 1, people base their immediate JOLs on their performance on the last test. Previously recalled items tend to get high JOLs, as their future recall seems very likely. Conversely, previously unrecalled items are assigned low JOLs, as people remember their failed recall attempt. What people fail to appreciate, though, is that additional learning occurs between the two tests. This additional learning means that some of these previously unrecalled items are recalled on a subsequent test, increasing the discrepancy between mean JOLs and mean recall performance, thus producing UWP. The MPT account localizes the UWP effect mostly in unduly low judgments assigned to previously unrecalled items. However, Koriat et al. (2002), Finn and Metcalfe (2007), and Hanczakowski et al. (2013) reported the presence of the UWP pattern for previously recalled items as well. Finn and Metcalfe argued that these items may contribute to the UWP effect because of variability present in JOLs. Subsequent recall of items that were successfully recalled on a previous cycle(s) is typically excellent and they attract very high JOLs. However, because the JOL scale ends at 100%, any variability is necessarily downward, resulting in mean JOLs that underestimate mean recall performance (i.e., underconfidence). But what produces this downward variability? Finn and Metcalfe remain agnostic of its source. One DISSOCIATIONS BETWEEN JOLS, BETTING, AND RECALL 5 option is that it may be simply random, not stemming systematically from any characteristics of the rated items. For example, people may be reluctant to use the 100% rating too often, therefore assigning lower ratings to some items even though they believe that they are extremely likely to be later recalled. It is equally plausible, however, that the JOL variance for recalled items depends on item-specific information (see Dougherty, Scheck, Nelson, & Narens, 2005, for similar considerations regarding random and systematic sources of variability in delayed scale JOLs). If item-specific information is responsible for the downward variability present in scale JOLs, then it is important to consider what cues might influence the assignment of these judgments. One type of information participants may rely on when predicting future recall of the previously recalled items is how many times these items were successfully retrieved. If JOLs are collected in a procedure employing at least three study-test cycles, then it stands to reason that items that were successfully retrieved twice, on both cycles 1 and 2, will receive higher JOLs than items that were successfully retrieved once, on either cycle 1 or cycle 2. As Vaughn and Rawson (2011) have shown, the number of successful retrievals in a learning-to-criterion procedure determines the magnitude of JOLs. Crucially, it seems also likely that such a systematic source of variance in JOLs for already recalled items will contribute to the pattern of UWP. As Finn and Metcalfe (2007) noted, any variance in JOLs for recalled items in the multi-cycle procedure used to investigate JOLs is likely to worsen calibration by removing JOLs further from the ceiling level of recall performance. Thus, if participants incorporate the information about the number of previous successful retrieval attempts into their JOLs, then items previously recalled once only should exhibit worse calibration than items previously recalled twice, DISSOCIATIONS BETWEEN JOLS, BETTING, AND RECALL 6 exaggerating UWP. The aim of the present study was to test this hypothesis by comparing the way in which people assign scale and binary ratings to previously recalled items in the UWP paradigm.