Exploring a Mnemonic Debiasing Account of the Underconfidence-With-Practice Effect Asher Koriat, Hilit Ma’ayan, Limor Sheffer University of Haifa Robert A. Bjork University of California, Los Angeles Judgments of learning (JOLs) underestimate the increase in recall that occurs with repeated study (the underconfidence-with-practice effect; UWP). The authors explore an account in terms of a foresight bias in which JOLs are inflated when the to-be-recalled target highlights aspects of the cue that are not transparent when the cue appears alone and the tendency of practice to alleviate bias by providing learners with cues pertinent to recall. In 3 experiments the UWP effect was strongest for items that induce a foresight bias, but delaying JOLs reduced the debiasing effects of practice, thereby moderating the UWP effect. This occurred when delayed JOLs were prompted by the cue alone (like during testing), not when prompted by the cue-target pair (like during study). Keywords: judgments of learning, underconfidence, debiasing, foresight bias Koriat, Sheffer, and Ma’ayan (2002) documented a phenome- non that they termed the underconfidence-with-practice (UWP) effect: When participants are presented with the same list of paired-associates for several study-test cycles, their judgments of learning (JOLs) exhibit relatively good calibration on the first study-test cycle, with a tendency for overconfidence. However, a shift toward marked underconfidence occurs from the second study-test cycle onward. The UWP effect was found to be very robust, surviving several experimental manipulations. It has also been replicated in subsequent experiments since (Dougherty & Barnes, 2003; Meeter & Nelson, 2003; Scheck & Nelson, 2005; Serra & Dunlosky, 2005; Simon, 2003;Tiede, Lee, & Leboe, 2004). The UWP effect is surprising for several reasons. First, it stands at odds with the general tendency for overconfidence that has been observed in a great many calibration studies involving retrospec- tive confidence (see Keren, 1991; Koriat, Lichtenstein, & Fisch- hoff, 1980; Lichtenstein, Fischhoff, & Phillips, 1982; McClelland & Bolger, 1994). Second, practice, if anything, would be expected to improve calibration because after the first study-test cycle, learners have a more concrete idea about the difficulty of the task and about their recall performance than they had before. Finally, the impairment in calibration contrasts sharply with the observa- tion that resolution actually improves steadily with practice. Whereas calibration (or absolute accuracy, see Nelson & Dunlo- sky, 1991) refers to the correspondence between mean JOLs and mean recall, and reflects the extent to which recall predictions are realistic, resolution (or relative accuracy) refers to the extent to which JOLs discriminate between recalled and unrecalled items. Thus, unlike the impairment in the JOL-recall (JR) calibration that occurs with practice, resolution, as indexed by the within-person JOL-recall gamma correlation, has been found to improve steadily with practice (e.g., King, Zechmeister, & Shaughnessy, 1980; Koriat, 1997; Koriat et al., 2002; Leonesio & Nelson, 1990; Lovelace, 1984; Mazzoni, Cornoldi, & Marchitelli, 1990). Several explanations of the UWP effect have been proposed. Koriat (1997) suggested that this effect might be a manifestation of a general tendency of learners to discount the effects of extrinsic factors—factors that pertain to the conditions of learning or to the encoding operations applied by the learner. Indeed, he found JOLs to underestimate the effects of presentation duration on recall (but see Dunlosky & Matvey, 2001). The UWP effect accords with this tendency because it implies that the effects of list repetition (an extrinsic factor) are underweighted in the computation of JOLs. This account, however, does not predict the specific JR correspon- dence pattern observed—increased underconfidence—and does not offer a process-type explanation of this pattern. Several additional accounts were explored by Koriat et al. (2002). One is that participants underestimate the correctness of the responses that they supplied on the preceding recall test, hence reporting lower JOLs on a subsequent presentation of the item. It was found, however, that giving participants feedback about the accuracy of their reported targets did not eliminate the UWP effect. Another account was that a fixed-rate presentation of the items might yield an underconfidence bias if learners wrongly estimate that the amount of allotted time was insufficient to memorize the items. However, the UWP effect was also observed when study time was self-paced. Finally, Koriat et al. examined the possibility that the UWP effect is another manifestation of the so called hard-easy effect—the tendency of participants to exhibit overcon- fidence in their answers for difficult items, but either no bias or even an underconfidence bias for easier items (e.g., Gigerenzer, Asher Koriat, Hilit Ma’ayan, and Limor Sheffer, Department of Psy- chology, University of Haifa, Israel; Robert A. Bjork, Department of Psychology, University of California, Los Angeles. This research was carried out at the Institute of Information Processing and Decision Making, University of Haifa. It was supported by a grant from the German Federal Ministry of Education and Research (BMBF) within the framework of German-Israeli Project Cooperation (DIP). We are grateful to Rinat Gil for her help in conducting the experiments. We are indebted to John Dunlosky, Janet Metcalfe, and Sverker Sik- stro ¨m for their comments on a previous draft. Correspondence concerning this article should be addressed to Asher Koriat, Department of Psychology, University of Haifa, Haifa 31905, Israel. E-mail: akoriat@research.haifa.ac.il Journal of Experimental Psychology: Copyright 2006 by the American Psychological Association Learning, Memory, and Cognition 2006, Vol. 32, No. 3, 595– 608 0278-7393/06/$12.00 DOI: 10.1037/0278-7393.32.3.595 595