The fluency of retrieval during a test of memory has been implicated as a cue for judgments of learning (JOLs), but little is known about how fluency affects JOLs. In three experiments, we investigated (1) whether the fluency of generation during study may be a cue for JOLs and (2) whether such fluency effects are mediated by an analytic or nonanalytic inference. To accomplish our goals, we used a learner-observer-judge method. While studying paired associates, learners generated some targets at study. For these items, their JOLs were negatively correlated with the time taken to generate targets. Observers watched learners generate targets and then predicted learners' memory performance. Judges also made JOLs but did not watch the learners generate targets. JOLs from all groups were negatively related to learners' latencies to generate targets, with the magnitude of the relationship equivalent for learners and observers and lower for judges. These and other findings are consistent with the conclusions that the fluency of generation at study is a cue for JOLs and that such fluency effects are partly mediated by an analytic inference about how fluency is related to memory.

A. Koriat's (1997) cue-utilization framework provided a significant advance in understanding how people make judgments of learning (JOLs). A major distinction is made between intrinsic and extrinsic cues. JOLs are predicted to be sensitive to intrinsic cues (e.g., item relatedness) and less sensitive to extrinsic cues (e.g., serial position) because JOLs are comparative across items in a list. The authors evaluated predictions by having people make JOLs after studying either related (poker-flush) or unrelated (dog-spoon) items. Although some outcomes confirmed these predictions, others could not be readily explained by the framework. Namely, relatedness influenced JOLs even when manipulated between participants, primacy effects were evident on JOLs, and the order in which blocks of items were presented (either all related items first or all unrelated items first) influenced JOLs. The authors discuss the framework in relation to these and other outcomes.

ABSTRACT. The authors explored the relations between predictions of the likelihood of recalling studied items (called judgments of learning, or JOLs) and second-order judgments (SOJs), in which one rates confidence in the accuracy of each JOL. Each participant studied paired-associate items and made JOLs. A given JOL was either immediate or delayed and was followed immediately by an SOJ. After all items were studied and judged, paired-associate recall occurred. The incorporation of SOJs into this standard method yielded numerous outcomes relevant to theory of metacognitive judgments. SOJs were greater for extreme JOLs (0, 100) than for intermediate JOLs (40, 50). Also, JOL accuracy was greater for delayed than for immediate JOLs, and, reflecting this delayed-JOL effect, SOJs were greater for delayed than for immediate JOLs. These and other outcomes support 2-process hypotheses of how people make JOLs and uncover some pitfalls in interpreting poor judgment accuracy.Key words: judgments of learning, metacognition, metamemory, second-order judgments IN THIS ARTICLE, we offer an advance for metacognitive research by introducing second-order judgments (SOJs), which provide a new perspective for exploring the processes and accuracy of metacognitive judgments in general and judgments of learning in particular. Judgments of learning (JOLs) are a person's predictions about the likelihood of correctly retrieving recently studied items on an upcoming test. These judgments have become one of the most intensely investigated of the metacognitive judgments (Koriat, 1997;Schwartz, 1994), partly because making them accurately can support the effective regulation of self-paced study (Thiede, 1999). Second-order judgments (SOJs) pertain to an individual's confidence in the JOLs themselves and have not yet been investigated in the lit-

Knowledge updating involves learning about cue effectiveness based on task experience. Priorresearch has yielded inconsistent conclusions regarding age and knowledge updating. To resolve this inconsistency, the authors analyzed the effects of aging within a single paradigm. Participants studied cue-target associates during 2 study-test trials. Cues included rhyme cues and highly effective category cues. On each study-test trial, different items were presented, and participants predicted recall performance, received a cued recall test, and postdicted performance. Knowledge updating was operationalized as an improvement in the accuracy of predictive judgments across trials. An age deficit was evident in improvements in absolute accuracy, whereas age equivalence was evident in relative accuracy. Evidence suggested that deficient inferential processes contributed to the age deficit in knowledge updating.

On-line monitoring during study can be influenced by the relatedness shared between the cue and target of a paired associate. We examined the effects on people's judgements of learning (JOLs) of a different kind of relatedness, which occurs in a list organised into sets of categorically related words and unrelated words. In two experiments, participants studied a list of words organised into a series of sets of four categorically related words or four unrelated words. In Experiment 1, JOLs were made immediately after each word had been studied, and JOL magnitude was greater for related than unrelated words. In Experiment 2, JOLs were delayed after study and, as expected, they were substantially greater for related sets of words. Serial position effects (an increase in JOL magnitude across the words of a related set) were evident with immediate JOLs but not with delayed JOLs. The relatedness effect was not present early in the list for immediate JOLs but was present throughout the list for delayed JOLs. We conclude by discussing some preliminary explanations for these new phenomena.