Boise State UniversityScholarWorksCurriculum, Instruction, and Foundational StudiesFaculty Publications and Presentations

Department of Curriculum, Instruction, andFoundational Studies

1-1-2013

MetamemoryJohn DunloskyKent State University

Keith W. ThiedeBoise State University

"Metamemory" by John Dunlosky and Keith W. Thiede (pp. 283-298) is from The Oxford Handbook of Cognitive Psychology, edited by Daniel Reisberg(2013), is reproduced by permission of Oxford University Press. doi: 10.1093/oxfordhb/9780195376746.013.0019

CHAPTER

Metamemory19John Dunlosky and Keith W Thiede

Abstract

Metamemory refers to people's beliefs about their memory and to how people monitor and controltheir learning and retrieval. In this chapter, we describe monitoring and control processes involvedin learning and retrieval, how these processes have been measured, and key outcomes relevant tohuman metamemory. Based on these outcomes, general conclusions include the following: (a) people'sjudgments of their memory are based on a variety of cues; hence (b) judgment accuracy arises fromthe diagnosticity of the cues, so that above-chance accuracy of any metamemory judgment only ariseswhen the available cues are predictive (or diagnostic) of criterion performance; and finally, (c) peopleuse their memory judgments to guide their study and retrieval. Thus, people's memory monitoringplays a pivotal role in the effectiveness of their self-regulated learning and retrieval, so a major aimof metamemory research is to discover techniques that yield high levels of judgment accuracy andoptimal regulation.

Key Words: metamemory, self-regulated learning, memory monitoring, control, judgments oflearning, feeling of knowing, confidence judgments, metacognition

Metamemory refers to people's thoughts abouttheir memory and how memory operates. Althoughthe term metamemory may seem esoteric to some,people rely on their metamemory as they performmany activities. When heading to a grocery store,one person may believe that he can remember all 10{ems that need to be purchased, whereas anotherperson with the same items to purchase may believethat her memory will fail and hence decides to take1 list along. An eyewitness may point the finger atthe accused, and do so with extreme confidence thatthe accused had committed the crime; however,such high confidence can be illusory, such as whenconfidence is inadvertently based on a memoryof seeing the accused in a lineup instead of actu-ally witnessing the accused commit the crime (forreal-life examples, see Loftus & Ketcham, 1991).

Unfortunately, jurors believe the testimony of ahighly confident eyewitness, regardless of whetherhis or her confidence is well placed. On a lighternote, when playing games (such as Trivial Pursuit),players may withhold answers when they believethat those answers are wrong, yet they will try topersuade their team to respond with an answerwhen their confidence in it is high. And when fail-ing to generate answers to some questions, an emo-tionally charged tip-of-the-tongue state may arisewhen they believe the sought-after answer is avail-able in memory. Being a student can put an evenhigher premium on the effective use of metamem-ory, because to learn efficiently, students must beable to accurately judge how well key concepts havebeen learned and make appropriate decisions aboutwhich concepts require further study.

hort list illustrates several metamemoryrocesses that can occur while people are learning, materials or are attempting to retrieve old ones.

ioreo\'er, rhey illustrate the multifaceted nature0- metarnernory, which includes knowledge (andbeliefs) abour memory, moniroring of memory, andcontrol of memory. Metamemory knowledge refers ro

lara rive knowledge or beliefs that an individualholds about how memory operates and whetherthose beliefs are accurate. In the earlier example, theindividual who decided ro write down the groceryu t may know from experience that memory tendsto fail after four or five items from a grocery list needto be remembered. The individual who went with-out a list may believe that he has a very good memo-ry-s-thar is, he has high memory self-efficacy-andhence will be able to recall even lengthier lists with-our having difficulties.

Memory monitoring refers to assessing progressduring learning or the current state of a previ-ously studied item. For example, when studyingfor an upcoming test, students may attempt tomonitor and evaluate their ongoing progresswhile studying. Their moniroring of memoryyields a confidence judgment about whether theywill remember the key concepts on the upcomingexam. Finally, memory control involves regulatingany aspect of learning or retrieval. For the con-rrol of learning, one example of a control processincludes deciding to spend more time studyingmaterials that one believes have not been welllearned. For retrieval, accusing a defendant ofcommitting a crime is a prime example, becausethe eyewitness allegedly must be highly confidentin his or her memory of the crime to potentiallycondemn the defendant.

Based on these examples, it may be evidentwhy so many people have become interested inmetamemory: Faulty metamemory can lead topoor memory and low achievement, whereasaccurate metamemory can enhance memory andachievement. For instance, if students inaccu-rately assess that they are ready for an upcomingtest, then they may prematurely stop studying.In this case, poor monitoring leads to a nonopti-mal control decision, which in turn would lead toless-than-expected performance on the examina-tion. To expand on this example, consider two stu-dents who are studying for an upcoming test in aclass of introductory psychology. Both students aretrying to learn the core concepts relevant to mem-ory. such as what is short-term memory, long-termmemory, and encoding specificity. Both students

METAMEMORY

judge thar they will be able to remember about90% of the concepts, which they believe is fine forthe test, so rhey stop studying to join each otherfor a late-night movie. Whereas Julie actually willretain 90% of the concepts (her judgments wereaccurate), Mike was highly overconfident and willremember only 50%. In this case, Mike was over-confident and hence he stopped studying before hemet his desired learning goal; unfortunately, Mikewill not perform well on the test, and he may eventell the teacher, "But I thought I knew all theseconcepts. Why did I earn such a poor grade onthe exam?"

Although the importance of accurate metamern-ory may be intuitive, research in this area did notbegin in earnest until about 1970 when John Flavellcoined the term "meramemory." In the 1970s, theterm "meta" began to appear in articles and in con-ference papers, and groundbreaking research wasalso conducted by John Flavell, Ellen Markman, andAnn Brown, among others. Perhaps most importantfor solidifying a specialized field of metamemorywas Flavell's (1979) American Psychologist article,called "Metacognition and Cognitive Moniroring:A New Area of Cognitive-Developmental Inquiry."This article has been highly influential because it ishere where Flavell defines core concepts for the field,such as metacognitive knowledge and metacogni-tive experiences (which are most closely alignedwith metamemory knowledge and memory moni-toting, as described earlier). As important, he devel-oped numerous testable hyporheses about how thedevelopment of merarnernory in childhood wouldin turn influence the developmental progress of corecognitive processes.

John Flavell and his colleagues-most nota-bly, Henry Wellman-first captured our attentionwith persuasive arguments about the importanceof rnetamemory, but it was Joseph Hart who pro-vided the first method to assess metamemory in anobjective manner. One can best understand Hart'sbreakthrough within the context of the introspec-tion method that was used in the infancy of psy-chological research. In the early 1900s, a researcherinterested in associations may ask a (trained) par-ticipant to introspect about the psychological pro-cesses that produce a free association. For instance,what is the first word that comes to mind whenyou read "guitar"? Perhaps you think "Stratocasrer,"but if you were an introspectionist, you would alsoneed to describe the cognitive processes that pre-ceded the thought "Stratocaster." Doing so mayseem difficult, but participants in these studies (at

rimes the experimenters themselves) were highlypracticed. A downfall of this method was that theintrospective reports were viewed as a windowinto the mind; that is, the reports were assumedby some to be accurate and complete. This viewis most evident in R. S. Woodworth's (1921) defi-nition of introspection as the "observation by anindividual of his own conscious action ... Noticethat it is a form of observation, and not specu-lation or reasoning from past experience. It is adirect observation of fact" (p. 10). This definitionemphasizes introspection as a metacognitive act,because introspection involves directly observinga mental action. Unfortunately, even in 1901, itvas evident that introspective methods would fall. 'ell short of direct observation of mental actionsinvolved in a response (for a historical review, seeHumphrey, 1951).

Six decades later, Joseph Hart introduced amethod to systematically explore distortions inpeople's introspections about their memory. In par-ticular, participants were asked general-informationquestions. When they did not answer a question,they were simply asked to predict whether they couldchoose the correct answer on a multiple-choice rec-ognition test. These predictions-which are called.7eling-ofknowing (FOK) judgments-are intro-pecrive reports, but in contrast to earlier intro-spection research, Hart (1965) did not assume thatthey were valid, Instead, he evaluated their accuracyagainst objective performance by administeringrecognition tests. By doing so, people's judgmentscould be validated against their performance: Whenthey were in a tip-of-the-tongue state and were surethey knew the correct answer even when they couldnot retrieve it, would they then recognize the cor-rect answer? Hart (1965, 1966) found that people's?OK judgments showed above-chance accuracy.That is, their judgments were higher for correctlyrecognized answers than for ones that were notecognized.

Hart's methods focused on the monitoringof retrieval, as measured by FOK judgments.The method is invaluable because it has allowedresearchers to systematically explore the biases inpeople's FOK judgments, which we consider insome detail later. As important, his methods can bereadily applied to evaluate how people monitor allaspects of learning-from study through retrieval,And extensions of his methods have been used toexplore people's control of study and retrieval. In:he remainder of this chapter, we first provide aird's eye view of the kinds of monitoring and

control processes investigated in the field, We thendiscuss current theory pertaining to two widelyinvestigated metamemory judgments relevant tostudy and retrieval: judgments of learning (JOLs)and FOK judgments, respectively.

In 1990, Nelson and Narens unified the field ofmetamemory by organizing the various monitoringand control processes into a single framework. Anexpanded version of this framework is presented inFigure 19.1, which illustrates measures of memorymonitoring and memory control that correspondto each phase of learning. For instance, duringstudy ("acquisition" in Fig. 19.1), one may monitormemory for to-be-learned items, which is measuredby having people judge how well an item has beenlearned. This JOL may also influence the control ofstudy, such as by informing people's decisions aboutwhen to terminate study of a given item. The mea-sures depicting memory monitoring are presentedin the top portion of Figure 19.1, and the measuresdepicting memory control are presented in the bot-tom portion. Definitions of each measure are inTable 19.1.

Modern research on metamemory largelyfocuses on answering just a few questions aboutmonitoring and control. Concerning memorymonitoring, one primary question is, How dopeople monitor various phases of learning?Answers to this question are obtained by investi-gating how people make the various monitoringjudgments. For instance, JOLs are investigated tounderstand how people monitor their learning,whereas confidence judgments and FOK judg-ments are investigated to understand how peoplemonitor their retrieval. The accuracy of these judg-ments is often of central interest, and taking thelead from Hart (1965), accuracy is measured bycomparing a given judgment to its correspondingcriterion measure. JOLs (which are predictions offuture performance) are compared to future per-formance, FOK judgments for predicting futurerecognition of currently unrecallable informationare compared to future recognition performance,and confidence judgments are compared to per-formance on the criterion test being judged.These and other judgments often demonstrate apositive correlation with criterion performance.For instance, Souchay, Moulin, Clarys, Taconnar,and Isingrini (2007) had older (M age = 72 years)and younger (M age = 27) adults attempt toanswer general-information questions and thenmake an FOK judgment for any question theydid not answer. After this judgment phase, the

DUNLOSKY, THIEDE 285

I - - - - - - - - - - - - - - - - - - Monitoring - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - 1

Acquisition

Feeling-of-knowingjudgments

Retention Retrieval

In advanceof learning

On-goinglearning

Self-directedsearch

Output ofresponse

Maintenanceof knowledge

Selectionof searchsrrategy

Control - - - - - - - - - - - - - - - - - - - - - - -

Figure 19.1 Overview of meramernory monitoring and control components that occur throughout study (acquisition), retention, andretrieval. (Adapted from Nelson & arens, 1990, to include judgments that were absent from their original framework.)

Name

Table 19.1 Names and Common Definitions of Metacognitive Judgments and Control Processes

Definition

Metacognitiue judgments

Ease-of-learning (EOL) judgments Judgments of how easy to-be-studied items will be to learn.

Judgments oflearning GOLs) Judgments of the likelihood of remembering recently studied items on anupcoming test

Feeling-of-knowing (FOK)

judgmentsJudgments of the likelihood of recognizing currently unrecallable answers on anupcoming test

Source-monitoring judgments Judgments made during a criterion test pertaining to the source of a particularmemory

Confidence in retrieved answers Judgments of the likelihood that a response on a test is correct. Often referred toas retrospective confidence (RC) judgments

Control processes

Selection of kind of processing Selection of strategies to employ when arrempting to commit an item tomemory

Item selection Decision about whether to study an item on an upcoming trial

Termination of study Decision to stop studying an item currently being studied

Selection of search strategy Selecting a particular strategy in order to produce a correct response during a test

Termination of search Decisions to terminate searching for a response

SoUTU: From Dunlosky, J., Serra, M., & Baker,J. M. C. (2007). Metamemory applied. In F. Durso er aI. (2nd ed.) Handbook a/applied~Jtion. New York:Wiley.

2 6 METAMEMORY

participants completed a recognition test foreach of the unrecalled questions. Means acrossintraindividual correlations between FOK judg-ments and recognition performance were .35 foryounger adults and .46 for older adults, indicat-ing that people of all ages can accurately predictfuture recognition performance for currentlyunrecallable information (for extended primerson metamemory measurement and analyses, seeDunlosky & Bjork, 2008; Dunlosky & Metcalfe,2009).

Concerning memory control, a primary ques-tion is, How do people use monitoring to controltheir memory? To address this question, researchersoften examine the relationship between a measure ofmonitoring and a measure of control. For instance,to examine control of study, JOLs made duringan initial study trial may be compared to subse-quent self-paced study time (Mazzoni, Cornoldi,& Marchitelli, 1990). JOLs are often negativelyrelated to later self-paced study time-that is, learn-ers tend to spend more time studying informationthat has been less well learned (lower JOLs) thanbetter learned (higher JOLs). As we discuss later,explaining this negative relationship between JOLsand self-paced study times-and exceptions to it-has become a principal focus of current theories ofself-paced study.

Theory and Data of Basic Monitoring andControl Processes

The literature on rnetarnernory includes basicresearch on metamemory knowledge, monitoring,and control; analyses of changes in metamemorythat occur throughout the life span; and the neu-roscience and neuropsychology of monitoring andcontrol processes (Dunlosky & Metcalfe, 2009).Given the size of this literature, our overview heremust be limited in scope, so we have decided tofocus largely on twO questions: How do peoplemake metamemory judgments during study andretrieval? And how do people control their studyand retrieval? To answer them, we could rely onthe literature of almost any of the measures pre-sented in Figure 19.1. Presently, however, we con-sider only JOLs and FOK judgments, because theliterature in both cases is extensive and offers somebasic principles that apply to all metamemoryjudgments. A more thorough introduction to theliterature on each of the monitoring and controlfunctions in Figure 19.1 is provided by Dunloskyand Metcalfe (2009) and Dunlosky, Serra, andBaker (2007).

Monitoring of Study and RetrievalHOW DO PEOPLE MAKE JUDGMENTS OF

LEARNING AND FEELING-OF-KNOWING

JUDGMENTS?Some of the earliest theoretical work on the

bases of FOK judgments explored the degree towhich they directly measure memory. Accordingto this direct-access view, people's judgments reflectthe underlying strength of an item in memory(cf. trace-access mechanisms in Nelson, Gerler, &Narens, 1984). So, if you made aJOL after you hadjust studied "dog-spoon," your judgment woulddirectly tap how well the association between "dog"and "spoon" was stored in memory. Such directaccess echoes the early introspectionists' belief thatself-evaluations arise from a "direct observation offact," and perhaps not surprisingly, predictions fromthe direct-access view have been disconfirmed withregard to JOLs and FOK judgments.

Let's just consider a disconfirmation of thisview from the JOL literature, which will also serveto introduce cue-based accounts for metamemoryjudgments. Benjamin, Bjork, and Schwartz (1998)had participants answer general-knowledge ques-tions that were moderate to easy in difficulty level,so participants could generate an answer to mostquestions. Immediately after they generated ananswer, participants made a JOL concerning thelikelihood that they would recall the answer ona test of free recall; for this test, they were givena blank sheet of paper and had to recall only thepreviously retrieved answers (without the questioncues). JOLs were related to the fluency of initialretrieval-the faster they could generate an answerto a general-knowledge question, the more likelythey judged that they would be able to freely recallit on the criterion test. By contrast, the oppositewas true for the cri terion test; that is, the faster theanswer was retrieved during the initial phase, theless likely they were to recall it during the free-recallcriterion test. Thus, people's JOLs did not directlyaccess how well each target answer was stored inmemory. Instead, JOLs tracked retrieval fluency:As they retrieved answers more quickly during theinitial test, they judged the answer would be easierto remember, even though this assessment did notreflect actual storage strength (as assessed by freerecall on the criterion test).

This outcome, among many others, suggeststhat people use processing fluency as a cue tomake JOLs. In fact, people tend to use fluency foralmost all metamemory judgments when process-ing fluency differs across to-be-judged items (Alter

DUNLOSKY, THIEDE 287

& Oppenheimer, 2009). This conclusion not onlypertains to retrieval Ruency (as in Benjamin et al.,1998) but other forms of processing fluency aswell. Another example involves Ruency of gener-ating encoding strategies. In one study (Hertzog,Dunlosky, Robinson, & Kidder, 2003), college stu-dents studied word pairs and were asked to generatean interactive image to associate both words. Duringstudy, participants pressed the space bar once theyhad generated an image. The time from an item'spresentation and this key press was a measure of theRuency of image generation. Across three experi-ments, encoding Ruency was negatively related toJOLs; that is, the more Ruently participants gener-ated an image (faster latency), the higher people'sjudgment that they would correctly recall the item.Even subtle aspects of the stimulus environment caninfluence encoding Ruency and JOLs. For instance,Rhodes and Castel (2009) manipulated the Hu-ency of perceptually processing items. Participantslistened to items that were presented at differentvolumes and made a JOL after each item was pre-sented. JOLs were higher for items presented loudlythan those presented quietly, presumably becausethe former were more Ruently processed. Thus,both encoding Ruency and retrieval Ruency influ-ence people's JOLs.

People also use Ruency when making FOK judg-ments. The potential influence of this cue is mostevident in Koriat's (1997) accessibility hypothesis.According to this hypothesis, when people make anFOK judgment for an unrecalled item, it is based onthe information that they had accessed while search-ing for an answer. So, if you are asked, "What cityis the capital of Washington State?" you may fail tocome up with an answer, but in trying to retrieve onefrom memory, you may remember "emerald," that itbegins with an "s" and that it is in the far northwestportion of the state. For another question, "Whowrote Moll Flanders?" you do not respond but thinkthe author's name begins with a "D." A predictionfrom this hypothesis is that the more informationyou access-and the more Ruently you access it-prior to making an FOK judgment, the higher yourFOK is expected to be. So you may judge that youwill absolutely recognize the capital of WashingtonState but will unlikely recognize the name of theauthor who wrote Moll Flanders.

Importantly, Koriat (1993) proposed that thequality of what is accessed is irrelevant-even if a lotof incorrect information is Ruently accessed, then ahigh FOK will result. In the present example, notethat all the accessed information about Washington's

METAMEMORY

capital pointed toward Seattle (the "EmeraldCity"), when in fact, the correct answer is Olympia.By contrast, "D" is the first letter of the author'sfirst and last names (Daniel Defoe). Nevertheless,according to the accessibility hypothesis, the sheeramount retrieved and the Huency of retrieval aremost influential, so FOK judgments are expected tobe greater for the question about Washington's capi-tal, even though the partial information retrievedabout it was incorrect. In a series of creative studies,Koriat (1993) demonstrated the powerful inRuenceof accessibility on FOK judgments. In one experi-ment, participants studied tetragrams-four-letterstrings that were nonwords, such as RFSC andFKRD. A tetragram was briefly presented for study,which was followed by a short retention intervaland a retrieval attempt. After the retrieval attempt,an FOK judgment was made for that tetragram, andthen after all tetragrams were studied and judged,a recognition test was administered so that FOKaccuracy could be measured. Most relevant for now,people's FOK judgments were related to how manyletters were recalled prior to making the judgment,even when those letters were not in the studied tet-ragram. Put differently, FOK judgments were basedon how much information was accessed, regardlessof whether the information recalled was correct.

We have focused almost exclusively on people'suse of fluency as a basis for their judgments. Moregenerally, however, people appear to base theirjudgments on many cues that appear to distinguishbetween items in a manner that may be relevant tocriterion performance. For instance, another cuethat is available when people make FOK judgmentsis the familiarity of the cue used to prompt theretrieval attempt. The cue for "What city is the capi-tal of Washington State?" is the question itself. Fora variety of reasons (e.g., perhaps you recently readan article about Washington State), you may findthe cue itself familiar and hence judge that you arelikely to recognize the correct answer. Research hasestablished that FOK judgments are influenced bycue familiarity in this manner (Metcalfe, Schwartz,& Joaquim, 1993; Miner & Reder, 1994).

Given that metamemory judgments can bebased on multiple cues, researchers are beginning toexplore how various cues are integrated into a judg-ment. For FOK judgments, Koriat and Levy-Sadot(2001) proposed that cue familiarity has the firstinfluence. If your familiarity with a cue is low, thenyou would spend little, if any, time attempting toretrieve the correct answer and your FOK would below. However, if you are familiar with the cue, then

you would search for an answer, and any informa-tion you accessed during this search would in turninfluence your FOK judgment. That is, accessibil-iry is expected to have a greater influence on FOKjudgments when cue familiariry is high than when itis low. For instance, if you were asked, "Who is themayor of Ravenna, Ohio?" you would likely haveno familiariry with this particular cue and quicklyrespond with the lowest FOK judgment withouteven attempting to search for the answer (cf Kolers& Palef, 1976). However, if you were asked, "Whowas the second president of the United States?" ifyou had familiariry with the presidents of the UnitedStates, then you would likely search for the correctanswer. While searching, any information that youaccessed along the way would then influence yoursubsequent FOK judgments. Across three experi-ments, Koriat and Levy-Sadot (2001) providedevidence consistent with this interactive hypothesis.Moreover, Benjamin (2005) demonstrated howcue familiary and target accessibiliry both influ-ence people's JOLs, but the time course for theirinfluence differs. When JOLs are made under timepressure, cue familiariry has an influence. Withouttime pressure, people have more time to attemptretrieval, and hence this cue has a larger influenceon people's JOLs.

To conclude this section, we will rerum to ourinitial question, How do people make JOLs andFOK judgments? Processing fluency and cue famil-iariry have a joint influence on these judgments, butthey also can be influenced by any number of cuesthat are available when the judgments are made.For instance, in some experiments on JOLs, par-ticipants study paired associates consisting of eitherunrelated (dog-spoon) or related (king-crown)words. JOLs are much higher for related than unre-lated items (e.g., Carroll, Nelson, & Kirwan, 1997).In some cases, people's JOLs are influenced by theserial position of items on a list, with higher judg-ments being made for items in the first positionsthe primacy items) than for those in the middle ofthe list (Castel, 2008; Dunlosky & Marvey, 2001).

Given the number of cues that do influenceJOLs, it is perhaps surprising that some relativelyobvious ones have a rather minimal (or inconsis-tent) influence. For instance, when studying pairedassociates, people can use numerous strategies, suchas interactive imagery (for "dog-spoon," imaginga dog paddling in a large spoon filled with milk)or repetition (repeating "dog-spoon" togetherduring study). The strategies differ in their effective-ness, with final recall performance typically being

much higher after interactive imagery than roterepetition. When people judge items studied underthese two different strategies, however, their JOLsrypically do not differ for items that were studiedusing imagery rather than repetition (Rabinowitz,Ackerman, Craik, & Hinchley, 1982; Shaughnessy,1981). Likewise, other studies have demonstratedthat people's JOLs can be insensitive to retentioninterval and to the number of learning trials (e.g.,Carroll er al., 1997; Koriat, Bjork, Sheffer, & Bar,2004; Kornell & Bjork, 2008, respectively), whichoften have a substantial influence on criterion recallperformance.

Why are some cues influential (e.g., fluency),whereas other cues (e.g., encoding strategy or num-ber of learning trials) are nor? One answer is basedon the assumption that all judgments are inferen-tial. That is, people use various cues to infer thelikelihood of correctly performing on the criteriontest. What makes one cue more potent than anothercan partly be ascertained from Koriat and col-leagues' (Koriat, Nussinson, Bless, & Shaked, 2008)inference-based account that emphasizes a distinc-tion between whether the source of an inference isexperience based (EB) or information based (IB). IB(or theory-based) judgments rely on people's declar-ative knowledge and beliefs about how a particularcue influences memory. For instance, as noted ear-lier, JOLs do not distinguish between items that arestudied by imagery versus rote repetition, and thisnull effect presumably arises because most peopledo not know that imagery is a more effective strat-egy. Once people are given experience using bothstrategies, however, they obtain declarative knowl-edge that "imagery is better," and in subsequenttrials, JOLs (and other metmamemory judgments)are higher after imagery than repetition (Hertzog,Price, & Dunlosky, 2008).

EB judgments involve two processing stages:"first a process that gives rise to a sheer subjectivefeeling and second a process that uses that feelingas a basis for memory predictions" (Koriat et al.,2008, p. 118). The idea here is that various cuestrigger feelings, which in turn boost metamemoryjudgments. According to Koriat et al. (2008), EBjudgments for JOLs rely on the ease with whichitems are encoded or retrieved during learning,and EB judgments for FOKs rely on the fluency ofaccessing partial information when searching for atarget answer. Put differently, differences in the flu-ency of processing presumably give rise to differentfeelings that then influence people's metamemoryjudgments.

DUNLOSKY, THIEDE 289

This taxonomy does provide some insight intowhy various cues influence-or do not influence-people's metamemory judgments. For instance, ifa participant does not believe a particular cue willinfluence memory (IE) and if the cue does not giverise to a metacognitive experience (EB), then itshould not influence the judgments. This situationpresumably holds for different encoding strategiesthat most participants would not have experienceusing in a novel associative task, such as imagery andrepetition. Another possibility is that a cue whichproduces a strong metacognitive experience (e.g.,processing fluency) could override the influence ofother IE cues. Thus, when people study "archer-tree" and then make a ]OL for it, they may be cap-tured by the ease of developing an image for theword pair (and hence make a high ]OL), and alto-gether ignore the fact that the word pair is slated tobe tested weeks after study (and hence should prob-ably receive a low ]OL). In this way, even obviousIE cues (e.g., retention interval) may not influence]OLs when other strong EB cues are available in theenvironment. This taxonomy is useful for explain-ing the influence of various cues on metamemoryjudgments, and it has intuitive appeal. Even so, thetaxonomy is not entirely orthogonal, because peo-ple apparently have theories about fluency cues; thatis, learners believe that fluently processed informa-tion is easier to remember. So this EB cue (fluency)may actually have its influence on metamemoryjudgments vis-a-vis IE processing (as in Marvey,Dunlosky, & Guttentag, 2001).

WHEN WILL METAMEMORY JUDGMENTS BE

ACCURATE?

Hart's groundbreaking research demonstratedthat the accuracy of people's FOK judgments wasabove chance, and the seminal work on ]OLs byArbuckle and Cuddy (1969) also revealed that theyhad above-chance accuracy. Why? One straight-forward answer is offered by cue-based accountsof metamernory judgments, which are them-selves founded on Egon Brunswik's lens model ofperceptual judgments (Brunswik, 1956). Theseaccounts highlight the importance of the threerelationships among a cue, judgment, and crite-rion performance-which are illustrated in Figure19.2. The relationship between cue and judgmentis referred to as cue utilization, which represents thedegree to which a cue influences (or is related to) thejudgment. k described earlier, cue utilization forprocessing fluency is usually high for metamemoryjudgment. A second relationship occurs between

290 I METAMEMORY

Cueutilization

Judgmentaccuracy

Cuediagnosticity

Figure 19.2 Relationships among cues, judgments, and recallthat emphasize the central role of cue utilization and diagnostic-ity in determining judgment accuracy.

judgment and criterion performance, which rep-resents judgment accuracy. Finally, the relationshipbetween cue and performance is referred to as cuediagnosticity and represents the degree to which aparticular cue predicts criterion test performance(cf ecological validity, Brunswik, 1956).

According to the cue-based accounts, judgmentaccuracy is a function of cue urilization and cuediagnosticiry, and hence, above-chance accuracyoccurs whenever (a) a cue influences (or is relatedto) a judgment and (b) the cue itself accuratelypredicts-or rather, is diagnostic of-criterion per-formance. Above-chance accuracy is not ensured,because even when a cue is used to make a judg-ment, the cue itself may have nil diagnosticity oreven negative diagnosticiry, In the latter case, thejudgments would show below-chance accuracy,with higher confidence in performance (e.g., higher]OLs) being indicative of lower criterion perfor-mance. In fact, the most impressive support forsuch inference-based accounts of accuracy comesfrom studies in which the accuracy of people's judg-ments was zero or negative.

One excellent example comes from Benjaminet al. (1998), which was described earlier. Recall thatparticipants used initial fluency of retrieving answersto general-information questions as a cue to make]OLs for predicting subsequent free recall of theanswers. In this case, the cue of retrieval fluency wasnegatively related to cued recall: The more quicklythey retrieved answers to general-information ques-tions (the cue), the less likely they would later recallthe answers on the criterion test. Such negative cuediagnosticity in turn led to inaccurate ]OLs. Thegeneral point here is that when people use a cue, itsdiagnosticity will in part drive judgment accuracy.

Thus, for Benjamin er al. (1998), if retrieval fluencyon the first test had been predictive of performanceon the second (criterion) test, then JOL accuracyshould be above chance. This prediction has receivedempirical support. In particular, when the first testis a cued recall test (as in Benjamin et al., 1998) butthe second criterion test is also cued recall, then (a)JOLs are still based on the fluency of retrieval on thefirst test, but (b) now this cue of retrieval fluency ispredictive of final recall performance. As the cue isdiagnostic of criterion performance, JOL accuracyis relatively high (Serra & Dunlosky, 2005).

This example also suggests one way that theaccuracy of JOLs and FOK judgments can beimproved or fine-tuned to match levels of perfor-mance. Namely, to assess memory well, one mustevaluate it in contexts that yield diagnostic cues.For instance, when studying paired associates (e.g.,dog-spoon), two kinds oOOLs can be made. Animmediate JOL occurs immediately after an itemis studied. So one would study "dog-spoon," andthen ask, "Will I later recall spoon when shown"dog" on the test?" Immediate JOLs are not highlyaccurate because the cues available for the JOL(e.g., processing fluency) usually are not highlypredictive of criterion performance. By contrast,a delayed JOL occurs after other items have beenstudied, so when the prompt for JOL is presented(i.e., "dog-?"), one must try to recall the responsefrom memory (i.e., "spoon"). For paired associates,the outcome of this delayed retrieval attempt ishighly diagnostic of future memory performance,so delayed JOLs are highly accurate (Nelson &Dunlosky, 1991).

Another avenue to improve judgment accuracyis to provide people with experience about thediagnosticity of various cues. For instance, as notedearlier, people's JOLs do not differ for paired associ-ates studied by imagery more than for those studiedby rote repetition, even though this cue (strategyused) is highly diagnostic of performance. That is,criterion recall performance is much higher afterimagery than repetition. When people are givenexperience using these strategies during an initialstudy-test trial and then make JOLs on a secondstudy-test trial in which imagery and repetition areused, people's JOLs begin to favor imagery items(Hertzog et al., 2008). Thus, even if people's JOLsand FOK judgments are not initially accurate,they can be tuned to better reflect memory perfor-mance (for further discussion of a variety of tacticsto improve judgment accuracy, see Dunlosky &Metcalfe, 2009).

THE BASES AND ACCURACY OF OTHER

METAMEMORYJUDGMENTS

Although we have focused exclusively on JOLsand FOK judgments, the ideas described earlierpertain to any of the judgments listed in the tophalf of Figure 19.1. Each of the judgments is pre-sumably inferential in nature in that people inferthe likelihood of criterion performance based onavailable cues. Even so, they are not always empiri-cally correlated because they also differ in otherways. Concerning their intercorrelations, Nelsonand Leonesio (1988) had participants make mul-tiple judgments during a multiphase experiment-including ease-of-learning (EOL) judgments,JOLs, and FOK judgments. Within individualparticipants, the judgments were not highly cor-related: +.19 between EOL judgments and JOLs,+.12 between EOL judgments and FOKs, and +.17between JOLs and FOK judgments. One explana-tion for these small intercorrelations follows fromcue-based accounts of metamemory judgments,because the cues available can differ as people judgedifferent aspects of memory. As the cues differ, therelationship between the judgments will diverge;however, when the same cues are available for differ-ent judgments, the judgments will show a strongerrelationship.

These points can be illustrated with a simpleexperiment. Participants study 40 paired associates;20 pairs consist of related words (king-crown) and20 consist of unrelated words (turtle-bean). Beforestudying any pairs, participants are shown each pairand asked to make this ease-of-learning (EOL) judg-ment: How difficult will this item be to learn (0 =difficult to 100 = easy)?Then, each pair is presentedindividually for 6 seconds, and a JOL is made (0 =

0% likelihood of recall to 100 = 100% likelihood).Finally, a cued recall test is administered across allpairs, and for those in which a correct responsecould not be recalled, participants made a FOKjudgment for a subsequent recognition test. Noticethat some cues are similar but others differ acrossthe judgments. For EOL judgments and JOLs, therelatedness of pairs is available as a cue and it hasa major influence: Both judgments are higher forrelated than unrelated pairs. In this case, the judg-ments are highly correlated (e.g., +.71 and +.65in two different conditions; Dunlosky & Matvey,2001). By contrast, note that pair relatedness isnot available when people make FOK judgments,because these judgments are made when peoplecannot retrieve the response to a pair. Thus, the cueof pair relatedness would not be available for FOK

DUNLOSKY, THIEDE 291

j gmenLS, and hence other cues would be used toinfer the likelihood that the correct answers willbe recognized. This conrext more closely matchesthe method used by Nelson and Leonesio (1988),which yielded low correlations.

Given the usefulness of a cue-based accounr forunderstanding metamemory judgmenrs, a generalagenda for research has been to identify (a) the cuesthat are available when any particular judgmenr ismade, (b) the degree to which people use a givencue when making a judgmenr, (c) the diagnostic-ity of the available cues, and, of course, (d) thedegree to which the available cues support above-chance accuracy. Any particular research study typi-cally only explores a subset of these issues, but theenrire enrerprise has consistently led to convergingconclusions about all the metamemory judgmenrslisted in Table 19.1. In particular, when monitor-ing our memories, we do not possess the skill todirectly assess how well a memory has been storedor how accurately it can be retrieved, and hence theaccuracy of people's metamemory judgmenrs is notensured. Instead, judgmenr accuracy depends onthe available cues, the degree to which those cuesare used in constructing a metamemory judgmenr,and the diagnosticity of the cues for predicting cri-terion performance.

Control of MemoryA major focus of metamemory research con-

cerns how people conrrol their study and retrievalprocesses. In the next rwo sections, we review theextensive literature on these control processes.The main issues that we address here are whetherand how monitoring is used to make controldecisions.

HOW DO LEARNERS CONTROL THEIR STUDY?

To begin answering this question, we describe astandard method used to investigate the conrrol ofstudy. Participants are asked to learn a list of wordpairs, and during an initial trial, each pair (e.g., cat-boat) is presented at a fixed rate (e.g., 2 seconds anitem). Participanrs then make a ]OL for each item.Afterward, they have another chance to study eachpair. During this self-paced study trial, participantsmay be shown all the pairs in an array (such as in atextbook page of foreign language translation equiv-alenrs), and then they select pairs that they wanr torestudy. Alternatively, each pair may be presenredindividually, and participants study each pair aslong as they want. A key question is, Do people usetheir monitoring (as measured by ]OLs) to make

METAMEMORY

decisions about which items to select for restudy orto decide how long to study each item?

The modal (and nearly universal) finding fromthe earliest studies using this method was a negativerelationship berween ]OLs and subsequent studytime (either item selection or self-paced study).That is, as people rated items as more difficult toremember (lower ]OLs), they spem more timestudying (for reviews, see Son & Kornell, 2008; Son& Metcalfe, 2000). This relationship suggests thatpeople do use monitoring to conrrol their studytime. Nevertheless, it is correlational, and hencesome other variable may be responsible for why]OLs are related to the allocation of study time. Toevaluate whether ]OLs are causally related to alloca-tion, Metcalfe and Finn (2008) used a varianr of theaforememioned method. On an initial study trial,participants either studied pairs (cat-boat) once orthree times; a cued recall test then occurred (e.g.,cat-?) , and of course, recall was greater for pairspresented thrice than once. Next, a second studytrial occurred in which items initially presentedonce were now presented three times, whereas itemsinitially presenred three times were presenred onlyonce. After each pair was studied (either once orthrice), participanrs made a ]OL for the pair andthen made a decision about whether to restudy thepair.

On this judgmenr-selection trial, memory per-formance for the rwo sets of items was statisticallyequivalent, because all items were studied fourtimes, either being studied once during the firsttrial and three times during the second, or viceversa. Thus, when people made decisions aboutwhich items to restudy, on average, memory acrossall items was the same. However, people's ]OLswere different for the rwo sets of items: On thesecond trial, people made higher ]OLs for itemsthat were initially studied three times on the firsttrial than for those studied only once. Note that]OLs made on the second study trial are inaccu-rate here: People believe they will remember itemsstudied more time on the first trial, when in fact.all items were studied the same number of timeswhen ]OLs were made, so recall did not differfor them. Thus, if people's ]OLs drive study-timeallocation, then they should more often select torestudy those items studied once on the first trial.because people believe those items have been lesswell learned. Across three experiments, Metcalfeand Finn (2008) demonstrated that people usetheir ]OLs to control their allocation of study time(see also, Hines, Touron, & Hertzog, 2009).

Now that we have established that people douse their monitoring of learning to guide srudy, anintriguing issue becomes how it is used in allocatingstudy time. According to the discrepancy-reductionmodel, people set a goal for learning items, and thencontinue srudying each one until the goal is met.That is, learners continue srudying an item untilthe discrepancy between current learning and thegoal is completely reduced. As it takes longer toreduce this discrepancy for more difficult-to-learnitems than easier ones, people use more time srudy-ing the more difficult items. In this manner, thediscrepancy-reduction model predicts the modaloutcome in which people spend more time study-ing items given lower than higher ]OLs (Son &Metcalfe, 2000, 2005).

Although discrepancy reduction may underliesome srudy behaviors (Benjamin & Bird, 2006),people's allocarion of study time is more flexible. Wedo more than (perhaps mindlessly) continue srudy-ing items until they all meet the same preset goal. Insome situations, learners use monitoring to priori-tize the easiest items for srudy and do not even srudythe more difficult items (Son & Metcalfe, 2000;Thiede & Dunlosky, 1999), and in other siruations,they abandon memory monitoring in deciding howto allocate study time across items (Ariel, Dunlosky,& Bailey, 2009). Let's consider the former casefirst. The modal outcome (i.e., more srudy to itemsjudged as most difficult) has typically been foundwhen participants are encouraged to learn all theitems on a list. To learn them all (i.e., meet a mas-tery goal), more time would likely need to be spentsrudying the more difficult items. In some sirua-tions, however, learners do not have mastery goals:Perhaps they have little time to srudy and cannotmaster all the items, or perhaps they do not desireto master the list (e.g., students who only need topass a final exam to obtain their desired grade). Iflearners are trying to efficiently meet their goals, inboth cases, they should focus on the easiest items,because using extra effort on the more difficult oneswould be a waste of time. Consistent with such pos-sibilities, when srudy time is limited (e.g., 15 sec-onds to srudy 30 items) and when people set lowperformance goals (e.g., learn only 6 of 30 items),they spend the most time srudying a subset of theeasiest-to-learn items (Dunlosky & Thiede, 2004;Son & Metcalfe, 2000).

Note that in these cases, people are using moni-toring of learning to decide which items should beselected and prioritized for srudy. Beyond moni-toring, people also tend to focus their resrudy on

items that they expect will provide the most reward.For instance, college students more often select tosrudy (a) items that are more likely (than less likely)to appear on the criterion test and (b) items thatwill be worth more points (or reward) if correctlyrecalled on the criterion test (Ariel et al., 2009;Thiede & Dunlosky, 1999). In both cases, collegestudents prioritize the more highly valued items forstudy, regardless of whether they are difficult or easyto learn.

Thus, students often appear to attempt to maxi-mize their performance in the most efficient man-ner, which is a central premise of the agenda-basedregulation (ABR) framework of study-time alloca-tion (Ariel er al., 2009; Dunlosky, Ariel, & Thiede,2011). According to the ABR framework, peopleconstruct and execute agenda-or simple plans-for allocaring srudy time to meet these dual criteria:meeting current task goals and doing so efficiently.Although this framework highlights the importanceof agendas, factors other than agenda constructionand execution are expected to influence how (andhow well) learners allocate their srudy time. Forinstance, individual differences in working-memoryabilities can limit the quality of agenda use thatleads to the dysregulation of study time (Dunlosky& Thiede, 2004). Bottom-up processes that are trig-gered by prepotent responses to the stimulus envi-ronment also can influence srudy rime (Rhodes &Castel,2009).

The ABR framework is an example of just one ofseveral theories of people's allocation of srudy time(see also, Koriar, Ma'ayan, & Nussinson, 2006;Metcalfe, 2009; Winne & Hadwin, 1998). Thesetheories differ with respect to their explanatoryscope and are not mutually exclusive. For instance,Metcalfe's (2009) region-ofproximal-Iearning frame-work emphasizes how students in some cases pri-oririze the easiest of the unlearned items for study,because focusing on those items presumably wouldincrease the likelihood that students would effi-ciently obtain their learning goals (Kornel! &Metcalfe, 2006). In the context of the ABR frame-work, prioritizing items within this region of proxi-mal learning constitutes just one of many agendasthat people can use to allocate their time acrossitems. Given that research on self-regulated learningis largely in its infancy, new evidence will likely helpto shape these recent theories as we move towarda general theory of study-rime allocation that canboth describe how people control their srudy as wellas prescribe how they should control srudy to effi-ciently meet their goals.

DUNLOSKY, THIEDE 293

CONTROL OF OTHER FACETS OF LEARNING

Researchers have just begun exploring how stu-dents control other facets of learning, such as mak-ing decisions to mass or space their study (Son,2004; Toppino, Cohen, Davis, & Moors, 2009) orwhether they elect to test themselves during study(Karpicke, 2009; Kornell & Bjork, 2007). Decidinghow (or whether) to use these study strategies isimportant, because they can boost performance ifused properly. For instance, memory performanceis often better when study of each item is spacedthroughout a study session than when it is massedtogether; fortunately, when given the choice, sru-dents often decide to space their study (Pyc &Dunlosky, 2010). Memory performance also ben-efits from testing; in fact, testing oneself on previ-ously studied items typically is better for memorythan is restudying the same items again (Roediger& Karpicke, 2006). Unfortunately, college stu-dents tend to underuse testing as a study strategy(Kornell & Bjork, 2007). Understanding why stu-dents appropriately use some effective strategies andun de ruse others is becoming an important goal formetamemory research, partly because it has obviousimplications for improving student scholarship.

HOW DO PEOPLE CONTROL THEIR RETRIEVAL?

A great way to begin answering this questioninvolves asking another one: Which country wonthe World Cup in 2006? If you are a soccer fan, youwould be familiar with the World Cup. Thus, if youcannot recall the answer, you may spend quite a bitof time trying to retrieve it. If your first thoughtwas, "What is the World Cup?" you may not evensearch for an answer. Understanding how peoplecontrol their retrieval has been central to metamem-ory research on retrieval processes. Just like researchon the control of study, much of the research on thecontrol of retrieval has sought to understand howpeople use their monitoring to control the time. Asimplied by the earlier example, FOK judgments aswell as confidence in retrieved answers may playarole in people's decisions about how long to searchfor a response ("termination of search") and whetherto output a response that had been retrieved.

Concerning termination of search, Singer andTiede (2008) investigated whether FOK judgmentsare related to search times for general-knowledgequestions that lead to "don't know" responses. Thatis, when people ultimately cannot retrieve answersto questions, do their feelings of knowing drivethem to use more or less time while searching? Toanswer this question, some participants were used to

METAMEMORY

develop FOK norms for a set of general-knowledgequestions. These participants attempted to answereach question and then rated their confidence inrecognizing the correct answer (1 = sure I wouldnot recognize it to 9 = sure I would recognize it).An average FOK for each item was computed, andsuch norms across participants were positively cor-related with individual FOK judgments from singleindividuals. Thus, using them allowed Singer andTiede (2007) to estimate the FOK judgments of asecond group of participants who were instructedto merely retrieve the answers. The prediction wasthat the normative FOKs would be positively cor-related with retrieval latencies (i.e., the time takento retrieve an answer before giving up) when partici-pants could not retrieve an answer. Consistent withthis prediction, the correlation between FOKs andretrieval latencies for unrecalled answers was +.50.

This outcome is intuitive: When we really feelwe know an answer, we typically spend more timetrying to pull it from memory. Even when we doretrieve an answer to a question, however, it doesnot mean we will share it with others. In fact, with-holding incorrect responses is often just as impor-tant as responding correctly, such as when a witnessis admonished to "tell the truth, and nothing butthe truth." Koriar and Goldsmith (1996) proposedan influential model of memory retrieval thatexplained why (and when) people would either vol-unteer a response or withhold it (see also Higham &Arnold, 2007). Importantly, this model illustrateshow accurate monitoring is critical for the qual-ity of people's decision to output a response. Theirmodel is presented in Figure 19.3. After being askeda question ("Input Question" in Fig. 19.3), peopleattempt to retrieve candidate answers from memory.For the question "Which country won the WorldCup in 2006?" you may have retrieved UnitedStates, Brazil, and Italy. As answers are retrieved,you would evaluate their quality, such as by judg-ing your confidence in the retrieved answer from"0" (meaning there is no way it is correct) to "1.0"(meaning you are sure it is correct). Perhaps youfeel that there is only a .10 probability that UnitedStates is correct, a .70 chance that Brazil is correct,and a .40 chance that Italy is correct. In the model,these are the assessed probabilities (Pa). If you didnot retrieve any other candidates, you would choosethe best one (in this case, Brazil), but even now,you would not necessarily respond with "Brazil."Instead, you would compare your assessed probabil-ity (Pa = .70) with a response criterion (Prc). Thiscriterion can be influenced by a variety of factors,

Input question

retrieve <=>G <=> monitor

Situational - report optiondemands / payoffs - accuracy incentive

Figure 19.3 Model of self-regulated retrieval that explains whether someone will report an answer that is retrieved when asked a giveninput question. See text for details. (From Koriat & Goldsmith, 1996.)

such as the current set of payoffs for responses. Forinstance, you may be taking a test over sports triviawhere the reward for a correct response is high andwhere the penalty for an incorrect response is small.In this case, your response criterion would likely below. By contrast, if you are on the witness stand andwere told not to report any incorrect evidence, youmay set a very high response criterion. Let's assumethat you set a Pre at .40, and hence you would vol-unteer the response "Brazil," because your assessedprobability exceeds your response criterion; that is,Pa (.70) > Pre (.40).

Based on this example, it should be clear whymonitoring accuracy is critical for the quality (andquantity) of what responses people will volunteer.Concerning accuracy, if your assessed probabili-ties do not reflect the actual probabilities of beingcorrect (i.e., poor judgment accuracy), then incor-rect answers could be volunteered. This mistake isillustrated in the earlier example: Italy is the cor-rect answer, so the assessed probabilities were notaccurate. Thus, your overconfidence in Brazil'steam led you to respond incorrectly. If instead, Italyhad a much higher Pa (e.g., close to 1.0), then theenhanced judgment accuracy would have led to thecorrect response.

To evaluate this implication of the model, Koriatand Goldsmith (1996) used the following method.Participants were forced to answer general-knowledgequestions so that the experimenters would knowwhat each participant's best candidate answer wasfor every question; for each answer, participantsprovided a confidence judgment, which was used as

a measure of Pa. Next, the participants answered thequestions again, but this time, they could withholdanswers-that is, they were not forced ro respond.As important, two different sets of questions wereused. Some were considered standard questions andsupported high levels of judgment accuracy andsome were deceptive and led to inaccurate judg-ments. The latter consisted of questions that peopleoften answer incorrectly with high confidence; forinstance, "What is the capital of Australia?" is oftenconfidently answered with "Syndey" when the cor-rect answer is "Canberra." During the first round ofquestions, performance was greater for the standardquestions (28%) than for the deceptive ones (12%).This outcome is not surprising, given that it merelydemonstrates that the deceptive questions are infact deceptive. The more important outcomes con-cern performance for the free-report phase. If par-ticipants use their confidence judgments to makedecisions about which answers to withhold, theirperformance should increase for standard questionsbur may not change for deceptive ones. As expected,the percentage of accurate recall for the free-reportphase was 75% for the standard questions, but it wasonly 21% for the deceptive questions. According tothe model, people had difficulties with the deceptivequestions because they would have failed to with-hold incorrect responses that were inadvertentlyheld with high confidence.

How long one persists in trying to retrieve anitem and whether one elects to respond are not theonly facets of retrieval that are under people's con-trol. Others include (a) whether to even attempt to

DUNLOSKY, THIEDE 295

reuieve an answer or to use some other means toobtain one and (b) decisions abour the grain size ofinformation to report. For the former, Reder andRiner (1992) investigated factors that influencepeople's decision to retrieve an answer to a difficultmath problem or to compute it. People attemptedretrieval when they believed problems were famil-iar (even when they were actually not familiarwith them) and elected to compure answers whenfamiliarity was low. For the latter, the instructionto "speak the truth and nothing but the truth" mayinfluence the grain size of reports; in this case, peoplewill tend to give more general reports that are morelikely to be true and withhold details that have ahigher chance of not being accurate (cf. Goldsmith,Koriar, & Weinberg-Eliezer, 2002). According toAckerman and Goldsmith (2008), people controlthe grain size of their reporrs (either by providingmore precise or more coarse answers) in order to beboth informative and confident in their responses.Although we have much to learn about how peoplecontrol their retrieval and whether such controlcan be optimal (for discussions of oprimaliry, seeHigham, 2007), one conclusion is certain: Peopledo use their monitoring to make decisions abouthow long to search for answers and whether to vol-unteer responses. Thus, as with the control of study,accurate monitoring is essential for effective self-regulation.

ConclusionIn the present chapter, we provided an introduc-

tion to many of the common measures, methods,and issues that drive research on metamemory. Somemajor conclusions included that people's monitoringjudgments can be accurate, but only when the cuesthat are used to construct the judgments are diag-nostic of criterion performance. As cue diagnosticityincreases, then the upper limit on judgment accu-racy also increases. Regardless of judgment accuracy,monitoring is also used in the service of controllingmemory, whether it be in deciding which items toprioritize for study or when to volunteer (or with-hold) a response during retrieval. Our overview,however, did not cover the entire field, which hassought to answer many intriguing questions aboutmetamemory. How does metamernory develop inchildhood and does it decline as we age? Do animalshave metamemory-cio they know when they knowand when they do not? How do individual differ-ences in metamemory influence educational our-comes? What are the neurological underpinningsof metamernory, and what kinds of brain insult

METAMEMORY

and psychological disorders lead to disruptionsof metamemory? For those interested in furtherexploring metamemory, answers to these and otherquestions are available in a variety of more extensivesources (e.g., Dunlosky & Metcalfe, 2009; Hacker,Dunlosky, & Graesser, 2009; Higham & Leboe,2011; Terrace & Metcalfe, 2005).

ReferencesAlter, A. L., & Oppenheimer, D. M. (2009). Uniting the tribes

of Ruency ro form a metacognitive nation. Personality andSocial Psychology Review, 13, 219-235.

Ackerman, R., & Goldsmith, M. (2008). Control over grain sizein memory reporting: With and without satisficing knowl-edge. jou77U11of Experimental Psychology: Learning, Memory,and Cognition, 34, 1224-1245.

Arbuckle, T. Y., & Cuddy, L. L. (1969). Discrimination of itemstrength at time of presentation. journal of ExperimentalPsychology, 81, 126-131.

Ariel, R., Dunlosky, ]., & Bailey, H. (2009). Agenda-basedregulation of study-time allocation: When agendas overrideitem-based moniroring. journal of Experimental Psychology:General, 138,432-447.

Benjamin, A. S. (2005). Response speeding mediates the contri-butions of cue familiarity and target rerrievabiliry ro metam-nemonic judgments. Psychonomic Bulletin and Review, 12,874-879.

Benjamin, A. 5., & Bird, R. D. (2006). Meracognirive controlof the spacing of study repetitions. journal of Memory andLanguage, 55, 126-137.

Benjamin, A. 5., Bjork, R. A., & Schwartz, B. L. (1998). Themismeasure of memory: When retrieval fluency is misleadingas a metamnemonic index.journal of Experimental Psychology:General, 127, 55-68.

Brunswik, E. (1956). Perception and representative design of psy-chological experiments. Berkeley: University of CaliforniaPress.

Carroll, M., Nelson, T. 0., & Kirwan, A. (1997). Tradeoff ofsemantic relatedness and degree of overlearning: Differentialeffects on metamemory and on long-term retention. ActaPsycbologica, 95, 239-253.

Castel, A. D. (2008). Meracognition and learning about primacyand recency effects in free recall: The utilization of intrin-sic and extrinsic cues when making judgments of learning.Memory and Cognition, 36,429-437.

Dunlosky, ]., Ariel, R., & Thiede, K. W. (2011). Agenda-basedregulation of study-time allocation. In P. Higham & ]. P.Leboe (Eds.), Constructions of remembering and metacognition(pp, 182-200). Hampshire, England: Palgrave Macmillan.

Dunlosky, ]., & Bjork, R. A. (Eds.). (2008). A handbook ofmetamemory and memory. New York: Psychology Press.

Dunlosky, ]., & Marvey, G. (2001). Empirical analysis of theintrinsic-extrinsic distinction of judgments of learningQOLs): Effects of relatedness and serial position on ]OLs.journal of Experimental Psychology: Learning, Memory, andCognition, 27, 1180-1191.

Dunlosky, ]., & Metcalfe,]. (2009). Metacognition. Los Angeles:Sage.

Dunlosky, j., Serra, M. ]., & Baker, ]. M. C. (2007).Metamemory. In F. T. Durso, Nickerson, R. 5., Dumais,S. T., Lewandowsky, 5., & Perfect, T. j. (Eds.), Handbook

of applied cognition (2nd ed., pp. 137-160). Hoboken, N]:Wiley.

Dunlosky, ]., & Thiede, K. W (2004). Causes and constraintsof the shift-to-easier-materials effect in the control of study.Memory and Cognition, 32, 779-788.

Flavell,]. H. (1979). Metacognition and cognitive monitoring:A new area of cognitive-developmental inquiry. AmericanPsychologist,34, 906-9 J l.

Goldsmith, M., Koriat, A., & Weinberg-Eliezer, A. (2002).Strategic regulation of grain size memory reporting. Journalof Experimental Psychology:General, 131,73-95.

Hacker, D. j., Dunlosky, j., & Graesser, A. C. (Eds.). (2009).Handbook of metacognition and self regulated learning.Mahwah, N]: Erlbaum.

Hart,]. T. (1965). Memory and the feeling-of-knowing experi-ence. Journal of Educational Psychology,56, 208-216.

Hart, J. T. (1966). Methodological note on feeling-of-knowingexperiments. Journal of Educational Psychology, 57,347-349.

Hertzog, C, Dunlosky, J., Robinson, A. E., & Kidder, D.p. (2003). Encoding fluency is a cue used for judgmentsabout learning. Journal of Experimental Psychology:Learning,Memory, and Cognition, 29, 22-34.

Hertzog, c., Price, ]., & Dunlosky,]. (2008). How is knowledgegenerated about memory encoding strategy effectiveness?Learning and Individual Di/firences, 18,430--445.

Higham, P. A. & Leboe, ]. P. (Ed.). (2011). Constructions ofremembering and metacognition. Hampshire, England:Palgrave Macmillan.

Higham, P. A. (2007). No Special K! A signal detection frame-work for the strategic regulation of memory accuracy. Journalof Experimental Psychology:General, 136, 1-22.

Higham, P.A., & Arnold, M. M. (2007). How many questionsshould I answer? Using bias profiles to estimate optimalbias and maximum score on formula-scored tests. EuropeanJournal of Cognitive Psychology,19,718-742.

Hines.]. C;Touron, D. R., & Hertzog, C. (2009). Metacognitiveinfluences on study time allocation in an associative recogni-tion task: An analysis of adult age differences. PsychologyandAging,24,462--475.

Humphrey, G. (1951). Thinking: an introduction to its experimen-tal psychology. New York: Wiley.

Karpicke,]. D. (2009). Metacognitive control and strategy selec-tion: Deciding to practice retrieval during learning. [ournal ofExperimental Psychology:General, 138, 469--486.

Kolers, P.A., & Palef, S. R. (1976). Knowing not. Memory andCognition, 4, 553-558.

Koriat, A. (1993). How do we know that we know) The acces-sibility model of the feeling of knowing. PsychologicalReview,100,609-639.

Koriat, A. (1997). Moniroring one's own knowledge duringstudy: A cue-utilization approach to judgments of learn-ing. Journal of Experimental Psychology: General, 126,349-370.

Koriat, A., Bjork, R. A., Sheffer, L., & Bar, S. K. (2004).Predicting one's own forgetting: The role of experience-basedand theory-based processes. Journal of ExperimentalPsychology:General, 133,643-656

Koriat, A., & Goldsmith, M. (1996). Monitoring and controlprocesses in the strategic regulation of memory accuracy.PsychologicalReview, 103,490-517.

Koriat, A., & Levy-Sador, R. (2001). The combined contri-butions of the cue-familiarity and accessibility heuristics

to feelings of knowing. Journal of Experimental Psychology:Learning, Memory, and Cognition, 27, 34--53.

Koriar, A., Ma'ayan, H., & Nussinson, R. (2006). The intricarerelationships between monitoting and control in rneracog-nition: Lessons for the cause-and-effect relation betweensubjective experience and behavior. Journal of ExperimentalPsychology:General, 135,36-69.

Koriat, A., Nussinson, R., Bless, H., & Shaked, . (200 ).Information-based and experience-based metacognitive judg-ments: Evidence from subjective confidence. In]. Dunlosky& R. A. Bjork (Eds.), Handbook of metamemory and memory(pp. 117-135). New York: Psychology Press.

Kornell, N., & Bjork, R. A. (2007). The promise and perils ofself regulated study. Psychonomic Bulletin and Review, 14,219-224.

Kornell, N., & Bjork, R. A. (2008). Optimizing self-regulatedstudy: The benefits-and costs-of dropping flashcards.Memory, 16, 125-136.

Kornell, N., & Metcalfe, J. (2006). Study efficacy and the regionof proximal learning framework. Journal of ExperimentalPsychology:Learning, Memory, and Cognition, 32, 609-622.

Loftus, E., & Ketchum, K. (1991). Witness fin the defense: Theaccused, the eyewitness, and the expert who puts memory ontrial. New York: St. Martin's Press.

Marvey, G., Dunlosky, ]., & Guttentag, R. (2001). Fluencyof retrieval at study affects judgments of learning (JOLs):An anlaytic or nonanalytic basis for ]OLs? Memory andCognition, 29, 222-232.

Mazzoni, G., Cornoldi, c., & Marchitelli, G. (1990). Do mem-orability ratings affect study-time allocation? Memory andCognition, 18, 196-204.

Metcalfe,]. (2002). Is study time allocated selectively to a regionof proximal learning? Journal of Experimental Psychology:General, 131,349-363.

Metcalfe, ]. (2009). Metacognitive judgments and controlof study. Current Directions in Psychological Science, 18,159-163.

Metcalfe, ]., & Finn, B. (2008). Evidence that judgments oflearning are causally related to study choice. PsychonomicBulletin and Review, 15, 174--179.

Metcalfe, ]., Schwartz, B. L., & Joaquim, S. G. (1993). Thecue-familiarity heuristic in metacognition. Journal ofExperimental Psychology: Learning, Memory, and Cognition,19,851-86l.

Miner, A. C; & Reder, L. M. (1994). A new look at feeling ofknowing: Its metacognitive role in regulating question answer-ing. In J. Metcalfe & A. Shirnamura (Eds.), Metacognition:Knowing about knowing (pp. 47-70). Cambridge, MA: MITPress.

Nelson, T. 0., & Dunlosky,]. (1991). When people's judgmentsoflearning (JOLs) are extremely accurate at predicting subse-quent recall: The "delayed-]OL effect." PsychologicalScience,2,267-270.

Nelson, T. 0., & Narens, L. (1990). Metamemory: A theoreti-cal framework and new findings. In G. H. Bower (Ed.), Thepsychology of learning and motivation (Vol. 26, pp. 125-141).New York: Academic Press.

elson, T. 0., Gerler, D., & Narens, L. (1984). Accuracy of feel-ing of knowing judgments for predicting perceptual identi-fication and relearning. Journal of Experimental Psychology:General, 113,282-300.

Nelson, T. 0., & Leonesio, R.]. (1988). Allocation of self-pacedstudy time and the "labor-in-vain effect." Journal of

DUNLOSKY, THIEDE 297

Experimental Psychology: Learning, Memory, and Cognition,14,676-686.

Pyc, M. A. & Dunlosky, J. (2010). Toward an understandingof students' allocation of study time; Why do they decideto mass or space their practice? Memory and Cognition, 38,431-440.

Rabinowitz, J. c. Ackerman, B. P., Craik, F. l. M., & Hinchley,J. L. (1982). Aging and metamemory; The roles of related-ness and imagery. journal of Gerontology, 37, 688-695.

Reder, L. M., & Ritter, F. E. (1992). What determines initialfeeling of knowing? Familiarity with question terms, notwith the answer. journal of Experimental Psychology: Learning,Memory, and Cognition, 18, 435-451.

Rhodes, M. G., & Castel, A. D. (2009). Mecacognitive illusionsfor auditory information; Effects on monitoring and control.Psychonomic Bulletin and Review, 16,550-554.

Roediger, H. L., III, & Karpicke, J. D. (2006). The power of test-ing memory; Basic research and implications for educationalpractice. Perspectives on Psychological Science, 1,181-210.

Serra, M. J., & Dunlosky, J. (2005). Does retrieval fluency con-tribute to the underconlidence-with-practice effect? journalof Experimental Psychology: Learning, Memory, and Cognition,31, 1258-1266.

Shaughnessy, J. J. (1981). Memory monitoring accuracy andmodification of rehearsal strategies. journal ofVerbal Learningand Verbal Behavior, 20, 216--230.

Singer, M., & Tiede, H. L. (2008). Feeling of knowing and dura-tion of unsuccessful memory search. Memory and Cognition,36, 588-597.

Son, L. K (2004). Spacing one's study; Evidence for a metacog-nitive con trol strategy. journal of Experimental Psychology:Learning, Memory, and Cognition, 30, 601--604.

~9 METAMEMORY

Son, L. K., & Kornell, N. (2008). Research on the allocationof study time; Key studies from 1890 ro the present (andbeyond). In J. Dunlosky & R. A. Bjork (Eds.), Handbookof metamemory and memory (pp, 333-351). New York:Psychology Press.

Son, L., & Metcalfe, J. (2005). Judgments of learning; Evidencefor a two-stage process. Memory and Cognition, 33,1116--1129.

Son, L. K, & Metcalfe, J. (2000). Metacognitive and controlstrategies in study-time allocation. journal of ExperimentalPsychology: Learning, Memory, and Cognition, 26, 204-221.

Souchay, C; Moulin, C. J. A., Clarys, D., Taconnat, L., &Isingrini, M. (2007). Diminished episodic memoty aware-ness in older adults; Evidence from feeling-of-knowing andrecollection. Consciousness and Cognition, 16, 769-784.

Terrace, H. 5., & Metcalfe, J. (Eds.). (2005). The missing linkin cognition: Origins of self-reflective consciousness. New York:Oxford University Press.

Thiede, K W, & Dunlosky,J. (1999). Toward a general model ofself-regulated study; An analysis of selection of items for studyand self-paced study time. journal of Experimental Psychology:Learning, Memory, and Cognition, 25, 1024-1037.

Toppino, T. c., Cohen, M.S., Davis, M. L., & Moors, A. C.(2009). Metacognitive control over the distribution of prac-tice; When is spacing preferred? journal of ExperimentalPsychology: Learning, Memory, and Cognition, 5, 1352-1358.

Winne, P.H., & Hadwin, A. F. (1998). Studying as self-regulatedlearning. In D. Hacker, J. Dunlosky, & A. C. Graesser(Eds.), Metacognition in educational theory and practice (pp.277-304). Hillsdale, NJ; Erlbaum.

Woodworth, R. S. (1921). Psychology: A study of mentallife. NewYork; Henry Holt.