when knowing more means less the effect of fan

8/12/2019 When Knowing More Means Less the Effect of Fan

1/8

Journal of Experimental Psychology:Learning, Memory, and Cognition2001, Vol. 27, No. 5, 1172-1179Copyright 2001 by the American Psychological Association, Inc.O278-7393/O1/S5.00 DO I: 10.1037//02 78-7393.27.5.1172

When Knowing More Means Less: The Effect of Fanon Metamemory JudgmentsMichael J. McGuire and Ruth H. MakiTexas Tech University

The authors used a fan paradigm (J. R. Anderson , 1974) to test the accessibility and competitionmodels of metamemory using judgments of learning (JOLs). JOLs in this study reflect one's confidencelevel in subsequen tly recognizing newly learned material. The number of facts, or fan, associated withJOL-qu eried conce pts varied from 1 to 3 associates. Results of 3 experimen ts indicated that as the levelof fan increased, the magnitude of JOLs decreased. This finding was observed even when the fan effect(i.e., slower recognition as number of facts increase) was attenuated on a verification task in 2 of theexperiments by manipulating the organization of the multiple concepts. The results supported thecompetition hypothesis (T. A. Schreiber, 1998; T. A. Schreiber & D. L. Nelson, 1998) as an importantdeterminant of JO Ls.

Will more information learned about concepts increase or de-crease the magnitude of metamemory prediction judgments forsuch concepts? We tested these two opposite predictions derivedfrom the accessibility (Koriat, 1993, 1995) and competition(Schreiber, 1998; Schreiber & Nelson, 1998) hypotheses formetamemory judgments. In an associative interference, or fanparadigm (Anderson, 1974), our participants learned facts contain-ing objects and locations. The num ber of facts learned about targetconcepts varied from one to three, which defined the magnitude offan.Koriat's (1993, 1995) accessibility hypothesis predicts higherjudgments when more facts can be retrieved. In the 1993 study,participants who could produce more letters of a letter-string

response gave higher judgments than those who could producefewer letters. Participants in Koriat's (1995) study gave higherjudgments to general knowledge questions that produced more

Michael J. McGuire and Ruth H. Maki, Department of Psychology,Texas Tech University.This research represents a major portion of Michael J. McGuire'sdissertation at Texas Tech University. The findings for Experiment werepresented at the meeting of the Psychonomic Society, Dallas, Texas,November 1998, and the findings for Experiment 2 were presented at themeeting of the Psychonomic S ociety, Los An geles, California, N ovember1999. Findings for Experiment 3 were presented at the meeting of thePsychonomic Society, New Orleans, Louisiana, November 2000.We thank G. A. Radvansky for an ongoing correspondence dealing withsituation models and the fan effect as well as comments pertaining tovarious points in this article. We thank Asher Koriat, Janet Metcalfe, andtwo anonymous reviewers for their helpful comments on this article.Special thanks go to undergraduate research assistants Kate Haskew,Joseph Johnson, and Laura Powell for testing participants in Experiment 1,and to Jenni Brumelle, Jill Bulgeron, and Keith Pennick for testing partic-ipants in Experiment 3.Correspondence concerning this article should be addressed to eitherMichael J. McGuire, who is now at the Department of Psychology,Northern Michigan University, Marquette, Michigan 49855, orRuth H. Maki, Department of Psychology, Texas Tech University, Lab-

bock, Texas 7940 9-2051. Electronic mail may be sent to [email protected] or [email protected].

answers than they did to questions that produced fewer answers.Thus, an extension of the accessibility hypothesis predicts highermetamemory judgments on concepts for which more informationhas been learned and can be retrieved, that is, judgments willincrease with fan.Conversely, an extension of the competition hypothesis(Schreiber, 1998; Schreiber & Nelson, 1998) predicts lowermetamemory judgments for targets associated with more facts.Using an associative cuing paradigm, Schreiber and Nelson (1998)found that metamemory judgments were higher for cues with asmaller number of associates in semantic memory than for cueswith a larger number of associates. Schreiber (1998) found evi-dence for competition in similar experiments when he varied theassociative set size for the to-be-remembered targets. Maki (1999)used retroactive interference paradigms to investigate the c ompe-tition versus accessibility hypotheses for newly learned material.She found that stimuli paired with two responses produced lowermetamemory judgments than stimuli paired with only one re-sponse. The same was true for responses; pairs in which theresponse was paired with two stimuli were given lower judgmentsthan pairs in which the response was paired with only one stimu-

lus.Thus, M aki's results, like those of Schreiber (1998; Schreiber& Nelson, 1998), supported competition rather than accessibilityas the basis for metamemory judgments.The accessibility hypothesis was developed primarily forfeeling-of-knowing (FOK) judgments, in which participants pre-

dicted recognition for nonrecalled items. However, Koriat (1995)also had participants make predictions for all items before recall,that is, make judgments similar to what Nelson and Narens (1990)called judgments of learning (JOLs). Koriat found that both JO Lsand FOKs w ere higher when more information could be retrieved.Schreiber (1998; Schreiber & Nelson, 1998) also used both FOKsand JOLs and found the opposite, that targets with more preex-perimental associates resulted in lower judgments.We investigated the accessibility and competition hypothesesusing JOLs in an associative interference (fan) paradigm modeledafter the studies of Anderson (1974) and Radvansky, Spieler, andZacks (1993). Participants learned facts containing object andlocation concepts (e.g.,The exit sign is in the airport).The number1172


2/8

EFFECTOFFANON METAMEMORY 1173of facts learned about targets varied from one to three, whichdefined the fan. For example, if the target concept (in this example,exit sign)was a Fan Level 3 concept, then it would be associatedwith three sentences containing three different locations (e.g.,airport, lounge, library).Therefore, target concepts (e.g.,exit sign)varied in the amount of information learned about them. Once allof the facts were learned to a criterion, participants were presentedwith target concepts that varied by fan. The participants then madeJOLs by predicting how well they would correctly recognizestudied facts containing the target concepts on a later recognitiontask. The accessibility hypothesis predicts higher JOLs for Fan 3concepts relative to Fan 1 and 2 concepts because more informa-tion is accessible for Fan 3 concepts. The competition hypothesispredicts higher JOLs for Fan 1 concepts relative to Fan 2 and 3concepts because there are fewer sources of potential interferencefor Fan 1 concepts.

General MethodParticipants

All participants were recruited fromtheTexas Tech University Gene ralPsychology participant pool. They received creditfortheir course require-mentsinGeneral Psychologyinexchangefor their participation.Materials

The stimuli were modeled after Radvansky et al.'s(1993) materialsinwhich sentences consisted of object-location pairings.Thestructureforeach sentence was,The [object] is in the [location]. For example, aparticipant might see The banner is in the gym.Participants were presentedwith24 sentences to learn. Sentences variedby fan, or thenumberoffacts (from one to three) participants learned about the target concepts(hereafter referred to asJOL-queried concepts). Fan mayalsobeinter-pretedas thenumberof sentencesinwhicha given JOL-queried conceptappeared. Example sentences varyingby fan can befoundinTable 1 (thelocation manipulation willbe discussed inExperiment 2). Forexample,accordingtoTable 1 theexit sign is a Fan 3 concept, whereasthe ceilingfan is a Fan 1concept.

Learning phase. Participants were instructed to learn sentences asefficiently aspossible. The experiments w ere conductedonIBM-PC com-patible computers, using Micro Experimental Laboratory (MEL) (Ver-sion 2.0) software (Schneider, 1995). On acomputer monitor, 24 randomlyordered sentences were displayed for 7 s each. After all sentences werepresented, participants were givenacued-recall test in which they typed thesingle correct answerto questions suchasWhatis located intheairport?Participants received feedback after typingin their response. Once partic-ipants correctly answered every question in ablockof 24questions, they

advancedto theprediction pha se. Otherwise , participants were presentedwiththesentencesfor study again, followedby thecued-recall task.

Prediction JOL)phase. After learning the sentences, participantsmade predictions for JOL-queried concepts. For example, participantswould see thefollowing:Howconfidentare you that you will recognizestudied fact(s) involving the exit sign?Under the query wasa7-point ratingscale with 1 representing not at all confident and 7 representing veryconfident.Verificationphase. For the final phase, participants differentiated stud-ied from nonstudied sentences. They were instructed to respond as quicklyas possible while remaining accurate. Before the test trials, participantswere given 10practice trialsto become familiar with theprocedure.Foreachofthe practice trials,aswellasthe test trials, participants p laced theirforefingers on the Z andforward slash k eys, respectively. Practicesen-tences were This sentence is studied or This sentence is notstudied.Pressing the Z key meant thatthe sentence was not a studied sentence,whereas pressing the forward slash key meant that the sentencewasstudied.

Experiment1The primary purpose of Experiment 1 was to assess whether fanaffected JOLs. If JOLs for higher fan concepts resulted in higher

predictions, then the accessibility hypothesis would be supported.The competition hypothesis would be supported if the resultsshowed lower JOLs for higher fan concepts due to increasedcompeting activation for such concepts. We attempted to reducethe effect of fan by enhancing the integration of studied facts. Tothis end, we introduced an instructional manipulation emphasizingimagery. We expected that participants using imagery would beable to integrate information more efficiently than participants notusing imagery. Therefore, we expected that participants in theimagery groups would show a reduced fan effect in reaction times(RTs) because of less interference with integrated materials. Ifmetamemory is equally sensitive to competition, then the fan effectwould also be reduced for JOLs.

MethodParticipants. Atotalof66 participants were randomly assigned to one

of three groups receivingadifferent setofinstructions (no imagery, simpleimagery,andinteractive im agery), resultingin three groupsof22 partici-pants each.

Design. The experimental design formed a 3 X 3 mixed-model facto-rial with imagery instructions (no imagery, simple imagery, and interactiveimagery) manipulated between subjectsand fan (1-3) manipulated withinsubject.The dependent measures included: (a)proportionoftrials correct

Table 1Exemplar Sentences Participants Studied in Fan Experiments

LocationFan Multiple Single

Th e ceilingfanis in thebarbershop.Th e coke machine is in theclubspa.Th e coke machineis in theschool.Th e exit sign is in theairport.The exit signis in thelounge.Th e exit sign is in thelibrary.

Th ebigdeskis in theoffice.Th epayphoneis in thehospital.The file cabinetis in thehospital.The marble benchis in thehotel.Th eoakdeskis in thehotel.The revolving dooris in thehotel.Note. Italicized words representthe fanconceptsforwhich participants made judgmentsoflearning.


3/8

1174 McGUIRE AND MAKITable2Basic Schemata of Word Pairings for the Design of Experiments

Condition

Multiple locations

Single locationMultiple locations

Fan 1

O , - L ,O 2 -L 2O 3 -L 3O4-L4

O,-L,O 2-L 2Ou -L,

O 5-L 5O 8 - L '

Fan 2Experiment1

O 5-L 6O 6 -L 81 Og-L 12

Experiments 2 and3O 3 -L 3 O4 -L 3O 5 -L 4 O6 -L 4

Ojj-Lp Oi5-L o

O9-L13O 1 0 -L 1 6o 2-L22

O 7-L 50 1 0 - L 6O , 7 -L 1 3

Fan 3

O 9 -L 1 4o l L o0 2- 23

O 8-L 50 , , - L ,O 1 7-L 1 4O 1 8 -L I 7

O 9-L 1 5

O 9-L 50,2-LsO 1 7 -L 1 5O ,8 -L 1 8

Note. O=object concept; L= location concept. Subscripts refer to specific objects and locations. Whenasubscript is repeated, that means that the same concept is repeated as well. For nonstudied sentences, eachconcept was re-paired with an alternative associate from the same cell. For example, Object would be re-pairedwith one of the three locations within its cell (i.e., Location 2, 3, or 4).

during the learning phase, (b) JOLs, and (c) RT and error rates on theverification task.

Materials. The 24 sentences to be remembered were composed of 12objects (the same used in Radvansky et al., 1993) and 24 locations 1 (halfof which came from Radvansky et al.). The top half of Table 2 gives anillustration of how objects and locations were paired for the sentences.

Thirty different sentences were used in the verification task. FollowingAnderson's (1974) rationale, we did not include all studied sentences sothat Fan 3 sentences were not presented more often than Fan 1 sentencesduring the verification phase. Nine critical studied sentences were pre-sented. Another 9 critical nonstudied sentences consisted of novel combi-nations of the objects and locations from the studied sentences. The criticalsentences were each tested twice, giving 36 trials for which data wereanalyzed. So that participants could keep all the material active in memory,we included 12 additional filler sentences using the concepts not includedin the critical set. Half of these filler sentences were studied and half werenonstudied. Verification times for the filler sentences were not analyzed.

Procedure. The main difference between procedures in this experi-ment and those of the general method section was an instructional manip-ulation presented atthe beginningofthe learning phas e. The verbalinstructions varied according to which group participants were randomlyassigned. All participants were instructed to mem orize the materials asefficiently as possible. Simple and interactive imagery groups were alsoinstructed to form distinct mental image s of the conce pts. In addition, theinteractive image ry group was instructed to form mental images with theto-be-remembered concepts actively interacting with one another... al-lowing sentence objects to act as pegs upon which you can attach loca-tions. Several specific exam ples of such interactive imagery were given inboth imagery conditions.

Results and DiscussionLearning phase. We first examined blocks to criterion, that is,how many study-test cycles were necessary before the 24 ques-tions were answered correctly on a single test. The mean numberof blocks to reach criterion for the no imagery, simple imagery,and interactive imagery groups was 4.68, 4.05, and 3.86 blocks,respectively(overall, participants took an average of 4.2 0 blocks,SD =1.60, to reach criterion). An analysis of variance (ANOV A)showed no significant differences among imagery groups, F(2,63) = 1.07, MSE = 2.80,p > .05 (the level of significance usedthroughout this article unless noted otherwise).

Next, we calculated the proportion of trials on which each objectwas correctly recalled on the cued-recall tests. Mean proportioncorrect for Fan 1, Fan 2, and Fan3conditions is displayed inTable 3.A 3 (imagery)X 3 (fan) mixed ANOVA showedasignificant main effect for fan, F(2, 126)=3.27,M SE =0.008.Post hoc pairwise comparisons with a Tu key correction for TypeIerror indicated only that Fan 3 sentences w ere correct more oftenthan Fan 2 sentences, F(l, 63)=7.63,MSE = 0.007. The othercomparisons resulted in values less than the Tukey critical Fvalue of5.78:Fan 1 versus Fan 2, F(l , 63)=2.08,MS E = 0 .001;Fan 1 versus Fan 3,


4/8

EFFECT OF FAN ON METAMEMORY 1175Table 3Mean Proportion Correct (and Standard Deviation) DuringLearning for Experiments 1, 2, and 3 by Location and F an

Experiment123

LocationMultipleSingleMultipleSingleMultiple

1.76 (.12).77 (.22).78 (.20).78 (.19).80 (.17)

Fa n2

.74 (.11).77 (.14).77 (.16).77 (.18).77 (.16)

3.78 (.10).78 (.12).78 (.13).77 (.15).80 (.13)

pairwise comparisons using a Tukey correction (F T = 5.78) indi-cated that Fan RTs were faster than Fan 3 RTs, F( l, 63) = 12.08,M SE = 157,794.82. Fan RTs did not differ from Fan 2 RTs, F (l ,63) = 5.51,M SE = 149,365.56, and RTs between Fan 2 and Fan 3did not differ, F( l, 63) = 1.38, M SE = 162,972 .17. Effect ofstudied-nonstudied and imagery were nonsignificant (F


5/8

1176 McGUIRE AND MAKJmodels necessary in the ML condition (for differing interpretationsof fan effects, see Anderson & Reder, 1999a, 1999b; Radvansky,1999a, 1999b).

As shown in Table 1, three objects can be in one location, in thiscase the hotel, at the same time. Thus, in the SL condition, onemental representation can accommodate one, two, or three objects,so a single representation would be activated for Fan 1, Fan 2, andFan 3 concepts. However, in the ML c ondition, it is unrealistic tohave the same object, like an exit sign, in three locations at thesame time. Thus, in the ML, Fan 3 condition, three differentmental representations, one for each object-location pairing,would be created. All three representations would need to beactivated in order to verify Fan 3 sentences. This would contrastwith the Fan 1 and Fan 2 conditions, in which only one or tworepresentations would need to be activated for verification. Thus,a greater fan effect should be seen in the ML than in the SLcondition. The question of primary interest to us was whether asimilar reduction in the fan effect would occur in JOLs.Method

The design was a 2 (location) X 3 (fan) within-subject design. Thematerials consisted of sentences2 that adhered to the same format as inExperiment 1, namely, The [object] is in the [location]. Four differentobject-location lists were used across participants. As shown in the bottomof Table 2, half of the sentences in each list were SL sentences and theother half were ML sentences. A total of 20 participants was randomlyassigned to the four lists, with 5 participants per list.

The procedure was identical to that used in Experiment with two majorexceptions. First, we included the within-subject materials manipulationreferred to as location (SL vs. ML). Second, three more blocks of verifi-cation trials were conducted, so that there were 192 trials for each fan-by-location condition. The increase in blocks of trials is consistent withRadvansky et al. 's (1993) procedure.

Results and DiscussionLearning phase. On average, participants took 3.65 blocks

(SD = 1.14) to reach criterion. Mean proportions correct, asdisplayed in Table 3, on the cued-recall test during learning did notvary by fan or by location. A 2 (location) X 3 (fan) repeatedmeasures ANOVA was conducted to evaluate the effects of fanand location on learning. No significant effects were obtained(Fs < 1 for fan, location, and the Fan X Loca tion interaction).JOL magnitude. Figure 2 displays the median ratings of par-

7.00

6.756.506.256.005.75

1 2 3Fan

Figure 2. Judgment of learning (JOL) medians as a function of fan andlocation for Experiment 2.

ticipants' JOLs for each level of fan. These data were entered intoa 2 (location) X 3 (fan) repeated measures ANOVA yielding thefollowing results: fan was significant, F(2, 38) = 6.45,MSE =0.504; both the main effect of location and the Location XFan interaction were nonsignificant (both Fs < 1).

Verification task performance. Figure 3 displays the meanRTs for verifying whether a sentence had been studied for SL andML sentences. These data were analyzed in a 2 (studied vs.nonstudied) X 2 (location) X 3 (fan) mixed model ANOV A. Meanresponse latency for studied sentences (1,542 ms) was faster thanthat for nonstudied sentences (1,635 ms), F(l, 19) = 11.89,MSE =43,720.11.Overall, participants responded more quickly toSL condition sentences (1,550 ms) than to ML condition sentences(1,627 m s), F( l, 19) = 10.98,MSE =32,496.19. As fan increased,mean response latency increased as well (1,479 ms, 1,610ms,1,678 ms for Fans 1, 2, and 3, respectively), F(2, 32) = 13.85,MSE = 59,128.64.

The three-way interaction was nonsignificant ( F < 1). However,the critical Location X Fan interaction that can be seen in Figure4 was significant, F(2 , 38) = 4.99,MSE = 72,716.54. A follow-upone-way repeated measures ANO VA for the ML condition yieldeda significant effect for fan, F(2, 38) = 17.06, M SE = 30,813.65.Pairwise comparisons with a Tukey correction (F T = 6.44) indi-cated that Fan RTs were faster than Fan 2 RTs, F (l , 19) = 18.39,MSE = 73556.61, and faster than Fan 3 RTs, F(l, 19) = 24.64,MSE = 71928.78. RTs between Fan 2 and Fan 3 did not differ(F < 1). There was no significant fan effect in the SL condition,F(2 , 38) = 1.86, MSE = 35,108.95. Thus, a fan effect wasobserved in the ML condition but not in the SL condition, repli-cating Radvansky et al. (1993). This indicates that there wasgreater interference with an increased number of facts in the MLcondition but not in the SL condition.

An analysis of participants' error rates indicated that overall,participants made fewer errors for studied sentences (3.1%, SD =.04) than nonstudied sentences (7.2% ,SD = .12), F( l, 19) = 6.81,MSE = .01.Mean error rates for Fan 1, Fan 2, and Fan 3 in the SLcondition were 2.5%, 12.0%, and 3.2%, respectively. In the MLcondition, these values were 2 .2%, 3.9%, and 6.9%. Although themain effect of fan was significant, F(2, 38) = 3.83, MSE - 0 .01 ,there was also a significant Fan X Location interaction, F(2,38) = 4.56,MSE = .02. Errors were particularly high in the SL,Fan 2 condition, but it is unclear why this occurred.

The results of Experiment 2 replicated the finding that JOLsdecreased as a function of fan regardless of whether JOL-queriedconcepts stemmed from SL or ML sentences. Furthermore, SLsentences displayed less interference as measured by verificationthan did ML sentences, as evidenced by the interaction betweenfan and location for RTs. Participants verified SL sentences morequickly than ML senten ces, but there was no significant differencebetween the JOL s in the two conditions. Evidence for com petition

2 Sentences were normed for sensibility in an earlier study in which allpossible combinations for 32 objects and 32 locations were constructed toyield a total of 1,024 sentences. Participants in this norming study re-sponded to the query, How sensible is this sentenc e? on a Likert-typescale from 1 (least sensible) to 7 (most sensible). The sentences with thehighest overall ratings were then selected. The lists were constructed suchthat each list represented a unique set of sentences in terms of locationcondition (SL/ML) and fan. Sentences can be obtained from the authors.


6/8

EFFECT OF FAN ON METAMEMORY 11771800 .1700f

g 150014001300

o SingleStudied single Norstudled o M J H p l e Studied

uWpleNonstudied

1 2 3Fan

Figure 3. Mean reaction times for both studied and nonstudied sentencesas a function of fan and location on the verification task for Experiment 2.

at the metacognitive level (i.e., JOLs) but not at a cognitive level(i.e., verification RTs) was attained in the SL condition.

Ex p er imen t 3We conducted a third experiment manipulating the instructionsfor making JOLs. In Experiments 1 and 2, participants madepredictions about their ability to correctly recognize studied factscontaining JOL-queried concepts. We assumed that participantswould interpret this as a llstudied facts associated with the queriedconcept, but this may not have been the case. For this experiment,we explicitly stated what participants should have in mind whenmaking predictions. One group of participants w as told to indicatehow confident they were in recognizing at least one studiedsentence involving the fan concepts w hen making their JOLs . The

other group indicated their confidence in recognizing al l studiedsentences involving the fan concepts. Predicting future recognitionof at least one studied sentence for Fan 3 concepts may result inhigher JOL ratings than for a Fan 1 concept because more factsmean that at least one should be retrievable at test; that is, there isgreater accessibility to the facts. Phrasing the query in terms of a llfacts may focus participants' attention on greater competition forFan 3 concepts than for Fan 1 concepts. Thus, we thought that wemight see increased predictions as fan increased in the at least onecondition, supporting accessibility, and decreased predictions inth e al lcondition, supporting competition.

10..1700I 1600^ 1500| 1 4 O OS1300

SrgetonstudtedM JpteStudted MjtpeNonstudted

2Fan

Figure 4. Mean reaction times for both studied and nonstudied sentencesas a function of fan and location on the verification task for Experiment 3.

MethodTwenty participants were randomly assigned to the at least one condi-

tion, referred to hereafter as the JOL-one condition, and the other 20participants were randomly assigned to the a ll condition, referred to here-after as the JOL-all condition.

The materials, procedure, and design were identical to that used inExperimen t 2 with the exception of two different sets of instructions for theJOL phase. One group of participants, JOL-one, made predictions forcorrectly recognizing at least one studied sentence that contained theJOL-queried concept, whereas the other group, JOL-all, made predictionsfor correctly recognizing all studied sentences involving a concept.Results and Discussion

Learning phase. Participants took an average of 4.57(SD = 1.31) blocks to reach criterion. Neither fan nor locationsignificantly affected mean proportions correct on the cued-recalltest during learning (see Table 3). All effects were nonsignificant,largest F(l, 39) = 1.59, MSE = 0.0 1, for location.

JOL magnitude. Table 4 displays the median ratings of par-ticipants' JOLs for each level of fan as a function of location andJOL condition (one vs. all). These data were entered into a 2 (JOLcondition) X 3 (fan) X 2 (location) mixed-model ANOVA yield-ing only two significant main effects: JOL condition, F(l,38) = 8.51,MSE =2.83, and fan, F(2 , 76) = 15.66,M SE = 0.53.The interactions were all nonsignificant: Location X Fan X JOLCondition, F(2 , 76) = 1.67, MSE =0.55; Location X Fan,F


7/8

1178 McGUIRE AND MAKIbut Fan 1 and Fan 2 concepts were responded to faster than Fan 3concepts, F(l, 39) = 45.63, MSE = 32,443.67, and F(l ,39) = 8.79,M SE = 83,479.49, respectively. Pairwise comparisonsfor the ML sentences showed that Fan 1 concepts were respondedto faster than both Fan 2 concepts and Fan 3 concepts, Fs(l,39) = 21.02 and 34.31, respectively (MSEs = 80,124.62and 75,322.57). RTs for Fan 2 did not differ from Fan 3 RTs, F (l ,39) = 1.27, MSE = 75,499.02. All other interactions were notsignificant, largest F (l , 39) = 1.91, MSE = 36,333.75, for Stud-ied-Nonstudied X Location.

As fan increased, participants' error rates increased as well withmean error rates at 1.8%, 3.6%, and 4.9% for Fan 1, 2, and 3sentences, respectively. Analysis of participants' error rates re-vealed a significant fan effect, F(2, 78) = 9.84,M SE = 0.004. Noother ef fects were s ignif icant , largest F( l , 39) = 3 .31,M SE = 0.002, for the location effect.

The primary purpose of Experiment 3 w as to determine whethera JOL instruction to predict performance on at least one fact versusall facts related to a concept would provide evidence for accessi-bility and competition, respectively. Participants instructed to pre-dict future verification performance for at least one sentencecontaining a JOL-queried concept were more confident overallthan participants who were instructed to base their JOLs on rec-ognizing all studied sentences containing the queried concept. Forboth conditions, though, JOLs decreased as a function of fan, andthe absence of an interaction between instructions and fan suggeststhat the size of the fan effect was similar in the two instructionalconditions. This outcome is inconsistent with predictions from theaccessibility account of JOLs. Instead, our results support the roleof competition in JOLs.

General DiscussionThere are three major points resulting from our study. First,despite having learned all material to approximately the samelevel, participants' JO Ls varied by fan. For all three experiments,JOLs decreased in magnitude as fan increased. Second, we pro-duced an attenuated fan effect in Experiments 2 and 3 for verifi-cation RTs but not for JOLs. Finally, in a direct test of theaccessibility hypothesis using an instructional manipulation, JOLsdecreased as fan increased, whether participants were basing JOLson retrieval of all facts or just one fact for each sentence.Previous investigations of metamemory have reported conflict-ing findings. Koriat (1993, 1995) demonstrated that accessibilityplayed a critical role in predictions of future memory performance.Accordingly, more information retrieved about queried conceptsresulted in higher judgments. In contrast, Schreiber and Nelson

(1998; Schreiber, 1998) demonstrated that more information asso-ciated with queried items in memory resulted in lower judgments.Maki (1999) corroborated Schreiber and Nelson 's findings, show-ing that competition in the form of retroactive interference resultedin lower judgments. On the one hand, studies associated with theaccessibility hypothesis have shown that more information resultsin higher judgments; on the other hand, studies in a competitionframework have shown that more information results in lowerjudgme nts. Our study attempted to resolve these contrasting resultsby manipulating the integration of information to reveal whetherintegration determines the importance of accessibility and com pe-tition in JOLs.However, our manipulation (SL vs. ML sentences)did not produce a significant reduction in the fan effect for

metamemory judgments, although it reduced the fan effect forverification RTs. Still, integration warrants further investigationbecause a higher degree of integration may be necessary beforemore information results in an increase in JOL magnitude.A third model of metamemory, cue familiarity (Metcalfe,Schwartz, & Joaquim, 1 993; Schwartz & Metcalfe, 1992), was not

discussed earlier because it was developed for nonrecalled itemsthat were not well learned. However, it is worth noting that ourresults are inconsistent with an extension of the cue familiarityaccount. Participants were exposed to Fan 3 concepts more thaneither Fan 2 or Fan 1 concepts both during study and during thetest trials. In fact, the queried concepts from Fan 3 sentences werepresented six times during each study-test block, whereas theFan 1 concepts were presented only twice. More exposure shouldhave resulted in an increase of familiarity for such concepts, but itdid not result in an increase in JOLs. JOLs were lower for theFan 3 than for the Fan 1 concepts in all three experiments.Our findings suggest that we do not necessarily monitor theactual contents of memory when making predictions but mayinstead rely on nonanalytic inferences. This line of thought issimilar to the more formal accessibility hypothesis (Koriat, 1993,1995) as well as to the cue-familiarity hypothesis (Metcalfe et al.

1993; Schwartz & Metcalfe, 1992). The competition hypothesisalso suggests the use of a nonanalytic heuristic, that is, thatmetamemory judgments are based on the total amount of activa-tion of associates in memory (Schreiber, 1998). Thus, all threemodels posit that metamemory judgments are based on inferentialheuristics (Koriat, 1998) and not on direct access to memory.Competition was the heuristic that was used in making JOLs inthe fan paradigm. However, JOLs were not based on the amount ofcompetition directly affecting memory, because the fan effect wasreduced for verification RTs in the SL condition relative to the M Lcondition, but there was not a similar reduction for JOLs. F urther-

more, proportion correct during learning did not vary consistentlywith fan, but JOLs decreased consistently with fan. Our findingsfurther support the conclusion that a competition-like mechanismoperates more strongly at the metamemory level than at the levelof memory. This result extends the findings of several earlierstudies (Maki, 1999; Schreiber, 1998; Schreiber & Nelson, 1998)that supported the role of competition in metamem ory judgments.

R efe ren cesAnderson, J. R. (1974). Retrieval of prepositional information from long-term memory. Cognitive Psychology, 6,451474.Anderson, J. R., & Reder, L. M. (1999a). The fan effect: New results andnew theories.Journal of Experimental Psychology: General, 128, 1 8 6 -197.Anderson, J. R., & Reder, L. M. (1999b). Process, not representation:Reply to Radvansky (1999).Journal of Experimental Psychology: Gen-

eral, 128, 207-210 .Koriat, A. (1993). How do we know that we know? T he accessibility modelof the feeling of knowing. Psychological Review, 100, 6 0 9 - 6 3 9 .Koriat, A. (1995). Dissociating knowing and the feeling of knowing:Further evidence for the accessibility model. Journal of Experimental

Psychology: General, 124,311-333.Koriat, A. (1998). Metamem oiy: The feeling of knowing and its vagaries.In M. Sabourin & F. Craik (Eds.), Advances in psychological science:

Vol. 2. Biological and cognitive aspects (pp. 461479). East Sussex,UK: Psychology Press.Maki, R. H. (1999). The roles of competition, target accessibility, and cue


8/8

EFFECT OF FAN ON METAMEMORY 1179familiarity in metamemory for word pairs. Journal of ExperimentalPsychology: Learning, Memory, and Cognition, 25, 1011-1023.

Metcalfe, J., Schwartz, B. L., & Joaquim, S. G. (1993). The cue-familiarityheuristic in metacognition.Journal of Experimental Psychology: Learn-ing, Memory, and Cognition, 19, 8 5 1 - 8 6 1 .

Nelson, T. O., & N arens, L. (1990). Metamemory: A theoretical frameworkand new findings. In G. H. Bower (Ed.), hepsychology of learning andmotivation: Advances in research and theory (Vol. 26, pp. 125-173).San Diego, CA: Academic Press.

Nelson, T. O., Sheck, P., Dunlosky, J., & Narens, L. (1999, November).What is the basis for judgments of learning (JOLs)for recallable items?Paper presented at the meeting of the Psychonomic S ociety, Los An ge-les, CA.

Radvansky, G. A. (1999a). The fan effect: A tale of two theories. Journalof Experimental Psychology: General, 128, 198-206.

Radvansky, G. A. (1999b). Memory retrieval and suppression: The inhi-bition of situation models. Journal of Experimental Psychology: Gen-eral, 128, 1-17.

Radvansky, G. A., Spieler, D. H., & Zacks, R. T. (1993). Mental modelorganization. Journal of Experimental Psychology: Learning, Memory,and Cognition, 19, 95-114.

Schneider, W. (1995). MEL Professional User s Guide. Pittsburgh, PA:Psychology Software Tools.

Schreiber, T. A. (1998). Effects of target set size on feelings of knowingand cued recall: Implications for the cue effectiveness and partial-retrieval hypotheses. Memory and Cognition, 26, 553-571.Schreiber, T. A., & Nelson, D. L. (1998). The relation between feelings ofknowing and the number of neighboring concepts linked to the test cue.Memory and Cognition, 26, 8 6 9 - 8 8 3 .

Schwartz, B. L., & Metcalfe, J. (1992). Cue familiarity but not targetretrievability enhances feeling-of-knowing judgments. Journal of Exper-imental Psychology: Learning, Mem ory, and Cognition, 1 8, 1074-1083.

Received February 16, 2000Revision received November 22, 2000

Accepted February 1, 2001

Mem bers of Underrepresented G roups:Reviewers for Journal Manuscripts WantedIf you are interested in reviewing manuscripts for APA journals, the APA Publicationsand Com mu nications Board would like to invite your participation. Manu script re-viewers are vital to the publications proc ess. As a reviewe r, you would gain valuableexperienc e in publishing. The P&C Board is particularly interested in encouragingmembers of underrepresented groups to participate more in this process.If you are interested in reviewing manuscripts, please write to Demarie Jackson at theaddress below. Please note the following important points: To be selected as a reviewe r, you must have published articles in peer-reviewedjour nals. The experience of publishing provides a reviewer with the basis for

preparing a thorough, objective review. To be selected, it is critical to be a regular reader of the five to six empirical jour-nals that are most central to the area or journ al for which you w ould like to review.Current knowledge of recently published research provides a reviewer with theknowledge base to evaluate a new submission within the context of existing re-search. To select the appropriate reviewers for each man uscript, the editor needs detailedinformation. Please include with your letter your vita. In your letter, please iden-tify which APA journal(s) you are interested in, and describe your area of exper-

tise. Be as specific as possible. For exam ple, social psycho logy is not suffi-cient you would need to specify social cogn ition or attitude cha nge as well. Reviewing a manuscript takes time (1 -4 hours per manuscript reviewed). If youare selected to review a manuscript, be prepared to invest the necessary time toevaluate the manuscript thoroughly.Write to Demarie Jackson, Journals Office, American Psychological Association, 750First Street, NE, Washington, DC 20002-4242.

when knowing more means less the effect of fan

Documents