n adi

Ar

ry Co

isione 9 N

Abstract

Introduction

& McDermott, 1995; Tulving, 1985). The overall high

levels of false recognition and recall, as well as the factthat participants often claim to consciously recollect lure

not seem to diminish levels of false recognition (Gallo& Roediger, 2002), and may even increase levels of falserecognition (Arndt & Hirshman, 1998). Similarly,repeating study items diminishes, but does not eliminate,false recognition of lure items (Benjamin, 2001; Seamon

0749-596X/$ - see front matter 2005 Elsevier Inc. All rights reserved.

* Fax: +1 802 443 2072.E-mail address: jarndt@middlebury.edu.

Journal of Memory and Language

Journal ofMemory andRecent interest in memory errors can be traced to areport by Roediger and McDermott (1995). Roedigerand McDermott, reviving a paradigm originated byDeese (1959), demonstrated that when given a list ofstudy items that are semantically related to a single,unstudied word (referred to as the lure item), partici-pants will both falsely recall and recognize the unstudiedlure item. Further, participants claimed to be able toconsciously recollect the lure items presentation on thestudy list, as indexed by remember responses (Roediger

items on recognition memory tests has been taken as evi-dence that these errors reect fundamental memoryprocesses.

Despite the hypothesis that errors in this paradigm(referred to as the DRM paradigm below) derive frombasic memory processes, one particularly striking fea-ture of performance in the DRM paradigm is howstrongly lure false alarms persist in the face of manipu-lations intended to reduce memory errors. For example,increasing presentation duration for study items doesFour experiments evaluated the role of encoding-based and retrieval-based factors in the production of false recog-nition. The association of unusual fonts with study items, the match between study and test font, and the duration ofretrieval time allotted to subjects to make recognition memory decisions were varied in order to examine the role encod-ing and retrieval play. The combined results of these experiments suggest that both encoding and retrieval are importantfor understanding false recognition. Additionally, the results of these studies suggest that visual features encountered atencoding can become associated with representations of unstudied items. This association, in turn, can lead to inatedlevels of false recognition when unstudied items are tested in a visual format experienced at encoding or when partic-ipants utilize monitoring processes to search memory for evidence of perceptual information encountered duringencoding. 2005 Elsevier Inc. All rights reserved.

Keywords: False memory; Recognition; Distinctiveness; Encoding; RetrievalDistinctive informatioThe contribution of enco

Jason

Department of Psychology, Middlebu

Received 2 June 2005; revAvailable onlindoi:10.1016/j.jml.2005.08.003nd false recognition:ng and retrieval factors

ndt *

llege, Middlebury, VT 05753, USA

received 22 August 2005ovember 2005

54 (2006) 113130

www.elsevier.com/locate/jml

Language

114 J. Arndt / Journal of Memory and Language 54 (2006) 113130et al., 2002). Finally, even providing participants withexplicit warnings about the nature of the paradigm, giv-ing them an example of a study list, and instructing themto monitor for and attempt to reject lure items on a rec-ognition memory test does little to reduce lure falsealarms when given after study (Gallo, Roediger, &McDermott, 2001) and does not fully eliminate lure er-rors when given prior to study (Gallo et al., 2001). Thus,it is of particular interest to examine variables that sub-stantially reduce lure false alarms. Doing so will bothenhance our understanding of the genesis of false recog-nition and the kinds of information people are able toutilize to reduce false recognition. The exploration ofone such manipulation, presentation of study items inunusual-looking fonts (Arndt & Reder, 2003), will bethe central focus of the research reported in this paper.

A second stimulus for the present research derivesfrom contemporary theoretical perspectives that suggestfalse memory is the joint product of encoding-based andretrieval-based factors (e.g., Brainerd, Reyna, Wright, &Mojardin, 2003; Roediger, Watson, McDermott, &Gallo, 2001). While multiple two-factor explanationsfor the generation of false memory exist, the one mostgermane to the present set of studies is activation-mon-itoring theory (Roediger et al., 2001). Thus, this theoryand its propositions are outlined in detail, along withits characterization of the properties of encoding-basedprocesses and retrieval-based processes that producefalse memory.

Roediger et al. (2001) suggest that false memory iscaused by two factors, activation and monitoring. Whileactivation processes are not solely operative at encodingand monitoring processes are not solely operative atretrieval, the focus of the studies reported here exploreactivation processes that are the product of encoding-based inuences and monitoring processes that occurat retrieval. Thus, the discussion of activation processesis limited to the encoding factors that are assumed toinuence lure activation and the discussion of monitor-ing processes is limited to their role at retrieval.

In activation-monitoring theory, increased lure itemactivation is assumed to increase false memory. Roedi-ger et al. suggest that the extent to which study condi-tions produce item-specic and relational processing(Hunt & Einstein, 1981) determines how strongly a lureitems representation will become activated. Relationalprocessing will tend to increase activation levels of lureitems representations by drawing participants encodingresources to the commonalities among study items. Incontrast, item-specic processing draws participantsencoding resources and attention to elements of studyitems that distinguish them from one another, whichmay allow participants to better distinguish studieditems from unstudied, but related, items at retrieval.Thus, the extent to which encoding conditions empha-

size relational processing over item-specic processingshould lead to enhanced levels of lure item activation,and subsequently, increased levels of false memory.

The characterization of monitoring processes in acti-vation-monitoring theory borrows notions from thesource-monitoring framework (Johnson, Hashtroudi,& Lindsay, 1993). Monitoring processes are assumedto be enhanced by the encoding of information that isparticular to a study episode, which in turn allows sub-jects to distinguish studied items from unstudied items atretrieval. Thus, monitoring processes will tend to be en-hanced by the encoding of presentation characteristicsof study items (e.g., item-specic information such asvisual presentation features and auditory features), men-tal operations that were engaged during study, or anyother factor that is uniquely associated with the experi-ence of encoding study items (Johnson et al., 1993). Atretrieval, monitoring processes then provide participantswith improved ability to reject unstudied items becauseitems that were not encountered during an episode areunlikely to be associated with the details of a studyepisode.

While activation-monitoring theory borrows mostheavily on the source monitoring framework to charac-terize how retrieval-based processes can be used to re-duce false memory, it is important to note that othertheoretical accounts of monitoring processes exist, eachof which may also provide an account of monitoring-based reductions in false memory. In the false memoryliterature, the two most prominent explanations of mon-itoring processes are the distinctiveness heuristic (e.g.,Dodson & Schacter, 2001, 2002a, 2002b; Gallo, Weiss,& Schacter, 2004; Hege & Dodson, 2004; Schacter, Cen-dan, Dodson, & Cliord, 2001; Schacter, Israel, & Ra-cine, 1999) and recall-to-reject processing (e.g.,Brainerd et al., 2003; Gallo, 2004; Rotello, Macmillan,& Van Tassel, 2000).

The distinctiveness heuristic hypothesis suggests thatcertain encoding experiences are strongly indicative ofthe source of a memory (e.g., encoding of pictorial infor-mation; Gallo et al., 2004; Schacter et al., 1999, 2001).These salient memories of study items in turn lead peo-ple to expect that they will be able to retrieve distinctiveinformation at test. In the absence of being able toretrieve distinctive information about a test item, partic-ipants will reject that item. Given that unstudied itemssuch as lures in the DRM paradigm were never encoun-tered at study and therefore should not have distinctiveinformation associated with their representations, thedistinctiveness heuristic allows subjects to eectivelydistinguish between studied items and related lure items.The distinctiveness heuristic, while suggesting very simi-lar processes to that outlined in the source monitoringframework, proposes one essential dierence in howmonitoring processes operate. The use of a distinctive-ness heuristic has been suggested to operate on a global,

rather than a local basis (Dodson & Schacter, 2001;

unique font conditions. A recall-to-reject explanationwould claim that a font, when used to present a lureitem, can cue the retrieval of study items that were pre-sented in that font, which then serves as a basis forrejecting lure items. In the unique font condition, thefont in which a lure item was presented at test was onlyused to present a single study item. Thus, when a lurewas presented at test, that font potentially cued theretrieval of the only study item presented in that font,producing rejection of the lure (i.e., it enabled disquali-fying monitoring; Gallo, 2004). In the correlated fontcondition, on the other hand, the font in which a lureitem was presented at test was used to present multiple(12) study items. Thus, retrieval of study items presented

J. Arndt / Journal of Memory and Language 54 (2006) 113130 115Schacter et al., 1999). Thus, if participants study experi-ence leads to the application of the distinctiveness heu-ristic, (s)he will apply it to all test items, regardless ofwhether or not all study items were presented in a dis-tinctive format at encoding, a proposal that has beensupported by numerous studies (e.g., Dodson & Schact-er, 2001, 2002a, 2002b; Schacter et al., 1999, 2001).

Recall-to-reject processing highlights the notion thatthe retrieval of a study item related to a lure item providesevidence of whether or not a lure was studied (e.g., Gallo,2004; Rotello et al., 2000). Thus, retrieving a study itemwhen a lure item is tested can conceivably provide discrim-inative information that allows a participant to decide onthe studied status of a related, but unstudied, lure item.Gallo (2004) recently suggested that recall-to-reject pro-cessing can operate in one of twoways, referred to as diag-nostic monitoring and disqualifying monitoring.Diagnostic monitoring suggests that as the number ofstudy items recalled or remembered improves, the abilityto reject lure items improves as well, whether due to im-proved sourcemonitoring (Johnson et al., 1993; Roedigeret al., 2001) or simply better overall knowledge of the stud-ied members of a theme. Disqualifying monitoring is amore restricted view of the usefulness of recall-to-rejectmonitoring processes. This type of monitoring suggeststhat the only way in which recall-to-reject processing willallow for the rejection of a lure item is when one can accu-rately retrieve all of the studied associates of a lure item.Thus, disqualifying monitoring would be unlikely to beeective in reducing false memory in the DRMparadigm,except in cases where relatively few associates were stud-ied. Important for application of recall-to-reject process-ing to the current studies is that Gallo (2004) foundevidence that only disqualifying monitoring was eectiveat reducing false recognition in the DRM paradigm.

The present studies

The present research seeks to examine the contribu-tion of encoding and retrieval processes to the false rec-ognition reduction produced by presenting study itemsin unusual-looking fonts (see Fig. 1 for examples). Arndtand Reder (2003) presented themes in unusual-lookingfonts, and manipulated the association of the font infor-mation to themes related to lure items. In one condition,all items within a theme were presented in the same,unusual-looking font (referred to as the correlated fontcondition below), while in a comparison condition, eachitem within a theme was presented in a dierent, unusu-al-looking font (referred to as the unique font conditionbelow). Arndt and Reder found that participantsshowed reduced false recognition in the unique fontcondition relative to the correlated font condition.Further, they found that lower levels of false recognitionin the unique condition replicated across between- and

within-subjects designs, and that the levels of falserecognition found in the correlated condition were com-parable to the levels found under standard recognitionconditions in the DRM paradigm, where study and testitems are presented in commonly encountered fonts.

Arndt and Reder argued that their results were bestunderstood in terms of encoding-based inuences, suchas the extent to which study conditions strengthenedthe activation of lure items representations. Specically,Arndt and Reder argued that the correlated font condi-tion inspired more relational processing than the uniquefont condition (Hunt & Einstein, 1981), producinggreater lure item activation in the correlated font condi-tion. While their results can be explained by the opera-tion of encoding processes, it is plausible that Arndtand Reders results can also be understood in terms ofthe operation of retrieval mechanisms. This alternativeexplanation is plausible because Arndt and Rederalways presented lure items in a font that was used topresent a study item related to it. Thus, if a lure itemwas assigned to the correlated condition, it was present-ed in the same font used to present all of the study itemsrelated to it. Similarly, if a lure item was assigned to theunique condition, it was presented in one of the fontsused to present one of the study items related to it.

Given Arndt and Reders procedures, a variant of re-call-to-reject processes can also explain the dierences infalse recognition observed between the correlated and

Fig. 1. Examples of fonts used to present stimulus items.in that font would not be an eective basis for rejecting a

tion claims that when participants encounter the salientvisual information in the unique font condition, they

116 J. Arndt / Journal of Memory and Language 54 (2006) 113130develop the expectation that they will be able to retrievesalient font information on a later memory test. Thus,on a memory test, participants expect that they shouldbe able to retrieve evidence that a word was studied in adistinctive font. When they are unable to retrieve distinc-tive visual information associatedwith theirmemory for atest item, they will tend to reject that item as unstudied.

Although both propose that retrieval processes makea contribution to reducing false recognition, the distinc-tiveness heuristic diers from the recall-to-reject hypoth-esis regarding how retrieval processes operate to reducefalse memory in the unique font condition compared tothe correlated font condition. Specically, the recall-to-reject hypothesis claims that participants use informa-tion available in the retrieval cue provided at test (e.g.,the font in which a test item is presented) to retrieve alures associate(s) that were also presented in that font,potentially facilitating the use of disqualifying monitor-ing when a font is associated with a single study item.The distinctiveness heuristic hypothesis, on the otherhand, suggests that the presence or absence of a fontat test is irrelevant to the use of retrieval monitoringto reduce false recognition in the unique font condition.Rather, the distinctiveness heuristic hypothesis suggeststhat participants search memory for distinctive informa-tion that is indicative of study item presentation, and doso regardless of whether or not such information is pre-sented at retrieval (e.g., Schacter et al., 1999).

In the present set of studies, Arndt and Reders (2003)encoding-based explanation is contrasted with the recall-to-reject explanation and the distinctiveness heuristicsexplanation of the eects of font condition on false recog-nition. In the rst two experiments, the match betweenstudy and test font presentation is manipulated. In thethird experiment, a retrieval time manipulation is em-ployed (Benjamin, 2001; Heit, Brockdor, & Lamberts,2004). The fourth experiment examines the extent towhich reinstatement of a studied font increases lure falsealarms. The logic behind each experiments test of the con-tribution of encoding- and retrieval-based processes tofalse recognition is elaborated prior to each study.

Experiment 1

In this experiment, an approach similar to Reder,Donavos, and Erickson (2002) is adopted. Reder et al.studied font-matching eects in standard item recogni-lure item because the font was associated with multiplestudy items, reducing or eliminating the ability to usedisqualifying monitoring.

The other retrieval-based explanation that is tested inthese studies is the distinctiveness heuristic. This explana-tion by manipulating whether test items were re-present-ed in the same font used to present that item at study(the match condition), or in a font used to present a dif-ferent study item (the mismatch condition). Reder et al.demonstrated that items in the match condition showedsuperior recognition compared to items in the mismatchcondition, suggesting the contribution of retrieval mech-anisms to recognition memory. Additionally, Rederet al. found that the eect of matching fonts from studyto test was greater when a font was associated with a sin-gle study item relative to when it was associated withmultiple (12) study items. Reder et al. argued this resultreected dierences in contextual interference across theone vs. multiple presentation font conditions, such thata font used to present a single study item was a betterretrieval cue for that item relative to a font used to pres-ent multiple study items. Note that the relative ecacyof fonts as retrieval cues for study items when they areused to present a single study item and present multiplestudy items is common to Reder et al.s (2002) explana-tion and the recall-to-reject explanation of Arndt andReders (2003) data outlined above. This same approachcan be used to study the extent to which encoding andretrieval mechanisms contribute to false recognitionreduction. In particular, both font condition (correlatedvs. unique) and study-test match (match vs. mismatch)were manipulated in Experiment 1.

Given this design, the encoding-based explanationoered by Arndt and Reder (2003) and recall-to-rejecttheories expect dierent patterns of results. Encoding-based theories propose that manipulating study-testmatch should not substantially alter the inuence ofthe dierent font conditions on false recognition. Thisis because a strictly encoding-based view argues thatthe activation of lure items representations, and there-fore levels of false recognition, result from the extentto which font conditions inspire dierent levels ofitem-specic and relational processing at encoding.Thus, from the perspective of pure encoding-based viewsof false recognition, test probes only serve the purposeof providing participants with an assessment of the lurerepresentations level of activation. Therefore, alteringthe font information available at retrieval should nothave a substantial inuence on lure false recognition.On the other hand, a recall-to-reject explanation wouldpropose that study-test match should lead to dramatical-ly dierent levels of false recognition. Specically, a re-call-to-reject explanation would expect lower falserecognition in the unique font condition relative to thecorrelated font condition when lure items are presentedin a font that was used to present one of its studiedassociates (the match condition). In contrast to anencoding-based theory, this explanation would expecta substantial reduction or elimination of the false recog-nition advantage for items in the unique font conditiononly when lure items are presented in a font not used

to present one of its studied associates (the mismatch

J. Arndt / Journal of Memory and Language 54 (2006) 113130 117condition). This prediction is based on the idea that re-call-to-reject processing will only be eective when astudied item that is related to a lure is retrieved (Brain-erd et al., 2003; Gallo, 2004; Rotello et al., 2000). Thus,presenting lure items in a font that was not used to dis-play its studied associates will at best produce retrievalof an unrelated study item, which will not aid subjectsin deciding the study status of a lure item.

Finally, font condition was manipulated in two dif-ferent study list compositions (pure vs. mixed) in orderto replicate the eects of between- and within-subjectsmanipulations demonstrated in Arndt and Reder(2003). Manipulating font condition in pure lists (i.e.,all themes within a study list assigned to the correlatedcondition or the unique condition) can serve as a proxyfor a between-subjects manipulation, because partici-pants are free to alter their decision criterion as a func-tion of changes in the composition of memory(Hirshman, 1995). Similarly, manipulating font condi-tion in mixed lists (i.e., some themes assigned to the cor-related condition and some themes assigned to theunique condition) replicates the manner in which with-in-subjects manipulations have been implemented in pri-or research (Arndt & Reder, 2003; Dodson & Schacter,2001; Schacter et al., 1999).

Manipulating font condition in both pure and mixedlists allows for an evaluation of the explanation oeredby encoding-based theories against that oered by thedistinctiveness heuristic. Recall from above that the dis-tinctiveness heuristic is a special case of source monitor-ing in which participants employ a strategy where theyapply the heuristic to all test items as long as salient dis-tinctive information is encountered at study. Thus, tothe extent the heuristic is employed, it will be used to re-duce false recognition for all test items, regardless ofwhether their associated study items were experiencedin a distinctive format or not. One method that has beenused to test the distinctiveness heuristic is to comparethe eects of an independent variable when it is manip-ulated between subjects relative to when it is manipulat-ed within subjects (Arndt & Reder, 2003; Dodson &Schacter, 2001; McCabe, Presmanes, Robertson, &Smith, 2004; Schacter et al., 1999). The prediction ofthe distinctiveness heuristic in this case is that levels offalse memory are free to vary when a variable is manip-ulated between subjects, but not within subjects. Thisprediction bears strong similarity to the notion thatwhile recognition memory decision criteria are free tovary when conditions are manipulated between subjectsor test lists (Hirshman, 1995; Stretch & Wixted, 1998;Wixted & Stretch, 2000), participants have a strong ten-dency to maintain a single decision criterion when con-ditions are manipulated within subjects and test lists(Stretch & Wixted, 1998; Wixted & Stretch, 2000). Thebasis for this occurrence according to the distinctiveness

heuristic is that subjects are generally assumed to applythe heuristic to all test items, regardless of whether allstudy items were encoded in a distinctive manner ornot. Thus, when a variable is manipulated within sub-jects and test lists, the heuristic will be applied to testitems regardless of which encoding condition they werepresented in, producing lower levels of false recognitionfor lure items related to study items presented in a dis-tinctive format at study as well as lure items related tostudy items not presented in a distinctive format. As aresult, the distinctiveness heuristic predicts that variablesshould dierentially reduce false memory when they aremanipulated between subjects (or lists), but not whenthey are manipulated within subjects and lists. In con-trast, from the perspective of a strictly encoding-basedtheory, there is no reason to believe that altering listcomposition should alter the amount of item-specicand relational processing that occurs in each font condi-tion. Thus, encoding-based theories predict that the ef-fects of font condition should be comparable in mixedand pure lists.

The comparison of mixed and pure lists has the add-ed advantage that the distinctiveness heuristic can beevaluated against encoding-based explanations more rig-orously than by comparing the eects of a manipulationacross within- and between-subjects designs as has beendone in past research (Arndt & Reder, 2003; Dodson &Schacter, 2001; McCabe et al., 2004; Schacter et al.,1999). Specically, by manipulating a variable both be-tween and within subjects, one is left to rely on qualita-tive comparisons, where either the eects of anindependent variable on false recognition qualitativelyreplicate, favoring an encoding-based explanation, orthey disappear, favoring the explanation of the distinc-tiveness heuristic. The advantage of the present, entirelywithin-subjects, approach is that it allows one to betterassess the comparability of the eects of independentvariables across pure and mixed lists designs. Thus, thesimilarity between levels of false recognition across pure-and mixed-lists designs provides a sense for the role thatencoding-based processes play in false recognitionreduction, while the dierence provides an estimate ofthe extent to which the distinctiveness heuristic contrib-utes to false recognition reduction.

In sum, the encoding-based explanation would pre-dict that there should be fewer false alarms in the uniquefont condition than in the correlated font condition, andthat the inuence of font condition on false recognitionshould be comparable in the match and mismatch con-ditions, as well as when the unique and correlated fontconditions are presented in mixed and pure lists atencoding. The recall-to-reject explanation expects thatthe unique font condition will produce lower falsealarms to lure items when the font in which a lure is pre-sented is one that was used to study one of its associates(the match condition), but not when the font in which a

lure is presented is one that was used to present a study

118 J. Arndt / Journal of Memory and Language 54 (2006) 113130item that is unrelated to the lure (the mismatch condi-tion). Finally, the distinctiveness heuristic explanationexpects that the dierence between lure false alarms inthe unique and correlated font conditions should besmaller in mixed lists relative to pure lists.

Method

Participants

Thirty-two students enrolled in introduction to psy-chology at Middlebury College participated as part ofa research appreciation requirement.

Materials and design

Stimulus items were selected from the Nelson, McE-voy, and Schreiber (1998) database. Ninety-six sets often items each (referred to as themes below) were chosento serve as study items. Stimuli were selected such thatall items within a theme produced the same lureitem with a nonzero probability (mean = .398; range =.033.960; range of mean associative strength forthemes = .267.685). For example, the stimulus itemssaddle, pony, mare, gallop, colt, stallion, rodeo, stable,trot, and unicorn were chosen as study items and allproduce the lure item horse in free association.

Study and test stimuli were presented in a variety ofunusual-looking fonts. One hundred and forty-fourfonts were selected from various internet sources. Selec-tion criteria were similar to those used by Arndt and Re-der (2003) as well as Reder et al. (2002). The fontsutilized in this experiment were ones that participantswere unlikely to have encountered previously, did notresemble fonts one typically sees in everyday life (e.g.,Helvetica, Courier), were easily readable, and were sim-ilar in size to one another. Examples of the fonts selectedfor the current experiment are presented in Fig. 1.

For each participant, 48 of the 96 themes were chosento serve as study materials, with the remaining 48 serv-ing as distractor materials for the recognition memorytests. Within each set of 48 themes chosen to be studyitems, six themes were assigned to one of eight experi-mental conditions formed by the crossing of three exper-imental factors. The rst factor (referred to as FontCondition below) manipulated whether the study itemsin a theme were all presented in the same font (the cor-related condition) or were all presented in a dierent font(the unique condition). In the correlated font condition, asingle font was utilized to present all of the study itemswithin a given theme, and fonts diered across themes.In the unique font condition, each of the study itemswithin a theme was displayed in a dierent font. The sec-ond factor (referred to as Study-Test Match Conditionbelow) manipulated whether test items from a studiedtheme (both studied and lure items) were presented inthe same font used to display the study items from that

theme (the match condition), or in a font that was used topresent the study items from a dierent theme (the mis-match condition). Thus, in the match condition, studyitems were presented in the font in which they were stud-ied, while lure items were presented in a font used topresent a thematically related study item. In the mis-match condition, study items were presented in a fontused to present a dierent theme in the same study list,while lure items were presented in a font used to presentits associates at test (i.e., in a font not used to present itsassociates at study). Further, in the mismatch condition,study items and lure items were presented in a font thatwas assigned to the same level of Font Condition asthose study and lure items were. The third factor (re-ferred to as List Composition below) manipulatedwhether all themes in a study list were assigned to thesame level of the Font Condition factor (pure lists) orif half of the themes assigned to a given study list werein the correlated condition, and the other half of thethemes assigned to a given study list in the unique con-dition (mixed lists).

Finally, fonts were assigned to new items and newlures in an analogous manner to that utilized to assignfonts to study items in each condition. Thus, all newitems and new lures were presented in a font encoun-tered at study, and the manner in which those fonts wereassigned to new items and new lures depended on thefont condition to which an unstudied theme wasassigned.

The four experimental conditions were implementedin four dierent study lists. In one of the study lists,all themes were presented in the correlated condition,while a second study list presented all themes in the un-ique condition. The nal two study lists were composedof six themes presented in the correlated condition, andsix themes presented in the unique condition. In both ofthe mixed lists, themes assigned to the correlated condi-tion and themes assigned to the unique condition wererandomly intermixed.

Procedure

Participants completed four study-test trials. Atstudy, participants were informed that they would beshown a list of words one at a time on a computerscreen, and that their task was to do their best to remem-ber the words because their memory would be tested lat-er in the experiment. Further, participants were asked torate how appropriate each font was for the meaning ofthe word on a scale ranging from one (not very appro-priate) to four (very appropriate). This task was intend-ed to direct participants attention to both the visualproperties of the font as well as the semantic propertiesof the word (Arndt & Reder, 2003; Reder et al., 2002).Participants were further provided with an instance ofa situation in which one might judge the correspondencebetween the visual properties of a font and the meaning

of a word as appropriate, and an instance in which the

correspondence might be judged as inappropriate. Par-ticipants were instructed that if the study item waswindy and the font appeared to be italicized or wasbent over towards one side as wind might do to an ob-ject, that correspondence might be judged appropriate.On the other hand, if the font appeared to be rigid ordid not contain wavy lines characteristic of the depictionof windiness they might judge that correspondence tobe inappropriate. Participants then progressed throughthe study list at their own pace.

For each study list, the study items from 12 themes(120 items total) were presented serially in the centerof a computer screen, blocked by theme. Following thepresentation of each study list, participants were givenan oldnew recognition memory test. On the recognitionmemory test, participants were presented with studieditems (ve per theme), the lure items corresponding toeach studied theme, 5 items from a comparable set of12 themes that had not been presented at study (referredto as new items below), and the lure items corresponding

ond, third, and fourth an equal number of timesacross participants. Further, themes served as studiedand new themes an equal number of times across partic-ipants, and each theme served in each of the experimen-tal conditions an equal number of times acrossparticipants. Finally, fonts were assigned to each ofthe experimental conditions an equal number of timesacross participants.

Results and discussion

The top half of Table 1 presents hits and lure falsealarms as a function of Font Condition, Study-TestMatch, and List Composition, while the bottom halfof Table 1 presents new item and new lure false alarmsas a function of Font Condition and List Composition.Hits and lure false alarms were analyzed with separate2 2 2 within-subjects ANOVAs. New item falsealarms and new lure false alarms were analyzed withseparate 2 2 within-subjects ANOVAs. The alpha level

Exp

Mism

.79 (

.81 (

.61 (

.68 (

.11 (

.11 (

.15 (

.17 (

J. Arndt / Journal of Memory and Language 54 (2006) 113130 119to the unstudied themes (referred to as new lures below).Thus, the test list length was 144 items (ve items fromeach of the 12 studied and 12 new themes, as well as eachthemes corresponding lure items). Test items were ran-domly presented. Following the completion of the rstrecognition memory task, participants were instructedthey would be shown another list of words on the com-puter screen, and that their task was again to do theirbest to remember those items. Further, participants wereinstructed that they no longer needed to remember thewords that were presented in the prior study-test trial.

Order of presentation of the four study and test lists(pure correlated, pure unique, and the two mixed lists)was counterbalanced with a Latin Square design to en-sure that each type of list condition occurred rst, sec-

Table 1Hits, false alarms to lure items, and false alarms to new items in

List Composition

Correlated

Match

Hits

Pure .87 (.025)Mixed .89 (.025)

Lure false alarms

Pure .63 (.041)Mixed .70 (.046)

New false alarms

Pure N/AMixed N/A

New lure false alarms

Pure N/AMixed N/A

Standard error of the mean in parenthesis.used for all analyses was .05.The analysis of hits indicated that the Font Condi-

tion main eect was reliable, F(1,31) = 5.77,MSE = .009, such that hits were higher in the correlatedcondition (M = .84) than in the unique condition(M = .81). Further, the Study-Test Match main eectwas reliable, F(1,31) = 48.89, MSE = .012, indicatingthat hits were higher when study and test font matched(M = .88) than when study and test font did not match(M = .78). Finally, the interaction between Font Condi-tion and Study-Test Match approached signicance,F(1,31) = 3.98, MSE = .006, p < .06. The pattern ofmeans producing this interaction suggests that hits forstudy items in the unique font condition diminishedmore when study and test font mismatched (.87 vs. .75

eriment 1

Font Condition

Unique

atch Match Mismatch

.030) .88 (.016) .76 (.023)

.026) .86 (.028) .75 (.028)

.048) .47 (.045) .50 (.045)

.046) .43 (.048) .48 (.050)

.014) N/A .13 (.018)

.016) N/A .11 (.015)

.023) N/A .15 (.026)

.029) N/A .13 (.019)

Thus, this weaker variant of the recall to reject hypoth-esis cannot be ruled out by the results of Experiment 1,

120 J. Arndt / Journal of Memory and Language 54 (2006) 113130in the match and mismatch conditions, respectively) rel-ative to the correlated font condition (.88 vs. .80 in thematch and mismatch conditions, respectively). Thistrend is consistent with prior studies showing thatstudy-test font match eects in recognition are largerwhen fonts are uniquely associated with a single item rel-ative to when a font is associated with multiple studyitems (Reder et al., 2002). No other main eects or inter-actions were reliable (all Fs < 1).

The analysis of lure false alarms produced a main ef-fect of Font Condition, F(1,31) = 47.23, MSE = .05,indicating that lure item errors were higher in the corre-lated condition (M = .66) than in the unique condition(M = .47). The only other eect that approached signif-icance was the interaction between Font Condition andList Composition, F(1,31) = 2.97, MSE = .05, p < .10.This marginal interaction hints at a trend for the eectof Font Condition to be slightly larger in mixed lists(.69 vs. .46 in the correlated and unique conditions,respectively) than in pure lists (.62 vs. .49 in the correlat-ed and unique conditions, respectively). None of theother main eects or interactions approached signi-cance (largest F = 1.82).

Finally, both the analysis of new item false alarmsand new lure false alarms failed to produce any signi-cant eects. The only eect that approached signicancewas the interaction between Font Condition and ListComposition (p < .10). The ordinal direction of themeans producing this trend suggested that there was aslightly higher probability of false alarms to new itemsin the correlated font condition (M = .13) than the un-ique font condition (M = .11) when font condition wasmanipulated in mixed lists, but not when font conditionwas manipulated between lists (M = .10 and .11 in thecorrelated and unique font conditions, respectively).

The results of this experiment are consistent with theexpectations of an encoding-based explanation, and areinconsistent with both a recall-to-reject and a distinctive-ness heuristic explanation. Specically, the nding thatStudy-Test Match did not alter rates of false alarms tolure items is inconsistent with the recall-to-reject expla-nation of false recognition reduction (Rotello et al.,2000). It is important to note that this result is not sim-ply due to the study-test match manipulation being tooweak to inuence performance, because hits were reli-ably higher when the font of presentation matched be-tween study and test. Further, hit analyses showedpatterns of results qualitatively consistent with thosefound by Reder et al. (2002), where study-test matchhad a larger inuence on performance when items werepresented in the unique font condition relative to thecorrelated condition.

Replicating Arndt and Reder (2003), it mattered littlewhether Font Condition was manipulated in pure listsor mixed lists, a result that is inconsistent with the dis-

tinctiveness heuristic explanation. Although the interac-because the font utilized to present lures at test alwaysserved in the font condition in which the lures associ-ates were presented. Experiment 2 tested this alterna-tive recall-to-reject hypothesis by presenting test itemsassigned to the mismatch condition in a common fontthat was not experienced in the study list (Helvetica).The logic behind this change in procedure is that ifthe weaker variant of the recall-to-reject hypothesisoutlined above is able to account for some or all ofthe eects of font condition on false recognition, thedierence in false recognition between the unique andcorrelated font conditions should be reduced when testitems are presented in a font that was never experi-enced at encoding. As with the rst experiment, theencoding-based view would still predict that the eectsof Font Condition on false recognition should not beinuenced by the font used to present lure items at test.Finally, Experiment 2 again manipulated List Compo-sition (pure vs. mixed) as a test of the distinctivenesstion between Font Condition and List Compositionapproached signicance in the lure false alarm analyses,the qualitative pattern of performance suggested that thedierence in false alarms between the unique and corre-lated font conditions was greater in mixed than in purelists. This pattern is the opposite of that expected bythe distinctiveness heuristic explanation, which predictsthat the eects of Font Condition should diminish whenmanipulated in the same study (and test) list. Thus, giv-en the similarity in magnitude of eect of Font Condi-tion on false recognition across mixed and pure lists,these results appear in accord with an account highlight-ing the inuence of encoding-based processes and theirinuence on lure item activation (e.g., item specic andrelational processing; Arndt & Reder, 2003; Hunt &Einstein, 1981; Roediger et al., 2001). In the secondexperiment, the encoding-based and retrieval-basedexplanations were again assessed, this time with a stron-ger manipulation of study-test match.

Experiment 2

Experiment 1 provided evidence inconsistent withone variant of a recall-to-reject explanation of font ef-fects on false recognition. However, a weaker versionof the recall-to-reject hypothesis could claim that theretrieval of any study item in the unique conditioncan be used as a basis for rejection of a lure, and notsimply the retrieval of a related item. In Experiment1, lure items whose corresponding study items werepresented in the unique and correlated font conditionswere always tested in a font that was also assigned tothe unique and correlated font conditions, respectively.heuristic hypothesis.

Method

Participants

Thirty-two students enrolled in introduction to psy-chology at Middlebury College participated as part ofa research appreciation requirement.

Materials, design, and procedure

The materials, design, and procedure for Experiment2 were identical to Experiment 1, with the exception thatitems presented in the mismatch test condition were allpresented in a common, everyday font (Helvetica).

Results and discussion

The top half of Table 2 presents hits and lure falsealarms as a function of Font Condition, Study-TestMatch, and List Composition, while the bottom halfof Table 2 presents new item and new lure false alarmsas a function of Font Condition and List Composition.

MSE = .003. This interaction indicates that the hit rateadvantage for the correlated condition was restrictedto pure study lists (M = .91 and M = .86 for the corre-lated and unique conditions in pure study lists;M = .89 andM = .89 for the correlated and unique con-ditions in mixed study lists). The second interaction wasbetween Font Condition and Study-Test Match,F(1,31) = 4.23, MSE = .004. This interaction indicatesthat the size of the font matching eect was slightly larg-er in the unique condition (M = .95 vs. M = .80 in thematch and mismatch conditions, respectively) than inthe correlated condition (M = .96 vs. M = .84 in thematch and mismatch conditions, respectively). Finally,the three-way interaction between Font Condition,Study-Test Match, and List Composition was signi-cant, F(1,31) = 9.59, MSE = .003. This interaction sug-gests that the Font Condition X Study-Test Matchinteraction occurred in mixed lists (F(1,31) = 11.95,MSE = .004), but not in pure lists (F < 1).

The analysis of false alarms to lure items indicated a

Exp

Mism

.82 (

.86 (

.64 (

.64 (

.08 (

.08 (

.13 (

.12 (

J. Arndt / Journal of Memory and Language 54 (2006) 113130 121Hits and lure false alarms were analyzed with separate2 2 2 within-subjects ANOVAs. New item falsealarms and new lure false alarms were analyzed withseparate 2 2 within-subjects ANOVAs.

The analysis of hit rates indicated that the main eectof Font Condition was signicant, F(1,31) = 12.80,MSE = .003, suggesting that hits were higher in the cor-related condition (M = .90) than in the unique condition(M = .88). There was also a main eect of Study-TestMatch, F(1,31) = 79.60, MSE = .016, indicating thathits were higher when the font matched between studyand test (M = .96) than when the font did not match be-tween study and test (M = .82). Further, there werethree signicant interactions. The rst was between FontCondition and List Composition, F(1,31) = 8.07,

Table 2Hits, false alarms to lure items, and false alarms to new items in

List Composition

Correlated

Match

Hits

Pure .96 (.008)Mixed .96 (.011)

Lure false alarms

Pure .70 (.048)Mixed .72 (.041)

New false alarms

Pure N/AMixed N/A

New lure false alarms

Pure N/AMixed N/A

Standard error of the mean in parenthesis.main eect of Font Condition, F(1,31) = 96.73,MSE = .025, where lure false alarms were higher inthe correlated condition (M = .67) than in the uniquecondition (M = .48). There was also a reliable main ef-fect of Study-Test Match, F(1,31) = 5.31, MSE = .071,where lure false alarms were higher when lure items werepresented in a font presented at study (M = .61) than ina font not presented at study (M = .54). No other maineects or interactions were signicant (all F < 1). Final-ly, the analysis of new item and new lure false alarmsfailed to produce any signicant eects (all F < 1).

In many respects, the results of this experiment repli-cate those of Experiment 1, and again pose problems forthe recall-to-reject and distinctiveness heuristic explana-tions. Most fundamentally, the size of the eect of Font

eriment 2

Font Condition

Unique

atch Match Mismatch

.025) .96 (.013) .82 (.020)

.021) .95 (.011) .77 (.027)

.040) .53 (.040) .43 (.043)

.041) .51 (.053) .44 (.046)

.014) N/A .09 (.023)

.015) N/A .08 (.013)

.025) N/A .12 (.024)

.022) N/A .14 (.022)

cue the retrieval of related study items, producing areduction in false recognition. Finally, the similarity in

122 J. Arndt / Journal of Memory and Language 54 (2006) 113130the eects of Font Condition across mixed and pure listsis inconsistent with the distinctiveness heuristic explana-tion, replicating the results of Arndt and Reder (2003) aswell as Experiment 1. The similarity in the size of the ef-fects of Font Condition on false recognition across levelsof Study-Test Match and List Composition are, howev-er, within the explanatory domain of an account thatproposes that font condition alters the extent to whichlure items representations are activated at encoding inthe unique and correlated font conditions.

One result from this experiment that does not twithin the explanation oered by a strict encoding-basedview is that levels of lure false recognition were loweroverall when lure items were presented in the mismatchcondition. This result suggests the intriguing possibilitythat presenting lure items in a font that was experiencedat study actually increases illusory recognition in theDRM paradigm, rather than providing a useful basisfor discriminating lures from studied items. One expla-nation for this result is that font information encoun-tered at study becomes associated with lure itemrepresentations, producing a stronger match betweenlure item representations and the contents of memorywhen font information is re-presented at test. Animplication of this hypothesis will be evaluated in Exper-iment 4. In the next experiment, encoding-based andretrieval-based explanations of false memory are furtherevaluated via the use of a retrieval time manipulation(Benjamin, 2001; Heit et al., 2004).

Experiment 3

In this experiment, encoding-based and retrieval-based explanations of the inuence of Font Conditionon false recognition were contrasted via the use of aretrieval time manipulation (e.g., Benjamin, 2001; Heitet al., 2004). The logic of using a retrieval time manipu-lation to test encoding- and retrieval-based explanationsof false memory is that activation levels of lure items willbe available at an early stage in retrieval, while retrieval-based monitoring processes should only be available lat-Condition on false recognition was not altered bymanipulating the font information available at retrieval.This result is inconsistent with the recall-to-rejecthypothesis, which would expect presenting test items ina font not shown at study to reduce or eliminate the dif-ferences in false recognition across the font conditions.Further, presenting test items in a non-studied font pro-duced a reduction in false recognition, rather than an in-crease in false recognition. This result is the opposite ofthat expected by the recall-to-reject explanation, whichwould expect the presence of distinctive fonts at test toer in retrieval because they require deliberate, strategicmechanisms (Heit et al., 2004; Johnson et al., 1993;Roediger et al., 2001). In support of this assumption,Heit et al. (2004) demonstrated that lure false alarms de-creased as retrieval time increased, a result that is consis-tent with improved ability to use monitoring processesas retrieval time increased. Further support for the ideathat activation and monitoring processes are separableby a retrieval time manipulation comes from Benjamin(2001). Benjamin (2001) showed that when retrievalwas speeded, lure false alarms increased with study itemrepetition, while in unspeeded retrieval conditions, lurefalse alarms decreased with study repetition. These tworesults support the idea that there are two separate pro-cesses that inuence lure false alarm rates, and that theyare available at dierent time points in retrieval.

Given the above assumptions about the availabilityof activation and monitoring processes at dierent timepoints in retrieval, encoding- and retrieval-basedhypotheses provide dierent expectations for how fontcondition should inuence speeded and unspeeded rec-ognition decisions. Specically, an encoding-basedhypothesis suggests that the inuence of font conditionon false recognition should primarily inuence activa-tion processes. Therefore, the eects of font conditionshould be evident in the early stages of retrieval, andof comparable magnitude at both fast and slow retrievaldurations. In contrast, a retrieval-based view would sug-gest that the lower level of false memory in the uniquecondition should only be present at slower retrievaldurations. In addition, standard retrieval-based theories,such as the distinctiveness heuristic, disqualifying moni-toring, and other recall-to-reject processes tend to viewmonitoring as a memory editing process. Thus, thesehypotheses would suggest that the unique conditionshould produce lower levels of false recognition inunspeeded retrieval conditions relative to speededretrieval conditions, in addition to the expectation ofgeneral monitoring-based explanations that the eectsof font condition should be larger in unspeeded retrievalconditions compared to speeded retrieval conditions.

Method

Participants

Thirty-two students enrolled in introduction to psy-chology at Middlebury College participated as part ofa research appreciation requirement.

Materials, design, and procedure

The materials, design, and procedure for Experiment3 were identical to the mismatch condition of Experi-ment 2, with four exceptions. First, test duration wasmanipulated within subjects such that participantsresponded to half of the items in a given test list within750 ms (referred to as the speeded retrieval condition be-

low), while participants were given an unlimited amount

of time to make a recognition decision for the other halfof the items in each test list (referred to as the unspeededretrieval condition below; Benjamin, 2001). Second, testduration was blocked within each test list. Third, testduration was counterbalanced such that half of the par-ticipants received the speeded retrieval condition fol-lowed by the unspeeded retrieval condition on eachrecognition memory test, and half of the participants re-ceived the unspeeded retrieval condition followed by thespeeded retrieval condition on each recognition memorytest. Fourth, the list composition factor was not manip-ulated. All study lists were presented in a mixed fashion.

Results and discussion

The results of Experiment 3 are presented in Table 3in terms of hits, false alarms to lure items, false alarms tonew items, and false alarms to new lures. Further, thedata from Experiment 3 are also presented in terms of

J. Arndt / Journal of Memory and Language 54 (2006) 113130 123two measures of recognition memory discrimination,which will be referred to here as oldnew A 0 and lureA 0. Oldnew A 0 is a standard computation of discrimina-bility used in recognition memory studies (Donaldson,1992). Lure A 0 is a computation intended to reect thelevel of error-proneness to lure items in the speededand unspeeded conditions. Specically, lure A 0 was com-puted by treating the lure false alarm rate as the hit rate,and the false alarm rate for new lures as the false alarmrate (see Arndt & Hirshman, 1998; Arndt & Reder,2003; Schacter et al., 1999 for similar computations).Fig. 2 presents oldnew A 0, while Fig. 3 presents lure A 0.

The reason for analyzing the data in terms of bothraw response probabilities as well as a derived measureof discriminability is that retrieval duration producedoverall dierences in false alarms to new items andnew lures, in addition to dierences in hits and lure false

Table 3Hits, false alarms to lure items, and false alarms to new items inExperiment 3

Retrieval Speed Font Condition

Correlated Unique

Hits

Speeded .65 (.026) .64 (.031)Unspeeded .83 (.020) .79 (.024)

Lure false alarms

Speeded .49 (.035) .45 (.039)Unspeeded .53 (.038) .36 (.033)

Retrieval Speed

Speeded Unspeeded

Unrelated item false alarms

New theme items .18 (.020) .11 (.014)New lure items .24 (.023) .17 (.020)

Standard error of the mean in parenthesis.alarms. In instances when there are condition dierencesin baseline false alarm rates (e.g., new lure false alarms)as well as lure false alarm rates, it is dicult to interpretthe basis for the decline in lure false alarm rates as achange in memory strength, because that decline couldalso reect dierences in decision criterion across condi-tions (Green & Swets, 1966). Rotello and Heit (1999)noted a similar point in their studies of the use of re-call-to-reject processing to reduce false memory to simi-lar lures. Specically, Rotello and Heit (1999) arguedthat in order for one to nd evidence of recall-to-rejectprocessing for similar foils (e.g., plurality-reversed luresor similar pseudowords), one must not only demonstratelower levels of false alarms to similar foils, but also thatthe change in foil endorsement can not be explained bysimple changes in decision criteria that participants useacross retrieval durations. Thus, Rotello and Heit(1999) argued that it was essential to use a derived mea-sure of memory strength or discrimination such as A 0,which provides a measure of memory strength indepen-dent of changes in recognition decision criterion changesacross conditions. Therefore, lure A 0 can be interpretedas a measure of the amount of evidence in memory forlure false recognition that is not inuenced by decisioncriterion dierences across retrieval speed conditions.

The analysis of hits indicated a main eect of TestDuration, F(1,31) = 80.58, MSE = .010, indicating thathits were higher in the unspeeded condition (M = .81)compared to the speeded condition (M = .65). Further,the interaction between Font Condition and Test Dura-tion approached signicance, F(1,31) = 3.95, MSE =.003, p < .06. This trend results from the tendency forthe hit rate dierence between the unique and correlatedconditions to occur in the unspeeded retrieval condition(M = .83 and M = .79 in the correlated and uniquecondition, respectively), but not in the speeded retrievalcondition (M = .645 andM = .644 in the correlated andunique condition, respectively).

The analysis of lure false alarms indicated a main ef-fect of Font Condition, F(1,31) = 19.22. MSE = .020,where false alarms were higher in the correlated condi-tion (M = .51) than in the unique condition (M = .40).Further, this analysis indicated an interaction betweenFont Condition and Test Duration, F(1,31) = 6.76,MSE = .022. This interaction occurred because the ef-fect of Font Condition on lure false alarms was largerin the unspeeded retrieval condition (M = .53 andM = .36 in the correlated and unique condition, respec-tively, t(31) = 5.11), than in the speeded retrieval condi-tion (M = .49 and M = .45 in the correlated and uniquecondition, respectively, t(31) = 1.08, ns). Importantly,lure false alarms decreased when retrieval was unspee-ded (M = .36) relative to when it was speeded(M = .45) in the unique condition (t(31) = 2.45), butshowed a trend toward increasing when retrieval was

unspeeded (M = .53) relative to when retrieval was

124 J. Arndt / Journal of Memory and Language 54 (2006) 113130speeded (M = .49) in the correlated condition (t(31) =1.33, ns). Finally, as noted above, false alarms to newitems and new lures were reliably inuenced by TestDuration (t(31) = 4.16 and 3.44 for new items and newlures, respectively). For both false alarm eects, falsealarms were higher in the speeded condition comparedto the unspeeded condition. Because of these dierences,true and false recognition were also analyzed withA 0 and lure A 0 in order to assess the extent to whichinformation in memory supports accurate (A 0) and false(lure A 0) recognition.

The analysis of oldnew A 0 indicated a main eect ofTest Duration, F(1,31) = 79.20, MSE = .003, whereoldnew A 0 was greater in the unspeeded condition(M = .91) than in the speeded condition (M = .82). Nei-ther the main eect of Font Condition nor the interac-tion of Font Condition and Test Duration reached the

Fig. 2. A 0 in Experiment 3 as a function of Font Condition and

Fig. 3. Lure A 0 in Experiment 3 as a function of Font Condition ancriterion of signicance (both F < 2.0). The analysis oflure A 0 revealed reliable main eects of Font Condition,F(1,31) = 23.37, MSE = .009, and Test Duration,F(1,31) = 6.35, MSE = .013, as well as an interactionbetween Font Condition and Test DurationF(1,31) = 5.24, MSE = .008. As can be seen in Fig. 3,the interaction between Font Condition and Test Dura-tion arises because the dierence between the unique andcorrelated conditions was greater when retrieval wasunspeeded (mean dierence = .12, t(31) = 5.01, p < .05)relative to when retrieval was speeded (mean dier-ence = .04, t(31) = 1.84, p > .05). Importantly, lure itemerrors increased when retrieval was unspeeded relative towhen it was speeded in the correlated condition(t(31) = 3.48), but not in the unique condition(t(31) = .50). Thus, although the analyses of lure A 0

and lure false alarms both showed an interaction

Retrieval Time. Standard error of the mean in parenthesis.

d Retrieval Time. Standard error of the mean in parenthesis.

all of a lures associates. In contrast, fonts in the uniquecondition were only experienced a single time. Thus, lure

J. Arndt / Journal of Memory and Language 54 (2006) 113130 125between Font Condition and Test Duration, the specif-ics of that interaction diered across the two measuresof false memory. Specically, analyses of lure falsealarms in the unique font condition showed a reductionin false alarms in the unspeeded condition compared tothe speeded condition, as well as a trend toward in-creased false alarms in the correlated font conditionwhen retrieval was unspeeded compared to speeded. Incontrast, analyses of lure A 0 showed an increase in falserecognition in the correlated condition when retrievalwas unspeeded compared to when it was speeded, a re-sult that did not occur in the unique font condition.

The results of this experiment contradict bothencoding-based explanations and editing-based retrievalexplanations of the dierence between the unique andcorrelated font conditions. Encoding-based views expect-ed that font condition would lead to dierent levels oflure item activation, which should have been evidenteven when retrieval was speeded. Editing-based retrievalexplanations expected the interaction between font con-dition and retrieval time for lure A 0, but anticipated adierent form of that interaction, where additionalretrieval time would allow for reductions in false recogni-tion of test items in the unique font condition. Althoughsuch a reduction was found in analyses of lure false alarmrates, it is likely that those dierences reect generalchanges in recognition memory decision criteria acrossretrieval durations and not reductions in the amount ofevidence in memory supporting false recognition of lures.When general criterion dierences are taken in toaccount by computing a measure of the evidence in mem-ory supporting lure false recognition, lure A 0, the patternof results suggests that false recognition was actuallyincreased in the correlated condition when participantswere given time to use strategic monitoring processes.

Adopting the general characterization of retrievaldurations provided above, the results of this study areconsistent with the notion that the unique font conditionproduces lower levels of lure item errors because the cor-related condition impairs monitoring processes, ratherthan the unique condition producing lower levels of lureitem activation or improved retrieval-based monitoring.While this may seem to be a somewhat counterintuitiveinterpretation, it does fall within the explanatory boundsof two major characterizations of monitoring processesin the DRM paradigm, activation-monitoring theory(Roediger et al., 2001), and the Source-MonitoringFramework (Johnson et al., 1993). Monitoring notionsencapsulated in these frameworks suggest that whilemonitoring is generally associated with reductions inmemory errors, under conditions where unstudied itemsare associated with presentation characteristics ofstudied items, the ability to employ monitoring will becompromised. Specic to the DRM paradigm, activa-tion-monitoring theory suggests that conditions that

produce activation of a lures representation may alsoitem representations would become associated with mul-tiple dierent visual formats, and in a weaker form thanthe repeated visual format experienced in the correlatedcondition. Then, on a recognition memory test, onemust assume that participants attempt to utilize the abil-ity to retrieve details of study items presentation in aneort to decide if a test item was encountered at study.Participants would be more likely to retrieve physicalcharacteristics associated with lure items in the correlat-ed condition, which in turn would lead them to make agreater number of false alarms to lure items in the cor-related condition. Critically, the relative dierence infalse alarm rates between the unique and correlated con-ditions would be stronger after participants had su-cient time to search their memories and retrievephysical characteristics associated with lure items. Fur-ther, the dierence between the unique and correlatedconditions would be expected to emerge as an increasein false recognition in the correlated condition, ratherthan a decrease in false recognition in the unique condi-tion, precisely the result found in this experiment. Exper-iment 4 provides a further test of the notion that lurerepresentations become associated with presentationcharacteristics of their studied associates, a claim centralto the explanation of the results of Experiments 2 and 3.

Experiment 4

This study examined an implication of the explana-tions advanced for the results of Experiments 2 and 3.Specically, in both of those studies, there were resultsthat are consistent with the idea that the presentationcharacteristics of study items become associated withlure item representations. This association in turn ren-ders participants more error-prone when the lure is pre-sented in a visual format encountered during study andwhen an association between a font and a lure should bestronger, such as when a single font was used to presentall of the associates of a lure. This conjecture providescause the lure to become associated with the presenta-tion characteristics of its studied associates. Thus, atretrieval, monitoring processes can make lure itemsharder to reject, given the proviso that the lures have be-come associated with characteristics of the studyexperience.

This general notion can be applied to the results ofExperiment 3 by assuming that participants associatemore generalized details of study items presentationwith lure items in the correlated condition than in theunique condition. This assumption seems reasonablegiven that fonts in the correlated condition were experi-enced repeatedly and the same font was used to presentan explanation of both the nding in Experiment 2 that

126 J. Arndt / Journal of Memory and Language 54 (2006) 113130lure false alarms were higher when a font experienced atstudy was re-presented at retrieval, and the nding fromExperiment 3 that the eects of font condition on falserecognition only emerged late in retrieval, results thatchallenge a pure encoding-based view of the eects offont condition on false memory.

Experiment 4 seeks to further test the claim that fontinformation becomes associated with lure item represen-tations, and that the inuence of that association on lurefalse recognition involves the contribution of retrievalprocesses. The test provided in Experiment 4 was de-rived from Roediger et al.s (2001) activation-monitor-ing theory because it describes how encoding andretrieval processes interact in the DRM paradigm as wellas the conditions under which an unstudied lure itemshould become associated with the presentation charac-teristics of its studied associates. Specically, activation-monitoring theory suggests that lure item activationplays a role in determining the extent to which lures be-come associated with presentation characteristics oftheir studied associates, such that the more a lures rep-resentation is activated at encoding, the greater the like-lihood that its representation becomes associated withpresentation characteristics of the associates that pro-duced that activation. Thus, this explanation suggeststhat if a font is associated in a systematic way with a fac-tor that inuences lure activation, lure representationsshould be more likely to become associated with thecharacteristics of that font.

This theory was tested by utilizing two fonts to pres-ent the study items related to each lureone font wasused to present study items that had a high degree ofmean backward association strength (MBAS) to thelure item, and a second font was used to present thestudy items with a low degree of MBAS to the lureitem. On the recognition test, lure items were presentedin either the font used to present its high-MBAS asso-ciates or its low-MBAS associates. If the level of acti-vation of lure items is important for determining thelikelihood that lure items become associated with pre-sentation characteristics of their studied associates,the data should reveal that presenting a lure item ina font used to present the high-MBAS study items willlead to greater levels of false recognition than present-ing a lure item in a font used to present the low-MBASstudy items. This result would be consistent with acti-vation-monitoring theory, and would also further clar-ify the role that encoding and retrieval processes playin the production of false recognition. Specically, sucha result would suggest that the activation level of alures representation helps to determine the extent towhich a lure is associated with the presentation formatof study items. This association, when retrieved at test,should increase the likelihood that participants believelures were encountered at study, increasing false

recognition.Method

Participants

Thirty-two students enrolled in introduction to psy-chology at Middlebury College participated as part ofa research appreciation requirement.

Materials and design

The materials for Experiment 4 were 48 themes takenfrom Nelson et al. (1998). For each theme, four high-MBAS and four low-MBAS associates were selected tobe presented in the study list. High-MBAS study itemswere dened as the four strongest associates of a lure,with a minimum probability of producing the lure itemin free association of .15 (mean MBAS = .585;range = .168.960). Low-MBAS study items were de-ned as four associates of a lure with a probability ofproducing the lure item in free association between .02and .10 (mean MBAS = .063; range = .036.099). Thefonts used to present study and test items were 96 ofthe same fonts used in Experiments 13.

For each participant, the 48 themes were broken in totwo sets of 24 themes for presentation in two study lists,producing a study list length of 192 (24 themes 8 itemsper theme). In each study list, high- and low-MBAS itemswere presented blocked by theme. Within each theme, allhigh-MBAS items were presented in a single font and alllow-MBAS items were also presented in a single font,but one that diered from the font used to present thehigh-MBAS items. Further, the font used to presenthigh-MBAS and low-MBAS items diered across themes.Thus, the total number of fonts utilized to present studyitems was 96 (2 fonts per theme 48 themes).

Test lists consisted of one high-MBAS and one low-MBAS study item per theme, the lure item for eachtheme, and 24 new items that were unrelated to any ofthe items presented at study, producing a test list lengthof 96 (24 high-MBAS study items, 24 low-MBAS studyitems, 24 lure items, and 24 new items). Unrelated newitems had a word frequency count between 50 and 150(Kucera & Francis, 1967). Studied items were presentedin the same font utilized to present them at study. Lureitems were presented in either the font utilized to presenttheir high-MBAS associates or their low-MBAS associ-ates. Unrelated new items were presented equally oftenin a font used to present high-MBAS study items andlow-MBAS study items. Stimulus counterbalancing en-sured that each lure item was presented equally oftenin the font used to display its high-MBAS and low-MBAS associates at study, and that each font was usedto present high-MBAS and low-MBAS study itemsequally often across subjects.

Procedure

The procedure was identical to that of the rst two

experiments, as well as the unspeeded condition of

J. Arndt / Journal of Memory and Language 54 (2006) 113130 127Experiment 3, with the exception that participants wereonly run through two study-test cycles rather than four.

Results and discussion

Hits to studied items did not dier as a function ofMBAS, t(31) = 0.97 (M = 0.91 and M = 0.91 for high-and low-MBAS study items, respectively). In addition,false alarms to new items did not dier as a functionof the font shown at test, t(31) = 0.787 (M = 0.14 andM = 0.12 for high-MBAS fonts and low-MBAS fonts,respectively). In contrast, false alarms to lure items werehigher when lures were presented in the font used topresent its high-MBAS associates at study, t(31) = 3.46(M = 0.40 and M = 0.33 for lure items presented inhigh-MBAS fonts and low-MBAS fonts, respectively).

Consistent with arguments outlined in activation-monitoring theory, and the explanation proposed forunexpected results from Experiments 2 and 3, the resultsof this experiment suggest that the level of activation oflure items produced by their associates at encodingdetermines the extent to which lure representations be-come associated with presentation features encounteredduring encoding. When lures were tested in a font thatwas used to present their high-MBAS associates, sub-jects showed increased levels of false recognition, sug-gesting that font information becomes associated withlure item representations despite the fact that lures werenever presented at encoding. This result is inconsistentwith strict encoding-based views of false recognition,which would not expect the font reinstated at retrievalto make a dierence in the level of false recognition ob-served. However, theories that include the contributionof retrieval processes are able to explain this result byclaiming that participants are retrieving associationsthat were built during the study episode, and that theassociation between a lure item and the presentationcharacteristics of studied associates is stronger for theassociates that provide the majority of activation ofthe lures representation (i.e., high-MBAS associates).

General discussion

The results of these studies further clarify the rolethat encoding- and retrieval-based processes play inthe production of false recognition. In particular, thecombined implications of these four studies suggest theimportance of considering the role of both encoding-based and retrieval-based processes in fully understand-ing the inuence of distinctiveness on false recognition.Thus, these studies suggest that pure versions ofencoding-based theories and retrieval-based theoriesencounter diculty explaining the full range of false rec-ognition phenomena that result from the presentation ofstudy items in a distinctive format. Specically, encod-

ing-based theories fail to account for the observationsthat (1) reinstating studied fonts at retrieval producedhigher levels of false recognition than presenting testitems in previously unseen fonts (Experiment 2); (2) dif-ferences in false recognition between the unique and cor-related conditions were greater when subjects were givenunlimited retrieval time (Experiment 3); and (3) present-ing lure items in a font utilized to present its high-MBASassociates led to higher levels of false recognition thanpresenting lure items in a font utilized to present itslow-MBAS associates (Experiment 4). Strictly retriev-al-based theories, on the other hand, fail to accountfor the observations that presenting lure items in a fontnot used to present its studied associates did not increasefalse recognition and also did not alter the dierence infalse recognition between the unique and correlated con-ditions (Experiments 1 and 2). Further, without the con-sideration of encoding-based contributions to falserecognition, retrieval-based theories are unable toexplain the dierence in false recognition betweenhigh-MBAS and low-MBAS fonts on lure false alarmsobserved in Experiment 4. Finally, several prominentretrieval-based theories that suggest retrieval processesare used to reduce false memory encounter dicultyexplaining the results of Experiment 3. In particular,these views are unable to explain why false recognitionincreased in the correlated condition and did notdecrease in the unique condition when participants wereallotted more retrieval time, a condition where theyshould be better able to utilize retrieval monitoringprocesses (Benjamin, 2001; Heit et al., 2004).

Although pure encoding or retrieval perspectivesencounter diculty explaining all of the data from thesestudies, theories that emphasize the combined inuenceof encoding and retrieval processes on false memory,such as activation-monitoring theory (McDermott &Watson, 2001; Roediger et al., 2001) are capable ofaccounting for the range of results reported here. Acti-vation-monitoring theorys account of these data sug-gests that during encoding, lure activation generatesassociations between the fonts used to present studyitems related to a lure and the lures representation.On a later recognition memory test, if the perceptual fea-tures associated with a lures representation are used topresent the lure item, the perceptual overlap betweenthe test cue and the lures representation can produce in-creased levels of false recognition, as evidenced by theresults of Experiments 2 and 4. Similarly, even in the ab-sence of lures being presented in a font that matches onepresented during encoding, participants may engageretrieval monitoring processes in order to search memo-ry for characteristics of the study episode such as thefont in which a study item was presented. When they re-trieve evidence that a test item was associated with anunusual-looking font, they will tend to endorse that testitem as studied. While the use of monitoring processes is

generally a benecial technique for ensuring accuracy of

ble by the use of a distinctiveness heuristic, an hypoth-esis that has found success explaining reductions in

128 J. Arndt / Journal of Memory and Language 54 (2006) 113130ones memory, when a lures representation was alsoassociated with presentation characteristics of the studyepisode, discrimination between studied lure items isrendered ineective. Indeed, as evidenced by the resultsof the third experiment reported here, monitoring pro-cesses can contribute to the exacerbation of memory er-rors (see also Hicks & Marsh, 2001).

Although it is not common to view monitoringprocesses as having a negative inuence on memoryaccuracy, the characterization of monitoring processesoutlined above may account for results in the DRMliterature beyond those reported in these experiments.First, when subjective states of recollection and famil-iarity are measured (e.g., with the remember-knowparadigm; Tulving, 1985), subjects tend to claim thatthey recollect the presentation of lure items on thestudy list (Roediger & McDermott, 1995). While thereare accounts of this result that attribute it to represen-tations that are fundamentally dierent from thosesupporting recollection of study items, (e.g., verystrong gist representations; Brainerd, Wright, Reyna,& Mojardin, 2001), the monitoring-based view sug-gests that the basis for the subjective experience ofrecollection is comparable for both studied items andluresretrieval of source information from the studylist. Second, the emphasis on monitoring processesmay provide an explanation of why post-encodingwarnings are not fully eective in reducing false mem-ory in the DRM paradigm (e.g., Gallo et al., 2001).Assuming that post-encoding warnings sometimeslead subjects to use monitoring processes, this viewwould suggest that post-encoding instructions will beunable to reduce lure false memory to the extent thatlure item activation produces associations betweenlures representations and study list presentationcharacteristics.

In addition to general support for the activation-monitoring view of false memory, the results of thesestudies reinforce and expand upon research suggestingthat source monitoring may not always reduce falsememory (Hicks & Marsh, 1999, 2001) as is typically as-sumed in theories highlighting the relationship betweenmonitoring processes and false recognition (Brainerdet al., 2003; Dodson & Schacter, 2001, 2002a, 2002b;Gallo, 2004; Rotello et al., 2000; Schacter et al., 1999).Indeed, some evidence suggests that using a standardsource monitoring test instead of a recognition memorytest impairs the ability to distinguish between studieditems and related, but unstudied, lure items (Hicks &Marsh, 2001; but see Mather, Henkel, & Johnson,1997). The results of the third experiment reported hereare consistent with the view that monitoring-based inu-ences can increase false recognition of lure items in theDRM paradigm, particularly when the visual formatused to present the lure item was presented with many

of the lures associates at study.false memory in the DRM paradigm. Thus, the useof fonts to manipulate stimulus distinctiveness appearsto inuence false memory via a dierent mechanism orset of mechanisms than other methods of producingstimulus distinctiveness. The reason for this dierenceis not immediately apparent, although one possibilityis that perceptual details that are an intrinsic part ofstimulus presentation (e.g., a font or voice; Roediger,McDermott, Pisoni, & Gallo, 2004) are easier for sub-jects to falsely integrate with lure representations thanwhen study items are associated with mental operations(e.g., saying words aloud; Dodson & Schacter, 2001) orpresented as complex stimuli such as pictures (Schacteret al., 1999).

In closing, three particular points warrant empha-sizing. First, these studies add to the growing litera-ture showing that a complete understanding of falserecognition in the DRM paradigm requires the consid-eration of both encoding and retrieval processes (e.g.,Benjamin, 2001; Gallo et al., 2001; Hege & Dodson,2004; Roediger et al., 2001). Thus, a complete under-standing of the nature of false memory may be bestserved by studying the role that encoding and retrievalprocesses play in the production of false memory,both individually and jointly. Second, these studiessuggest that perceptual features encountered at encod-ing, even salient visual information such as the unusu-al fonts used in these experiments, can becomeassociated with lure item representations, despite thefact that lures are never visually perceived duringstudy. Third, these results suggest the intriguing possi-bility that not only can monitoring processes fail toaid rejection of lure items (Hicks & Marsh, 1999),but that the use of monitoring processes may actuallyimpair the ability to reduce memory errors (Hicks &Marsh, 2001). This impairment will be exacerbatedwhen study conditions are conducive to the generationof associations between unstudied items and featuresof an encoding episode (Hicks & Marsh, 2001; John-son et al., 1993; Roediger et al., 2001), which not onlyrenders monitoring processes ineective, but may alsomislead subjects to making an increased number ofmemory errors.

Acknowledgment

I thank Jacob Carney, Karen Lee, Ashley Osborne,Meghan Stone, and John Carney for assistance withOne notable dimension of the results of these stud-ies, as well as those reported by Arndt and Reder(2003) is that the eects of font manipulations on falsememory in the DRM paradigm are not readily explica-data collection.

J. Arndt / Journal of Memory and Language 54 (2006) 113130 129References

Arndt, J., & Hirshman, E. (1998). True and false recognition inMINERVA2: Explanations from a global-matching per-spective. Journal of Memory and Language, 39, 371391.

Arndt, J., & Reder, L. M. (2003). The eect of distinctive visualinformation on false recognition. Journal of Memory andLanguage, 48, 115.

Benjamin, A. S. (2001). On the dual eects of repetition on falserecognition. Journal of Experimental Psychology: Learning,Memory, and Cognition, 27, 941947.

Brainerd, C. J., Reyna, V. F., Wright, R., & Mojardin, A. H.(2003). Recollection rejection: False-memory editing inchildren and adults. Psychological Review, 110, 762784.

Brainerd, C. J., Wright, R., Reyna, V. F., & Mojardin, A. H.(2001). Conjoint recognition and phantom recollection.Journal of Experimental Psychology: Learning, Memory, and

Cognition, 27, 307327.Deese, J. (1959). On the prediction of occurrence of particular

verbal intrusions in immediate recall. Journal of Experimen-tal Psychology, 58, 1722.

Dodson, C. S., & Schacter, D. L. (2001). If I had said it Iwould have remembered it: Reducing false memories witha distinctiveness heuristic. Psychonomic Bulletin & Review,8, 155161.

Dodson, C. S., & Schacter, D. L. (2002a). When falserecognition meets metacognition: The distinctiveness heu-ristic. Journal of Memory and Language, 46, 782803.doi:10.1006/jmla.2001.2882.

Dodson, C. S., & Schacter, D. L. (2002b). Aging and strategicretrieval processes: Reducing false memories with a distinc-tiveness heuristic. Psychology and Aging, 17, 405415.

Donaldson, W. (1992). Measuring recognition memory. Journalof Experimental Psychology: General, 121, 275277.

Gallo, D. A. (2004). Using recall to reduce false recognition:Diagnostic and disqualifying monitoring. Journal of Exper-imental Psychology: Learning, Memory, and Cognition, 30,120128.

Gallo, D. A., & Roediger, H. L. III, (2002). Variability amongword lists in evoking memory illusions: Evidence forassociative activation and monitoring. Journal of Memoryand Language, 47, 469497.

Gallo, D. A., Roediger, H. L., III, & McDermott, K. B. (2001).Associative false recognition occurs without strategic crite-rion shifts. Psychonomic Bulletin & Review, 8, 579586.

Gallo, D. A., Weiss, J. A., & Schacter, D. L. (2004). Reducingfalse recognition with criterial recollection tests: Distinc-tiveness heuristic versus criterion shifts. Journal of Memoryand Language, 51, 473493.

Green, D. M., & Swets, J. A. (1966). Signal detection theory andpsychophysics. New York: Wiley.

Hicks, J. L., & Marsh, R. (1999). Attempts to reduce theincidence of false recall with source monitoring. Journal ofExperimental Psychology: Learning, Memory, and Cogni-

tion, 25, 11951209.Hicks, J. L., & Marsh, R. (2001). False recognition occurs more

frequently during source identication than during oldnewrecognition. Journal of Experimental Psychology: Learning,Memory, and Cognition, 27, 375383.

Hirshman, E. (1995). Decision processes in recognition mem-ory: Criterion shifts and the list-strength paradigm. Journalof Experimental Psychology: Learning, Memory and Cogni-

tion, 21, 302313.Heit, E., Brockdor, N., & Lamberts, K. (2004). Strategic

processes in false recognition memory. Psychonomic Bulletin& Review, 11, 380386.

Hege, A., & Dodson, C. S. (2004). Why distinctive informationreduces false memories: Evidence for both impoverishedrelational-encoding and distinctiveness heuristic accounts.Journal of Experimental Psychology: Learning, Memory, and

Cognition, 30, 787795.Hunt, R. R., & Einstein, G. (1981). Relational and item-specic

information in memory. Journal of Verbal Learning andVerbal Behavior, 20, 497514.

Johnson, M. K., Hashtroudi, S., & Lindsay, D. S. (1993).Source monitoring. Psychological Bulletin, 114, 328.

Kucera, H., & Francis, W. N. (1967). Computational analysis ofpresent day American English. Providence, RI: BrownUniversity Press.

Mather, M., Henkel, L. A., & Johnson, M. K. (1997).Evaluating characteristics of false memories: Remember/know judgments and memory characteristics questionnairecompared. Memory & Cognition, 25, 826837.

McCabe, D. P., Presmanes, A. G., Robertson, C. L., & Smith,A. D. (2004). Item-specic processing reduces false memo-ries. Psychonomic Bulletin & Review, 11, 10741079.

McDermott, K. B., & Watson, J. M. (2001). The rise and fall offalse recall: The impact of presentation duration. Journal ofMemory and Language, 45, 160176.

Nelson, D. L., McEvoy, C. L., & Schreiber, T. A. (1998). TheUniversity of South Florida word association, rhyme, andword fragment norms. Available from http://www.usf.edu/FreeAssociation/.

Reder, L. M., Donavos, D., & Erickson, M. A. (2002).Perceptual match eects in direct tests of memory: Therole of contextual fan. Memory & Cognition, 30,312323.

Roediger, H. L., III, & McDermott, K. B. (1995). Creating falsememories: Remembering words not presented in lists.Journal of Experimental Psychology: Learning, Memory,

and Cognition, 21, 803814.Roediger, H. L., III, McDermott, K. B., Pisoni, D. P., & Gallo,

D. A. (2004). Illusory recollection of voices. Memory, 12,586602.

Roediger, H. L., III, Watson, J. M., McDermott, K. B., &Gallo, D. A. (2001). Factors that determine false recall: Amultiple regression analysis. Psychonomic Bulletin & Review,8, 385407.

Rotello, C. M., & Heit, E. (1999). Two-process models ofrecognition memory: Evidence for recall-to-reject? Journalof Memory and Language, 40, 432453.

Rotello, C. M., Macmillan, N. A., & Van Tassel, G. (2000).Recall to reject in recognition: Evidence from ROC curves.Journal of Memory and Language, 43, 6788.

Schacter, D. L., Cendan, D. L., Dodson, C. S., & Cliord, E. R.(2001). Retrieval conditions and false recognition: Testingthe distinc