Psychonomic Bulletin & Review 2001, 8 (3), 385-407 In 1894, in the first volume of Psychological Review, E. A. Kirkpatrick performed “an experimental study of memory,” in which he showed that items presented as vi- sual objects are better remembered than are words and that imagery aids retention of verbal materials. Those features of his paper have been cited, off and on, through- out the years. However, toward the end of the paper, Kirkpatrick reported briefly the results of a side experi- ment. It is worth quoting part of the relevant paragraph to help set the stage for the present inquiry: About a week previously in experimenting upon mental imagery I had pronounced to the students ten common words. Many of these were recalled and placed with the memory list. Again, it appears that when such words as “spool,” “thimble,” “knife,” were pronounced many stu- dents at once thought of “thread, “needle”, “fork,” which are so frequently associated with them. The result was that many gave those words as belonging to the list. This is an excellent illustration of how things suggested to a person by an experience may be honestly reported by him as part of the experience. (Kirkpatrick, 1894, p. 608) Kirkpatrick’s informal observations have recently been confirmed by many investigators (e.g., Roediger & McDermott, 1995), and false recall and false recognition owing to associative processes have become central top- ics in the field (see Roediger, McDermott, & Robinson, 1998, for a review). Roediger and McDermott’s (1995) research built on prior work by Deese (1959b). Deese presented subjects with 36 lists of 12 associatively re- lated words, in which the list words (e.g., shoe, hand, toe, kick, etc.) were all associates of a word not presented in the list ( foot, in this case), with a free recall test given after each list. For some of the lists, Deese (1959b) showed that subjects intruded the word associated to the list items at relatively high rates. However, Deese was in- terested in how associations affected recall (e.g., Deese, 1959a), not in false recall per se, and his paper and tech- nique were relatively neglected (see Bruce & Winograd, 1998). Many of the lists Deese created elicited no or very few intrusions, which served his needs perfectly well be- cause he was interested in how associative factors would affect both veridical and false recall (Deese, 1959a, 1959b). It is perhaps for this reason that Deese never made much of his intrusion findings. Indeed, in his 1965 book, The Structure of Associations in Language and Thought , which summarized much of his important work pub- lished over the previous 10 years, the intrusion paper (Deese, 1959b) was not even cited. 385 Copyright 2001 Psychonomic Society, Inc. This research was supported by a contract from the U.S. Office of Technical Services to H.L.R. and K.B.M., as well as NIMH Grant 1RO3 MH59034-01 to K.B.M. The authors thank Martha Storandt for advice regarding the regression analysis, Michael A. Stadler for providing nor- mative data, Douglas Nelson for advice in collecting our norms, David A. Balota for advice on various aspects of this research, Randy Buck- ner for calling our attention to Kirkpatrick’s work, and Cathy McEvoy, Douglas Nelson, and Tamaku Komatsu for providing additional data. Michael Cortese, Patrick Dolan, Stephanie Lee, Elizabeth Marsh, Michelle Meade, Douglas Nelson, and Valerie Reyna provided helpful comments. Correspondence concerning this article may be sent to H. L. Roediger III, Department of Psychology, Box 1125, Washington Uni- versity, One Brookings Drive, St. Louis, MO 63130-4899 (e-mail: [email protected]). Factors that determine false recall: A multiple regression analysis HENRY L. ROEDIGER III, JASON M. WATSON, KATHLEEN B. MCDERMOTT, and DAVID A. GALLO Washington University, St. Louis, Missouri In the Deese–Roediger–McDermott (DRM) paradigm, subjects study lists of words that are designed to elicit the recall of an associatively related critical item. The 55 lists we have developed provide lev- els of false recall ranging from .01 to .65, and understanding this variability should provide a key to un- derstanding this memory illusion. Using a simultaneous multiple regression analysis, we assessed the contribution of seven factors in creating false recall of critical items in the DRM paradigm. This analy- sis accounted for approximately 68% of the variance in false recall, with two main predictors: associa- tive connections from the study words to the critical item (r 51.73; semipartial r 51.60) and re- callability of the lists (r 5 .43; semipartial r 5 .34). Taken together, the variance in false recall captured by these predictors accounted for 84% of the variance that can be explained, given the relia- bility of the false recall measures (r 5 .90). Therefore, the results of this analysis strongly constrain the- ories of false memory in this paradigm, suggesting that at least two factors determine the propensity of DRM lists to elicit false recall. The results fit well within the theoretical framework postulating that both semantic activation of the critical item and strategic monitoring processes influence the proba- bility of false recall and false recognition in this paradigm.
23
Embed
Factors that determine false recall: A multiple regression ...
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Psychonomic Bulletin amp Review2001 8 (3) 385-407
In 1894 in the first volume of Psychological ReviewE A Kirkpatrick performed ldquoan experimental study ofmemoryrdquo in which he showed that items presented as vi-sual objects are better remembered than are words andthat imagery aids retention of verbal materials Thosefeatures of his paper have been cited off and on through-out the years However toward the end of the paperKirkpatrick reported briefly the results of a side experi-ment It is worth quoting part of the relevant paragraphto help set the stage for the present inquiry
About a week previously in experimenting upon mentalimagery I had pronounced to the students ten commonwords Many of these were recalled and placed with thememory list Again it appears that when such words asldquospoolrdquo ldquothimblerdquo ldquokniferdquo were pronounced many stu-dents at once thought of ldquothread ldquoneedlerdquo ldquoforkrdquo whichare so frequently associated with them The result was that
many gave those words as belonging to the list This is anexcellent illustration of how things suggested to a personby an experience may be honestly reported by him as partof the experience (Kirkpatrick 1894 p 608)
Kirkpatrickrsquos informal observations have recentlybeen confirmed by many investigators (eg Roediger ampMcDermott 1995) and false recall and false recognitionowing to associative processes have become central top-ics in the field (see Roediger McDermott amp Robinson1998 for a review) Roediger and McDermottrsquos (1995)research built on prior work by Deese (1959b) Deesepresented subjects with 36 lists of 12 associatively re-lated words in which the list words (eg shoe hand toekick etc) were all associates of a word not presented inthe list ( foot in this case) with a free recall test givenafter each list For some of the lists Deese (1959b)showed that subjects intruded the word associated to thelist items at relatively high rates However Deese was in-terested in how associations affected recall (eg Deese1959a) not in false recall per se and his paper and tech-nique were relatively neglected (see Bruce amp Winograd1998) Many of the lists Deese created elicited no or veryfew intrusions which served his needs perfectly well be-cause he was interested in how associative factors wouldaffect both veridical and false recall (Deese 1959a 1959b)It is perhaps for this reason that Deese never made muchof his intrusion findings Indeed in his 1965 book TheStructure of Associations in Language and Thoughtwhich summarized much of his important work pub-lished over the previous 10 years the intrusion paper(Deese 1959b) was not even cited
385 Copyright 2001 Psychonomic Society Inc
This research was supported by a contract from the US Office ofTechnical Services to HLR and KBM as well as NIMH Grant 1RO3MH59034-01 to KBM The authors thank Martha Storandt for adviceregarding the regression analysis Michael A Stadler for providing nor-mative data Douglas Nelson for advice in collecting our norms DavidA Balota for advice on various aspects of this research Randy Buck-ner for calling our attention to Kirkpatrickrsquos work and Cathy McEvoyDouglas Nelson and Tamaku Komatsu for providing additional dataMichael Cortese Patrick Dolan Stephanie Lee Elizabeth MarshMichelle Meade Douglas Nelson and Valerie Reyna provided helpfulcomments Correspondence concerning this article may be sent to H LRoediger III Department of Psychology Box 1125 Washington Uni-versity One Brookings Drive St Louis MO 63130-4899 (e-mailroedigerartsciwustledu)
Factors that determine false recall A multiple regression analysis
HENRY L ROEDIGER III JASON M WATSON KATHLEEN B MCDERMOTT and DAVID A GALLO
Washington University St Louis Missouri
In the DeesendashRoedigerndashMcDermott (DRM) paradigm subjects study lists of words that are designedto elicit the recall of an associatively related critical item The 55 lists we have developed provide lev-els of false recall ranging from 01 to 65 and understanding this variability should provide a key to un-derstanding this memory illusion Using a simultaneous multiple regression analysis we assessed thecontribution of seven factors in creating false recall of critical items in the DRM paradigm This analy-sis accounted for approximately 68 of the variance in false recall with two main predictors associa-tive connections from the study words to the critical item (r 5 173 semipartial r 5 160) and re-callability of the lists (r 5 43 semipartial r 5 34) Taken together the variance in false recallcaptured by these predictors accounted for 84 of the variance that can be explained given the relia-bility of the false recall measures (r 5 90) Therefore the results of this analysis strongly constrain the-ories of false memory in this paradigm suggesting that at least two factors determine the propensityof DRM lists to elicit false recall The results fit well within the theoretical framework postulating thatboth semantic activation of the critical item and strategic monitoring processes influence the proba-bility of false recall and false recognition in this paradigm
386 ROEDIGER WATSON MCDERMOTT AND GALLO
Roediger and McDermott (1995) developed a para-digm for the creation and experimental investigation offalse memories that was an adaptation of Deesersquos tech-nique Following a suggestion made by Endel Tulvingthis procedure is now referred to as the DRM paradigman acronym for Deese Roediger and McDermott Thebasic procedure is straightforward Subjects hear lists of15 words that are the strongest associates to a missingword or critical item in word association norms (NelsonMcEvoy amp Schreiber 1999 Russell amp Jenkins 1954)For example subjects might be presented with a list ofwords like bed rest awake tired dream wake snoozeblanket doze slumber snore nap peace yawn drowsymdashall of which are associated to the (nonpresented) criticalitem sleep Immediately following the presentation of astudy list subjects recall as many of the list words aspossible without respect to their order (ie single-trialfree recall) In addition subjects are instructed to be surethat each word they write down was indeed on the pre-ceding list Despite this warning against guessing sub-jects recall these critical nonpresented items with aboutthe same probability as items appearing in the middle ofthe list (ie excluding the primacy and recency effects)At the end of the experiment a recognition test that cov-ers many lists is usually given (often with metamemoryjudgments) Recognition of the critical lures typicallyequals or exceeds recognition of the studied words Theseremarkably high levels of false recall and false recogni-tion in the DRM paradigm have been widely replicated(eg Payne Elie Blackwell amp Neuschatz 1996 andRoediger McDermott amp Pisoni 2001 among manyothers)
The DRM paradigm is quite different from many otherpopular techniques used to study false memories Mostdemonstrations of false memory phenomena have usedrelatively naturalistic materials such as stories or visualscenes fairly complex techniques such as the presenta-tion of misleading information or relatively long delaysbetween the study of the material and its later testing Infact the underlying assumption (beginning at least withBartlettrsquos 1932 influential pronouncements) seems tobe that these conditions are necessary to create illusorymemories because the use of such materials as wordlists encourages reliance on reproductive rather than re-constructive memory However the levels of false re-membering with the DRM paradigm are among the mostrobust ever obtained in the experimental literature eventhough they occur under conditions of immediate test-ing with no misleading information with lists of wordsand with strong warnings against guessing and makingerrors Therefore Roediger and McDermott (1995) sug-gested that the distinction between reproductive and re-constructive modes of retention should be abandoned Insome sense all recollection even immediate recall andrecognition of lists of words is reconstructive
Although the DRM paradigm is a relative newcomerto the investigation of false memories considerable re-search has been conducted using it and variations on its
basic theme The purpose of this paper is to consider anoverlooked issue in the growing literature on false mem-oriesmdashnamely the characteristics of the materials usedto produce the effect In Roediger and McDermottrsquos(1995) first experiment they selected the 6 lists used byDeese (1959b) that produced the highest levels of intru-sion in his experiment and they replicated his effects Intheir second experiment they used a total of 24 falsememory lists that included the 6 lists from their first ex-periment along with 18 new lists Some of the new listshad been used by Deese whereas Roediger and McDer-mott (1995) created some of the others using similar pro-cedures
One remarkable outcome that we have noticed repeat-edly while working with these lists is the striking vari-ability they exhibit in eliciting false recall and falserecognition of the associatively related critical itemThese differences occur despite the fact that all the listsare created in the same manner to include with someminor variation the first 15 associates of a single wordthat is itself not presented With few exceptions later re-searchers have used Roediger and McDermottrsquos (1995)original 24 lists to study various interesting issues col-lapsing the resulting data across the lists as did Roedi-ger and McDermott (1995) However few investigatorshave been concerned with the fundamental issue of iden-tifying characteristics of lists that influence the effectThat is why do some lists work so well in producing thefalse memory effect whereas others do not when all ofthe lists are constructed in the same general manner Inaddition examination of the lists often reveals no obvi-ous reasons for their variations in effectiveness at elicit-ing false recall For example consider the lists associ-ated with bitter (sweet sour taste chocolate rice coldlemon angry hard mad acid almonds herbs grapefruit) and sweet (sour candy sugar bitter good tastetooth nice honey soda chocolate heart cake tart pie)Despite the fact that both words refer to concepts in thesame domain represent the same part of speech and haveoverlapping words in the lists they differ dramatically inthe effectiveness with which they elicit false recall Themean probability of false recall of bitter when the rele-vant list was presented was only 01 whereas the meanfalse recall of sweet following the relevant list was 54
Stadler Roediger and McDermott (1999) obtainednorms on 36 DRM lists the 24 used by Roediger andMcDermott (1995) and 12 other lists False recall rangedfrom greater than 60 intrusion rates to lists structuredaround the nonpresented words window sleep smelland doctor to 20 and 10 respectively for the listsstructured around fruit and king False recognitionshowed similarly wide variations Although these listswere originally constructed to yield high levels of falseresponding some did not The purpose of the presentstudy was to attempt to identify those characteristics ofthe lists that cause their differential efficacy in produc-ing false recall and false recognition through a simul-taneous multiple regression analysis By identifying
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 387
these factors we should gain evidence bearing on themechanisms that underlie the creation of false memoriesin the DRM paradigm Before turning to this analysiswe review leads provided by prior research in uncover-ing the factors at work
Deesersquos (1959a 1959b) early work produced two im-portant clues about factors that affect both veridical andfalse recall The two factors he identified are associativeconnections among the study list words themselves andassociative connections between the study list words andthe critical item Deese (1959a) provided subjects withlists that varied in what he termed interitem associativestrength ldquodefined as the average relative frequency withwhich all items in a list tend to elicit all other items in thesame list as free associatesrdquo (p 235) He showed that thismeasure correlated highly (r 5 188) with the totalnumber of words recalled from the listsmdashthe strongerthe associative bonds between items in the list the bet-ter that list was recalled More important for present pur-poses interitem associative strength correlated nega-tively with the number of extralist intrusions (r 5 48)However the stronger the associative bonds among listitems the more likely the subjects were to produce thesame intrusion (r 5 155) Therefore associationsamong list items increased recall of list items and de-creased overall intrusions yet when intrusions did occurfor strongly interconnected lists they tended to be thesame item across subjects
Deese (1959b) also examined the influence of asso-ciative connections from the list words to the critical itemon the false recall of the critical item Subjects in this ex-periment heard 36 lists each of which consisted of the12 highest associates to the nonpresented critical itemFalse recall with these lists ranged from 42 for the crit-ical item needle to 0 for the critical item butterfly Deese(1959b) found that one predictor could account for themajority of the variance in the recall of critical itemsacross his 36 lists This variable was the mean backwardassociative strength (BAS) or the average tendency forwords in the study list to elicit the critical item on a freeassociation test In fact Deese (1959b) reported that thecorrelation between a listrsquos mean BAS and the probabil-ity of recall of its associated critical item was 187Roediger and McDermott (1995) argued that this out-come supports the notion that encoding factors are partlyresponsible for the effect That is lists high in BAS mightactivate the critical item consciously or unconsciouslyas the list items are presented Subjects are confrontedwith a reality-monitoring dilemma (Johnson amp Raye1981) during the test ldquoDid I hear the critical item or didI only think of it while listening to the list itemsrdquo
Other data support this interpretation of the produc-tion of false memories in the DRM paradigm Subjectsclaim to remember the occurrence of the critical items inthe study list which may mean that the item was con-sciously aroused during encoding (Roediger amp McDer-mott 1995) or at least unconsciously activated throughspreading activation (Roediger Balota amp Watson 2001)
In addition subjects claim to remember which of twospeakers said the nonpresented critical item when thelists are presented in both a male and a female voice(Payne et al 1996 Roediger McDermott amp Pisoni2001) Similarly when some lists are presented auditorilyand others are presented visually subjects attribute theillusory item as having occurred in the modality in whichthe list had been presented to the same extent as theymake attributions for list items (Gallo McDermott Perceramp Roediger 2001) Furthermore subjects have beenshown to report having rehearsed the critical item duringthe presentation of the study list (Mather Henkel ampJohnson 1997) Finally McDermott (1997) showed thatpresentation of the lists yields priming of the nonpre-sented critical item on implicit memory tests of word-stem and word-fragment completion albeit at a lessermagnitude than would be the case if the critical itemshad actually been presented Because priming on thesetests usually depends on lexical activation (Weldon1991) McDermott (1997) argued that her data were con-sistent with the notion that the presentation of DRM listsproduces lexical activation of the critical item at leastfor some lists
Of course all memory phenomena depend on theinteraction of encoding and retrieval processes (Tulving1974 1979) and we do not intend to suggest that re-trieval processes are unimportant In fact Roediger andMcDermott (1995) argued that the consideration of re-trieval and decision processes are crucial to under-standing this memory illusion and others They cited Ja-cobyrsquos attributional approach (eg Jacoby Kelley ampDywan 1989) and Johnsonrsquos source-monitoring theory(eg Johnson Hashtroudi amp Lindsay 1993) as promis-ing frameworks to guide this arena of inquiry Israel andSchacter (1997) and Smith and Hunt (1998) have bothargued that the distinctiveness of studied items at re-trieval can explain why some manipulations reduce falserecall and false recognition in this paradigm Neverthe-less given the strong activation of the critical item by itssemantic associates at encoding list context clearlyplays an important role in eliciting false memories in theDRM paradigm
To our knowledge McEvoy Nelson and Komatsu(1999) are the only other contemporary researchers whohave attempted to determine the list-based factors thatunderlie the creation of false memories with associativelists In their experiments McEvoy et al manipulatedthe two factors that Deesersquos (1959a 1959b) earlier workhad implicated interitem associative strength (whichthey call connectivity) and the mean BAS of the listwords to the critical item Consistent with Deesersquos (1959a)research they found that lists with high interitem asso-ciative strength yielded higher recall of list words andlower false recall of critical items relative to lists withlower interitem associative strength In addition and alsoconsistent with Deese (1959b) they found that lists withhigh mean BAS yielded more false recall than did listswith low mean BAS They concluded that although con-
388 ROEDIGER WATSON MCDERMOTT AND GALLO
nectivity and mean BAS had opposite influences on the re-call of critical items both variables played an importantrole in creating false memories in the DRM paradigm
In sum both Deese (1959a 1959b) and McEvoy et al(1999) argued that at least two factors are important inpredicting veridical and false recall Associative con-nections from study words to the critical item increasethe probability of false memories whereas associativeconnections among study words decrease the probabilityof false memories However other factors besides meanBAS and connectivity may underlie the recall of criticalitems in the DRM paradigm For example consider theperformance on the king list (ie queen England crownprince George dictator palace throne chess rule sub-jects monarch royal leader reign) Because it has a rel-atively high mean BAS it would be predicted to have ahigh level of false recall However this list elicited thelowest level of recall in the entire set of 36 DRM listsnormed by Stadler Roediger and McDermott (1999)
Gallo and Roediger (2001 Experiment 1) attemptedto produce lists that were constructed in the same generalway as those of Roediger and McDermott (1995) but thatwould like the king list produce low levels of false re-call and false recognition However unlike the king listthe lists created by Gallo and Roediger all had low BASOf the 19 new lists they constructed many had very lowlevels of false recall and false recognition However sev-eral of the lists (eg wish justice) produced relativelyhigh levels of false recall and false recognition muchlike the lists developed by Roediger and McDermott(1995) Clearly BAS is not the entire reason for the widedifferences in the effectiveness of the lists in creating falsememories The aim of the present study is to gain furtherinformation on the characteristics of lists and criticalitems that affect false recall and false recognition
OVERVIEW OF THE PRESENT STUDY
In the present study we used a simultaneous multipleregression analysis to identify the factors predicting re-call of critical items in the DRM paradigm We consid-ered factors that pertain to the critical item itself as wellas factors characterizing the study lists Specificallywith respect to the critical items we used the followingthree variables word length word frequency and con-creteness On the basis of an examination of the Stadleret al (1999) norms and Deesersquos (1959b) original find-ings we hypothesized that factors making a critical itemdistinct might produce low levels of false recall For ex-ample butterfly is a long low-frequency critical itemInterestingly it was never recalled in Deesersquos (1959b) ex-periment and it was recalled by only 1 of Gallo andRoedigerrsquos (2001) subjects
We also assessed the ability of several variables re-lated to the study words to predict the recall of criticalitems in the DRM paradigm In particular we used thefollowing four variables an index of the strength of theassociative connections from the critical item to the
study words or forward associative strength (FAS ieassociations from sleep to bed to rest to awake etc) anindex of the strength of the associative connections fromthe study words to the critical item or BAS (associationsfrom bed rest to sleep) the average number of as-sociative connections among the study words (interitemassociative strength or connectivity) and the averageprobability of recall of the studied words within each ofthe false memory lists (veridical recall) On the basis ofprior research by Deese (1959a 1959b) and McEvoy et al(1999) we hypothesized that BAS and connectivitywould be the best predictors of the differences observedin false recall among critical items More specificallyrecall of critical items should increase as the strength ofassociative connections from the study words to the crit-ical item increases however recall of critical items shoulddecrease as the average number of associative connec-tions among study words increases (As will be shownin our data set only the first of these two predictions wasfulfilled) In sum the goal of the present research was toidentify the factors that underlie the creation of falsememories in the DRM paradigm by explaining the widevariability in the effectiveness of our 55 lists to elicitfalse recall The multiple regression analysis that led toour answers is explained more fully in the Method andResults sections
METHOD
MaterialsIn the present study we obtained data for 55 DRM lists which in-
cluded 24 lists from Roediger and McDermott (1995) 12 others fromthe Stadler et al (1999) norms and 19 lists from the Gallo andRoediger (2001) norms The study words and the associated criticalitem for each list are presented in Appendix A along with the BASand FAS of each item in relation to the critical item The words withinlists are presented in descending order with respect to their strengthof association to the critical item That is the first list item is typicallythe item most often generated as an associate to the critical item Bothveridical and false recall across our 55 lists are quite stable The split-half correlation of veridical recall for the 55 lists is +86 and the split-half correlation for the lists in eliciting false recall is +90 Thereforewe could potentially explain 81 of the variance in false recall
VariablesAs was described in the introduction the variables we used can
be considered as belonging to two categories properties of the crit-ical item itself and properties of the word lists A complete listingof the values of all seven variables (as well as levels of false recalland false recognition) for all 55 lists appears in Appendix B
Critical item variables With respect to the critical items weused the following three variables word length log frequency andconcreteness We consider each below
1 Word length was defined by the number of letters in each crit-ical item and represents the relative distinctiveness of each wordform (longer words being more distinct) For example butterfly hasnine letters whereas sleep has only five letters We also obtained asubjective measure of orthographic distinctiveness 1 and an objec-tive measure of orthographic neighborhood size2 but because ofthe high degree of multicollinearity among these three variables 3
we included only word length of the critical item in the multiple re-gression analyses
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 389
2 The raw frequency of each critical item was obtained from theKu Iumlcera and Francis (1967) norms and is the number of instancesfound in print per million words For example the raw frequency ofthe critical item sleep is 65 and the raw frequency of the criticalitem butterfly is 2 To correct for skewness in the frequency distri-bution for each critical item raw frequencies were transformedinto log frequencies using the following formula log (5 1 rawfrequency) For example the log frequency of sleep is 182 [ie log(655)] whereas the log frequency of butterfly is 40 [ie log (25)]
3 Concreteness ratings for each critical item were obtained fromthe word association norms of Nelson et al (1999) which were inturn obtained in part from the Paivio Yuille and Madigan (1968)and the Toglia and Battig (1978) norms These ratings are obtainedon a scale of 1ndash7 with 1 being the least concrete and 7 the mostconcrete To carry through with our examples sleep is slightlyabove the midpoint in concreteness (with a rating of 474) and but-terfly is more concrete (with a rating of 591)
List variables With respect to the study lists we used the fol-lowing four variables FAS BAS interitem associative strengthand veridical recall We consider each below
1 Forward associative strength Indices of the strength of asso-ciative connections from the critical item to the study words wereobtained from the word association norms of Nelson et al (1999)Specifically for each critical item we obtained the probability thatit elicited each word in its corresponding study list as an associatea measure Deese (1959b) referred to as the FAS For example as isshown in the third column of Appendix A for the critical item sleepwe identified the strength of its connections to bed rest awaketired and the remaining study words We then averaged these asso-ciations to obtain the mean FAS value for each list In the case ofsleep the mean FAS value for its 15 study words is 047 whereasin the case of butterfly the mean FAS value is 033 Note that all 15list items were given as a response to the critical item in the Russelland Jenkins (1954) norms (from which the lists were generated)However owing to discrepancies between these norms and those ofNelson et al (1999) the number of forward connections for eachlist could have theoretically ranged from 0 to 15
2 Backward associative strength An index of the strength of as-sociative connections from the study words to the critical item wasobtained in large part from the Nelson et al (1999) word associa-tion norms However for some of the 55 DRM lists there were afew list words that were missing from these norms For these wordswe collected our own norms using the procedures outlined by Nel-son and colleagues Briefly we gave groups of over 75 undergrad-uates one of three sheets of 80 to-be-normed words the subjectswere instructed to write next to each word the first word it broughtto mind In those cases in which more than one word from a partic-ular study list needed to be normed these words were placed onseparate norming sheets or were separated by at least 30 items onthe same norming sheet At least 86 observations were obtained foreach normed item and 150 or more observations were obtained foran item in some cases
With these two sets of norms we took each study word and de-termined its connection strength to its corresponding critical itemthe measure Deese (1959b) referred to as the BAS For example asis shown in the second column of Appendix A we obtained for eachstudy word (eg bed rest awake tired etc) the probability thatit elicited its critical item (eg sleep) as an associate Thus thismeasure examines connections in the direction opposite to that usedto obtain the FAS measures We then averaged these associations toobtain the mean BAS value for each list For example the meanBAS value for the sleep list is 431 By comparison the mean BASvalue for the butterfly list is 045
3 Interitem associative strength or connectivity The average in-teritem associative strength (Deese 1959a) or connectivity (Mc-Evoy et al 1999) of the study words was obtained for each listusing two steps First following a procedure outlined by McEvoy
et al we constructed a 15 3 15 matrix of study items for each listIn Appendix C we have illustrated the connectivity matrix for thesleep list Second using the Nelson et al (1999) norms and insome instances our own word association norms we identified thepresence of a connection for all 225 combinations of study wordpairs in each listrsquos matrix excluding the 15 self-connections whichwere not determined (listed as ND) For each word pair a nonzeroconnection strength was scored as a 1 and a zero connectionstrength was scored as a 0 In the case of sleep a total of 27 nonzeroconnections exists among the study words (or an average of 180connections per study word) In the case of butterfly a total of 29nonzero connections exists among the study words (or an averageof 193 connections per study word) High mean connectivityscores indicate a greater number of associative relationships amongthe study words whereas low mean connectivity scores indicate alesser number of associative relationships among the study words
4 Veridical recall of the list One final predictor was the averageprobability of recall of the studied words collapsing across serialpositions 1ndash15 for each list These data were obtained from Stadleret al (1999) and Gallo and Roediger (2001) In the case of sleepstudied items were recalled at a 61 level whereas in the case ofbutterfly studied items were recalled at a 70 level As was notedearlier the split-half correlation of the veridical recall measure wasreasonably high at +86
ProcedureTaking the seven factors that were just described along with the
probability of false recall and the probability of false recognitionfor each of the 55 lists we constructed the data matrix shown inAppendix B for our simultaneous multiple regression analysis4
Each of the 55 lists used in the regression analysis is represented byits respective critical item each of which is listed in alphabeticalorder in the f irst column of the matrix The second and thirdcolumns of the matrix contain the probability of false recall andfalse recognition respectively of each critical item from the Stadleret al (1999) and Gallo and Roediger (2001) norms The next threecolumns contain properties of the critical item itself length log fre-quency and concreteness The next four columns contain propertiesof the list that was constructed for each critical item mean FASmean BAS the mean connectivity of the study words and the av-erage recall of the studied items The last three columns contain thecritical item predictors that were excluded from the multiple re-gression analyses raw frequency orthographic distinctiveness andorthographic neighborhood size (Coltheartrsquos N )
In multiple regression analysis the independent variables can beentered either simultaneously (in a single step) or hierarchically (withsome variables entered into the analysis before others) In the for-mer case all independent variables are allowed to compete forshared variance in the dependent variable (eg false recall) in a sin-gle step so that no single variable is given greater priority than isanother In the latter case those independent variables entered inearlier steps are allowed to capture variance in the absence of com-petition from variables to be entered subsequently In this way in ahierarchical analysis earlier variables are given greater prioritythan later ones usually for theoretical reasons For the purpose ofthe present analysis there exists no definitive theoretical justifica-tion for ordering the entry of critical item variables and list vari-ables so we opted for the simultaneous entry of all variables in asingle step
RESULTS
We consider first the bivariate correlations betweenour seven factors and false recall and then we turn to thesimultaneous multiple regression analysis and its find-ings The entire matrix of correlations (Pearsonrsquos r in all
390 ROEDIGER WATSON MCDERMOTT AND GALLO
cases) among the seven predictors and false recall isshown in Table 1
Bivariate CorrelationsIn describing the results we report significant bivari-
ate correlations between critical item variables and falserecall followed by significant correlations between listvariables and false recall To determine these values eachof the critical item and list variables were entered into aregression equation as the sole predictor of false recallOnly three of our seven variables were significantly cor-related ( p lt 05) with false recall the critical item vari-able of length and the list variables of BAS and recall ofthe studied items (Figures 1A and 1B respectively)
Length The length of the critical item and false recallwere negatively correlated [r 5 37 t(53) 5 287]
Backward associative strength Mean BAS or thestrength of the connections between study items andtheir corresponding critical items was positively corre-lated with false recall r 5 173 t(53) 5 773 As isshown in Figure 1A the greater the strength of associa-tion between the list items and the critical item the morelikely is false recall This outcome replicates Deesersquos(1959b) finding of a 187 correlation between BAS andfalse recall which he obtained with a set of thirty-six 12-item lists Although the 173 correlation between thesetwo variables in our set of f ifty-f ive 15-item lists issomewhat smaller than the 187 correlation obtained byDeese BAS is still the strongest correlate of false recallin our entire data set
Veridical recall Recall of list items and false recallwere negatively correlated [r 5 43 t(53) 5 346]As is shown in Figure 1B increasing levels of accuraterecall are associated with decreasing levels of false re-call This outcome is contrary to some ideas about whyfalse recall occurs as will be brought out in the Discus-sion section
As is shown in Table 1 no other variables were corre-lated with false recall at the bivariate level which is sur-prising in some cases For example in terms of criticalitem variables one might have expected word frequencyor concreteness to be correlated with false recall but nei-ther did (Deese 1959b also failed to find a relation be-tween word frequency and false recall) In terms of listvariables one might have expected connectivity to cor-relate with false recall but it did not The lack of a rela-
tion between connectivity and false recall is noteworthybecause Deese (1959a) and McEvoy et al (1999) re-ported a significant relationship between these variablesso that lists with greater connectivity were less likely toelicit false recall of the critical item In the Discussionsection we consider connectivity or interitem associa-tive strength more fully We turn now to the results of thesimultaneous multiple regression analysis
Simultaneous Multiple Regression AnalysisIn the multiple regression analysis the three variables
based on the characteristics of the critical items them-selves (ie length log frequency and concreteness) andthe four variables based on the characteristics of the listsdesigned to elicit the critical items (ie FAS and BASmeasures interitem associative strength or connectivityand the level of recall of list items) were entered simul-taneously into the regression equation to predict false re-call The results of the analysis are shown in Table 2 in-cluding beta weights (b) semipartial correlations (sr)and significance tests for each predictor For signifi-cance tests a t(47) was used for each predictor and theresultant t statistic is expressed as the absolute valuemdashthat is frac12tfrac12 In addition Table 2 displays the results of thesignificance test for the multiple R2 for the entire set ofseven predictor variables
As is shown in Table 2 the results of the regressionanalysis indicated that two list variables were signifi-cantly related to false recall Specifically BAS was thestrongest predictor of false recall (b 5 170 sr 5160 t 5 717) Therefore consistent with our findingsat the bivariate level the greater the strength of associa-tion between study items and their corresponding criti-
Table 1 Correlation Matrix Among Eight Variables False Recall Three Critical Item Variables and Four List Variables
False Log VeridicalRecall Length Frequency Concreteness FAS BAS Connectivity Recall
NotemdashR2 5 68 F(747) 5 1406 MSe 5 0013 p lt 05 p lt 05
Figure 1 Probability of false recall of the critical item from 55 lists (A) as a functionof associative strength of the list items to the critical item or backward associativestrength and (B) as a function of veridical recall of studied items from the lists
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 391
cal item on a free association test the more probable isfalse recall of the critical item Veridical recall was alsoa strong predictor of false recall (b 5 40 sr 5 34t 5 411) As we found at the bivariate level the morewords recalled from a list the less likely that the criticalitem for that list was falsely recalled None of the re-maining critical item or list variables including length ofthe critical item connectivity and FAS explained aunique portion of the total variance in false recall afterthe influence of other predictors had been residualized inthe context of the regression equation Therefore BASand recall of list items appeared to be the major contrib-utors to the multiple R of 182 and the associated mul-tiple R2 of 68 [F(747) 5 1406 MSe 5 0013] We fur-ther explore the implications of these effects in theDiscussion section
Factors Affecting False RecognitionIn our analyses thus far we have concentrated on
false recall exclusively However the norms for our 55lists (collected by Stadler et al 1999 and Gallo ampRoediger 2001) also include values for false recogni-tion The recognition data in these norms were collectedin a final recognition test after lists had been studied andrecalled This confounding of recognition by prior recallalmost certainly affected both veridical and false recog-nition However in experiments directly examining theeffects of recall on recognition in this paradigm and oth-ers the finding is that recall of a list may increase itsrecognition by a moderate amount (typically in the4ndash10 range eg Roediger amp McDermott 1995Roediger McDermott amp Pisoni 2001) In fact some re-searchers have failed to observe any effect at all of recall
392 ROEDIGER WATSON MCDERMOTT AND GALLO
on recognition (Schacter Verfaellie amp Pradere 1996 seeRoediger et al 1998 for a discussion) We conducted thesame regression analyses with our seven predictor vari-ables on false recognition as we did on false recall Wesummarize the basic results and conclusions as follows
There were three significant bivariate correlationswith false recognition including length (r 5 27)BAS (r 5 143) and veridical recall (r 5 52) Over-all these correlations with false recognition were simi-lar to those obtained with false recall However the pos-itive correlation between BAS and false recognition wassmaller than that in the recall data and the negative cor-relation between veridical recall and false recognitionwas larger than that in the recall data To maintain con-sistency with the previous regression analysis on falserecall we used veridical recall rather than veridicalrecognition in this analysis In addition recognition es-timates were based on only three itemslist and wereprobably affected by prior recall
Turning to the results of the simultaneous multiple re-gression analysis on false recognition BAS and veridi-cal recall of list items appeared to be the major contrib-utors to the multiple R of 169 and the associatedmultiple R2 of 48 [F(747) 5 611 MSe 5 002] Table 3shows the summary of the regression analyses Althoughthe regression analysis accounted for less total variancein false recognition than in false recall (48 as opposedto 68 respectively) the same factors seem to be re-sponsible for predicting false recognition and false re-call Of course because measures of recognition alwaysfollowed recall this similar pattern may reflect a carry-over effect of false recall on false recognition Nonethe-less within the limits of this study the factors responsi-ble for false recognition across lists seem to be the sameas those producing false recall
DISCUSSION
To summarize the main findings our results implicatetwo primary factors (of the seven variables we exam-ined) as accounting for most of the variance in false re-call across lists The split-half correlation of our lists was190 so the amount of potentially explainable variancein our dependent variable of false recall is 81 The fac-
tors in our multiple regression accounted for 68 of thetotal variance in false recall or 84 of the explainablevariance (68 4 81 3 100 5 84) The analysis forfalse recognition revealed the same factors at work al-though they accounted for less overall variance Thestrongest factor was BAS consistent with prior work byDeese (1959b) and McEvoy et al (1999) The other fac-tor was the number of items recalled accurately from thelist which is negatively correlated with false recall
We turn next to the theoretical implications of ourfindings We first consider the implications for the acti-vationmonitoring framework that we have developed toexplain false recall and false recognition in the DRMparadigm We then consider the implications for othertheories and f inally turn to several further issues onwhich our results shed light
ActivationMonitoring TheoryIn the theroretical framework that we (eg McDer-
mott amp Watson 2001 Roediger Balota amp Watson 2001Roediger amp McDermott 1995 2000) have used to explainvariations in the probability of false recall and false recog-nition in the DRM paradigm two sets of processes havebeen proposed Specifically we have proposed that pro-cesses occurring during both encoding and retrieval af-fect the probability of false recall and false recognitionin the DRM paradigm Two critical sets of processes in-volve activation and monitoring Although there is a nat-ural alignment of activation to encoding and monitoringto retrieval we note that both processes can potentiallyoccur both during encoding and during retrieval as willbe discussed below The ideas we employ have many pre-cursors which we acknowledge below although theirparticular arrangement here may be unique
A critical distinction in understanding encoding pro-cesses and their interaction with retrieval processes isbetween item-specific and relational processing (Ander-son 1972 Hunt amp Einstein 1981 Hunt amp McDaniel1993) Briefly the notion is that emphasis can be given tothe encoding of individual elements and their features(what Anderson called node-taggingand Hunt and his col-leagues called item-specific processing) or to the encod-ing of relationships among elements (pathway-taggingor relational processing) Item-specific processing should
Table 3 Summary of Multiple Regression Analysis
on False Recognition With Critical Item Variablesand List Variables Used as Predictors
NotemdashR2 5 48 F(747) 5 611 MSe 5 002 p lt 05 p lt 05
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 393
make individual elements more memorable and lead tomore distinctive memories such processing should helpdistinguish veridical memories from false memories be-cause list items would have features of presentation as-sociated with them whereas the critical item sometimeswould not (However if the critical item is strongly acti-vated during encoding it may take on many of the fea-tures of list items) Relational processing (extractingthemes or schemas or constructing networks of associa-tions) should increase the probability of false recall andfalse recognition by leading to inferences or associationsof events that may not actually have occurred
We assume that in encoding and interpreting experi-ences people go beyond the information given in thestimulus (Bartlett 1932 chap 2 Bruner 1957) and ac-tivate related information through inferential processingEach personrsquos experience is coded in terms of his or herown schemata (or associative networks or knowledgestructures) When schemata are activated a process offilling in of ldquomissingrdquo information may occur wherebyitems strongly implied by the schema (but not actuallypresent in the stimulus) may become activated (Bartlett1932) In encoding lists of words in the DRM paradigmpresentation of the list can create activation that spreadsthroughout the lexicalsemantic system and can createimplicit associative responses (Underwood 1965) Thisactivation may arouse concepts that are not presented but that are associated to concepts that are presented(eg Anderson amp Pirolli 1984 Collins amp Loftus 1975Meyer amp Schvaneveldt 1971) The associative activa-tion in this approach can arise from automatic process-ing or from more controlled elaborative processing(Neely 1977) In addition the associative activation ofconcepts that were not presented may occur consciously(subjects may become aware of the concept of sleep andrehearse it during presentation of the list words bed restawake etc) or the concept may be strongly activatedbut never come to conscious awareness The debate overwhether arousal of the critical item in the DRM para-digm is conscious or unconscious is interesting but dif-ficult to decide empirically Some evidence has impli-cated conscious arousal of the concepts (McDermott1997) whereas other evidence implicates unconsciousactivation (Seamon Luo amp Gallo 1998) The correctanswer to the question of whether activation is consciousor unconscious is probably ldquobothrdquomdashthat is the criticalitem probably sometimes comes consciously to mindand even if it does not relevant information may never-theless be activated unconsciously
The high correlation between BAS and false recall aswell as the strong weighting of BAS in the multiple re-gression analysis strongly implicates activation of criti-cal items as an important factor in explaining false recalland false recognition in the DRM paradigm The morestrongly associated list items are to critical items themore activation would be expected to spread to the crit-ical items More colloquially to the extent that the wordson the list spark the associative connection to the critical
item the more likely the critical item is to be falsely re-called This activation factor would seem to be one mainreason why ldquothings suggested to a person by an experi-ence may be honestly reported by him as part of the ex-periencerdquo (Kirkpatrick 1894 p 608) These associatedlists probably all tend to elicit relational processing butonly those with strong BAS to the critical item lead tofalse recall
The second set of factors proposed in activationmon-itoring theory is concerned with the monitoring of mem-ory accuracy Monitoring of encoding processes canoccur during the study phase of an experiment espe-cially under intentional learning conditions In experi-ments in which subjects are warned about false memoryphenomena and are told to pay close attention to lists andto remember only the presented items and not the criti-cal items monitoring is enhanced (Gallo Roberts amp Sea-mon 1997 Gallo Roediger amp McDermott 2001 Mc-Dermott amp Roediger 1998) Monitoring of memorieshas typically been discussed with respect to retrieval dur-ing which time a primary goal is to distinguish fromamong the considerable information brought to con-sciousness that which refers to perception of past eventsfrom that which does not a process referred to as realitymonitoring (Johnson amp Raye 1981) The experience ofremembering events involves mentally traveling back intime and seeming to reexperience the events (Tulving1985) During retrieval recovered memories that appar-ently carry with them details of having been experiencedare generally attributed to be memories of actual pastevents although sometimes this attribution may be inerror (Jacoby et al 1989)
Johnson et al (1993) have described in detail the typesof information that may be used to distinguish eventsthat actually happened from ones that were only imag-ined thought about or read about For example eventsthat actually occurred have more features from the ex-ternal world associated with them (sights sounds etc)whereas those that were internally generated are lesslikely to have these features but more likely to have as-sociated reflections and cognitive operations In theDRM paradigm the fact that the critical item has beenstrongly activated during encoding means that duringretrieval the subject is likely to retrieve the critical itemas a list member To the extent that the critical items arestrongly activated during encoding they will take onsome features similar to those of the list items this is es-pecially true if the critical items come consciously tomind during encoding and are rehearsed in the samemanner as list items To the extent that the features of thecritical item are similar to those of actual list items theprobability of false recall or false recognition is increasedHowever if the list items systematically differ from thecritical item on some dimensions or features retrievalheuristics may be brought into play that can reduce falserecall and false recognition (Israel amp Schacter 1997Schacter Israel amp Racine 1999) Considerable evidencesupports this general source-reality-monitoring frame-
394 ROEDIGER WATSON MCDERMOTT AND GALLO
work that we advocate here (eg Johnson et al 1993Schacter et al 1999) as well as the attributional ap-proach to remembering (Jacoby et al 1989)
The other significant factor arising from our multipleregression analysis may operate during retrieval by aid-ing the monitoring processes that reduce false recall Thenegative correlation between veridical recall of list itemsand false recall of the critical item may indicate that thebetter encoded list items are the more easily they can bedistinguished from the illusory critical item Subjectsmay use item-specific information to aid recall of listitems and suppress recall of the critical items whichwould not carry as much item-specific information Whenexperimental manipulations increase veridical recall ei-ther through longer study time at relatively slow rates(Gallo amp Roediger 2001 McDermott amp Watson 2001Toglia Neuschatz amp Goodwin 1999) or through re-peated study and test opportunities (McDermott 1996Schacter et al 1999) false recall is often decreasedTherefore the negative relation between veridical recalland false recall can be produced experimentally as wellas appearing in our correlationregression analysis How-ever veridical and false recall can be positively related as afunction of other variables too as was shown by McDer-mott (1996) Toglia et al (1999) and Thapar amp McDermott(2001) In these cases greater activation of the criticalitem from relational processing or from greater seman-tic processing is probably at work
The fact that the length andor distinctiveness of thecritical item are negatively correlated with false recallcan be similarly interpreted as the effect of monitoringduring retrieval because the greater distinctiveness ofthe critical item simplifies the monitoring process Forexample if words in the list are generally shorter andhave more regular orthography than does the criticalitem then during retrieval the critical items that are longand distinctive (eg butterfly cabbage) should be moreeasily discriminated from list items as indeed seems tobe the case in the bivariate correlations Interestingly thelengthdistinctiveness factor played a role in determin-ing false recall even though the lists in our norms werepresented auditorily and the orthographic informationat least should not be particularly accessible In addi-tion presenting the DRM lists visually rather than audi-torily reduces false recall and false recognition at leaston tests in which recognition items are presented visually(as opposed to auditorily) or recall is written rather thanspoken (Gallo McDermott et al 2001 Kellogg 2001Smith amp Hunt 1998) This modality difference has alsobeen interpreted as arising from the retrieval of distinctiveinformation operating during the test (Gallo McDermottet al 2001 Smith amp Hunt 1998)
As was noted above activation processes may occurduring retrieval as well as during encoding The morelist items a subject recalls the more primed the criticalitem might become However the present analyses un-dercut the simplest form of this idea If the recalling oflist items activated or primed false recall the simplest
prediction would be that there should be a positive cor-relation between veridical recall and false recall Our datarevealed the opposite pattern showing a negative corre-lation between veridical and false recall Marsh McDer-mott and Roediger (2001) reported other experimentsthat also call into question the idea that priming duringthe test plays a role in false recall and false recognition
Balota et al (1999) obtained further evidence sup-porting the activationmonitoring interpretation of falsememories in the DRM paradigm They compared per-formance of younger adults healthy older adults andolder adults with Alzheimerrsquos disease The latter twogroups have been shown to have intact activation pro-cesses as measured by semantic priming paradigms(Balota amp Duchek 1988 1991) yet they are deficient insource-monitoring processes which is probably attribut-able to impaired frontal lobe functioning (see BalotaDolan amp Duchek 2000) Therefore one might expectthat whereas veridical recall would be worse in older adultsand Alzheimerrsquos patients (relative to younger adults)false recall and false recognition would occur at the samelevel or perhaps be increased in these groups IndeedBalota et al (1999) reported exactly this pattern A sim-ilar outcome has been obtained in other laboratories(eg Budsen Daffner Desikan amp Schacter 2000 Nor-man amp Schacter 1997 Tun Wingfield Rosen amp Blan-chard 1998) This pattern is just as predicted by the ac-tivation monitoring framework
In short the present results and many others (seeRoediger Balota amp Watson 2001 Roediger et al 1998)fit comfortably within the activationmonitoring frame-work In the remainder of our discussion we will con-sider briefly the implications of our present results forother accounts of the DRM paradigm and for the role ofconnectivity (or interitem associative strength) in pro-ducing veridical and false recall
Fuzzy-Trace TheoryAnother leading theory that has been used to explain
false memory phenomena is fuzzy-trace theory (egReyna amp Brainerd 1995) Payne et al (1996) and Schac-ter et al (1996) among many others have used the basictheory to account for false recall and false recognitiondata within the DRM paradigm and the theory has beenextended in interesting ways to account for new varia-tions on the paradigm (eg Brainerd Wright Reyna ampMojardin 2001) The fundamental assumption of fuzzy-trace theory is that experience leaves two types of mem-ory traces gist traces (which capture the meaning of ex-perience but without specific attributes) and verbatimtraces (which represent specific attributes) False memo-ries are attributed to gist processing When the meaning ofa recognition probe matches the gist of the encoded expe-rience false recognition occurs The theory has been di-rected primarily to false recognition and not false recall
In trying to explain the present results one difficultyis that the operational definition of gist has never beenspecified that is within the DRM paradigm given a list
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 395
of 15 related words how does one specify the gist repre-sentation of the list and quantify how strong it is Onecould use patterns of false recognitionmdashto the extentthat a concept is falsely recognized it was represented aspart of the gist representation This definition is per-fectly suitable for some purposes but clearly explainingfalse recognition is not one of them owing to the prob-lem of circularity To explain why some lists lead to highlevels of false recall and false recognition (whereas otherlists lead to low levels) within fuzzy-trace theory oneneeds an independent specification of gist which to ourknowledge has not been provided in this context Whydoes the sweet list have so much gist that it leads to falserecall whereas the bitter list does not One plausible in-terpretation of gist is BAS The more the list items areassociated to the critical item the more a gist represen-tation of the critical item is created This solution to adefinition of gist works within the DRM paradigm butof course the interpretation of how gist is created be-comes isomorphic to activation processes within the ac-tivationmonitoring framework
Interpretation of the other factor that was significantwithin our multiple regression analysis may be morestraightforward within fuzzy-trace theory The negativecorrelation between veridical recall and false recall (andfalse recognition) may be accounted for by increased re-liance on verbatim traces That is if increased recall of listitems is due to increased verbatim information subjectsmay rely less heavily on gist information and hence showless false recall and false recognition during respondingSimilarly for critical items that are long andor ortho-graphically distinct recall of list items that relies on thesefeatures may lead to suppression of false recall and falserecognition of the critical items because they may differsystematically from other list items These factors pos-tulated within the context of fuzzy-trace theory are quitesimilar to those within the activationmonitoring frame-work outlined above At some level of generality there aretwo sets of factors in each account and both theories canexplain many empirical results
Shifting Criterion TheoryMiller and Wolford (1999) proposed that false recog-
nition in the DRM paradigm could be explained by sub-jectsrsquo shifting their criteria differentially to list items andto critical items during a recognition test (They ex-tended their explanation to false recall but somewhattenuously by proposing a generaterecognize theory)They postulated that subjects encoded the structure ofthe list during its presentation and then during the recog-nition test responded with especially liberal criteria tocritical items but not to list items The criterion shift issupposed to occur to the degree that the test item appearsto be related to the list items To support their argumentthey reported that a measure of bias from signal detec-tion theory was indeed more liberal for critical itemsthan for list items However Wickens and Hirshman(2000) and Wixted and Stretch (2000) pointed out that
within signal detection theory differences in bias mea-sures do not necessarily implicate a criterion shiftRather differences in measured bias could arise fromdifferences among the underlying distributions Wixtedand Stretch argued that this latter interpretation was themore likely and that the criterion shift interpretation wasuntenable (also see Roediger amp McDermott 1999)
The present evidence also weighs heavily against acriterion shift account Miller and Wolford (1999) ap-pealed to subjectsrsquo metaknowledge of the structure of thelist Within the criterion shift account we can assumethat the likelihood that a critical item cues metaknowl-edge of the list depends on its FAS or how well the crit-ical item will cue the list If so FAS (the association ofthe critical item to the list items) should be the primarydeterminant of false recognition and false recall How-ever the correlation between FAS and false recall in ourdata was negligible (08) and the same was true in analy-ses of false recognition BAS was actually the primarypredictor of false recall which implicates activation pro-cesses during encoding as being a primary cause of theeffect There would be no reason to expect BAS to haveany impact according to the idea that a simple criterionshift during the test causes false recall and false recog-nition Of course all our lists were relatively high androughly equivalent in terms of FAS because they werecreated from forward associations to the critical item sorestriction of range may account for the low correlationEven so the same problem exists for the shifting crite-rion theory to explain the results If FAS is supposed todetermine false recall and false recognition by this ac-count and if all our lists were high and roughly equal onthis feature why were so many lists ineffective in elicitingfalse recognition and false recall In short either way theshifting criterion theory cannot account for the results
Of course within the activationmonitoring frame-work the monitoring processes occurring during retrievalcould be conceived as conservative shifts in response cri-teria However in our theory these monitoring processesdetermine the degree to which a given level of activationat retrieval will be judged sufficient to endorse a criticalitem as having been studied thereby producing a falsememory In contrast in Miller and Wolfordrsquos (1999) ac-count there was no activation of the critical item duringencoding and shifting criteria during the test were saidto account entirely for the false recognition phenomenon(see Wixted amp Stretch 2000) In the activationmonitoringaccount the memory illusion provoked by the DRM listsis assumed to be a subjectively powerful phenomenonthat can be ameliorated (but rarely eliminated) when listitems are made more distinctive relative to criticalitems Under conditions in which the subject is warnedat test (but not before the encoding phase) to monitortheir memories for the presence or absence of the criti-cal item and is discouraged from making criterion shiftsthe false memory effect is neither eliminated nor greatlyreduced (Gallo Roediger amp McDermott 2001) Hicksand Marsh (2001) showed that under conditions in which
396 ROEDIGER WATSON MCDERMOTT AND GALLO
subjects are explicitly told to monitor their memories dur-ing the test (and are asked to judge whether each wordhad been externally presented or internally generated)false recognition in the DRM paradigm actually increasesrather than decreases In fact the false recognition effectpersists even when subjects are fully informed about thenature of the phenomenon before the study episode andare therefore encouraged to enhance monitoring processesduring both encoding and retrieval in order to attempt todiminish or eliminate the effect (see Gallo et al1997Gallo Roediger amp McDermott 2001 McDermott ampRoediger 1998) For these and other reasons the notionthat the DRM memory illusion is caused by a general shiftof criteria can be eliminated as a potential explanation
Interitem Associative Strength or ConnectivityOur data pose something of a puzzle about the role of
interitem associative strength (Deese 1959a) or connec-tivity (McEvoy et al 1999) in false recall As was dis-cussed in the introduction Deese (1959a) found a sig-nificant negative correlation between this construct andthe probability of false recall In a later paper Deese(1961) obtained a weaker negative correlation that failedto reach conventional levels of significance so the issueis somewhat in doubt from Deesersquos (1959a 1961) researchMcEvoy et al created lists that varied in connectivity(interitem associative strength) and in BAS and manip-ulated these factors orthogonally They replicated thefinding of Deese (1959a) in that the density of inter-connections of list words was negatively related to falserecall However in our analysis of 55 lists (different fromthose used by McEvoy et al 1999) we found no relationbetween connectivity and false recall (r 5 04) or be-tween connectivity and false recognition (r 5 03)
Further research will be necessary to determine thetrue state of affairs although a case can be made that thelack of correlation we obtained is accurate First we had55 lists that varied widely in their ability to elicit false re-call as compared with 24 lists used in McEvoy et al(1999) Our lists also represented a relatively full rangeof connectivity or interitem associative strength relativeto those of McEvoy et al McEvoy et al used lists thathad a somewhat wider range of connectivity because theyused extreme values whereas our larger set of lists pro-vided a fuller body of the middle ranges of connectivityGiven our results it seems at least possible that interitemassociative strength (connectivity) is not a factor in de-termining whether lists of associative words lead to falserecall although the issue remains in doubt
We did find that interitem associative strength or con-nectivity is positively and significantly correlated withveridical recall (r 5 132) This outcome is consistentwith that of Deese (1959a) and McEvoy et al (1999)The greater the associative bonds among list items thegreater is list recall This outcome is also consistent withvirtually any associative theory of recall (eg Anderson1972 1983 Nelson Schreiber amp McEvoy 1992)
CONCLUSION
The present study has uncovered two primary factorsthat are significantly related to false recall and falserecognition in the DRM paradigm BAS is positivelycorrelated with false recall whereas veridical recall ofthe list items is negatively related to false recall The ac-tivationmonitoring account provides a natural interpre-tation of these factors Any viable theory posited to ex-plain the occurrence of false memories in the DRMparadigm must explain the variability in false recall andfalse recognition across lists and the two primary factorsresponsible for this variability
REFERENCES
Anderson J R (1972) FRAN A simulation model of free recall InG H Bower (Ed) The psychology of learning and motivation (Vol 5pp 315-378) New York Academic Press
Anderson J R (1983) The architecture of cognition CambridgeMA Harvard University Press
Anderson J R amp Pirolli P L (1984) Spread of activation Journalof Experimental Psychology Learning Memory amp Cognition 10791-798
Balota D A Cortese M J Duchek J M Adams D RoedigerH L III McDermott K B amp Yerys B E (1999) Veridical andfalse memories in healthy older adults and in dementia of the Alz-heimers type Cognitive Neuropsychology 16 361-384
Balota D A Dolan P O amp Duchek J M (2000) Memorychanges in healthy older adults In E Tulving amp F I M Craik (Eds)The Oxford handbook of memory (pp 395-409) Oxford Oxford Uni-versity Press
Balota D A amp Duchek J M (1988) Age-related differences inlexical access spreading activation and simple pronunciation Psy-chology and Aging 3 84-93
Balota D A amp Duchek J (1991) Semantic priming effects lexicalrepetition effects and contextual disambiguation effects in healthyaged individuals and individuals with senile dementia of theAlzheimer type Brain amp Language 40 181-201
Bartlett F C (1932) Remembering A study in experimental and so-cial psychology Cambridge Cambridge University Press
Brainerd C J Wright R Reyna V F amp Mojardin A H (2001)Conjoint recognition and phantom recollection Journal of Experi-mental Psychology Learning Memory amp Cognition 27 307-327
Bruce D amp Winograd E (1998) Remembering Deesersquos 1959 arti-cles The Zeitgeist the sociology of science and false memoriesPsychonomic Bulletin amp Review 5 615-624
Bruner J S (1957) On perceptual readiness Psychological Review64 123-152
Budson A E Daffner K R Desikan R amp Schacter D L(2000) When false recognition is unopposed by true recognitionGist-based memory distortion in Alzheimerrsquos disease Neuropsy-chology 14 277-287
Collins A M amp Loftus E F (1975) A spreading-activation theoryof semantic memory Psychological Review 82 407-428
Coltheart M Davelaar E Jonasson J T amp Besner D (1977)Access to the internal lexicon In S Dormic (Ed) Attention and per-formance VI (pp 535-555) Hillsdale NJ Erlbaum
Deese J (1959a) Influence of interitem associative strength upon im-mediate free recall Psychological Reports 5 235-241
Deese J (1959b) On the prediction of occurrence of particular verbalintrusions in immediate recall Journal of Experimental Psychology58 17-22
Deese J (1961) Associative structure and the serial reproduction ex-periment Journal of Abnormal amp Social Psychology 63 95-100
Deese J (1965) The structure of associations in language and thoughtBaltimore Johns Hopkins University Press
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 397
Gallo D A McDermott K B Percer J M amp Roediger H LIII (2001) Modality effects in false recall and false recognitionJournal of Experimental Psychology Learning Memory amp Cogni-tion 27 339-353
Gallo D A Roberts M J amp Seamon J G (1997) Rememberingwords not presented in lists Can we avoid creating false memoriesPsychonomic Bulletin amp Review 4 271-276
Gallo D A amp Roediger H L III (2001) Variability among wordlists in evoking associative memory illusions Manuscript submittedfor publication
Gallo D A Roediger H L III amp McDermott K B (2001) As-sociative false recognition occurs without strategic criterion shiftsPsychonomic Bulletin amp Review 8 579-586
Hicks J L amp Marsh R L (2001) False recognition occurs more fre-quently during source identification than during oldndashnew recogni-tion Journal of Experimental Psychology Learning Memory ampCognition 27 375-383
Hunt R R amp Einstein G O (1981) Relational and item-specific in-formation in memory Journal of Verbal Learning amp Verbal Behav-ior 19 497-514
Hunt R R amp McDaniel M A (1993) The enigma of organizationand distinctiveness Journal of Memory amp Language 32 421-445
Israel L amp Schacter D L (1997) Pictorial encoding reduces falserecognition of semantic associates Psychonomic Bulletin amp Review4 577-581
Jacoby L L Kelley C M amp Dywan J (1989) Memory attribu-tions In H L Roediger III amp F I M Craik (Eds) Varieties of mem-ory and consciousness Essays in honour of Endel Tulving (pp 391-422) Hillsdale NJ Erlbaum
Johnson M K Hashtroudi S amp Lindsay D S (1993) Sourcemonitoring Psychological Bulletin 114 3-28
Johnson M K amp Raye C L (1981) Reality monitoring Psycho-logical Review 88 67-85
Kellogg RT (2001) Presentation modality and mode of recall in ver-bal false memory Journal of Experimental Psychology LearningMemory amp Cognition 27 913-919
Kirkpatrick E A (1894) An experimental study of memory Psy-chological Review 1 602-609
Ku Iumlcera H amp Francis W (1967) Computational analysis of present-day American English Providence RI Brown University Press
Marsh E J McDermott K B amp Roediger H L III (2001) Doretrieval factors play a role in the creation of of false memories Man-uscript submitted for publication
Mather M Henkel L A amp Johnson M J (1997) Evaluatingcharacteristics of false memories Rememberknow judgments andmemory characteristics questionnaire compared Memory amp Cogni-tion 25 826-837
McDermott K B (1996) The persistence of false memories in list re-call Journal of Memory amp Language 35 212-230
McDermott K B (1997) Priming on perceptual implicit memorytests can be achieved through presentation of associates Psycho-nomic Bulletin amp Review 4 582-586
McDermott K B amp Roediger H L III (1998) Attempting to avoidillusory memories Robust false recognition of associates persistsunder conditions of explicit warnings and immediate testing Journalof Memory amp Language 39 508-520
McDermott K B amp Watson J M (2001) The rise and fall of falserecall The impact of presentation duration Journal of Memory ampLanguage 45 160-176
McEvoy C L Nelson D L amp Komatsu T (1999) What is the con-nection between true and false memories The differential roles of in-teritem associations in recall and recognition Journal of Experimen-tal Psychology Learning Memory amp Cognition 25 1177-1194
Meyer D E amp Schvaneveldt R W (1971) Facilitation in recog-nizing pairs of words Evidence of a dependence between retrievaloperations Journal of Experimental Psychology 90 227-234
Miller M B amp Wolford G L (1999) Theoretical commentaryThe role of criterion shift in false memory Psychological Review106 398-405
Neely J H (1977) Semantic priming and retrieval from semanticmemory Roles of inhibitionless spreading activation and limited-
capacity attention Journal of Experimental Psychology General106 226-254
Nelson D L McEvoy C L amp Schreiber T A (1999) The Uni-versity of South Florida word association rhyme and word fragmentnorms Unpublished manuscript University of South Florida Tampa
Nelson D L Schreiber TA amp McEvoy C L (1992) Processing im-plicit and explicit representations Psychological Review 99 322-348
Norman K A amp Schacter D L (1997) False recognition in youngerand older adults Exploring the characteristics of illusory memoriesMemory amp Cognition 25 838-848
Paivio A Yuille J C amp Madigan S A (1968) Concreteness im-agery and meaningfulness values for 925 nouns Journal of Experi-mental Psychology Monographs 76 (1 Pt 2)
Payne D G Elie C J Blackwell J M amp Neuschatz J S (1996)Memory illusions Recalling recognizing and recollecting events thatnever occurred Journal of Memory amp Language 35 261-285
Reyna V F amp Brainerd C J (1995) Fuzzy-trace theory An interimsynthesis Learning amp Individual Differences 7 1-75
Roediger H L III Balota D A amp Watson J M (2001) Spreadingactivation and the arousal of false memories In H L Roediger III J SNairne I Neath amp A M Surprenant (Eds) The nature of remem-bering Essays in honor of Robert G Crowder (pp 95-115) Wash-ington DC American Psychological Association
Roediger H L III amp McDermott K B (1995) Creating false mem-ories Remembering words not presented in lists Journal of Experi-mental Psychology Learning Memory amp Cognition 21 803-814
Roediger H L III amp McDermott K B (1999) False alarms aboutfalse memories Psychological Review 106 406-410
Roediger H L III amp McDermott K B (2000) Tricks of memoryCurrent Directions in Psychological Science 9 123-127
Roediger H L III McDermott K B amp Pisoni D B (2001) Rec-ollection of illusory voices Manuscript submitted for publication
Roediger H L III McDermott K B amp Robinson K J (1998)The role of associative processes in creating false memories In M AConway S E Gathercole amp C Cornoldi (Eds) Theories of memoryII (pp 187-245) Hove UK Psychological Press
Russell W A amp Jenkins J J (1954) The complete Minnesotanorms for responses to 100 words from the KentndashRosanoff Word As-sociation Test (Tech Rep No 11 Contract N8 ONR 66216 Officeof Naval Research) University of Minnesota Minneapolis
Schacter D L Israel L amp Racine C (1999) Suppressing falserecognition in younger and older adults The distinctiveness heuris-tic Journal of Memory amp Language 40 1-24
Schacter D L Verfaellie M amp Pradere D (1996) The neuro-psychology of memory illusions False recall and recognition in am-nesic patients Journal of Memory amp Language 35 319-334
Seamon J G Luo C R amp Gallo D A (1998) Creating false mem-ories of words with or without recognition of list items Evidence fornonconscious processes Psychological Science 9 20-26
Smith R E amp Hunt R R (1998) Presentation modality affects falsememory Psychonomic Bulletin amp Review 5 710-715
Stadler M A Roediger H L III amp McDermott K B (1999)Norms for word lists that create false memories Memory amp Cogni-tion 27 494-500
Thapar A amp McDermott K B (2001) False recall and false recog-nition induced by presentation of associated words Effects of retentioninterval and level of processing Memory amp Cognition 29 424-432
Toglia M P amp Battig W F (1978) Handbook of semantic wordnorms Hillsdale NJ Erlbaum
Toglia M P Neuschatz J S amp Goodwin K A (1999) Recall ac-curacy and illusory memories When more is less Memory 7 233-256
Tulving E (1974) Cue-dependent forgetting American Scientist 6274-82
Tulving E (1979) Relation between encoding specificity and levelsof processing In L S Cermak amp F I M Craik (Eds) Levels of pro-cessing in human memory (pp 405-428) Hillsdale NJ Erlbaum
Tulving E (1985) Memory and consciousness Canadian Psycholo-gist 26 1-12
Tun P A Wingfield A Rosen M J amp Blanchard L (1998)Response latencies for false memories Gist-based processes in nor-mal aging Psychology amp Aging 13 230-241
398 ROEDIGER WATSON MCDERMOTT AND GALLO
Underwood B J (1965) False recognition produced by implicit ver-bal responses Journal of Experimental Psychology 70 122-129
Weldon M S (1991) Mechanisms underlying priming on perceptualtests Journal of Experimental Psychology Learning Memory ampCognition 17 526-541
Wickens T D amp Hirshman E (2000) False memories and statisticaldecision theory Comment on Miller and Wolford (1999) and Roedi-ger and McDermott (1999) Psychological Review 107 377-383
Wixted J T amp Stretch V (2000) The case against a criterion-shiftaccount of false memory Psychological Review 107 368-376
Zechmeister E B (1969) Orthographic distinctiveness Journal ofVerbal Learning amp Verbal Behavior 8 754-761
NOTES
1 The orthographic distinctiveness of a critical item was obtained bypresenting 72 subjects with all 55 critical items visually and havingthem rate the words in accordance with procedures developed by Zech-meister (1969) Briefly the subjects were instructed to rate each wordfor its physical appearance in terms of distinctiveness in the languageon a 9-point scale with 1 being not very distinctive 5 corresponding toaverage distinctiveness and 9 being most distinctive To avoid orderingeffects three different randomized orderings of words were used acrossroughly equal groups of subjects In the case of the critical item sleep
for example the distinctiveness is 515 Corresponding values for but-terfly and for man are 744 and 213 respectively Although ortho-graphic distinctiveness was not included in our multiple regressionanalyses this predictor is included in Appendix C
2 The orthographic neighborhood size or Coltheartrsquos N of each crit-ical item was defined as the number of words that can be formed bychanging a single letter in the word (Coltheart Davelaar Jonasson ampBesner 1977) For example because it has no orthographic neighborsColtheartrsquos N for the critical item butterfly is zero In contrast sleep hasfive orthographic neighbors (ie sheep sleek sleet steep and sweep)The reference set of words used were those in Ku Iumlcera and Francisrsquos(1967) norms Although Coltheartrsquos N was not included in our multipleregression analyses this predictor is included in Appendix C
3 Word length orthographic distinctiveness and orthographic neigh-borhood size of the critical item were all highly correlated Specificallythe bivariate correlations among these three predictors were as followslength and orthographic distinctiveness r 5 184 t(53) 5 1117 p lt05 orthographic distinctiveness and orthographic neighborhood sizer 5 66 t(53) 5 640 p lt 05 length and orthographic neighbor-hood size r 5 63 t(53) 5 582 p lt 05 Of these three variablesword length had the largest bivariate correlation with false recall (r 5
37) so it was the predictor used in the multiple regression analyses4 The multiple regression data matrix is available at wwwiacwustl
edunmclWeb
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 399
APPENDIX A The Fifty-Five 15-Word Lists and Their Associated Critical Items
With Backward Associative Strength (BAS) and Forward Associative Strength (FAS)Values Used in the Multiple Regression Analysis of False Recall
Item Word BAS FAS Item Word BAS FAS
Anger mad 393 412 Black white 655 557fear 020 059 dark 111 100hate 028 109 cat 000 043rage 541 042 charred 023 000temper 182 000 night 000 021fury 306 000 funeral 034 000ire 179 000 color 074 050wrath 128 000 grief 000 000happy 000 042 blue 028 000fight 034 000 death 016 014hatred 070 000 ink 020 000mean 090 000 bottom 000 000calm 000 000 coal 288 000emotion 000 000 brown 338 000enrage 378 000 gray 365 000
Mean 157 044 Mean 130 052Army Navy 543 500 Bread butter 364 487
NotemdashUnless otherwise indicated with an asterisk all BAS and FAS values are drawn from the associative norms of Nel-son McEvoy and Schreiber (1999) Those values shown with an asterisk were obtained by the authors by using normingprocedures similar to those of Nelson et al (1999) The corresponding mean for these BAS and FAS values for each list arealso represented
406 ROEDIGER WATSON MCDERMOTT AND GALLO
APPENDIX B Data Matrix of the Fifty-Five 15-Word Lists and Associated Critical Items
Used in the Multiple Regression Analysis of False Recall
Critical False False Log Vrd Raw Ortho Item Recall Recgn Length Freq Concrete FAS BAS Connect Recall Freq Dist Colt N
(Manuscript received July 10 2000revision accepted for publication February 23 2001)
386 ROEDIGER WATSON MCDERMOTT AND GALLO
Roediger and McDermott (1995) developed a para-digm for the creation and experimental investigation offalse memories that was an adaptation of Deesersquos tech-nique Following a suggestion made by Endel Tulvingthis procedure is now referred to as the DRM paradigman acronym for Deese Roediger and McDermott Thebasic procedure is straightforward Subjects hear lists of15 words that are the strongest associates to a missingword or critical item in word association norms (NelsonMcEvoy amp Schreiber 1999 Russell amp Jenkins 1954)For example subjects might be presented with a list ofwords like bed rest awake tired dream wake snoozeblanket doze slumber snore nap peace yawn drowsymdashall of which are associated to the (nonpresented) criticalitem sleep Immediately following the presentation of astudy list subjects recall as many of the list words aspossible without respect to their order (ie single-trialfree recall) In addition subjects are instructed to be surethat each word they write down was indeed on the pre-ceding list Despite this warning against guessing sub-jects recall these critical nonpresented items with aboutthe same probability as items appearing in the middle ofthe list (ie excluding the primacy and recency effects)At the end of the experiment a recognition test that cov-ers many lists is usually given (often with metamemoryjudgments) Recognition of the critical lures typicallyequals or exceeds recognition of the studied words Theseremarkably high levels of false recall and false recogni-tion in the DRM paradigm have been widely replicated(eg Payne Elie Blackwell amp Neuschatz 1996 andRoediger McDermott amp Pisoni 2001 among manyothers)
The DRM paradigm is quite different from many otherpopular techniques used to study false memories Mostdemonstrations of false memory phenomena have usedrelatively naturalistic materials such as stories or visualscenes fairly complex techniques such as the presenta-tion of misleading information or relatively long delaysbetween the study of the material and its later testing Infact the underlying assumption (beginning at least withBartlettrsquos 1932 influential pronouncements) seems tobe that these conditions are necessary to create illusorymemories because the use of such materials as wordlists encourages reliance on reproductive rather than re-constructive memory However the levels of false re-membering with the DRM paradigm are among the mostrobust ever obtained in the experimental literature eventhough they occur under conditions of immediate test-ing with no misleading information with lists of wordsand with strong warnings against guessing and makingerrors Therefore Roediger and McDermott (1995) sug-gested that the distinction between reproductive and re-constructive modes of retention should be abandoned Insome sense all recollection even immediate recall andrecognition of lists of words is reconstructive
Although the DRM paradigm is a relative newcomerto the investigation of false memories considerable re-search has been conducted using it and variations on its
basic theme The purpose of this paper is to consider anoverlooked issue in the growing literature on false mem-oriesmdashnamely the characteristics of the materials usedto produce the effect In Roediger and McDermottrsquos(1995) first experiment they selected the 6 lists used byDeese (1959b) that produced the highest levels of intru-sion in his experiment and they replicated his effects Intheir second experiment they used a total of 24 falsememory lists that included the 6 lists from their first ex-periment along with 18 new lists Some of the new listshad been used by Deese whereas Roediger and McDer-mott (1995) created some of the others using similar pro-cedures
One remarkable outcome that we have noticed repeat-edly while working with these lists is the striking vari-ability they exhibit in eliciting false recall and falserecognition of the associatively related critical itemThese differences occur despite the fact that all the listsare created in the same manner to include with someminor variation the first 15 associates of a single wordthat is itself not presented With few exceptions later re-searchers have used Roediger and McDermottrsquos (1995)original 24 lists to study various interesting issues col-lapsing the resulting data across the lists as did Roedi-ger and McDermott (1995) However few investigatorshave been concerned with the fundamental issue of iden-tifying characteristics of lists that influence the effectThat is why do some lists work so well in producing thefalse memory effect whereas others do not when all ofthe lists are constructed in the same general manner Inaddition examination of the lists often reveals no obvi-ous reasons for their variations in effectiveness at elicit-ing false recall For example consider the lists associ-ated with bitter (sweet sour taste chocolate rice coldlemon angry hard mad acid almonds herbs grapefruit) and sweet (sour candy sugar bitter good tastetooth nice honey soda chocolate heart cake tart pie)Despite the fact that both words refer to concepts in thesame domain represent the same part of speech and haveoverlapping words in the lists they differ dramatically inthe effectiveness with which they elicit false recall Themean probability of false recall of bitter when the rele-vant list was presented was only 01 whereas the meanfalse recall of sweet following the relevant list was 54
Stadler Roediger and McDermott (1999) obtainednorms on 36 DRM lists the 24 used by Roediger andMcDermott (1995) and 12 other lists False recall rangedfrom greater than 60 intrusion rates to lists structuredaround the nonpresented words window sleep smelland doctor to 20 and 10 respectively for the listsstructured around fruit and king False recognitionshowed similarly wide variations Although these listswere originally constructed to yield high levels of falseresponding some did not The purpose of the presentstudy was to attempt to identify those characteristics ofthe lists that cause their differential efficacy in produc-ing false recall and false recognition through a simul-taneous multiple regression analysis By identifying
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 387
these factors we should gain evidence bearing on themechanisms that underlie the creation of false memoriesin the DRM paradigm Before turning to this analysiswe review leads provided by prior research in uncover-ing the factors at work
Deesersquos (1959a 1959b) early work produced two im-portant clues about factors that affect both veridical andfalse recall The two factors he identified are associativeconnections among the study list words themselves andassociative connections between the study list words andthe critical item Deese (1959a) provided subjects withlists that varied in what he termed interitem associativestrength ldquodefined as the average relative frequency withwhich all items in a list tend to elicit all other items in thesame list as free associatesrdquo (p 235) He showed that thismeasure correlated highly (r 5 188) with the totalnumber of words recalled from the listsmdashthe strongerthe associative bonds between items in the list the bet-ter that list was recalled More important for present pur-poses interitem associative strength correlated nega-tively with the number of extralist intrusions (r 5 48)However the stronger the associative bonds among listitems the more likely the subjects were to produce thesame intrusion (r 5 155) Therefore associationsamong list items increased recall of list items and de-creased overall intrusions yet when intrusions did occurfor strongly interconnected lists they tended to be thesame item across subjects
Deese (1959b) also examined the influence of asso-ciative connections from the list words to the critical itemon the false recall of the critical item Subjects in this ex-periment heard 36 lists each of which consisted of the12 highest associates to the nonpresented critical itemFalse recall with these lists ranged from 42 for the crit-ical item needle to 0 for the critical item butterfly Deese(1959b) found that one predictor could account for themajority of the variance in the recall of critical itemsacross his 36 lists This variable was the mean backwardassociative strength (BAS) or the average tendency forwords in the study list to elicit the critical item on a freeassociation test In fact Deese (1959b) reported that thecorrelation between a listrsquos mean BAS and the probabil-ity of recall of its associated critical item was 187Roediger and McDermott (1995) argued that this out-come supports the notion that encoding factors are partlyresponsible for the effect That is lists high in BAS mightactivate the critical item consciously or unconsciouslyas the list items are presented Subjects are confrontedwith a reality-monitoring dilemma (Johnson amp Raye1981) during the test ldquoDid I hear the critical item or didI only think of it while listening to the list itemsrdquo
Other data support this interpretation of the produc-tion of false memories in the DRM paradigm Subjectsclaim to remember the occurrence of the critical items inthe study list which may mean that the item was con-sciously aroused during encoding (Roediger amp McDer-mott 1995) or at least unconsciously activated throughspreading activation (Roediger Balota amp Watson 2001)
In addition subjects claim to remember which of twospeakers said the nonpresented critical item when thelists are presented in both a male and a female voice(Payne et al 1996 Roediger McDermott amp Pisoni2001) Similarly when some lists are presented auditorilyand others are presented visually subjects attribute theillusory item as having occurred in the modality in whichthe list had been presented to the same extent as theymake attributions for list items (Gallo McDermott Perceramp Roediger 2001) Furthermore subjects have beenshown to report having rehearsed the critical item duringthe presentation of the study list (Mather Henkel ampJohnson 1997) Finally McDermott (1997) showed thatpresentation of the lists yields priming of the nonpre-sented critical item on implicit memory tests of word-stem and word-fragment completion albeit at a lessermagnitude than would be the case if the critical itemshad actually been presented Because priming on thesetests usually depends on lexical activation (Weldon1991) McDermott (1997) argued that her data were con-sistent with the notion that the presentation of DRM listsproduces lexical activation of the critical item at leastfor some lists
Of course all memory phenomena depend on theinteraction of encoding and retrieval processes (Tulving1974 1979) and we do not intend to suggest that re-trieval processes are unimportant In fact Roediger andMcDermott (1995) argued that the consideration of re-trieval and decision processes are crucial to under-standing this memory illusion and others They cited Ja-cobyrsquos attributional approach (eg Jacoby Kelley ampDywan 1989) and Johnsonrsquos source-monitoring theory(eg Johnson Hashtroudi amp Lindsay 1993) as promis-ing frameworks to guide this arena of inquiry Israel andSchacter (1997) and Smith and Hunt (1998) have bothargued that the distinctiveness of studied items at re-trieval can explain why some manipulations reduce falserecall and false recognition in this paradigm Neverthe-less given the strong activation of the critical item by itssemantic associates at encoding list context clearlyplays an important role in eliciting false memories in theDRM paradigm
To our knowledge McEvoy Nelson and Komatsu(1999) are the only other contemporary researchers whohave attempted to determine the list-based factors thatunderlie the creation of false memories with associativelists In their experiments McEvoy et al manipulatedthe two factors that Deesersquos (1959a 1959b) earlier workhad implicated interitem associative strength (whichthey call connectivity) and the mean BAS of the listwords to the critical item Consistent with Deesersquos (1959a)research they found that lists with high interitem asso-ciative strength yielded higher recall of list words andlower false recall of critical items relative to lists withlower interitem associative strength In addition and alsoconsistent with Deese (1959b) they found that lists withhigh mean BAS yielded more false recall than did listswith low mean BAS They concluded that although con-
388 ROEDIGER WATSON MCDERMOTT AND GALLO
nectivity and mean BAS had opposite influences on the re-call of critical items both variables played an importantrole in creating false memories in the DRM paradigm
In sum both Deese (1959a 1959b) and McEvoy et al(1999) argued that at least two factors are important inpredicting veridical and false recall Associative con-nections from study words to the critical item increasethe probability of false memories whereas associativeconnections among study words decrease the probabilityof false memories However other factors besides meanBAS and connectivity may underlie the recall of criticalitems in the DRM paradigm For example consider theperformance on the king list (ie queen England crownprince George dictator palace throne chess rule sub-jects monarch royal leader reign) Because it has a rel-atively high mean BAS it would be predicted to have ahigh level of false recall However this list elicited thelowest level of recall in the entire set of 36 DRM listsnormed by Stadler Roediger and McDermott (1999)
Gallo and Roediger (2001 Experiment 1) attemptedto produce lists that were constructed in the same generalway as those of Roediger and McDermott (1995) but thatwould like the king list produce low levels of false re-call and false recognition However unlike the king listthe lists created by Gallo and Roediger all had low BASOf the 19 new lists they constructed many had very lowlevels of false recall and false recognition However sev-eral of the lists (eg wish justice) produced relativelyhigh levels of false recall and false recognition muchlike the lists developed by Roediger and McDermott(1995) Clearly BAS is not the entire reason for the widedifferences in the effectiveness of the lists in creating falsememories The aim of the present study is to gain furtherinformation on the characteristics of lists and criticalitems that affect false recall and false recognition
OVERVIEW OF THE PRESENT STUDY
In the present study we used a simultaneous multipleregression analysis to identify the factors predicting re-call of critical items in the DRM paradigm We consid-ered factors that pertain to the critical item itself as wellas factors characterizing the study lists Specificallywith respect to the critical items we used the followingthree variables word length word frequency and con-creteness On the basis of an examination of the Stadleret al (1999) norms and Deesersquos (1959b) original find-ings we hypothesized that factors making a critical itemdistinct might produce low levels of false recall For ex-ample butterfly is a long low-frequency critical itemInterestingly it was never recalled in Deesersquos (1959b) ex-periment and it was recalled by only 1 of Gallo andRoedigerrsquos (2001) subjects
We also assessed the ability of several variables re-lated to the study words to predict the recall of criticalitems in the DRM paradigm In particular we used thefollowing four variables an index of the strength of theassociative connections from the critical item to the
study words or forward associative strength (FAS ieassociations from sleep to bed to rest to awake etc) anindex of the strength of the associative connections fromthe study words to the critical item or BAS (associationsfrom bed rest to sleep) the average number of as-sociative connections among the study words (interitemassociative strength or connectivity) and the averageprobability of recall of the studied words within each ofthe false memory lists (veridical recall) On the basis ofprior research by Deese (1959a 1959b) and McEvoy et al(1999) we hypothesized that BAS and connectivitywould be the best predictors of the differences observedin false recall among critical items More specificallyrecall of critical items should increase as the strength ofassociative connections from the study words to the crit-ical item increases however recall of critical items shoulddecrease as the average number of associative connec-tions among study words increases (As will be shownin our data set only the first of these two predictions wasfulfilled) In sum the goal of the present research was toidentify the factors that underlie the creation of falsememories in the DRM paradigm by explaining the widevariability in the effectiveness of our 55 lists to elicitfalse recall The multiple regression analysis that led toour answers is explained more fully in the Method andResults sections
METHOD
MaterialsIn the present study we obtained data for 55 DRM lists which in-
cluded 24 lists from Roediger and McDermott (1995) 12 others fromthe Stadler et al (1999) norms and 19 lists from the Gallo andRoediger (2001) norms The study words and the associated criticalitem for each list are presented in Appendix A along with the BASand FAS of each item in relation to the critical item The words withinlists are presented in descending order with respect to their strengthof association to the critical item That is the first list item is typicallythe item most often generated as an associate to the critical item Bothveridical and false recall across our 55 lists are quite stable The split-half correlation of veridical recall for the 55 lists is +86 and the split-half correlation for the lists in eliciting false recall is +90 Thereforewe could potentially explain 81 of the variance in false recall
VariablesAs was described in the introduction the variables we used can
be considered as belonging to two categories properties of the crit-ical item itself and properties of the word lists A complete listingof the values of all seven variables (as well as levels of false recalland false recognition) for all 55 lists appears in Appendix B
Critical item variables With respect to the critical items weused the following three variables word length log frequency andconcreteness We consider each below
1 Word length was defined by the number of letters in each crit-ical item and represents the relative distinctiveness of each wordform (longer words being more distinct) For example butterfly hasnine letters whereas sleep has only five letters We also obtained asubjective measure of orthographic distinctiveness 1 and an objec-tive measure of orthographic neighborhood size2 but because ofthe high degree of multicollinearity among these three variables 3
we included only word length of the critical item in the multiple re-gression analyses
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 389
2 The raw frequency of each critical item was obtained from theKu Iumlcera and Francis (1967) norms and is the number of instancesfound in print per million words For example the raw frequency ofthe critical item sleep is 65 and the raw frequency of the criticalitem butterfly is 2 To correct for skewness in the frequency distri-bution for each critical item raw frequencies were transformedinto log frequencies using the following formula log (5 1 rawfrequency) For example the log frequency of sleep is 182 [ie log(655)] whereas the log frequency of butterfly is 40 [ie log (25)]
3 Concreteness ratings for each critical item were obtained fromthe word association norms of Nelson et al (1999) which were inturn obtained in part from the Paivio Yuille and Madigan (1968)and the Toglia and Battig (1978) norms These ratings are obtainedon a scale of 1ndash7 with 1 being the least concrete and 7 the mostconcrete To carry through with our examples sleep is slightlyabove the midpoint in concreteness (with a rating of 474) and but-terfly is more concrete (with a rating of 591)
List variables With respect to the study lists we used the fol-lowing four variables FAS BAS interitem associative strengthand veridical recall We consider each below
1 Forward associative strength Indices of the strength of asso-ciative connections from the critical item to the study words wereobtained from the word association norms of Nelson et al (1999)Specifically for each critical item we obtained the probability thatit elicited each word in its corresponding study list as an associatea measure Deese (1959b) referred to as the FAS For example as isshown in the third column of Appendix A for the critical item sleepwe identified the strength of its connections to bed rest awaketired and the remaining study words We then averaged these asso-ciations to obtain the mean FAS value for each list In the case ofsleep the mean FAS value for its 15 study words is 047 whereasin the case of butterfly the mean FAS value is 033 Note that all 15list items were given as a response to the critical item in the Russelland Jenkins (1954) norms (from which the lists were generated)However owing to discrepancies between these norms and those ofNelson et al (1999) the number of forward connections for eachlist could have theoretically ranged from 0 to 15
2 Backward associative strength An index of the strength of as-sociative connections from the study words to the critical item wasobtained in large part from the Nelson et al (1999) word associa-tion norms However for some of the 55 DRM lists there were afew list words that were missing from these norms For these wordswe collected our own norms using the procedures outlined by Nel-son and colleagues Briefly we gave groups of over 75 undergrad-uates one of three sheets of 80 to-be-normed words the subjectswere instructed to write next to each word the first word it broughtto mind In those cases in which more than one word from a partic-ular study list needed to be normed these words were placed onseparate norming sheets or were separated by at least 30 items onthe same norming sheet At least 86 observations were obtained foreach normed item and 150 or more observations were obtained foran item in some cases
With these two sets of norms we took each study word and de-termined its connection strength to its corresponding critical itemthe measure Deese (1959b) referred to as the BAS For example asis shown in the second column of Appendix A we obtained for eachstudy word (eg bed rest awake tired etc) the probability thatit elicited its critical item (eg sleep) as an associate Thus thismeasure examines connections in the direction opposite to that usedto obtain the FAS measures We then averaged these associations toobtain the mean BAS value for each list For example the meanBAS value for the sleep list is 431 By comparison the mean BASvalue for the butterfly list is 045
3 Interitem associative strength or connectivity The average in-teritem associative strength (Deese 1959a) or connectivity (Mc-Evoy et al 1999) of the study words was obtained for each listusing two steps First following a procedure outlined by McEvoy
et al we constructed a 15 3 15 matrix of study items for each listIn Appendix C we have illustrated the connectivity matrix for thesleep list Second using the Nelson et al (1999) norms and insome instances our own word association norms we identified thepresence of a connection for all 225 combinations of study wordpairs in each listrsquos matrix excluding the 15 self-connections whichwere not determined (listed as ND) For each word pair a nonzeroconnection strength was scored as a 1 and a zero connectionstrength was scored as a 0 In the case of sleep a total of 27 nonzeroconnections exists among the study words (or an average of 180connections per study word) In the case of butterfly a total of 29nonzero connections exists among the study words (or an averageof 193 connections per study word) High mean connectivityscores indicate a greater number of associative relationships amongthe study words whereas low mean connectivity scores indicate alesser number of associative relationships among the study words
4 Veridical recall of the list One final predictor was the averageprobability of recall of the studied words collapsing across serialpositions 1ndash15 for each list These data were obtained from Stadleret al (1999) and Gallo and Roediger (2001) In the case of sleepstudied items were recalled at a 61 level whereas in the case ofbutterfly studied items were recalled at a 70 level As was notedearlier the split-half correlation of the veridical recall measure wasreasonably high at +86
ProcedureTaking the seven factors that were just described along with the
probability of false recall and the probability of false recognitionfor each of the 55 lists we constructed the data matrix shown inAppendix B for our simultaneous multiple regression analysis4
Each of the 55 lists used in the regression analysis is represented byits respective critical item each of which is listed in alphabeticalorder in the f irst column of the matrix The second and thirdcolumns of the matrix contain the probability of false recall andfalse recognition respectively of each critical item from the Stadleret al (1999) and Gallo and Roediger (2001) norms The next threecolumns contain properties of the critical item itself length log fre-quency and concreteness The next four columns contain propertiesof the list that was constructed for each critical item mean FASmean BAS the mean connectivity of the study words and the av-erage recall of the studied items The last three columns contain thecritical item predictors that were excluded from the multiple re-gression analyses raw frequency orthographic distinctiveness andorthographic neighborhood size (Coltheartrsquos N )
In multiple regression analysis the independent variables can beentered either simultaneously (in a single step) or hierarchically (withsome variables entered into the analysis before others) In the for-mer case all independent variables are allowed to compete forshared variance in the dependent variable (eg false recall) in a sin-gle step so that no single variable is given greater priority than isanother In the latter case those independent variables entered inearlier steps are allowed to capture variance in the absence of com-petition from variables to be entered subsequently In this way in ahierarchical analysis earlier variables are given greater prioritythan later ones usually for theoretical reasons For the purpose ofthe present analysis there exists no definitive theoretical justifica-tion for ordering the entry of critical item variables and list vari-ables so we opted for the simultaneous entry of all variables in asingle step
RESULTS
We consider first the bivariate correlations betweenour seven factors and false recall and then we turn to thesimultaneous multiple regression analysis and its find-ings The entire matrix of correlations (Pearsonrsquos r in all
390 ROEDIGER WATSON MCDERMOTT AND GALLO
cases) among the seven predictors and false recall isshown in Table 1
Bivariate CorrelationsIn describing the results we report significant bivari-
ate correlations between critical item variables and falserecall followed by significant correlations between listvariables and false recall To determine these values eachof the critical item and list variables were entered into aregression equation as the sole predictor of false recallOnly three of our seven variables were significantly cor-related ( p lt 05) with false recall the critical item vari-able of length and the list variables of BAS and recall ofthe studied items (Figures 1A and 1B respectively)
Length The length of the critical item and false recallwere negatively correlated [r 5 37 t(53) 5 287]
Backward associative strength Mean BAS or thestrength of the connections between study items andtheir corresponding critical items was positively corre-lated with false recall r 5 173 t(53) 5 773 As isshown in Figure 1A the greater the strength of associa-tion between the list items and the critical item the morelikely is false recall This outcome replicates Deesersquos(1959b) finding of a 187 correlation between BAS andfalse recall which he obtained with a set of thirty-six 12-item lists Although the 173 correlation between thesetwo variables in our set of f ifty-f ive 15-item lists issomewhat smaller than the 187 correlation obtained byDeese BAS is still the strongest correlate of false recallin our entire data set
Veridical recall Recall of list items and false recallwere negatively correlated [r 5 43 t(53) 5 346]As is shown in Figure 1B increasing levels of accuraterecall are associated with decreasing levels of false re-call This outcome is contrary to some ideas about whyfalse recall occurs as will be brought out in the Discus-sion section
As is shown in Table 1 no other variables were corre-lated with false recall at the bivariate level which is sur-prising in some cases For example in terms of criticalitem variables one might have expected word frequencyor concreteness to be correlated with false recall but nei-ther did (Deese 1959b also failed to find a relation be-tween word frequency and false recall) In terms of listvariables one might have expected connectivity to cor-relate with false recall but it did not The lack of a rela-
tion between connectivity and false recall is noteworthybecause Deese (1959a) and McEvoy et al (1999) re-ported a significant relationship between these variablesso that lists with greater connectivity were less likely toelicit false recall of the critical item In the Discussionsection we consider connectivity or interitem associa-tive strength more fully We turn now to the results of thesimultaneous multiple regression analysis
Simultaneous Multiple Regression AnalysisIn the multiple regression analysis the three variables
based on the characteristics of the critical items them-selves (ie length log frequency and concreteness) andthe four variables based on the characteristics of the listsdesigned to elicit the critical items (ie FAS and BASmeasures interitem associative strength or connectivityand the level of recall of list items) were entered simul-taneously into the regression equation to predict false re-call The results of the analysis are shown in Table 2 in-cluding beta weights (b) semipartial correlations (sr)and significance tests for each predictor For signifi-cance tests a t(47) was used for each predictor and theresultant t statistic is expressed as the absolute valuemdashthat is frac12tfrac12 In addition Table 2 displays the results of thesignificance test for the multiple R2 for the entire set ofseven predictor variables
As is shown in Table 2 the results of the regressionanalysis indicated that two list variables were signifi-cantly related to false recall Specifically BAS was thestrongest predictor of false recall (b 5 170 sr 5160 t 5 717) Therefore consistent with our findingsat the bivariate level the greater the strength of associa-tion between study items and their corresponding criti-
Table 1 Correlation Matrix Among Eight Variables False Recall Three Critical Item Variables and Four List Variables
False Log VeridicalRecall Length Frequency Concreteness FAS BAS Connectivity Recall
NotemdashR2 5 68 F(747) 5 1406 MSe 5 0013 p lt 05 p lt 05
Figure 1 Probability of false recall of the critical item from 55 lists (A) as a functionof associative strength of the list items to the critical item or backward associativestrength and (B) as a function of veridical recall of studied items from the lists
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 391
cal item on a free association test the more probable isfalse recall of the critical item Veridical recall was alsoa strong predictor of false recall (b 5 40 sr 5 34t 5 411) As we found at the bivariate level the morewords recalled from a list the less likely that the criticalitem for that list was falsely recalled None of the re-maining critical item or list variables including length ofthe critical item connectivity and FAS explained aunique portion of the total variance in false recall afterthe influence of other predictors had been residualized inthe context of the regression equation Therefore BASand recall of list items appeared to be the major contrib-utors to the multiple R of 182 and the associated mul-tiple R2 of 68 [F(747) 5 1406 MSe 5 0013] We fur-ther explore the implications of these effects in theDiscussion section
Factors Affecting False RecognitionIn our analyses thus far we have concentrated on
false recall exclusively However the norms for our 55lists (collected by Stadler et al 1999 and Gallo ampRoediger 2001) also include values for false recogni-tion The recognition data in these norms were collectedin a final recognition test after lists had been studied andrecalled This confounding of recognition by prior recallalmost certainly affected both veridical and false recog-nition However in experiments directly examining theeffects of recall on recognition in this paradigm and oth-ers the finding is that recall of a list may increase itsrecognition by a moderate amount (typically in the4ndash10 range eg Roediger amp McDermott 1995Roediger McDermott amp Pisoni 2001) In fact some re-searchers have failed to observe any effect at all of recall
392 ROEDIGER WATSON MCDERMOTT AND GALLO
on recognition (Schacter Verfaellie amp Pradere 1996 seeRoediger et al 1998 for a discussion) We conducted thesame regression analyses with our seven predictor vari-ables on false recognition as we did on false recall Wesummarize the basic results and conclusions as follows
There were three significant bivariate correlationswith false recognition including length (r 5 27)BAS (r 5 143) and veridical recall (r 5 52) Over-all these correlations with false recognition were simi-lar to those obtained with false recall However the pos-itive correlation between BAS and false recognition wassmaller than that in the recall data and the negative cor-relation between veridical recall and false recognitionwas larger than that in the recall data To maintain con-sistency with the previous regression analysis on falserecall we used veridical recall rather than veridicalrecognition in this analysis In addition recognition es-timates were based on only three itemslist and wereprobably affected by prior recall
Turning to the results of the simultaneous multiple re-gression analysis on false recognition BAS and veridi-cal recall of list items appeared to be the major contrib-utors to the multiple R of 169 and the associatedmultiple R2 of 48 [F(747) 5 611 MSe 5 002] Table 3shows the summary of the regression analyses Althoughthe regression analysis accounted for less total variancein false recognition than in false recall (48 as opposedto 68 respectively) the same factors seem to be re-sponsible for predicting false recognition and false re-call Of course because measures of recognition alwaysfollowed recall this similar pattern may reflect a carry-over effect of false recall on false recognition Nonethe-less within the limits of this study the factors responsi-ble for false recognition across lists seem to be the sameas those producing false recall
DISCUSSION
To summarize the main findings our results implicatetwo primary factors (of the seven variables we exam-ined) as accounting for most of the variance in false re-call across lists The split-half correlation of our lists was190 so the amount of potentially explainable variancein our dependent variable of false recall is 81 The fac-
tors in our multiple regression accounted for 68 of thetotal variance in false recall or 84 of the explainablevariance (68 4 81 3 100 5 84) The analysis forfalse recognition revealed the same factors at work al-though they accounted for less overall variance Thestrongest factor was BAS consistent with prior work byDeese (1959b) and McEvoy et al (1999) The other fac-tor was the number of items recalled accurately from thelist which is negatively correlated with false recall
We turn next to the theoretical implications of ourfindings We first consider the implications for the acti-vationmonitoring framework that we have developed toexplain false recall and false recognition in the DRMparadigm We then consider the implications for othertheories and f inally turn to several further issues onwhich our results shed light
ActivationMonitoring TheoryIn the theroretical framework that we (eg McDer-
mott amp Watson 2001 Roediger Balota amp Watson 2001Roediger amp McDermott 1995 2000) have used to explainvariations in the probability of false recall and false recog-nition in the DRM paradigm two sets of processes havebeen proposed Specifically we have proposed that pro-cesses occurring during both encoding and retrieval af-fect the probability of false recall and false recognitionin the DRM paradigm Two critical sets of processes in-volve activation and monitoring Although there is a nat-ural alignment of activation to encoding and monitoringto retrieval we note that both processes can potentiallyoccur both during encoding and during retrieval as willbe discussed below The ideas we employ have many pre-cursors which we acknowledge below although theirparticular arrangement here may be unique
A critical distinction in understanding encoding pro-cesses and their interaction with retrieval processes isbetween item-specific and relational processing (Ander-son 1972 Hunt amp Einstein 1981 Hunt amp McDaniel1993) Briefly the notion is that emphasis can be given tothe encoding of individual elements and their features(what Anderson called node-taggingand Hunt and his col-leagues called item-specific processing) or to the encod-ing of relationships among elements (pathway-taggingor relational processing) Item-specific processing should
Table 3 Summary of Multiple Regression Analysis
on False Recognition With Critical Item Variablesand List Variables Used as Predictors
NotemdashR2 5 48 F(747) 5 611 MSe 5 002 p lt 05 p lt 05
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 393
make individual elements more memorable and lead tomore distinctive memories such processing should helpdistinguish veridical memories from false memories be-cause list items would have features of presentation as-sociated with them whereas the critical item sometimeswould not (However if the critical item is strongly acti-vated during encoding it may take on many of the fea-tures of list items) Relational processing (extractingthemes or schemas or constructing networks of associa-tions) should increase the probability of false recall andfalse recognition by leading to inferences or associationsof events that may not actually have occurred
We assume that in encoding and interpreting experi-ences people go beyond the information given in thestimulus (Bartlett 1932 chap 2 Bruner 1957) and ac-tivate related information through inferential processingEach personrsquos experience is coded in terms of his or herown schemata (or associative networks or knowledgestructures) When schemata are activated a process offilling in of ldquomissingrdquo information may occur wherebyitems strongly implied by the schema (but not actuallypresent in the stimulus) may become activated (Bartlett1932) In encoding lists of words in the DRM paradigmpresentation of the list can create activation that spreadsthroughout the lexicalsemantic system and can createimplicit associative responses (Underwood 1965) Thisactivation may arouse concepts that are not presented but that are associated to concepts that are presented(eg Anderson amp Pirolli 1984 Collins amp Loftus 1975Meyer amp Schvaneveldt 1971) The associative activa-tion in this approach can arise from automatic process-ing or from more controlled elaborative processing(Neely 1977) In addition the associative activation ofconcepts that were not presented may occur consciously(subjects may become aware of the concept of sleep andrehearse it during presentation of the list words bed restawake etc) or the concept may be strongly activatedbut never come to conscious awareness The debate overwhether arousal of the critical item in the DRM para-digm is conscious or unconscious is interesting but dif-ficult to decide empirically Some evidence has impli-cated conscious arousal of the concepts (McDermott1997) whereas other evidence implicates unconsciousactivation (Seamon Luo amp Gallo 1998) The correctanswer to the question of whether activation is consciousor unconscious is probably ldquobothrdquomdashthat is the criticalitem probably sometimes comes consciously to mindand even if it does not relevant information may never-theless be activated unconsciously
The high correlation between BAS and false recall aswell as the strong weighting of BAS in the multiple re-gression analysis strongly implicates activation of criti-cal items as an important factor in explaining false recalland false recognition in the DRM paradigm The morestrongly associated list items are to critical items themore activation would be expected to spread to the crit-ical items More colloquially to the extent that the wordson the list spark the associative connection to the critical
item the more likely the critical item is to be falsely re-called This activation factor would seem to be one mainreason why ldquothings suggested to a person by an experi-ence may be honestly reported by him as part of the ex-periencerdquo (Kirkpatrick 1894 p 608) These associatedlists probably all tend to elicit relational processing butonly those with strong BAS to the critical item lead tofalse recall
The second set of factors proposed in activationmon-itoring theory is concerned with the monitoring of mem-ory accuracy Monitoring of encoding processes canoccur during the study phase of an experiment espe-cially under intentional learning conditions In experi-ments in which subjects are warned about false memoryphenomena and are told to pay close attention to lists andto remember only the presented items and not the criti-cal items monitoring is enhanced (Gallo Roberts amp Sea-mon 1997 Gallo Roediger amp McDermott 2001 Mc-Dermott amp Roediger 1998) Monitoring of memorieshas typically been discussed with respect to retrieval dur-ing which time a primary goal is to distinguish fromamong the considerable information brought to con-sciousness that which refers to perception of past eventsfrom that which does not a process referred to as realitymonitoring (Johnson amp Raye 1981) The experience ofremembering events involves mentally traveling back intime and seeming to reexperience the events (Tulving1985) During retrieval recovered memories that appar-ently carry with them details of having been experiencedare generally attributed to be memories of actual pastevents although sometimes this attribution may be inerror (Jacoby et al 1989)
Johnson et al (1993) have described in detail the typesof information that may be used to distinguish eventsthat actually happened from ones that were only imag-ined thought about or read about For example eventsthat actually occurred have more features from the ex-ternal world associated with them (sights sounds etc)whereas those that were internally generated are lesslikely to have these features but more likely to have as-sociated reflections and cognitive operations In theDRM paradigm the fact that the critical item has beenstrongly activated during encoding means that duringretrieval the subject is likely to retrieve the critical itemas a list member To the extent that the critical items arestrongly activated during encoding they will take onsome features similar to those of the list items this is es-pecially true if the critical items come consciously tomind during encoding and are rehearsed in the samemanner as list items To the extent that the features of thecritical item are similar to those of actual list items theprobability of false recall or false recognition is increasedHowever if the list items systematically differ from thecritical item on some dimensions or features retrievalheuristics may be brought into play that can reduce falserecall and false recognition (Israel amp Schacter 1997Schacter Israel amp Racine 1999) Considerable evidencesupports this general source-reality-monitoring frame-
394 ROEDIGER WATSON MCDERMOTT AND GALLO
work that we advocate here (eg Johnson et al 1993Schacter et al 1999) as well as the attributional ap-proach to remembering (Jacoby et al 1989)
The other significant factor arising from our multipleregression analysis may operate during retrieval by aid-ing the monitoring processes that reduce false recall Thenegative correlation between veridical recall of list itemsand false recall of the critical item may indicate that thebetter encoded list items are the more easily they can bedistinguished from the illusory critical item Subjectsmay use item-specific information to aid recall of listitems and suppress recall of the critical items whichwould not carry as much item-specific information Whenexperimental manipulations increase veridical recall ei-ther through longer study time at relatively slow rates(Gallo amp Roediger 2001 McDermott amp Watson 2001Toglia Neuschatz amp Goodwin 1999) or through re-peated study and test opportunities (McDermott 1996Schacter et al 1999) false recall is often decreasedTherefore the negative relation between veridical recalland false recall can be produced experimentally as wellas appearing in our correlationregression analysis How-ever veridical and false recall can be positively related as afunction of other variables too as was shown by McDer-mott (1996) Toglia et al (1999) and Thapar amp McDermott(2001) In these cases greater activation of the criticalitem from relational processing or from greater seman-tic processing is probably at work
The fact that the length andor distinctiveness of thecritical item are negatively correlated with false recallcan be similarly interpreted as the effect of monitoringduring retrieval because the greater distinctiveness ofthe critical item simplifies the monitoring process Forexample if words in the list are generally shorter andhave more regular orthography than does the criticalitem then during retrieval the critical items that are longand distinctive (eg butterfly cabbage) should be moreeasily discriminated from list items as indeed seems tobe the case in the bivariate correlations Interestingly thelengthdistinctiveness factor played a role in determin-ing false recall even though the lists in our norms werepresented auditorily and the orthographic informationat least should not be particularly accessible In addi-tion presenting the DRM lists visually rather than audi-torily reduces false recall and false recognition at leaston tests in which recognition items are presented visually(as opposed to auditorily) or recall is written rather thanspoken (Gallo McDermott et al 2001 Kellogg 2001Smith amp Hunt 1998) This modality difference has alsobeen interpreted as arising from the retrieval of distinctiveinformation operating during the test (Gallo McDermottet al 2001 Smith amp Hunt 1998)
As was noted above activation processes may occurduring retrieval as well as during encoding The morelist items a subject recalls the more primed the criticalitem might become However the present analyses un-dercut the simplest form of this idea If the recalling oflist items activated or primed false recall the simplest
prediction would be that there should be a positive cor-relation between veridical recall and false recall Our datarevealed the opposite pattern showing a negative corre-lation between veridical and false recall Marsh McDer-mott and Roediger (2001) reported other experimentsthat also call into question the idea that priming duringthe test plays a role in false recall and false recognition
Balota et al (1999) obtained further evidence sup-porting the activationmonitoring interpretation of falsememories in the DRM paradigm They compared per-formance of younger adults healthy older adults andolder adults with Alzheimerrsquos disease The latter twogroups have been shown to have intact activation pro-cesses as measured by semantic priming paradigms(Balota amp Duchek 1988 1991) yet they are deficient insource-monitoring processes which is probably attribut-able to impaired frontal lobe functioning (see BalotaDolan amp Duchek 2000) Therefore one might expectthat whereas veridical recall would be worse in older adultsand Alzheimerrsquos patients (relative to younger adults)false recall and false recognition would occur at the samelevel or perhaps be increased in these groups IndeedBalota et al (1999) reported exactly this pattern A sim-ilar outcome has been obtained in other laboratories(eg Budsen Daffner Desikan amp Schacter 2000 Nor-man amp Schacter 1997 Tun Wingfield Rosen amp Blan-chard 1998) This pattern is just as predicted by the ac-tivation monitoring framework
In short the present results and many others (seeRoediger Balota amp Watson 2001 Roediger et al 1998)fit comfortably within the activationmonitoring frame-work In the remainder of our discussion we will con-sider briefly the implications of our present results forother accounts of the DRM paradigm and for the role ofconnectivity (or interitem associative strength) in pro-ducing veridical and false recall
Fuzzy-Trace TheoryAnother leading theory that has been used to explain
false memory phenomena is fuzzy-trace theory (egReyna amp Brainerd 1995) Payne et al (1996) and Schac-ter et al (1996) among many others have used the basictheory to account for false recall and false recognitiondata within the DRM paradigm and the theory has beenextended in interesting ways to account for new varia-tions on the paradigm (eg Brainerd Wright Reyna ampMojardin 2001) The fundamental assumption of fuzzy-trace theory is that experience leaves two types of mem-ory traces gist traces (which capture the meaning of ex-perience but without specific attributes) and verbatimtraces (which represent specific attributes) False memo-ries are attributed to gist processing When the meaning ofa recognition probe matches the gist of the encoded expe-rience false recognition occurs The theory has been di-rected primarily to false recognition and not false recall
In trying to explain the present results one difficultyis that the operational definition of gist has never beenspecified that is within the DRM paradigm given a list
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 395
of 15 related words how does one specify the gist repre-sentation of the list and quantify how strong it is Onecould use patterns of false recognitionmdashto the extentthat a concept is falsely recognized it was represented aspart of the gist representation This definition is per-fectly suitable for some purposes but clearly explainingfalse recognition is not one of them owing to the prob-lem of circularity To explain why some lists lead to highlevels of false recall and false recognition (whereas otherlists lead to low levels) within fuzzy-trace theory oneneeds an independent specification of gist which to ourknowledge has not been provided in this context Whydoes the sweet list have so much gist that it leads to falserecall whereas the bitter list does not One plausible in-terpretation of gist is BAS The more the list items areassociated to the critical item the more a gist represen-tation of the critical item is created This solution to adefinition of gist works within the DRM paradigm butof course the interpretation of how gist is created be-comes isomorphic to activation processes within the ac-tivationmonitoring framework
Interpretation of the other factor that was significantwithin our multiple regression analysis may be morestraightforward within fuzzy-trace theory The negativecorrelation between veridical recall and false recall (andfalse recognition) may be accounted for by increased re-liance on verbatim traces That is if increased recall of listitems is due to increased verbatim information subjectsmay rely less heavily on gist information and hence showless false recall and false recognition during respondingSimilarly for critical items that are long andor ortho-graphically distinct recall of list items that relies on thesefeatures may lead to suppression of false recall and falserecognition of the critical items because they may differsystematically from other list items These factors pos-tulated within the context of fuzzy-trace theory are quitesimilar to those within the activationmonitoring frame-work outlined above At some level of generality there aretwo sets of factors in each account and both theories canexplain many empirical results
Shifting Criterion TheoryMiller and Wolford (1999) proposed that false recog-
nition in the DRM paradigm could be explained by sub-jectsrsquo shifting their criteria differentially to list items andto critical items during a recognition test (They ex-tended their explanation to false recall but somewhattenuously by proposing a generaterecognize theory)They postulated that subjects encoded the structure ofthe list during its presentation and then during the recog-nition test responded with especially liberal criteria tocritical items but not to list items The criterion shift issupposed to occur to the degree that the test item appearsto be related to the list items To support their argumentthey reported that a measure of bias from signal detec-tion theory was indeed more liberal for critical itemsthan for list items However Wickens and Hirshman(2000) and Wixted and Stretch (2000) pointed out that
within signal detection theory differences in bias mea-sures do not necessarily implicate a criterion shiftRather differences in measured bias could arise fromdifferences among the underlying distributions Wixtedand Stretch argued that this latter interpretation was themore likely and that the criterion shift interpretation wasuntenable (also see Roediger amp McDermott 1999)
The present evidence also weighs heavily against acriterion shift account Miller and Wolford (1999) ap-pealed to subjectsrsquo metaknowledge of the structure of thelist Within the criterion shift account we can assumethat the likelihood that a critical item cues metaknowl-edge of the list depends on its FAS or how well the crit-ical item will cue the list If so FAS (the association ofthe critical item to the list items) should be the primarydeterminant of false recognition and false recall How-ever the correlation between FAS and false recall in ourdata was negligible (08) and the same was true in analy-ses of false recognition BAS was actually the primarypredictor of false recall which implicates activation pro-cesses during encoding as being a primary cause of theeffect There would be no reason to expect BAS to haveany impact according to the idea that a simple criterionshift during the test causes false recall and false recog-nition Of course all our lists were relatively high androughly equivalent in terms of FAS because they werecreated from forward associations to the critical item sorestriction of range may account for the low correlationEven so the same problem exists for the shifting crite-rion theory to explain the results If FAS is supposed todetermine false recall and false recognition by this ac-count and if all our lists were high and roughly equal onthis feature why were so many lists ineffective in elicitingfalse recognition and false recall In short either way theshifting criterion theory cannot account for the results
Of course within the activationmonitoring frame-work the monitoring processes occurring during retrievalcould be conceived as conservative shifts in response cri-teria However in our theory these monitoring processesdetermine the degree to which a given level of activationat retrieval will be judged sufficient to endorse a criticalitem as having been studied thereby producing a falsememory In contrast in Miller and Wolfordrsquos (1999) ac-count there was no activation of the critical item duringencoding and shifting criteria during the test were saidto account entirely for the false recognition phenomenon(see Wixted amp Stretch 2000) In the activationmonitoringaccount the memory illusion provoked by the DRM listsis assumed to be a subjectively powerful phenomenonthat can be ameliorated (but rarely eliminated) when listitems are made more distinctive relative to criticalitems Under conditions in which the subject is warnedat test (but not before the encoding phase) to monitortheir memories for the presence or absence of the criti-cal item and is discouraged from making criterion shiftsthe false memory effect is neither eliminated nor greatlyreduced (Gallo Roediger amp McDermott 2001) Hicksand Marsh (2001) showed that under conditions in which
396 ROEDIGER WATSON MCDERMOTT AND GALLO
subjects are explicitly told to monitor their memories dur-ing the test (and are asked to judge whether each wordhad been externally presented or internally generated)false recognition in the DRM paradigm actually increasesrather than decreases In fact the false recognition effectpersists even when subjects are fully informed about thenature of the phenomenon before the study episode andare therefore encouraged to enhance monitoring processesduring both encoding and retrieval in order to attempt todiminish or eliminate the effect (see Gallo et al1997Gallo Roediger amp McDermott 2001 McDermott ampRoediger 1998) For these and other reasons the notionthat the DRM memory illusion is caused by a general shiftof criteria can be eliminated as a potential explanation
Interitem Associative Strength or ConnectivityOur data pose something of a puzzle about the role of
interitem associative strength (Deese 1959a) or connec-tivity (McEvoy et al 1999) in false recall As was dis-cussed in the introduction Deese (1959a) found a sig-nificant negative correlation between this construct andthe probability of false recall In a later paper Deese(1961) obtained a weaker negative correlation that failedto reach conventional levels of significance so the issueis somewhat in doubt from Deesersquos (1959a 1961) researchMcEvoy et al created lists that varied in connectivity(interitem associative strength) and in BAS and manip-ulated these factors orthogonally They replicated thefinding of Deese (1959a) in that the density of inter-connections of list words was negatively related to falserecall However in our analysis of 55 lists (different fromthose used by McEvoy et al 1999) we found no relationbetween connectivity and false recall (r 5 04) or be-tween connectivity and false recognition (r 5 03)
Further research will be necessary to determine thetrue state of affairs although a case can be made that thelack of correlation we obtained is accurate First we had55 lists that varied widely in their ability to elicit false re-call as compared with 24 lists used in McEvoy et al(1999) Our lists also represented a relatively full rangeof connectivity or interitem associative strength relativeto those of McEvoy et al McEvoy et al used lists thathad a somewhat wider range of connectivity because theyused extreme values whereas our larger set of lists pro-vided a fuller body of the middle ranges of connectivityGiven our results it seems at least possible that interitemassociative strength (connectivity) is not a factor in de-termining whether lists of associative words lead to falserecall although the issue remains in doubt
We did find that interitem associative strength or con-nectivity is positively and significantly correlated withveridical recall (r 5 132) This outcome is consistentwith that of Deese (1959a) and McEvoy et al (1999)The greater the associative bonds among list items thegreater is list recall This outcome is also consistent withvirtually any associative theory of recall (eg Anderson1972 1983 Nelson Schreiber amp McEvoy 1992)
CONCLUSION
The present study has uncovered two primary factorsthat are significantly related to false recall and falserecognition in the DRM paradigm BAS is positivelycorrelated with false recall whereas veridical recall ofthe list items is negatively related to false recall The ac-tivationmonitoring account provides a natural interpre-tation of these factors Any viable theory posited to ex-plain the occurrence of false memories in the DRMparadigm must explain the variability in false recall andfalse recognition across lists and the two primary factorsresponsible for this variability
REFERENCES
Anderson J R (1972) FRAN A simulation model of free recall InG H Bower (Ed) The psychology of learning and motivation (Vol 5pp 315-378) New York Academic Press
Anderson J R (1983) The architecture of cognition CambridgeMA Harvard University Press
Anderson J R amp Pirolli P L (1984) Spread of activation Journalof Experimental Psychology Learning Memory amp Cognition 10791-798
Balota D A Cortese M J Duchek J M Adams D RoedigerH L III McDermott K B amp Yerys B E (1999) Veridical andfalse memories in healthy older adults and in dementia of the Alz-heimers type Cognitive Neuropsychology 16 361-384
Balota D A Dolan P O amp Duchek J M (2000) Memorychanges in healthy older adults In E Tulving amp F I M Craik (Eds)The Oxford handbook of memory (pp 395-409) Oxford Oxford Uni-versity Press
Balota D A amp Duchek J M (1988) Age-related differences inlexical access spreading activation and simple pronunciation Psy-chology and Aging 3 84-93
Balota D A amp Duchek J (1991) Semantic priming effects lexicalrepetition effects and contextual disambiguation effects in healthyaged individuals and individuals with senile dementia of theAlzheimer type Brain amp Language 40 181-201
Bartlett F C (1932) Remembering A study in experimental and so-cial psychology Cambridge Cambridge University Press
Brainerd C J Wright R Reyna V F amp Mojardin A H (2001)Conjoint recognition and phantom recollection Journal of Experi-mental Psychology Learning Memory amp Cognition 27 307-327
Bruce D amp Winograd E (1998) Remembering Deesersquos 1959 arti-cles The Zeitgeist the sociology of science and false memoriesPsychonomic Bulletin amp Review 5 615-624
Bruner J S (1957) On perceptual readiness Psychological Review64 123-152
Budson A E Daffner K R Desikan R amp Schacter D L(2000) When false recognition is unopposed by true recognitionGist-based memory distortion in Alzheimerrsquos disease Neuropsy-chology 14 277-287
Collins A M amp Loftus E F (1975) A spreading-activation theoryof semantic memory Psychological Review 82 407-428
Coltheart M Davelaar E Jonasson J T amp Besner D (1977)Access to the internal lexicon In S Dormic (Ed) Attention and per-formance VI (pp 535-555) Hillsdale NJ Erlbaum
Deese J (1959a) Influence of interitem associative strength upon im-mediate free recall Psychological Reports 5 235-241
Deese J (1959b) On the prediction of occurrence of particular verbalintrusions in immediate recall Journal of Experimental Psychology58 17-22
Deese J (1961) Associative structure and the serial reproduction ex-periment Journal of Abnormal amp Social Psychology 63 95-100
Deese J (1965) The structure of associations in language and thoughtBaltimore Johns Hopkins University Press
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 397
Gallo D A McDermott K B Percer J M amp Roediger H LIII (2001) Modality effects in false recall and false recognitionJournal of Experimental Psychology Learning Memory amp Cogni-tion 27 339-353
Gallo D A Roberts M J amp Seamon J G (1997) Rememberingwords not presented in lists Can we avoid creating false memoriesPsychonomic Bulletin amp Review 4 271-276
Gallo D A amp Roediger H L III (2001) Variability among wordlists in evoking associative memory illusions Manuscript submittedfor publication
Gallo D A Roediger H L III amp McDermott K B (2001) As-sociative false recognition occurs without strategic criterion shiftsPsychonomic Bulletin amp Review 8 579-586
Hicks J L amp Marsh R L (2001) False recognition occurs more fre-quently during source identification than during oldndashnew recogni-tion Journal of Experimental Psychology Learning Memory ampCognition 27 375-383
Hunt R R amp Einstein G O (1981) Relational and item-specific in-formation in memory Journal of Verbal Learning amp Verbal Behav-ior 19 497-514
Hunt R R amp McDaniel M A (1993) The enigma of organizationand distinctiveness Journal of Memory amp Language 32 421-445
Israel L amp Schacter D L (1997) Pictorial encoding reduces falserecognition of semantic associates Psychonomic Bulletin amp Review4 577-581
Jacoby L L Kelley C M amp Dywan J (1989) Memory attribu-tions In H L Roediger III amp F I M Craik (Eds) Varieties of mem-ory and consciousness Essays in honour of Endel Tulving (pp 391-422) Hillsdale NJ Erlbaum
Johnson M K Hashtroudi S amp Lindsay D S (1993) Sourcemonitoring Psychological Bulletin 114 3-28
Johnson M K amp Raye C L (1981) Reality monitoring Psycho-logical Review 88 67-85
Kellogg RT (2001) Presentation modality and mode of recall in ver-bal false memory Journal of Experimental Psychology LearningMemory amp Cognition 27 913-919
Kirkpatrick E A (1894) An experimental study of memory Psy-chological Review 1 602-609
Ku Iumlcera H amp Francis W (1967) Computational analysis of present-day American English Providence RI Brown University Press
Marsh E J McDermott K B amp Roediger H L III (2001) Doretrieval factors play a role in the creation of of false memories Man-uscript submitted for publication
Mather M Henkel L A amp Johnson M J (1997) Evaluatingcharacteristics of false memories Rememberknow judgments andmemory characteristics questionnaire compared Memory amp Cogni-tion 25 826-837
McDermott K B (1996) The persistence of false memories in list re-call Journal of Memory amp Language 35 212-230
McDermott K B (1997) Priming on perceptual implicit memorytests can be achieved through presentation of associates Psycho-nomic Bulletin amp Review 4 582-586
McDermott K B amp Roediger H L III (1998) Attempting to avoidillusory memories Robust false recognition of associates persistsunder conditions of explicit warnings and immediate testing Journalof Memory amp Language 39 508-520
McDermott K B amp Watson J M (2001) The rise and fall of falserecall The impact of presentation duration Journal of Memory ampLanguage 45 160-176
McEvoy C L Nelson D L amp Komatsu T (1999) What is the con-nection between true and false memories The differential roles of in-teritem associations in recall and recognition Journal of Experimen-tal Psychology Learning Memory amp Cognition 25 1177-1194
Meyer D E amp Schvaneveldt R W (1971) Facilitation in recog-nizing pairs of words Evidence of a dependence between retrievaloperations Journal of Experimental Psychology 90 227-234
Miller M B amp Wolford G L (1999) Theoretical commentaryThe role of criterion shift in false memory Psychological Review106 398-405
Neely J H (1977) Semantic priming and retrieval from semanticmemory Roles of inhibitionless spreading activation and limited-
capacity attention Journal of Experimental Psychology General106 226-254
Nelson D L McEvoy C L amp Schreiber T A (1999) The Uni-versity of South Florida word association rhyme and word fragmentnorms Unpublished manuscript University of South Florida Tampa
Nelson D L Schreiber TA amp McEvoy C L (1992) Processing im-plicit and explicit representations Psychological Review 99 322-348
Norman K A amp Schacter D L (1997) False recognition in youngerand older adults Exploring the characteristics of illusory memoriesMemory amp Cognition 25 838-848
Paivio A Yuille J C amp Madigan S A (1968) Concreteness im-agery and meaningfulness values for 925 nouns Journal of Experi-mental Psychology Monographs 76 (1 Pt 2)
Payne D G Elie C J Blackwell J M amp Neuschatz J S (1996)Memory illusions Recalling recognizing and recollecting events thatnever occurred Journal of Memory amp Language 35 261-285
Reyna V F amp Brainerd C J (1995) Fuzzy-trace theory An interimsynthesis Learning amp Individual Differences 7 1-75
Roediger H L III Balota D A amp Watson J M (2001) Spreadingactivation and the arousal of false memories In H L Roediger III J SNairne I Neath amp A M Surprenant (Eds) The nature of remem-bering Essays in honor of Robert G Crowder (pp 95-115) Wash-ington DC American Psychological Association
Roediger H L III amp McDermott K B (1995) Creating false mem-ories Remembering words not presented in lists Journal of Experi-mental Psychology Learning Memory amp Cognition 21 803-814
Roediger H L III amp McDermott K B (1999) False alarms aboutfalse memories Psychological Review 106 406-410
Roediger H L III amp McDermott K B (2000) Tricks of memoryCurrent Directions in Psychological Science 9 123-127
Roediger H L III McDermott K B amp Pisoni D B (2001) Rec-ollection of illusory voices Manuscript submitted for publication
Roediger H L III McDermott K B amp Robinson K J (1998)The role of associative processes in creating false memories In M AConway S E Gathercole amp C Cornoldi (Eds) Theories of memoryII (pp 187-245) Hove UK Psychological Press
Russell W A amp Jenkins J J (1954) The complete Minnesotanorms for responses to 100 words from the KentndashRosanoff Word As-sociation Test (Tech Rep No 11 Contract N8 ONR 66216 Officeof Naval Research) University of Minnesota Minneapolis
Schacter D L Israel L amp Racine C (1999) Suppressing falserecognition in younger and older adults The distinctiveness heuris-tic Journal of Memory amp Language 40 1-24
Schacter D L Verfaellie M amp Pradere D (1996) The neuro-psychology of memory illusions False recall and recognition in am-nesic patients Journal of Memory amp Language 35 319-334
Seamon J G Luo C R amp Gallo D A (1998) Creating false mem-ories of words with or without recognition of list items Evidence fornonconscious processes Psychological Science 9 20-26
Smith R E amp Hunt R R (1998) Presentation modality affects falsememory Psychonomic Bulletin amp Review 5 710-715
Stadler M A Roediger H L III amp McDermott K B (1999)Norms for word lists that create false memories Memory amp Cogni-tion 27 494-500
Thapar A amp McDermott K B (2001) False recall and false recog-nition induced by presentation of associated words Effects of retentioninterval and level of processing Memory amp Cognition 29 424-432
Toglia M P amp Battig W F (1978) Handbook of semantic wordnorms Hillsdale NJ Erlbaum
Toglia M P Neuschatz J S amp Goodwin K A (1999) Recall ac-curacy and illusory memories When more is less Memory 7 233-256
Tulving E (1974) Cue-dependent forgetting American Scientist 6274-82
Tulving E (1979) Relation between encoding specificity and levelsof processing In L S Cermak amp F I M Craik (Eds) Levels of pro-cessing in human memory (pp 405-428) Hillsdale NJ Erlbaum
Tulving E (1985) Memory and consciousness Canadian Psycholo-gist 26 1-12
Tun P A Wingfield A Rosen M J amp Blanchard L (1998)Response latencies for false memories Gist-based processes in nor-mal aging Psychology amp Aging 13 230-241
398 ROEDIGER WATSON MCDERMOTT AND GALLO
Underwood B J (1965) False recognition produced by implicit ver-bal responses Journal of Experimental Psychology 70 122-129
Weldon M S (1991) Mechanisms underlying priming on perceptualtests Journal of Experimental Psychology Learning Memory ampCognition 17 526-541
Wickens T D amp Hirshman E (2000) False memories and statisticaldecision theory Comment on Miller and Wolford (1999) and Roedi-ger and McDermott (1999) Psychological Review 107 377-383
Wixted J T amp Stretch V (2000) The case against a criterion-shiftaccount of false memory Psychological Review 107 368-376
Zechmeister E B (1969) Orthographic distinctiveness Journal ofVerbal Learning amp Verbal Behavior 8 754-761
NOTES
1 The orthographic distinctiveness of a critical item was obtained bypresenting 72 subjects with all 55 critical items visually and havingthem rate the words in accordance with procedures developed by Zech-meister (1969) Briefly the subjects were instructed to rate each wordfor its physical appearance in terms of distinctiveness in the languageon a 9-point scale with 1 being not very distinctive 5 corresponding toaverage distinctiveness and 9 being most distinctive To avoid orderingeffects three different randomized orderings of words were used acrossroughly equal groups of subjects In the case of the critical item sleep
for example the distinctiveness is 515 Corresponding values for but-terfly and for man are 744 and 213 respectively Although ortho-graphic distinctiveness was not included in our multiple regressionanalyses this predictor is included in Appendix C
2 The orthographic neighborhood size or Coltheartrsquos N of each crit-ical item was defined as the number of words that can be formed bychanging a single letter in the word (Coltheart Davelaar Jonasson ampBesner 1977) For example because it has no orthographic neighborsColtheartrsquos N for the critical item butterfly is zero In contrast sleep hasfive orthographic neighbors (ie sheep sleek sleet steep and sweep)The reference set of words used were those in Ku Iumlcera and Francisrsquos(1967) norms Although Coltheartrsquos N was not included in our multipleregression analyses this predictor is included in Appendix C
3 Word length orthographic distinctiveness and orthographic neigh-borhood size of the critical item were all highly correlated Specificallythe bivariate correlations among these three predictors were as followslength and orthographic distinctiveness r 5 184 t(53) 5 1117 p lt05 orthographic distinctiveness and orthographic neighborhood sizer 5 66 t(53) 5 640 p lt 05 length and orthographic neighbor-hood size r 5 63 t(53) 5 582 p lt 05 Of these three variablesword length had the largest bivariate correlation with false recall (r 5
37) so it was the predictor used in the multiple regression analyses4 The multiple regression data matrix is available at wwwiacwustl
edunmclWeb
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 399
APPENDIX A The Fifty-Five 15-Word Lists and Their Associated Critical Items
With Backward Associative Strength (BAS) and Forward Associative Strength (FAS)Values Used in the Multiple Regression Analysis of False Recall
Item Word BAS FAS Item Word BAS FAS
Anger mad 393 412 Black white 655 557fear 020 059 dark 111 100hate 028 109 cat 000 043rage 541 042 charred 023 000temper 182 000 night 000 021fury 306 000 funeral 034 000ire 179 000 color 074 050wrath 128 000 grief 000 000happy 000 042 blue 028 000fight 034 000 death 016 014hatred 070 000 ink 020 000mean 090 000 bottom 000 000calm 000 000 coal 288 000emotion 000 000 brown 338 000enrage 378 000 gray 365 000
Mean 157 044 Mean 130 052Army Navy 543 500 Bread butter 364 487
NotemdashUnless otherwise indicated with an asterisk all BAS and FAS values are drawn from the associative norms of Nel-son McEvoy and Schreiber (1999) Those values shown with an asterisk were obtained by the authors by using normingprocedures similar to those of Nelson et al (1999) The corresponding mean for these BAS and FAS values for each list arealso represented
406 ROEDIGER WATSON MCDERMOTT AND GALLO
APPENDIX B Data Matrix of the Fifty-Five 15-Word Lists and Associated Critical Items
Used in the Multiple Regression Analysis of False Recall
Critical False False Log Vrd Raw Ortho Item Recall Recgn Length Freq Concrete FAS BAS Connect Recall Freq Dist Colt N
(Manuscript received July 10 2000revision accepted for publication February 23 2001)
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 387
these factors we should gain evidence bearing on themechanisms that underlie the creation of false memoriesin the DRM paradigm Before turning to this analysiswe review leads provided by prior research in uncover-ing the factors at work
Deesersquos (1959a 1959b) early work produced two im-portant clues about factors that affect both veridical andfalse recall The two factors he identified are associativeconnections among the study list words themselves andassociative connections between the study list words andthe critical item Deese (1959a) provided subjects withlists that varied in what he termed interitem associativestrength ldquodefined as the average relative frequency withwhich all items in a list tend to elicit all other items in thesame list as free associatesrdquo (p 235) He showed that thismeasure correlated highly (r 5 188) with the totalnumber of words recalled from the listsmdashthe strongerthe associative bonds between items in the list the bet-ter that list was recalled More important for present pur-poses interitem associative strength correlated nega-tively with the number of extralist intrusions (r 5 48)However the stronger the associative bonds among listitems the more likely the subjects were to produce thesame intrusion (r 5 155) Therefore associationsamong list items increased recall of list items and de-creased overall intrusions yet when intrusions did occurfor strongly interconnected lists they tended to be thesame item across subjects
Deese (1959b) also examined the influence of asso-ciative connections from the list words to the critical itemon the false recall of the critical item Subjects in this ex-periment heard 36 lists each of which consisted of the12 highest associates to the nonpresented critical itemFalse recall with these lists ranged from 42 for the crit-ical item needle to 0 for the critical item butterfly Deese(1959b) found that one predictor could account for themajority of the variance in the recall of critical itemsacross his 36 lists This variable was the mean backwardassociative strength (BAS) or the average tendency forwords in the study list to elicit the critical item on a freeassociation test In fact Deese (1959b) reported that thecorrelation between a listrsquos mean BAS and the probabil-ity of recall of its associated critical item was 187Roediger and McDermott (1995) argued that this out-come supports the notion that encoding factors are partlyresponsible for the effect That is lists high in BAS mightactivate the critical item consciously or unconsciouslyas the list items are presented Subjects are confrontedwith a reality-monitoring dilemma (Johnson amp Raye1981) during the test ldquoDid I hear the critical item or didI only think of it while listening to the list itemsrdquo
Other data support this interpretation of the produc-tion of false memories in the DRM paradigm Subjectsclaim to remember the occurrence of the critical items inthe study list which may mean that the item was con-sciously aroused during encoding (Roediger amp McDer-mott 1995) or at least unconsciously activated throughspreading activation (Roediger Balota amp Watson 2001)
In addition subjects claim to remember which of twospeakers said the nonpresented critical item when thelists are presented in both a male and a female voice(Payne et al 1996 Roediger McDermott amp Pisoni2001) Similarly when some lists are presented auditorilyand others are presented visually subjects attribute theillusory item as having occurred in the modality in whichthe list had been presented to the same extent as theymake attributions for list items (Gallo McDermott Perceramp Roediger 2001) Furthermore subjects have beenshown to report having rehearsed the critical item duringthe presentation of the study list (Mather Henkel ampJohnson 1997) Finally McDermott (1997) showed thatpresentation of the lists yields priming of the nonpre-sented critical item on implicit memory tests of word-stem and word-fragment completion albeit at a lessermagnitude than would be the case if the critical itemshad actually been presented Because priming on thesetests usually depends on lexical activation (Weldon1991) McDermott (1997) argued that her data were con-sistent with the notion that the presentation of DRM listsproduces lexical activation of the critical item at leastfor some lists
Of course all memory phenomena depend on theinteraction of encoding and retrieval processes (Tulving1974 1979) and we do not intend to suggest that re-trieval processes are unimportant In fact Roediger andMcDermott (1995) argued that the consideration of re-trieval and decision processes are crucial to under-standing this memory illusion and others They cited Ja-cobyrsquos attributional approach (eg Jacoby Kelley ampDywan 1989) and Johnsonrsquos source-monitoring theory(eg Johnson Hashtroudi amp Lindsay 1993) as promis-ing frameworks to guide this arena of inquiry Israel andSchacter (1997) and Smith and Hunt (1998) have bothargued that the distinctiveness of studied items at re-trieval can explain why some manipulations reduce falserecall and false recognition in this paradigm Neverthe-less given the strong activation of the critical item by itssemantic associates at encoding list context clearlyplays an important role in eliciting false memories in theDRM paradigm
To our knowledge McEvoy Nelson and Komatsu(1999) are the only other contemporary researchers whohave attempted to determine the list-based factors thatunderlie the creation of false memories with associativelists In their experiments McEvoy et al manipulatedthe two factors that Deesersquos (1959a 1959b) earlier workhad implicated interitem associative strength (whichthey call connectivity) and the mean BAS of the listwords to the critical item Consistent with Deesersquos (1959a)research they found that lists with high interitem asso-ciative strength yielded higher recall of list words andlower false recall of critical items relative to lists withlower interitem associative strength In addition and alsoconsistent with Deese (1959b) they found that lists withhigh mean BAS yielded more false recall than did listswith low mean BAS They concluded that although con-
388 ROEDIGER WATSON MCDERMOTT AND GALLO
nectivity and mean BAS had opposite influences on the re-call of critical items both variables played an importantrole in creating false memories in the DRM paradigm
In sum both Deese (1959a 1959b) and McEvoy et al(1999) argued that at least two factors are important inpredicting veridical and false recall Associative con-nections from study words to the critical item increasethe probability of false memories whereas associativeconnections among study words decrease the probabilityof false memories However other factors besides meanBAS and connectivity may underlie the recall of criticalitems in the DRM paradigm For example consider theperformance on the king list (ie queen England crownprince George dictator palace throne chess rule sub-jects monarch royal leader reign) Because it has a rel-atively high mean BAS it would be predicted to have ahigh level of false recall However this list elicited thelowest level of recall in the entire set of 36 DRM listsnormed by Stadler Roediger and McDermott (1999)
Gallo and Roediger (2001 Experiment 1) attemptedto produce lists that were constructed in the same generalway as those of Roediger and McDermott (1995) but thatwould like the king list produce low levels of false re-call and false recognition However unlike the king listthe lists created by Gallo and Roediger all had low BASOf the 19 new lists they constructed many had very lowlevels of false recall and false recognition However sev-eral of the lists (eg wish justice) produced relativelyhigh levels of false recall and false recognition muchlike the lists developed by Roediger and McDermott(1995) Clearly BAS is not the entire reason for the widedifferences in the effectiveness of the lists in creating falsememories The aim of the present study is to gain furtherinformation on the characteristics of lists and criticalitems that affect false recall and false recognition
OVERVIEW OF THE PRESENT STUDY
In the present study we used a simultaneous multipleregression analysis to identify the factors predicting re-call of critical items in the DRM paradigm We consid-ered factors that pertain to the critical item itself as wellas factors characterizing the study lists Specificallywith respect to the critical items we used the followingthree variables word length word frequency and con-creteness On the basis of an examination of the Stadleret al (1999) norms and Deesersquos (1959b) original find-ings we hypothesized that factors making a critical itemdistinct might produce low levels of false recall For ex-ample butterfly is a long low-frequency critical itemInterestingly it was never recalled in Deesersquos (1959b) ex-periment and it was recalled by only 1 of Gallo andRoedigerrsquos (2001) subjects
We also assessed the ability of several variables re-lated to the study words to predict the recall of criticalitems in the DRM paradigm In particular we used thefollowing four variables an index of the strength of theassociative connections from the critical item to the
study words or forward associative strength (FAS ieassociations from sleep to bed to rest to awake etc) anindex of the strength of the associative connections fromthe study words to the critical item or BAS (associationsfrom bed rest to sleep) the average number of as-sociative connections among the study words (interitemassociative strength or connectivity) and the averageprobability of recall of the studied words within each ofthe false memory lists (veridical recall) On the basis ofprior research by Deese (1959a 1959b) and McEvoy et al(1999) we hypothesized that BAS and connectivitywould be the best predictors of the differences observedin false recall among critical items More specificallyrecall of critical items should increase as the strength ofassociative connections from the study words to the crit-ical item increases however recall of critical items shoulddecrease as the average number of associative connec-tions among study words increases (As will be shownin our data set only the first of these two predictions wasfulfilled) In sum the goal of the present research was toidentify the factors that underlie the creation of falsememories in the DRM paradigm by explaining the widevariability in the effectiveness of our 55 lists to elicitfalse recall The multiple regression analysis that led toour answers is explained more fully in the Method andResults sections
METHOD
MaterialsIn the present study we obtained data for 55 DRM lists which in-
cluded 24 lists from Roediger and McDermott (1995) 12 others fromthe Stadler et al (1999) norms and 19 lists from the Gallo andRoediger (2001) norms The study words and the associated criticalitem for each list are presented in Appendix A along with the BASand FAS of each item in relation to the critical item The words withinlists are presented in descending order with respect to their strengthof association to the critical item That is the first list item is typicallythe item most often generated as an associate to the critical item Bothveridical and false recall across our 55 lists are quite stable The split-half correlation of veridical recall for the 55 lists is +86 and the split-half correlation for the lists in eliciting false recall is +90 Thereforewe could potentially explain 81 of the variance in false recall
VariablesAs was described in the introduction the variables we used can
be considered as belonging to two categories properties of the crit-ical item itself and properties of the word lists A complete listingof the values of all seven variables (as well as levels of false recalland false recognition) for all 55 lists appears in Appendix B
Critical item variables With respect to the critical items weused the following three variables word length log frequency andconcreteness We consider each below
1 Word length was defined by the number of letters in each crit-ical item and represents the relative distinctiveness of each wordform (longer words being more distinct) For example butterfly hasnine letters whereas sleep has only five letters We also obtained asubjective measure of orthographic distinctiveness 1 and an objec-tive measure of orthographic neighborhood size2 but because ofthe high degree of multicollinearity among these three variables 3
we included only word length of the critical item in the multiple re-gression analyses
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 389
2 The raw frequency of each critical item was obtained from theKu Iumlcera and Francis (1967) norms and is the number of instancesfound in print per million words For example the raw frequency ofthe critical item sleep is 65 and the raw frequency of the criticalitem butterfly is 2 To correct for skewness in the frequency distri-bution for each critical item raw frequencies were transformedinto log frequencies using the following formula log (5 1 rawfrequency) For example the log frequency of sleep is 182 [ie log(655)] whereas the log frequency of butterfly is 40 [ie log (25)]
3 Concreteness ratings for each critical item were obtained fromthe word association norms of Nelson et al (1999) which were inturn obtained in part from the Paivio Yuille and Madigan (1968)and the Toglia and Battig (1978) norms These ratings are obtainedon a scale of 1ndash7 with 1 being the least concrete and 7 the mostconcrete To carry through with our examples sleep is slightlyabove the midpoint in concreteness (with a rating of 474) and but-terfly is more concrete (with a rating of 591)
List variables With respect to the study lists we used the fol-lowing four variables FAS BAS interitem associative strengthand veridical recall We consider each below
1 Forward associative strength Indices of the strength of asso-ciative connections from the critical item to the study words wereobtained from the word association norms of Nelson et al (1999)Specifically for each critical item we obtained the probability thatit elicited each word in its corresponding study list as an associatea measure Deese (1959b) referred to as the FAS For example as isshown in the third column of Appendix A for the critical item sleepwe identified the strength of its connections to bed rest awaketired and the remaining study words We then averaged these asso-ciations to obtain the mean FAS value for each list In the case ofsleep the mean FAS value for its 15 study words is 047 whereasin the case of butterfly the mean FAS value is 033 Note that all 15list items were given as a response to the critical item in the Russelland Jenkins (1954) norms (from which the lists were generated)However owing to discrepancies between these norms and those ofNelson et al (1999) the number of forward connections for eachlist could have theoretically ranged from 0 to 15
2 Backward associative strength An index of the strength of as-sociative connections from the study words to the critical item wasobtained in large part from the Nelson et al (1999) word associa-tion norms However for some of the 55 DRM lists there were afew list words that were missing from these norms For these wordswe collected our own norms using the procedures outlined by Nel-son and colleagues Briefly we gave groups of over 75 undergrad-uates one of three sheets of 80 to-be-normed words the subjectswere instructed to write next to each word the first word it broughtto mind In those cases in which more than one word from a partic-ular study list needed to be normed these words were placed onseparate norming sheets or were separated by at least 30 items onthe same norming sheet At least 86 observations were obtained foreach normed item and 150 or more observations were obtained foran item in some cases
With these two sets of norms we took each study word and de-termined its connection strength to its corresponding critical itemthe measure Deese (1959b) referred to as the BAS For example asis shown in the second column of Appendix A we obtained for eachstudy word (eg bed rest awake tired etc) the probability thatit elicited its critical item (eg sleep) as an associate Thus thismeasure examines connections in the direction opposite to that usedto obtain the FAS measures We then averaged these associations toobtain the mean BAS value for each list For example the meanBAS value for the sleep list is 431 By comparison the mean BASvalue for the butterfly list is 045
3 Interitem associative strength or connectivity The average in-teritem associative strength (Deese 1959a) or connectivity (Mc-Evoy et al 1999) of the study words was obtained for each listusing two steps First following a procedure outlined by McEvoy
et al we constructed a 15 3 15 matrix of study items for each listIn Appendix C we have illustrated the connectivity matrix for thesleep list Second using the Nelson et al (1999) norms and insome instances our own word association norms we identified thepresence of a connection for all 225 combinations of study wordpairs in each listrsquos matrix excluding the 15 self-connections whichwere not determined (listed as ND) For each word pair a nonzeroconnection strength was scored as a 1 and a zero connectionstrength was scored as a 0 In the case of sleep a total of 27 nonzeroconnections exists among the study words (or an average of 180connections per study word) In the case of butterfly a total of 29nonzero connections exists among the study words (or an averageof 193 connections per study word) High mean connectivityscores indicate a greater number of associative relationships amongthe study words whereas low mean connectivity scores indicate alesser number of associative relationships among the study words
4 Veridical recall of the list One final predictor was the averageprobability of recall of the studied words collapsing across serialpositions 1ndash15 for each list These data were obtained from Stadleret al (1999) and Gallo and Roediger (2001) In the case of sleepstudied items were recalled at a 61 level whereas in the case ofbutterfly studied items were recalled at a 70 level As was notedearlier the split-half correlation of the veridical recall measure wasreasonably high at +86
ProcedureTaking the seven factors that were just described along with the
probability of false recall and the probability of false recognitionfor each of the 55 lists we constructed the data matrix shown inAppendix B for our simultaneous multiple regression analysis4
Each of the 55 lists used in the regression analysis is represented byits respective critical item each of which is listed in alphabeticalorder in the f irst column of the matrix The second and thirdcolumns of the matrix contain the probability of false recall andfalse recognition respectively of each critical item from the Stadleret al (1999) and Gallo and Roediger (2001) norms The next threecolumns contain properties of the critical item itself length log fre-quency and concreteness The next four columns contain propertiesof the list that was constructed for each critical item mean FASmean BAS the mean connectivity of the study words and the av-erage recall of the studied items The last three columns contain thecritical item predictors that were excluded from the multiple re-gression analyses raw frequency orthographic distinctiveness andorthographic neighborhood size (Coltheartrsquos N )
In multiple regression analysis the independent variables can beentered either simultaneously (in a single step) or hierarchically (withsome variables entered into the analysis before others) In the for-mer case all independent variables are allowed to compete forshared variance in the dependent variable (eg false recall) in a sin-gle step so that no single variable is given greater priority than isanother In the latter case those independent variables entered inearlier steps are allowed to capture variance in the absence of com-petition from variables to be entered subsequently In this way in ahierarchical analysis earlier variables are given greater prioritythan later ones usually for theoretical reasons For the purpose ofthe present analysis there exists no definitive theoretical justifica-tion for ordering the entry of critical item variables and list vari-ables so we opted for the simultaneous entry of all variables in asingle step
RESULTS
We consider first the bivariate correlations betweenour seven factors and false recall and then we turn to thesimultaneous multiple regression analysis and its find-ings The entire matrix of correlations (Pearsonrsquos r in all
390 ROEDIGER WATSON MCDERMOTT AND GALLO
cases) among the seven predictors and false recall isshown in Table 1
Bivariate CorrelationsIn describing the results we report significant bivari-
ate correlations between critical item variables and falserecall followed by significant correlations between listvariables and false recall To determine these values eachof the critical item and list variables were entered into aregression equation as the sole predictor of false recallOnly three of our seven variables were significantly cor-related ( p lt 05) with false recall the critical item vari-able of length and the list variables of BAS and recall ofthe studied items (Figures 1A and 1B respectively)
Length The length of the critical item and false recallwere negatively correlated [r 5 37 t(53) 5 287]
Backward associative strength Mean BAS or thestrength of the connections between study items andtheir corresponding critical items was positively corre-lated with false recall r 5 173 t(53) 5 773 As isshown in Figure 1A the greater the strength of associa-tion between the list items and the critical item the morelikely is false recall This outcome replicates Deesersquos(1959b) finding of a 187 correlation between BAS andfalse recall which he obtained with a set of thirty-six 12-item lists Although the 173 correlation between thesetwo variables in our set of f ifty-f ive 15-item lists issomewhat smaller than the 187 correlation obtained byDeese BAS is still the strongest correlate of false recallin our entire data set
Veridical recall Recall of list items and false recallwere negatively correlated [r 5 43 t(53) 5 346]As is shown in Figure 1B increasing levels of accuraterecall are associated with decreasing levels of false re-call This outcome is contrary to some ideas about whyfalse recall occurs as will be brought out in the Discus-sion section
As is shown in Table 1 no other variables were corre-lated with false recall at the bivariate level which is sur-prising in some cases For example in terms of criticalitem variables one might have expected word frequencyor concreteness to be correlated with false recall but nei-ther did (Deese 1959b also failed to find a relation be-tween word frequency and false recall) In terms of listvariables one might have expected connectivity to cor-relate with false recall but it did not The lack of a rela-
tion between connectivity and false recall is noteworthybecause Deese (1959a) and McEvoy et al (1999) re-ported a significant relationship between these variablesso that lists with greater connectivity were less likely toelicit false recall of the critical item In the Discussionsection we consider connectivity or interitem associa-tive strength more fully We turn now to the results of thesimultaneous multiple regression analysis
Simultaneous Multiple Regression AnalysisIn the multiple regression analysis the three variables
based on the characteristics of the critical items them-selves (ie length log frequency and concreteness) andthe four variables based on the characteristics of the listsdesigned to elicit the critical items (ie FAS and BASmeasures interitem associative strength or connectivityand the level of recall of list items) were entered simul-taneously into the regression equation to predict false re-call The results of the analysis are shown in Table 2 in-cluding beta weights (b) semipartial correlations (sr)and significance tests for each predictor For signifi-cance tests a t(47) was used for each predictor and theresultant t statistic is expressed as the absolute valuemdashthat is frac12tfrac12 In addition Table 2 displays the results of thesignificance test for the multiple R2 for the entire set ofseven predictor variables
As is shown in Table 2 the results of the regressionanalysis indicated that two list variables were signifi-cantly related to false recall Specifically BAS was thestrongest predictor of false recall (b 5 170 sr 5160 t 5 717) Therefore consistent with our findingsat the bivariate level the greater the strength of associa-tion between study items and their corresponding criti-
Table 1 Correlation Matrix Among Eight Variables False Recall Three Critical Item Variables and Four List Variables
False Log VeridicalRecall Length Frequency Concreteness FAS BAS Connectivity Recall
NotemdashR2 5 68 F(747) 5 1406 MSe 5 0013 p lt 05 p lt 05
Figure 1 Probability of false recall of the critical item from 55 lists (A) as a functionof associative strength of the list items to the critical item or backward associativestrength and (B) as a function of veridical recall of studied items from the lists
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 391
cal item on a free association test the more probable isfalse recall of the critical item Veridical recall was alsoa strong predictor of false recall (b 5 40 sr 5 34t 5 411) As we found at the bivariate level the morewords recalled from a list the less likely that the criticalitem for that list was falsely recalled None of the re-maining critical item or list variables including length ofthe critical item connectivity and FAS explained aunique portion of the total variance in false recall afterthe influence of other predictors had been residualized inthe context of the regression equation Therefore BASand recall of list items appeared to be the major contrib-utors to the multiple R of 182 and the associated mul-tiple R2 of 68 [F(747) 5 1406 MSe 5 0013] We fur-ther explore the implications of these effects in theDiscussion section
Factors Affecting False RecognitionIn our analyses thus far we have concentrated on
false recall exclusively However the norms for our 55lists (collected by Stadler et al 1999 and Gallo ampRoediger 2001) also include values for false recogni-tion The recognition data in these norms were collectedin a final recognition test after lists had been studied andrecalled This confounding of recognition by prior recallalmost certainly affected both veridical and false recog-nition However in experiments directly examining theeffects of recall on recognition in this paradigm and oth-ers the finding is that recall of a list may increase itsrecognition by a moderate amount (typically in the4ndash10 range eg Roediger amp McDermott 1995Roediger McDermott amp Pisoni 2001) In fact some re-searchers have failed to observe any effect at all of recall
392 ROEDIGER WATSON MCDERMOTT AND GALLO
on recognition (Schacter Verfaellie amp Pradere 1996 seeRoediger et al 1998 for a discussion) We conducted thesame regression analyses with our seven predictor vari-ables on false recognition as we did on false recall Wesummarize the basic results and conclusions as follows
There were three significant bivariate correlationswith false recognition including length (r 5 27)BAS (r 5 143) and veridical recall (r 5 52) Over-all these correlations with false recognition were simi-lar to those obtained with false recall However the pos-itive correlation between BAS and false recognition wassmaller than that in the recall data and the negative cor-relation between veridical recall and false recognitionwas larger than that in the recall data To maintain con-sistency with the previous regression analysis on falserecall we used veridical recall rather than veridicalrecognition in this analysis In addition recognition es-timates were based on only three itemslist and wereprobably affected by prior recall
Turning to the results of the simultaneous multiple re-gression analysis on false recognition BAS and veridi-cal recall of list items appeared to be the major contrib-utors to the multiple R of 169 and the associatedmultiple R2 of 48 [F(747) 5 611 MSe 5 002] Table 3shows the summary of the regression analyses Althoughthe regression analysis accounted for less total variancein false recognition than in false recall (48 as opposedto 68 respectively) the same factors seem to be re-sponsible for predicting false recognition and false re-call Of course because measures of recognition alwaysfollowed recall this similar pattern may reflect a carry-over effect of false recall on false recognition Nonethe-less within the limits of this study the factors responsi-ble for false recognition across lists seem to be the sameas those producing false recall
DISCUSSION
To summarize the main findings our results implicatetwo primary factors (of the seven variables we exam-ined) as accounting for most of the variance in false re-call across lists The split-half correlation of our lists was190 so the amount of potentially explainable variancein our dependent variable of false recall is 81 The fac-
tors in our multiple regression accounted for 68 of thetotal variance in false recall or 84 of the explainablevariance (68 4 81 3 100 5 84) The analysis forfalse recognition revealed the same factors at work al-though they accounted for less overall variance Thestrongest factor was BAS consistent with prior work byDeese (1959b) and McEvoy et al (1999) The other fac-tor was the number of items recalled accurately from thelist which is negatively correlated with false recall
We turn next to the theoretical implications of ourfindings We first consider the implications for the acti-vationmonitoring framework that we have developed toexplain false recall and false recognition in the DRMparadigm We then consider the implications for othertheories and f inally turn to several further issues onwhich our results shed light
ActivationMonitoring TheoryIn the theroretical framework that we (eg McDer-
mott amp Watson 2001 Roediger Balota amp Watson 2001Roediger amp McDermott 1995 2000) have used to explainvariations in the probability of false recall and false recog-nition in the DRM paradigm two sets of processes havebeen proposed Specifically we have proposed that pro-cesses occurring during both encoding and retrieval af-fect the probability of false recall and false recognitionin the DRM paradigm Two critical sets of processes in-volve activation and monitoring Although there is a nat-ural alignment of activation to encoding and monitoringto retrieval we note that both processes can potentiallyoccur both during encoding and during retrieval as willbe discussed below The ideas we employ have many pre-cursors which we acknowledge below although theirparticular arrangement here may be unique
A critical distinction in understanding encoding pro-cesses and their interaction with retrieval processes isbetween item-specific and relational processing (Ander-son 1972 Hunt amp Einstein 1981 Hunt amp McDaniel1993) Briefly the notion is that emphasis can be given tothe encoding of individual elements and their features(what Anderson called node-taggingand Hunt and his col-leagues called item-specific processing) or to the encod-ing of relationships among elements (pathway-taggingor relational processing) Item-specific processing should
Table 3 Summary of Multiple Regression Analysis
on False Recognition With Critical Item Variablesand List Variables Used as Predictors
NotemdashR2 5 48 F(747) 5 611 MSe 5 002 p lt 05 p lt 05
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 393
make individual elements more memorable and lead tomore distinctive memories such processing should helpdistinguish veridical memories from false memories be-cause list items would have features of presentation as-sociated with them whereas the critical item sometimeswould not (However if the critical item is strongly acti-vated during encoding it may take on many of the fea-tures of list items) Relational processing (extractingthemes or schemas or constructing networks of associa-tions) should increase the probability of false recall andfalse recognition by leading to inferences or associationsof events that may not actually have occurred
We assume that in encoding and interpreting experi-ences people go beyond the information given in thestimulus (Bartlett 1932 chap 2 Bruner 1957) and ac-tivate related information through inferential processingEach personrsquos experience is coded in terms of his or herown schemata (or associative networks or knowledgestructures) When schemata are activated a process offilling in of ldquomissingrdquo information may occur wherebyitems strongly implied by the schema (but not actuallypresent in the stimulus) may become activated (Bartlett1932) In encoding lists of words in the DRM paradigmpresentation of the list can create activation that spreadsthroughout the lexicalsemantic system and can createimplicit associative responses (Underwood 1965) Thisactivation may arouse concepts that are not presented but that are associated to concepts that are presented(eg Anderson amp Pirolli 1984 Collins amp Loftus 1975Meyer amp Schvaneveldt 1971) The associative activa-tion in this approach can arise from automatic process-ing or from more controlled elaborative processing(Neely 1977) In addition the associative activation ofconcepts that were not presented may occur consciously(subjects may become aware of the concept of sleep andrehearse it during presentation of the list words bed restawake etc) or the concept may be strongly activatedbut never come to conscious awareness The debate overwhether arousal of the critical item in the DRM para-digm is conscious or unconscious is interesting but dif-ficult to decide empirically Some evidence has impli-cated conscious arousal of the concepts (McDermott1997) whereas other evidence implicates unconsciousactivation (Seamon Luo amp Gallo 1998) The correctanswer to the question of whether activation is consciousor unconscious is probably ldquobothrdquomdashthat is the criticalitem probably sometimes comes consciously to mindand even if it does not relevant information may never-theless be activated unconsciously
The high correlation between BAS and false recall aswell as the strong weighting of BAS in the multiple re-gression analysis strongly implicates activation of criti-cal items as an important factor in explaining false recalland false recognition in the DRM paradigm The morestrongly associated list items are to critical items themore activation would be expected to spread to the crit-ical items More colloquially to the extent that the wordson the list spark the associative connection to the critical
item the more likely the critical item is to be falsely re-called This activation factor would seem to be one mainreason why ldquothings suggested to a person by an experi-ence may be honestly reported by him as part of the ex-periencerdquo (Kirkpatrick 1894 p 608) These associatedlists probably all tend to elicit relational processing butonly those with strong BAS to the critical item lead tofalse recall
The second set of factors proposed in activationmon-itoring theory is concerned with the monitoring of mem-ory accuracy Monitoring of encoding processes canoccur during the study phase of an experiment espe-cially under intentional learning conditions In experi-ments in which subjects are warned about false memoryphenomena and are told to pay close attention to lists andto remember only the presented items and not the criti-cal items monitoring is enhanced (Gallo Roberts amp Sea-mon 1997 Gallo Roediger amp McDermott 2001 Mc-Dermott amp Roediger 1998) Monitoring of memorieshas typically been discussed with respect to retrieval dur-ing which time a primary goal is to distinguish fromamong the considerable information brought to con-sciousness that which refers to perception of past eventsfrom that which does not a process referred to as realitymonitoring (Johnson amp Raye 1981) The experience ofremembering events involves mentally traveling back intime and seeming to reexperience the events (Tulving1985) During retrieval recovered memories that appar-ently carry with them details of having been experiencedare generally attributed to be memories of actual pastevents although sometimes this attribution may be inerror (Jacoby et al 1989)
Johnson et al (1993) have described in detail the typesof information that may be used to distinguish eventsthat actually happened from ones that were only imag-ined thought about or read about For example eventsthat actually occurred have more features from the ex-ternal world associated with them (sights sounds etc)whereas those that were internally generated are lesslikely to have these features but more likely to have as-sociated reflections and cognitive operations In theDRM paradigm the fact that the critical item has beenstrongly activated during encoding means that duringretrieval the subject is likely to retrieve the critical itemas a list member To the extent that the critical items arestrongly activated during encoding they will take onsome features similar to those of the list items this is es-pecially true if the critical items come consciously tomind during encoding and are rehearsed in the samemanner as list items To the extent that the features of thecritical item are similar to those of actual list items theprobability of false recall or false recognition is increasedHowever if the list items systematically differ from thecritical item on some dimensions or features retrievalheuristics may be brought into play that can reduce falserecall and false recognition (Israel amp Schacter 1997Schacter Israel amp Racine 1999) Considerable evidencesupports this general source-reality-monitoring frame-
394 ROEDIGER WATSON MCDERMOTT AND GALLO
work that we advocate here (eg Johnson et al 1993Schacter et al 1999) as well as the attributional ap-proach to remembering (Jacoby et al 1989)
The other significant factor arising from our multipleregression analysis may operate during retrieval by aid-ing the monitoring processes that reduce false recall Thenegative correlation between veridical recall of list itemsand false recall of the critical item may indicate that thebetter encoded list items are the more easily they can bedistinguished from the illusory critical item Subjectsmay use item-specific information to aid recall of listitems and suppress recall of the critical items whichwould not carry as much item-specific information Whenexperimental manipulations increase veridical recall ei-ther through longer study time at relatively slow rates(Gallo amp Roediger 2001 McDermott amp Watson 2001Toglia Neuschatz amp Goodwin 1999) or through re-peated study and test opportunities (McDermott 1996Schacter et al 1999) false recall is often decreasedTherefore the negative relation between veridical recalland false recall can be produced experimentally as wellas appearing in our correlationregression analysis How-ever veridical and false recall can be positively related as afunction of other variables too as was shown by McDer-mott (1996) Toglia et al (1999) and Thapar amp McDermott(2001) In these cases greater activation of the criticalitem from relational processing or from greater seman-tic processing is probably at work
The fact that the length andor distinctiveness of thecritical item are negatively correlated with false recallcan be similarly interpreted as the effect of monitoringduring retrieval because the greater distinctiveness ofthe critical item simplifies the monitoring process Forexample if words in the list are generally shorter andhave more regular orthography than does the criticalitem then during retrieval the critical items that are longand distinctive (eg butterfly cabbage) should be moreeasily discriminated from list items as indeed seems tobe the case in the bivariate correlations Interestingly thelengthdistinctiveness factor played a role in determin-ing false recall even though the lists in our norms werepresented auditorily and the orthographic informationat least should not be particularly accessible In addi-tion presenting the DRM lists visually rather than audi-torily reduces false recall and false recognition at leaston tests in which recognition items are presented visually(as opposed to auditorily) or recall is written rather thanspoken (Gallo McDermott et al 2001 Kellogg 2001Smith amp Hunt 1998) This modality difference has alsobeen interpreted as arising from the retrieval of distinctiveinformation operating during the test (Gallo McDermottet al 2001 Smith amp Hunt 1998)
As was noted above activation processes may occurduring retrieval as well as during encoding The morelist items a subject recalls the more primed the criticalitem might become However the present analyses un-dercut the simplest form of this idea If the recalling oflist items activated or primed false recall the simplest
prediction would be that there should be a positive cor-relation between veridical recall and false recall Our datarevealed the opposite pattern showing a negative corre-lation between veridical and false recall Marsh McDer-mott and Roediger (2001) reported other experimentsthat also call into question the idea that priming duringthe test plays a role in false recall and false recognition
Balota et al (1999) obtained further evidence sup-porting the activationmonitoring interpretation of falsememories in the DRM paradigm They compared per-formance of younger adults healthy older adults andolder adults with Alzheimerrsquos disease The latter twogroups have been shown to have intact activation pro-cesses as measured by semantic priming paradigms(Balota amp Duchek 1988 1991) yet they are deficient insource-monitoring processes which is probably attribut-able to impaired frontal lobe functioning (see BalotaDolan amp Duchek 2000) Therefore one might expectthat whereas veridical recall would be worse in older adultsand Alzheimerrsquos patients (relative to younger adults)false recall and false recognition would occur at the samelevel or perhaps be increased in these groups IndeedBalota et al (1999) reported exactly this pattern A sim-ilar outcome has been obtained in other laboratories(eg Budsen Daffner Desikan amp Schacter 2000 Nor-man amp Schacter 1997 Tun Wingfield Rosen amp Blan-chard 1998) This pattern is just as predicted by the ac-tivation monitoring framework
In short the present results and many others (seeRoediger Balota amp Watson 2001 Roediger et al 1998)fit comfortably within the activationmonitoring frame-work In the remainder of our discussion we will con-sider briefly the implications of our present results forother accounts of the DRM paradigm and for the role ofconnectivity (or interitem associative strength) in pro-ducing veridical and false recall
Fuzzy-Trace TheoryAnother leading theory that has been used to explain
false memory phenomena is fuzzy-trace theory (egReyna amp Brainerd 1995) Payne et al (1996) and Schac-ter et al (1996) among many others have used the basictheory to account for false recall and false recognitiondata within the DRM paradigm and the theory has beenextended in interesting ways to account for new varia-tions on the paradigm (eg Brainerd Wright Reyna ampMojardin 2001) The fundamental assumption of fuzzy-trace theory is that experience leaves two types of mem-ory traces gist traces (which capture the meaning of ex-perience but without specific attributes) and verbatimtraces (which represent specific attributes) False memo-ries are attributed to gist processing When the meaning ofa recognition probe matches the gist of the encoded expe-rience false recognition occurs The theory has been di-rected primarily to false recognition and not false recall
In trying to explain the present results one difficultyis that the operational definition of gist has never beenspecified that is within the DRM paradigm given a list
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 395
of 15 related words how does one specify the gist repre-sentation of the list and quantify how strong it is Onecould use patterns of false recognitionmdashto the extentthat a concept is falsely recognized it was represented aspart of the gist representation This definition is per-fectly suitable for some purposes but clearly explainingfalse recognition is not one of them owing to the prob-lem of circularity To explain why some lists lead to highlevels of false recall and false recognition (whereas otherlists lead to low levels) within fuzzy-trace theory oneneeds an independent specification of gist which to ourknowledge has not been provided in this context Whydoes the sweet list have so much gist that it leads to falserecall whereas the bitter list does not One plausible in-terpretation of gist is BAS The more the list items areassociated to the critical item the more a gist represen-tation of the critical item is created This solution to adefinition of gist works within the DRM paradigm butof course the interpretation of how gist is created be-comes isomorphic to activation processes within the ac-tivationmonitoring framework
Interpretation of the other factor that was significantwithin our multiple regression analysis may be morestraightforward within fuzzy-trace theory The negativecorrelation between veridical recall and false recall (andfalse recognition) may be accounted for by increased re-liance on verbatim traces That is if increased recall of listitems is due to increased verbatim information subjectsmay rely less heavily on gist information and hence showless false recall and false recognition during respondingSimilarly for critical items that are long andor ortho-graphically distinct recall of list items that relies on thesefeatures may lead to suppression of false recall and falserecognition of the critical items because they may differsystematically from other list items These factors pos-tulated within the context of fuzzy-trace theory are quitesimilar to those within the activationmonitoring frame-work outlined above At some level of generality there aretwo sets of factors in each account and both theories canexplain many empirical results
Shifting Criterion TheoryMiller and Wolford (1999) proposed that false recog-
nition in the DRM paradigm could be explained by sub-jectsrsquo shifting their criteria differentially to list items andto critical items during a recognition test (They ex-tended their explanation to false recall but somewhattenuously by proposing a generaterecognize theory)They postulated that subjects encoded the structure ofthe list during its presentation and then during the recog-nition test responded with especially liberal criteria tocritical items but not to list items The criterion shift issupposed to occur to the degree that the test item appearsto be related to the list items To support their argumentthey reported that a measure of bias from signal detec-tion theory was indeed more liberal for critical itemsthan for list items However Wickens and Hirshman(2000) and Wixted and Stretch (2000) pointed out that
within signal detection theory differences in bias mea-sures do not necessarily implicate a criterion shiftRather differences in measured bias could arise fromdifferences among the underlying distributions Wixtedand Stretch argued that this latter interpretation was themore likely and that the criterion shift interpretation wasuntenable (also see Roediger amp McDermott 1999)
The present evidence also weighs heavily against acriterion shift account Miller and Wolford (1999) ap-pealed to subjectsrsquo metaknowledge of the structure of thelist Within the criterion shift account we can assumethat the likelihood that a critical item cues metaknowl-edge of the list depends on its FAS or how well the crit-ical item will cue the list If so FAS (the association ofthe critical item to the list items) should be the primarydeterminant of false recognition and false recall How-ever the correlation between FAS and false recall in ourdata was negligible (08) and the same was true in analy-ses of false recognition BAS was actually the primarypredictor of false recall which implicates activation pro-cesses during encoding as being a primary cause of theeffect There would be no reason to expect BAS to haveany impact according to the idea that a simple criterionshift during the test causes false recall and false recog-nition Of course all our lists were relatively high androughly equivalent in terms of FAS because they werecreated from forward associations to the critical item sorestriction of range may account for the low correlationEven so the same problem exists for the shifting crite-rion theory to explain the results If FAS is supposed todetermine false recall and false recognition by this ac-count and if all our lists were high and roughly equal onthis feature why were so many lists ineffective in elicitingfalse recognition and false recall In short either way theshifting criterion theory cannot account for the results
Of course within the activationmonitoring frame-work the monitoring processes occurring during retrievalcould be conceived as conservative shifts in response cri-teria However in our theory these monitoring processesdetermine the degree to which a given level of activationat retrieval will be judged sufficient to endorse a criticalitem as having been studied thereby producing a falsememory In contrast in Miller and Wolfordrsquos (1999) ac-count there was no activation of the critical item duringencoding and shifting criteria during the test were saidto account entirely for the false recognition phenomenon(see Wixted amp Stretch 2000) In the activationmonitoringaccount the memory illusion provoked by the DRM listsis assumed to be a subjectively powerful phenomenonthat can be ameliorated (but rarely eliminated) when listitems are made more distinctive relative to criticalitems Under conditions in which the subject is warnedat test (but not before the encoding phase) to monitortheir memories for the presence or absence of the criti-cal item and is discouraged from making criterion shiftsthe false memory effect is neither eliminated nor greatlyreduced (Gallo Roediger amp McDermott 2001) Hicksand Marsh (2001) showed that under conditions in which
396 ROEDIGER WATSON MCDERMOTT AND GALLO
subjects are explicitly told to monitor their memories dur-ing the test (and are asked to judge whether each wordhad been externally presented or internally generated)false recognition in the DRM paradigm actually increasesrather than decreases In fact the false recognition effectpersists even when subjects are fully informed about thenature of the phenomenon before the study episode andare therefore encouraged to enhance monitoring processesduring both encoding and retrieval in order to attempt todiminish or eliminate the effect (see Gallo et al1997Gallo Roediger amp McDermott 2001 McDermott ampRoediger 1998) For these and other reasons the notionthat the DRM memory illusion is caused by a general shiftof criteria can be eliminated as a potential explanation
Interitem Associative Strength or ConnectivityOur data pose something of a puzzle about the role of
interitem associative strength (Deese 1959a) or connec-tivity (McEvoy et al 1999) in false recall As was dis-cussed in the introduction Deese (1959a) found a sig-nificant negative correlation between this construct andthe probability of false recall In a later paper Deese(1961) obtained a weaker negative correlation that failedto reach conventional levels of significance so the issueis somewhat in doubt from Deesersquos (1959a 1961) researchMcEvoy et al created lists that varied in connectivity(interitem associative strength) and in BAS and manip-ulated these factors orthogonally They replicated thefinding of Deese (1959a) in that the density of inter-connections of list words was negatively related to falserecall However in our analysis of 55 lists (different fromthose used by McEvoy et al 1999) we found no relationbetween connectivity and false recall (r 5 04) or be-tween connectivity and false recognition (r 5 03)
Further research will be necessary to determine thetrue state of affairs although a case can be made that thelack of correlation we obtained is accurate First we had55 lists that varied widely in their ability to elicit false re-call as compared with 24 lists used in McEvoy et al(1999) Our lists also represented a relatively full rangeof connectivity or interitem associative strength relativeto those of McEvoy et al McEvoy et al used lists thathad a somewhat wider range of connectivity because theyused extreme values whereas our larger set of lists pro-vided a fuller body of the middle ranges of connectivityGiven our results it seems at least possible that interitemassociative strength (connectivity) is not a factor in de-termining whether lists of associative words lead to falserecall although the issue remains in doubt
We did find that interitem associative strength or con-nectivity is positively and significantly correlated withveridical recall (r 5 132) This outcome is consistentwith that of Deese (1959a) and McEvoy et al (1999)The greater the associative bonds among list items thegreater is list recall This outcome is also consistent withvirtually any associative theory of recall (eg Anderson1972 1983 Nelson Schreiber amp McEvoy 1992)
CONCLUSION
The present study has uncovered two primary factorsthat are significantly related to false recall and falserecognition in the DRM paradigm BAS is positivelycorrelated with false recall whereas veridical recall ofthe list items is negatively related to false recall The ac-tivationmonitoring account provides a natural interpre-tation of these factors Any viable theory posited to ex-plain the occurrence of false memories in the DRMparadigm must explain the variability in false recall andfalse recognition across lists and the two primary factorsresponsible for this variability
REFERENCES
Anderson J R (1972) FRAN A simulation model of free recall InG H Bower (Ed) The psychology of learning and motivation (Vol 5pp 315-378) New York Academic Press
Anderson J R (1983) The architecture of cognition CambridgeMA Harvard University Press
Anderson J R amp Pirolli P L (1984) Spread of activation Journalof Experimental Psychology Learning Memory amp Cognition 10791-798
Balota D A Cortese M J Duchek J M Adams D RoedigerH L III McDermott K B amp Yerys B E (1999) Veridical andfalse memories in healthy older adults and in dementia of the Alz-heimers type Cognitive Neuropsychology 16 361-384
Balota D A Dolan P O amp Duchek J M (2000) Memorychanges in healthy older adults In E Tulving amp F I M Craik (Eds)The Oxford handbook of memory (pp 395-409) Oxford Oxford Uni-versity Press
Balota D A amp Duchek J M (1988) Age-related differences inlexical access spreading activation and simple pronunciation Psy-chology and Aging 3 84-93
Balota D A amp Duchek J (1991) Semantic priming effects lexicalrepetition effects and contextual disambiguation effects in healthyaged individuals and individuals with senile dementia of theAlzheimer type Brain amp Language 40 181-201
Bartlett F C (1932) Remembering A study in experimental and so-cial psychology Cambridge Cambridge University Press
Brainerd C J Wright R Reyna V F amp Mojardin A H (2001)Conjoint recognition and phantom recollection Journal of Experi-mental Psychology Learning Memory amp Cognition 27 307-327
Bruce D amp Winograd E (1998) Remembering Deesersquos 1959 arti-cles The Zeitgeist the sociology of science and false memoriesPsychonomic Bulletin amp Review 5 615-624
Bruner J S (1957) On perceptual readiness Psychological Review64 123-152
Budson A E Daffner K R Desikan R amp Schacter D L(2000) When false recognition is unopposed by true recognitionGist-based memory distortion in Alzheimerrsquos disease Neuropsy-chology 14 277-287
Collins A M amp Loftus E F (1975) A spreading-activation theoryof semantic memory Psychological Review 82 407-428
Coltheart M Davelaar E Jonasson J T amp Besner D (1977)Access to the internal lexicon In S Dormic (Ed) Attention and per-formance VI (pp 535-555) Hillsdale NJ Erlbaum
Deese J (1959a) Influence of interitem associative strength upon im-mediate free recall Psychological Reports 5 235-241
Deese J (1959b) On the prediction of occurrence of particular verbalintrusions in immediate recall Journal of Experimental Psychology58 17-22
Deese J (1961) Associative structure and the serial reproduction ex-periment Journal of Abnormal amp Social Psychology 63 95-100
Deese J (1965) The structure of associations in language and thoughtBaltimore Johns Hopkins University Press
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 397
Gallo D A McDermott K B Percer J M amp Roediger H LIII (2001) Modality effects in false recall and false recognitionJournal of Experimental Psychology Learning Memory amp Cogni-tion 27 339-353
Gallo D A Roberts M J amp Seamon J G (1997) Rememberingwords not presented in lists Can we avoid creating false memoriesPsychonomic Bulletin amp Review 4 271-276
Gallo D A amp Roediger H L III (2001) Variability among wordlists in evoking associative memory illusions Manuscript submittedfor publication
Gallo D A Roediger H L III amp McDermott K B (2001) As-sociative false recognition occurs without strategic criterion shiftsPsychonomic Bulletin amp Review 8 579-586
Hicks J L amp Marsh R L (2001) False recognition occurs more fre-quently during source identification than during oldndashnew recogni-tion Journal of Experimental Psychology Learning Memory ampCognition 27 375-383
Hunt R R amp Einstein G O (1981) Relational and item-specific in-formation in memory Journal of Verbal Learning amp Verbal Behav-ior 19 497-514
Hunt R R amp McDaniel M A (1993) The enigma of organizationand distinctiveness Journal of Memory amp Language 32 421-445
Israel L amp Schacter D L (1997) Pictorial encoding reduces falserecognition of semantic associates Psychonomic Bulletin amp Review4 577-581
Jacoby L L Kelley C M amp Dywan J (1989) Memory attribu-tions In H L Roediger III amp F I M Craik (Eds) Varieties of mem-ory and consciousness Essays in honour of Endel Tulving (pp 391-422) Hillsdale NJ Erlbaum
Johnson M K Hashtroudi S amp Lindsay D S (1993) Sourcemonitoring Psychological Bulletin 114 3-28
Johnson M K amp Raye C L (1981) Reality monitoring Psycho-logical Review 88 67-85
Kellogg RT (2001) Presentation modality and mode of recall in ver-bal false memory Journal of Experimental Psychology LearningMemory amp Cognition 27 913-919
Kirkpatrick E A (1894) An experimental study of memory Psy-chological Review 1 602-609
Ku Iumlcera H amp Francis W (1967) Computational analysis of present-day American English Providence RI Brown University Press
Marsh E J McDermott K B amp Roediger H L III (2001) Doretrieval factors play a role in the creation of of false memories Man-uscript submitted for publication
Mather M Henkel L A amp Johnson M J (1997) Evaluatingcharacteristics of false memories Rememberknow judgments andmemory characteristics questionnaire compared Memory amp Cogni-tion 25 826-837
McDermott K B (1996) The persistence of false memories in list re-call Journal of Memory amp Language 35 212-230
McDermott K B (1997) Priming on perceptual implicit memorytests can be achieved through presentation of associates Psycho-nomic Bulletin amp Review 4 582-586
McDermott K B amp Roediger H L III (1998) Attempting to avoidillusory memories Robust false recognition of associates persistsunder conditions of explicit warnings and immediate testing Journalof Memory amp Language 39 508-520
McDermott K B amp Watson J M (2001) The rise and fall of falserecall The impact of presentation duration Journal of Memory ampLanguage 45 160-176
McEvoy C L Nelson D L amp Komatsu T (1999) What is the con-nection between true and false memories The differential roles of in-teritem associations in recall and recognition Journal of Experimen-tal Psychology Learning Memory amp Cognition 25 1177-1194
Meyer D E amp Schvaneveldt R W (1971) Facilitation in recog-nizing pairs of words Evidence of a dependence between retrievaloperations Journal of Experimental Psychology 90 227-234
Miller M B amp Wolford G L (1999) Theoretical commentaryThe role of criterion shift in false memory Psychological Review106 398-405
Neely J H (1977) Semantic priming and retrieval from semanticmemory Roles of inhibitionless spreading activation and limited-
capacity attention Journal of Experimental Psychology General106 226-254
Nelson D L McEvoy C L amp Schreiber T A (1999) The Uni-versity of South Florida word association rhyme and word fragmentnorms Unpublished manuscript University of South Florida Tampa
Nelson D L Schreiber TA amp McEvoy C L (1992) Processing im-plicit and explicit representations Psychological Review 99 322-348
Norman K A amp Schacter D L (1997) False recognition in youngerand older adults Exploring the characteristics of illusory memoriesMemory amp Cognition 25 838-848
Paivio A Yuille J C amp Madigan S A (1968) Concreteness im-agery and meaningfulness values for 925 nouns Journal of Experi-mental Psychology Monographs 76 (1 Pt 2)
Payne D G Elie C J Blackwell J M amp Neuschatz J S (1996)Memory illusions Recalling recognizing and recollecting events thatnever occurred Journal of Memory amp Language 35 261-285
Reyna V F amp Brainerd C J (1995) Fuzzy-trace theory An interimsynthesis Learning amp Individual Differences 7 1-75
Roediger H L III Balota D A amp Watson J M (2001) Spreadingactivation and the arousal of false memories In H L Roediger III J SNairne I Neath amp A M Surprenant (Eds) The nature of remem-bering Essays in honor of Robert G Crowder (pp 95-115) Wash-ington DC American Psychological Association
Roediger H L III amp McDermott K B (1995) Creating false mem-ories Remembering words not presented in lists Journal of Experi-mental Psychology Learning Memory amp Cognition 21 803-814
Roediger H L III amp McDermott K B (1999) False alarms aboutfalse memories Psychological Review 106 406-410
Roediger H L III amp McDermott K B (2000) Tricks of memoryCurrent Directions in Psychological Science 9 123-127
Roediger H L III McDermott K B amp Pisoni D B (2001) Rec-ollection of illusory voices Manuscript submitted for publication
Roediger H L III McDermott K B amp Robinson K J (1998)The role of associative processes in creating false memories In M AConway S E Gathercole amp C Cornoldi (Eds) Theories of memoryII (pp 187-245) Hove UK Psychological Press
Russell W A amp Jenkins J J (1954) The complete Minnesotanorms for responses to 100 words from the KentndashRosanoff Word As-sociation Test (Tech Rep No 11 Contract N8 ONR 66216 Officeof Naval Research) University of Minnesota Minneapolis
Schacter D L Israel L amp Racine C (1999) Suppressing falserecognition in younger and older adults The distinctiveness heuris-tic Journal of Memory amp Language 40 1-24
Schacter D L Verfaellie M amp Pradere D (1996) The neuro-psychology of memory illusions False recall and recognition in am-nesic patients Journal of Memory amp Language 35 319-334
Seamon J G Luo C R amp Gallo D A (1998) Creating false mem-ories of words with or without recognition of list items Evidence fornonconscious processes Psychological Science 9 20-26
Smith R E amp Hunt R R (1998) Presentation modality affects falsememory Psychonomic Bulletin amp Review 5 710-715
Stadler M A Roediger H L III amp McDermott K B (1999)Norms for word lists that create false memories Memory amp Cogni-tion 27 494-500
Thapar A amp McDermott K B (2001) False recall and false recog-nition induced by presentation of associated words Effects of retentioninterval and level of processing Memory amp Cognition 29 424-432
Toglia M P amp Battig W F (1978) Handbook of semantic wordnorms Hillsdale NJ Erlbaum
Toglia M P Neuschatz J S amp Goodwin K A (1999) Recall ac-curacy and illusory memories When more is less Memory 7 233-256
Tulving E (1974) Cue-dependent forgetting American Scientist 6274-82
Tulving E (1979) Relation between encoding specificity and levelsof processing In L S Cermak amp F I M Craik (Eds) Levels of pro-cessing in human memory (pp 405-428) Hillsdale NJ Erlbaum
Tulving E (1985) Memory and consciousness Canadian Psycholo-gist 26 1-12
Tun P A Wingfield A Rosen M J amp Blanchard L (1998)Response latencies for false memories Gist-based processes in nor-mal aging Psychology amp Aging 13 230-241
398 ROEDIGER WATSON MCDERMOTT AND GALLO
Underwood B J (1965) False recognition produced by implicit ver-bal responses Journal of Experimental Psychology 70 122-129
Weldon M S (1991) Mechanisms underlying priming on perceptualtests Journal of Experimental Psychology Learning Memory ampCognition 17 526-541
Wickens T D amp Hirshman E (2000) False memories and statisticaldecision theory Comment on Miller and Wolford (1999) and Roedi-ger and McDermott (1999) Psychological Review 107 377-383
Wixted J T amp Stretch V (2000) The case against a criterion-shiftaccount of false memory Psychological Review 107 368-376
Zechmeister E B (1969) Orthographic distinctiveness Journal ofVerbal Learning amp Verbal Behavior 8 754-761
NOTES
1 The orthographic distinctiveness of a critical item was obtained bypresenting 72 subjects with all 55 critical items visually and havingthem rate the words in accordance with procedures developed by Zech-meister (1969) Briefly the subjects were instructed to rate each wordfor its physical appearance in terms of distinctiveness in the languageon a 9-point scale with 1 being not very distinctive 5 corresponding toaverage distinctiveness and 9 being most distinctive To avoid orderingeffects three different randomized orderings of words were used acrossroughly equal groups of subjects In the case of the critical item sleep
for example the distinctiveness is 515 Corresponding values for but-terfly and for man are 744 and 213 respectively Although ortho-graphic distinctiveness was not included in our multiple regressionanalyses this predictor is included in Appendix C
2 The orthographic neighborhood size or Coltheartrsquos N of each crit-ical item was defined as the number of words that can be formed bychanging a single letter in the word (Coltheart Davelaar Jonasson ampBesner 1977) For example because it has no orthographic neighborsColtheartrsquos N for the critical item butterfly is zero In contrast sleep hasfive orthographic neighbors (ie sheep sleek sleet steep and sweep)The reference set of words used were those in Ku Iumlcera and Francisrsquos(1967) norms Although Coltheartrsquos N was not included in our multipleregression analyses this predictor is included in Appendix C
3 Word length orthographic distinctiveness and orthographic neigh-borhood size of the critical item were all highly correlated Specificallythe bivariate correlations among these three predictors were as followslength and orthographic distinctiveness r 5 184 t(53) 5 1117 p lt05 orthographic distinctiveness and orthographic neighborhood sizer 5 66 t(53) 5 640 p lt 05 length and orthographic neighbor-hood size r 5 63 t(53) 5 582 p lt 05 Of these three variablesword length had the largest bivariate correlation with false recall (r 5
37) so it was the predictor used in the multiple regression analyses4 The multiple regression data matrix is available at wwwiacwustl
edunmclWeb
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 399
APPENDIX A The Fifty-Five 15-Word Lists and Their Associated Critical Items
With Backward Associative Strength (BAS) and Forward Associative Strength (FAS)Values Used in the Multiple Regression Analysis of False Recall
Item Word BAS FAS Item Word BAS FAS
Anger mad 393 412 Black white 655 557fear 020 059 dark 111 100hate 028 109 cat 000 043rage 541 042 charred 023 000temper 182 000 night 000 021fury 306 000 funeral 034 000ire 179 000 color 074 050wrath 128 000 grief 000 000happy 000 042 blue 028 000fight 034 000 death 016 014hatred 070 000 ink 020 000mean 090 000 bottom 000 000calm 000 000 coal 288 000emotion 000 000 brown 338 000enrage 378 000 gray 365 000
Mean 157 044 Mean 130 052Army Navy 543 500 Bread butter 364 487
NotemdashUnless otherwise indicated with an asterisk all BAS and FAS values are drawn from the associative norms of Nel-son McEvoy and Schreiber (1999) Those values shown with an asterisk were obtained by the authors by using normingprocedures similar to those of Nelson et al (1999) The corresponding mean for these BAS and FAS values for each list arealso represented
406 ROEDIGER WATSON MCDERMOTT AND GALLO
APPENDIX B Data Matrix of the Fifty-Five 15-Word Lists and Associated Critical Items
Used in the Multiple Regression Analysis of False Recall
Critical False False Log Vrd Raw Ortho Item Recall Recgn Length Freq Concrete FAS BAS Connect Recall Freq Dist Colt N
(Manuscript received July 10 2000revision accepted for publication February 23 2001)
388 ROEDIGER WATSON MCDERMOTT AND GALLO
nectivity and mean BAS had opposite influences on the re-call of critical items both variables played an importantrole in creating false memories in the DRM paradigm
In sum both Deese (1959a 1959b) and McEvoy et al(1999) argued that at least two factors are important inpredicting veridical and false recall Associative con-nections from study words to the critical item increasethe probability of false memories whereas associativeconnections among study words decrease the probabilityof false memories However other factors besides meanBAS and connectivity may underlie the recall of criticalitems in the DRM paradigm For example consider theperformance on the king list (ie queen England crownprince George dictator palace throne chess rule sub-jects monarch royal leader reign) Because it has a rel-atively high mean BAS it would be predicted to have ahigh level of false recall However this list elicited thelowest level of recall in the entire set of 36 DRM listsnormed by Stadler Roediger and McDermott (1999)
Gallo and Roediger (2001 Experiment 1) attemptedto produce lists that were constructed in the same generalway as those of Roediger and McDermott (1995) but thatwould like the king list produce low levels of false re-call and false recognition However unlike the king listthe lists created by Gallo and Roediger all had low BASOf the 19 new lists they constructed many had very lowlevels of false recall and false recognition However sev-eral of the lists (eg wish justice) produced relativelyhigh levels of false recall and false recognition muchlike the lists developed by Roediger and McDermott(1995) Clearly BAS is not the entire reason for the widedifferences in the effectiveness of the lists in creating falsememories The aim of the present study is to gain furtherinformation on the characteristics of lists and criticalitems that affect false recall and false recognition
OVERVIEW OF THE PRESENT STUDY
In the present study we used a simultaneous multipleregression analysis to identify the factors predicting re-call of critical items in the DRM paradigm We consid-ered factors that pertain to the critical item itself as wellas factors characterizing the study lists Specificallywith respect to the critical items we used the followingthree variables word length word frequency and con-creteness On the basis of an examination of the Stadleret al (1999) norms and Deesersquos (1959b) original find-ings we hypothesized that factors making a critical itemdistinct might produce low levels of false recall For ex-ample butterfly is a long low-frequency critical itemInterestingly it was never recalled in Deesersquos (1959b) ex-periment and it was recalled by only 1 of Gallo andRoedigerrsquos (2001) subjects
We also assessed the ability of several variables re-lated to the study words to predict the recall of criticalitems in the DRM paradigm In particular we used thefollowing four variables an index of the strength of theassociative connections from the critical item to the
study words or forward associative strength (FAS ieassociations from sleep to bed to rest to awake etc) anindex of the strength of the associative connections fromthe study words to the critical item or BAS (associationsfrom bed rest to sleep) the average number of as-sociative connections among the study words (interitemassociative strength or connectivity) and the averageprobability of recall of the studied words within each ofthe false memory lists (veridical recall) On the basis ofprior research by Deese (1959a 1959b) and McEvoy et al(1999) we hypothesized that BAS and connectivitywould be the best predictors of the differences observedin false recall among critical items More specificallyrecall of critical items should increase as the strength ofassociative connections from the study words to the crit-ical item increases however recall of critical items shoulddecrease as the average number of associative connec-tions among study words increases (As will be shownin our data set only the first of these two predictions wasfulfilled) In sum the goal of the present research was toidentify the factors that underlie the creation of falsememories in the DRM paradigm by explaining the widevariability in the effectiveness of our 55 lists to elicitfalse recall The multiple regression analysis that led toour answers is explained more fully in the Method andResults sections
METHOD
MaterialsIn the present study we obtained data for 55 DRM lists which in-
cluded 24 lists from Roediger and McDermott (1995) 12 others fromthe Stadler et al (1999) norms and 19 lists from the Gallo andRoediger (2001) norms The study words and the associated criticalitem for each list are presented in Appendix A along with the BASand FAS of each item in relation to the critical item The words withinlists are presented in descending order with respect to their strengthof association to the critical item That is the first list item is typicallythe item most often generated as an associate to the critical item Bothveridical and false recall across our 55 lists are quite stable The split-half correlation of veridical recall for the 55 lists is +86 and the split-half correlation for the lists in eliciting false recall is +90 Thereforewe could potentially explain 81 of the variance in false recall
VariablesAs was described in the introduction the variables we used can
be considered as belonging to two categories properties of the crit-ical item itself and properties of the word lists A complete listingof the values of all seven variables (as well as levels of false recalland false recognition) for all 55 lists appears in Appendix B
Critical item variables With respect to the critical items weused the following three variables word length log frequency andconcreteness We consider each below
1 Word length was defined by the number of letters in each crit-ical item and represents the relative distinctiveness of each wordform (longer words being more distinct) For example butterfly hasnine letters whereas sleep has only five letters We also obtained asubjective measure of orthographic distinctiveness 1 and an objec-tive measure of orthographic neighborhood size2 but because ofthe high degree of multicollinearity among these three variables 3
we included only word length of the critical item in the multiple re-gression analyses
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 389
2 The raw frequency of each critical item was obtained from theKu Iumlcera and Francis (1967) norms and is the number of instancesfound in print per million words For example the raw frequency ofthe critical item sleep is 65 and the raw frequency of the criticalitem butterfly is 2 To correct for skewness in the frequency distri-bution for each critical item raw frequencies were transformedinto log frequencies using the following formula log (5 1 rawfrequency) For example the log frequency of sleep is 182 [ie log(655)] whereas the log frequency of butterfly is 40 [ie log (25)]
3 Concreteness ratings for each critical item were obtained fromthe word association norms of Nelson et al (1999) which were inturn obtained in part from the Paivio Yuille and Madigan (1968)and the Toglia and Battig (1978) norms These ratings are obtainedon a scale of 1ndash7 with 1 being the least concrete and 7 the mostconcrete To carry through with our examples sleep is slightlyabove the midpoint in concreteness (with a rating of 474) and but-terfly is more concrete (with a rating of 591)
List variables With respect to the study lists we used the fol-lowing four variables FAS BAS interitem associative strengthand veridical recall We consider each below
1 Forward associative strength Indices of the strength of asso-ciative connections from the critical item to the study words wereobtained from the word association norms of Nelson et al (1999)Specifically for each critical item we obtained the probability thatit elicited each word in its corresponding study list as an associatea measure Deese (1959b) referred to as the FAS For example as isshown in the third column of Appendix A for the critical item sleepwe identified the strength of its connections to bed rest awaketired and the remaining study words We then averaged these asso-ciations to obtain the mean FAS value for each list In the case ofsleep the mean FAS value for its 15 study words is 047 whereasin the case of butterfly the mean FAS value is 033 Note that all 15list items were given as a response to the critical item in the Russelland Jenkins (1954) norms (from which the lists were generated)However owing to discrepancies between these norms and those ofNelson et al (1999) the number of forward connections for eachlist could have theoretically ranged from 0 to 15
2 Backward associative strength An index of the strength of as-sociative connections from the study words to the critical item wasobtained in large part from the Nelson et al (1999) word associa-tion norms However for some of the 55 DRM lists there were afew list words that were missing from these norms For these wordswe collected our own norms using the procedures outlined by Nel-son and colleagues Briefly we gave groups of over 75 undergrad-uates one of three sheets of 80 to-be-normed words the subjectswere instructed to write next to each word the first word it broughtto mind In those cases in which more than one word from a partic-ular study list needed to be normed these words were placed onseparate norming sheets or were separated by at least 30 items onthe same norming sheet At least 86 observations were obtained foreach normed item and 150 or more observations were obtained foran item in some cases
With these two sets of norms we took each study word and de-termined its connection strength to its corresponding critical itemthe measure Deese (1959b) referred to as the BAS For example asis shown in the second column of Appendix A we obtained for eachstudy word (eg bed rest awake tired etc) the probability thatit elicited its critical item (eg sleep) as an associate Thus thismeasure examines connections in the direction opposite to that usedto obtain the FAS measures We then averaged these associations toobtain the mean BAS value for each list For example the meanBAS value for the sleep list is 431 By comparison the mean BASvalue for the butterfly list is 045
3 Interitem associative strength or connectivity The average in-teritem associative strength (Deese 1959a) or connectivity (Mc-Evoy et al 1999) of the study words was obtained for each listusing two steps First following a procedure outlined by McEvoy
et al we constructed a 15 3 15 matrix of study items for each listIn Appendix C we have illustrated the connectivity matrix for thesleep list Second using the Nelson et al (1999) norms and insome instances our own word association norms we identified thepresence of a connection for all 225 combinations of study wordpairs in each listrsquos matrix excluding the 15 self-connections whichwere not determined (listed as ND) For each word pair a nonzeroconnection strength was scored as a 1 and a zero connectionstrength was scored as a 0 In the case of sleep a total of 27 nonzeroconnections exists among the study words (or an average of 180connections per study word) In the case of butterfly a total of 29nonzero connections exists among the study words (or an averageof 193 connections per study word) High mean connectivityscores indicate a greater number of associative relationships amongthe study words whereas low mean connectivity scores indicate alesser number of associative relationships among the study words
4 Veridical recall of the list One final predictor was the averageprobability of recall of the studied words collapsing across serialpositions 1ndash15 for each list These data were obtained from Stadleret al (1999) and Gallo and Roediger (2001) In the case of sleepstudied items were recalled at a 61 level whereas in the case ofbutterfly studied items were recalled at a 70 level As was notedearlier the split-half correlation of the veridical recall measure wasreasonably high at +86
ProcedureTaking the seven factors that were just described along with the
probability of false recall and the probability of false recognitionfor each of the 55 lists we constructed the data matrix shown inAppendix B for our simultaneous multiple regression analysis4
Each of the 55 lists used in the regression analysis is represented byits respective critical item each of which is listed in alphabeticalorder in the f irst column of the matrix The second and thirdcolumns of the matrix contain the probability of false recall andfalse recognition respectively of each critical item from the Stadleret al (1999) and Gallo and Roediger (2001) norms The next threecolumns contain properties of the critical item itself length log fre-quency and concreteness The next four columns contain propertiesof the list that was constructed for each critical item mean FASmean BAS the mean connectivity of the study words and the av-erage recall of the studied items The last three columns contain thecritical item predictors that were excluded from the multiple re-gression analyses raw frequency orthographic distinctiveness andorthographic neighborhood size (Coltheartrsquos N )
In multiple regression analysis the independent variables can beentered either simultaneously (in a single step) or hierarchically (withsome variables entered into the analysis before others) In the for-mer case all independent variables are allowed to compete forshared variance in the dependent variable (eg false recall) in a sin-gle step so that no single variable is given greater priority than isanother In the latter case those independent variables entered inearlier steps are allowed to capture variance in the absence of com-petition from variables to be entered subsequently In this way in ahierarchical analysis earlier variables are given greater prioritythan later ones usually for theoretical reasons For the purpose ofthe present analysis there exists no definitive theoretical justifica-tion for ordering the entry of critical item variables and list vari-ables so we opted for the simultaneous entry of all variables in asingle step
RESULTS
We consider first the bivariate correlations betweenour seven factors and false recall and then we turn to thesimultaneous multiple regression analysis and its find-ings The entire matrix of correlations (Pearsonrsquos r in all
390 ROEDIGER WATSON MCDERMOTT AND GALLO
cases) among the seven predictors and false recall isshown in Table 1
Bivariate CorrelationsIn describing the results we report significant bivari-
ate correlations between critical item variables and falserecall followed by significant correlations between listvariables and false recall To determine these values eachof the critical item and list variables were entered into aregression equation as the sole predictor of false recallOnly three of our seven variables were significantly cor-related ( p lt 05) with false recall the critical item vari-able of length and the list variables of BAS and recall ofthe studied items (Figures 1A and 1B respectively)
Length The length of the critical item and false recallwere negatively correlated [r 5 37 t(53) 5 287]
Backward associative strength Mean BAS or thestrength of the connections between study items andtheir corresponding critical items was positively corre-lated with false recall r 5 173 t(53) 5 773 As isshown in Figure 1A the greater the strength of associa-tion between the list items and the critical item the morelikely is false recall This outcome replicates Deesersquos(1959b) finding of a 187 correlation between BAS andfalse recall which he obtained with a set of thirty-six 12-item lists Although the 173 correlation between thesetwo variables in our set of f ifty-f ive 15-item lists issomewhat smaller than the 187 correlation obtained byDeese BAS is still the strongest correlate of false recallin our entire data set
Veridical recall Recall of list items and false recallwere negatively correlated [r 5 43 t(53) 5 346]As is shown in Figure 1B increasing levels of accuraterecall are associated with decreasing levels of false re-call This outcome is contrary to some ideas about whyfalse recall occurs as will be brought out in the Discus-sion section
As is shown in Table 1 no other variables were corre-lated with false recall at the bivariate level which is sur-prising in some cases For example in terms of criticalitem variables one might have expected word frequencyor concreteness to be correlated with false recall but nei-ther did (Deese 1959b also failed to find a relation be-tween word frequency and false recall) In terms of listvariables one might have expected connectivity to cor-relate with false recall but it did not The lack of a rela-
tion between connectivity and false recall is noteworthybecause Deese (1959a) and McEvoy et al (1999) re-ported a significant relationship between these variablesso that lists with greater connectivity were less likely toelicit false recall of the critical item In the Discussionsection we consider connectivity or interitem associa-tive strength more fully We turn now to the results of thesimultaneous multiple regression analysis
Simultaneous Multiple Regression AnalysisIn the multiple regression analysis the three variables
based on the characteristics of the critical items them-selves (ie length log frequency and concreteness) andthe four variables based on the characteristics of the listsdesigned to elicit the critical items (ie FAS and BASmeasures interitem associative strength or connectivityand the level of recall of list items) were entered simul-taneously into the regression equation to predict false re-call The results of the analysis are shown in Table 2 in-cluding beta weights (b) semipartial correlations (sr)and significance tests for each predictor For signifi-cance tests a t(47) was used for each predictor and theresultant t statistic is expressed as the absolute valuemdashthat is frac12tfrac12 In addition Table 2 displays the results of thesignificance test for the multiple R2 for the entire set ofseven predictor variables
As is shown in Table 2 the results of the regressionanalysis indicated that two list variables were signifi-cantly related to false recall Specifically BAS was thestrongest predictor of false recall (b 5 170 sr 5160 t 5 717) Therefore consistent with our findingsat the bivariate level the greater the strength of associa-tion between study items and their corresponding criti-
Table 1 Correlation Matrix Among Eight Variables False Recall Three Critical Item Variables and Four List Variables
False Log VeridicalRecall Length Frequency Concreteness FAS BAS Connectivity Recall
NotemdashR2 5 68 F(747) 5 1406 MSe 5 0013 p lt 05 p lt 05
Figure 1 Probability of false recall of the critical item from 55 lists (A) as a functionof associative strength of the list items to the critical item or backward associativestrength and (B) as a function of veridical recall of studied items from the lists
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 391
cal item on a free association test the more probable isfalse recall of the critical item Veridical recall was alsoa strong predictor of false recall (b 5 40 sr 5 34t 5 411) As we found at the bivariate level the morewords recalled from a list the less likely that the criticalitem for that list was falsely recalled None of the re-maining critical item or list variables including length ofthe critical item connectivity and FAS explained aunique portion of the total variance in false recall afterthe influence of other predictors had been residualized inthe context of the regression equation Therefore BASand recall of list items appeared to be the major contrib-utors to the multiple R of 182 and the associated mul-tiple R2 of 68 [F(747) 5 1406 MSe 5 0013] We fur-ther explore the implications of these effects in theDiscussion section
Factors Affecting False RecognitionIn our analyses thus far we have concentrated on
false recall exclusively However the norms for our 55lists (collected by Stadler et al 1999 and Gallo ampRoediger 2001) also include values for false recogni-tion The recognition data in these norms were collectedin a final recognition test after lists had been studied andrecalled This confounding of recognition by prior recallalmost certainly affected both veridical and false recog-nition However in experiments directly examining theeffects of recall on recognition in this paradigm and oth-ers the finding is that recall of a list may increase itsrecognition by a moderate amount (typically in the4ndash10 range eg Roediger amp McDermott 1995Roediger McDermott amp Pisoni 2001) In fact some re-searchers have failed to observe any effect at all of recall
392 ROEDIGER WATSON MCDERMOTT AND GALLO
on recognition (Schacter Verfaellie amp Pradere 1996 seeRoediger et al 1998 for a discussion) We conducted thesame regression analyses with our seven predictor vari-ables on false recognition as we did on false recall Wesummarize the basic results and conclusions as follows
There were three significant bivariate correlationswith false recognition including length (r 5 27)BAS (r 5 143) and veridical recall (r 5 52) Over-all these correlations with false recognition were simi-lar to those obtained with false recall However the pos-itive correlation between BAS and false recognition wassmaller than that in the recall data and the negative cor-relation between veridical recall and false recognitionwas larger than that in the recall data To maintain con-sistency with the previous regression analysis on falserecall we used veridical recall rather than veridicalrecognition in this analysis In addition recognition es-timates were based on only three itemslist and wereprobably affected by prior recall
Turning to the results of the simultaneous multiple re-gression analysis on false recognition BAS and veridi-cal recall of list items appeared to be the major contrib-utors to the multiple R of 169 and the associatedmultiple R2 of 48 [F(747) 5 611 MSe 5 002] Table 3shows the summary of the regression analyses Althoughthe regression analysis accounted for less total variancein false recognition than in false recall (48 as opposedto 68 respectively) the same factors seem to be re-sponsible for predicting false recognition and false re-call Of course because measures of recognition alwaysfollowed recall this similar pattern may reflect a carry-over effect of false recall on false recognition Nonethe-less within the limits of this study the factors responsi-ble for false recognition across lists seem to be the sameas those producing false recall
DISCUSSION
To summarize the main findings our results implicatetwo primary factors (of the seven variables we exam-ined) as accounting for most of the variance in false re-call across lists The split-half correlation of our lists was190 so the amount of potentially explainable variancein our dependent variable of false recall is 81 The fac-
tors in our multiple regression accounted for 68 of thetotal variance in false recall or 84 of the explainablevariance (68 4 81 3 100 5 84) The analysis forfalse recognition revealed the same factors at work al-though they accounted for less overall variance Thestrongest factor was BAS consistent with prior work byDeese (1959b) and McEvoy et al (1999) The other fac-tor was the number of items recalled accurately from thelist which is negatively correlated with false recall
We turn next to the theoretical implications of ourfindings We first consider the implications for the acti-vationmonitoring framework that we have developed toexplain false recall and false recognition in the DRMparadigm We then consider the implications for othertheories and f inally turn to several further issues onwhich our results shed light
ActivationMonitoring TheoryIn the theroretical framework that we (eg McDer-
mott amp Watson 2001 Roediger Balota amp Watson 2001Roediger amp McDermott 1995 2000) have used to explainvariations in the probability of false recall and false recog-nition in the DRM paradigm two sets of processes havebeen proposed Specifically we have proposed that pro-cesses occurring during both encoding and retrieval af-fect the probability of false recall and false recognitionin the DRM paradigm Two critical sets of processes in-volve activation and monitoring Although there is a nat-ural alignment of activation to encoding and monitoringto retrieval we note that both processes can potentiallyoccur both during encoding and during retrieval as willbe discussed below The ideas we employ have many pre-cursors which we acknowledge below although theirparticular arrangement here may be unique
A critical distinction in understanding encoding pro-cesses and their interaction with retrieval processes isbetween item-specific and relational processing (Ander-son 1972 Hunt amp Einstein 1981 Hunt amp McDaniel1993) Briefly the notion is that emphasis can be given tothe encoding of individual elements and their features(what Anderson called node-taggingand Hunt and his col-leagues called item-specific processing) or to the encod-ing of relationships among elements (pathway-taggingor relational processing) Item-specific processing should
Table 3 Summary of Multiple Regression Analysis
on False Recognition With Critical Item Variablesand List Variables Used as Predictors
NotemdashR2 5 48 F(747) 5 611 MSe 5 002 p lt 05 p lt 05
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 393
make individual elements more memorable and lead tomore distinctive memories such processing should helpdistinguish veridical memories from false memories be-cause list items would have features of presentation as-sociated with them whereas the critical item sometimeswould not (However if the critical item is strongly acti-vated during encoding it may take on many of the fea-tures of list items) Relational processing (extractingthemes or schemas or constructing networks of associa-tions) should increase the probability of false recall andfalse recognition by leading to inferences or associationsof events that may not actually have occurred
We assume that in encoding and interpreting experi-ences people go beyond the information given in thestimulus (Bartlett 1932 chap 2 Bruner 1957) and ac-tivate related information through inferential processingEach personrsquos experience is coded in terms of his or herown schemata (or associative networks or knowledgestructures) When schemata are activated a process offilling in of ldquomissingrdquo information may occur wherebyitems strongly implied by the schema (but not actuallypresent in the stimulus) may become activated (Bartlett1932) In encoding lists of words in the DRM paradigmpresentation of the list can create activation that spreadsthroughout the lexicalsemantic system and can createimplicit associative responses (Underwood 1965) Thisactivation may arouse concepts that are not presented but that are associated to concepts that are presented(eg Anderson amp Pirolli 1984 Collins amp Loftus 1975Meyer amp Schvaneveldt 1971) The associative activa-tion in this approach can arise from automatic process-ing or from more controlled elaborative processing(Neely 1977) In addition the associative activation ofconcepts that were not presented may occur consciously(subjects may become aware of the concept of sleep andrehearse it during presentation of the list words bed restawake etc) or the concept may be strongly activatedbut never come to conscious awareness The debate overwhether arousal of the critical item in the DRM para-digm is conscious or unconscious is interesting but dif-ficult to decide empirically Some evidence has impli-cated conscious arousal of the concepts (McDermott1997) whereas other evidence implicates unconsciousactivation (Seamon Luo amp Gallo 1998) The correctanswer to the question of whether activation is consciousor unconscious is probably ldquobothrdquomdashthat is the criticalitem probably sometimes comes consciously to mindand even if it does not relevant information may never-theless be activated unconsciously
The high correlation between BAS and false recall aswell as the strong weighting of BAS in the multiple re-gression analysis strongly implicates activation of criti-cal items as an important factor in explaining false recalland false recognition in the DRM paradigm The morestrongly associated list items are to critical items themore activation would be expected to spread to the crit-ical items More colloquially to the extent that the wordson the list spark the associative connection to the critical
item the more likely the critical item is to be falsely re-called This activation factor would seem to be one mainreason why ldquothings suggested to a person by an experi-ence may be honestly reported by him as part of the ex-periencerdquo (Kirkpatrick 1894 p 608) These associatedlists probably all tend to elicit relational processing butonly those with strong BAS to the critical item lead tofalse recall
The second set of factors proposed in activationmon-itoring theory is concerned with the monitoring of mem-ory accuracy Monitoring of encoding processes canoccur during the study phase of an experiment espe-cially under intentional learning conditions In experi-ments in which subjects are warned about false memoryphenomena and are told to pay close attention to lists andto remember only the presented items and not the criti-cal items monitoring is enhanced (Gallo Roberts amp Sea-mon 1997 Gallo Roediger amp McDermott 2001 Mc-Dermott amp Roediger 1998) Monitoring of memorieshas typically been discussed with respect to retrieval dur-ing which time a primary goal is to distinguish fromamong the considerable information brought to con-sciousness that which refers to perception of past eventsfrom that which does not a process referred to as realitymonitoring (Johnson amp Raye 1981) The experience ofremembering events involves mentally traveling back intime and seeming to reexperience the events (Tulving1985) During retrieval recovered memories that appar-ently carry with them details of having been experiencedare generally attributed to be memories of actual pastevents although sometimes this attribution may be inerror (Jacoby et al 1989)
Johnson et al (1993) have described in detail the typesof information that may be used to distinguish eventsthat actually happened from ones that were only imag-ined thought about or read about For example eventsthat actually occurred have more features from the ex-ternal world associated with them (sights sounds etc)whereas those that were internally generated are lesslikely to have these features but more likely to have as-sociated reflections and cognitive operations In theDRM paradigm the fact that the critical item has beenstrongly activated during encoding means that duringretrieval the subject is likely to retrieve the critical itemas a list member To the extent that the critical items arestrongly activated during encoding they will take onsome features similar to those of the list items this is es-pecially true if the critical items come consciously tomind during encoding and are rehearsed in the samemanner as list items To the extent that the features of thecritical item are similar to those of actual list items theprobability of false recall or false recognition is increasedHowever if the list items systematically differ from thecritical item on some dimensions or features retrievalheuristics may be brought into play that can reduce falserecall and false recognition (Israel amp Schacter 1997Schacter Israel amp Racine 1999) Considerable evidencesupports this general source-reality-monitoring frame-
394 ROEDIGER WATSON MCDERMOTT AND GALLO
work that we advocate here (eg Johnson et al 1993Schacter et al 1999) as well as the attributional ap-proach to remembering (Jacoby et al 1989)
The other significant factor arising from our multipleregression analysis may operate during retrieval by aid-ing the monitoring processes that reduce false recall Thenegative correlation between veridical recall of list itemsand false recall of the critical item may indicate that thebetter encoded list items are the more easily they can bedistinguished from the illusory critical item Subjectsmay use item-specific information to aid recall of listitems and suppress recall of the critical items whichwould not carry as much item-specific information Whenexperimental manipulations increase veridical recall ei-ther through longer study time at relatively slow rates(Gallo amp Roediger 2001 McDermott amp Watson 2001Toglia Neuschatz amp Goodwin 1999) or through re-peated study and test opportunities (McDermott 1996Schacter et al 1999) false recall is often decreasedTherefore the negative relation between veridical recalland false recall can be produced experimentally as wellas appearing in our correlationregression analysis How-ever veridical and false recall can be positively related as afunction of other variables too as was shown by McDer-mott (1996) Toglia et al (1999) and Thapar amp McDermott(2001) In these cases greater activation of the criticalitem from relational processing or from greater seman-tic processing is probably at work
The fact that the length andor distinctiveness of thecritical item are negatively correlated with false recallcan be similarly interpreted as the effect of monitoringduring retrieval because the greater distinctiveness ofthe critical item simplifies the monitoring process Forexample if words in the list are generally shorter andhave more regular orthography than does the criticalitem then during retrieval the critical items that are longand distinctive (eg butterfly cabbage) should be moreeasily discriminated from list items as indeed seems tobe the case in the bivariate correlations Interestingly thelengthdistinctiveness factor played a role in determin-ing false recall even though the lists in our norms werepresented auditorily and the orthographic informationat least should not be particularly accessible In addi-tion presenting the DRM lists visually rather than audi-torily reduces false recall and false recognition at leaston tests in which recognition items are presented visually(as opposed to auditorily) or recall is written rather thanspoken (Gallo McDermott et al 2001 Kellogg 2001Smith amp Hunt 1998) This modality difference has alsobeen interpreted as arising from the retrieval of distinctiveinformation operating during the test (Gallo McDermottet al 2001 Smith amp Hunt 1998)
As was noted above activation processes may occurduring retrieval as well as during encoding The morelist items a subject recalls the more primed the criticalitem might become However the present analyses un-dercut the simplest form of this idea If the recalling oflist items activated or primed false recall the simplest
prediction would be that there should be a positive cor-relation between veridical recall and false recall Our datarevealed the opposite pattern showing a negative corre-lation between veridical and false recall Marsh McDer-mott and Roediger (2001) reported other experimentsthat also call into question the idea that priming duringthe test plays a role in false recall and false recognition
Balota et al (1999) obtained further evidence sup-porting the activationmonitoring interpretation of falsememories in the DRM paradigm They compared per-formance of younger adults healthy older adults andolder adults with Alzheimerrsquos disease The latter twogroups have been shown to have intact activation pro-cesses as measured by semantic priming paradigms(Balota amp Duchek 1988 1991) yet they are deficient insource-monitoring processes which is probably attribut-able to impaired frontal lobe functioning (see BalotaDolan amp Duchek 2000) Therefore one might expectthat whereas veridical recall would be worse in older adultsand Alzheimerrsquos patients (relative to younger adults)false recall and false recognition would occur at the samelevel or perhaps be increased in these groups IndeedBalota et al (1999) reported exactly this pattern A sim-ilar outcome has been obtained in other laboratories(eg Budsen Daffner Desikan amp Schacter 2000 Nor-man amp Schacter 1997 Tun Wingfield Rosen amp Blan-chard 1998) This pattern is just as predicted by the ac-tivation monitoring framework
In short the present results and many others (seeRoediger Balota amp Watson 2001 Roediger et al 1998)fit comfortably within the activationmonitoring frame-work In the remainder of our discussion we will con-sider briefly the implications of our present results forother accounts of the DRM paradigm and for the role ofconnectivity (or interitem associative strength) in pro-ducing veridical and false recall
Fuzzy-Trace TheoryAnother leading theory that has been used to explain
false memory phenomena is fuzzy-trace theory (egReyna amp Brainerd 1995) Payne et al (1996) and Schac-ter et al (1996) among many others have used the basictheory to account for false recall and false recognitiondata within the DRM paradigm and the theory has beenextended in interesting ways to account for new varia-tions on the paradigm (eg Brainerd Wright Reyna ampMojardin 2001) The fundamental assumption of fuzzy-trace theory is that experience leaves two types of mem-ory traces gist traces (which capture the meaning of ex-perience but without specific attributes) and verbatimtraces (which represent specific attributes) False memo-ries are attributed to gist processing When the meaning ofa recognition probe matches the gist of the encoded expe-rience false recognition occurs The theory has been di-rected primarily to false recognition and not false recall
In trying to explain the present results one difficultyis that the operational definition of gist has never beenspecified that is within the DRM paradigm given a list
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 395
of 15 related words how does one specify the gist repre-sentation of the list and quantify how strong it is Onecould use patterns of false recognitionmdashto the extentthat a concept is falsely recognized it was represented aspart of the gist representation This definition is per-fectly suitable for some purposes but clearly explainingfalse recognition is not one of them owing to the prob-lem of circularity To explain why some lists lead to highlevels of false recall and false recognition (whereas otherlists lead to low levels) within fuzzy-trace theory oneneeds an independent specification of gist which to ourknowledge has not been provided in this context Whydoes the sweet list have so much gist that it leads to falserecall whereas the bitter list does not One plausible in-terpretation of gist is BAS The more the list items areassociated to the critical item the more a gist represen-tation of the critical item is created This solution to adefinition of gist works within the DRM paradigm butof course the interpretation of how gist is created be-comes isomorphic to activation processes within the ac-tivationmonitoring framework
Interpretation of the other factor that was significantwithin our multiple regression analysis may be morestraightforward within fuzzy-trace theory The negativecorrelation between veridical recall and false recall (andfalse recognition) may be accounted for by increased re-liance on verbatim traces That is if increased recall of listitems is due to increased verbatim information subjectsmay rely less heavily on gist information and hence showless false recall and false recognition during respondingSimilarly for critical items that are long andor ortho-graphically distinct recall of list items that relies on thesefeatures may lead to suppression of false recall and falserecognition of the critical items because they may differsystematically from other list items These factors pos-tulated within the context of fuzzy-trace theory are quitesimilar to those within the activationmonitoring frame-work outlined above At some level of generality there aretwo sets of factors in each account and both theories canexplain many empirical results
Shifting Criterion TheoryMiller and Wolford (1999) proposed that false recog-
nition in the DRM paradigm could be explained by sub-jectsrsquo shifting their criteria differentially to list items andto critical items during a recognition test (They ex-tended their explanation to false recall but somewhattenuously by proposing a generaterecognize theory)They postulated that subjects encoded the structure ofthe list during its presentation and then during the recog-nition test responded with especially liberal criteria tocritical items but not to list items The criterion shift issupposed to occur to the degree that the test item appearsto be related to the list items To support their argumentthey reported that a measure of bias from signal detec-tion theory was indeed more liberal for critical itemsthan for list items However Wickens and Hirshman(2000) and Wixted and Stretch (2000) pointed out that
within signal detection theory differences in bias mea-sures do not necessarily implicate a criterion shiftRather differences in measured bias could arise fromdifferences among the underlying distributions Wixtedand Stretch argued that this latter interpretation was themore likely and that the criterion shift interpretation wasuntenable (also see Roediger amp McDermott 1999)
The present evidence also weighs heavily against acriterion shift account Miller and Wolford (1999) ap-pealed to subjectsrsquo metaknowledge of the structure of thelist Within the criterion shift account we can assumethat the likelihood that a critical item cues metaknowl-edge of the list depends on its FAS or how well the crit-ical item will cue the list If so FAS (the association ofthe critical item to the list items) should be the primarydeterminant of false recognition and false recall How-ever the correlation between FAS and false recall in ourdata was negligible (08) and the same was true in analy-ses of false recognition BAS was actually the primarypredictor of false recall which implicates activation pro-cesses during encoding as being a primary cause of theeffect There would be no reason to expect BAS to haveany impact according to the idea that a simple criterionshift during the test causes false recall and false recog-nition Of course all our lists were relatively high androughly equivalent in terms of FAS because they werecreated from forward associations to the critical item sorestriction of range may account for the low correlationEven so the same problem exists for the shifting crite-rion theory to explain the results If FAS is supposed todetermine false recall and false recognition by this ac-count and if all our lists were high and roughly equal onthis feature why were so many lists ineffective in elicitingfalse recognition and false recall In short either way theshifting criterion theory cannot account for the results
Of course within the activationmonitoring frame-work the monitoring processes occurring during retrievalcould be conceived as conservative shifts in response cri-teria However in our theory these monitoring processesdetermine the degree to which a given level of activationat retrieval will be judged sufficient to endorse a criticalitem as having been studied thereby producing a falsememory In contrast in Miller and Wolfordrsquos (1999) ac-count there was no activation of the critical item duringencoding and shifting criteria during the test were saidto account entirely for the false recognition phenomenon(see Wixted amp Stretch 2000) In the activationmonitoringaccount the memory illusion provoked by the DRM listsis assumed to be a subjectively powerful phenomenonthat can be ameliorated (but rarely eliminated) when listitems are made more distinctive relative to criticalitems Under conditions in which the subject is warnedat test (but not before the encoding phase) to monitortheir memories for the presence or absence of the criti-cal item and is discouraged from making criterion shiftsthe false memory effect is neither eliminated nor greatlyreduced (Gallo Roediger amp McDermott 2001) Hicksand Marsh (2001) showed that under conditions in which
396 ROEDIGER WATSON MCDERMOTT AND GALLO
subjects are explicitly told to monitor their memories dur-ing the test (and are asked to judge whether each wordhad been externally presented or internally generated)false recognition in the DRM paradigm actually increasesrather than decreases In fact the false recognition effectpersists even when subjects are fully informed about thenature of the phenomenon before the study episode andare therefore encouraged to enhance monitoring processesduring both encoding and retrieval in order to attempt todiminish or eliminate the effect (see Gallo et al1997Gallo Roediger amp McDermott 2001 McDermott ampRoediger 1998) For these and other reasons the notionthat the DRM memory illusion is caused by a general shiftof criteria can be eliminated as a potential explanation
Interitem Associative Strength or ConnectivityOur data pose something of a puzzle about the role of
interitem associative strength (Deese 1959a) or connec-tivity (McEvoy et al 1999) in false recall As was dis-cussed in the introduction Deese (1959a) found a sig-nificant negative correlation between this construct andthe probability of false recall In a later paper Deese(1961) obtained a weaker negative correlation that failedto reach conventional levels of significance so the issueis somewhat in doubt from Deesersquos (1959a 1961) researchMcEvoy et al created lists that varied in connectivity(interitem associative strength) and in BAS and manip-ulated these factors orthogonally They replicated thefinding of Deese (1959a) in that the density of inter-connections of list words was negatively related to falserecall However in our analysis of 55 lists (different fromthose used by McEvoy et al 1999) we found no relationbetween connectivity and false recall (r 5 04) or be-tween connectivity and false recognition (r 5 03)
Further research will be necessary to determine thetrue state of affairs although a case can be made that thelack of correlation we obtained is accurate First we had55 lists that varied widely in their ability to elicit false re-call as compared with 24 lists used in McEvoy et al(1999) Our lists also represented a relatively full rangeof connectivity or interitem associative strength relativeto those of McEvoy et al McEvoy et al used lists thathad a somewhat wider range of connectivity because theyused extreme values whereas our larger set of lists pro-vided a fuller body of the middle ranges of connectivityGiven our results it seems at least possible that interitemassociative strength (connectivity) is not a factor in de-termining whether lists of associative words lead to falserecall although the issue remains in doubt
We did find that interitem associative strength or con-nectivity is positively and significantly correlated withveridical recall (r 5 132) This outcome is consistentwith that of Deese (1959a) and McEvoy et al (1999)The greater the associative bonds among list items thegreater is list recall This outcome is also consistent withvirtually any associative theory of recall (eg Anderson1972 1983 Nelson Schreiber amp McEvoy 1992)
CONCLUSION
The present study has uncovered two primary factorsthat are significantly related to false recall and falserecognition in the DRM paradigm BAS is positivelycorrelated with false recall whereas veridical recall ofthe list items is negatively related to false recall The ac-tivationmonitoring account provides a natural interpre-tation of these factors Any viable theory posited to ex-plain the occurrence of false memories in the DRMparadigm must explain the variability in false recall andfalse recognition across lists and the two primary factorsresponsible for this variability
REFERENCES
Anderson J R (1972) FRAN A simulation model of free recall InG H Bower (Ed) The psychology of learning and motivation (Vol 5pp 315-378) New York Academic Press
Anderson J R (1983) The architecture of cognition CambridgeMA Harvard University Press
Anderson J R amp Pirolli P L (1984) Spread of activation Journalof Experimental Psychology Learning Memory amp Cognition 10791-798
Balota D A Cortese M J Duchek J M Adams D RoedigerH L III McDermott K B amp Yerys B E (1999) Veridical andfalse memories in healthy older adults and in dementia of the Alz-heimers type Cognitive Neuropsychology 16 361-384
Balota D A Dolan P O amp Duchek J M (2000) Memorychanges in healthy older adults In E Tulving amp F I M Craik (Eds)The Oxford handbook of memory (pp 395-409) Oxford Oxford Uni-versity Press
Balota D A amp Duchek J M (1988) Age-related differences inlexical access spreading activation and simple pronunciation Psy-chology and Aging 3 84-93
Balota D A amp Duchek J (1991) Semantic priming effects lexicalrepetition effects and contextual disambiguation effects in healthyaged individuals and individuals with senile dementia of theAlzheimer type Brain amp Language 40 181-201
Bartlett F C (1932) Remembering A study in experimental and so-cial psychology Cambridge Cambridge University Press
Brainerd C J Wright R Reyna V F amp Mojardin A H (2001)Conjoint recognition and phantom recollection Journal of Experi-mental Psychology Learning Memory amp Cognition 27 307-327
Bruce D amp Winograd E (1998) Remembering Deesersquos 1959 arti-cles The Zeitgeist the sociology of science and false memoriesPsychonomic Bulletin amp Review 5 615-624
Bruner J S (1957) On perceptual readiness Psychological Review64 123-152
Budson A E Daffner K R Desikan R amp Schacter D L(2000) When false recognition is unopposed by true recognitionGist-based memory distortion in Alzheimerrsquos disease Neuropsy-chology 14 277-287
Collins A M amp Loftus E F (1975) A spreading-activation theoryof semantic memory Psychological Review 82 407-428
Coltheart M Davelaar E Jonasson J T amp Besner D (1977)Access to the internal lexicon In S Dormic (Ed) Attention and per-formance VI (pp 535-555) Hillsdale NJ Erlbaum
Deese J (1959a) Influence of interitem associative strength upon im-mediate free recall Psychological Reports 5 235-241
Deese J (1959b) On the prediction of occurrence of particular verbalintrusions in immediate recall Journal of Experimental Psychology58 17-22
Deese J (1961) Associative structure and the serial reproduction ex-periment Journal of Abnormal amp Social Psychology 63 95-100
Deese J (1965) The structure of associations in language and thoughtBaltimore Johns Hopkins University Press
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 397
Gallo D A McDermott K B Percer J M amp Roediger H LIII (2001) Modality effects in false recall and false recognitionJournal of Experimental Psychology Learning Memory amp Cogni-tion 27 339-353
Gallo D A Roberts M J amp Seamon J G (1997) Rememberingwords not presented in lists Can we avoid creating false memoriesPsychonomic Bulletin amp Review 4 271-276
Gallo D A amp Roediger H L III (2001) Variability among wordlists in evoking associative memory illusions Manuscript submittedfor publication
Gallo D A Roediger H L III amp McDermott K B (2001) As-sociative false recognition occurs without strategic criterion shiftsPsychonomic Bulletin amp Review 8 579-586
Hicks J L amp Marsh R L (2001) False recognition occurs more fre-quently during source identification than during oldndashnew recogni-tion Journal of Experimental Psychology Learning Memory ampCognition 27 375-383
Hunt R R amp Einstein G O (1981) Relational and item-specific in-formation in memory Journal of Verbal Learning amp Verbal Behav-ior 19 497-514
Hunt R R amp McDaniel M A (1993) The enigma of organizationand distinctiveness Journal of Memory amp Language 32 421-445
Israel L amp Schacter D L (1997) Pictorial encoding reduces falserecognition of semantic associates Psychonomic Bulletin amp Review4 577-581
Jacoby L L Kelley C M amp Dywan J (1989) Memory attribu-tions In H L Roediger III amp F I M Craik (Eds) Varieties of mem-ory and consciousness Essays in honour of Endel Tulving (pp 391-422) Hillsdale NJ Erlbaum
Johnson M K Hashtroudi S amp Lindsay D S (1993) Sourcemonitoring Psychological Bulletin 114 3-28
Johnson M K amp Raye C L (1981) Reality monitoring Psycho-logical Review 88 67-85
Kellogg RT (2001) Presentation modality and mode of recall in ver-bal false memory Journal of Experimental Psychology LearningMemory amp Cognition 27 913-919
Kirkpatrick E A (1894) An experimental study of memory Psy-chological Review 1 602-609
Ku Iumlcera H amp Francis W (1967) Computational analysis of present-day American English Providence RI Brown University Press
Marsh E J McDermott K B amp Roediger H L III (2001) Doretrieval factors play a role in the creation of of false memories Man-uscript submitted for publication
Mather M Henkel L A amp Johnson M J (1997) Evaluatingcharacteristics of false memories Rememberknow judgments andmemory characteristics questionnaire compared Memory amp Cogni-tion 25 826-837
McDermott K B (1996) The persistence of false memories in list re-call Journal of Memory amp Language 35 212-230
McDermott K B (1997) Priming on perceptual implicit memorytests can be achieved through presentation of associates Psycho-nomic Bulletin amp Review 4 582-586
McDermott K B amp Roediger H L III (1998) Attempting to avoidillusory memories Robust false recognition of associates persistsunder conditions of explicit warnings and immediate testing Journalof Memory amp Language 39 508-520
McDermott K B amp Watson J M (2001) The rise and fall of falserecall The impact of presentation duration Journal of Memory ampLanguage 45 160-176
McEvoy C L Nelson D L amp Komatsu T (1999) What is the con-nection between true and false memories The differential roles of in-teritem associations in recall and recognition Journal of Experimen-tal Psychology Learning Memory amp Cognition 25 1177-1194
Meyer D E amp Schvaneveldt R W (1971) Facilitation in recog-nizing pairs of words Evidence of a dependence between retrievaloperations Journal of Experimental Psychology 90 227-234
Miller M B amp Wolford G L (1999) Theoretical commentaryThe role of criterion shift in false memory Psychological Review106 398-405
Neely J H (1977) Semantic priming and retrieval from semanticmemory Roles of inhibitionless spreading activation and limited-
capacity attention Journal of Experimental Psychology General106 226-254
Nelson D L McEvoy C L amp Schreiber T A (1999) The Uni-versity of South Florida word association rhyme and word fragmentnorms Unpublished manuscript University of South Florida Tampa
Nelson D L Schreiber TA amp McEvoy C L (1992) Processing im-plicit and explicit representations Psychological Review 99 322-348
Norman K A amp Schacter D L (1997) False recognition in youngerand older adults Exploring the characteristics of illusory memoriesMemory amp Cognition 25 838-848
Paivio A Yuille J C amp Madigan S A (1968) Concreteness im-agery and meaningfulness values for 925 nouns Journal of Experi-mental Psychology Monographs 76 (1 Pt 2)
Payne D G Elie C J Blackwell J M amp Neuschatz J S (1996)Memory illusions Recalling recognizing and recollecting events thatnever occurred Journal of Memory amp Language 35 261-285
Reyna V F amp Brainerd C J (1995) Fuzzy-trace theory An interimsynthesis Learning amp Individual Differences 7 1-75
Roediger H L III Balota D A amp Watson J M (2001) Spreadingactivation and the arousal of false memories In H L Roediger III J SNairne I Neath amp A M Surprenant (Eds) The nature of remem-bering Essays in honor of Robert G Crowder (pp 95-115) Wash-ington DC American Psychological Association
Roediger H L III amp McDermott K B (1995) Creating false mem-ories Remembering words not presented in lists Journal of Experi-mental Psychology Learning Memory amp Cognition 21 803-814
Roediger H L III amp McDermott K B (1999) False alarms aboutfalse memories Psychological Review 106 406-410
Roediger H L III amp McDermott K B (2000) Tricks of memoryCurrent Directions in Psychological Science 9 123-127
Roediger H L III McDermott K B amp Pisoni D B (2001) Rec-ollection of illusory voices Manuscript submitted for publication
Roediger H L III McDermott K B amp Robinson K J (1998)The role of associative processes in creating false memories In M AConway S E Gathercole amp C Cornoldi (Eds) Theories of memoryII (pp 187-245) Hove UK Psychological Press
Russell W A amp Jenkins J J (1954) The complete Minnesotanorms for responses to 100 words from the KentndashRosanoff Word As-sociation Test (Tech Rep No 11 Contract N8 ONR 66216 Officeof Naval Research) University of Minnesota Minneapolis
Schacter D L Israel L amp Racine C (1999) Suppressing falserecognition in younger and older adults The distinctiveness heuris-tic Journal of Memory amp Language 40 1-24
Schacter D L Verfaellie M amp Pradere D (1996) The neuro-psychology of memory illusions False recall and recognition in am-nesic patients Journal of Memory amp Language 35 319-334
Seamon J G Luo C R amp Gallo D A (1998) Creating false mem-ories of words with or without recognition of list items Evidence fornonconscious processes Psychological Science 9 20-26
Smith R E amp Hunt R R (1998) Presentation modality affects falsememory Psychonomic Bulletin amp Review 5 710-715
Stadler M A Roediger H L III amp McDermott K B (1999)Norms for word lists that create false memories Memory amp Cogni-tion 27 494-500
Thapar A amp McDermott K B (2001) False recall and false recog-nition induced by presentation of associated words Effects of retentioninterval and level of processing Memory amp Cognition 29 424-432
Toglia M P amp Battig W F (1978) Handbook of semantic wordnorms Hillsdale NJ Erlbaum
Toglia M P Neuschatz J S amp Goodwin K A (1999) Recall ac-curacy and illusory memories When more is less Memory 7 233-256
Tulving E (1974) Cue-dependent forgetting American Scientist 6274-82
Tulving E (1979) Relation between encoding specificity and levelsof processing In L S Cermak amp F I M Craik (Eds) Levels of pro-cessing in human memory (pp 405-428) Hillsdale NJ Erlbaum
Tulving E (1985) Memory and consciousness Canadian Psycholo-gist 26 1-12
Tun P A Wingfield A Rosen M J amp Blanchard L (1998)Response latencies for false memories Gist-based processes in nor-mal aging Psychology amp Aging 13 230-241
398 ROEDIGER WATSON MCDERMOTT AND GALLO
Underwood B J (1965) False recognition produced by implicit ver-bal responses Journal of Experimental Psychology 70 122-129
Weldon M S (1991) Mechanisms underlying priming on perceptualtests Journal of Experimental Psychology Learning Memory ampCognition 17 526-541
Wickens T D amp Hirshman E (2000) False memories and statisticaldecision theory Comment on Miller and Wolford (1999) and Roedi-ger and McDermott (1999) Psychological Review 107 377-383
Wixted J T amp Stretch V (2000) The case against a criterion-shiftaccount of false memory Psychological Review 107 368-376
Zechmeister E B (1969) Orthographic distinctiveness Journal ofVerbal Learning amp Verbal Behavior 8 754-761
NOTES
1 The orthographic distinctiveness of a critical item was obtained bypresenting 72 subjects with all 55 critical items visually and havingthem rate the words in accordance with procedures developed by Zech-meister (1969) Briefly the subjects were instructed to rate each wordfor its physical appearance in terms of distinctiveness in the languageon a 9-point scale with 1 being not very distinctive 5 corresponding toaverage distinctiveness and 9 being most distinctive To avoid orderingeffects three different randomized orderings of words were used acrossroughly equal groups of subjects In the case of the critical item sleep
for example the distinctiveness is 515 Corresponding values for but-terfly and for man are 744 and 213 respectively Although ortho-graphic distinctiveness was not included in our multiple regressionanalyses this predictor is included in Appendix C
2 The orthographic neighborhood size or Coltheartrsquos N of each crit-ical item was defined as the number of words that can be formed bychanging a single letter in the word (Coltheart Davelaar Jonasson ampBesner 1977) For example because it has no orthographic neighborsColtheartrsquos N for the critical item butterfly is zero In contrast sleep hasfive orthographic neighbors (ie sheep sleek sleet steep and sweep)The reference set of words used were those in Ku Iumlcera and Francisrsquos(1967) norms Although Coltheartrsquos N was not included in our multipleregression analyses this predictor is included in Appendix C
3 Word length orthographic distinctiveness and orthographic neigh-borhood size of the critical item were all highly correlated Specificallythe bivariate correlations among these three predictors were as followslength and orthographic distinctiveness r 5 184 t(53) 5 1117 p lt05 orthographic distinctiveness and orthographic neighborhood sizer 5 66 t(53) 5 640 p lt 05 length and orthographic neighbor-hood size r 5 63 t(53) 5 582 p lt 05 Of these three variablesword length had the largest bivariate correlation with false recall (r 5
37) so it was the predictor used in the multiple regression analyses4 The multiple regression data matrix is available at wwwiacwustl
edunmclWeb
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 399
APPENDIX A The Fifty-Five 15-Word Lists and Their Associated Critical Items
With Backward Associative Strength (BAS) and Forward Associative Strength (FAS)Values Used in the Multiple Regression Analysis of False Recall
Item Word BAS FAS Item Word BAS FAS
Anger mad 393 412 Black white 655 557fear 020 059 dark 111 100hate 028 109 cat 000 043rage 541 042 charred 023 000temper 182 000 night 000 021fury 306 000 funeral 034 000ire 179 000 color 074 050wrath 128 000 grief 000 000happy 000 042 blue 028 000fight 034 000 death 016 014hatred 070 000 ink 020 000mean 090 000 bottom 000 000calm 000 000 coal 288 000emotion 000 000 brown 338 000enrage 378 000 gray 365 000
Mean 157 044 Mean 130 052Army Navy 543 500 Bread butter 364 487
NotemdashUnless otherwise indicated with an asterisk all BAS and FAS values are drawn from the associative norms of Nel-son McEvoy and Schreiber (1999) Those values shown with an asterisk were obtained by the authors by using normingprocedures similar to those of Nelson et al (1999) The corresponding mean for these BAS and FAS values for each list arealso represented
406 ROEDIGER WATSON MCDERMOTT AND GALLO
APPENDIX B Data Matrix of the Fifty-Five 15-Word Lists and Associated Critical Items
Used in the Multiple Regression Analysis of False Recall
Critical False False Log Vrd Raw Ortho Item Recall Recgn Length Freq Concrete FAS BAS Connect Recall Freq Dist Colt N
(Manuscript received July 10 2000revision accepted for publication February 23 2001)
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 389
2 The raw frequency of each critical item was obtained from theKu Iumlcera and Francis (1967) norms and is the number of instancesfound in print per million words For example the raw frequency ofthe critical item sleep is 65 and the raw frequency of the criticalitem butterfly is 2 To correct for skewness in the frequency distri-bution for each critical item raw frequencies were transformedinto log frequencies using the following formula log (5 1 rawfrequency) For example the log frequency of sleep is 182 [ie log(655)] whereas the log frequency of butterfly is 40 [ie log (25)]
3 Concreteness ratings for each critical item were obtained fromthe word association norms of Nelson et al (1999) which were inturn obtained in part from the Paivio Yuille and Madigan (1968)and the Toglia and Battig (1978) norms These ratings are obtainedon a scale of 1ndash7 with 1 being the least concrete and 7 the mostconcrete To carry through with our examples sleep is slightlyabove the midpoint in concreteness (with a rating of 474) and but-terfly is more concrete (with a rating of 591)
List variables With respect to the study lists we used the fol-lowing four variables FAS BAS interitem associative strengthand veridical recall We consider each below
1 Forward associative strength Indices of the strength of asso-ciative connections from the critical item to the study words wereobtained from the word association norms of Nelson et al (1999)Specifically for each critical item we obtained the probability thatit elicited each word in its corresponding study list as an associatea measure Deese (1959b) referred to as the FAS For example as isshown in the third column of Appendix A for the critical item sleepwe identified the strength of its connections to bed rest awaketired and the remaining study words We then averaged these asso-ciations to obtain the mean FAS value for each list In the case ofsleep the mean FAS value for its 15 study words is 047 whereasin the case of butterfly the mean FAS value is 033 Note that all 15list items were given as a response to the critical item in the Russelland Jenkins (1954) norms (from which the lists were generated)However owing to discrepancies between these norms and those ofNelson et al (1999) the number of forward connections for eachlist could have theoretically ranged from 0 to 15
2 Backward associative strength An index of the strength of as-sociative connections from the study words to the critical item wasobtained in large part from the Nelson et al (1999) word associa-tion norms However for some of the 55 DRM lists there were afew list words that were missing from these norms For these wordswe collected our own norms using the procedures outlined by Nel-son and colleagues Briefly we gave groups of over 75 undergrad-uates one of three sheets of 80 to-be-normed words the subjectswere instructed to write next to each word the first word it broughtto mind In those cases in which more than one word from a partic-ular study list needed to be normed these words were placed onseparate norming sheets or were separated by at least 30 items onthe same norming sheet At least 86 observations were obtained foreach normed item and 150 or more observations were obtained foran item in some cases
With these two sets of norms we took each study word and de-termined its connection strength to its corresponding critical itemthe measure Deese (1959b) referred to as the BAS For example asis shown in the second column of Appendix A we obtained for eachstudy word (eg bed rest awake tired etc) the probability thatit elicited its critical item (eg sleep) as an associate Thus thismeasure examines connections in the direction opposite to that usedto obtain the FAS measures We then averaged these associations toobtain the mean BAS value for each list For example the meanBAS value for the sleep list is 431 By comparison the mean BASvalue for the butterfly list is 045
3 Interitem associative strength or connectivity The average in-teritem associative strength (Deese 1959a) or connectivity (Mc-Evoy et al 1999) of the study words was obtained for each listusing two steps First following a procedure outlined by McEvoy
et al we constructed a 15 3 15 matrix of study items for each listIn Appendix C we have illustrated the connectivity matrix for thesleep list Second using the Nelson et al (1999) norms and insome instances our own word association norms we identified thepresence of a connection for all 225 combinations of study wordpairs in each listrsquos matrix excluding the 15 self-connections whichwere not determined (listed as ND) For each word pair a nonzeroconnection strength was scored as a 1 and a zero connectionstrength was scored as a 0 In the case of sleep a total of 27 nonzeroconnections exists among the study words (or an average of 180connections per study word) In the case of butterfly a total of 29nonzero connections exists among the study words (or an averageof 193 connections per study word) High mean connectivityscores indicate a greater number of associative relationships amongthe study words whereas low mean connectivity scores indicate alesser number of associative relationships among the study words
4 Veridical recall of the list One final predictor was the averageprobability of recall of the studied words collapsing across serialpositions 1ndash15 for each list These data were obtained from Stadleret al (1999) and Gallo and Roediger (2001) In the case of sleepstudied items were recalled at a 61 level whereas in the case ofbutterfly studied items were recalled at a 70 level As was notedearlier the split-half correlation of the veridical recall measure wasreasonably high at +86
ProcedureTaking the seven factors that were just described along with the
probability of false recall and the probability of false recognitionfor each of the 55 lists we constructed the data matrix shown inAppendix B for our simultaneous multiple regression analysis4
Each of the 55 lists used in the regression analysis is represented byits respective critical item each of which is listed in alphabeticalorder in the f irst column of the matrix The second and thirdcolumns of the matrix contain the probability of false recall andfalse recognition respectively of each critical item from the Stadleret al (1999) and Gallo and Roediger (2001) norms The next threecolumns contain properties of the critical item itself length log fre-quency and concreteness The next four columns contain propertiesof the list that was constructed for each critical item mean FASmean BAS the mean connectivity of the study words and the av-erage recall of the studied items The last three columns contain thecritical item predictors that were excluded from the multiple re-gression analyses raw frequency orthographic distinctiveness andorthographic neighborhood size (Coltheartrsquos N )
In multiple regression analysis the independent variables can beentered either simultaneously (in a single step) or hierarchically (withsome variables entered into the analysis before others) In the for-mer case all independent variables are allowed to compete forshared variance in the dependent variable (eg false recall) in a sin-gle step so that no single variable is given greater priority than isanother In the latter case those independent variables entered inearlier steps are allowed to capture variance in the absence of com-petition from variables to be entered subsequently In this way in ahierarchical analysis earlier variables are given greater prioritythan later ones usually for theoretical reasons For the purpose ofthe present analysis there exists no definitive theoretical justifica-tion for ordering the entry of critical item variables and list vari-ables so we opted for the simultaneous entry of all variables in asingle step
RESULTS
We consider first the bivariate correlations betweenour seven factors and false recall and then we turn to thesimultaneous multiple regression analysis and its find-ings The entire matrix of correlations (Pearsonrsquos r in all
390 ROEDIGER WATSON MCDERMOTT AND GALLO
cases) among the seven predictors and false recall isshown in Table 1
Bivariate CorrelationsIn describing the results we report significant bivari-
ate correlations between critical item variables and falserecall followed by significant correlations between listvariables and false recall To determine these values eachof the critical item and list variables were entered into aregression equation as the sole predictor of false recallOnly three of our seven variables were significantly cor-related ( p lt 05) with false recall the critical item vari-able of length and the list variables of BAS and recall ofthe studied items (Figures 1A and 1B respectively)
Length The length of the critical item and false recallwere negatively correlated [r 5 37 t(53) 5 287]
Backward associative strength Mean BAS or thestrength of the connections between study items andtheir corresponding critical items was positively corre-lated with false recall r 5 173 t(53) 5 773 As isshown in Figure 1A the greater the strength of associa-tion between the list items and the critical item the morelikely is false recall This outcome replicates Deesersquos(1959b) finding of a 187 correlation between BAS andfalse recall which he obtained with a set of thirty-six 12-item lists Although the 173 correlation between thesetwo variables in our set of f ifty-f ive 15-item lists issomewhat smaller than the 187 correlation obtained byDeese BAS is still the strongest correlate of false recallin our entire data set
Veridical recall Recall of list items and false recallwere negatively correlated [r 5 43 t(53) 5 346]As is shown in Figure 1B increasing levels of accuraterecall are associated with decreasing levels of false re-call This outcome is contrary to some ideas about whyfalse recall occurs as will be brought out in the Discus-sion section
As is shown in Table 1 no other variables were corre-lated with false recall at the bivariate level which is sur-prising in some cases For example in terms of criticalitem variables one might have expected word frequencyor concreteness to be correlated with false recall but nei-ther did (Deese 1959b also failed to find a relation be-tween word frequency and false recall) In terms of listvariables one might have expected connectivity to cor-relate with false recall but it did not The lack of a rela-
tion between connectivity and false recall is noteworthybecause Deese (1959a) and McEvoy et al (1999) re-ported a significant relationship between these variablesso that lists with greater connectivity were less likely toelicit false recall of the critical item In the Discussionsection we consider connectivity or interitem associa-tive strength more fully We turn now to the results of thesimultaneous multiple regression analysis
Simultaneous Multiple Regression AnalysisIn the multiple regression analysis the three variables
based on the characteristics of the critical items them-selves (ie length log frequency and concreteness) andthe four variables based on the characteristics of the listsdesigned to elicit the critical items (ie FAS and BASmeasures interitem associative strength or connectivityand the level of recall of list items) were entered simul-taneously into the regression equation to predict false re-call The results of the analysis are shown in Table 2 in-cluding beta weights (b) semipartial correlations (sr)and significance tests for each predictor For signifi-cance tests a t(47) was used for each predictor and theresultant t statistic is expressed as the absolute valuemdashthat is frac12tfrac12 In addition Table 2 displays the results of thesignificance test for the multiple R2 for the entire set ofseven predictor variables
As is shown in Table 2 the results of the regressionanalysis indicated that two list variables were signifi-cantly related to false recall Specifically BAS was thestrongest predictor of false recall (b 5 170 sr 5160 t 5 717) Therefore consistent with our findingsat the bivariate level the greater the strength of associa-tion between study items and their corresponding criti-
Table 1 Correlation Matrix Among Eight Variables False Recall Three Critical Item Variables and Four List Variables
False Log VeridicalRecall Length Frequency Concreteness FAS BAS Connectivity Recall
NotemdashR2 5 68 F(747) 5 1406 MSe 5 0013 p lt 05 p lt 05
Figure 1 Probability of false recall of the critical item from 55 lists (A) as a functionof associative strength of the list items to the critical item or backward associativestrength and (B) as a function of veridical recall of studied items from the lists
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 391
cal item on a free association test the more probable isfalse recall of the critical item Veridical recall was alsoa strong predictor of false recall (b 5 40 sr 5 34t 5 411) As we found at the bivariate level the morewords recalled from a list the less likely that the criticalitem for that list was falsely recalled None of the re-maining critical item or list variables including length ofthe critical item connectivity and FAS explained aunique portion of the total variance in false recall afterthe influence of other predictors had been residualized inthe context of the regression equation Therefore BASand recall of list items appeared to be the major contrib-utors to the multiple R of 182 and the associated mul-tiple R2 of 68 [F(747) 5 1406 MSe 5 0013] We fur-ther explore the implications of these effects in theDiscussion section
Factors Affecting False RecognitionIn our analyses thus far we have concentrated on
false recall exclusively However the norms for our 55lists (collected by Stadler et al 1999 and Gallo ampRoediger 2001) also include values for false recogni-tion The recognition data in these norms were collectedin a final recognition test after lists had been studied andrecalled This confounding of recognition by prior recallalmost certainly affected both veridical and false recog-nition However in experiments directly examining theeffects of recall on recognition in this paradigm and oth-ers the finding is that recall of a list may increase itsrecognition by a moderate amount (typically in the4ndash10 range eg Roediger amp McDermott 1995Roediger McDermott amp Pisoni 2001) In fact some re-searchers have failed to observe any effect at all of recall
392 ROEDIGER WATSON MCDERMOTT AND GALLO
on recognition (Schacter Verfaellie amp Pradere 1996 seeRoediger et al 1998 for a discussion) We conducted thesame regression analyses with our seven predictor vari-ables on false recognition as we did on false recall Wesummarize the basic results and conclusions as follows
There were three significant bivariate correlationswith false recognition including length (r 5 27)BAS (r 5 143) and veridical recall (r 5 52) Over-all these correlations with false recognition were simi-lar to those obtained with false recall However the pos-itive correlation between BAS and false recognition wassmaller than that in the recall data and the negative cor-relation between veridical recall and false recognitionwas larger than that in the recall data To maintain con-sistency with the previous regression analysis on falserecall we used veridical recall rather than veridicalrecognition in this analysis In addition recognition es-timates were based on only three itemslist and wereprobably affected by prior recall
Turning to the results of the simultaneous multiple re-gression analysis on false recognition BAS and veridi-cal recall of list items appeared to be the major contrib-utors to the multiple R of 169 and the associatedmultiple R2 of 48 [F(747) 5 611 MSe 5 002] Table 3shows the summary of the regression analyses Althoughthe regression analysis accounted for less total variancein false recognition than in false recall (48 as opposedto 68 respectively) the same factors seem to be re-sponsible for predicting false recognition and false re-call Of course because measures of recognition alwaysfollowed recall this similar pattern may reflect a carry-over effect of false recall on false recognition Nonethe-less within the limits of this study the factors responsi-ble for false recognition across lists seem to be the sameas those producing false recall
DISCUSSION
To summarize the main findings our results implicatetwo primary factors (of the seven variables we exam-ined) as accounting for most of the variance in false re-call across lists The split-half correlation of our lists was190 so the amount of potentially explainable variancein our dependent variable of false recall is 81 The fac-
tors in our multiple regression accounted for 68 of thetotal variance in false recall or 84 of the explainablevariance (68 4 81 3 100 5 84) The analysis forfalse recognition revealed the same factors at work al-though they accounted for less overall variance Thestrongest factor was BAS consistent with prior work byDeese (1959b) and McEvoy et al (1999) The other fac-tor was the number of items recalled accurately from thelist which is negatively correlated with false recall
We turn next to the theoretical implications of ourfindings We first consider the implications for the acti-vationmonitoring framework that we have developed toexplain false recall and false recognition in the DRMparadigm We then consider the implications for othertheories and f inally turn to several further issues onwhich our results shed light
ActivationMonitoring TheoryIn the theroretical framework that we (eg McDer-
mott amp Watson 2001 Roediger Balota amp Watson 2001Roediger amp McDermott 1995 2000) have used to explainvariations in the probability of false recall and false recog-nition in the DRM paradigm two sets of processes havebeen proposed Specifically we have proposed that pro-cesses occurring during both encoding and retrieval af-fect the probability of false recall and false recognitionin the DRM paradigm Two critical sets of processes in-volve activation and monitoring Although there is a nat-ural alignment of activation to encoding and monitoringto retrieval we note that both processes can potentiallyoccur both during encoding and during retrieval as willbe discussed below The ideas we employ have many pre-cursors which we acknowledge below although theirparticular arrangement here may be unique
A critical distinction in understanding encoding pro-cesses and their interaction with retrieval processes isbetween item-specific and relational processing (Ander-son 1972 Hunt amp Einstein 1981 Hunt amp McDaniel1993) Briefly the notion is that emphasis can be given tothe encoding of individual elements and their features(what Anderson called node-taggingand Hunt and his col-leagues called item-specific processing) or to the encod-ing of relationships among elements (pathway-taggingor relational processing) Item-specific processing should
Table 3 Summary of Multiple Regression Analysis
on False Recognition With Critical Item Variablesand List Variables Used as Predictors
NotemdashR2 5 48 F(747) 5 611 MSe 5 002 p lt 05 p lt 05
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 393
make individual elements more memorable and lead tomore distinctive memories such processing should helpdistinguish veridical memories from false memories be-cause list items would have features of presentation as-sociated with them whereas the critical item sometimeswould not (However if the critical item is strongly acti-vated during encoding it may take on many of the fea-tures of list items) Relational processing (extractingthemes or schemas or constructing networks of associa-tions) should increase the probability of false recall andfalse recognition by leading to inferences or associationsof events that may not actually have occurred
We assume that in encoding and interpreting experi-ences people go beyond the information given in thestimulus (Bartlett 1932 chap 2 Bruner 1957) and ac-tivate related information through inferential processingEach personrsquos experience is coded in terms of his or herown schemata (or associative networks or knowledgestructures) When schemata are activated a process offilling in of ldquomissingrdquo information may occur wherebyitems strongly implied by the schema (but not actuallypresent in the stimulus) may become activated (Bartlett1932) In encoding lists of words in the DRM paradigmpresentation of the list can create activation that spreadsthroughout the lexicalsemantic system and can createimplicit associative responses (Underwood 1965) Thisactivation may arouse concepts that are not presented but that are associated to concepts that are presented(eg Anderson amp Pirolli 1984 Collins amp Loftus 1975Meyer amp Schvaneveldt 1971) The associative activa-tion in this approach can arise from automatic process-ing or from more controlled elaborative processing(Neely 1977) In addition the associative activation ofconcepts that were not presented may occur consciously(subjects may become aware of the concept of sleep andrehearse it during presentation of the list words bed restawake etc) or the concept may be strongly activatedbut never come to conscious awareness The debate overwhether arousal of the critical item in the DRM para-digm is conscious or unconscious is interesting but dif-ficult to decide empirically Some evidence has impli-cated conscious arousal of the concepts (McDermott1997) whereas other evidence implicates unconsciousactivation (Seamon Luo amp Gallo 1998) The correctanswer to the question of whether activation is consciousor unconscious is probably ldquobothrdquomdashthat is the criticalitem probably sometimes comes consciously to mindand even if it does not relevant information may never-theless be activated unconsciously
The high correlation between BAS and false recall aswell as the strong weighting of BAS in the multiple re-gression analysis strongly implicates activation of criti-cal items as an important factor in explaining false recalland false recognition in the DRM paradigm The morestrongly associated list items are to critical items themore activation would be expected to spread to the crit-ical items More colloquially to the extent that the wordson the list spark the associative connection to the critical
item the more likely the critical item is to be falsely re-called This activation factor would seem to be one mainreason why ldquothings suggested to a person by an experi-ence may be honestly reported by him as part of the ex-periencerdquo (Kirkpatrick 1894 p 608) These associatedlists probably all tend to elicit relational processing butonly those with strong BAS to the critical item lead tofalse recall
The second set of factors proposed in activationmon-itoring theory is concerned with the monitoring of mem-ory accuracy Monitoring of encoding processes canoccur during the study phase of an experiment espe-cially under intentional learning conditions In experi-ments in which subjects are warned about false memoryphenomena and are told to pay close attention to lists andto remember only the presented items and not the criti-cal items monitoring is enhanced (Gallo Roberts amp Sea-mon 1997 Gallo Roediger amp McDermott 2001 Mc-Dermott amp Roediger 1998) Monitoring of memorieshas typically been discussed with respect to retrieval dur-ing which time a primary goal is to distinguish fromamong the considerable information brought to con-sciousness that which refers to perception of past eventsfrom that which does not a process referred to as realitymonitoring (Johnson amp Raye 1981) The experience ofremembering events involves mentally traveling back intime and seeming to reexperience the events (Tulving1985) During retrieval recovered memories that appar-ently carry with them details of having been experiencedare generally attributed to be memories of actual pastevents although sometimes this attribution may be inerror (Jacoby et al 1989)
Johnson et al (1993) have described in detail the typesof information that may be used to distinguish eventsthat actually happened from ones that were only imag-ined thought about or read about For example eventsthat actually occurred have more features from the ex-ternal world associated with them (sights sounds etc)whereas those that were internally generated are lesslikely to have these features but more likely to have as-sociated reflections and cognitive operations In theDRM paradigm the fact that the critical item has beenstrongly activated during encoding means that duringretrieval the subject is likely to retrieve the critical itemas a list member To the extent that the critical items arestrongly activated during encoding they will take onsome features similar to those of the list items this is es-pecially true if the critical items come consciously tomind during encoding and are rehearsed in the samemanner as list items To the extent that the features of thecritical item are similar to those of actual list items theprobability of false recall or false recognition is increasedHowever if the list items systematically differ from thecritical item on some dimensions or features retrievalheuristics may be brought into play that can reduce falserecall and false recognition (Israel amp Schacter 1997Schacter Israel amp Racine 1999) Considerable evidencesupports this general source-reality-monitoring frame-
394 ROEDIGER WATSON MCDERMOTT AND GALLO
work that we advocate here (eg Johnson et al 1993Schacter et al 1999) as well as the attributional ap-proach to remembering (Jacoby et al 1989)
The other significant factor arising from our multipleregression analysis may operate during retrieval by aid-ing the monitoring processes that reduce false recall Thenegative correlation between veridical recall of list itemsand false recall of the critical item may indicate that thebetter encoded list items are the more easily they can bedistinguished from the illusory critical item Subjectsmay use item-specific information to aid recall of listitems and suppress recall of the critical items whichwould not carry as much item-specific information Whenexperimental manipulations increase veridical recall ei-ther through longer study time at relatively slow rates(Gallo amp Roediger 2001 McDermott amp Watson 2001Toglia Neuschatz amp Goodwin 1999) or through re-peated study and test opportunities (McDermott 1996Schacter et al 1999) false recall is often decreasedTherefore the negative relation between veridical recalland false recall can be produced experimentally as wellas appearing in our correlationregression analysis How-ever veridical and false recall can be positively related as afunction of other variables too as was shown by McDer-mott (1996) Toglia et al (1999) and Thapar amp McDermott(2001) In these cases greater activation of the criticalitem from relational processing or from greater seman-tic processing is probably at work
The fact that the length andor distinctiveness of thecritical item are negatively correlated with false recallcan be similarly interpreted as the effect of monitoringduring retrieval because the greater distinctiveness ofthe critical item simplifies the monitoring process Forexample if words in the list are generally shorter andhave more regular orthography than does the criticalitem then during retrieval the critical items that are longand distinctive (eg butterfly cabbage) should be moreeasily discriminated from list items as indeed seems tobe the case in the bivariate correlations Interestingly thelengthdistinctiveness factor played a role in determin-ing false recall even though the lists in our norms werepresented auditorily and the orthographic informationat least should not be particularly accessible In addi-tion presenting the DRM lists visually rather than audi-torily reduces false recall and false recognition at leaston tests in which recognition items are presented visually(as opposed to auditorily) or recall is written rather thanspoken (Gallo McDermott et al 2001 Kellogg 2001Smith amp Hunt 1998) This modality difference has alsobeen interpreted as arising from the retrieval of distinctiveinformation operating during the test (Gallo McDermottet al 2001 Smith amp Hunt 1998)
As was noted above activation processes may occurduring retrieval as well as during encoding The morelist items a subject recalls the more primed the criticalitem might become However the present analyses un-dercut the simplest form of this idea If the recalling oflist items activated or primed false recall the simplest
prediction would be that there should be a positive cor-relation between veridical recall and false recall Our datarevealed the opposite pattern showing a negative corre-lation between veridical and false recall Marsh McDer-mott and Roediger (2001) reported other experimentsthat also call into question the idea that priming duringthe test plays a role in false recall and false recognition
Balota et al (1999) obtained further evidence sup-porting the activationmonitoring interpretation of falsememories in the DRM paradigm They compared per-formance of younger adults healthy older adults andolder adults with Alzheimerrsquos disease The latter twogroups have been shown to have intact activation pro-cesses as measured by semantic priming paradigms(Balota amp Duchek 1988 1991) yet they are deficient insource-monitoring processes which is probably attribut-able to impaired frontal lobe functioning (see BalotaDolan amp Duchek 2000) Therefore one might expectthat whereas veridical recall would be worse in older adultsand Alzheimerrsquos patients (relative to younger adults)false recall and false recognition would occur at the samelevel or perhaps be increased in these groups IndeedBalota et al (1999) reported exactly this pattern A sim-ilar outcome has been obtained in other laboratories(eg Budsen Daffner Desikan amp Schacter 2000 Nor-man amp Schacter 1997 Tun Wingfield Rosen amp Blan-chard 1998) This pattern is just as predicted by the ac-tivation monitoring framework
In short the present results and many others (seeRoediger Balota amp Watson 2001 Roediger et al 1998)fit comfortably within the activationmonitoring frame-work In the remainder of our discussion we will con-sider briefly the implications of our present results forother accounts of the DRM paradigm and for the role ofconnectivity (or interitem associative strength) in pro-ducing veridical and false recall
Fuzzy-Trace TheoryAnother leading theory that has been used to explain
false memory phenomena is fuzzy-trace theory (egReyna amp Brainerd 1995) Payne et al (1996) and Schac-ter et al (1996) among many others have used the basictheory to account for false recall and false recognitiondata within the DRM paradigm and the theory has beenextended in interesting ways to account for new varia-tions on the paradigm (eg Brainerd Wright Reyna ampMojardin 2001) The fundamental assumption of fuzzy-trace theory is that experience leaves two types of mem-ory traces gist traces (which capture the meaning of ex-perience but without specific attributes) and verbatimtraces (which represent specific attributes) False memo-ries are attributed to gist processing When the meaning ofa recognition probe matches the gist of the encoded expe-rience false recognition occurs The theory has been di-rected primarily to false recognition and not false recall
In trying to explain the present results one difficultyis that the operational definition of gist has never beenspecified that is within the DRM paradigm given a list
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 395
of 15 related words how does one specify the gist repre-sentation of the list and quantify how strong it is Onecould use patterns of false recognitionmdashto the extentthat a concept is falsely recognized it was represented aspart of the gist representation This definition is per-fectly suitable for some purposes but clearly explainingfalse recognition is not one of them owing to the prob-lem of circularity To explain why some lists lead to highlevels of false recall and false recognition (whereas otherlists lead to low levels) within fuzzy-trace theory oneneeds an independent specification of gist which to ourknowledge has not been provided in this context Whydoes the sweet list have so much gist that it leads to falserecall whereas the bitter list does not One plausible in-terpretation of gist is BAS The more the list items areassociated to the critical item the more a gist represen-tation of the critical item is created This solution to adefinition of gist works within the DRM paradigm butof course the interpretation of how gist is created be-comes isomorphic to activation processes within the ac-tivationmonitoring framework
Interpretation of the other factor that was significantwithin our multiple regression analysis may be morestraightforward within fuzzy-trace theory The negativecorrelation between veridical recall and false recall (andfalse recognition) may be accounted for by increased re-liance on verbatim traces That is if increased recall of listitems is due to increased verbatim information subjectsmay rely less heavily on gist information and hence showless false recall and false recognition during respondingSimilarly for critical items that are long andor ortho-graphically distinct recall of list items that relies on thesefeatures may lead to suppression of false recall and falserecognition of the critical items because they may differsystematically from other list items These factors pos-tulated within the context of fuzzy-trace theory are quitesimilar to those within the activationmonitoring frame-work outlined above At some level of generality there aretwo sets of factors in each account and both theories canexplain many empirical results
Shifting Criterion TheoryMiller and Wolford (1999) proposed that false recog-
nition in the DRM paradigm could be explained by sub-jectsrsquo shifting their criteria differentially to list items andto critical items during a recognition test (They ex-tended their explanation to false recall but somewhattenuously by proposing a generaterecognize theory)They postulated that subjects encoded the structure ofthe list during its presentation and then during the recog-nition test responded with especially liberal criteria tocritical items but not to list items The criterion shift issupposed to occur to the degree that the test item appearsto be related to the list items To support their argumentthey reported that a measure of bias from signal detec-tion theory was indeed more liberal for critical itemsthan for list items However Wickens and Hirshman(2000) and Wixted and Stretch (2000) pointed out that
within signal detection theory differences in bias mea-sures do not necessarily implicate a criterion shiftRather differences in measured bias could arise fromdifferences among the underlying distributions Wixtedand Stretch argued that this latter interpretation was themore likely and that the criterion shift interpretation wasuntenable (also see Roediger amp McDermott 1999)
The present evidence also weighs heavily against acriterion shift account Miller and Wolford (1999) ap-pealed to subjectsrsquo metaknowledge of the structure of thelist Within the criterion shift account we can assumethat the likelihood that a critical item cues metaknowl-edge of the list depends on its FAS or how well the crit-ical item will cue the list If so FAS (the association ofthe critical item to the list items) should be the primarydeterminant of false recognition and false recall How-ever the correlation between FAS and false recall in ourdata was negligible (08) and the same was true in analy-ses of false recognition BAS was actually the primarypredictor of false recall which implicates activation pro-cesses during encoding as being a primary cause of theeffect There would be no reason to expect BAS to haveany impact according to the idea that a simple criterionshift during the test causes false recall and false recog-nition Of course all our lists were relatively high androughly equivalent in terms of FAS because they werecreated from forward associations to the critical item sorestriction of range may account for the low correlationEven so the same problem exists for the shifting crite-rion theory to explain the results If FAS is supposed todetermine false recall and false recognition by this ac-count and if all our lists were high and roughly equal onthis feature why were so many lists ineffective in elicitingfalse recognition and false recall In short either way theshifting criterion theory cannot account for the results
Of course within the activationmonitoring frame-work the monitoring processes occurring during retrievalcould be conceived as conservative shifts in response cri-teria However in our theory these monitoring processesdetermine the degree to which a given level of activationat retrieval will be judged sufficient to endorse a criticalitem as having been studied thereby producing a falsememory In contrast in Miller and Wolfordrsquos (1999) ac-count there was no activation of the critical item duringencoding and shifting criteria during the test were saidto account entirely for the false recognition phenomenon(see Wixted amp Stretch 2000) In the activationmonitoringaccount the memory illusion provoked by the DRM listsis assumed to be a subjectively powerful phenomenonthat can be ameliorated (but rarely eliminated) when listitems are made more distinctive relative to criticalitems Under conditions in which the subject is warnedat test (but not before the encoding phase) to monitortheir memories for the presence or absence of the criti-cal item and is discouraged from making criterion shiftsthe false memory effect is neither eliminated nor greatlyreduced (Gallo Roediger amp McDermott 2001) Hicksand Marsh (2001) showed that under conditions in which
396 ROEDIGER WATSON MCDERMOTT AND GALLO
subjects are explicitly told to monitor their memories dur-ing the test (and are asked to judge whether each wordhad been externally presented or internally generated)false recognition in the DRM paradigm actually increasesrather than decreases In fact the false recognition effectpersists even when subjects are fully informed about thenature of the phenomenon before the study episode andare therefore encouraged to enhance monitoring processesduring both encoding and retrieval in order to attempt todiminish or eliminate the effect (see Gallo et al1997Gallo Roediger amp McDermott 2001 McDermott ampRoediger 1998) For these and other reasons the notionthat the DRM memory illusion is caused by a general shiftof criteria can be eliminated as a potential explanation
Interitem Associative Strength or ConnectivityOur data pose something of a puzzle about the role of
interitem associative strength (Deese 1959a) or connec-tivity (McEvoy et al 1999) in false recall As was dis-cussed in the introduction Deese (1959a) found a sig-nificant negative correlation between this construct andthe probability of false recall In a later paper Deese(1961) obtained a weaker negative correlation that failedto reach conventional levels of significance so the issueis somewhat in doubt from Deesersquos (1959a 1961) researchMcEvoy et al created lists that varied in connectivity(interitem associative strength) and in BAS and manip-ulated these factors orthogonally They replicated thefinding of Deese (1959a) in that the density of inter-connections of list words was negatively related to falserecall However in our analysis of 55 lists (different fromthose used by McEvoy et al 1999) we found no relationbetween connectivity and false recall (r 5 04) or be-tween connectivity and false recognition (r 5 03)
Further research will be necessary to determine thetrue state of affairs although a case can be made that thelack of correlation we obtained is accurate First we had55 lists that varied widely in their ability to elicit false re-call as compared with 24 lists used in McEvoy et al(1999) Our lists also represented a relatively full rangeof connectivity or interitem associative strength relativeto those of McEvoy et al McEvoy et al used lists thathad a somewhat wider range of connectivity because theyused extreme values whereas our larger set of lists pro-vided a fuller body of the middle ranges of connectivityGiven our results it seems at least possible that interitemassociative strength (connectivity) is not a factor in de-termining whether lists of associative words lead to falserecall although the issue remains in doubt
We did find that interitem associative strength or con-nectivity is positively and significantly correlated withveridical recall (r 5 132) This outcome is consistentwith that of Deese (1959a) and McEvoy et al (1999)The greater the associative bonds among list items thegreater is list recall This outcome is also consistent withvirtually any associative theory of recall (eg Anderson1972 1983 Nelson Schreiber amp McEvoy 1992)
CONCLUSION
The present study has uncovered two primary factorsthat are significantly related to false recall and falserecognition in the DRM paradigm BAS is positivelycorrelated with false recall whereas veridical recall ofthe list items is negatively related to false recall The ac-tivationmonitoring account provides a natural interpre-tation of these factors Any viable theory posited to ex-plain the occurrence of false memories in the DRMparadigm must explain the variability in false recall andfalse recognition across lists and the two primary factorsresponsible for this variability
REFERENCES
Anderson J R (1972) FRAN A simulation model of free recall InG H Bower (Ed) The psychology of learning and motivation (Vol 5pp 315-378) New York Academic Press
Anderson J R (1983) The architecture of cognition CambridgeMA Harvard University Press
Anderson J R amp Pirolli P L (1984) Spread of activation Journalof Experimental Psychology Learning Memory amp Cognition 10791-798
Balota D A Cortese M J Duchek J M Adams D RoedigerH L III McDermott K B amp Yerys B E (1999) Veridical andfalse memories in healthy older adults and in dementia of the Alz-heimers type Cognitive Neuropsychology 16 361-384
Balota D A Dolan P O amp Duchek J M (2000) Memorychanges in healthy older adults In E Tulving amp F I M Craik (Eds)The Oxford handbook of memory (pp 395-409) Oxford Oxford Uni-versity Press
Balota D A amp Duchek J M (1988) Age-related differences inlexical access spreading activation and simple pronunciation Psy-chology and Aging 3 84-93
Balota D A amp Duchek J (1991) Semantic priming effects lexicalrepetition effects and contextual disambiguation effects in healthyaged individuals and individuals with senile dementia of theAlzheimer type Brain amp Language 40 181-201
Bartlett F C (1932) Remembering A study in experimental and so-cial psychology Cambridge Cambridge University Press
Brainerd C J Wright R Reyna V F amp Mojardin A H (2001)Conjoint recognition and phantom recollection Journal of Experi-mental Psychology Learning Memory amp Cognition 27 307-327
Bruce D amp Winograd E (1998) Remembering Deesersquos 1959 arti-cles The Zeitgeist the sociology of science and false memoriesPsychonomic Bulletin amp Review 5 615-624
Bruner J S (1957) On perceptual readiness Psychological Review64 123-152
Budson A E Daffner K R Desikan R amp Schacter D L(2000) When false recognition is unopposed by true recognitionGist-based memory distortion in Alzheimerrsquos disease Neuropsy-chology 14 277-287
Collins A M amp Loftus E F (1975) A spreading-activation theoryof semantic memory Psychological Review 82 407-428
Coltheart M Davelaar E Jonasson J T amp Besner D (1977)Access to the internal lexicon In S Dormic (Ed) Attention and per-formance VI (pp 535-555) Hillsdale NJ Erlbaum
Deese J (1959a) Influence of interitem associative strength upon im-mediate free recall Psychological Reports 5 235-241
Deese J (1959b) On the prediction of occurrence of particular verbalintrusions in immediate recall Journal of Experimental Psychology58 17-22
Deese J (1961) Associative structure and the serial reproduction ex-periment Journal of Abnormal amp Social Psychology 63 95-100
Deese J (1965) The structure of associations in language and thoughtBaltimore Johns Hopkins University Press
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 397
Gallo D A McDermott K B Percer J M amp Roediger H LIII (2001) Modality effects in false recall and false recognitionJournal of Experimental Psychology Learning Memory amp Cogni-tion 27 339-353
Gallo D A Roberts M J amp Seamon J G (1997) Rememberingwords not presented in lists Can we avoid creating false memoriesPsychonomic Bulletin amp Review 4 271-276
Gallo D A amp Roediger H L III (2001) Variability among wordlists in evoking associative memory illusions Manuscript submittedfor publication
Gallo D A Roediger H L III amp McDermott K B (2001) As-sociative false recognition occurs without strategic criterion shiftsPsychonomic Bulletin amp Review 8 579-586
Hicks J L amp Marsh R L (2001) False recognition occurs more fre-quently during source identification than during oldndashnew recogni-tion Journal of Experimental Psychology Learning Memory ampCognition 27 375-383
Hunt R R amp Einstein G O (1981) Relational and item-specific in-formation in memory Journal of Verbal Learning amp Verbal Behav-ior 19 497-514
Hunt R R amp McDaniel M A (1993) The enigma of organizationand distinctiveness Journal of Memory amp Language 32 421-445
Israel L amp Schacter D L (1997) Pictorial encoding reduces falserecognition of semantic associates Psychonomic Bulletin amp Review4 577-581
Jacoby L L Kelley C M amp Dywan J (1989) Memory attribu-tions In H L Roediger III amp F I M Craik (Eds) Varieties of mem-ory and consciousness Essays in honour of Endel Tulving (pp 391-422) Hillsdale NJ Erlbaum
Johnson M K Hashtroudi S amp Lindsay D S (1993) Sourcemonitoring Psychological Bulletin 114 3-28
Johnson M K amp Raye C L (1981) Reality monitoring Psycho-logical Review 88 67-85
Kellogg RT (2001) Presentation modality and mode of recall in ver-bal false memory Journal of Experimental Psychology LearningMemory amp Cognition 27 913-919
Kirkpatrick E A (1894) An experimental study of memory Psy-chological Review 1 602-609
Ku Iumlcera H amp Francis W (1967) Computational analysis of present-day American English Providence RI Brown University Press
Marsh E J McDermott K B amp Roediger H L III (2001) Doretrieval factors play a role in the creation of of false memories Man-uscript submitted for publication
Mather M Henkel L A amp Johnson M J (1997) Evaluatingcharacteristics of false memories Rememberknow judgments andmemory characteristics questionnaire compared Memory amp Cogni-tion 25 826-837
McDermott K B (1996) The persistence of false memories in list re-call Journal of Memory amp Language 35 212-230
McDermott K B (1997) Priming on perceptual implicit memorytests can be achieved through presentation of associates Psycho-nomic Bulletin amp Review 4 582-586
McDermott K B amp Roediger H L III (1998) Attempting to avoidillusory memories Robust false recognition of associates persistsunder conditions of explicit warnings and immediate testing Journalof Memory amp Language 39 508-520
McDermott K B amp Watson J M (2001) The rise and fall of falserecall The impact of presentation duration Journal of Memory ampLanguage 45 160-176
McEvoy C L Nelson D L amp Komatsu T (1999) What is the con-nection between true and false memories The differential roles of in-teritem associations in recall and recognition Journal of Experimen-tal Psychology Learning Memory amp Cognition 25 1177-1194
Meyer D E amp Schvaneveldt R W (1971) Facilitation in recog-nizing pairs of words Evidence of a dependence between retrievaloperations Journal of Experimental Psychology 90 227-234
Miller M B amp Wolford G L (1999) Theoretical commentaryThe role of criterion shift in false memory Psychological Review106 398-405
Neely J H (1977) Semantic priming and retrieval from semanticmemory Roles of inhibitionless spreading activation and limited-
capacity attention Journal of Experimental Psychology General106 226-254
Nelson D L McEvoy C L amp Schreiber T A (1999) The Uni-versity of South Florida word association rhyme and word fragmentnorms Unpublished manuscript University of South Florida Tampa
Nelson D L Schreiber TA amp McEvoy C L (1992) Processing im-plicit and explicit representations Psychological Review 99 322-348
Norman K A amp Schacter D L (1997) False recognition in youngerand older adults Exploring the characteristics of illusory memoriesMemory amp Cognition 25 838-848
Paivio A Yuille J C amp Madigan S A (1968) Concreteness im-agery and meaningfulness values for 925 nouns Journal of Experi-mental Psychology Monographs 76 (1 Pt 2)
Payne D G Elie C J Blackwell J M amp Neuschatz J S (1996)Memory illusions Recalling recognizing and recollecting events thatnever occurred Journal of Memory amp Language 35 261-285
Reyna V F amp Brainerd C J (1995) Fuzzy-trace theory An interimsynthesis Learning amp Individual Differences 7 1-75
Roediger H L III Balota D A amp Watson J M (2001) Spreadingactivation and the arousal of false memories In H L Roediger III J SNairne I Neath amp A M Surprenant (Eds) The nature of remem-bering Essays in honor of Robert G Crowder (pp 95-115) Wash-ington DC American Psychological Association
Roediger H L III amp McDermott K B (1995) Creating false mem-ories Remembering words not presented in lists Journal of Experi-mental Psychology Learning Memory amp Cognition 21 803-814
Roediger H L III amp McDermott K B (1999) False alarms aboutfalse memories Psychological Review 106 406-410
Roediger H L III amp McDermott K B (2000) Tricks of memoryCurrent Directions in Psychological Science 9 123-127
Roediger H L III McDermott K B amp Pisoni D B (2001) Rec-ollection of illusory voices Manuscript submitted for publication
Roediger H L III McDermott K B amp Robinson K J (1998)The role of associative processes in creating false memories In M AConway S E Gathercole amp C Cornoldi (Eds) Theories of memoryII (pp 187-245) Hove UK Psychological Press
Russell W A amp Jenkins J J (1954) The complete Minnesotanorms for responses to 100 words from the KentndashRosanoff Word As-sociation Test (Tech Rep No 11 Contract N8 ONR 66216 Officeof Naval Research) University of Minnesota Minneapolis
Schacter D L Israel L amp Racine C (1999) Suppressing falserecognition in younger and older adults The distinctiveness heuris-tic Journal of Memory amp Language 40 1-24
Schacter D L Verfaellie M amp Pradere D (1996) The neuro-psychology of memory illusions False recall and recognition in am-nesic patients Journal of Memory amp Language 35 319-334
Seamon J G Luo C R amp Gallo D A (1998) Creating false mem-ories of words with or without recognition of list items Evidence fornonconscious processes Psychological Science 9 20-26
Smith R E amp Hunt R R (1998) Presentation modality affects falsememory Psychonomic Bulletin amp Review 5 710-715
Stadler M A Roediger H L III amp McDermott K B (1999)Norms for word lists that create false memories Memory amp Cogni-tion 27 494-500
Thapar A amp McDermott K B (2001) False recall and false recog-nition induced by presentation of associated words Effects of retentioninterval and level of processing Memory amp Cognition 29 424-432
Toglia M P amp Battig W F (1978) Handbook of semantic wordnorms Hillsdale NJ Erlbaum
Toglia M P Neuschatz J S amp Goodwin K A (1999) Recall ac-curacy and illusory memories When more is less Memory 7 233-256
Tulving E (1974) Cue-dependent forgetting American Scientist 6274-82
Tulving E (1979) Relation between encoding specificity and levelsof processing In L S Cermak amp F I M Craik (Eds) Levels of pro-cessing in human memory (pp 405-428) Hillsdale NJ Erlbaum
Tulving E (1985) Memory and consciousness Canadian Psycholo-gist 26 1-12
Tun P A Wingfield A Rosen M J amp Blanchard L (1998)Response latencies for false memories Gist-based processes in nor-mal aging Psychology amp Aging 13 230-241
398 ROEDIGER WATSON MCDERMOTT AND GALLO
Underwood B J (1965) False recognition produced by implicit ver-bal responses Journal of Experimental Psychology 70 122-129
Weldon M S (1991) Mechanisms underlying priming on perceptualtests Journal of Experimental Psychology Learning Memory ampCognition 17 526-541
Wickens T D amp Hirshman E (2000) False memories and statisticaldecision theory Comment on Miller and Wolford (1999) and Roedi-ger and McDermott (1999) Psychological Review 107 377-383
Wixted J T amp Stretch V (2000) The case against a criterion-shiftaccount of false memory Psychological Review 107 368-376
Zechmeister E B (1969) Orthographic distinctiveness Journal ofVerbal Learning amp Verbal Behavior 8 754-761
NOTES
1 The orthographic distinctiveness of a critical item was obtained bypresenting 72 subjects with all 55 critical items visually and havingthem rate the words in accordance with procedures developed by Zech-meister (1969) Briefly the subjects were instructed to rate each wordfor its physical appearance in terms of distinctiveness in the languageon a 9-point scale with 1 being not very distinctive 5 corresponding toaverage distinctiveness and 9 being most distinctive To avoid orderingeffects three different randomized orderings of words were used acrossroughly equal groups of subjects In the case of the critical item sleep
for example the distinctiveness is 515 Corresponding values for but-terfly and for man are 744 and 213 respectively Although ortho-graphic distinctiveness was not included in our multiple regressionanalyses this predictor is included in Appendix C
2 The orthographic neighborhood size or Coltheartrsquos N of each crit-ical item was defined as the number of words that can be formed bychanging a single letter in the word (Coltheart Davelaar Jonasson ampBesner 1977) For example because it has no orthographic neighborsColtheartrsquos N for the critical item butterfly is zero In contrast sleep hasfive orthographic neighbors (ie sheep sleek sleet steep and sweep)The reference set of words used were those in Ku Iumlcera and Francisrsquos(1967) norms Although Coltheartrsquos N was not included in our multipleregression analyses this predictor is included in Appendix C
3 Word length orthographic distinctiveness and orthographic neigh-borhood size of the critical item were all highly correlated Specificallythe bivariate correlations among these three predictors were as followslength and orthographic distinctiveness r 5 184 t(53) 5 1117 p lt05 orthographic distinctiveness and orthographic neighborhood sizer 5 66 t(53) 5 640 p lt 05 length and orthographic neighbor-hood size r 5 63 t(53) 5 582 p lt 05 Of these three variablesword length had the largest bivariate correlation with false recall (r 5
37) so it was the predictor used in the multiple regression analyses4 The multiple regression data matrix is available at wwwiacwustl
edunmclWeb
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 399
APPENDIX A The Fifty-Five 15-Word Lists and Their Associated Critical Items
With Backward Associative Strength (BAS) and Forward Associative Strength (FAS)Values Used in the Multiple Regression Analysis of False Recall
Item Word BAS FAS Item Word BAS FAS
Anger mad 393 412 Black white 655 557fear 020 059 dark 111 100hate 028 109 cat 000 043rage 541 042 charred 023 000temper 182 000 night 000 021fury 306 000 funeral 034 000ire 179 000 color 074 050wrath 128 000 grief 000 000happy 000 042 blue 028 000fight 034 000 death 016 014hatred 070 000 ink 020 000mean 090 000 bottom 000 000calm 000 000 coal 288 000emotion 000 000 brown 338 000enrage 378 000 gray 365 000
Mean 157 044 Mean 130 052Army Navy 543 500 Bread butter 364 487
NotemdashUnless otherwise indicated with an asterisk all BAS and FAS values are drawn from the associative norms of Nel-son McEvoy and Schreiber (1999) Those values shown with an asterisk were obtained by the authors by using normingprocedures similar to those of Nelson et al (1999) The corresponding mean for these BAS and FAS values for each list arealso represented
406 ROEDIGER WATSON MCDERMOTT AND GALLO
APPENDIX B Data Matrix of the Fifty-Five 15-Word Lists and Associated Critical Items
Used in the Multiple Regression Analysis of False Recall
Critical False False Log Vrd Raw Ortho Item Recall Recgn Length Freq Concrete FAS BAS Connect Recall Freq Dist Colt N
(Manuscript received July 10 2000revision accepted for publication February 23 2001)
390 ROEDIGER WATSON MCDERMOTT AND GALLO
cases) among the seven predictors and false recall isshown in Table 1
Bivariate CorrelationsIn describing the results we report significant bivari-
ate correlations between critical item variables and falserecall followed by significant correlations between listvariables and false recall To determine these values eachof the critical item and list variables were entered into aregression equation as the sole predictor of false recallOnly three of our seven variables were significantly cor-related ( p lt 05) with false recall the critical item vari-able of length and the list variables of BAS and recall ofthe studied items (Figures 1A and 1B respectively)
Length The length of the critical item and false recallwere negatively correlated [r 5 37 t(53) 5 287]
Backward associative strength Mean BAS or thestrength of the connections between study items andtheir corresponding critical items was positively corre-lated with false recall r 5 173 t(53) 5 773 As isshown in Figure 1A the greater the strength of associa-tion between the list items and the critical item the morelikely is false recall This outcome replicates Deesersquos(1959b) finding of a 187 correlation between BAS andfalse recall which he obtained with a set of thirty-six 12-item lists Although the 173 correlation between thesetwo variables in our set of f ifty-f ive 15-item lists issomewhat smaller than the 187 correlation obtained byDeese BAS is still the strongest correlate of false recallin our entire data set
Veridical recall Recall of list items and false recallwere negatively correlated [r 5 43 t(53) 5 346]As is shown in Figure 1B increasing levels of accuraterecall are associated with decreasing levels of false re-call This outcome is contrary to some ideas about whyfalse recall occurs as will be brought out in the Discus-sion section
As is shown in Table 1 no other variables were corre-lated with false recall at the bivariate level which is sur-prising in some cases For example in terms of criticalitem variables one might have expected word frequencyor concreteness to be correlated with false recall but nei-ther did (Deese 1959b also failed to find a relation be-tween word frequency and false recall) In terms of listvariables one might have expected connectivity to cor-relate with false recall but it did not The lack of a rela-
tion between connectivity and false recall is noteworthybecause Deese (1959a) and McEvoy et al (1999) re-ported a significant relationship between these variablesso that lists with greater connectivity were less likely toelicit false recall of the critical item In the Discussionsection we consider connectivity or interitem associa-tive strength more fully We turn now to the results of thesimultaneous multiple regression analysis
Simultaneous Multiple Regression AnalysisIn the multiple regression analysis the three variables
based on the characteristics of the critical items them-selves (ie length log frequency and concreteness) andthe four variables based on the characteristics of the listsdesigned to elicit the critical items (ie FAS and BASmeasures interitem associative strength or connectivityand the level of recall of list items) were entered simul-taneously into the regression equation to predict false re-call The results of the analysis are shown in Table 2 in-cluding beta weights (b) semipartial correlations (sr)and significance tests for each predictor For signifi-cance tests a t(47) was used for each predictor and theresultant t statistic is expressed as the absolute valuemdashthat is frac12tfrac12 In addition Table 2 displays the results of thesignificance test for the multiple R2 for the entire set ofseven predictor variables
As is shown in Table 2 the results of the regressionanalysis indicated that two list variables were signifi-cantly related to false recall Specifically BAS was thestrongest predictor of false recall (b 5 170 sr 5160 t 5 717) Therefore consistent with our findingsat the bivariate level the greater the strength of associa-tion between study items and their corresponding criti-
Table 1 Correlation Matrix Among Eight Variables False Recall Three Critical Item Variables and Four List Variables
False Log VeridicalRecall Length Frequency Concreteness FAS BAS Connectivity Recall
NotemdashR2 5 68 F(747) 5 1406 MSe 5 0013 p lt 05 p lt 05
Figure 1 Probability of false recall of the critical item from 55 lists (A) as a functionof associative strength of the list items to the critical item or backward associativestrength and (B) as a function of veridical recall of studied items from the lists
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 391
cal item on a free association test the more probable isfalse recall of the critical item Veridical recall was alsoa strong predictor of false recall (b 5 40 sr 5 34t 5 411) As we found at the bivariate level the morewords recalled from a list the less likely that the criticalitem for that list was falsely recalled None of the re-maining critical item or list variables including length ofthe critical item connectivity and FAS explained aunique portion of the total variance in false recall afterthe influence of other predictors had been residualized inthe context of the regression equation Therefore BASand recall of list items appeared to be the major contrib-utors to the multiple R of 182 and the associated mul-tiple R2 of 68 [F(747) 5 1406 MSe 5 0013] We fur-ther explore the implications of these effects in theDiscussion section
Factors Affecting False RecognitionIn our analyses thus far we have concentrated on
false recall exclusively However the norms for our 55lists (collected by Stadler et al 1999 and Gallo ampRoediger 2001) also include values for false recogni-tion The recognition data in these norms were collectedin a final recognition test after lists had been studied andrecalled This confounding of recognition by prior recallalmost certainly affected both veridical and false recog-nition However in experiments directly examining theeffects of recall on recognition in this paradigm and oth-ers the finding is that recall of a list may increase itsrecognition by a moderate amount (typically in the4ndash10 range eg Roediger amp McDermott 1995Roediger McDermott amp Pisoni 2001) In fact some re-searchers have failed to observe any effect at all of recall
392 ROEDIGER WATSON MCDERMOTT AND GALLO
on recognition (Schacter Verfaellie amp Pradere 1996 seeRoediger et al 1998 for a discussion) We conducted thesame regression analyses with our seven predictor vari-ables on false recognition as we did on false recall Wesummarize the basic results and conclusions as follows
There were three significant bivariate correlationswith false recognition including length (r 5 27)BAS (r 5 143) and veridical recall (r 5 52) Over-all these correlations with false recognition were simi-lar to those obtained with false recall However the pos-itive correlation between BAS and false recognition wassmaller than that in the recall data and the negative cor-relation between veridical recall and false recognitionwas larger than that in the recall data To maintain con-sistency with the previous regression analysis on falserecall we used veridical recall rather than veridicalrecognition in this analysis In addition recognition es-timates were based on only three itemslist and wereprobably affected by prior recall
Turning to the results of the simultaneous multiple re-gression analysis on false recognition BAS and veridi-cal recall of list items appeared to be the major contrib-utors to the multiple R of 169 and the associatedmultiple R2 of 48 [F(747) 5 611 MSe 5 002] Table 3shows the summary of the regression analyses Althoughthe regression analysis accounted for less total variancein false recognition than in false recall (48 as opposedto 68 respectively) the same factors seem to be re-sponsible for predicting false recognition and false re-call Of course because measures of recognition alwaysfollowed recall this similar pattern may reflect a carry-over effect of false recall on false recognition Nonethe-less within the limits of this study the factors responsi-ble for false recognition across lists seem to be the sameas those producing false recall
DISCUSSION
To summarize the main findings our results implicatetwo primary factors (of the seven variables we exam-ined) as accounting for most of the variance in false re-call across lists The split-half correlation of our lists was190 so the amount of potentially explainable variancein our dependent variable of false recall is 81 The fac-
tors in our multiple regression accounted for 68 of thetotal variance in false recall or 84 of the explainablevariance (68 4 81 3 100 5 84) The analysis forfalse recognition revealed the same factors at work al-though they accounted for less overall variance Thestrongest factor was BAS consistent with prior work byDeese (1959b) and McEvoy et al (1999) The other fac-tor was the number of items recalled accurately from thelist which is negatively correlated with false recall
We turn next to the theoretical implications of ourfindings We first consider the implications for the acti-vationmonitoring framework that we have developed toexplain false recall and false recognition in the DRMparadigm We then consider the implications for othertheories and f inally turn to several further issues onwhich our results shed light
ActivationMonitoring TheoryIn the theroretical framework that we (eg McDer-
mott amp Watson 2001 Roediger Balota amp Watson 2001Roediger amp McDermott 1995 2000) have used to explainvariations in the probability of false recall and false recog-nition in the DRM paradigm two sets of processes havebeen proposed Specifically we have proposed that pro-cesses occurring during both encoding and retrieval af-fect the probability of false recall and false recognitionin the DRM paradigm Two critical sets of processes in-volve activation and monitoring Although there is a nat-ural alignment of activation to encoding and monitoringto retrieval we note that both processes can potentiallyoccur both during encoding and during retrieval as willbe discussed below The ideas we employ have many pre-cursors which we acknowledge below although theirparticular arrangement here may be unique
A critical distinction in understanding encoding pro-cesses and their interaction with retrieval processes isbetween item-specific and relational processing (Ander-son 1972 Hunt amp Einstein 1981 Hunt amp McDaniel1993) Briefly the notion is that emphasis can be given tothe encoding of individual elements and their features(what Anderson called node-taggingand Hunt and his col-leagues called item-specific processing) or to the encod-ing of relationships among elements (pathway-taggingor relational processing) Item-specific processing should
Table 3 Summary of Multiple Regression Analysis
on False Recognition With Critical Item Variablesand List Variables Used as Predictors
NotemdashR2 5 48 F(747) 5 611 MSe 5 002 p lt 05 p lt 05
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 393
make individual elements more memorable and lead tomore distinctive memories such processing should helpdistinguish veridical memories from false memories be-cause list items would have features of presentation as-sociated with them whereas the critical item sometimeswould not (However if the critical item is strongly acti-vated during encoding it may take on many of the fea-tures of list items) Relational processing (extractingthemes or schemas or constructing networks of associa-tions) should increase the probability of false recall andfalse recognition by leading to inferences or associationsof events that may not actually have occurred
We assume that in encoding and interpreting experi-ences people go beyond the information given in thestimulus (Bartlett 1932 chap 2 Bruner 1957) and ac-tivate related information through inferential processingEach personrsquos experience is coded in terms of his or herown schemata (or associative networks or knowledgestructures) When schemata are activated a process offilling in of ldquomissingrdquo information may occur wherebyitems strongly implied by the schema (but not actuallypresent in the stimulus) may become activated (Bartlett1932) In encoding lists of words in the DRM paradigmpresentation of the list can create activation that spreadsthroughout the lexicalsemantic system and can createimplicit associative responses (Underwood 1965) Thisactivation may arouse concepts that are not presented but that are associated to concepts that are presented(eg Anderson amp Pirolli 1984 Collins amp Loftus 1975Meyer amp Schvaneveldt 1971) The associative activa-tion in this approach can arise from automatic process-ing or from more controlled elaborative processing(Neely 1977) In addition the associative activation ofconcepts that were not presented may occur consciously(subjects may become aware of the concept of sleep andrehearse it during presentation of the list words bed restawake etc) or the concept may be strongly activatedbut never come to conscious awareness The debate overwhether arousal of the critical item in the DRM para-digm is conscious or unconscious is interesting but dif-ficult to decide empirically Some evidence has impli-cated conscious arousal of the concepts (McDermott1997) whereas other evidence implicates unconsciousactivation (Seamon Luo amp Gallo 1998) The correctanswer to the question of whether activation is consciousor unconscious is probably ldquobothrdquomdashthat is the criticalitem probably sometimes comes consciously to mindand even if it does not relevant information may never-theless be activated unconsciously
The high correlation between BAS and false recall aswell as the strong weighting of BAS in the multiple re-gression analysis strongly implicates activation of criti-cal items as an important factor in explaining false recalland false recognition in the DRM paradigm The morestrongly associated list items are to critical items themore activation would be expected to spread to the crit-ical items More colloquially to the extent that the wordson the list spark the associative connection to the critical
item the more likely the critical item is to be falsely re-called This activation factor would seem to be one mainreason why ldquothings suggested to a person by an experi-ence may be honestly reported by him as part of the ex-periencerdquo (Kirkpatrick 1894 p 608) These associatedlists probably all tend to elicit relational processing butonly those with strong BAS to the critical item lead tofalse recall
The second set of factors proposed in activationmon-itoring theory is concerned with the monitoring of mem-ory accuracy Monitoring of encoding processes canoccur during the study phase of an experiment espe-cially under intentional learning conditions In experi-ments in which subjects are warned about false memoryphenomena and are told to pay close attention to lists andto remember only the presented items and not the criti-cal items monitoring is enhanced (Gallo Roberts amp Sea-mon 1997 Gallo Roediger amp McDermott 2001 Mc-Dermott amp Roediger 1998) Monitoring of memorieshas typically been discussed with respect to retrieval dur-ing which time a primary goal is to distinguish fromamong the considerable information brought to con-sciousness that which refers to perception of past eventsfrom that which does not a process referred to as realitymonitoring (Johnson amp Raye 1981) The experience ofremembering events involves mentally traveling back intime and seeming to reexperience the events (Tulving1985) During retrieval recovered memories that appar-ently carry with them details of having been experiencedare generally attributed to be memories of actual pastevents although sometimes this attribution may be inerror (Jacoby et al 1989)
Johnson et al (1993) have described in detail the typesof information that may be used to distinguish eventsthat actually happened from ones that were only imag-ined thought about or read about For example eventsthat actually occurred have more features from the ex-ternal world associated with them (sights sounds etc)whereas those that were internally generated are lesslikely to have these features but more likely to have as-sociated reflections and cognitive operations In theDRM paradigm the fact that the critical item has beenstrongly activated during encoding means that duringretrieval the subject is likely to retrieve the critical itemas a list member To the extent that the critical items arestrongly activated during encoding they will take onsome features similar to those of the list items this is es-pecially true if the critical items come consciously tomind during encoding and are rehearsed in the samemanner as list items To the extent that the features of thecritical item are similar to those of actual list items theprobability of false recall or false recognition is increasedHowever if the list items systematically differ from thecritical item on some dimensions or features retrievalheuristics may be brought into play that can reduce falserecall and false recognition (Israel amp Schacter 1997Schacter Israel amp Racine 1999) Considerable evidencesupports this general source-reality-monitoring frame-
394 ROEDIGER WATSON MCDERMOTT AND GALLO
work that we advocate here (eg Johnson et al 1993Schacter et al 1999) as well as the attributional ap-proach to remembering (Jacoby et al 1989)
The other significant factor arising from our multipleregression analysis may operate during retrieval by aid-ing the monitoring processes that reduce false recall Thenegative correlation between veridical recall of list itemsand false recall of the critical item may indicate that thebetter encoded list items are the more easily they can bedistinguished from the illusory critical item Subjectsmay use item-specific information to aid recall of listitems and suppress recall of the critical items whichwould not carry as much item-specific information Whenexperimental manipulations increase veridical recall ei-ther through longer study time at relatively slow rates(Gallo amp Roediger 2001 McDermott amp Watson 2001Toglia Neuschatz amp Goodwin 1999) or through re-peated study and test opportunities (McDermott 1996Schacter et al 1999) false recall is often decreasedTherefore the negative relation between veridical recalland false recall can be produced experimentally as wellas appearing in our correlationregression analysis How-ever veridical and false recall can be positively related as afunction of other variables too as was shown by McDer-mott (1996) Toglia et al (1999) and Thapar amp McDermott(2001) In these cases greater activation of the criticalitem from relational processing or from greater seman-tic processing is probably at work
The fact that the length andor distinctiveness of thecritical item are negatively correlated with false recallcan be similarly interpreted as the effect of monitoringduring retrieval because the greater distinctiveness ofthe critical item simplifies the monitoring process Forexample if words in the list are generally shorter andhave more regular orthography than does the criticalitem then during retrieval the critical items that are longand distinctive (eg butterfly cabbage) should be moreeasily discriminated from list items as indeed seems tobe the case in the bivariate correlations Interestingly thelengthdistinctiveness factor played a role in determin-ing false recall even though the lists in our norms werepresented auditorily and the orthographic informationat least should not be particularly accessible In addi-tion presenting the DRM lists visually rather than audi-torily reduces false recall and false recognition at leaston tests in which recognition items are presented visually(as opposed to auditorily) or recall is written rather thanspoken (Gallo McDermott et al 2001 Kellogg 2001Smith amp Hunt 1998) This modality difference has alsobeen interpreted as arising from the retrieval of distinctiveinformation operating during the test (Gallo McDermottet al 2001 Smith amp Hunt 1998)
As was noted above activation processes may occurduring retrieval as well as during encoding The morelist items a subject recalls the more primed the criticalitem might become However the present analyses un-dercut the simplest form of this idea If the recalling oflist items activated or primed false recall the simplest
prediction would be that there should be a positive cor-relation between veridical recall and false recall Our datarevealed the opposite pattern showing a negative corre-lation between veridical and false recall Marsh McDer-mott and Roediger (2001) reported other experimentsthat also call into question the idea that priming duringthe test plays a role in false recall and false recognition
Balota et al (1999) obtained further evidence sup-porting the activationmonitoring interpretation of falsememories in the DRM paradigm They compared per-formance of younger adults healthy older adults andolder adults with Alzheimerrsquos disease The latter twogroups have been shown to have intact activation pro-cesses as measured by semantic priming paradigms(Balota amp Duchek 1988 1991) yet they are deficient insource-monitoring processes which is probably attribut-able to impaired frontal lobe functioning (see BalotaDolan amp Duchek 2000) Therefore one might expectthat whereas veridical recall would be worse in older adultsand Alzheimerrsquos patients (relative to younger adults)false recall and false recognition would occur at the samelevel or perhaps be increased in these groups IndeedBalota et al (1999) reported exactly this pattern A sim-ilar outcome has been obtained in other laboratories(eg Budsen Daffner Desikan amp Schacter 2000 Nor-man amp Schacter 1997 Tun Wingfield Rosen amp Blan-chard 1998) This pattern is just as predicted by the ac-tivation monitoring framework
In short the present results and many others (seeRoediger Balota amp Watson 2001 Roediger et al 1998)fit comfortably within the activationmonitoring frame-work In the remainder of our discussion we will con-sider briefly the implications of our present results forother accounts of the DRM paradigm and for the role ofconnectivity (or interitem associative strength) in pro-ducing veridical and false recall
Fuzzy-Trace TheoryAnother leading theory that has been used to explain
false memory phenomena is fuzzy-trace theory (egReyna amp Brainerd 1995) Payne et al (1996) and Schac-ter et al (1996) among many others have used the basictheory to account for false recall and false recognitiondata within the DRM paradigm and the theory has beenextended in interesting ways to account for new varia-tions on the paradigm (eg Brainerd Wright Reyna ampMojardin 2001) The fundamental assumption of fuzzy-trace theory is that experience leaves two types of mem-ory traces gist traces (which capture the meaning of ex-perience but without specific attributes) and verbatimtraces (which represent specific attributes) False memo-ries are attributed to gist processing When the meaning ofa recognition probe matches the gist of the encoded expe-rience false recognition occurs The theory has been di-rected primarily to false recognition and not false recall
In trying to explain the present results one difficultyis that the operational definition of gist has never beenspecified that is within the DRM paradigm given a list
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 395
of 15 related words how does one specify the gist repre-sentation of the list and quantify how strong it is Onecould use patterns of false recognitionmdashto the extentthat a concept is falsely recognized it was represented aspart of the gist representation This definition is per-fectly suitable for some purposes but clearly explainingfalse recognition is not one of them owing to the prob-lem of circularity To explain why some lists lead to highlevels of false recall and false recognition (whereas otherlists lead to low levels) within fuzzy-trace theory oneneeds an independent specification of gist which to ourknowledge has not been provided in this context Whydoes the sweet list have so much gist that it leads to falserecall whereas the bitter list does not One plausible in-terpretation of gist is BAS The more the list items areassociated to the critical item the more a gist represen-tation of the critical item is created This solution to adefinition of gist works within the DRM paradigm butof course the interpretation of how gist is created be-comes isomorphic to activation processes within the ac-tivationmonitoring framework
Interpretation of the other factor that was significantwithin our multiple regression analysis may be morestraightforward within fuzzy-trace theory The negativecorrelation between veridical recall and false recall (andfalse recognition) may be accounted for by increased re-liance on verbatim traces That is if increased recall of listitems is due to increased verbatim information subjectsmay rely less heavily on gist information and hence showless false recall and false recognition during respondingSimilarly for critical items that are long andor ortho-graphically distinct recall of list items that relies on thesefeatures may lead to suppression of false recall and falserecognition of the critical items because they may differsystematically from other list items These factors pos-tulated within the context of fuzzy-trace theory are quitesimilar to those within the activationmonitoring frame-work outlined above At some level of generality there aretwo sets of factors in each account and both theories canexplain many empirical results
Shifting Criterion TheoryMiller and Wolford (1999) proposed that false recog-
nition in the DRM paradigm could be explained by sub-jectsrsquo shifting their criteria differentially to list items andto critical items during a recognition test (They ex-tended their explanation to false recall but somewhattenuously by proposing a generaterecognize theory)They postulated that subjects encoded the structure ofthe list during its presentation and then during the recog-nition test responded with especially liberal criteria tocritical items but not to list items The criterion shift issupposed to occur to the degree that the test item appearsto be related to the list items To support their argumentthey reported that a measure of bias from signal detec-tion theory was indeed more liberal for critical itemsthan for list items However Wickens and Hirshman(2000) and Wixted and Stretch (2000) pointed out that
within signal detection theory differences in bias mea-sures do not necessarily implicate a criterion shiftRather differences in measured bias could arise fromdifferences among the underlying distributions Wixtedand Stretch argued that this latter interpretation was themore likely and that the criterion shift interpretation wasuntenable (also see Roediger amp McDermott 1999)
The present evidence also weighs heavily against acriterion shift account Miller and Wolford (1999) ap-pealed to subjectsrsquo metaknowledge of the structure of thelist Within the criterion shift account we can assumethat the likelihood that a critical item cues metaknowl-edge of the list depends on its FAS or how well the crit-ical item will cue the list If so FAS (the association ofthe critical item to the list items) should be the primarydeterminant of false recognition and false recall How-ever the correlation between FAS and false recall in ourdata was negligible (08) and the same was true in analy-ses of false recognition BAS was actually the primarypredictor of false recall which implicates activation pro-cesses during encoding as being a primary cause of theeffect There would be no reason to expect BAS to haveany impact according to the idea that a simple criterionshift during the test causes false recall and false recog-nition Of course all our lists were relatively high androughly equivalent in terms of FAS because they werecreated from forward associations to the critical item sorestriction of range may account for the low correlationEven so the same problem exists for the shifting crite-rion theory to explain the results If FAS is supposed todetermine false recall and false recognition by this ac-count and if all our lists were high and roughly equal onthis feature why were so many lists ineffective in elicitingfalse recognition and false recall In short either way theshifting criterion theory cannot account for the results
Of course within the activationmonitoring frame-work the monitoring processes occurring during retrievalcould be conceived as conservative shifts in response cri-teria However in our theory these monitoring processesdetermine the degree to which a given level of activationat retrieval will be judged sufficient to endorse a criticalitem as having been studied thereby producing a falsememory In contrast in Miller and Wolfordrsquos (1999) ac-count there was no activation of the critical item duringencoding and shifting criteria during the test were saidto account entirely for the false recognition phenomenon(see Wixted amp Stretch 2000) In the activationmonitoringaccount the memory illusion provoked by the DRM listsis assumed to be a subjectively powerful phenomenonthat can be ameliorated (but rarely eliminated) when listitems are made more distinctive relative to criticalitems Under conditions in which the subject is warnedat test (but not before the encoding phase) to monitortheir memories for the presence or absence of the criti-cal item and is discouraged from making criterion shiftsthe false memory effect is neither eliminated nor greatlyreduced (Gallo Roediger amp McDermott 2001) Hicksand Marsh (2001) showed that under conditions in which
396 ROEDIGER WATSON MCDERMOTT AND GALLO
subjects are explicitly told to monitor their memories dur-ing the test (and are asked to judge whether each wordhad been externally presented or internally generated)false recognition in the DRM paradigm actually increasesrather than decreases In fact the false recognition effectpersists even when subjects are fully informed about thenature of the phenomenon before the study episode andare therefore encouraged to enhance monitoring processesduring both encoding and retrieval in order to attempt todiminish or eliminate the effect (see Gallo et al1997Gallo Roediger amp McDermott 2001 McDermott ampRoediger 1998) For these and other reasons the notionthat the DRM memory illusion is caused by a general shiftof criteria can be eliminated as a potential explanation
Interitem Associative Strength or ConnectivityOur data pose something of a puzzle about the role of
interitem associative strength (Deese 1959a) or connec-tivity (McEvoy et al 1999) in false recall As was dis-cussed in the introduction Deese (1959a) found a sig-nificant negative correlation between this construct andthe probability of false recall In a later paper Deese(1961) obtained a weaker negative correlation that failedto reach conventional levels of significance so the issueis somewhat in doubt from Deesersquos (1959a 1961) researchMcEvoy et al created lists that varied in connectivity(interitem associative strength) and in BAS and manip-ulated these factors orthogonally They replicated thefinding of Deese (1959a) in that the density of inter-connections of list words was negatively related to falserecall However in our analysis of 55 lists (different fromthose used by McEvoy et al 1999) we found no relationbetween connectivity and false recall (r 5 04) or be-tween connectivity and false recognition (r 5 03)
Further research will be necessary to determine thetrue state of affairs although a case can be made that thelack of correlation we obtained is accurate First we had55 lists that varied widely in their ability to elicit false re-call as compared with 24 lists used in McEvoy et al(1999) Our lists also represented a relatively full rangeof connectivity or interitem associative strength relativeto those of McEvoy et al McEvoy et al used lists thathad a somewhat wider range of connectivity because theyused extreme values whereas our larger set of lists pro-vided a fuller body of the middle ranges of connectivityGiven our results it seems at least possible that interitemassociative strength (connectivity) is not a factor in de-termining whether lists of associative words lead to falserecall although the issue remains in doubt
We did find that interitem associative strength or con-nectivity is positively and significantly correlated withveridical recall (r 5 132) This outcome is consistentwith that of Deese (1959a) and McEvoy et al (1999)The greater the associative bonds among list items thegreater is list recall This outcome is also consistent withvirtually any associative theory of recall (eg Anderson1972 1983 Nelson Schreiber amp McEvoy 1992)
CONCLUSION
The present study has uncovered two primary factorsthat are significantly related to false recall and falserecognition in the DRM paradigm BAS is positivelycorrelated with false recall whereas veridical recall ofthe list items is negatively related to false recall The ac-tivationmonitoring account provides a natural interpre-tation of these factors Any viable theory posited to ex-plain the occurrence of false memories in the DRMparadigm must explain the variability in false recall andfalse recognition across lists and the two primary factorsresponsible for this variability
REFERENCES
Anderson J R (1972) FRAN A simulation model of free recall InG H Bower (Ed) The psychology of learning and motivation (Vol 5pp 315-378) New York Academic Press
Anderson J R (1983) The architecture of cognition CambridgeMA Harvard University Press
Anderson J R amp Pirolli P L (1984) Spread of activation Journalof Experimental Psychology Learning Memory amp Cognition 10791-798
Balota D A Cortese M J Duchek J M Adams D RoedigerH L III McDermott K B amp Yerys B E (1999) Veridical andfalse memories in healthy older adults and in dementia of the Alz-heimers type Cognitive Neuropsychology 16 361-384
Balota D A Dolan P O amp Duchek J M (2000) Memorychanges in healthy older adults In E Tulving amp F I M Craik (Eds)The Oxford handbook of memory (pp 395-409) Oxford Oxford Uni-versity Press
Balota D A amp Duchek J M (1988) Age-related differences inlexical access spreading activation and simple pronunciation Psy-chology and Aging 3 84-93
Balota D A amp Duchek J (1991) Semantic priming effects lexicalrepetition effects and contextual disambiguation effects in healthyaged individuals and individuals with senile dementia of theAlzheimer type Brain amp Language 40 181-201
Bartlett F C (1932) Remembering A study in experimental and so-cial psychology Cambridge Cambridge University Press
Brainerd C J Wright R Reyna V F amp Mojardin A H (2001)Conjoint recognition and phantom recollection Journal of Experi-mental Psychology Learning Memory amp Cognition 27 307-327
Bruce D amp Winograd E (1998) Remembering Deesersquos 1959 arti-cles The Zeitgeist the sociology of science and false memoriesPsychonomic Bulletin amp Review 5 615-624
Bruner J S (1957) On perceptual readiness Psychological Review64 123-152
Budson A E Daffner K R Desikan R amp Schacter D L(2000) When false recognition is unopposed by true recognitionGist-based memory distortion in Alzheimerrsquos disease Neuropsy-chology 14 277-287
Collins A M amp Loftus E F (1975) A spreading-activation theoryof semantic memory Psychological Review 82 407-428
Coltheart M Davelaar E Jonasson J T amp Besner D (1977)Access to the internal lexicon In S Dormic (Ed) Attention and per-formance VI (pp 535-555) Hillsdale NJ Erlbaum
Deese J (1959a) Influence of interitem associative strength upon im-mediate free recall Psychological Reports 5 235-241
Deese J (1959b) On the prediction of occurrence of particular verbalintrusions in immediate recall Journal of Experimental Psychology58 17-22
Deese J (1961) Associative structure and the serial reproduction ex-periment Journal of Abnormal amp Social Psychology 63 95-100
Deese J (1965) The structure of associations in language and thoughtBaltimore Johns Hopkins University Press
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 397
Gallo D A McDermott K B Percer J M amp Roediger H LIII (2001) Modality effects in false recall and false recognitionJournal of Experimental Psychology Learning Memory amp Cogni-tion 27 339-353
Gallo D A Roberts M J amp Seamon J G (1997) Rememberingwords not presented in lists Can we avoid creating false memoriesPsychonomic Bulletin amp Review 4 271-276
Gallo D A amp Roediger H L III (2001) Variability among wordlists in evoking associative memory illusions Manuscript submittedfor publication
Gallo D A Roediger H L III amp McDermott K B (2001) As-sociative false recognition occurs without strategic criterion shiftsPsychonomic Bulletin amp Review 8 579-586
Hicks J L amp Marsh R L (2001) False recognition occurs more fre-quently during source identification than during oldndashnew recogni-tion Journal of Experimental Psychology Learning Memory ampCognition 27 375-383
Hunt R R amp Einstein G O (1981) Relational and item-specific in-formation in memory Journal of Verbal Learning amp Verbal Behav-ior 19 497-514
Hunt R R amp McDaniel M A (1993) The enigma of organizationand distinctiveness Journal of Memory amp Language 32 421-445
Israel L amp Schacter D L (1997) Pictorial encoding reduces falserecognition of semantic associates Psychonomic Bulletin amp Review4 577-581
Jacoby L L Kelley C M amp Dywan J (1989) Memory attribu-tions In H L Roediger III amp F I M Craik (Eds) Varieties of mem-ory and consciousness Essays in honour of Endel Tulving (pp 391-422) Hillsdale NJ Erlbaum
Johnson M K Hashtroudi S amp Lindsay D S (1993) Sourcemonitoring Psychological Bulletin 114 3-28
Johnson M K amp Raye C L (1981) Reality monitoring Psycho-logical Review 88 67-85
Kellogg RT (2001) Presentation modality and mode of recall in ver-bal false memory Journal of Experimental Psychology LearningMemory amp Cognition 27 913-919
Kirkpatrick E A (1894) An experimental study of memory Psy-chological Review 1 602-609
Ku Iumlcera H amp Francis W (1967) Computational analysis of present-day American English Providence RI Brown University Press
Marsh E J McDermott K B amp Roediger H L III (2001) Doretrieval factors play a role in the creation of of false memories Man-uscript submitted for publication
Mather M Henkel L A amp Johnson M J (1997) Evaluatingcharacteristics of false memories Rememberknow judgments andmemory characteristics questionnaire compared Memory amp Cogni-tion 25 826-837
McDermott K B (1996) The persistence of false memories in list re-call Journal of Memory amp Language 35 212-230
McDermott K B (1997) Priming on perceptual implicit memorytests can be achieved through presentation of associates Psycho-nomic Bulletin amp Review 4 582-586
McDermott K B amp Roediger H L III (1998) Attempting to avoidillusory memories Robust false recognition of associates persistsunder conditions of explicit warnings and immediate testing Journalof Memory amp Language 39 508-520
McDermott K B amp Watson J M (2001) The rise and fall of falserecall The impact of presentation duration Journal of Memory ampLanguage 45 160-176
McEvoy C L Nelson D L amp Komatsu T (1999) What is the con-nection between true and false memories The differential roles of in-teritem associations in recall and recognition Journal of Experimen-tal Psychology Learning Memory amp Cognition 25 1177-1194
Meyer D E amp Schvaneveldt R W (1971) Facilitation in recog-nizing pairs of words Evidence of a dependence between retrievaloperations Journal of Experimental Psychology 90 227-234
Miller M B amp Wolford G L (1999) Theoretical commentaryThe role of criterion shift in false memory Psychological Review106 398-405
Neely J H (1977) Semantic priming and retrieval from semanticmemory Roles of inhibitionless spreading activation and limited-
capacity attention Journal of Experimental Psychology General106 226-254
Nelson D L McEvoy C L amp Schreiber T A (1999) The Uni-versity of South Florida word association rhyme and word fragmentnorms Unpublished manuscript University of South Florida Tampa
Nelson D L Schreiber TA amp McEvoy C L (1992) Processing im-plicit and explicit representations Psychological Review 99 322-348
Norman K A amp Schacter D L (1997) False recognition in youngerand older adults Exploring the characteristics of illusory memoriesMemory amp Cognition 25 838-848
Paivio A Yuille J C amp Madigan S A (1968) Concreteness im-agery and meaningfulness values for 925 nouns Journal of Experi-mental Psychology Monographs 76 (1 Pt 2)
Payne D G Elie C J Blackwell J M amp Neuschatz J S (1996)Memory illusions Recalling recognizing and recollecting events thatnever occurred Journal of Memory amp Language 35 261-285
Reyna V F amp Brainerd C J (1995) Fuzzy-trace theory An interimsynthesis Learning amp Individual Differences 7 1-75
Roediger H L III Balota D A amp Watson J M (2001) Spreadingactivation and the arousal of false memories In H L Roediger III J SNairne I Neath amp A M Surprenant (Eds) The nature of remem-bering Essays in honor of Robert G Crowder (pp 95-115) Wash-ington DC American Psychological Association
Roediger H L III amp McDermott K B (1995) Creating false mem-ories Remembering words not presented in lists Journal of Experi-mental Psychology Learning Memory amp Cognition 21 803-814
Roediger H L III amp McDermott K B (1999) False alarms aboutfalse memories Psychological Review 106 406-410
Roediger H L III amp McDermott K B (2000) Tricks of memoryCurrent Directions in Psychological Science 9 123-127
Roediger H L III McDermott K B amp Pisoni D B (2001) Rec-ollection of illusory voices Manuscript submitted for publication
Roediger H L III McDermott K B amp Robinson K J (1998)The role of associative processes in creating false memories In M AConway S E Gathercole amp C Cornoldi (Eds) Theories of memoryII (pp 187-245) Hove UK Psychological Press
Russell W A amp Jenkins J J (1954) The complete Minnesotanorms for responses to 100 words from the KentndashRosanoff Word As-sociation Test (Tech Rep No 11 Contract N8 ONR 66216 Officeof Naval Research) University of Minnesota Minneapolis
Schacter D L Israel L amp Racine C (1999) Suppressing falserecognition in younger and older adults The distinctiveness heuris-tic Journal of Memory amp Language 40 1-24
Schacter D L Verfaellie M amp Pradere D (1996) The neuro-psychology of memory illusions False recall and recognition in am-nesic patients Journal of Memory amp Language 35 319-334
Seamon J G Luo C R amp Gallo D A (1998) Creating false mem-ories of words with or without recognition of list items Evidence fornonconscious processes Psychological Science 9 20-26
Smith R E amp Hunt R R (1998) Presentation modality affects falsememory Psychonomic Bulletin amp Review 5 710-715
Stadler M A Roediger H L III amp McDermott K B (1999)Norms for word lists that create false memories Memory amp Cogni-tion 27 494-500
Thapar A amp McDermott K B (2001) False recall and false recog-nition induced by presentation of associated words Effects of retentioninterval and level of processing Memory amp Cognition 29 424-432
Toglia M P amp Battig W F (1978) Handbook of semantic wordnorms Hillsdale NJ Erlbaum
Toglia M P Neuschatz J S amp Goodwin K A (1999) Recall ac-curacy and illusory memories When more is less Memory 7 233-256
Tulving E (1974) Cue-dependent forgetting American Scientist 6274-82
Tulving E (1979) Relation between encoding specificity and levelsof processing In L S Cermak amp F I M Craik (Eds) Levels of pro-cessing in human memory (pp 405-428) Hillsdale NJ Erlbaum
Tulving E (1985) Memory and consciousness Canadian Psycholo-gist 26 1-12
Tun P A Wingfield A Rosen M J amp Blanchard L (1998)Response latencies for false memories Gist-based processes in nor-mal aging Psychology amp Aging 13 230-241
398 ROEDIGER WATSON MCDERMOTT AND GALLO
Underwood B J (1965) False recognition produced by implicit ver-bal responses Journal of Experimental Psychology 70 122-129
Weldon M S (1991) Mechanisms underlying priming on perceptualtests Journal of Experimental Psychology Learning Memory ampCognition 17 526-541
Wickens T D amp Hirshman E (2000) False memories and statisticaldecision theory Comment on Miller and Wolford (1999) and Roedi-ger and McDermott (1999) Psychological Review 107 377-383
Wixted J T amp Stretch V (2000) The case against a criterion-shiftaccount of false memory Psychological Review 107 368-376
Zechmeister E B (1969) Orthographic distinctiveness Journal ofVerbal Learning amp Verbal Behavior 8 754-761
NOTES
1 The orthographic distinctiveness of a critical item was obtained bypresenting 72 subjects with all 55 critical items visually and havingthem rate the words in accordance with procedures developed by Zech-meister (1969) Briefly the subjects were instructed to rate each wordfor its physical appearance in terms of distinctiveness in the languageon a 9-point scale with 1 being not very distinctive 5 corresponding toaverage distinctiveness and 9 being most distinctive To avoid orderingeffects three different randomized orderings of words were used acrossroughly equal groups of subjects In the case of the critical item sleep
for example the distinctiveness is 515 Corresponding values for but-terfly and for man are 744 and 213 respectively Although ortho-graphic distinctiveness was not included in our multiple regressionanalyses this predictor is included in Appendix C
2 The orthographic neighborhood size or Coltheartrsquos N of each crit-ical item was defined as the number of words that can be formed bychanging a single letter in the word (Coltheart Davelaar Jonasson ampBesner 1977) For example because it has no orthographic neighborsColtheartrsquos N for the critical item butterfly is zero In contrast sleep hasfive orthographic neighbors (ie sheep sleek sleet steep and sweep)The reference set of words used were those in Ku Iumlcera and Francisrsquos(1967) norms Although Coltheartrsquos N was not included in our multipleregression analyses this predictor is included in Appendix C
3 Word length orthographic distinctiveness and orthographic neigh-borhood size of the critical item were all highly correlated Specificallythe bivariate correlations among these three predictors were as followslength and orthographic distinctiveness r 5 184 t(53) 5 1117 p lt05 orthographic distinctiveness and orthographic neighborhood sizer 5 66 t(53) 5 640 p lt 05 length and orthographic neighbor-hood size r 5 63 t(53) 5 582 p lt 05 Of these three variablesword length had the largest bivariate correlation with false recall (r 5
37) so it was the predictor used in the multiple regression analyses4 The multiple regression data matrix is available at wwwiacwustl
edunmclWeb
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 399
APPENDIX A The Fifty-Five 15-Word Lists and Their Associated Critical Items
With Backward Associative Strength (BAS) and Forward Associative Strength (FAS)Values Used in the Multiple Regression Analysis of False Recall
Item Word BAS FAS Item Word BAS FAS
Anger mad 393 412 Black white 655 557fear 020 059 dark 111 100hate 028 109 cat 000 043rage 541 042 charred 023 000temper 182 000 night 000 021fury 306 000 funeral 034 000ire 179 000 color 074 050wrath 128 000 grief 000 000happy 000 042 blue 028 000fight 034 000 death 016 014hatred 070 000 ink 020 000mean 090 000 bottom 000 000calm 000 000 coal 288 000emotion 000 000 brown 338 000enrage 378 000 gray 365 000
Mean 157 044 Mean 130 052Army Navy 543 500 Bread butter 364 487
NotemdashUnless otherwise indicated with an asterisk all BAS and FAS values are drawn from the associative norms of Nel-son McEvoy and Schreiber (1999) Those values shown with an asterisk were obtained by the authors by using normingprocedures similar to those of Nelson et al (1999) The corresponding mean for these BAS and FAS values for each list arealso represented
406 ROEDIGER WATSON MCDERMOTT AND GALLO
APPENDIX B Data Matrix of the Fifty-Five 15-Word Lists and Associated Critical Items
Used in the Multiple Regression Analysis of False Recall
Critical False False Log Vrd Raw Ortho Item Recall Recgn Length Freq Concrete FAS BAS Connect Recall Freq Dist Colt N
(Manuscript received July 10 2000revision accepted for publication February 23 2001)
Figure 1 Probability of false recall of the critical item from 55 lists (A) as a functionof associative strength of the list items to the critical item or backward associativestrength and (B) as a function of veridical recall of studied items from the lists
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 391
cal item on a free association test the more probable isfalse recall of the critical item Veridical recall was alsoa strong predictor of false recall (b 5 40 sr 5 34t 5 411) As we found at the bivariate level the morewords recalled from a list the less likely that the criticalitem for that list was falsely recalled None of the re-maining critical item or list variables including length ofthe critical item connectivity and FAS explained aunique portion of the total variance in false recall afterthe influence of other predictors had been residualized inthe context of the regression equation Therefore BASand recall of list items appeared to be the major contrib-utors to the multiple R of 182 and the associated mul-tiple R2 of 68 [F(747) 5 1406 MSe 5 0013] We fur-ther explore the implications of these effects in theDiscussion section
Factors Affecting False RecognitionIn our analyses thus far we have concentrated on
false recall exclusively However the norms for our 55lists (collected by Stadler et al 1999 and Gallo ampRoediger 2001) also include values for false recogni-tion The recognition data in these norms were collectedin a final recognition test after lists had been studied andrecalled This confounding of recognition by prior recallalmost certainly affected both veridical and false recog-nition However in experiments directly examining theeffects of recall on recognition in this paradigm and oth-ers the finding is that recall of a list may increase itsrecognition by a moderate amount (typically in the4ndash10 range eg Roediger amp McDermott 1995Roediger McDermott amp Pisoni 2001) In fact some re-searchers have failed to observe any effect at all of recall
392 ROEDIGER WATSON MCDERMOTT AND GALLO
on recognition (Schacter Verfaellie amp Pradere 1996 seeRoediger et al 1998 for a discussion) We conducted thesame regression analyses with our seven predictor vari-ables on false recognition as we did on false recall Wesummarize the basic results and conclusions as follows
There were three significant bivariate correlationswith false recognition including length (r 5 27)BAS (r 5 143) and veridical recall (r 5 52) Over-all these correlations with false recognition were simi-lar to those obtained with false recall However the pos-itive correlation between BAS and false recognition wassmaller than that in the recall data and the negative cor-relation between veridical recall and false recognitionwas larger than that in the recall data To maintain con-sistency with the previous regression analysis on falserecall we used veridical recall rather than veridicalrecognition in this analysis In addition recognition es-timates were based on only three itemslist and wereprobably affected by prior recall
Turning to the results of the simultaneous multiple re-gression analysis on false recognition BAS and veridi-cal recall of list items appeared to be the major contrib-utors to the multiple R of 169 and the associatedmultiple R2 of 48 [F(747) 5 611 MSe 5 002] Table 3shows the summary of the regression analyses Althoughthe regression analysis accounted for less total variancein false recognition than in false recall (48 as opposedto 68 respectively) the same factors seem to be re-sponsible for predicting false recognition and false re-call Of course because measures of recognition alwaysfollowed recall this similar pattern may reflect a carry-over effect of false recall on false recognition Nonethe-less within the limits of this study the factors responsi-ble for false recognition across lists seem to be the sameas those producing false recall
DISCUSSION
To summarize the main findings our results implicatetwo primary factors (of the seven variables we exam-ined) as accounting for most of the variance in false re-call across lists The split-half correlation of our lists was190 so the amount of potentially explainable variancein our dependent variable of false recall is 81 The fac-
tors in our multiple regression accounted for 68 of thetotal variance in false recall or 84 of the explainablevariance (68 4 81 3 100 5 84) The analysis forfalse recognition revealed the same factors at work al-though they accounted for less overall variance Thestrongest factor was BAS consistent with prior work byDeese (1959b) and McEvoy et al (1999) The other fac-tor was the number of items recalled accurately from thelist which is negatively correlated with false recall
We turn next to the theoretical implications of ourfindings We first consider the implications for the acti-vationmonitoring framework that we have developed toexplain false recall and false recognition in the DRMparadigm We then consider the implications for othertheories and f inally turn to several further issues onwhich our results shed light
ActivationMonitoring TheoryIn the theroretical framework that we (eg McDer-
mott amp Watson 2001 Roediger Balota amp Watson 2001Roediger amp McDermott 1995 2000) have used to explainvariations in the probability of false recall and false recog-nition in the DRM paradigm two sets of processes havebeen proposed Specifically we have proposed that pro-cesses occurring during both encoding and retrieval af-fect the probability of false recall and false recognitionin the DRM paradigm Two critical sets of processes in-volve activation and monitoring Although there is a nat-ural alignment of activation to encoding and monitoringto retrieval we note that both processes can potentiallyoccur both during encoding and during retrieval as willbe discussed below The ideas we employ have many pre-cursors which we acknowledge below although theirparticular arrangement here may be unique
A critical distinction in understanding encoding pro-cesses and their interaction with retrieval processes isbetween item-specific and relational processing (Ander-son 1972 Hunt amp Einstein 1981 Hunt amp McDaniel1993) Briefly the notion is that emphasis can be given tothe encoding of individual elements and their features(what Anderson called node-taggingand Hunt and his col-leagues called item-specific processing) or to the encod-ing of relationships among elements (pathway-taggingor relational processing) Item-specific processing should
Table 3 Summary of Multiple Regression Analysis
on False Recognition With Critical Item Variablesand List Variables Used as Predictors
NotemdashR2 5 48 F(747) 5 611 MSe 5 002 p lt 05 p lt 05
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 393
make individual elements more memorable and lead tomore distinctive memories such processing should helpdistinguish veridical memories from false memories be-cause list items would have features of presentation as-sociated with them whereas the critical item sometimeswould not (However if the critical item is strongly acti-vated during encoding it may take on many of the fea-tures of list items) Relational processing (extractingthemes or schemas or constructing networks of associa-tions) should increase the probability of false recall andfalse recognition by leading to inferences or associationsof events that may not actually have occurred
We assume that in encoding and interpreting experi-ences people go beyond the information given in thestimulus (Bartlett 1932 chap 2 Bruner 1957) and ac-tivate related information through inferential processingEach personrsquos experience is coded in terms of his or herown schemata (or associative networks or knowledgestructures) When schemata are activated a process offilling in of ldquomissingrdquo information may occur wherebyitems strongly implied by the schema (but not actuallypresent in the stimulus) may become activated (Bartlett1932) In encoding lists of words in the DRM paradigmpresentation of the list can create activation that spreadsthroughout the lexicalsemantic system and can createimplicit associative responses (Underwood 1965) Thisactivation may arouse concepts that are not presented but that are associated to concepts that are presented(eg Anderson amp Pirolli 1984 Collins amp Loftus 1975Meyer amp Schvaneveldt 1971) The associative activa-tion in this approach can arise from automatic process-ing or from more controlled elaborative processing(Neely 1977) In addition the associative activation ofconcepts that were not presented may occur consciously(subjects may become aware of the concept of sleep andrehearse it during presentation of the list words bed restawake etc) or the concept may be strongly activatedbut never come to conscious awareness The debate overwhether arousal of the critical item in the DRM para-digm is conscious or unconscious is interesting but dif-ficult to decide empirically Some evidence has impli-cated conscious arousal of the concepts (McDermott1997) whereas other evidence implicates unconsciousactivation (Seamon Luo amp Gallo 1998) The correctanswer to the question of whether activation is consciousor unconscious is probably ldquobothrdquomdashthat is the criticalitem probably sometimes comes consciously to mindand even if it does not relevant information may never-theless be activated unconsciously
The high correlation between BAS and false recall aswell as the strong weighting of BAS in the multiple re-gression analysis strongly implicates activation of criti-cal items as an important factor in explaining false recalland false recognition in the DRM paradigm The morestrongly associated list items are to critical items themore activation would be expected to spread to the crit-ical items More colloquially to the extent that the wordson the list spark the associative connection to the critical
item the more likely the critical item is to be falsely re-called This activation factor would seem to be one mainreason why ldquothings suggested to a person by an experi-ence may be honestly reported by him as part of the ex-periencerdquo (Kirkpatrick 1894 p 608) These associatedlists probably all tend to elicit relational processing butonly those with strong BAS to the critical item lead tofalse recall
The second set of factors proposed in activationmon-itoring theory is concerned with the monitoring of mem-ory accuracy Monitoring of encoding processes canoccur during the study phase of an experiment espe-cially under intentional learning conditions In experi-ments in which subjects are warned about false memoryphenomena and are told to pay close attention to lists andto remember only the presented items and not the criti-cal items monitoring is enhanced (Gallo Roberts amp Sea-mon 1997 Gallo Roediger amp McDermott 2001 Mc-Dermott amp Roediger 1998) Monitoring of memorieshas typically been discussed with respect to retrieval dur-ing which time a primary goal is to distinguish fromamong the considerable information brought to con-sciousness that which refers to perception of past eventsfrom that which does not a process referred to as realitymonitoring (Johnson amp Raye 1981) The experience ofremembering events involves mentally traveling back intime and seeming to reexperience the events (Tulving1985) During retrieval recovered memories that appar-ently carry with them details of having been experiencedare generally attributed to be memories of actual pastevents although sometimes this attribution may be inerror (Jacoby et al 1989)
Johnson et al (1993) have described in detail the typesof information that may be used to distinguish eventsthat actually happened from ones that were only imag-ined thought about or read about For example eventsthat actually occurred have more features from the ex-ternal world associated with them (sights sounds etc)whereas those that were internally generated are lesslikely to have these features but more likely to have as-sociated reflections and cognitive operations In theDRM paradigm the fact that the critical item has beenstrongly activated during encoding means that duringretrieval the subject is likely to retrieve the critical itemas a list member To the extent that the critical items arestrongly activated during encoding they will take onsome features similar to those of the list items this is es-pecially true if the critical items come consciously tomind during encoding and are rehearsed in the samemanner as list items To the extent that the features of thecritical item are similar to those of actual list items theprobability of false recall or false recognition is increasedHowever if the list items systematically differ from thecritical item on some dimensions or features retrievalheuristics may be brought into play that can reduce falserecall and false recognition (Israel amp Schacter 1997Schacter Israel amp Racine 1999) Considerable evidencesupports this general source-reality-monitoring frame-
394 ROEDIGER WATSON MCDERMOTT AND GALLO
work that we advocate here (eg Johnson et al 1993Schacter et al 1999) as well as the attributional ap-proach to remembering (Jacoby et al 1989)
The other significant factor arising from our multipleregression analysis may operate during retrieval by aid-ing the monitoring processes that reduce false recall Thenegative correlation between veridical recall of list itemsand false recall of the critical item may indicate that thebetter encoded list items are the more easily they can bedistinguished from the illusory critical item Subjectsmay use item-specific information to aid recall of listitems and suppress recall of the critical items whichwould not carry as much item-specific information Whenexperimental manipulations increase veridical recall ei-ther through longer study time at relatively slow rates(Gallo amp Roediger 2001 McDermott amp Watson 2001Toglia Neuschatz amp Goodwin 1999) or through re-peated study and test opportunities (McDermott 1996Schacter et al 1999) false recall is often decreasedTherefore the negative relation between veridical recalland false recall can be produced experimentally as wellas appearing in our correlationregression analysis How-ever veridical and false recall can be positively related as afunction of other variables too as was shown by McDer-mott (1996) Toglia et al (1999) and Thapar amp McDermott(2001) In these cases greater activation of the criticalitem from relational processing or from greater seman-tic processing is probably at work
The fact that the length andor distinctiveness of thecritical item are negatively correlated with false recallcan be similarly interpreted as the effect of monitoringduring retrieval because the greater distinctiveness ofthe critical item simplifies the monitoring process Forexample if words in the list are generally shorter andhave more regular orthography than does the criticalitem then during retrieval the critical items that are longand distinctive (eg butterfly cabbage) should be moreeasily discriminated from list items as indeed seems tobe the case in the bivariate correlations Interestingly thelengthdistinctiveness factor played a role in determin-ing false recall even though the lists in our norms werepresented auditorily and the orthographic informationat least should not be particularly accessible In addi-tion presenting the DRM lists visually rather than audi-torily reduces false recall and false recognition at leaston tests in which recognition items are presented visually(as opposed to auditorily) or recall is written rather thanspoken (Gallo McDermott et al 2001 Kellogg 2001Smith amp Hunt 1998) This modality difference has alsobeen interpreted as arising from the retrieval of distinctiveinformation operating during the test (Gallo McDermottet al 2001 Smith amp Hunt 1998)
As was noted above activation processes may occurduring retrieval as well as during encoding The morelist items a subject recalls the more primed the criticalitem might become However the present analyses un-dercut the simplest form of this idea If the recalling oflist items activated or primed false recall the simplest
prediction would be that there should be a positive cor-relation between veridical recall and false recall Our datarevealed the opposite pattern showing a negative corre-lation between veridical and false recall Marsh McDer-mott and Roediger (2001) reported other experimentsthat also call into question the idea that priming duringthe test plays a role in false recall and false recognition
Balota et al (1999) obtained further evidence sup-porting the activationmonitoring interpretation of falsememories in the DRM paradigm They compared per-formance of younger adults healthy older adults andolder adults with Alzheimerrsquos disease The latter twogroups have been shown to have intact activation pro-cesses as measured by semantic priming paradigms(Balota amp Duchek 1988 1991) yet they are deficient insource-monitoring processes which is probably attribut-able to impaired frontal lobe functioning (see BalotaDolan amp Duchek 2000) Therefore one might expectthat whereas veridical recall would be worse in older adultsand Alzheimerrsquos patients (relative to younger adults)false recall and false recognition would occur at the samelevel or perhaps be increased in these groups IndeedBalota et al (1999) reported exactly this pattern A sim-ilar outcome has been obtained in other laboratories(eg Budsen Daffner Desikan amp Schacter 2000 Nor-man amp Schacter 1997 Tun Wingfield Rosen amp Blan-chard 1998) This pattern is just as predicted by the ac-tivation monitoring framework
In short the present results and many others (seeRoediger Balota amp Watson 2001 Roediger et al 1998)fit comfortably within the activationmonitoring frame-work In the remainder of our discussion we will con-sider briefly the implications of our present results forother accounts of the DRM paradigm and for the role ofconnectivity (or interitem associative strength) in pro-ducing veridical and false recall
Fuzzy-Trace TheoryAnother leading theory that has been used to explain
false memory phenomena is fuzzy-trace theory (egReyna amp Brainerd 1995) Payne et al (1996) and Schac-ter et al (1996) among many others have used the basictheory to account for false recall and false recognitiondata within the DRM paradigm and the theory has beenextended in interesting ways to account for new varia-tions on the paradigm (eg Brainerd Wright Reyna ampMojardin 2001) The fundamental assumption of fuzzy-trace theory is that experience leaves two types of mem-ory traces gist traces (which capture the meaning of ex-perience but without specific attributes) and verbatimtraces (which represent specific attributes) False memo-ries are attributed to gist processing When the meaning ofa recognition probe matches the gist of the encoded expe-rience false recognition occurs The theory has been di-rected primarily to false recognition and not false recall
In trying to explain the present results one difficultyis that the operational definition of gist has never beenspecified that is within the DRM paradigm given a list
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 395
of 15 related words how does one specify the gist repre-sentation of the list and quantify how strong it is Onecould use patterns of false recognitionmdashto the extentthat a concept is falsely recognized it was represented aspart of the gist representation This definition is per-fectly suitable for some purposes but clearly explainingfalse recognition is not one of them owing to the prob-lem of circularity To explain why some lists lead to highlevels of false recall and false recognition (whereas otherlists lead to low levels) within fuzzy-trace theory oneneeds an independent specification of gist which to ourknowledge has not been provided in this context Whydoes the sweet list have so much gist that it leads to falserecall whereas the bitter list does not One plausible in-terpretation of gist is BAS The more the list items areassociated to the critical item the more a gist represen-tation of the critical item is created This solution to adefinition of gist works within the DRM paradigm butof course the interpretation of how gist is created be-comes isomorphic to activation processes within the ac-tivationmonitoring framework
Interpretation of the other factor that was significantwithin our multiple regression analysis may be morestraightforward within fuzzy-trace theory The negativecorrelation between veridical recall and false recall (andfalse recognition) may be accounted for by increased re-liance on verbatim traces That is if increased recall of listitems is due to increased verbatim information subjectsmay rely less heavily on gist information and hence showless false recall and false recognition during respondingSimilarly for critical items that are long andor ortho-graphically distinct recall of list items that relies on thesefeatures may lead to suppression of false recall and falserecognition of the critical items because they may differsystematically from other list items These factors pos-tulated within the context of fuzzy-trace theory are quitesimilar to those within the activationmonitoring frame-work outlined above At some level of generality there aretwo sets of factors in each account and both theories canexplain many empirical results
Shifting Criterion TheoryMiller and Wolford (1999) proposed that false recog-
nition in the DRM paradigm could be explained by sub-jectsrsquo shifting their criteria differentially to list items andto critical items during a recognition test (They ex-tended their explanation to false recall but somewhattenuously by proposing a generaterecognize theory)They postulated that subjects encoded the structure ofthe list during its presentation and then during the recog-nition test responded with especially liberal criteria tocritical items but not to list items The criterion shift issupposed to occur to the degree that the test item appearsto be related to the list items To support their argumentthey reported that a measure of bias from signal detec-tion theory was indeed more liberal for critical itemsthan for list items However Wickens and Hirshman(2000) and Wixted and Stretch (2000) pointed out that
within signal detection theory differences in bias mea-sures do not necessarily implicate a criterion shiftRather differences in measured bias could arise fromdifferences among the underlying distributions Wixtedand Stretch argued that this latter interpretation was themore likely and that the criterion shift interpretation wasuntenable (also see Roediger amp McDermott 1999)
The present evidence also weighs heavily against acriterion shift account Miller and Wolford (1999) ap-pealed to subjectsrsquo metaknowledge of the structure of thelist Within the criterion shift account we can assumethat the likelihood that a critical item cues metaknowl-edge of the list depends on its FAS or how well the crit-ical item will cue the list If so FAS (the association ofthe critical item to the list items) should be the primarydeterminant of false recognition and false recall How-ever the correlation between FAS and false recall in ourdata was negligible (08) and the same was true in analy-ses of false recognition BAS was actually the primarypredictor of false recall which implicates activation pro-cesses during encoding as being a primary cause of theeffect There would be no reason to expect BAS to haveany impact according to the idea that a simple criterionshift during the test causes false recall and false recog-nition Of course all our lists were relatively high androughly equivalent in terms of FAS because they werecreated from forward associations to the critical item sorestriction of range may account for the low correlationEven so the same problem exists for the shifting crite-rion theory to explain the results If FAS is supposed todetermine false recall and false recognition by this ac-count and if all our lists were high and roughly equal onthis feature why were so many lists ineffective in elicitingfalse recognition and false recall In short either way theshifting criterion theory cannot account for the results
Of course within the activationmonitoring frame-work the monitoring processes occurring during retrievalcould be conceived as conservative shifts in response cri-teria However in our theory these monitoring processesdetermine the degree to which a given level of activationat retrieval will be judged sufficient to endorse a criticalitem as having been studied thereby producing a falsememory In contrast in Miller and Wolfordrsquos (1999) ac-count there was no activation of the critical item duringencoding and shifting criteria during the test were saidto account entirely for the false recognition phenomenon(see Wixted amp Stretch 2000) In the activationmonitoringaccount the memory illusion provoked by the DRM listsis assumed to be a subjectively powerful phenomenonthat can be ameliorated (but rarely eliminated) when listitems are made more distinctive relative to criticalitems Under conditions in which the subject is warnedat test (but not before the encoding phase) to monitortheir memories for the presence or absence of the criti-cal item and is discouraged from making criterion shiftsthe false memory effect is neither eliminated nor greatlyreduced (Gallo Roediger amp McDermott 2001) Hicksand Marsh (2001) showed that under conditions in which
396 ROEDIGER WATSON MCDERMOTT AND GALLO
subjects are explicitly told to monitor their memories dur-ing the test (and are asked to judge whether each wordhad been externally presented or internally generated)false recognition in the DRM paradigm actually increasesrather than decreases In fact the false recognition effectpersists even when subjects are fully informed about thenature of the phenomenon before the study episode andare therefore encouraged to enhance monitoring processesduring both encoding and retrieval in order to attempt todiminish or eliminate the effect (see Gallo et al1997Gallo Roediger amp McDermott 2001 McDermott ampRoediger 1998) For these and other reasons the notionthat the DRM memory illusion is caused by a general shiftof criteria can be eliminated as a potential explanation
Interitem Associative Strength or ConnectivityOur data pose something of a puzzle about the role of
interitem associative strength (Deese 1959a) or connec-tivity (McEvoy et al 1999) in false recall As was dis-cussed in the introduction Deese (1959a) found a sig-nificant negative correlation between this construct andthe probability of false recall In a later paper Deese(1961) obtained a weaker negative correlation that failedto reach conventional levels of significance so the issueis somewhat in doubt from Deesersquos (1959a 1961) researchMcEvoy et al created lists that varied in connectivity(interitem associative strength) and in BAS and manip-ulated these factors orthogonally They replicated thefinding of Deese (1959a) in that the density of inter-connections of list words was negatively related to falserecall However in our analysis of 55 lists (different fromthose used by McEvoy et al 1999) we found no relationbetween connectivity and false recall (r 5 04) or be-tween connectivity and false recognition (r 5 03)
Further research will be necessary to determine thetrue state of affairs although a case can be made that thelack of correlation we obtained is accurate First we had55 lists that varied widely in their ability to elicit false re-call as compared with 24 lists used in McEvoy et al(1999) Our lists also represented a relatively full rangeof connectivity or interitem associative strength relativeto those of McEvoy et al McEvoy et al used lists thathad a somewhat wider range of connectivity because theyused extreme values whereas our larger set of lists pro-vided a fuller body of the middle ranges of connectivityGiven our results it seems at least possible that interitemassociative strength (connectivity) is not a factor in de-termining whether lists of associative words lead to falserecall although the issue remains in doubt
We did find that interitem associative strength or con-nectivity is positively and significantly correlated withveridical recall (r 5 132) This outcome is consistentwith that of Deese (1959a) and McEvoy et al (1999)The greater the associative bonds among list items thegreater is list recall This outcome is also consistent withvirtually any associative theory of recall (eg Anderson1972 1983 Nelson Schreiber amp McEvoy 1992)
CONCLUSION
The present study has uncovered two primary factorsthat are significantly related to false recall and falserecognition in the DRM paradigm BAS is positivelycorrelated with false recall whereas veridical recall ofthe list items is negatively related to false recall The ac-tivationmonitoring account provides a natural interpre-tation of these factors Any viable theory posited to ex-plain the occurrence of false memories in the DRMparadigm must explain the variability in false recall andfalse recognition across lists and the two primary factorsresponsible for this variability
REFERENCES
Anderson J R (1972) FRAN A simulation model of free recall InG H Bower (Ed) The psychology of learning and motivation (Vol 5pp 315-378) New York Academic Press
Anderson J R (1983) The architecture of cognition CambridgeMA Harvard University Press
Anderson J R amp Pirolli P L (1984) Spread of activation Journalof Experimental Psychology Learning Memory amp Cognition 10791-798
Balota D A Cortese M J Duchek J M Adams D RoedigerH L III McDermott K B amp Yerys B E (1999) Veridical andfalse memories in healthy older adults and in dementia of the Alz-heimers type Cognitive Neuropsychology 16 361-384
Balota D A Dolan P O amp Duchek J M (2000) Memorychanges in healthy older adults In E Tulving amp F I M Craik (Eds)The Oxford handbook of memory (pp 395-409) Oxford Oxford Uni-versity Press
Balota D A amp Duchek J M (1988) Age-related differences inlexical access spreading activation and simple pronunciation Psy-chology and Aging 3 84-93
Balota D A amp Duchek J (1991) Semantic priming effects lexicalrepetition effects and contextual disambiguation effects in healthyaged individuals and individuals with senile dementia of theAlzheimer type Brain amp Language 40 181-201
Bartlett F C (1932) Remembering A study in experimental and so-cial psychology Cambridge Cambridge University Press
Brainerd C J Wright R Reyna V F amp Mojardin A H (2001)Conjoint recognition and phantom recollection Journal of Experi-mental Psychology Learning Memory amp Cognition 27 307-327
Bruce D amp Winograd E (1998) Remembering Deesersquos 1959 arti-cles The Zeitgeist the sociology of science and false memoriesPsychonomic Bulletin amp Review 5 615-624
Bruner J S (1957) On perceptual readiness Psychological Review64 123-152
Budson A E Daffner K R Desikan R amp Schacter D L(2000) When false recognition is unopposed by true recognitionGist-based memory distortion in Alzheimerrsquos disease Neuropsy-chology 14 277-287
Collins A M amp Loftus E F (1975) A spreading-activation theoryof semantic memory Psychological Review 82 407-428
Coltheart M Davelaar E Jonasson J T amp Besner D (1977)Access to the internal lexicon In S Dormic (Ed) Attention and per-formance VI (pp 535-555) Hillsdale NJ Erlbaum
Deese J (1959a) Influence of interitem associative strength upon im-mediate free recall Psychological Reports 5 235-241
Deese J (1959b) On the prediction of occurrence of particular verbalintrusions in immediate recall Journal of Experimental Psychology58 17-22
Deese J (1961) Associative structure and the serial reproduction ex-periment Journal of Abnormal amp Social Psychology 63 95-100
Deese J (1965) The structure of associations in language and thoughtBaltimore Johns Hopkins University Press
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 397
Gallo D A McDermott K B Percer J M amp Roediger H LIII (2001) Modality effects in false recall and false recognitionJournal of Experimental Psychology Learning Memory amp Cogni-tion 27 339-353
Gallo D A Roberts M J amp Seamon J G (1997) Rememberingwords not presented in lists Can we avoid creating false memoriesPsychonomic Bulletin amp Review 4 271-276
Gallo D A amp Roediger H L III (2001) Variability among wordlists in evoking associative memory illusions Manuscript submittedfor publication
Gallo D A Roediger H L III amp McDermott K B (2001) As-sociative false recognition occurs without strategic criterion shiftsPsychonomic Bulletin amp Review 8 579-586
Hicks J L amp Marsh R L (2001) False recognition occurs more fre-quently during source identification than during oldndashnew recogni-tion Journal of Experimental Psychology Learning Memory ampCognition 27 375-383
Hunt R R amp Einstein G O (1981) Relational and item-specific in-formation in memory Journal of Verbal Learning amp Verbal Behav-ior 19 497-514
Hunt R R amp McDaniel M A (1993) The enigma of organizationand distinctiveness Journal of Memory amp Language 32 421-445
Israel L amp Schacter D L (1997) Pictorial encoding reduces falserecognition of semantic associates Psychonomic Bulletin amp Review4 577-581
Jacoby L L Kelley C M amp Dywan J (1989) Memory attribu-tions In H L Roediger III amp F I M Craik (Eds) Varieties of mem-ory and consciousness Essays in honour of Endel Tulving (pp 391-422) Hillsdale NJ Erlbaum
Johnson M K Hashtroudi S amp Lindsay D S (1993) Sourcemonitoring Psychological Bulletin 114 3-28
Johnson M K amp Raye C L (1981) Reality monitoring Psycho-logical Review 88 67-85
Kellogg RT (2001) Presentation modality and mode of recall in ver-bal false memory Journal of Experimental Psychology LearningMemory amp Cognition 27 913-919
Kirkpatrick E A (1894) An experimental study of memory Psy-chological Review 1 602-609
Ku Iumlcera H amp Francis W (1967) Computational analysis of present-day American English Providence RI Brown University Press
Marsh E J McDermott K B amp Roediger H L III (2001) Doretrieval factors play a role in the creation of of false memories Man-uscript submitted for publication
Mather M Henkel L A amp Johnson M J (1997) Evaluatingcharacteristics of false memories Rememberknow judgments andmemory characteristics questionnaire compared Memory amp Cogni-tion 25 826-837
McDermott K B (1996) The persistence of false memories in list re-call Journal of Memory amp Language 35 212-230
McDermott K B (1997) Priming on perceptual implicit memorytests can be achieved through presentation of associates Psycho-nomic Bulletin amp Review 4 582-586
McDermott K B amp Roediger H L III (1998) Attempting to avoidillusory memories Robust false recognition of associates persistsunder conditions of explicit warnings and immediate testing Journalof Memory amp Language 39 508-520
McDermott K B amp Watson J M (2001) The rise and fall of falserecall The impact of presentation duration Journal of Memory ampLanguage 45 160-176
McEvoy C L Nelson D L amp Komatsu T (1999) What is the con-nection between true and false memories The differential roles of in-teritem associations in recall and recognition Journal of Experimen-tal Psychology Learning Memory amp Cognition 25 1177-1194
Meyer D E amp Schvaneveldt R W (1971) Facilitation in recog-nizing pairs of words Evidence of a dependence between retrievaloperations Journal of Experimental Psychology 90 227-234
Miller M B amp Wolford G L (1999) Theoretical commentaryThe role of criterion shift in false memory Psychological Review106 398-405
Neely J H (1977) Semantic priming and retrieval from semanticmemory Roles of inhibitionless spreading activation and limited-
capacity attention Journal of Experimental Psychology General106 226-254
Nelson D L McEvoy C L amp Schreiber T A (1999) The Uni-versity of South Florida word association rhyme and word fragmentnorms Unpublished manuscript University of South Florida Tampa
Nelson D L Schreiber TA amp McEvoy C L (1992) Processing im-plicit and explicit representations Psychological Review 99 322-348
Norman K A amp Schacter D L (1997) False recognition in youngerand older adults Exploring the characteristics of illusory memoriesMemory amp Cognition 25 838-848
Paivio A Yuille J C amp Madigan S A (1968) Concreteness im-agery and meaningfulness values for 925 nouns Journal of Experi-mental Psychology Monographs 76 (1 Pt 2)
Payne D G Elie C J Blackwell J M amp Neuschatz J S (1996)Memory illusions Recalling recognizing and recollecting events thatnever occurred Journal of Memory amp Language 35 261-285
Reyna V F amp Brainerd C J (1995) Fuzzy-trace theory An interimsynthesis Learning amp Individual Differences 7 1-75
Roediger H L III Balota D A amp Watson J M (2001) Spreadingactivation and the arousal of false memories In H L Roediger III J SNairne I Neath amp A M Surprenant (Eds) The nature of remem-bering Essays in honor of Robert G Crowder (pp 95-115) Wash-ington DC American Psychological Association
Roediger H L III amp McDermott K B (1995) Creating false mem-ories Remembering words not presented in lists Journal of Experi-mental Psychology Learning Memory amp Cognition 21 803-814
Roediger H L III amp McDermott K B (1999) False alarms aboutfalse memories Psychological Review 106 406-410
Roediger H L III amp McDermott K B (2000) Tricks of memoryCurrent Directions in Psychological Science 9 123-127
Roediger H L III McDermott K B amp Pisoni D B (2001) Rec-ollection of illusory voices Manuscript submitted for publication
Roediger H L III McDermott K B amp Robinson K J (1998)The role of associative processes in creating false memories In M AConway S E Gathercole amp C Cornoldi (Eds) Theories of memoryII (pp 187-245) Hove UK Psychological Press
Russell W A amp Jenkins J J (1954) The complete Minnesotanorms for responses to 100 words from the KentndashRosanoff Word As-sociation Test (Tech Rep No 11 Contract N8 ONR 66216 Officeof Naval Research) University of Minnesota Minneapolis
Schacter D L Israel L amp Racine C (1999) Suppressing falserecognition in younger and older adults The distinctiveness heuris-tic Journal of Memory amp Language 40 1-24
Schacter D L Verfaellie M amp Pradere D (1996) The neuro-psychology of memory illusions False recall and recognition in am-nesic patients Journal of Memory amp Language 35 319-334
Seamon J G Luo C R amp Gallo D A (1998) Creating false mem-ories of words with or without recognition of list items Evidence fornonconscious processes Psychological Science 9 20-26
Smith R E amp Hunt R R (1998) Presentation modality affects falsememory Psychonomic Bulletin amp Review 5 710-715
Stadler M A Roediger H L III amp McDermott K B (1999)Norms for word lists that create false memories Memory amp Cogni-tion 27 494-500
Thapar A amp McDermott K B (2001) False recall and false recog-nition induced by presentation of associated words Effects of retentioninterval and level of processing Memory amp Cognition 29 424-432
Toglia M P amp Battig W F (1978) Handbook of semantic wordnorms Hillsdale NJ Erlbaum
Toglia M P Neuschatz J S amp Goodwin K A (1999) Recall ac-curacy and illusory memories When more is less Memory 7 233-256
Tulving E (1974) Cue-dependent forgetting American Scientist 6274-82
Tulving E (1979) Relation between encoding specificity and levelsof processing In L S Cermak amp F I M Craik (Eds) Levels of pro-cessing in human memory (pp 405-428) Hillsdale NJ Erlbaum
Tulving E (1985) Memory and consciousness Canadian Psycholo-gist 26 1-12
Tun P A Wingfield A Rosen M J amp Blanchard L (1998)Response latencies for false memories Gist-based processes in nor-mal aging Psychology amp Aging 13 230-241
398 ROEDIGER WATSON MCDERMOTT AND GALLO
Underwood B J (1965) False recognition produced by implicit ver-bal responses Journal of Experimental Psychology 70 122-129
Weldon M S (1991) Mechanisms underlying priming on perceptualtests Journal of Experimental Psychology Learning Memory ampCognition 17 526-541
Wickens T D amp Hirshman E (2000) False memories and statisticaldecision theory Comment on Miller and Wolford (1999) and Roedi-ger and McDermott (1999) Psychological Review 107 377-383
Wixted J T amp Stretch V (2000) The case against a criterion-shiftaccount of false memory Psychological Review 107 368-376
Zechmeister E B (1969) Orthographic distinctiveness Journal ofVerbal Learning amp Verbal Behavior 8 754-761
NOTES
1 The orthographic distinctiveness of a critical item was obtained bypresenting 72 subjects with all 55 critical items visually and havingthem rate the words in accordance with procedures developed by Zech-meister (1969) Briefly the subjects were instructed to rate each wordfor its physical appearance in terms of distinctiveness in the languageon a 9-point scale with 1 being not very distinctive 5 corresponding toaverage distinctiveness and 9 being most distinctive To avoid orderingeffects three different randomized orderings of words were used acrossroughly equal groups of subjects In the case of the critical item sleep
for example the distinctiveness is 515 Corresponding values for but-terfly and for man are 744 and 213 respectively Although ortho-graphic distinctiveness was not included in our multiple regressionanalyses this predictor is included in Appendix C
2 The orthographic neighborhood size or Coltheartrsquos N of each crit-ical item was defined as the number of words that can be formed bychanging a single letter in the word (Coltheart Davelaar Jonasson ampBesner 1977) For example because it has no orthographic neighborsColtheartrsquos N for the critical item butterfly is zero In contrast sleep hasfive orthographic neighbors (ie sheep sleek sleet steep and sweep)The reference set of words used were those in Ku Iumlcera and Francisrsquos(1967) norms Although Coltheartrsquos N was not included in our multipleregression analyses this predictor is included in Appendix C
3 Word length orthographic distinctiveness and orthographic neigh-borhood size of the critical item were all highly correlated Specificallythe bivariate correlations among these three predictors were as followslength and orthographic distinctiveness r 5 184 t(53) 5 1117 p lt05 orthographic distinctiveness and orthographic neighborhood sizer 5 66 t(53) 5 640 p lt 05 length and orthographic neighbor-hood size r 5 63 t(53) 5 582 p lt 05 Of these three variablesword length had the largest bivariate correlation with false recall (r 5
37) so it was the predictor used in the multiple regression analyses4 The multiple regression data matrix is available at wwwiacwustl
edunmclWeb
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 399
APPENDIX A The Fifty-Five 15-Word Lists and Their Associated Critical Items
With Backward Associative Strength (BAS) and Forward Associative Strength (FAS)Values Used in the Multiple Regression Analysis of False Recall
Item Word BAS FAS Item Word BAS FAS
Anger mad 393 412 Black white 655 557fear 020 059 dark 111 100hate 028 109 cat 000 043rage 541 042 charred 023 000temper 182 000 night 000 021fury 306 000 funeral 034 000ire 179 000 color 074 050wrath 128 000 grief 000 000happy 000 042 blue 028 000fight 034 000 death 016 014hatred 070 000 ink 020 000mean 090 000 bottom 000 000calm 000 000 coal 288 000emotion 000 000 brown 338 000enrage 378 000 gray 365 000
Mean 157 044 Mean 130 052Army Navy 543 500 Bread butter 364 487
NotemdashUnless otherwise indicated with an asterisk all BAS and FAS values are drawn from the associative norms of Nel-son McEvoy and Schreiber (1999) Those values shown with an asterisk were obtained by the authors by using normingprocedures similar to those of Nelson et al (1999) The corresponding mean for these BAS and FAS values for each list arealso represented
406 ROEDIGER WATSON MCDERMOTT AND GALLO
APPENDIX B Data Matrix of the Fifty-Five 15-Word Lists and Associated Critical Items
Used in the Multiple Regression Analysis of False Recall
Critical False False Log Vrd Raw Ortho Item Recall Recgn Length Freq Concrete FAS BAS Connect Recall Freq Dist Colt N
(Manuscript received July 10 2000revision accepted for publication February 23 2001)
392 ROEDIGER WATSON MCDERMOTT AND GALLO
on recognition (Schacter Verfaellie amp Pradere 1996 seeRoediger et al 1998 for a discussion) We conducted thesame regression analyses with our seven predictor vari-ables on false recognition as we did on false recall Wesummarize the basic results and conclusions as follows
There were three significant bivariate correlationswith false recognition including length (r 5 27)BAS (r 5 143) and veridical recall (r 5 52) Over-all these correlations with false recognition were simi-lar to those obtained with false recall However the pos-itive correlation between BAS and false recognition wassmaller than that in the recall data and the negative cor-relation between veridical recall and false recognitionwas larger than that in the recall data To maintain con-sistency with the previous regression analysis on falserecall we used veridical recall rather than veridicalrecognition in this analysis In addition recognition es-timates were based on only three itemslist and wereprobably affected by prior recall
Turning to the results of the simultaneous multiple re-gression analysis on false recognition BAS and veridi-cal recall of list items appeared to be the major contrib-utors to the multiple R of 169 and the associatedmultiple R2 of 48 [F(747) 5 611 MSe 5 002] Table 3shows the summary of the regression analyses Althoughthe regression analysis accounted for less total variancein false recognition than in false recall (48 as opposedto 68 respectively) the same factors seem to be re-sponsible for predicting false recognition and false re-call Of course because measures of recognition alwaysfollowed recall this similar pattern may reflect a carry-over effect of false recall on false recognition Nonethe-less within the limits of this study the factors responsi-ble for false recognition across lists seem to be the sameas those producing false recall
DISCUSSION
To summarize the main findings our results implicatetwo primary factors (of the seven variables we exam-ined) as accounting for most of the variance in false re-call across lists The split-half correlation of our lists was190 so the amount of potentially explainable variancein our dependent variable of false recall is 81 The fac-
tors in our multiple regression accounted for 68 of thetotal variance in false recall or 84 of the explainablevariance (68 4 81 3 100 5 84) The analysis forfalse recognition revealed the same factors at work al-though they accounted for less overall variance Thestrongest factor was BAS consistent with prior work byDeese (1959b) and McEvoy et al (1999) The other fac-tor was the number of items recalled accurately from thelist which is negatively correlated with false recall
We turn next to the theoretical implications of ourfindings We first consider the implications for the acti-vationmonitoring framework that we have developed toexplain false recall and false recognition in the DRMparadigm We then consider the implications for othertheories and f inally turn to several further issues onwhich our results shed light
ActivationMonitoring TheoryIn the theroretical framework that we (eg McDer-
mott amp Watson 2001 Roediger Balota amp Watson 2001Roediger amp McDermott 1995 2000) have used to explainvariations in the probability of false recall and false recog-nition in the DRM paradigm two sets of processes havebeen proposed Specifically we have proposed that pro-cesses occurring during both encoding and retrieval af-fect the probability of false recall and false recognitionin the DRM paradigm Two critical sets of processes in-volve activation and monitoring Although there is a nat-ural alignment of activation to encoding and monitoringto retrieval we note that both processes can potentiallyoccur both during encoding and during retrieval as willbe discussed below The ideas we employ have many pre-cursors which we acknowledge below although theirparticular arrangement here may be unique
A critical distinction in understanding encoding pro-cesses and their interaction with retrieval processes isbetween item-specific and relational processing (Ander-son 1972 Hunt amp Einstein 1981 Hunt amp McDaniel1993) Briefly the notion is that emphasis can be given tothe encoding of individual elements and their features(what Anderson called node-taggingand Hunt and his col-leagues called item-specific processing) or to the encod-ing of relationships among elements (pathway-taggingor relational processing) Item-specific processing should
Table 3 Summary of Multiple Regression Analysis
on False Recognition With Critical Item Variablesand List Variables Used as Predictors
NotemdashR2 5 48 F(747) 5 611 MSe 5 002 p lt 05 p lt 05
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 393
make individual elements more memorable and lead tomore distinctive memories such processing should helpdistinguish veridical memories from false memories be-cause list items would have features of presentation as-sociated with them whereas the critical item sometimeswould not (However if the critical item is strongly acti-vated during encoding it may take on many of the fea-tures of list items) Relational processing (extractingthemes or schemas or constructing networks of associa-tions) should increase the probability of false recall andfalse recognition by leading to inferences or associationsof events that may not actually have occurred
We assume that in encoding and interpreting experi-ences people go beyond the information given in thestimulus (Bartlett 1932 chap 2 Bruner 1957) and ac-tivate related information through inferential processingEach personrsquos experience is coded in terms of his or herown schemata (or associative networks or knowledgestructures) When schemata are activated a process offilling in of ldquomissingrdquo information may occur wherebyitems strongly implied by the schema (but not actuallypresent in the stimulus) may become activated (Bartlett1932) In encoding lists of words in the DRM paradigmpresentation of the list can create activation that spreadsthroughout the lexicalsemantic system and can createimplicit associative responses (Underwood 1965) Thisactivation may arouse concepts that are not presented but that are associated to concepts that are presented(eg Anderson amp Pirolli 1984 Collins amp Loftus 1975Meyer amp Schvaneveldt 1971) The associative activa-tion in this approach can arise from automatic process-ing or from more controlled elaborative processing(Neely 1977) In addition the associative activation ofconcepts that were not presented may occur consciously(subjects may become aware of the concept of sleep andrehearse it during presentation of the list words bed restawake etc) or the concept may be strongly activatedbut never come to conscious awareness The debate overwhether arousal of the critical item in the DRM para-digm is conscious or unconscious is interesting but dif-ficult to decide empirically Some evidence has impli-cated conscious arousal of the concepts (McDermott1997) whereas other evidence implicates unconsciousactivation (Seamon Luo amp Gallo 1998) The correctanswer to the question of whether activation is consciousor unconscious is probably ldquobothrdquomdashthat is the criticalitem probably sometimes comes consciously to mindand even if it does not relevant information may never-theless be activated unconsciously
The high correlation between BAS and false recall aswell as the strong weighting of BAS in the multiple re-gression analysis strongly implicates activation of criti-cal items as an important factor in explaining false recalland false recognition in the DRM paradigm The morestrongly associated list items are to critical items themore activation would be expected to spread to the crit-ical items More colloquially to the extent that the wordson the list spark the associative connection to the critical
item the more likely the critical item is to be falsely re-called This activation factor would seem to be one mainreason why ldquothings suggested to a person by an experi-ence may be honestly reported by him as part of the ex-periencerdquo (Kirkpatrick 1894 p 608) These associatedlists probably all tend to elicit relational processing butonly those with strong BAS to the critical item lead tofalse recall
The second set of factors proposed in activationmon-itoring theory is concerned with the monitoring of mem-ory accuracy Monitoring of encoding processes canoccur during the study phase of an experiment espe-cially under intentional learning conditions In experi-ments in which subjects are warned about false memoryphenomena and are told to pay close attention to lists andto remember only the presented items and not the criti-cal items monitoring is enhanced (Gallo Roberts amp Sea-mon 1997 Gallo Roediger amp McDermott 2001 Mc-Dermott amp Roediger 1998) Monitoring of memorieshas typically been discussed with respect to retrieval dur-ing which time a primary goal is to distinguish fromamong the considerable information brought to con-sciousness that which refers to perception of past eventsfrom that which does not a process referred to as realitymonitoring (Johnson amp Raye 1981) The experience ofremembering events involves mentally traveling back intime and seeming to reexperience the events (Tulving1985) During retrieval recovered memories that appar-ently carry with them details of having been experiencedare generally attributed to be memories of actual pastevents although sometimes this attribution may be inerror (Jacoby et al 1989)
Johnson et al (1993) have described in detail the typesof information that may be used to distinguish eventsthat actually happened from ones that were only imag-ined thought about or read about For example eventsthat actually occurred have more features from the ex-ternal world associated with them (sights sounds etc)whereas those that were internally generated are lesslikely to have these features but more likely to have as-sociated reflections and cognitive operations In theDRM paradigm the fact that the critical item has beenstrongly activated during encoding means that duringretrieval the subject is likely to retrieve the critical itemas a list member To the extent that the critical items arestrongly activated during encoding they will take onsome features similar to those of the list items this is es-pecially true if the critical items come consciously tomind during encoding and are rehearsed in the samemanner as list items To the extent that the features of thecritical item are similar to those of actual list items theprobability of false recall or false recognition is increasedHowever if the list items systematically differ from thecritical item on some dimensions or features retrievalheuristics may be brought into play that can reduce falserecall and false recognition (Israel amp Schacter 1997Schacter Israel amp Racine 1999) Considerable evidencesupports this general source-reality-monitoring frame-
394 ROEDIGER WATSON MCDERMOTT AND GALLO
work that we advocate here (eg Johnson et al 1993Schacter et al 1999) as well as the attributional ap-proach to remembering (Jacoby et al 1989)
The other significant factor arising from our multipleregression analysis may operate during retrieval by aid-ing the monitoring processes that reduce false recall Thenegative correlation between veridical recall of list itemsand false recall of the critical item may indicate that thebetter encoded list items are the more easily they can bedistinguished from the illusory critical item Subjectsmay use item-specific information to aid recall of listitems and suppress recall of the critical items whichwould not carry as much item-specific information Whenexperimental manipulations increase veridical recall ei-ther through longer study time at relatively slow rates(Gallo amp Roediger 2001 McDermott amp Watson 2001Toglia Neuschatz amp Goodwin 1999) or through re-peated study and test opportunities (McDermott 1996Schacter et al 1999) false recall is often decreasedTherefore the negative relation between veridical recalland false recall can be produced experimentally as wellas appearing in our correlationregression analysis How-ever veridical and false recall can be positively related as afunction of other variables too as was shown by McDer-mott (1996) Toglia et al (1999) and Thapar amp McDermott(2001) In these cases greater activation of the criticalitem from relational processing or from greater seman-tic processing is probably at work
The fact that the length andor distinctiveness of thecritical item are negatively correlated with false recallcan be similarly interpreted as the effect of monitoringduring retrieval because the greater distinctiveness ofthe critical item simplifies the monitoring process Forexample if words in the list are generally shorter andhave more regular orthography than does the criticalitem then during retrieval the critical items that are longand distinctive (eg butterfly cabbage) should be moreeasily discriminated from list items as indeed seems tobe the case in the bivariate correlations Interestingly thelengthdistinctiveness factor played a role in determin-ing false recall even though the lists in our norms werepresented auditorily and the orthographic informationat least should not be particularly accessible In addi-tion presenting the DRM lists visually rather than audi-torily reduces false recall and false recognition at leaston tests in which recognition items are presented visually(as opposed to auditorily) or recall is written rather thanspoken (Gallo McDermott et al 2001 Kellogg 2001Smith amp Hunt 1998) This modality difference has alsobeen interpreted as arising from the retrieval of distinctiveinformation operating during the test (Gallo McDermottet al 2001 Smith amp Hunt 1998)
As was noted above activation processes may occurduring retrieval as well as during encoding The morelist items a subject recalls the more primed the criticalitem might become However the present analyses un-dercut the simplest form of this idea If the recalling oflist items activated or primed false recall the simplest
prediction would be that there should be a positive cor-relation between veridical recall and false recall Our datarevealed the opposite pattern showing a negative corre-lation between veridical and false recall Marsh McDer-mott and Roediger (2001) reported other experimentsthat also call into question the idea that priming duringthe test plays a role in false recall and false recognition
Balota et al (1999) obtained further evidence sup-porting the activationmonitoring interpretation of falsememories in the DRM paradigm They compared per-formance of younger adults healthy older adults andolder adults with Alzheimerrsquos disease The latter twogroups have been shown to have intact activation pro-cesses as measured by semantic priming paradigms(Balota amp Duchek 1988 1991) yet they are deficient insource-monitoring processes which is probably attribut-able to impaired frontal lobe functioning (see BalotaDolan amp Duchek 2000) Therefore one might expectthat whereas veridical recall would be worse in older adultsand Alzheimerrsquos patients (relative to younger adults)false recall and false recognition would occur at the samelevel or perhaps be increased in these groups IndeedBalota et al (1999) reported exactly this pattern A sim-ilar outcome has been obtained in other laboratories(eg Budsen Daffner Desikan amp Schacter 2000 Nor-man amp Schacter 1997 Tun Wingfield Rosen amp Blan-chard 1998) This pattern is just as predicted by the ac-tivation monitoring framework
In short the present results and many others (seeRoediger Balota amp Watson 2001 Roediger et al 1998)fit comfortably within the activationmonitoring frame-work In the remainder of our discussion we will con-sider briefly the implications of our present results forother accounts of the DRM paradigm and for the role ofconnectivity (or interitem associative strength) in pro-ducing veridical and false recall
Fuzzy-Trace TheoryAnother leading theory that has been used to explain
false memory phenomena is fuzzy-trace theory (egReyna amp Brainerd 1995) Payne et al (1996) and Schac-ter et al (1996) among many others have used the basictheory to account for false recall and false recognitiondata within the DRM paradigm and the theory has beenextended in interesting ways to account for new varia-tions on the paradigm (eg Brainerd Wright Reyna ampMojardin 2001) The fundamental assumption of fuzzy-trace theory is that experience leaves two types of mem-ory traces gist traces (which capture the meaning of ex-perience but without specific attributes) and verbatimtraces (which represent specific attributes) False memo-ries are attributed to gist processing When the meaning ofa recognition probe matches the gist of the encoded expe-rience false recognition occurs The theory has been di-rected primarily to false recognition and not false recall
In trying to explain the present results one difficultyis that the operational definition of gist has never beenspecified that is within the DRM paradigm given a list
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 395
of 15 related words how does one specify the gist repre-sentation of the list and quantify how strong it is Onecould use patterns of false recognitionmdashto the extentthat a concept is falsely recognized it was represented aspart of the gist representation This definition is per-fectly suitable for some purposes but clearly explainingfalse recognition is not one of them owing to the prob-lem of circularity To explain why some lists lead to highlevels of false recall and false recognition (whereas otherlists lead to low levels) within fuzzy-trace theory oneneeds an independent specification of gist which to ourknowledge has not been provided in this context Whydoes the sweet list have so much gist that it leads to falserecall whereas the bitter list does not One plausible in-terpretation of gist is BAS The more the list items areassociated to the critical item the more a gist represen-tation of the critical item is created This solution to adefinition of gist works within the DRM paradigm butof course the interpretation of how gist is created be-comes isomorphic to activation processes within the ac-tivationmonitoring framework
Interpretation of the other factor that was significantwithin our multiple regression analysis may be morestraightforward within fuzzy-trace theory The negativecorrelation between veridical recall and false recall (andfalse recognition) may be accounted for by increased re-liance on verbatim traces That is if increased recall of listitems is due to increased verbatim information subjectsmay rely less heavily on gist information and hence showless false recall and false recognition during respondingSimilarly for critical items that are long andor ortho-graphically distinct recall of list items that relies on thesefeatures may lead to suppression of false recall and falserecognition of the critical items because they may differsystematically from other list items These factors pos-tulated within the context of fuzzy-trace theory are quitesimilar to those within the activationmonitoring frame-work outlined above At some level of generality there aretwo sets of factors in each account and both theories canexplain many empirical results
Shifting Criterion TheoryMiller and Wolford (1999) proposed that false recog-
nition in the DRM paradigm could be explained by sub-jectsrsquo shifting their criteria differentially to list items andto critical items during a recognition test (They ex-tended their explanation to false recall but somewhattenuously by proposing a generaterecognize theory)They postulated that subjects encoded the structure ofthe list during its presentation and then during the recog-nition test responded with especially liberal criteria tocritical items but not to list items The criterion shift issupposed to occur to the degree that the test item appearsto be related to the list items To support their argumentthey reported that a measure of bias from signal detec-tion theory was indeed more liberal for critical itemsthan for list items However Wickens and Hirshman(2000) and Wixted and Stretch (2000) pointed out that
within signal detection theory differences in bias mea-sures do not necessarily implicate a criterion shiftRather differences in measured bias could arise fromdifferences among the underlying distributions Wixtedand Stretch argued that this latter interpretation was themore likely and that the criterion shift interpretation wasuntenable (also see Roediger amp McDermott 1999)
The present evidence also weighs heavily against acriterion shift account Miller and Wolford (1999) ap-pealed to subjectsrsquo metaknowledge of the structure of thelist Within the criterion shift account we can assumethat the likelihood that a critical item cues metaknowl-edge of the list depends on its FAS or how well the crit-ical item will cue the list If so FAS (the association ofthe critical item to the list items) should be the primarydeterminant of false recognition and false recall How-ever the correlation between FAS and false recall in ourdata was negligible (08) and the same was true in analy-ses of false recognition BAS was actually the primarypredictor of false recall which implicates activation pro-cesses during encoding as being a primary cause of theeffect There would be no reason to expect BAS to haveany impact according to the idea that a simple criterionshift during the test causes false recall and false recog-nition Of course all our lists were relatively high androughly equivalent in terms of FAS because they werecreated from forward associations to the critical item sorestriction of range may account for the low correlationEven so the same problem exists for the shifting crite-rion theory to explain the results If FAS is supposed todetermine false recall and false recognition by this ac-count and if all our lists were high and roughly equal onthis feature why were so many lists ineffective in elicitingfalse recognition and false recall In short either way theshifting criterion theory cannot account for the results
Of course within the activationmonitoring frame-work the monitoring processes occurring during retrievalcould be conceived as conservative shifts in response cri-teria However in our theory these monitoring processesdetermine the degree to which a given level of activationat retrieval will be judged sufficient to endorse a criticalitem as having been studied thereby producing a falsememory In contrast in Miller and Wolfordrsquos (1999) ac-count there was no activation of the critical item duringencoding and shifting criteria during the test were saidto account entirely for the false recognition phenomenon(see Wixted amp Stretch 2000) In the activationmonitoringaccount the memory illusion provoked by the DRM listsis assumed to be a subjectively powerful phenomenonthat can be ameliorated (but rarely eliminated) when listitems are made more distinctive relative to criticalitems Under conditions in which the subject is warnedat test (but not before the encoding phase) to monitortheir memories for the presence or absence of the criti-cal item and is discouraged from making criterion shiftsthe false memory effect is neither eliminated nor greatlyreduced (Gallo Roediger amp McDermott 2001) Hicksand Marsh (2001) showed that under conditions in which
396 ROEDIGER WATSON MCDERMOTT AND GALLO
subjects are explicitly told to monitor their memories dur-ing the test (and are asked to judge whether each wordhad been externally presented or internally generated)false recognition in the DRM paradigm actually increasesrather than decreases In fact the false recognition effectpersists even when subjects are fully informed about thenature of the phenomenon before the study episode andare therefore encouraged to enhance monitoring processesduring both encoding and retrieval in order to attempt todiminish or eliminate the effect (see Gallo et al1997Gallo Roediger amp McDermott 2001 McDermott ampRoediger 1998) For these and other reasons the notionthat the DRM memory illusion is caused by a general shiftof criteria can be eliminated as a potential explanation
Interitem Associative Strength or ConnectivityOur data pose something of a puzzle about the role of
interitem associative strength (Deese 1959a) or connec-tivity (McEvoy et al 1999) in false recall As was dis-cussed in the introduction Deese (1959a) found a sig-nificant negative correlation between this construct andthe probability of false recall In a later paper Deese(1961) obtained a weaker negative correlation that failedto reach conventional levels of significance so the issueis somewhat in doubt from Deesersquos (1959a 1961) researchMcEvoy et al created lists that varied in connectivity(interitem associative strength) and in BAS and manip-ulated these factors orthogonally They replicated thefinding of Deese (1959a) in that the density of inter-connections of list words was negatively related to falserecall However in our analysis of 55 lists (different fromthose used by McEvoy et al 1999) we found no relationbetween connectivity and false recall (r 5 04) or be-tween connectivity and false recognition (r 5 03)
Further research will be necessary to determine thetrue state of affairs although a case can be made that thelack of correlation we obtained is accurate First we had55 lists that varied widely in their ability to elicit false re-call as compared with 24 lists used in McEvoy et al(1999) Our lists also represented a relatively full rangeof connectivity or interitem associative strength relativeto those of McEvoy et al McEvoy et al used lists thathad a somewhat wider range of connectivity because theyused extreme values whereas our larger set of lists pro-vided a fuller body of the middle ranges of connectivityGiven our results it seems at least possible that interitemassociative strength (connectivity) is not a factor in de-termining whether lists of associative words lead to falserecall although the issue remains in doubt
We did find that interitem associative strength or con-nectivity is positively and significantly correlated withveridical recall (r 5 132) This outcome is consistentwith that of Deese (1959a) and McEvoy et al (1999)The greater the associative bonds among list items thegreater is list recall This outcome is also consistent withvirtually any associative theory of recall (eg Anderson1972 1983 Nelson Schreiber amp McEvoy 1992)
CONCLUSION
The present study has uncovered two primary factorsthat are significantly related to false recall and falserecognition in the DRM paradigm BAS is positivelycorrelated with false recall whereas veridical recall ofthe list items is negatively related to false recall The ac-tivationmonitoring account provides a natural interpre-tation of these factors Any viable theory posited to ex-plain the occurrence of false memories in the DRMparadigm must explain the variability in false recall andfalse recognition across lists and the two primary factorsresponsible for this variability
REFERENCES
Anderson J R (1972) FRAN A simulation model of free recall InG H Bower (Ed) The psychology of learning and motivation (Vol 5pp 315-378) New York Academic Press
Anderson J R (1983) The architecture of cognition CambridgeMA Harvard University Press
Anderson J R amp Pirolli P L (1984) Spread of activation Journalof Experimental Psychology Learning Memory amp Cognition 10791-798
Balota D A Cortese M J Duchek J M Adams D RoedigerH L III McDermott K B amp Yerys B E (1999) Veridical andfalse memories in healthy older adults and in dementia of the Alz-heimers type Cognitive Neuropsychology 16 361-384
Balota D A Dolan P O amp Duchek J M (2000) Memorychanges in healthy older adults In E Tulving amp F I M Craik (Eds)The Oxford handbook of memory (pp 395-409) Oxford Oxford Uni-versity Press
Balota D A amp Duchek J M (1988) Age-related differences inlexical access spreading activation and simple pronunciation Psy-chology and Aging 3 84-93
Balota D A amp Duchek J (1991) Semantic priming effects lexicalrepetition effects and contextual disambiguation effects in healthyaged individuals and individuals with senile dementia of theAlzheimer type Brain amp Language 40 181-201
Bartlett F C (1932) Remembering A study in experimental and so-cial psychology Cambridge Cambridge University Press
Brainerd C J Wright R Reyna V F amp Mojardin A H (2001)Conjoint recognition and phantom recollection Journal of Experi-mental Psychology Learning Memory amp Cognition 27 307-327
Bruce D amp Winograd E (1998) Remembering Deesersquos 1959 arti-cles The Zeitgeist the sociology of science and false memoriesPsychonomic Bulletin amp Review 5 615-624
Bruner J S (1957) On perceptual readiness Psychological Review64 123-152
Budson A E Daffner K R Desikan R amp Schacter D L(2000) When false recognition is unopposed by true recognitionGist-based memory distortion in Alzheimerrsquos disease Neuropsy-chology 14 277-287
Collins A M amp Loftus E F (1975) A spreading-activation theoryof semantic memory Psychological Review 82 407-428
Coltheart M Davelaar E Jonasson J T amp Besner D (1977)Access to the internal lexicon In S Dormic (Ed) Attention and per-formance VI (pp 535-555) Hillsdale NJ Erlbaum
Deese J (1959a) Influence of interitem associative strength upon im-mediate free recall Psychological Reports 5 235-241
Deese J (1959b) On the prediction of occurrence of particular verbalintrusions in immediate recall Journal of Experimental Psychology58 17-22
Deese J (1961) Associative structure and the serial reproduction ex-periment Journal of Abnormal amp Social Psychology 63 95-100
Deese J (1965) The structure of associations in language and thoughtBaltimore Johns Hopkins University Press
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 397
Gallo D A McDermott K B Percer J M amp Roediger H LIII (2001) Modality effects in false recall and false recognitionJournal of Experimental Psychology Learning Memory amp Cogni-tion 27 339-353
Gallo D A Roberts M J amp Seamon J G (1997) Rememberingwords not presented in lists Can we avoid creating false memoriesPsychonomic Bulletin amp Review 4 271-276
Gallo D A amp Roediger H L III (2001) Variability among wordlists in evoking associative memory illusions Manuscript submittedfor publication
Gallo D A Roediger H L III amp McDermott K B (2001) As-sociative false recognition occurs without strategic criterion shiftsPsychonomic Bulletin amp Review 8 579-586
Hicks J L amp Marsh R L (2001) False recognition occurs more fre-quently during source identification than during oldndashnew recogni-tion Journal of Experimental Psychology Learning Memory ampCognition 27 375-383
Hunt R R amp Einstein G O (1981) Relational and item-specific in-formation in memory Journal of Verbal Learning amp Verbal Behav-ior 19 497-514
Hunt R R amp McDaniel M A (1993) The enigma of organizationand distinctiveness Journal of Memory amp Language 32 421-445
Israel L amp Schacter D L (1997) Pictorial encoding reduces falserecognition of semantic associates Psychonomic Bulletin amp Review4 577-581
Jacoby L L Kelley C M amp Dywan J (1989) Memory attribu-tions In H L Roediger III amp F I M Craik (Eds) Varieties of mem-ory and consciousness Essays in honour of Endel Tulving (pp 391-422) Hillsdale NJ Erlbaum
Johnson M K Hashtroudi S amp Lindsay D S (1993) Sourcemonitoring Psychological Bulletin 114 3-28
Johnson M K amp Raye C L (1981) Reality monitoring Psycho-logical Review 88 67-85
Kellogg RT (2001) Presentation modality and mode of recall in ver-bal false memory Journal of Experimental Psychology LearningMemory amp Cognition 27 913-919
Kirkpatrick E A (1894) An experimental study of memory Psy-chological Review 1 602-609
Ku Iumlcera H amp Francis W (1967) Computational analysis of present-day American English Providence RI Brown University Press
Marsh E J McDermott K B amp Roediger H L III (2001) Doretrieval factors play a role in the creation of of false memories Man-uscript submitted for publication
Mather M Henkel L A amp Johnson M J (1997) Evaluatingcharacteristics of false memories Rememberknow judgments andmemory characteristics questionnaire compared Memory amp Cogni-tion 25 826-837
McDermott K B (1996) The persistence of false memories in list re-call Journal of Memory amp Language 35 212-230
McDermott K B (1997) Priming on perceptual implicit memorytests can be achieved through presentation of associates Psycho-nomic Bulletin amp Review 4 582-586
McDermott K B amp Roediger H L III (1998) Attempting to avoidillusory memories Robust false recognition of associates persistsunder conditions of explicit warnings and immediate testing Journalof Memory amp Language 39 508-520
McDermott K B amp Watson J M (2001) The rise and fall of falserecall The impact of presentation duration Journal of Memory ampLanguage 45 160-176
McEvoy C L Nelson D L amp Komatsu T (1999) What is the con-nection between true and false memories The differential roles of in-teritem associations in recall and recognition Journal of Experimen-tal Psychology Learning Memory amp Cognition 25 1177-1194
Meyer D E amp Schvaneveldt R W (1971) Facilitation in recog-nizing pairs of words Evidence of a dependence between retrievaloperations Journal of Experimental Psychology 90 227-234
Miller M B amp Wolford G L (1999) Theoretical commentaryThe role of criterion shift in false memory Psychological Review106 398-405
Neely J H (1977) Semantic priming and retrieval from semanticmemory Roles of inhibitionless spreading activation and limited-
capacity attention Journal of Experimental Psychology General106 226-254
Nelson D L McEvoy C L amp Schreiber T A (1999) The Uni-versity of South Florida word association rhyme and word fragmentnorms Unpublished manuscript University of South Florida Tampa
Nelson D L Schreiber TA amp McEvoy C L (1992) Processing im-plicit and explicit representations Psychological Review 99 322-348
Norman K A amp Schacter D L (1997) False recognition in youngerand older adults Exploring the characteristics of illusory memoriesMemory amp Cognition 25 838-848
Paivio A Yuille J C amp Madigan S A (1968) Concreteness im-agery and meaningfulness values for 925 nouns Journal of Experi-mental Psychology Monographs 76 (1 Pt 2)
Payne D G Elie C J Blackwell J M amp Neuschatz J S (1996)Memory illusions Recalling recognizing and recollecting events thatnever occurred Journal of Memory amp Language 35 261-285
Reyna V F amp Brainerd C J (1995) Fuzzy-trace theory An interimsynthesis Learning amp Individual Differences 7 1-75
Roediger H L III Balota D A amp Watson J M (2001) Spreadingactivation and the arousal of false memories In H L Roediger III J SNairne I Neath amp A M Surprenant (Eds) The nature of remem-bering Essays in honor of Robert G Crowder (pp 95-115) Wash-ington DC American Psychological Association
Roediger H L III amp McDermott K B (1995) Creating false mem-ories Remembering words not presented in lists Journal of Experi-mental Psychology Learning Memory amp Cognition 21 803-814
Roediger H L III amp McDermott K B (1999) False alarms aboutfalse memories Psychological Review 106 406-410
Roediger H L III amp McDermott K B (2000) Tricks of memoryCurrent Directions in Psychological Science 9 123-127
Roediger H L III McDermott K B amp Pisoni D B (2001) Rec-ollection of illusory voices Manuscript submitted for publication
Roediger H L III McDermott K B amp Robinson K J (1998)The role of associative processes in creating false memories In M AConway S E Gathercole amp C Cornoldi (Eds) Theories of memoryII (pp 187-245) Hove UK Psychological Press
Russell W A amp Jenkins J J (1954) The complete Minnesotanorms for responses to 100 words from the KentndashRosanoff Word As-sociation Test (Tech Rep No 11 Contract N8 ONR 66216 Officeof Naval Research) University of Minnesota Minneapolis
Schacter D L Israel L amp Racine C (1999) Suppressing falserecognition in younger and older adults The distinctiveness heuris-tic Journal of Memory amp Language 40 1-24
Schacter D L Verfaellie M amp Pradere D (1996) The neuro-psychology of memory illusions False recall and recognition in am-nesic patients Journal of Memory amp Language 35 319-334
Seamon J G Luo C R amp Gallo D A (1998) Creating false mem-ories of words with or without recognition of list items Evidence fornonconscious processes Psychological Science 9 20-26
Smith R E amp Hunt R R (1998) Presentation modality affects falsememory Psychonomic Bulletin amp Review 5 710-715
Stadler M A Roediger H L III amp McDermott K B (1999)Norms for word lists that create false memories Memory amp Cogni-tion 27 494-500
Thapar A amp McDermott K B (2001) False recall and false recog-nition induced by presentation of associated words Effects of retentioninterval and level of processing Memory amp Cognition 29 424-432
Toglia M P amp Battig W F (1978) Handbook of semantic wordnorms Hillsdale NJ Erlbaum
Toglia M P Neuschatz J S amp Goodwin K A (1999) Recall ac-curacy and illusory memories When more is less Memory 7 233-256
Tulving E (1974) Cue-dependent forgetting American Scientist 6274-82
Tulving E (1979) Relation between encoding specificity and levelsof processing In L S Cermak amp F I M Craik (Eds) Levels of pro-cessing in human memory (pp 405-428) Hillsdale NJ Erlbaum
Tulving E (1985) Memory and consciousness Canadian Psycholo-gist 26 1-12
Tun P A Wingfield A Rosen M J amp Blanchard L (1998)Response latencies for false memories Gist-based processes in nor-mal aging Psychology amp Aging 13 230-241
398 ROEDIGER WATSON MCDERMOTT AND GALLO
Underwood B J (1965) False recognition produced by implicit ver-bal responses Journal of Experimental Psychology 70 122-129
Weldon M S (1991) Mechanisms underlying priming on perceptualtests Journal of Experimental Psychology Learning Memory ampCognition 17 526-541
Wickens T D amp Hirshman E (2000) False memories and statisticaldecision theory Comment on Miller and Wolford (1999) and Roedi-ger and McDermott (1999) Psychological Review 107 377-383
Wixted J T amp Stretch V (2000) The case against a criterion-shiftaccount of false memory Psychological Review 107 368-376
Zechmeister E B (1969) Orthographic distinctiveness Journal ofVerbal Learning amp Verbal Behavior 8 754-761
NOTES
1 The orthographic distinctiveness of a critical item was obtained bypresenting 72 subjects with all 55 critical items visually and havingthem rate the words in accordance with procedures developed by Zech-meister (1969) Briefly the subjects were instructed to rate each wordfor its physical appearance in terms of distinctiveness in the languageon a 9-point scale with 1 being not very distinctive 5 corresponding toaverage distinctiveness and 9 being most distinctive To avoid orderingeffects three different randomized orderings of words were used acrossroughly equal groups of subjects In the case of the critical item sleep
for example the distinctiveness is 515 Corresponding values for but-terfly and for man are 744 and 213 respectively Although ortho-graphic distinctiveness was not included in our multiple regressionanalyses this predictor is included in Appendix C
2 The orthographic neighborhood size or Coltheartrsquos N of each crit-ical item was defined as the number of words that can be formed bychanging a single letter in the word (Coltheart Davelaar Jonasson ampBesner 1977) For example because it has no orthographic neighborsColtheartrsquos N for the critical item butterfly is zero In contrast sleep hasfive orthographic neighbors (ie sheep sleek sleet steep and sweep)The reference set of words used were those in Ku Iumlcera and Francisrsquos(1967) norms Although Coltheartrsquos N was not included in our multipleregression analyses this predictor is included in Appendix C
3 Word length orthographic distinctiveness and orthographic neigh-borhood size of the critical item were all highly correlated Specificallythe bivariate correlations among these three predictors were as followslength and orthographic distinctiveness r 5 184 t(53) 5 1117 p lt05 orthographic distinctiveness and orthographic neighborhood sizer 5 66 t(53) 5 640 p lt 05 length and orthographic neighbor-hood size r 5 63 t(53) 5 582 p lt 05 Of these three variablesword length had the largest bivariate correlation with false recall (r 5
37) so it was the predictor used in the multiple regression analyses4 The multiple regression data matrix is available at wwwiacwustl
edunmclWeb
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 399
APPENDIX A The Fifty-Five 15-Word Lists and Their Associated Critical Items
With Backward Associative Strength (BAS) and Forward Associative Strength (FAS)Values Used in the Multiple Regression Analysis of False Recall
Item Word BAS FAS Item Word BAS FAS
Anger mad 393 412 Black white 655 557fear 020 059 dark 111 100hate 028 109 cat 000 043rage 541 042 charred 023 000temper 182 000 night 000 021fury 306 000 funeral 034 000ire 179 000 color 074 050wrath 128 000 grief 000 000happy 000 042 blue 028 000fight 034 000 death 016 014hatred 070 000 ink 020 000mean 090 000 bottom 000 000calm 000 000 coal 288 000emotion 000 000 brown 338 000enrage 378 000 gray 365 000
Mean 157 044 Mean 130 052Army Navy 543 500 Bread butter 364 487
NotemdashUnless otherwise indicated with an asterisk all BAS and FAS values are drawn from the associative norms of Nel-son McEvoy and Schreiber (1999) Those values shown with an asterisk were obtained by the authors by using normingprocedures similar to those of Nelson et al (1999) The corresponding mean for these BAS and FAS values for each list arealso represented
406 ROEDIGER WATSON MCDERMOTT AND GALLO
APPENDIX B Data Matrix of the Fifty-Five 15-Word Lists and Associated Critical Items
Used in the Multiple Regression Analysis of False Recall
Critical False False Log Vrd Raw Ortho Item Recall Recgn Length Freq Concrete FAS BAS Connect Recall Freq Dist Colt N
(Manuscript received July 10 2000revision accepted for publication February 23 2001)
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 393
make individual elements more memorable and lead tomore distinctive memories such processing should helpdistinguish veridical memories from false memories be-cause list items would have features of presentation as-sociated with them whereas the critical item sometimeswould not (However if the critical item is strongly acti-vated during encoding it may take on many of the fea-tures of list items) Relational processing (extractingthemes or schemas or constructing networks of associa-tions) should increase the probability of false recall andfalse recognition by leading to inferences or associationsof events that may not actually have occurred
We assume that in encoding and interpreting experi-ences people go beyond the information given in thestimulus (Bartlett 1932 chap 2 Bruner 1957) and ac-tivate related information through inferential processingEach personrsquos experience is coded in terms of his or herown schemata (or associative networks or knowledgestructures) When schemata are activated a process offilling in of ldquomissingrdquo information may occur wherebyitems strongly implied by the schema (but not actuallypresent in the stimulus) may become activated (Bartlett1932) In encoding lists of words in the DRM paradigmpresentation of the list can create activation that spreadsthroughout the lexicalsemantic system and can createimplicit associative responses (Underwood 1965) Thisactivation may arouse concepts that are not presented but that are associated to concepts that are presented(eg Anderson amp Pirolli 1984 Collins amp Loftus 1975Meyer amp Schvaneveldt 1971) The associative activa-tion in this approach can arise from automatic process-ing or from more controlled elaborative processing(Neely 1977) In addition the associative activation ofconcepts that were not presented may occur consciously(subjects may become aware of the concept of sleep andrehearse it during presentation of the list words bed restawake etc) or the concept may be strongly activatedbut never come to conscious awareness The debate overwhether arousal of the critical item in the DRM para-digm is conscious or unconscious is interesting but dif-ficult to decide empirically Some evidence has impli-cated conscious arousal of the concepts (McDermott1997) whereas other evidence implicates unconsciousactivation (Seamon Luo amp Gallo 1998) The correctanswer to the question of whether activation is consciousor unconscious is probably ldquobothrdquomdashthat is the criticalitem probably sometimes comes consciously to mindand even if it does not relevant information may never-theless be activated unconsciously
The high correlation between BAS and false recall aswell as the strong weighting of BAS in the multiple re-gression analysis strongly implicates activation of criti-cal items as an important factor in explaining false recalland false recognition in the DRM paradigm The morestrongly associated list items are to critical items themore activation would be expected to spread to the crit-ical items More colloquially to the extent that the wordson the list spark the associative connection to the critical
item the more likely the critical item is to be falsely re-called This activation factor would seem to be one mainreason why ldquothings suggested to a person by an experi-ence may be honestly reported by him as part of the ex-periencerdquo (Kirkpatrick 1894 p 608) These associatedlists probably all tend to elicit relational processing butonly those with strong BAS to the critical item lead tofalse recall
The second set of factors proposed in activationmon-itoring theory is concerned with the monitoring of mem-ory accuracy Monitoring of encoding processes canoccur during the study phase of an experiment espe-cially under intentional learning conditions In experi-ments in which subjects are warned about false memoryphenomena and are told to pay close attention to lists andto remember only the presented items and not the criti-cal items monitoring is enhanced (Gallo Roberts amp Sea-mon 1997 Gallo Roediger amp McDermott 2001 Mc-Dermott amp Roediger 1998) Monitoring of memorieshas typically been discussed with respect to retrieval dur-ing which time a primary goal is to distinguish fromamong the considerable information brought to con-sciousness that which refers to perception of past eventsfrom that which does not a process referred to as realitymonitoring (Johnson amp Raye 1981) The experience ofremembering events involves mentally traveling back intime and seeming to reexperience the events (Tulving1985) During retrieval recovered memories that appar-ently carry with them details of having been experiencedare generally attributed to be memories of actual pastevents although sometimes this attribution may be inerror (Jacoby et al 1989)
Johnson et al (1993) have described in detail the typesof information that may be used to distinguish eventsthat actually happened from ones that were only imag-ined thought about or read about For example eventsthat actually occurred have more features from the ex-ternal world associated with them (sights sounds etc)whereas those that were internally generated are lesslikely to have these features but more likely to have as-sociated reflections and cognitive operations In theDRM paradigm the fact that the critical item has beenstrongly activated during encoding means that duringretrieval the subject is likely to retrieve the critical itemas a list member To the extent that the critical items arestrongly activated during encoding they will take onsome features similar to those of the list items this is es-pecially true if the critical items come consciously tomind during encoding and are rehearsed in the samemanner as list items To the extent that the features of thecritical item are similar to those of actual list items theprobability of false recall or false recognition is increasedHowever if the list items systematically differ from thecritical item on some dimensions or features retrievalheuristics may be brought into play that can reduce falserecall and false recognition (Israel amp Schacter 1997Schacter Israel amp Racine 1999) Considerable evidencesupports this general source-reality-monitoring frame-
394 ROEDIGER WATSON MCDERMOTT AND GALLO
work that we advocate here (eg Johnson et al 1993Schacter et al 1999) as well as the attributional ap-proach to remembering (Jacoby et al 1989)
The other significant factor arising from our multipleregression analysis may operate during retrieval by aid-ing the monitoring processes that reduce false recall Thenegative correlation between veridical recall of list itemsand false recall of the critical item may indicate that thebetter encoded list items are the more easily they can bedistinguished from the illusory critical item Subjectsmay use item-specific information to aid recall of listitems and suppress recall of the critical items whichwould not carry as much item-specific information Whenexperimental manipulations increase veridical recall ei-ther through longer study time at relatively slow rates(Gallo amp Roediger 2001 McDermott amp Watson 2001Toglia Neuschatz amp Goodwin 1999) or through re-peated study and test opportunities (McDermott 1996Schacter et al 1999) false recall is often decreasedTherefore the negative relation between veridical recalland false recall can be produced experimentally as wellas appearing in our correlationregression analysis How-ever veridical and false recall can be positively related as afunction of other variables too as was shown by McDer-mott (1996) Toglia et al (1999) and Thapar amp McDermott(2001) In these cases greater activation of the criticalitem from relational processing or from greater seman-tic processing is probably at work
The fact that the length andor distinctiveness of thecritical item are negatively correlated with false recallcan be similarly interpreted as the effect of monitoringduring retrieval because the greater distinctiveness ofthe critical item simplifies the monitoring process Forexample if words in the list are generally shorter andhave more regular orthography than does the criticalitem then during retrieval the critical items that are longand distinctive (eg butterfly cabbage) should be moreeasily discriminated from list items as indeed seems tobe the case in the bivariate correlations Interestingly thelengthdistinctiveness factor played a role in determin-ing false recall even though the lists in our norms werepresented auditorily and the orthographic informationat least should not be particularly accessible In addi-tion presenting the DRM lists visually rather than audi-torily reduces false recall and false recognition at leaston tests in which recognition items are presented visually(as opposed to auditorily) or recall is written rather thanspoken (Gallo McDermott et al 2001 Kellogg 2001Smith amp Hunt 1998) This modality difference has alsobeen interpreted as arising from the retrieval of distinctiveinformation operating during the test (Gallo McDermottet al 2001 Smith amp Hunt 1998)
As was noted above activation processes may occurduring retrieval as well as during encoding The morelist items a subject recalls the more primed the criticalitem might become However the present analyses un-dercut the simplest form of this idea If the recalling oflist items activated or primed false recall the simplest
prediction would be that there should be a positive cor-relation between veridical recall and false recall Our datarevealed the opposite pattern showing a negative corre-lation between veridical and false recall Marsh McDer-mott and Roediger (2001) reported other experimentsthat also call into question the idea that priming duringthe test plays a role in false recall and false recognition
Balota et al (1999) obtained further evidence sup-porting the activationmonitoring interpretation of falsememories in the DRM paradigm They compared per-formance of younger adults healthy older adults andolder adults with Alzheimerrsquos disease The latter twogroups have been shown to have intact activation pro-cesses as measured by semantic priming paradigms(Balota amp Duchek 1988 1991) yet they are deficient insource-monitoring processes which is probably attribut-able to impaired frontal lobe functioning (see BalotaDolan amp Duchek 2000) Therefore one might expectthat whereas veridical recall would be worse in older adultsand Alzheimerrsquos patients (relative to younger adults)false recall and false recognition would occur at the samelevel or perhaps be increased in these groups IndeedBalota et al (1999) reported exactly this pattern A sim-ilar outcome has been obtained in other laboratories(eg Budsen Daffner Desikan amp Schacter 2000 Nor-man amp Schacter 1997 Tun Wingfield Rosen amp Blan-chard 1998) This pattern is just as predicted by the ac-tivation monitoring framework
In short the present results and many others (seeRoediger Balota amp Watson 2001 Roediger et al 1998)fit comfortably within the activationmonitoring frame-work In the remainder of our discussion we will con-sider briefly the implications of our present results forother accounts of the DRM paradigm and for the role ofconnectivity (or interitem associative strength) in pro-ducing veridical and false recall
Fuzzy-Trace TheoryAnother leading theory that has been used to explain
false memory phenomena is fuzzy-trace theory (egReyna amp Brainerd 1995) Payne et al (1996) and Schac-ter et al (1996) among many others have used the basictheory to account for false recall and false recognitiondata within the DRM paradigm and the theory has beenextended in interesting ways to account for new varia-tions on the paradigm (eg Brainerd Wright Reyna ampMojardin 2001) The fundamental assumption of fuzzy-trace theory is that experience leaves two types of mem-ory traces gist traces (which capture the meaning of ex-perience but without specific attributes) and verbatimtraces (which represent specific attributes) False memo-ries are attributed to gist processing When the meaning ofa recognition probe matches the gist of the encoded expe-rience false recognition occurs The theory has been di-rected primarily to false recognition and not false recall
In trying to explain the present results one difficultyis that the operational definition of gist has never beenspecified that is within the DRM paradigm given a list
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 395
of 15 related words how does one specify the gist repre-sentation of the list and quantify how strong it is Onecould use patterns of false recognitionmdashto the extentthat a concept is falsely recognized it was represented aspart of the gist representation This definition is per-fectly suitable for some purposes but clearly explainingfalse recognition is not one of them owing to the prob-lem of circularity To explain why some lists lead to highlevels of false recall and false recognition (whereas otherlists lead to low levels) within fuzzy-trace theory oneneeds an independent specification of gist which to ourknowledge has not been provided in this context Whydoes the sweet list have so much gist that it leads to falserecall whereas the bitter list does not One plausible in-terpretation of gist is BAS The more the list items areassociated to the critical item the more a gist represen-tation of the critical item is created This solution to adefinition of gist works within the DRM paradigm butof course the interpretation of how gist is created be-comes isomorphic to activation processes within the ac-tivationmonitoring framework
Interpretation of the other factor that was significantwithin our multiple regression analysis may be morestraightforward within fuzzy-trace theory The negativecorrelation between veridical recall and false recall (andfalse recognition) may be accounted for by increased re-liance on verbatim traces That is if increased recall of listitems is due to increased verbatim information subjectsmay rely less heavily on gist information and hence showless false recall and false recognition during respondingSimilarly for critical items that are long andor ortho-graphically distinct recall of list items that relies on thesefeatures may lead to suppression of false recall and falserecognition of the critical items because they may differsystematically from other list items These factors pos-tulated within the context of fuzzy-trace theory are quitesimilar to those within the activationmonitoring frame-work outlined above At some level of generality there aretwo sets of factors in each account and both theories canexplain many empirical results
Shifting Criterion TheoryMiller and Wolford (1999) proposed that false recog-
nition in the DRM paradigm could be explained by sub-jectsrsquo shifting their criteria differentially to list items andto critical items during a recognition test (They ex-tended their explanation to false recall but somewhattenuously by proposing a generaterecognize theory)They postulated that subjects encoded the structure ofthe list during its presentation and then during the recog-nition test responded with especially liberal criteria tocritical items but not to list items The criterion shift issupposed to occur to the degree that the test item appearsto be related to the list items To support their argumentthey reported that a measure of bias from signal detec-tion theory was indeed more liberal for critical itemsthan for list items However Wickens and Hirshman(2000) and Wixted and Stretch (2000) pointed out that
within signal detection theory differences in bias mea-sures do not necessarily implicate a criterion shiftRather differences in measured bias could arise fromdifferences among the underlying distributions Wixtedand Stretch argued that this latter interpretation was themore likely and that the criterion shift interpretation wasuntenable (also see Roediger amp McDermott 1999)
The present evidence also weighs heavily against acriterion shift account Miller and Wolford (1999) ap-pealed to subjectsrsquo metaknowledge of the structure of thelist Within the criterion shift account we can assumethat the likelihood that a critical item cues metaknowl-edge of the list depends on its FAS or how well the crit-ical item will cue the list If so FAS (the association ofthe critical item to the list items) should be the primarydeterminant of false recognition and false recall How-ever the correlation between FAS and false recall in ourdata was negligible (08) and the same was true in analy-ses of false recognition BAS was actually the primarypredictor of false recall which implicates activation pro-cesses during encoding as being a primary cause of theeffect There would be no reason to expect BAS to haveany impact according to the idea that a simple criterionshift during the test causes false recall and false recog-nition Of course all our lists were relatively high androughly equivalent in terms of FAS because they werecreated from forward associations to the critical item sorestriction of range may account for the low correlationEven so the same problem exists for the shifting crite-rion theory to explain the results If FAS is supposed todetermine false recall and false recognition by this ac-count and if all our lists were high and roughly equal onthis feature why were so many lists ineffective in elicitingfalse recognition and false recall In short either way theshifting criterion theory cannot account for the results
Of course within the activationmonitoring frame-work the monitoring processes occurring during retrievalcould be conceived as conservative shifts in response cri-teria However in our theory these monitoring processesdetermine the degree to which a given level of activationat retrieval will be judged sufficient to endorse a criticalitem as having been studied thereby producing a falsememory In contrast in Miller and Wolfordrsquos (1999) ac-count there was no activation of the critical item duringencoding and shifting criteria during the test were saidto account entirely for the false recognition phenomenon(see Wixted amp Stretch 2000) In the activationmonitoringaccount the memory illusion provoked by the DRM listsis assumed to be a subjectively powerful phenomenonthat can be ameliorated (but rarely eliminated) when listitems are made more distinctive relative to criticalitems Under conditions in which the subject is warnedat test (but not before the encoding phase) to monitortheir memories for the presence or absence of the criti-cal item and is discouraged from making criterion shiftsthe false memory effect is neither eliminated nor greatlyreduced (Gallo Roediger amp McDermott 2001) Hicksand Marsh (2001) showed that under conditions in which
396 ROEDIGER WATSON MCDERMOTT AND GALLO
subjects are explicitly told to monitor their memories dur-ing the test (and are asked to judge whether each wordhad been externally presented or internally generated)false recognition in the DRM paradigm actually increasesrather than decreases In fact the false recognition effectpersists even when subjects are fully informed about thenature of the phenomenon before the study episode andare therefore encouraged to enhance monitoring processesduring both encoding and retrieval in order to attempt todiminish or eliminate the effect (see Gallo et al1997Gallo Roediger amp McDermott 2001 McDermott ampRoediger 1998) For these and other reasons the notionthat the DRM memory illusion is caused by a general shiftof criteria can be eliminated as a potential explanation
Interitem Associative Strength or ConnectivityOur data pose something of a puzzle about the role of
interitem associative strength (Deese 1959a) or connec-tivity (McEvoy et al 1999) in false recall As was dis-cussed in the introduction Deese (1959a) found a sig-nificant negative correlation between this construct andthe probability of false recall In a later paper Deese(1961) obtained a weaker negative correlation that failedto reach conventional levels of significance so the issueis somewhat in doubt from Deesersquos (1959a 1961) researchMcEvoy et al created lists that varied in connectivity(interitem associative strength) and in BAS and manip-ulated these factors orthogonally They replicated thefinding of Deese (1959a) in that the density of inter-connections of list words was negatively related to falserecall However in our analysis of 55 lists (different fromthose used by McEvoy et al 1999) we found no relationbetween connectivity and false recall (r 5 04) or be-tween connectivity and false recognition (r 5 03)
Further research will be necessary to determine thetrue state of affairs although a case can be made that thelack of correlation we obtained is accurate First we had55 lists that varied widely in their ability to elicit false re-call as compared with 24 lists used in McEvoy et al(1999) Our lists also represented a relatively full rangeof connectivity or interitem associative strength relativeto those of McEvoy et al McEvoy et al used lists thathad a somewhat wider range of connectivity because theyused extreme values whereas our larger set of lists pro-vided a fuller body of the middle ranges of connectivityGiven our results it seems at least possible that interitemassociative strength (connectivity) is not a factor in de-termining whether lists of associative words lead to falserecall although the issue remains in doubt
We did find that interitem associative strength or con-nectivity is positively and significantly correlated withveridical recall (r 5 132) This outcome is consistentwith that of Deese (1959a) and McEvoy et al (1999)The greater the associative bonds among list items thegreater is list recall This outcome is also consistent withvirtually any associative theory of recall (eg Anderson1972 1983 Nelson Schreiber amp McEvoy 1992)
CONCLUSION
The present study has uncovered two primary factorsthat are significantly related to false recall and falserecognition in the DRM paradigm BAS is positivelycorrelated with false recall whereas veridical recall ofthe list items is negatively related to false recall The ac-tivationmonitoring account provides a natural interpre-tation of these factors Any viable theory posited to ex-plain the occurrence of false memories in the DRMparadigm must explain the variability in false recall andfalse recognition across lists and the two primary factorsresponsible for this variability
REFERENCES
Anderson J R (1972) FRAN A simulation model of free recall InG H Bower (Ed) The psychology of learning and motivation (Vol 5pp 315-378) New York Academic Press
Anderson J R (1983) The architecture of cognition CambridgeMA Harvard University Press
Anderson J R amp Pirolli P L (1984) Spread of activation Journalof Experimental Psychology Learning Memory amp Cognition 10791-798
Balota D A Cortese M J Duchek J M Adams D RoedigerH L III McDermott K B amp Yerys B E (1999) Veridical andfalse memories in healthy older adults and in dementia of the Alz-heimers type Cognitive Neuropsychology 16 361-384
Balota D A Dolan P O amp Duchek J M (2000) Memorychanges in healthy older adults In E Tulving amp F I M Craik (Eds)The Oxford handbook of memory (pp 395-409) Oxford Oxford Uni-versity Press
Balota D A amp Duchek J M (1988) Age-related differences inlexical access spreading activation and simple pronunciation Psy-chology and Aging 3 84-93
Balota D A amp Duchek J (1991) Semantic priming effects lexicalrepetition effects and contextual disambiguation effects in healthyaged individuals and individuals with senile dementia of theAlzheimer type Brain amp Language 40 181-201
Bartlett F C (1932) Remembering A study in experimental and so-cial psychology Cambridge Cambridge University Press
Brainerd C J Wright R Reyna V F amp Mojardin A H (2001)Conjoint recognition and phantom recollection Journal of Experi-mental Psychology Learning Memory amp Cognition 27 307-327
Bruce D amp Winograd E (1998) Remembering Deesersquos 1959 arti-cles The Zeitgeist the sociology of science and false memoriesPsychonomic Bulletin amp Review 5 615-624
Bruner J S (1957) On perceptual readiness Psychological Review64 123-152
Budson A E Daffner K R Desikan R amp Schacter D L(2000) When false recognition is unopposed by true recognitionGist-based memory distortion in Alzheimerrsquos disease Neuropsy-chology 14 277-287
Collins A M amp Loftus E F (1975) A spreading-activation theoryof semantic memory Psychological Review 82 407-428
Coltheart M Davelaar E Jonasson J T amp Besner D (1977)Access to the internal lexicon In S Dormic (Ed) Attention and per-formance VI (pp 535-555) Hillsdale NJ Erlbaum
Deese J (1959a) Influence of interitem associative strength upon im-mediate free recall Psychological Reports 5 235-241
Deese J (1959b) On the prediction of occurrence of particular verbalintrusions in immediate recall Journal of Experimental Psychology58 17-22
Deese J (1961) Associative structure and the serial reproduction ex-periment Journal of Abnormal amp Social Psychology 63 95-100
Deese J (1965) The structure of associations in language and thoughtBaltimore Johns Hopkins University Press
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 397
Gallo D A McDermott K B Percer J M amp Roediger H LIII (2001) Modality effects in false recall and false recognitionJournal of Experimental Psychology Learning Memory amp Cogni-tion 27 339-353
Gallo D A Roberts M J amp Seamon J G (1997) Rememberingwords not presented in lists Can we avoid creating false memoriesPsychonomic Bulletin amp Review 4 271-276
Gallo D A amp Roediger H L III (2001) Variability among wordlists in evoking associative memory illusions Manuscript submittedfor publication
Gallo D A Roediger H L III amp McDermott K B (2001) As-sociative false recognition occurs without strategic criterion shiftsPsychonomic Bulletin amp Review 8 579-586
Hicks J L amp Marsh R L (2001) False recognition occurs more fre-quently during source identification than during oldndashnew recogni-tion Journal of Experimental Psychology Learning Memory ampCognition 27 375-383
Hunt R R amp Einstein G O (1981) Relational and item-specific in-formation in memory Journal of Verbal Learning amp Verbal Behav-ior 19 497-514
Hunt R R amp McDaniel M A (1993) The enigma of organizationand distinctiveness Journal of Memory amp Language 32 421-445
Israel L amp Schacter D L (1997) Pictorial encoding reduces falserecognition of semantic associates Psychonomic Bulletin amp Review4 577-581
Jacoby L L Kelley C M amp Dywan J (1989) Memory attribu-tions In H L Roediger III amp F I M Craik (Eds) Varieties of mem-ory and consciousness Essays in honour of Endel Tulving (pp 391-422) Hillsdale NJ Erlbaum
Johnson M K Hashtroudi S amp Lindsay D S (1993) Sourcemonitoring Psychological Bulletin 114 3-28
Johnson M K amp Raye C L (1981) Reality monitoring Psycho-logical Review 88 67-85
Kellogg RT (2001) Presentation modality and mode of recall in ver-bal false memory Journal of Experimental Psychology LearningMemory amp Cognition 27 913-919
Kirkpatrick E A (1894) An experimental study of memory Psy-chological Review 1 602-609
Ku Iumlcera H amp Francis W (1967) Computational analysis of present-day American English Providence RI Brown University Press
Marsh E J McDermott K B amp Roediger H L III (2001) Doretrieval factors play a role in the creation of of false memories Man-uscript submitted for publication
Mather M Henkel L A amp Johnson M J (1997) Evaluatingcharacteristics of false memories Rememberknow judgments andmemory characteristics questionnaire compared Memory amp Cogni-tion 25 826-837
McDermott K B (1996) The persistence of false memories in list re-call Journal of Memory amp Language 35 212-230
McDermott K B (1997) Priming on perceptual implicit memorytests can be achieved through presentation of associates Psycho-nomic Bulletin amp Review 4 582-586
McDermott K B amp Roediger H L III (1998) Attempting to avoidillusory memories Robust false recognition of associates persistsunder conditions of explicit warnings and immediate testing Journalof Memory amp Language 39 508-520
McDermott K B amp Watson J M (2001) The rise and fall of falserecall The impact of presentation duration Journal of Memory ampLanguage 45 160-176
McEvoy C L Nelson D L amp Komatsu T (1999) What is the con-nection between true and false memories The differential roles of in-teritem associations in recall and recognition Journal of Experimen-tal Psychology Learning Memory amp Cognition 25 1177-1194
Meyer D E amp Schvaneveldt R W (1971) Facilitation in recog-nizing pairs of words Evidence of a dependence between retrievaloperations Journal of Experimental Psychology 90 227-234
Miller M B amp Wolford G L (1999) Theoretical commentaryThe role of criterion shift in false memory Psychological Review106 398-405
Neely J H (1977) Semantic priming and retrieval from semanticmemory Roles of inhibitionless spreading activation and limited-
capacity attention Journal of Experimental Psychology General106 226-254
Nelson D L McEvoy C L amp Schreiber T A (1999) The Uni-versity of South Florida word association rhyme and word fragmentnorms Unpublished manuscript University of South Florida Tampa
Nelson D L Schreiber TA amp McEvoy C L (1992) Processing im-plicit and explicit representations Psychological Review 99 322-348
Norman K A amp Schacter D L (1997) False recognition in youngerand older adults Exploring the characteristics of illusory memoriesMemory amp Cognition 25 838-848
Paivio A Yuille J C amp Madigan S A (1968) Concreteness im-agery and meaningfulness values for 925 nouns Journal of Experi-mental Psychology Monographs 76 (1 Pt 2)
Payne D G Elie C J Blackwell J M amp Neuschatz J S (1996)Memory illusions Recalling recognizing and recollecting events thatnever occurred Journal of Memory amp Language 35 261-285
Reyna V F amp Brainerd C J (1995) Fuzzy-trace theory An interimsynthesis Learning amp Individual Differences 7 1-75
Roediger H L III Balota D A amp Watson J M (2001) Spreadingactivation and the arousal of false memories In H L Roediger III J SNairne I Neath amp A M Surprenant (Eds) The nature of remem-bering Essays in honor of Robert G Crowder (pp 95-115) Wash-ington DC American Psychological Association
Roediger H L III amp McDermott K B (1995) Creating false mem-ories Remembering words not presented in lists Journal of Experi-mental Psychology Learning Memory amp Cognition 21 803-814
Roediger H L III amp McDermott K B (1999) False alarms aboutfalse memories Psychological Review 106 406-410
Roediger H L III amp McDermott K B (2000) Tricks of memoryCurrent Directions in Psychological Science 9 123-127
Roediger H L III McDermott K B amp Pisoni D B (2001) Rec-ollection of illusory voices Manuscript submitted for publication
Roediger H L III McDermott K B amp Robinson K J (1998)The role of associative processes in creating false memories In M AConway S E Gathercole amp C Cornoldi (Eds) Theories of memoryII (pp 187-245) Hove UK Psychological Press
Russell W A amp Jenkins J J (1954) The complete Minnesotanorms for responses to 100 words from the KentndashRosanoff Word As-sociation Test (Tech Rep No 11 Contract N8 ONR 66216 Officeof Naval Research) University of Minnesota Minneapolis
Schacter D L Israel L amp Racine C (1999) Suppressing falserecognition in younger and older adults The distinctiveness heuris-tic Journal of Memory amp Language 40 1-24
Schacter D L Verfaellie M amp Pradere D (1996) The neuro-psychology of memory illusions False recall and recognition in am-nesic patients Journal of Memory amp Language 35 319-334
Seamon J G Luo C R amp Gallo D A (1998) Creating false mem-ories of words with or without recognition of list items Evidence fornonconscious processes Psychological Science 9 20-26
Smith R E amp Hunt R R (1998) Presentation modality affects falsememory Psychonomic Bulletin amp Review 5 710-715
Stadler M A Roediger H L III amp McDermott K B (1999)Norms for word lists that create false memories Memory amp Cogni-tion 27 494-500
Thapar A amp McDermott K B (2001) False recall and false recog-nition induced by presentation of associated words Effects of retentioninterval and level of processing Memory amp Cognition 29 424-432
Toglia M P amp Battig W F (1978) Handbook of semantic wordnorms Hillsdale NJ Erlbaum
Toglia M P Neuschatz J S amp Goodwin K A (1999) Recall ac-curacy and illusory memories When more is less Memory 7 233-256
Tulving E (1974) Cue-dependent forgetting American Scientist 6274-82
Tulving E (1979) Relation between encoding specificity and levelsof processing In L S Cermak amp F I M Craik (Eds) Levels of pro-cessing in human memory (pp 405-428) Hillsdale NJ Erlbaum
Tulving E (1985) Memory and consciousness Canadian Psycholo-gist 26 1-12
Tun P A Wingfield A Rosen M J amp Blanchard L (1998)Response latencies for false memories Gist-based processes in nor-mal aging Psychology amp Aging 13 230-241
398 ROEDIGER WATSON MCDERMOTT AND GALLO
Underwood B J (1965) False recognition produced by implicit ver-bal responses Journal of Experimental Psychology 70 122-129
Weldon M S (1991) Mechanisms underlying priming on perceptualtests Journal of Experimental Psychology Learning Memory ampCognition 17 526-541
Wickens T D amp Hirshman E (2000) False memories and statisticaldecision theory Comment on Miller and Wolford (1999) and Roedi-ger and McDermott (1999) Psychological Review 107 377-383
Wixted J T amp Stretch V (2000) The case against a criterion-shiftaccount of false memory Psychological Review 107 368-376
Zechmeister E B (1969) Orthographic distinctiveness Journal ofVerbal Learning amp Verbal Behavior 8 754-761
NOTES
1 The orthographic distinctiveness of a critical item was obtained bypresenting 72 subjects with all 55 critical items visually and havingthem rate the words in accordance with procedures developed by Zech-meister (1969) Briefly the subjects were instructed to rate each wordfor its physical appearance in terms of distinctiveness in the languageon a 9-point scale with 1 being not very distinctive 5 corresponding toaverage distinctiveness and 9 being most distinctive To avoid orderingeffects three different randomized orderings of words were used acrossroughly equal groups of subjects In the case of the critical item sleep
for example the distinctiveness is 515 Corresponding values for but-terfly and for man are 744 and 213 respectively Although ortho-graphic distinctiveness was not included in our multiple regressionanalyses this predictor is included in Appendix C
2 The orthographic neighborhood size or Coltheartrsquos N of each crit-ical item was defined as the number of words that can be formed bychanging a single letter in the word (Coltheart Davelaar Jonasson ampBesner 1977) For example because it has no orthographic neighborsColtheartrsquos N for the critical item butterfly is zero In contrast sleep hasfive orthographic neighbors (ie sheep sleek sleet steep and sweep)The reference set of words used were those in Ku Iumlcera and Francisrsquos(1967) norms Although Coltheartrsquos N was not included in our multipleregression analyses this predictor is included in Appendix C
3 Word length orthographic distinctiveness and orthographic neigh-borhood size of the critical item were all highly correlated Specificallythe bivariate correlations among these three predictors were as followslength and orthographic distinctiveness r 5 184 t(53) 5 1117 p lt05 orthographic distinctiveness and orthographic neighborhood sizer 5 66 t(53) 5 640 p lt 05 length and orthographic neighbor-hood size r 5 63 t(53) 5 582 p lt 05 Of these three variablesword length had the largest bivariate correlation with false recall (r 5
37) so it was the predictor used in the multiple regression analyses4 The multiple regression data matrix is available at wwwiacwustl
edunmclWeb
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 399
APPENDIX A The Fifty-Five 15-Word Lists and Their Associated Critical Items
With Backward Associative Strength (BAS) and Forward Associative Strength (FAS)Values Used in the Multiple Regression Analysis of False Recall
Item Word BAS FAS Item Word BAS FAS
Anger mad 393 412 Black white 655 557fear 020 059 dark 111 100hate 028 109 cat 000 043rage 541 042 charred 023 000temper 182 000 night 000 021fury 306 000 funeral 034 000ire 179 000 color 074 050wrath 128 000 grief 000 000happy 000 042 blue 028 000fight 034 000 death 016 014hatred 070 000 ink 020 000mean 090 000 bottom 000 000calm 000 000 coal 288 000emotion 000 000 brown 338 000enrage 378 000 gray 365 000
Mean 157 044 Mean 130 052Army Navy 543 500 Bread butter 364 487
NotemdashUnless otherwise indicated with an asterisk all BAS and FAS values are drawn from the associative norms of Nel-son McEvoy and Schreiber (1999) Those values shown with an asterisk were obtained by the authors by using normingprocedures similar to those of Nelson et al (1999) The corresponding mean for these BAS and FAS values for each list arealso represented
406 ROEDIGER WATSON MCDERMOTT AND GALLO
APPENDIX B Data Matrix of the Fifty-Five 15-Word Lists and Associated Critical Items
Used in the Multiple Regression Analysis of False Recall
Critical False False Log Vrd Raw Ortho Item Recall Recgn Length Freq Concrete FAS BAS Connect Recall Freq Dist Colt N
(Manuscript received July 10 2000revision accepted for publication February 23 2001)
394 ROEDIGER WATSON MCDERMOTT AND GALLO
work that we advocate here (eg Johnson et al 1993Schacter et al 1999) as well as the attributional ap-proach to remembering (Jacoby et al 1989)
The other significant factor arising from our multipleregression analysis may operate during retrieval by aid-ing the monitoring processes that reduce false recall Thenegative correlation between veridical recall of list itemsand false recall of the critical item may indicate that thebetter encoded list items are the more easily they can bedistinguished from the illusory critical item Subjectsmay use item-specific information to aid recall of listitems and suppress recall of the critical items whichwould not carry as much item-specific information Whenexperimental manipulations increase veridical recall ei-ther through longer study time at relatively slow rates(Gallo amp Roediger 2001 McDermott amp Watson 2001Toglia Neuschatz amp Goodwin 1999) or through re-peated study and test opportunities (McDermott 1996Schacter et al 1999) false recall is often decreasedTherefore the negative relation between veridical recalland false recall can be produced experimentally as wellas appearing in our correlationregression analysis How-ever veridical and false recall can be positively related as afunction of other variables too as was shown by McDer-mott (1996) Toglia et al (1999) and Thapar amp McDermott(2001) In these cases greater activation of the criticalitem from relational processing or from greater seman-tic processing is probably at work
The fact that the length andor distinctiveness of thecritical item are negatively correlated with false recallcan be similarly interpreted as the effect of monitoringduring retrieval because the greater distinctiveness ofthe critical item simplifies the monitoring process Forexample if words in the list are generally shorter andhave more regular orthography than does the criticalitem then during retrieval the critical items that are longand distinctive (eg butterfly cabbage) should be moreeasily discriminated from list items as indeed seems tobe the case in the bivariate correlations Interestingly thelengthdistinctiveness factor played a role in determin-ing false recall even though the lists in our norms werepresented auditorily and the orthographic informationat least should not be particularly accessible In addi-tion presenting the DRM lists visually rather than audi-torily reduces false recall and false recognition at leaston tests in which recognition items are presented visually(as opposed to auditorily) or recall is written rather thanspoken (Gallo McDermott et al 2001 Kellogg 2001Smith amp Hunt 1998) This modality difference has alsobeen interpreted as arising from the retrieval of distinctiveinformation operating during the test (Gallo McDermottet al 2001 Smith amp Hunt 1998)
As was noted above activation processes may occurduring retrieval as well as during encoding The morelist items a subject recalls the more primed the criticalitem might become However the present analyses un-dercut the simplest form of this idea If the recalling oflist items activated or primed false recall the simplest
prediction would be that there should be a positive cor-relation between veridical recall and false recall Our datarevealed the opposite pattern showing a negative corre-lation between veridical and false recall Marsh McDer-mott and Roediger (2001) reported other experimentsthat also call into question the idea that priming duringthe test plays a role in false recall and false recognition
Balota et al (1999) obtained further evidence sup-porting the activationmonitoring interpretation of falsememories in the DRM paradigm They compared per-formance of younger adults healthy older adults andolder adults with Alzheimerrsquos disease The latter twogroups have been shown to have intact activation pro-cesses as measured by semantic priming paradigms(Balota amp Duchek 1988 1991) yet they are deficient insource-monitoring processes which is probably attribut-able to impaired frontal lobe functioning (see BalotaDolan amp Duchek 2000) Therefore one might expectthat whereas veridical recall would be worse in older adultsand Alzheimerrsquos patients (relative to younger adults)false recall and false recognition would occur at the samelevel or perhaps be increased in these groups IndeedBalota et al (1999) reported exactly this pattern A sim-ilar outcome has been obtained in other laboratories(eg Budsen Daffner Desikan amp Schacter 2000 Nor-man amp Schacter 1997 Tun Wingfield Rosen amp Blan-chard 1998) This pattern is just as predicted by the ac-tivation monitoring framework
In short the present results and many others (seeRoediger Balota amp Watson 2001 Roediger et al 1998)fit comfortably within the activationmonitoring frame-work In the remainder of our discussion we will con-sider briefly the implications of our present results forother accounts of the DRM paradigm and for the role ofconnectivity (or interitem associative strength) in pro-ducing veridical and false recall
Fuzzy-Trace TheoryAnother leading theory that has been used to explain
false memory phenomena is fuzzy-trace theory (egReyna amp Brainerd 1995) Payne et al (1996) and Schac-ter et al (1996) among many others have used the basictheory to account for false recall and false recognitiondata within the DRM paradigm and the theory has beenextended in interesting ways to account for new varia-tions on the paradigm (eg Brainerd Wright Reyna ampMojardin 2001) The fundamental assumption of fuzzy-trace theory is that experience leaves two types of mem-ory traces gist traces (which capture the meaning of ex-perience but without specific attributes) and verbatimtraces (which represent specific attributes) False memo-ries are attributed to gist processing When the meaning ofa recognition probe matches the gist of the encoded expe-rience false recognition occurs The theory has been di-rected primarily to false recognition and not false recall
In trying to explain the present results one difficultyis that the operational definition of gist has never beenspecified that is within the DRM paradigm given a list
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 395
of 15 related words how does one specify the gist repre-sentation of the list and quantify how strong it is Onecould use patterns of false recognitionmdashto the extentthat a concept is falsely recognized it was represented aspart of the gist representation This definition is per-fectly suitable for some purposes but clearly explainingfalse recognition is not one of them owing to the prob-lem of circularity To explain why some lists lead to highlevels of false recall and false recognition (whereas otherlists lead to low levels) within fuzzy-trace theory oneneeds an independent specification of gist which to ourknowledge has not been provided in this context Whydoes the sweet list have so much gist that it leads to falserecall whereas the bitter list does not One plausible in-terpretation of gist is BAS The more the list items areassociated to the critical item the more a gist represen-tation of the critical item is created This solution to adefinition of gist works within the DRM paradigm butof course the interpretation of how gist is created be-comes isomorphic to activation processes within the ac-tivationmonitoring framework
Interpretation of the other factor that was significantwithin our multiple regression analysis may be morestraightforward within fuzzy-trace theory The negativecorrelation between veridical recall and false recall (andfalse recognition) may be accounted for by increased re-liance on verbatim traces That is if increased recall of listitems is due to increased verbatim information subjectsmay rely less heavily on gist information and hence showless false recall and false recognition during respondingSimilarly for critical items that are long andor ortho-graphically distinct recall of list items that relies on thesefeatures may lead to suppression of false recall and falserecognition of the critical items because they may differsystematically from other list items These factors pos-tulated within the context of fuzzy-trace theory are quitesimilar to those within the activationmonitoring frame-work outlined above At some level of generality there aretwo sets of factors in each account and both theories canexplain many empirical results
Shifting Criterion TheoryMiller and Wolford (1999) proposed that false recog-
nition in the DRM paradigm could be explained by sub-jectsrsquo shifting their criteria differentially to list items andto critical items during a recognition test (They ex-tended their explanation to false recall but somewhattenuously by proposing a generaterecognize theory)They postulated that subjects encoded the structure ofthe list during its presentation and then during the recog-nition test responded with especially liberal criteria tocritical items but not to list items The criterion shift issupposed to occur to the degree that the test item appearsto be related to the list items To support their argumentthey reported that a measure of bias from signal detec-tion theory was indeed more liberal for critical itemsthan for list items However Wickens and Hirshman(2000) and Wixted and Stretch (2000) pointed out that
within signal detection theory differences in bias mea-sures do not necessarily implicate a criterion shiftRather differences in measured bias could arise fromdifferences among the underlying distributions Wixtedand Stretch argued that this latter interpretation was themore likely and that the criterion shift interpretation wasuntenable (also see Roediger amp McDermott 1999)
The present evidence also weighs heavily against acriterion shift account Miller and Wolford (1999) ap-pealed to subjectsrsquo metaknowledge of the structure of thelist Within the criterion shift account we can assumethat the likelihood that a critical item cues metaknowl-edge of the list depends on its FAS or how well the crit-ical item will cue the list If so FAS (the association ofthe critical item to the list items) should be the primarydeterminant of false recognition and false recall How-ever the correlation between FAS and false recall in ourdata was negligible (08) and the same was true in analy-ses of false recognition BAS was actually the primarypredictor of false recall which implicates activation pro-cesses during encoding as being a primary cause of theeffect There would be no reason to expect BAS to haveany impact according to the idea that a simple criterionshift during the test causes false recall and false recog-nition Of course all our lists were relatively high androughly equivalent in terms of FAS because they werecreated from forward associations to the critical item sorestriction of range may account for the low correlationEven so the same problem exists for the shifting crite-rion theory to explain the results If FAS is supposed todetermine false recall and false recognition by this ac-count and if all our lists were high and roughly equal onthis feature why were so many lists ineffective in elicitingfalse recognition and false recall In short either way theshifting criterion theory cannot account for the results
Of course within the activationmonitoring frame-work the monitoring processes occurring during retrievalcould be conceived as conservative shifts in response cri-teria However in our theory these monitoring processesdetermine the degree to which a given level of activationat retrieval will be judged sufficient to endorse a criticalitem as having been studied thereby producing a falsememory In contrast in Miller and Wolfordrsquos (1999) ac-count there was no activation of the critical item duringencoding and shifting criteria during the test were saidto account entirely for the false recognition phenomenon(see Wixted amp Stretch 2000) In the activationmonitoringaccount the memory illusion provoked by the DRM listsis assumed to be a subjectively powerful phenomenonthat can be ameliorated (but rarely eliminated) when listitems are made more distinctive relative to criticalitems Under conditions in which the subject is warnedat test (but not before the encoding phase) to monitortheir memories for the presence or absence of the criti-cal item and is discouraged from making criterion shiftsthe false memory effect is neither eliminated nor greatlyreduced (Gallo Roediger amp McDermott 2001) Hicksand Marsh (2001) showed that under conditions in which
396 ROEDIGER WATSON MCDERMOTT AND GALLO
subjects are explicitly told to monitor their memories dur-ing the test (and are asked to judge whether each wordhad been externally presented or internally generated)false recognition in the DRM paradigm actually increasesrather than decreases In fact the false recognition effectpersists even when subjects are fully informed about thenature of the phenomenon before the study episode andare therefore encouraged to enhance monitoring processesduring both encoding and retrieval in order to attempt todiminish or eliminate the effect (see Gallo et al1997Gallo Roediger amp McDermott 2001 McDermott ampRoediger 1998) For these and other reasons the notionthat the DRM memory illusion is caused by a general shiftof criteria can be eliminated as a potential explanation
Interitem Associative Strength or ConnectivityOur data pose something of a puzzle about the role of
interitem associative strength (Deese 1959a) or connec-tivity (McEvoy et al 1999) in false recall As was dis-cussed in the introduction Deese (1959a) found a sig-nificant negative correlation between this construct andthe probability of false recall In a later paper Deese(1961) obtained a weaker negative correlation that failedto reach conventional levels of significance so the issueis somewhat in doubt from Deesersquos (1959a 1961) researchMcEvoy et al created lists that varied in connectivity(interitem associative strength) and in BAS and manip-ulated these factors orthogonally They replicated thefinding of Deese (1959a) in that the density of inter-connections of list words was negatively related to falserecall However in our analysis of 55 lists (different fromthose used by McEvoy et al 1999) we found no relationbetween connectivity and false recall (r 5 04) or be-tween connectivity and false recognition (r 5 03)
Further research will be necessary to determine thetrue state of affairs although a case can be made that thelack of correlation we obtained is accurate First we had55 lists that varied widely in their ability to elicit false re-call as compared with 24 lists used in McEvoy et al(1999) Our lists also represented a relatively full rangeof connectivity or interitem associative strength relativeto those of McEvoy et al McEvoy et al used lists thathad a somewhat wider range of connectivity because theyused extreme values whereas our larger set of lists pro-vided a fuller body of the middle ranges of connectivityGiven our results it seems at least possible that interitemassociative strength (connectivity) is not a factor in de-termining whether lists of associative words lead to falserecall although the issue remains in doubt
We did find that interitem associative strength or con-nectivity is positively and significantly correlated withveridical recall (r 5 132) This outcome is consistentwith that of Deese (1959a) and McEvoy et al (1999)The greater the associative bonds among list items thegreater is list recall This outcome is also consistent withvirtually any associative theory of recall (eg Anderson1972 1983 Nelson Schreiber amp McEvoy 1992)
CONCLUSION
The present study has uncovered two primary factorsthat are significantly related to false recall and falserecognition in the DRM paradigm BAS is positivelycorrelated with false recall whereas veridical recall ofthe list items is negatively related to false recall The ac-tivationmonitoring account provides a natural interpre-tation of these factors Any viable theory posited to ex-plain the occurrence of false memories in the DRMparadigm must explain the variability in false recall andfalse recognition across lists and the two primary factorsresponsible for this variability
REFERENCES
Anderson J R (1972) FRAN A simulation model of free recall InG H Bower (Ed) The psychology of learning and motivation (Vol 5pp 315-378) New York Academic Press
Anderson J R (1983) The architecture of cognition CambridgeMA Harvard University Press
Anderson J R amp Pirolli P L (1984) Spread of activation Journalof Experimental Psychology Learning Memory amp Cognition 10791-798
Balota D A Cortese M J Duchek J M Adams D RoedigerH L III McDermott K B amp Yerys B E (1999) Veridical andfalse memories in healthy older adults and in dementia of the Alz-heimers type Cognitive Neuropsychology 16 361-384
Balota D A Dolan P O amp Duchek J M (2000) Memorychanges in healthy older adults In E Tulving amp F I M Craik (Eds)The Oxford handbook of memory (pp 395-409) Oxford Oxford Uni-versity Press
Balota D A amp Duchek J M (1988) Age-related differences inlexical access spreading activation and simple pronunciation Psy-chology and Aging 3 84-93
Balota D A amp Duchek J (1991) Semantic priming effects lexicalrepetition effects and contextual disambiguation effects in healthyaged individuals and individuals with senile dementia of theAlzheimer type Brain amp Language 40 181-201
Bartlett F C (1932) Remembering A study in experimental and so-cial psychology Cambridge Cambridge University Press
Brainerd C J Wright R Reyna V F amp Mojardin A H (2001)Conjoint recognition and phantom recollection Journal of Experi-mental Psychology Learning Memory amp Cognition 27 307-327
Bruce D amp Winograd E (1998) Remembering Deesersquos 1959 arti-cles The Zeitgeist the sociology of science and false memoriesPsychonomic Bulletin amp Review 5 615-624
Bruner J S (1957) On perceptual readiness Psychological Review64 123-152
Budson A E Daffner K R Desikan R amp Schacter D L(2000) When false recognition is unopposed by true recognitionGist-based memory distortion in Alzheimerrsquos disease Neuropsy-chology 14 277-287
Collins A M amp Loftus E F (1975) A spreading-activation theoryof semantic memory Psychological Review 82 407-428
Coltheart M Davelaar E Jonasson J T amp Besner D (1977)Access to the internal lexicon In S Dormic (Ed) Attention and per-formance VI (pp 535-555) Hillsdale NJ Erlbaum
Deese J (1959a) Influence of interitem associative strength upon im-mediate free recall Psychological Reports 5 235-241
Deese J (1959b) On the prediction of occurrence of particular verbalintrusions in immediate recall Journal of Experimental Psychology58 17-22
Deese J (1961) Associative structure and the serial reproduction ex-periment Journal of Abnormal amp Social Psychology 63 95-100
Deese J (1965) The structure of associations in language and thoughtBaltimore Johns Hopkins University Press
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 397
Gallo D A McDermott K B Percer J M amp Roediger H LIII (2001) Modality effects in false recall and false recognitionJournal of Experimental Psychology Learning Memory amp Cogni-tion 27 339-353
Gallo D A Roberts M J amp Seamon J G (1997) Rememberingwords not presented in lists Can we avoid creating false memoriesPsychonomic Bulletin amp Review 4 271-276
Gallo D A amp Roediger H L III (2001) Variability among wordlists in evoking associative memory illusions Manuscript submittedfor publication
Gallo D A Roediger H L III amp McDermott K B (2001) As-sociative false recognition occurs without strategic criterion shiftsPsychonomic Bulletin amp Review 8 579-586
Hicks J L amp Marsh R L (2001) False recognition occurs more fre-quently during source identification than during oldndashnew recogni-tion Journal of Experimental Psychology Learning Memory ampCognition 27 375-383
Hunt R R amp Einstein G O (1981) Relational and item-specific in-formation in memory Journal of Verbal Learning amp Verbal Behav-ior 19 497-514
Hunt R R amp McDaniel M A (1993) The enigma of organizationand distinctiveness Journal of Memory amp Language 32 421-445
Israel L amp Schacter D L (1997) Pictorial encoding reduces falserecognition of semantic associates Psychonomic Bulletin amp Review4 577-581
Jacoby L L Kelley C M amp Dywan J (1989) Memory attribu-tions In H L Roediger III amp F I M Craik (Eds) Varieties of mem-ory and consciousness Essays in honour of Endel Tulving (pp 391-422) Hillsdale NJ Erlbaum
Johnson M K Hashtroudi S amp Lindsay D S (1993) Sourcemonitoring Psychological Bulletin 114 3-28
Johnson M K amp Raye C L (1981) Reality monitoring Psycho-logical Review 88 67-85
Kellogg RT (2001) Presentation modality and mode of recall in ver-bal false memory Journal of Experimental Psychology LearningMemory amp Cognition 27 913-919
Kirkpatrick E A (1894) An experimental study of memory Psy-chological Review 1 602-609
Ku Iumlcera H amp Francis W (1967) Computational analysis of present-day American English Providence RI Brown University Press
Marsh E J McDermott K B amp Roediger H L III (2001) Doretrieval factors play a role in the creation of of false memories Man-uscript submitted for publication
Mather M Henkel L A amp Johnson M J (1997) Evaluatingcharacteristics of false memories Rememberknow judgments andmemory characteristics questionnaire compared Memory amp Cogni-tion 25 826-837
McDermott K B (1996) The persistence of false memories in list re-call Journal of Memory amp Language 35 212-230
McDermott K B (1997) Priming on perceptual implicit memorytests can be achieved through presentation of associates Psycho-nomic Bulletin amp Review 4 582-586
McDermott K B amp Roediger H L III (1998) Attempting to avoidillusory memories Robust false recognition of associates persistsunder conditions of explicit warnings and immediate testing Journalof Memory amp Language 39 508-520
McDermott K B amp Watson J M (2001) The rise and fall of falserecall The impact of presentation duration Journal of Memory ampLanguage 45 160-176
McEvoy C L Nelson D L amp Komatsu T (1999) What is the con-nection between true and false memories The differential roles of in-teritem associations in recall and recognition Journal of Experimen-tal Psychology Learning Memory amp Cognition 25 1177-1194
Meyer D E amp Schvaneveldt R W (1971) Facilitation in recog-nizing pairs of words Evidence of a dependence between retrievaloperations Journal of Experimental Psychology 90 227-234
Miller M B amp Wolford G L (1999) Theoretical commentaryThe role of criterion shift in false memory Psychological Review106 398-405
Neely J H (1977) Semantic priming and retrieval from semanticmemory Roles of inhibitionless spreading activation and limited-
capacity attention Journal of Experimental Psychology General106 226-254
Nelson D L McEvoy C L amp Schreiber T A (1999) The Uni-versity of South Florida word association rhyme and word fragmentnorms Unpublished manuscript University of South Florida Tampa
Nelson D L Schreiber TA amp McEvoy C L (1992) Processing im-plicit and explicit representations Psychological Review 99 322-348
Norman K A amp Schacter D L (1997) False recognition in youngerand older adults Exploring the characteristics of illusory memoriesMemory amp Cognition 25 838-848
Paivio A Yuille J C amp Madigan S A (1968) Concreteness im-agery and meaningfulness values for 925 nouns Journal of Experi-mental Psychology Monographs 76 (1 Pt 2)
Payne D G Elie C J Blackwell J M amp Neuschatz J S (1996)Memory illusions Recalling recognizing and recollecting events thatnever occurred Journal of Memory amp Language 35 261-285
Reyna V F amp Brainerd C J (1995) Fuzzy-trace theory An interimsynthesis Learning amp Individual Differences 7 1-75
Roediger H L III Balota D A amp Watson J M (2001) Spreadingactivation and the arousal of false memories In H L Roediger III J SNairne I Neath amp A M Surprenant (Eds) The nature of remem-bering Essays in honor of Robert G Crowder (pp 95-115) Wash-ington DC American Psychological Association
Roediger H L III amp McDermott K B (1995) Creating false mem-ories Remembering words not presented in lists Journal of Experi-mental Psychology Learning Memory amp Cognition 21 803-814
Roediger H L III amp McDermott K B (1999) False alarms aboutfalse memories Psychological Review 106 406-410
Roediger H L III amp McDermott K B (2000) Tricks of memoryCurrent Directions in Psychological Science 9 123-127
Roediger H L III McDermott K B amp Pisoni D B (2001) Rec-ollection of illusory voices Manuscript submitted for publication
Roediger H L III McDermott K B amp Robinson K J (1998)The role of associative processes in creating false memories In M AConway S E Gathercole amp C Cornoldi (Eds) Theories of memoryII (pp 187-245) Hove UK Psychological Press
Russell W A amp Jenkins J J (1954) The complete Minnesotanorms for responses to 100 words from the KentndashRosanoff Word As-sociation Test (Tech Rep No 11 Contract N8 ONR 66216 Officeof Naval Research) University of Minnesota Minneapolis
Schacter D L Israel L amp Racine C (1999) Suppressing falserecognition in younger and older adults The distinctiveness heuris-tic Journal of Memory amp Language 40 1-24
Schacter D L Verfaellie M amp Pradere D (1996) The neuro-psychology of memory illusions False recall and recognition in am-nesic patients Journal of Memory amp Language 35 319-334
Seamon J G Luo C R amp Gallo D A (1998) Creating false mem-ories of words with or without recognition of list items Evidence fornonconscious processes Psychological Science 9 20-26
Smith R E amp Hunt R R (1998) Presentation modality affects falsememory Psychonomic Bulletin amp Review 5 710-715
Stadler M A Roediger H L III amp McDermott K B (1999)Norms for word lists that create false memories Memory amp Cogni-tion 27 494-500
Thapar A amp McDermott K B (2001) False recall and false recog-nition induced by presentation of associated words Effects of retentioninterval and level of processing Memory amp Cognition 29 424-432
Toglia M P amp Battig W F (1978) Handbook of semantic wordnorms Hillsdale NJ Erlbaum
Toglia M P Neuschatz J S amp Goodwin K A (1999) Recall ac-curacy and illusory memories When more is less Memory 7 233-256
Tulving E (1974) Cue-dependent forgetting American Scientist 6274-82
Tulving E (1979) Relation between encoding specificity and levelsof processing In L S Cermak amp F I M Craik (Eds) Levels of pro-cessing in human memory (pp 405-428) Hillsdale NJ Erlbaum
Tulving E (1985) Memory and consciousness Canadian Psycholo-gist 26 1-12
Tun P A Wingfield A Rosen M J amp Blanchard L (1998)Response latencies for false memories Gist-based processes in nor-mal aging Psychology amp Aging 13 230-241
398 ROEDIGER WATSON MCDERMOTT AND GALLO
Underwood B J (1965) False recognition produced by implicit ver-bal responses Journal of Experimental Psychology 70 122-129
Weldon M S (1991) Mechanisms underlying priming on perceptualtests Journal of Experimental Psychology Learning Memory ampCognition 17 526-541
Wickens T D amp Hirshman E (2000) False memories and statisticaldecision theory Comment on Miller and Wolford (1999) and Roedi-ger and McDermott (1999) Psychological Review 107 377-383
Wixted J T amp Stretch V (2000) The case against a criterion-shiftaccount of false memory Psychological Review 107 368-376
Zechmeister E B (1969) Orthographic distinctiveness Journal ofVerbal Learning amp Verbal Behavior 8 754-761
NOTES
1 The orthographic distinctiveness of a critical item was obtained bypresenting 72 subjects with all 55 critical items visually and havingthem rate the words in accordance with procedures developed by Zech-meister (1969) Briefly the subjects were instructed to rate each wordfor its physical appearance in terms of distinctiveness in the languageon a 9-point scale with 1 being not very distinctive 5 corresponding toaverage distinctiveness and 9 being most distinctive To avoid orderingeffects three different randomized orderings of words were used acrossroughly equal groups of subjects In the case of the critical item sleep
for example the distinctiveness is 515 Corresponding values for but-terfly and for man are 744 and 213 respectively Although ortho-graphic distinctiveness was not included in our multiple regressionanalyses this predictor is included in Appendix C
2 The orthographic neighborhood size or Coltheartrsquos N of each crit-ical item was defined as the number of words that can be formed bychanging a single letter in the word (Coltheart Davelaar Jonasson ampBesner 1977) For example because it has no orthographic neighborsColtheartrsquos N for the critical item butterfly is zero In contrast sleep hasfive orthographic neighbors (ie sheep sleek sleet steep and sweep)The reference set of words used were those in Ku Iumlcera and Francisrsquos(1967) norms Although Coltheartrsquos N was not included in our multipleregression analyses this predictor is included in Appendix C
3 Word length orthographic distinctiveness and orthographic neigh-borhood size of the critical item were all highly correlated Specificallythe bivariate correlations among these three predictors were as followslength and orthographic distinctiveness r 5 184 t(53) 5 1117 p lt05 orthographic distinctiveness and orthographic neighborhood sizer 5 66 t(53) 5 640 p lt 05 length and orthographic neighbor-hood size r 5 63 t(53) 5 582 p lt 05 Of these three variablesword length had the largest bivariate correlation with false recall (r 5
37) so it was the predictor used in the multiple regression analyses4 The multiple regression data matrix is available at wwwiacwustl
edunmclWeb
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 399
APPENDIX A The Fifty-Five 15-Word Lists and Their Associated Critical Items
With Backward Associative Strength (BAS) and Forward Associative Strength (FAS)Values Used in the Multiple Regression Analysis of False Recall
Item Word BAS FAS Item Word BAS FAS
Anger mad 393 412 Black white 655 557fear 020 059 dark 111 100hate 028 109 cat 000 043rage 541 042 charred 023 000temper 182 000 night 000 021fury 306 000 funeral 034 000ire 179 000 color 074 050wrath 128 000 grief 000 000happy 000 042 blue 028 000fight 034 000 death 016 014hatred 070 000 ink 020 000mean 090 000 bottom 000 000calm 000 000 coal 288 000emotion 000 000 brown 338 000enrage 378 000 gray 365 000
Mean 157 044 Mean 130 052Army Navy 543 500 Bread butter 364 487
NotemdashUnless otherwise indicated with an asterisk all BAS and FAS values are drawn from the associative norms of Nel-son McEvoy and Schreiber (1999) Those values shown with an asterisk were obtained by the authors by using normingprocedures similar to those of Nelson et al (1999) The corresponding mean for these BAS and FAS values for each list arealso represented
406 ROEDIGER WATSON MCDERMOTT AND GALLO
APPENDIX B Data Matrix of the Fifty-Five 15-Word Lists and Associated Critical Items
Used in the Multiple Regression Analysis of False Recall
Critical False False Log Vrd Raw Ortho Item Recall Recgn Length Freq Concrete FAS BAS Connect Recall Freq Dist Colt N
(Manuscript received July 10 2000revision accepted for publication February 23 2001)
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 395
of 15 related words how does one specify the gist repre-sentation of the list and quantify how strong it is Onecould use patterns of false recognitionmdashto the extentthat a concept is falsely recognized it was represented aspart of the gist representation This definition is per-fectly suitable for some purposes but clearly explainingfalse recognition is not one of them owing to the prob-lem of circularity To explain why some lists lead to highlevels of false recall and false recognition (whereas otherlists lead to low levels) within fuzzy-trace theory oneneeds an independent specification of gist which to ourknowledge has not been provided in this context Whydoes the sweet list have so much gist that it leads to falserecall whereas the bitter list does not One plausible in-terpretation of gist is BAS The more the list items areassociated to the critical item the more a gist represen-tation of the critical item is created This solution to adefinition of gist works within the DRM paradigm butof course the interpretation of how gist is created be-comes isomorphic to activation processes within the ac-tivationmonitoring framework
Interpretation of the other factor that was significantwithin our multiple regression analysis may be morestraightforward within fuzzy-trace theory The negativecorrelation between veridical recall and false recall (andfalse recognition) may be accounted for by increased re-liance on verbatim traces That is if increased recall of listitems is due to increased verbatim information subjectsmay rely less heavily on gist information and hence showless false recall and false recognition during respondingSimilarly for critical items that are long andor ortho-graphically distinct recall of list items that relies on thesefeatures may lead to suppression of false recall and falserecognition of the critical items because they may differsystematically from other list items These factors pos-tulated within the context of fuzzy-trace theory are quitesimilar to those within the activationmonitoring frame-work outlined above At some level of generality there aretwo sets of factors in each account and both theories canexplain many empirical results
Shifting Criterion TheoryMiller and Wolford (1999) proposed that false recog-
nition in the DRM paradigm could be explained by sub-jectsrsquo shifting their criteria differentially to list items andto critical items during a recognition test (They ex-tended their explanation to false recall but somewhattenuously by proposing a generaterecognize theory)They postulated that subjects encoded the structure ofthe list during its presentation and then during the recog-nition test responded with especially liberal criteria tocritical items but not to list items The criterion shift issupposed to occur to the degree that the test item appearsto be related to the list items To support their argumentthey reported that a measure of bias from signal detec-tion theory was indeed more liberal for critical itemsthan for list items However Wickens and Hirshman(2000) and Wixted and Stretch (2000) pointed out that
within signal detection theory differences in bias mea-sures do not necessarily implicate a criterion shiftRather differences in measured bias could arise fromdifferences among the underlying distributions Wixtedand Stretch argued that this latter interpretation was themore likely and that the criterion shift interpretation wasuntenable (also see Roediger amp McDermott 1999)
The present evidence also weighs heavily against acriterion shift account Miller and Wolford (1999) ap-pealed to subjectsrsquo metaknowledge of the structure of thelist Within the criterion shift account we can assumethat the likelihood that a critical item cues metaknowl-edge of the list depends on its FAS or how well the crit-ical item will cue the list If so FAS (the association ofthe critical item to the list items) should be the primarydeterminant of false recognition and false recall How-ever the correlation between FAS and false recall in ourdata was negligible (08) and the same was true in analy-ses of false recognition BAS was actually the primarypredictor of false recall which implicates activation pro-cesses during encoding as being a primary cause of theeffect There would be no reason to expect BAS to haveany impact according to the idea that a simple criterionshift during the test causes false recall and false recog-nition Of course all our lists were relatively high androughly equivalent in terms of FAS because they werecreated from forward associations to the critical item sorestriction of range may account for the low correlationEven so the same problem exists for the shifting crite-rion theory to explain the results If FAS is supposed todetermine false recall and false recognition by this ac-count and if all our lists were high and roughly equal onthis feature why were so many lists ineffective in elicitingfalse recognition and false recall In short either way theshifting criterion theory cannot account for the results
Of course within the activationmonitoring frame-work the monitoring processes occurring during retrievalcould be conceived as conservative shifts in response cri-teria However in our theory these monitoring processesdetermine the degree to which a given level of activationat retrieval will be judged sufficient to endorse a criticalitem as having been studied thereby producing a falsememory In contrast in Miller and Wolfordrsquos (1999) ac-count there was no activation of the critical item duringencoding and shifting criteria during the test were saidto account entirely for the false recognition phenomenon(see Wixted amp Stretch 2000) In the activationmonitoringaccount the memory illusion provoked by the DRM listsis assumed to be a subjectively powerful phenomenonthat can be ameliorated (but rarely eliminated) when listitems are made more distinctive relative to criticalitems Under conditions in which the subject is warnedat test (but not before the encoding phase) to monitortheir memories for the presence or absence of the criti-cal item and is discouraged from making criterion shiftsthe false memory effect is neither eliminated nor greatlyreduced (Gallo Roediger amp McDermott 2001) Hicksand Marsh (2001) showed that under conditions in which
396 ROEDIGER WATSON MCDERMOTT AND GALLO
subjects are explicitly told to monitor their memories dur-ing the test (and are asked to judge whether each wordhad been externally presented or internally generated)false recognition in the DRM paradigm actually increasesrather than decreases In fact the false recognition effectpersists even when subjects are fully informed about thenature of the phenomenon before the study episode andare therefore encouraged to enhance monitoring processesduring both encoding and retrieval in order to attempt todiminish or eliminate the effect (see Gallo et al1997Gallo Roediger amp McDermott 2001 McDermott ampRoediger 1998) For these and other reasons the notionthat the DRM memory illusion is caused by a general shiftof criteria can be eliminated as a potential explanation
Interitem Associative Strength or ConnectivityOur data pose something of a puzzle about the role of
interitem associative strength (Deese 1959a) or connec-tivity (McEvoy et al 1999) in false recall As was dis-cussed in the introduction Deese (1959a) found a sig-nificant negative correlation between this construct andthe probability of false recall In a later paper Deese(1961) obtained a weaker negative correlation that failedto reach conventional levels of significance so the issueis somewhat in doubt from Deesersquos (1959a 1961) researchMcEvoy et al created lists that varied in connectivity(interitem associative strength) and in BAS and manip-ulated these factors orthogonally They replicated thefinding of Deese (1959a) in that the density of inter-connections of list words was negatively related to falserecall However in our analysis of 55 lists (different fromthose used by McEvoy et al 1999) we found no relationbetween connectivity and false recall (r 5 04) or be-tween connectivity and false recognition (r 5 03)
Further research will be necessary to determine thetrue state of affairs although a case can be made that thelack of correlation we obtained is accurate First we had55 lists that varied widely in their ability to elicit false re-call as compared with 24 lists used in McEvoy et al(1999) Our lists also represented a relatively full rangeof connectivity or interitem associative strength relativeto those of McEvoy et al McEvoy et al used lists thathad a somewhat wider range of connectivity because theyused extreme values whereas our larger set of lists pro-vided a fuller body of the middle ranges of connectivityGiven our results it seems at least possible that interitemassociative strength (connectivity) is not a factor in de-termining whether lists of associative words lead to falserecall although the issue remains in doubt
We did find that interitem associative strength or con-nectivity is positively and significantly correlated withveridical recall (r 5 132) This outcome is consistentwith that of Deese (1959a) and McEvoy et al (1999)The greater the associative bonds among list items thegreater is list recall This outcome is also consistent withvirtually any associative theory of recall (eg Anderson1972 1983 Nelson Schreiber amp McEvoy 1992)
CONCLUSION
The present study has uncovered two primary factorsthat are significantly related to false recall and falserecognition in the DRM paradigm BAS is positivelycorrelated with false recall whereas veridical recall ofthe list items is negatively related to false recall The ac-tivationmonitoring account provides a natural interpre-tation of these factors Any viable theory posited to ex-plain the occurrence of false memories in the DRMparadigm must explain the variability in false recall andfalse recognition across lists and the two primary factorsresponsible for this variability
REFERENCES
Anderson J R (1972) FRAN A simulation model of free recall InG H Bower (Ed) The psychology of learning and motivation (Vol 5pp 315-378) New York Academic Press
Anderson J R (1983) The architecture of cognition CambridgeMA Harvard University Press
Anderson J R amp Pirolli P L (1984) Spread of activation Journalof Experimental Psychology Learning Memory amp Cognition 10791-798
Balota D A Cortese M J Duchek J M Adams D RoedigerH L III McDermott K B amp Yerys B E (1999) Veridical andfalse memories in healthy older adults and in dementia of the Alz-heimers type Cognitive Neuropsychology 16 361-384
Balota D A Dolan P O amp Duchek J M (2000) Memorychanges in healthy older adults In E Tulving amp F I M Craik (Eds)The Oxford handbook of memory (pp 395-409) Oxford Oxford Uni-versity Press
Balota D A amp Duchek J M (1988) Age-related differences inlexical access spreading activation and simple pronunciation Psy-chology and Aging 3 84-93
Balota D A amp Duchek J (1991) Semantic priming effects lexicalrepetition effects and contextual disambiguation effects in healthyaged individuals and individuals with senile dementia of theAlzheimer type Brain amp Language 40 181-201
Bartlett F C (1932) Remembering A study in experimental and so-cial psychology Cambridge Cambridge University Press
Brainerd C J Wright R Reyna V F amp Mojardin A H (2001)Conjoint recognition and phantom recollection Journal of Experi-mental Psychology Learning Memory amp Cognition 27 307-327
Bruce D amp Winograd E (1998) Remembering Deesersquos 1959 arti-cles The Zeitgeist the sociology of science and false memoriesPsychonomic Bulletin amp Review 5 615-624
Bruner J S (1957) On perceptual readiness Psychological Review64 123-152
Budson A E Daffner K R Desikan R amp Schacter D L(2000) When false recognition is unopposed by true recognitionGist-based memory distortion in Alzheimerrsquos disease Neuropsy-chology 14 277-287
Collins A M amp Loftus E F (1975) A spreading-activation theoryof semantic memory Psychological Review 82 407-428
Coltheart M Davelaar E Jonasson J T amp Besner D (1977)Access to the internal lexicon In S Dormic (Ed) Attention and per-formance VI (pp 535-555) Hillsdale NJ Erlbaum
Deese J (1959a) Influence of interitem associative strength upon im-mediate free recall Psychological Reports 5 235-241
Deese J (1959b) On the prediction of occurrence of particular verbalintrusions in immediate recall Journal of Experimental Psychology58 17-22
Deese J (1961) Associative structure and the serial reproduction ex-periment Journal of Abnormal amp Social Psychology 63 95-100
Deese J (1965) The structure of associations in language and thoughtBaltimore Johns Hopkins University Press
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 397
Gallo D A McDermott K B Percer J M amp Roediger H LIII (2001) Modality effects in false recall and false recognitionJournal of Experimental Psychology Learning Memory amp Cogni-tion 27 339-353
Gallo D A Roberts M J amp Seamon J G (1997) Rememberingwords not presented in lists Can we avoid creating false memoriesPsychonomic Bulletin amp Review 4 271-276
Gallo D A amp Roediger H L III (2001) Variability among wordlists in evoking associative memory illusions Manuscript submittedfor publication
Gallo D A Roediger H L III amp McDermott K B (2001) As-sociative false recognition occurs without strategic criterion shiftsPsychonomic Bulletin amp Review 8 579-586
Hicks J L amp Marsh R L (2001) False recognition occurs more fre-quently during source identification than during oldndashnew recogni-tion Journal of Experimental Psychology Learning Memory ampCognition 27 375-383
Hunt R R amp Einstein G O (1981) Relational and item-specific in-formation in memory Journal of Verbal Learning amp Verbal Behav-ior 19 497-514
Hunt R R amp McDaniel M A (1993) The enigma of organizationand distinctiveness Journal of Memory amp Language 32 421-445
Israel L amp Schacter D L (1997) Pictorial encoding reduces falserecognition of semantic associates Psychonomic Bulletin amp Review4 577-581
Jacoby L L Kelley C M amp Dywan J (1989) Memory attribu-tions In H L Roediger III amp F I M Craik (Eds) Varieties of mem-ory and consciousness Essays in honour of Endel Tulving (pp 391-422) Hillsdale NJ Erlbaum
Johnson M K Hashtroudi S amp Lindsay D S (1993) Sourcemonitoring Psychological Bulletin 114 3-28
Johnson M K amp Raye C L (1981) Reality monitoring Psycho-logical Review 88 67-85
Kellogg RT (2001) Presentation modality and mode of recall in ver-bal false memory Journal of Experimental Psychology LearningMemory amp Cognition 27 913-919
Kirkpatrick E A (1894) An experimental study of memory Psy-chological Review 1 602-609
Ku Iumlcera H amp Francis W (1967) Computational analysis of present-day American English Providence RI Brown University Press
Marsh E J McDermott K B amp Roediger H L III (2001) Doretrieval factors play a role in the creation of of false memories Man-uscript submitted for publication
Mather M Henkel L A amp Johnson M J (1997) Evaluatingcharacteristics of false memories Rememberknow judgments andmemory characteristics questionnaire compared Memory amp Cogni-tion 25 826-837
McDermott K B (1996) The persistence of false memories in list re-call Journal of Memory amp Language 35 212-230
McDermott K B (1997) Priming on perceptual implicit memorytests can be achieved through presentation of associates Psycho-nomic Bulletin amp Review 4 582-586
McDermott K B amp Roediger H L III (1998) Attempting to avoidillusory memories Robust false recognition of associates persistsunder conditions of explicit warnings and immediate testing Journalof Memory amp Language 39 508-520
McDermott K B amp Watson J M (2001) The rise and fall of falserecall The impact of presentation duration Journal of Memory ampLanguage 45 160-176
McEvoy C L Nelson D L amp Komatsu T (1999) What is the con-nection between true and false memories The differential roles of in-teritem associations in recall and recognition Journal of Experimen-tal Psychology Learning Memory amp Cognition 25 1177-1194
Meyer D E amp Schvaneveldt R W (1971) Facilitation in recog-nizing pairs of words Evidence of a dependence between retrievaloperations Journal of Experimental Psychology 90 227-234
Miller M B amp Wolford G L (1999) Theoretical commentaryThe role of criterion shift in false memory Psychological Review106 398-405
Neely J H (1977) Semantic priming and retrieval from semanticmemory Roles of inhibitionless spreading activation and limited-
capacity attention Journal of Experimental Psychology General106 226-254
Nelson D L McEvoy C L amp Schreiber T A (1999) The Uni-versity of South Florida word association rhyme and word fragmentnorms Unpublished manuscript University of South Florida Tampa
Nelson D L Schreiber TA amp McEvoy C L (1992) Processing im-plicit and explicit representations Psychological Review 99 322-348
Norman K A amp Schacter D L (1997) False recognition in youngerand older adults Exploring the characteristics of illusory memoriesMemory amp Cognition 25 838-848
Paivio A Yuille J C amp Madigan S A (1968) Concreteness im-agery and meaningfulness values for 925 nouns Journal of Experi-mental Psychology Monographs 76 (1 Pt 2)
Payne D G Elie C J Blackwell J M amp Neuschatz J S (1996)Memory illusions Recalling recognizing and recollecting events thatnever occurred Journal of Memory amp Language 35 261-285
Reyna V F amp Brainerd C J (1995) Fuzzy-trace theory An interimsynthesis Learning amp Individual Differences 7 1-75
Roediger H L III Balota D A amp Watson J M (2001) Spreadingactivation and the arousal of false memories In H L Roediger III J SNairne I Neath amp A M Surprenant (Eds) The nature of remem-bering Essays in honor of Robert G Crowder (pp 95-115) Wash-ington DC American Psychological Association
Roediger H L III amp McDermott K B (1995) Creating false mem-ories Remembering words not presented in lists Journal of Experi-mental Psychology Learning Memory amp Cognition 21 803-814
Roediger H L III amp McDermott K B (1999) False alarms aboutfalse memories Psychological Review 106 406-410
Roediger H L III amp McDermott K B (2000) Tricks of memoryCurrent Directions in Psychological Science 9 123-127
Roediger H L III McDermott K B amp Pisoni D B (2001) Rec-ollection of illusory voices Manuscript submitted for publication
Roediger H L III McDermott K B amp Robinson K J (1998)The role of associative processes in creating false memories In M AConway S E Gathercole amp C Cornoldi (Eds) Theories of memoryII (pp 187-245) Hove UK Psychological Press
Russell W A amp Jenkins J J (1954) The complete Minnesotanorms for responses to 100 words from the KentndashRosanoff Word As-sociation Test (Tech Rep No 11 Contract N8 ONR 66216 Officeof Naval Research) University of Minnesota Minneapolis
Schacter D L Israel L amp Racine C (1999) Suppressing falserecognition in younger and older adults The distinctiveness heuris-tic Journal of Memory amp Language 40 1-24
Schacter D L Verfaellie M amp Pradere D (1996) The neuro-psychology of memory illusions False recall and recognition in am-nesic patients Journal of Memory amp Language 35 319-334
Seamon J G Luo C R amp Gallo D A (1998) Creating false mem-ories of words with or without recognition of list items Evidence fornonconscious processes Psychological Science 9 20-26
Smith R E amp Hunt R R (1998) Presentation modality affects falsememory Psychonomic Bulletin amp Review 5 710-715
Stadler M A Roediger H L III amp McDermott K B (1999)Norms for word lists that create false memories Memory amp Cogni-tion 27 494-500
Thapar A amp McDermott K B (2001) False recall and false recog-nition induced by presentation of associated words Effects of retentioninterval and level of processing Memory amp Cognition 29 424-432
Toglia M P amp Battig W F (1978) Handbook of semantic wordnorms Hillsdale NJ Erlbaum
Toglia M P Neuschatz J S amp Goodwin K A (1999) Recall ac-curacy and illusory memories When more is less Memory 7 233-256
Tulving E (1974) Cue-dependent forgetting American Scientist 6274-82
Tulving E (1979) Relation between encoding specificity and levelsof processing In L S Cermak amp F I M Craik (Eds) Levels of pro-cessing in human memory (pp 405-428) Hillsdale NJ Erlbaum
Tulving E (1985) Memory and consciousness Canadian Psycholo-gist 26 1-12
Tun P A Wingfield A Rosen M J amp Blanchard L (1998)Response latencies for false memories Gist-based processes in nor-mal aging Psychology amp Aging 13 230-241
398 ROEDIGER WATSON MCDERMOTT AND GALLO
Underwood B J (1965) False recognition produced by implicit ver-bal responses Journal of Experimental Psychology 70 122-129
Weldon M S (1991) Mechanisms underlying priming on perceptualtests Journal of Experimental Psychology Learning Memory ampCognition 17 526-541
Wickens T D amp Hirshman E (2000) False memories and statisticaldecision theory Comment on Miller and Wolford (1999) and Roedi-ger and McDermott (1999) Psychological Review 107 377-383
Wixted J T amp Stretch V (2000) The case against a criterion-shiftaccount of false memory Psychological Review 107 368-376
Zechmeister E B (1969) Orthographic distinctiveness Journal ofVerbal Learning amp Verbal Behavior 8 754-761
NOTES
1 The orthographic distinctiveness of a critical item was obtained bypresenting 72 subjects with all 55 critical items visually and havingthem rate the words in accordance with procedures developed by Zech-meister (1969) Briefly the subjects were instructed to rate each wordfor its physical appearance in terms of distinctiveness in the languageon a 9-point scale with 1 being not very distinctive 5 corresponding toaverage distinctiveness and 9 being most distinctive To avoid orderingeffects three different randomized orderings of words were used acrossroughly equal groups of subjects In the case of the critical item sleep
for example the distinctiveness is 515 Corresponding values for but-terfly and for man are 744 and 213 respectively Although ortho-graphic distinctiveness was not included in our multiple regressionanalyses this predictor is included in Appendix C
2 The orthographic neighborhood size or Coltheartrsquos N of each crit-ical item was defined as the number of words that can be formed bychanging a single letter in the word (Coltheart Davelaar Jonasson ampBesner 1977) For example because it has no orthographic neighborsColtheartrsquos N for the critical item butterfly is zero In contrast sleep hasfive orthographic neighbors (ie sheep sleek sleet steep and sweep)The reference set of words used were those in Ku Iumlcera and Francisrsquos(1967) norms Although Coltheartrsquos N was not included in our multipleregression analyses this predictor is included in Appendix C
3 Word length orthographic distinctiveness and orthographic neigh-borhood size of the critical item were all highly correlated Specificallythe bivariate correlations among these three predictors were as followslength and orthographic distinctiveness r 5 184 t(53) 5 1117 p lt05 orthographic distinctiveness and orthographic neighborhood sizer 5 66 t(53) 5 640 p lt 05 length and orthographic neighbor-hood size r 5 63 t(53) 5 582 p lt 05 Of these three variablesword length had the largest bivariate correlation with false recall (r 5
37) so it was the predictor used in the multiple regression analyses4 The multiple regression data matrix is available at wwwiacwustl
edunmclWeb
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 399
APPENDIX A The Fifty-Five 15-Word Lists and Their Associated Critical Items
With Backward Associative Strength (BAS) and Forward Associative Strength (FAS)Values Used in the Multiple Regression Analysis of False Recall
Item Word BAS FAS Item Word BAS FAS
Anger mad 393 412 Black white 655 557fear 020 059 dark 111 100hate 028 109 cat 000 043rage 541 042 charred 023 000temper 182 000 night 000 021fury 306 000 funeral 034 000ire 179 000 color 074 050wrath 128 000 grief 000 000happy 000 042 blue 028 000fight 034 000 death 016 014hatred 070 000 ink 020 000mean 090 000 bottom 000 000calm 000 000 coal 288 000emotion 000 000 brown 338 000enrage 378 000 gray 365 000
Mean 157 044 Mean 130 052Army Navy 543 500 Bread butter 364 487
NotemdashUnless otherwise indicated with an asterisk all BAS and FAS values are drawn from the associative norms of Nel-son McEvoy and Schreiber (1999) Those values shown with an asterisk were obtained by the authors by using normingprocedures similar to those of Nelson et al (1999) The corresponding mean for these BAS and FAS values for each list arealso represented
406 ROEDIGER WATSON MCDERMOTT AND GALLO
APPENDIX B Data Matrix of the Fifty-Five 15-Word Lists and Associated Critical Items
Used in the Multiple Regression Analysis of False Recall
Critical False False Log Vrd Raw Ortho Item Recall Recgn Length Freq Concrete FAS BAS Connect Recall Freq Dist Colt N
(Manuscript received July 10 2000revision accepted for publication February 23 2001)
396 ROEDIGER WATSON MCDERMOTT AND GALLO
subjects are explicitly told to monitor their memories dur-ing the test (and are asked to judge whether each wordhad been externally presented or internally generated)false recognition in the DRM paradigm actually increasesrather than decreases In fact the false recognition effectpersists even when subjects are fully informed about thenature of the phenomenon before the study episode andare therefore encouraged to enhance monitoring processesduring both encoding and retrieval in order to attempt todiminish or eliminate the effect (see Gallo et al1997Gallo Roediger amp McDermott 2001 McDermott ampRoediger 1998) For these and other reasons the notionthat the DRM memory illusion is caused by a general shiftof criteria can be eliminated as a potential explanation
Interitem Associative Strength or ConnectivityOur data pose something of a puzzle about the role of
interitem associative strength (Deese 1959a) or connec-tivity (McEvoy et al 1999) in false recall As was dis-cussed in the introduction Deese (1959a) found a sig-nificant negative correlation between this construct andthe probability of false recall In a later paper Deese(1961) obtained a weaker negative correlation that failedto reach conventional levels of significance so the issueis somewhat in doubt from Deesersquos (1959a 1961) researchMcEvoy et al created lists that varied in connectivity(interitem associative strength) and in BAS and manip-ulated these factors orthogonally They replicated thefinding of Deese (1959a) in that the density of inter-connections of list words was negatively related to falserecall However in our analysis of 55 lists (different fromthose used by McEvoy et al 1999) we found no relationbetween connectivity and false recall (r 5 04) or be-tween connectivity and false recognition (r 5 03)
Further research will be necessary to determine thetrue state of affairs although a case can be made that thelack of correlation we obtained is accurate First we had55 lists that varied widely in their ability to elicit false re-call as compared with 24 lists used in McEvoy et al(1999) Our lists also represented a relatively full rangeof connectivity or interitem associative strength relativeto those of McEvoy et al McEvoy et al used lists thathad a somewhat wider range of connectivity because theyused extreme values whereas our larger set of lists pro-vided a fuller body of the middle ranges of connectivityGiven our results it seems at least possible that interitemassociative strength (connectivity) is not a factor in de-termining whether lists of associative words lead to falserecall although the issue remains in doubt
We did find that interitem associative strength or con-nectivity is positively and significantly correlated withveridical recall (r 5 132) This outcome is consistentwith that of Deese (1959a) and McEvoy et al (1999)The greater the associative bonds among list items thegreater is list recall This outcome is also consistent withvirtually any associative theory of recall (eg Anderson1972 1983 Nelson Schreiber amp McEvoy 1992)
CONCLUSION
The present study has uncovered two primary factorsthat are significantly related to false recall and falserecognition in the DRM paradigm BAS is positivelycorrelated with false recall whereas veridical recall ofthe list items is negatively related to false recall The ac-tivationmonitoring account provides a natural interpre-tation of these factors Any viable theory posited to ex-plain the occurrence of false memories in the DRMparadigm must explain the variability in false recall andfalse recognition across lists and the two primary factorsresponsible for this variability
REFERENCES
Anderson J R (1972) FRAN A simulation model of free recall InG H Bower (Ed) The psychology of learning and motivation (Vol 5pp 315-378) New York Academic Press
Anderson J R (1983) The architecture of cognition CambridgeMA Harvard University Press
Anderson J R amp Pirolli P L (1984) Spread of activation Journalof Experimental Psychology Learning Memory amp Cognition 10791-798
Balota D A Cortese M J Duchek J M Adams D RoedigerH L III McDermott K B amp Yerys B E (1999) Veridical andfalse memories in healthy older adults and in dementia of the Alz-heimers type Cognitive Neuropsychology 16 361-384
Balota D A Dolan P O amp Duchek J M (2000) Memorychanges in healthy older adults In E Tulving amp F I M Craik (Eds)The Oxford handbook of memory (pp 395-409) Oxford Oxford Uni-versity Press
Balota D A amp Duchek J M (1988) Age-related differences inlexical access spreading activation and simple pronunciation Psy-chology and Aging 3 84-93
Balota D A amp Duchek J (1991) Semantic priming effects lexicalrepetition effects and contextual disambiguation effects in healthyaged individuals and individuals with senile dementia of theAlzheimer type Brain amp Language 40 181-201
Bartlett F C (1932) Remembering A study in experimental and so-cial psychology Cambridge Cambridge University Press
Brainerd C J Wright R Reyna V F amp Mojardin A H (2001)Conjoint recognition and phantom recollection Journal of Experi-mental Psychology Learning Memory amp Cognition 27 307-327
Bruce D amp Winograd E (1998) Remembering Deesersquos 1959 arti-cles The Zeitgeist the sociology of science and false memoriesPsychonomic Bulletin amp Review 5 615-624
Bruner J S (1957) On perceptual readiness Psychological Review64 123-152
Budson A E Daffner K R Desikan R amp Schacter D L(2000) When false recognition is unopposed by true recognitionGist-based memory distortion in Alzheimerrsquos disease Neuropsy-chology 14 277-287
Collins A M amp Loftus E F (1975) A spreading-activation theoryof semantic memory Psychological Review 82 407-428
Coltheart M Davelaar E Jonasson J T amp Besner D (1977)Access to the internal lexicon In S Dormic (Ed) Attention and per-formance VI (pp 535-555) Hillsdale NJ Erlbaum
Deese J (1959a) Influence of interitem associative strength upon im-mediate free recall Psychological Reports 5 235-241
Deese J (1959b) On the prediction of occurrence of particular verbalintrusions in immediate recall Journal of Experimental Psychology58 17-22
Deese J (1961) Associative structure and the serial reproduction ex-periment Journal of Abnormal amp Social Psychology 63 95-100
Deese J (1965) The structure of associations in language and thoughtBaltimore Johns Hopkins University Press
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 397
Gallo D A McDermott K B Percer J M amp Roediger H LIII (2001) Modality effects in false recall and false recognitionJournal of Experimental Psychology Learning Memory amp Cogni-tion 27 339-353
Gallo D A Roberts M J amp Seamon J G (1997) Rememberingwords not presented in lists Can we avoid creating false memoriesPsychonomic Bulletin amp Review 4 271-276
Gallo D A amp Roediger H L III (2001) Variability among wordlists in evoking associative memory illusions Manuscript submittedfor publication
Gallo D A Roediger H L III amp McDermott K B (2001) As-sociative false recognition occurs without strategic criterion shiftsPsychonomic Bulletin amp Review 8 579-586
Hicks J L amp Marsh R L (2001) False recognition occurs more fre-quently during source identification than during oldndashnew recogni-tion Journal of Experimental Psychology Learning Memory ampCognition 27 375-383
Hunt R R amp Einstein G O (1981) Relational and item-specific in-formation in memory Journal of Verbal Learning amp Verbal Behav-ior 19 497-514
Hunt R R amp McDaniel M A (1993) The enigma of organizationand distinctiveness Journal of Memory amp Language 32 421-445
Israel L amp Schacter D L (1997) Pictorial encoding reduces falserecognition of semantic associates Psychonomic Bulletin amp Review4 577-581
Jacoby L L Kelley C M amp Dywan J (1989) Memory attribu-tions In H L Roediger III amp F I M Craik (Eds) Varieties of mem-ory and consciousness Essays in honour of Endel Tulving (pp 391-422) Hillsdale NJ Erlbaum
Johnson M K Hashtroudi S amp Lindsay D S (1993) Sourcemonitoring Psychological Bulletin 114 3-28
Johnson M K amp Raye C L (1981) Reality monitoring Psycho-logical Review 88 67-85
Kellogg RT (2001) Presentation modality and mode of recall in ver-bal false memory Journal of Experimental Psychology LearningMemory amp Cognition 27 913-919
Kirkpatrick E A (1894) An experimental study of memory Psy-chological Review 1 602-609
Ku Iumlcera H amp Francis W (1967) Computational analysis of present-day American English Providence RI Brown University Press
Marsh E J McDermott K B amp Roediger H L III (2001) Doretrieval factors play a role in the creation of of false memories Man-uscript submitted for publication
Mather M Henkel L A amp Johnson M J (1997) Evaluatingcharacteristics of false memories Rememberknow judgments andmemory characteristics questionnaire compared Memory amp Cogni-tion 25 826-837
McDermott K B (1996) The persistence of false memories in list re-call Journal of Memory amp Language 35 212-230
McDermott K B (1997) Priming on perceptual implicit memorytests can be achieved through presentation of associates Psycho-nomic Bulletin amp Review 4 582-586
McDermott K B amp Roediger H L III (1998) Attempting to avoidillusory memories Robust false recognition of associates persistsunder conditions of explicit warnings and immediate testing Journalof Memory amp Language 39 508-520
McDermott K B amp Watson J M (2001) The rise and fall of falserecall The impact of presentation duration Journal of Memory ampLanguage 45 160-176
McEvoy C L Nelson D L amp Komatsu T (1999) What is the con-nection between true and false memories The differential roles of in-teritem associations in recall and recognition Journal of Experimen-tal Psychology Learning Memory amp Cognition 25 1177-1194
Meyer D E amp Schvaneveldt R W (1971) Facilitation in recog-nizing pairs of words Evidence of a dependence between retrievaloperations Journal of Experimental Psychology 90 227-234
Miller M B amp Wolford G L (1999) Theoretical commentaryThe role of criterion shift in false memory Psychological Review106 398-405
Neely J H (1977) Semantic priming and retrieval from semanticmemory Roles of inhibitionless spreading activation and limited-
capacity attention Journal of Experimental Psychology General106 226-254
Nelson D L McEvoy C L amp Schreiber T A (1999) The Uni-versity of South Florida word association rhyme and word fragmentnorms Unpublished manuscript University of South Florida Tampa
Nelson D L Schreiber TA amp McEvoy C L (1992) Processing im-plicit and explicit representations Psychological Review 99 322-348
Norman K A amp Schacter D L (1997) False recognition in youngerand older adults Exploring the characteristics of illusory memoriesMemory amp Cognition 25 838-848
Paivio A Yuille J C amp Madigan S A (1968) Concreteness im-agery and meaningfulness values for 925 nouns Journal of Experi-mental Psychology Monographs 76 (1 Pt 2)
Payne D G Elie C J Blackwell J M amp Neuschatz J S (1996)Memory illusions Recalling recognizing and recollecting events thatnever occurred Journal of Memory amp Language 35 261-285
Reyna V F amp Brainerd C J (1995) Fuzzy-trace theory An interimsynthesis Learning amp Individual Differences 7 1-75
Roediger H L III Balota D A amp Watson J M (2001) Spreadingactivation and the arousal of false memories In H L Roediger III J SNairne I Neath amp A M Surprenant (Eds) The nature of remem-bering Essays in honor of Robert G Crowder (pp 95-115) Wash-ington DC American Psychological Association
Roediger H L III amp McDermott K B (1995) Creating false mem-ories Remembering words not presented in lists Journal of Experi-mental Psychology Learning Memory amp Cognition 21 803-814
Roediger H L III amp McDermott K B (1999) False alarms aboutfalse memories Psychological Review 106 406-410
Roediger H L III amp McDermott K B (2000) Tricks of memoryCurrent Directions in Psychological Science 9 123-127
Roediger H L III McDermott K B amp Pisoni D B (2001) Rec-ollection of illusory voices Manuscript submitted for publication
Roediger H L III McDermott K B amp Robinson K J (1998)The role of associative processes in creating false memories In M AConway S E Gathercole amp C Cornoldi (Eds) Theories of memoryII (pp 187-245) Hove UK Psychological Press
Russell W A amp Jenkins J J (1954) The complete Minnesotanorms for responses to 100 words from the KentndashRosanoff Word As-sociation Test (Tech Rep No 11 Contract N8 ONR 66216 Officeof Naval Research) University of Minnesota Minneapolis
Schacter D L Israel L amp Racine C (1999) Suppressing falserecognition in younger and older adults The distinctiveness heuris-tic Journal of Memory amp Language 40 1-24
Schacter D L Verfaellie M amp Pradere D (1996) The neuro-psychology of memory illusions False recall and recognition in am-nesic patients Journal of Memory amp Language 35 319-334
Seamon J G Luo C R amp Gallo D A (1998) Creating false mem-ories of words with or without recognition of list items Evidence fornonconscious processes Psychological Science 9 20-26
Smith R E amp Hunt R R (1998) Presentation modality affects falsememory Psychonomic Bulletin amp Review 5 710-715
Stadler M A Roediger H L III amp McDermott K B (1999)Norms for word lists that create false memories Memory amp Cogni-tion 27 494-500
Thapar A amp McDermott K B (2001) False recall and false recog-nition induced by presentation of associated words Effects of retentioninterval and level of processing Memory amp Cognition 29 424-432
Toglia M P amp Battig W F (1978) Handbook of semantic wordnorms Hillsdale NJ Erlbaum
Toglia M P Neuschatz J S amp Goodwin K A (1999) Recall ac-curacy and illusory memories When more is less Memory 7 233-256
Tulving E (1974) Cue-dependent forgetting American Scientist 6274-82
Tulving E (1979) Relation between encoding specificity and levelsof processing In L S Cermak amp F I M Craik (Eds) Levels of pro-cessing in human memory (pp 405-428) Hillsdale NJ Erlbaum
Tulving E (1985) Memory and consciousness Canadian Psycholo-gist 26 1-12
Tun P A Wingfield A Rosen M J amp Blanchard L (1998)Response latencies for false memories Gist-based processes in nor-mal aging Psychology amp Aging 13 230-241
398 ROEDIGER WATSON MCDERMOTT AND GALLO
Underwood B J (1965) False recognition produced by implicit ver-bal responses Journal of Experimental Psychology 70 122-129
Weldon M S (1991) Mechanisms underlying priming on perceptualtests Journal of Experimental Psychology Learning Memory ampCognition 17 526-541
Wickens T D amp Hirshman E (2000) False memories and statisticaldecision theory Comment on Miller and Wolford (1999) and Roedi-ger and McDermott (1999) Psychological Review 107 377-383
Wixted J T amp Stretch V (2000) The case against a criterion-shiftaccount of false memory Psychological Review 107 368-376
Zechmeister E B (1969) Orthographic distinctiveness Journal ofVerbal Learning amp Verbal Behavior 8 754-761
NOTES
1 The orthographic distinctiveness of a critical item was obtained bypresenting 72 subjects with all 55 critical items visually and havingthem rate the words in accordance with procedures developed by Zech-meister (1969) Briefly the subjects were instructed to rate each wordfor its physical appearance in terms of distinctiveness in the languageon a 9-point scale with 1 being not very distinctive 5 corresponding toaverage distinctiveness and 9 being most distinctive To avoid orderingeffects three different randomized orderings of words were used acrossroughly equal groups of subjects In the case of the critical item sleep
for example the distinctiveness is 515 Corresponding values for but-terfly and for man are 744 and 213 respectively Although ortho-graphic distinctiveness was not included in our multiple regressionanalyses this predictor is included in Appendix C
2 The orthographic neighborhood size or Coltheartrsquos N of each crit-ical item was defined as the number of words that can be formed bychanging a single letter in the word (Coltheart Davelaar Jonasson ampBesner 1977) For example because it has no orthographic neighborsColtheartrsquos N for the critical item butterfly is zero In contrast sleep hasfive orthographic neighbors (ie sheep sleek sleet steep and sweep)The reference set of words used were those in Ku Iumlcera and Francisrsquos(1967) norms Although Coltheartrsquos N was not included in our multipleregression analyses this predictor is included in Appendix C
3 Word length orthographic distinctiveness and orthographic neigh-borhood size of the critical item were all highly correlated Specificallythe bivariate correlations among these three predictors were as followslength and orthographic distinctiveness r 5 184 t(53) 5 1117 p lt05 orthographic distinctiveness and orthographic neighborhood sizer 5 66 t(53) 5 640 p lt 05 length and orthographic neighbor-hood size r 5 63 t(53) 5 582 p lt 05 Of these three variablesword length had the largest bivariate correlation with false recall (r 5
37) so it was the predictor used in the multiple regression analyses4 The multiple regression data matrix is available at wwwiacwustl
edunmclWeb
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 399
APPENDIX A The Fifty-Five 15-Word Lists and Their Associated Critical Items
With Backward Associative Strength (BAS) and Forward Associative Strength (FAS)Values Used in the Multiple Regression Analysis of False Recall
Item Word BAS FAS Item Word BAS FAS
Anger mad 393 412 Black white 655 557fear 020 059 dark 111 100hate 028 109 cat 000 043rage 541 042 charred 023 000temper 182 000 night 000 021fury 306 000 funeral 034 000ire 179 000 color 074 050wrath 128 000 grief 000 000happy 000 042 blue 028 000fight 034 000 death 016 014hatred 070 000 ink 020 000mean 090 000 bottom 000 000calm 000 000 coal 288 000emotion 000 000 brown 338 000enrage 378 000 gray 365 000
Mean 157 044 Mean 130 052Army Navy 543 500 Bread butter 364 487
NotemdashUnless otherwise indicated with an asterisk all BAS and FAS values are drawn from the associative norms of Nel-son McEvoy and Schreiber (1999) Those values shown with an asterisk were obtained by the authors by using normingprocedures similar to those of Nelson et al (1999) The corresponding mean for these BAS and FAS values for each list arealso represented
406 ROEDIGER WATSON MCDERMOTT AND GALLO
APPENDIX B Data Matrix of the Fifty-Five 15-Word Lists and Associated Critical Items
Used in the Multiple Regression Analysis of False Recall
Critical False False Log Vrd Raw Ortho Item Recall Recgn Length Freq Concrete FAS BAS Connect Recall Freq Dist Colt N
(Manuscript received July 10 2000revision accepted for publication February 23 2001)
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 397
Gallo D A McDermott K B Percer J M amp Roediger H LIII (2001) Modality effects in false recall and false recognitionJournal of Experimental Psychology Learning Memory amp Cogni-tion 27 339-353
Gallo D A Roberts M J amp Seamon J G (1997) Rememberingwords not presented in lists Can we avoid creating false memoriesPsychonomic Bulletin amp Review 4 271-276
Gallo D A amp Roediger H L III (2001) Variability among wordlists in evoking associative memory illusions Manuscript submittedfor publication
Gallo D A Roediger H L III amp McDermott K B (2001) As-sociative false recognition occurs without strategic criterion shiftsPsychonomic Bulletin amp Review 8 579-586
Hicks J L amp Marsh R L (2001) False recognition occurs more fre-quently during source identification than during oldndashnew recogni-tion Journal of Experimental Psychology Learning Memory ampCognition 27 375-383
Hunt R R amp Einstein G O (1981) Relational and item-specific in-formation in memory Journal of Verbal Learning amp Verbal Behav-ior 19 497-514
Hunt R R amp McDaniel M A (1993) The enigma of organizationand distinctiveness Journal of Memory amp Language 32 421-445
Israel L amp Schacter D L (1997) Pictorial encoding reduces falserecognition of semantic associates Psychonomic Bulletin amp Review4 577-581
Jacoby L L Kelley C M amp Dywan J (1989) Memory attribu-tions In H L Roediger III amp F I M Craik (Eds) Varieties of mem-ory and consciousness Essays in honour of Endel Tulving (pp 391-422) Hillsdale NJ Erlbaum
Johnson M K Hashtroudi S amp Lindsay D S (1993) Sourcemonitoring Psychological Bulletin 114 3-28
Johnson M K amp Raye C L (1981) Reality monitoring Psycho-logical Review 88 67-85
Kellogg RT (2001) Presentation modality and mode of recall in ver-bal false memory Journal of Experimental Psychology LearningMemory amp Cognition 27 913-919
Kirkpatrick E A (1894) An experimental study of memory Psy-chological Review 1 602-609
Ku Iumlcera H amp Francis W (1967) Computational analysis of present-day American English Providence RI Brown University Press
Marsh E J McDermott K B amp Roediger H L III (2001) Doretrieval factors play a role in the creation of of false memories Man-uscript submitted for publication
Mather M Henkel L A amp Johnson M J (1997) Evaluatingcharacteristics of false memories Rememberknow judgments andmemory characteristics questionnaire compared Memory amp Cogni-tion 25 826-837
McDermott K B (1996) The persistence of false memories in list re-call Journal of Memory amp Language 35 212-230
McDermott K B (1997) Priming on perceptual implicit memorytests can be achieved through presentation of associates Psycho-nomic Bulletin amp Review 4 582-586
McDermott K B amp Roediger H L III (1998) Attempting to avoidillusory memories Robust false recognition of associates persistsunder conditions of explicit warnings and immediate testing Journalof Memory amp Language 39 508-520
McDermott K B amp Watson J M (2001) The rise and fall of falserecall The impact of presentation duration Journal of Memory ampLanguage 45 160-176
McEvoy C L Nelson D L amp Komatsu T (1999) What is the con-nection between true and false memories The differential roles of in-teritem associations in recall and recognition Journal of Experimen-tal Psychology Learning Memory amp Cognition 25 1177-1194
Meyer D E amp Schvaneveldt R W (1971) Facilitation in recog-nizing pairs of words Evidence of a dependence between retrievaloperations Journal of Experimental Psychology 90 227-234
Miller M B amp Wolford G L (1999) Theoretical commentaryThe role of criterion shift in false memory Psychological Review106 398-405
Neely J H (1977) Semantic priming and retrieval from semanticmemory Roles of inhibitionless spreading activation and limited-
capacity attention Journal of Experimental Psychology General106 226-254
Nelson D L McEvoy C L amp Schreiber T A (1999) The Uni-versity of South Florida word association rhyme and word fragmentnorms Unpublished manuscript University of South Florida Tampa
Nelson D L Schreiber TA amp McEvoy C L (1992) Processing im-plicit and explicit representations Psychological Review 99 322-348
Norman K A amp Schacter D L (1997) False recognition in youngerand older adults Exploring the characteristics of illusory memoriesMemory amp Cognition 25 838-848
Paivio A Yuille J C amp Madigan S A (1968) Concreteness im-agery and meaningfulness values for 925 nouns Journal of Experi-mental Psychology Monographs 76 (1 Pt 2)
Payne D G Elie C J Blackwell J M amp Neuschatz J S (1996)Memory illusions Recalling recognizing and recollecting events thatnever occurred Journal of Memory amp Language 35 261-285
Reyna V F amp Brainerd C J (1995) Fuzzy-trace theory An interimsynthesis Learning amp Individual Differences 7 1-75
Roediger H L III Balota D A amp Watson J M (2001) Spreadingactivation and the arousal of false memories In H L Roediger III J SNairne I Neath amp A M Surprenant (Eds) The nature of remem-bering Essays in honor of Robert G Crowder (pp 95-115) Wash-ington DC American Psychological Association
Roediger H L III amp McDermott K B (1995) Creating false mem-ories Remembering words not presented in lists Journal of Experi-mental Psychology Learning Memory amp Cognition 21 803-814
Roediger H L III amp McDermott K B (1999) False alarms aboutfalse memories Psychological Review 106 406-410
Roediger H L III amp McDermott K B (2000) Tricks of memoryCurrent Directions in Psychological Science 9 123-127
Roediger H L III McDermott K B amp Pisoni D B (2001) Rec-ollection of illusory voices Manuscript submitted for publication
Roediger H L III McDermott K B amp Robinson K J (1998)The role of associative processes in creating false memories In M AConway S E Gathercole amp C Cornoldi (Eds) Theories of memoryII (pp 187-245) Hove UK Psychological Press
Russell W A amp Jenkins J J (1954) The complete Minnesotanorms for responses to 100 words from the KentndashRosanoff Word As-sociation Test (Tech Rep No 11 Contract N8 ONR 66216 Officeof Naval Research) University of Minnesota Minneapolis
Schacter D L Israel L amp Racine C (1999) Suppressing falserecognition in younger and older adults The distinctiveness heuris-tic Journal of Memory amp Language 40 1-24
Schacter D L Verfaellie M amp Pradere D (1996) The neuro-psychology of memory illusions False recall and recognition in am-nesic patients Journal of Memory amp Language 35 319-334
Seamon J G Luo C R amp Gallo D A (1998) Creating false mem-ories of words with or without recognition of list items Evidence fornonconscious processes Psychological Science 9 20-26
Smith R E amp Hunt R R (1998) Presentation modality affects falsememory Psychonomic Bulletin amp Review 5 710-715
Stadler M A Roediger H L III amp McDermott K B (1999)Norms for word lists that create false memories Memory amp Cogni-tion 27 494-500
Thapar A amp McDermott K B (2001) False recall and false recog-nition induced by presentation of associated words Effects of retentioninterval and level of processing Memory amp Cognition 29 424-432
Toglia M P amp Battig W F (1978) Handbook of semantic wordnorms Hillsdale NJ Erlbaum
Toglia M P Neuschatz J S amp Goodwin K A (1999) Recall ac-curacy and illusory memories When more is less Memory 7 233-256
Tulving E (1974) Cue-dependent forgetting American Scientist 6274-82
Tulving E (1979) Relation between encoding specificity and levelsof processing In L S Cermak amp F I M Craik (Eds) Levels of pro-cessing in human memory (pp 405-428) Hillsdale NJ Erlbaum
Tulving E (1985) Memory and consciousness Canadian Psycholo-gist 26 1-12
Tun P A Wingfield A Rosen M J amp Blanchard L (1998)Response latencies for false memories Gist-based processes in nor-mal aging Psychology amp Aging 13 230-241
398 ROEDIGER WATSON MCDERMOTT AND GALLO
Underwood B J (1965) False recognition produced by implicit ver-bal responses Journal of Experimental Psychology 70 122-129
Weldon M S (1991) Mechanisms underlying priming on perceptualtests Journal of Experimental Psychology Learning Memory ampCognition 17 526-541
Wickens T D amp Hirshman E (2000) False memories and statisticaldecision theory Comment on Miller and Wolford (1999) and Roedi-ger and McDermott (1999) Psychological Review 107 377-383
Wixted J T amp Stretch V (2000) The case against a criterion-shiftaccount of false memory Psychological Review 107 368-376
Zechmeister E B (1969) Orthographic distinctiveness Journal ofVerbal Learning amp Verbal Behavior 8 754-761
NOTES
1 The orthographic distinctiveness of a critical item was obtained bypresenting 72 subjects with all 55 critical items visually and havingthem rate the words in accordance with procedures developed by Zech-meister (1969) Briefly the subjects were instructed to rate each wordfor its physical appearance in terms of distinctiveness in the languageon a 9-point scale with 1 being not very distinctive 5 corresponding toaverage distinctiveness and 9 being most distinctive To avoid orderingeffects three different randomized orderings of words were used acrossroughly equal groups of subjects In the case of the critical item sleep
for example the distinctiveness is 515 Corresponding values for but-terfly and for man are 744 and 213 respectively Although ortho-graphic distinctiveness was not included in our multiple regressionanalyses this predictor is included in Appendix C
2 The orthographic neighborhood size or Coltheartrsquos N of each crit-ical item was defined as the number of words that can be formed bychanging a single letter in the word (Coltheart Davelaar Jonasson ampBesner 1977) For example because it has no orthographic neighborsColtheartrsquos N for the critical item butterfly is zero In contrast sleep hasfive orthographic neighbors (ie sheep sleek sleet steep and sweep)The reference set of words used were those in Ku Iumlcera and Francisrsquos(1967) norms Although Coltheartrsquos N was not included in our multipleregression analyses this predictor is included in Appendix C
3 Word length orthographic distinctiveness and orthographic neigh-borhood size of the critical item were all highly correlated Specificallythe bivariate correlations among these three predictors were as followslength and orthographic distinctiveness r 5 184 t(53) 5 1117 p lt05 orthographic distinctiveness and orthographic neighborhood sizer 5 66 t(53) 5 640 p lt 05 length and orthographic neighbor-hood size r 5 63 t(53) 5 582 p lt 05 Of these three variablesword length had the largest bivariate correlation with false recall (r 5
37) so it was the predictor used in the multiple regression analyses4 The multiple regression data matrix is available at wwwiacwustl
edunmclWeb
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 399
APPENDIX A The Fifty-Five 15-Word Lists and Their Associated Critical Items
With Backward Associative Strength (BAS) and Forward Associative Strength (FAS)Values Used in the Multiple Regression Analysis of False Recall
Item Word BAS FAS Item Word BAS FAS
Anger mad 393 412 Black white 655 557fear 020 059 dark 111 100hate 028 109 cat 000 043rage 541 042 charred 023 000temper 182 000 night 000 021fury 306 000 funeral 034 000ire 179 000 color 074 050wrath 128 000 grief 000 000happy 000 042 blue 028 000fight 034 000 death 016 014hatred 070 000 ink 020 000mean 090 000 bottom 000 000calm 000 000 coal 288 000emotion 000 000 brown 338 000enrage 378 000 gray 365 000
Mean 157 044 Mean 130 052Army Navy 543 500 Bread butter 364 487
NotemdashUnless otherwise indicated with an asterisk all BAS and FAS values are drawn from the associative norms of Nel-son McEvoy and Schreiber (1999) Those values shown with an asterisk were obtained by the authors by using normingprocedures similar to those of Nelson et al (1999) The corresponding mean for these BAS and FAS values for each list arealso represented
406 ROEDIGER WATSON MCDERMOTT AND GALLO
APPENDIX B Data Matrix of the Fifty-Five 15-Word Lists and Associated Critical Items
Used in the Multiple Regression Analysis of False Recall
Critical False False Log Vrd Raw Ortho Item Recall Recgn Length Freq Concrete FAS BAS Connect Recall Freq Dist Colt N
(Manuscript received July 10 2000revision accepted for publication February 23 2001)
398 ROEDIGER WATSON MCDERMOTT AND GALLO
Underwood B J (1965) False recognition produced by implicit ver-bal responses Journal of Experimental Psychology 70 122-129
Weldon M S (1991) Mechanisms underlying priming on perceptualtests Journal of Experimental Psychology Learning Memory ampCognition 17 526-541
Wickens T D amp Hirshman E (2000) False memories and statisticaldecision theory Comment on Miller and Wolford (1999) and Roedi-ger and McDermott (1999) Psychological Review 107 377-383
Wixted J T amp Stretch V (2000) The case against a criterion-shiftaccount of false memory Psychological Review 107 368-376
Zechmeister E B (1969) Orthographic distinctiveness Journal ofVerbal Learning amp Verbal Behavior 8 754-761
NOTES
1 The orthographic distinctiveness of a critical item was obtained bypresenting 72 subjects with all 55 critical items visually and havingthem rate the words in accordance with procedures developed by Zech-meister (1969) Briefly the subjects were instructed to rate each wordfor its physical appearance in terms of distinctiveness in the languageon a 9-point scale with 1 being not very distinctive 5 corresponding toaverage distinctiveness and 9 being most distinctive To avoid orderingeffects three different randomized orderings of words were used acrossroughly equal groups of subjects In the case of the critical item sleep
for example the distinctiveness is 515 Corresponding values for but-terfly and for man are 744 and 213 respectively Although ortho-graphic distinctiveness was not included in our multiple regressionanalyses this predictor is included in Appendix C
2 The orthographic neighborhood size or Coltheartrsquos N of each crit-ical item was defined as the number of words that can be formed bychanging a single letter in the word (Coltheart Davelaar Jonasson ampBesner 1977) For example because it has no orthographic neighborsColtheartrsquos N for the critical item butterfly is zero In contrast sleep hasfive orthographic neighbors (ie sheep sleek sleet steep and sweep)The reference set of words used were those in Ku Iumlcera and Francisrsquos(1967) norms Although Coltheartrsquos N was not included in our multipleregression analyses this predictor is included in Appendix C
3 Word length orthographic distinctiveness and orthographic neigh-borhood size of the critical item were all highly correlated Specificallythe bivariate correlations among these three predictors were as followslength and orthographic distinctiveness r 5 184 t(53) 5 1117 p lt05 orthographic distinctiveness and orthographic neighborhood sizer 5 66 t(53) 5 640 p lt 05 length and orthographic neighbor-hood size r 5 63 t(53) 5 582 p lt 05 Of these three variablesword length had the largest bivariate correlation with false recall (r 5
37) so it was the predictor used in the multiple regression analyses4 The multiple regression data matrix is available at wwwiacwustl
edunmclWeb
FALSE RECALL A MULTIPLE REGRESSION ANALYSIS 399
APPENDIX A The Fifty-Five 15-Word Lists and Their Associated Critical Items
With Backward Associative Strength (BAS) and Forward Associative Strength (FAS)Values Used in the Multiple Regression Analysis of False Recall
Item Word BAS FAS Item Word BAS FAS
Anger mad 393 412 Black white 655 557fear 020 059 dark 111 100hate 028 109 cat 000 043rage 541 042 charred 023 000temper 182 000 night 000 021fury 306 000 funeral 034 000ire 179 000 color 074 050wrath 128 000 grief 000 000happy 000 042 blue 028 000fight 034 000 death 016 014hatred 070 000 ink 020 000mean 090 000 bottom 000 000calm 000 000 coal 288 000emotion 000 000 brown 338 000enrage 378 000 gray 365 000
Mean 157 044 Mean 130 052Army Navy 543 500 Bread butter 364 487
NotemdashUnless otherwise indicated with an asterisk all BAS and FAS values are drawn from the associative norms of Nel-son McEvoy and Schreiber (1999) Those values shown with an asterisk were obtained by the authors by using normingprocedures similar to those of Nelson et al (1999) The corresponding mean for these BAS and FAS values for each list arealso represented
406 ROEDIGER WATSON MCDERMOTT AND GALLO
APPENDIX B Data Matrix of the Fifty-Five 15-Word Lists and Associated Critical Items
Used in the Multiple Regression Analysis of False Recall
Critical False False Log Vrd Raw Ortho Item Recall Recgn Length Freq Concrete FAS BAS Connect Recall Freq Dist Colt N
NotemdashUnless otherwise indicated with an asterisk all BAS and FAS values are drawn from the associative norms of Nel-son McEvoy and Schreiber (1999) Those values shown with an asterisk were obtained by the authors by using normingprocedures similar to those of Nelson et al (1999) The corresponding mean for these BAS and FAS values for each list arealso represented
406 ROEDIGER WATSON MCDERMOTT AND GALLO
APPENDIX B Data Matrix of the Fifty-Five 15-Word Lists and Associated Critical Items
Used in the Multiple Regression Analysis of False Recall
Critical False False Log Vrd Raw Ortho Item Recall Recgn Length Freq Concrete FAS BAS Connect Recall Freq Dist Colt N
NotemdashUnless otherwise indicated with an asterisk all BAS and FAS values are drawn from the associative norms of Nel-son McEvoy and Schreiber (1999) Those values shown with an asterisk were obtained by the authors by using normingprocedures similar to those of Nelson et al (1999) The corresponding mean for these BAS and FAS values for each list arealso represented
406 ROEDIGER WATSON MCDERMOTT AND GALLO
APPENDIX B Data Matrix of the Fifty-Five 15-Word Lists and Associated Critical Items
Used in the Multiple Regression Analysis of False Recall
Critical False False Log Vrd Raw Ortho Item Recall Recgn Length Freq Concrete FAS BAS Connect Recall Freq Dist Colt N
NotemdashUnless otherwise indicated with an asterisk all BAS and FAS values are drawn from the associative norms of Nel-son McEvoy and Schreiber (1999) Those values shown with an asterisk were obtained by the authors by using normingprocedures similar to those of Nelson et al (1999) The corresponding mean for these BAS and FAS values for each list arealso represented
406 ROEDIGER WATSON MCDERMOTT AND GALLO
APPENDIX B Data Matrix of the Fifty-Five 15-Word Lists and Associated Critical Items
Used in the Multiple Regression Analysis of False Recall
Critical False False Log Vrd Raw Ortho Item Recall Recgn Length Freq Concrete FAS BAS Connect Recall Freq Dist Colt N
NotemdashUnless otherwise indicated with an asterisk all BAS and FAS values are drawn from the associative norms of Nel-son McEvoy and Schreiber (1999) Those values shown with an asterisk were obtained by the authors by using normingprocedures similar to those of Nelson et al (1999) The corresponding mean for these BAS and FAS values for each list arealso represented
406 ROEDIGER WATSON MCDERMOTT AND GALLO
APPENDIX B Data Matrix of the Fifty-Five 15-Word Lists and Associated Critical Items
Used in the Multiple Regression Analysis of False Recall
Critical False False Log Vrd Raw Ortho Item Recall Recgn Length Freq Concrete FAS BAS Connect Recall Freq Dist Colt N
NotemdashUnless otherwise indicated with an asterisk all BAS and FAS values are drawn from the associative norms of Nel-son McEvoy and Schreiber (1999) Those values shown with an asterisk were obtained by the authors by using normingprocedures similar to those of Nelson et al (1999) The corresponding mean for these BAS and FAS values for each list arealso represented
406 ROEDIGER WATSON MCDERMOTT AND GALLO
APPENDIX B Data Matrix of the Fifty-Five 15-Word Lists and Associated Critical Items
Used in the Multiple Regression Analysis of False Recall
Critical False False Log Vrd Raw Ortho Item Recall Recgn Length Freq Concrete FAS BAS Connect Recall Freq Dist Colt N
NotemdashUnless otherwise indicated with an asterisk all BAS and FAS values are drawn from the associative norms of Nel-son McEvoy and Schreiber (1999) Those values shown with an asterisk were obtained by the authors by using normingprocedures similar to those of Nelson et al (1999) The corresponding mean for these BAS and FAS values for each list arealso represented
406 ROEDIGER WATSON MCDERMOTT AND GALLO
APPENDIX B Data Matrix of the Fifty-Five 15-Word Lists and Associated Critical Items
Used in the Multiple Regression Analysis of False Recall
Critical False False Log Vrd Raw Ortho Item Recall Recgn Length Freq Concrete FAS BAS Connect Recall Freq Dist Colt N
NotemdashUnless otherwise indicated with an asterisk all BAS and FAS values are drawn from the associative norms of Nel-son McEvoy and Schreiber (1999) Those values shown with an asterisk were obtained by the authors by using normingprocedures similar to those of Nelson et al (1999) The corresponding mean for these BAS and FAS values for each list arealso represented
406 ROEDIGER WATSON MCDERMOTT AND GALLO
APPENDIX B Data Matrix of the Fifty-Five 15-Word Lists and Associated Critical Items
Used in the Multiple Regression Analysis of False Recall
Critical False False Log Vrd Raw Ortho Item Recall Recgn Length Freq Concrete FAS BAS Connect Recall Freq Dist Colt N