Imagination inflation is a fact, not an artifact: A reply to Pezdek and Eddy

Memory amp Cognition2001 29 (5) 719-729

A Western traveler encountering an Oriental philosopherasks him to describe the nature of the world

ldquoIt is a great ball resting on the flat back of the worldturtlerdquo

ldquoAh yes but what does the world turtle stand onrdquo

ldquoOn the back of a still larger turtlerdquo

ldquoYes but what does he stand onrdquo

ldquoA very perceptive question But itrsquos no use mister itrsquosturtles all the way downrdquo (Sagan 1979 p 293)

Like Pezdek and Eddy (2001) we begin our reply withan instructive quote This well-known parable here toldby Carl Sagan illustrates the dangers of building an argu-ment by stacking faulty thinking on top of faulty thinkingAlthough Pezdek and Eddy do raise some valid points inthe end their criticisms of imagination inflation are a tur-ret of turtles

Let us begin with their very first sentence Contrary toPezdek and Eddyrsquos claims their paper does not addressldquothe conditions under which false autobiographical eventsare likely to be planted in memoryrdquo (p 707) Imaginationinflation is not about implanting memories Pezdek andEddy still continue to assert that it is even though wehave stressed this point with them before (Garry amp Lof-tus 2000) Imagination inflation is an increase in confi-dence that a fictional or hypothetical event that was merelyimagined was actually experienced This point is not triv-ial because other imagination inflation researchers havebeen careful to distinguish between imagination inflationand an implanted memory For instance in two differentpapers Heaps and Nash (1999 in press) described imag-

ination inflation variouslymdashand accuratelymdashas a changein subjective likelihood a false belief about the past ora mechanism by which ldquoimagination can serve to raise sub-sequent likelihood judgmentsrdquo (Heaps amp Nash in press)Mazzoni (2001) has clearly captured the difference be-tween confidence that an event actually happened and amemory for that event We are all confident they note thatour births occurred as our mothers have explained themto us although we do not actually have memories for them

When we change peoplersquos confidence that an eventmight have happened to them have we affected theirmemory We think the answer is yes because we havechanged the way they think about and report their pastBut is that an implanted memory No because the imag-ination inflation effect does not meet the criteria othershave proposed for implanting a false memory Take Hy-man and colleaguesrsquo (Hyman amp Kleinknecht 1999 Hy-man amp Loftus 1998) proposal that implanting falsememories involves three conditions First subjects mustaccept that the event is plausible Second they must con-struct details about the false event that include an imageplus a narrative Finally subjects must misattribute thesource of that information wrongly ascribing the mentalinformation to genuine experience rather than to theirown imagination Increased confidence that the targetevent did happenmdashthe definition of imagination infla-tionmdashmeets the first criterion but there is no evidencethat it meets the remaining two Of course this is not tosay that some subjects who participate in an imaginationinflation experiment do not develop false memories aboutthe events they imagine but our aim has been to examineonly the change in confidence

The title of the original paper ldquoImagination InflationImagining a Childhood Event Inflates Confidence ThatIt Occurredrdquo (Garry Manning Loftus amp Sherman 1996)makes the definition of imagination inflation clear andso does quoting Garry and Polaschek (2000) more accu-rately than Pezdek and Eddy did Pezdek and Eddy wrote

719 Copyright 2001 Psychonomic Society Inc

Kimberley Wade is supported by a Victoria University Targeted PhDScholarship We thank Dave Kenny Don Read Scott Brown DevonPolaschek Matt Gerrie Beth Loftus and our colleagues in the VictoriaUniversity Cognitive Workshop Correspondence should be addressed toM Garry School of Psychology Victoria University of Wellington Box600 Wellington New Zealand (e-mail maryannegarryvuwacnz)

Imagination inflation is a fact not an artifactA reply to Pezdek and Eddy

MARYANNE GARRY STEFANIE J SHARMAN KIMBERLEY A WADEMAREE J HUNT and PETER J SMITH

Victoria University of Wellington Wellington New Zealand

Pezdek and Eddy (2001) claim to prove that imagination inflation is a spurious effect caused by re-gression to the mean (RTM) They make four predictions about what patterns of data would demon-strate a genuine effect for imagination versus those that would be explainable by RTM We review eachof those predictions and demonstrate significant problems with them We conclude that imaginationinflation is a genuine effect and that Pezdek and Eddyrsquos work has contributed to the growing researchshowing that when people imagine fictitious events from long ago they become more confident thatthose false events were genuine experiences

720 GARRY SHARMAN WADE HUNT AND SMITH

In summarizing the results of this experiment and severalothers using the imagination inflation paradigm Garry andPolaschek (2000) concluded that ldquoa growing body of lit-erature shows that imagining contrary-to-truth experiencescan change memoryrdquo (p 6) Furthermore ldquowhen peoplethink about or imagine a false event entire false memoriescan be implanted Imagination inflation can occur evenwhen there is no overt social pressure and when hypothet-ical events are imagined only brieflyrdquo (p 708ndash709)

While Garry and Polaschekrsquos (2000) review paper fo-cused on imagination inflation it did so in the context ofa larger review on imagination and memory In thatlarger context the few specific lines Pezdek and Eddyquoted from the Garry and Polaschek abstract make senseThus when the Garry and Polaschek abstract foreshad-ows that ldquowhen people think about or imagine a falseevent entire false memories can be implantedrdquo it refersto the opening section of the Garry and Polaschek paperwhich reviews what we might call the ldquoLost in the Mallrdquogenre of studies Loftus (1993) Loftus and Pickrell (1995)and Hyman and Pentland (1996) Later on in the paperwhen the focus is on imagination inflation Garry andPolaschek refer only to the ldquoeffect of imagination onmemoryrdquo (p 7) and even define imagination inflationrather tightly as the ldquoconfidence-boosting effect of imag-inationrdquo (p 7)

Now that we know what effect we are talking about letus turn to the heart of Pezdek and Eddyrsquos argument Putbluntly they assert that Garry et alrsquos (1996) results mightnot show any confidence-inflating effect of imaginationat all Instead Pezdek and Eddy say these ldquofindings couldbe explained as simply a case of regression toward themeanrdquo (p 709 hereafter RTM) Pezdek and Eddy putforth a number of arguments as evidence of their claimbut we believe each argument is flawed Below we sum-marize Pezdek and Eddyrsquos experiment review their fourpredictions and examine the strength of their argumentsin light of the evidence First however we begin with abasic overview of RTM

REGRESSION TO THE MEANTHE (VERY) SHORT COURSE

Every time researchers measure some construct theyunwittingly drag some error into their measurement Takefor instance the case each of us might face when ad-ministering an Introductory Psychology exam There arestudents for whom error depresses their score and stu-dents for whom error inflates their score So what is theconscientious college professor to do Suppose you gavethe exam again to the whole class Of course you wouldgive a parallel form of the exam but it would not be per-fectly correlated with the first exam What might hap-pen The end result would be that some students wouldfind that their retest scores increased while others wouldf ind that their scores decreased The mean howeverwould stay the same as would the standard deviation and

the shape of the distribution in general With each reteststudents would change and test scores would change aslong as tests were not perfectly correlated In short withsubsequent tests error would be shuffled around differ-ently The shuffling around would tend to work so thatscores farther away from the mean moved closer to themean on a retest Instead of referring to this shufflingaround under the curve by a tedious name such as ldquoshuf-fling around under the curverdquo we call it RTM

Several other interesting features of RTM are outsidethe scope of this article and the reader wanting more thanthis (very) short course could do no better than to readCampbell and Kennyrsquos (1999) book

Before we turn our attention to Pezdek and Eddyrsquos spec-ific predictions there is an important point to noteThroughout their paper Pezdek and Eddy seem to mis-take regression to the mean of the items with regressionto the midpoint of the scale and this confusion under-pins much of their thinking

PEZDEK AND EDDYrsquoS PREDICTIONS

Pezdek and Eddy hypothesize that if imagination in-flation is a real effect we should expect to see the follow-ing pattern of results

1 The residuals between actual posttest scores andRTM-adjusted (predicted) posttest scores should begreater than zero

2 There should be a test occasion 3 condition inter-action

3 All mean likelihood ratings should increase from thepre- to posttest not just those initially rated as 1ndash4

4 There should be a time 3 imagination 3 age inter-action

Below we examine the logic behind each predictionthe data on which Pezdek and Eddy rely and the variousconclusions that can be responsibly drawn

1A The Prediction Actual Versus RTM-Adjusted Residuals Should Be Greater Than Zero

Pezdek and Eddy argue that if imagination inflation isa real effect posttest confidence for imagined events willbe above and beyond the increase in confidence producedby RTM alone To analyze their data according to thishypothesis they did the following analysis

First they found the correlation between pretest andposttest scores for not imagined (control) target eventsand used the correlation to estimate the effects of RTMIn other words they constructed a regression line to pre-dict the change in posttest confidence scores that are at-tributable only to RTM Second Pezdek and Eddy usedthis regression line to predict the posttest scores for theimagined events again assuming that only RTM was re-sponsible for changing these scores Finally they com-pared the predicted imagined posttest scores with the ob-tained imagined posttest scores to produce residual scoresWhere the residuals are positive there is undoubtedly an

IMAGINATION INFLATION 721

effect over and above RTM However what does it meanwhen the residuals are not different from zero Pezdekand Eddy argue that it means imagination has no effectand that imagination inflation is nothing more than RTMdisguised as imagination inflation But they are wrong

1B The Turtle Their Analysis Does Not WorkContrary to Pezdek and Eddyrsquos claim their residual-

ized change scores analysis is not a strong statistical basisfor drawing any conclusion There are two broad prob-lems with the attempt to predictmdashand then correct formdashRTM using residualized change scores The first is ageneral problem Several researchers have noted that themethod is fraught with difficulties (Campbell amp Kenny1999 Rogosa Brandt amp Zimowski 1982 Rogosa ampWillet 1985) In the end Campbell and Kenny recom-mend that if the purpose of the research is to examinecauses of change then ldquoit is inadvisable ever to use resid-ualized change scores or estimated true scoresrdquo (p 99)The second problem is specific to imagination inflationPezdek and Eddyrsquos use of residualized change scores justdoes not work Because this article is about imaginationinflation in particular let us concentrate on the secondproblem We shall see that their formula actually fails todetect imagination inflation in some circumstancesBelow we illustrate this problematic situation

Hypothetical situation Failing to detect imagina-tion inflation Suppose we give a sample of people thesame test on two different occasions If there is no inter-vening treatment no measurement error and no otherbehavioral change among our subjects then each per-sonrsquos pretest score and posttest score should be the sameIn other words pretest and posttest scores would be per-fectly correlated (r 5 1) as in Figure 1A However amore realistic scenario is that our measurements will in-clude some error andor some of our subjectsrsquo behaviorwill change over time Therefore we actually expect a shiftaway from the perfect correlation line and toward themean Figure 1A shows how the predicted y line shifts tow-ard the mean The amount of regression can be estimatedby looking at the distance between the perfect correlationline and this predicted line (Campbell amp Kenny 1999)

Suppose we now assume in this hypothetical examplethat imagination inflation has a consistent and genuineeffect on confidence Indeed such an effect is Pezdekand Eddyrsquos fundamental thesis that imagination shouldincrease all scores by the same amount not just lowscores Such a constant effect would produce a parallelline above the predicted y line1 Let us consider the firstof Pezdek and Eddyrsquos recommended analyses that a testoccasion 3 imagination condition interaction is evidenceof imagination inflation In our hypothetical examplethere will be a significant interaction Now let us con-sider the secondmdashand what Pezdek and Eddy claim istheir most sensitive techniquemdashthe residualized changescores analysis What will this analysis f ind It willnever be able to find the effect because it is unable to de-

tect imagination inflation when the shape of the distrib-utions does not change from pretest to posttest The effectbecomes clear if we replot Figure 1A and use standard-ized scores to produce Figure 1B What will happen tothe plot The regression and imagination inflation lineswill in fact collapse on top of one another and both gothrough the y-intercept at zero In short no matter howlarge the effect size the technique touted by Pezdek andEddy as very sensitive to RTM is insensitive to imaginationinflation if imagination inflation is a homogenous effect

In sum Pezdek and Eddy claim that their residualanalysis ldquoprovides the strongest statistical basis for con-cluding that regression to the meanmdashnot imagination in-flationrdquo (p 715)mdashcaused their results However wehave demonstrated that when the shape of the distribu-tion does not change from pretest to posttest the analy-sis will not actually detect an effect

2A The Prediction There Should Be a Test Occasion and Condition Interaction

According to Pezdek and Eddy the mean change inconfidence ratings for imagined events should be greaterthan the mean change for not imagined events from pre-test to posttest A lack of interaction would signify no ef-fect over and above RTM

2B The Turtle An ANOVA Is Not Always the Best Way to Go

In some ways the interaction between test occasionand imagination condition is intuitively appealing In-deed the difference between the imagined and not imag-ined change scores is what is evaluated in the interactionterm When it is different from zero there is an inter-action However the extent to which the analysis of vari-ance (ANOVA) reveals an interaction depends not onlyon whether there is an interaction but on other factors aswell It matters for example to what extent the data meetthe ANOVA assumptions When some assumptions areviolated the ANOVA will not reveal an effect that is therebut when other assumptions are violated it might signalan effect that is not really there For example if the dataare not normally distributed the ANOVA will be less ableto fit a model to the data and it may not be sensitive toan effect should one exist In imagination inflation re-search target events are deliberately constructed so thatthe majority of subjects rate them low at pretest thusthe data are skewed Outliers pose an additional problemas does a change in variance between pre- and posttestand across event(s) If the variance changes dramaticallythe ANOVA copes by fitting a common interpolatedvariance If for instance one set of measurements has alarge variance and the other has a small variance theANOVA assumes a moderate variance overall This mod-erate assumption causes the ANOVA to assume a lowervariance than is true in one of the subsets which mightproduce false significance for effects in these parts ofthe data In addition it will assume a higher variance than


is true in the other subset which might fail to detect gen-uine effects found in the other parts of the data Take forexample Pezdek and Eddyrsquos own Table 2 where theirstandard deviations vary by about a factor of 10 In shortevaluating the interaction term in an ANOVA might be agood choice in some situations but not in others

One solution to the violations of the ANOVA assump-tions is the ldquoproportion increaserdquo analysis that Garry et al(1996) used (reproduced below as Figure 2) It elimi-nates the problems caused by changing variance acrossevents the standard deviations in Garry et al varied by

a factor of nearly 5 thus their choice of dependent mea-sure (treating events as subjects) made the heterogeneityof variance across events irrelevant As a reminder Fig-ure 2 shows the percentage of subjects inflating for eachevent classified according to whether that event was orwas not imagined A paired t test treating events as casesshowed that events were more likely to show inflationwhen subjects imagined them than when they did notimagine them [t(7) 5 548 p lt 001] The effect size(J Cohen 1988) was 1942 Garry et al relied mainly ona categorical dependent measuremdashthe percentage of sub-

Figure 1 (A) Perfect correlation line imagination inflation line and regression line (B) Fig-ure 1A replotted with standardized scores

A

B


jects who inflated on each eventmdashand then comparedthat categorical change using a parametric t test

Pezdek and Eddy seem to argue that the pattern demon-strated in Figure 2 is somehow less interesting or mean-ingful than if Garry et al (1996) had shown evidence ofa test occasion 3 condition interaction In fact for someresearch questions the Garry et al type of analysis is theone that answers the research question best Take for in-stance a pharmaceutical company that believes it hasdeveloped a new hand cream designed to prevent handsfrom drying and cracking in the winter cold To gatherdata about the hand creamrsquos safety and efficacy the com-pany begins trials with human subjects Suppose one as-pect the company needs to evaluate is the nature and ex-tent of side effects such as the incidence and severity ofallergic reaction in the population One kind of allergicreaction it might look for consists of swelling of the handsThus the company will gather wrist measurements forall its subjects before applying the hand cream to half ofthem and a placebo cream to the other half Then they willwait one hour and measure everyonersquos wrists again If thehand cream groupsrsquo wrists inflate more than the controlgroupsrsquo do that is evidence of an allergic reaction

What kind of data analysis would be more meaningfulin this example the time 3 condition interaction or theproportion of subjects in each condition who showed in-flated wrists The time 3 condition interaction would beuninformative if a sizeable minority of subjects showedan allergic reaction because the substantial number ofsubjects who showed no change in wrist size would over-shadow the minority of those who might have shown size-able reactions By contrast looking at the data to see theproportion of subjects in each condition who experienced

some wrist inflation is much more informativemdashand isanalogous to what Garry et al (1996) did

In fact the parallels with the hand cream example arenumerous In the hand cream trials the most commonchange in wrist size should be no change Likewise inthe imagination inflation paradigm the most commonchange is no change Such a finding makes sense in lightof Heaps and Nashrsquos (1999) work They found a predis-position to hypnotic suggestion and dissociationmdashten-dencies held by a minority of the population (Waller ampRoss 1997)mdashpredicted imagination inflation a findingsupported by the work of Paddock et al (1998) In bothinstances the ldquoinflatersrsquordquo mean scores are already heav-ily affected by the zero scores These zero scores maskthe change whether one analyzes the data using subjectsas cases or events as cases (subjects 3 critical items ineach condition) yet in both instances the minority ldquoin-flatersrdquo are the more interesting group to study Finallythere is another obvious and important parallel hereClearly some subjects in the hand cream trial who re-ceive only a placebo will still show inflated wrists becausetheir hands swell as the day goes on or because of someexpectancy effect and so forth But the key question isthis When people experience an allergic reaction to thehand cream by how much do their wrists inflate in com-parison with those of controls Similarly if we ask theanalogous question of the Garry et al (1996) data wefind that 42152 imagined events inflated with a meaninflation of 300 (198) and 38152 not imagined eventsinflated with a mean inflation of 250 (135)

The problem as we see it is to find the right combi-nation of design measures and analytical technique todetect the influence of imagination On that count the pa-

Figure 2 Percentage of subjects who increased on each item (adapted from Garry et al 1996)


pers by Garry et al (1996) Heaps and Nash (1999) andPezdek and Eddy all fall short and future research onimagination inflation should look at ways to increase thesize of the effect modify the method and come up withbetter ways of analyzing the data

3A The Prediction All Mean Likelihood RatingsShould Increase From the Pretest to Posttest

If imagination inflation is a real effect all mean con-fidence ratings for imagined events including those rated5ndash8 should inflate from pretest to posttest

3B The Turtle There Is No Basisfor the Prediction

To address this third prediction Pezdek and Eddy an-alyze their data in several different ways First for eachcritical event they split the data into two groups basedon pretest responses Events that were initially rated 1ndash4were put into one group and events initially rated 5ndash8were put into the other For the sake of simplicity let usrefer to these two groups as the low and high pretestgroups respectively Then for each event in each groupPezdek and Eddy calculated the percentage of subjectswhose confidence increased decreased or stayed thesame from the pretest to posttest Life Events Inventory(LEI) Within each group they collapsed these changescores across all subjects and all events and classifiedthem according to whether the event was imagined or notimagined This analysis is what Garry et al (1996) didbut they confined their analysis to low events explainingthat the purpose of their study was to examine what hap-pens when people imagine an unlikely event not recall alikely one

Pezdek and Eddyrsquos analysis of their data and the com-parable analysis of the Garry et al (1996) data can besummarized into three broad findings

Finding 1 For low pretest events ratings increasedfor both imagined and not imagined events Garryet al (1996) found that for low events confidence rat-ings inflated when they were imagined and when theywere not imagined 34 of imagined events showed in-creased confidence and 25 of not imagined eventsshowed increased confidence Pezdek and Eddy found asimilar pattern in their data as did Heaps and Nash (1999)although Heaps and Nash did not present their findingsthis way in their paper we have reanalyzed their dataalong similar lines shown in Figure 3

Pezdek and Eddy say that ldquoregression toward the meanmight explain the upward shift in scores that occurred inboth conditionsrdquo (p 709) Of course they are absolutelycorrect on this point RTM is almost certainly movinglow scores upward in both conditions However they goon to say that ldquoif the results simply reflect regression to-ward the mean then (1) likelihood ratings for events ini-tially rated 1ndash 4 would be more likely to increase andlikelihood ratings for events initially rated 5ndash8 would bemore likely to decrease from Time 1 to Time 2rdquo (p 709)This prediction is an example of Pezdek and Eddyrsquos ten-dency to confuse the event mean with the scale midpointIn fact RTM would move confidence to less than 4mdashbe-cause the critical events have overall means of less than4mdashand an analysis purporting to detect RTM effects needsto take as its reference the mean not 4

Finally the alert reader will note that all three experi-ments had both a control and a treatment group Thusthe difference between the upward shifts in the twogroups constitutes an effect We are heartened to see thatPezdek and Eddyrsquos findings contribute to the growinglist of those who have replicated imagination inflation

Figure 3 Direction of change for subjects who initially re-sponded 1ndash4 across three studies


(Heaps amp Nash 1999 Paddock et al1998 Paddocket al 1999)

Finding 2 For high pretest events ratings weremore likely to decrease than do anything else Con-sider now the top and middle panels of Figure 4 BothGarry et al (1996) and Pezdek and Eddy found that whenhigh pretest events were imagined confidence tended todecrease or stay the same How are we to make sense ofthese results Pezdek and Eddy make sense of them bysaying ldquothese findings are exactly what would be pre-dicted by regression toward the meanrdquo (p 709) It is truethat RTM predicts that high scores should move closer tothe mean on retesting (as should low scores) Howeverif we now add Heaps and Nashrsquos (1999) reanalyzed data

into the equationmdashon the bottom of Figure 4mdashthe pic-ture gets cloudier because in their experiment confi-dence rarely decreased

Let us return to the behavior of high pretest events andask another question If we assume that imagination in-flates confidence that the low ldquoprobably didnrsquot happenrdquoevents did happen what should be the effect of imagina-tion on confidence for the high ldquoprobably did happenrdquoend of the scale We might make a number of predic-tions On the one hand Read and Lindsay (2000) haveshown that when people think about genuine events theyremember more about them over time On the basis oftheir research we might expect that if high pretest con-fidence events indicate recall of genuine experiencesimagining these events will cause subjects to becomeeven more confident about them than about high pretestevents that are not imagined On the other hand it mightwell be the case that imagining counterfactual detailsabout genuine experiences ultimately makes subjects lessconfident about what is real and what is imagined Theissue of how imagining ldquoprobably did happenrdquo events af-fects memory for those events is a question worthy ofstudy in its own right and it is unlikely to be answeredby post hoc analysis and speculation As matters standnow we have no idea what might be the effect of imag-ining events that subjects believe were probably genuineexperiences Pezdek and Eddyrsquos assertion that if imagi-nation inflation is a real effect it should affect confidenceabout events no matter what their pretest ratings is un-tenable given how little we know about the effects ofimagination on memory

A concrete example should make this point clearerAn antidepressant might work systematically to elevatethe mood of depressed people but might have a range ofeffects on the already well adjusted If the drug works bysomehow sucking up a bit of a biochemical that depressedpeople have in excess most well-adjusted people willnot have that excess biochemical and the drug will notaffect them If however the drug works by bringingeveryone who takes it up to some level of general goodmood perhaps some well-adjusted people might becomeeven happier whereas the chronically happy might al-ready be at ceiling The point is we have no idea whichif any of these mechanisms drives imagination inflation

Of course looking only at the decreasing scores in Fig-ure 4 gives an incomplete picture We should return toFigure 3 and see what happened to low pretest eventsWhen they changed were they more likely to increasethan do anything else The answer of course is ldquonordquo Garryet al (1996) Pezdek and Eddy and Heaps and Nashrsquos(1999) data all show that the most common change forlow pretest scores is no change Here Pezdek and Eddywant it both ways They assert that imagination inflationis really RTM dressed up as an effect and as evidencethey show that the most common change for high pretestevents is a decrease Yet they do not expect the mostcommon change for low pretest events to be an increaseMoreover although Figure 3 shows a consistent pattern



across low pretest items for Garry et al Heaps and Nashand Pezdek and Eddy Figure 4 shows no consistent pat-tern across high pretest items We might conclude there-fore that imagination operates more consistently on low-confidence events to produce imagination inflation thanit does on high-confidence events This is not to say thatimagination has no effect on some high-confidence eventsit is merely to say that the data across the three studiesare inconclusive

Finding 3 Items tend to move toward the mean atposttest Pezdek and Eddy examined their data accord-ing to the following rationale ldquoif regression toward themean is operative in the imagination inflation paradigmit should be evidenced with the nontarget events as wellas with the target eventsrdquo (p 711) On this point we haveno dispute with Pezdek and Eddy On the contrary webelieve that of course the data show evidence of RTMIndeed interpretation of any pretestndashposttest designmust take into account RTM The best way to take RTMinto account is to use a control group or control itemsthat ldquocan be used to estimate the effect of regression tothe meanrdquo so that ldquoany difference observed between thetwo groups can be attributed to the treatmentrdquo (Kotz ampJohnson 1982 p 707) This interpretation also accountsfor the results Pezdek and Eddy found when they lookedat change in confidence ratings for events rated 1ndash23ndash 4 5ndash6 and 7ndash8 at pretest They found that extremescores showed more change than did scores already closerto the mean Again we are unsurprised by this pattern ofdata and we agree with Pezdek and Eddyrsquos claim thatthese results support ldquothe significant role of regressiontoward the mean in accounting for the results of theimagination inflation paradigmrdquo (p 712) because re-gression to the mean plays a role in the overall resultsHowever their results do not account for the imaginationinflation effect

Finding 4 Only the low pretest group showed in-flation We were also heartened to see that Pezdek andEddy replicated another of Garry et alrsquos (1996) findingsThat is when low events were treated as cases there wasmore inflation (the percentage of positive change) forimagined events than for not imagined events It is inter-esting to note that even though Pezdek and Eddy usedfewer critical itemsmdashand thus had fewer degrees of free-dom in their t testmdashthey still showed imagination infla-tion Pezdek and Eddy then did the same analysis forhigh pretest events and found no evidence of imaginationinflation We agree with part of their conclusion thatldquothese results suggest a role of imagination inflation overand above the effect of regression toward the meanrdquo(p 712) However they temper their conclusion by say-ing that it does not square with their magnitude ofchange analysis part of which appears in their Table 2

Pezdek and Eddyrsquos Table 2 presents the magnitude ofchange in confidence ratings for low events They con-clude that because the increases are small and becauseldquolargerdquo changes were more likely to occur for not imag-ined events than for imagined events the changes are

ldquonot a product of the subjectsrsquo imagining the target eventsrdquo(p 713) We disagree with this conclusion Yes thechanges are small but they are consistent A paired t testtreating these four events as cases showed more inflationfor imagined events (M 5 120 SD 5 060 SEM 5030) than for not imagined events (M 5 076 SD 5055 SEM 5 027) This difference was significant witha one-tailed test [t(3) 5 256 p 504] In fact in J Cohenrsquos(1988) terms the effect is not small it is an impressived 5 126

4A The Prediction There Should Bean Age and Imagination Interaction

On the basis of G Cohen and Faulknerrsquos (1989) find-ing that older adults perform more poorly than youngeradults at distinguishing between externally and inter-nally generated information (watched vs imaginedevents) Pezdek and Eddy claim that if imagination in-flation is a real effect there should be an age 3 imagi-nation interaction when one compares the effects ofimagination on college aged and on older adults How-ever if RTM is behind imagination inflation there willbe no interaction because RTM will affect both agegroups equally

4B The Turtle Pezdek and Eddy Failedto Consider Other Mechanisms

Although Pezdek and Eddy describe their predictionin a couple of sentences we think it is worth more thana brief overview There is nothing wrong with this predic-tion per se but in focusing on the source-monitoring ex-planation (Johnson Hashtroudi amp Lindsay 1993) Pezdekand Eddy have failed to take into account other equallylikely explanations such as a familiarity-based explana-tion Imagination inflation can also occur if the famil-iarity of an event is misattributed to the actual occurrenceof the event rather than the imagining of the event

Although the source-monitoring account suggests thatthe poor perceptual qualities of an imagined event areeasily confusable with a long-ago faded memory of areal event and that source judgments become less accu-rate with age (Schacter Koutstaal Johnson Gross ampAngell 1997) a familiarity account of imagination in-flation makes no such prediction Familiarity does notchange with age With the use of Jacobyrsquos (1991) processdissociation procedure it has been shown in many stud-ies that the effects of familiarity (an automatic process)are not influenced by age whereas the effects of the searchcomponent (a controlled process) are (Jennings amp Jacoby1993 Schmitter-Edgecombe 1999 Titov amp Knight1997) Therefore if familiarity is the main contributorto imagination inflation we would expect to see no age 3imagination interaction because both younger and olderadults are able to use familiarity judgments equally well

Pezdek and Eddyrsquos failure to consider other mecha-nisms for imagination inflation is not the only possiblereason why they found no age 3 imagination interactionAlong with her colleagues Pezdek herself has shown that


event plausibility is a crucial factor in changing whatpeople believe about their past (Pezdek Finger amp Hodge1997 Pezdek amp Hodge 1999) Furthermore Anderson(1983) and Sherman Cialdini Schwartzman and Rey-noldsrsquos (1985) research on imagination and future likeli-hood judgements suggest that imagination is more likelyto affect judgments about easy-to-imagine events thanabout hard-to-imagine ones therefore it might be rea-sonable to assume that implausible events are also diffi-cult to imagine Thus we should expect imagination in-flation to be influenced by the plausibility of an event

Pezdek and Eddy failed to take plausibility into ac-count when they included ldquoFound a $10 bill in a parkinglotrdquo as one of only two critical events imagined by half ofthe subjects Their college subjects (mean age 209 years)probably believed that finding a $10 bill in the parkinglot in 1988 (when these subjects were approximately 10years old) was quite plausible However consider the sit-uation for the older adults (mean age 757 years) Find-ing a $10 bill in 1933 (when the older subjects were ap-proximately 10) would be the equivalent of finding nearly$120 today according to the Bureau of Labor StatisticsrsquoConsumer Price Index (2000) Finding such a significantsum of money is not very plausible at all particularlyduring the Great Depression A second plausibility prob-lem caused by using the ldquo$10 billrdquo event is the frequencywith which subjects might have been in a parking lotWhen the college students were 10 years old there wereapproximately 137 million cars in the United States (Bu-reau of the Census 1988) However when the older adultswere 10 there were only about 1 million cars andmdashpre-sumablymdashnot as great a need for somewhere to park them(Bureau of the Census 1934) Thus we may concludethat Pezdek and Eddyrsquos older adults might have found theidea of being in a parking lot as a child to be a rather un-usual event The conjunction of these two situationsmdashfinding the equivalent of $120 and finding it in a park-ing lotmdashwould be even less likely than either one alone

In short we believe that the 50 of older subjects whoimagined finding a $10 bill in a parking lot may wellhave found the experience implausible andor difficultto imagine These factors might have offset any tendencyof these older subjects for increased source confusion(which by itself might have caused more imagination in-flation) resulting in no effect for age on imagination in-flation Therefore Pezdek and Eddyrsquos claim that no age3 imagination interaction shows that imagination infla-tion is nothing more than RTM is premature

OTHER TURTLES

Do Pezdek and Eddy offer imagination inflation re-searchers a way out of the methodological quagmireNo In fact their discussion forecasts doom for any re-searcher who uses a pretestndashposttest design Note theirtreatment of Paddock et alrsquos (1998) imagination infla-tion research Paddock et al extended the Garry et al(1996) analysis by not examining the low pretest sub-

jects separately from the high pretest subjects In otherwords they simply compared posttest means overall(having removed subjects whose pretest responses were8) for imagined and not imagined events Paddock et alfound that imagined events inflated more than not imag-ined events However Pezdek and Eddy suggest that be-cause imagination inflation studies tend to rely on pretestdistributions that are loaded heavily with low confidenceratings ldquoanalyses that combine results for all events arenot sensitive to the direction of effect for events with ini-tial high likelihood ratingsrdquo (p 717)

Although at face value that statement is true Pezdekand Eddy seem to use it to prop up two claims that arenot true The first is that because the overwhelming ma-jority of pretest scores are low RTM will cause the over-whelming majority of scores to go up The second is thatbecause only a tiny portion of the pretest scores are highRTM will cause only a tiny portion of the scores to godown The net effect is an increase in mean confidenceor a spurious imagination inflation result

Their warning reveals a misunderstanding of RTM forseveral reasons First if there is no effect for imagina-tion their scenario should apply to the control conditionas well After all RTM should not be promoted by imag-ination indeed Pezdek and Eddyrsquos fundamental thesis isthat imagination has no effect Second when we con-sider only the effect of RTM on pretest and posttest scoreswhat should happen to the mean Nothing The overallmean stays the same Third when we consider only theeffect of RTM on pretest and posttest scores what shouldhappen to the shape of the distribution Nothing Does itmatter that the pretest conf idence ratings are not nor-mally distributed No If it were the case that repeatedadministrations of a test caused a mean increase in anydistribution that was heavily weighted toward lower scoreswe would find that the massive increase in the incidenceof depression (Weissman Livingston-Bruce Leaf Flo-rio amp Holzer 1991) could be attributed to repeated ad-ministrations of the Beck Depression Inventory

Moreover their warning assumes yet again that RTMoperates to cause regression to the midpoint of the LEIscale not to the mean of items It is important for us tonote here that for theoretical reasons imagination infla-tion researchers sometimes analyze data by splitting it atthe LEI scale midpoint A midpoint split we think cor-responds to confidence that the event probably did nothappen (1ndash 4) or probably did happen (5ndash8) HoweverPezdek and Eddy continue to disregard the theoreticalbasis for the split and throughout their paper they assumethat splitting data on the midpoint is the equivalent tosplitting data on the mean Of course these are not thesame thing

In short an analysis that combines data points fromsubjects without considering separately high and lowscores can be a reasonable approach if researchers use acontrol versus experimental design to control for RTMand other factors such as time effects Even if imagina-tion operates mostly on low pretest scores combining


data affords the researcher a wider range of statisticaltechniques with the downside probably limited to a lossof statistical power which can usually be compensatedfor by boosting the sample size Besides Paddock et al(1998) others have taken this approach (Garry Frameamp Loftus 1999 Hayes 1999) As an alternative to sim-ply increasing sample size researchers might considerusing Heaps and Nashrsquos (1999) findings showing whichfactors predispose people to imagination inflation toscreen potential subjectsrsquo involvement in an imaginationinflation experiment Such an approach would boost theeffect size permitting combined data analysis with asmaller n

FINAL COMMENTS

We have illustrated several shortcomings in Pezdekand Eddyrsquos paper These shortcomings range from therelatively minor instances of misrepresenting what imag-ination inflation is and what others have said about it tomajor problems with their predictions Perhaps the mostimportant problem in their arguments is the lack of ap-preciation that comparisons between control and exper-imental conditions are used to adjust for a variety of pos-sible factors including RTM Still we believe that Pezdekand Eddy have made a contribution to the research onthe effects of imagination and memory in much thesame way as McCloskey and Zaragoza (1985) did formisinformation research (Loftus Miller amp Burns 1978)McCloskey and Zaragoza attempted to demonstrate thatthere was no evidence that misleading postevent infor-mation had an effect on memory Although they did notclaim that previous evidence of postevent memory dis-tortion was simply RTM disguised as an effect they didclaim something similar that the standard forced-choicememory test Loftus and colleagues used biased subjectsto respond in such a way as to give the appearance of aneffect Later Belli (1989) showed that McCloskey andZaragozarsquos modified test was actually insensitive to mem-ory impairment In many ways McCloskey and Zaragozarsquosmost important contribution was to spark renewed inter-est in the area the end result being that this new researchadvanced our knowledge of how postevent suggestionsaffect memory (Abeles amp Morton 1999 Belli 1989Frost 2000 Lindsay 1990 Loftus amp Hoffman 1989Zaragoza amp Lane 1998 Zaragoza amp McCloskey 1989)What we discovered grew quickly and by leaps andbounds We are grateful to Pezdek and Eddy for provid-ing researchers with a similar opportunity to discover theways in which imagination can influence our sense ofourselves and our past Let us hope our research advancesfaster than a turtle does

REFERENCES

Abeles P amp Morton J (1999) Avoiding misinformation Reinstat-ing target modality Quarterly Journal of Experimental Psychology52A 581-592

Anderson C A (1983) Imagination and expectation The effect of

imagining behavioral scripts on personal influences Journal of Per-sonality amp Social Psychology 45 293-305

Belli R F (1989) Influences of misleading postevent informationMisinformation interference and acceptance Journal of Experimen-tal Psychology General 118 72-85

Bureau of the Census US Department of Commerce (1934) Sta-tistical Abstracts of the United States (57th ed) Washington DCGovernment Printing Office


Bureau of Labor Statistics (2000) Consumer price index homepage[On line] Available httpwwwstatsblsgovcpihomehtml

Campbell D T amp Kenny D A (1999) A primer on regression arti-facts New York Guilford

Cohen G amp Faulkner D (1989) Age differences in source forget-ting Effects on reality monitoring and on eyewitness testimony Psy-chology amp Aging 4 10-17

Cohen J (1988) Statistical power analysis for the behavioral sciences(2nd ed) Hillsdale NJ Erlbaum

Frost P (2000) The quality of false memory over time Is memory formisinformation ldquorememberedrdquo or ldquoknownrdquo Psychonomic Bulletin ampReview 7 531-536

Garry M Frame S amp Loftus E F (1999) Lie down and let me tellyou about your childhood In S D Sala (Ed) Mind myths Explor-ing popular assumptions about the mind and brain (pp 113-124)Chichester UK Wiley

Garry M amp Loftus E F (2000 March) Imagination inflation is nota statistical artifact Paper presented at the meeting of the AmericanPsychology-Law Society New Orleans

Garry M Manning C G Loftus E F amp Sherman S J (1996)Imagination inflation Imagining a childhood event inflates confidencethat it occurred Psychonomic Bulletin amp Review 3 208-214

Garry M amp Polaschek D L L (2000) Imagination and memoryCurrent Directions in Psychological Science 9 6-10

Hayes J E (1999) The effect of imagined time-distance on imagina-tion inflation for both children and adults Unpublished masterrsquos the-sis Victoria University of Wellington New Zealand

Heaps C amp Nash M (1999) Individual differences in imagination in-flation Psychonomic Bulletin amp Review 6 313-318

Heaps C amp Nash M (in press) Comparing recollective experiencein true and false autobiographical memories Journal of Experimen-tal Psychology Learning Memory amp Cognition

Hyman I E Jr amp Kleinknecht E E (1999) False childhood mem-ories In L M Williams amp V L Banyard (Eds) Trauma and mem-ory (pp 178-188) Thousand Oaks CA Sage

Hyman I E Jr amp Loftus E F (1998) Errors in autobiographica lmemory Clinical Psychology General 117 371-376

Hyman I E Jr amp Pentland J (1996) The role of mental imageryin the creation of false childhood memories Journal of Memory ampLanguage 35 101-117

Jacoby L L (1991) A process dissociation framework Separating au-tomatic from intentional uses of memory Journal of Memory amp Lan-guage 30 513-541

Jennings J M amp Jacoby L L (1993) Automatic versus intentionaluses of memory Aging attention and control Psychology amp Aging8 283-293

Johnson M K Hashtroudi S amp Lindsay D S (1993) Sourcemonitoring Psychological Bulletin 114 3-28

Kotz S amp Johnson N L (1982) Encyclopedia of statistical sci-ences New York Wiley

Lindsay D S (1990) Misleading suggestions can impair eyewitnessesrsquoability to remember event details Journal of Experimental Psychol-ogy Learning Memory amp Cognition 16 1077-1083

Loftus E F (1993) The reality of repressed memories American Psy-chologist 48 518-537

Loftus E F amp Hoffman H G (1989) Misinformation and memoryThe creation of new memories Journal of Experimental PsychologyGeneral 118 100-104

Loftus E F Miller D G amp Burns H J (1978) Semantic inte-


gration of verbal information into a visual memory Journal of Ex-perimental Psychology Human Learning amp Memory 4 19-31

Loftus E F amp Pickrell J E (1995) The formation of false mem-ories Psychiatric Annals 25 720-725

Mazzoni G (2001 June) Time course in the development of false be-liefs and memories after imagination Paper presented at the meetingof the Society for Applied Research in Memory and CognitionKingston ON

McCloskey M amp Zaragoza M (1985) Misleading postevent in-formation and memory for events Arguments and evidence againstmemory impairment hypotheses Journal of Experimental Psychol-ogy General 114 1-16

Paddock J R Joseph A L Chan F M Terranova S Man-ning C amp Loftus E F (1998) When guided visualization proce-dures may backfire Imagination inflation and predicting individualdifferences in suggestibility Applied Cognitive Psychology 12S63-S75

Paddock J R Noel M Terranova S Eber H W ManningC G amp Loftus E F (1999) Imagination inflation and the perils ofguided visualization Journal of Psychology 133 581-595

Pezdek K amp Eddy R M (2001) Imagination inflation A statisticalartifact of regression toward the mean Memory amp Cognition 29707-718

Pezdek K Finger K amp Hodge D (1997) Planting false childhoodmemories The role of event plausibility Psychological Science 8437-441

Pezdek K amp Hodge D (1999) Planting false childhood memoriesin children The role of event plausibility Child Development 70887-895

Read J D amp Lindsay D S (2000) ldquoAmnesiardquo for summer camps andhigh school graduation Memory work increases reports of prior pe-riods of remembering less Journal of Traumatic Stress 13 129-147

Rogosa D Brandt D amp Zimowski M (1982) A growth curve ap-proach to the measurement of change Psychological Bulletin 92726-748

Rogosa D R amp Willett J B (1985) Understanding correlates ofchange by modeling individual differences in growth Psychome-trika 50 203-228

Sagan C (1979) Brocarsquos brain Reflections on the romance of scienceNew York Random House

Schacter D L Koutstaal W Johnson M K Gross M S ampAngell K A (1997) False recollection induced by photographs A

comparison of older and younger adults Psychology amp Aging 12203-215

Schmitter-Edgecombe M (1999) Effects of divided attention andtime course on automatic and controlled components of memory inolder adults Psychology amp Aging 14 331-345

Sherman S J Cialdini R B Schwartzman D F amp ReynoldsK D (1985) Imagining can heighten or lower the perceived likeli-hood of contracting a disease The mediating effect of ease of im-agery Personality amp Social Psychology Bulletin 11 118-127

Titov N amp Knight R G (1997) Adult age differences in controlledand automatic memory processing Psychology amp Aging 12 565-573

Waller N G amp Ross C A (1997) The prevalence and biometricstructure of pathological dissociation in the general population taxo-metric and behavior genetic findings Journal of Abnormal Psychol-ogy 106 499-510

Weissman N M Livingston-Bruce M L Leaf P J Florio L Pamp Holzer C E (1991) Affective disorders In L N Robbins ampD A Reiger (Eds) Psychiatric disorders in America The epidemi-ologic catchment area study (pp 53-80) New York Free Press

Zaragoza M S amp Lane S M (1998) Processing resources and eye-witness suggestibility Legal amp Criminological Psychology 3 305-320

Zaragoza M S amp McCloskey M (1989) Misleading postevent in-formation and the memory impairment hypothesis Comment onBelli and reply to Tversky and Tuchin Journal of Experimental Psy-chology General 118 92-99

NOTES

1 Although the scenario that we describe is obviously possible yetextreme we would not expect a parallel line when pretest scores wereabove 4 because we do not know how imagination affects confidenceratings for events judged as probably having happened However theissue of high pretest ratings is not a problem in our scenario becausePezdek and Eddy included only scores below the mean For thosescores we assume that the line might be parallel

2 Garry et al (1996) reported effect size using a different formulathe mean differenceSDnot imagined

(Manuscript received January 26 2001revision accepted for publication May 30 2001)


In summarizing the results of this experiment and severalothers using the imagination inflation paradigm Garry andPolaschek (2000) concluded that ldquoa growing body of lit-erature shows that imagining contrary-to-truth experiencescan change memoryrdquo (p 6) Furthermore ldquowhen peoplethink about or imagine a false event entire false memoriescan be implanted Imagination inflation can occur evenwhen there is no overt social pressure and when hypothet-ical events are imagined only brieflyrdquo (p 708ndash709)

While Garry and Polaschekrsquos (2000) review paper fo-cused on imagination inflation it did so in the context ofa larger review on imagination and memory In thatlarger context the few specific lines Pezdek and Eddyquoted from the Garry and Polaschek abstract make senseThus when the Garry and Polaschek abstract foreshad-ows that ldquowhen people think about or imagine a falseevent entire false memories can be implantedrdquo it refersto the opening section of the Garry and Polaschek paperwhich reviews what we might call the ldquoLost in the Mallrdquogenre of studies Loftus (1993) Loftus and Pickrell (1995)and Hyman and Pentland (1996) Later on in the paperwhen the focus is on imagination inflation Garry andPolaschek refer only to the ldquoeffect of imagination onmemoryrdquo (p 7) and even define imagination inflationrather tightly as the ldquoconfidence-boosting effect of imag-inationrdquo (p 7)

Now that we know what effect we are talking about letus turn to the heart of Pezdek and Eddyrsquos argument Putbluntly they assert that Garry et alrsquos (1996) results mightnot show any confidence-inflating effect of imaginationat all Instead Pezdek and Eddy say these ldquofindings couldbe explained as simply a case of regression toward themeanrdquo (p 709 hereafter RTM) Pezdek and Eddy putforth a number of arguments as evidence of their claimbut we believe each argument is flawed Below we sum-marize Pezdek and Eddyrsquos experiment review their fourpredictions and examine the strength of their argumentsin light of the evidence First however we begin with abasic overview of RTM

REGRESSION TO THE MEANTHE (VERY) SHORT COURSE

Every time researchers measure some construct theyunwittingly drag some error into their measurement Takefor instance the case each of us might face when ad-ministering an Introductory Psychology exam There arestudents for whom error depresses their score and stu-dents for whom error inflates their score So what is theconscientious college professor to do Suppose you gavethe exam again to the whole class Of course you wouldgive a parallel form of the exam but it would not be per-fectly correlated with the first exam What might hap-pen The end result would be that some students wouldfind that their retest scores increased while others wouldf ind that their scores decreased The mean howeverwould stay the same as would the standard deviation and

the shape of the distribution in general With each reteststudents would change and test scores would change aslong as tests were not perfectly correlated In short withsubsequent tests error would be shuffled around differ-ently The shuffling around would tend to work so thatscores farther away from the mean moved closer to themean on a retest Instead of referring to this shufflingaround under the curve by a tedious name such as ldquoshuf-fling around under the curverdquo we call it RTM

Several other interesting features of RTM are outsidethe scope of this article and the reader wanting more thanthis (very) short course could do no better than to readCampbell and Kennyrsquos (1999) book

Before we turn our attention to Pezdek and Eddyrsquos spec-ific predictions there is an important point to noteThroughout their paper Pezdek and Eddy seem to mis-take regression to the mean of the items with regressionto the midpoint of the scale and this confusion under-pins much of their thinking

PEZDEK AND EDDYrsquoS PREDICTIONS

Pezdek and Eddy hypothesize that if imagination in-flation is a real effect we should expect to see the follow-ing pattern of results

1 The residuals between actual posttest scores andRTM-adjusted (predicted) posttest scores should begreater than zero

2 There should be a test occasion 3 condition inter-action

3 All mean likelihood ratings should increase from thepre- to posttest not just those initially rated as 1ndash4

4 There should be a time 3 imagination 3 age inter-action

Below we examine the logic behind each predictionthe data on which Pezdek and Eddy rely and the variousconclusions that can be responsibly drawn

1A The Prediction Actual Versus RTM-Adjusted Residuals Should Be Greater Than Zero

Pezdek and Eddy argue that if imagination inflation isa real effect posttest confidence for imagined events willbe above and beyond the increase in confidence producedby RTM alone To analyze their data according to thishypothesis they did the following analysis

First they found the correlation between pretest andposttest scores for not imagined (control) target eventsand used the correlation to estimate the effects of RTMIn other words they constructed a regression line to pre-dict the change in posttest confidence scores that are at-tributable only to RTM Second Pezdek and Eddy usedthis regression line to predict the posttest scores for theimagined events again assuming that only RTM was re-sponsible for changing these scores Finally they com-pared the predicted imagined posttest scores with the ob-tained imagined posttest scores to produce residual scoresWhere the residuals are positive there is undoubtedly an


















A

B












































OTHER TURTLES









FINAL COMMENTS


REFERENCES





















































NOTES





















A

B












































OTHER TURTLES









FINAL COMMENTS


REFERENCES





















































NOTES















































OTHER TURTLES









FINAL COMMENTS


REFERENCES





















































NOTES






FINAL COMMENTS


REFERENCES





















































NOTES



























NOTES




Imagination inflation is a fact, not an artifact: A reply to Pezdek and Eddy

Documents

Transcript of Imagination inflation is a fact, not an artifact: A reply to Pezdek and Eddy