APA

PROOFS

Implicit Change Identification: A Replication of Fernandez-Duque andThornton (2003)

Cedric Laloyaux, Arnaud Destrebecqz, and Axel CleeremansUniversite Libre de Bruxelles

Using a simple change detection task involving vertical and horizontal stimuli, I. M. Thornton and D.Fernandez-Duque (2000) showed that the implicit detection of a change in the orientation of an iteminfluences performance in a subsequent orientation judgment task. However, S. R. Mitroff, D. J. Simons,and S. L. Franconeri (2002) were not able to replicate this finding after correcting for confounds and thusattributed Thornton and Fernandez-Duque’s results to methodological artifacts. Because Mitroff et al.’sfailure to replicate might in turn have stemmed from several methodological differences between theirstudy and those of Thornton and Fernandez-Duque (2000) and Fernandez-Duque and Thornton (2003),the current authors set out to conduct a further replication in which they corrected all known method-ological biases identified so far. The results suggest that implicit change detection indeed occurs:People’s conscious decisions about the orientation of an item appear to be influenced by previousundetected changes in the orientation of other items in the display. Implications of this finding in lightof current theories of visual awareness are discussed.

Keywords: change detection, change blindness, implicit change detection, awareness, attention

Change blindness and other related phenomena (e.g., inatten-tional blindness, the attentional blink) demonstrate that people’sexperience of the visual world is far more limited than one wouldthink: Large changes presented in the center of the visual field failto be detected, incongruous moving objects escape detection, andso forth (Rensink, 2002; Rensink, O’Regan, & Clark, 1997; Si-mons, 2000; Simons & Ambinder, 2005; Simons & Chabris,1999). However, there also is a substantial (but controversial)literature suggesting that people’s behavior can be influenced byaspects of the environment that they show little or no evidence ofhaving processed consciously (e.g., Greenwald, Abrams, Nac-cache, & Dehaene, 2003) or in the near absence of attention (e.g.,Li, VanRullen, Koch, & Perona, 2002). The phenomenon of con-textual cuing, for instance, in which people’s ability to locate avisual target presented among distractors is improved when thearray of distractors is predictive of the target’s location, occurseven in cases where subjects fail to detect the contingency betweenthe visual context set up by the distractors and the target’s location(Chun & Jiang, 1998). Likewise, sequence learning studies dem-onstrate that people can use the temporal context set up by previ-ous elements to speed up their responses in a choice reaction time(RT) task, even in cases where they show little or no ability to

report on the sequential contingencies (Cleeremans & McClelland,1991; Destrebecqz & Cleeremans, 2001). These phenomena sug-gest that behavior can be influenced, essentially through primingmechanisms, by information that subjects do not hold consciously.

Such findings raise the following issue in the context of changeblindness experiments: What is the fate of changes to objects thatpeople have failed to perceive? Are they represented beyond theretina? Can they influence subsequent processing? If this were thecase, it would constitute a clear instance of implicit perception:People claim not to have seen a stimulus, yet this stimulus demon-strably exerts subsequent causal effects and must hence be repre-sented somehow in the cognitive system.

The possibility that stimuli of which one remains unawarenevertheless influence subsequent processing has recently been thefocus of renewed debate in the domain of visual processing,through the paradigm of change detection (Fernandez-Duque &Thornton, 2000, 2003; Mitroff et al., 2002; Thornton & Fernandez-Duque, 2000). In this paradigm, observers are exposed to simplechanges, such as changes in the orientation of horizontal andvertical rectangles (see Fernandez-Duque & Thornton, 2000, 2003;Mitroff et al., 2002; Thornton & Fernandez-Duque, 2000). Suchsimple changes stand in contrast with other situations that typicallyinvolve complex scenes, such as the well-known flicker paradigm,first introduced by Rensink et al. (1997) and subsequently widelyused to explore the mechanisms underlying change detection (i.e.,Hollingworth & Henderson, 2002; O’Regan, Rensink, & Clark,1999; Rensink et al., 1997; Simons, Chabris, Schnur, & Levin,2002). Simple stimuli have also been used by Rensink in a para-digm that combines visual search with a repeated flicker paradigm(i.e., visual search for a change; Rensink, 2000b).

Thus, in this paradigm, subjects are exposed to a change detec-tion task involving vertical and horizontal bars. Thornton andFernandez-Duque (2000) showed that exposure to a change in theorientation of an item influenced performance in a subsequent

Cedric Laloyaux, Arnaud Destrebecqz, and Axel Cleeremans, CognitiveScience Research Unit, Universite Libre de Bruxelles, Brussels, Belgium.

Cedric Laloyaux is a scientific research worker with the National Fundfor Scientific Research (Belgium). Axel Cleeremans is a senior researchassociate with the same institution. This work was supported by an insti-tutional grant from the Universite Libre de Bruxelles.

Correspondence concerning this article should be addressed to CedricLaloyaux, Cognitive Science Research Unit, Universite Libre de BruxellesCP 191, Avenue F.-D. Roosevelt, 50, 1050, Brussels, Belgium. E-mail:claloyau@ulb.ac.be

Journal of Experimental Psychology: Copyright 2006 by the American Psychological AssociationHuman Perception and Performance2006, Vol. !, No. !, 000–000

0096-1523/06/$12.00 DOI: 10.1037/0096-1523.!.!.000

1

tapraid1/zfn-xhp/zfn-xhp/zfn00506/zfn2023d06g mazzam S!5 8/18/06 12:54 Art: 2005-0264-R

APA

PROOFS

orientation judgment task performed on the same material, evenwhen subjects claimed not to have consciously perceived thechange. However, Mitroff et al. (2002) were not able to replicatethis result after having corrected methodological biases and thustook Thornton et al.’s findings to be artifactual. These and otherstudies reviewed below now form the core of a controversialliterature that Simons and Rensink (2005) have recently charac-terized as addressing one of the most central issues raised bychange blindness research.

In this context, the main goal of our study was to conduct aconceptual replication of the original studies in such a manner thatall known methodological issues identified by previous authorswere addressed. We begin by describing the original findings andcontinue by surveying subsequent attempts to replicate these re-sults. We focus on studies that have specifically explored implicitchange identification (Fernandez-Duque & Thornton, 2003, Ex-periments 2–4; Mitroff et al., 2002, Experiments 4A and 4B;Thornton & Fernandez-Duque, 2000). Table 1 lists the experi-ments we have considered, their results, and important aspects oftheir methodology. It can serve as a roadmap to the followingsections.

Thornton and Fernandez-Duque (2000)

Thornton and Fernandez-Duque (2000) explored whether unde-tected changes can nevertheless be processed and influence sub-sequent processing. Observers were shown two successive dis-plays containing an array of eight black rectangles. On the firstdisplay, half of these items were oriented horizontally, and the

remaining half were oriented vertically (see Figure 1 for an exam-ple of the stimuli used in the present studies; these stimuli are verysimilar to those used in the previous studies). This display wasshown for 250 ms and was immediately followed by a blankscreen, again shown for 250 ms. Next, a second stimulus array, inwhich one of the rectangles might now appear rotated by 90°relative to its orientation in the first display, was shown for 250ms. Finally, a third display—the probe—appeared for a mere 20ms. This probe display was exactly the same as the second display,but one of the rectangles now appeared in white. Subjects had toperform two tasks: First, they had to make an orientation judgmenton the highlighted (white) rectangle, pressing one key when thisitem was oriented horizontally and another when it was orientedvertically (speeded orientation task). Next, they had to indicatewhether they had perceived the change in orientation that mighthave taken place between the first and second displays. Thissecond task (change detection task) involved a go/no-go response:Subjects pressed the space bar if they had perceived a change anddid nothing if they had not.

This design made it possible to compare performance in fourconditions defined by crossing two factors: congruency and valid-ity (see Figure 2). Validity refers to the location at which both thechange and the probe may be presented. Trials in which a changeoccurred between the first and second displays and in which thechange occurred at the location at which the probe stimulus ap-peared on the third display were labeled valid trials, based on thenotion that if people are sensitive to the change, their orientationresponse to the white rectangle would be facilitated (Posner,

Table 1Summary of the Previous Experiments Investigating Implicit Change Identification

Study Experiment ResultConfoundeliminated Remaining confounds

Thornton and Fernandez-Duque (2000) Experiment 2 Original findings:congruency andvalidity effects

None Spatial link

Number of itemsMitroff et al. (2002) Experiment 4A Replication of the

congruency effectNone Spatial link

No speeding beepNumber of items

Experiment 4B Nonreplication of thecongruency effect

Spatial link No speeding beep, yieldingvery slow RTs

Number of itemsFernandez-Duque and Thornton (2003) Experiment 2 Replication of

congruency effectSpatial link Number of items

Speeding beepExperiment 3 Replication of

congruency effectfor fast respondersonly

Spatial link Speeding beep

Number of itemsExperiment 4 Replication of

congruency effectNumber of items Spatial link

Laloyaux, Destrebecqz, and Cleeremans(current study)

Experiment 1 Replication ofcongruency effect

Spatial link None

Number of items

Note. The Result column is a brief reminder of the results found in the different experiments. The next column describes the confounds eliminated ascompared with the previous experiments, and the final column describes briefly the potential confounds still present in the experiment, which we correctedin the present study. RTs ! reaction times.

2 LALOYAUX, DESTREBECQZ, AND CLEEREMANS

T1

F1

F2

tapraid1/zfn-xhp/zfn-xhp/zfn00506/zfn2023d06g mazzam S!5 8/18/06 12:54 Art: 2005-0264-R

APA

PROOFS

1980). Trials in which the change occurred at a different locationfrom where the probe item appeared were labeled invalid trials. Ofnote, in Thornton and Fernandez-Duque’s (2000) experiment, thehighlighted rectangle always appeared either at the location of thechange or at the diametrically opposite location.

Congruency between the orientation of the probe item and thefinal orientation of the changed rectangle defined the second

factor of interest in this paradigm. Congruent trials are trials inwhich the probe item is oriented in the same direction as that ofthe changed rectangle presented on the second display; incon-gruent trials are those in which this condition does not hold.Thus, if the orientation of the changed rectangle was vertical onthe second display, the trial was congruent when the highlightedrectangle presented on the probe display was vertical but in-

Figure 1. Time course of one trial used by Thornton and Fernandez-Duque (2000).

Figure 2. Example of different types of trials used in the present study. Note that these stimuli are slightlydifferent from those used by Thornton and Fernandez-Duque (2000) because a “spatial link” has been eliminatedas compared with this initial study.

3IMPLICIT CHANGE IDENTIFICATION

tapraid1/zfn-xhp/zfn-xhp/zfn00506/zfn2023d06g mazzam S!5 8/18/06 12:54 Art: 2005-0264-R

APA

PROOFS

congruent when the highlighted rectangle was orientedhorizontally.

Using this paradigm, Thornton and Fernandez-Duque (2000)found a congruency effect: Observers responded more slowly andmade more errors for incongruent than for congruent trials whenthey were aware of the change. The authors also reported a validityeffect on the error rate when the change had been consciouslyreported: The error rate was lower for valid than for invalid trials.Of more interest, they also found a congruency effect on the errorrate when observers had reported being unaware of the change.This finding clearly suggests that the nature of a change can beregistered without reaching awareness and that this information(e.g., “verticality” when a horizontal rectangle is replaced by avertical one) can be activated sufficiently strongly to have an effecton a subsequent response. It must be noted that the congruencyeffect implies that it is indeed the nature of the change that isactivated, and not merely its localization, because the effect holdseven for invalid trials, that is, for trials in which the changeactually occurred at a different location than the one at whichpeople have to respond.

However, this interpretation of Thornton and Fernandez-Duque’s (2000) data, and thus the very existence of implicitchange detection, was challenged by Mitroff et al. (2002). In thenext section, we describe their alternative explanation of Thorntonand Fernandez-Duque’s results, according to which “implicit”change detection can simply be attributed to explicit strategies.

Mitroff et al. (2002)

A simple, alternative explanation of the implicit change detec-tion effect reported by Thornton and Fernandez-Duque (2000) isthat subjects are using conscious strategies to anticipate the loca-tion of the probe based on regularities contained in the stimulusmaterial. Building on this idea, Mittroff et al. (2002) replicatedThornton and Fernandez-Duque’s experiments and also other stud-ies purporting to demonstrate implicit change registration (Wil-liams & Simons, 2000) or localization (Fernandez-Duque &Thornton, 2000; Smilek, Eastwood, & Merikle, 2000), and claimedthat all the behavioral evidence for implicit change detection orregistration could be explained in terms of the deployment ofconscious strategies rather than on unconscious sensitivity tochange.

Thus, Mitroff et al. (2002) pointed out that in the case of anorientation change in Thornton and Fernandez-Duque’s (2000)experiments, the probe (i.e., the item that changed color) waseither the changed item (i.e., the item that changed orientation in50% of the trials) or the item that was located at the diametricallyopposite location (in the remaining 50% of the trials). Mitroff et al.therefore suggested that participants could quickly learn this spa-tial relationship between the probe and the changed item and hencecome to pay more attention to the item opposite to the probe. Incongruent trials, the probe and the changed item have the sameorientation in the third display, whereas they have different orien-tations in incongruent trials. The congruency effect reported byThornton and Fernandez-Duque would thus depend on the rela-tionship between the probe and the changed item in the finaldisplay rather than on the relationship between the probe and thechange that occurred between the first and the second displays.

To explore this hypothesis, Mitroff et al. (2002) first replicatedThornton and Fernandez-Duque’s (2002) experiment and wereable to reproduce the congruency effect both when subjects wereaware and when they were unaware of the change (Mitroff et al.,2002, Experiment 4A). In a second replication, however, Mitroff etal. eliminated the spatial relationship between the item thatchanged orientation and the item that changed color (the probe)(Mitroff et al., 2002, Experiment 4B). With the exception of thevalid trials (in which the probe appeared at the same location as theorientation change), the probe could then be any of the sevenremaining elements. Significantly, under these better controlledconditions, Mitroff et al. were not able to replicate the congruencyeffect reported by Thornton and Fernandez-Duque, either for theRTs or for the error rates and regardless of whether participantsreported being aware of having perceived the change. Also in linewith their interpretation, Mitroff et al. reported that most observerswere conscious of the spatial relationship between the probe andthe changed item in Experiment 4A. Mitroff et al. therefore con-cluded that the effects described by Thornton and Fernandez-Duque were entirely artifactual and hence that there is no basis forthe concept of implicit change detection. Theoretically, they fur-ther argued that the two visual representations—before and afterthe orientation change—could not be compared in the absence ofan explicit comparison process, which is assumed to requireattention.

It must be noted, however, that Mitroff et al. (2002) failed toobserve any congruency effect, even in the “aware” trials in whichparticipants noticed the orientation change. In a reply to Mitroff etal., Fernandez-Duque and Thornton (2003) thus claimed that thisfailure to obtain a congruency effect suggests that the modifiedversion of the paradigm used by Mitroff et al. was not sensitiveenough. We examine this study in the next section.

Fernandez-Duque and Thornton (2003)

According to Fernandez-Duque and Thornton (2003), Mitroff etal. (2002) changed the paradigm so much that it made it simplyimpossible to obtain any congruency effect. In other words,Fernandez-Duque and Thornton consider that observing a congru-ency effect in aware trials constitutes a prerequisite to obtain thesame effect in unaware trials. In the absence of the former, thefailure to replicate the latter effect should therefore not be con-strued as a demonstration that implicit change detection does notexist.

According to Fernandez-Duque and Thornton (2003), there areat least two reasons why the congruency effect was completelyabsent in Mitroff et al.’s (2002) Experiment 4B. First, accuracywas only 60% in that experiment, whereas it was about 80% inprevious experiments demonstrating a significant congruency ef-fect (even for the first replication of Mitroff et al., 2002: 77%).Given that chance level is 50%, a hit rate of 60% might be too lowto reveal any significant effect, simply because the number ofcorrect trials on which to measure the effect is too small.

Second, RTs were very slow in Mitroff et al.’s (2002) experi-ment, and congruency effects are known to depend on globalresponse speed. It has indeed been reported that congruency effectscan disappear when RTs are too slow (Dejong, Liang, & Lauber,1994). In contrast to the initial experiments demonstrating animplicit congruency effect (Fernandez-Duque & Thornton, 2003;

4 LALOYAUX, DESTREBECQZ, AND CLEEREMANS

tapraid1/zfn-xhp/zfn-xhp/zfn00506/zfn2023d06g mazzam S!5 8/18/06 12:54 Art: 2005-0264-R

APA

PROOFS

Thornton & Fernandez-Duque, 2000, Mitroff et al. (2002, Exper-iment 4B) did not use an auditory signal to speed up and constrainRTs in their experiment. The mean RT difference between Mitroffet al.’s nonreplication and Thornton and Fernandez-Duque’s(2000) original findings was about 300 ms. In contrast, RTs weresignificantly faster in Mitroff et al.’s Experiment 4A than in theirExperiment 4B, which probably allowed them to successfullyproduce a reliable congruency effect.

Fernandez-Duque and Thornton (2003) also replicated theirinitial experiment (Thornton & Fernandez-Duque, 2000), this timeeliminating the potential bias uncovered by Mitroff et al. (2002),that is, the spatial relationship between the changed and the high-lighted item. They also made sure that their subjects would re-spond more quickly than in Mitroff et al.’s Experiment 4B. Underthese conditions, they again found a congruency effect on the errorrates, for both aware and unaware trials. Moreover, they observeda congruency effect on the RTs, for both aware and unawaretrials—a result they had failed to find in their original study.However, it should be noted that Fernandez-Duque and Thornton(2003) argued that finding a congruency effect for the accuracyand not for the RTs in the unaware trials might suggest that thereis a qualitative difference between these two measures. Moregenerally, they argued that any kind of dissociation between awareand unaware trials would be a strong argument in favor of theinterpretation that two different processes (one explicit and anotherone implicit) are involved in the task.

To sum up, Fernandez-Duque and Thornton’s (2003) new re-sults were even stronger than their previous findings and also ruledout the alternative explanation proposed by Mitroff et al. (2002).However, it is worth noting that Fernandez-Duque and Thorntonincreased the duration of the probe to 40 ms (instead of 20 ms) inthis new experiment so as to increase accuracy. This introducesanother modification with respect to the original task.

In another experiment aimed at further elucidating differencesbetween the two series of experiments, Fernandez-Duque andThornton (2003) reset the display duration to its initial value of 20ms and eliminated the speeding beep. They still found congruencyeffects on error rates, in both aware and unaware trials. Furtheranalyses showed that the effect was present only for so-called fastresponders (RT " 1,000 ms), which again suggests that the slowRTs in Mitroff et al. (2002) might explain their failure to replicate.

Finally, Fernandez-Duque and Thornton (2003) again replicatedtheir first experiment correcting for another potential confound. Inevery experiment reviewed so far, the first stimulus screen alwaysdisplayed 4 horizontal (H) and 4 vertical (V) rectangles. Thesecond display, in which one of the rectangles had changed itsorientation, was then always composed of five rectangles of onecategory and three of the other category. This necessarily impliesthat one of the rectangles has changed its orientation between thefirst and the second display. The presence of such regularity in thestimulus material may lead participants to apply an explicit (oreven implicit) strategy based on counting so as to identify thenature of the change. As a consequence, Fernandez-Duque andThornton conducted another experiment in which this potentialbias was controlled for by using as a first stimulus screen either a4V/4H display or a 3V(or H)/5H (or V) display. Under theseconditions, Fernandez-Duque and Thornton reported a congruencyeffect on the RTs, in both aware and unaware trials. It must benoted, however, that in this last experiment, Fernandez-Duque and

Thornton reintroduced another potential bias—that is, the system-atic spatial relationship between the changed item and the probe.

To summarize this rather long and intricate story, none of theexperiments conducted by the two labs were successful in correct-ing all known potential confounds possibly influencing the occur-rence of a congruency effect in unaware trials. As a result, thepossibility for changes to be processed unconsciously remains anopen question. As Thornton and Fernandez-Duque (2002) sug-gested, converging evidence is needed to establish implicit changedetection. We therefore attempted to replicate their findings in abetter controlled experimental setting.

In this controversial context, our main goal was thus simply toreplicate the original findings in such a way that all previouslyidentified methodological biases were controlled for. Indeed, al-though Fernandez-Duque and Thornton (2003) did replicate theirown previous findings, they also failed to correct every potentialconfound in a single experiment. In particular, they controlledeither for the number of vertical and horizontal items (thus pre-venting an explicit counting strategy) or for the spatial relationshipbetween prime and target (thus preventing strategies based on thelearning of this relationship), but never for both. In addition, theirstudies all involved rather small numbers of participants. In ourreplication, we controlled for both biases and used a larger sampleof 24 participants. We now turn to describing our conceptualreplication.

Method

Participants

A total of 24 students from the Universite Libre de Bruxelles partici-pated in this experiment for course credit. All had normal or corrected-to-normal vision and were naive as to the hypotheses under investigation.

Material

Stimulus presentation and data acquisition were conducted using a G4Macintosh computer running PsyScope (http://psy.ck.sissa.it/) for OS Xand connected to a 17-in. 100-Hz CRT monitor.

Stimuli

Each stimulus display was a circle of rectangles arranged in a clock facedesign such that any particular item was equidistant from a central fixationpoint. Rectangles were colored in black on a gray background. The size ofa rectangle was 10 # 30 pixels on a 17-in. screen at a 1,024 # 768-pixelscreen resolution, which subtended approximately 1.15° # 0.38° of visualangle at a 50-cm viewing distance. The complete ring subtended 4.3° ofvisual angle from the fixation point, a value similar to those used byThornton and Fernandez-Duque (2000) and Fernandez-Duque and Thorn-ton (2003) (4.6°) and by Mitroff et al. (2000) (4.46°).

For each trial, the initial ring of rectangles was composed either of fourhorizontally oriented rectangles and four vertically oriented ones or of fiverectangles of one category and three rectangles of the other category. Thisinitial display was presented for 250 ms and subsequently replaced by ablank screen, again displayed for 250 ms. A second array of rectangles wasthen presented for 250 ms. This second array could either be the same asthe first one (in the no-change, catch trials) or differ from the first array bya single rectangle, the orientation of which was rotated by 90°. This secondarray could thus also be composed either of four vertical rectangles andfour horizontal rectangles or of five rectangles of one category and threerectangles of the other category. Immediately after this second display, a

5IMPLICIT CHANGE IDENTIFICATION

tapraid1/zfn-xhp/zfn-xhp/zfn00506/zfn2023d06g mazzam S!5 8/18/06 12:54 Art: 2005-0264-R

APA

PROOFS

third was presented, this time only for 40 ms. This third array was identicalto the second array except that one rectangle was now colored in whiteinstead of black. A central fixation cross was presented for 1,500 ms beforethe onset of the first display.

Experimental Design

One central aspect of our experimental design concerned the relationshipbetween the changed item and the features of the highlighted item. Each ofthe eight possible locations was probed in 64 trials for a total of 512 trials.

Among the trials, 25% were “catch” trials, in which no change occurredbetween the first and the second array. Among those catch trials, 50%included four horizontal and four vertical rectangles, 25% included threevertical and five horizontal rectangles, and the remaining 25% includedfive vertical and three horizontal rectangles (see Table 2 for a globaloverview).

Among the trials, 25% were “valid change” trials, in which anorientation change occurred between the first and the second array (e.g.,a vertical rectangle rotates 90°) and in which the highlighted rectanglein the third screen was the same rectangle that had changed orientationbetween the first and second arrays. Half of these valid change trialsbegan with four horizontal and four vertical (4H/4V) rectangles in thefirst display and finished in either a 3V/5H or a 5V/3H configuration onthe third display (each configuration occurred in 50% of the cases). Forthe other half of the valid change trials, 50% used an initial 3V/5Hconfiguration and a final 4V/4H distribution, and the other 50% usedthe reverse pattern.

The remaining 50% of the trials were “invalid” trials, in which thelocation of the change and the location of the probe differed. Half of thoseinvalid trials were congruent: The final orientation of the changed item andthe probe’s orientation were identical. The other half of the invalid trialswere incongruent: The final orientation of the changed item and the probe’sorientation differed. Finally, the different possible configurations of verti-cal and horizontal rectangles (3V/5H, 5V/3H, 4V/4H) were also counter-balanced for the initial and final displays of the invalid congruent andinvalid incongruent trials. Another important feature of the procedure isthat the total number of vertical and horizontal probes was equal over theentire experiment.

Procedure

Participants performed a total of 512 trials. In addition, they were firsttrained for 20 trials with the probe shown for 200 ms and then with 40 trialswith the probe displayed for 40 ms in order to learn the task progressively.Observers were asked to stare at a central fixation cross until a probeappeared. They were instructed that in some cases, a change of orientationwould occur and that they had to try to detect it. They were also told thatone rectangle—the probe—would change color (from black to white) in thefinal display and that they had to press the S or L key with the index fingersof their left or right hands, respectively, to indicate the orientation of the

probe. For half of the participants, the S key coded for horizontal orienta-tion and the L key coded for vertical orientation. The reverse mapping wasused for the other participants. They were instructed to respond as quicklyand as accurately as possible and were told that a speeding beep wouldoccur after 1,200 ms if they had failed to respond.

Participants were then asked whether they had noticed the orientationchange of one rectangle between the first and second displays. They had topress the space bar only if they had seen the change and not to press anykey if they had not seen the change. It was, then, a go/no-go response. Aftereach trial, the next one began 2 s after the orientation response. We useda very liberal criterion for awareness: Participants were instructed to saythat they had perceived the change even if they were not sure at all. Wechose this procedure to ensure that change identification for unaware trialscould safely be attributed to implicit processes rather than to consciousknowledge held with low confidence. Hence, even if participants had justa tiny feeling that a change might have happened, they had to say that theyhad seen the change. We did not give feedback concerning the accuracy ofthe orientation response, as in Fernandez-Duque and Thornton (2003) andMitroff et al. (2002).

Results

We removed 3 participants from the analysis either becausetheir false alarm rate was higher than their hit rate or because bothrates were strictly equivalent, suggesting poor understanding of thetask or uncooperative behavior. Analyses were thus conducted onthe data from the remaining 21 participants.

Before analyzing potential congruency effects, we had to ascer-tain that participants could perceive the stimulus, at least in somecases, and that their accuracy differed from chance level. To do so,we inspected hit and false alarm rates. The average hit rate was56% and the false alarm rate was 27%, yielding an average d$ of0.86 (average of all the individual d$ values). A t test performed ond$ compared with random performance (d$ ! 0) showed thatparticipants were able to discriminate between change and no-change trials, t(20) ! 8.70, p " .001.

Before going further in the analyses, it is important to determinewhether overall performance level is comparable to that in previ-ous studies. Accuracy was about 89%, and average RT was 750ms. As compared with previous studies, both RTs and accuracyrate seem to be roughly in the same range as in Fernandez-Duqueand Thornton’s (2003) second experiment. Note, however, that asin our study, Fernandez-Duque and Thornton used a 40-ms probein their second experiment, whereas Mitroff et al. (2002) used a20-ms probe.

Table 2Number of Repetitions of a Same Configuration (H and V distribution) for One Participant

% Change category

Configuration of the firstdisplay

Configuration of the seconddisplay

4V/4H 3V/5H 5V/3H 4V/4H 5V/3H 3V/5H

25 Catch trials 64 32 32 64 32 3225 Valid change 64 32 32 64 32 3225 Invalid congruent change 64 32 32 64 32 3225 Invalid congruent change 64 32 32 64 32 32

Note. H ! horizontal rectangle; V ! vertical rectangle.

6 LALOYAUX, DESTREBECQZ, AND CLEEREMANS

T2

tapraid1/zfn-xhp/zfn-xhp/zfn00506/zfn2023d06g mazzam S!5 8/18/06 12:54 Art: 2005-0264-R

APA

PROOFS

Orientation Task Accuracy

Congruency effect. Data from invalid trials were submitted toan analysis of variance (ANOVA) with two within-subject vari-ables: awareness of the change (two levels: seen vs. not seen) andcongruency (two levels: congruent vs. incongruent trials). Wefound a significant effect of awareness, F(1, 20) ! 23.25, p ".001, with accuracy being significantly higher for the unawaretrials than for the aware trials, and a significant congruency effect,F(1, 20) ! 14.32, p " .001, with accuracy being higher forcongruent trials than for incongruent trials. There was no signifi-cant Awareness # Congruency interaction, F(1, 20) ! 1.80, p !.194. We also analyzed aware and unaware trials separately andfound a congruency effect both for unaware, F(1, 20) ! 6.80, p ".01, and for aware trials, F(1, 20) ! 8.50, p " .008 (see Figure 3).

Validity effect. Data from congruent trials were submitted to awithin-subject ANOVA with two variables: awareness of thechange (seen vs. not seen) and validity (valid vs. invalid trials).The effect of awareness was marginally significant, F(1, 20) !4.24, p ! .053, which again suggests that the accuracy was higherfor unaware than for aware trials. However, there was no maineffect of validity, F(1, 20) ! 0.05, p ! .826. We also found asignificant Awareness # Validity interaction, F(1, 20) ! 11.81,p " .002. Analyzing aware and unaware trials separately, wefound a significant “reversed” validity effect for the unaware trials,F(1, 20) ! 5.36, p " .03, and a validity effect for the aware trials,F(1, 23) ! 5.54, p " .029: Mean accuracy was higher for invalidcongruent trials than for valid congruent trials when the changewas not perceived consciously, whereas the opposite pattern ofresults was observed for the aware trials (see Figure 3).

RT

Congruency effect. Data from invalid trials were submitted toan ANOVA with two within-subject variables: awareness of thechange (seen vs. not seen) and congruency (congruent vs. incon-gruent trials). We found a significant effect of awareness, F(1,20) ! 29.93, p " .001, with RTs being much longer for awaretrials than for unaware trials, and a significant congruency effect,

F(1, 20) ! 8.40, p " .008, with RTs being longer for incongruenttrials than for congruent trials. There was no significant interactionbetween those two variables, F(1, 20) ! 0.01, p ! .945. However,we found a congruency effect for the unaware trials, F(1, 20) !14.13, p " .001, but not for the aware trials, F(1, 20) ! 2.35, p !.142 (see Figure 4).

Validity effect. Data from congruent trials were submitted toan ANOVA with two within-subject variables: awareness of thechange (seen vs. not seen) and validity (valid vs. invalid trials). Wefound a significant effect of awareness, F(1, 20) ! 14.49, p ".001, but no validity effect, F(1, 20) ! 2.90, p ! .104. Moreover,we found a reliable interaction between these variables, F(1, 20) !10.89, p " .003. There was a highly significant validity effect forthe aware trials, F(1, 20) ! 9.11, p " .006, but not for unawaretrials, F(1, 20) ! 1.85, p ! .189 (see Figure 4). To summarize, wefound a significant main congruency effect for accuracy. Post hocplanned comparisons showed that this congruency effect wassignificant both for aware and unaware trials. Concerning RTs, wefound a significant main congruency effect and no interaction withawareness. Moreover, post hoc planned comparisons showed thatthe effect is reliable for the unaware trials. The validity effect wasfound both in the accuracy and in the RT data, but only for awaretrials. It therefore seems that a robust congruency effect was foundfor both aware and unaware trials on both accuracy and RTs,whereas a validity effect was found for aware trials only.

Far–Close Analysis

An important issue is to determine whether the congruencyeffect we obtained is modulated by the location at which thechange occurred. To find out, we followed Fernandez-Duque andThornton’s (2003) method and split our data into “far” and “close”trials. Close trials are invalid trials for which the change hadoccurred within 45° of the location at which the probe appeared.Far trials comprise all the remaining invalid trials, that is, trials forwhich the change had been located at 90°, 135°, or 180° from thelocation at which the probe appeared.

Accuracy for the orientation task. Data from invalid trialswere submitted to an ANOVA with three within-subject variables:location of the probe relative to the location of the change (closevs. far change, described in previous paragraph), awareness of thechange (seen vs. not seen), and congruency (congruent vs. incon-gruent trials). We failed to find a significant effect of location, F(1,20) ! 1.67, p ! .210, but the analysis revealed significant effectsof awareness, F(1, 20) ! 31.03, p " .001, and of congruency, F(1,20) ! 15.00, p " .001. None of the two-way or three-wayinteractions reached significance (see Figure 5).

To ascertain whether congruency effects were significant forclose and far trials, we conducted planned comparisons. Therationale of this analysis is that if the congruency effect is trulybased on the identification of the change and not merely onidentification of its location, we should observe a significantcongruency effect for both far and close trials. The first plannedanalysis compared accuracy between congruent and incongruenttrials for the close and aware trials. This analysis revealed only atrend for a congruency effect, F(1, 20) ! 3.60, p ! .072. Thesecond planned analysis compared accuracy between congruentand incongruent trials for close but unaware trials. This analysisdid not reach significance, F(1, 20) ! 0.66, p " .425. A third

Figure 3. Accuracy for the orientation judgment task as a function of theconditions for aware and unaware trials. Error bars represent standard errorof the mean. VC ! valid change trials; IC ! invalid congruent trials; II !invalid incongruent trials; NC ! no-change trials.

7IMPLICIT CHANGE IDENTIFICATION

F3

F4

F5

tapraid1/zfn-xhp/zfn-xhp/zfn00506/zfn2023d06g mazzam S!5 8/18/06 12:54 Art: 2005-0264-R

APA

PROOFS

planned comparison, performed on far and aware trials, revealed asignificant congruency effect, F(1, 20) ! 7.18, p " .01. Finally, afourth planned comparison, performed on far and unaware trials,also revealed a significant congruency effect, F(1, 20) ! 7.63, p ".01. Hence, we observed congruency effects only for far trials,regardless of whether participants had become aware of thechange. Presumably, our failure to find similar effects for closetrials stems from a lack of power.

RT. Data from invalid trials were submitted to a three-wayANOVA with three within-subject variables: location of the proberelative to the location of the change (close vs. far change),awareness of the change (seen vs. not seen), and congruency(congruent vs. incongruent trials). We failed to find a significanteffect of location, F(1, 20) ! 1.70, p ! .207, but the main effectof awareness was significant, F(1, 20) ! 38.41, p " .001. We alsofound a significant congruency effect, F(1, 20) ! 12.94, p " .001.None of the two-way or three-way interactions reached signifi-

cance (see Figure 6). As for the accuracy measure, we performeda series of planned comparisons to determine whether congruencyeffects were present for close and far positions. The first plannedcomparison revealed a significant congruency effect for the closeand aware trials, F(1, 20) ! 6.73, p " .01. The second plannedcomparison also revealed a significant congruency effect for closeand unaware trials, F(1, 20) ! 9.70, p " .005. A third plannedcomparison, performed on far and aware trials, was not significant,F(1, 20) ! 1.35, p ! .258. Finally, a fourth planned comparisonrevealed a significant congruency effect for far and unaware trials,F(1, 20) ! 6.38, p " .02. As for our accuracy analyses, thissuggests that there is a congruency effect for the unaware and fartrials. In addition, we also found a congruency effect for theunaware and close trials. This suggests that the congruency effectdoes not interact with the location of the change.

Number of Stimuli: 4/4 Versus 3/5

As discussed in the introduction, in previous studies (Mitroff etal., 2002; Thornton & Fernandez-Duque, 2000; Fernandez-Duque& Thornton, 2003, Experiments 1–3), the last frame of the trialsalways contained an unequal number of horizontal and verticalrectangles (except in Fernandez-Duque & Thornton, 2003, Exper-iment 4). Participants’ performance may therefore be biased whenthey have to decide the orientation of the probe in the third display.For instance, they could respond faster when the probe is horizon-tal not because they (consciously or unconsciously) noticed theorientation change between the first and second displays butmerely because the number of horizontal rectangles exceeds thenumber of vertical rectangles. To control for this potential con-found, we introduced the initial configuration of the rectangles asan independent variable in a new series of analyses. We expect acongruency effect regardless of the configuration of horizontal andvertical rectangles used.

Accuracy for the orientation task. Data from invalid trialswere submitted to a three-way ANOVA with three within-subjectvariables: the initial configuration of the display (4/H vs. 3/5),

Figure 4. Reaction times (RTs) for the orientation judgment task as afunction of the conditions for aware and unaware trials. Error bars repre-sent standard error of the mean. VC ! valid change trials; IC ! invalidcongruent trials; II ! invalid incongruent trials; NC ! no-change trials.

Figure 5. Accuracy for the orientation judgment task as a function of theconditions for aware and unaware trials and for close and far changes ascompared with the location of the probe. Error bars represent standard errorof the mean. CC ! close and congruent trials; CI ! close and incongruenttrials; FC ! far and congruent trials; FI ! far and incongruent trials.

Figure 6. Reaction times (RTs) for the orientation judgment task as afunction of the conditions for aware and unaware trials and for close andfar changes as compared with the location of the probe. Error bars representstandard error of the mean. CC ! close and congruent trials; CI ! closeand incongruent trials; FC ! far and congruent trials; FI ! far andincongruent trials.

8 LALOYAUX, DESTREBECQZ, AND CLEEREMANS

F6

AQ: 1

tapraid1/zfn-xhp/zfn-xhp/zfn00506/zfn2023d06g mazzam S!5 8/18/06 12:54 Art: 2005-0264-R

APA

PROOFS

awareness of the change (seen vs. not seen), and congruency(congruent vs. incongruent trials). We failed to find a significantmain effect of configuration, F(1, 20) ! 0.02, p ! .880. The maineffect of awareness, however, was significant, F(1, 20) ! 22.11,p " .001, as was the congruency effect, F(1, 20) ! 13.16, p ".001. None of the two-way interactions reached significance, butthere was a trend for the three-way Configuration # Awareness #Congruency interaction, F(1, 20) ! 3.62, p ! .07 (see Figure 7).We performed planned comparisons to determine whether congru-ency effects were present for the 4/4 and the 3/5 initial configu-rations. The first planned comparison revealed a significant con-gruency effect for the 4/4 initial configuration and aware trials,F(1, 20) ! 12.65, p " .001. The second planned comparison,performed on the 4/4 initial configuration for unaware trials,revealed only a trend for a congruency effect, F(1, 20) ! 3.61, p !.071. A third comparison, performed on the 3/5 initial configura-tion for aware trials, was not significant, F(1, 20) ! 2.66, p !.118. Finally, a fourth planned comparison, performed on unawaretrials with a 3/5 initial configuration, indicated a significant con-gruency effect, F(1, 20) ! 6.33, p " .02.

RT. Data from invalid trials were submitted to a three-wayANOVA with three within-subject variables: the initial configura-tion of the display (4V/4H vs. 3/5), awareness of the change (seenvs. not seen), and congruency (congruent vs. incongruent trials).The main effect of the initial configuration was not significant,F(1, 20) ! 1.70, p ! .207, but we found a significant effect ofawareness, F(1, 20) ! 32.27, p " .001, and a significant congru-ency effect, F(1, 20) ! 9.33, p " .006. None of the two-way orthree-way interactions was significant (see Figure 8). We per-formed planned comparisons to determine whether congruencyeffects were present for the 4/4 and the 3/5 starting configurations.The first planned comparison contrasted congruent and incongru-ent trials for the 4/4 initial configuration and aware trials. It failedto reveal a congruency effect, F(1, 20) ! 1.25, p ! .276. Thesecond planned comparison evaluated the congruency effect forthe 4/4 initial configuration among unaware trials, and it also failedto be reliable, F(1, 20) ! 1.25, p ! .276. A third comparison

checked for congruency effects for the 3/5 initial configuration inaware trials. This was significant, F(1, 20) ! 7.08, p " .01.Finally, a fourth planned comparison concerned the 3/5 startingconfiguration and unaware trials, and it was also significant, F(1,20) ! 22.83, p " .001.

Comparison Between Fast and Slow Responders

In a final analysis, we compared performance for fast and slowresponders. Indeed, if the congruency effect is reliably stronger forfast responders than for slow responders, it would strengthen thenotion that fast RTs are necessary to obtain a congruency effect, aspreviously shown by Fernandez-Duque and Thornton (2003).1 Afirst important point in this respect is that in our experiment, allparticipants should be considered fast responders according tocriterion used by Fernandez-Duque and Thornton, as they allresponded within 1,000 ms. We therefore decided to set ourcriterion for separating fast from slow responders as the medianvalue of the mean RTs across every condition (751 ms). Thisresulted in 11 participants considered to be fast responders and 10considered to be slow responders. For accuracy, we obtainedsimilar patterns of congruency effects for the very fast (RT ! 751ms) and slower (RT % 751 ms) responders, F(1, 10) ! 5.95, p ".034, and F(1, 9) ! 8.12, p " .019, respectively. We also observedan effect of awareness in both groups, F(1, 10) ! 10.21, p " .009,and F(1, 9) ! 13.53, p " .005, respectively. The Congruency #Awareness interaction failed to reach significance in either group:fast responders, F(1, 10) ! 1.61, p ! .232; slow responders, F(1,9) ! 0.20, p ! .662. For the RTs, the congruency effect was highlysignificant for fast responders, F(1, 10) ! 10.08, p " .009, but notfor slow responders, F(1, 9) ! 0.81, p ! .39. The effect ofawareness was significant in both groups, F(1, 10) ! 13.93, p ".003, and F(1, 9) ! 14.72, p " .003. The Congruency # Aware-ness interaction was significant neither for the fast responders, F(1,

1 We thank Ron Rensing for suggesting this analysis.

Figure 7. Accuracy for the orientation judgment task as a function of theconditions for aware and unaware trials and for 4 vertical (V)/4 horizontal(H) or 3V (or H)/3H (or V). Error bars represent standard error of the mean.44C ! 4V/4H and congruent trials; 44I ! 4V/4H and incongruent trials;35C ! 3V (or H)/3H (or V) and congruent trials; 35I ! 3V (or H)/3H (orV) and incongruent trials.

Figure 8. Reaction times for the orientation judgment task as a functionof the conditions for aware and unaware trials and for 4 vertical (V)/4horizontal (H) or 3V (or H)/3H (or V). Error bars represent standard errorof the mean. 44C ! 4V/4H and congruent trials; 44I ! 4V/4H andincongruent trials; 35C ! 3V (or H)/3H (or V) and congruent trials; 35I !3V (or H)/3H (or V) and incongruent trials.

9IMPLICIT CHANGE IDENTIFICATION

F7

F8

Fn1

tapraid1/zfn-xhp/zfn-xhp/zfn00506/zfn2023d06g mazzam S!5 8/18/06 12:54 Art: 2005-0264-R

APA

PROOFS

10) ! 0.79, p ! .395, nor for the slow responders, F(1, 9) ! 1.83,p ! .208. These results therefore reinforce the notion that fastresponses are necessary in this paradigm and suggest that Mitroffet al.’s (2002) failure to replicate was most likely due to the overallslower RTs of their participants.

Summary

The goal of this study was to explore whether implicit changedetection would occur when all known potential biases previouslydescribed in the literature were controlled for. Our results repeat-edly indicated a congruency effect, not only for aware but also forunaware trials, and are therefore clearly in favor of the implicitchange detection hypothesis previously defended by Fernandez-Duque and Thornton (2003). More specifically, we replicated acongruency effect for the unaware trials, both in terms of accuracyand in terms of RT. More detailed analyses concerned interactionsbetween the location of the change and the occurrence of a con-gruency effect. We could not replicate Fernandez-Duque andThornton’s three-way interaction here, suggesting that congruencyeffects might be dependent on both “closeness” and awareness ofthe change. However, in contrast to Fernandez-Duque and Thorn-ton, we found a congruency effect for “far” unaware trials on theaccuracy data. This rules out purely spatial accounts of the con-gruency effect and suggests instead that congruency effects trulydepend on the identity of the change rather than simply on itslocalization. We also found this effect in various other conditions,and additionally for the RTs, which suggests that the congruencyeffect we observed for unaware trials is relatively robust.

We also explored the extent to which the initial configuration ofstimuli influences performance—a factor that we had identified asa potential confound in previous studies. Here, we found only atrend for congruency effects in the accuracy data associated with4/4 initial configurations when participants had reported beingunaware of the change, but the effect was significant for awaretrials. In the case of 3/5 initial configurations, however, we founda significant congruency effect for unaware trials but not for awaretrials. A slightly different pattern emerged from the RT data, butaltogether we found congruency effects over a range of conditionsin which such effects were not supposed to emerge according toMitroff et al.’s (2002) account that participants use explicit, con-scious strategies in this situation.

Discussion

Can changes that occur in visual displays and that fail to beregistered by participants nevertheless influence subsequent deci-sions? What is the fate of undetected changes? Such questions aregood starting points to think about the differences between con-scious and unconscious information processing. A simple way toaddress these issues is to ascertain whether an undetected changecan nevertheless prime subsequent processing. Thornton andFernandez-Duque (2000) reported just such an effect using aparadigm in which a change in the orientation of one of severalitems could prime a subsequent orientation decision about anotheritem (a congruency effect) even when participants reported havingfailed to perceive the change.

Because this original finding later proved to be controversial, weset out to explore whether we could replicate Thornton and

Fernandez-Duque’s (2000) “implicit change detection” effect. Ourmain approach has been to integrate, in a single experiment,controls for all known methodological biases identified in previousstudies. Using this improved design, we found congruency effects(i.e., faster RTs and fewer errors) even for those trials for whichparticipants reported not having perceived the change. Our resultstherefore replicate Fernandez-Duque and Thornton’s (2003) find-ings and are thus congruent with the notion that changes in visualdisplays can be processed unconsciously. In the remainder of thisdiscussion, we would like to address three central issues. First, doour findings indeed suggest that changes in visual displays may beprocessed unconsciously up to fairly sophisticated levels that in-volve, for instance, identifying the nature of the change? Second,what are the mechanisms involved? Third, what other evidence isthere to support the notion that undetected changes can influencesubsequent processing? Here, we focus on converging evidenceobtained through functional magnetic resonance imaging (fMRI)and physiological measures in addition to behavioral methods.

Is Implicit Change Detection Implicit?

Fernandez-Duque and Thornton’s (2003) interpretation of im-plicit change detection requires not only implicit registration butalso implicit identification of the change. Three elements seemcrucial for change identification to occur: Representations of bothpre- and postchange displays must exist (which seems to be thecase; see Mitroff, Simons, & Levin, 2004), and a comparisonprocess capable of indicating the nature of the change, if any, mustoperate on these two representations. For implicit change identi-fication, all of this also needs to occur outside of consciousawareness. Fernandez-Duque and Thornton’s interpretation there-fore postulates relatively sophisticated implicit processes.

This being said, other evidence suggests that complex visualprocesses might operate outside conscious or attentional processes.As discussed in the introduction, the context created by an array ofunattended distractors can cue the location of the target in a visualsearch task (Chun & Jiang, 1998). Moore and Egeth (1997) alsoshowed that observers continued to be sensitive to illusory arraysof dots even while failing to experience the illusion because theirattention was engaged elsewhere.

In this context, the replication of the implicit perception of thenature of a change is an important result, not only because itsuggests that change perception can occur implicitly but alsobecause this form of implicit perception can involve sophisticatedprocesses and representations reflecting the identity of a changeand not its mere localization. Our data are particularly convincingin this respect insofar as the congruency effect for unaware trialswas more reliable when the change occurred at a location far fromthe probe. Fernandez-Duque and Thornton (2003), by contrast,reported a congruency effect for unaware trials only when thechange occurred at a location close to the probe. Contrary to ours,this result might suggest that participants were (consciously) pay-ing attention to the localization of the probe when the orientationchange occurred.

One could argue, however, that change detection was at leastpartly supported by conscious perception left undetected by therather insensitive awareness test that we used in our experiment.Indeed, if the awareness test lacks sensitivity, change detectionmight be attributed to unconscious processes not because it indeed

10 LALOYAUX, DESTREBECQZ, AND CLEEREMANS

tapraid1/zfn-xhp/zfn-xhp/zfn00506/zfn2023d06g mazzam S!5 8/18/06 12:54 Art: 2005-0264-R

APA

PROOFS

took place outside of awareness but simply because the awarenesstest did not compel participants to report the actual consciousinformation on which their performance was based. It might be thecase for instance that, given the difficulty of the task, consciousperception was associated with a very low level of confidence andwas therefore not reported by the participants. It is interesting tonote that the very same issues continue to elicit debate in relatedfields, such as implicit learning (Shanks & St. John, 1994) orsubliminal perception (Draine & Greenwald, 1998; Hannula, Si-mons, & Cohen, 2005; Holender, 1986; Merikle & Reingold,1998; Snodgrass & Shevrin, in press).

In our experiment, we used, like Fernandez-Duque and Thorn-ton (2003), a rather limited measure of awareness: We simplyasked participants whether they had seen the change. However, weinstructed them to use a liberal criterion, that is, to respond thatthey had not seen the change only when they were certain that theyhad not seen anything. We used this particular set of instructionsto ensure that “unaware” trials effectively corresponded to cases inwhich participants had failed to consciously perceive any aspect ofthe orientation change. Moreover, this set of instructions also tendsto increase the level of confidence associated with responses to theunaware trials as compared with the aware trials (Mitroff et al.,2002).

This procedure might explain why we observed an awarenesseffect in RT and accuracy (i.e., more errors and slower RTs foraware than for unaware trials). As participants only responded“unaware” when they were confident of not having seen thechange, they did not have to thoughtfully consider whether theyhad perceived something. This process, which requires both timeand resources, must take place for the aware trials. Another inter-pretation would be that aware trials are slower because it takessome time for change perception to reach conscious awareness.This latter interpretation, however, would lead one to consideraware trials as cases in which the orientation change was effec-tively perceived consciously, whereas, given the difficulty of thetask, it is in fact more plausible that participants were uncertain oftheir perception for most of the aware trials. This interpretation isalso supported by the important rate of false alarms (27%) ob-served for the aware trials.

We are confident that our use of a liberal criterion compensatesfor the relative insensitivity of the awareness test in our experi-ment, and that the unaware trials indeed correspond to cases inwhich there was no level or, at most, extremely low levels ofconscious perception. We also believe, however, that the sensitiv-ity of the awareness test should be improved in order to provide amore precise assessment of implicit change detection. An interest-ing way to extend awareness measurement in this task would be toadd a subjective confidence rating to the change detection report.Such a procedure has been previously used in the domains ofsubliminal perception (Cheesman & Merikle, 1984) and implicitlearning (Destrebecqz & Cleeremans, 2003; Dienes, Altmann,Kwan, & Goode, 1995; Dienes & Berry, 1997; Shanks & St. John,1994) and consists in asking participants to rate on a graded scale,from guess to certain, how confident they were in their responses.

According to Cheesman and Merikle (1984), perception is un-conscious when it is under the subjective threshold, that is, whenparticipants are able to identify the target at above chance perfor-mance while stating that they were guessing and did not perceiveit consciously. Reingold and Merikle (1990), however, have in-

sisted that subjective measurement of unconscious perception mustbe interpreted with caution because it depends on the participant’sinterpretation of the task instructions. For instance, participantsmight give a more liberal interpretation to the term guess than theexperimenter does or might assign a high level of expectancy tothe experimenter and therefore tend to claim that they were guess-ing while in fact basing their responding on low-confidence, frag-mentary, but nevertheless conscious knowledge.

A method of particular interest in this context has been intro-duced by Kunimoto, Miller, and Pashler (2001). As a possiblesolution to the potential bias of subjective measurement, theyproposed to use a two-alternative forced-choice confidence ratingassociated with the awareness test. They argued that this methodmakes it possible to use signal detection theory, which can dealwith biases in criterion. The idea is to compute d$ combiningdetection performance and confidence ratings. In this framework,a correct response given with high confidence is considered a hit;a correct response associated with low confidence is viewed as amiss; an incorrect response given with high confidence is consid-ered a false alarm; and an incorrect response associated with lowconfidence is classified as a correct rejection. This method allowedKunimoto et al. to find evidence for subliminal perception even forunbiased measurement of participants’ awareness. This methodallows one to truly dissociate perception from awareness, in thesense that subjects can discriminate different kinds of stimuli at abetter-than-chance level while their confidence rating is not pre-dictive of their performance, hence suggesting that subjects haveno conscious access to the information they use. The added valueof this method is thus that response biases are neutralized thanks tosignal detection theory.

This method could be nicely applied to the implicit changedetection paradigm. It might provide new information regardingthe confidence of the observers and allow control for the potentialbiases described above. It is important to note that this idea isrooted in the notion that consciousness should be considered agradual phenomenon. Indeed, it acknowledges the fact that insome cases, fragmentary or imperfect perception can induce phe-nomenal states that fall between full awareness and complete lackof awareness of some features of the material—states that maylead participants to consider themselves as partially aware of astimulus.

In conclusion, we are quite confident that the information per-ceived by our participants in unaware trials was so weak and noisythat they can be described as truly unaware. Two aspects of ourresults specifically support this conclusion. First, the fact thataccuracy is higher for unaware than for aware performance seemshard to explain by assuming that performance on unaware trialscan be accounted for by weak and fleeting conscious perception ofthe change, for if this were the case, one would expect perfor-mance on such trials to be lower than performance on aware trials.What we found instead is increased accuracy on unaware trials.Second, though we did not expect to observe an interaction be-tween validity and awareness, such a qualitative difference be-tween aware and unaware trials (i.e., a validity effect for the awaretrials and a reverse validity effect for the unaware trials) suggeststhat conscious perception and unconscious sensitivity appeal todifferent mechanisms in this task (Reingold & Merikle, 1990).Though establishing the extent to which observed dissociationsdepend on separate, overlapping, or unitary information-

11IMPLICIT CHANGE IDENTIFICATION

tapraid1/zfn-xhp/zfn-xhp/zfn00506/zfn2023d06g mazzam S!5 8/18/06 12:54 Art: 2005-0264-R

APA

PROOFS

processing systems is notoriously difficult (Dunn & Kirsner,1988), we nevertheless offer a few reflections on these issues in thenext section.

What Are the Mechanisms of Implicit Change Detection?

It seems difficult to explain our data by means of the dominantmodels defended in the change blindness or the visual short-termmemory (VSTM) literature. Indeed, in the field of change blind-ness research, it is often claimed that change detection requiresfocused attention that can be allocated only to a few items at a time(see, e.g., Mitroff et al., 2002; Rensink, 2000a; Rensink et al.,1997). Similarly, in the VSTM literature, the dominant viewclaims that the VSTM capacity limitations strongly restrict thenumber of items that can be encoded at once (e.g., Luck & Vogel,1997). From this perspective, only a few items, typically three orfour, can be encoded from a visual display and then compared withthe subsequent view.

However, it is important to note that some authors have pro-posed that implicit and explicit processes are subtended by differ-ent mechanisms. Thus, according to Rensink (2002) and Thorntonand Fernandez-Duque (2002), focused attention is required for achange to be consciously detected. Thornton and Fernandez-Duque(2002) further suggested that the representation of a change builtby attentional processes can remain unavailable to consciousness.In their view, attention and awareness are thus not synonymous.Attention can be seen as a functional process modulating informa-tion processing, whereas awareness is defined as an “attribute ofthe represented stimulus” (Thornton & Fernandez-Duque, 2002, p.103). Other researchers have also claimed that attention and aware-ness should be distinguished from each other (i.e., Lamme, 2003).

Our results indicate that conscious access does not take place inunaware change identification trials. We cannot say, however,whether attention is involved in these cases.

An alternative perspective proposed by Wilken and Ma (2004),however, offers a very different conceptualization of visual per-ception, one that is based on the notion that neural activity isinherently noisy. Thus, Wilken and Ma claim that an observer whoattempts to detect a change at n locations is monitoring n noisychannels. A change in an item will produce a change in a channel,but because the signal to each channel is noisy, there is a proba-bility that the signal coming from a no-change channel will beabove threshold and produce a false alarm (Wilken & Ma, 2004).

Thus, in this framework, the entire visual display is assumed tobe encoded, but in such a way that the corresponding representa-tions contain a certain level of noise. Note also that Landman,Spekreijse, and Lamme (2003) likewise defend the view that theentire visual display is encoded and stored; however, they considerthat overwriting of the prechange representation by the postchangerepresentation is what causes change blindness (see Lakha &Wright, 2004, for similar findings and discussion). When a changeoccurs, the prechange representation will be compared with thepostchange representation within each channel. Of note, theserepresentations are graded rather than all or none: They vary inquality depending, among other factors, on the level of noise,which in turn depends on the time available for processing. Im-plicit change detection, in this framework, would thus take placewhenever the changes that occur in a channel are of a sufficientstrength to produce priming yet are weak enough not to reach

awareness (see Cleeremans, 2005; Cleeremans & Jimenez, 2001,for similar ideas).

The notion of noisy representations is also defended by Baldassiand Burr (2000), who proposed a “featured-based integration oforientation signals in visual search” (p. 1293). In this perspective,observers could access the different orientations in a display con-taining up to eight grating patches displayed for 100 ms. They alsofound that when noise was added to the display, the identificationof the orientation of a target was at a much lower threshold than itslocation. Hence, the observers can discriminate the orientation ofrectangles without being able to locate them. The authors sug-gested that performance was based on a simple integration mech-anism that quickly extracts the global aspect of the display.

In the same vein, using a different paradigm, Ariely (2001)showed that observers can extract information about the averagecharacteristics of a set of items presented for 500 ms but cannotaccess specific information concerning the individual items withinthe sets. He concluded that when presented with sets of items, thevisual system might create representations reflecting the generalcharacteristics of the set but discard information about the indi-vidual items. However, we cannot determine whether such a globalcomparison mechanism of some average characteristics betweenthe first and the second display is operating in our experiment orwhether a more focused mechanism centered on individual items isat play. Nevertheless, our results clearly suggest that this mecha-nism is implicit. We now turn to converging evidence from func-tional imaging studies of change blindness and change detection.

Evidence From Imaging and Electrophysiological Studies

Our results suggest not only that nonconsciously perceivedchanges can nevertheless be represented but that they can alsoinfluence subsequent behavior. Numerous recent studies have ad-dressed the relationships between subjective experience of a stim-ulus and its neural correlates in an attempt to elucidate the neuralbases of conscious awareness.

As a case in point, a recent study by Haynes and Rees (2005)suggests that sometimes, subjects’ brains seem to know more thanthey can tell. Haynes and Rees used sophisticated signal-processing techniques to demonstrate that it is possible to predict(albeit not perfectly) what stimulus (gratings oriented either to theleft or to the right) a subject has been exposed to on the basis of asingle fMRI image of activity in the visual cortex—a method theydubbed “mindreading.” Strikingly, this was the case regardless ofwhether subjects had actually consciously perceived the stimulus.Indeed, whereas in a first experiment Haynes and Rees used visiblegratings, in a second they used subliminal gratings rendered in-visible by masking—gratings whose orientation subjects wereunable to detect despite prolonged exposure lasting up to 15 s. Ofinterest, Haynes and Rees found graded differences in V1, V2, andV3 activity between the two conditions. V1’s activity, for instance,was predictive of the stimulus regardless of whether subjects hadperceived it, but more so when they had than when they had not.V2 and V3, on the other hand, were active only when subjects hadconsciously perceived the stimulus. Haynes and Rees concludedby pointing out that “whether to be represented in consciousexperience information has to cross a threshold level of activity, orperhaps needs to be relayed to another region of the brain, is aninteresting question for further research” (Haynes & Rees, 2005, p.

12 LALOYAUX, DESTREBECQZ, AND CLEEREMANS

tapraid1/zfn-xhp/zfn-xhp/zfn00506/zfn2023d06g mazzam S!5 8/18/06 12:54 Art: 2005-0264-R

APA

PROOFS

689). Thus, it seems possible that a stimulus that is not perceivedconsciously nevertheless produces brain activation similar to thatproduced by the same stimulus perceived consciously. We nowturn to studies exploring the neural correlates of undetectedchanges in change blindness situations.

A recent fMRI change blindness experiment showed that spe-cific regions of the extrastriate cortex were implicated when par-ticipants were exposed to a change that they failed to detectconsciously as compared with a situation in which no changeoccurred in the display (Beck, Rees, Frith, & Lavie, 2001). More-over, conscious detection of a change was associated with in-creased activation in bilateral parietal and in right dorsolateralfrontal areas as compared with situations in which the subjectsfailed to consciously report the change. The specific pattern ofactivation observed when the change was not consciously reportedsuggests that implicit change detection might occur. However,further studies should replicate this result, as it was found only fora subset of the stimuli used in the study (pictures of faces) and onlyin some subjects (6 out of 10). Another study, using event-relatedpotential methods, found a specific pattern of brain activity asso-ciated with the presence of a change without conscious detection,which was different from the pattern associated with explicitchange detection (Fernandez-Duque, Grossi, Thornton, & Neville,2003). The authors concluded that this specific activation reflectsimplicit change detection. Other researchers concluded, on thebasis of similar differences, that it merely reflects differences inthe level of subjects’ confidence regarding their response ratherthan a difference in the availability to consciousness of the per-ceived change (i.e., participants should be more confident whenthey see the change than when they do not see it) (Eimer & Mazza,2005).

To sum up, these results using different brain imaging methodsand the behavioral findings seem to indicate that implicit changedetection can occur, as a change that is nonconsciously detected iscapable of (a) influencing further behavior and (b) yielding apattern of brain activity that is different from the pattern of activityresulting from a nonchanging stimulus.

Conclusion

We replicated Thornton and Fernandez-Duque’s (2000) originalfindings under better controlled experimental conditions. Thisresult suggests that weak and noisy spatiotemporal representationscan nevertheless influence subsequent behavior. Because partici-pants remain unaware of such representations and of their influ-ence on their behavior, they can be described as implicit. Ourreplication findings extend previous results and, interestingly,elaborate on subliminal perception findings, for in contrast tosubliminal perception, implicit change perception requires integra-tion of two different displays over time.

References

Ariely, D. (2001). Seeing sets: Representation by statistical properties.Psychological Science, 12, 157–162.

Baldassi, S., & Burr, D. C. (2000). Feature-based integration of orientationsignals in visual search. Vision Research, 40, 1293–1300.

Beck, D. M., Rees, G., Frith, C. D., & Lavie, N. (2001). Neural correlatesof change detection and change blindness. Nature Neuroscience, 4,645–650.

Cheesman, J., & Merikle, P. M. (1984). Priming with and without aware-ness. Perception & Psychophysics, 36, 387–395.

Chun, M. M., & Jiang, Y. (1998). Contextual cueing: Implicit learning andmemory of visual context guides spatial attention. Cognitive Psychology,36, 28–71.

Cleeremans, A. (2005). Computational correlates of consciousness.Progress in Brain Research, 150, 81–98.

Cleeremans, A., & Jimenez, L. (2001). Implicit learning and conscious-ness: A graded, dynamic perspective. In R. M. French & A. Cleeremans(Eds.), Implicit learning and consciousness: An empirical, computa-tional and philosophical consensus in the making? (pp. 1–40). Hove,England: Psychology Press.

Cleeremans, A., & McClelland, J. L. (1991). Learning the structure ofevent sequences. Journal of Experimental Psychology: General, 120,235–253.

Dejong, R., Liang, C. C., & Lauber, E. (1994). Conditional and uncondi-tional automaticity: A dual-process model of effects of spatial stimulus–response correspondence. Journal of Experimental Psychology: HumanPerception and Performance, 20, 731–750.

Destrebecqz, A., & Cleeremans, A. (2001). Can sequence learning beimplicit? New evidence with the process dissociation procedure. Psy-chonomic Bulletin & Review, 8, 343–350.

Destrebecqz, A., & Cleeremans, A. (2003). Temporal factors in sequencelearning. In L. Jimenez (Ed.), Attention and implicit learning (pp.181–213). Philadelphia: John Benjamins.

Dienes, Z., Altmann, G. T. M., Kwan, L., & Goode, A. (1995). Uncon-scious knowledge of artificial grammars is applied strategically. Journalof Experimental Psychology: Learning, Memory, and Cognition, 21,1322–1338.

Dienes, Z., & Berry, D. C. (1997). Implicit learning: Below the subjectivethreshold. Psychonomic Bulletin & Review, 4, 3–23.

Draine, S. C., & Greenwald, A. G. (1998). Replicable unconscious seman-tic priming. Journal of Experimental Psychology: General, 127, 286–303.

Dunn, J. C., & Kirsner, K. (1988). Discovering functionally independentmental processes: The principle of reversed association. PsychologicalReview, 95, 91–101.

Eimer, M., & Mazza, V. (2005). Electrophysiological correlates of changedetection. Psychophysiology, 42, 328–342.

Fernandez-Duque, D., Grossi, G., Thornton, I. M., & Neville, H. J. (2003).Representation of change: Separate electrophysiological markers of at-tention, awareness, and implicit processing. Journal of Cognitive Neu-roscience, 15, 491–507.

Fernandez-Duque, D., & Thornton, I. M. (2000). Change detection withoutawareness: Do explicit reports underestimate the representation ofchange in the visual system? Visual Cognition, 7, 323–344.

Fernandez-Duque, D., & Thornton, I. M. (2003). Explicit mechanisms donot account for implicit localization and identification of change: Anempirical reply to Mitroff et al. (2002). Journal of Experimental Psy-chology: Human Perception and Performance, 29, 846–858.

Greenwald, A. G., Abrams, R. L., Naccache, L., & Dehaene, S. (2003).Long-term semantic memory versus contextual memory in unconsciousnumber processing. Journal of Experimental Psychology: Learning,Memory, and Cognition, 29, 235–247.

Hannula, D. E., Simons, D. J., & Cohen, N. J. (2005). Opinion: Imagingimplicit perception: Promise and pitfalls. Nature Reviews Neuroscience,6, 247–255.

Haynes, J.-D., & Rees, G. (2005). Predicting the orientation of invisiblestimuli from activity in human primary visual cortex. Nature Neuro-science, 8, 686–691.

Holender, D. (1986). Semantic activation without conscious activation indichotic listening, parafoveal vision, and visual masking: A survey andappraisal. Behavioral and Brain Sciences, 9, 1–23.

Hollingworth, A., & Henderson, J. M. (2002). Accurate visual memory for

13IMPLICIT CHANGE IDENTIFICATION

tapraid1/zfn-xhp/zfn-xhp/zfn00506/zfn2023d06g mazzam S!5 8/18/06 12:54 Art: 2005-0264-R

APA

PROOFS

previously attended objects in natural scenes. Journal of ExperimentalPsychology: Human Perception and Performance, 28, 113–136.

Kunimoto, C., Miller, J., & Pashler, H. (2001). Confidence and accuracy ofnear-threshold discrimination responses. Consciousness and Cognition,10, 294–340.

Lakha, L., & Wright, M. J. (2004). Capacity limitations of visual memoryin two-interval comparison of Gabor arrays. Vision Research, 44, 1707–1716.

Lamme, V. A. F. (2003). Why visual attention and awareness are different.Trends in Cognitive Sciences, 7, 12–18.

Landman, R., Spekreijse, H., & Lamme, V. A. F. (2003). Large capacitystorage of integrated objects before change blindness. Vision Research,43, 149–164.

Li, F. F., VanRullen, R., Koch, C., & Perona, P. (2002). Rapid naturalscene categorization in the near absence of attention. Proceedings of theNational Academy of Sciences, US A, 99, 9596–9601.

Luck, S. J., & Vogel, E. K. (1997, November 20). The capacity of visualshort-term memory for features and conjunctions. Nature, 309, 279–281.

Merikle, P. M., & Reingold, E. M. (1998). On demonstrating unconsciousperception: Comment on Draine and Greenwald (1998). Journal ofExperimental Psychology: General, 127, 304–310.

Mitroff, S. R., Simons, D. J., & Franconeri, S. L. (2002). The siren song ofimplicit change detection. Journal of Experimental Psychology: HumanPerception and Performance, 28, 798–815.

Mitroff, S. R., Simons, D. J., & Levin, D. T. (2004). Nothing compares 2views: Change blindness can occur despite preserved access to thechanged information. Perception & Psychophysics, 66, 1268–1281.

Moore, C. M., & Egeth, H. (1997). Perception without attention: Evidenceof grouping under conditions of inattention. Journal of ExperimentalPsychology: Human Perception and Performance, 23, 339–352.

O’Regan, J. K., Rensink, R. A., & Clark, J. J. (1999, March 4). Change-blindness as a result of “mudsplashes.” Nature, 398, 34.

Posner, M. I. (1980). Orienting of attention. Quarterly Journal of Exper-imental Psychology, 32, 3–25.

Reingold, E. M., & Merikle, P. M. (1990). On the inter-relatedness oftheory and measurement in the study of unconscious processes. Mind &Language, 5, 9–28.

Rensink, R. A. (2000a). The dynamic representation of scenes. VisualCognition, 7, 17–42.

Rensink, R. A. (2000b). Visual search for change: A probe into the natureof attentional processing. Visual Cognition, 7, 345–376.

Rensink, R. A. (2002). Change detection. Annual Review of Psychology,53, 245–277.

Rensink, R. A., O’Regan, J. K., & Clark, J. J. (1997). To see or not to see:The need for attention to perceive changes in scenes. PsychologicalScience, 8, 368–373.

Shanks, D. R., & St. John, M. F. (1994). Characteristics of dissociablehuman learning systems. Behavioral and Brain Sciences, 17, 367–447.

Simons, D. J. (2000). Current approaches to change blindness. VisualCognition, 7, 1–15.

Simons, D. J., & Ambinder, M. S. (2005). Change blindness: Theory andconsequences. Current Directions in Psychological Science, 14, 44–48.

Simons, D. J., & Chabris, C. F. (1999). Gorillas in our midst: Sustainedinattentional blindness for dynamic events. Perception, 28, 1059–1074.

Simons, D. J., Chabris, C. F., Schnur, T., & Levin, D. T. (2002). Evidencefor preserved representations in change blindness. Consciousness andCognition, 11, 78–97.

Simons, D. J., & Rensink, R. A. (2005). Change blindness: Past, present,and future. Trends in Cognitive Sciences, 9, 16–20.

Smilek, D., Eastwood, J. D., & Merikle, P. M. (2000). Does unattendedinformation facilitate change detection? Journal of Experimental Psy-chology: Human Perception and Performance, 26, 480–487.

Snodgrass, M., & Shevrin, H. (2006). Unconscious inhibition and facili-tation at the objective detection threshold: Replicable and qualitativelydifferent unconscious perceptual effects. Cognition, 101, 43–79.

Thornton, I. M., & Fernandez-Duque, D. (2000). An implicit measure ofundetected change. Spatial Vision, 14, 21–44.

Thornton, I. M., & Fernandez-Duque, D. (2002). Change blindness andtranssaccadic integration. In J. Hyona, D. P. Munoz, W. Heide, & R.Radach (Eds.), Progress in brain research (Vol. 140, pp. 99–118).Amsterdam: Elsevier Science.

Wilken, P., & Ma, W. J. (2004). A detection theory account of changedetection. Journal of Vision, 4, 1120–1135.

Williams, P., & Simons, D. J. (2000). Detecting changes in novel, complexthree-dimensional objects. Visual Cognition, 7, 297–322.

Received September 9, 2005Revision received January 7, 2006

Accepted January 18, 2006 "

14 LALOYAUX, DESTREBECQZ, AND CLEEREMANS

tapraid1/zfn-xhp/zfn-xhp/zfn00506/zfn2023d06g mazzam S!5 8/18/06 12:54 Art: 2005-0264-R