On the electrophysiological evidence for the capture of visual attention

On the Electrophysiological Evidence for the Capture of Visual Attention

John J. McDonald, Jessica J. Green, Ali Jannati, and Vincent Di LolloSimon Fraser University

The presence of a salient distractor interferes with visual search. According to the salience-drivenselection hypothesis, this interference is because of an initial deployment of attention to the distractor.Three event-related potential (ERP) findings have been regarded as evidence for this hypothesis: (a)salient distractors were found to elicit an ERP component called N2pc, which reflects attentionalselection; (b) with target and distractor on opposite sides, a distractor N2pc was reported to precede thetarget N2pc (N2pc flip); (c) the distractor N2pc on slow-response trials was reported to occur particularlyearly, suggesting that the fastest shifts of attention were driven by salience. This evidence is equivocal,however, because the ERPs were noisy (b, c) and were averaged across all trials, thereby making itdifficult to know whether attention was deployed directly to the target on some trials (a, b). Wereevaluated this evidence using a larger sample size to reduce noise and by analyzing ERPs separatelyfor fast- and slow-response trials. On fast-response trials, the distractor elicited a contralateral positivity(PD)—an index of attentional suppression—instead of an N2pc. There was no N2pc flip or earlydistractor N2pc. As it stands, then, there is no ERP evidence for the salience-driven selection hypothesis.

Keywords: visual search, attention capture, additional singleton paradigm, N2pc, PD

People must often search cluttered and continually changing visualenvironments for objects of interest (targets). The search for the targetis particularly challenging when another highly salient object is pres-ent in the search array. Visual search for a target in the presence of asalient but irrelevant item (the distractor) has been studied in thelaboratory using the additional-singleton paradigm (Lamy & Yashar,2008; Pinto, Olivers, & Theeuwes, 2005; Theeuwes, 1991, 1992). Inthis paradigm, observers search covertly for a target singleton whiletrying to ignore an irrelevant singleton that is also present in thedisplay on a subset of trials. The target and distractor “pop out” fromthe rest of the items because they each possess a unique feature. Inmost experiments of this sort, the distractor is chosen to pop out evenmore than the target, so that distractor’s salience and the observer’sintention are in opposition.

Early studies with the additional-singleton paradigm (Theeuwes,1991, 1992) gave rise to the salience-driven selection hypothesis inwhich the initial visual selection is said to be determined entirelyby bottom-up activations based on stimulus salience (for a recentreview, see Theeuwes, 2010).1 According to this hypothesis, vo-litional control comes online only after the most salient item in thedisplay has been selected. Thus, if the target is less salient than the

distractor, observers must select the distractor first, determine thatit is irrelevant, and only then can they select the target. By thisaccount, automatic selection of the distractor delays search for thetarget by as much as 150 ms when the features of the target and thedistractor swap randomly from trial to trial (mixed-feature search;Theeuwes, 1991).

Although there is general agreement that salience computationsin early vision can drive selection in a bottom-up fashion, the ideathat salience-driven selection is obligatory has met stiff opposition.One of the main competing perspectives—the contingent-capturehypothesis—holds that an irrelevant singleton can capture atten-tion only when its features are relevant to the task at hand (Folk,Remington, & Johnston, 1992). Evidence for contingent capturehas been obtained with a modified spatial cueing paradigm inwhich the conventional abrupt-onset stimuli are replaced by cueand target search displays. The cue display contains an irrelevantdistractor singleton whose location is completely independentfrom the location of the subsequent target singleton. In this para-digm, the location of the distractor singleton is found to influencethe time required to find the target singleton only when thedefining feature of the distractor singleton matches that of thetarget singleton (Folk & Remington, 1998, 1999; Folk et al., 1992;Folk, Remington & Wright, 1994). This pattern of results suggeststhat salience-driven selection is contingent on whether the featuresof the most salient item match the observer’s attentional set.

On the face of it, the pictures emerging from the additional-singleton and the cueing paradigms are clear-cut and diametricallyopposed: Data from the additional-singleton paradigm have gen-

1 The salience-driven selection hypothesis has been modified over theyears to account for new findings (for a review, see Theeuwes, 2010).Herein, we refer to the strong version of the hypothesis, according to whichattention must be deployed to the location of the most physically salientitem in the display before it can be redeployed to the location lessphysically salient items.

John J. McDonald, Jessica J. Green, Ali Jannati, and Vincent Di Lollo,Department of Psychology, Simon Fraser University, British Columbia,Canada.

This study was supported by a grant from the Natural Sciences andEngineering Research Council of Canada [JJM]. We thank Greg Christie,Christina Hull, and T. J. Radonjic for assistance.

Dr. Jessica J. Green is now at the Department of Psychology andMcCausland Center for Brain Imaging, University of South Carolina, USA.

Correspondence concerning this article should be addressed to John J.McDonald, Department of Psychology, Simon Fraser University, 8888University Drive, Burnaby, BC V5A 1S6, Canada. E-mail: [email protected]

Journal of Experimental Psychology:Human Perception and Performance

© 2012 American Psychological Association

2012, Vol. 38, No. 6, 0000096-1523/12/$12.00 DOI: 10.1037/a0030510

1

erally supported the bottom-up, salience-driven selection hypoth-esis, whereas data from the cuing paradigm have generally sup-ported the top-down, contingent-capture hypothesis. However,several alternative explanations have emerged to challenge theinterpretations of the results obtained in each paradigm. On the onehand, the apparent lack of salience-driven selection in the cueingparadigm has been attributed to rapid disengagement of attentionfrom cue singletons that are easily distinguishable from the target(Theeuwes, Atchley, & Kramer, 2000). On the other hand, it hasbeen argued that in the additional-singleton paradigm a salientdistractor might delay target selection without capturing attention(Folk & Remington, 1998; LaBerge, 2002). For example, Folk andRemington (1998) suggested that when multiple singletons arepresent, additional nonspatial filtering is required to determinewhich of the singletons should be attended (cf. Kahneman, Treis-man, & Burkell, 1983).

Given the possibility of these alternative explanations, no con-sensus has been reached as to whether salient-but-irrelevant stim-uli capture attention. Part of the problem is that the two competinghypotheses now make similar predictions regarding reaction time(RT) performance in additional-singleton and cueing tasks. Thisproblem stems in part from the difficulty in pinpointing the neu-rocognitive stages that mediate the RT effects. Namely, it isdifficult to link RT effects—such as the RT interference observedin the additional singleton paradigm—to a specific processingstage because RTs reflect the cumulative result of multiple pro-cessing stages. Moreover, in both paradigms, processing of thedistractor must be inferred from the ways in which that stimulusinfluences responses to the target. In other words, distractor pro-cessing is measured indirectly, which makes it difficult to knowwhether salient distractors capture attention or influence RTs with-out capturing attention.

Researchers have begun to use ERPs in conjunction with be-havioral measures to determine whether salient distractors captureattention automatically or contingently. ERPs reflect moment-to-moment changes in postsynaptic potentials that are related tosensory, cognitive, and motor events (Luck, 2005). Stimulus-driven ERPs can be measured even when participants make noovert response to the eliciting stimulus, making the ERP methodideally suited for the investigation of distractor processing in theadditional-singleton and cueing paradigms.

Most ERP studies of attention capture have focused on a com-ponent of the visual ERP that is known to reflect the selection ofattended items in visual search. This component—known as theN2 posterior contralateral (N2pc) —is apparent in the ERPs re-corded over lateral occipital scalp regions, approximately 175–300ms after the onset of a multi-item display. During that timeinterval, the ERP waveform recorded contralateral to the attendeditem is more negative than the ERP waveform recorded ipsilateralto the attended item. This difference is believed to reflectattentional-filtering operations that enable selective processing ofthe attended item (Boehler, Tsotsos, Schoenfeld, Heinze, & Hopf,2011; Eimer, 1996; Hickey, Di Lollo, & McDonald, 2009; Hopf,Boelmans, Schoenfeld, Heinze, & Luck, 2002; Luck, Girelli, Mc-Dermott, & Ford, 1997; Luck & Hillyard, 1994a, 1994b). Al-though there is some debate as to whether the N2pc reflectssuppression of unattended signals (e.g., Luck & Hillyard, 1994b;Luck et al., 1997), selection of attended signals (e.g., Eimer, 1996),or some combination of target enhancement and distractor sup-

pression (Hickey et al., 2009), there is a general consensus that theN2pc is a good measure of attentional selection (Eimer & Kiss,2010; Hickey et al., 2009; Woodman & Luck, 2003).

On the ERP Evidence for Attention Capture

There are three ERP findings that have been regarded as evi-dence for the salience-driven selection hypothesis. The first wasreported by Hickey, McDonald, and Theeuwes (2006), who con-ducted a mixed-feature search task (cf. Theeuwes, 1991) andexamined lateralized ERPs to various additional-singleton searchdisplays. On some trials, the search display contained a salientdistractor to the left or right of fixation and a less-salient targetabove or below fixation. With this configuration, lateralized ERPcomponents such as the N2pc can be attributed to the lateralsingleton since the midline singleton would activate both corticalhemispheres equally (Hickey et al., 2009, 2006; Woodman &Luck, 2003). Critically, Hickey et al. found an N2pc contralateralto the salient distractor when the target was on the vertical midline(henceforth we refer to the distractor-elicited N2pc as distractorN2pc and the target-elicited N2pc as target N2pc). This demon-strates that observers deployed attention directly to the distractor,at least on a portion of trials.

The second ERP finding consistent with salience-driven atten-tion capture is illustrated in Figure 4c of the Hickey et al. (2006)study, which presented the ERPs to displays containing a targetand a more salient distractor on opposite sides of fixation. TheERP was initially more negative contralateral to the distractor inthe N2pc interval (220–265 ms). Following this distractor N2pc,the ERP became more negative contralateral to the target. Thissequence of events, to which we refer as the N2pc flip, wasregarded by Hickey et al. as evidence that observers attended to thesalient-but-irrelevant singleton before attending to the target. In-terpretation of this N2pc flip, however, is complicated by consid-erations of noise in the data. Namely, the magnitude of the dis-tractor N2pc was no larger than the contralateral-ipsilateraldifferences in the prestimulus baseline.2 As has been pointedlynoted by Woodman (2010), for an ERP effect to be regarded asreal its magnitude must exceed that of the prestimulus noise.Because of these considerations, the reliability of the N2pc flip inthe Hickey et al. (2006) study is open to question. Thus, theconclusion that observers attended to the salient distractor beforeattending to the target is also questionable.

The third ERP finding consistent with salience-driven attentioncapture hypothesis was reported by Hickey, van Zoest, and Theeu-wes (2010) who reanalyzed the data from Experiment 1 of Hickeyet al. (2006). Specifically, the trials were subdivided into quartiles,according to RTs, with the ERPs averaged separately for the fastestand slowest quartiles. It was found that the distractor N2pc ob-

2 We analyzed ERP data from Hickey et al.’s (2006) Experiment 2 tocompare the distractor N2pc with the pre-stimulus noise in the lateral-target, contralateral-distractor display configuration. The peak amplitude ofthe early distractor N2pc was 0.36 !V. For comparison, we computed thepeak amplitudes of the largest positive and negative noise peaks in thepre-stimulus baseline ("0.45 !V at "66 ms and #0.33 !V at "4 ms) andthen averaged the absolute values of these peak amplitudes. The resultingestimate of peak noise amplitude in the pre-stimulus baseline was 0.39 !V.In other words, the reported distractor N2pc was smaller than the largestnoise peaks in the prestimulus baseline.

2 MCDONALD, GREEN, JANNATI, AND DI LOLLO

tained on the slowest quartile occurred earlier than all other targetand distractor N2pc waves. On the basis of this finding, Hickey etal. (2010) concluded that the fastest shifts of attention—as indexedby the latency of the N2pc—were directed to the salient distractor,which delayed target-discrimination responses.

It should be noted that Experiment 1 of Hickey et al. (2006), onwhich the reanalysis by Hickey et al. (2010) was based, wasacknowledged to contain a confound that may have led partici-pants to search for the distractor voluntarily as a cue for thetarget’s location. The same confound, therefore, may have biasedthe outcome of the reanalysis. In addition, the problem with noiselevel, noted above, was exacerbated by the quartile split thatincluded only one quarter of the original trials in the fastest andslowest trials. This can be best seen in Figure 3 of Hickey et al.(2010), in which the noise peaks during the initial 130 ms of theERP plots (100 ms prestimulus and the initial 30 ms poststimulus,in which there is no visually evoked activity) were in excess of 0.4!V. Only one of the N2pc peaks (target N2pc on fast trials) wasconsiderably larger. These considerations question the reliabilityof the timing and amplitude measures of the N2pc peaks seen inHickey et al. (2010, Figure 3). This concern is compounded by thelack of any statistical evidence in the Hickey et al. (2010) paperthat those components were significantly different from zero.

In sum, three ERP findings have been regarded as supporting thesalience-driven selection hypothesis, but two of these findings arequestionable because of low signal-to-noise ratio in the ERPrecordings. The first finding—the existence of a distractor N2pc—does provide convincing evidence that participants deployed at-tention to the salient distractor in Hickey et al.’s (2006) mixed-feature search. However, an additional finding reported by Hickeyet al. (2006) is inconsistent with the notion of salience-drivenattention capture. Specifically, the salience-driven selection hy-pothesis predicts that the distractor N2pc (with the target on themidline) should occur before the target N2pc (with the distractoron the midline). Contrary to this prediction, however, the targetand distractor N2pc waves occurred in the same time range in bothexperiments (Hickey et al., 2006, Figures 2 and 3). This suggeststhat attention was oriented directly to the target on some trials andto the distractor on other trials, or that it was deployed to bothstimuli simultaneously. Any evidence that participants deploy at-tention directly to the target without first attending to the moresalient distractor would be difficult to reconcile with the salience-driven selection hypothesis.

The present study was designed to reexamine the evidence forsalience-driven selection in the mixed-feature variant of the addi-tional singleton search task. In this pursuit, we were guided by theidea that selective processing may be governed by different factorson different trials. For example, it is plausible that deployingattention initially to the salient distractor would lead to slowerresponses whereas deploying attention directly to the target wouldlead to faster responses. In that case, ERPs averaged across alltrials might obscure the search processes that take place on anyindividual trial. This is because the ERPs averaged across all trials(herein referred to as all-trials ERPs) would reflect the algebraicsummation of the waveforms elicited on fast- and slow-responsetrials and would, therefore, obscure the ERPs attributable to eachtype of trial separately. Thus, it is important to sort trials intosubsets of “like” trials so that the ERPs averaged within eachsubset reflect more accurately the processing of target and distrac-

tor on those trials. Accordingly, in the present work, the trials foreach participant were subdivided into fast-response and slow-response subsets, depending on whether the RT was shorter orlonger than the median RT for that display configuration.

The original data from the 14 participants used in Hickey et al.’sExperiment 2 were used for the present study, along with 26 newparticipants. The number of participants was increased to 40 toreduce the level of noise in the ERPs and therefore allow for amore accurate assessment of the N2pc waves, particularly for theanalysis of ERPs obtained on fast- and slow-response subsets oftrials. Experiment 2 did not suffer from the methodological con-found that marred Hickey et al.‘s Experiment 1.

Method

The Research Ethics Board at Simon Fraser University ap-proved the experimental procedures used in this study.

Participants

Forty-nine neurologically typical volunteers from Simon FraserUniversity participated in this experiment after giving informedconsent. They were either paid for their participation or receivedcourse credit. Data from nine participants were excluded from theanalyses because of excessive blinks or eye movements. Each ofthe remaining 40 participants reported normal or corrected-to-normal visual acuity and normal color vision. Demographic datafor eight participants were not obtained. Of the remaining 32participants (ages 18–40 years, M $ 23.6 years), 14 were women,and two were left-handed. A previously published study was basedon data from 14 of these participants (Hickey et al., 2006, Exper-iment 2). In the present study, the original data from these 14participants were combined with data from 26 new participants.

Apparatus

The experiment was conducted in a sound-attenuated and elec-trically shielded chamber that contained a 19-in CRT monitor withthe screen resolution set to 800 % 600 pixels. Participants sat in anadjustable chair and viewed the monitor from a distance of 60 cm.A Windows-based computer running Presentation (Neurobehav-ioral Systems Inc., Albany, CA) controlled stimulus presentationand registered the participants’ button presses. A secondWindows-based computer running custom software (Acquire) con-trolled EEG acquisition. The acquisition computer housed a 64-channel A-to-D board (PCI 6071e, National Instruments, Austin,TX) that was connected to an EEG amplifier system with highinput impedance (SA Instrumentation, San Diego, CA). Tin elec-trodes mounted in an elastic cap (Electro-cap International, Eaton,OH) were used to record EEG.

Stimuli and Procedure

Search displays consisted of 10 items spaced equally around animaginary circle (9.1° radius) that was centered upon a fixationpoint. Items were unfilled diamonds (4.2° % 4.2°) and circles (1.7°radius) with thin (0.3°) red or green outlines. Each display con-tained nine identically shaped items (either all circles or all dia-monds) and one shape singleton (a diamond among circles or acircle among diamonds). On two thirds of the trials, one of the nine

3ELECTOPHYSIOLOGICAL EVIDENCE FOR CAPTURE

identically shaped objects was a color singleton (either a red itemamong green items or a green item among red items). A horizontalor vertical gray line (0.3° % 1.5°) was centered within each object.

On each trial, a fixation point appeared for 600–1,600 ms andwas followed by a search display, which remained on screen for100 ms after a response was registered. The participants’ task wasto press one of two mouse buttons with their dominant hand toindicate the orientation of the line contained in the shape singleton(the target). The shapes and colors of the 10 objects changedpseudorandomly across trials within each block (mixed-featuresearch condition). Distractor-present and distractor-absent trialswere also intermixed within the same blocks of trials. Participantscompleted 30 blocks of 48 trails, for a total of 1,440 experimentaltrials, after performing at least 48 practice trials.

The target shape-singleton appeared at one of the eight lateralpositions or one of the two positions on the vertical midline. Whenpresent, the distractor also appeared either at a lateral position ora midline position, resulting in the following display configura-

tions: lateral target, midline distractor (&17%); lateral distractor,midline target (&17%), lateral target, no distractor (&33%); lateraltarget, contralateral distractor (&17%); lateral target, ipsilateraldistractor (&17%). The relevant configurations are illustrated inFigure 1.

Behavioral Analyses

Median RTs were computed for distractor-present and distractor-absent trials for each participant, after excluding trials on whichparticipants responded incorrectly, too quickly (RT '100 ms) or tooslowly (RT (2,000 ms). A paired t test was used to assess thedifference between the resulting median RTs. This was done togauge the overall RT interference effect (distractor-present RTs vs.distractor-absent RTs). Following this RT analysis, median RTswere computed separately for each of five display configurationsof interest. These median RTs were used to analyze ERPs associ-

ipsilateral to targetcontralateral to target

Original Sample (N=14)

ipsilateral to lateral singletoncontralateral to lateral singleton

Full Sample (N=40)

f

j

b

d

h

PD

800 ms

+2 µV

c

i -

a

e

g

PD

SPCNN2pc

ALL-TRIALS ERPs

Figure 1. Event-related potentials (ERPs) elicited by the display configurations of interest, averaged across alltrials, for both the original sample used by Hickey, McDonald, & Theeuwes (2006, Experiment 2) and the fullsample. The shaded boxes in Panels g and h represent the distractor and target N2pc time windows used inHickey et al.’s (2006) Experiment 2. The remaining shaded boxes represent the main N2pc time window(225–325 ms). The small, filled squares on the x-axis represent time intervals during which the mean amplitudeof the N2pc was statistically significant (see text for details). Negative voltages are plotted up, by convention.

4 MCDONALD, GREEN, JANNATI, AND DI LOLLO

ated with the various search displays as a function of responsespeed.

Electrophysiological Recording

EEG signals were recorded using our standard procedures, in-cluding a semiautomated routine for rejection of trials contami-nated by blinks or eye movements (for additional details, seeGreen, Conder, & McDonald, 2008). Eye movements were mon-itored using bipolar recordings from electrodes positioned lateralto the left and right external canthi (horizontal electro-oculogram,HEOG). Blinks were detected using scalp electrode FP1, posi-tioned over the left eye. Electrode impedances were kept below 10k). All signals were amplified with a gain of 20,000, were filteredwith a pass-band of 0.1–100 Hz, and were digitized at 500 Hz.

ERP Analyses: N2pc and Beyond

ERPs and average HEOGs time-locked to the various search-display configurations were computed separately, based onartifact-free trials. The averaged waveforms were digitally low-pass filtered with a half-power cutoff at 25 Hz to remove high-frequency noise produced by muscle activity and external electri-cal sources and were digitally rereferenced to the average of theleft and right mastoids.

We focused on ERPs elicited by search displays that contained:(a) lateral target and midline distractor; (b) lateral distractor andmidline target; (c) lateral target and no distractor; (d) lateral targetand contralateral distractor; (e) lateral target and ipsilateral dis-tractor. The third configuration enabled us to examine target pro-cessing in the absence of a salient distractor. The first and secondconfigurations enabled us to measure N2pc (and other lateralizedERPs) to the lateral singleton when paired with a competingstimulus on the midline (cf. Hickey et al., 2006, 2009; Woodman& Luck, 2003).3 The remaining two configurations enabled us toinvestigate the effect of relative target-distractor separation (oppo-site side vs. same sides) and, in particular, to look for the N2pc flipreported by Hickey et al. (2006).

For each participant, the ERP waveforms were collapsed acrossleft and right visual hemifields and left and right electrode sites tocreate waveforms recorded contralateral and ipsilateral to a lateraltarget or distractor. This was done in two steps. In the first step, weaveraged ERPs to the search displays of interest irrespective of RT(i.e., all-trials ERPs). In the second step, we averaged ERPsseparately for fast-response and slow-response trials. Our mainRT-based ERP analysis was performed by computing the medianRT for each display configuration, separately for each participant.Individual trials with RTs falling below or above the relevantmedian RT were defined as fast-response and slow-response trials,respectively. This median-split analysis was followed by aquartile-based analysis, which isolated the shortest and longest RTquartiles on a participant-by-participant basis.

ERP amplitudes were computed in specific time windows cen-tered on the peaks observed in the relevant contralateral-ipsilateraldifference waves, relative to a 200-ms prestimulus baseline period.We measured three lateralized ERP components in order to tracktarget and distractor processing through multiple stages associatedwith attention and working memory: (a) the N2pc, an index ofattentional selection (Luck & Hillyard, 1994a, 1994b); (b) a con-

tralateral distractor positivity (PD), an index of attentional sup-pression (Hickey et al., 2009); (c) a sustained posterior contralat-eral negativity (SPCN), an index of the maintenance of visualinformation in working memory (Jolicœur, Brisson, & Robitaille,2008). Except where noted, the N2pc, PD, and SPCN were mea-sured in the 225–325 ms, 340–390 ms, and 700–800 ms timewindows, respectively. N2pc latencies were measured as the timeat which the voltage reached 70% of the peak amplitude within thetime window of interest. A wider window (75–325 ms) was usedfor this fractional peak latency measure than for the mean ampli-tude measure to avoid computational problems. Fractional peaklatency measures were based on jackknife-average ERPs ratherthan the individual-subject ERPs, and the results of statistical testswere adjusted accordingly (Ulrich & Miller, 2001).

Results and Discussion

Behavior

Approximately 17% of trials were discarded because of incor-rect target discrimination (10.6%) or excessively fast or slowresponse (RT '100 ms or RT (2,000 ms; 6.7%). In addition,15.1% of trials were excluded because of contamination by EEG/EOG artifacts. On the remaining trials, the presence of the salientdistractor slowed responses considerably: The average medianRTs for distractor-present and distractor-absent trials were 965 msand 841 ms, respectively. This 124-ms distractor interferenceeffect was statistically significant, t(39) $ 13.9, p ' .0001.4 Afollow-up analysis was performed to compare the median RTs forlateral-target, no distractor displays (configuration 3 in Table 1) tothe median RTs for distractor-present displays containing lateraltargets (configurations 1, 4, and 5 in Table 1). This analysisrevealed an RT-interference effect of 128 ms (833 ms vs. 962 ms,respectively), t(39) $ 13.53, p ' .001. Table 1 presents the across-participant averages of the median RTs and error rates associated withall five critical display configurations. There was also a small butstatistically significant difference in the error rates across distractor-present (10.4% * 1.8 SD) and distractor-absent (11.0% * 2.2 SD)trials, t(39) $ 2.04, p $ .048.

Electrophysiology

In the Introduction, we reviewed three key pieces of ERPevidence for the salience-driven selection hypothesis. Here weevaluate that evidence in light of the present results.

Do salient distractors elicit N2pc? Hickey et al. (2006)found that a lateral distractor elicited an N2pc when the target wason the vertical meridian. This can be seen in Figure 1, whichpresents the ERPs averaged across the subset of 14 participants inHickey et al.’s Experiment 2 and across the full sample of 40

3 The attentional selection of a midline singleton is presumed to involveboth visual cortical regions in the left and right hemisphere; consequently,these midline stimuli elicit no net N2pc.

4 Hickey et al. (2006) reported a substantially larger interference effectin Experiment 2 (321 ms). This report was in error, however. A re-analysisof the RT data in the original sample (N $ 14) of participants revealed thatthe mean average RTs for distractor-present and distractor-absent trialswere 929 ms and 812 ms, respectively. This 117-ms difference wassignificantly different from zero, t(13) $ 8.66, p ' .0001.

5ELECTOPHYSIOLOGICAL EVIDENCE FOR CAPTURE

participants. The ERPs to the lateral-distractor, midline-target dis-play are shown in Figures 1c and 1d. The distractor N2pc isevidenced by the finding that the ERP is more negative contralat-eral to the distractor than ipsilateral to it in the time range of the N2(225–325 ms; shaded boxes). The presence of a distractor N2pc ledHickey et al. to conclude that attention was often oriented to thesalient distractor. However, in both the Hickey et al. study and inthe present study, a lateral target was also found to elicit an N2pcin the same time range (Figures 1a and 1b). This temporal coin-cidence is inconsistent with the salience-driven selection hypoth-esis, according to which attention must be deployed to the moresalient distractor before it can be redeployed to the less salienttarget.

To examine the time courses of the target and distractor N2pcwaves, we analyzed the mean N2pc amplitudes in five consecutive25-ms intervals starting at 200 ms, separately for each displayconfiguration. The results of these analyses are presented in theright-hand panels of Figure 1 where the filled boxes on the time-line indicate the 25-ms intervals at which the N2pc to the lateralsingleton was significant. Contrary to predictions from thesalience-driven selection hypothesis, there was considerable over-

lap in the time courses of target and distractor N2pc waves, withno evidence of an early distractor N2pc. In fact, the interval inwhich the target N2pc became significant preceded the interval inwhich the distractor N2pc became significant (Figures 1b and 1d,respectively). This suggests that the onset of the target N2pc mayhave preceded that of the distractor N2pc.

The temporal overlap of target and distractor N2pc waves mayhave resulted from the averaging of two different subsets of trials:one in which attention was oriented initially to the target (leadingto the target N2pc) and the other in which attention was orientedinitially to the distractor (leading to the distractor N2pc). If thiswere the case, the ERPs averaged across all trials would notaccurately reflect the processes mediating visual search in eithersubset. With this in mind, we separated two subsets of trials on thebasis of RT: Trials with RTs falling below or above the median RTwere classified as fast-response and slow-response trials, respec-tively. This was done on the assumption that orienting attentioninitially to the target would lead to fast responses whereas orient-ing attention initially to the distractor would lead to slow re-sponses.

We compared ERPs on fast-response and slow-response trialsfor two display configurations: (a) lateral target, midline distractor,and (b) lateral distractor, midline target. The relevant statisticalanalyses are set out in Table 2. When a lateral target was pairedwith a midline distractor, a target N2pc was in evidence onfast-response trials but not on slow-response trials (Figures 2a and2b, shaded boxes). In contrast, when a lateral distractor was pairedwith a midline target, a distractor N2pc was in evidence onslow-response trials but not on fast-response trials (Figures 2c and2d, shaded boxes). There was a hint of a distractor N2pc onfast-response trials as well, but this contralateral-ipsilateral differ-ence did not approach significance (see Table 2) and was no larger

Table 1Interparticipant Averages of Median Response Times(in Milliseconds) for All Five Search-Display Configurations

Display configurationResponse time

(SE M)

1. Lateral target, midline distractor 954 (30)2. Midline target, lateral distractor 998 (29)3. Lateral target, no distractor 833 (24)4. Lateral target; contralateral distractor 919 (29)5. Lateral target, ipsilateral distractor 1,013 (28)

Table 2Results of Statistical Tests for Lateralized Event-Related Potential (ERP) Components at LateralOccipital Electrodes (PO7|PO8)

Analysis N2pc PD SPCN

Display t p t p t p

Fast-response1. T-lateral; D-midline 4.50 .0000* 1.80 .0789 6.13 .0000*

2. T-midline; D-lateral 1.22 .2314 3.73 .0006* 0.29 .77283. T-lateral; D-none 6.30 .0000* 0.31 .7557 6.85 .0000*

4. T-lateral; D-contra 2.95 .0054* 2.03 .0496* 3.58 .0009*

5. T-lateral; D-ipsi 4.60 .0000* 2.66 .0112* 6.94 .0000*

Slow-response1. T-lateral; D-midline 0.22 .8286 0.81 .4252 4.64 .0000*

2. T-midline; D-lateral 2.38 .0222* 1.96 .0578 1.77 .08493. T-lateral; D-none 3.67 .0007* 0.12 .9082 6.80 .0000*

4. T-lateral; D-contra 1.18 .2434 1.86 .0704 3.57 .0010*

5. T-lateral; D-ipsi 2.92 .0057* 0.98 .3325 5.89 .0000*

All trials1. T-lateral; D-midline 2.82 .0075* 0.68 .5016 6.73 .0000*

2. T-midline; D-lateral 2.53 .0155* 3.35 .0018* 1.17 .24863. T-lateral; D-none 5.41 .0000* 0.28 .7820 7.79 .0000*

4. T-lateral; D-contra 1.92 .0625 2.18 .0353* 4.59 .0000*

5. T-lateral; D-ipsi 3.57 .0010* 1.11 .2726 7.82 .0000*

Note. T $ target; D $ distractor. Data are from the median-split and all-trials ERPs. All t-tests were performedwith 39 degrees of freedom. ! Denotes statistically significant at the p $ .05 level.

6 MCDONALD, GREEN, JANNATI, AND DI LOLLO

than the contralateral-ipsilateral differences that occurred in theprestimulus baseline.

In the absence of a salient distractor, a target N2pc was inevidence on fast-response as well as slow-response trials (Figures2e and 2f). On fast-response trials, the target N2pc appeared to belarger when the distractor was absent (Figure 2e) than when it wason the midline (Figure 2a). The difference between the two N2pcwaves was marginally significant, t(39) $ 1.94, p $ .06. Thissuggests that attention was deployed to the location of the target onmost fast-response trials but was deployed to the location of thedistractor on a small subset of those trials. Averaging the twosubset of trials would then result in a modest reduction in target-N2pc amplitude. The small but nonsignificant distractor N2pc onfast-response trials (Figure 2c) is consistent with this conjecture.

The answer to the question of whether salient distractors elicitthe N2pc is that they do, but only on slow-response trials. Onfast-response trials, only the target elicits the N2pc. This pattern ofresults does not invalidate Hickey et al.’s (2006) finding of adistractor N2pc—indeed, a distractor N2pc was once again inevidence in the all-trials ERPs (Figure 1). Rather, what needs to be

qualified is the conclusion that attention was often deployed to thedistractor. Here, we show that attention was deployed to thedistractor on about half the trials. The finding that attention wasdeployed to the target on the remaining half of the trials isinconsistent with the salience-driven selection hypothesis.5

This raises an important question: Why did attention go to thedistractor on about half of the trials? There are at least twoplausible answers. First, participants might have deployed atten-tion to the distractor when it had been the target on the precedingtrial. Currently, there is considerable debate as to whether suchintertrial effects are bottom-up (Pinto et al., 2005; Theeuwes,2010) or top-down (e.g., Geyer, Müller, & Krummenacher, 2008).Second, participants might have deployed attention to the distrac-

5 The fact that attention was deployed to the distractor on half the trialsmight seem to argue against the contingent-capture hypothesis. It should benoted, however, that because target and distractor features varied randomlyfrom trial to trial, participants would have been unable to set themselves forany particular target feature. Therefore, the present study does not addressthe issue of contingent capture.

ipsilateral to targetcontralateral to target

Fast Response

800 ms

2 µV

Slow Response

ipsilateral to lateral singletoncontralateral to lateral singleton

d

e

i j -

a b

c

f

g h

PD

SPCNN2pc N2pc

MEDIAN-SPLIT ERPs

Figure 2. Event-related potentials (ERPs) elicited by the display configurations of interest, averaged separatelyfor fast-response and slow-response trials. The shaded boxes in Panels g and h represent the distractor and targetN2pc time windows used in Hickey et al.’s (2006) Experiment 2. The remaining shaded boxes represent the mainN2pc time window (225–325 ms).

7ELECTOPHYSIOLOGICAL EVIDENCE FOR CAPTURE

tor on trials in which top-down control was weak. fMRI evidenceconsistent with this option has been reported by Leber (2008), whofound that the strength of the pretrial fMRI signal in middle frontalgyrus predicted the magnitude of distractor interference. Specifi-cally, weak pretrial signals were associated with considerable RTinterference, whereas strong pretrial signals were associated withreduced RT interference. These results suggest that salient single-tons can interfere with search when the activity in frontal cortex islow but that observers can exert top-down control to prevent suchinterference when activity in frontal cortex is high.

Is there an N2pc flip? The N2pc flip refers to an ERPsequence in which an N2pc is observed initially contralateral toone side of a display and then contralateral to the other side. Thisflip provides evidence for sequential shifts of attention to items onopposite sides of fixation (Woodman & Luck, 1999, 2003). Hickeyet al. (2006) reported such a flip in the additional singletonparadigm, when the target and distractor were on opposite sides offixation: The initial N2pc was contralateral to the distractor and thesubsequent N2pc was contralateral to the target. Theeuwes (2010)regarded this N2pc flip as the most important ERP evidence for thesalience-driven selection hypothesis because it appeared to showthat the initial deployment of attention was aimed at the location ofthe most salient item in the display (i.e., the distractor).

Figures 1g and 1h present the ERPs elicited by displays in whichthe target and the distractor were on opposite sides. The twoshaded boxes in each of these figures correspond to the timeintervals used by Hickey et al. (2006) to test for an early distractorN2pc (220–265ms; left boxes) and a late target N2pc (275–350ms; right boxes). Consistent with the original report, the late targetN2pc was significant in the full sample (Figure 1h), t(39) $ 3.04,p $ .004. Critically, the early distractor N2pc seen in the original-sample ERPs (Figure 1g) is absent from the full-sample ERPs(Figure 1h; t ' 1). This discrepancy raises an obvious question:Why was an N2pc flip in evidence in the original sample but notin the full sample?

In answering this question, it needs to be emphasized that thepresent study is not an independent attempt at replication: it is thesame experiment as Hickey et al.’s Experiment 2, it was conductedin the same laboratory, and it included the original 14 participants.The only two differences were the inclusion of 26 additionalparticipants in the full sample and the use of a lower low-pass filtercutoff (25 Hz vs. 40 Hz). To ensure that the change in filter settingdid not eliminate the distractor N2pc, we reprocessed the ERPsusing the original 40-Hz filter and still found no distractor N2pc,t ' 1. The increase in sample size improved the signal-to-noiseratio (SNR), as evidenced in Figure 3, which shows contralateral-minus-ipsilateral difference waves for both samples. The peaks inthe prestimulus baselines highlight the residual noise in the wave-forms, while the 95% confidence intervals around the differencewaves highlight the variability across participants in each sample.Both the residual noise and the variability are substantially reducedin the full sample, relative to the original sample. Notably, thedownward peak at 250 ms in the original-sample waveforms(Figure 3a, shaded box)—which Hickey et al. regarded as thedistractor N2pc—is absent in the full-sample waveforms (Figure3b, shaded box). From this we conclude that the downward peak inthe original-sample waveforms was residual noise, not a distractorN2pc.

An alternative explanation for the contrasting N2pc-flip resultsis that there might have been important differences between the 14observers in the original sample and the 26 new observers. Forinstance, the original sample might have been skewed to lowworking memory capacity. It is known that low-capacity individ-uals are less capable of resisting attention capture by salient-but-irrelevant stimuli (Fukuda & Vogel, 2009). On this option, theparticipants in the original sample would have been more prone todeploy attention to the distractor before shifting attention to thetarget. We regard this explanation as unlikely, however, on twogrounds. First, the present study replicated the original results inall respects except for the critical N2pc flip. Second, the extantevidence for the N2pc flip is not robust: no N2pc flip was inevidence in either Hickey et al.’s (2006) Experiment 1 or a morerecent study that used a different variant of the additional singletonparadigm (Kiss, Grubert, Petersen, & Eimer, 2012). According toHickey et al., the absence of the N2pc flip in their Experiment 1suggested either that the distractor captured attention on only asubset of trials or that the distractor captured attention withouteliciting the N2pc consistently. The former conclusion is consis-tent with the median-split results outlined in the previous section,but is inconsistent with the salience-driven selection hypothesis,according to which the most salient item should capture attentionreliably.

In the previous section, we showed that the distractor N2pcoccurred only on slow-response trials and that the all-trials ERPsdid not accurately reflect the processes that mediate search ineither subset of trials. The same reasoning may apply to the N2pcflip. Specifically, the flip might have occurred only on slow-response trials. To address this possibility, we looked for the N2pcflip in the fast- and slow-response ERPs separately. Figures 2g and2h show the ERPs obtained when the target and the distractor wereon opposite sides of fixation, for fast- and slow-response trials,respectively. On slow-response trials—where an N2pc flip wouldbe most likely to occur—there was neither an early distractorN2pc, t(39) $ 1.32, p $ .19, nor a late target N2pc, t(39) $ 1.05,p $ .30 (Figure 2h, shaded boxes). The absence of either N2pcsuggests that the target and distractor N2pc waves cancelled eachother out over the slow-response trials. For this to be true, how-ever, the latencies of the two N2pc waves would have to be thesame, arguing against an early distractor N2pc. On fast-responsetrials, there was no hint of early distractor N2pc (Figure 2g, leftbox). In fact, a target N2pc was seen to emerge in this timeinterval, although it was not statistically significant, t(39) $ 1.31,

Figure 3. Contralateral-minus-ipsilateral difference waveforms (with95% confidence intervals) associated with search displays that containedtarget and distractor on opposite sides of fixation. The shaded box in eachplot represents the time at which the distractor N2pc was reported to occurin Hickey et al.’s (2006) study. (a) Results from the original sample ofparticipants (N $ 14; cf. Hickey et al., 2006, Experiment 2). (b) Resultsfrom the full sample of participants (N $ 40).

8 MCDONALD, GREEN, JANNATI, AND DI LOLLO

p $ .20. A target N2pc was observed in the late interval (Figure2g, right box), t(39) $ 3.63, p $ .001, as well as in the main N2pctime range (see Table 2). This is compelling evidence for atten-tional processing at the location of the target with no precedingattentional processing at the location of the distractor.

The explanation for the absence of an early distractor N2pcoffered above—the summation of opposite-polarity target anddistractor N2pc waves—seems at odds with the ERPs obtainedwhen one of the two singletons was presented on the verticalmeridian (Figures 2b and 2d). On those slow-response trials, thedistractor elicited an N2pc (Figure 2d), but the target did not(Figure 2b), suggesting that attention was deployed initially to thedistractor on the majority of those trials (see previous section).However, this does not mean that attention was deployed initiallyto the distractor on the majority of slow-response trials when thetarget and distractor were on opposite sides of fixation. A plausiblehypothesis is that the proportion of trials on which attention isdeployed initially to the distractor decreases as the spatial separa-tion between the two singletons is increased. This would predict asmaller distractor N2pc and less RT interference when the targetand distractor are on opposite sides of fixation (maximum spatialseparation) than when either singleton is on the vertical meridian(less separation). The latter prediction is confirmed by the RT datain Table 1: there was significantly less RT interference when thetarget and distractor were on opposite sides than when the targetwas on the vertical meridian (86 ms vs. 164 ms), t(39) $ 5.99,p ' .001, or when the distractor was on the vertical meridian (86ms vs. 121 ms), t(39) $ 3.56, p $ .001.

In brief, contrary to a critical prediction from the salience-drivenselection hypothesis (cf. Hickey et al., 2006, p. 606), there was noN2pc flip in the present study. This finding is consistent with theresults from Experiment 1 of Hickey et al.’s (2006) study. There-fore, the answer to the question of whether there is an N2pc flip inthe additional singleton paradigm appears to be “no.”

Are the fastest shifts of attention directed to the distractor?The median-split analyses outlined so far indicate that observersdeployed attention to the target on approximately half the trials and tothe distractor on the other half. Thus, there appears to have been nooverall bias to select the more salient singleton first. It might still bethe case, however, that the fastest (i.e., the shortest-latency) deploy-ments of attention were aimed at the location of the distractor, as hasbeen reported by Hickey et al. (2010). Had this been the case, thedistractor N2pc that was in evidence on slow-response trials shouldhave occurred earlier than the target N2pc on fast-response trials. TheERPs illustrated in Figure 2 show no evidence for any such differencein N2pc latency, but in order to evaluate this possibility, we isolatedthe N2pc by subtracting ipsilateral ERPs from corresponding con-tralateral ERPs and then measured the latencies of the N2pc in theresulting difference waveforms. This latency analysis was restricted todisplays that contained one lateral singleton and one midline single-ton, resulting in the four difference waveforms illustrated in eachpanel of Figure 4 (corresponding to ERPs illustrated in Figures2a–2d). These difference waveforms are plotted so that upward de-flections in the 200–400 ms time range correspond to N2pc (anddownward deflections in the 300–400 ms range to PD; see subsequentsection). The plots in Figure 4 include a relatively long prestimulusinterval in order to gauge the N2pc relative to the residual noisepresent in the baseline.

Consistent with the statistical analyses of N2pc amplitudes, thetarget N2pc was clearly larger than the baseline noise peaks onfast-response trials but not on slow-response trials, while theopposite was true for the distractor N2pc (Figure 4). Consequently,we focused the N2pc latency analysis on the peaks that werestatistically significant and clearly stood out from the noise: thetarget/fast-response and distractor/slow-response pairings. Criti-cally, these N2pc waves appeared to have the same onset latency.This was confirmed by defining the onset latency as the time atwhich the voltage reached 70% of peak amplitude and by usingjackknife procedures to test for statistical differences betweenpairings (see Method; cf. Kiesel et al., 2008). The latencies of thetarget/fast-response N2pc (240 ms) and distractor/slow-responseN2pc (249 ms) were statistically indistinguishable, t ' 1. Thus,there is no evidence to support Hickey et al.’s (2010) claim that thecovert deployment of attention is initially driven by the physicalsalience of the stimuli.

The results of this N2pc latency analysis are at odds with thoseof Hickey et al. (2010), who reported that the onset of thedistractor/slow-response N2pc preceded that of the target/fast-response N2pc by about 35 ms. We considered whether the pro-cedures for separating fast- and slow-response trials in the twostudies was responsible for this discrepancy. In particular, sinceHickey et al. examined target and distractor N2pc waves on thefastest and slowest quartiles, we considered whether we failed toadequately isolate trials on which attentional selection was drivenby stimulus salience (i.e., the slowest quartile rather than theslower half). At the outset, we chose to use a median split rather

a. median split

b. quartile split

target, fasttarget, slow

distractor, fastdistractor, slow

-1 µV

0 400 ms

PD

N2pc

N2pc

PD

Figure 4. Contralateral-minus-ipsilateral difference waveforms obtainedfrom the target/fast-response event-related potentials (ERPs; Figure 2a)and the distractor/slow-response ERPs (Figure 2d). (a) Results of themedian-split analysis. (b) Results of a follow-up quartile-split analysis.

9ELECTOPHYSIOLOGICAL EVIDENCE FOR CAPTURE

than a quartile split so that twice as many trials would contributeto each ERP waveform, thereby increasing the SNR. Following themedian-split analysis, however, we performed a quartile-split anal-ysis to increase comparability with the Hickey et al. (2010) study.As expected, the resulting contralateral-ipsilateral differencewaves were noisier, as evidenced by the increased amplitude of theresidual noise peaks in the prestimulus baseline (Figure 4b). Thetarget N2pc on the fastest quartile was considerably larger than itwas on the fast-response trials in the median-split analysis, t(39) $3.34 p $ .002. The distractor N2pc on the slowest quartile nolonger stood out from the residual noise (".53 !V vs. ".54 !V,respectively; methods similar to those detailed in Footnote 2).Despite the increased noise present in the quartile-split ERPs, theN2pc latency analysis revealed similar findings: The N2pc onsetlatencies were 236 ms and 239 ms for the target on fast-responsetrials and the distractor on slow-response trials, respectively. This3-ms difference was not significant, t ' 1.

The follow-up quartile-split analysis rules out the possibilitythat we missed an early distractor N2pc in the median-split pro-cedure. Obviously, some other factor is responsible for the dis-crepancy between the results of the present study and those of theHickey et al. (2010) study. Three potential factors come to mind.First, the early distractor N2pc reported by Hickey et al. (on theslowest quartile) may have been because of the summation of anN2pc wave (i.e., the signal) and residual alpha activity (i.e., noise),leading to a spurious early peak. Second, the data that Hickey et al.(2010) reanalyzed came from Hickey et al.’s (2006) Experiment 1,which was acknowledged to have a methodological confound thatmight have led participants to seek out the distractor in a voluntarymanner (target and distractor never appeared on the same side offixation). Third, participants in Experiment 1 of the Hickey et al.(2006) study responded more quickly than did the participants inthe present study. For example, on distractor-absent trials, themean RT in Hickey et al.’s (2006) Experiment 1 was 245 msshorter than in the same condition of the present study. It ispossible that the fast-acting participants were given more practiceand as a result approached the search task in a different way.

Bearing in mind that this is a null result, the answer to thequestion of whether the fastest shifts of attention are directed to thedistractor appears to be “no.”

Beyond the N2pc. In addition to the N2pc, two other lateral-ized components were clearly visible in the all-trials ERPs (Figure1): (a) a PD contralateral to the distractor in the 300–400 ms timerange; (b) an SPCN contralateral to the target in the 400–800 mstime range.

In the all-trials ERPs, the PD was statistically significant only forthe configuration that contained a lateral distractor and midlinetarget (Table 2; Figures 1c and 1d). This is consistent with thenotion that the PD represents distractor suppression (Hickey et al.,2009). Although not discussed by Hickey et al. (2006), a substan-tial PD was evident in Figure 3b of their study (see also the presentFigure 1c). A similar PD was in evidence in a more recent study(Kiss et al., 2012). The SPCN, on the other hand, was statisticallysignificant for each display containing a lateral target but wasabsent when the display contained a lateral distractor and midlinetarget. This is consistent with the idea that the SPCN representsactivity related to the transfer and maintenance of task-relevantinformation into visual working memory (Jolicœur et al., 2008).The absence of a distractor SPCN indicates that the distractor did

not gain access to visual working memory, likely because it wassuppressed (as evidenced by the PD).

The results of a median-split analysis were in line with theseconclusions. When the distractor was absent, a lateral target elic-ited the SPCN on both fast- and slow-response trials (Figures 2eand 2f). A similar pattern was observed when a midline distractoraccompanied the lateral target (Figures 2a and 2b) and when thetwo singletons were on the same side of fixation (Figures 2i and2j), although in these cases the SPCN began later on slow-responsetrials than on fast-response trials. This delay, coupled with theabsence of target N2pc and presence of distractor N2pc (Figures2b and 2d), provides compelling evidence that on slow-responsetrials, attention was oriented initially to the distractor, thus delay-ing selection of the target.

The PD was larger on fast-response trials (0.82 !V) than onslow-response trials (0.33 !V), t(39) $ 2.52, p $ .016, althoughit was marginally significant on slow-response trials as well (p $.0578; see Table 2). This pattern of results is consistent withHickey et al.’s (2009) suggestion that the PD is an index ofdistractor suppression. On this idea, suppression of the irrelevantsingleton would bias the competition in favor of the target, therebyresulting in faster responses. Furthermore, it appears that thesuppression indexed by the PD helps prevent the distractor fromaccessing subsequent processing stages involved in working mem-ory. This conclusion is based on three findings. First, the targetelicited successive lateralized selection negativities—the N2pc andSPCN—with no hint of PD. On the evidence linking the SPCN toworking memory processes (e.g., Jolicœur et al., 2008), this se-quence suggests that the target was selected and then transferred toworking memory even on slow-response trials. Second, the dis-tractor elicited a PD and no SPCN on fast-response trials, suggest-ing that information about the distractor was prevented fromaccessing working memory. Third, the distractor elicited an N2pcand a marginally significant SPCN (p $ .0849; Table 2) onslow-response trials, when the PD was significantly reduced. Thus,on some of these trials, it appears that the suppression was insuf-ficient to prevent the distractor from accessing working memory.

Summary and Conclusions

The purpose of the present study was to evaluate three ERPfindings that have been regarded as key pieces of evidence for thesalience-driven selection hypothesis: an N2pc elicited by a salientdistractor (Hickey et al., 2006), a temporal N2pc flip from thelocation of the distractor to that of the target (Hickey et al., 2006),and an early distractor N2pc on slow-response trials (Hickey et al.,2010). We questioned these findings on two grounds. First, be-cause of the relatively small sample sizes of the previous studies,the signal-to-noise ratio of the ERP waveforms was low, making itdifficult to reach unambiguous conclusions about the small ERPeffects (Woodman, 2010). Second, the first two findings werebased on ERPs averaged across all trials, making it impossible toknow whether the distractor elicited N2pc on most trials, as wouldbe predicted by the salience-driven selection hypothesis. We cir-cumvented these problems by using a larger sample (that includedthe participants in Hickey et al.‘s, 2006, Experiment 2) and byanalyzing separately the ERPs obtained in subsets of “like” tri-als—that is, fast- and slow-response trials.

10 MCDONALD, GREEN, JANNATI, AND DI LOLLO

The results obtained with the improved methodology stand incontrast to those obtained in the previous studies (Hickey et al.,2006, 2010). First, although the distractor N2pc was present onslow-response trials, it was notably absent on fast-response trials.This indicates that, on approximately half the trials, attentionalprocessing occurred at the location of the target with no earlierattentional processing at the location of the distractor. This isinconsistent with the salience-driven selection hypothesis. In fact,on fast-response trials, the distractor elicited a PD – an index ofattentional suppression—but no subsequent SPCN. On thestrength of these findings, we hypothesize that the suppressionindexed by the PD prevents irrelevant information from accessingworking memory. Second, no N2pc flip was observed when thetarget and distractor were on opposite sides of fixation; on thesetrials, only the target elicited the N2pc (Figure 2). The absence ofan initial distractor N2pc is consistent with the results of Hickey etal.’s (2006) Experiment 1 and of Kiss et al.’s (2012) recent study.We conclude that the N2pc flip reported by Hickey et al. (2006)was because of noise. Third, the distractor N2pc observed onslow-response trials was no earlier than the target N2pc observedon fast-response trials. This finding runs counter to the claim thatthe fastest shifts of attention are aimed at the location of thedistractor.

On the other hand, it has been suggested that the absence of adistractor N2pc does not rule out an initial deployment of attentionto the location of the distractor (Theeuwes, 2010). In the context ofthe present study, one may argue that, on fast-response trials,attention was deployed initially to the location of the distractor, butwas disengaged rapidly without triggering an N2pc or any otherERP index of selective distractor processing. This fleeting-captureaccount is difficult to falsify because it presupposes that attentionalselection may occur without leaving any ERP trace. In an attemptto document such a trace, we reasoned that shifting attention to adistractor, even though fleetingly, should delay the deployment ofattention to the target. We examined this possibility by comparingthe latency of the target N2pc when the distractor was on themidline (Figure 1a) to that of the target N2pc when the distractorwas absent (Figure 1f). If attention was deployed initially to thedistractor, the target N2pc should occur later when the displaycontained a distractor. In fact, the target N2pc was no later whenthe distractor was on the midline (240 ms) than when it was absent(243 ms), t ' 1. In other words, if the distractor captured attentionfleetingly on fast-response trials, it must have done so withoutaffecting the time required to deploy attention to the target.

As it stands, then, there is no ERP evidence for the salience-driven selection hypothesis.

References

Boehler, C. N., Tsotsos, J. K., Schoenfeld, M. A., Heinze, H. J., & Hopf,J.-M. (2011). Neural mechanisms of surround attenuation and distractorcompetition in visual search. The Journal of Neuroscience, 31, 5213–5224. doi:10.1523/JNEUROSCI.6406-10.2011

Eimer, M. (1996). The N2pc component as an indicator of attentionalselectivity. Electroencephalography & Clinical Neurophysiology, 99,225–234. doi:10.1016/0013-4694(96)95711-9

Eimer, M., & Kiss, M. (2010). The top-down control of visual selectionand how it is linked to the N2pc component. Acta Psychologica, 135,100–102. doi:10.1016/j.actpsy.2010.04.010

Folk, C. L., & Remington, R. (1998). Selectivity in distraction by irrelevantfeatural singletons: Evidence for two forms of attentional capture. Jour-nal of Experimental Psychology: Human Perception and Performance,24, 847–858. doi:10.1037/0096-1523.24.3.847

Folk, C. L., & Remington, R. (1999). Can new objects override attentionalcontrol settings? Perception & Psychophysics, 61, 727–739. doi:10.3758/BF03205541

Folk, C. L., Remington, R. W., & Johnston, J. C. (1992). Involuntarycovert orienting is contingent on attentional control settings. Journal ofExperimental Psychology: Human Perception and Performance, 18,1030–1044. doi:10.1037/0096-1523.18.4.1030

Folk, C. L., Remington, R. W., & Wright, J. H. (1994). The structure ofattentional control: Contingent attentional capture by apparent motion,abrupt onset, and colour. Journal of Experimental Psychology: HumanPerception and Performance, 20, 317–329. doi:10.1037/0096-1523.20.2.317

Fukuda, K., & Vogel, E. K. (2009). Human variation in overriding atten-tional capture. The Journal of Neuroscience, 29, 8726–8733. doi:10.1523/JNEUROSCI.2145-09.2009

Geyer, T., Müller, H. J., & Krummenacher, J. (2008). Expectancies mod-ulate attentional capture by salient color singletons. Vision Research, 48,1315–1326. doi:10.1016/j.visres.2008.02.006

Green, J. J., Conder, J. A., & McDonald, J. J. (2008). Lateralized frontalactivity elicited by attention-directing visual and auditory cues. Psycho-physiology, 45, 579–587.

Hickey, C., & Di Lollo, V., & McDonald, J. J. (2009). Electrophysiologicalindices of target and distractor processing in visual search. Journal ofCognitive Neuroscience, 21, 760–775. doi:10.1162/jocn.2009.21039

Hickey, C., McDonald, J. J., & Theeuwes, J. (2006). Electrophysiologicalevidence of the capture of attention. Journal of Cognitive Neuroscience,18, 604–613. doi:10.1162/jocn.2006.18.4.604

Hickey, C., van Zoest, W., & Theeuwes, J. (2010). The time course ofexogenous and endogenous control of covert attention. ExperimentalBrain Research, 201, 789–796. doi:10.1007/s00221-009-2094-9

Hopf, J. M., Boelmans, K., Schoenfeld, A. M., Heinze, H. J., & Luck, S. J.(2002). How does attention attenuate target-distractor interference invision? Evidence from magnetoencephalographic recordings. CognitiveBrain Research, 15, 17–29. doi:10.1016/S0926-6410(02)00213-6

Jolicœur, P., Brisson, B., & Robitaille, N. (2008). Dissociation of the N2pcand sustained posterior contralateral negativity in a choice response task.Brain Research, 1215, 160–172. doi:10.1016/j.brainres.2008.03.059

Kahneman, D., Treisman, A., & Burkell, J. (1983). The cost of visualfiltering. Journal of Experimental Psychology: Human Perception &Performance, 9, 510–522. doi:10.1037/0096-1523.9.4.510

Kiesel, A., Miller, J., Jolicœur, P., & Brisson, B. (2008). Measurement ofERP latency differences: A comparison of single-participant andjackknife-based scoring methods. Psychophysiology, 45, 250–274. doi:10.1111/j.1469-8986.2007.00618.x

Kiss, M., Grubert, A., Petersen, A., & Eimer, M. (2012). Attentionalcapture by salient distractors during visual search is determined bytemporal task demands. Journal of Cognitive Neuroscience, 24, 749–759. doi:10.1162/jocn_a_00127

LaBerge, D. (2002). Attentional control: Brief and prolonged. Psycholog-ical Research, 66, 220–233. doi:10.1007/s00426-002-0097-2

Lamy, D., & Yashar, A. (2008). Intertrial target-feature changes do notlead to more distraction by singletons: Target uncertainty does. VisionResearch, 48, 1274–1279. doi:10.1016/j.visres.2008.02.021

Leber, A. B. (2008). Neural predictors of within-subject fluctuations inattentional control. Journal of Neuroscience, 30, 11458–11465. doi:10.1523/JNEUROSCI.0809-10.2010

Luck, S. J. (2005). An introduction to the event-related potential technique.Cambridge, MA: The MIT Press.

Luck, S. J., Girelli, M., McDermott, M. T., & Ford, M. A. (1997). Bridgingthe gap between monkey neurophysiology and human perception: An

11ELECTOPHYSIOLOGICAL EVIDENCE FOR CAPTURE

ambiguity resolution theory of visual selective attention. Cognitive Psy-chology, 33, 64–87. doi:10.1006/cogp.1997.0660

Luck, S. J., & Hillyard, S. A. (1994a). Electrophysiological correlates offeature analysis during visual search. Psychophysiology, 31, 291–308.doi:10.1111/j.1469-8986.1994.tb02218.x

Luck, S. J., & Hillyard, S. A. (1994b). Spatial filtering during visualsearch: Evidence from human electrophysiology. Journal of Experimen-tal Psychology: Human Perception and Performance, 20, 1000–1014.doi:10.1037/0096-1523.20.5.1000

Pinto, Y., Olivers, C. N. L., & Theeuwes, J. (2005). Target uncertainty doesnot lead to more distraction by singletons: Intertrial priming does.Perception & Psychophysics, 67, 1354–1361. doi:10.3758/BF03193640

Theeuwes, J. (1991). Cross-dimensional perceptual selectivity. Perception& Psychophysics, 50, 184–193. doi:10.3758/BF03212219

Theeuwes, J. (1992). Perceptual selectivity for colour and form. Perception& Psychophysics, 51, 599–606. doi:10.3758/BF03211656

Theeuwes, J. (2010). Top–down and bottom–up control of visual selec-tion. Acta Psychologica, 135, 77–99. doi:10.1016/j.actpsy.2010.02.006

Theeuwes, J., Atchley, P., & Kramer, A. F. (2000). On the time course oftop-down and bottom-up control of visual attention. In S. Monsell & J.Driver (Eds.), Attention & performance (Vol. 18, pp.105–125).Cambridge, MA: MIT Press.

Ulrich, R., & Miller, J. O. (2001). Using the jackknife-based scoringmethod for measuring LRP onset effects in factorial designs. Psycho-physiology, 38, 816–827. doi:10.1111/1469-8986.3850816

Woodman, G. F. (2010). A brief introduction to the use of event-relatedpotentials in studies of perception and attention. Attention, Perception, &Psychophysics, 72, 2031–2046.

Woodman, G. F., & Luck, S. J. (1999). Electrophysiological measurementof rapid shifts of attention during visual search. Nature, 400, 867–869.doi:10.1038/23698

Woodman, G. F., & Luck, S. J. (2003). Serial deployment of attentionduring visual search. Journal of Experimental Psychology: Human Per-ception and Performance, 29, 121–138. doi:10.1037/0096-1523.29.1.121

Received April 16, 2012Revision received June 18, 2012

Accepted July 23, 2012 !

12 MCDONALD, GREEN, JANNATI, AND DI LOLLO