Lavie Perceptual Load Selective Attn JEPHPP95

8/12/2019 Lavie Perceptual Load Selective Attn JEPHPP95

1/18

Journal of Experimental Psychology:Human Perception and Performance1995, Vol. 21, No. 3, 451-468

Copyright 1995 by the American Psychological Association, Inc.0096-1523/95/S3.00

Perceptual Load as a Necessary Condition for Selective Attention

Nilli LavieMedical Research Council-Applied Psychology Unit

Th e early an d late selection debate may be resolved i f perceptual load o f relevant informationdetermines the selective processing of irrelevant information. This hypothesis was tested in3 studies; all used a variation of the response competition paradigm to measure irrelevantprocessing when load in the relevant processing was varied. Perceptual load was manipulatedby relevant display set size or by different processing requirements fo r identical displays.These included the requirement to process conjunctions versus isolated features and therequirement to perform simple detection of a character's presence versus difficult identifi-cation of its size and position. Distractors' interference was found only under low-loadconditions. Because t he distractor w as usually clearly distinct from th e target, it is concludedthat physical separation is not a sufficient condition fo r selective perception; overloadingperception is also required. This allows a compromise between early and late selection viewsand resolves apparent discrepancies in previous work.

W hile no one disputes the importance of selective mech-anism s in menta l processing, the locus of selection in thesequence of processing from perception to action remains tobe resolved (see, for example, Francolini & Egeth, 1980;Johnston & Dark, 1982; Lambert, 1985; Pashler, 1984). Theearly selection approach, originally proposed by Broadbent(1958) and developed further by Treisman (1969; Treisman& Geffen, 1967), claims that perception is a limited processthat requires selective attention to proceed. Consequently,attentional selection occurs early, after the rudimentaryanalysis of physical features, which are used to distinguishbetween selected and nonselected stimuli. As a result, un-attended stimuli are not fully perceived. By contrast, t he lateselection approach advanced by Deutch and Deutch (1963,1967) and Norman (1968) assumes that perception is anunlimited process, which can be performed in an automaticparallel fashion without need fo r selection. Selection ac -cording to this approach occurs only late in the process,after full perception, in order to provide the relevantresponse.

Although the debate on the locus of selection has stimu-lated a great deal of empirical work, this research ha ssucceeded only in moving the pendulum of the debate fromone side to the other. While initial studies offered supportfor the early selection approach (see, for example, Neisser,1969; Snyder, 1972; Sperling, 1960; Treisman & Riley,1969; Von Wright, 1970), from the late seventies onwardthe consensus shifted toward the late selection view (e.g.,Keele & Neill, 1978; LaBerge, 1975; Logan, 1988; Miller,1987; Posner & Snyder, 1975). As a result, the question ofthe locus of selection came to be restated as whether early

This research was supported in part by the Miller Institute fo rBasic Research in Science, University of California, Berkeley. Iam indebted to Jon Driver, Anne Treisman, an d Jehoshua Tsal fortheir most helpful advice.

Correspondence concerning this article should be addressed toNilli Lavie, Medical Research Council-Applied Psychology Unit,15 Chaucer Road, Cambridge CB2 2EF United Kingdom.

selection was possible at all (e.g., Miller, 1991; Yantis &Johnston, 1990).

Kahneman and Treisman (1984) suggested that thischange in emphasis from early to late selection was theresult of a paradigmatic shift within the field of attention.They claimed that the experimental situations that charac-terized the pioneering studies of attention (e.g., Cherry,1953; Sperling, 1960; Treisman & Geffen, 1967) weretypically m ore complex tha n those of modern research (e.g.,Keele & Neill, 1978; N eely , 1977; Posner, 1980; Posne r,Snyder, & Davidson, 1980; Posner, Nissen, & Ogden, 1978;Schneider & Shiffrin, 1977). They presented a list of dif-ferences between the filtering p aradigm (characterizing theearly research) and the selective set paradigm (characteriz-ing the modern research) an d argued that the marked dif-ferences between these tw o paradigms may lead to theoperation of different attentional mechanisms, thus, pre-cluding any meaningful generalization regarding the locusof selection from one paradigm to the other.

In this article I examine further Kahneman an d Treis-man's (1984) suggestions by considering the role of per-ceptual load in selective attention tasks. I suggest that per-ceptual load is the major determinant of the occurrence ofearly or late selection and that consideration of this factorcan resolve the apparent discrepancies between previousstudies of the locus of selection.

Perceptual Load as a Determinant of theLocus of Selection

The assumption of some limitation or bottleneck inprocessing is crucial in the early and late selection ap-proaches, as it is this limitation that is thought to producethe requirement fo r selection. The dispute is over the locusof this bottleneck in the sequence of information processing.However, even though Broadbent's (1958) classic filtermodel had a limited capacity channel as its central con-struct, the emphasis in subseq uent research has tended to be

451


2/18

452 NILLI LA VIE

on the role of physical distinctions between relevant andirrelevant information rather than o n information load (e.g.,Duncan, 1981,1984; Hu m phr eys, 1981; Nissen , 1985; Sn y-der, 1972; Tsal & Lavie, 1988, 1993). Clear physical dis-tinction, however, proved to be insufficient fo r selectiveprocessing as there were numerous demonstrations of se-mantic effects from distractors that were clearly distinctfrom the target by means of location, color, or size, or acombination of these attributes (e.g., Eriksen & Schultz,1979; Gatti & Egeth, 1978; Hagenaar & van der Heijden,1986).

Accordingly, I suggest that physical distinction is a nec-essary but not a sufficient cond ition fo r selective processing.Clear physical distinctions allow relevant stimuli to bereadily differentiated from irrelevant stimuli so that only th eformer are responded to . However, this differentiation doesnot in itself en tail that the irrelevant items will be excludedfrom full perceptual processing. Th e cause fo r such exclu-sion is not physical distinction per se but overload of theperceptual system. Once the capacity limit is exceeded,selection of the information to be processed will berequired.1

I suggest that it is only if the latter condition is met thatselective processing will occur. Thus, according to thisaccount, early selection is both the inevitable outcome ofallocating attention from a limited pool (e.g., Kahneman,1973; Navon & Gopher, 1979) and impossible to achievewhen the capacity is not exceeded. When th e relevant stim-uli do not demand all of the available attentional capacity,irrelevant stimuli will unintentionally capture spare capac-ity, consequently enabling their processing. Hence, a sub-stantial relevant information load constitutes a necessarycondition fo r early selection.

This view proposes a compromise between the early andlate selection approaches because it combines the assump-t ion (of the early selection approach) that perception is alimited process with the assumption (of the late selectionapproach) that perception is an automatic process to theextent that there remains available capacity. Perception isautomatic in this approach not in the sense that it does notrequire attention but in the sense that it is not subject tocomplete voluntary control (e.g., Jon ides, 1981; Lo gan,1988; Posner & Snyder, 1975). Perceptual processing islimited, but it proceeds automatically until the mechanismruns out of capacity. Hence, selection on this view is thenatural consequence of allocating attention. To account fo rselective attention, we may just assume that voluntary con-trol is restricted to determining priorities in the allocation ofattention between relevant and irrelevant information(Bundesen, 1987, 1990; Yantis & John son, 1990; Y antis &Jones, 1991). However, any spare capacity beyond thattaken by the high-priority relevant stimuli is automaticallyallocated to the irrelevant stimuli. Perceivers cannot reduceth e amount of attention paid by inhibiting th e allocation ofattention. Such active inhibition may be restricted to pro-cesses that are later than perception 2 (e.g., Low e, 1985;Neely, 1977; Neill, 1977; Neill & W estberr y, 1987; Tipper,1985; Tipper, MacQueen, & Brehaut , 1988).

This proposal allows more predictability for selectiveattention tasks: Conditions of load will determine wh etherperceptual processing will be selective or not, only loadedprocesses will be selective. Conditions of clear physicaldistinction between th e relevant and irrelevant stimuli willdetermine only whe ther the selection w ill be the appropriateone.

Previous Research Reconsidered in Terms ofPerceptual Load

The approach developed above explains why most re-search conducted since the late seventies has supported thelate selection view; th e reason is that th e experimentalsituations involved low perceptual load, as defined by smalldisplay set size. By using variations of the Stroop task,Eriksen and Eriksen (1974), Gatti and Egeth (1978), Hage-naar and van der Heijden (1986), Kahneman and Henik(1981), Keren, O'Hara, and Skelton (1977), Miller (1987),Paquet and Lortie (1990), and others have found that irrel-evant flankers are often identified. These studies used dis-plays with usually n o more than tw o different items: a targetand a distractor. Research into th e negative priming phe-nomenon has been conducted under analogous conditions(e.g., Allport, Tipper, & Ch miel, 1985; T ipper, 1985; Tipper& Cranston, 1985) and established similar results. Thepresent claim is that all of these studies showed failure ofearly selection because they were conducted in situations oflow load, which did not require early selection. As dis-cussed previously, selection does not occur until capacity isexceeded. A more detailed account and analysis of contem-porary results in the light of perceptual load is given inanother article (Lavie & Tsal, 1994).

Manipulation of P erceptual Load in SelectiveAttention Tasks

A similar proposal to the present one was made by Treis-m an (1969). She suggested that the nervou s system is forcedto use whatever discriminative systems it has available,unless these are already fully occupied with other tests or

1 Note that the phrase exceeding the capacity limit need not

imply that perceptual selection operates on the basis of a threshold

point of transition from early to late selection. A gradual dilutionin the distractor processing may be charac teristic of load effectsinstead. The present study was not designed to investigate thisissue but rather to establish initially that there is strong depende ncebetween load of relevant information and selective perception ofirrelevant info rm ation . Thus, I focused only on low and high levelsof load and did not include levels that seemed intermediate.

2 Perhaps it should be clarified at this point that perception asused here follows the conventional usage in the early versus lateselection debate, namely referring to processes that lead to stim-ulus identification. Elaborative semantic activation, memory, re-sponse selection, and response execution are conceived as post-perceptual processes from this perspective. See Pashler (1989) andPashler and Johnston (1989) for discussion of distinctions betweenperception, in this sense, and later processes.


3/18

PERCEPTUAL LOAD AND SELECTIVE ATTENTION 453

inputs so we tend to use our perceptual capa city to the fullon whatever sense data reach the receptors (Treisman,1969, p. 296).

However, few attempts have been made to manipulateperceptual load in focused-attention tasks in connectionwith the question of early selection. One exception is apaper by Kahneman and Chajczyk (1983), who varied per-ceptual load in a version of the Stroop color task. Theyfound that the compatibility effect of an irrelevant word wasmarkedly reduced when either a second neutral word oreven an array of xs was added to the display. Kahnem an andChajczyk concluded that th e addition of an irrelevant neu-tral item reduced the resources available for processing ofthe incompatible distractor. This conclusion reflects theo-retical reasoning similar to the present account, namely thatirrelevant processing is limited and load dependent. How-ever, the interpretation of their findings in terms of percep-tual load is not unequivocal. In the displays used in Kahne-man and Chajczyk experiments, th e addition of a neutralirrelevant object not only increased perceptual load but mayalso have reduced the perceptual saliency of the irrelevantdistractor. A single distractor in the periphery may havemore power to attract attention than one of two peripheraldistractors (Jonides, 1981). Similar results established undersimilar conditions were found by Dark, Johnston, Myles-Worsley, and Farah (1985).

Tw o recent studies searched for the optimal situation fo rthe occurrence of early selection. The first manipulated thedistinctiveness of the relevant stimuli by using the cuingparadigm. Yantis and Johnston (1990) measured identityeffects of irrelevant items between valid and invalid cuesthat indicated target position in circular arrays loaded with

eight different letters. When the target was validly cued,early selection was indeed obtained, as defined by theminimal effects from the identity of other letters in thedisplay. However, it is difficult to reach firm conclusions onthe role of perceptual load in this study as load was notvaried; rather, a single high-load condition was used. In afurther study, Johnston an d Yantis (1990) m anipulated th eperceptual load of the display in a similar cuing paradigm tothe study described above. They demonstrated that cuingthe position of a target eliminated compatibility effects fo rirrelevant letters only under th e condition of high load(circular array of eight letters); under the low-load condition(array of two letters), the cue was not effective in reducing

the processing of the distractor. They concluded that clut-tered display s we re im porta nt for selection. This conclu-sion is in accordance with the load hypothesis advancedhere. However, it remains possible that embedding thecritical distractor letter in a compact circle w ith m any othersprimarily had the effect of reducing its perceptual saliencyor orienting pull.

Miller (1991) more directly investigated the effect ofperceptual load by comparing th e compatibility effects offlanking distractors on response to a central target embed-ded in displays of two, four, or eight items. M iller found thatthe compatibility of the flanker affected response times onlyunder conditions of low perceptual load, w ith a display sizeof tw o items. Th e elimination of the flanker compatibility

effect cannot be attributed in this study to its reducedperceptual saliency with large display sizes because theletters were added only in a central circular array compris-ing the target and additional nontargets used to manipulateload, while the critical flanker appeared outside this circlean d was very large.

All of these previously mentioned studies manipulatedperceptual load by increasing display size. As a result, theyusually involved substantial changes in the appearance ofthe displays, leading to difficulty in interpretation. Th epurpose of the present study was to achieve a more gen eralan d clearer conclusion concerning the role of perceptualload in the processing of irrelevant information .

General Method

In the interest of generality, in the following experimentsI used several different manipulations of load, with the

anticipation that they would al l result in the elimination ofdistractor interference in high-load displays, thus providingconverging evidence for the influence o f perceptual load onirrelevant processing. In all of the experiments I used theEriksen paradigm (e.g., Eriksen & Eriksen, 1974), whichhas been accepted as diagnostic of early or late selection,with the innovation that load in the relevant processing wasmanipulated. The participants were required to make achoice response to the identity of a target letter. This targetalways appeared in the central region of the display. Acritical distractor that could be incompatible, neutral, orcompatible in relation to the target response was located farfrom th e target, above or below th e center. Reaction times(RTs) to the target were measured a s a function of the natureof th e critical distractor and the load of the relevant pro-cessing in the task.

All the manipulations of load used in the present studychanged th e nature of the task in the center, witho ut touch-ing the peripheral distractor. The aim was to show that loadin the processing of the relevant inform ation determines thedegree of processing of the irrelevant information and thatthis determination is not dependent on any change in theperceptual saliency of the irrelevant information (see earlierdiscussion in the Manipulation of Perceptual Load in Se-lective A ttention Tasks section.

Th e critical distractor was always located relatively farfrom the target letter to allow a test of my claim that clearphysical distinctions between relevant an d irrelevant infor-mation are not sufficient alone fo r early selection to occur.This claim implies that even when a clear distinction be-tween target an d distractor exists, whether or not earlyselection takes place will depend mainly on the perceptualload of the task an d will obtain only when load is high.

Operationalizing Perceptual Load

Perceptual load, though a term often used in attentionresearch, is an abstract, if not vague, concept. The aim of thepresent research was not to investigate any specific formu-lation of load but to rely on conventional operational defi-


4/18

454 NILLI LA V IE

nitions. Because with the current hypothesis early selectionrequires load in perceptual processing, it seemed impo rtantto choose operational definitions that were more likely toload perceptual routines rather than postperceptual ones. 3

Thus, in Experiment 1, the relevant display set size of i temsamo ng which the target appeared w as m anipu lated (i.e., thenumber of items among which the target appeared). InExperiments 2A, 2B, and 3, I kept th e high- an d low-loaddisplays physically the same but varied the processing thatwas required. All of the displays in the latter studies con-tained an additional shape situated near th e target letter inthe center, and the choice response to the letter was depen-dent on different requirements for the perceptual processingof this additional shape's properties. In Experiments 2A and2B, the shape was colored, and load was m anipu lated inaccordance with th e feature integration theory of attention(e.g., Treisman & G elade, 1980; Treisman & Sato , 1990;Treisman & Schmidt, 1982). E ither the conjunction of thisshape's color and form or just its color had to be processed

in addition to the target letter to decide which responseshould be ma de to the target. In Experime nt 2B I tested thehypothesis that physical distinction is a necessary albeit notsufficient condition fo r selective processing by comparingth e effect of a near distractor (within 1 with the target) an da fa r distractor (more than 1 of separation from th e target)under high and low relevant loads.

In Experiment 3 I tested the load hypothesis by using thetraditional distinction between detection and identificationas an additional load manipulation (e.g, Bonnel, Possama-mai, & Schmitt, 1987; Bonnel, Stein, & Be rtucc i, 1992;Graham, 1989; Sagi & Julesz, 1985; Uttal, 1987). Partici-pants were required to detect the presence of either a circle

or a bar shape under the low-load condition and to identifythe size an d exact position of the circle and the bar shapeunder th e high-load condition.

In all of these exp eriments I tested th e hyp othesis that th ecompatibility effects from th e critical distractor would varyas a function of the perceptual load conditions of the rele-vant processing in each task. The prediction was that th eidentity of the critical distractor would affect the RTs to therelevant target only under conditions of low load in thecentral task; namely u nder conditions of small display size,simple detect ion, and s ingle-feature processing ratherthan high display size, f ine discr iminat ion, and conjunc-tion processing. U n d e r th e low-load conditions, th e clear

physical dis t inct ion betwe en targe t and cr i tical dis t ractorshould not suffice for selective processing, which shouldonly be observed when there was also a high load inrelevant processing.

In al l cases, th e processing of the distractor was measuredby comparing conditions in which its identity was compat-ible, incom patible, or neu tral with respect to the appropriateresponse for the current target letter. Incompatible distrac-tors would be expected to produce interference relative tothe neutral baseline when they are identified because ofresponse competition. The predictions for compatible dis-tractors were less clear. Because our com patible distractorswere a Jwa ys identical to the target, their identification couldbe based on an ear ly an d presemantic detection of the

physical match to the target. This stage of processing maynot require full identification and may not, therefore, indi-cate true late selection. Moreover, it is unclear what effectson RTs should be predicted when th e distractor is identicalto the target. Matching of the physical features may causeboth interference because of feature-specific inhibition (cf.Bjork & M ur ra y, 1977; E stes, 1972; S antee & Eg eth, 1982)an d fa cilitation because of redundancy gain effects on signalactivation, response selection, or both (e.g., Eriksen & Erik-sen, 1979; Eriksen, Gottel, St. James, & Fournier, 1989;Flowers & W ilcox, 1982; G rice, Can ham , & Gwynne 1984;Miller, 1982; Santee & Egeth, 1982). Fortunately, theseproblems of interpretation do not apply to the incomp atibleconditions in which interference as a result of distractoridentification can unam biguously be predicted. Accord-ingly, my conclusions are primarily based on the effects ofincompatible distractors as compared with neutral distrac-tors. Compatible distractors had to be included to preventan y predictable relation between distractors from the re-sponse set and the pa rticular target shown because any suchcorrelations are kno wn to affect perform ance (M iller, 1987).

Experiment 1

The traditional way to increase task load is by h avinglarger display sizes (see Duncan, 1980; Johnsen & Briggs,1973; Kerr, 1973; Navon, 1989). Displays with a largernum ber of items involve a direct increase in the am ount ofthe information presented. The purpose of Experiment 1was to investigate the effect of an irrelevant distractor in aversion of the Eriksen paradigm (Eriksen & Erik sen, 1974)

that involved manipulation of the display se t size for thetarget task. Each display contained a target and a criticalirrelevant distractor. Th e distractor could be compatible,incompatible, or neutral in relation to the target response,an d it was ph ysically distinct from the targe t by being largerin size an d situated in remote an d irrelevant positions. Th evariable of perceptual load had two levels: lo w perceptualload consisted of relevant display size one (Condition SI),and high perceptual load consisted of relevant display sizesix (Condition S6). In Condition SI, th e target appeared inone out of six possible positions, whereas the other fivepositions were empty. I n Condition S6, the five other posi-tions were occupied by five neutral letter nontargets (with

3 Manipulations of memory load, and of number of alternativesfor response, were not chosen for this reason. It is possible toadvance a more general version of the load hypothesis that wouldalso apply to processes that ar e considered to be later than per-ception. For example, the extent of semantic activation and ofmemory for irrelevant information may also depend on load inrelevant processing at these levels. It seemed, however, morerelevant for the early versus late selection debate to establish firstthe load hypothesis for the process of shape identification. Notealso that a load manipulation such as increasing th e number ofalternatives might weaken th e stimulus-response associations inthe task an d thus reduce distractor compatibility effects withoutnecessarily affecting distractor perception.


5/18


no associated response in the experiment). Examples of thedisplays are given in Figure 1.

The hypothesis of this experiment w as that the load in therelevant task would determine the ability to ignore theirrelevant distractor. Under Condition SI, the proce ssing ofthe target should leave spare capacity, which will be unin-tentionally allocated for the processing of the irrelevantdistractor. Under Condition S6, searching for the targetamong five nontargets should load attentional capacity an dleave no spare capacity for the irrelevant processing. There-fore, interference from the critical distractor was expectedunder SI and was not expected under S6.

Method

Participants. The participants were 14 undergraduates fromthe University of California, Berkeley, who participated to fulfill acourse requirement. All had normal or corrected-to-normal vision.

Apparatus and stimuli. An IBM PC compatible computer at-

tached to a VGA color monitor presented the stimuli and recordedthe response latencies. The software used for creating and runningthe experiments was Micro Experimental Laboratory (MEL), soldby Psychology Software Tools, Inc. (Schneider, 1988).

A target letter, which could be either x or z, appeared randomlyand equiprobable at one of six positions arranged in a row, locatedat the center of the display. The target letter appeared alone inCondition SI or was accompanied by five nontarget letters (k, s, m,v, and n ) in Condition S6. These five nontargets occupied the otherpositions in the central row equally often in a random order. Thetarget and nontarget letters were presented in lowercase, and at a

viewing distance of 60 cm, they subtended a visual angle of 0.70vertically and 0.4 horizontally and were separated by 0.65 fromnearest edge to edge. A critical distractor of a larger size, subtend-ing a visual angle of 0.96 vertically and 0.48 horizontally,appeared randomly and equiprobable either above or below thecenter. The distance between the distractor edge and the fixation

point was 1.9 so that the separation between the distractor and thecentral row letters (from edge to edge) varied from 1.40 of dis-tance for the two central letters to 2.10 and 2.90 of separationfor the two intermediate and two end letters, respectively. All ofthe letters were presented in a light grey color (No. 7 in theMEL color palette) on a black background. The critical distrac-to r was equally likely to be incompatible (the capital letter Xwhen the target letter was z, or vice versa), compatible (the cap-ital letter Z when the target letter was z, or likewise for x), or itcould be neutral in relation to the targets (the capital letter P,which had no defined response in the experiment). Each of thedistractor categories appeared equally often with each of the tar-get positions. Seventy-two displays were created according tothese specifications.

Procedure. Before each display, a light grey fixation dot ap-peared at the center of the display for 1 s. This was immediatelyreplaced by the target display, which appeared for approximately50 ms. Participants used the numerical keys on the right-hand sideof the computer keyboard for their responses. They were requiredto press the zero key with their thumb for the target letter z and thenumber two key with their index finger for the target letter x as fastas they could while avoiding errors. Feedback on errors was givenby a 500-ms computer tone. Participants were also emphaticallyinstructed to ignore the irrelevant distractor and were informedabout the occasional presence of incompatible distractors and their

Center

Lo w load Relevant Se t size 1

Target position:

Intermediate End

z

X z

Z

Z

x

Incompatible Neutral Compatible Incompatible Neutral Compatible

High load Relevant set size 6

Target position:

Center Intermediate End

mvznsk

P

ksmx nv vzskmn

Z

Z

skvnxm zmvnks

nkmsvx

Incompatible Neutra l Compatible Incompatible Neutra l Compatible

Figure 1. Examples

of stimuli used

in Experiment

1.


6/18

456 NILLI LA VIE

possible detr imental effect on performance if they were not ig-nored. Th e SI and S6 conditions appeared in separate blocks inalternate fashion. Half of the participants got the S1-S6 ord er, a ndhalf of the participants got the reverse S6-S1 order. Each partic-ipant passed through 10 blocks of 72 trials each. The first 2 blockswere discarded as practice and their results were excluded fromanalysis. An intermission of 5 min was given after th e first 6blocks. T he participants initiated each block by pressing th e spacebar. The order of the trials within blocks was randomized with norepetitions.

Results and Discussion

Median RTs were computed for each participant as afunction of the display set size and the nature of the criticaldistractor (compatible, neutral, or incompatible). Responselatencies longer than 2 s were excluded from analysis. Th eaverages across participants are presented in Table 1 to-gether with the associated error rates.

A two-way within-subject analysis o f variance (ANOVA;Set Size X Distractor Type) on the RT data showed a maineffect of the set size, F(l, 13) = 144, p < .001, confirmingthat perceptual load was effectively manipulated. To revealthe possible interference and facilitation effects of the dis-tractor identity, I conducted pairwise comparisons of in-compatible with neutral distractors and compatible withneutral distractors by an ANOVA. As noted previously, mymain concern was with the possible interference effects ofincompatible distractors relative to the neutral conditionbecause only these unambiguously specify the level ofprocessing of the distractor.

In the analysis of incompatible versus neutral conditions,set size interacted significantly with the type of distractor,F(l, 13) = 5.93, p < .03, in accordance with the loadhypothesis. The interference effect from the incompatibledistractor was significant only under the low-load conditionof SI, F(l, 13) = 7.62, p < .02 (F < 1, in the S6 condition).

In th e separate comparison of identical distractors withth e neutral baseline, the identical distractor produced only a

Table 1The Intersubject Means of Reaction Times (inMilliseconds), Standard Errors, and Percentage of Errorsas a Function of Distractor Compatibility and Task Load

Distractor identity

Task load N I N N C

LowMSE% E

HighMSE% E

50122

4.1 a

613195.4

461 40 a

131.9

609 422

2.9

45211

1.5

594174.4

9 a

15

Note. I = incompatible; N = neutral ; C = compatible; % E =percentage of error.a Indicates significant effects of the pairwise com parisons w ith theneutral conditions.

main effect of fac ilitation, F(l, 13) = 7.38, p < .02, whichdid not interact with the set size (F < 1). How ever, thefacilitation effect was more consistently obtained under th eSI condition, as it was on ly significant fo r this, F(l, 13) =8.68, p < .01, (p > .10, under S6).

Analysis of errors. A two-way ANOVA (Set Size XDistractor Type) showed a main effect of load on thenumber of errors, F(l, 13) = 10.52, p < .006.

Pairwise comparisons of the incompatible and of thecompatible distractor with the neutral baseline showed thatth e interference effect from incom patible distractors on thepercentage of errors was obtained only under the low-loadcondition, F(l, 13) = 9.66, p < .01. There were no o the rsignificant effects in the analysis of errors (p > .10 in all ofth e other comparisons).

E ff e c t of target positions on distractor processing. Th etarget letter could appear in six different positions. Therewere tw o central positions: two end positions and twoin termediate positions. The central targets were situated0.53 from f ixation, the intermediate and end targets were1.6 and 2.65, respectively, from fixation. The variable oftarget positions also involved a difference in separationbetween target an d distractor, which varied from 1.3, 2.1,and 2.9 from edge to edge for the central, intermediate, andend positions, respectively. A three-way ANOVA w as con-ducted on the variables of target position, load, and distrac-tor type for incompatible versus neutral distractors. 4 Thethree-way interaction between the variables of target posi-tion, load, and distractor type did not reach significance(p > .10), which is why the results of the load effect ondistractor processing were reported pooled across the tar-get positions. However, target position did have a main

effect on the overall RTs F(2, 12) = 19.94, p < .001,which tend to increase as the foveality of the target de -creased. Th e overall RTs were 511, 552, and 606 for thecentral, intermediate, and end positions, respectively. Ta-ble 2 presents th e distractor interference effects as a func-tion of the target position.

A significant two-way interaction was foun d between thetarget position and the distractor effect, F(2, 12) = 17.36,p < .001. As can be seen from Table 2, the irrelevantdistractor tended to produce more interference in the case ofmore peripheral targets.

Thus, the results of the position analyses seem to be inagreement with Eriksen an d Schultz 's (1979) conclusion

that distractor effects tend to be stronger for degradedtargets. They showed that reducing the target size or con-trast increased th e interfering effects of irrelevant distrac-tors. The present results are important in showing the con-trast between th e effects of relevant information load an ddata quality. Loading the perceptual system with more in-formation enabled participants to avoid irrelevant process-ing in the present study. However, the reduction in retinal

4 A similar three-w ay AN OV A (Target Position X Load XDistractor) w as also conducted on the compatible distractor versusth e neutral baseline. Only the main effect of facilitation and theinteraction of position an d load were significant in this analysis(p > .10 in all the other comparisons).


7/18


Table 2Distractor Interference on Reaction Times (inMilliseconds) as a Function of Load an d Target Positions

Low load High load

Position N I N N I N

C e n t e r

I n t e r m e d i a t e

Edge

47519

50423

55831

453 5

46414

476 5

22

40

82

55520

62621

70529

56220

61427

68432

-7

12

21

Note. I = incompatible; N = neutral.

acuity of more peripheral targets made them more prone toirrelevant interference. Because more peripheral targets

took a longer time to be processed, it may be that th eprolonging of the relevant processing increased the chanceof distractor intrusion during the accumulated time (e.g.,Eriksen & Schultz, 1979; Na vo n, 1989).

The fact that the significant increase in overall RTs be-cause of target eccentricity had an independent effect thatwas the reverse of the load effect in the present study seemsto be in clear contrast with a claim by Miller (1991). Heclaimed that it is the higher overall RT, which is typicallyconfounded with increases in the display size, rather thanth e higher load itself that causes the elimination of distractoreffects because of a dissipation of the effect in time. Millershowed that when the distractors' appearance was delayed,

they were processed under high-load conditions. However,his stimulus onset asynchrony manipulation, intended toretard the processing of the flanker, w as confounded with i tsabrupt onset or offset, which ha s been shown to captureattention (Jonides & Yan tis, 1988; Ka hne m an, Treisman, &Burkell, 1983; Yantis & Jonides, 1984, 1990). Thus, theprocessing of the flanker could have resulted from theautomatic allocation of attention to its location.

In this study, the increase in overall RTs with moreperipheral target locations did not involve an y change in theappearance of the distractor. Under these conditions a sig-nificant increase in the overall RTs actually increased thedistractor effect rather than eliminating it, as long as anincrease in load was not involved.

Th e target-to-distractor distance w as confounded withretinal acuity in this study, preventing any clear conclusionabout its influence on the distractor processing. How ever,th e present results do demonstrate that clear physical dis-tinction between target and distractor (defined by theirspatial separation an d size) is not sufficient in itself toeliminate distractor processing. The distractor was pro-cessed under a ll conditions of target-to-distractor separationin th e low-load displays. Note that th e nearest target-to-distractor separation was still far in spotlight terms (e.g.,Eriksen & Hoffman, 1972,1973). The spotlight results wereusually obtained in crowded displays equivalent to the high-load conditions of the present study. The results obtained

under the low-load cond ition of the present study seem to bein accordance with the studies of Murphy and Eriksen(1987), Hagenaar and van der Heijden (1986), and Merikleand Gorewich (1979), which showed no effect of target-to-distractor separation on the distractor effect. Note that all ofthese studies were characterized by low-load situations.

In summary, the pattern of results of the present studygave clear support to the hypothesis that the load involvedin target processing determines the extent of processing foran irrelevant distractor. This conclusion was found to beindependent of target-to-distractor separation, the stimulusquality of the target, and the overall RTs.

Experiments 2A and 2B

The following experiments further tested the load hypoth-esis by using load manipulations that did not involve anychange in the appearance of the displays. In Experiment 1,the manipulation of the relevant set size did involve achange in the appearance of the displays. This changeshould no t have had any effect on the saliency of theperipheral distractor given its distance from th e centraltarget region. However, some confounding variables maystill have been involved. One might say, fo r example, thatth e perceptual difference between th e target and the irrele-vant distractor was more pronounced in S6. The neutralnontargets accompanying the target in the center of the S6displays might be grouped perceptually with the target byproximity and similarity in size, resulting in a strongerperceptual segregation between the target and the criticaldistractor (see Driver & B aylis, 1989; Kah nem an & Henik,

1977; Prin zm etal, 1981; Prin zm etal & Banks, 1977).In th e following experiments I manipulated th e process-in g demands for d isplays that were iden tical in their appear-ance. This was intended to emphasize the role of processingload in resource terms, with n o effect on the data conditionsof the stimuli. Th e manipulation of load was adapted fromfeature integration theory (Treisman & Gelad e, 1980; Treis-m an & Sato, 1990). According to feature integration theory,perception of features is load free, and it is only theirconjunctions that require the focusing o f attention and there-fore impose perceptual load. Treisman has used severalconverging operations that provide support fo r this claim:The phenomenon of illusory conjunctions (Treisman &

Paterson, 1984; Treisman & Schmidt, 1982) and studies ofvisual search (Treisman, 1991; Treisman & Gormican,1988; Treisman & Gelade, 1980) have generated the mostwidespread and persuasive examples. Although it has beensometimes disputed (e.g., Navon, 1990; Tsal, 1989; Wolfe,Cave, & Franzel, 1989), feature integration theory seemscurrently the best explanation for a wide range of selectiveattention studies. Thus, manipulating the requirement toprocess features versus conjunctions has the advantage ofimplementing a well-established operational formulation ofperceptual load.

I manipulated the perceptual load in the present experi-ment by requiring two different forms of processing for anadditional shape presented next to the target letter while


8/18

458 NILLI LAVIE

maintaining an identical display. The target appeared in thecenter of the display with to its right or left side a coloredshape that could either be a circle or a square, with either ablue or a red color. As in Experiment 1, the participantsmade a choice response to the target letter. A n irrelevantcritical distractor that could be compatible, neutral, or in-compatible in relation to the target response appeared aboveor below the center, as before. However, in this experimentth e response to the target w as also dependent on the colorfeature of the closely flanking shape (in the condition offeature demand ) or on the conjun ction of its shape and colorfeatures (in the conjunction demand condition). In the fea-ture demand condition participants had to make their choiceresponse to the target letter wh en the color of the additionalshape was blue and to withhold response when it was red,no m atter whether the shape w as a circle o r a square; tha t is,a go/no-go procedure contingent on the additional itemdetermined whether th e choice response should be made tothe target letter. On 25% of the trials, randomly intermixedwith the other trials, the no-go color appeared. In the con-junction demand condition, either a blue square or a redcircle had to be present for a response to be appropriate, andparticipants had to withhold response to the opposite con-junctions (i.e., a blue circle or a red square). Thus, partici-pants now had to process the specific conjunction of colorand shape for the additional item rather than simply it scolor. On 25% o f the trials, the no-go con junctions appeared(see Figure 2).

I predicted that th e level of perceptual load manipulated

by th e demand of the task would determine the degree ofprocessing of the irrelevant distractor. While the under-standing of the mechanisms of feature integration and ofw hy it poses such a demand on attention may not be clear,few would dispute that th e conjunctive processing of twofeatures in addition to the target letter should be moredeman ding than processing a single color feature. Thus, inth e feature demand condition, th e color feature should bedetected with little or no increase in the load of relevantprocessing, leaving spare capacity to spill over automati-cally to the critical distractor. Under th e conjunction de-mand condition, th e task of recognizing th e appropriateconjunctions in the very same displays should impose muchmore of a demand on attentional capacity, leaving consid-erably less for the irrelevant distractor and hence reducinginterference effects.

Experiment 2A

Method

Participants. The participants were 14 undergrad uates fromTel-Aviv U niver sity wh o participated to fulfill a course require-ment. All had normal or corrected-to-normal vision.

Apparatus and stimuli. The stimuli were presented by a four-field Gerbrands tachistoscope. A colored shape and a black targetletter were situated in the center of each display, with 0.70 ofcontour-to-contour separation between them. Each of these stimuliappeared with equal probability and in random arrangement either

Low load Color features

Go : presence of a blue color No go: presence of a red color

Incompatible Neutral Compatible Incompatible Neutral Compatible

High load Conjunctionsof colors and shapes

Go presence of eitherred circle or blue square

Incompatible Neu tral Compatible

No go: presence of eitherred square or blue circle

Incompatible Neutra l Compatible

Figure 2. Examples of stimuli used in Experiment 2A. Black = red; dotted texture = blue.


9/18


to th e left or to the right side o f fixation. The target letter w as eitherthe capital letter H or U, and it subtended a visual angle of 0.36vertically an d 0.25 horizontally. An irrelevant distractor of alarger size, subtending a visual angle of 0.50 vertically and 0.36horizontally, appeared randomly and equiprobably either above orbelow the center, separated by 1.30 of visual angle from th e

nearest contour of the central stimuli. The distractor letter wasequally likely to be incompatible (the letter U when the target letterwas H, or vice versa), compatible (the letter H when the targetletter was H, or likewise for Us), or it could be neutral in relationto the targets (the letter X, which had no defined response in theexperiment). In the feature demand condition, th e color of theshape was blue on 75% of the trials or red on 25% of the trials. Foreach of the colors, the shape could equally probably be a square ora circle. In the conjunction deman d condition, either a blue squareor a red circle appeared on 75% of the trials, and either a red squ areor a blue circle appeared on 25% of the trials. The order of thecolors, shapes, their locations, and each of their combinations wasrandom with the constraint that none w as repeated more than threeconsecutive times.

Procedure. Before each display a black fixation point appearedat the center of the display for 1 s. This was immediately replacedby the target display that appeared for 100 ms. The participantswere required to press one of two buttons as fast as they could ongo trials while avoiding errors. Half of the participants were askedto respond with their dom inant hand to the target letter H and withth e nondominant hand to the target letter U (the other ha lf receivedthe opposite instruction). Participants were also emphatically in-structed to ignore th e irrelevant distractor. Th e conditions of taskdemand were arranged in blocks consisting of 48 display cards.Each condition was run in one session consisting of three repeti-tions of the condition block. An intermission of 5 m in separatedthe two sessions. The order of the sessions was counterbalancedbetween subjects. In addition to the choice RT instructions, par-ticipants were requested to pay attention to the color of the shape

(in the feature demand condition) or to the relevant conjunctions ofcolor and shape (in the conjunction demand condition) and torespond to the target only when the appropriate feature or con-junctions appeared. This instruction was emphasized and partici-pants received feedback on their errors. Each session began with24 practice trials, except for the participants who got the conjunc-tion demand condition in the first session, fo r which they received48 practice trials.


The results of Experiment 2A are discussed with theresults of Experiment 2B .

Experiment 2B

In Experiment 2B I tested whether selective processingunder h igh-load conditions is dependent on the spatial sep-aration between the target and distractor and can only beobtained with far distractors. Spotlight studies have foundthat selective perception did not occur fo r items that werewithin an area of 1 from the target (e.g., Eriksen & Hoff-man, 1972, 1973). These results were obtained under situ-ations of high load as defined by set size. Thus, it may bethat a minimal physical separation is required for elimina-tion of distractor interference in addition to high load. Inother words, it may be that physical separation between

stimuli is a necessary condition for selective attention tooccur, even though it is not sufficient without high load inthe relevant processing.

If adequate distractor separation is a necessary conditionfor selective processing, then a very near distractor shouldbe processed even when the processing load is high.

Method

Participants. Eighteen new participants from the same pool asExperiment 2A were tested in Experiment 2B. (None of theseparticipants participated in Experiment 2A.)

Stimuli and procedure. The stimuli and procedure were iden-tical to those of Experiment 2A, except for the following changes.In Experiment 2B I reversed the role of go and no-go attributes,thus participants were required to respond to the red color underth e feature dem and condition and to either a red square or a bluecircle under the conjunction demand condition.

Half of the displays included a very near distractor that w asplaced above or below fixation and was separated by 0.7 of visualangle from the nearest contour of the central stimuli an d subtended0.43 vertically and 0.29 horizontally. These displays were ran-domly mixed with th e original displays from Experiment 2A .

Each participant passed through four blocks of 48 trials for eachtask. Half th e participants performed th e feature task in the firstsession, whereas half began with the conjunction task.


The mean s of the RTs w ere computed for each participantof Experiment 2A as a function of the task demand (featuredemand or conjunction demand), distractor identity, andpresentation order, which was a between-subjects variableaccording to whether the function demand session or theconjunction demand session w as first.5 A three-way mixedANOVA conducted on these variables showed no effect ofthe presentation order and no interaction involving thisvariable (p > .10 in all cases). Thus, the presentation orderswere pooled fo r further analy ses. Table 3 shows the aver-ages of RTs and error rates across participants and orders.

To increase the statistical pow er, I included in the currentanalyses the conditions with the far distractor fo r bothfeature demand and conjunction demand tasks of Exp eri-ment 2B, which were equivalent to those of Experiment 2A .A two-way AN OV A (Task Demand X Distractor Type) ofthe RT data showed a main effect of the task demand, F(l,

31) = 507.7, p < .001, demonstrating that perceptual loadwas effectively manipulated.In the analysis of the incompatible distractors versus the

neutral baseline, there was an interaction between task de -mand an d the distractor type, F(l, 31) = 3.94, p < .056. Inaccordance w ith the p redictions of the load hypothesis, theinterference effect was significant only under the low-load

5 In the following experiments I used the mean RTs for each

participant rather than th e medians, as the increase in number ofgo/no-go errors under th e high-load conditions resulted in a dif-ferent number of responses in between conditions (see Miller,1988).


10/18

460 NILLI LAVIE

Table 3Mean Reaction Times (in Milliseconds), Standard Errors,an d Percentage of Errors as a Function of DistractorCompatibility and Task Load in Experiments 2A and 2B

Distractor identity

Task load N I N N CLow

E

High

E

63315

6

95925

2 6

61715

0 7

96625

2 8

16a 61214

3

-7 994

271 9

5

-28a

Note. = 32. I = incompatible; N = neutral; C = compatible;% E = percentage of error.a Indicates significant effects of the pairwise comparisons with theneutral conditions.

condition of feature demand, F(l, 31) = 6.8, p < .01, (F .10 in all of the comparisons), thus, the presentation orderswere pooled for further analyses. A two-way ANOVA onth e RT data showed a main effect of task demand, F(l,17) = 162.75, p < .001. Pairwise analyses showed a sig-nificant interaction between the task demand and the incom-patible distractor interference, F(l, 31) = 4.55, p < .048.The interference effect was significant only in the low-loaddetection task, F(l, 17) = 19.39, p < .001, (F < 1, in the

discrimination task). This result clearly supports the loadhypothesis.

The compatible distractor did not have any significanteffects under any of the load conditions (p > .10). However,th e opposed direction of the trends under the two loadconditions (i.e., facilitation under the high load and inter-ference under low load) was significant, F(l, 17) = 5.03,p < .04. Further inspection of the individual data revealedthat the trend of facilitation under the high load was largelydue to the slowest participants of the pool. Three partici-pants were slower by more than 230 ms relative to theaverage RT, which was 924 ms (they were at 1.61.7standard deviations from the mean). Their overall RTs were1,181, 1,178, and 1,163 ms. Exclusion of these participants

from analysis eliminated completely the facilitation effectof the identical distractor, under the high load (from 31 msto 2 ms), and did not change the strength of the interferenceeffect of the incompatible distractor (which was 30 mswithout these slowest participants).

Table 4Mean Reaction Times (in Milliseconds), Standard Errors,and Percentage of Errors as a Function of DistractorCompatibility and Task Load in Experiment 3

Distractor identity

Task load

LowMSE% E

HighMSE% E

I

73734

3.1

1,15749

8.5

N I - NDetection

705 32 a

322.6

Identification

1,153 452

9.3

C

72033

2.8

1,12239

8.8

N C

15

31

Note. I = incompatible; N = neutral; C = compatible; %E =percentage of error.a Indicates significant effects of the pairwise com parisons with theneutral conditions.


13/18


Thus, the facilitation effect of compatible distractors inthe high-load task does no t seem to be a robust effect in thepresent study. The somewhat erratic effects of compatibledistractors are discussed further below. As in the previousexperiments, no such difficulties of interpretation apply tothe interference effects from incompatible distractors.

Analyses of errors. A three-way mixed ANOVA (TaskDemand X Distractor Type X O rder) showed a main effectof load on the number of errors, F(l, 13) = 10.52, p < .006.There were no other effects in the analysis of the choiceresponse errors.

The analysis of errors in the go/no-go task (i.e., failure torespond to the presence of the character in the low-loadcondition or to its appropriate position or size in the high-load conditions, plus false-alarm responses in the absence ofthe appropriate character) demonstrated the difficulty of theidentification task. The average accuracy for the go/no-gotask requiring detection was 98%, and the correspondingaccuracy of the identification task dropped to 71%. Thissuggests that the identification task overloaded capacity.

Inspection of the individual data did not show any differ-ences in the pattern of performance for the identificationtask between participants whose overall go/no-go perfor-mance w as relatively accurate or inaccurate. As a result, theonly criterion I applied to go/no-go performance was toreplace participants that were less than 50% accurate in thistask.

Discussion of the identical distractors' e ffec t . The iden-tical distractors in the far locations seemed to have minoran d inconsistent effects on the performance in the presentstudies (the near distractors are discussed separately). Therather strange trends observed under the last tw o experi-

ments are not without precedent. Previous studies haveshown that identical distractors can sometimes interferewith performance rather than facilitate it, presumably be-cause of early feature-specific inhibition (cf. B jork & M ur-ray, 1977; Estes, 1972). Other studies have shown facilita-tion effects (e.g., Eriksen & Eriksen, 1979; Flowers &Wilcox, 1982; Santee & Egeth, 1982) or no effects at all foridentical distractors (e.g., Miller, 1991; Flowers & Wilcox,1982). This inconsistency can be explained by the assump-tion that identical distractors may have tw o opposing effectsof both interference an d facilitation (perhaps at differentlevels of processing, namely an early interference effect atthe feature extraction level an d later facilitation effects on

signal, response activation, or both). These componentsmay be emphasized differently in different situations (e.g.,with different exposure times, stimulus onset asynchronies,and relative emphases on accuracy versus speed; see Flow-ers & W ilcox, 1982; S antee & Egeth, 1982). The currentdata shed little real light on these previous discrepancies.They merely show that inform ation load does not seem to bean important variable in their explanation.

Load effects on the processing of incompatible distractorsas compared with the neutral ones are easier to interpretbecause previous research ha s shown consistent interferenceeffects (see Driver & Baylis, 1989; Eriksen & Eriksen,1974; Eriksen & Hoffman, 1972; Miller, 1991). Note alsothat measuring the effect of load on distractor interference

tests a more surprising and counterintuitive prediction,namely, that easy tasks are more prone to interference fromirrelevant information than ar e more difficult ones.

General DiscussionThe results of the present study suggest that perceptual

load plays a causal role in determining the efficiency ofselective attention. The different manipulations of load con-verged to show th at interference from irrelevant distractorswas found only under conditions of low perceptual load inthe relevant processing and was eliminated under conditionsof high load. This pattern of results was observed with threedifferent manipulations of load in the target task: relevantdisplay set size of one versus six items, a demand to processcolor alone versus conjunctions of color and shape, andsimple detection of characters' presence rather than identi-fication of their exact position and size. Hence, the resultsprovide converging evidence for the concept of perceptualload and show that the ability to ignore irrelevant informa-tion is directly related to the load in the processing of therelevant information.

These conclusions derive further support from previouswork, discussed earlier, showing a decrease in flanker ef-fects in more loaded situations (Dark et al., 1985; Johnston& Yantis, 1990; Kahneman & Chajczyk, 1983; Miller,1991; Yantis & Johnston, 1990). All of these studiesshowed a reduction in the flanker effect with an increase indisplay set size. The present study generalized these find-ings by m anipulating load w ithout affecting the data quality

or salience of distracting stimuli and by relating manipula-tions in display set size to other manipulations of load.The variety o f m anipulations used allows us to generalize

the conclusion on the role of perceptual load in selectiveattention across different levels of overall RT and across theaccompanying differences in variance. Note that the overallRT for the low-load conditions in E xperiments 2 and 3 (621ms and 721 ms, respectively) w ere actually higher than theoverall R Ts of the high-load condition in Experiment 1 (605ms). Moreover, the variance of the low-load condition ofExperiment 3 (SE = 33) was higher than th e variance of thehigh-load conditions of Experiments 1 and 2 (SE = 19 andSE = 26 , respectively). Thus, taken together, the results o f

the present experiments show that the degree of interferencefrom incompatible distractors was more consistently asso-ciated with manipulations of load than with either overallRTs or statistical variance.

This point is further underlined by the detailed pattern ofresults in Experiment 1, which deconfounded load fromoverall RT by means of target eccentricity. The processingdelay caused by the reduced acuity for eccentric targetsactually had an opposite effect on distractor processing toincreases in load, even though both produced longer RTs.Distractor interference was greater for peripheral versuscentral targets but was reduced by higher load for each ofthe target positions. This pattern of results clarifies therelation between the load hypothesis and the overall RT in


14/18

464 NILLI LA V IE

a particular task. On the one hand , in the absence of strictlyquantitative definitions of perceptual load, th e increase inoverall RTs typically found under high-load conditions (ashere) serves to verify that a task indeed became more dif-ficult. On the other hand, the theoretical meaning of loadencompasses more than just delayed RT or increased dif-ficulty. The concept of load implies that the system mustcarry out further operations or must apply operations toadditional units. It is these additional demands on capac-ity, rather than mere delay in processing, that shouldblock low-priority, irrelevant items from consumingscarce capacity.

By contrast, the mere prolonging of relevant processing(e.g., by acuity limits), without imposing additional opera-tions or items, should actually increase ra ther than decreasesusceptibility to irrelevant distraction. Stimulus evidencefor th e distractor m ay gradually accum ulate over time (Erik-sen & Schultz, 1979), and the statistical chance for intrusio nfrom irrelevant stimuli (at any particular point in t ime) is

inevitably greater with an increased time window (seeNavon, 1989). The pattern of results in Experiment 1 isconsistent with these analyses. When relevant processingw as prolonged by degrading th e data quality for the target(i.e., by eccentricity), distractor interference increased. Bycontrast, loading the relevant processing eliminated thisinterference for targets at every position.

The Nature of Capacity Limits That DetermineSelective Processing

In the present framework I have used a very general

concept of perceptual capacity limits to explain when se-lective processing occurs. However, previous attempts toaccount fo r divided atten tion (rather than selective attention,as here) in terms of general capacity limits have ultimatelyha d to incorporate various structural (or content-specific)constraints as well (e.g., Allport, 1980; Kahneman, 1973;Navon & G opher, 1979; W ickens, 1980). Thus, multipleresources, and specific cross-talk conflicts w ithin hy poth et-ical network implementations, have been proposed to ex-plain a range of content-specific effects in dual-task perfor-mance (see, for example, Navon, 1984).

How do such proposals relate to the present hypothesis,which concerns the role of capacity limitations in selective

rather than divided-attention tasks? It seems that th e notionof limitation used in the present study might be consistentwith any of the above accounts, as they a ll predict increasedlimitation with an increase in the load of processing (e.g.,cross talk in pathways is more likely to arise when moreinformation is transmitted in them). Hence, the predictionfor th e effect of load on irrelevant processing remainsessentially the same for all perspectives (i.e., reduced irrel-evant processing with greater load).

Note, however, that from the specific-resource or cross-talk perspectives the load hypoth esis migh t only apply whenrelevant and irrelevant processes share a common routine.Thus, to exclude irrelevant stimuli from particular percep-tual operations, one might have to load these very same

operations wi th th e relevant stimuli. In the present experi-ments, the load manipulations were all directed at identifi-cation processes, which were presumed to be involved inboth th e re levant and the irrelevant processing. It mightpossibly be argued for an even greater specificity of load,namely, in shape discrimination. Thus, in Experiment 1,more different shapes had to be coded fo r relev ant stimuli inth e high-load condition. In Experiment 2, high load meantthat th e shape and the color of the go/no-go stimulus had tobe coded. Finally, in Experiment 3, the shape, location, andsize of the go versus no-go stimulus rather than its merepresence had to be processed in the high-load condition.Thus, th e present results are at least co nsistent w ith a ratherspecific process, nam ely shape discrimination , being subjectto capacity limits that lead to selective processing underhigh load.

Alternatively, it might be argued that both specific an dgeneral limitations apply fo r each routine. Hum an attentionhas recently been described as a multiple system , consistingof both specialized processing routines and a central exec-utive controller that sets up and coordinates priorities andgoals (see Carr, 1992; Posner & Petersen, 1990, for recentneuropsychological evidence). On the present hypothesis,selective processing might in principle be achieved by load-in g either specific subsystems or a general-purpose execu-tive con trol system. The present study was not designed todistinguish these possibilities bu t rather to establish a gen-eral principle fo r selective processing that can be exploredin future research. At this juncture, it might be speculatedthat m y first experiment specifically loaded th e perceptualprocess of search, whereas th e second and third study m ayhave loaded executive processes, by means of a more com-

plex decision rule in the high-load condition, in addition toan y lower level changes in perceptual processing. Furtherexperiments designed to m anipulate load in one componentof processing versus another ar e required to decide suchissues.

The Relation Between Load and Physical Separationin Selective Processing

In the present study I proposed a resolution to discrepan-cies in previous findings concerning th e role of target-distractor spatial separation in selective processing. Studies

conducted in the f ramework of the spotlight metaphor havesuggested that irrelevant items that are separated by morethan 1 of visual angle from the target are not processed(e.g., Eriksen & Hoffman, 1972, 1973). However, studiesconducted in the f ramework of the late selection approach(e.g., Gatti & Egeth, 1978; Hagenaar & van der Heijden,1986; Merikle & Gorewich, 1979) have found that distrac-tors separated by considerably more than 1 from the targetwere processed.

In the present set of experiments, results tha t accord withspotlight claims were obtained under the high-load condi-tions, whereas results supporting late selection claims wereobtained un der the low-load conditions. M oreover, previousspotlight studies have typically used high-load displays,


15/18


consisting of 8 to 12 items, whereas results favoring lateselection have typically been obtained under low-load con-ditions. (A more detailed review of task load in variouscontemporary studies appears in Lavie and Tsal, 1994.)Thus, previous apparent discrepancies may now be attrib-uted to variations in perceptual load.

Furthermore, the load hypothesis may allow a clearercharacterization for the role of physical distinctions betweenrelevant an d irrelevant stimuli in determining whether pro-cessing is selective. The specific suggestion is that suchdistinctions will be more effective in ensuring the appropri-ate allocation of attention under situations of high load.Note that in the low-load conditions of the present experi-ments, the distractor was processed in spite of its clearseparation from the target in terms of location (and in sizeas well for some studies). Traditional late selection ap -proaches would take such results as evidence that percep-tion is not a selective process, supporting the unli mit edcapacity view. How ever, with the present account, the pro-cessing of a clearly distinct distractor under low-load con-ditions leads to the conclusion that physical distinctionalone is not sufficient to ensure selective processing. Highload in the relevant processing is also required.

The facilitation effect from the very near compatibledistractor under the high-load condition of Experiment 2Blends some support to the suggestion that physical separa-tion plays a necessary if not a sufficient role in allowingselective processing. However, interpretation of these re -sults is com plicated by the inconsistency of the compatibledistractor effects when they were presented far from th etarget. It should also be noted that in Exp eriment 2B I usedonly one man ipulation of the physical distinction between

target an d distractor (namely, their location), which may nothave been the strongest means for linking the distractorwith the target items (see, for example, Driver & Baylis,1989; Kramer & Jacobson, 1991). A more definitive con-clusion as to whether clear physical separation is a neces-sary condition for selection under conditions of high loadshould await more extensive study of the interaction be-tween these variables.

Elimination of Interference and the Locusof Selection

Th e present study suggests a compromise between tradi-tional early and late selection m odels because it show ed thatlate selection results are found under the conditions of lowload and early selection results under conditions of highload. In the present view both results are produced by thesame attentional mechanism, which allocates resources forthe processing of information until it runs out of capacity.Thus, the suggested resolution combines the assumption ofthe early selection approach that perception is a limited-resource process with the assumption of the late selectionapproach that perception is an automatic process (i.e., can-no t volun tarily be prevented) to the exten t that there remainsavailable capacity.

The current model can therefore explain selective pro-

cessing without requiring any active mechanism (e.g., inhi-bition) for rejecting irrelevant information (see Neisser,1976, for a similar view). On the present account, distractorsare simply not processed when relevan t processing con-sumes full perceptual capacity. No specific mechanism isrequired for this early selectivity beyond the intrinsic ca-pacity limitations of perception. These limitations will nec-essarily result in selective processing whenever the relevantinformation exhausts available capacity.

How would this view accommodate the evidence thatinhibitory mechanisms can be involved in attention, such asthe various phenomena of negative priming (e.g., Tipper,1985)? Perhaps active inhibition of distractor information isonly necessary when the intrinsic means of selection (i.e.,by means of capacity limits) fail to prevent distractor pro-cessing. On this view, negative priming should only befound under conditions of low load (because spare capacitywill spill over to distractors) rather than under high load(because selectivity will emerge naturally from resourcelimits). The characteristically small stimulus set size (i.e.,low load) used in typical negative priming paradigms seemsconsistent with this suggestion (see Lavie & Tsal, 1994, fora more detailed review).

Note that the prediction derived above from the loadhypothesis, namely for more negative priming under low-load conditions, is the opposite to what would follow fromconventional accounts of negative priming that posit inhi-bition as the prim ary mean s of selection. C onsider thepresent experiments, which found greater selectivity (i.e.,less distractor interference) under high-load conditions. Ifinhibition were the m eans of this selectivity, there sho uld bemore negative priming under high load, corresponding to

the greater selectivity. Such a finding would imply that thereduction in interference I have observed under high loadwas caused by a later selectivity than the one I have ar-gued for. It would show that the noninterfering irrelevantdistractor was actually processed under high load andthen inhibited (see Driver & Tipper, 1989). More defini-tive conclusions on the exact nature of selection awaitfurther indexes of processing (e.g., priming in addition tointerference measures).

For the moment, I can conclude that high task load playsa role in selective attention by prohibiting interference thatwould otherwise arise. This seems to be an important con-clusion in itself, from both theoretical and practical aspects.

Obviously, the practical implications of finding the condi-tions for avoiding distraction from irrelevant informationare important. Paradoxically, my data suggest that moredifficult tasks (i.e., those with high load) can be performedbetter in this sense (i.e., show less interference). Theoreti-cally, the load hypothesis combines tasks of focused anddivided attention under the same explanation. On thepresent view, performance unde r focused-attention tasks isno t essentially different from performance under divided-attention tasks (i.e., spare capacity from the primary rele-vant perception task was allocated to the secondary irrele-vant perception task). The focused-attention tasks of thepresent study differed from divided-attention tasks only inthat participants were strongly requested to ignore the irrel-


16/18

466 NILLI LA VIE

evant stimuli. Yet, they succeeded in rejecting irrelevantinformation only when the relevant task load was high.Thus, it seems tha t perceivers can only regulate order ofpriority in the allocation of attention (Yantis & Johnson,1990; Yantis & Jones, 1991). However, from this point on,processing is entirely determinate . Whether selective pro-cessing will occur is at the mercy of the perceptual loadimposed by external events

References

Allport, D. A. (1980). Patterns an d actions: Cognitive mechanismsare content-specific. In G. Claxtone (Ed.), Cognitive psychol-ogy: Ne w directions (pp. 26-64). London: Routledge & KeganPaul.

Allport, D. A ., Tipper, S. P., & Chm iel, N. J. C. (1985). Perceptualintegration and post categorical filtering. In M. I. Posner & O. S.Marin (Eds.), Attention and performance XL Hillsdale, NJ:Erlbaum.

Bjork, E. L., & Mu rray, J. T. (1977). On the nature of inputchannels in visual processing. Psychological Review, 84 , 472-484.

Bonnel, A., Possamamai, C. A., & Schmitt, M. (1987). Earlymodulations of visual input: A study of attentional strategies.The Quarterly Journal of Experimental Psychology, 39A, 757-776.

Bonnel, A. M., Stein, J. F., & Bertucci, P. (1992). Does attentionmodulate the perception of luminance changes? The QuarterlyJournal of Experimental Psychology, 44, 601-626.

Broadbent, D. E. (1958). Perception an d communication. London:Pergamon Press.

Bundesen, C. (1987). Visual attention: Race models for selectionfrom multielement displays. Psychological Research, 49 , 113-121.

Bundesen, C. (1990). A theory of visual attention. PsychologicalReview, 97 , 523-547.

Carr, T. H. (1992). Automaticity an d cognitive anatomy: Is wordrecognition autom atic ? American Journal of Psychology, 105,201-237.

Cherry, E. C. (1953). Some experiments on the recognition ofspeech with one and with two ears. Journal of the AcousticalSociety of America, 25 , 975-979.

Dark, V. J. , Johnston, W. A ., Myles-Worsley, M., & Farah, M. J.(1985). Levels of selection and capacity limits. Journal of Ex-perimental Psychology: General, 114, 472-497.

Deutch, J. A ., & Deutch, D. (1963). Attention: Some theoreticalconsiderations. Psychological Review, 70 , 80-90.

Deutch, J. A ., & Deutch, D. (1967). Comments on selective

attention: Perception or response? The Quarterly Journal ofExperimental Psychology, 19, 362-363.

Driver, J ., & Bay lis, G. C. (1989). Movement and visual attention:The spotlight metaphor breaks down. Journal of ExperimentalPsychology: Human Perception an d Performance, 15, 448-456.

Driver, J., & Tipper, S. P. (1989). On the nonselectivity of selec-tive seeing: Contrasts between interference and priming inselective attention. Journal of Experimental Psychology: HumanPerception an d Performance, 15, 304-314.

Duncan, J. (1980). The locus of interference in the perception ofsimultaneous stimuli. Psychological Review, 87 , 272-300.

Duncan, J. (1981). Direction attention in the visual field. Percep-tion Psychophysics, 33 , 2028.

Duncan, J. (1984). Selective attention and the organization of

visual information. Journal of Experimental Psychology: Gen-eral, 113, 501-517.

Duncan, J., & Humphreys, G. W. (1989). Visual search and stim-ulus similarity. Psychological Review, 96, 433-458.

Eriksen, B. A ., & Eriksen, C. W. (1974). Effects of noise lettersupon the identification of a target letter in a non search task.

Perception Psychophysics, 16 , 143-149.Eriksen, C. W ., & Eriksen, B. A. (1979). Target redundancy invisual search: Do repetitions of target within the display impairprocessing? Perception Psychophysics, 26 , 195-205.

Eriksen, C. W., G ottel, B., St. James, J. D ., & Fournier, L. R.(1989). Processing redundant signals: Coactivation, divided at-tention, or what? Perception Psychophysics, 45, 356-370.

Eriksen, C. W., & Hoffman, J. E. (1972). Temporal and spatialcharacteristics of selective encoding from visual displays. Per-ception Psychophysics, 12, 201-204.

Eriksen, C. W., & Hoffman, J. E. (1973). The extent of processingof noise elements during selective encoding from visual dis-plays. Perception Psychophysics, 14 , 155-160.

Eriksen, C. W., & Schultz, D. W. (1979). Information processingin visual search: A continuous flow conception and experimentalresults. Perception Psychophysics, 25 , 249-263.

Eriksen, C. W ., & St. James, J. D. (1986). Visual attention withinan d around th e field of focal attention: A zoom lens model.Perception Psychophysics, 40 , 225-240.

Eriksen, C. W ., & Yeh, Y. (1985). Allocation of attention in thevisual field. Journal of Experimental Psychology: Human Per-ception and Performance, 11, 583-597.

Estes, W. K. (1972). Interactions of signal and background vari-ables in visual processing. Perception Psychophysics, 12 ,278-286.

Flowers, J. H ., & Wilcox, N. (1982). Th e effect of flanking contexton visual classification: The joint contribution o f interactions atdifferent processing levels. Perception Psychophysics, 32,581-591.

Francolini, C. M., & Egeth, H. E. (1980). On the nonautom aticityof autom atic activation : Evidence of selective seeing. Percep-tion Psychophysics, 27 , 331-342.

Gatti, S. V., & Egeth, H. E. (1978). Failure of spatial selectivity invision. Bulletin of the Psychonomic Society, 11 , 181-184.

Graham, N. V. S. (1989). Visual pattern analyzers. Ne w York:Oxford University Press.

Grice, G. R., Canham, L., & Gwynne, J. W. (1984). Absence of aredundant signals effect in reaction time task with divided at -tention. Perception Psychophysics, 36, 565-570.

Hagenaar, R., & van der Heijden, A. H. C. (1986). Target-noiseseparation in visual selective attention. Acta Psychologica, 62,161-176.

Humphreys, G. (1981). Flexibility of attention between stimulus

dimensions. Perception Psychophysics, 30 , 291-302.Johnsen, A. M., & Briggs, G. E. (1973). On the locus of displayload effects on choice reactions. Journal of Experimental Psy-chology, 99, 266-271.

Johnston, W., & Dark, V. (1982). In defense of intraperceptualtheories of attention. Journal of Experimental Psychology: Hu-man Perception an d Performance, 8, 407-421.

Johnston, J., & Yantis, S. (1990, November). Failure to processunattended o bjects: The search for b oundary cond itions. Paperpresented at the 31st ann ual meeting of the Psychonomic Soci-ety, New Orleans, LA.

Jonides, J. (1981). Voluntary versus automatic control over themind's ey e movement. In J. B. Long & A. D. Baddeley (Eds.),Attention and performa nce IX (pp. 187-204). Hillsdale, NJ:Erlbaum.


17/18


Jonides, J., & Yantis, S. (1988). Uniqueness of abrupt visual onsetas an attention-capturing p roperty. Perception Psychophysics,43 , 346-354.

Kahneman, D. (1973). Attention and e ffor t . Englewood Cliffs, NJ:Prentice Hall.

Kahneman, D., & Chajczyk, D. (1983). Tests of the automaticity

of reading: Dilution of Stroop effects b y color-irrelevant stimuli.Journal of Experimental Psychology: Human Perception an dPerformance, 9 , 497-509.

Kahneman, D., & Henik, A. (1977). Effects of visual grouping onimmediate recall an d selective attention. In S. Domic (Ed.),Attention and performance VI (pp. 307-322) Hillsdale, NJ:Erlbaum.

Kahneman, D., & Henik, A. (1981). Perceptual organization andattention. In M. Kubovy & J. R. Pomerantz (Eds.), Perceptualorganization (pp. 181-211). Hillsdale, NJ: Erlbaum.

Kahneman, D., & Treisman, A. (1984). Changing views of atten-tion an d autom aticity. In R. Parasuraman & D. R. Davies (Eds.),Varieties of attention (pp. 29-61). New York: Academic Press.

Kahneman, D., Treisman, A., & Burkell, J. (1983). Th e cost of

visual filtering. Journal of Experimental Psychology: HumanPerception and Performance, 9, 510-522.Keele, S. W., & Neill, W. T. (1978). M echanisms of attention. In

E. C. Carterette & M. P. Friedman (Eds.), Handbook of percep-tion (Vol. 9, pp. 3-47). New York: Academic Press.

Keren, G., O'Hara, W. P., & Skelton, J. M. (1977). Levels of noiseprocessing and attentional control. Journal of Experimental Psy-chology: Human Perception and Performance, 3, 653664.

Kerr, B . (1973). Processing demands during mental operations.Memory Cognition, I, 401-412.

Kramer, A. F., & Jacobson, A. (1991). Perceptual organization andfocused attention: Th e role of objects an d proximity in visualprocessing. Perception Psychophysics, 50, 267-284.

LaBerge, D. (1975). Acquisition of automatic processing in per-

ceptual an d associative learning. In P. M. A. Rabbit & S. Domic(Eds.), Attention and performance V (pp. 50-64). New York:Academic Press.

Lambert, A. J. (1985). Selectivity an d stages of processing: Anenduring controversy in attentional theory: A review. CurrentPsychology Research and Reviews, 4, 239-256.

Lavie, N., & Tsal, Y . (1994). Perceptual load as a major determi-nant of the locus of selection in visual attention. Perception Psychophysics, 56, 183-197.

Logan, G. D. (1988). Toward an instance theo ry of automatization.Psychological Review, 95, 492-527.

Lowe, D. G. (1985). Further investigations of inhibitory mecha-nisms in attention. Memory Cognition, 13, 74-80.

Merikle, P. M ., & Gorewich, N. J. (1979). Spatial selectivity invisual attention: Field size depends upon noise size. Bulletin ofthe Psychonomic Society, 14, 343-346.

Miller, J . (1982). Divided attention: Evidence fo r coactivation withredundant signals. Cognitive Psychology, 14, 247-279.

Miller, J. (1987). Priming is not necessary for selective attentionfailures: Semantic effects of unattended, unprimed letters. Per-ception Psychophysics, 41 , 419-434.

Miller, J. (1988). A warning about median reaction time. Journalof Experimental Psychology: Human Perception an d Perfor-mance, 14, 539-543.

Miller, J. (1991). T he flanker compatibility effect as a function ofvisual angle attentional focus, visual transients, and perceptualload: A search for boundary conditions. Perception Psycho-physics, 49, 270-288.

Murphy, T. D., & Eriksen, C . W. (1987). Temporal changes in the

distribution of attention in the visual field in response to precues.Perception Psychophysics, 42, 576-586.

Navon, D. (1984). ResourcesA theoretical soup stone? Psycho-logical Review, 91, 216-234.

Navon, D. (1989). The locus of attentional selection: Is it early,late, or neither? European Journal of Cognitive Psychology, 1 ,

47-68.Navon, D. (1990). Does attention serve to integrate features.Psychological Review, 97, 453-459.

Navon, D., & G opher, D. (1979). On the economy of the human-processing system. Psychological Review, 86, 214-255.

Neely, J. H. (1977). Semantic priming an d retrieval from lexicalmemory: Role of inhibitionless spreading activation an d limited-capacity attention. Journal of Experimental Psychology: Gen-eral, 106, 226-254.

Neill, W. T. (1977). Inhibitory an d facilitatory processes in atten-tion. Journal of Experimental Psychology: Human Perceptionan d Performance, 3, 444-450.

Neill, W. T., & W estberry, R. L. (1987). Selective attention andthe suppression of cognitive noise. Journal of Experimental

Psychology: Learning, Memory, an d Cognition, 13, 327-334.Neisser, U. (1969). Selective reading: A m ethod of the study ofvisual attention. Paper presented at the 19th International Con-gress of Psychology, London .

Neisser, U. (1976). Cognition and reality: Principles and impli-cations of cognitive psychology. San Francisco: Freeman.

Nissen, M. J. (1985). Accessing features an d objects: Is locationspecial? In M. I. Posner & O. S. Marin (Eds.), Attention andperformance XI (pp. 205-219). Hillsdale, NJ: Erlbaum.

Norman, D. A. (1968). Towards a theory o f memory a nd attention.Psychological Review, 75 , 522-536.

Paquet, L., & Lortie, C . (1990). Evidence for early selection:Precuing target location reduces interference from same-cate-gory distractors. Perception Psychophysics, 48 , 382-388.

Pashler, H. (1984). Evidence against late selection: Stimulus qual-ity effects in previewed displays. Journal of Experimental Psy-chology: Human Perception an d Performance, 10, 429-448.

Pashler, H. (1989). Dissociations an d dependencies between speedan d accuracy: Evidence fo

Documents

Lavie Perceptual Load Selective Attn JEPHPP95