selective attention and the organization of visual … · american psychological association, inc....

Journal of Experimental Psychology: General1984, Vol. 113, No. 4, 501-517.

Copyright 1984 by theAmerican Psychological Association, Inc.

Selective Attention and the Organization of Visual Information

John DuncanMedical Research Council, Applied Psychology Unit, Cambridge, England

Theories of visual attention deal with the limit on our ability to see (and laterreport) several things at once. These theories fall into three broad classes. Object-based theories propose a limit on the number of separate objects :that can beperceived simultaneously. Discrimination-based theories propose a limit on thenumber of separate discriminations that can be made. Space-based theories pro-pose a limit on the spatial area from which information can be taken up. Todistinguish these views, the present experiments used small (< 1 °), brief, fovealdisplays, each consisting of two overlapping objects (a box with a line struckthrough it). It was found that two judgments that concern the same object can bemade simultaneously without loss of accuracy, whereas two judgments that con-cern different objects cannot. Neither the similarity nor the difficulty of requireddiscriminations, nor the spatial distribution of information, could account for theresults. The experiments support a view in which parallel, preattentive processesserve to segment the field into separate objects, followed by a pfocess of focalattention that deals with only one object at a time. This view is also able to accountfor results taken to support both discrimination-based and space-based theories.

Object-Based Theories of Visual Attention

Theories of visual attention are concernedwith the limit on our ability to see (and laterreport) several things at once. This articledeals with what I call object-based theories(e.g., Neisser, 1967), which propose that thislimit concerns the number of separate objectsthat can be seen. Here some predictions ofthis view are tested, and object-based theoriesare contrasted with discrimination-basedtheories (e.g., Allport, 1971, 1980) and withspaced-based theories (e.g., Hoffman & Nel-son, 1981; Posner, Snyder, & Davidson,1980).

The work of Neisser (1967) illustrates theobject-based approach. Neisser (1967) pro-posed that perceptual analysis of the visualworld takes place in two successive stages.The first, preattentive, stage segments thefield into separate objects on the basis of suchGestalt properties as spatial proximity, conti-nuity of contour, shared color or movement,and so on. The second stage, focal attention,analyzes a particular object in more detail.Neisser (1967) supposed that, whereas preat-

Requests for reprints should be sent to John Duncan,Medical Research Council, Applied Psychology Unit, 15Chaucer Road, Cambridge CB2 2EF, England.

tentive processing is parallel across objectssimultaneously present in the visual field,focal attention is serial and thus is responsiblefor the limit on our ability to see several ob-jects at once. Similar positions are taken byKahneman and Henik (1981) and by Treis-man, Kahneman, and Burkell (1983),

This article is concerned with some pre-dictions following from such a theory. If sub-jects must report two aspects of a brief visualdisplay, performance should depend onwhether these aspects concern the same ordifferent objects. Reporting two aspects ofone object should be no more difficult thanreporting only one, because focal attention ispaid to the object as a whole. In contrast,reporting aspects of two different objectsshould be less successful, reflecting competi-tion between these objects for focal attention.

A relevant experiment was reported byLappin (1967). Tachistoscopic displays con-tained nine circles, each varying in size,color, and angle of an inscribed diameter. Inone condition, a single circle was cued by abar marker, and the subject was to report itssize, color, and angle. In a second condition,three circles were cued by separate barmarkers, and the subject was to report thesize of one, the color of the second, and theangle of the third. Although reports were in-deed more accurate in the single-object con-

so i

502 JOHN DUNCAN

dition, the result as it stands cannot be takenas strong support for the object-based theory.First, locating and using one bar marker maybe easier than locating and using three. Sec-ond, reporting aspects of three different ob-jects required uptake of information from awider spatial area than did reporting threeaspects of one object, thai is, the number ofobjects to be attended was confounded withthe spatial distribution of information.Third, in the different-objects case, the ruledescribing which property to report fromwhich object was necessarily complex, andusing it to extract and to remember only therelevant information from the display musthave been difficult.

A second line of evidence sometimes takento support the object-based theory concernseffects of visual grouping. Identifying severalstimuli in a complex display is improved ifthey are made to form a "strong" perceptualgroup. For example, Treisman et al. (1983)presented displays consisting of a word andan outline box. Identifying the word and theposition of a break in the box's contour si-multaneously was better when the box sur-rounded the word than when it appeared onthe opposite side of fixation, even though thespatial separation between the word and criti-cal break in the box's contour was equal inthe two cases. Merikle (1980) showed thatreporting the letters in one row or column ofa 4X4 matrix is better when they aregrouped by common color than when colorgrouping produces a conflicting organiza-tion. Similar effects of spatial grouping havebeen reported by Fryklund (1975) and byKahneman and Henik (1977, 1981).

To what extent do these results support theobject-based theory? Certainly the theorydepends on the importance of perceptualgrouping processes, because it is these thatpreattentively define what is treated as oneobject: And in the displays of Treisman et al.(1983), for example, it might be sensible tosuggest that the box arid the word were seenmore as parts of "one object" when the boxsurrounded the word than when it did not.On the other hand, one may hesitate to saythat a box surrounding a word or that twoletters grouped by common color are attri-butes of one object in quite the same way as,for example, a letter's color and shape. The

relation between these cases is dealt withlater: For the moment, it seems best to leaveopen the precise relevance of grouping phe-nomena to the object-based theory.

It should be noted that in the context of thistheory, the concept of the object serves onlyas an approximation to a construct whoseprecise details must be filled^in empirically.The idea is that Gestalt grouping processesserve to package information preattentivelyinto chunks, which then serve as units forfocal attention. As a first approximation,such chunks should correspond to our intu-itions concerning discrete objects; if they donot, the validity of the object-based approachwill be seriously in doubt. But the precisedetails of preattentive information packagingmust ultimately be an empirical matter. If theobject-based theory is correct; then the studyof visual attention and of perceptual organi-zation must proceed together.

Other Theories

Object-based theories propose that thelimit on our ability to deal simultaneouslywith several sources of visual informationconcerns the number of separate objects thatcan be seen. In contrast, discrimination-based theories propose a limit on the numberof separate discriminations that can be made,while space-based theories propose a limit onthe spatial area from which information canbe picked up. >

It will be shown that these; three types oftheories are not mutually exclusive. At thesame time, evidence taken to support dis-crimination-based and space-abased views isnot entirely compelling.

Discrimination-Based Theories

One discrimination-based theory was pro-posed by Allport (1971). Perceptual analysisis supposed to take place in systems of ana-lyzers, each dealing with a particular stimulusattribute such as color or forta. If two dis-criminations involve different analyzers,they may be undertaken simultaneouslywithout mutual interference; but when bothinvolve the same analyzer (e.g., two form dis-criminations), they may not.

This proposal is not a logical alternative to

SELECTIVE ATTENTION AND VISUAL ORGANIZATION 503

an object-based theory. Limits on our abilityto see several things at once may reflect bothdifficulties in dealing with several objects anddifficulties in making several similar discrim-inations. Nevertheless, one may ask whetherthe major empirical support for the analyzertheory—provided in experiments by Allport(1971) and Wing and Allport (1972)—distinguishes the two views. Allport (1971)presented tachistoscopic displays consistingof three colored outline shapes, each contain-ing a small black numeral. In three dividedattention conditions, subjects reported colorsand shapes, colors and numerals, or shapesand numerals. From the analyzer theory,Allport (1971) predicted poor performancewhen reporting shapes and numerals becauseboth required form analysis. The object-based theory, in contrast, predicts poor per-formance when reporting either shapes andnumerals or colors and numerals becauseboth involved attention to different objects.The results were unclear. As predicted byboth theories, performance when reportingcolors and shapes was as good as perform-ance in single-report control conditions (re-porting only colors or only shapes). Also, aspredicted by both theories, performance waspoor when reporting shapes and numerals. Inthe crucial condition, report of colors andnumerals, control levels of performance werenot quite attained, though the loss was muchless dramatic than for report of shapes andnumerals.1 No firm conclusion seems possi-ble. In a second experiment Wing and All-port (1972) presented circular patches ofgrating made up of parallel straight lines. Thegrating varied in spatial frequency and in ori-entation, and additionally, a break gave theimpression of a white band of different orien-tation, lying across the grating patch. As pre-dicted by the analyzer theory, no loss fromcontrol performance occurred when subjectsreported both grating frequency and gratingorientation, while a substantial loss occurredwhen subjects reported grating orientationand break orientation. In this case, however,the object-based theory predicts exactly thesame results, assuming that the display wasseen as two objects (a white bar lying acrossthe background of the grating). The condi-tion producing differential predictions—report of grating frequency and break

orientation—was not run. As things stand,evidence for the analyzer theory does notseem especially strong.

Different discrimination-based theoriesare possible. One proposal may be that allperceptual discriminations call for resourcesfrom some common pool, perhaps in pro-portion to their difficulty. If total resourcesare insufficient, then simultaneous discrimi-nations interfere. This would be a naturalextension of early accounts of selectiveattention based on information theory(Broadbent, 1958) and resembles, for exam-ple, the view of Moray (1967). Here, similar-ity between simultaneous discriminationsmight be less important than their difficulty,but again, the discrimination rather than theobject would be the appropriate unit for theanalysis of attention.

Space-Based Theories

Mental spotlight theories (Eriksen & Hoff-man, 1973; Hoffman & Nelson, 1981;Posneret al., 1980) propose that at any given mo-ment, attention is focused on a particulararea of visual space, and only stimuli withinthis area receive full perceptual analysis.Sometimes it is proposed that the spotlighthas a radius of about 1 ° (Eriksen & Hoffman,1973). Or it may be suggested that as with azoom lens, the beam can be made larger atthe cost of a loss in definition (Treisman &Schmidt, 1982). But in any case, the limit onour ability to see several things at once is seenas a limit on the spatial area from which in-formation can be picked up.

Again, space-based and object-basedtheories are not mutually exclusive. For ex-ample, focused attention in vision could beachieved by the combined action of two sys-tems, one selecting a particular area of spaceand the other selecting a particular object

1 In assessing this it should be noted that even whensubjects reported shapes and numerals, losses from con-trol performance occurred only for the attribute re-ported second. As color was always reported second inthe report of colors and numerals, a loss from controlperformance might have been expected only for this at-tribute. At each of three exposure durations, color re-ports were worse when reporting colors and numeralsthan when reporting either colors alone or colors andshapes.

504 JOHN DUNCAN

within that area. One may ask, nevertheless,how clear is the evidence for the mental spot-light position. Effects of advance knowledgeof spatial position, of response competitionfrom irrelevant stimuli, and of spatial separa-tion in divided attention will be considered inturn.

The detection or identification of a stimu-lus is usually facilitated by advance knowl-edge of its position (Bashinski & Bacharach,1980; Eriksen & Hoffman, 1974; Posner,Nissen, & Ogden, 1978). A common inter-pretation is that with advance knowledge, themental spotlight can be turned to the correctposition before the stimulus is presented.However, advance knowledge of many stim-ulus properties other than position can havesimilar facilitatory effects. Detection can beimproved by advance knowledge of form(Barber & Folkard, 1972; Teichner & Krebs,1974) or of spatial frequency (Davis, Kramer,& Graham, 1983). Even when position isknown in advance, identification can be fur-ther facilitated by advance knowledge ofcolor (Humphreys, 198 la). Several factorsprobably contribute to such effects, includingincreased efficiency of selecting relevantfrom irrelevant information (Duncan, 1981)and reduced chance for false alarms (Davis etal., 1983). To justify the conclusion that vi-sual attention behaves as a specifically spatialmental spotlight, one would need to showthat effects of advance knowledge of positionare in some way different from these moregeneral effects. This has not been done.

Turning to response competition, Eriksenand Hoffman (1973) showed that reaction toa target letter is slowed by the presence in thevisual field of irrelevant letters associatedwith the wrong response. Importantly, thiseffect rapidly diminishes as letters associatedwith the wrong response are moved fartheraway from the target. The effect occurs withvarious types of stimuli (Gatti & Egeth, 1978;Humphreys, 1981 b; Kahneman & Henik,1981). Depending on the stimuli used, someeffect of response competition can remainwith distances between the target and the ir-relevant stimuli at least up to 5° (Gatti &Egeth, 1978), but a diminution in the effectwith increasing distance is the rule. Again, amental spotlight theory gives a good accountof the results. The farther from the target ir-

relevant stimuli are placed, the less likely theyare to fall within the spotlight and thus to beprocessed to the point at which their compet-ing responses become strongly activated.Again, however, analogous results involvingmanipulation of color rather than spacebring a different perspective. Humphreys(198 la) showed that effects of competing re-sponses are diminished when the irrelevantstimulus is different in color from the target(and target color is known in advance). Thegeneral point is that when a person must dis-tinguish relevant and irrelevant stimuli in adisplay, any increase in the difference be-tween the two is likely to be helpful. When atarget is to be selected on the basis of its posi-tion (e.g., adjacency to a bar! marker), it isdoubtless easier to tell that an irrelevant stim-ulus is not the target, the farther from thetarget position it is. Similarly, it will be easierto tell that a stimulus is not the target if it hasthe wrong color. Again, the conclusion thatattention behaves as a specifically spatialmental spotlight could only be justified byshowing that effects of spatial separation areadditional to those predictable on more gen-eral grounds.

The third line of evidence concerns effectsof spatial separation of stimuli that must bothbe identified. In an experiment by Hoffmanand Nelson (1981), displays of four letterswere searched to determine which of twospecified targets they contained. At the sametime, a small box with one side missing waspresented near one of the letters, and for thisstimulus, the task was to locate the missingside. The letter search task was more accu-rately performed when the box was adjacentto the target than when it was not, a resultreplicated and extended by Hoffman, Nel-son, and Houck (1983). Although Hoffmanand Nelson (1981) suggested that the boxdrew the mental spotlight to its spatial vicin-ity, improving the detection of adjacent tar-gets, this account is not in line with resultsreported by Skelton and Eriksen (1976). Cir-cles of eight letters were presented, with barmarkers indicating a pair of letters to bejudged same or different. Reaction timeswere shortest for adjacent and; diametricallyopposite pairs. It does not seerh generally tobe true that the identification of two stimuli isbest when they are close together.


One possible interpretation of these resultsrests on a point raised already: Identifyingseveral stimuli in a complex display is bestwhen they form a strong perceptual group. Itseems clear that in the displays of Hoffmanand Nelson (19 81), the box and the letter tar-get would have grouped together moststrongly when they were adjacent. I haveconfirmed that in displays like those of Skel-ton and Eriksen (1976), subjects rate adja-cent and diametrically opposite characters asmost strongly grouped. Supportive datacome from an experiment by Fryklund(1975). Five letters, colored red, were to bereported from a 5 X 5 matrix; the remainder,colored black, were to be ignored. Perform-ance was better when the five targets formed arow than when they were adjacent but ar-ranged in no regular pattern. It seems quitelikely that strength of perceptual grouping,rather than adjacency per se, determines per-formance in these experiments.

Thus, there are doubts over all three linesof evidence taken to support the mental spot-light theory. Overall, the evidence for thistheory is not compelling.

Experiment 1The technique of the present experiments

was based on the work of Rock and Gutman(1981). These authors presented displayscontaining two overlapping outline shapes,one red and one green, with one shape (speci-fied by color) to be rated for pleasingness, theother to be ignored. Each display was about3.5° in size and was presented for 1 s. Evenwhen tested immediately after presentation,subjects showed very poor memory for theunattended shape. Similar results were ob-tained by Neisser and Becklen (1975), usingsuperimposed visual events (video scenes ofgames being played) each lasting 1 min,.

As they stand, these results are consistentwith any of the theories just discussed. Anobject-based theory would propose that onlythe target object received focal attention. Adiscrimination-based theory would proposethat only discriminations concerning the tar-get shape were made. Even a space-basedtheory might handle the results by proposingthat in these rather large displays, the mentalspotlight was focused only on areas of mostrelevance for target identification.

Consider, however, displays like thoseshown in the top half of Figure 1. Each con-sists of a box varying in size (small or large)and varying in position of a gap in its contour(right or left), with a line struck through itvarying in tilt (clockwise or counterclockwisefrom the vertical) and varying in texture(dotted or dashed). Assuming that such a dis-play is in fact seen as two separate objects, abox and a line (cues distinguishing the twoinclude texture, line continuation, majoraxis, and perhaps instructions), it allows adirect test of the object-based theory. With abrief stimulus exposure, identifying twoaspects of one object should be as easy asidentifying either aspect alone, whereas iden-tifying aspects of two different objects shouldbe more difficult. Such results could not beexplained either by a discrimination-basedtheory or by a space-based theory. As for theanalyzer theory of Allport (1971), there is noreason to suppose that judgments concerningthe same object would involve different ana-lyzers, whereas those concerning differentobjects would involve the same analyzer. Adiscrimination-based theory based on the

Figure 1. Displays used in Experiment 1. (Top left: Maintask—small box with gap on right, dotted line with tiltclockwise. Top right: Main task—large box with gap onleft, dashed line with tilt counterclockwise. Bottom left:Initial task—longer line on left. Bottom right: Mask.)

506 JOHN DUNCAN

difficulty rather than on the similarity of dis-criminations would have no more success:Here the extent of interference between twojudgments would 'depend only on their indi-vidual difficulties. Finally, such results couldnot be explained by the mental spotlighttheory's proposal that simultaneous judg-ments are best made when the critical sourcesof information are close together. Notice, inparticular, that the critical information forbox size occurs only at the top and bottom ofthe display (box width is fixed), whereas thecritical information for gap position occursonly in the middle of the box's sides. Thespatial separation of these two is at least asgreat as separations for judgments concern-ing different objects. It seems that with thistype of display, the object-based theory pre-dicts a unique pattern of results.

MethodSubjects. Forty-eight subjects, aged between 18 and

40, were recruited from the paid subject panel of theApplied Psychology Unit. Twenty-eight were female.

Apparatus. The experiment was run on-line on aCambridge Electronic Design laboratory computer sys-tem. Stimuli were presented on a Hewlett-Packard cath-ode-ray tube (CRT) display (X-Y, Model 1332A) withP24 phosphor (fast decay, bluish color). Responses weremade on two pairs of keys; One pair operated with themiddle and index fingers of the left hand and the other,with the corresponding fingers of the right hand. A footswitch was used to control stimulus presentation. Thesubject sat alone in a semidarkened room and viewed thescreen from a fixed chin rest.

Procedure. Each subject served in two sessions ofabout an hour each, on different days. The first sessionwas split into an initial phase to determine a suitablestimulus exposure duration for the particular subject,followed by practice on the task of the experimentproper. Data were collected only in the second session.

In the initial phase, each stimulus display was made upof two vertical lines (see Figure 1, bottom left),2 At theviewing distance of approximately 65 cm, one line wascentered 3.5' of visual angle to the left of fixation; theother, 3.5' of visual angle to the right. One line (at ran-dom either the left or the right) was 23' of visual angle inlength; the other, 20' of visual angle. The task was todecide which line was longer. Responses were made withthe pair of keys operated by the left hand. Within thepair, the subject pressed the key on the same side as theline judged to be longer. Responses were to be made asaccurately as possible, without emphasis on speed.

Each trial began with a single fixation dot in the centerof the screen. When ready, the subject pressed the footswitch to give an immediate presentation of the stimulusdisplay. This remained for a predetermined exposureduration, then was at once replaced by a masking dis-play. This mask, in turn, remained until the response

was made, then was replaced by the fixation dot of thenext trial. The mask (see Figure 1, bottom right) was anirregular pattern of horizontal, vertical, and oblique linescovering an approximately rectangular area of 52' ofvisual angle vertically X 32' of visual angle horizontally,centered on fixation.

The task was performed in blocks of 24 trials. Theorder of stimuli was random except that each alternativeappeared 12 times per block. Exposure duration wasfixed throughout a block, but changed from block toblock on the following schedule. A duration of 220 mswas used for the first block. After each block, if the prob-ability correct in that block was greater than .85, thenexposure duration was decreased; if probability correctwas between .65 and .85, then exposure duration wasunchanged; if probability correct was less than .65, thenexposure duration was increased. This procedure wascontinued1 until a criterion was reached of three blocks(not necessarily successive) at a particular exposure du-ration, each with a probability correct between .65and .85.

The absolute size of decreases and increases in expo-sure duration was determined as follows. Consider thismaster sequence of exposure duration? (in milliseconds):220-180-140-100-80-60-40. Starting at 220, sub-jects worked down this sequence either until criterionwas reached or until one block had a probability correctof less than .65. In the latter case, the size of the steps inthe master sequence was then halved, and thereafter in-creases and decreases in exposure duration took place inthese half steps until criterion was reached.

Thus, each subject ended the initial phase having es-tablished a criterion exposure duration that ranged from50 ms to 100 ms, with a mean of 79.2 ms. Pilot worksuggested that this criterion exposure duration, whenused for data collection in the experiment proper, wouldgive a suitable overall level of performance.

In the experimental task, each stiihulus display wasmade up of a rectangular outline box with an obliqueline struck through it (see Figure 1, top left and right).Both the box and the line were centered on fixation. Bothvaried in two dimensions.

The box, 20' of visual angle in width, was either 30' or37' of visual angle in height. It had a gap of 2.5' of visualangle in the center of either its left or its right side.

The line, 46' of visual angle in length, was tilted 4° 1'either counterclockwise or clockwise from the vertical.In texture it was either dotted or dashed. In either case,the line was made up of a total of 12 illuminated dots. Inthe dotted case, these dots were spaced evenly along theline's length. In the dashed case, they were grouped into6 clearly separate dashes. Both textures appeared quitedifferent from the texture of the box. The box too, ofcourse, was made up of illuminated dots, but adjacentdots in its outline were sufficiently close (separation was0.45 times that of the dotted line) to give the impressionof a continuous line more than that of a line of separatedots.

Though stimulus displays potentially varied alongfour dimensions termed box(size), bdx(gap), line(tilt),

2 Detailed display descriptions are available from theauthor.


and line(texture), for any given subject only two dimen-sions were relevant, and only these two varied. When notvarying, box(size) was fixed at 30' ,of visual angle,box(gap) was fixed on the right, line(tilt) was fixed coun-terclockwise, and line(texture) was fixed at dashed,

Trials resembled those of the initial phase. When fix-ated, the subject pressed the foot switch to initiate thestimulus display. Following a predetermined exposureduration, the mask (see Figure 1, bottom right) appearedand remained until the response was complete. Subjectswere asked to respond as accurately as possible, withoutemphasis on speed.

Responses were made as follows. One of a subject'stwo relevant dimensions (termed the first-reported di-mension) was assigned to the pair of keys operated by theleft hand, the other (the second-reported dimension), tothe pair operated by the right. For any 1 subject, thisassignment was fixed throughput the experiment. Insome blocks (single-judgment blocks), subjects made,ajudgment about only one dimension and correspond-ingly responded only with one hand. In other blocks(double-judgment blocks), judgments were made forboth dimensions, In these blocks order of report wasfixed, and the left-hand response was always made first.Within each pair of keys, let the left key be termed Key 1and the right key be termed Key 2. For judgments con-cerning box(size), a response on Key 1 indicated a judg-ment of small, and a response on Key 2, a judgment oflarge. For box(gap), Key 1 indicated left, and Key 2indicated right. For line(tilt), Key 1 indicated counter-clockwise, and Key 2 indicated clockwise. For line(tex-ture), Key 1 indicated dotted, and Key 2 indicateddashed.

When the initial phase of determining an appropriateexposure duration was complete, the remainder of thesubject's first session was devoted to practice in the ex-perimental task. Practice was given under three condi-tions: single-judgment conditions for each of the sub-ject's two relevant dimensions and the double-judgmentcondition. Performance was blocked by condition, andthe order of conditions was counterbalanced across sub-jects (orthogonally to other factors). Within each condi-tion, there were six blocks of 24 trials, with the order ofstimuli random in each block. The first block was run atan exposure duration three steps up in the master se-quence from the subject's criterion duration (the +3duration); the second, at the +2 duration; the third, atthe +1 duration; and the remaining three, at the crite-rion duration. If the criterion duration was at a half stepin the master sequence (e.g., 90 ms), the +1 durationwas taken to be the first step above (100 ms). This sched-ule of practice was designed to teach each discriminationinitially under conditions of relatively good visibility.The percentage correct was shown to the subject at theend of each block.

In the second session, the subject again performed inthree conditions, in the same order as before. Withineach condition, two practice blocks of 24 trials each, thefirst at the + 3 duration and the second at the criterionduration, were followed by six experimental blocks of 48trials each, all at the criterion duration. Only data fromthe experimental blocks were used in later analyses.Within each 48-trial block, each of the four stimulus al-ternatives (defined by the two possible values of the sub-ject's two varying dimensions) occurred 12 times. Other-

Table 1Experiment 1: Proportion Correctin Each Condition

First-reporteddimension

Second-reporteddimension

Group Single Double Single Double

Same-object

Different-objects

.844

.786

.834

.788

.849

.796

.834

.715

Note. Single = single-judgment. Double = double-judg-ment.

wise, stimulus sequences were random. Again, thepercentage correct was shown at the end of each block,

In single-judgment conditions, the importance of at-tending only to the one dimension to be judged wasemphasized, while in double-judgment conditions, itwas stated that the two dimensions were equally impor-tant, though one was always to be reported first. Re-peated emphasis was given to the importance of carefulfixation before stimulus initiation and to the avoidanceof keyboard errors.

Design. Subjects were divided equally into two groups.In the different-objects group, the subject's two relevantdimensions concerned different objects, that is, one con-cerned the box and the other, the line. In the same-objectgroup, the subject's two relevant dimensions concernedthe same object.

Each group was divided equally into subgroups on thebasis of the particular dimensions assigned to be rele-vant. The different-objects group was divided into eightsubgroups of 3 subjects each, corresponding to the fourpossible pairs of dimensions from different objects[box(size) plus line(tilt), box(size) plus line(texture),box(gap) plus line(tilt), box(gap) plus line(texture)] mul-tiplied by the two possible orders of report (i.e., assign-ment of dimensions to hands) within each pair. Thesame-dbject group was divided into four sub-groups of6 subjects each, corresponding to the two possible pairsof dimensions from the same, object [box(size) plusbox(gap), line(tilt) plus line(texture)] multiplied by thetwo possible orders of report. Note that each displaydimension appeared 6 times in each of the four cells'(first- vs. second-reported dimension, different-objectsgroup vs. same-object group) of the design.

Verbal descriptions of displays. After the experimentwas complete, 10 new subjects were each asked to give aspontaneous verbal description of an experimental dis-play. In all 10 cases, the display was said to show a boxwith a line through it. This reinforces the assumptionthat these displays were seen to contain two separateobjects.

Results and Discussion

The major results are shown in Table 1,which gives the mean proportion correct in

508 JOHN DUNCAN

each condition. For the s^me-pbject group,performance was very similar in the single-judgment and double-judgment conditions.For the different-objects group, a decline inperformance occurred in the double-judg-ment condition, though it was confined en-tirely to the second-reported dimension.

The data were examined by analysis ofvariance (ANOVA), with group (same-objectvs. different-object) as a between-subjectsfactor, and with dimension (first-reported vs.second-reported) and condition (single-judg-ment vs. double-judgment) as within-sub-jects factors. There were significant maineffects of group, F(l, 46) = 5.2, p < .05, andof condition, F(l, 46) = 7.0, p < .02; andsignificant interactions of Dimension XCondition, F( 1,46) = 6.7, p<.02, andof Group X Dimension X Condition, F(l,46) = 5.1, p < .05. This last interaction is theimportant one and shows that effects of con-dition were confined to the second-reporteddimension of the different-objects group.

These conclusions were confirmed by sup-plementary analyses that treated each groupseparately. For the same-object group, therewas no significant effect of dimension, condi-tion, or their interaction (all Jf*is < 1). For thedifferent-objects group, there were signifi-cant effects of condition, F(l, 23) = 8.8,/? <.01, and of Dimension X Condition, F(l,23) = 8.7,/?<.01.

As a whole, these data confirm the predic-tions of the object-based theory. If two judg-ments concern the same object, they do notinterfere; but if they concern two differentobjects, these may compete for focal atten-tion. The data suggest that the object whoseproperty is to be reported first is favored, andperformance on the other is allowed to de-cline.

Further breakdowns of the data, however,revealed two troublesome details. The firstconcerned the different-objects group. Al-though in this group as a whole there was anadvantage for single- over double-judgments,this trend vanished for those subjects whosesecond-reported dimension was line(tex-ture). For these 6 subjects, the mean single-judgment advantage (in terms of proportioncorrect) was — .020 for the first-reported di-mension and was -000 for the second-re-ported dimension. Although this result could

well have been due to noise rathe data, it doessuggest a possible interpretation in terms of avariant of Allport's (1971) analyzer theory.Whereas the dimensions box(size), box(gap),and line(tilt) all concern the broad questionof where lines are in the visual field, line(tex-ture) concerns the quality of a line. It may besuggested that judgments concerning differ-ent objects interfere only when they involvethe broad question of where ;lines are. Datafor those 6 subjects having jine(texture) astheir/zrtf-reported dimension contradict thisproposal. Mean single-judgment advantageswere .017 and .102 for first- and second-re-ported judgments, respectively. Neverthe-less, the proposal seemed worth a further test.Experiment 2 was run to check on the resultsobtained for line(texture) and to see whetherthey would generalize to another dimensionconcerning line quality—brightness.

The second troublesome detail of the re-sults concerned the same-object group. Thedata were quite clear for those subjects whosejudgments concerned the line. Averagingover first- and second-reported dimensions,the mean advantage of single-judgments overdouble-judgments was -.008, with a stan-dard error of .011. Though the conclusionthat performance was equal in the two condi-tions rests on a failure to reject the null hy-pothesis, any "true" advantage for single-judgments can at most have been very small.Data were less clear, however^ for those sub-jects whose judgments concerned the box.The mean advantage of single-judgments was.034, with a standard error of .026. Thesemore noisy data do not allow a strong con-clusion that performance was effectivelyequal in single- and double-judgment condi-tions. In fact, there was a hint that resultsdepended on order of report. It was the sub-group for whom box(size) was first-reportedand box(gap) was second-reported whoshowed evidence of better performance insingle-judgments (a mean advantage overdouble-judgments of .063). This trend heldonly for 4 of the 6 subjects in this subgroup,but should be remembered for comparisonwith later results.

It is crucial to know whether judgments ofbox(size) and box(gap) can be made togetherwithout mutual interference. Line(tilt) andline(texture) are both represented in precisely


the same place—along the whole length ofthe line—whereas box(size) and box(gap)are not. If only the former pair of judgmentscan be made without interference, the con-clusion may be, in line with the mental spot-light theory, that interference occurs when-ever the information for two judgments issipatially separate, not when the two concerndifferent objects. Box judgments were exam-ined further in Experiments 3 and 4.

Experiment 2Part 1

In Part 1 of Experiment 2, each displayagain consisted of a box with a line struckthrough it. Again, only two dimensions wererelevant for a single subject, this time alwaysconcerning different objects. For half of thesubjects, one dimension concerned tilt andthe other concerned texture. Counterbalanc-ing was used to determine whether the linevaried in tilt and the box varied in texture orvice versa. For the remaining subjects the di-mensions of tilt and brightness were similarlymanipulated. Order of report was now a cru-cial variable, because the results of Experi-ment 1 suggested that performance losses inthe double-judgment condition might occuronly if tilt was the second-reported dimen-sion.

Part 2 of Experiment 2 was a control ex-perim'ent in which tilt and brightness variedfor the same object, either the box or the line.

The data for individual subjects in Experi-ment 1 suggested that little had been gainedby establishing a separate exposure durationfor each. In Experiment 2 the same exposureduration was used for all subjects.

MethodSubjects. Twenty-four subjects, aged between 23 and

41, were recruited as before. Nineteen were female.Procedure. Each stimulus-display was made up gf a

rectangular outline box (33' of visual angle vertically X18' of visual angle horizontally) with a line (46' of visualangle in length) struck through it. For half of the subjects(texture group), one object varied only in texture (dottedor dashed), whereas the other varied only in tilt (counter-clockwise or clockwise from the vertical). For the re-maining subjects (brightness group), one object variedonly in brightness (dim or bright), whereas the othervaried only in tilt (counterclockwise or clockwise).Again, only two dimensions of the display varied for agiven subject.

In the texture group, the object varying in texture wasalways vertical. The dotted and dashed textures them-selves were made up in the same way as in Experiment 1.The other object, varying in tilt, was always drawn withthe dotted texture. As in Experiment 1, the tilt was 4" 1'counterclockwise or clockwise from the vertical.

In the brightness group, the object varying in bright-ness was always vertical. Brightness was manipulated byvarying the spacing between adjacent dots in a line. Al-though absolute brightnesses were not measured (anddue to limitations of the display device were not preciselyreplicable across sessions), the ratio of spacings in dimand bright lines was 1.55; or in other words, for a givenline length, a bright line contained 1.55 times as manydots as a dim one. In both cases, spacings were suffi-ciently short to give the appearance of a continuous linemore than that of lines of separate dots, and the differ-ence between the two cases did accordingly appear to beone of brightness rather than one of spacing. The otherobject, varying in tilt, was always drawn with the dottedtexture of Experiment 1 (spacing was twice that of thedim lines). Tilt, again, was 4" 1' clockwise or counter-clockwise from the vertical.

Responses were made in the same way as in Experi-ment 1. Within a pair of keys, response assignments fortexture and tilt judgments were as before, while forbrightness, Key 1 indicated a judgment of dim, and Key2 indicated a judgment of bright.

Events and timing on each trial were similar to thoseof Experiment 1, with one exception. During Experi-ment 1, some subjects had complained of feeling"dazed," "hypnotized," and so on. Mot work suggestedthese sensations could be minimized by forcing.shortpauses between trials, giving the subject an opportunityto blink if desired. Accordingly, mask offset on each trial,was followed by a blank interval of 1 s before the onset ofthe fixation point for the next trial. Subjects were en-couraged to blink between trials if they were having trou-ble "focusing or concentrating."

Each subject served in two sessions of about an houreach, on different days. In each session, the subject per-formed under three conditions: single-judgment condi-tions for each of the two relevant dimensions and thedouble-judgment condition. Performance was blockedby condition; the order of conditions, counterbalancedacross subjects (orthogonally to other factors), was thesame for a given subject across sessions.

In the first session, there were seven blocks of 24 trialsper condition, and exposure durations (in milliseconds)were 220,180,140,100,80,80, and 80, respectively. Inthe second session, performance in each condition con-sisted of two practice blocks of 12 trials each, with expo-sure durations of 180 ms and 80 ms, respectively, fol-lowed by four experimental blocks of 48 trials each, withan exposure duration of 80 ms. Stimulus sequences wereconstructed as in Experiment 1, and again the percent-age correct was shown at the end of each block. Again,only data from the 48-trial blocks of the second sessionwere analyzed.

Design. The 12 subjects in the texture group weredi-vided into two equal subgroups. For 6 subjects, tilt wasthe first-reported dimension and texture was the second-reported dimension, an arrangement reversed for theother 6 subjects. Each subgroup was further subdividedinto two sets of 3 subjects each, according to whether the

510 JOHN DUNCAN

box varied in tilt and the line varied in texture or viceversa. The 12 subjects in the brightness group were sub-divided similarly.

Part 2

The results of Experiment 1 showed thatwhen judgments of tilt and texture concernthe same object, performance is equal in sin-gle-judgment and double-judgment condi-tions. Part 2 of Experiment 2 was run tocheck that the same applies to judgments oftilt and brightness.

Part 2 was similar to Part 1 (brightnessgroup) except that for a given subject, tilt andbrightness both varied for the same object,either the box or the line. The object not vary-ing was always vertical and was drawn withdotted texture. Note that in Part 1 of Experi-ment 2, the object that varied in tilt wasalways drawn with a dotted texture, whichmade tilt harder to determine than in Part 2because there were relatively few dots perunit line length. To compensate for this, inPart 2 the exposure duration was reduced.The sequence (in milliseconds) used for eachcondition in the first session was 180-140-100-80-60-60-60, whereas durations of140 ms (first practice block) and 60 ms (re-maining blocks) were used for each conditionin the second session.

Twelve subjects, aged between 18 and 41,were recruited as before. Six were female.

Results and Discussion

Experiment 2: Part 1

The proportion correct in each conditionof Part 1 is shown in Table 2. Because prelim-inary analyses showed that results were thesame whether the line varied in tilt and thebox varied in quality (texture or brightness)or vice versa, this factor is not consideredfurther. Data are shown separately for thetexture and the brightness groups and for thetwo orders of report. For all groups perform-ance was better in single-judgment condi-tions than in double-judgment conditions.For all groups except that reporting tilt firstand texture second, the single-judgment ad-vantage was greater for the second-reporteddimension.

The data were examined by an ANOVAwith group (texture vs. brightness) and order

Table 2Experiment 2, Part 1: Proportion Correctin Each Condition

Group

First-reported Second-reporteddimension dimension

Single Double Single Double

TextureTilt firstTilt second

BrightnessTilt firstTilt second

.910

.883

.744,871

.844

.834

.722

.837

.959

.884

.872

.812

.937

.794

.792

.673


(tilt first vs. tilt second) as between-subjectsfactors and with dimension (first-reportedvs. second-reported) and condition (single-judgment vs. double-judgment) as within-subjects factors. The effect of condition wassignificant, F(\, 20) = 29.1, p < .001, indi-cating the overall superiority of single-judg-ments. The Dimension X Condition interac-tion was marginal, F( 1,20) = 3.8, p<.07, as was the Group X Dimension X Condi-tion interaction, F(\, 20) = 4.1, p < .06, hint-ing at a stronger Dimension X Condition in-teraction for the brightness group. Neitherthe Order X Dimension X Condition interac-tion, F( 1,20)=2.7, nor the Group X Order XDimension X Condition interaction, F(l,20) = 0.2, was significant. Other significanteffects were group, F(\, 20) =• 10.9, p < .01,showing the overall superiority of the texturegroup, and the Order X Dimension interac-tion, F(\, 20) = 5.B,p < .05, showing that tiltjudgments tended to be less accurate thanjudgments of either texture or brightness.

In a supplementary analysis, the data forthe group having tilt as their first-reporteddimension and texture as their second-re-ported dimension were considered alone.Even for this group the effect of conditionwas significant, F(l, 5) = 8.4, p < .05.

Experiment 2: Part 2

The proportion correct in each conditionof Part 2 is shown in Table 3. The data con-firm that when two judgments concern thesame object, performance is equal in single-judgment and double-judgment conditions.


Table 3Experiment 2, Part 2: Proportion Correctin Each Condition

Group

• First-reported Second-reporteddimension dimension


Tilt firstTilt second

.928

.824.920.817

.750

.939.774.933


The data were examined by an ANOVA,with order as a between-subjects factor anddimension and condition as within-subjectsfactors. The only significant effect was theOrder X Dimension interaction, F(l, 10) =53.8, p < .001, showing that tilt was judgedmore accurately than brightness.

In terms of proportion correct, the meanadvantage of single-judgment over double-judgment conditions was — .001, with a stan-dard error of .014. Again, though the conclu-sion that performance was equal in the twoconditions reflects a failure to reject the nullhypothesis, we may be confident that anytrue advantage for the single-judgment casecan at most have been small.

A further ANOVA compared these resultswith those obtained for the brightness groupin Part 1 of Experiment 2. Between-subjectsfactors were experiment (Part 1 vs. Part 2)and order, and within-subjects factors weredimension and condition. The main effect ofcondition was significant, F(l, 20) = 8.8,p < .01, and there were significant interac-tions of Experiment X Condition, f(l,20) = 9.1, p < .01, andkof Experiment XDimension X Condition, F(l, 20) = 7.5,p < .02, showing that the effect of conditionwas confined to Part 1, and in Part 1 occurredespecially for the second-reported dimen-sion. The remaining significant effects wereexperiment, f (1, 20) = 5.4, p < .05, reflect-ing better performance overall in Part 2, andthe Experiment X Order X Dimension in-teraction, F(l, 20) = 24.0,p < .001, showingthat tilt was judged less accurately thanbrightness in Part 1, but more accurately inPart 2. The reason for this—that the tiltedobject was drawn with a dotted texture inPart 1 —has already been described.

Taken together, the results of Parts 1 and 2of Experiment 2 provide further support forthe object-based theory. In particular theydisconfirm the idea that simultaneous judg-ments concerning different objects may in-terfere only when both concern the broadquestion of where lines are. There seems novariant of Allport's (1971) analyzer theorythat could account for the results. Simulta-neous judgments of tilt and brightness inter-fere when they concern different objects, butnot when they concern the same object.Though the result is weaker, the same proba-bly holds for judgments of tilt and texture.

In all likelihood, there is something un-usual about texture judgments. A furthergroup of 6 subjects run under the conditionsof Experiment 2, Part 1 with tilt as the first-reported dimension and texture as the sec-ond-reported dimension, gave proportionscorrect of .874 and .887, respectively, forfirst- and second-reported dimensions in sin-gle-judgment conditions, compared with.871 and .866, respectively, for double judg-ments. These data suggest again that the sin-gle-judgment advantage is reduced when thesecond-reported dimension is texture. Nev-ertheless, Experiment 2, Part 1 showed thateven then this advantage can remain signifi-cant. Although further research on texturejudgments may be interesting, the generalprinciple stands that performance sufferswhen a person must make simultaneousjudgments concerning different objects.

Experiment 3

In Experiment 1 it was not fully clear thatjudgments of box(size) and box(gap) can bemade together without mutual interference.The issue is especially important becausethese judgments, though concerning thesame object, rest on information at separatespatial locations. Experiment 3 reexaminedperformance on these judgments, using theimproved technique (forced pause of 1 s be-tween trials) of Experiment 2. This time thebox was displayed alone, and for all subjects,size and gap were the relevant dimensions.

MethodSubjects. Twenty-four subjects, aged between 18 and

42, were recruited as before. Fifteen were female.

512 JOHN DUNCAN

Design and procedure. Each stimulus display con-sisted of a box varying in size and side of gap, exactly as inExperiment 1. There was no line struck through. For halfof the subjects, size was the first-reported dimension andgap was the secpnd-reported dimension, an arrangementreversed for the other half.

The procedure was copied exactly from Experiment 2,as were the events and the timing of a single trial. In anattempt to maximize sensitivity, two different exposuredurations were used. For half of the subjects, a durationof 70 ms was substituted wherever 80 ms had been usedin Experiment 2, including the experimental blocks ofthe second session from which data were analyzed. Forthe remaining subjects, the duration of 80 ms was re-tained. This factor was varied orthogonally to the others;otherwise, counterbalancing was as before.

ResultsThe mean proportion correct in each con-

dition is presented in Table 4. Data are shownseparately for each order of report and expo-sure duration. In terms of proportion correct,the mean advantage of single-judgments overdouble-judgments was .014, but again, as inExperiment 1, this advantage seemed to beconfined to those subjects reporting size firstand gap second.

The data were examined by an ANOVA,with order (gap first vs. gap second) and ex-posure duration (70 ms vs. 80 ms) as be-tween-subjects factors and dimension (first-reported vs. second-reported) and condition(single-judgment vs. double-judgment) aswithin-subjects factors. Neither the effect ofcondition, F(l, 20) = 2.8, nor the Order XCondition interaction, F(\, 20) = 2.9, wassignificant. The only significant effect was theOrder X Dimension interaction, F(\, 20) =15.7, p < .001, indicating that gap was more'accurately judged than size.

For the 12 subjects reporting gap first, themean advantage of single-judgments overdouble-judgments was .000, with a standarderror of .008. Again, any true single-judg-ment advantage can at most have been verysmall.

For the 12 subjects reporting gap second,the mean advantage of single-judgments overdouble-judgments was .027, with a standarderror of .013. Ten subjects showed better per-formance in the single-judgment condition.Again it seems unwise, taking these data inconjunction with those of Experiment 1, toconclude that with this order of report, simul-taneous judgments of size and gap can bemade without mutual interference.

Table 4Experiment 3: Proportion Correctin Each Condition

Group

First-reported Second-reporteddimension dimension


Gap first70 ms80ms

Gap second70ms80ms

.913

.931

.872

.886

.916

.920

.852

.839

.754

.845

.917

.941

.766

.842

.884

.932


Experiment 4

What special difficulty could arise when ajudgment of size must be reported before ajudgment of gap? Experience in the task sug-gested a possible effect of response compati-bility.

When both properties of the box are at-tended, subjectively one "sees" the size andthe side of the gap and then must choose aresponse. Because the side of the gap is itselfcoded in terms of left and right, there is aparticularly compelling mapping onto thechoice between left and right keys to be madewithin the pair operated by each hand. Re-porting size before gap, I became aware that itwas persistently distracting to be unable tomake this compelling response first, on theleft pair of keys.

The importance of this may be tested sim-ply by turning the display on its side. Now,neither gap nor size would have an especiallycompelling mapping onto response key andin this respect would resemble all other judg-ments used in the present work, Accordingly,the advantage of single-judgments over dou-ble-judgments should vanish.

MethodSubjects. Twelve subjects, aged between 19 and 38,

were recruited as before. Six were female.Design and procedure. Experiment 4 was similar to

Experiment 3 in all respects except that (a) size wasalways the first-reported dimension, and gap was thesecond-reported; (b) the display device was turned on itsside (counter-clockwise) so that the box varied in widthrather than height, and the gap was in the top or bottom.Within the right-hand pair of keys, Key 1 indicated a


judgment of top, and Key 2, a judgment of bottom.Exposure duration was varied as in Experiment 3, takingvalues of either 70 ms or 80 ms in the experimentalblocks of the second session.

Results and DiscussionThe mean proportion correct in each con-

dition is shown in Table 5. There was no ad-vantage for single-judgments over double-judgments.

The data were examined by an ANOVA,with exposure duration (70 ms vs. 80 ms) as abetween-subjects factor and with dimension(first-reported vs. second-reported) and con-dition (single-judgment vs. double-judg-ment) as within-subjects factors. The effect ofcondition was not significant, F( 1,10) = 3.4,and in any case was in the direction of anadvantage for double-judgments. The onlysignificant effect was dimension, F(l, 10) =9.6, p < .02, showing that gap was judgedmore accurately than size.

In terms of proportion correct, the meanadvantage of single-judgments over double-judgments was — .017, with a standard errorof .009. Again, it is safe to conclude that anytrue advantage for single-judgments can atmost have been very small.

A further ANOVA compared these datawith those obtained when size was reportedfirst in Experiment 3. Factors were as before,with the addition of experiment (3 vs. 4) vary-ing between subjects. The Experiment XCondition interaction was significant, F(l,20) = 7.1, p < .02, confirming that turningthe display on its side reduced the advantageof the single-judgment condition. The onlyOther significant effect was dimension, F(i,20) = 7.7, p < .02, again showing that gapwas judged more accurately than was size.

In all probability, the results of Experi-ments 1 and 3^-suggesting a special diffi-culty in reporting box(size) and box(gap) inthat order—were due to a compatibility ef-fect. When this effect is avoided, judgmentsof the two attributes can be made togetherwithout mutual interference.

In these experiments, an important generalquestion concerns the possibility of informa-tion loss during output. When two responsesare to be made, does it happen that the sec-ond answer is forgotten while the first is givenor that confusion over key assignments

TablesExperiment 4: Proportion Correctin Each Condition

Exposureduration

70 ms80ms

First-reporteddimension

Single

.840

.885

Double

.847

.912

Second-reporteddimension

Single

.874

.931

Double

.885

.955


causes mistakes to be made? The results ofsame-object groups suggest that the answerusually is no: If compatibility effects areavoided, two key-press responses can bemade in turn without loss of accuracy. Itshould be remembered that accuracy wasstressed in the experimental instructions, andsubjects were encouraged to take as muchtime as they needed to make their responses.It seems likely that this procedure was suc-cessful in producing performance dependentonly on the events of stimulus input, not onresponse output.

General DiscussionThe present results confirm the predictions

of the object-based theory. Two judgmentsconcerning the same object can be made to-gether without mutual interference, whereastwo judgments concerning different objectscannot. The results cannot be explained onthe basis of either the similarity of discri-minations (Experiment 2) or the spatial dis-tribution of information (Experiments 3and 4).

A rider should be added to the conclusionon judgments concerning the same object.Although the present results agree with thoseof Allport (1971) and Wing and Allport(1972), different findings were described byEgeth and Pachella (1969). The stimulus wasa dot varying in horizontal and vertical posi-tions within a square, with 15 alternative po-sitions along each dimension. With stimulusexposures up to 2 s, judgments on each di-mension suffered when the two were requiredsimultaneously. When the number of re-sponse alternatives is large (the absolute

514 JOHN DUNCAN

judgment task), the crucial problem wouldseem to be to map one's perceptual impres-sion onto the response categories. In the ex-periment of Egeth and Pachella (1969), forexample, this would require something like14 separate (stored) response criteria for eachdimension and a mapping of perceptualinput onto these criteria. It may be impossi-ble to carry out such a complex, rather de-liberative mapping for two dimensions si-multaneously even though they may concernthe same object. This problem deserves fur-ther study.

It should be clear that at present we have nomore than the skeleton of an object-basedtheory. The present data confirm that focalattention acts on packages of information de-nned preattentively and that these packagesseem to correspond, at least to a first approxi-mation, to our intuitions concerning discreteobjects. As discussed before, however, theexact details of preattentive packaging mustbe filled in empirically. The present resultssuggest that an object-based theory is alongthe right lines; but it will take much furtherwork to refine our knowledge of preattentiveperceptual organization beyond this firstcrude idea of the "discrete object."

Earlier, evidence was reviewed to showthat the simultaneous identification of sev-eral stimuli is best when they form a strongperceptual group. It was noted that this evi-dence is sometimes taken to support the ob-ject-based theory, which does depend on theimportance of preattentive grouping. Howdoes such evidence fit with the present find-ings?

First, there is now empirical justificationfor our earlier hesitation to refer to two lettersgrouped by spatial arrangement or color asparts of "one object," in quite the same senseas, for example, a box's tilt and its brightness.Although it is true that letters groupedstrongly together are identified better thanthose grouped less strongly, it is not true thatthey can be identified without any mutualinterference. Adding extra letters to a group,for example, causes performance to decline(Kahneman & Henik, 1977).

With this in mind, there seem to be twopossible approaches to the integration ofpresent and previous results. The first wouldinterpret grouping strength as a simple con-

tinuum. To explain why there is no interfer-ence at all in the simultaneous identificationof, for example, a box's tilt and its brightness,it would simply be proposed that the group-ing together of these two features is evenstronger than the grouping of separate lettersby spatial arrangement or common color.This approach would interpret the presentdata as simply an extreme case of stronggrouping effects.

The second approach is based on the hier-archical organization of visual information.Although less parsimonious than the first, ithas the advantage of capturing the intuitionof a qualitative difference between the case oftwo letters grouped together, which remaintwo objects, and the case of a box's tilt and itsbrightness, which are two aspects of a singleobject. Consider the following problem forthe object-based theory. Although the claimis that focal attention deals with whole ob-jects, common sense strongly suggests thatattention to a whole skyscraper would be in-appropriate for determining whether there isa crack in the third window from the left onthe 13th floor. What we need is an idea of thesorts of information likely to be coded in the"chunks" corresponding to "objects" at dif-ferent levels of a hierarachy: building, win-dow, and so forth. It seems highly likely thatproperties such as its height, shape, and tex-ture would be coded in the representation of awhole building, whereas a crack in one of itswindows would not. Although such ideas ob-viously should be checked by experiment,they suggest the following approach to effectsof grouping. In a mixed display, for example,of red and blue letters, there are objectspresent at (at least) two levels. At the higherlevel would be the shape formed by the wholeset of red letters. At the lower level would bethe shapes of individual letters. Focal atten-tion at the higher level might be sufficient todetermine the shape formed by the whole setof red letters; but at this level, the identities ofindividual letters would not be coded. Toidentify and to report individual letters, itmight be necessary to direct focal attention toeach in turn. To account for grouping effectslike those of Kahneman and Henik (1977)and Merikle (1980), one would propose thatdirecting focal attention to several objects inturn is easier when these form part of a single


object at a higher level. (This might be analo-gous to paying attention first to a whole ob-ject and then "zooming in" for closer exami-nation of a particular part.) However, aseparate act of focal attention is still neededfor each, producing mutual interference be-tween them. To account for effects like thoseobtained in the present experiments, onewould propose that when focal attention ispaid to an object, all features coded as proper-ties of the whole (size, shape, color, etc.) areperceived without mutual interference. Thisapproach to integrating the present findingswith work on perceptual grouping seems wellworth further work.

As discussed previously, object-based, dis-crimination-based, and space-based theoriesare not mutually exclusive. It is possible thatthere are several separate contributions to thelimit on our ability to deal with severalsources of visual information at once. Thus,although it is clear in retrospect that the re-sults of Allport (1971) and Wing and Allport(1972) do not lend unambiguous support tothe analyzer theory, such support may wellarise in an experiment that orthogonally ma-nipulates the similarity between discrimina-tions and their reference to the same or todifferent objects. Similarly, although doubtshave been raised over the evidence currentlytaken to support a mental spotlight theory,experiments should be done that vary thespatial separation of information for discrim-inations concerning both the same and dif-ferent objects.

Two variations of the mental spotlighttheory are worth particular consideration.First, it may be suggested that this spotlight isoriented not in two-dimensional space but inthree-dimensional space. If this were the caseand if the present displays were interpreted asdepicting two objects separated in depth,then the results might be explained withoutreference to an object-based theory. Judg-ments concerning the same object would reston information at the same (perceived) loca-tion in depth, whereas those concerning dif-ferent objects would not.3 The difficulty withthis account is that subjectively the displaysappeared quite flat. Of 10 subjects asked togive verbal descriptions of these displays (seeExperiment 1), none mentioned spontane-ously a separation in depth between the box

and the line, and when specifically asked, allpreferred to describe displays as flat ratherthan as containing separation in depth. In thesemidarkened room, there were ample cuesto the true depth of the screen and the dis-play. Any illusory separation of objects indepth would have had to have overriddenveridical information. For this reason, an ac-count of the results in terms of a three-di-mensional mental spotlight seems quite un-likely, though not completely impossible.

In a different elaboration of the space-based view, Treisman (Treisman, 1982;Treisman & Gelade, 1980; Treisman &Schmidt, 1982; Treisman et al., 1983) sug-gested that visual attention acts to take up allinformation from a particular area of spacebut that the shape of this area can be deter-mined by a prior stage of Gestalt groupingand segmentation of the visual field, verymuch as envisioned by the object-basedtheory, For example, it may well be possiblefor such a mental spotlight to assume theshape of either the box's contour or the line,in the present displays; and indeed this maybe expected if grouping cues indicate that thebox and the line are separate objects. At firstsight, it seems that this is better viewed as anobject-based theory rather than a space-based theory. The chunk of informationdealt with by focal attention is determinedby Gestalt grouping, not by anything specifi-cally spatial. A further point is missed by sucha conclusion, however. Treisman proposedthat the function of focal attention is to linktogether information concerning an object'sdifferent attributes. Preattentively, separate"maps" of the visual field are formed for dif-ferent stimulus attributes: color, size, aspectsof shape, and so forth. Within each map,Gestalt grouping factors can operate to indi-cate candidate areas for focal attention. Focalattention then acts on a particular candidatearea and links together into a single perceivedobject the information from this area in allthe separate maps. Now, although it is truethat the chunk of information dealt with byfocal attention is determined by Gestaltgrouping (cf. an object-based theory), it isalso true that an object's different features are

11 am grateful (?) to Mike Posner for pointing this out.

516 JOHN DUNCAN

conjoined specifically by virtue of shared lo-cation. This has testable consequences: Forexample, it should only be possible to knowwhich color goes with which shape in a dis-play if it is also known in which location bothoccur (Treisman & Gelade, 1980). It is quitepossible that the object-based theory is cor-rect with regard to the chunk of informationdealt with by focal attention, but that in theprocess of focal attention, an object's differ-ent attributes are conjoined by virtue ofshared location.

It may be noted, however, that anotherclaim of Treisman's theory is somewhat in-consistent with the present results. Treismansuggested that focal attention can be by-passed if it is not important to know howattributes in a display are conjoined: Infor-mation concerning several objects can be ob-tained in parallel. Suppose, for example, thaton6 is presented with a green T and a red O.Focal attention is not required to tell that thedisplay contains green, red, a T, and an O. Itis only required to distinguish a green T and ared O from a green O and a red T. Similarly,Treisman suggested that focal attention canbe bypassed if prior knowledge is sufficient todetermine correct conjunctions. Now con-sider, as an example, a subject judgingbox(tilt) and line(brightness) in the presentExperiment 2. Strictly speaking, conjunctioninformation was unnecessary; To performcorrectly, the subject needed only to know,for example, that somewhere in the displaythere was a contour tilted slightly clockwiseand that somewhere there was a contour thatwas bright. Furthermore, prior knowledgewas sufficient to tell that only the box couldbe tilted, and that only the line could be dimor bright. Similar arguments apply to othercases in which subjects made simultaneousjudgments concerning different objects in thepresent experiments. The data suggest thatfocal attention is not optional in the processof reporting judgments concerning visualstimuli. Even when conjunction informationwas both unnecessary and predictable, sub-jects either could not or did not deal withdifferent objects in parallel.

In conclusion, the present results show thatthe limit on our ability tb deal simulta-neously with several, sources of visual infor-mation is at least in part a limit on the num-

ber of separate objects that can be seen. It isfor future work to show whether this in fact isthe only limit responsible for the phenomenaof visual attention.

ReferencesAllport, D. A. (1971). Parallel encoding within and be-

tween elementary stimulus dimensions. Perception &Psychophysics, 10, 104-108.

Allport, D. A. (1980). Attention and performance. In G.Claxton (Ed.), Cognitive psychology: New directions(pp. 112-153). London: Routledge & Kegan Paul.

Barber, P. J., &Folkard, S. (1972). Reaction time understimulus uncertainty with respons^ certainty. Journalof Experimental Psychology, 93, 138-142.

Bashinski, H. S., & Bacharach, V. Rj (1980). Enhance-ment of perceptual sensitivity as the result of selec-tively attending to spatial locations. Perception & Psy-chophysics, 28, 241 -248.

Broadbent, D. E. (1958). Perception and communica-tion. London: Pergamon Press. •

Davis, E. T., Kramer, P., & Graham, N. (1983). Uncer-tainty about spatial frequency, spatial position, orcontrast of visual patterns. Perception & Psychophy-sics, 33, 20-28.

Duncan, J. (1981). Directing attention in the visual field.Perception & Psychophysics, 30, 90-93.

Egeth, H., & Pachella, R. (1969). Multidimensionalstimulus identification. Perception & Psychophysics,5,341-346.

Eriksen, C. W., & Hoffman, J. E. (1973). The extent ofprocessing of noise elements during selective encodingfrom visual displays. Perception & Psychophysics, 14,155-160, ;

Eriksen, C. W., & Hoffman, J. E. (1974). Selective.atten-tion: Noise suppression or signal enhancement? Bul-letin of the Psychonomic Society, 4,- 587-589.

Fryklund, I. (1975). Effects of cued-set spatial arrange-ment and target-background similarity in the partial-report paradigm. Perception & Psychophysics, 17,375-386.

Gatti, S. V., & Egeth, H. E. (1978). Failure of spatialselectivity in vision. Bulletin of 'the Psychonomic Soci-ety, 11, 181-184. ;

Hoffman, J. E., & Nelson, B. (1981). Spatial selectivity invisual search. Perception & Psychophysics, 30, 283-290.

Hoffman, J. E., Nelson, B., & Houck, M. R. (1983). Therole of attentional resources in automatic detection.Cognitive Psychology, 15, 379-410.

Humphreys, G. W. (198 la). Flexibility of attention be-tween stiinulus dimensions. Perception & Psychophy-sics, 50,291-302.

Humphreys, G. W. (1981b). On varying the span of vi-sual attention: Evidence for two modes of spatial at-tention. Quarterly Journal of Experimental Psychol-ogy, 33A, 17-30.

Kahneman, D., & Henik, A. (1977)., Effects of visualgrouping on immediate recall and selective attention.

. In S. Dornic (Ed.), Attention and performance K/(pp.307-332). Hillsdale, NJ: Erlbaum.

Kahneman, D., & Henik, A. (1981). Perceptual organi-


zation and attention. In M. Kubovy & J. R. Pomer-antz (Eds.), Perceptual organization (pp. 181-211).Hillsdale, NJ: Erlbaum.

Lappin, J. S. (1967). Attention in the identification ofstimuli in complex displays. Journal of ExperimentalPsychology, 75, 321-328.

Merikle, P, M. (1980). Selection from visual persistenceby perceptual groups and category membership. Jour-nal of Experimental Psychology: General, 109, 279-295.

Moray, N. (1967). Where is capacity limited? A surveyand a model. In A. F. Sanders (Ed.), Attention andperformance I (pp. 84-92). Amsterdam: North-Hol-land.

Neisser, L(. (1967). Cognitive psychology. New York:Appleton-Century-Crofts.

Neisser, U., & Becklen, R. (1975). Selective looking: At-tending to visually specified events. Cognitive Psy-chology, 7, 480-494,

Posner, M. I., Nissen, M. J., & Ogden, W. C. (1978).Attended and unattended processing modes: The roleof set for spatial location. In H. L. Pick & E. J. Saltz-man (Eds.), Modes of perceiving and processing infor-mation (pp. 137-157). Hillsdale, NJ: Erlbaum.

Posner, M. I., Snyder, C. R. R., & Davidson, B. J. (1980).Attention and the detection of signals. Journal of Ex-perimental Psychology: General, 109, 160-174.

Rock, I., & Gutman, D. (1981). The effect of inattentionon form perception. Journal of Experimental Psychol-

ogy: Human Perception and Performance, 7, 275 -285.

Skeltpn, J. M., & Eriksen, C. W. (1976). Spatial charac-teristics of selective attention in letter matching. Bul-letin of the Psychonomic Society, 7, 136-138.

Teichner, W. H., & Krebs, M. J. (1974). Laws of visualchoice reaction time. Psychological Review, 81, 75-98.

Treisman, A. (1982). Perceptual grouping and attentionin visual search for features and for objects. Journal ofExperimental Psychology: Human Perception andPerformance, 8,194-214.

Treisman, A., & Gelade, G. (1980). A feature integrationtheory of attention. Cognitive Psychology, 12, 97-136;

Treisman, A., Kahneman, D., & Burkell, J. (1983). Per-ceptual objects and the cost of filtering. Perception &Psychophysics, 33, 527-532.

Treisman, A., & Schmidt, H. (1982). Illusory conjunc-tions in the perception of objects. Cognitive Psychol-ogy, 14, 107-141.

Wing, A., & Allport, D. A. (1972). Multidimensionalencoding of visual form. Perception & Psychophysics,12, 474-476.

Received December 27, 1983Revision received March 27, 1984

Domjan Appointed Editor, 1986-1991

The Publications and Communications Board of the American Psychological Asso-ciation announces the appointment of Michael Domjan, University of Texas atAustin, as editor of the Journal'of Experimental Psychology: Animal BehaviorProcesses for a 6-year term beginning in 1986. As of January 1, 1985, manuscriptsshould be directed to:

i Michael Domjan, EditorDepartment of PsychologyUniversity of TexasAustin, Texas 78712

Manuscript submission patterns for JEP: Animal make the precise date of completionof the 1985 volume uncertain. The current editor, Donald Blough, will receive andconsider manuscripts until December 31, 1984. Should the 1985 volume becompleted before that date, manuscripts will be redirected to Domjan for considerationin the 1986 volume.

selective attention and the organization of visual … · american psychological association, inc....

Documents