attention and the detection of signals - hanover...

Journal of Experimental Psychology: General1980, Vol. 109, No. 2, 160-174

Attention and the Detection of Signals

Michael I. Posner, Charles R. R. Snyder, and Brian J. DavidsonUniversity of Oregon

SUMMARY

Detection of a visual signal requires information to reach a system capable ofeliciting arbitrary responses required by the experimenter. Detection latencies arereduced when subjects receive a cue that indicates where in the visual field thesignal will occur. This shift in efficiency appears to be due to an alignment (orient-ing) of the central attentional system with the pathways to be activated by thevisual input.

It would also be possible to describe these results as being due to a reducedcriterion at the expected target position. However, this description ignores impor-tant constraints about the way in which expectancy improves performance. First,when subjects are cued on each trial, they show stronger expectancy effects thanwhen a probable position is held constant for a block, indicating the active natureof the expectancy. Second, while information on spatial position improves per-formance, information on the form of the stimulus does not. Third, expectancymay lead to improvements in latency without a reduction in accuracy. Fourth,there appears to be little ability to lower the criterion at two positions that arenot spatially contiguous.

A framework involving the employment of a limited-capacity attentional mech-anism seems to capture these constraints better than the more general language ofcriterion setting. Using this framework, we find that attention shifts are not closelyrelated to the saccadic eye movement system. For luminance detection the retinaappears to be equipotential with respect to attention shifts, since costs to unex-pected stimuli are similar whether foveal or peripheral. These results appear toprovide an important model system for the study of the relationship between atten-tion and the structure of the visual system.

Detecting the presence of a clear signal controlling our awareness of environmentalin an otherwise noise-free environment is events. Although there are a number of em-probably the simplest perceptual act of which pirical approaches to the study of detection,the human is capable. For this reason it may most have not clearly separated between at-serve as an ideal model task for investigating tentional factors and sensory factors and arethe role of sensory and attentional factors in thus incapable of providing an analysis of the

relationship between the two.The classical psychophysical approach to

the use ofUniversity of Oregon. Portions of the data were near-threshold signals (e.g., Hecht, Schlaer,adapted from Chronometrk Explorations of Mind & Pirenne, 1942). This approach has been(Posner, 1978). Parts of these experiments were concerned with such stimulus factors as in-

rdllV116 Psychonomk Society> November tensity, duration, wavelength, and sensoryReqests for reprints should be sent to Michael I. organismic factors such as the degree of dark

Posner, Psychology Department, University of Ore- adaptation, retinal position of the stimulus,gon, Eugene, Oregon 97403. and so on. Evidence that a signal has been

Copyright 1980 by the American Psychological Association, Inc. 0096-3445/80/0902-0160$00.75

160

ATTENTION AND DETECTION 161

detected usually involves verbal reports bythe subjects as an indication that they areaware of the event. An effort is made to op-timize the state of attention, and it is assumedthat the organism has its attention aligned tothe input channel over which the event occurs.

A different approach to signal detection isrepresented by a body of research using thecomponents of the orienting reflex ratherthan verbal reports as an indicant of de-tection. Research on the orienting reflex isconcerned with both stimulus and contextualfactors controlling its elicitation. Sokolov(1963) has suggested that subjects build up aneural model (i.e., an expectancy) of the re-peated signal that blocks elicitation of thereflex by stimuli resembling the model. Littleis known about whether the reflex is priorto or only follows our awareness of the sig-nal. Indeed, the relatively slow times of somecomponents of the orienting reflex, such asvasodilation and galvanic skin response(GSR), may prevent precise specification ofthe temporal relation of the orienting reflexto awareness of the signal. Some componentsof the orienting reflex, such as alignment ofthe eyes, may well precede our awareness ofthe signal, whereas other components of theorienting reflex, such as changes in GSR andvasoconstriction, almost surely must follow it.

The theory of signal detection (Green &Swets, 1974) has greatly influenced studiesof detecting stimuli. One needs to distinguishbetween the mathematical theory and itspsychological application. The mathematicaltheory of signal detection is a powerful toolfor the analysis of many problems. It is anormative theory that may be used to de-scribe a large number of psychological situa-tions. However, like many tools it often pro-duces in its users some implicit assumptions.

The use of detection has involved situationsall the way from separating a pure tone inwhite noise (Green & Swets, 1974) to thetask of a radiologist locating a tumor (Green& Birdsall, 1978). It seems unlikely that thesame processes are involved in these situa-tions. Often, in addition to detecting thepresence of a stimulus, a person must identifyit in order to discriminate it from complex

backgrounds. Accordingly, sometimes it hasbeen concluded that attention aids detectionmore than would be expected from an idealobserver (Sekuler & Ball, 1977), and some-times no effects of attention are found (Lap-pin & Uttal, 1976). The one task in whichsignal detection theory is not applied is wherethere is a clear above-threshold signal in un-cluttered background. Since the signal wouldbe detected 100% of the time, the methoddoes not apply. Yet in many ways this is theperfect task for understanding the roles oforienting and detecting in their simplestforms.

Users of signal detection theory often as-sume a two-stage model of information pro-cessing in which sensory systems are coupledin series to a central statistical decision pro-cess. This view runs counter to studies thathave forced the distinction between physical,phonetic, and semantic codes of letters andwords (LaBerge & Samuels, 1974; Posner,1978). In systems involving multiple codes,changes induced in the criterion within onecode can affect the inflow of evidence to othersystems.

Another approach to the detection of sig-nals has been developing in the last severalyears. It applies the methods of mental chro-nometry (Posner, 1978) through the use ofevoked potentials, poststimulus latency histo-grams of single cells, or reaction time to tryto determine when and where central at-tentional states influence the input messageproduced by a signal. It has been shown, forexample, that independent of eye position,the instruction to attend to a particular posi-tion in visual space affects occipital recordingfor that event in comparison to a controlstimulus arising at another position withinthe first ISO msec after input (Eason, Harter,& White, 1969; Von Voorhis & Hillyard,1977). Similarly, enhancement of single cellswhose receptive field is the target of an eyemovement occurs well within 100 msec afterinput (Goldberg & Wurtz, 1972; Wurtz &Mohler, 1976). These enhancements arenot necessarily coupled to the eye movementbut are unique to the stimulus toward whichthe eye will be moved. All these studies showevidence of interaction of central systems

162 M. POSNER, C. SNYDER, AND B. DAVIDSON

with input processing. They suggest that cen-tral control modifies the stimulus evidencerather than merely providing a criterion forchoices among fixed states of evidence. Thesechronometric studies suggest that it may bepossible to study the detailed processes in-volved in the detection of a suprathresholdsignal even in an empty visual field.

By detection, we will mean the entry of in-formation concerning the presence of a sig-nal into a system that allows the subject toreport the existence of the signal by an ar-bitrary response indicated by the experi-menter. We mean to distinguish detection inthis sense from more limited automatic re-sponses that may occur to the event. Orient-ing, as we will use the term, involves themore limited process of aligning sensory(e.g., eyes) or central systems with the in-put channel over which the signal is to occur.Thus it is possible to entertain the hypothesisthat subjects may orient toward a signalwithout having first detected it. This wouldmean simply that the signal was capable ofeliciting certain kinds of responses (e.g.,eye movements or shifts of attention) but hasnot yet reached systems capable of generatingresponses not habitual for that type of signal.

The purpose of this article is to examinethe relationship of the two component pro-cesses, orienting and detecting, in the taskof reporting the presence of a visual signal.

In the course of the article, we will try toshow that central processes can seriously af-fect the efficiency with which we detect stim-uli in even the most simple of detection tasksand that the nature of these changes in ef-ficiency is such that it implies a separate at-tentional system in close interaction with thevisual system. The article is structured interms of four propositions. First, knowledgeof the location of a clear visual signal can beshown to affect the efficiency of processingsignals that arise from that location. Second,this improved efficiency is not due to a gen-eral tendency for any kind of information toimprove performance nor to an improvementin speed at the expense of accuracy, but im-plies a centrally controlled attentional system.Third, the attentional system cannot be al-located freely but can be directed only over

contiguous portions of the visual field. Fi-nally, this attentional mechanism appears notto be closely coupled to the structure of thesaccadic eye movement system nor to dif-fer between fovea and periphery.

Knowledge of Spatial Position AffectsPerformance

Evoked potential and single-cell resultsshow that when a signal occurs at a positionfor which the subject is prepared, electricalactivity is enhanced in the first 100 msecfollowing input. This result suggests that itshould be possible to observe this enhance-ment in terms of changes in detection. Thereis much evidence that knowledge of where astimulus will occur affects processing effici-ency in a complex visual field (Engle, 1971) ;Sperling & Melchner, 1978). However,there has been a great deal of dispute aboutthis fact when above-threshold signals havebeen used in an empty field. Posner, Nissen,and Ogden (1978) provided subjects with aprecue as to whether a given event wouldoccur to the left or right of fixation. One sec-ond following the cue, a .5° square wasplotted on the cathode ray tube. As shownin Figure 1, when the stimulus occurred atthe expected position (.8 probability), sub-jects' detection (simple reaction time) re-sponses were faster than following the neu-tral cue (.5 probability each side) and whenthe stimulus occurred at an unexpected posi-tion (.2 probability), they were slower. Care-ful monitoring of eye position and the use ofa single response key insure that neitherchanges in eye position nor differential prep-aration of responses could be responsible forsuch a result.1

In addition to the data reported above,

1 After having found that movements of the eyesof more than one degree occurred on less than 4%of the trials (Posner et al., 1978) and that thesetrials did not in any way change the cost-benefitresults of the study, we did not maintain carefulmonitoring of eye position in all subsequent stud-ies, although we used the same instructions andtraining to suppress movements. When monitoringwas instituted in some of the later studies, resultswere not substantially altered by the eye movementsthat were detected.


some other performance experiments havealso shown improvement in performance atexpected spatial positions. These experimentsinclude the use of signal detection measures(rf'; Smith & Blaha, Note 1), vocal reactiontime (Eriksen & Hoffman, 1973), and per-cent correct identifications (Shaw & Shaw,1977).

Nonetheless, it has been difficult in manyexperiments to obtain significant benefitsfrom knowledge of spatial position (Grindley& Townsend, 1968; Mertens, 1956; Mowrer,1941; Shiffrin & Gardner, 1972). There maybe many reasons why some studies have beensuccessful in showing improved performancefrom expected spatial positions and othersnot. One of the reasons that seemed mostlikely to us was that most experiments, otherthan ours, examined only the benefits in-volved when subjects knew something aboutthe location of a visual object when comparedto a condition where no such knowledge waspresent. Our design showed about equal costsand benefits. However, in our design, sub-jects received a cue on each trial indicatingthe most likely position of the target, whereasin most studies subjects prepared for an ex-pected position for a block of trials. We foundthat it was difficult for subjects to maintaina differential preparation for a particular lo-cation and suspected that many of the studiesexamining benefits due to knowledge ofvisual location did not find .them because thesubjects did not continue to set themselvesfor the position in space at which the signalwas most expected. To test this view, wecompared our standard cuing condition witha method in which noncued blocks were used.

Experiment 1

Method

Subjects. Six volunteers were recruited throughthe subject pool of the Center for Cognitive andPerceptual Research at the University of Oregon.All were college age and possessed normal hear-ing and vision. The subjects were run individuallyin two 1-hour sessions on consecutive days andwere paid $2 for each session.

Apparatus. All testing was conducted in anacoustical chamber. Subjects were seated approxi-mately 1.3 m in front of a cathode-ray tube (CRT)on which fixation markers, warning signals, and

300

250

IH

200

LEFT

INVALID20%

NEUTRAL50%

VALID80%

POSITION UNCERTAINTY

Figure 1. Reaction time (RT) to expected, unex-pected, and neutral signals that. occur 7° to theleft or right of fixation. (Benefits are calculatedby subtracting expected RTs from neutral, andcosts by subtracting neutral from unexpected.)

feedback occurred. The displays were viewed bi-nocularly. Four red light-emitting diodes (LEDs)were arrayed horizontally immediately below theCRT. Two LEDs were positioned 24° left andright of fixation (far stimuli). The other two stim-uli were 8° to either side of fixation (near stimuli).The LEDs were driven by 15 V through eithera S600 or a l.SJ) resistor, producing two supra-threshold intensities. Subjects indicated their re-sponses by pressing a key-operated microswitchwith the right index finger. A PDP-9 computercontrolled the timing, stimulus presentation, andcollection.

Procedure. The experimental task was a simplereaction time (RT) to the onset of an LED. Trialblocks consisted of 120 trials, including 20 catchtrials. Stimulus trials consisted of a visual warn-ing signal, a stimulus (LED), subject's response,feedback, and an intertrial interval (ITI). Sub-jects were asked to fixate the center of the CRTwhere a 1" square was displayed. Warning sig-nals, either a plus sign (+) or a digit from 1 to4, indicating one of the stimulus locations fromleft to right, were presented in the square. Follow-ing a warning interval of 1 sec, the stimulus waspresented. Subjects were encouraged to respondquickly, but not so quickly that they anticipated thestimulus. The response terminated the warningsignal and stimulus display. Feedback was the RTin milliseconds unless an anticipation had occurred,in which case the word ERROR was presented. TO re-


360

320

C 280I-tr

240 -

• • CUED

0 o BLOCKED

•O

7% 25% 79%


Figure 2. Reaction times (RT) for events of vary-ing probability. (79% = expected, 25% = neutral,and 7% = unexpected for blocked presentation andpresentation where cues are presented on eachtrial.)

duce anticipatory responses, no stimulus occurredon approximately 20 trials per block. These catchtrials consisted of only a warning signal and anITI. The proportion of catch trials was constantacross experimental conditions.

The central objective was to compare detectionlatencies when the stimulus location was cued oneach trial (mixed blocks) to a noncued situationin which subjects prepared for one location for ablock of trials (pure blocks). Two conditions wereused in the pure blocks. In the equal condition, thewarning signal was always a plus sign, and eachof the four locations was equally probable. In theunequal condition the warning signal was also aplus sign, but during each trial block one locationwas presented 79% of the time, and the other threelocations occurred 7% of the time each. At the be-ginning of each unequal block, subjects were in-formed of the most likely stimulus location. In themixed blocks (i.e., the cued condition), 20% ofthe warning signals were plus signs, indicating thatthe four locations were equally probable for thattrial. On the remaining 80% of the trials, the warn-ing signal was a digit (1, 2, 3, or 4), indicating themost probable location (79%) for that trial. Eachof the non-cued locations was equally probable(7%). In both the unequal and cued conditionssubjects were encouraged to set themselves forthe expected stimulus but not to move their eyesfrom the warning cue. No actual monitoring of eyeposition was used, since previous work had shown

that costs and benefits were not dependent onchanges in eye position (Posner et al., 1978).

Design. Each subject was tested in the threeconditions (equal, unequal, and cued) on 2 days inan ABC-CBA order, with order balanced acrosssubjects. There were two blocks in the equal con-dition, four blocks in the unequal condition (onefor each position most likely), and two blocks inthe cued condition on each day. Each session foreach subject contained a different random order ofthe four blocks within the unequal condition.

Results and Discussion

The results of these mixed and pure blocksare shown in Figure 2. Note that once againwhen the cuing technique was used, we ob-tained very significant costs and benefits overa neutral condition. However, in the pureblock technique, only the costs were signifi-cantly different from the neutral condition.There was no evidence of benefit.2

We attribute this failure to find benefit inthe expected position over the neutral condi-tion to the tendency of subjects to avoid thetask of placing their attention at the expectedposition when they were not cued to do soon each trial. It is not clear why benefits aremore labile than costs. However, since bothcosts and benefits are aspects of our knowl-edge of the position of an expected signal, itis clear that the difference between the bene-fit trials and the cost trials is a legitimate wayof asking whether expectancy changes theefficiency of performance of signals arrivingfrom expected versus unexpected conditions.The failure of most other paradigms to ex-amine the cost of unexpected positions inspace makes them far less sensitive than thetechniques we have outlined. This, togetherwith the general use of blocking rather thancuing, helps to reconcile several of the con-flicts in the literature.

Experiment 2: Attention Is Involved

Investigators using the signal detectiontheory to guide work on detecting signalsoften argue that any information provided to

2 This experiment was subsequently replicatedwith 12 additional subjects in the same design, ex-cept that the LEDs were 2° and 8° from fixation.The results were identical.


the subject about a signal will be useful indisentangling the signal from backgroundnoise. For this reason, evidence that someparticular type of information, for example,about the location of a signal, improves per-formance is not taken to mean that there isany special mechanism associated with theutilization of that information. Lappin andUttal (1976) have argued that knowledge ofany orthogonal stimulus parameter ought toimprove detection of that stimulus (p. 368).In their experiments they use a high level ofbackground noise and ask the subjects to de-tect a line within the noise. Detection in-volves a difficult discrimination betweenbackground and signal. They find that thesubject's information about the location ofthe line does improve performance, but notmore than would be expected from a modelin which no attentional assumptions are used.From this they conclude that the demonstra-tions of costs and benefits of the type indi-cated above are not evidence in favor ofspecific attentional mechanisms.

According to our view, evidence that onlysome types of information serve to improveperformance would indicate that our effectsare not due to general knowledge serving toallow separation of signal from noise. For ex-ample, consider a comparison of providingsubjects with information about the shape ofa stimulus with providing information aboutthe location of the stimulus. It seems clearthat in the Lappin and Uttal experiment,knowledge of the target shape would affectperformance. This fits with the notion thatinformation about the target's shape servesto disentangle the signal from noise. On theother hand, in our experiment it seems some-what unlikely that information about shapewould improve detection of signals.

Method

Subjects. Twelve volunteers were recruited inthe same manner as in Experiment 1. The sub-jects were paid $2 for each of three 1-hour sessionsrun on consecutive days. Each session consisted ofeight blocks of 130 trials.

Stimuli. Warning signals, stimuli, and feedbackwere presented on the CRT. Warning signals oc-curred at the center of the CRT. The stimulus,one of 10 capital letters selected at random, was

presented 7° to the left or right of the warningsignal.

Procedure. The experimental task was a simpleRT to the occurrence of a letter. Each trial beganwith a warning signal indicating either the formor location of the stimulus. The stimulus was pre-sented after a variable warning interval that rangedbetween 800 and 1200 msec, Subjects were en-couraged to respond quickly but not to anticipatethe stimulus. About 25% of the trials were catchtrials in which no letter was presented. As in Ex-periment 1, catch trial rates were constant acrossconditions. Feedback consisted of the RT in milli-seconds, or the word ERROR if an anticipation hadoccurred.

The primary objective was to compare the ef-fectiveness of location and form cues on simpledetection. Location cues were left or right arrows.Following a location cue, the stimulus occurred onthe indicated side on 80% of the trials. The neu-tral location cue was a plus sign. Following thiscue, each location was equally probable. The formcue was one of the letters that were used as stim-uli. On 80% of the trials following this cue, thestimulus was the indicated letter. The form cuealways occurred with either a neutral or an infor-mative location cue. Half of the form cues werepresented slightly below a plus sign. This warn-ing signal indicated that the cued letter was themost probable stimulus but did not indicate itslocation (i.e., each location was equally probable).On the remaining trials the form cue occurred be-low the left or right arrow, informing subjects ofboth the form and location of the stimulus. Sincethe location and form cues were each valid on 80%of the trials, the combined form and location cuewas valid on 64% of the trials. On 16% of the trials,the cued letter occurred in the unexpected location.On 16% of the trials, an unexpected letter occurredin the cued location. Finally, on 4% of the trials,an unexpected letter occurred in the unexpected lo-cation. Each type of warning signal (plus signalone, arrow alone, letter with plus sign, letter witharrow) occurred equally often. Subjects were en-couraged to use the warning signals to prepare forthe stimulus but not to move their eyes from thewarning cue.

Results and Discussion

The results of this experiment are shownin Table 1. Clearly, information about the lo-cation of the letter improves performance,but information about the form does not.3

3 It should be noted that the prime reduced theletter uncertainty from 10 alternatives, whereasthe spatial uncertainty had only 2 alternatives. Itseems unlikely that a different result would haveobtained had only 2 letters been used, however.


Table 1Mean Reaction Time for Expected,Unexpected, and Neutral Form andLocation Cues

Location

Form Expected Neutral Unexpected M

ExpectedNeutralUnexpectedM

247252248249

263271263266

292299299297

267274270

Note. Time is measured in milliseconds.

Another form of objection to our studiesis to suppose that changes in the latency ofprocessing the stimulus arising at the ex-pected location are a result of changes inthe amount of information that the subjectssampled from the expected location. Considera comparison of the neutral trials with thetrials cued by an arrow. In the latter, sub-jects may decide to reduce their criterion forpressing the key at the risk of making an in-creased number of anticipations. Indeed, wegenerally find that conditions involving thearrow do show an increased number of an-ticipatory responses over those times whenthe plus sign is used. This is evidence of ashift in amount of evidence that the subjectsrequire to respond. However, this kind ofshift cannot account for differences in costplus benefit, since both of these RTs are fromthe arrow conditions, and subjects cannotdifferentially prepare prior to making the re-sponse. One might suppose that in some waythe subjects are able to reduce their criterionwhen the stimulus arises from the particularposition in space that was cued. It is possibleto test whether improvement in reaction timeobtained from knowledge of the location ofthe stimulus is accompanied by an increasein error. To do this we used a choice reactiontime task.

Experiment 3

We modified our standard simple reactiontime method (Posner et al., 1978) by pro-viding the subjects with a toggle switch thatmoved up or down. The cues were left orright arrows or a plus sign. The imperative

stimulus was a .5° square of light that oc-curred 7° from fixation and either below orabove the line on the scope indicated by thecue stimulus. If it occurred above, the subjectwas required to move the toggle switch up,and if below, the toggle switch was to bemoved down. This was a highly compatiblestimulus-response combination that did notrequire a great deal of learning by the sub-ject. Eight subjects were run for five blocksof 96 trials on each of the 2 days.

The results for Day 2 are shown in Fig-ure 3. It is clear that we did find costs andbenefits in reaction time in the same direc-tion but not in as great a magnitude as hadbeen found in the simple reaction time detec-tion experiments. Analysis of variance indi-cated that both costs and benefits are signifi-cant.* There clearly is no significant differ-ence in the error rates. Error rates on costtrials are somewhat larger than error rates inneutral or benefit trials. There is no evidencethat the reaction time results are producedby an opposite effect on errors. A speed-accuracy tradeoff is not a necessary factorin producing the costs and benefits found inour experiment.

Experiment 4

It seemed important to determine the re-lationship of our results using luminancedetection to those obtained when subjects arerequired to identify a target. In Experiment2 it was shown that a subject's knowledgeabout the form of the target did not influenceluminance detection. These findings suggestthat luminance detection may be a simplerdomain in which to examine the effects ofset on performance than the more frequentlystudied tasks in which it is important to iden-tify or match forms.

4 There is also an interaction apparent in thegraph between the target location (up vs. down)and position uncertainty. This probably resultsfrom a tendency to associate an unexpected posi-tion with a downward response. Presumably thereis also a tendency to associate an expected positionwith an upward response. Despite this complica-tion, the main result of the experiment is to showhighly significant effects of knowledge of spatialposition for both choice responses.


To examine this question we displayedfour boxes arrayed around a central fixationpoint. The maximum visual angle of the dis-play was about 1.5° so that all stimuli werefoveal. Subjects saw either a neutral warn-ing cue or an arrow pointing to one of thefour positions. Following the neutral cue, astimulus was equally likely to appear at anyof the four positions and following an arrowthe target appeared at the cued position 79%of the time and at the other positions 7% ofthe time. The stimulus could be the digits4 or 7 or the letters D or Q. The subjects'task was to respond to designated targetstimuli.

In pilot research we provided subjectswith only a single key that they were topress whenever a digit was presented. If aletter was presented, they were to refrainfrom pressing a key. In this paradigm RTsto the expected position were very fast, buterror rates were always much higher than inunexpected or neutral trials. Subjects foundit very difficult to withhold responding whena nontarget occurred in the expected posi-tion. This result indicates that there is astrong tendency to react with a false alarmto a visual event occurring in an expectedposition. Subjectively, it felt as if one wereall set to respond when an event ocurred inthe indicated position, and it was very frus-trating to inhibit the response while waitingto determine if it was a digit (target). Whenan event occurred in an unexpected position,it felt as though the answer was already pres-ent by the time one was ready to make a re-sponse. These subjective impressions fit verywell with the idea that the attentional systemis responsible for releasing the responserather than for the accrual of informationrelevant to the decision that a target waspresent.

It was relatively easy to show that costsand benefits were not due entirely to rapidbut inaccurate responses. We simply pro-vided subjects with a second key so that oneach trial they were required to decidewhether the target was a letter or digit. Fig-ure 4 indicates the results from 14 subjectsrun in such a study for 2 days. Half the sub-jects were presented with a brief target

480

460

440

400 -O O DOWN• • UP

(% ERROR)

30% 50% 80%


Figure 3. Reaction times (RT) for expectedneutral (50%), and unexpected (20%) stimulus lo-cations. (The task is to determine if the stimulusis above [up response] or below [down response]the center line. Error rates are in parentheses.)

masked after 40 msec (short duration) andhalf with a target remaining present untilthe response. For reaction time there areclear costs when the stimulus occurs in theunexpected position and benefits when it oc-curs in the expected position in comparisonwith the neutral control. On the other hand,error rates are constant over the variouspositions. These results argue clearly thatsubjects did not simply sacrifice accuracyfor speed when the stimulus occurred in theexpected location. This finding is incompati-ble with the view that central decision pro-cesses are responsible for setting a criterionfor the response, since that implies thatmore rapid responding will be associatedwith increased error.

The results of the pilot study and of Ex-periment 4 also indicate to us that there arequite different processes present when sub-jects are required merely to detect a lumi-nance change from those present when theymust identify the stimulus. The false alarmsfound in the pilot study were far greater thanwere found in any study involving the detec-


460

440

I 420H(E

400

(.08)

I • LONG DURATION

r7% 25%" 79%


Figure 4. Reaction times (RT) for expected (79%),neutral (25%), and unexpected (7%) positions.(The task requires separate responses for lettersand digits. Short duration = 40 msec masked presen-tation. Long duration = stimulus present until re-sponse is made. Error rates are in parentheses.)

tion of a stimulus. It is as though the occur-rence of a luminance change at the expectedposition gives rise to detection of an event.It is the speed of this detection that we havebeen measuring in our previous work. If thesubject is given only one key, there is a verystrong bias to use the act of detecting theevent as the basis for pressing the key. If thekey press is to be made to only one class ofstimuli, it is difficult to withhold the response.On the other hand, if subjects have to makea choice between keys, they are able to in-hibit rapid responses and still obtain benefitsfrom the cue.

These results all suggest that luminancedetection is facilitated when subjects knowwhere in space a stimulus will occur. Theyalso indicate that such facilitation is not dueto a bias introduced by the tendency to re-spond quickly and inaccurately to stimuli oc-curring at the expected position. On theother hand, they also suggest that the resultsof luminance detection cannot necessarily begeneralized to studies in which subjects are

required to identify the form present at aparticular position. Although attention isquickly available at the expected position,this may result either in quick but error-prone reactions or in improved speed withoutincreases in error, depending upon how thetask is structured.

Although we have not done any formalcomparisons, it seems obvious that the sizeof the effects in the choice RT tasks are muchsmaller than we have typically obtained insimple RT. This may seem counterintuitive,since the actual RTs are much greater in thechoice tasks. We believe that this is due tothe necessity of the subject's switching at-tention from the spatial location indicated bythe cue to the internal lookup processes thatidentify (e.g., digit) or determine (e.g.,above) the discriminative responses. Spatialcues are very effective for simple RT to lu-minance increments because this task doesnot require determining what the event is be-fore responding, since subjects are requiredto respond to any event. Whether a spatialcue is effective in a more complex task willdepend upon the details of the task and thecompeting stimuli. Spatial cues will be ofgreat help in complex cluttered fields becausethey tell the subjects which stimuli are to bedealt with; in an empty field they may ormay not help, depending upon the difficultyof reorienting from the location to the in-ternal lookup of item identity.

Another perspective on our results fromthe point of view of signal detection theoryis to suppose that they depend upon the re-ciprocal nature of the stimulus conditionsthat we impose upon the subjects (Duncan,1980). If subjects follow the correlations inthe experiment, they may seek to raise theircriterion at the unexpected position andlower it at the expected position. This mighthave nothing to do with capacity or atten-tional limitations but would simply be anadaptation to the experimental contingencies.This view is more difficult to deal with. It ispossible to design a study without intro-ducing a contingency by, for example, pre-senting a stimulus that occurs with equallikelihood to the left, right, or both positions.However, such an experiment is probably


not sufficient to dispose of the more generalidea that performance in these tasks is medi-ated by independent shifts in criterion at dif-ferent positions in space and not by the al-location of any central mechanism. It is tothis question that the next section of thearticle is addressed.

Attention Cannot Be Allocated at Will

Recently, Shaw and Shaw (1977) haveproposed that subjects can allocate their at-tention pretty much at will over the visualfield. Shaw and Shaw presented letters atone of eight positions in a circular array. Inone condition, the positions varied in theprobability with which a target would occur.Performance was compared with a conditionin which targets occurred at all eight posi-tions with equal likelihood. Subjects showedsignificant costs and benefits in detectionaccording to the assigned probabilities. Fromthis, Shaw and Shaw argued for a model inwhich subjects were able to allocate a limited-capacity attentional resource to differentareas of the visual field. While their resultsare consistent with allocation of a limited-capacity mechanism, they would also be con-sistent with the sort of view discussed in thelast paragraph. It could be that subjects areable to set criteria for different positions inthe visual field according to the probabilitythat those positions will be sampled. How-ever, there is a serious problem with this in-terpretation. The results of Shaw and Shawcould also be obtained if subjects sometimesattend to one position in space and some-times to another, and these probabilitiesmatch those assigned to target presentation.

Our goal was to determine whether sub-jects were able to allocate their attention todifferent positions on a given trial. To dothis we gave subjects both a most frequentposition and a second most frequent positionon each trial. We examined their RTs to thesecond most likely position in comparison tolower frequency positions to see if they couldallocate attention simultaneously both to themost frequent and the second most frequentevents.

MethodExperiment 5

Subjects. Twelve subjects participated in two1-hour sessions on consecutive days. Experiment5A involved 12 additional subjects and Experi-ment SB 7 subjects. All were paid for their par-ticipation.

Apparatus. The apparatus from Experiment 1,including the LED displays, was used in this study.The LEDs were positioned either 2° (foveal stim-uli) or 8° (peripheral stimuli) from fixation, withtwo LEDs on either side of fixation.

Procedure. The experimental task was a simpleRT to the onset of an LED. Trial blocks con-sisted of 100 trials, including catch trials. Subjectsfixated a 1° square in the center of the cathode-ray tube. Warning signals, either a plus sign or adigit from 1 to 4 indicating one of the stimulus lo-cations from left to right, were presented in thesquare. After a variable warning interval, the stim-ulus occurred. Approximately 25% of the trialswere catch trials in which no LED was presented.The feedback consisted of the RT in millisecondsor, in the event of an anticipation, the word ERROR.

On Day 1 subjects were seated in the test cham-ber and allowed to adapt to the dark for about 5minutes before testing was begun. Prior to eachblock a most likely (65%) and next most likely(25%) stimulus location were indicated on thecathode-ray tube. Subjects were asked to rememberthese positions throughout the block and to try toprepare for stimuli at these locations on those trials(80%) when a digit appeared as the warning sig-nal. The digit indicated the most likely positionduring that block, with the stimulus positions num-bered from 1 to 4 from left to right. On trials pre-ceded by a plus sign as a warning signal (20%),subjects were told that all four stimuli would beequally likely to occur and were asked to preparethemselves accordingly. Subjects were also in-formed that the first four blocks would be prac-tice, and the final three blocks on Day 1, plus nineblocks on Day 2, would be test blocks.

On Day 2, subjects were again shown the ap-paratus, task instructions were reviewed, and about5 min. were allowed for dark adaptation. Followingtesting, subjects were asked for their impressions ofthe helpfulness of advance information concerningstimulus location and whether they had felt theycould prepare for stimuli at two locations.

Experiment 5A was an exact replication of Ex-periment 5 except that blocks of trials in which onesignal had a probability of .64 and the other threehad probabilities of .12 were also included. In ad-dition, Experiment 5A was run under light-adaptedconditions. Experiment SB was identical to 5A butrun under conditions of dark adaptation as in Ex-periment 5.

Design. Each subject received the same set offour practice blocks, which sampled the four posi-tions as most likely and as next most likely. Each


280

270

260

250

O

'O Non-adjacent

Adjacent

Overall

r5% 23% 65%


Figure 5. Reaction times (RT) to unexpected (5%),second most likely (25%), and most likely (65%)events as a function of the adjacency of the lessprobable events to the most probable (65%) event.

subject was then tested at all twelve combinationsof most likely and next most likely stimuluslocations.

The most likely position was cued on each trialby the central digit while the next most likely posi-tion remained constant for three consecutive blocks,with order of the four positions counterbalancedacross subjects. Within each three-block set, orderof the most likely positions was also counter-balanced.6

Resists and Discussion

The data of all three experiments are givenin Table 2. The statistical analysis of Experi-ment 5 showed that both the most likely andthe second most likely target position weresignificantly faster in reaction time than thetwo least likely positions. In addition, fovealevents showed some advantage over periph-eral events, and intense stimuli showed someadvantage over weak stimuli.

However, the important result is a com-parison of the reaction times to the secondmost likely position (25%) and least likelyposition (5%) when the former was eitheradjacent to or remote from the 65% position.This is shown in Figure 5. The results are

really quite clear-cut. When the second mostlikely target position was adjacent to themost likely target position, its RT resembledthe most likely targets. There was a slight(5 msec) nonsignificant advantage to themost likely over the second most likely posi-tion in this condition. However, when thesecond most likely was separated by a posi-tion from the most likely, its reaction timeresembled the least likely (5%) position.This constellation of results was independentof whether the two most likely events oc-curred at the central position or whether oneof them occurred at the periphery. These re-sults suggest that for detection-, it is not pos-sible for subjects to split their attentionalmechanism so that it is allocated to two sepa-rated positions in space.

Experiment 5 did not contain a conditionin which there was only one likely event.Thus we were unable to tell whether subjectswere reducing their efficiency in detecting themost likely event. This condition was presentin both Experiment 5A and SB. From Ta-ble 2 it is clear that the requirement to giveattention to a second most likely event hadno effect on RT to the most likely event. Inboth experiments the blocks in which therewas and was not a secondary focus had thesame detection RTs for the most likely event.

In other ways Experiments 5A and SB area replication of Experiment 5 except theinteraction between adjacency and proba-bility (5% vs. 25%) was not statistically sig-nificant. Overall, the 25% event is only 5msec faster than the 5% event when they areboth remote from the most likely event.There is no evidence of an ability to divideattention. When the events are adjacent tothe 65% event, the 25% event has a 16-msecadvantage. This advantage is statistically sig-nificant in each study.

8 The use of a blocking rather than a cuingtechnique for the second most likely event was madenecessary by the difficulty we found in getting sub-jects to process two target position cues on eachtrial. Since the 25% target position (second mostlikely) is compared in RT to the 5% target posi-tion, differences should reflect a sum of costs andbenefits that do show up in the blocking method asshown in Experiment 1.


Table 2Splitting Attention: Reaction Time as a Function of Stimulus Event and Expectancy Condition.

Expectancy

Experiment One only Most likely 25% adj. 5% adj. 25% non-adj. 5% non-adj.

Stimulus central (Positions 2-3)

55ASB

M

55ASB

M

M

256270263

255301278

270

252253269258

Stimulus266256289274

266

257273275268

peripheral (Positions272272312285

Overall276

270281280277

1-4)272284328295

286

272284296284

288293328303

294

276294295288

288303337309

299

Note. Time is measured in milliseconds. Adj. = adjacent.

Overall our results suggest severe limitsin the ability of subjects to assign attentionto a secondary focus in addition to a pri-mary focus. Clear evidence for such an abilityoccurs only when the secondary focus isadjacent to the primary focus.

This finding favors the view of a unifiedattentional mechanism under the conditionsof this experiment. These conditions includethe use of a luminance detection task and theblocking of the second most likely targetposition.

Attention and Visual System Structure

The results summarized so far argue thatsubjects' knowledge about where in space thesignal will occur does affect processing effi-ciency both in facilitating latencies at the ex-pected position and retarding them at theunexpected positions. Our results suggest anattentional mechanism that cannot be allo-cated freely to positions in space but appearsto have a central focus that may vary in sizeaccording to the requirements of the experi-ment. These findings are consonant with theidea of attention as an internal eye or spot-light. The metaphor of attention as a kind ofspotlight has been used by Norman (1968),among others.

It seems useful to summarize the relation-

ship between the attentional spotlight andfeatures of vision such as saccadic movementsand foveal versus peripheral acuity. Our re-sults have shown that orienting is not de-pendent upon actually moving the eye. More-over, the extent of benefit to a signal is notaffected by its distance from the fovea (Pos-ner, 1978, Figure 7.9) from .5°-25° of visualangle. This finding for detection differs mark-edly from one obtained by Engle (1971) fora task demanding a high level of acuity.Engle required subjects to find a single formembedded in a complex visual field. He pro-vided both a fixation point and a point awayfrom fixation where attention was to be con-centrated. He found that the field of highacuity (conspicuity) for the ability to iden-tify the target stimulus included the foveabut was elongated in the direction of the sub-ject's attention. This result contrasts sharplywith our results. In our detection experi-ments when subjects are told to attend awayfrom the fovea, the point of maximum speedof reaction shifts to surround the area of at-tention and does not include any specialability at the fovea itself. Costs of unexpectedfoveal stimuli are quite comparable to thosewith unexpected peripheral stimuli (Posner,1978, Figure 7.9).

This equipotentiality of attention with re-spect to visual detection shows that the at-


tentional spotlight is not related to the fieldof clear foveal vision. Moreover, taken withEngle's study it shows that attention cannotcompensate for structural deficiencies in acu-ity. Attending away from the fovea does notcompensate for the lack of acute vision inthat part of the retina, though it does pro-duce a complete shift in the speed of detectionof luminance changes in that area of thevisual field.

Our results may seem paradoxical becauseof the strong belief that attention is tiedclosely to the fovea. In the real world, we arealways moving our eyes to stimuli that in-terest us, and thus we are habitually payingattention to the stimuli to which we are look-ing. We found that this belief affected thestrategies our subjects employed when eventscould be either foveal or peripheral in mixedblocks (Posner, 1978, Figures 7.10 and7.11). When subjects were cued as to whichside of the field was most likely, they uni-formly prepared for the peripheral (7°)stimulus and not the foveal (.5°) stimulus.The costs and benefits for peripheral stimulusin such mixed blocks were the same as whenonly peripheral stimuli could occur (pureblocks). The benefits for foveal events weregreatly reduced in blocks when they weremixed with peripheral events. This showsthat subjects behave as though peripheralevents benefit from attention, whereas fovealevents do not require attention. This strategyis quite wrong in our task, since both fovealand peripheral events show equal costs andbenefits in pure blocks. Nonetheless, it is areasonable strategy to carry over from thereal world in which attention is closely as-sociated with the fovea.

Conclusions

The conclusions from this series of experi-ments are of two kinds. The first kind issomewhat general and concerns the theoreti-cal framework most appropriate for the studyof detection. In the introduction we outlinedfour alternative approaches based uponwhether a distinction is made between centraldecision and sensory processes and, if it is,whether the two are thought to be indepen-

dent and serial or interactive. Our experi-ments have shown clearly that the subject'sknowledge about where in space a stimuluswill occur affects the efficiency of detection.Moreover, the kind of effect one finds (costsalone or costs and benefits) depends uponwhether a general set is maintained overmany trials or is precued on each trial. Thesetwo results indicate that central factors influ-ence the efficiency of detection. By them-selves these results merely reinforce a pointmade at the advent of signal detection theoryconcerning the importance of taking centralfactors into explicit consideration as a part ofunderstanding sensory processes.

How shall these central factors or cogni-tive factors be viewed ? The idea of separatesensory and decision stages suggests an es-sentially noninteractive mode. Cognitive ef-fects are seen to establish logical criteria forthe selection of sensory evidence. These se-lection criteria modify our reports about theevidence but not the evidence itself. Our datasuggest an interactive framework, becausethey show serious constraints upon the wayknowledge of a signal can aid detection. Ithelps us to know where a signal will occurbut not the form in which it occurs. It helpsto know that a stimulus will occur in ad-jacent regions of space, but we cannot pre-pare efficiently for two separated regions.Knowledge of where a stimulus will occurproduces benefits when it is used actively(cued) but not when it is used to maintaina general set (blocked). None of these re-sults disprove the signal detection languagebut all suggest constraints upon how ourknowledge affects processing that go beyonda general improvement to be found by a logi-cal selection criterion. Thus, our data seem tolead one to view detection as an interactionbetween the structure of the visual systemand the structure of the attentional system.

The second set of conclusions deals withthe structure of the attentional system im-plied by our experimental results. It is herethat our findings are more specific. Attentioncan be likened to a spotlight that enhancesthe efficiency of detection of events within itsbeam. Unlike when acuity is involved, the ef-


feet of the beam is not related to the fovea.When the fovea is unilluminated by atten-tion, its ability to lead to detection is dimin-ished, as would be the case with any otherarea of the visual system. Subjects' assump-tion that the fovea is closely coupled to at-tentional systems is a correlation they carryover from everyday life. It is usually appro-priate, because we move our eyes to thosethings in which we are interested, but whenthis correlation is broken, the fovea has nospecial connection to attention. Nor are wegood at dividing the attentional beam so asto simultaneously illuminate different cornersof our visual space. This failure to find anability to divide attention contrasts sharplywith views arising from more complex tasks(Moray, 1967; Shaw & Shaw, 1977) thatstress attentional allocation. Perhaps the dif-ference lies in the complex pathway-activa-tion processes involved when linguistic stim-uli are to be identified before responding andin their use of more than one stimulus event.

How is this attentional system brought tobear upon stimulus input? We distinguishbetween two different aspects of the atten-tional system. The first we call orienting.Orienting involves the direction in which at-tention is pointed. Since the visual cortex isorganized by spatial position, orienting canbe viewed as the selection of a position inspace. However, orienting may also involvethe selection of a modality, and within mo-dalities it may differ based upon the natureof the organization of information in thatsensory system (Posner, 1978). When inputinvolves more than one modality, it is pos-sible to compare orienting by modality withorienting by position in real space. Whenthis is done (Posner et al., 1978), modalityinformation dominates over spatial position,supporting the view that the sensory path-ways matter more than a reconstructed in-ternal model of space. Orienting, as we havedescribed it, may be an entirely central phe-nomenon without any overt change in eyeposition. Usually the eyes do follow the di-rection of our attention, however. Orienting,as we have described it, cannot be identifiedwith the orienting reflex. The orienting re-flex doubtless includes orienting in the sense

we have used it, but it also involves the op-eration that we call detecting. By detectionwe mean the contact between the attentionalsystem and the input signal, such that arbi-trary response to it can be made.

In our experiments we provide the sub-jects with cues that allow them to performthe act of orienting. When this is done, de-tection proceeds more quickly. In the realworld it is usual for a signal to produce bothorienting (covert and often overt) and detec-tion. Since the efficiency of detection is af-fected by orienting, orienting must either bein parallel or precede detecting. It mightseem paradoxical that orienting toward asignal could precede or occur at the same timeas detecting the signal. This paradox is simi-lar to the problem of subception. How can weorient to something that is as yet undetected ?The answer to both paradoxes lies in thespecific nature of the attentive system thatunderlies detection. Much of our informationprocessing does not depend upon this system.It is now well documented that complexsemantic analysis can go on outside this sys-tem (Posner, 1978). Attention is importantfor nonhabitual responses such as are impliedby detection responses. Habitual responsessuch as orienting the eyes to an event oraligning attention to the stimulated pathwaydo not appear to require support from thissystem.

Our experiments also suggest several di-rections for the analysis of detection. If themovement of attention can be time-locked toan input event, it should be possible to deter-mine (a) the latency with which attentioncan be switched, (b) whether the time toreach the target is a function of distance, and(c) how such attention switches relate to thearticulation of visual space and to the move-ment of the eyes.6 It is clear that the generalframework for viewing detection experi-ments outlined in this article is quite con-

6 While this article has been in press much of thework outlined here has been accomplished. For adiscussion of movements of covert attention seeShulman, Remington, and McLean (1979). Abroader treatment of the relationship between overtand covert attention movements may be found inPosner (1980).


sistent with the ideas developing from evokedpotential and single-cell work. A more de-tailed integration of the two approaches mayeventually enhance our knowledge of thenature of attention.

Reference Note

1. Smith, S. W., & Blaha, J. Preliminary reportsummarising the results of location uncertainty:Experiments I-Vll. Unpublished experiments,Ohio State University, 1969.

References

Duncan, J. The demonstration of capacity limita-tion. Cognitive Psychology, 1980, 12, 75-96.

Eason, R. G., Harter, R., & White, C. T. Effectsof attention and arousal on visually evoked cor-tical potentials and reaction time in man. Physi-ology and Behavior, 1969, 4, 283-289.

Engle, F. L. Visual conspicuity, direction attentionand retinal locus. Vision Research, 1971, 11,563-576.

Eriksen, C. W., & Hoffman, J. E. The extent ofprocessing of noise elements during selective en-coding from visual displays. Perception & Psy-cophysics, 1973, 14, 155-160.

Goldberg, M. E., & Wurtz, R. H. Activity of su-perior colliculus in behaving monkeys : II. Effectsof attention on neuronal response. Journal ofNewophysiology, 1972, 35, 560-574.

Green, D. M., & Birdsall, R. G. Detection and rec-ognition. Psychological Review, 1978, 85, 192-206.

Green, D. M., & Swets, J. A. Signal detectiontheory and psychophysics. Huntington, N.Y.:Krieger, 1974.

Grindley, C. G., & Townsend, V. Voluntary at-tention in peripheral vision and its effects onacuity and differential thresholds. QuarterlyJournal of Experimental Psychology, 1968, 20,11-19.

Hecht, S., Schlaer, S., & Pirenne, M. H. Energy,quanta and vision. Journal of General Psychology,1942, 25, 819-840.

LaBerge, D. S., & Samuels, J. Toward a theoryof automatic information processing in reading.Cognitive Psychology, 1974, 6, 293-323.

Lappin, J. S., & Uttal, W. R. Does prior knowl-edge facilitate the detection of visual targets inrandom noise? Perception &• Psychophysics, 1976,20, 367-374.

Mertens, J. J. Influence of knowledge of target lo-cation upon the probability of observation ofperipherally observable test flashes. Journal ofthe Optical Society of America, 1956, 46, 1069-1070.

Moray, N. Where is attention limited? A surveyand a model. Ada Psychologica, 1967, 27, 84-92.

Mowrer, O. H. Preparatory set (expectancy) —Further evidence of its central locus. Journal ofExperimental Psychology, 1941, 28, 116-133.

Norman, D. A. Toward a theory of memory andattention. Psychological Review, 1968, 75, 522-536.

Posner, M. I. Chronometric explorations of mind.Hillsdale, N.J.: Erlbaum 1978.

Posner, M. I. Orienting of attention. QuarterlyJournal of Experimental Psychology, 1980, 32,3-25.

Posner, M. I., Nissen, M. J., & Ogden, W. C.Attended and unattended processing modes: Therole of set for spatial location. In H. L. Pick &I. J. Saltzman (Eds.), Modes of perceiving andprocessing information. Hillsdale, N.J.: Erlbaum,1978.

Sekuler, R., & Ball, K. Mental set alters visibilityof moving targets. Science, 1977, 198, 60-62.

Shaw, M., & Shaw, P. Optimal allocation of cog-nitive resources to spatial location. Journal ofExperimental Psychology: Human Perceptionand Performance, 1977, 3, 201-211.

Shiffrin, R. M., & Gardner, G. T. Visual processingcapacity and attentional control. Journal of Ex-perimental Psychology, 1972, 93, 72-82.

Shulman, G. L., Remington, R. W., & McLean,J. P. Moving attention through space. Journal ofExperimental Psychology: Human Perceptionand Performance, 1979, 5, 522-526.

Sokolov, E. N. Perception and the conditioned re-flex. New York: Macmillan, 1963.

Sperling, G., & Melchner, M. J. Visual search,visual attention and the attention operating char-acteristic. In J. Requin (Ed.), Attention and per-formance VII. New York: Academic Press, 1978.

Von Voorhis, S., & Hillyard, S. A. Visual evokedpotentials and selective attention to points inspace. Perception & Psychophysics, 1977, 22,54-62.

Wurtz, R. H., & Mohler, C W. Organization ofmonkey superior colliculus: Enhanced visual re-sponse of superficial layer cells. Journal of Neu-rophysiology, 1976, 39, 745-765.

Received October 30, 1978 •

attention and the detection of signals - hanover...

Documents