attention in character-classification tasks: evidence for ... · attention in...

32
Journal of Experimental Psychology: General 1978, Vol. 107, No. 1, 32-63 Attention in Character-Classification Tasks: Evidence for the Automaticity of Component Stages Gordon D. Logan Erindale College, University of Toronto, Mississauga, Canada SUMMARY Most theories of attention predict that engaging a subject in demanding con- current activity will increase choice reaction time by an amount proportional to the attention demands of the choice task. The attention demands of mental op- erations within the choice task can be assessed by extending Sternberg's additive- factors method. Parameters associated with those processing stages that demand attention will interact with the amount of concurrent activity (i.e., will have greater effects with concurrent activity than without), but parameters associated with automatic processing stages will not (i.e., the joint effects will be additive). This analysis was applied to eight character-classification experiments. The classification task was either performed alone or in the retention interval of a short-term memory task which required ordered recall of 7 digits. According to the extended additive-factors method, the important effects are the interactions of classification-task parameters with memory load. The first two experiments reported validate the method with a memory-search task. Interactions were obtained between memory load and target-set size whose magnitude depended on the amount of practice with specific target sets. Inferences about attention demand and automaticity were shown to be consistent with those drawn from another criterion for automaticity, based on a different aspect of the data. The remaining six experiments examined four stages sufficient to perform a visual-search task (encoding, comparison, decision, and response selection). Memory load did not interact with parameters associated with encoding, decision and response-selection stages, and of three parameters associated with the com- parison stage (target-set size, array size, and the presence or absence of a bar marker indicating the target's position), only target-set size interacted with memory load (Experiments 1 and 2). It was concluded that most of the process- ing involved in character classification does not require attention, and this was taken as evidence against models that identify attention with specific processing- structures. The results were interpreted as supporting the view that attention is a limited capacity to activate processing structures internally, and the role of at- tention in preparing, maintaining, and executing mental operations is discussed. Copyright 1978 by the American Psychological Association, Inc. All rights of reproduction in any form reserved. 32

Upload: phungcong

Post on 08-Sep-2018

227 views

Category:

Documents


0 download

TRANSCRIPT

Journal of Experimental Psychology: General1978, Vol. 107, No. 1, 32-63

Attention in Character-Classification Tasks: Evidence forthe Automaticity of Component Stages

Gordon D. LoganErindale College, University of Toronto, Mississauga, Canada

SUMMARY

Most theories of attention predict that engaging a subject in demanding con-current activity will increase choice reaction time by an amount proportional tothe attention demands of the choice task. The attention demands of mental op-erations within the choice task can be assessed by extending Sternberg's additive-factors method. Parameters associated with those processing stages that demandattention will interact with the amount of concurrent activity (i.e., will havegreater effects with concurrent activity than without) , but parameters associatedwith automatic processing stages will not (i.e., the joint effects will be additive).

This analysis was applied to eight character-classification experiments. Theclassification task was either performed alone or in the retention interval of ashort-term memory task which required ordered recall of 7 digits. According to theextended additive-factors method, the important effects are the interactions ofclassification-task parameters with memory load.

The first two experiments reported validate the method with a memory-searchtask. Interactions were obtained between memory load and target-set size whosemagnitude depended on the amount of practice with specific target sets. Inferencesabout attention demand and automaticity were shown to be consistent with thosedrawn from another criterion for automaticity, based on a different aspect ofthe data.

The remaining six experiments examined four stages sufficient to perform avisual-search task (encoding, comparison, decision, and response selection).Memory load did not interact with parameters associated with encoding, decisionand response-selection stages, and of three parameters associated with the com-parison stage (target-set size, array size, and the presence or absence of a barmarker indicating the target's position), only target-set size interacted withmemory load (Experiments 1 and 2). It was concluded that most of the process-ing involved in character classification does not require attention, and this wastaken as evidence against models that identify attention with specific processing-structures. The results were interpreted as supporting the view that attention is alimited capacity to activate processing structures internally, and the role of at-tention in preparing, maintaining, and executing mental operations is discussed.

Copyright 1978 by the American Psychological Association, Inc. All rights of reproduction in any form reserved.

32

AUTOMATIC PROCESSING IN SEARCH TASKS 33

Often in everyday behavior we notice thatsome of the things we do demand moreattention than others. In a more analyticmood, we may notice that some componentsof individual acts seem more demandingthan others. These intuitions point to a ques-tion of considerable breadth and specificityin more formal studies of attention—thequestion of the involvement of attention inmental operations.

Mental operations can he considered in-dividually or as components of a series ofprocessing stages involved in producing a

The experiments reported here were conductedat McGill University, Queen's University, and theUniversity of Waterloo; the final report was pre-pared at Erindale College, University of Toronto.I am grateful for space and facilities so generouslygiven at McGill, Queen's, and Waterloo, and forthe stimulation provided by the members of theirrespective psychology departments.

This research was supported by the followinggrants from the National Research Council ofCanada: Grant A0127 to A. S. Bregman at McGillUniversity; Grant APA-318 to D. J. K. Mewhortat Queen's University; a Principal's Grant toQueen's University; Grant A95 to M. P. Brydenat the University of Waterloo; and Grant APA-231 to P. M. Merikle at the University of Water-loo.

Experiments 3, 6, and 8 were part of a doctoraldissertation submitted by the author to McGillUniversity, May 1975. I am grateful to Al Bregmanfor supervising the dissertation and providing criti-cal discussion.

Experiment 3 was presented at the 46th AnnualMeeting of the Eastern Psychological Association,New York, April 1975. Experiments 6 and 7 werepresented at the Canadian Psychological Associa-tion meeting, Quebec, P.Q., June 197S. Experi-ment 4 was presented at the Canadian Psycho-logical Association meeting, Toronto, Ontario,June 1976. Experiments 1 and 2 were presented atthe 48th Annual Meeting of the Eastern Psycho-logical Association, Boston, April 1977.

I would like to thank Jane Zbrodoff for con-ducting Experiments 6 and 8, helping with the dataanalysis, and providing critical discussion through-out the development of the ideas, particularly whileI was writing the article. The reviewers also sug-gested many improvements for which I am grateful.Finally, I must acknowledge D. Kahneman's ex-cellent book, Attention and Effort (1973), as aprimary source of inspiration.

Requests for reprints should be sent to GordonD. Logan, Department of Psychology, ErindaleCollege, University of Toronto, Mississauga, On-tario, Canada LSL 1C6.

response. The involvement of attention inspecific mental operations is currently con-troversial. For instance, attentional involve-ment in operations that prepare a set of tar-get items for comparison with a visual array(Comstock, 1973; Millar, 1975; Posner &Boies, 1971; Posner & Klein, 1973) and insubsequent operations that perform the com-parison (Atkinson, Holmgren, & Juola, 1969;Estes, 1972, 1974; Gardner, 1973a; Rumel-hart, 1970 ; Schneider & Shiffrin, 1977 ; Shif-frin & Schneider, 1977) has been discussedextensively. The issues are important becauseof their implications about the control ofmental operations during performance. Whenmental operations are considered as com-ponent processing stages, the pattern ofattentional involvement across processingstages becomes important. Attention may beconsidered either as a property of the struc-ture of the system or as a capacity for acti-vating structures. Since attention as a struc-tural property must have an identifiablelocus within the structure, the structuralview implies a particular pattern of atten-tional involvement across stages. By con-trast, the capacity view makes no implica-tions about the locus of attentional involve-ment, since capacity is an energistic propertyapart from structure.

The present experiments examine the in-volvement of attention in mental operationsunderlying performance in a character-clas-sification task. The first two experimentsestablish and validate a method for inferringattention demands, which focuses on changesin the effects of parameters of the classifica-tion task corresponding to changes in theavailability of attention. The classificationtask is performed alone or in the retentioninterval of a short-term memory task de-signed to occupy the subject's attention. In-teractions between classification-task param-eters and the presence or absence of a con-current memory load are interpreted asindicating that mental operations associatedwith those parameters require attention.This method is shown to be both internallyconsistent and consistent with another cur-rent method for inferring attentional in-volvement.

34 GORDON D. LOGAN

The remaining six experiments examineattentional involvement in four processingstages sufficient to perform the visual-searchversion of the character-classification task:encoding, comparison, decision, and responseselection (Smith, 1968). The pattern of in-volvement across stages is discussed in termsof structural and capacity models, then thecomparison stage is considered in greaterdetail. The contrast between comparisons invisual search and memory search leads to adiscussion of attentional control and suggeststhat attending might involve the use ofcapacity to prepare an organized structureto deal most efficiently with an expectedstimulus.

Attention, Mental Operations, andReaction Time

Structural models. The prototypic struc-tural model includes a limited-capacity chan-nel and a filter. The function of the limited-capacity channel is to provide task-relevantinformation to the effectors. The filter pre-cedes the limited-capacity channel in theinformation flow so that it may select rele-vant attributes from the stimulus array andthereby reduce the load on the channel. Theoperation of the filter depends on prelimi-nary processes outside the channel that pro-vide the informational basis for selection(Broadbent, 1958; Deutsch & Deutsch,1963; Neisser, 1967).

In structural models, the attention de-mands of mental operations may be charac-terized by the time and space they requirewithin the limited-capacity channel (Posner& Keele, Note 1). Because space is limited,an increase in the demands on attention willincrease the time spent in the channel. Afurther possibility explored in this article isto reduce the space available by imposingadditional, irrelevant demands. With spacethus reduced, the same demand should re-quire more time.

Tn structural models, operations outsidethe channel require time but not space, andreaction time is the sum of the time require-ments inside and outside of the channel.Reaction time will thus increase as attentiondemands increase whether or not the de-mands arc relevant to the task.

Capacity models. Capacity models ex-plain attention in terms of the activation ofstructures, not in terms of the structuresthemselves. Attention is identified with theselective activation of mental structures bythe allocation of limited central processingcapacity. The distinguishing characteristic ofthese models is the assertion that structureand central processing capacity are separateconstituents of the mind : Capacity repre-sents energy, and structure represents thedevices that consume energy to produce be-havior (Kahneman, 1970, 1973; see Moray,1967, and Posner & Boies, 1971, for relatedviews).

These models generally agree that ca-pacity is not always necessary to activatestructures. Structures may be activated in-ternally by the allocation of capacity or ex-ternally by the presence of a stimulus on areceptor surface. The attention demands ofmental structures thus reflect the extent towhich capacity is necessary to activate them.Some automatic structures may only requirethe presence of a stimulus to be activated;other structures may require capacity aswell as stimulation. Theorists have arguedfor a dimension of automaticity representingall possible degrees of attentional involve-ment (Kahneman, 1973; LaBerge & Sam-uels, 1974; Posner & Snyder, 1975a).

In capacity models, reaction time is de-termined by the rate of processing withinmental operations and the amount of workrequired of them. Attention has its effectsthrough the momentary relation between ca-pacity supplied and capacity demanded(Kahneman, 1973). Structural factors im-pose a limiting rate of processing on anoperation that can be approached only whenthe supply of capacity equals or exceeds thedemand. Tn automatic operations, the de-mand is essentially zero, so they operate attheir limiting rates regardless of the supply.When the supply is less than the demand, asis often the case for operations that requirecapacity, the rate of processing is slowerthan the limiting rate by an amount propor-tional to the deficit. Deficits are producedby increasing the demands of other mentaloperations to reduce the overall supply. Theother mental operations may reside in other

AUTOMATIC PROCESSING IN SEARCH TASKS 35

processing stages of the same task or inother, independent tasks performed concur-rently. Concurrent-task performance is es-pecially interesting as both structural andcapacity models predict that the load im-posed by the additional task will increaseprocessing time by an amount proportionalto the demands of the initial task.

Additive-lactors method. Under certainconditions, concurrent-task situations mayreveal the attention demands of mental op-erations. The additive-factors method forisolating processing stages in reaction timedata specifies one set of conditions: Stern-berg (1969a, 1969b) has argued that meanreaction time represents the sum of the du-rations of a series of processing stages in-volved in producing an appropriate response.This implies that two parameters affectingdifferent stages will have additive effectswhen varied concurrently, whereas parame-ters affecting the same stage will interactstatistically (i.e., the rate of processing ofone parameter depends on the value of theother). Thus, when a parameter of one task,identified with a particular processing stage(or mental operation), interacts with a pa-rameter of a concurrent task known to affectthe availability of attention, the magnitudeof the interaction will indicate the attentiondemands of the stage (or operation). Addi-tive effects in this situation are important,for they indicate that the operation requiresno attention to function. Thus, additivitymay be used to identify automatic processes.1

Before we can interpret data in this way,we must consider three inherent difficultieswith the additive-factors method: First,Sternberg (1969a, 1969b) has raised thepossibility that parameters affecting differentstages may interact, and parameters affectingthe same stage may have additive effects.Thus, additivity and interaction cannot beinterpreted without some ambiguity. This ismost problematic in exploratory research inwhich there are no a priori reasons for ac-cepting one interpretation or the other. Inthe present studies, however, the associationof additivity with automaticity and interac-tion with demand derives from both struc-tural and capacity models of attention. Fur-thermore, the use of the additive-factors

method to define specific processing stagesin terms of the particular parameters em-ployed in the studies for the most part repli-cates previous work (e.g., Smith, 1968;Sternberg, 1969b, 197S).

Statistical considerations provide a sec-ond difficulty (Pachella, 1974) : The evi-dence for automaticity and for the separa-bility of processing stages is the nonsignifi-cance of interactions ; the interpretation restson the acceptance of a null hypothesis. Weshall see that the data obtained were suffi-ciently reliable and regular that the nullhypothesis of no interaction seems the bestdescription of the data. In most cases, the/'" ratios for the interaction terms were closeto unity, while those for the contributingmain effects were highly significant.

A third difficulty derives from Taylor's(1976) stage analysis of reaction time. Tay-lor argues that subjects may compensate forincreased processing time in one stage byreducing processing time in another, andthe right amount of compensation may maska true interaction, producing effects best de-scribed as additive. Compensation is wellknown in the reaction time literature in theform of the speed—accuracy trade-off (e.g.,Pachella, 1974) ; it is apparent that savingsin time can only be bought at the cost ofaccuracy. Thus, we will look to the errordata for evidence of a compensatory strategy.

1 This discussion assumes that attention can bedivided so that both tasks are attended simultane-ously. It is possible that attention cannot be di-vided and must be switched from task to task. Ifthe switch occurs during the reaction time interval,reaction time will increase by an amount that re-flects switching time, [f switching time is constant(e.g., Broadbent, 19S8), no interactions will be ob-tained, and the method proposed here is inappro-priate (but see Experiments 1 and 2). If switchingtime increases with the attention demands of bothtasks (e.g., LaBerge, 1973a), the parameters, andthus stages, contributing to the attention demandsof the classification task can be identified: Sinceprocessing time and switching time will both in-crease with the values of these parameters, inter-actions will indicate that a stage demands atten-tion. Parameters that increase processing time butnot switching time may be associated with auto-matic processing stages. Their effects will be addi-tive with respect to concurrent activity.

36 GORDON D. LOGAN

Retention as a Concurrent Task

Following this application of the additive-factors method, reaction time was measuredas a function of parameters of a classificationtask identified with particular processingstages. On half of the trials, the attentionavailable to the classification task was re-duced by inserting it in the retention inter-val of a short-term memory task. The atten-tion demands of the stage under study werethus inferred from patterns of interactionand additivity found between memory loadand the defining parameters.

Retention was chosen as a concurrent ac-tivity for several reasons. First, short-termretention has been treated theoretically asan attention-demanding activity (Atkinson& Shiffrin, 1968; Broadbent, 1958; Kahne-man, 1973). Empirically, it is clear thatretention requires attention, since a widevariety of tasks will interfere with it or areinterfered with by it when performed con-currently (e.g., Brown, 1958; Crowder,1967; Johnston, Greenberg, Fisher, & Mar-tin, 1970; Peterson, 1969; Peterson & Peter-son, 1959; Posner & Rossrnan, 1965; Salt-house, 1975; Watkins, Watkins, Craik, &Mazuryk, 1973). Moreover, the requirementthat subjects retain seven or more items dur-ing a trial interferes reliably with severalvisual tasks, including partial and whole re-port (Henderson, 1972; Scarborough, 1972),detection (Shulman & Greenberg, 1971),search (Broadbent & Heron, 1962), andchoice reaction time (Shulman & Greenberg,1971; Shulman, Greenberg, & Martin, 1971).In general, memory loads of five itemsor less produce no interference (Darley,Klatzky, & Atkinson, 1972 ; Boost & Turvey,1971; Turvey, 1966; Wattenbarger & Pa-chella, 1972), whereas loads of seven itemsor more produce interference in proportionto load (Shulman & Greenberg, 1971; Shul-man et al., 1971). However, loads of lessthan five items have been shown to produceinterference when the array for the visualtask is presented less than 1 sec after thememory load (Chow & Murdock, 1975,1976; Connor, 1972).

Retention was also attractive as a con-current activity because the short-term mem-

ory task and the classification task seemedto require different mental operations (Estes& Taylor, 1964; Gardner, 1973a). Opera-tionally, the memory stimuli (spoken digitsor words) were categorically different fromthe visual stimuli (letters) and involved aseparate sensory modality. Thus interferencebetween tasks is not likely to be attributableto overloading some peripheral structurecommon to both tasks (cf. Brooks, 1968;Kahneman, 1973; Norman & Bobrow, 1975)but rather to overloading attention.

The Experiments

To review the argument so far, we haveseen that most models of attention predictthat concurrent activity will increase reactiontime relative to single-task controls. Extend-ing Sternberg's (1969a, 1969b ) additive-factors method, the attention demands ofmental operations may be identified: Con-current activity will interact with parame-ters associated with processing stages thatdemand attention but will have additive ef-fects with parameters associated with auto-matic processing stages. The first two ex-periments varied target-set size and decisiontype (yes or no) in the memory-search ver-sion of the character-classification task, andthe task was performed both alone and inthe retention interval of a concurrent short-term memory task. Patterns of interactionand additivity obtained between target-setsize and concurrent memory load were suf-ficient to validate the present extension ofthe additive-factors method. Further, the as-sociated inferences of attention demand andautomaticity were corroborated by anothercriterion for automaticity (i.e., Posner &Snyder, 1975a).

The method thus validated was applied inthe next six experiments to identify theattention demands of parameters associatedwith encoding, comparison, decision, andresponse-selection stages of the visual-searchversion of the character-classification task.

General Method

This section describes aspects of the method com-mon to all experiments. When specific experiments

AUTOMATIC PROCESSING IN SEARCH TASKS 37

depart from the general method, the departure willbe noted in the method sections for those experi-ments.

Subjects

Each experiment involved paid subjects who re-ported normal or corrected-to-normal vision. Sub-jects were tested individually in 1-hr sessions. Thenumber of subjects in each experiment appears incolumn 7 of Table 1, and the number of sessionseach subject completed appears in column 6.

Apparatus and Stimuli

In all experiments, the classification task re-quired decisions about visually presented arrays.In the memory-search experiments (1 and 2) thearrays contained single letters, and in the visual-search experiments (3 through 8) the arrays con-tained 4, 8, and 12 different letters. Details of thecomposition of the arrays may be found in themethod section of Experiment 1 for the memory-search studies and of Experiment 3 for the visual-search studies.

In all experiments, the arrays were made ofblack letters on white cards and were exposed ina tachistoscope. The exposure of each array waspreceded and followed by a fixation field contain-ing a small black dot in the center of a white field.The luminance of fixation and stimulus field wasmatched for each experiment; the luminance val-ues appear in column 1 of Table 1.

In all experiments, there was a constant delaybetween the initiation of the tachistoscope timers

Table 1Parameter Values for each Experiment

and the onset of the array. The sharp click of theswitch initiating the timers was audible and servedas a warning signal. The durations of the delays(foreperiods) appear in column 2 of Table 1.

After the foreperiod, the arrays were exposedfor the durations appearing in column 3 of Table1. The fixation field returned when the arraysterminated and remained on for the remainder ofthe retention interval. At the end of the retentioninterval, the fixation field was turned off for 500msec as a signal for the subject to begin recall.The intervals between the termination of the arrayand the onset of the signal to recall are given incolumn 4 of Table 1. Note the value of zero forcue delay in Experiment 3. Here, recall was im-mediate, so no cue was given—the fixation fieldreturned after the array and remained on until thearray was exposed on the next trial.

Reaction time was measured in msec from theonset of the array, using a digital timer. Thetimer was started by the tachistoscope timers andstopped when the subject pressed one of two but-tons attached to microswitchcs mounted on a panelin front of him or her (all experiments except 6and 8) or spoke into a microphone, thus closinga voice-activated relay (Experiments 6, 7, and 8).Pressing each button illuminated a separate lightvisible only to the experimenter so that accuracycould be monitored.

In all except Experiment 8, the memory stimuliwere strings of 7 unique digits presented at anapproximate rate of .75 sec per digit (the memorystimuli for Experiment 8 will he described later).In Experiment 3 the digits were presented orally,while in all other experiments they were recordedand played back through a speaker. A different

Parameter

Experi-ment

1

2

3456

7

8

Characteristicvariable

Target-set size

Target-set size

Bar markerMaskYes-no decisionStimulus-response

compatibility

Response type

Memory load

1

Lumi-nance(ft L.)

14

14

82988

8

8

2

Fore-period(msec)

500

500

100500500500

500

500

3

Exposure(msec)

500

500

5001,500

500500

500

500

4

Delay ofrecallcue(sec)

4.5

4.5

0455

5

5

5

Practice(trials)

8 per set sizeper day

8 per set sizeper day

01010

10 percompatibilitycondition

10 perresponse type

10

6

Sessions

6-7

6

4414

4

1

7

Sub-jects

6

6

88

168

8

16

38 GORDON D. LOGAN

list was presented on each memory-load trial, andeach day involved a new set of lists. Each listwas preceded by a warning signal ( Y O U R NEXT LISTis) and followed by a ready signal (HEADY) , whichoccurred about 1.5 sec after the last digit in eachstring. Since the arrays were typically exposedabout 1 sec after the ready signal, the retentioninterval was roughly 3 sec in Experiment 3 androughly 8 sec, in all others (i.e., 2.5 sec plus thesum of columns 3 and 4 of Table 1).

Procedure

In all experiments, trials involving a memoryload consisted of the following sequence of events:(a) the presentation of the memory l is t ; (b) thepresentation of a verbal ready signal; (c) the sub-ject's affirmative reply to the ready signal, indi-cating that the fixation point was in sharp focus;(d) the foreperiod between the click of the initiat-ing switch and the exposure of the array; (e) theexposure of the array; (f) the subject's responseto the array; (g) a delay and the dark interval.signalling recall ; and (h) the subject's recall ofthe 7 digits in the order in which they were pre-sented. Trials not involving a memory load omit-ted events (a) and ( h ) .

Experiments 1 and 2 manipulated target-set size(1, 2, or 4 letters), decision type (yes or no), andmemory load (0 or 7 digits). Each session con-sisted of six blocks, one for each factorial combi-nation of target-set size and memory load, anddecision type varied randomly within blocks.

The primary variables in the remaining experi-ments (except Experiment 8) were array size (4,8, or 12 letters), memory load (0 or 7 digits),and a parameter (two levels) defining the stageunder study. Memory-load conditions and valuesof the defining parameters were combined fac-torially and run in separate blocks. Each sessionconsisted of four blocks, one for each combination,in which array size varied randomly.

The orders of conditions across subjects eachday and within subjects over days were deter-mined by balanced Latin squares, fn Experiments1 and 2 arrays within blocks (and thus decision-type conditions) were presented in two (inverse)orders, and subjects alternated orders over days,half beginning with one (i.e., A R A B ) and halfwith the other (i.e., BABA). Memory-load condi-tions alternated over blocks and subjects in asimilar fashion, and assignment of decision typesto response buttons was balanced over subjects aswell. Because of the small sample size (Table f,column 7), these assignments were necessarilyconfounded.

In Experiment 3 the arrays (and thus array-size conditions) were presented to all subjects inthe same order each day. Experiments 4 through8 used two orders of arrays, the one used in Ex-periment 3 and its inverse, alternating over subjectsand days as in Experiments 1 and 2. Assignmentto these orders was orthogonal to the assignment

to orders of memory-load and defining-parameterconditions. In experiments that involved manualresponses, the assignment of hands to responsebuttons was balanced across subjects. This assign-ment was always orthogonal to the assignmentto orders of memory-load and defining-parameterconditions, and wherever possible, orthogonal tothe assignment to orders of arrays.

Each session consisted of 144 trials divided intosix 24-trial blocks in Experiments 1 and 2, andinto four 36-trial blocks in Experiments 3 through8. The number of sessions per subject in each ex-periment appears in column 6 of Table 1. In ex-periments involving more than one session persubject, each subject served in one session per dayat the same time each day plus or minus 1 hr.Whenever possible, sessions were run on consecu-tive days, and no more than 2 days elapsed be-tween consecutive sessions.

Instructions first described the visual task, ex-plaining the procedure and describing the experi-mental conditions, using examples where appro-priate. Subjects were told to respond to the arrayas quickly as possible without making errors.Thus an example of a digit string was given, andthe sequence of events on a memory-load trial wasdescribed. Subjects were instructed to concentrateon the memory task during memory-load trials.They were told that it was the most important taskand that they should optimize their performanceon it. Thus, according to instructions, the memorytask was primary and the visual task was secondary.

fn general, the instructions were followed by aseries of practice trials, then testing began. Thenumber of practice trials in each experiment ap-pears in column 5 of Table 1. Practice was giveneach day in Experiments 1 and 2 but only on thefirst day in the remaining experiments; the se-quence of conditions for that day was described,and testing began after a brief delay.

Subjects were not given feedback regardingtheir speed or accuracy in the visual task. In thememory task of Experiments 1 and 2, subjectswere told the number of digits they recalled cor-rectly after each trial, but no feedback about recallaccuracy was given in the remaining experiments.

Data Analysis for the Visual Task

In each session, each subject completed 12 trialsfor each combination of visual-task conditions.Mean reaction times for correct responses in eachcondition were computed for each session for eachsubject. Reaction times exceeding 1,900 msec werescored as errors to reduce skew in the individualreaction time distribution. These means were sub-mitted to a four-way within-subjects analysis ofvariance, with target-set size, decision type, mem-ory load, and days as factors in Experiments 1and 2, and with array size, memory load, defining-parameter conditions, and days (where appro-priate) as factors in Experiments 3 through 8.The mean error rate (in proportions) is reported

AUTOMATIC PROCESSING IN SEARCH TASKS 39

for each condition, but the error frequencies weregenerally too low for statistical analysis.

Data Analysis for the Memory Task

In each session each subject was required torecall 12 lists of 7 digits in each combination ofvisual-task conditions. The mean number of digitsrecalled in correct order in each list was computedfor each subject in each condition. These meanswere then subjected to a three-way within-subjectsanalysis of variance, with target-set size, decisiontype, and days as factors in Experiments 1 and 2,and with array size, defining-parameter conditions,and days (where appropriate) as factors in Ex-periments .3 through 8.

Criteria for Autoniaticity

The experiments reported in this sectionexamined attentional involvement in com-parison and decision stages of a memory-search task. Patterns of interaction and addi-tivity were obtained between memory loadand target-set size that, first, validated theextended additive-factors method as a tech-nique for assessing attentional involvementby demonstrating the development of automa-ticity with practice and, second, provided newinsight into the comparison stage, particularlyin contrast with related findings in the follow-ing experiments on visual search. Since thetechnique must be validated before it can beused inferentially, this section will focus onvalidation, and discussion of implications forthe nature of the comparison stage will bedeferred until later (see Comparison StageRevisited, p. 58).

The memory-search task is an appropriatevehicle for validation, as it allows the assess-ment of automaticity by three different cri-teria. The task itself requires that the sub-ject decide whether or not an array containsa member of a set of target items presentedearlier, and in general, reaction time in-creases linearly with the size of the targetset with the same slope for both decisionoutcomes (for reviews, see Nickerson, 1972;Sternberg, 1969a, 1969b, 1975).

The first criterion identifies automaticsearch with zero slope in the target-set-sizefunction. It was first proposed by Neisser andhis colleagues (Neisser, Novick, & Lazar,1963) upon finding that with enough practice

subjects could search for 10 targets as quicklyas for 1 (i.e., zero slope). They argued thatsearch was automatic, since decisions aboutindividual targets seemed independent of oneanother. Subsequent replications have re-vealed sufficient conditions for approxi-mately this effect; extended practice withnested target sets and consistent mapping ofstimuli onto responses will produce slopesthat approach zero but rarely reach it (Cor-ballis, Koldan, & Zbrodoff, 1974; Graboi,1971; Kristofferson, 1972a, 1972b; Neisser,1974; Schneider & Shiff r in , 1977). Auto-niaticity, by this criterion, seems a rareevent, and although Shiffrin and Schneider(1977) have used it to build a general theoryof attentional control, it applies directlyonly to the comparison stage of classificationtasks.

Posner and Snycler (1975a) have pro-posed more general criteria for automaticity:The}- argue that automatic structures areactivated without intention and without in-terference from concurrent activity in otherstructures. The latter is essentially the ex-tension of the additive-factors method devel-oped earlier, and it is readily applicable tomemory search. The amount of interferencemay be assessed from the magnitude of in-teractions between memory load and pa-rameters of the search task—target-set sizein particular. The former may be applied tothe memory-search task as well: Tf a stim-ulus automatically elicits a tendency towarda response onto which it has been consist-ently mapped, the tendency should be elicitedeven when the mapping is changed so thatthe stimulus now requires the opposite re-sponse, and it should inhibit responses basedon the new mapping. Similar interference isfound when subjects name the color of theink in which a conflicting color word iswritten (Stroop, 1935; for a review seeDyer, 1973). Attention is directed elsewhere,yet the representation in permanent memoryis activated automatically by its character-istic stimulus (Keele, 1972).

Experiments / and 2

Experiments 1 and 2 were designed tosearch for agreement among the three cri-

40 GORDON D. LOGAN

teria for automalicity. Tn Experiment 1, sub-jects searched one list of 1-, 2-, and 4-lettcrtarget sets for 6 days, then switched to an-other list in which half of the no items hadbeen yes on the previous list. On half ofthe trials each day, the task was performedin the rentention interval of a short-termmemory task requiring ordered recall of 7digits. According to the first criterion, searchwill be automatic when the slope of thetarget-set-size function reaches zero. Ac-cording to the second, search will be auto-matic, when memory load no longer increasesthe slope of the target-set-size function (i.e.,when memory load and target-set size nolonger interact), and according to the third,search will have become automatic if itemsthat had received yes responses consistentlyfor 6 days take longer to respond "no" tothan other items, when the mapping ischanged on the 7th day.

Experiment 2 was an exact replication ofExperiment 1 except that a different list oftarget sets was used each clay. Its purposewas to provide baseline conditions underwhich automaticity should not develop,against which to evaluate the effects in Ex-periment 1.

Method

There were several departures from the generalmethod. First, subjects were run for at least six1-hr sessions on consecutive days (see GeneralMethod). Second, subjects received feedback abouttheir memory performance. At the end of each trialthey were told the number of digits they recalledcorrectly and the positions and nature of their

Table 2Positive and Negative Target Sets Used inExperiment 1 (Lists 1 and 2) andExperiment 2 (Lists 1-6)

Positivetarget-set size

List 1 Negative set

123456

HDMTZs

NRB.TCVXKLFPM

SKXPFZTGLWYHRBGWCMVYDQNC

BCnFGJLMTVYZCHKLMKPRSVWYBDFGKNPQRSXZCDFHLMNPSVYZDGHKNPQRSTWXBFGKLRTVWXYZ

errors. The feedback was to ensure concentrationon the memory task. Third, subjects were not re-quired to reply to the verbal ready signal presentedbefore each trial (i.e., Event c of the general pro-cedure was omitted). Fourth, subjects were given8 practice trials whenever target-set size changed,and practice was given every day.

Apparatus and stimuli. A Scientific Prototypethree-field lachistoscope (Model GB) was used toexpose the arrays. Each array consisted of a singleblack capital letter (Lctraset #287) mounted cen-trally on white cards so as to appear on top ofthe fixation point. Each letter subtended about 11'X 11' of visual angle. Temporal and luminance pa-rameters are given in Table 1.

The target sets were drawn from a restrictedalphabet of 21 letters (A, E, I, 0, and U wereexcluded). Positive sets (those requiring yes re-sponses) contained 1, 2, or 4 different letters. Thesame 12-letter negative set was used for eachpositive set in a list. Six different lists were pre-pared (see Table 2). Tn Experiment 1, one list(List 1) was used for the first 6 days, and 4 sub-jects returned for a 7th session, which used List 2.(Note that half of the no items in List 2 were yesitems in List 1.) In Experiment 2, all 6 lists wereused. Each subject dealt with a different list eachday, and each day each subject had a different list.Assignment of lists to subjects and days wasbased on a 6 X 6 balanced Latin square. SinceList 2 did not follow List 1 consistently in thisexperiment, List 2 data can be compared with thoseof Experiment 1 as a control for specific practicewith List 1.

Procedure, Each session consisted of 144 trialsdivided into six blocks of 24 trials, one for eachcombination of target-set-size and memory-loadconditions. Each target-set size was tested for twoconsecutive blocks, one with memory load and onewithout. The order of memory-load and no-mem-ory-load conditions varied within subjects overdays and between subjects within days. Over days,each subject received all six possible orders oftarget-set-size conditions, and each day each ofthe six subjects received a different order. Deci-sion outcome (yes or no) varied randomly withinblocks. In each block there were 12 yes and 12no responses.

Results and Discussion: Experiment 1

Visual task. Mean reaction times for eachcombination of target-set-size, decision-out-come, and memory-load conditions (exclud-ing Day 7) are displayed in Figure 1. Eachpoint in the figure is based on 432 observa-tions. In an analysis of variance performedon mean reaction times, significant effectswere found for days, F(5, 25) = 15.00,p < .01, target-set size, F(2, 10) = 80.61,p < .01, and decision outcome, F(_l, 5) =

AUTOMATIC PROCESSING IN SEARCH TASKS 41

600

~£ 500UJ

400

5300

TARGET SET SIZE

Figure 7. Mean reaction times in the visual taskof Experiment 1 as a function of target-set size.(Decision-type and memory load conditions areparameters : Y = yes ; N = no ; L = memory load;UL = no memory load.)

7.45, p < .05. Reaction time decreased overdays, increased with target-set size, and waslonger for no than for yes decisions.

Though a concurrent memory load in-creased reaction time overall by 54 msec,the effect was not significant, F{\, 5) = 4.21,p> .05. However, the interaction of memoryload with days was significant, -P(10, 50)= 4.40, p < .01, indicating a large initialeffect (109 msec) which diminished overdays.

The interaction between memory load andtarget-set size was not significant, F(2, 10)< 1, but the three-way interaction betweenmemory load, target-set size, and days was,F(10, SO) = 3.97, p < .01. The meaning ofthis interaction and the one between target-set size and days, F(10, 50) = 4.40, p<.01, may be better expressed in terms of theslopes of the functions relating reaction timeto target-set size. Slopes and interceptsof the best-fitting linear functions were com-puted each day for each subject individually,and the mean values appear, respectively, inthe top and bottom panels of Figure 2. Notethat an apparent main effect in the slopedata is equivalent to a linear interaction withtarget-set size.

Figure 2 shows a general reduction inslope over days, reaching an asymptoticvalue of about 20 msec per item on Days

4, 5, and 6. According to the zero-slopecriterion, search had not become automaticafter 6 days of practice. The figure alsoshows that memory load and target-set sizeinteracted at early stages of practice, butnot later on. Considering the yes data, thememory-load slopes were substantially higherthan no-memory-load slopes for Days 1, 2,and 3, but were slightly lower on Days 4,5, and 6. In the no data, memory-load slopeswere only higher than no-memory-loadslopes on Day 1. According to the extendedadditive-factors logic, the interaction earlyin practice indicates that search demandedattention initially, and the gradual transi-tion to additivity indicates the developmentof automaticity with practice.

To see whether this automaticity had de-veloped from specific practice, four sub-jects were induced to return for a 7th day tobe tested with a different list of target sets.Their data, plotted above Day 7 in Figure 2,show elevated 'memory-load slopes at leastfor yes decisions. For no decisions, the no-memory-load slopes remained higher thanthe memory-load slopes on Day 7, continuinga trend that had developed on Day 4. Thedata, then, provide some support for thehypothesis that automaticity had developedfrom specific practice. The hypothesis re-

80

8 60

u 40

9 20(fl

0

600

500

400

300

. NU.NUf> - •-..

YULo = — ̂ ___ -̂ n^H

1 2 3 4 5

DAYS

— : g - f i y f

6 7

Figitre 2. Mean slopes (top panel) and intercepts(bottom panel) of the best-fitting linear functionsof target-set size from the reaction time data fromExperiment 1. (Decision-type and memory-loadconditions are parameters : Y = yes; N — no; L =memory load ; UL — no memory load.)

42 GORDON D. LOGAN

Table 3Mean Reaction Times (in msec) and Proportions of Errors to No Items from List 2

Memory load

Experiment 1, Day 7 Experiment 2

Type of item

OldOld

yesno

0

RT

486441

digits

Errorrate

.04

.01

7

RT

S37469

digits

Errorrate

.10

.00

0

RT

495490

digits

Errorrate

.03

.01

7

RT

584600

digits

Errorrate

.02

.02

Note. Old-yes items were yes items on List 1, and old-no items were no items on List 1. RT = reaction time.

ceivcs a more stringent test in Experi-ment 2.

Clearly, these two criteria for automaticityprovide conflicting interpretations of theslope data; the zero-slope criterion suggeststhat automaticity had not been obtained,while the additive-factors criterion suggeststhat automaticity had developed by Day 4.The data from Day 7 provide a test by thethird criterion of automaticity, unintentionalactivation.

The no items from Day 7 (List 2) weredivided into old-yes and old-no items ac-cording to their assigned mapping on Days 1to 6 (List 1). Mean reaction times to theseitems with and without a memory load aredisplayed in Table 3.

According to the unintentional-activationcriterion, old-yes reaction times should belonger than old-no reaction times if auto-

Table 4Proportion of Errors in the Visual Task andRecall Accuracy (in parentheses) in theMemory Task (Across Siibjects and Days)as a Function of Visual-Task Conditionsin Experiment 1

j-v • •

type

Yes0 digits7 digits

No0 digits7 digits

.02

.02

.02

.02

Target-set size

1 2

.04(6.26) .05 (6.29) .

.02(6.33) .05 (6.23) .

0408

0505

4

(6.22)

(6.25)

maticity had developed over the 6 days ofpractice with List 1. On the average, old-yes reaction times were 56 msec longer thanold-no reaction times, and the difference wassignificant, t (3) - 2.44, /> < .05, one-tailed.The data thus indicate that automaticity haddeveloped, corroborating the interpretationof the slope data given the additive-factorscriterion.

The error data from Day 7 are presentedin Table 3; the error data from Days 1 to 6appear in Table 4. Tn general, the error datareflect the patterning of the reaction timedata.

Memory task. The mean numbers ofdigits recalled in correct order for eachcombination of target-set-size and decision-outcome conditions also appear in Table 4.The only significant effect in an analysisof variance performed on these data was thatof clays, F(5, 25) = 3.89, /; < .01, indicatingimprovement in recall accuracy withpractice.

Results and Discussion: Experiment 2

Visual task. Mean reaction times for eachcombination of target-set-size, decision-out-come, and memory-load conditions are dis-played in Figure 3. Each point in the figureis hased on 432 observations. The only sig-nificant effects in an analysis of varianceperformed on mean reaction times were themain effects of days, F(5, 25) = 3,25, p< .05, target-set size, F(2, 10) = 85.16, p< .01, and decision outcome, F ( ] , 5) =29.83, p < .01. Again, reaction time de-

AUTOMATIC PROCESSING IN SEARCH TASKS 43

creased over days, increased with target-setsize, and was larger for no decisions thanfor yes decisions.

Overall, the requirement that subjects re-tain 7 digits increased reaction time by 97msec, but the memory-load main effect wasnot significant, F(l, S) = 4.82, /> > .05,and although the Memory Load X Target-Set Size interaction appears substantial inFigure 3, it was not significant either, F(2,10) = 3.45, p > .05. However, there werelarge individual differences in the memory-load main effect (ranging from —1 to 299msec), which when combined with the smallsample size (six subjects), may have in-flated the error term and masked the inter-action. To remove this source of influence,slopes and intercepts of best-fitting linearfunctions were computed for each subjecteach clay, and the slopes were submitted toan analysis of variance. Again, a main ef-fect in the slope data represents a linear in-teraction with target-set size.

Mean slopes and intercepts in each con-dition each day appear, respectively, in thetop and bottom panels of Figure 4. Theslope data in the figure show a consistentinteraction of memory load with target-setsize: The memory-load slopes for both de-cision outcomes are consistently higherthan the no-memory-load slopes each day.

700

Z 600

500

400

1

TARGET SET SIZE

Figure 3, Mean reaction times in the visual task ofExperiment 2 as a function of target-set size. (De-cision-type and memory-load conditions are pa-rameters : Y = yes; N = no; L = memory load;UL = no memory load.)

80

I 6°w•§ 40

201 0

\ 600

•500

0400

Si!Z 300

YUlf

1 2 3 4 5 6

DAYS

Figure 4. Mean slopes (top panel) and intercepts(bottom panel) of the best-fitting linear functionsof target-set size from the reaction time data fromExperiment 2, (Decision-type and memory-loadconditions are parameters: Y = yes; N = no; L =memory load; UL = no memory load.)

This is supported by the significant memory-load main effect in the analysis of varianceon the slope data, F(l, 5) = 8,28, p < .05,which was the only significant effect in theanalysis.

According to both the zero-slope criterionand the additive-factors criterion, the dataindicate that search was not automatic, Theslopes were consistently above zero, and thememory-load slopes were consistently greaterthan the no-memory-load slopes, The un-intentional-activation criterion may be ap-plied to the List 2 no data presented inTable 3, divided into old-yes and old-no itemsaccording to their mapping on List L Sincesubjects in this experiment had not receivedextended practice with List 1 before theywere tested with List 2, it is not likely thatold-yes items would automatically elicit com1

peting response tendencies, so old-yes andold-no reaction times should be approxi-mately equal. In fact, old-no reaction timeswere 6 msec longer than old-yes, and thedifference was not significant, £(S) < 1.According to the unintentional-activationcriterion, automatic search had not de-veloped.

44 GORDON D. LOGAN

Table 5Proportion of Errors in the Visual Task andRecall Accuracy (in parentheses) in theMemory Task (Across Subjects and Days)as a Function of Visual-Task Condi-lionsin Experiment 2

Decisiontype

Yes0 digits7 digits

No0 digits7 digits

Target-set size

1

.01

.01 (6.52)

.02

.01 (6.53)

2

.02

.05 (6.59)

.02

.02 (6.51)

.06

.07

.02

.03

4

(6.47)

(6.55)

Thus, the three criteria agree in assessingattentional involvement in search in Ex-periment 2. The tests by the additive-factorsand unintentional-activation criteria confirmthe earlier conclusion that specific practiceis necessary to develop automaticity.

The error data are presented in Tables 3and 5.

Memory task. The mean number of digitsrecalled in correct order for each combina-tion of target-set-size and decision-outcomeconditions appear in Table 5. Again, theonly significant effect in the analysis ofvariance performed on these data was thatof days, F(S, 25) - 15.94, p < .01, indicat-ing an improvement in recall accuracy withpractice.

Discussion of Experiments 1 and 2

Experiments 1 and 2 have shown aninteraction between concurrent memory loadand target-set size that varies in magnitudeaccording to the degree of practice withspecific target sets. Three days of specificpractice were enough to eliminate the inter-action. These findings bear on the validationof the extended additive-factors logic as amethod for assessing attentional involve-ment in mental operations. First, they showthat the method is internally consistent inthat additivity and interaction both occurin real data and their pattern can be inter-preted meaningfully. In particular, the di-minishing interaction in Experiment 1 sug-

gests the development of automaticity overdays. Second, the inferences based on thesefindings agree with those 'based on anothercriterion for automaticity—the unintentional-activation criterion—which was assessedfrom different aspects of the data (i.e., thedifference between old-yes and old-no re-action times). When search had becomeautomatic by the additive-factors criterion,there was evidence of unintentional activa-tion (Experiment 1 ) ; when search re-quired attention'by the additive-factors cri-terion, there was no evidence of uninten-tional activation (Experiment 2).

The disagreement between these two cri-teria and the zero-slope criterion is interest-ing. According to the zero-slope criterion,search was not automatic in either experi-ment. This difference may reflect two dis-tinct concepts of automaticity. In this articleand elsewhere (Kahncman, 1973; LaBerge& Samuels, 1974; Posner & Snycler, 1975a,1975b), automaticity refers to the with-drawal of attentional control over mentalstructures. Tn other contexts, automaticityrefers to a change in the composition ofstructures assembled to perform a task infavor of more efficient performance (Kolers,1975). The additive-factors and uninten-tional-activation criteria refer explicitly tothe former concept (see Posner & Snyder,1975a). In fact, the additive-factors methodassumes that the composition of task-rele-vant structures does not change when mem-ory load is manipulated. By contrast, thezero-slope criterion seems to refer to a changein the composition of structures (Neisser,1974; Shiffrin & Schneider, 1977). Since weare concerned directly with the involvementof attention in the activation of mental struc-tures, the disagreement with the zero-slopecriterion is of no consequence, however in-teresting it may be in other contexts. Theexperiments have shown that the extendedadditive-factors method provides a valid as-sessment of attentional involvement, so wemay use it to investigate the demands ofother processing stages.

The remaining experiments examine en-coding, comparison, decision, and response-selection stages of the visual-search versionof the classification task. Since these repre-

AUTOMATIC PROCESSING IN SEARCH TASKS 45

sent an assembly of operations sufficientto produce a stimulus-controlled response(Smith, 1968; Sternberg, 1969a, 1969b), wemay construct a relatively complete pictureof the nature and loci of the demands ofthe task.

The Comparison Stage

In general, classification tasks require thesubject to decide whether or not an arrayof items contains one or more of a set oftarget items presented some time earlier (fora review, see Nickerson, 1972). Logically,the decision depends on the outcome of com-parisons between the items in the array andthe target set, so it is natural that the com-parison stage has received the lion's shareof empirical and theoretical attention. Thestage is usually identified with two parame-ters, the number of items in the array (arraysize) and the number in the target set.

Experiments that vary target-set size usu-ally employ single-item arrays and are saidto study memory search; experiments thatvary array size usually involve single-itemtarget sets and are said to study visitalsearch. It is not clear whether the two typesof search involve different comparison pro-cesses. The strong interactions obtained be-tween array size and target-set size maysuggest they do (Briggs & Blaha, 1969;Briggs & Johnsen, 1973 ; Briggs & Swanson,1970; Johnsen & Briggs, 1973; Nickerson,1966; Schneider & Shiffrin, 1977; Stern-berg, Note 2), but these effects may be ex-plained as well by models that assume sepa-rate comparison processes (e.g., Townsend& Roos, 1973) as by models that assume asingle process (e.g., Schneider & Shiffrin,1977). I will offer an opinion at a laterpoint (see Comparison Stage Revisited},but for now, let us consider each task sepa-rately. The focus on visual comparison inthe remaining experiments reflects my ini-tial interest in visual search (see Logan,1975).

Two versions of the visual-search taskappear in the literature, yes-no> and forcedchoice. The yes-no version requires a yesresponse if a target item appears in thearray, a no response otherwise. The forced-

choice version involves two target items,each associated with a separate response.One of the targets is presented on each trial,and the task is to decide which of the twoit was. The forced-choice version offerstwice as many observations per data pointin the same number of trials, and so it waschosen over yes-no in the present experi-ments.

There has been some speculation thatforced choice may involve different processesthan the yes-no version (Holmgren, 1974b),but the supportive evidence is not compel-ling (cf. Experiment 5). On the contrary,the similarities are quite striking: Reactiontime increases with array size in both yes-no (Atkinson et al., 1969; Egeth, Atkin-son, Gilmore, & Marcus, 1973; Novik &Katz, 1971; Schneider & Shiffrin, 1977;Townsend & Roos, 1973) and forced-choiceversions (Estes, 1972; Estes & Wessel, 1966 ;Holmgren, 1974b; Logan, 1976), and whenthe number of items relevant to the com-parison process is reduced to one by indicat-ing the position of the target with a barmarker, reaction times are reduced in propor-tion to array size in both yes-no (Holmgren,1974a) and forced-choice versions (Logan,Withey, & Cowan, 1977). Indeed, the inter-action between array size and the presenceor absence of a bar marker common to bothversions suggests that they do, in fact, affectthe same comparison stage.

Experiment 3In this experiment, the comparison stage

was denned by two parameters, array sizeand the presence or absence of a bar marker,both of which determined the number ofitems relevant to the comparison process.The visual task was performed both aloneand in the retention interval of a memorytask that required ordered recall of 7 digits,The attention demands of the comparisonstage were to be indicated by interactionsbetween memory load and the defining pa-rameters : An effect that increased mono-tonically with the number of relevant com-parisons would indicate a demanding com-parison process, whereas a constant effectover all numbers of relevant items wouldindicate an automatic one.

46 GORDON D. LOGAN

Method

Experiments 3-8 departed from the previousmethod in two major ways: First, array size wasvaried, so the visual stimuli differed as describedbelow, and second, subjects were given no feed-back regarding their performance (i.e., the mem-ory-task feedback was omitted). Nevertheless, thememory task was still defined as primary, and thesubjects were exhorted to do their best on it.

Apparatus and stimuli. In Experiments 3-8 thevisual stimuli were arrays containing 4, 8, or 12different letters equally spaced around an imagi-nary circle centered on the fixation point. Eacharray contained one target letter, an A or a V.Each array size was represented by 48 differentarrays in which each target letter appeared ineach position equally often. As far as samplinglimitations permitted, each nontarget letter (all re-maining letters except Q) appeared in each posi-

H

M

Figure 5. Examples of 4-, 8-, and 12-letter arrayswith a bar marker (Experiment 3) on the 8-letterarray and a noise mask (Experiment 5) on the12-letter array.

tion equally often. The arrays were made fromblack uppercase Letraset letters (#727) mountedon white cards. Examples of each array size areshown in Figure 5.

Experiments 3, 5, 6, 7, and 8 used a Gerbrandsthree-field tachistoscope (Model T-3B-1) with aviewing distance of 80 cm. At this distance, eachletter subtended about 26' X 26' of visual angle, andthe diameter of the imaginary circle subtended 4°of visual angle. A Scientific Prototype three-fieldtachistoscope (Model GB) was used in Experi-ment 4. Since its viewing distance was substan-tially longer, the stimuli appeared smaller: Eachletter subtended about 19' X 19' of visual angle, andthe diameter of the imaginary circle subtendedabout 2° S3' of visual angle.

In Experiment 3, a bar marker indicating theposition of the target was presented on half ofthe trials. The bar was made from a capital /(Letraset #727), and it appeared as an extensionof an imaginary radius outside the circle formedby the letters (see Figure 5). The bar subtendedabout 26' in length and 5' in width, and the sepa-ration between the bar and the target letter (A orV) was about 20' of visual angle.

The bar markers were exposed for 100 msec im-mediately before the arrays. To control for thealerting function of the bar marker (cf. Posner &Boies, 1971), in the no-bar-marker conditions, ablack dot was presented in the same interval priorto the arrays. It was made by filling in a capitalO (Letraset #727), and it appeared superimposedon the fixation point.

Procedure. Experiment 3 differed from the gen-eral procedure in that the memory digits wereread out loud by the experimenter, and recall waspermitted as soon as the subject responded to thevisual task.

Results

Visual task. The mean reaction times ineach combination of bar-marker, array-size,and memory-load conditions are displayedin Figure 6. Each point in the figure is basedon 384 observations. Inspection of the figuremakes it clear that reaction time increasedwith array size, F(2, 14) = 50.28, p < .01.However, the presence of a bar marker indi-cating the target's position reduced reactiontime substantially, F(l, 7) = 45.43, p < .01,and diminished the effect of array size sothat the interaction between them was sig-nificant, F(2, 14) = 33.12, p < .01. Theseeffects replicate previous work (e.g., Holm-gren, 1974a; Logan, Withey, & Cowan,1977) and serve to define comparison as aprocessing stage.

AUTOMATIC PROCESSING IN SEARCH TASKS 47

O 1100UJ</)S 1000ZUl 900s*" 800zop 700O2 600c

4 8 12

ARRAY SIZE

Figure 6. Mean reaction times in the visual taskin Experiment 3 as a function of array size. (Bar-marker and memory-load conditions are param-eters : B = bar marker ; NB = no bar marker ; L= memory load; UL = no memory load.)

The requirement that subjects retain 7digits during a visual trial increased reactiontimes overall by 59 msec, a difference thatwas highly significant, F(l, 7) = 35.75, p <.01. However, there was one perturbation inthis effect: Memory load increased reactiontimes to 8-letter arrays in the no-bar-markercondition by 95 msec, nearly twice the meanincrease of 52 msec for the other five com-parisons (which ranged from 46 to 57msec). This perturbation was large enoughto produce significant interactions betweenmemory load and array size, F(2, 14) =4.98, p < .05, and between memory-load andbar-marker conditions, F(l, 7) = 6.39, p <.05. The interpretation of these interactionsis deferred to the Discussion section.

Reaction time decreased significantly overdays, F(3, 31) = 57.71, p < .01, indicatinga practice effect that interacted only withbar-marker conditions, F(3, 21) = 7.11, p< .01. No-bar-marker conditions improvedmore than bar-marker conditions from Day1 to Day 2, but at the same rate on subse-quent days. The proportion of errors in eachcondition appears in Table 6.

Memory task. The mean number of digitsrecalled in each combination of array-sizeand bar-marker conditions appears in Table6. There were no significant effects in theanalysis of variance performed on these data.

Discussion

Overall, the results support the hypothesisthat comparisons between an array and a

target set can be carried out without theinvolvement of attention. Concurrent mem-ory load increased reaction time by a con-stant amount when 1 (the bar-marker con-ditions), 4, or 12 items were relevant tothe comparison process. Whereas the hy-pothesis that comparisons require attentionpredicted larger memory-load effects withmore items relevant to comparison, the hy-pothesis that comparisons occur automati-cally predicted the observed invariance.

The weak interactions between memoryload, bar-marker conditions, and array sizeshould not be troublesome because the pat-tern of the interactions was not predicted byeither hypothesis: The memory-load effectwas constant for all except 8-letter, no-bar-marker arrays, where it was larger. Non-monotonic interactions such as these are notinterpretable in the extended additive-fac-tors method, so the preferred interpretationis that they were spurious. Indeed, they werenot replicated in the remaining experiments(also see Logan, 1976).

The invariance of the memory-load effectbears on two current issues: First, theoriesof visual search are divided on the issue ofattcntional involvement in the comparisonprocess; some argue for attentional involve-ment (e.g., Atkinson et al., 1969; Rumel-hart, 1970; Schneider & Shiffrin, 1977), andsome argue against it (e.g., Estes, 1972,1974; Gardner, 1973a, 1973b; Shiffrin &Geisler, 1973). The invariant memory-loadeffect supports the argument against atten-

Table 6Proportion of Errors in the Visual Task andRecall Accuracy (in parentheses) in theMemory Task (Across Subjects and Days)as a Function of Visual-Task Conditionsin Experiment 3

Barmarker

Present0 digits7 digits

Absent0 digits7 digits

4

.04

.08 (6.12)

.06

.08 (6.16)

Array size

8

.04

.04 (6.33)

.OS

.09 (6.04)

12

.06

.06 (6.15)

.18

.19 (6.15)

GORDON D. LOGAN

tional involvement (also see Logan, 1976).Second, a number of recent studies haveused cuing techniques like the present bar-marker manipulation to assess attentional in-volvement. The effects of uncuecl items (orattributes) on performance have been of-fered as evidence of automatic processing,since attention is directed to the cued item(e.g., Butler, 1973, 1974; Dick, 1972;Keren, 1976; Mewhort, 1967; Willows,1974; Willows & MacKinnon, 1973; Neis-ser, Note 3). The invariant memory-loadeffect suggests that such cues may not directattention. Most theories agree that directingattention both reduces the demands onattention and improves processing (e.g.,Broadbent, 1958; Kahneman, 1973; Keele,1973; Treisman, 1969). Experiment 3 hasshown that the bar marker improved per-formance but did not reduce the demandson attention as measured by the memory-load effect (also see Gardner, 1973b).

The Encoding Stage

In the comparison stage, it is an internalrepresentation of the array that is comparedwith a memorized target set. Thus, it isnecessary to postulate the existence of men-tal operations prior to comparison whosefunction is to provide a suitable representa-titon. This is the function that is identifiedwith the term encoding.

Evidence that encoding is itself a process-ing stage, distinguishable from the compari-son stage, comes from memory-search ex-periments. Sternberg (1967) and Johnsenand Briggs (1973) have shown that de-grading an array by superimposing a visual-noise pattern (see Figure 5) increases reac-tion time by a constant amount, regardlessof target-set size. Significantly, the presenceof noise has no (interactive) effect on thearray-size function or the interaction be-tween array size and target-set size (John-sen & Briggs, 1973), Precisely this addi-tivity would be expected were encoding aseparate processing stage whose duration isaffected by stimulus quality.

Experiment 4

This experiment examined the automaticityof the encoding stage by varying stimulus

quality, array size, and the extent of a con-current memory load. Low-quality stimuliwere produced by superimposing a noisemask on the arrays. These were comparedwith normal arrays, without a superimposedmask.

On the hypothesis that encoding requiresattention, a larger memory-load effect waspredicted for low-quality stimuli than forhigh-quality ones (i.e., a significant inter-action between memory-load and maskingconditions). On the hypothesis that encod-ing is automatic, constant memory-load ef-fects were predicted for both high- and low-quality stimuli (i.e., no significant interactionbetween memory-load and masking condi-tions) .

Method

The major methodological departure here wasthe presentation of a masking stimulus superim-posed on the arrays on one half of the trials. Themasking stimulus was a random-noise mask (Let-ratone LT134) mounted on clear Plexiglas andplaced directly in front of the arrays to appearsuperimposed on top of them, filling the entirefield (see Figure 5). The arrays were exposed for1,500 msec to ensure a high level of accuracy inthe masking conditions, otherwise all methodo-logical details were as described in the generalmethod.

Results

Visual task. The mean reaction times foreach combination of masking, array-size, andmemory-load conditions are displayed inFigure 7. Each point in the figure is basedon 384 observations. Again, reaction timeincreased with array size, F(2, 14) = 63.88,p < ,01, and memory load, F(\, 7) = 32.26,p < ,01, but their effects were clearly addi-tive; the interaction between them was non-significant, F(2, 14) < 1, This supports theconclusion drawn from Experiment 3 thatthe comparison process in visual search isautomatic.

The presence of a superimposed noisemask increased reaction time overall by 68msec, F(l, 7} = 16,71, p < ,01, In the pres-ence of this substantial main effect, the non-significance of the Masking X Array Sizeinteraction, F(2, 14) < 1, suggests that en-

AUTOMATIC PROCESSING IN SEARCH TASKS 49

1100

I 1000

z 900

Ijj 800

§ 700

§ 6OO

tt 500

12ARRAY SIZE

Figure 7. Mean reaction times in the visual taskin Experiment 4 as a function of array size.(Masking and memory-load conditions are param-eters : M = mask ; UM = no mask ; L — memoryload; UL = no memory load.)

coding and comparison ma}' be treated asseparate processing stages. The nonsignifi-cance of the Masking X Memory Load in-teraction, F(l, 7) < 1, suggests further thatthe encoding stage functions without theinvolvement of central processing capacity.

As in Experiment 3, reaction time de-creased over days, F(3, 21) =27.97, p <.01. The interaction between array size andclays was significant, F(6, 42) = 2.81, p <.05, indicating a small but stable reductionin the array-size effect over days. The pro-portion of errors in each experimental con-dition appears in Table 7.

Memory task. The mean number of digitsrecalled in each combination of array-sizeand masking conditions appears in Table 7,The only significant effect in the analysis ofthe memory data was that of days, F(3, 21)— 20.15, p < .01, indicating a gradual im-provement in memory performance withpractice. The nonsignificance of all main ef-fects and interactions involving parametersof the visual task indicates that a constantamount of attention was allocated to thevisual task in all memory-load conditions.

Discussion

The mutual additivity of the effects ofarray size, masking conditions, and memoryload suggests that encoding and comparisoninvolve separate processing stages, each ofwhich functions automatically. The additiv-ity of the effects of array size and memory

load found here corroborates the interpreta-tion of Experiment 3, namely, that the in-teractions involving array size and memoryload were spurious.

The conclusions regarding the automatic-ity of the encoding stage must be qualifiedsomewhat. Encoding has been construed aspreparing a suitable representation for com-parison. Clearly, what is suitable depends onthe nature of the target set. Tasks that re-quire translation of the array to the formof the target set may well involve differentmental operations than those required here(e.g., Clifton, Cruse, & Gutschera, 1973),and such operations may require attention.Indeed, complementary translations on thetarget set would appear to demand attention(Millar, 1975; Posner & Klein, 1973).

Thus the nature of the task may be an im-portant determinant of the capacity demandsof component stages inasmuch as task de-mands determine the construction of eachstage. The demands of encoding, then, de-pend on the task. For simple search tasks(i.e., the present experiments), however,encoding appears to be automatic.

The Decision Stage

Once the encoded array has been com-pared with the target set, the evidence ac-cumulated in the comparison stage must beassigned to one of two equivalence classesdefined by instructions (i.e., A versus V).The decision stage is most often studied in

Table 7Proportion of Errors in the Visual Task andRecall Accuracy (in parentheses) in theMemory Task (Across Subjects and Days)as a Function of Visual-Task Conditionsin Experiment 4

Maskingcondition

Mask0 digits7 digits

No mask0 digits7 digits

4

.02

.04 (5.98)

.03

.03 (6.14)

Array size

8

.07

.08 (6.16)

.04

.04 (6.08)

12

.13

.17 (6.9S)

.08

.10 (6.16)

50 GORDON D. LOGAN

the yes-no version of character-classificationtasks by comparing reaction times associatedwith the different decision outcomes (i.e.,yes and no). Tn general, no decisions takelonger than yes decisions, and the differenceis consistent across array size or target-setsize (Nickerson, 1972; Steinberg, 1975),suggesting that comparison and decision in-volve separate stages.

Experiment 5

Array size, decision outcome, and memoryload were varied factorially to examine theautomaticity of comparison and decisionstages. On the hypothesis that the decisionstage requires attention, it was predictedthat memory load and decision outcomewould interact. On the hypothesis that thedecision stage functions automatically, addi-tive effects were predicted.

Method

The major departure in this experiment wasprocedural: Subjects were instructed to respondyes if an array contained an A, and to respondno otherwise. Although the letter V was presenton all arrays receiving no responses, subjects ap-parently did not take it into account in responding.After the experiment, subjects were asked how theydecided there was an A present, and none indi-cated knowledge of the (negative) correlation be-tween A and V. When the correlation was subse-quently described, 2 of 16 subjects reported no-ticing a preponderance of Ks, and one reportednoticing the correlation. No subject reportedsearching for a V to decide upon a no response.

iy 1000

- 900ul3= 800h-

Z 700O

t> 600

<*• 500

....-• NU

..." NUL

12

ARRAY SIZE

Figure S. Mean reaction times in the visual taskin Experiment 5 as a function of array size. (De-cision-type and memory-load conditions are pa-rameters : Y — yes; N = No; L — memory load;UL = no memory load.)

Table 8Proportion of Errors in the Visual Task andRecall Accuracy (in parentheses) in theMemory Task (Across Subjects') as a Functionof Visual-Task Conditions in Experiment 5

Decisiontype

Yes0 digits7 digits

No0 digits7 digits

4

.05

.09 (5.38)

.02

.02 (5.24)

Array size

8

.07

.07 (5.13)

.03

.02 (4.92)

12

.17

.17 (4.96)

.08

.15 (5.25)

Note that each subject served in only one session(see Table 1).

Decision outcome (yes-no) and array size variedrandomly within blocks. The basic four-block pro-cedure was retained, and memory-load conditionswere run in alternating blocks. Half of the subjectsbegan with a memory-load block, and half beganwith a no-memory-load block.

Results

Visual task. The mean reaction times foreach decision-outcome, array-size, and mem-ory-load condition are displayed in Figure 8.Each point in the figure is based on 192 ob-servations. Once again, reaction time in-creased significantly with array size, P(2,30) = 129.08, p < .01, and with memoryload, F(l, 15) = 27.91, p < .01, and onceagain, the interaction between them was notsignificant, F(2, 30) < 1. These findingscorroborate Experiments 3 and 4 but shouldbe interpreted with caution because of anom-alous results in the memory task.

Characteristic of the yes-no task, reactiontimes were significantly longer for no de-cisions than for yes decisions, F(l, 15) =74.16, p < .01. The slope of the no functionwas larger than the slope of the yes function,so much so that the interaction between de-cision outcome and array size was highlysignificant, F(2, 30) = 24.98, p < .01.

The requirement that subjects retain 7digits in memory increased yes reactiontimes by 86 msec and no reaction times by95 msec. The interaction between memoryload and decision outcome was not sig-nificant, F(l, 15) < 1, providing some sup-

AUTOMATIC PROCESSING IN SEARCH TASKS 51

port for the hypothesis that the decisionstage functions automatically. The propor-tion of errors in each condition is presentedin Table 8.

Memory task. The mean number of digitsrecalled in each combination of array sizeand decision outcome appears in Table 8.While the effect of decision outcome wasnot significant, F(\t I S ) < 1, both the maineffect of array size and the interaction ofarray size with decision outcome were sig-nificant, F(2, 30) = 8.78 and 3.78, respec-tively, p < .05. A closer inspection of thedata reveals a monotonic reduction in recallaccuracy as array size increased for yes de-cisions, and a more complex pattern for nodecisions: Recall accuracy was high with 4-and 12-letter arrays but low with 8-letterarrays.

Discussion

The reaction time data, considered bythemselves, are relatively clear: Memoryload did not interact with array size or de-cision type, although all three variables pro-duced substantial main effects. The invari-ance of the memory-load effect suggests thateach condition demanded the same amountof attention. The independence of attentiondemands and processing time, in turn, sug-gests that the stages defined by these pa-rameters function automatically. Moreover,the additivity of memory-load and decision-outcome effects replicates the additivityfound in Experiments 1 and 2 with memorysearch.

The memory data, however, are not soclear. The dependence of memory perform-ance on array size and decision outcome(jointly) suggests a trade-off between taskssuch that some visual conditions receivedmore attention than others. The pattern ofthe interaction, however, is difficult to inter-pret in terms of a systematic trade-off strat-egy that is sensitive to visual-task conditions,so it is possible that the effects were spuri-ous. Indeed, array size did not affect mem-ory performance in any other experiment(especially Experiment 8), nor did decisionoutcome in memory-search experiments.

The interaction between array size and

decision outcome is noteworthy, since theireffects are typically additive (e.g., Atkin-son et al., 1969; Holmgren, 1974a). Addi-tivity is often interpreted as indicating thatall items are compared before a decision ismade, that is, search is exhaustive, whilesteeper slopes for no than for yes decisionsoften suggest that the comparison processstops once sufficient evidence for a decisionhas accumulated, that is, search is self-ter-minating (see Sternberg, 1969a, 1975). Thedifference between the present results andearlier findings (Atkinson et al., 1969;Holmgren, 1974a) might reflect a differencein strategy: The earlier studies used smallerarrays (1-5 items) which would not benefitmuch from a self-terminating strategy. As-suming a search rate of 20 msec per item,the maximum benefit would be 80 msec. Bycontrast, the maximum benefit with the largearrays of the present study would be 220msec. Thus, experiments involving smallarrays may induce an exhaustive strategy,while large-array studies may induce a self-terminating strategy. Indeed, with arrays aslarge as those studied by Neisser and hiscolleagues (50 lines of 6 items), efficientsearch is necessarily self-terminating (e.g.,Neisser, 1963; Neisser et al., 1963).

The evidence for self-terminating searchis inconsistent with Holmgren's (1974b)claim that yes-no search is characteristicallyexhaustive and involves different operationsthan forced-choice search because forced-choice search is self-terminating. There isnothing in the present data to suggest thatyes-no and forced-choice search involvesubstantially different operations.

The Response Selection Stage

The fourth stage in the classification taskis the response-selection stage. At this stage,the subject has already classified the stimu-lus in terms of instructed equivalence classesand need only execute an appropriate re-sponse. The response is thought to be se-lected from a response set by the applicationof situation-specific rules for mapping equiv-alence classes onto the response set.

In Experiments 6 and 7 the response-selection stage was identified with the effects

52 GORDON D. LOGAN

of variations in the compatibility of stimu-lus-response (S-R) relations (Deininger &Pitts, 1955; Keelc, 1973; Welford, 1968).The S-R compatibility manipulation wasattractive because it had been associatedwith the response-selection stage, using theadditive-factors method (Eiedcrman & Kap-lan, 1970; Rabbitt, 1967), and because ithas been implicated as a determinant of theattention demands of tasks in concurrent-task situations (Broadbent & Gregory,1965; Crowder, 1967; Keele, 1967). Experi-ment 6 examined the compatibility of rela-tions between one stimulus set and one re-sponse set, while Experiment 7 comparedthe compatibility of two response sets withone stimulus set.

Experiment 6

This experiment used vocal responses. Inthe compatible condition (high S—R com-patibility), subjects spoke the name of thetarget letter present in the array. The in-compatible condition (low S-R compati-bility) was produced by simply reversingthe S-R mapping. Whereas "A" was thecorrect response to an A in the array in thecompatible condition, "V" became the cor-rect response in the incompatible condition.The advantage of this arrangement was thatstimulus conditions, response uncertainty,and the responses themselves remained un-changed as S-R compatibility was varied.

1100o 1000

i2 900

I 800

i 7°°O 600

c 500

8

ARRAY SIZE

12

Fiyitre 9. Mean reaction times in the visual taskiu Experiment 6 as a function of array size. (Stim-ulus-response compatibility and memory-load con-ditions are parameters : C = compatible ; 1C = in-compatible ; L — memory load; UL — no memoryload.)

Only the S-R mapping changed (c.f. Alluisi,Strain, & Thurmond, 1964; Broadbent &Gregory, 1962; Hawkins, MacKay, Holley,Friedin, & Cohen, 1973, Experiment 1).

On the hypothesis that response selectionrequires attention, it was predicted that thedifference between the S-R compatibilityconditions would increase when the task wasperformed with a concurrent memory load(i.e., a significant interaction between mem-ory-load and S—R compatibility conditions).On the hypothesis that response selectiondoes not require attention, it was predictedthat the difference between the S-R com-patibility conditions would not change asmemory load was manipulated (i.e., no sig-nificant interaction between memory loadand S-R compatibility).

Method

This experiment differed from the previous onesin that vocal responses were used. Again, the two-alternative (A or V) forced-choice procedure wasused.

Results

Visual task. The mean reaction times foreach combination of array-size, memory-load, and S-R compatibility conditions aredisplayed in Figure 9. Each point in the fig-ure is based on 384 observations. Reactiontime increased significantly with array size,P(2, 14) - 101.24, p < .01, and was'longerwith incompatible S-R relations, F(l, 7) =79.18, p < .01. The contention that arraysize and S-R compatibility affect differentstages was supported by the nonsignificanceof the interaction between them, F(2, 14)= 2.34, p > .05, and the nonsignificance ofall higher order interactions involving bothof them.

Unlike the previous experiments, thememory-load effect was small (26 msec)and failed to reach significance, F(l, 7) =3.74, .05 < p < .10. However, the interac-tion of memory load and clays was signifi-cant, F(3, 21) = 3.58, p < .05, indicating asubstantial (74 msec) memory-load effect onthe first day, but a negligible effect (10msec) on all subsequent days. While reac-tion times in all conditions decreased signifi-

AUTOMATIC PROCESSING IN SEARCH TASKS 53

cantly over days, F(3, 21) = 9.80, p < .01,the reduction was greatest in conditions in-volving a memory load.

In this experiment, as in the previous ex-periments, the simple interaction betweenmemory load and array size failed to reachsignificance, F(2, 14) < 1 ; however, thethree-way interaction between array size,memory load, and clays was significant, F(6,42) = 3.54, /) < .01. It reflects a smallermemory-load effect for larger arrays on thefirst day, whereas the memory-load effectwas consistent (and small) across array sizeon subsequent days. This pattern of differ-ences would not be predicted on the hy-pothesis that the comparison stage requiresattention, and it need not concern us further.

The interaction between memory load andS-R compatibility was also nonsignificant,/?(!, 7) = 3.45, p > .05. If anything, therewas a tendency for the memory-load effectto be smaller in the incompatible conditionon the first day, clearly contradicting theprediction that the attention demands of atask are greater the lower the degree of S-Rcompatibility (c.f. Broaclbent & Gregory,1965; Crowder, 1967, Keele, 1967). Ofcourse, the present experiment exploitedonly one type of S-R compatibility (Dein-inger & Fitts, 1955). The proportion of er-rors in each condition (across subjects anddays) is presented in Table 9.

Memory task. Table 9 contains the meannumber of digits recalled for each combina-tion of array-size and S-R compatibilityconditions. In the analysis, only the effectof days was significant, F(3, 21) = 17.23.Memory performance improved over daysby the same amount in each condition. Thusthere is no evidence here suggesting a trade-off between recall accuracy and reactiontime.

Discussion

The contention that S-R compatibilityaffects a response-selection stage separatefrom the comparison stage was supported bythe additivity of the effects of array sizeand S-R compatibility. In this context, theadditivity found between memory load andS-R compatibility becomes meaningful: In-

Table 9Proportion of Errors in the Visual Task andRecall A ccuracy (in parentheses') in theMemory Task (Across Subjects and Days)as a Function of Visual-Task Conditionsin Experiment 6

Stimulus-

compatibility 4

Array size

8 12

Compatible0 digits .02 .06 .167 digits .01 (5.95) .03 (5.89) .15 (5.94)

Incompatible0 digits .03 .08 .177 digits .01 (5.80) .06 (5.88) .13 (5.98)

compatible S-R relations may increase thetime required for response selection withoutaffecting the attention it demands. This con-clusion is further supported by the findingthat the compatibility effect remained con-stant over days while the memory-load effect(the index of attention demands) diminished.

Experiment 7

This experiment compared the attentiondemands of vocal and manual responses.Subjects either spoke the name of the targetletter presented (A or V} or pressed the ap-propriate button. A significant interaction be-tween memory load and response type wouldindicate differential attention demands, whileadditivity would indicate equivalent atten-tion demands.

Method

This experiment differed from the previous onesin that two response buttons as well as a voice-activated relay were connected to a millisecondtimer so that it would be slopped by vocal ormanual responses. Vocal- and manual-responseconditions were run in separate blocks in themanner described in the General Method. Again,the two-alternative (A or V) forced-choice pro-cedure was used.

Results

Visual task. Mean reaction times foreach combination of array-size, memory-load, and response-type conditions are dis-

54 GORDON D. LOGAN

a 100°| 900

- 800sp 700

O 600

< 5OO

ARRAY SIZE

Figure 10. Mean reaction times in the visual taskof Experiment 7 as a function of array size. (Re-sponse-type and memory-load conditions are pa-rameters : V = vocal; M = manual; L = memoryload; UL — no memory load.)

played in Figure 10. Each point in the figureis based on 384 observations. Once again,reaction time increased with array size, F(2,14) = 38.58, p < .01, and with memory load,77(1, 7) - 21.18, p < .01, and once again,the interaction between array size and mem-ory load was not significant, F(2, 14) < 1,further supporting the possible automaticityof the comparison stage.

Overall, vocal reaction times were 25msec longer than manual reaction times, butthe difference was not significant, P(l, 7)— 2.52, p > .05, nor were the interactionsbetween response type and memory load,I ' ( I , 7) < 1, and response type and arraysize, F(2, 14) < 1. Reaction time decreasedover days, F(3, 21) - 36.40, p < .01, andthe array-size effect diminished with prac-tice, F(6, 42) = 7.19, p < .01. The propor-tion of errors in each combination (acrosssubjects and clays) is displayed in Table 10.

Memory task. The mean number of digitsrecalled in each combination of array-sizeand response-type conditions appears inTable 10. Once again, only the effect of dayswas significant in the analysis, F(3, 21) =18.90, p < .01.

Discussion

This experiment provides no evidence thatvocal and manual responses differ in thedemands they place on attention, which, to-gether with the results of Experiment 6,suggests that- the response-selection stage

functions automatically. Thus, it appearsthat no stage in visual search requires atten-tion. The consistent main effect of memoryload suggests that the task demands atten-tion, but the demands canot be localized inany stage. This surprising conclusion willbe discussed presently; but now let us con-sider a final investigation of memory-loadeffects.

Experiment 8

It is significant that none of the param-eters of the visual task had reliable effects onmemory performance. This is further evi-dence that none of the stages associated withthe parameters take attention. However, thehigh level of recall accuracy over all experi-ments (84%) is problematic in that ceilingeffects may have 'masked true differences inmemory performance. Experiment 8 wasrun as a parametric variation of memoryload and array size designed to produce datafree of ceiling effects so that any hidden in-fluence of visual-task parameters couldemerge.

Method

The experimental conditions were the factorialcombinations of three levels of array size (4, 8, or12 letters) and four levels of memory load (0, 4, 8,or 12 words). The experiment was run as a single-session 144-trial experiment divided into fourblocks of 36 trials. Array size varied randomlywithin blocks, and memory load varied between

Table 10Proportion of Errors in the Visual Task andRecall Accuracy (in parentheses) in theMemory Task (Across Subjects and Days)as a Function of Visual-Task Conditionsin Experiment 7

Responsetype

Vocal0 digits7 digits

Manual0 digits7 digits

4

.08

.09 (6.07)

.01

.00 (5.78)

Array size

8

.10

.06 (6.09)

.03

.03 (S.85)

12

.18

.15 (6.09)

.14

.13 (5.91)

AUTOMATIC PROCESSING IN SEARCH TASKS 55

blocks in four orders derived from a balanced4 X 4 Latin square (see General Method),

The memory stimuli were drawn from theToronto word pool, a pool of 1,056 two-syllablewords from the AA and A categories of theThorndike-Lorge (1944) word-frequency count.Lists of 4, 8, and 12 words were recorded at anapproximate rate of 1.24 sec per word, precededby a warning signal and followed by a readysignal about 2.5 sec after the last word. The reten-tion interval was about 8.5 sec.

Results and Discussion

Visual task. Mean reaction times for eachcombination of array-size and memory-loadconditions are displayed in Figure 11. Eachpoint in the figure is based on 192 observa-tions. The effects of both array size andmemory load were significant, F(2, 30) =123.74, and F(3, 45) = 5.74, respectively,p < .01, but the interaction between themwas not, F(6, 90) < 1. This replicates theprevious findings of additivity under moresevere loads. Notice that reaction time in-creased monotonically with memory loadfor loads of 0, 4, and 8 words but wasslightly (not significantly) reduced for loadsof 12 words. The monotonic increase rep-licates the results of Shulman and his col-leagues (Shulman & Greenberg, 1971; Shul-man et al., 1971), but the reduction with12-word loads does not. However, Shulman'sexperiments used a maximum load of 10items, and it is possible that the effect reachesasymptote between 8 and 12 items.

Memory task. The mean numbers of

sz

I900

800

700

600

500

4 8 12

ARRAY SIZE

Figure 11. Mean reaction times in the visual taskof Experiment 8 as a function of array size. (Thenumber of words concurrently in memory is theparameter.)

Table 11Proportion of Errors in the Visual Task andRecall Accuracy (in parentheses) in theMemory Task (Across Subjects) as a Functionof Visual-Task Conditions in Experiment 8

Memoryload

0 words4 words8 words

12 words

4

.01

.00 (3.80)

.02 (4.60)

.00 (4.89)

Array size

8

.07

.09 (3.75)

.08 (4.63)

.12 (4.84)

12

.22

.23 (3.75)

.22 (4.47)

.20 (4.74)

words recalled in each array-size and mem-ory-load condition appear in Table 11. Sincefew subjects recalled more than one wordin correct order, the order requirementswere relaxed and a word was scored correctif it appeared in any position in the list.

The number of words recalled increasedwith the number presented, F(2, 30) =43.72, p < .01, but was unaffected by arraysize. Neither the main effect of array size,F(2, 30) = 1.67, nor the interaction of arraysize with memory load, F(4, 60) < 1, ap-proached significance. Notice that the lackof array-size effects with loads of 8 and 12words cannot be attributed to ceiling effects,as the number of words recalled in each ofthese conditions was considerably less thanthe number presented.

A final point of interest is that signifi-cantly more words were recalled from 12-word lists than from 8-word lists, £(30)= 3.17, p < .01, yet 12-word lists had aslightly smaller effect on reaction time. Theincrease in recall, however, was small (.25words) and might not be expected to in-fluence reaction time, especially near theasymptote of the memory-load effect.

General Discussion

Most theories of attention would predictthat a concurrent primary task will increasevisual reaction time by an amount propor-tional to the attention demands of the visualtask. Sternberg's (1969a, 1969b) additive-factors method was used in an attempt toassociate observable attention demands with

56 GORDON D. LOGAN

particular processing stages. The amount ofinteraction between concurrent task difficulty(memory load) and parameters known toinfluence a particular processing stage wasa measure of the attention demands of thatstage. The method was shown to be bothinternally consistent and consistent withanother method that assesses attentional in-volvement. It was then applied to the analysisof four stages sufficient to perform a visual-search task.

Memory load did not interact with param-eters associated with encoding, decision, andresponse-selection stages, and of threeparameters associated with the comparisonstage (array size, presence or absence of abar marker, and target-set size), only target-set size interacted significantly with memoryload. Thus it appears that a great deal ofprocessing in classification tasks is auto-matic.2 The implications of this conclusionare discussed in the next four sections.

Structural and Capacity Models of Attention

Structural models of attention consist ofa limited-capacity channel, a filter to protectthe channel from overload, and preliminaryprocesses to provide the informational basisfor selection. Since the preliminary processesand the filter are outside the channel, theyare not subject to capacity limitations. Sincethey precede the channel in the informationflow, a clear pattern of attentional involve-ment across processing stages may be pre-dicted : Early stages will not demand at-tention, but stages closer to the effectors will.

The point of transition represents theboundary of the limited-capacity channel,and all subsequent stages should require at-tention, as they are inside the channel.3

Thus, structural models predict a step func-tion of attentional involvement across stages.

Capacity models, on the other hand, makeno predictions about the locus of attentionalinvolvement, as attention is an energisticproperty apart from structure. Any patternof attentional involvement across stages isconsistent with capacity models.

The results of the present experiments areinconsistent with structural models. The

visual-search experiments (Experiments 3~8) examined four stages sufficient to performthe task and found no evidence of attentionalinvolvement in any of them. The memory-search experiments (Experiments 1 and 2)showed some evidence of attentional involve-ment in the comparison stage, but none inthe subsequent decision stage. Thus the ob-served pattern of attentional involvementacross stages was not the step function pre-dicted by structural models. Of course, theobserved pattern is consistent with capacitymodels.

If we accept that the same comparisonstage is involved in visual search and mem-ory search, the experiments have demon-strated that all parameters that affect a stagedo not contribute equally to its attention de-mands. The demands of comparison dependon target-set size but not array size or bar-marker conditions (also see Howard, 1975,and Logan, 1976). Furthermore, since therange of parameters that might affect thestages studied was not sampled exhaustively,it is possible that some new parameter (s)may show evidence of attention demands inone (or more) of the stages. The demandsof response selection, for example, may de-pend on the number of possible responses,but not on the compatibility of various S-Rarrangements. A more conservative con-clusion, then, is that some aspects of allprocessing stages are automatic.

This challenges the structural view of aunified stage of attentive analysis in that allcomponents of a stage need not contributeto its attention demands. By contrast, ca-pacity models are comfortable with this sug-

2 Howard Egeth (1977) lias also found evidencefor automatic encoding, discrimination, and re-sponse selection in choice reaction time using amemory-load procedure.

3 The boundary of the limited-capacity channelhas not yet been established unequivocally. Somebelieve it occurs after processing of physical fea-tures (e.g., Broadbent, 19S8), and others believeit occurs after semantic analysis (e.g., Deutsch &Deutsch, 1963). For the present purposes, the locusis not as important as the transition itself, since asharp transition in attention demands is evidencesupporting structural models. The locus mattersonly if structural models provide the best descrip-tion of the data.

AUTOMATIC PROCESSING IN SEARCH TASKS 57

gestion. Attention refers to the activation ofstructural units though the units themselvesmay not be denned. The subsequent discus-sion suggests steps toward definition.

What Takes Attention?

It was clear throughout the experimentsthat the visual task required attention—themain effect of memory load was quite reli-able—yet there was no clear evidence thatattention was involved in the execution ofany stage's function. The components ofthe task seem automatic, but the task itselfis not. Perhaps the stage structure of a taskshould be viewed as a temporary assemblyof automatic, special-purpose processors or-ganized by attention to deal with the task athand. This is not unreasonable, given therather arbitrary nature of the sequence ofstages we have studied. It may take atten-tion, then, to prepare an appropriate se-quence of structures and maintain it until astimulus has appeared.

There is currently a great deal of sym-pathy for this notion among attentiontheorists (e.g., LaBerge, 1973a, 1973b;Posner & Boies, 1971 ; Posner & Snyder,1975a, 1975b; Thomas, 1974), for the em-pirical properties of preparation suggest anattentional process. Preparation improvesreaction time in that prepared reactionsare faster than unprepared ones (Beller,1971; Bertelson, 1967; Posner & Boies,1971), but the improvement with prepara-tion is limited to a narrow range of stimuliand is bought at the cost of less efficientprocessing for stimuli outside this range(Gottsdanker, 1975; LaBerge, 1973a; Pos-ner & Snyder, 197Sb; Rosch, 1975; Rosch,Simpson, & Miller, 1976).

Clear evidence for the involvement of at-tention in maintaining preparation is avail-able in Comstock's (1975) doctoral thesis.Her subjects performed a primary letter-matching task in which a warning signaland two letters to be matched were pre-sented in succession. The momentary at-tention demands of the letter-matching taskwere assessed from reaction times to anauditory probe presented at various timesthroughout the time course of the visual

task. As is typically found in these studies,probe reaction times were uniform through-out the warning interval and the presenta-tion of the first letter but rose substantiallyaround the presentation of the second letter,presumably reflecting the greater attentiondemands of response-related processes (Corn-stock, 1973; Millar, 1975; Posner & Boies,1971; Posner & Klein, 1973). When theprobe task was run as a single-task control(i.e., the visual events were the same butsubjects ignored them and responded onlyto the probes), reaction times were uniformthroughout the warning interval and thepresentation of both letters. Interestingly,single-task probe reaction times were con-sistently faster than their dual-task counter-parts, even during the intertrial interval,when the visual task could not have de-manded attention lor execution, Becker(1976) found a similar elevation of reactiontime when subjects expected a second taskbut did not have to perform it.

The memory-load effects in the presentexperiments may be explained in terms ofpreparation. The primary (memory) taskmay have required so much attention thatthe amount remaining was not sufficient tooptimally prepare the secondary (visual)task, and consequently reaction times in-creased.4 Target-set size was the only param-eter to interact with memory load becauseit was the only one to vary the amount ofpreparation necessary to perform the task(i.e., 1, 2, or 4 targets). Indeed, the memory-search experiments were conducted afterthe visual-search studies to test the hy-pothesis that attention demands depend onthe amount to be prepared not on the amountto be processed.

Capacity models provide a theoretical linkbetween attention and preparation. Prep-aration may result from allocating capacity

* An interesting implication of Comstock's (197S)findings is thai subjects need not have executeddemanding operations in the memory task duringthe reaction time interval for reaction time to in-crease ; preparing to execute the operations mayhave been sufficient. Note that this excludes anyinterpretation based on switching attention fromthe memory task to the visual task.

58 GORDON D. LOGAN

to activate task-relevant structures beforethe stimulus appears. Stimulus presentationwill then provide a pattern of activationthat when combined with the preparatoryattentional activation will determine an ap-propriate response. Attentional activationmay organize a sequence of structures toform a "path of least resistance" which mo-mentarily dominates habitual response tend-encies. This principle is developed in greaterdetail in the next section.

Comparison Stage Revisited

A model of the comparison stage mustaccount for the seemingly demanding natureof memory search and the automatic natureof visual search found in the present experi-ments. Similar findings have been reportedwhen subjects change the instructed S-Rmapping in response to a signal in the arraywhile engaged in search: Changing the S-Rmapping increases the slope of the target-set-size function (Howard, 1975) but notthe array-size function (Logan, 1976). Theprinciple of preparation may be developedto provide an appropriate model.

The structure underlying the comparisonstage might involve units, for example, setsof feature detectors, capable of representingitems from the visual array and the targetset by characteristic patterns of activation.The activation of the array items is presum-ably an automatic consequence of stimuluspresentation; indeed, this is what the presentstudies suggest (particularly Experiment 4).By contrast, target-set items require an in-ternal source of activation, since the targetset is usually presented long before thearray. Central processing capacity is a likelycandidate here, particularly in view of theresults of Experiments 1 and 2 (also seeRaddeley & Ecob, 1973; Corballis, 1975).

The comparison process might involve theintegration of the two sources of activationin the structure once the array is presentedso that units representing items common tothe array and the target set will be activatedmore than the other nonmatching items fromeither set. The additional activation is thusa signal to be detected in the presence of

noise from the activation of nonmatchingitems, and a response can be emitted oncethe signal-to-noise ratio exceeds a threshold.If the integration process is extended intime, the time to reach threshold (and thusreaction time) will be longer the lower thesignal-to-noise ratio (see Anderson, 1973,for a formal development, and Estes, 1972,1975a, 1975b, for similar views). Such aprocess can produce both the separate andthe joint effects of array size and target-setsize reviewed earlier.

The consistent additivity of memory loadand array size in the present studies sug-gests that integration is an automatic proc-ess, initiated by stimulus presentation (alsosee Logan, 1976). The interaction of mem-ory load and target-set size, then, reflectsthe use of attention in preparing the targetset. If memory load reduced the preparatoryactivation of the target set, lower signal-to-noise ratios would result. The effect wouldbe greater with larger target sets, thus theinteractions. Alternatively, the target-setitems may be optimally activated under allconditions so that their demands producedeficits in other parts of the structure. This,too, would yield appropriate interactions.The automatizing of memory search withspecific practice seen in Experiment 1 pre-sumably reflects the automatizing of pre-paratory processing. Perhaps with sufficientpractice, the target set may be prepared asa unit so attention demands are equal forall set sizes.

However, this explanation is by no meansunique. Most of the effects could be ac-counted for by a model that assumes twoseparate serial comparison processes, one forthe array and one for the target set (Town-send & Roos, 1973). Indeed, the interactionof concurrent activity with target-set size(also see Howard, 1975, and Sternbeirg,1969b, Experiment 5) but not with arraysize (also see Logan, 1976) supports thenotion of separate processes. Alternatively,Schneider and Shiffrin's (1977) single-process model can explain most of the re-sults except the different memory-load ef-fects in visual search and memory search.In their model, both visual search and mem-

AUTOMATIC PROCESSING IN SEARCH TASKS 59

ory search are under attentional control aslong as their respective slopes are greaterthan zero, and both should show interactionswith memory load. Of course, the modelcould be changed to accommodate the presentfindings.

Ultimately, the choice between thesemodels and the preparation model is amatter of parsimony and aesthetics. Theserial, self-terminating comparison processcentral to both dual- and single-processmodels would require considerable atten-tional control during the execution of com-parisons. In view of the present evidence forwidespread automaticity in classificationtasks, this is not a desirable feature. By con-trast, the preparation model performs thesame function without the attendant com-plexity. Only two components require at-tentional control, the activation of the targetset and, possibly, the threshold setting.Once these are prepared, the comparisonprocess will run off automatically whenevera stimulus is presented.

Broader Implications

In the preparation model, attentional con-trol was accomplished by activating part ofa structure. Activating the target set effec-tively "harnessed" the automatic processesof encoding and comparison so they wouldparticipate in activity directed toward pro-ducing a response. It may prove useful todistinguish points in the structure that at-tention can control from those that it cannot.The control points may well be the wordsof the language with which the mind talksto itself. Presumably, other control pointsmust also be activated to assemble an ap-propriate structure, and each would con-tribute to the demands of the task. If a num-ber of control points are consistently ac-tivated together, however, they may becomeassociated so that activating one of them isenough to activate all of them. The de-velopment of automaticity might reflect theassociative combination of control structures(cf. Experiment 1). Thus, the units of at-tention may be entire tasks as well as in-dividual mental operations.

The dual role of attention in preparationand execution suggests a hierarchy of at-tentional control. In performing a task, ca-pacity is first allocated to control structuresto prepare the task as a whole for execution.Upon stimulus presentation, control mayshift to the level of component operations;attention must be allocated to those opera-tions (if any) that require it for execution.This shift is apparent in Comstock's (1975)data discussed earlier: Concurrent-probereaction times were longer than single-taskcontrols throughout the warning intervaland the presentation of the first letter whenthe visual task could not have required at-tention for execution. This difference re-flects the demands of maintaining a set toperform the visual task. When the secondletter appeared, concurrent-probe reactiontimes increased sharply, but single-task con-trols did not. This further increase mightreflect the additional demands of execution.

Finally, the emphasis on preparation andthe maintenance of preparation is reminis-cent of the notion of mental set in the earlyattention literature (see Woodworth, 1938).Attention literally sets the mind to respondin a certain way to particular aspects of stim-ulation. Stimulus presentation effectivelyreleases the automatic processes comprisingthe set, and a response occurs without fur-ther attentional involvement. Historical an-tecendents may be found in discussions ofthe prepared reflex in the early introspectivestudies of reaction time (see Woodworth,1938). Little of the reaction itself was foundto be available to introspection, but the re-sponse to the instructions and the act ofpreparation preceding the reaction werereadily available. The major cognitive in-volvement seemed to occur before the stim-ulus was presented, and the reaction itselfseemed reflexive. Following Posner's specu-lative identification of capacity with con-sciousness (Posner & Klein, 1972; Posner& Snyder, 1975a), the evidence presentedhere for the automaticity of execution con-trasted with the demanding nature of prep-aration echoes the earlier findings: Behaviorin reaction time situations still appears tobe nothing more than a prepared reflex.

60 GORDON D. LOGAN

Conclusions

The experiments suggest that attention isbest construed as the selective allocation ofcentral processing capacity to activate mentalstructures, either individually or in concertas sets to perform a specific task. The auto-maticity of aspects of encoding, comparison,decision, and response-selection stages im-plicates preparation for the task as a wholeas an important determinant of attention de-mands. The interactions between memory-load and target-set size provide some sup-port for this view. Finally, the emphasis onmental set suggests that the most interestingthings in visual search might go on beforethe array is presented.

Reference Notes

1. Posner, M. L, & Keele, S. W. Time and spaceas measures of mental operations. Paper pre-sented at the 78th Annual Meeting of theAmerican Psychological Association, MiamiBeach, Florida, September 1970.

2. Sternberg, S. Scanning a persisting visual imageversus a memorised list. Paper presented at the38th Annual Meeting of the Eastern Psycho-logical Association, Boston, April 1967.

3. Neisser, U. Selective reading: A method for thestudy of visual attention. Paper presented at theXIX International Congress of Psychology,London, 1969.

References

Alluisi, E. A., Strain, G. S., & Thurmond, J. B.Stimulus-response compatibility and the rate ofgain of information. Psychonomic Science, 1964,1, 111-112.

Anderson, J. A. A theory for the recognition ofitems from short memorized lists. PsychologicalReview, 1973, 80, 417-438.

Atkinson, R. C, Holmgren, J. E., & Juola, J. F.Processing time as influenced by the number ofelements in a visual display. Perception & Psy-chophysics, 1969, 6, 321-326.

Atkinson, R. C., & Shiffrin, R. M. Human mem-ory : A proposed system and its control processes.In K. W. Spence & J. T. Spence (Eds.), Thepsychology of learning and motivation: Advancesin research and theory (Vol. 2). New York:Academic Press, 1968.

Baddcley, A. D., & Ecob, J. R. Reaction-time andshort-term memory: Implications of repetitioneffects for the high-speed exhausive scan hy-pothesis. Quarterly Journal of ExperimentalPsychology, 1973, 25, 229-240.

Becker, C. A. Allocation of attention during visualword recognition. Journal of Experimental Psy-chology: Human Perception and Performance,1976, 2, S56-S66.

Heller, H. K. Priming: Effects of advance infor-mation on matching. Journal of ExperimentalPsychology, 1971, 87, 176-182.

Bertelson, P. The time course of preparation.Quarterly Journal of Experimental Psychology,1967, 79, 272-279.

Biederman, I., & Kaplan, R. Stimulus discrimin-ability and S-R cotnpatability: Evidence forindependent effects in choice reaction time.Journal of Experimental Psychology, 1970, 86,434-439.

Briggs, G. E., & Blaha, J. Memory retrieval andcentral comparison times in information proc-essing. Joural of Experimental Psychology, 1969,79, 395-402.

Briggs, G. E., & Johnsen, A. M. On the nature ofcentral processing in choice reations. Memory &Cognition, 1973, 1, 91-100.

Briggs, G. E., & Swanson, J. M. Encoding, de-coding, and central functions in human in-formation processing. Journal of ExperimentalPsychology, 1970, 86, 296-308.

Broadbcnt, D. E. Perception and communication.New York: Pergamon Press, 1958.

Broadbent, D. E., & Gregory, M. Bonders' B-and C- reactions and S-R compatibility. Journalof Experimental Psychology, 1962, 63, 575-578.

Broadbent, D. E., & Gregory, M. On the inter-action of S-R compatibility with other variablesaffecting reaction time. British Journal of Psy-chology, 1965, 56, 61-67.

Broadbent, D. E., & Heron, A. Effects of a sub-sidiary task on performance involving immediatememory by younger and older men. BritishJournal of Psychology, 1962, 53, 189-198.

Brooks, L. R. Spatial and verbal components ofthe act of recall. Canadian Journal of Psy-chology, 1968, 22, 349-368.

Brown, J. Some tests of the decay theory of im-mediate memory. Quarterly Journal of Experi-mental Psychology, 1958, 10, 12-21.

Butler, B. Components of the familiarity effect intachistoscopic recognition. Journal of Experi-mental Psychology, 1973, 101, 367-372.

Butler, B. The limits of selective attention intachistoscopic recognition. Canadian Journal ofPsychology, 1974, 28, 199-213.

Chow, S. L., & Murdock, B. B., Jr. The effect ofa subsidiary task on iconic memory. Memory &Cognition, 1975, 3, 678-688.

Chow, S. L., & Murdock, B. B., Jr. Concurrentmemory load and the rate of readout from iconicmemory. Journal of Experimental Psychology:Human Perception and Performance, 1976, 2,179-190.

Clifton, Q, Cruse, D., & Gutschera, K. D. Reced-ing processes in recognition: Some effects ofpresentation rate. Memory & Cognition, 1973, 1,387-394.

AUTOMATIC PROCESSING IN SEARCH TASKS 61

Comstock, E. M. Processing capacity in a. letter-match task. Journal of Experimental Psychology,1973, 100, 63-72.

Comstock, E. M. Limited-capacity central at-tention mechanisms. Doctoral dissertation, Uni-versity of Massachusetts, 1975.

Connor, J. M. Effects of increased processing loadon parallel processing from visual displays.Perception & Psychophysics, 1972, 12, 121-128.

Corballis, M. C. Access to memory: An analysisof recognition times. In P. M. A. Rabbitt &5. Dornic (Eds.), Attention and performance V.New York: Academic Press, 1975.

Corballis, M. C., Roldan, C. E., & Zbrodoff, J.Response set effects in recognition memory.Memory &• Cognition, 1974, 2, S01-S08.

Crowder, R. G. Short-term memory for words witha perceptual-motor interpolated activity. Journalof Verbal Learning and Verbal Behavior, 1967,6, 7S3-761.

Darley, C. F., Klatzky, R. L., & Atkinson, R. C.Effects of memory load on reaction time. Journalof Experimental Psychology, 1972, 96, 232-234.

Deininger, R. L., & Fitts, P. M. Stimulus-responsecompatibility, information theory, and perceptual-motor performance. In H. Quastler (Ed.), In-formation theory in psychology. Glencoe, 111.:Free Press, 1955.

Deutsch, J. A., & Deutsch, D. Attention: Sometheoretical considerations. Psychological Revieiv,1963, 70, 80-90.

Dick, A. O. Parallel and serial processing in tach-istoscopic recognition: Two mechanisms. Journalof Experimental Psychology, 1972, 96, 60-66.

Doost, R., & Turvey, M. T. Iconic memory andcentral processing capacity. Perception & Psy-chophysics, 1971, 9, 269-274.

Dyer, F. N. The Stroop phenomenon and its usein the study of perceptual, cognitive, and re-sponse processes. Memory & Cognition, 1973, 1,106-120.

Egeth, H. Attention and preattention. In G. H.Bower (Ed.), The psychology of learning andmotivation (Vol. 11). 1977.

Egeth, H., Atkinson, J., Gilmore, G., & Marcus, N.Factors influencing processing mode in visualsearch. Perception &• Psychophysics, 1973, 13,394-402.

Estes, W. K. Interactions of signal and backgroundvariables in visual processing. Perception &Psychophysics, 1972, 12, 278-286.

Estes, W. K. Redundancy of noise elements andsignals in visual detection of letters. Perception&• Psychophysics, 1974, 16, 53-60.

Estes, W. K. Memory, perception and decision inletter identification. In R. L. Solso (Ed.), In-formation processing and cognition: The1 Loyolasymposium. Hillsdale, NJ.: Erlbaum, 1975. (a)

Estes, W. K. Structural aspects of associative mod-els for memory. In C. N. Cofer (Ed.), Thestructure of human memory. San Francisco:Freeman, 1975. (b)

Estes, W. K., & Taylor, H. A. A detection methodand probabilistic models for assessing informa-tion processing from brief visual displays. Pro-ceedings of the National Academy of Science,1964, 52, 446-454.

Estes, W. K., & Wessel, D. C. Reaction time inrelation to display size and correctness of re-sponse in forced-choice visual signal detection.Perception &• Psychophysics, 1966, 1, 369-373.

Gardner, G. T. Evidence for independent channelsin tachistoscopic perception. Cognitive Psychol-ogy, 1973, 4, 130-155. (a)

Gardner, G. T. Parallel perceptual processing anddecisional strategies: A reinterprctation of theShaw and LaBerge effect. Perception fr Psycho-physics, 1973, 13, 517-518. (b)

Gottsdanker, R. The attaining and maintaining ofpreparation. In P. M. A. Rabbitt & S. Dornic(Eds.), Attention and performance V. NewYork: Academic Press, 1975.

Graboi, D. Searching for targets: The effects ofspecific practice. Perception &• Psychophysics,1971, 10, 300-304.

Hawkins, H. L., MacKay, S. L., Holley, S. L.,Friedin, B. D., & Cohen, S. L. Locus of the rela-tive frequency effect in choice reaction time.Journal of Experimental Psychology, 1973, 101,90-99.

Henderson, L. Spatial and verbal codes and thecapacity of STM. Quarterly Journal of Experi-mental Psychology, 1972, 24, 485-495.

Holmgren, J. E. The effect of a visual indicator onrate of visual search: Evidence for processingcontrol. Perception & Psychophysics, 1974, 15,544-550. (a)

Holmgren, J. E. Visual search in a forced-choiceparadigm. Perception & Psychophysics, 1974, 16,253-258. (b)

Howard, J. H. The attentional demands of nega-tion in a memory-scanning task. Memory & Cog-nition, 1975, 3, 319-324.

Johnsen, A. M., & Briggs, G. E. On the locus ofdisplay load effects in choice reactions. Journalof Experimental Psychology, 1973, 99, 266-271.

Johnston, W. A., Greenberg, S. N., Fisher, R. P.,& Martin, D. W. Divided attention: A vehiclefor monitoring memory processes. Journal ofExperimental Psychology, 1970, 83, 164-171.

Kahneman, D. Remarks on attentional control.Acta Psychologica, 1970, 33, 118-131.

Kahneman, D. Attention and effort. EnglewoodCliffs, NJ.: Prentice-Hall, 1973.

Keele, S. W. Compatibility and time-sharing inserial reaction time. Journal of ExperimentalPsychology, 1967, 75, 529-S39.

Keele, S. W. Attention demands of memory re-trieval. Journal of Experimental Psychology,1972, 93, 245-248.

Keele, S. W. Attention and human performance.Pacific Palisades, Calif.: Goodyear, 1973.

Keren, G. Some considerations of two alleged kindsof selective attention. Journal of ExperimentalPsychology: General, 1976, 105, 349-374.

62 GORDON D. LOGAN

Kolcrs, P. A. Memorial consequences of automa-tized encoding. Journal of Experimental Psy-chology: Human Learning and Memory 1975,/, 689-701.

Kristofferson, M. W. Effects of practice on char-acter classification performance. Canadian Jour-nal of Psychology, 1972, 26, 54-60. (a)

Kristoffcrson, M. W. When item recognition andvisual search functions are similar. Perception &Psychaphysics, 1972, 12, 379-384. (h)

LaBergc, D. Identification of two components ofthe time to switch attention: A test of a serialand a parallel model of attention. In S. Korn-blum (Ed.), Attention and performance IV.New York: Academic Press, 1973. (a)

LaBerge, D. Attention and the measurement ofperceptual learning. Memory & Cognition, 1973,/ . 268-276. (b)

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

Logan, G. D. On the nature of capacity limitationsin visual search. Doctoral dissertation, McGillUniversity, 1975.

Logan, G. D. Converging evidence for automaticperceptual processing in visual search. CanadianJournal oj Psychology, 1976, 30, 193-200.

Logan, G. D., Withcy, M. J., & Cowan, W. B. Cuesearch and comparison processes in visual searchfor letters. Canadian Journal of Psychology,1977, 31, 113-121.

Mewhort, D. J. K. Familiarity of letter sequences,response uncertainty and the tachistoscopic rec-ognition experiment. Canadian Journal of Psy-chology. 1967, 21, 309-321.

Millar, K. Processing capacity requirements ofstimulus encoding. A eta Psychologica, 1975, 39,393-410.

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

Neisscr, U. Decision-time without reaction-time:Experiments in visual scanning. American Jour-nal of Psychology,' 1963, 76, 376-385.

Neisser, U. Cognitive psychology. New York:Appleton-Century-Crofts, 1967.

Neisser, U. Practiced card sorting for multipletargets. Memory &• Cognition, 1974, 2, 781-785.

Neisser, U., Novick, R., & Lazar, R. Searchingfor ten targets simultaneously. Perceptual andMotor Skills, 1963, 17, 955-961.

Nickerson, R. S. Response times with a memory-dependent decision task. Journal oj ExperimentalPsychology, 1966, 72, 761-769.

Nickerson, R. S. Binary-classification reactiontime: A review of some studies of human in-formation-processing capabilities. PsychonomicMonograph Supplements, 1972, 4, 275-318.

Norman, D. A., & Bobrow, D. G. On data-limitedand resource-limited processes. Cognitive Psy-chology, 1975, 7, 44-64.

Novik, N., & Katx, L. High-speed visual scanningof words and nonwords. Journal of ExperimentalPsychology, 1971, 91, 350-353.

Pachella, R. G. The interpretation of reaction timein information-processing research. In B. H.Kantowitz (Ed.), Human information process-ing: Tutorials in performance and cognition.Hillsdale, NJ.: Erlbaum, 1974.

Peterson, L. R. Concurrent verbal activity. Psycho-logical Review. 1969, 76, 376-386.

Peterson, L. R., & Peterson, M. J. Short-termretention of individual verbal items. Journal ojExperimental Psychology. 1959, 58. 193-198.

Posner, M. L, & Boies, S. J. Components of at-tention. Psychological Review, 1971, 78, 391-408.

Posner, M. L, & Klein, R. On the functions ofconsciousness. In S. Kornblum (Ed.), Attentionand performance IV. New York: AcademicPress, 1973.

Posner, M. I., & Rossman, E. Effect of size andlocation of informational transforms upon short-term retention. Journal of Experimental Psy-chology, 1965, 70, 496-505.

Posner, M. L, & Snyclcr, C. R. R. Attention andcognitive control. In R. L. Solso (Ed.), Infor-mation processing and cognition: The Loyolasymposium. Hillsdale, NJ. : Erlbaum, 1975. (a)

Posner, M. I., & Snyder, C. R. R. Facilitation andinhibition in the processing of signals. In P. M. A.Rabbitt & S. Dornic (Eds.), Attention andperformance V. New York: Academic Press,1975. (b)

Rabbitt, P. M. A. Signal discriminability, S-Rcompatibility and choice reaction time. Psycho-nomic Science, 1967, 7, 419-420.

Rosch, E. Cognitive representations of semanticcategories. Journal oj Experimental Psychology:General, 1975, 104, 192-233.

Rosch, E., Simpson, C., & Miller, R. S. Structuralbases of typicality effects. Journal oj Experi-mental Psychology: Human Perception and Per-formance,'1976, 2, 491-502.

Rumelhart, D. E. A multicomponcnt theory of theperception of briefly exposed visual displays.Journal oj Mathematical Psychology, 1970, 7,191-218.

Salthouse, T. A. Simultaneous processing of verbaland spatial information. Memory & Cognition,1975, 3, 221-225.

Scarborough, D. L. Memory for brief visual dis-plays of symbols. Cognitive Psychology, 1972, 3,408-429.

Schneider, W., & Shiffrin, R. M. Controlled andautomatic human information processing: I. De-tection, search, and attention. Psychological Re-view, 1977, 84, 1-66.

Shiffrin, R. M., & Geisler, W. S. Visual recogni-tion in a theory of information processing. InR. L. Solso (Ed.), Contemporary issues in cog-nitive psychology: The Loyola symposium.Washington, D.C.: Winston, 1973.

Shiffrin, R. M., & Schneider, W. Controlled andautomatic human information processing: II.Perceptual learning, automatic attending, and ageneral theory. Psychological Review, 1977, 84,127-190.

AUTOMATIC PROCESSING IN SEARCH TASKS 63

Shulman, H. G., & Greenberg, S. N. Perceptualdeficit due to division of attention between mem-ory and perception. Journal of ExperimentalPsychology, 1971, 88, 171-176.

Shulman, H. G., Greenberg, S. N., & Martin, J.Intertask delay as a parameter of perceptual defi-cit in divided attention. Journal of ExperimentalPsychology, 1971, 88, 439-440.

Smith, E. E. Choice reaction time : An analysis ofthe major theoretical positions. PsychologicalBulletin, 1968, 69, 77-110.

Sternberg, S. Two operations in character recog-nition : Some evidence from reaction-time mea-surements. Perception & Psychophysics, 1967, 2,45-53.

Sternhcrg, S. The discovery of processing stages:Extensions of Donders' method. In W. G. Koster(Ed.), Attention and performance II. Amster-dam : North Holland, 1969. (a)

Sternberg, S. Memory scanning: Mental processesrevealed by reaction-time experiments. AmericanScientist, 1969, 57, 421-457. (b)

Sternberg, S. Memory scanning: New findings andcurrent controversies. Quarterly Journal of Ex-perimental Psychology, 1975, 27, 1-32.

Stroop, J. R. Studies of interference in serial ver-bal reactions. Journal of Experimental Psychol-ogy, 1935, 18, 643-662.

Taylor, D. A. Stage analysis of reaction time.Psychological Bulletin, 1976, 83, 161-191.

Thomas, E. A. C. The selectivity of preparation.Psychological Review, 1974, 81, 442-464.

Thorndike, E. L., & Lorge, I. The teacher's wordbook of 30,000 ivords. New York: Columbia

University, Teachers College, Bureau of Publi-cations, 1944.

Townsend, J. T., & Roos, R. N. Search reactiontime for single targets in multiletter stimuli withbrief visual displays. Memory &• Cognition, 1973,1, 319-332.

Treisman, A. M. Strategies and models of selectiveattention. Psychological Review, 1969, 76, 282-299.

Turvey, M. T. The effects of rehearsing analyzedinformation upon the retrieval of unanalyzed in-formation. Psychonomic Science, 1966, 6, 365-366.

Watkins, M. J., Watkins, O. C., Craik, F. I. M.,& Mazuryk, C. Effect of nonverbal distractionon short-term storage. Journal of ExperimentalPsychology, 1973, 101, 296-300.

Wattenbarger, B. L., & Pachella, R. G. The effectof memory load on reaction time in characterclassification. Perception & Psychophysics, 1972,12, 100-102.

Welford, A. T. Fundamentals of skill. London:Methuen, 1968.

Willows, D. M. Reading between the lines: Se-lective attention in good and poor readers. ChildDevelopment, 1974, 45, 408-415.

Willows, D. M., & MacKinnon, G. E. Selectivereading: Attention to the "unattended" lines.Canadian Journal of Psychology, 1973, 27, 292-304.

Woodworth, R. S. Experimental psychology. NewYork: Holt, 1938.

Received September 23, 1976 •