Attention and Automaticity in the Processingof Self-Relevant Information

John A. BarghNew York University

A neglected aspect of the study of social cognition has been the way in whichpeople select information for further processing from the vast amount availablein social environments. A major contemporary model of attention holds thatthere are two separate types of processes that operate concurrently: a flexiblebut resource-limited control process that regulates the contents of consciousawareness, and a relatively inflexible automatic process that can attract attentionto stimuli without conscious intent. Passive automatic processes, can either fa-cilitate or inhibit active attentional processing, necessitating either less or moreattentional effort, depending on the characteristics of the information that iscurrently present. On a dichotic listening task in which subjects attended to orignored self-relevant stimuli, it was found that self-relevant information requiredless attentional resources when presented to the attended channel, but more whenpresented to the rejected channel, relative to neutral words. This differentialcapacity allocation occurred despite subjects' lack of awareness of the contentsof the rejected channel. The results supported the existence and interaction ofthe two processes of attention in social information processing.

of the principles of cognitive psy-chology have been successfully applied to thestudy of social perception. Until recently,however, most of this research has been con-cerned with the organization of already ac-quired social information in memory and itsconsequences for retrieval and judgmentalprocesses (Fiske & Linville, 1980; Hastie,1981). As a result, relatively little is knownabout the role of mental organizations ofsocial information in the selection of infor-mation for possible encoding and storage

This article is based on part of a doctoral dissertationin social psychology at the University of Michigan. Iwould like to express my gratitude to Hazel Markus,Stephen Kaplan, William Kunst-Wilson, and especiallyto Robert Zajonc for their guidance and support as mycommittee. I thank Robert Zajonc, Tory Higgins, andan anonymous reviewer for their detailed suggestionson earlier versions of the article. I am also indebted toWilliam Johnston for his advice and the use of this lab-oratory during the early stages of this research and toHoward Schultz for his invaluable help in assemblingthe reaction time apparatus.

Requests for reprints should be sent to John A. Bargh,Department of Psychology, New York University, 6Washington Place, Seventh Floor, New York, NewYork 10003.

from the vast array available at any onetime. Such knowledge is essential for thegeneration of falsifiable predictions of theinfluence of social constructs or categoriesin complex social environments (AUport,1979; Fiske & Linville, 1980; Taylor &Crocker, 1981; Wyer, 1980). Toward thisend, the present research is an attempt toapply a widely accepted contemporary modelof attention allocation to the domain of so-cial psychological phenomena.

Given the overall limit on one's informa-tion processing resources at any given time,one is able to attend to only a small fractionof the overwhelming amount of potential in-formation present in the environment (Bor-ing, 1933, p. 194; Kahneman, 1973; Miller,1956; Norman & Bobrow, 1976). Therefore,people must be selective, even miserly, intheir allocation of processing resources(Bateson, 1972; Erdelyi, 1974; Mischel,1979). Attention serves as a "gating pro-cess," delimiting the range of potential in-formation. This screening of the stimulusfield has been found to be an important de-terminant of the course of information pro-cessing. One general approach to causal rea-

Journal of Personality and Social Psychology 1982, Vol. 43, No. 3, 425-436Copyright 1982 by the American Psychological Association, Inc. 0022-3514/82/4303-0425S00.75

425

426 JOHN A. BARGH

soning has argued that the object on whichattention is focused is likely to be perceivedas the causal agent (Jones & Nisbett, 1972;McArthur & Post, 1977; Smith & Miller,1979; Storms, 1973; Taylor & Fiske, 1975,1978). Fiske (1980) found that the moretime subjects spent looking at a slide de-picting a stimulus person's behavior, themore the behavior influenced their finalimpressions of the person.

Active and PassivePerceptual Expectancies

What types of information are more likelythan others to attract attention? Psycholo-gists have long been aware that those stimulithat we are ready to perceive have an ad-vantage over others in gaining our attention(e.g., Bruner, 1957; Higgins & King, 1981;Kahneman, 1973; Wyer & Srull, 1981).This "perceptual readiness" can arise in twoways. People can have an active set, or con-scious expectancy, for the occurrence ofsome environmental event (e.g., Broadbent,1977; Neisser, 1967). This can be due to thecontext of the current situation, which bringsinto consciousness objects and events thatare typically part of that particular scenariothrough a process of spreading activation(Collins & Loftus, 1975). It can also bemediated by verbal cues, as when a friendtells you to "watch out for Joe—he's beenacting strangely today." Active sets, becausethey reside in consciousness, are temporaryperceptual states.

Second, people possess a variety of long-term, or chronic expectancies that haveevolved out of frequent and consistent ex-perience within specific environmental do-mains (Broadbent, 1977; Logan, 1979; Shif-frin & Dumais, 1981; Shiffrin & Schneider,1977). Considerable experience in a domainresults in associations between its constit-uent elements forming and strengthening'inlong-term memory (e.g., Hayes-Roth, 1977;Hebb, 1949). Because chronic expectanciesare relatively permanent characteristics ofthe perceptual system itself (Broadbent,1977), one's active direction is not necessaryfor them to influence the interpretation ofstimuli. In contrast to active sets, then, thesechronic, or passive, processes can occur with-

out the person's intent or awareness (Logan,1980; Posner & Snyder, 1975).

In an ambitious and rigorous set of ex-periments, Shiffrin and Schneider (1977;Schneider & Shiffrin, 1977) demonstratedthat automatic attention responses could belearned for very frequently encounteredstimuli, so that attention is automaticallydrawn to such stimuli upon their sensationwithout conscious intent or effort. Subjectssearched for the presence of targets (eithera consonant or a digit) in series of rapidlypresented consonant-digit sets. After thou-sands of trials, subjects for whom the targetswere always digits and the distractors alwaysconsonants became able to detect the targetsregardless of the size of the target array thathad to be held in working memory duringthe search. Memory load did negatively af-fect performance of subjects searching fordifferent targets on each trial for whom tar-gets and distractors were both consonants.The investigators concluded that, becauseautomatic processing requires no attention,processing resources can be directed to im-portant stimulation whatever the ongoingconscious activity (Shiffrin & Schneider,1977, p. 161).

Interaction of Attention and Automaticity

It is widely recognized that the processingof any stimulus event almost always involvesa mixture of such automatic and attentionalprocesses, and that the phenomenon of realinterest is how the two types interact (Logan,1980; Neely, 1977; Posner & Snyder, 1975;Shiffrin & Dumais, 1981; Shiffrin & Schnei-der, 1977). Two studies have recently inves-tigated the effects on processing of an activeset that is either compatible or incompatiblewith passive automatic responses.

Neely (1977) tested Posner and Snyder's(1975) hypothesis that the conscious-atten-tion mechanism inhibits automatically ac-tivated memory locations. Using the seman-tic priming paradigm, he presented subjectswith priming words that were of the samecategory as the target word that followed(e.g., BiRD-sparrow), or that through in-structions to the subject primed a differentcategory (e.g., subjects were told that thename of a body part meant to expect the

ATTENTION AND AUTOMATICITY 427

name of a part of a building). Subjects thushad an active set for a building part whenthey saw the BODY prime whereas, accordingto the Posner-Snyder theory, BODY shouldat the same time passively prime the cate-gory of body part names. Consistent withthis prediction, Neely (1977) found thatBODY facilitated the time needed to decideif body part names were words at short de-lays between prime and target (approxi-mately 250 msec), but inhibited the responsetime to body part names at longer delayswhen the active set had had time to focuson the actively expected BUILDING category.In other words, the influence of the passiveactivation of BODY category members wasoverridden by the active set for the BUILDINGcategory, so that the usually facilitative ef-fect of prior associations between the primeand target on response speed (Meyer &Schvaneveldt, 1976) actually became inhi-bitory.

Logan and Zbrodoff (1979) showed thatthe usual Stroop effect could be reversed inthe same way. The Stroop test requires sub-jects to report one dimension of a stimulusword (e.g., the color in which it appears).When unreported dimensions (e.g., theword's meaning) are incompatible with thereported dimension, time to respond is in-creased (see Kahneman, 1973; Logan, 1980).The subjects' task in the Logan and Zbro-doff (1979) study was to report the word thatappeared either above or below a fixationpoint. Subjects became much faster in re-sponding to the word ABOVE when it wasbelow the fixation point, and vice versa, whenthese conflicting trials were the more fre-quent in the experiment. Thus, prior chronicexpectancies were again muted by conscioustemporary expectancies.

Although the interaction of simultaneousactive and passive processes in social per-ception is likely to be the rule in real socialjudgments, it has yet to be directly investi-gated (Higgins & King, 1981). Passive pro-cesses have been implicated recently as im-portant determinants of the interpretationof social information, however. Both thosepersonality-trait constructs a person has re-cently used (Higgins, Rholes, & Jones, 1977;Srull & Wyer, 1979, 1980) and those thathe or she has used frequently in the past

(Higgins, King, & Mavin, 1982) are morelikely than others to be used in the inter-pretation and storage of person information.Given that both active and passive processesoperate in social perception, the present con-cern is their working relationship with eachother.

Attentional Versus Automatic Processingof Self-Relevant Stimuli

It is assumed here that people developautomatic attention responses to self-rele-vant information. A subject's own name, forexample, is one of the rare stimuli able tobreak through the attentional barrier of thedichotic listening task and be consciouslynoticed (Moray, 1959). Moreover, self-rel-evant stimuli easily attract one's attention,even when such diversion of attention is det-rimental to ongoing conscious activity(Brenner, 1973; Geller & Shaver, 1976; Hull& Levy, 1979; Jarvella & Collas, 1974). Forexample, in a study using the Stroop para-digm, Geller and Shaver (1976) found thatthe time taken to name the color of a stim-ulus word was greater for self-relevant thanfor neutral words. This greater inhibitoryeffect of self-relevant words relative to neu-tral words suggests that self-relevant stimulireceive a greater amount of automatic pro-cessing. On the Stroop test, this would re-quire additional attentional resources to bedirected to the relevant stimulus dimensionof ink color, resulting in longer responsetimes (see Logan, 1980).

The present experiment is analogous indesign to that of Geller and Shaver (1976).The relatively greater degree of automaticprocessing given to self-relevant informationwas again tested, but the interaction of at-tentional and automatic processing of self-relevant information was also examined. Thefocused attention paradigm was employedusing the dichotic listening task, in whichauditory stimuli are presented to both earssimultaneously through stereo headphones,with subjects instructed to attend to one earwhile ignoring stimuli presented to the otherear. When an obvious, characteristic distin-guishes the information to be attended andthat to be ignored in this task, as spatialorigin does here, subjects are quite able to

428 JOHN A. BARGH

keep the contents of the rejected channel outof consciousness (Cherry, 1953; Kahneman,1973, p. 114; Shiffrin & Schneider, 1977,p. 180), According to the dominant theo-retical positions discussed previously, how-ever, the contents of the unattended channelwill still be processed to some extent, butwill be inhibited from attaining conscious-ness (Kahneman, 1973; Posner & Snyder,1975; Shiffrin & Schneider, 1977; see alsoShallice, 1972, 1978). Moreover, the stron-ger the chronic expectancy for an item pre-sented to the unattended channel, the moreattention must be allocated to the targetchannel to prevent that item from disruptingfocused attention (Egeth, 1967; Kahneman,1973, p. 119; Logan, 1980). The greatestthreat to disruption of focused attention, re-quiring the greatest amount of attentionalresources to overcome, would be if stimulifor which automatic attention responses hadbeen learned were presented to the rejectedchannel (Shiffrin & Schneider, 1977, p.186). But unless the automatic attention re-sponse is sufficiently strong, even these stim-uli can be inhibited by the active set (Shiffrin& Schneider, 1977, p. 156).

Overview

In the dichotic listening task, subjectswere exposed to simultaneously-presentedword pairs: a noun in one channel and anadjective in the other. They were instructedto repeat out loud, or "shadow," each of thewords in a specified channel as soon as theyheard it while ignoring completely the con-tents of the other channel. To ensure thatthe ignored channel did not receive any con-scious attention, several measures of sub-jects' awareness of that channel were taken.One section of the adjective channel con-tained words related to the trait of indepen-dence, which was central to the self-descrip-tion of half of the subjects (schematics) andnot self-descriptive for the other half(aschematics). For half of each of thesegroups of subjects, the noun channel was tobe shadowed1 and the adjective channel ig-nored; the remaining subjects shadowed theadjective channel and ignored the nouns. Inthis way, the independence-relevant and in-dependence-irrelevant adjectives were pre-

sented to the unattended channel for somesubjects and to the attended channel for theothers.

The amount of attentional capacity takenby the shadowing task was assessed by theprobe reaction time (RT) technique (Brown,1964; Posner & Boies, 1971), in which thesubject is instructed to optimize perfor-mance of the primary task and to use onlythe remaining capacity to react as quicklyas possible to the periodically presentedprobe stimulus. Given the assumption of anoverall limit on one's attentional capacity atany given time, the RT to the probe stimulusis inversely related to the amount of atten-tion being given to the primary shadowingtask (Kahneman, 1973, p. 185; Logan, 1979).The less attention needed by the shadowingtask, the more attention available to respondto the probe stimulus. Thus, the greater thelatency of response to the probe, the greaterthe amount of attentional capacity being re-quired at the moment by the shadowing task.

If self-relevant information is processedautomatically to a greater extent than otherinformation, perhaps even to the point ofeliciting automatic attention responses, itshould facilitate the shadowing task whenpresented to the attended ear, and inhibitperformance when presented to the unat-tended ear, relative to information that is notself-relevant. In other words, the automaticprocessing of the self-relevant informationwhen it is being shadowed should necessitateless attentional resources for the shadowingtask than for stimuli receiving less or no au-tomatic processing, but should require moreattentional resources to be allocated to theshadowing task to keep it from consciousnesswhen it is in the rejected channel. Therefore,the major hypothesis is that for subjects forwhom independence-related information isself-relevant, probe RTs will be relativelylower when the relevant adjectives are pre-sented to the attended ear but relativelyhigher when the relevant adjectives are pre-sented to the unattended ear.

MethodDesign

As manipulation checks on subjects' degree of aware-ness of material in the unattended channel, the numberof errors made while shadowing the attended channel

ATTENTION AND AUTOMATICITY 429

were tabulated, and subjects' momentary awareness andrecognition memory of the trait adjectives were assessed.In addition, if the probe RT is to be a valid measureof spare attentional capacity, the shadowing task mayrestrict the amount of attention available for reactingto the probe, but the probe task cannot take away fromthe processing capacity available for the shadowing task(Kantowitz, 1974; Logan, 1979). Accordingly, a groupof subjects engaged in the shadowing task without per-forming the probe task. If the probe task is a valid mea-sure of capacity remaining from the primary task al-location, there should be no differences between theprobe and no-probe conditions on the dependent mea-sures. Finally, the ear to which the attended channelwas presented was included as a factor in the design,to test for possible differences between brain hemi-spheres in the processing of the stimuli.

The factors of schematicity (schematic vs. asche-matic), channel (adjective channel attended vs. nounchannel attended), hemisphere (adjective channel pre-'sented to left vs. right ear), subject gender, and probe(probe RTs taken vs. not taken) were completelycrossed, with repeated measures on the blocks factor(the two schema-consistent adjective channel blocks andthe two schema-irrelevant blocks). The repeated depen-dent measures were the number of shadowing errors andprobe RTs. Unattended-channel recognition memoryperformance measures were taken only once, at the con-clusion of the shadowing task.

A separate control group of 20 subjects (10 sche-matics, 10 aschematics) engaged in the momentaryawareness test, in which the only dependent measureswere awareness of the unattended and attended channelcontents (yes/no).

SubjectsParticipating in this experiment were 145 undergrad-

uates (68 male) enrolled in the introductory psychologycourse at the University of Michigan. Their participa-tion in the experiment partially fulfilled a course re-quirement. All but four subjects stated that they hadnormal hearing in both ears and that English was theirnative language; data for the remaining four subjectswere excluded from the analyses. Subjects were selectedfor the experiment based on their responses to a ques-tionnaire administered to most of the students in thecourse at the beginning of the term. The questionnaireconsisted of 25 personality characteristics, each of whichstudents rated on a separate 11-point scale for its degreeof self-descriptiveness and importance to their self-con-cept.

Two groups of subjects were selected on the basis oftheir responses to the independence-related and depen-dence-related personality characteristic items, followingthe procedure of Markus (1977).' Independent sche-matic subjects responded in the high end (from 1 to 4)of both the self-descriptiveness and the importancescales on at least two of the three independence-relevanttraits (independent, individualist, and leader), and inthe middle-to-low range (5-11) of these scales on atleast two of the three dependence-relevant traits (de-pendent, conformist, and follower). Aschematic sub-jects responded in the middle (5-7) of the self-descrip-

tiveness scales and the middle-to-low end of theimportance scales for at least two of the threeindependence-related traits and for at least two of thethree dependence-related traits. Using these criteria,Markus (1977) found independent schematics to be re-liably faster, more accurate, and more consistent in theprocessing of independence-related information thanwere aschematics, thereby validating the selection pro-cedure.

During debriefing, subjects were asked which of theirhands was dominant. Data for the 15 left-handed sub-jects were excluded from all analyses involving the hemi-sphere factor because of the less complete cerebral lat-eralization in left-handed people (Bradshaw & Taylor,1979; Hardyck & Petrinovich, 1977). These data wereincluded in all analyses reported here, however, due tothe nonsignificance of the hemisphere factor.

Apparatus and MaterialsExperimental room. The subjects sat in a chair at

a small table, facing a partition that hid the experi-menter and most apparatus from their view. On a tablein front of the experimenter were the tape recorder andRT apparatus (described later). On the table in frontof the subject were a pair of Koss KO-727B stereo head-phones and (except in the no-probe condition) a smallblack panel equipped with a light bulb and a buttonswitch (part of the RT apparatus). The 4 X 5.5 X 2.7m room was carpeted, and its ceiling was tiled withsound-absorbing material.

Stimulus tape. Separate lists of 80 words were rec-orded on each of two channels of a Sony TC-270 stereotape recorder, such that one word from each list wouldbe presented simultaneously. The rate of presentationwas 750 msec per trial. Both lists were spoken by thesame male voice.

The list recorded on one channel consisted of commonnouns, such as table, poker, and quarry. The list re-corded on the other channel consisted of adjectives usedby Anderson (1968) in impression formation research.The first 30 and final 10 adjectives on this channel wereunrelated to the trait of independence (e.g., troubled,silly, prideful). The 15 adjectives judged by subjects inthe Markus (1977) study to be highly related to thetrait of independence plus four additional synonymouswords2 were repeated in various random orders to makeup the middle block of 40 adjectives. (I considered itdesirable to lengthen the independence-related adjectivelist from 15 to 40 words in this way in order to allow

1 Although Markus (1977) also used the Gough-Heil-brun Adjective Check List in the selection procedure itsuse has been discontinued, as the trait questionnairealone proved to be a sufficient criterion in subsequentstudies (e.g., Markus & Smith, 1981).

2 The words used in the Markus (1977) study wereindependent, aloof, assertive, uninhibited, arrogant,egotistical, unconventional, self-confident, individual-istic, ambitious, adventurous, dominating, argumen-tative, outspoken, and aggressive. The four additionalwords used were self-assured, self-sufficient, noncon-formist, and leader.

430 JOHN A. BARGH

two probe RTs to be taken during the self-relevantblock.) The independence-related and independence-un-related word sets had similar frequencies in the Englishlanguage, as estimated by Carroll, Davies, and Richman(1971).3 The first 10 word pairs served as practice forthe subjects, because (a) this was a novel task for them,and (b) it has been shown that subjects need a fewseconds to focus their attention on one ear in a dichoticlistening task (Kahneman, 1973, p. 118; Treisman,Squire, & Green, 1974). Therefore, all dependent mea-sures taken during this initial 10-word block were dis-regarded in the analyses. The duration of the tape was60 sec.

Reaction time apparatus. A button was located infront of the experimenter that simultaneously illumi-nated a light bulb located in front of the subject andstarted a Lafayette Instruments model 54519 digitaltiming device when pressed. When the subject saw thelight go on, he or she pressed a button next to the lightbulb that turned off the bulb and stopped the timer. ALafayette Instruments model 56024 automatic outputdevice printed each RT to the nearest millisecond.

Recognition test. Two versions of the recognitionmemory test on the information presented in the un-attended channel were constructed, one each for theadjective list and the noun list. The format of both ver-sions was the same: a list of words, half of which hadbeen presented on the unattended channel, and half ofwhich had not been presented. Subjects were instructedto check off those items they thought had been pre-sented. The adjective test contained four sets of 15 itemseach: independence targets (independence-related itemsactually presented), independence foils (independence-related but not presented), control targets (indepen-dence-unrelated items actually presented), and controlfoils (independence-unrelated and not presented).

The noun test consisted of three sets of 10 items:independent targets (nouns actually presented while theindependence-related adjectives were presented simul-taneously on the attended channel), control targets(nouns actually presented while the control adjectiveswere on the attended channel), and foils (nouns notpresented on the unattended channel). For both versions,item order was randomized.

Procedure

Within each gender-schematicity combination, sub-jects were randomly assigned to one level of the channel,hemisphere, and probe factors. After arriving at a wait-ing room, subjects were shown into the experimentalroom and seated. They were informed as to the natureof the shadowing task, and instructed to concentratetheir attention on a specific ear, repeating out loud thewords they heard in that ear while ignoring the wordscoming into their other ear. The secondary probe RTtask was then explained to all subjects except those inthe no-probe condition. Instructions were given to pressthe button as soon as the light came on, but to considerthe shadowing task as the more important of the two.The headphones were fitted comfortably on the subject'shead.

The stimulus tape was then started. Each channel waspresented at 70 db. (SPL). The experimenter kept track

of the number of shadowing errors by crossing off allmissed or misstated words on a list. A strict criterionof accuracy was employed: the correct and completeitem had to be repeated. In addition, the experimenteractivated the RT apparatus (except for subjects in theno-probe condition) at five predetermined times: afterthe presentation of the 8th, 21st, 39th, 59th, and 78thword pairs. The first time of measurement was duringthe practice words, the next corresponded to the firstcontrol adjective block, the next two to the indepen-dence-adjective section, and the final RT to the secondcontrol adjective section. (It should be noted that thetwo independence-adjective section probes did not followthe presentation of the same independence-related word:the two words were independent and argumentative.)Each time the subject's response stopped the tinier, theexperimenter activated the tinier printer to recordthe RT.

Immediately after completing the shadowing task,subjects were given as much time as they needed tocomplete the recognition memory test on the unattendedchannel items. Subjects were informed that some of thetest items had been presented and the others had notbeen presented. Many subjects balked at completing thistest, on the grounds that they were certain they did notremember any of the words presented to the unattendedear. The experimenter asked them to complete it none-theless, guessing if they needed to. It was explained thatalthough they might think they could not remember anyof the words, they still might be able to guess fairlyaccurately.

Momentary awareness condition. Ten schematic and10 aschematic subjects participated in a control con-dition designed to test for momentary awareness of theinformation in both the attended and unattended chan-nels. Half of these subjects (equal numbers of sche-matics and aschematics) attended to the noun list andhalf to the adjective list. These subjects were treated inthe same way as subjects in the no-probe condition4 untilthe 45th word pair had been presented (this was in themiddle of the independence-related section of the ad-jective channel), when the tape was stopped. Subjectswere immediately asked to recall the last word in theunattended ear. If they could not recall it, they wereasked to name any of the words presented to that ear.If they were still unable to report any of the unattendedcontent, the experimenter informed them that all of thewords had been of a certain type—either nouns, verbs,or adjectives. Subjects were asked to guess to which ofthese three categories the unattended words belonged.Finally, subjects were asked the same series of questionsabout the information in the attended channel.

After the subjects completed the experimental tasks,they were fully debriefed and thanked for their partic-ipation.

3 Mean estimated frequencies per million words: In-dependence related = 9.18, independence unrelated =12.26, *(78) < 1.

4 If the probe RT task caused any distraction fromthe shadowing task it would lessen these subjects'chances of being momentarily aware of the unattendedchannel.

ATTENTION AND AUTOMATICITY 431

Results

In none of the analyses performed werethere any statistically reliable main effectsor interactions involving gender or hemi-sphere, so they are not mentioned further.

Probe RT Validation

In order to ensure that the secondaryprobe RT task was a true measure of spareprocessing capacity, it must be shown thatit did not itself take processing capacityaway from the primary shadowing task.Such interference would be reflected in dif-ferences in dependent measures between theprobe and no-probe conditions.

In all of the analyses reported .(except, ofcourse, the probe RT analysis), the probefactor was included. It had no reliable effecton the number of shadowing errors. In ad-dition, neither probe nor no-probe conditionsubjects were able to recognize indepen-dence-related or control target items on thememory test at better than the chance levelof 50%. There was a significant effect of theprobe factor on control item hit rate, F(l,113) = 8.87, p < .01, with the probe condi-tion subjects endorsing on the average morecontrol targets than did the no-probe con-dition (48% vs. 38%). The fact that neithergroup demonstrated any better than chancerecognition of the control targets, however,indicates that subjects in the two conditionsshared a lack of awareness of the controlwords in the unattended channel.

The lack of differences between the probeand no-probe conditions validates the inter-pretation of the probe RT as a measure ofspare processing capacity, an inverse indi-cation of the amount of attention allocatedto the auditory stimuli. The probe factor isnot included in the further reporting of theanalyses, which now focus on the schema-ticity and channel factors.

Awareness Measures

Recognition test. Better than chance rec-ognition of the information presented to theunattended ear would indicate that there hadbeen conscious processing of that informa-tion, because conscious awareness of a stim-ulus is generally considered necessary for itto be stored in long-term memory (Kahne-

Table 1Mean Proportion of Shadowing Errors byAttended Channel and Schema Relevanceof Adjective List Block

Adjective list block

First SecondAttended channel control Independence control

Nouns (n = 56)Adjectives (n = 65)

.39

.38.37.33

.32

.38

man, 1973; Posner, 1978; Shiffrin & Schnei-der, 1977). However, there was no statisti-•cally reliable main effect or interaction ofthe schematicity and channel factors on hitor false alarm rates for either independence-related or independence-Unrelated recogni-tion test items. For both types of items, andfor both schematics and aschematics, falsealarm rates were either equal to or greaterthan hit rates, indicating that subjects hadno memory for the stimuli presented to theunattended channel.

Shadowing errors. The lack of memoryfor the unattended stimuli does not prove theabsence of conscious awareness of them. Itis possible that one could be momentarilyaware of environmental stimuli and yet notremember them (Egeth, 1967; White, 1980).As a measure of momentary awareness, thenumber of shadowing errors was tabulatedfor each subject. A greater number of shad-owing errors made by one group of subjectsor a greater number made during the pre-sentation of one type of attended stimuliwould indicate a greater possibility that al-ternating of conscious attention betweenchannels had occurred (LaBerge & Samuels,1974; Posner, 1978, p. 96).

An analysis of variance of the number ofshadowing errors was performed, with re-peated measures on the blocks factor (errorsmade during the first control section, inde-pendence-relevant section, and second con-trol section). There was no reliable maineffect or interaction involving the schema-ticity factor. There was a significant maineffect of the blocks factor, F(2, 226) = 4.30,p < .025, and a significant interaction be-tween the blocks and channel factors, F(2,226) = 8.13, p<.001. As can be seen inTable 1, shadowing performance steadily

432 JOHN A. BARGH

improved over time for subjects repeatingthe noun list, clearly a result of practice withthe novel task. Subjects shadowing the ad-jective list, however, did much better on theindependence-related adjective section thanon the control sections. A difference in taskcharacteristics appears to be responsible forthis interaction. On the noun list, no singleitem was repeated, hence no active set couldbe developed to aid in shadowing. Duringthe independence-related section of 40 ad-jectives, however, the same 15 words wererepeatedly encountered, so that a temporaryexpectancy for these words would help per-formance. But when the final 10 control ad-jectives were encountered, this presumedtemporary expectancy worked to the sub-jects' disadvantage relative to noun-shad-owing subjects, who had no such expectancy.

A certain number of shadowing errors bysubjects are desirable to ensure that theshadowing task is sufficiently difficult to re-quire their full conscious attention (Norman,1969). The equivalent numbers of shadow-ing errors for the two groups are uninfor-mative as to the momentary awareness issue,however, because one cannot know from thenumber of shadowing errors alone whetherthey are due to a switching of focused at-tention to the unattended ear or solely totask difficulty. The results of the third andstrictest awareness measure provide a clearanswer.

Momentary awareness. None of the 20subjects in the momentary awareness con-dition were able to recall any of the wordspresented to the unattended ear. When the10 subjects for whom the adjectives werepresented in the unattended channel wereasked to guess the type of word that hadbeen presented, four guessed adjectives (twoschematics, two aschematics), a number ex-pected by chance alone. Only two of the sub-jects for whom the noun list was unattendedguessed correctly. Sixteen subjects were ableto report at least one of the final five attendedwords; the other four subjects reported wordspresented earlier in the list.

The unequivocal indication of the threeawareness measures is that subjects' con-scious attention was fully focused on the to-be-attended channel; they were not con-sciously aware of the information presented

in the unattended channel. This is consistentwith past research demonstrating subjects'ability to keep the contents of the rejectedchannel completely out of consciousnesswhen spatial origin distinguishes the twochannels (Cherry, 1953; Kahneman,' 1973,p. 114).

Amount of Processing

Inspection of the cell means and standarddeviations of the raw probe RTs showed thatthe higher variances were associated with thehigher means. Therefore, a natural loga-rithm transformation of the raw scores wasmade (Winer, 1971, p. 400).

An analysis of variance of the log RTs wasperformed, with schematicity and channelas the between-subjects factors and blocksthe within-subjects factor. The blocks factorproved significant, .F(3, 219) = 4.47,p < .01,with the mean log RT steadily .declining overthe four times it was taken. This is probablyanother manifestation of the subjects' in-creasing experience with the novel shadow-ing task; both shadowing errors and probeRTs decreased over the course of the task.

As the main hypothesis of the study is thereversal of relative probe RTs for schematicsand aschematics as a function of channel,the three-way interaction among the sche-maticity, channel, and blocks factors is ofcentral importance. This interaction was sta-tistically reliable, F(3, 219) = 3.02, p < .05.Figure 1 presents the mean RTs for subjectswho attended the adjective list (left panel)and the noun list (right panel).

A series of planned comparisons was con-ducted to determine the source of the sig-nificant interaction.5 Comparison of the in-dividual schematicity by channel cell meansat each of the four times of measurementindicated the only significant difference to

5 Because the interaction of the schematicity andchannel factors at single levels of the within-subjectsfactor was being tested, the denominator of the F sta-tistic for the contrasts was formed by pooling the be-tween-subjects and within-subjects mean square errors(Winer, 1971, pp. 559-563). Due to the unequal numberof subjects per cell, the numerator of the contrasts wasformed by squaring the difference between the con-trasted means and dividing this by the summed ratiosof the squared weights.and cell MS (Wirier, 1971,p. 215).

ATTENTION AND AUTOMATICITY 433

be on the first independence-related blockRT, F(1, 73) = 6.40, p < .025. The contraston the second independence-related blockRT was marginally reliable, F(l, 73) = 3.00,p < ,10. From, Figure 1 it can be seen thatschematics took less time than aschematicsto respond to the probe when attending tothe independence-related adjectives, but rel-atively more time when the independenceinformation was presented in the unattendedchannel, as predicted. Probe RTs of sche-matics and aschematics did not differ on ei-ther of the control block measures.

The two-way interaction between sche-maticity and blocks was assessed individu-ally for the two channel conditions. The twoRTs for both the control and independenceblocks were averaged to obtain a single andmore stable estimate of the amount of ca-pacity used during the relevant and irrele-vant adjective blocks. Both interactionsproved significant: F(l, 40) = 9.78, p < .01,for the adjective channel, with schematics'RTs shorter during the independence-adjec-tive block and longer during the control-ad-jective block relative to aschematics; F(l,33) = 4.30, p < .05, for the noun channel,with schematics and aschematics takingnearly identical lengths of time to react tothe control probes, but schematics takinglonger to react to the independence-adjectiveprobes.

Discussion

The results are consistent with the exper-imental literature on the interaction of at-tentional and automatic processing (Logan,1980): automatic processing of the self-rel-evant information facilitated the shadowingtask when part of the attended channel, andinhibited performance when on the rejectedchannel. An active set to attend to the nounchannel was able to overcome the passiveprocessing of the self-relevant informationpresented to the other channel, but at a costof attentional capacity. When these passiveand active processes were compatible—thatis, when the self-relevant words were to beshadowed—less attentional capacity was re-quired for the task. Thus, the present find-ings are a conceptual replication of Neely(1977) and provide support for the Posner-

WPO

_ «oo

& 800

„,ao£ 8091

soo0

contort Ind«p1 IndepZ controls coition Irxtopl Ind«p2 contnU

STIMULUS WORD BLOCK

Figure 1. Mean probe reaction times (RT) in millisec-onds. (The left panel shows the results of the adjectivechannel attended; the right panel shows the noun chan-nel attended. A = aschematics; S = schematics.)

Snyder (1975) model of attention within thedomain of social information.

Another interesting implication of theprobe RT data is that schematics apparentlymade automatic attention responses to theself-relevant adjectives. According to Shif-frin and Schneider (1977, p. 186), decre-ments in focused-attention ability occurwhen the to-be-ignored stimuli initiate au-tomatic attention responses. Moreover, Shif-ffin and Dumais (1981) defined as automatic"any process that always utilizes general re-sources and decreases general processing ca-pacity whenever a given set of external ini-tiating stimuli are presented, regardless ofa subject's attempt to ignore or bypass thedistraction" (p. 117). Thus, the decrease inspare attentional capacity for schematics re-sulting from the presentation of the inde-pendence adjectives to the rejected channel(Figure 1, right panel) demonstrates auto-matic processing of trait information bythose who have acquired chronically acces-sible constructs for it. The existence of au-tomatic attention responses to self-relevantinformation presented outside of the sub-jects' conscious awareness provides strongsupport for the hypothesis that mental rep-resentations of such stimuli play a direct rolein the selection of stimuli for further pro-cessing.

Why is self-relevant information capableof provoking automatic attention responses?The development of such responses requires

434 JOHN A. BARGH

frequent and relatively consistent experiencewith the environmental event or object (Lo-gan, 1979; Shiffrin & Dumais, 1981; Shif-frin & Schneider, 1977). Because peopleconstantly experience events with themselvesas the central focus (e.g., Greenwald, 1980),self-relevant information is likely to be thetype most frequently processed. This shouldlead to greater accessibility in perceptionand retrieval for mental representations ofvarious sources of self-relevant information(Bruner, 1957; Higgins & King, 1981), andindeed, information relevant to the self ismore easily and quickly accessed than othertypes of information (Markus, 1977; Ross& Sicoly, 1979). So it appears that the prin-ciple of greater cognitive skill arising fromgreater experience within an environmentaldomain (e.g., Newell & Rosenbloom, 1981)can account for the development of a chronicperceptual set favoring self-relevant stimuli.The present results thus favor the Higginsand King (1981) "energy cell" metaphor forthe development of social construct acces-sibility over the "storage bin" model of Wyerand Srull (1981), as the former stipulatesboth frequency and recency of activation asfactors, whereas the latter specifies only re-cency of activation.

A recent study by Nielsen and Sarason(1981) is pertinent to these issues. Using theshadowing task, the investigators presenteddifferent classes of "emotional" words to theignored channel for different groups of sub-jects. The emotional word groups either wereneutral, sexually explicit, related to hostility,related to university life, or related to edu-cational evaluation. Only the sexually ex-plicit information tended to intrude uponsubjects' focused attention, as was indicatedby a greater number of shadowing errors andbetter recognition memory for the explicitwords. That is, sexually explicit words wereable to automatically attract attention de-spite subjects' attempts to ignore them. Noother class of words was able to breach theattentional barrier.

These findings are entirely consistent withthose of the present study. Sexually explicitwords apparently call forth an automaticattention response that is too strong to beexcluded from awareness by a control pro-cess, unlike the attention response triggered

by self-relevant trait terms. The resultantdistraction of conscious attention in theNielsen and Sarason (1981) study caused adeficit in shadowing performance, whereasno such distraction occurred in the presentexperiment. The present use of the RT asa measure of spare attentional capacity al-lows an extension of the Nielsen and Sarasonresults by demonstrating that automatic at-tention responses can occur but be barredfrom awareness by an active set or controlprocess, and so remain undetected by theshadowing error measure. Thus, it is possiblethat the other classes of words studied byNielsen and Sarason were also processed,but were successfully inhibited prior toachieving consciousness.

Nielsen and Sarason (1981) interpretedtheir results as evidence that emotionallysalient information is able to intrude uponawareness. This would also provide a viablealternative explanation of the present re-sults. Information that implicates the self isemotionally charged to a great extent (e.g.,Zajonc, 1980), so it is possible that theemotional quality of the self-relevant infor-mation may underlie its attentional effects,not its frequency in perceptual experiencealone. Alternatively, the frequency expla-nation advanced here can also explain theNielsen and Sarason (1981) results: Sexualthemes are most likely very common inthought, such as in fantasizing and day-dreams (Singer & Antrobus, 1972), and soby virtue of their greater frequency come toautomatically attract attentional resources.The data do not yet permit a decision onwhether it is perceptual frequency or emo-tional quality or both that drives this differ-ential allocation of attentional resources.The frequency interpretation is preferredhere on the basis of parsimony; it does notnecessitate the postulation that "qualita-tively different kinds of information pro-cessing" occur for certain types of input,such as sexual information (Nielsen & Sar-ason, 1981, p. 957).

The present findings suggest that atten-tional and automatic processes operate inmuch the same way for social informationas they do for nonsocial information. Butother important issues have been raised. Per-haps most critical of all are the consequences

ATTENTION AND AUTOMATICITY 435

of automatic trait information processing.Self-relevant trait stimuli apparently can beprocessed outside of conscious awareness,but it is not known whether such automaticprocessing has any influence on the moment-to-moment interpretation of the social en-vironment. This issue deserves further study.If selective attention to certain categories ofperson information is able to proceed auto-matically and outside of one's consciousawareness there are considerable implica-tions for our understanding of social percep-tion.

ReferencesAllport, D. A. Conscious and unconscious cognition: A

computational metaphor for the mechanism of atten-tion and integration. In L. G. Nilsson (Ed.), Per-spectives on memory research. Hillsdale, N.J.: Erl-baum, 1979.

Anderson, N. H. Likableness ratings of 555 personalitytrait words. Journal of Personality and Social Psy-chology, 1968, 9, 272-279.

Bateson, G. Steps to an ecology of mind. San Francisco:Chandler, 1972.

Boring, E. G. The physical dimensions of conscious-ness. New York: Century, 1933.

Bradshaw, J. L., & Taylor, M. J. A word-naming deficitin nonfamilial sinistrals? Laterality effects of vocalresponses to tachistoscopically presented letter strings.Neuropsychologia, 1979, 17, 21-31.

Brenner, M. The next-in-line effect. Journal of VerbalLearning and Verbal Behavior, 1973, 12, 320-323.

Broadbent, D. E. The hidden pre-attentive process.American Psychologist, 1977, 32, 109-118.

Brown, I. D. The measurement of perceptual load andreserve capacity. Transactions of the Association ofIndustrial Medical Officers, 1964, 14, 44-49.

Bruner, J. S. On perceptual readiness. PsychologicalReview, 1957, 64, 123-152.

Carroll, J. B., Davies, P., & Richman, B. The AmericanHeritage word frequency book, New York: Hough-ton-Mifflin, 1971.

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

Collins, A. M., & Loftus, E. F. A spreading activationtheory of semantic processing. Psychological Review,1975, 82, 407-428.

Egeth, H. Selective attention. Psychological Bulletin,1967, 67, 41-57.

Erdelyi, M. H. A new look at the New Look: Perceptualdefense and vigilance. Psychological Review, 1974,81, 1-25.

Fiske, S. T. Attention and weight in person perception:The impact of negative and extreme behavior. Journalof Personality and Social Psychology, 1980,38,889-906.

Fiske, S. T., & Linville, P. W. What does the schemaconcept buy us? Personality and Social PsychologyBulletin, 1980, 6, 543-557.

Geller, V., & Shaver, P. Cognitive consequences of self-awareness. Journal of Experimental Social Psy-chology, 1976, 12, 99-108.

Greenwald, A. G. The totalitarian ego: Fabrication andrevision of personal history. American Psychologist,1980, 35, 603-618.

Hardyck, C., & Petrinovich, L. F. Left-handedness.Psychological Bulletin, 1977, 84, 385-404.

Hastie, R. Schematic principles in human memory. InE. T. Higgins, C. P. Herman, & M. P. Zanna (Eds.),Social cognition: The Ontario Symposium. Hillsdale,N.J.: Erlbaum, 1981. ,

Hayes-Roth, B. Evolution of cognitive structures andprocesses. Psychological Review, 1977, 84, 260-278.

Hebb, D. O. Organization of behavior. New York:Wiley, 1949.

Higgins, E. T., & King, G. Accessibility of social con-structs: Information-processing consequences of in-dividual and contextual variability. In N. Cantor &J. F. Kihlstrom (Eds.), Personality, cognition, andsocial interaction. Hillsdale, N.J.: Erlbaum, 1981.

Higgins, E. T., King, G. A., & Mavin, G. H. Individualconstruct accessibility and subjective impressions andrecall. Journal of Personality and Social Psychology,1982, 43, 35-47.

Higgins, E. T., Rholes, W. S., & Jones, C. R. Categoryaccessibility and impression formation. Journal ofExperimental Social Psychology, 1977,13,141-154.

Hull, J. G., & Levy, A. S. The organizational functionsof the self: An alternative to the Duval and Wicklundmodel of self-awareness. Journal of Personality andSocial Psychology, 1979, 37, 756-768.

Jarvella, R. J., & Collas, J. G. Memory for the inten-tions of sentences. Memory & Cognition, 1974, 2,185-188.

Jones, E. E., & Nisbett, R. E. The actor and the ob-server: Divergent perceptions of the causes of behav-ior. In E. E. Jones et al. (Eds.), Attribution: Per-ceiving the causes of behavior. Morristown, N.J.:General Learning Press, 1972.

Kahneman, D. Attention and effort. Englewood Cliffs,N.J.: Prentice-Hall, 1973.

Kantowitz, B. H. Double stimulation. In B. H. Kan-towitz (Ed.), Human information processing: Tuto-rials in performance and cognition. Hillsdale, N.J.:Erlbaum, 1974.

LaBerge, D., & Samuels, S. J. Toward a theory of au-tomatic information processing in reading. CognitivePsychology, 1974, 6, 293-323.

Logan, G. D. On the use of a concurrent memory loadto measure attention and automaticity. Journal ofExperimental Psychology: Human Perception andPerformance, 1979, J, 189-207.

Logan, G. D. Attention and automaticity in Stroop andpriming tasks: Theory and data. Cognitive Psychol-ogy, 1980, 12, 523-553.

Logan, G. D., & Zbrodoff, N. J. When it helps to bemisled: Facilitative effects of increasing the frequencyof conflicting stimuli in a Stroop-like task. Memory& Cognition, 1979, 7, 166-174.

Markus, H. Self-schemata and processing informationabout the self. Journal of Personality and SocialPsychology, 1977, 35, 63-78.

Markus, H., & Smith, J. M. The influence of self-sche-

436 JOHN A. BARGH

mas on the perception of others. In N. Cantor &J. F. Kihlstrom (Eds,), Personality, cognition, andsocial interaction. Hillsdale, N.J.: Erlbaum, 1981.

McArthur, L. Z., & Post, D. D. Figural emphasis andperson perception. Journal of Experimental SocialPsychology, 1977, 13, 520-535.

Meyer, D. E., & Schvaneveldt, R. W. Meaning, memorystructure, and mental processes. In C. N. Cofer (Ed.),The structure of human memory. San Francisco:Freeman, 1976.

Miller, G. A. The magical number seven, plus or minustwo: Some limits on our capacity for processing in-formation. Psychological Review, 1956, 63, 81-97.

Mischel, W. On the interface of cognition and person-ality: Beyond the person-situation debate. AmericanPsychologist, 1919, 34, 740-754.

Moray, N. Attention in dichotic listening: Affective cuesand the influence of instructions. Quarterly Journalof Experimental Psychology, 1959, / / , 56-60.

Neely, J. H. Semantic priming and retrieval from lexicalmemory: Roles of inhibitionless spreading activationand limited-capacity attention. Journal of Experi-mental Psychology: General, 1977, 106, 226-254.

Neisser, U. Cognitive psychology. Englewood Cliffs,N.J.: Prentice-Hall, 1967.

Newell, A., & Rosenbloom, P. S. Mechanisms of skillacquisition and the law of practice. In J. R. Anderson(Ed.), Cognitive skills and their acquisition. Hills-dale, N.J.: Erlbaum, 1981.

Nielsen, S. L., & Sarason, I. G. Emotion, personality,and selective attention. Journal of Personality andSocial Psychology, 1981, 41, 945-960.

Norman, D. A. Memory while shadowing. QuarterlyJournal of Experimental Psychology, 1969, 21, 85-93.

Norman, D. A., & Bobrow, D. G. On the role of activememory processes in perception and cognition. InC. N. Cofer (Ed,), The structure of human memory.San Francisco: Freeman, 1976.

Posner, M. I. Chronometric explorations of mind. Hills-dale, N.J.: Erlbaum, 1978.

Posner, M. I., & Boies, S. J. Components of attention.Psychological Review, 1971, 78, 391-408.

Posner, M. I., & Snyder, C. R. R. Attention and cog-nitive control. In R. L. Solso (Ed.), Information pro-cessing and cognition: The Loyola Symposium. Hills-dale, N.J.: Erlbaum, 1975.

Ross, M., & Sicoly, F. Egocentric biases in availabilityand attribution. Journal of Personality and SocialPsychology, 1979, 37, 322-336.

Schneider, W., & Shiffrin, R. M. Controlled and au-tomatic human information processing: I. Detection,search, and attention. Psychological Review, 1977,84, 1-66.

Shallice, T. Dual functions of consciousness. Psycho-logical Review, 1972, 79, 383-393.

Shallice, T. The dominant action system: An informa-tion-processing approach to consciousness. In K. S.Pope & J. L. Singer (Eds.), The stream of conscious-ness: Scientific investigations into the flow of humanexperience. New York: Plenum, 1978.

Shiffrin, R. M., & Dumais, S. T. The development ofautomatism. In J. R. Anderson (Ed.), Cognitive skillsand their acquisition. Hillsdale, N.J.: Erlbaum, 1981.

Shiffrin, R. M., & Schneider, W. Controlled and au-tomatic human information processing: II. Perceptuallearning, automatic attending, and a general theory.Psychological Review, 1977, 84, 127-190.

Singer, J. L., & Antrobus, J. S. Daydreaming, imaginalprocesses, and personality: A normative study. InP. W. Sheehan (Ed.), The function and nature ofimagery. New York: Academic Press, 1972.

Smith, E. R., & Miller, F. D. Salience and the cognitivemediation of attribution. Journal of Personality andSocial Psychology, 1979, 37, 2240-2252.

Srull, T. K., & Wyer, R. S., Jr. The role of categoryaccessibility in the interpretation of information aboutpersons: Some determinants and implications. Jour-nal of Personality and Social Psychology, 1979, 37,1660-1672.

Srull, T. K., & Wyer, R. S. Jr. Category accessibilityand social perception: Some implications for the studyof person memory and interpersonal judgments. Jour-nal of Personality and Social Psychology, 1980, 38,841-856.

Storms, M. D. Videotape and the attribution process:Reversing the actors' and observers' point of view.Journal of Personality and Social Psychology, 1973,27, 165-175.

Taylor, S. E., & Crocker, J. Schematic bases of socialinformation processing. In E. T. Higgins, C. P. Her-man, & M. P. Zanna (Eds.), Social cognition: TheOntario Symposium on personality and social psy-chology. Hillsdale, N.J.: Erlbaum, 1981.

Taylor, S. E., & Fiske, S. T. Point of view and percep-tions of causality. Journal of Personality and SocialPsychology, 1975, 32, 439-445.

Taylor, S. E., & Fiske, S. T. Salience, attention, andattribution: Top of the head phenomena. In L. Ber-kowitz (Ed.), Advances in experimental social psy-chology (Vol. 11). New York: Acacemic Press, 1978.

Treisman, A. M., Squire, R., & Green, J. Semanticprocessing in dichotic listening? A replication. Mem-ory & Cognition, 1974, 2, 641-646.

White, P. Limitations on verbal reports of internalevents: A refutation of Nisbett and Wilson and ofBern. Psychological Review, 1980,57, 105-112.

Winer, B. J. Statistical principles in experimental de-sign (2nd ed.). New York: McGraw-Hill, 1971.

Wyer, R. S., Jr. The acquisition and use of social knowl-edge: Basic postulates and representative research.Personality and Social Psychology Bulletin, 1980, 6,558-573.

Wyer, R. S., Jr., & Srull, T. K. Category accessibility:Some theoretical and empirical issues concerning theprocessing of social stimulus information. In E. T.Higgins, C. P. Herman, & M. P. Zanna (Eds.), Socialcognition: The Ontario Symposium (Vol. 1). Hills-dale, N.J.: Erlbaum, 1981.

Zajonc, R. B. Cognition and social cognition: A histor-ical perspective. In L. Festinger (Ed.), Retrospectionson social psychology. New York: Oxford UniversityPress, 1980.

Received June 19, 1981Revision received January 21, 1982