attitudinal effects of mere exposure

Journal of Personalityand Social PsychologyMonograph Supplement

Volume 9, No. 2, Partj2 June 1968

ATTITUDINAL EFFECTS OF MERE EXPOSURE1

ROBERT B. ZAJONC

University of Michigan

The hypothesis is offered that mere repeated exposure of the individual to astimulus object enhances his attitude toward it. By "mere" exposure is meanta condition making the stimulus accessible to the individual's perception.Support for the hypothesis consists of 4 types of evidence, presented andreviewed: (a) the correlation between affective connotation of words andword frequency; (b) the effect of experimentally manipulated frequency ofexposure upon the affective connotation of nonsense words and symbols; (c)the correlation between word frequency and the attitude to their referents;(d) the effects of experimentally manipulated frequency of exposure on atti-tude. The relevance for the exposure-attitude hypothesis of the explorationtheory and of the semantic satiation findings were examined.

On February 27, 1967, the Associated Presscarried the following story from Corvallis,Oregon:

A mysterious student has been attending a class atOregon State University for the past two monthsenveloped in a big black bag. Only his bare feetshow. Each Monday, Wednesday, and Friday at11:00 A.M. the Black Bag sits on a small table nearthe back of the classroom. The class is Speech 113—basic persuasion. . . . Charles Goetzinger, professorof the class, knows the identity of the person inside.None of the 20 students in the class do. Goetzingersaid the students' attitude changed from hostilitytoward the Black Bag to curiosity and finally tofriendship [italics added].

This monograph examines the general hy-pothesis implied by the above phenomenon:mere repeated exposure of the individual to astimulus is a sufficient condition for the en-hancement of his attitude toward it. By "mere

1 This work was supported by the National Insti-tute of Mental Health, Grant MH 12174. I wish tothank Christine Linder, Dik van Kreveld, and JamesJ. Taylor for their invaluable assistance in carryingout various phases of the experimental work. I amalso indebted to Elinore Cottrell and Robert P.Weeks for making their students available as judgesand subjects.

exposure" is meant a condition which justmakes the given stimulus accessible to the in-dividual's perception.

Even though the hypothesis seems to be inconflict with such celebrated laws as famili-arity breeds contempt and absence makes theheart grow fonder, it is not particularly origi-nal or recent (Fechner, 1876, pp. 240-243;James, 1890, p. 672; Maslow, 1937; Meyer,1903; Pepper, 1919). The foremost proponentof this hypothesis, the advertising industry,has always attributed to exposure formidableadvertising potential. But—apparently, in re-spect for the law of enhancement by associa-tion—it seldom dared to utilize mere exposure.The product, its name, or its hallmark isalways presented to the public in contiguitywith other and always attractive stimuli, com-monly females, exposed more boldly than theproduct itself. At the same time, however, theadvertising industry also likes to warn againstowerexposure, relying, it would appear, on theabove law of familiarity (Erdelyi, 1940;Wiebe, 1940).

It isn't altogether clear just what evidencesupports these advertising principles. And

© 1968 by the American Psychological Association, Inc.

ROBERT B. ZAJONC

direct evidence that attitudes are enhanced bymere exposure or mere contact with the stimu-lus object is scant. Moreover, it is the productof antiquated methods, and almost all of itconcerns music appreciation (Downey &Knapp, 1927; Krugman, 1943; Meyer, 1903;Moore & Gilliland, 1924; Mull, 19S7; Ver-veer, Barry, & Bousfield, 1933; Washburn,Child, & Abel, 1927). The problem of atti-tudinal effects of social contact and inter-action has also been of some interest in thestudy of interracial attitudes (Cook & Selltiz,1952). But these studies have invariably ex-amined the effects not of mere perceptual ex-posure of people to each other, but of proc-esses considerably more complex: prolongedsocial interaction, group interdependence, co-operation, etc. (Deutsch & Collins, 1951;Kramer, 1950; MacKenzie, 1948; Wilner,Walkley, & Cook, 1952). Although the inde-pendent variables in these studies have gen-erally been featured under the labels "contact"and "exposure," the effects they report can-not, because of confounding with a multitudeof other events (and with reinforcement inparticular), be regarded as produced alone bycontact or exposure. Thus, it has been knownfor some time that social interaction enhancesthe attitudes of interactors toward each other(Bovard, 1951; Festinger, 1951; Homans,1961; Newcomb, 1963). But it is not knownjust what contribution to the relationship be-tween social interaction and attitudes ismade by mere exposure on the one hand, andby the variety of psychologically significantprocesses that necessarily accompany mere ex-posure during the course of social interaction,on the other.

The main empirical support for the exposurehypothesis comes, therefore, not from workon interaction, interracial attitudes, or atti-tudes in general, but from an entirely differ-ent and seemingly unrelated area of research.It comes from some recent work on word fre-quencies. This recent research shows thatthere exists an intimate relationship betweenword frequency and meaning. And this rela-tionship, in my opinion (for which I shalllater present support), may be a special caseof the more general relationship between mereexposure and attitude enhancement.

The strength and pervasiveness of the rela-

tionship between word frequency and mean-ing—the evaluative aspect of meaning, in par-ticular—is truly remarkable. For, if there isany correspondence between the frequencywith which words are used and the actualpreponderance of the things and events forwhich these words stand, then we may con-gratulate ourselves on living in a most happyworld. According to the Thorndike-Lorgecount (1944), the word "happiness" occurs761 times, "unhappiness" occurs only 49times. "Beauty" is to be found at least 41times as often as "ugliness," and "wealth"outdoes "poverty" by a factor of 1.6. We"laugh" 2.4 times as often as we "cry"; we"love" almost 7 times more often than we"hate"; we are "in" at least 5 times moreoften than we are "out"; "up" twice as oftenas we are "down"; much more often "suc-cessful" than "unsuccessful"; and we "find"things 4.5 times more often than we "lose"them—all because most of us are "lucky"(220) rather than "unlucky" (17).

We have all the reasons in the world to be"happy" (1449) and "gay" (418) rather than"sad" (202) and "gloomy" (72), for thingsare 5 times more often "good" than "bad,"almost 3 times more often "possible" than"impossible," and about five times more"profitable" than "unprofitable." That is, per-haps, why "boom" and "prosperity" outdo"recession" by a factor of just about 30,"abundance" outdoes "scarcity" by at least3:1, and "affluence" is 6 times more preva-lent than "deprivation." Catering to ourcorporeal sensibilities, things are 3 times moreoften "fragrant" than they are "foul," 12times more often "fresh" than "stale," andalmost 7 times more often "sweet" than"sour," and everything that can be filled isthree times as often "full" as it is "empty." Ifwe have anything, we have "more" of it 6times more often than we have "less" of it,2

and 3 times more often "most" of it than"least" of it. And those things that we haveso frequently more of are 5 times more often"better" than they are "worse," 6 timesmore often "best" than "worst," and 4 timesmore often "superior" than "inferior." Still,

2 N.B. The more-less ratio in this text is 7:1 upto now.

THE ATTITUDINAL EFFECTS OF MERE EXPOSURE

they "improve" at least 25 times as often asthey "deteriorate."

These examples suffice to convince one thatthe world represented by a one-to-one cor-respondence with word frequencies is as unrealas it is spectacular. Bitterly aware of it,Sartre (1964) confessed in his autobiography," . . . as a result of discovering the worldthrough language, for a long time, I tooklanguage for the world [p. 182]."

But, while they are unfaithful in represent-ing reality, word frequencies are extraordi-narily accurate in representing real values:words that stand for good, desirable, and pre-ferred aspects of reality are more frequentlyused.

It isn't entirely clear who discovered thisremarkable relationship between word fre-quency and the evaluative dimension of wordmeaning. Postman (1953) seems to be one ofthe early workers to note its generality, whileHowes and Solomon (1950) observed in theircritique of McGinnies' (1949) perceptual de-fense experiment that the so-called "taboo"words he used as stimuli are particularly in-frequent. However, the first systematic re-search effort that demonstrates the word-frequency-word-value relationship is due toJohnson, Thomson, and Frincke (1960).These authors were the first, I believe, to col-lect empirical data showing that words with"positive" meaning have higher frequencycounts than words with "negative" meanings.They have also gathered experimental evi-dence showing that the repeated use of a non-sense word tends to enhance its rating on thegood-bad scale of the semantic differential.Johnson, Thomson, and Frincke (1960) havenot tried to explain either of these two as-pects of the frequency-value relationship, be-ing primarily concerned with its implicationsfor the study of word-recognition thresholds.

This paper examines the frequency-valuerelationship, proposing that it is considerablymore pervasive and general than implied bythe Johnson-Thomson-Frincke results, andthat it is, moreover, a special case of abroader and more basic phenomenon: the en-hancement of attitudes by mere repeated ex-posure. I shall first review evidence on thecorrelation between word frequency and wordvalue, and between stimulus frequency and

attitude. Experimental evidence on these tworelationships, and on the likely causal direc-tion, will then be examined.

Word Frequency-Word Value: CorrelationalEvidence

Johnson, Thomson, and Frincke (1960) ob-tained correlations of .63, .40, and .38 be-tween the L-count (Thorndike & Lorge, 1944)and the good-bad scale values for three sam-ples of randomly chosen words. In a furtherattempt, they constructed 30 pairs, each con-sisting of one frequent and one infrequentword. These pairs were given to a group ofsubjects with the instructions to "encircle themost pleasantly toned word of each pair." In87% of the pairs the majority of subjectsendorsed the more frequent word. Finally, 64nonsense syllables of low, medium, and highassociation were rated by a group of subjectson the good-bad scale of the semantic differ-ential. Johnson, Thomson, and Frincke re-ported a clear relationship between associationvalue and "goodness" ratings. The rationaleof this study invoked the assumed relationshipbetween association of the given nonsensesyllable and the probability of occurrence ofthe corresponding letter combination in mean-ingful words (Underwood, 1959).

In an attempt to examine the generality ofthis phenomenon, we studied the evaluationsof 154 antonym pairs. First, a large pool ofantonym pairs was amassed. From this poolall symmetric8 pairs were chosen in the fol-lowing manner. For each antonym pair 10judges, 1 at a time, were asked to give theantonym of one member of the pair. Tenother judges—independently of the first 10—were asked to give the antonym of the other

3 One finds in the course of this endeavor that theantonymic relation is seldom symmetric. Accordingto the standard sources, if Y is listed as the antonymof X, then chances are that not X but Z is listed asthe antonym of V. For instance, in the 1960 edi-tion of Webster's New Collegiate Dictionary, "ex-tend" is given as the antonym of "contract." Lookingup "extend" we find, however, that its antonym is"reduce." The antonym of "reduce," on the otherhand, is "increase." The antonym of "increase" is"decrease," the antonym of "decrease" is "amplify,"the antonym of "amplify" is "condense," and theantonym of "condense" is "expand." We can ulti-mately close the circle, because "contract," accordingto this source, is the antonym of "expand."

ROBERT B. ZAJONC

TABLE 1SEMANTIC PREFERENCE AND FREQUENCY OF 154 ANTONYM PAIRS

%agree-ment

1001001001001001001009999999999999999999999999999999998989898989898989898989898989898989898989898989797979797979797979797

197979797

[9796

Preferredalternative (a)

ableattentivebetterencouragefriendlyhonestpossibleadvancebestcleancomfortablefavorablegoodgratefulpeacepresentpureresponsiblerewardrightsmiletolerantvictoryaddadvantageagreeablecapabledesirablefindfortunateforwardfriendhighhonorablekindlegallifelovematuremoralpleasantpolitereliablesuccessvalidvoluntaryadequatecompetentfoundimportantlikelyonpatiencepatientpatientlypopularpositiveprofitablepromoteremembersatisfactorywillinglyabove

Nonpref erredalternative (6)

unableinattentiveworsediscourageunfriendlydishonestimpossibleretreatworstdirtyuncomfortableunfavorablebadungratefulwarabsentimpureirresponsiblepunishmentwrongfrownintolerantdefeatsubtractdisadvantagedisagreeableincapableundesirableloseunfortunatebackwardenemylowdishonorableunkindillegaldeathhateimmatureimmoralunpleasantimpoliteunreliablefailureinvalidinvoluntaryinadequateincompetentlostunimportantunlikelyoffimpatienceimpatientimpatientlyunpolularnegativeunprofitabledemoteforgetunsatisfactoryunwillinglybelow

Fre-quencyof (a)

93049

2354205357393

1289452

185078134893

5122194472

1075197267154

38742143

42118

2018404

58176160

2698136736

25531674

581521180

48045129

9127245711578

57322289569

28921130364

3022413939285

418925790

168215466

941

Fre-quencyof (W

2394

4501471941

45910529222111225

100113

1118654

3080

89021613

1666

41433042

593108139883

12248

3434

8157561719

11439

26256265923

10744025

36443979821228122

8823213

529

%agree-ment

969696969696969696969696969695959595959595959594949494949393939393939393939292929292929191919190909090908988888685858483837978


activeearlyfrontfulllivepresenceprobablerationalreasonableresolutelystrongsucceedsuperiortimelyacceptdirectincludeincreasemostpracticalregularlyrichwealthapproveconsciousleaderobedienttogetheragreementcertainfirstmajornormalregularunselfishupwardswidemorenowupupwardvisibleyesalwaysfamiliarmaximumoptimismagreenecessaryoversweetwholelightdeepsmoothwhiteinindependentfastcomedyfastendaydry

Nonpreferredalternative (&)

passivelatebackemptydieabsenceimprobableirrationalunreasonableirresolutelyweakfailinferioruntimelyrejectindirectexcludedecreaseleastimpracticalirregularlypoorpovertydisapproveunconsciousfollowerdisobedientapartdisagreementuncertainlastminorabnormalirregularselfishdownwardsnarrowlessthendowndownwardinvisiblenoneverunfamiliarminimumpessimismdisagreeunnecessaryundersourpartdarkshallowroughblackoutdependentslowtragedyunfastennightwet

Fre-quencyof (a)

1861022109411294307

2776433

15530

77026416627

667416533781

344334012265624317129937370

1835143800

5154366335340329

59380157665

11718111110

220232853454328

729715

7520679

16632387881346

266375253

134514126142

4549592

Fre-quencyof (6)

2928596587395

1079163149

564

276620406

51233886

1259125

85714645

116454

27621

1073517

834344

13740

3911357

102085534

2774

117425715

39861138

1072961

10215851005

104294

108313649

1843418916

3385319


TABLE I—Continued

%agree-ment

7878777472727070686867676564


longunshakenusuallyupstairsinnerinteriornearunlimitedinsidewrapinfiniteinternalcominginformal

Nonpreferredalternative (i)

shortshakenunusuallydownstairsouterexteriorfarlimitedoutsideunwrapfiniteexternalgoingformal

Fre-quencyof (a)

53626

718314143185

133843

6562937136

148664

Fre-quencyof (6)

8878391

2269748

183567

921172

264623

166

CO

agree-ment

6363615959585857555555545251


answermendifferentinwardmanhusbandusualoffensehotimportinwardlyinconspicuousplaymortal

Nonpreferredalternative (6)

questionwomensameoutwardwomenwifeunusualdefensecoldexpertoutwardlyconspicuousworkimmortal

Fre-quencyof (a)

213236141194

437355178851686

1006863233

260654

Fre-quencyof (6)

130225521747

5424311668273223

1092883359

272026

member of the pair. Only those pairs wereretained about which the 20 judges showedunanimous agreement with the dictionarysources. A list of 154 antonym pairs was thusobtained. These were given to 100 subjects,all college students, for judgments as to whichmember had "the more favorable meaning,represented the more desirable object, event,state of affairs, characteristic, etc." A differ-ent random order of the antonym pairs wasgiven to each subject, and the lateral positionsof the members of each pair were reversed atrandom for half of the group.

Table 1 shows the list of these 154 antonympairs, together with the "desirability" and thefrequency data (the Thorndike-Lorge L-count). The preferred member of each pair isalways listed first. The "desirability" figuresare simply the percentages of subjects choos-ing the left member of the pair as the pre-ferred alternative.

It is of some interest, however incidental,that there is considerable agreement aboutdesirability of the meanings. On half of theitems the agreement exceeded 95%. Agree-ment is high even for words which are notgenuinely evaluative. For instance, 97 of the100 students preferred "on" to "off," 98 pre-ferred "add" to subtract, 96 "above" to"below," and 92 "upward" to "downward."

For the overwhelming majority of the itemsthe preferred word is also the more frequentone. Only 28 of the 154 antonym pairs (18%)show a negative relationship between fre-quency and desirability. Moreover, these "re-versals" occur primarily for antonym pairs on

which there is relatively little agreement. Forpairs with agreement greater than 95% (i.e.,the upper half of the list) there are only sixreversals out of the 77 possible. It is signifi-cant, moreover, that in three of these sixantonym pairs the less desirable member(which in these cases is the more frequentone) has more meanings and linguistic usesthan the more desirable one. "Invalid" meansboth "not valid" and "cripple," but "valid"is just "valid." "Yes" is an adverb, but "no"is an adverb and an adjective. And "front"is a noun, a verb, and an adjective, while"back" is all that and an adverb to boot.

Toward the end of the list where the desir-ability preferences are divided fairly evenlybetween the two members of the antonympairs, the frequencies of the two antonymsoften are nearly the same. "Play" is preferredto "work" only by a majority of two (a curi-ous commentary on the contemporary collegepopulation!), and the respective frequencycounts of these antonyms are 2606 and 2720.The "hot-cold" preference is 55 to 45 andtheir frequency counts 1006 and 1092. The"husband-wife" preference is 58 to 42 andtheir respective frequencies, 1788 and 1668.

Three antonym items about which agree-ment was complete or nearly complete show acurious pattern of results. They are "good-bad" (5122:1001), "better-worse" (2354:450), and "best-worst" (1850:292). Since"better" is presumably better than "good,""worse" worse than "bad," and since "best"is presumably better than "better," and"worst" worse than "worse," we would expect

ROBERT B. ZAJONC

the greatest separation between the frequen-cies of "best" and "worst," smallest betweenthe frequencies of "good" and "bad," andmedium between the frequencies of "better"and "worse." Since absolute differences aredeceiving, we best take the ratios of the fre-quencies, which are 6.34, 5.23, and 5.12 for"best-worst," "better-worse," and "good-bad,"respectively. It is indeed the case that thefrequency ratios increase from "good-bad" to"best-worse." However, if frequency reflects"desirability," we would also expect the fre-quency of "best" to exceed the frequency of"better," and that of "better" to exceed thefrequency of "good." In fact, however, "good"is more frequent than "better," and "better"more frequent than "best!" But is "better"better than "good?" In an extensive study ofmeanings Hosier (1941) found that "good"was consistently rated as better than "better."

Startling as this may appear to grammarians, it ispsychologically sound, since GOOD is a positive asser-tion, whereas BETTER implies comparison with somestandard which might, in many cases, be itself un-favorable. Compare the often heard comment, "Heis getting better, but he is still far from good" [p.134].

For purposes of comparison the frequenciesof French, German, and Spanish equivalentsof some of the antonyms examined are givenin Table 2 below. Systematic data on indige-nous desirability ratings are unfortunatelynot available, but it would be surprising ifthe French, German, and Spanish judgmentsdiffered from those obtained in the UnitedStates. An informal inquiry among foreignvisitors marshalled a good deal of support forthis conjecture. Comparing the data in Tables1 and 2, the agreement is rather striking. In15 out of the 44 cases the frequency relationin the antonym pairs is the same in the threeforeign languages as in English: the morefavorable item is more frequent, a result ex-ceeding chance expectation by a large margin.The results in Table 2, furthermore, give aready expression to our favorite ethnic preju-dices. The relatively low frequency of thetwo Romance equivalents of "early" and thehigh frequency of these equivalents of "late,"in comparison to their Germanic counterparts,make generalizations about national charactertempting, as does the relatively low frequency

of the German equivalent of "reward." Theforeign equivalents of answer-question, hot-cold, import-export, peace-war, etc., however,show patterns of differences that may reflectmore than superficial linguistic idiosyncrasies.

Several questions can immediately be raisedabout the above results. First, are these fig-ures up to date? The Thorndike-Lorge countis based on samples of material published dur-ing the late twenties and the early thirties.The German equivalents come from a sourcedating to the late 19th century (Kading,1898). The French count was published in1929 (Van der Beke, 1929), and the Spanishin 1927 (Buchanan, 1927). Secondly, dothese results reflect general verbal habits?Word counts are based on printed materialalone. Do people show the same linguisticpredilections in ordinary speech as they doin writing? Admittedly, both questions indi-cate caution in generalizing from the aboveresults. But this caution needn't be excessive.Howes (1954) has recently asked Harvardand Antioch undergraduates to estimate theprobabilities of various words. The correla-tions between the students' estimates of sev-eral word samples and the L-count of theThorndike-Lorge source varied around .80.There is also evidence from word associationstudies showing that word counts do reflectgeneral verbal habits of the population. Aword which has a high frequency of occur-rence in print is also a highly probable asso-ciate. The association norms to 200 wordswere recently collected by Palermo and Jen-kins (1964) from a sample of 4,500 schoolchildren and college students in Minneapolis.The list of the 200 stimulus words representsa systematic sample of verbs, nouns, pro-nouns, adverbs, adjectives, participles, etc.,all having fairly high frequency on the Thorn-dike-Lorge counts. Since in the word associa-tion task each subject makes one response toeach stimulus word, Palermo and Jenkinscollected from their subjects 900,000 wordresponses. Among them "good" occurred 4890times, "bad" only 1956. The response "right"was given 477 times, the response "wrong"only 100 times. "Full" was found 431 timesamong the associations, "empty" only 62times. "Strong" was given 557, "weak" 96times. "Together" occurred 575 times, "apart"


TABLE 2FREQUENCY RANKS OP ENGLISH, FRENCH, GERMAN, AND SPANISH ANTONYM PAIRS

English

able (3)unable (9)

accept (3)reject (9)

active (6)passive (14)

answer (2)question (3)

better (2)worse (4)

certain (2)uncertain (9)

clean (3)dirty (7)

comedy (9)tragedy (9)

comfortable (5)uncomfortable (11)

day (2)night (2)

direct (3)indirect (12)

dry (3)wet (4)

early (2)late (2)

fast (2)slow (3)

find (2)lose (3)

friend (2)enemy (3)

full (2)empty (4)

good (2)bad (2)

high (2)low (2)

hot (2)cold (2)

husband (3)wife (3)

import (7)export (11)

increase (3)decrease (8)

independent (6)dependent (14)

life (2)death (2)

light (2)dark (2)

French

capable (3)incapable (4)

accepter (2)rejeter (5)

actif (6)passif (?)

reponse (4)question (2)

meilleur (2)pire (S)

certain (2)incertain (10)

propre (2)sale (7)

comedie (6)tragedie (9)

a 1'aise (4)inconfortable (9)

jour (2)nuit (2)

direct (6)indirect (12)

sec (3)mouil!6 (5)

t6t (3)tard (2)

vite (2)lent (4)

trouver (2)perdre (2)

ami (2)ennemi (2)

plein (2)vide (4)

bon (2)mauvais (2)

haut (2)bas (2)

chaud (3)froid (3)

mari (3)femme (2)

importation (11)exportation (10)

augmentation (10)reduction (11)

independent (7)ddpendent (?)

vie (2)mort (2)

clair (3)sombre (3)

German

fahig (4)unfahig (11)

annehmen (2)ablehnen (5)

tatig (S)untatig (?)

Ant wort (3)Frage (2)

besser (2)schlechter (?)

sicher (2)unsicher (9)

sauber (9)schmutzig (12)

Komodie (9)tragodie (11)

bequem (S)unbequem (10)

Tag (2)Nacht (2)

direkt (3)indirekt (8)

trocken (S)nass (9)

frtih (2)spat (2)

schnell (2)langsam (3)

finden (2)verlieren (2)

Freund (2)Feind (2)

voll (2)leer (4)

gut (2)schlecht (3)

hoch (2)niedrig (4)

heiss (S)kalt (3)

Mann (2)Frau (2)

Einfuhr (11)Ausfuhr (12)

Vermehrung (6)Verminderung (11)

selbststiindig (4)abhiingig (6)

Leben (2)Tod (2)

hell (4)dunkel (3)

Spanish

capaz (3)incapaz (7)

acceptar (3)rechazar (S)

activo (6)pasivo (10)

respuesta (4)pregunta (4)

mejor (2)peor (2)

cierto (2)incierto (9)

limpio (3)sucio (6)

comedia (4)tragedia (8)

coniodo (7)incomodo (10)

dia (2)noche (2)

directo (4)indirecto (8)

seco (3)mojado (6)

temprano (4)tarde (2)

pronto (2)lento (4)

encontrar (2)perder (2)

amigo (2)enemigo (2)

lleno (2)vacio (4)

buen (2)mal (2)

alto (2)bajo (2)

caliente (5)frio (2)

esposo (2)esposa (2)

importaci6n (?)exportaci6n (13)

aumento (5)diminuci6n (?)

independiente (S)dependiente (9)

vida (2)muerte (2)

claro (2)obscuro (2)

ROBERT B. ZAJONC

TABLE 2— Continued

English

life (2)die (2)long (2)short (2)

love (2)hate (4)

more (2)less (2)

near (2)far (2)

peace (3)war (2)

positive (9)negative (11)

possible (3)impossible (5)presence (4)absence (7)reward (6)punishment (6)

right (2)wrong (3)

strong (2)weak (3)

sweet (2)sour (9)

together (2)apart (4)victory (5)defeat (7)wealth (4)poverty (7)white (2)black (2)

wide (2)narrow (3)

French

vivre (2)mourir (2)

long (2)court (3)

aimer (2)hair (6)

plus (2)moins (2)

pres (3)loin (2)

paix (3)guerre (3)

positif (6)negatif (11)

possible (2)impossible (3)presence (2)absence (5)recompense (6)punition (12)juste (2)faux (3)

fort (2)faible (3)

doux (2)amer (4)

ensemble (2)s6par6 (2)

victoire (4)defaite (8)

richesse (5)pauvrete (12)

blanc (2)noir (2)

large (2)etroit (3)

German

leben (2)sterben (2)

lang (2)kurz (2)

lieben (2)hassen (6)

mehr (2)weniger (2)

nah (2)fern (2)

Friede (3)Krieg (2)

positiv (8)negativ (?)moglich (2)unmoglich (3)

Anwesenheit (9)Abwesenheit (9)Anerkennung (S)Strafe (4)

richtig (2)falsch (3)

stark (2)schwach (3)

suss (4)sauer (9)

zusammen (2)getrennt (3)

Sieg (4)Niederlage (8)Vermogen (4)Armut (10)weiss (2)schwartz (3)

breit (4)schmal (6)

Spanish

vivir (2)morir (2)

largo (2)corto (3)

amar (2)odiar (7)

mas (2)menos (2)

cerca (2)lejos (2)

paz (2)guerra (2)

positive (7)negative (7)

posible (2)imposible (2)

presencia (3)ausencia (4)

premio (4)castigo (4)

justo (3)mal (2)

fuerte (2)debil (4)

dulce (2)amargo (4)

junto (2)separado (3)victoria (5)derrota (9)

riqueza (3)pobreza (S)

bianco (2)negro (2)

ancho (3)angosto (8)

Note.—The figures in brackets indicate frequency ranks: (1) means that the word is among the 500 most frequent words, (2)it is among the 1000 most frequent words, (3) that it is among the 1500 most frequent words, etc. The source of these counts

iton (1940).thatis Eaton (1940).

29 times. "Light" was a response 8655 times(N.B., some subjects must have given it morethan once), "dark" 4274 times. But as in thecase of the Thorndike-Lorge count, "front"occurred 22 times, while "back" occurred 265times; "rich" was given 36 times, while "poor"was a response 95 times. "Near" was given981 times, "far" 1218. "Coming" was given166 times, "going" 714 times. And, as in L-count, "play" and "work" showed 791 and957 occurrences, respectively.

However, the best evidence about the rela-tionship between the individual's verbal hab-

its and the evaluative aspect of meaning isfound in a recent study by Siegel,* althoughit wasn't the purpose of her study to explorethis relationship. Siegel's experiment dealtwith the effects of verbal reinforcement on theemission of words differing in affective conno-tation and in frequency. Eighteen six-letterwords of known frequencies and previouslyjudged on the good-bad and the pleasant-unpleasant scales were selected from a larger

* Siegel, Felicia S. Effects of word frequency andaffective connotation on verbal responding duringextinction. (Mimeo)


sample. Six of these words were of high fre-quency (100 and more in 1 million), six ofmedium (20 to 30), and six of low frequency(1 to 5). Within each frequency class twowords were previously judged to be good, twoneutral, and two bad. Three groups of sub-jects, other than those involved in the affec-tive judgments, participated in the experi-ment, each having to deal with six words ofthe same frequency. The procedure consistedof presenting the subject with the list of sixwords, all high, medium, or low in frequency,depending on the condition in which he was in,and giving him at the same time a stack ofcards on which appeared illegible six-letter"words." Ostensibly, each card contained oneof the six words in the subjects' list. Actually,the "words" consisted of random sequencesof six letters, printed over several thicknessesof paper and one carbon. Their legibility wasfurther reduced by placing each card in anonionskin paper envelope. The subjects' taskwas to "read" or to guess what word appearedon each card. Of interest for the present pur-poses are the first SO trials which served to es-tablish operant rate, and during which, ofcourse, no reinforcement of any sort was given.Table 3 shows data on the guessing behaviorof Siegel's subjects as a function of word fre-quency and affective connotation. Reported ineach cell is the average number of times aword of a given frequency and affective valuewas used as a guess during the SO operanttrials. Since there are six words to choosefrom, 8.33 represents a chance response rate.It is clear, however, that both frequency andaffective connotation displace response rateaway from the chance level. High frequencyseems to result in overcalling, and low fre-

TABLE 3FREE RESPONSE EMISSION AS A FUNCTION or

WORD-FREQUENCY AND WORD VALUE"

... . .

GoodMediumBadX

Word frequency

Low

7.436.286.286.66

Medium

9.438.575.867.95

High

9.688.717.718.70

8.8S7.856.61

' From Siegel, I960.

quency in undercalling. But it is striking todiscover that affective connotation had aneven stronger effect on response emission, themarginals for that variable showing a some-what greater range of differences.

Some words in the language have primarilyan evaluative function. These words shouldshow the frequency-value relationship withparticular clarity. Several instances of thisrelationship are examined.

Let us first consider the scales of the Se-mantic Differential (Osgood, Suci, Si Tannen-baum, 1957). We have chosen only thosescales which have high and relatively pureloadings on one of the three main factors,evaluation, potency, and activity. Table 4shows the polar opposites of these scales, to-gether with their frequencies according to theThorndike-Lorge L-count. The left-hand polaropposites in the three columns are the favor-able, potent, and active ends of the scales. Itis significant that among the 19 evaluativescales the favorable polar opposite has alwayshigher frequency than the unfavorable oppo-site. For the scales which do not load high onthe evaluative factor the high frequencies aredivided fairly evenly among the potent andnonpotent opposites. In 9 of the IS potencyscales the highly potent end of the scale ismore frequent. In 3 of the 8 activity scalesthe active polar opposite is more frequent.

There are two other instances of a highcorrelation between frequency and value foradjectives. The first comes from the work byGough (1953). Gough has given the items ofhis Adjective Checklist to 30 judges who ratedeach adjective for favorability. The mostfavorable and the least favorable quartiles ofGough's checklist are reported in his publi-cation. The average word frequency of theupper quartile is 140, and of the lower quar-tile 48. The second illustration comes fromdata collected by Anderson (1964). A list ofSSS adjectives was recently used by Andersonin his work on impression formation. The listwas constructed out of a large sample ofitems. The 555 selected items were given byAnderson to a group of 100 subjects with theinstructions to rate on a 7-point scale "howmuch you yourself would like the person de-scribed by that word." We have simply com-puted the correlation between these likeabil-

10 ROBERT B. ZAJONC

TABLE 4POLAR OPPOSITES OF THE SEMANTIC DIFFERENTIAL

AND THEIR FREQUENCIES

Evaluative factor

beautifulcleanfairfragrantgoodgratefulhappyharmonioushonestkindnicepleasantpositivereputablesacredsuccessfulsweettruewise

uglydirtyunfairfoulbadungratefulsaddissonantdishonestcruelawfulunpleasantnegativedisreputableprofaneunsuccessfulsourfalsefoolish

987781561

665122

1941449

26393

1521630457

9223

102352679

1711420

1782215939

100113

2029

4116537011428211314

102209223

Potency factor

bassbravedeephardheavylargemasculinematureroughruggedseverestrongtenaciousthickwide

treblecowardlyshallowsoftlightsmallfeminineyouthfulsmoothdelicatelenientweakyieldingthinnarrow

28216881

1909680

16975491

29437

119770

22443593

1726

104549

10051818

4099

346248

9276

7646391

Activity factor

activebrightexcitablefasthereticalhotrashsharp

passivedarkcalmsloworthodoxcoldcautiousdull

514645

7514

21006

37324

4341005267434

211092

48289

ity ratings and the logarithm of the Thorn-dike-Lorge L-count.5 Figure 1 shows thisrelationship graphically, where means of log

5 Items for which there was no frequency informa-tion in the Thorndike-Lorge count were not includedin computing this coefficient. These items were pri-marily of the hyphenated form, such as open-minded,good-humored, well-spoken, fault-finding, ultra-criti-cal, wishy-washy, etc.

RE

QU

EN

CY

PE

R 4

,SO

O,O

OO

WO

RD

S (

L-C

OU

NT

—

01

01-

01

01

O

O

OO

O

0

0

O

O

— <X/

.>

^

ss

s

^s

s

r- 83

1

o-I.OO

I.OI- 2.01- 3.OI- 4.01- 5.OI-2.OO 3.OO 4.OO 5OO 6.OO

AVERAGE FAVORABILITY RATING

FIG. 1. Average frequencies of 555 adjectives ratedfor favorability. (Based on data from Anderson,1964.)

frequencies are plotted for six categories ofadjectives in increasing order of favorability.Considering that the reliabilities of the Thorn-dike-Lorge count and of Anderson's favor-ability ratings are less than perfect, the co-efficient of correlation of .83 is particularlyimpressive.

Miller, Newman, and Friedman (19S8)have shown that word frequency is a negativefunction of word length. The problem im-mediately arises, therefore, as to which ofthese two variables is critical for word valueand word meaning. In order to examine thispossible confounding between frequency andword length, the above correlation was recom-puted holding the number of letters constant.No appreciable change in the previously ob-tained coefficient was observed.

The relationship between word frequencyand word length is generally explained interms of the principle of least effort. Wordsthat require considerable effort in writing andin speech are less likely candidates for use.In an attempt to control for effort Frinckeand Johnson (1960) have asked subjects tochoose the "most pleasantly toned word" fromeach of 108 homophone pairs. The greatestmajority of these pairs consisted of words ofthe same length, and all pairs, of course, con-sisted of words that required the same effortin uttering them. Out of 3,132 possible choices,

THE ATTITUDINAL EFFECTS OF MERE EXPOSURE 11

the more frequent member of the pair waschosen 1,836 times.

Dixon and Dixon (1964) have given a listof 200 verbs (in past-tense form) to 60 fe-male and 60 male judges who rated them onan 11-point good-bad scale. The instructionswere to rate what "kind of impression thesubject thought a psychologist would get ofhim when he used each verb in a sentence."These impression ratings have correlationswith log frequencies (the Thorndike-Lorge L-count) equal to .48 for females and to .50for males. But it must be pointed out thatthese coefficients represent correlations se-verely attenuated by unreliability of the fre-quency variable. The Thorndike-Lorge countlists verbs in the present-tense form. If anadjectival form of the verb exists, then it isalso listed. In our own research, in computingcorrelation coefficients, only the present-tensefrequencies were used.

Miron (1961) had American and Japanesesubjects rate a sample of three-element pho-netic combinations on various scales of theSemantic Differential. The subjects also ratedthese stimulus materials for their familiarity.It is interesting that the correlations betweenfamiliarity and the composite of evaluativescales were .59 and .50 for the American andthe Japanese samples, respectively. But thecorrelations of familiarity with the compositesof the potency and activity factors were lowand negative.

As a final example of the relationship be-tween word frequency and the evaluative as-pect of meaning, two poems by William Blakeare called to the reader's attention:

Infant Joy

"I have no name:I am but two days old,"What shall I call thee?"I happy am,Joy is my name."Sweet joy befall thee!

Pretty joy!Sweet joy but two days old,Sweet joy I call thee:Thou dost smile,I sing the while,Sweet joy befall thee!

Infant Sorrow

My mother groaned! My father wept;Into the dangerous world I leapt;Helpless, naked, piping loud,Like a fiend hid in a cloud,

Struggling in my father's hands,Striving against my swadling bands,Bound and weary I thought bestTo sulk upon my mother's breast.

In these two poems, expressing opposite quali-ties of affect, the frequencies of the criticalwords (i.e., words which convey the majorcontent, and hence not articles, pronouns, orauxiliary verbs) were averaged. The averagefrequency of Infant Joy is 2,037. The averagefor Infant Sorrow is 1,116. Two formally simi-lar verses, one by Browning and the other byShelley, show the same pattern:

Song. R. Browning

The year's at the spring,And day's at the morn;Morning's at seven;The hillside's dew-pearled;The lark's on the wing;The snail's on the thorn;God's in his Heaven—All's right with the world.

Dirge. P. B. Shelley

Rough wind, that meanest loudGrief too sad for song;Wild wind, when sullen cloudKnells all the night long;Sad storm, whose tears are in vain,Bare woods, whose branches strain,Deep caves and dreary main—Wail, for the world's wrong.

The average word frequency of Browning'spoem is 1,380. The poem by Shelley—whichcomes to a rather different and sadder con-clusion—has an average frequency of 728.

Stimulus Frequency Attitude: CorrelationalEvidence

We may now turn to the more general ques-tion of the effect of exposure on attitude, stilllimiting ourselves to correlational studies.Here, less evidence exists, and the evidencewhich is available is often indirect. But the

12 ROBERT B. ZAJONC

results are quite similar to those just reviewed.For instance, Alluisi and Adams (1962) founda correlation of .843 between the preferencesubjects expressed for the appearance of let-ters and their frequency in the language.Strassburger and Wertheimer (1959) hadsubjects rate for "pleasantness" nonsensesyllables varying in association value. Higherassociation values consistently received higher"pleasantness" ratings. Wilson and Becknell(1961) and Braun (1962) successfully repli-cated these results. Braun also found thateight-letter pseudo-words, varying in theirorder of approximation to English (Miller,1951), show the same pattern. These twostudies differ from the similar ones by John-son, Thomson, and Frincke, discussed earlier,in that subjects in the former ones were askedto judge how pleasant were the stimuli them-selves, or how much subjects liked them (Wil-son & Becknell, 1961), while in the latterwhether they meant something close to "good"or close to "bad."

In 1947 the National Opinion ResearchCenter conducted an extensive survey on the"prestige" of various occupations and pro-fessions. Nearly 100 occupational categorieswere rated for "general standing." Twenty-four of these occupations are labeled by singlewords, such as "physician," "scientist," "jani-tor," etc. The remainder is described lesseconomically: "owner-operator of a printingshop," or "tenant farmer—one who owns live-stock and machinery and manages the farm."Thus, one is able to determine the frequencyof usage for only a part of this list—the 24single-word occupations. The correlation be-tween rated occupational prestige of these 24items and the log of frequency of usage is .55.

Similar to the ratings of occupational pres-tige are the social distance ratings of ethnicand racial groups, first developed by Bogardus(1925) over 30 years ago. Recent replicationsshow that these social distance ratings enjoyremarkable stability (Bogardus, 1959). Thecorrelation between the so-called "racial-dis-tance quotients," which are numerical equiva-lents of these ratings, and the log frequencyof usage of these ethnic labels is .33.

In order to explore relationships of thissort further, I have selected 10 countrieswhose names are found in the Thorndike-

Lorge L-count, and whose frequencies can bearranged in increasing order in approximatelyconstant log units. These countries were thengiven to high-school students with the instruc-tions to rank-order them in terms of liking.Table 5 shows the average rank each countryreceived and its frequency of usage accordingto the L-count. There seems to be little ques-tion about the frequency-attitude relationship.The same relationship is found with Americancities. Selected were 10 cities that (a) arelisted in the Thorndike-Lorge L-count, and(b) can be arranged in increasing order offrequency in approximately constant log units.University students were asked how muchthey would like to live in each of these 10cities. Their task, specifically, was to rank-order these cities according to their pref-erences "as a place to live." The averageranks, together with frequency counts of these10 cities, are shown in Table 5.

Other subjects, also high-school studentsin the Midwest, were asked to rate on a 7-point scale how much they liked various trees,fruits, vegetables, and flowers. In each case 10items were selected which were listed in theThorndike-Lorge count and which could beordered according to a constant log frequencyunit. Table 6 shows both the average ratings(0 = dislike; 6 = like) and the frequencycounts for the four types of items. The rankcorrelations between the frequency and ave-rage attitude are .89, .85, .84, .81, .85, and .89,for countries, cities, trees, fruits, vegetables,and flowers, respectively.

TABLE 5PREFERENCE RANKS AND FREQUENCY COUNTS

FOR 10 COUNTRIES AND 10 CITIES

Countries

Country

EnglandCanadaHollandGreeceGermanyArgentinaVenezuelaBulgariaHondurasSyria

Fre-quency

4971305931

224159314

Aver-age pref-erencerank

2.673.333.424.004.926.086.587.757.928.34

Cities

City

BostonChicagoMilwaukeeSan DiegoDaytonBaltimoreOmahaTampaEl PasoSaginaw

Fre-quency

255621124

9146828512

Aver-age pref-erencerank

2.753.083.834.255.756.087.087.087.507.58


TABLE 6PREFERENCE RATINGS or TREES, FRUITS, VEGETABLES, AND FLOWERS,

AND THEIR CORRESPONDING FREQUENCIES

Trees

pinewalnutoakrosewoodbirchfirsassafrasaloesyewacacia

f

17275

1258

3414

2134

APR

4.794.424.003.963.833.753.002.922.832.75

Fruits

applecherrystrawberrypeargrapefruitcantaloupeavocadopomegranategooseberrymango

f

2201671216233

1.516852

APR

5.135.004.834.384.003.752.712.632.632.38

Vegetables

cornpotatolettucecarrotradishasparaguscauliflowerbroccolileekparsnip

f

22738414296435

271838

APR

4.174.134.003.573.132.331.961.961.961.92

Flowers

roselilyvioletgeraniumdaisyhyacinthyuccawoodbineanemonecowslip

f

8011641092762161482

APR

5.554.794.583.833.793.082.882.872.542.54

Note.—f = frequency of usage; APR = average preference rating.

Of course, word counts do not faithfullyrepresent the frequencies with which one en-counters the above items. And it is difficultto discover precisely how often the averageMidwestern high school student encounters ayew, a cowslip, or a radish. But a fair indexof frequency of exposure can be found in farmproduction data. For seven of the vegetablesin Table 6 farm production figures for 1963are available, and they are shown below inthousands of tons:

corn (4.17)potatoes (4.13)lettuce (4.00)carrots (3.57)asparagus (2.33)cauliflower (1.96)broccoli (1.96)

2,340.913,777.1

1,937.6843.8187.8123.4123.9

Included also (in brackets) are average pref-erence ratings of these seven vegetables. Therank correlation between the production fig-ures and the average preference ratings is .96.

Of course, this impressive correlation co-efficient, like those we observed above, maynot reflect the effect of frequency on attitudebut the effect of attitude on frequency. Thus,it can be argued that many roses are grownbecause people like roses. But it can also beargued that people like roses because thereare many roses growing. There is less am-biguity, however, with regard to the correla-tion between frequency of letters and thepreference for their appearance (Alluisi &Adams, 1962). There aren't so many e's inEnglish just because we like the way e's look.Still, until there is experimental evidence, the

question of which is the cause and which theeffect remains a matter of conjecture. Weshall now turn, therefore, to such experimentalevidence.

Exposure—Meaning: Experimental Evidence

Experiment I. The first experimental study on therelationship between exposure and word meaningwas carried out by Johnson, Thomson, and Frincke(1960). These authors first asked subjects to rate anumber of nonsense words on the good-bad scaleof the semantic differential. The subjects were theninstructed that "this is an experiment concerningthe effectiveness of repetition in learning to pro-nounce strange words correctly." Some of thesewords were shown once, others twice, 5 times, or 10tunes. Subjects were required to look at these wordsand to pronounce them on each presentation. Fol-lowing this training procedure the words were againrated on the good-bad scale. A significant exposureeffect was obtained, with the words shown fre-quently increasing on the evaluative scale. Strangely,however, words which were seen only once in train-ing were judged afterwards not quite as "good" asbefore training. Thus, as a result of 2, 5, and 10exposures words improved in meaning, and as aresult of but 1 exposure they deteriorated. Thisfinding, however, may be an artifact of the before-after procedure used by Johnson, Thomson, andFrincke. Moreover, frequencies and stimuli werefully confounded in their study.

Our experiment used the same stimuli which, inci-dentally, came from the familiar experiment bySolomon and Postman (19S2) on the effects of wordfrequency on recognition threshold, but our designdiffered from the one used by Johnson, Thomson,and Frincke in several respects. In the Johnson-Thomson-Frincke experiment the same words alwaysappeared in the same frequencies to all subjects.Thus, the word "jandara," for instance, was given 10times to each subject, and the word "mecburi" wasgiven once to each subject. It is possible that theeffects these authors obtained are not due to the

14 ROBERT B. ZAJONC

frequency manipulation alone, but that they dependon the stimulus material with which the frequencyvariable was fully confounded. In our study wordsand training frequencies were, therefore, counter-balanced in a Latin-square design. Because wordsand the number of exposures were counterbalanced,an after-only design could be employed, requiringno premeasures. The effects of repeated exposurecould be observed for each word by comparing thefavorability rating it received after having beenexposed during training once, twice, five times, etc.Eliminating premeasures also eliminated for eachstimulus one full exposure that necessarily precededand therefore accompanied the frequency manipula-tion.

The present experiment differs from that of John-son, Thomson, and Frincke (1960) in several otherrespects which are less critical for the interpretationof results. The procedure of this experiment, there-fore, is described in some detail. Except for somespecific changes, the same general methodology isfollowed throughout this series of studies.

Twelve seven-letter "Turkish" words, shown inFigure 3, were counterbalanced against six frequen-cies (0, 1, 2, 5, 10, and 25) in six replications of theexperiment. Seventy-two subjects were run, one ata time, 12 subjects in each replication. The initialinstructions informed the subject that the experi-ment dealt with "pronouncing foreign words." Hewas told that he would be shown some foreign words,hear the experimenter pronounce them, and that hewould be required to pronounce them himself. Thewords were typed on 3 X S-inch cards. On eachtrial a card was shown to the subject for approxi-mately 2 seconds. Simultaneously the experimenterpronounced the word, requiring the subject to fol-low him. Since each frequency class contained twoword-stimuli, there were 86 trials altogether. Theposition of a given stimulus in the sequence of these86 trials was determined at random. Following theabove frequency training subjects were told thatthe words they had just learned to pronounce werein fact Turkish adjectives, and that their next taskwould be to guess what they meant. The experi-menter told the subject that he realized how nearlyimpossible this task was, and he therefore did notrequire him to guess the word meanings exactly.Instead, it would suffice if the subject indicated ona 7-point (0 to 6) good-bad scale whether each wordmeant something good or something bad and towhat extent, because these Turkish adjectives allmeant something good or bad. These ratings weremade of the 10 stimuli which the subject receivedduring the frequency training, and of 2 additionalones previously never seen by him.

The results of the experiment are shown in Figure2 and in Figure 3. In Figure 2 are shown the rat-ings of "goodness" averaged for each of the sixfrequencies, and plotted on a log scale. Each pointon that curve is based on 144 observations, and itis clear that a strong exposure effect was obtained(F = 5.64; d/= 5/355; /><.001). Figure 3 showsthe exposure effect for each of the 12 words sepa-

rately. The ratings of "goodness" were averaged foreach word when it was given during training withthe lower frequencies of 0, 1, and 2 (hatched bars),and when it was given with the higher frequenciesof S, 10, and 25 (solid bars). This was possiblebecause each word was used in each frequencyequally often but for different subjects. It is evidentfrom Figure 3 that some words are rated as havingmore positive meaning than others, and this effect isindeed significant (F = 8.35; # = 11/781; p<.001).Apparently, some of these words "sound better" thanothers. But independently of word content, subjectsconsistently rated the given word to mean some-thing "better" if they had seen it (and had said it)more often. This is true for all the 12 words used inthe experiment, a result that has a chance likelihoodequal to .00024.

Experiment II. Since the hypothesis proposedabove holds that it is mere exposure that is a suffi-cient condition of attitude change, the procedureused in Experiment I is not optimal for testing itsvalidity. Subjects in Experiment I were required topronounce the nonsense words during training, andit is possible that a decrease in difficulty in pro-nouncing the words associated with successive presen-tations was responsible for the results. In other words,subjects rated the frequent stimuli more favorablybecause they found them easier to pronounce thanstimuli which they saw and pronounced only once ortwice. And, there were stimuli which they neverpronounced and, in fact, did not really know howto pronounce. Wilson and Becknell (1961) suggestedthat the evaluative ratings of nonsense syllables ofhigh association value are higher than of low associ-

01]

CHINESE CHARACTERS -

NONSENSE WORDS

2 5

FREQUENCY

FIG. 2. Average rated affective connotation ofnonsense words and Chinese-like characters as afunction of frequency of exposure.

THE ATTITUDINAL EFFECTS OF MERE EXPOSURE IS

AFWORBU

NANSOMA

ENANWAL.

ZABULON

LOKANTA

LOW FREQUENCY

HIGH FREQUENCY

1 2 3 4 5

RATED "GOODNESS" OF MEANING

FIG. 3. Average rated affective connotation ofnonsense words exposed with low and high frequen-cies.

ation value because they are easier to pronounce. Inorder to follow up their suggestion a group of 22University of Michigan subjects were given theabove "Turkish adjectives" with the instructions torate them according to "how easy or difficult it is topronounce" them. Using a 7-point scale, a signifi-cant item-effect was revealed by an analysis of vari-ance (F= 14.28; d f = 11/263; p<.00l), showingthat there are indeed differences among the nonsensewords in the ease with which they can be pro-nounced upon their first presentation. The Wilson-Becknell conjecture is supported by a correlation of.46 between the average ease of pronouncing thewords and the evaluative scores obtained in Experi-ment I. These latter scores were obtained byaveraging for each word the rating it obtained in allfrequencies.

These results, however, in themselves do notpreclude a relationship between exposure and evalua-tive rating. With ease of pronouncing held constantthe exposure effect may still be obtained. This ex-pectation is strengthened by the results of a studydescribed above (Frincke & Johnson, 1960) in whichhomophone pairs differing in word-frequency wererated for "pleasantness." Since homophones do notdiffer in pronounciation, the obtained frequencyeffects show that ease of pronouncing may be a

sufficient factor in affecting evaluative ratings butnot a necessary one. In order to eliminate the pro-nounciation factor and to reduce the subjects' activeparticipation while exposure is being manipulated,the following experiment was carried out.

To meet the requirements of the definition of"mere exposure," Chinese characters were substitutedfor the nonsense words. These stimuli were takenfrom Hull's (1920) concept formation study, and Iam told that not only are most of them meaning-less, but that they are also far from the absolutelyminimal standards of Chinese caligraphy. Neverthe-less, they were quite adequate for our experimentalpurposes. The subjects were again told that theexperiment dealt with the learning of a foreignlanguage, but now they were not required to pro-nounce the characters. Nor were they able to pro-nounce them subvocally. They were simply in-structed to pay close attention to the characterswhenever they were exposed to them. In all otherrespects the experiment was identical to the oneemploying nonsense words. Now, too, followingtraining subjects were told that the characters stoodfor adjectives, and that their task was to guess theirmeaning on the good-bad scale. Characters and ex-posures were again counterbalanced. Figures 2 and 4show the results, and it is obvious that the exposure-favorability relationship previously found withnonsense words obtains (F = 4.72; dj = 5/335; p <.001) even if the individual's exposure to the stimu-lus consists of his passively looking at it for aperiod of about 2 seconds. Figure 4 shows that theexposure effect is found for all stimuli but one.

The above results add strength to the hypothesisthat mere exposure is a sufficient condition for atti-tude enhancement. But again the last experiment didnot succeed in completely eliminating a learning fac-tor from the exposure manipulation, for it is possi-ble that this manipulation is now confounded withthe ease of recognition. This danger of confounding,however, is probably minimal because at no timewere the subjects ever required to recognize ordiscriminate the idiograms.

The results of Experiments I and II are in anapparent conflict with results reported by Amsterand Glasman (1966). These researchers report anegative result using a procedure similar to thatemployed by Johnson, Thomson, and Frincke (1960).The experiment was similar in all respects exceptthat meaningful English words were substituted forthe nonsense stimuli. No exposure effect was ob-served by Amster and Glasman for these meaningfulwords. But this finding is not at all surprising. Noris it especially significant for the understanding ofexposure effects. Adding one more occasion (or even10 more occasions) to see and say a perfectly well-known English word to all the times this word hadbeen seen and uttered by the individual in the past—a figure often in the thousands—really shouldn't havemuch effect on the meaning he attributes to it. Theexpectation of a change in the evaluative aspect ofmeaning as a function of a few additional exposuresbecomes even less reasonable when we consider that

16 ROBERT B. ZAJONC

LOW FREQUENCY

HIGH FREQUENCY

I 2 3 4 S

RATED "GOODNESS" OF MEANING

FIG. 4. Average rated affective connotation ofChinese-like characters exposed with low and highfrequencies.

the change in affective connotation is a linear func-tion of the logarithm of frequency, as we noted inFigures 1 and 2. If n is the frequency of the subject'spreexperimental exposure to the word, then the com-parisons made by Amster and Glasman involved thefollowing four frequencies: n + 1, n-\-2, n + S, andn + 10. Since n is large, perhaps as large as 1000, thedifferences in exposures amounted to fractions of1%.

Word-Frequency-Word-Value Relationshipas a Special Case oj the Exposure-AttitudeRelationship

In the first section of this paper some evi-dence was presented suggesting that wordswith positive affective connotations are usedmore frequently (both in print and in speech)than words with negative affective connota-tions. In the second section evidence wasgiven to suggest that the affective connota-tion of a word improves with their repeated

use. Because the second item of evidence restson experimental proof, in which the fre-quency of usage was systematically and inde-pendently manipulated, one cannot questionthe causal direction implied in these data. Butfinding that the frequency of usage affectsmeaning needn't necessarily preclude the pos-sibility that meaning determines the frequencyof usage. It is necessary, therefore, to examinemore closely the results on the correlationalevidence between word-frequency and wordvalue.

Why are positive words used more fre-quently? Besides the rather wistful and un-likely explanation that there are more posi-tive than negative referents (i.e., we live in aparadise), one real possibility suggests itself.The evidence reviewed so far deals only withusage per word. The totality of "good" and"bad" usage, however, depends on the num-bers of different "good" and "bad" words inthe language. It is entirely possible, therefore,that the superiority of "good" words in fre-quency per word exists side by side with thesuperiority of "bad" words in their greatervariety. This possibility receives some supportfrom the fact that in English (and in a hostof other languages) prefixes and suffixes thatserve to negate or reverse meaning, such asanti, de, im, in, ir, less, un, etc., are mostcommonly attached to words having a posi-tive connotation. Once attached to a wordthey almost universally form a word with anegative affective connotation. Positive wordswith these prefixes or suffixes are exceptional:unselfish, independent, are some examples.

It would appear, therefore, that there areindeed more negative than positive words.And if there are more different negativewords, the usage per word would naturally beattenuated for these words, because the totalusage would be distributed among a largeruniverse of items.

Norman ° has asked a group of students toseparate a large sample of adjectives into"good" ones and "bad" ones. On the average2.31 more items were placed in the "bad" pilethan in the "good" pile. The frequency fig-ures in Table 1 above show a pattern con-sistent with Norman's independent finding.

0 Warren T. Norman, personal communication,1965.


The average frequency of the preferred anto-nyms is 2.3 times larger than the average fre-quency of the nonpreferred antonyms! There-fore, for the material considered here, theratio of total positive and negative usage isequal to unity.

If repeated usage enhances the affectivemeaning of words, a relatively large supply ofnegative words would in fact be needed. Itwould be equally reasonable to expect thatthere exist devices in language protectingwords from a deterioration of meaning. It isentirely possible that the prefixes and suf-fixes discussed above serve this function. Be-cause the negative qualities of these prefixesand suffixes are independent of their referents,because they are essentially abstract, and be-cause they derive their negativity from thesemantic function they perform, words formedby means of these prefixes and suffixes areperhaps better able than root words to resistan enhancement of affective connotation as aresult of repeated usage. I was unable to findevidence corroborating this point of view,although there is a good deal of philologicalliterature on both positive changes in meaning(see for instance vanDongen, 1933) and neg-ative changes in meaning (see, for instance,Schreuder, 1929). Most of the sources, how-ever, consider changes in meaning of rootwords only.

If there are many remaining doubts thatfrequency of words is a function of the valueof their referents, then the following frequen-cies of a few well-chosen but significant wordsshould once and for all dispel them:

PsychologistChemistEconomistSociologistAstronomerGeologistPhysicistGeographerBotanistBiologist

36323214129876S

Exposure-Attitude Relationship: ExperimentalEvidence

Experiment HI. In all the experiments above thequestion asked of the subjects in rating the stimuli

following exposure dealt with the evaluative aspectof their meaning. The subjects were never requiredto say just how much they "liked" the nonsensewords or "Chinese" characters. In all probability, theresults would have been the same if they were askeddirectly to state their attitude toward these wordsand characters, and the Wilson-Becknell (1961)results support this conjecture. But because theirstimuli were essentially verbal, subjects' answerscould in these studies be strongly influenced by se-mantic factors. This would have been less likely, ofcourse, in the case of Chinese characters than in thecase of nonsense words.

As was pointed out above, there is some directevidence on the attitudinal effects of mere exposure,dealing almost exclusively with music appreciation.Meyer (1Q03), for example, played to his studentsoriental music 12 to IS times in succession. In mostcases the students' introspective protocols indicateda better liking for the pieces on the last than onthe first presentation. One of the students who tookpart in Meyers' experiment (H. T. Moore), andwho showed enhancement effects of repeated ex-posure ("I liked the last time better than the first,because I became more used to the successivechords"), followed up this work in a study of hisown 20 years later. Moore and Gilliland (1924)played to their students jazz and classical recordsonce a week for 25 weeks. Liking for classical rec-ords increased, but no change was found for jazzmusic. Similar results are reported by other writers(Krugman, 1943; Verveer, Barry, & Bousfield, 1933;Washburn, Child, & Abel, 1927). Downey andKnapp (1927) played to 33 students a variety ofmusical selections (e.g., Tschaikowsky's MarcheSlave, Massenet's Meditation from "Thais," Colum-bia, The Gem of the Ocean, etc.) once a week forfive weeks. All pieces of music except one (Colum-bia, The Gem of the Ocean) became better liked atthe close of the sessions. Alpert (1953) presentedsubjects with sounds having unfamiliar rhythms. Hissubjects found these sounds at first unpleasant.After repeated presentations, however, the liking forthem increased. Additional exposures of subjects tothe tones resulted in increasing indifference on thepart of the listeners. More recently, Mull (1957)found that upon repeated exposure to their musicsubjects enjoyed Schoenberg and Hindemith more.

In the area of visual arts, Pepper (1919) foundthat repeated exposure resulted in more positiveesthetic judgments of unusual color combinations.Krugman and Hartley (1960), however, usingfamous paintings, could only find ambiguous re-sults. Maslow (1937) projected for 4 days in suc-cession 15 paintings of great masters. Six days fol-lowing the last presentation the 15 paintings werepresented once again, and interspersed among themwere 15 others (matched for the artist) which thesubjects had never seen. The results indicated agreater liking for the familiar paintings. Maslow(1937) also made tests of preference, frequentlywith similar results, for other familiar and unfa-miliar objects, such as rubber bands, paper clips,

18 ROBERT B. ZAJONC

blotters, pens, pencils, etc. A similar experiment tothe one with paintings, but using instead Russiangirls' names, showed the same results. The samesubjects were used in all these studies and the ses-sions took place in the same room, the subjects al-ways sitting in the same chairs. Toward the end ofthe testing program Maslow asked if anyone wouldlike to change seats. No one did, preferring, appar-ently, to remain in the familiar one.

Although the results of the above studies arefairly consistent, the conditions under which theywere carried out make their conclusions somewhatless than compelling. In the majority of instances,the circumstances of the repeated exposure werequite ambiguous. The experiments were usuallyconducted in classes, the instructor serving as theexperimenter. Subjects often responded aloud, thusbeing able to influence each other's judgments andopinions. Prior to the sessions the experimenteroften expressed his own preferences. The stimuli,repeatedly shown, were not always exposed underthe same conditions, and the material, exposures, andsequences were seldom counterbalanced. But in allof these experiments a pattern of results emergesshowing that the frequency manipulation has morepronounced attitude effects for stimuli that are novel,unfamiliar, or unusual than for familiar stimuli.This pattern is, of course, consistent with the ob-servation that attitude enhancement is a function ofthe logarithm of frequency.

Becknell, Wilson, and Baird (1963) have recentlyreported more convincing support for the exposure-attitude hypothesis. Slides of nonsense syllables werepresented with different frequencies (1, 4, 7, and 10).Following this exposure training (which also includedinterspersed presentations of slides with landscapesand with ads) female subjects were given pairs ofboxes containing nylon stockings, and they wereasked to choose the "brand" they preferred. These"brands" corresponded to the nonsense syllables pre-viously shown, and they were printed on the boxes.Each subject received two different pairs of boxesfor comparison. The paired-comparison data showeda tendency of subjects to prefer the box marked bythe more frequent syllable. Again, however, thesemantic component is not excluded from the effectsobtained in these two studies.

There is one more item of evidence, somewhatindirect, on the problem of the effects of exposure.In a study by Munsinger (1964) subjects were giventhe opportunity to present to themselves CVC tri-grams whose association value, evaluation scalevalue, and prepotency score (Handler, 19SS) werepreviously assessed. By pressing a response key thesubject would expose in a small window a trigramwhich he would then have to spell. The rate atwhich he key-pressed constituted the dependentmeasure. In one of Munsinger's experimental groupssubjects could expose to themselves, by means ofthat key response, trigrams that were matched forassociation and prepotency. All these trigrams, how-ever, previously scored low on the evaluative scalesof the semantic differential. After subjects reached

t 3.5

J3 .0

2.5 -

I 2 5FREQUENCY

10 25

FIG. 5, Average attitude toward photographs as afunction of frequency of exposure.

an asymptotic key-pressing rate, the experimentalconditions changed such that now the subjects' re-sponse would expose trigrams that were high inevaluation, although they were still matched forassociation and prepotency. A significant increase inkey-pressing rates is reported by Munsinger follow-ing the change in the affective value of the trigrams.Again, however, the semantic component is not en-tirely excluded from the effects obtained in thesetwo studies.

Because they are less a matter of semantic factors,we have chosen to manipulate interpersonal attitudesby means of exposure. Using the same experimentaldesign as with the Chinese characters, faces of men(photographs of graduating Michigan State Uni-versity seniors taken from the MSU Yearbook)were employed as attitude objects. The experimentwas introduced to subjects—all students at theUniversity of Michigan—as dealing with the prob-lem of "visual memory." Following the exposuremanipulation, which consisted of presenting eachphotograph a different number of times for a periodof 2 seconds, subjects were asked to rate on a 7-pointscale how much, they might like the man on eachphotograph. The results of this study are shown inFigures S and 6. While the exposure effect is not asclear as previously (only 9 of the 12 stimuli showit), it is still rather impressive (F = 9.96; dl = 5/355; p< .001).

The Exposure-attitude Hypothesis andRelated Theoretical Issues

The above results raise a series of empiri-cal and theoretical questions. Are all attitudes


*-~ *% & $ i •* 4* ?^ ^&*» Vfti x arfL \SfA

FIG. 6. Average attitude toward photographs exposed with low and high frequencies.

enhanced by mere repeated exposure? Isthere a number of repetitions beyond whichattitude begins to become negative? Does thisnumber vary systematically across attitudeobjects? Are these effects stable? These andsimilar questions can only be answered byfurther empirical work. On a theoretical levelthese questions address themselves primarilyto those psychological processes that mediateexposure effects.

Let us first consider a possible biologicalsignificance of an exposure-enhancementmechanism. A stimulus presented for the firsttime evokes in the organism an instinctivefear reaction. Lorenz (1956) noted that ayoung raven,

confronted with a new object, which may be acamera, an old bottle, a stuffed polecat, or any-thing else, first reacts with escape responses. He willfly up to an elevated perch, and, from this point ofvantage, stare at the object literally for hours. Afterthis he will begin to approach the object verygradually . . .

Biihler, Hetzer, and Mabel (1928) observedthat human infants reacted to a strange soundby crying out with fear. Upon the second ex-posure of the sound stimulus, movement andvocalization that indicated displeasure wereobserved. On the third exposure, the infantslistened to the sound showing some signs ofattention, but did not seem to show any dis-pleasure. On the fourth exposure, they lookedin the direction of the sound with detectable

interest. These facts, of course, are borne outby common observation. Hunt (1965) re-ported that young infants he observed pre-ferred a familiar mobile to a new one. Andthe "Black Bag" story cited in the introduc-tion represents another example of phenomenain this category. At the outset the "BlackBag," in fact, attracted a good deal of hos-tility. Cairns (1966) has recently presented avery convincing argument that the affiliativebehavior and social attachments among ani-mals are solely determined by the animals'exposure to one another. Examining evidenceon affiliative preferences of animals observedunder conditions of inter- and intraspecific co-habitation and of animals deprived of socialcontact, Cairns concluded that such affiliativepreferences vary directly with the length ofthe association and with the importance ofthe cues which are generated in the course ofthe association. Cairns, moreover, did notlimit his conclusion to inter-animal social at-tachments but prposed that "animals tend toremain in the presence of [any] objects towhich they have been continually exposed[p. 409]."

The survival value of an avoidance reflexto a novel stimulus is obvious. But there isno direct evidence that all organisms areequipped with an avoidance reaction occurringupon the encounter of a novel stimulus. How-ever, if we assume that they are, then theexposure-attitude hypothesis becomes more

20 ROBERT B. ZAJONC

reasonable. The first encounter with the novelstimulus produces fear reaction. If no nega-tive consequences are associated with thisfirst encounter, the avoidance reaction uponthe second encounter will naturally be weaker.If such encounters continue, and if no otherevents—negative in their consequences forthe organism—accompany these encounters,then the organism's attitude toward thestimulus must improve. To be sure, the hy-pothesis does not deny or preclude the effectsof reinforcement. The exposure of a stimuluscoupled with reward will strengthen the ani-mal's approach behavior; and the exposureof stimulus coupled with a noxious event willstrengthen his avoidance reactions. But inthe absence of reward or punishment, mereexposure will result in the enhancement ofthe organism's attitude toward the givenstimulus object.

If novel stimuli evoke fear, conflict, or un-certainty, one should be able to detect thesestates, and to observe their dissipation uponrepeated exposure. To the extent that GSRmeasures arousal that is associated with theabove states, we would expect greater GSRsupon the presentation of novel stimuli thanupon the presentation of familiar ones, andwe would also expect a drop in GSR reactiv-ity to be the consequence of repeated stimu-lus exposure.

Experiment IV. Changes in affective arousal thatoccur as a result of repeated exposure of a novel

stimulus were examined in an independent experi-ment. Fifteen subjects were presented with nonsensewords (the same as in Experiment I) in a series of86 trials. Two words appeared 25 times, two 10 times,two 5 times, two twice, and two once. The positionof a word in the series of trials was determined bya random device. Words were counterbalancedagainst frequencies in three experimental replica-tions. Due to a mechanical failure of the apparatus,data for one subject could not be used, and werenot included in the analysis.

On a given trial, the stimulus word was projectedonto a screen for a period of 2 seconds, and thesubject's GSR was recorded. The interstimulus in-terval was 20 seconds. GSR was measured onlyduring the first 10 seconds following stimulus onset,and all GSRs occurring during the last 12 secondsof the interstimulus interval were treated as arti-facts. The Kaplan-Hobart technique of GSR meas-urement (Kaplan & Hobart, 1964), which requirescurrent of only 10 microamperes, was employed.Zinc-zinc sulphate electrodes (Kaplan & Fisher, 1964;Lykken, 1959) were applied to the forefinger andthe middle finger of the subject's nonpreferred hand.

The results of the experiment are shown in Figure7 and Figure 8. In Figure 7 changes in conductanceare plotted for each successive presentation of thestimulus-word. For purposes of clarity data for onlyone stimulus are graphed in each frequency class.The results for the other set of stimuli are thesame. It can be seen that, in general, successivepresentations result in a lower autonomic reactiv-ity. After about seven or eight exposures a stableasymptote is reached. Hence, only stimuli shown 25and 10 times attain an asymptote. Words shownfive times, twice, or once generate greater changesin conductance even on their last exposure. Thiseffect is seen better in Figure 8 in which GSRs onthe last exposure of the stimuli were plotted. Asprior frequency of exposure increases there is a

.35

30

°.25

20

.15

.10

5.05Llj5

—• 25 Exposures. .o 10 Exposures—» 5 Exposures

—..7 Z ExposuresI Exposures

5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85

FIG. 7. GSR obtained upon repeated exposures of nonsense words.


40

1 2 5 10 2 5FREQUENCY OF PAST EXPOSURE (LOG SCALE)

FIG. 8. GSR obtained on the last exposure of non-sense words exposed with various frequencies.

lesser change in conductance upon stimulus presenta-tion. This effect is significant at the .OS level (F =4.01; df = 4/117).

These results cannot be due to an overall adapta-tion that may be occurring as the series of the 86trials progresses. It is clear that the stimulus shownjust once generated a substantial GSR although itoccurred as late as on the 67th trial. As a matter offact, its GSR does not differ from the GSR obtainedfor the first presentation of the other stimuli. Thelast presentation of the stimulus word that had justone prior exposure also resulted in a substantialGSR although it occurred on the 64th trial. And onthe whole, although there is variability over trials, atany one point during the series GSR's are higher forstimuli with infrequent prior exposures than forstimuli previously seen by the subject many times.It seems, therefore, that with increased exposurethere is a genuine reduction in stimulus-evokedarousal. These findings are in agreement with thosereported by Berlyne, Craw, Salapatek, and Lewis(1963). In their study subjects were shown visualpatterns differing in complexity and in incongruity.The patterns were presented for 3 seconds on threesuccessive occasions. A significant drop in GSR wasobtained between the first and the second presenta-tion. But no significant GSR effects were associatedwith complexity or incongruity.

DISCUSSION

While the bulk of the results presented andreviewed in this monograph supports thehypothesis that repeated exposure is a suffi-cient condition of attitude enhancement, thereare findings and theoretical formulations which

appear to be in conflict with the hypothesis.The most pronounced source of ostensiblycontradictory results is in the area of ex-ploration and curiosity. There is impressiveevidence today that in a free situation thesubject (human or animal) will turn towarda novel stimulus in preference to a familiarone (e.g., Berlyne, 1960). If such orientingand exploratory "approach" behavior is asymptom of a favorable attitude toward thestimulus object, the wealth of data on ex-ploration and orienting behavior (Berlyne,1950, 19SS; Berlyne & Slater, 1957; Dember&Milbrook, 1956; Montgomery, 1953; Thies-sen & McGaugh, 1958; Thomson & Solomon,1954) stands in clear contradiction to thosereported in this monograph.

But there is at present no direct evidenceto support the above assumption. And, on thecontrary, it is more likely that orienting to-ward a novel stimulus in preference to afamiliar one may indicate that it is less likedrather than it is better liked. Ordinarily,when confronted with a novel stimulus theanimal's orienting response enables it to dis-cover if the novel stimulus constitutes asource of danger. It need not explore familiarstimuli in this respect. Novelty is thus com-monly associated with uncertainty and withconflict—states that are more likely to pro-duce negative than positive affect. Most recentwork by Harrison (1967) indicates quiteclearly that exploration and favorable atti-tudes are in fact negatively related. Usingnonsense words, Chinese characters, andphotographs of men's faces, Harrison ob-tained measures of liking from one group ofsubjects and measures of exploration from an-other group. The correlations between ex-ploration and liking were —.69, —.69, and— .60 for nonsense words, characters, andphotographs, respectively. If the function oforienting behavior is eventually to change thenovel stimulus into a familiar one, it is also itsconsequence to render the stimulus objecteventually more attractive (or perhaps merelyless repulsive).

In his research Harrison also obtained dataon the behavioral consequences of exposure,and in particular, on response conflict thatnovelty seems to arouse. It is a truism, of

22 ROBERT B. ZAJONC

course, to assert that novel stimulus is one towhich no specific response, beside orienting,has as yet been attached. But the novel stim-ulus cannot fail being similar to an entire hostof other stimuli that the individual had en-countered in the past, and to which he hadattached specific responses. And it is entirelylikely that some of these generalized responsetendencies that the novel stimulus simul-taneously excites are mutually incompatible;that is, they cannot all be emitted at the sametime. This antecedent condition is what makesresponse conflict (of some, however small,magnitude) a necessary concomitant of nov-elty.

Using latency of free associations as ameasure of response conflict, Harrison wasable to demonstrate that response conflict ismarkedly reduced upon repeated exposure ofa novel stimulus. Chinese ideographs wereshown to a group of subjects once, twice, Stimes, 10 times, and 25 times, in a mannersimilar to that used in other experiments de-scribed above. Following this exposure manip-ulation, Harrison presented each stimulusonce again requiring the subject to respond"with the first thing that came to mind." In-cluded now were also Chinese ideographs thatthe subject had never seen. The latency ofthese free associations was obtained on eachsuch trial, and the results revealed a sys-tematic and significant drop in latencies as afunction of the frequency of prior exposure.

Another set of data which may also be ofsome consequence for the exposure-attitudehypothesis is to be found in the area of se-mantic satiation. In a typical semantic-satia-tion experiment the subject is asked to re-peat words, two or three per second, for aperiod of IS seconds. The general findings inthis area indicate that following this sort ofrapid repetition the word seems to "lose" itsmeaning (for a review of the literature seeAmster, 1964). Loss of meaning is measuredby a departure from polarity on semanticdifferential scales, such as good-bad, strong-weak, etc. (Lambert & Jakobovits, 1960).When repeated in rapid succession and ratedon some semantic differential scale, immedi-ately thereafter the words tend to be placedneither toward one (e.g., good) nor the other

(e.g., bad) end of the scale, but are ratedtoward the neutral point of the scale. Whileseveral studies have demonstrated a reduc-tion of polarization following rapid repetitionof a word (Das, 1964; Kanungo & Lambert,1963a, 1963b; Messer, Jakobovitz, Kanungo,& Lambert, 1964), there is an equal amountof conflicting evidence (Amster & Glasman,1965; Floyd, 1962; Reynierse & Barch, 1963;Schulz, Weaver, & Radtke, 1965). Yelen andSchulz (1963) attribute satiation findings toa regression artifact. A reduction of polarityof positive words as a result of repetitionwould indeed be embarrassing for the ex-posure-attitude hypothesis. A reduction of po-larity of negative words (i.e., words with asemantic score below the neutral point)would, of course, be entirely in agreement withthe present results, for they would simply beshowing an enhancement effect along theevaluative dimension. It should be noted thatall our stimuli initially received negative rat-ings, that is, below the neutral point, 3, onthe 0-6 good-bad scale. Given no exposure atall, the evaluative ratings for the Turkishnonsense words, Chinese-like characters, andphotographs were 2.56, 2.67, and 2.79, re-spectively. In terms of polarity these averagesare —.44, —.33, and —.21. Contrary to thesemantic satiation hypothesis, absolute po-larities increased after 25 exposures, for theywere +.61, +.78, and +.61, respectively.Admittedly, the controversy within semanticsatiation literature may have to be resolvedbefore clear implications for the exposureeffect can be drawn. Parenthetically it shouldbe noted, however, that the form of exposureused in the semantic satiation paradigm (i.e.,30 to 45 repetitions in 15 seconds) is not whathas been above defined as the sufficient condi-tion of attitude enhancement. Moreover, insemantic satiation studies the stimulus is com-monly a verbal response made by the subjecthimself. Whether such a response-producedstimulus should constitute what is meant bymere exposure is a matter of some doubt.

The above hypothesis and data seem to beconsistent with the theory of reinforcementrecently proposed by Premack (1959). Thisnew and engaging approach to reinforcementeffects, for which a good deal of impressive


evidence has already been accumulated (Pre-mack, 1961, 1962, 1963, 1965; Premack &Bahwell, 1959; Premack & Collier, 1962),holds that if the emission of one response, A,is made contingent upon the emission ofanother response, B, and if A occurs withgreater frequency than B, B will gain in therate of emission. A can, therefore, be consid-ered as having positive reinforcement valuewhich seems to depend alone upon its fre-quency of occurrence. Contingencies betweenresponses can be introduced by controllingmanipulanda or the availability of goal ob-jects. Thus, for instance, in studying theseeffects in Cebus monkeys, Premack (1963)used four manipulanda: a plunger, a hingeddoor, a vertically operated lever, and a hori-zontally operated lever. It was possible tomake any manipulandum inoperable at anytime, and to make it operable only in the caseof a prior manipulation of another manipu-landum. Thus, each manipulandum couldserve for the reinforcwg or for the reinforcedresponse. After establishing for each monkeythe probabilities of operating each item underfree access to all, contingencies were arrangedbetween pairs of items, such that, for in-stance, the vertical lever could not be op-erated unless the animal pulled the plunger,or the hinged door remained locked unlessthe subject pressed the horizontal lever, etc.The introduction of these contingencies re-sulted in the predicted effects. In general,responses that were less probable increased inthe rate of emission when more probableresponses were made contingent upon them.When two responses were equal in probabilityof emission little or no change in responserates was observed.

Viewed in the present context, the indi-vidual's response probabilities can be takenas an indication of his "attitudinal liking"or "attraction" to the goal objects of theseresponses, or to the instrumental stimuli as-sociated with them. To the extent that thereward value of a given response, A, which iscontingent upon another response, B, is adirect function of A's probability of emission,then increasing its probability of emission willincrease its reward value. Premack (1961)was able to demonstrate such an effect in rats

for licking and bar pressing. Changing thelanguage somewhat, it may be said that theindividual's "attitudinal liking" for the goalobject of the response A increases with theindividual's exposure to the goal object of A.In Premack's work continuous reinforcementschedules are used and, hence, the individual'sfrequency of exposure to the goal object of Ais equal to the frequency of emission of A.

There are important differences betweenthe phenomena observed by Premack andthose reported in the present paper. In mostof his experiments the frequency of occurrenceof responses results from an independent, per-haps genetically given or previously acquired,response preference of the subject. In the ex-perimental work on attitudinal effects of ex-posure, the frequency of occurrence of thestimulus-object is deliberately manipulated.Moreover, the subject usually does not havea prior preference for the stimulus exposed.On the contrary, he may often manifestavoidance tendencies. But the parallel is com-pelling.

While there is a great deal in common be-tween Premack's work on reinforcement andthe research on attitudinal effects of exposure,a clear understanding of the implications ofone for the other requires a systematic de-termination of what "reinforcement value"and "attitudes" have in common that makesthem both equally vulnerable to simple fre-quency effects.

CONCLUSION

The balance of the experimental results re-viewed and reported in this paper is in favorof the hypothesis that mere repeated exposureof an individual to a stimulus object en-hances his attitude toward it. But, as yet,the account books cannot be closed. Furtherresearch must examine the boundary condi-tions of the exposure-attitude relationship, forit is possible that the neat linear log-fre-quency-attitude relationship, repeatedly ob-served here, may well break down under someconditions. This future research must, in par-ticular, concentrate on the effects of largefrequencies of exposure, on duration of ex-posure, on interexposure intervals, and on

24 ROBERT B. ZAJONC

many other similar parameters of mere ex-posure. This research must also assess theapplicability of the exposure-attitude rela-tionship to a greater variety of stimulus ob-jects. The question of generalization of spe-cific exposure effects is of equal theoreticalimportance. Does repeated exposure to agiven stimulus result in the enhancement ofattitudes toward similar or related stimuli?

Because the above effects seem to be a func-tion of the logarithm of frequency of expo-sure, they are more apparent and more pro-nounced for differences among small frequen-cies than for differences among large fre-quencies. For the same reason, attitudinal en-hancement produced by means of exposurewill be more readily effected for novel objectsthan for familiar ones. It is likely that ex-posure effects for very familiar objects areabsent completely or are so small that theycannot be detected at all by methods nowavailable. As we have seen above, Amster andGlasman (1966) failed to obtain the exposureeffect for common English words. It will beimportant for future research, therefore, todetermine the range of familiarity for whichthe exposure effect is obtained.

Mere exposure is a necessary preconditionof a vast variety of experimental manipula-tions. For example, in attempts to changeattitudes by means of persuasive communica-tions the attitude object is mentioned repeat-edly, regardless of whether the attempt is di-rected toward making the attitude more favor-able or toward making it less favorable. Mak-ing attitudes more favorable should, there-fore, be easier than making them less favor-able. It is interesting that studies on theeffectiveness of persuasion in attitude changeseldom try to effect a negative change, andalmost never compare the relative success ofa pro-persuasion with the success of a con-persuasion. In an attitude-change study Tan-nenbaum and Gengel (1966) have recentlyobtained only positive shifts, although botha positive as well as a negative manipulationwere employed.

The partial reinforcement manipulation,too, is subject to possible confounding withthe number of stimulus presentations. Erle-bacher and Archer (1961), for instance, re-ported the curious result that at the comple-

tion of training greater numbers of correctresponses were associated with smaller per-centages of reinforcement. However, in thevarious conditions of reinforcement subjectsworked until they performed in succession apredetermined number of correct responses,the same for all percentages of reinforcement.Therefore, percentage of reinforcement was inthis study completely confounded with thenumber of stimulus exposures (and also withthe number of reinforcements). Althoughmany authors have tried to cope with thisconfounding in one way or another (e.g.,Festinger, 1961; Kanfer, 1954; O'Connell,196S), the methodological difficulties have notbeen completely overcome. None of the fourvariables that are associated with the partialreinforcement effect-—percentage of reinforcedtrials, number of trials, number of positivereinforcements, number of nonreinforcements—can be studied independently of the others.

REFERENCES

AIXUISI, E. A., & ADAMS, O. S. Predicting letterpreferences: Aesthetics and filtering in man. Per-ceptual and Motor Skills, 1962, 14, 123-131.

ALPERT, R. Perceptual determinants of affect. Un-published master's thesis, Wesleyan University,1953.

AMSTER, H. Semantic satiation and generation:Learning? Adaptation? Psychological Bulletin, 1964,62, 273-286.

AMSTER, H., & GLASMAN, L. D. Verbal repetitionand connotative change. Journal of ExperimentalPsychology, 1966, 71, 389-395.

ANDERSON, N. H. Likeableness ratings of 555 person-ality-trait adjectives. Los Angeles: University ofCalifornia, 1964. (Mimeo)

BECKNEIX, J. C., JR., WILSON, W. R., & BAIRD, J. C.The effect of frequency of presentation on thechoice of nonsense syllables. Journal of Psychology,1963, 56, 165-170.

BERLYNE, D. E. Novelty and curiosity as determi-nants of exploratory behavior. British Journal ofPsychology, 1950, 41, 68-80.

BERLYNE, D. E. The arousal and satiation of per-ceptual curiosity in the rat. Journal of Compara-tive and Physiological Psychology, 1955, 48, 232-246.

BERLYNE, D. E. Conflict, curiosity, and exploratorybehavior. New York: McGraw-Hill, 1960.

BERLYNE, D. E., CRAW, M. A., SALAPATEK, P. H., &LEWIS, J. L. Novelty, complexity, incongruity, ex-trinsic motivation, and the GSR. Journal of Ex-perimental Psychology, 1963, 66, 560-567.

BERLYNE, D. E., & SLATER, J. Perceptual curiosity,exploratory behavior, and maze learning. Journal


of Comparative and Physiological Psychology, 1957,50, 228-232.

BOGAEDUS, E. S. Measuring social distance. Journalof Applied Sociology, 1925, 9, 299-308.

BOGARDUS, E. S. Social distance. Yellow Springs,Ohio: Antioch Press, 1959.

BOVAED, E, W., JR. Group structure and perception.Journal of Abnormal and Social Psychology, 1951,46, 398-405.

BRAUN, J. R. Three tests of the McClelland dis-crepancy hypothesis. Psychological Reports, 1962,10, 271-274.

BUCHANAN, M. A. Graded Spanish word book. To-ronto: Toronto University Press, 1927.

BIJHLER, C., HETZER, H., & MABEL, F. Die Affekt-wirksamkeit von Fremheitseindrticken im erstenLebensjahr. Zeitschrift fur Psychologie, 1928, 107,30-49.

CAIRNS, R. B. Attachment behavior of mammals.Psychological Review, 1966, 73, 409-426.

COOK, S. W., & SELLTIZ, C. Contact and intergroupattitudes: Some theoretical considerations. NewYork: Research Center for Human Relations,1952.

DAS, J. P. Hypnosis, verbal satiation, vigilance, andpersonality factors: A correlational study. Journalof Abnormal and Social Psychology, 1964, 68, 72-78.

DEMBER, W. N., & MILLBROOK, B. A. Free choice bythe rat of the greater of two brightness changes.Psychological Reports, 1956, 2, 465-467.

DEUTSCH, M., & COLLINS, M. E. Interracial housing:A psychological evaluation of a social experiment.Minneapolis: University of Minnesota Press, 1951.

DIXON, T. R., & DIXON, J. F. The impression valueof verbs. Journal of Verbal Learning and VerbalBehavior, 1964, 3, 161-165.

DOWNEY, J. E., & KNAPP, G. E. The effect on amusical programme of familiarity and of sequenceof selections. In M. Schoen (Ed.), The effects ofmusic. New York: Harcourt, Brace, 1927.

EATON, H. S. An English, French, German, Spanishword frequency dictionary. New York: Dover,1940.

ERDELYI, M. The relation between "Radio Plugs"and sheet sales of popular music. Journal of Ap-plied Psychology, 1940, 24, 696-702.

ERLEBACHER, A., & ARCHER, E. J. Perseveration as afunction of degree of learning and percentage ofreinforcement in card sorting. Journal of Experi-mental Psychology, 1961, 62, 510-517.

FECHNER, G. T. Vorschule der Aesthetik. Leipzig:Breitkopf & Hartel, 1876.

FESTINGER, L. Group attraction and membership.Journal of Social Issues, 1951, 7, 152-163.

FESTINGER, L. The psychological effects of insufficientreward. American Psychologist, 1961, 16, 1-11.

FLOYD, R. L. Semantic satiation: Replication andtest of further implications. Psychological Reports,1962, 11, 274.

FRINCKE, G., & JOHNSON, R. C. Word value andword frequency in homophone pairs. PsychologicalReports, 1960, 7, 470,

GOUGH, H. G. Reference handbook for the GoughAdjective Check List. Berkeley: University ofCalifornia Institute of Personality Assessment andResearch, April 1955. (Mimeo)

HARRISON, A. A. Response competition and attitudechange as a function of repeated stimulus ex-posure. Unpublished doctoral dissertation, Uni-versity of Michigan, 1967.

ROMANS, G. C. Social behavior: Its elementaryforms. New York: Harcourt, Brace, 1961.

HOWES, D. On the interpretation of word frequencyas a variable affecting speed of recognition. Jour-nal of Experimental Psychology, 1954, 48, 106-112.

HOWES, D. H., & SOLOMON, R. L. A note on Mc-Ginnies' "Emotionality and perceptual defense."Psychological Review, 1950, 57, 229-234.

HULL, C. L. Quantitative aspects of the evolution ofconcepts. Psychological Monographs, 1920, 28(1,Whole No. 123).

HUNT, J. McV. Traditional personality theory in thelight of recent evidence. American Scientist, 1965,53, 80-96.

JAMES, W. The principles of psychology. Vol. 2.New York: Holt, 1890.

JOHNSON, R. C., THOMSON, C. W., & FRINCKE, G.Word values, word frequency, and visual durationthresholds. Psychological Review, 1960, 67, 332-342.

KADING, F. W. Haufigkeitsworterbuch der deutschenSprache. Berlin: Mittler, 1948.

KANFER, F. H. The effect of partial reinforcement inacquisition and extinction of a class of verbalresponses. Journal of Experimental Psychology,1954, 48, 424-432.

KANUNGO, R. N., & LAMBERT, W. E. Paired-associatelearning as a function of stimulus and responsesatiation. British Journal of Psychology, 1963, 54,135-144 (a).

KANUNGO, R. N., & LAMBERT, W. E. Semantic satia-tion and meaningfulness. American Journal ofPsychology, 1963, 76, 421-428 (b).

KAPLAN, S., & FISHER, G. R. A modified design forthe Lykken zinc electrodes. Psychophysiology,1964, 1, 88-89.

KAPLAN, S., & HOBART, J. L. A versatile device forthe measurement of skin resistance in rats andhumans. American Journal of Psychology, 1964,77, 309-310.

KRAMER, M. Residential contact as a determinant ofattitudes toward Negroes. Unpublished doctoraldissertation, University of Chicago, 1950.

KRUOMAN, H. E. Affective response to music as afunction of familiarity. Journal of Abnormal andSocial Psychology, 1943, 38, 388-392.

KRUGMAN, H. E., & HARTLEY, E. C. The learning oftastes. Public Opinion Quarterly, 1960, 24, 621-631.

LAMBERT, W. E., & JAKOBOVITS, L. A. Verbal satia-tion and changes in the intensity of meaning.Journal of Experimental Psychology, 1960, 60,376-383.

LORENZ, K. L'instinct dans le comportement de I'ani-mal et de I'homme. Paris: Masson, 1956.

26 ROBERT B. ZAJONC

LYKKEN, D. T. Properties of electrodes used inelectrodermal measurements. Journal of Compara-tive and Physiological Psychology, 19S9, 52, 629-634.

MACKENZIE, B. K. The importance of contact in de-termining attitudes toward Negroes. Journal ofAbnormal and Social Psychology, 1948, 43, 4.

HANDLER, G. Associative frequency and associativeprepotency as measures of response to nonsensesyllables. American Journal of Psychology, 19SS,68, 662-66S.

MASLOW, A. H. The influence of familiarization onpreference. Journal of Experimental Psychology,1937, 21, 162-180.

MC&NNIES, E. Emotionality and perceptual defense.Psychological Review, 1949, 56, 244-2S1.

MESSER, S., JAKOBOVITS, L. A., KANUNGO, R. N., &LAMBERT, W. A. Semantic satiation of words andnumbers. British Journal of Psychology, 1964, 55,156-163.

MEYER, M. Experimental studies in the psychologyof music. American Journal oj Psychology, 1903,14, 456-476.

MILLER, G. A., NEWMAN, E. B., & FRIEDMAN, E. A.Length-frequency statistics for written English.Information and Control, 1958, 1, 370-398.

MIRON, M. S. A cross-linguistic investigation of pho-netic symbolism. Journal of Abnormal and SocialPsychology, 1961, 62, 623-630.

MONTGOMERY, K. C. Exploratory behavior as afunction of "similarity" of stimulus situations.Journal of Comparative and Physiological Psychol-ogy, 1953, 46, 129-133.

MOORE, H. T., & GILLILAND, A. R. The immediateand long time effects of classical and popularphonograph selections. Journal of Applied Psy-chology, 1924, 8, 309-323.

MOSIER, C. I. A psychometric study of meaning.Journal of Social Psychology, 1941, 13, 123-140.

MULL, H. K. The effect of repetition upon enjoy-ment of modern music. Journal oj Psychology,1957, 43, 155-162.

MUNSINGER, H. L. Meaningful symbols as reinforcingstimuli. Journal oj Abnormal and Social Psychol-ogy, 1964, 68, 665-668.

National Opinion Research Center. Jobs and occu-pations: A popular evaluation. In R. Bendix & S.B. Lipsit (Eds.), Class, status, and power: Aresearch in social stratification. Glencoe, 111.: FreePress, 1953.

NEWCOMB, T. M. Stabilities underlying changes ininterpersonal attraction. Journal of Abnormal andSocial Psychology, 1963, 66, 376-386.

O'CoNNELL, D. C. Concept learning and verbal con-trol under partial reinforcement and subsequentreversal and non-reversal shifts. Journal of Ex-perimental Psychology, 1965, 69, 144-151.

OSGOOD, C. E., Suci, G. J., & TANNENBATJM, P. H.The measurement of meaning. Urbana: Universityof Illinois Press, 1957.

PALERMO, D. S., & JENKINS, J. J. Word associationnorms—grade school through college. Minneapolis:University of Minnesota Press, 1964.

PEPPER, S. C. Changes of appreciation for color com-binations. Psychological Review, 1919, 26, 389-396.

POSTMAN, L. The experimental analysis of motiva-tional factors in perception. In J. S. Brown (Ed.),Current theory and research in motivation. Lin-coln: University of Nebraska Press, 1953.

PREMACK, D. Toward empirical behavior laws: I.Positive reinforcement. Psychological Review, 1959,66, 219-233.

PREMACK, D. Predicting instrumental performancefrom the independent rate of the contingent re-sponse. Journal oj Experimental Psychology, 1961,61, 163-171.

PREMACK, D. Reversibility of the reinforcement re-lation. Science, 1962, 136, 255-257.

PREMACK, D. Rate differential reinforcement in mon-key manipulation. Journal of the ExperimentalAnalysis oj Behavior, 1963, 6, 81-89.

PREMACK, D. Reinforcement theory. Nebraska Sym-posium on motivation, 1965, 13, 123-180.

PREMACK, D., & BAI-IWELL, R. Operant-level leverpressing by a monkey as a function of interestinterval. Journal of the Experimental Analysis ofBehavior, 1959, 2, 127-131.

PREMACK, D., & COLLIER, G. Joint effects of stimu-lus deprivation and intersession interval: Analysisof nonreinforcement variables affecting responseprobability. Psychological Monographs, 1962, 76(5,Whole No. 524).

REYNIERSE, J. H., & BARCH, A. M. Semantic satia-tion and generalization. Psychological Reports,1963, 13, 790.

SARTRE, J. P. Words. New York: Braziller, 1964.SCHREUDER, H. Pejorative sense development.

Groningen: Noordhoff, 1929.SCHULZ, R. W., WEAVER, G. E., & RADTKE, R. C.

Verbal satiation?? Psychonomic Science, 1965, 2,43-44.

SOLOMON, R. L., & POSTMAN, L. Usage as a determi-nant of visual duration thresholds of words.Journal of Experimental Psychology, 1952, 43, 195-201.

STRASSBURGER, F., & WERTHEIMER, M. The discrep-ancy hypothesis of affect and association value ofnonsense syllables. Psychological Reports, 1959, 5,528.

TANNENBAUM, P. H., & GENOEL, R. W. Generalizationof attitude change through congruity principlerelationships. Journal of Personality and SocialPsychology, 1966, 3, 299-304.

THIESSEN, D. D., & McGAUGH, J. L. Conflict andcuriosity in the rat. Paper presented at the meet-ing of the Western Psychological Association,Monterey, California, 1958.

THOMSON, W. R., & SOLOMON, L. M. Spontaneouspattern discrimination in the rat. Journal of Com-parative and Physiological Psychology, 1954, 47,104-107.

THORNDIKE, E. L., & LORGE, I. The teacher's word-book of 30,000 words. New York: Teachers Col-lege, Columbia University, 1944.


UNDERWOOD, B. J. Verbal learning and the educativeprocess. Harvard Educational Review, 1959, 29,107-117.

VAN DER BEKE, G. E. French word book. New York:Macmillan, 1929.

VAN DONGEN, G. A. Ameliorative! in English. Rotter-dam: De Vries, 1933.

VERVEER, E. M., BARRY, H., JR., & BOUSFIELD, W. A.Change in affectivity with repetition. AmericanJournal of Psychology, 1933, 45, 130-134.

WASHBURN, M. F., CHILD, M. S., & ABEL, T. M. Theeffects of immediate repetition on the pleasantnessor unpleasantness of music. In M. Schoen (Ed.),The effects of music. New York: Harcourt, Brace,1927.

WIEBE, G. The effects of radio plugging on students'opinions of popular songs. Journal of Applied Psy-chology, 1940, 24, 721-727.

WILNER, D. M., WALKLEY, R. P., & COOK, S. W.Residential proximity and intergroup relations inpublic housing projects. Journal of Social Issues,8, 1, 19S2.

WLLSON, L. R., & BECKNELL, J. C. The relation be-tween the association value, pronounciability, andaffectively of nonsense syllables. Journal of Psy-chology, 1961, 52, 47-49.

YELEN, D. R., & SCHULZ, R. W. Verbal satiation?Journal of Verbal Learning and Verbal Behavior,1963, 1, 372-377.

(Received April 11, 1967)

attitudinal effects of mere exposure

Documents