Journal of Personality and Social Psychology1980, Vol. 39, No. 5, 752-766

Heuristic Versus Systematic Information Processing andthe Use of Source Versus Message Cues in Persuasion

Shelly ChaikenUniversity of Toronto, Toronto, Ontario, Canada

In Experiment 1, subjects read a persuasive message from a likable or unlikablecommunicator who presented six or two arguments concerning one of two topics.High response involvement subjects anticipated discussing the message topic ata future experimental session, whereas low involvement subjects anticipated dis-cussing a different topic. For high involvement subjects, opinion change wassignificantly greater given six arguments but was unaffected by communicatorlikability. For low involvement subjects, opinion change was significantly greatergiven a likable communicator but was unaffected by the arguments manipulation.In Experiment 2, high issue involvement subjects showed slightly greater opinionchange when exposed to five arguments from an unlikable (vs. one argumentfrom a likable) communicator, whereas low involvement subjects exhibited sig-nificantly greater persuasion in response to one argument from a likable (vs. fivearguments from an unlikable) communicator. These findings support the ideathat high involvement leads message recipients to employ a systematic informa-tion processing strategy in which message-based cognitions mediate persuasion,whereas low involvement leads recipients to use a heuristic processing strategyin which simple decision rules mediate persuasion. Support was also obtainedfor the hypothesis that content-mediated (vs. source-mediated) opinion changewould shower greater persistence.

This research distinguishes between a sys- According to a systematic view, recipientstematic and a heuristic view of persuasion, exert considerable cognitive effort in perform-Both conceptualizations regard message re- ing this task: They actively attempt to com-cipients as concerned with assessing the prehend and evaluate the message's argumentsvalidity of the message's overall conclusion, as well as to assess their validity in relation

to the message's conclusion. In contrast, ac-cording to a heuristic view of persuasion, re-

This article is based on a dissertation submitted cipients exert comparatively little effort inin partial fulfillment of the doctoral requirements judging message validity: Rather than pro-at the University of Massachusetts-Amherst. The cessing argumentation, recipients may relyauthor is grateful to Alice H. Eagly and Icek Ajzen Qn (typlcally) more accessible informationfor their counsel and to John W. Fee III and John , ,, , ., ... .,Keenan, who served as experimenters in Experiment such as the source s identity or other non-i. John Bassili, Alice H. Eagly, Johnathan L. Freed- content cues in deciding to accept a message'sman, Melvin Manis, and Wendy Wood provided conclusion. In essence, a systematic view ofhelpful comments on earlier drafts of this manu- persuasion emphasizes detailed processing ofscript. A report on Experiment 1 was presented at CQ d fa } {the meeting of the American Psychological Associ- , ,b . . , ,. . , . , 6

ation, Toronto, Canada, August 1978. Experiment 2 based cognitions in mediating opinion change,was partially funded by a grant from the Humanities whereas a heuristic view de-emphasizes de-and Social Sciences Committee of the Research tailed information processing and focuses onBoard of the University of Toronto. the role of simple rules Qr cognitive heuristics

Requests for reprints should be sent to Shelly jn mediati persuasion.Chaiken, Department of Psychology, Ermdale Col- ° *lege, University of Toronto, Mississauga, Ontario, Stimulated largely by traditional theoriz-Canada LSL 1C6. ing (e.g., Hovland, Janis, & Kelley, 1953;

Copyright 1980 by the American Psychological Association, Inc. 0022-3514/80/3905-0752J00.75

7S2

SOURCE VERSUS MESSAGE CUES 753

McGuire, 1968, 1969), much persuasion re-search has assumed that source, message, andaudience variables affect opinion change tothe extent that they influence either messagereception (attention, comprehension) or yield-ing to what is received (cf. McGuire, 1968).To better understand the distinction betweensystematic versus heuristic processing, it isuseful to differentiate yielding to persuasiveargumentation from yielding to a message'soverall conclusion. In a systematic view, in-dependent variables (e.g., source factors)indirectly affect yielding to a message's con-clusion and therefore persuasion via their di-rect impact on reception or argument-accept-ance processes. According to a heuristic view,such variables may directly influence therecipient's willingness to accept the message'sconclusion without necessarily influencing re-ception or acceptance of argumentation.

Research suggesting that good comprehen-sion of persuasive argumentation often facili-tates opinion change (e.g., Chaiken & Eagly,1976; Eagly, 1974; Eagly & Warren, 1976)and research suggesting that distraction oftenenhances persuasion by interfering with re-cipients' abilities to critically evaluate per-suasive argumentation (e.g., Osterhouse &Brock, 1970; Petty, Wells, & Brock, 1976)illustrate a systematic view of persuasion.Although little if any research has been ex-plicitly guided by a heuristic view of per-suasion, such a conceptualization seems war-ranted by research suggesting that recipientsoften agree or disagree with a message pri-marily on the basis of their reactions to non-content cues such as communicator credibility(Miller, Maruyama, Beaber, & Valone, 1976),perceived audience opinion (e.g., Landy,1972; Silverthorne & Mazmanian, 1975), orexternal feedback about recipients' own in-ternal states (e.g., Giesen & Hendrick, 1974;Hendrick & Giesen, 1976; Mintz & Mills,1971). For example, Miller et al. (1976)found that faster (communicator) speechrates enhanced both perceptions of communi-cator credibility and persuasion. Finding noevidence that the persuasive impact of speechrate was mediated either by comprehensioneffects or by counterargument disruption, theauthors concluded that subjects may havepredicated their opinion judgments simply on

the basis of whether the communicatorseemed credible. Such a conclusion calls tomind McGuire's (1969) "lazy organism" mes-sage-recipient who utilizes source-linked in-formation as a cue for accepting or rejectingthe message "without really absorbing thearguments used" (p. 198).

Heuristic information processing may in-volve the use of relatively general rules(scripts, schemata) developed by individualsthrough their past experiences and observa-tions (cf . Abelson, 1976; Stotland & Canon,1972) . For example, persons may possess andutilize the rule or categorical script (Abel-son, 1976) that statements by experts canusually be considered veridical, or that state-ments by generally trustworthy persons prob-ably reflect those persons' honest opinions.With respect to source attractiveness, recipi-ents may agree with attractive communicatorsbecause they employ the rule that "peoplegenerally agree with people they like." Sucha script may derive from past experienceswith others or, alternatively, may stem froma lower-order rule suggesting a fairly con-sistent association between the concepts ofliking and interpersonal similarity (Stotland& Canon, 1972). Other heuristics may under-lie the persuasive impact of other cues. Forexample, the finding that recipients are lesspersuaded when an overheard audience ex-presses disapproval (vs. approval) of a mes-sage (Landy, 1972; Silverthorne & Mazma-nian, 1975) might reflect their use of a con-sensus heuristic: Recipients may reject sucha message because most other recipients findit unacceptable.

When will recipients employ a systematicrather than heuristic processing strategy? Aheuristic strategy has the economic advantageof requiring a minimum of cognitive effort.Certainly, judging message acceptability onthe basis of noncontent cues is less effortfulthan receiving and analyzing persuasive argu-mentation. Countering this advantage, aheuristic strategy may be a less reliable meansof judging message validity. In the long run,overreliance on simple decision rules may in-flate Type 1 and Type 2 errors: Recipientsmay sometimes accept (reject) message con-clusions they might otherwise have (cor-rectly) rejected (accepted) had they invested

754 SHELLY CHAIKEN

the time and effort to receive and scrutinizeargumentation. A functional perspective sug-gests that recipients will employ a systematicstrategy when reliability concerns outweigheconomic concerns, and a heuristic strategywhen economic concerns predominate.

Reliability concerns should be paramountand a systematic strategy therefore employedwhen recipients perceive that it is importantto formulate a highly accurate or veridicalopinion judgment. Recipients are likely tohold such a perception under conditions ofhigh "issue involvement" (Kiesler, Collins, &Miller, 1969) or high "response involvement"(Zimbardo, 1960). That is, when recipientsreceive messages on personally importanttopics or when recipients feel that their opin-ion judgments have important consequencesfor themselves (e.g., recipients may expect todiscuss or defend their opinions, or to engagein behavior congruent with their expressedopinions) or for others (e.g., jurors' verdicts).When asked for an opinion on an unimpor-tant topic or when one's opinion judgment isperceived as inconsequential, recipients maygive economic concerns greater weight andemploy a heuristic processing strategy.

This analysis has implications regardingthe relative impact of source and messagevariables on persuasion. In the systematicview, recipients focus primarily on messagecontent. Although source or other noncontentcues may sometimes be used as aids in assess-ing the validity of persuasive argumentation,such cues may be used in only a secondarymanner. Thus, when recipients employ a sys-tematic processing strategy, message charac-teristics (e.g., amount, comprehensibility,validity of persuasive argumentation) mayexert a stronger impact on persuasion thansource characteristics (e.g., credibility, lik-ability). Conversely, in the heuristic view ofpersuasion, recipients avoid detailed process-ing of message content and instead rely oninformation such as the source's identity injudging message acceptability. Thus, whenrecipients employ a heuristic strategy, sourcecharacteristics may exert a greater impact onpersuasion than message characteristics.

Consistent with the above reasoning, pre-vious researchers have shown that sourcecredibility significantly affects persuasion un-

der conditions of low, but not high, issueinvolvement (Rhine & Severence, 1970) andresponse involvement (Johnson & Scileppi,1969). Also, Petty and Capioppo (1979)recently found that a manipulation of argu-ment strength had a greater persuasive im-pact when issue involvement was high. Thepresent research systematically explored theutility of the systematic versus heuristicanalysis of persuasion and its implications re-garding the persuasive impact of source andmessage cues.

Experiment 1Subjects read a persuasive message contain-

ing six or two arguments from a likable or un-likable communicator under conditions ofhigh or low response involvement ("conse-quences"). It was expected that high involve-ment subjects would employ a systematicprocessing strategy and would therefore showgreater opinion change in response to mes-sages containing six arguments but would beunaffected by the likability manipulation. Incontrast, it was expected that low involvementsubjects would employ a heuristic strategyand would therefore express greater agree-ment with the likable communicator butwould be unaffected by the arguments manip-ulation.

The experiment also explored opinion per-sistence. It was assumed that opinion changewould persist to the extent that it was bol-stered by topic-relevant cognitions. It wasalso assumed that recipients who adopted abelief on the basis of who the communicatorwas would possess fewer supportive cogni-tions than recipients who adopted a belief onthe basis of what the communicator said.These assumptions led to the hypothesis thatthe presumably content-mediated (initial)opinion change manifested by high involve-ment subjects would show greater persistencethan the presumably source-mediated opinionchange manifested by low involvement sub-jects. To test this hypothesis, subjects' opin-ions were reassessed approximately 10 daysafter their laboratory participation.

MethodSubjects

Subjects were 207 male and female University ofMassachusetts undergraduates who participated for

SOURCE VERSUS MESSAGE CUES 755

extra course credit in sessions containing up to sixsubjects. Nine subjects were eliminated because theysuspected an influence attempt (three) or questionedthe cover story (six). Data from IS others werediscarded because they were not administered thedelayed opinion posttest (13), or because they as-sociated the posttest with their laboratory partici-pation (two).1

various student organizations. After praising or in-sulting undergraduates in response to a questionfrom the interviewer, the communicator was askedfor his opinion on his "assigned" topic (sleep habitsor trimester system). The remainder of the tran-script consisted of the communicator's statement ofhis overall position (see below) and his presentationof various supportive arguments.

Procedure

At the first (in actuality, the only) session of a"two-session experiment on opinions and group dis-cussions," the experimenter explained that subjectswould receive "discussion topics," give their opinionson these topics, and respond to another question-naire. Subjects learned that at "the second session"they would be individually interviewed about theiropinions on their assigned topics and then discusstheir opinions in groups. After this introduction,each subject received a list of five topics and, via asham random drawing, was assigned one of twotopics (sleep habits or the trimester system) fromthis list.

Justifying the persuasive message, the experi-menter stated that subjects would read an opinioninterview conducted in a related study to "get anidea of what their own interview would be like."The experimenter noted that the related study hadused the same topics but had employed universityadministrators and faculty, rather than undergradu-ates, as subjects. Subjects then received an inter-view transcript (persuasive message, see below) toread. Afterwards, subjects completed a questionnairethat solicited their opinions on all five topics.

"As a second experimental focus" subjects nextcompleted a questionnaire assessing their "reactionsto the interview transcripts." This questionnairecontained the remaining dependent measures (seebelow). Finally, subjects were excused after beingtold that they would be scheduled for "Session 2"later in the semester.

Approximately 10 days later (mean delay = 10.39days; range = 8-15 days), subjects were telephonedby an experimenter who was blind to experimentalcondition. Under the guise of conducting a campusopinion survey, the experimenter solicited subjects'agreement with various opinion statements, two ofwhich corresponded to the positions advocated inthe messages presented at the laboratory session.After probing for suspicion regarding the relation-ship between the posttest and the laboratory session,the experimenter thanked subjects for cooperatingin the telephone survey. Subsequently, debriefingletters were mailed to "all participants.

Interview Transcripts

The transcripts began with an interviewer askingan interviewee (communicator) for background in-formation. The communicator was portrayed as amale university administrator who worked with

Independent Variables

Perceived consequences. Response involvement,or perceived consequences, was manipulated by pre-senting subjects with a message on a topic identicalto or different from the topic they were assignedto be interviewed on and discuss at the "secondexperimental session." Some subjects assigned to thesleep/trimester topic received a sleep/trimester mes-sage (high consequences), whereas others received atrimester/sleep message (low consequences).

Communicator likability. Likability was manipu-lated via the communicator's response to the inter-viewer's question, "How do you like working withundergraduates?" (cf. Eagly & Chaiken, 197S;Jones & Brehm, 1967). This response appears below.Phrases specific to the likable or the unlikable ver-sion are enclosed in parentheses or brackets, respec-tively.

Well, as a matter of fact, I (really enjoy it a lot)[don't really enjoy it very much]. When I firststarted my job here at the university I was alittle apprehensive about the idea of working somuch with undergraduates. Over the years, (how-ever, I've realized that my apprehension was un-justified) [I'm sorry to say, I think that my ap-prehension has been justified]. The undergraduateswho I've met both in my work with variousstudent organizations and in other settings as well,strike me as being pretty (responsible and mature)[irresponsible and immature]. They're really (con-cerned) [unconcerned], I think, with their rolein society. I don't know, of course, but sometimesI think that the public too often (underestimates)[overestimates] the ability and maturity of today'scollege student. They (just don't give undergradu-ates enough credit) [give undergraduates morecredit than they deserve]. Anyway, (it's no wonderthat) [sometimes I wonder why] I continue todo the work I do. ... For me, working withundergraduates (has been pretty) [really hasn'tbeen very] rewarding.

Topic and position advocated. Two topics, sleephabits and the trimester system, were used. Thepositions advocated in the persuasive messages were"People should sleep much less than 8 hours per

1 Including these IS subjects in analyses performedon initial opinion change and other measures as-sessed at the laboratory session yielded findingsvirtually identical to those reported in the text.

756 SHELLY CHAIKEN

night," and "The university should switch from itstwo-semester system to a trimester system." Topicsand positions were selected after pretesting with anadditional 50 subjects who indicated their agree-ment with various opinion statements. The majorselection criterion was that the two statements becounterattitudinal and have similar mean ratings onthe 15-point agreement scale (Ms = 10.72 and 10.54for sleep and trimester topics, respectively, where1 signified strong agreement). Because of the study'sfocus on response involvement (consequences) ratherthan issue involvement, an additional criterion re-quired the topics to be rated similarly and not ex-tremely on personal importance. The sleep and tri-mester topics were considered "neither important norunimportant" (Ms — 8.07 and 7.34, respectively, ona 15-point scale) by pilot subjects (N — 2 2 ) .

Number of arguments. Amount of persuasiveargumentation was varied by preparing messagescontaining six or two supportive arguments. Tworenditions of each message type were written. Eachrendition of the longer message presented the samesix arguments in a different (randomly selected)order. Each rendition of the shorter message con-tained two different arguments drawn randomlyfrom the pool of six.

Measuring Instruments

Opinions. Immediately after message exposure,subjects indicated their agreement with the overallpositions stated in the persuasive messages (seeabove) by marking 15-point scales anchored byagree strongly and disagree strongly. During thedelayed telephone posttest, subjects indicated theiragreement with similarly worded opinion state-ments by responding orally to 5-point scales an-chored by definitely agree and definitely disagree.

Thoughts. On the questionnaire assessing their"reactions to the interview transcript," subjects werefirst given 3 minutes to list their "thoughts" aboutthe communicator's statements. Thoughts werescored by two independent raters as message-ori-ented or communicator-oriented (M, C) and aspositively, negatively, or neutrally valenced (+,—, 0). Examples of statements placed in each cate-gory along with interrater reliability coefficients(Pearson rs) are: M + (r = .86): "The economicadvantages of the trimester agree with me"; M —(r =.&&): "Reason for trimester not sound logi-cally"; M0(r =.86): "He said REM (sleep) canbe controlled"; C + (r = .85): "He was verypolite"; C - (r = .82): "He was a little close-minded"; C O (r = .7S): "Hesitant about talkingabout self." ~

Source perception. Subjects next rated the com-municator on 15-point bipolar adjective scales. Posi-tive poles of the 12 scales used were warm, knowl-cdgable, modest, intelligent, approachable, compe-tent, likable, trustworthy, pleasing, sincere, friendly,and unbiased.

Message comprehension. Next, subjects wereasked to write down the message topic, its overall

position, and each supportive argument that thecommunicator had presented. An argument wasscored correct if in the opinion of two independentraters (r = .86) it accurately summarized an argu-ment appearing in the message. All subjects recalledthe message's topic and all but four (retained in theanalyses) accurately specified its overall position.

Other measures. To measure the effort subjectsexpended processing message content, the experi-menter covertly recorded the time each subjectspent reading the interview transcript. On the lastpage of the reactions questionnaire, subjects ratedthe message topic's importance, their desire to bewell informed on the topic, the effort they hadspent reading the message, the relative amount oftime they had spent thinking about the communi-cator's arguments (vs. his personal characteristics),and their interest in the communicator's arguments(all on 15-point scales). Subjects also indicatedtheir age, sex, and assigned topics. In the laboratory,subjects wrote down their interpretations of thestudy; over the telephone, they were asked if they'dbeen in any similar opinion surveys. These responseswere coded for suspicion.

Resets

The design included two levels each of per-ceived consequences, communicator likability,number of arguments, message topic, subjectsex, and message rendition (nested withinlevels of topic and number of arguments).Since preliminary analyses yielded few effectsdue to message rendition, all reported analysesignored this variable. Of the few sex and topiceffects yielded by the reported analyses, onlythose relevant to the major hypotheses arepresented here.

2 Statements that could not be placed into any ofthe six categories were coded as "other thoughts"and were not analyzed. It should be noted that thepresent thought-scoring procedure parallels to a cer-tain extent scoring procedures typically used by per-suasion researchers (e.g., Osterhouse & Brock, 1970).Statements traditionally referred to as counter-arguments and source derogations are, in the presentscheme, termed negative message-oriented thoughtsand negative communicator-oriented thoughts, re-spectively. Statements commonly referred to as sup-portive thoughts are, in the p'resent scheme, termedpositive message-oriented or communicator-orientedthoughts depending on whether they refer more tothe message or more to the communicator. For eachtype of thought, each independent rater judged thenumber of such thoughts listed by each subject.Interrater reliability was calculated separately foreach thought category by determining the correlationbetween the two raters' judgments.

SOURCE VERSUS MESSAGE CUES 757

Check on Experimental Conditions

The communicator who praised (vs. in-sulted) undergraduates was judged more lik-able, Ms = 4.22 versus 8.83, F(l , 151) =125.14, p < .001. Subjects recalled more per-suasive arguments when the message con-tained six (vs. two) arguments, Ms = 2.81versus 1.56, F(l , 151) = 76.18, p< .001.Finally, the consequences manipulation suc-cessfully induced the conditions thought nec-essary to foster systematic versus heuristic in-formation processing: High (vs. low) conse-quences subjects expressed a greater desire tobe well-informed on the message topic, Ms =6.24 versus 7.05, F(l, 151) = 3.69, p < .06.

It is important to note that assignment totopics did not lead subjects to "strategically"shift their opinions (cf., Cialdini, Levy, Her-man, Kozlowski, & Petty, 1976): Topic As-signed X Topic Received analyses of varianceon subjects' initial and delayed opinions onthe two topics yielded no significant effectsdue to topic assignment (Fs < 1.0). Thetopic-received main effect was significant onall four opinion measures (ps < .05), indi-cating that the persuasive messages success-fully shifted subjects' opinions.

Opinions

Mean opinion change scores for the pri-mary experimental conditions appear in Table1.3 Analysis on subjects' initial opinion changescores revealed main effects due to communi-cator likability, F(l, 151) =4.16, p < .05;number of arguments received, F(l, 151) =6.14, p < .05; and message topic, F(l, 151)= 4.56, p < .05. Overall, greater initialchange occurred given the likable (vs. un-likable) communicator (Ms = 2.54 vs. 1.51),six (vs. two) persuasive arguments (Ms =2.67 vs. 1.47), and trimester (vs. sleep)messages (Ms = 2.60 vs. 1.59).

Planned comparisons on initial opinionchange yielded findings supportive of thestudy's major hypothesis. As Figure 1 illus-trates, high consequences subjects showedgreater initial opinion change in response tosix (vs. two) arguments, F(l, 151) =4.62,p < .05, but were unaffected by communi-cator likability, F(l , 151) < 1.0. In contrast,low consequences subjects showed greater

Table 1Initial and Delayed Opinion Change as aFunction of Perceived Future Consequences,Communicator Likability, and Number ofArguments Presented: Experiment 1

Likablecommunicator

Opinionchange

Sixargu-ments

Twoargu-ments

Unlikablecommunicator

Sixargu-ments

Twoargu-ments

High perceived consequences

InitialDelayed

2.793.02

2.202.48

3.282.15

.48

.94

Low perceived consequences

InitialDelayed

3.162.53

1.98.70

1.411.54

.95

.47

Note. Cell ns range from 20 to 26.

initial opinion change given the likable (vs.unlikable) communicator, F(l, 151) = 3.92,p < .05, but were unaffected by the amountof argumentation provided, F(l , 151) = 1.48,ns. Although these results conformed to pre-dictions, the differences between consequencesconditions proved statistically weak on anoverall basis. Of the interactions implied bythe major hypothesis, neither the Conse-quences X Arguments nor Consequences XLikability interactions reached significance inthe full analysis of variance (ps — .24 and .25,

3 Initial opinions were measured on IS-pointscales and delayed opinions on 5-point scales. Forcomparability, subjects' delayed opinion scores weretransformed to 15-point scales (see Minium, 1970,p. 115). Opinion change scores were formed bysubtracting the mean opinion expressed by an in-ternal control group from each subject's (initial ordelayed) opinion. All subjects receiving a sleepmessage formed an internal control group on thetrimester topic for subjects reading a trimestermessage, and all subjects who received a trimestermessage served as internal controls on the sleeptopic for subjects reading a sleep message. Theopinions expressed by internal control subjects didnot differ significantly from those expressed by anexternal control group of opinion-only pilot subjects:For the sleep topic, M = 10.23 versus M = 10.72for internal and external controls, respectively,t (137)=.80; for the trimester topic, M- 10.32versus M = 10.54 for internal and external controls,respectively, ((142) = .34.

758 SHELLY CHAIKEN

LUCD

TO

-rO2Q.o_1

p

3.0

•3 e

2.0

1.5

1.0

.5

o

3.0

2.5

2.0

1 5

1.0

.5

n

A

[

~"*^^^^^~~~—

~ — o

• • LIKEABLE COMMUNICATOR0 O UNLIKEABLE COMMUNICATOR

1 I

B ^^

~"~"~--\.

• • SIX ARGUMENTSO O TWO ARGUMENTS

i i

HIGHCONSEQUENCES

LOWCONSEQUENCES

Figure 1. Panel A depicts initial opinion change asa function of communicator likability and perceivedconsequences; Panel B depicts initial opinion changeas a function of number of arguments presented andperceived consequences.

respectively), and the Consequences X Argu-ments X Likability interaction attained onlymarginal significance (p < .10).

To explore opinion persistence, a repeatedmeasures analysis of variance employing timeof posttest (initial vs. delayed) as an addi-tional design factor was performed. A time ofposttest main effect, F(l, 151) =3.96, p<.05, revealed an overall trend for opinionchange to dissipate over the 10-day posttestinterval (Ms = 2.07 vs. 1.76)." The Time XConsequences X Likability interaction, F(l,151) = 4.00, p < .05, showed a data patternroughly in accord with the hypothesis thatgreater persistence would be shown by highconsequences subjects. Regardless of com-municator likability, the opinion change mani-fested by high consequences subjects remainedrelatively stable between posttests: Ms =2.48 versus 2.78, F< 1.0, for likable com-municator; Ms = 1.88 versus 1.54, F< 1.0,for unlikable communicator. In contrast,

opinion change declined significantly for lowconsequences subjects exposed to the likablecommunicator, Ms = 2.59 versus 1.65, F(l,151) = 7.24, p < .01, although not for thoseexposed to the unlikable communicator, Ms= 1.18 versus 1.00, F < 1.0. The nonsignifi-cant decrement in the latter condition prob-ably reflects the fact that initial change inthis cell was low and only marginally greaterthan zero, *(43) = 1.99, p < .10.

Paralleling the initial opinion change find-ings, the repeated measures analysis alsoyielded between-subjects effects due to lik-ability (p < .05), arguments (p < .05), anda marginal effect due to message topic (p <.10). In addition, a Consequences X LikabilityX Arguments X Topic interaction (p < .05)was obtained. This interaction indicated thatthe primary experimental variables exerted astronger impact on opinions within trimester(vs. sleep) message conditions even thoughthe patterning of opinion change means wasgenerally consistent within both topics.

Comprehension and Thoughts

In addition to the manipulation check notedearlier, the analysis on argument recall yieldeda consequences effect (p < .005) and a Con-sequences X Arguments interaction (p < .05).Overall, high (vs. low) consequences subjectsrecalled more arguments (Ms = 2.40 vs.1.98), and this difference was most pro-nounced given six persuasive arguments.Subjects reading trimester (vs. sleep) mes-sages also recalled more arguments (p <.001).

Analyses were performed on each type ofthought emitted by subjects (M +, M —,

••Subjects' opinions were reassessed between 8 and15 days after the initial posttest (mean delay =10.39 days). Analysis of variance on days betweenposttest yielded no significant effects. Thus all ex-perimental groups were subjected to approximatelythe same delay interval. Also, unless they wereplaying the role of "cooperative" subjects to thehilt, the low rate of suspicion or innocent mentionof their earlier laboratory experience expressed bysubjects during the "telephone opinion survey" sug-gested that the experiment had been successful indivorcing the context of the delayed posttest fromthe context of the initial opinion posttest.

SOURCE VERSUS MESSAGE CUES 759

MO, C +, C -, CO) and also on three de-rived indices: Message-oriented minus com-municator-oriented thoughts (regardless ofvalence), positive minus negative thoughts(regardless of orientation), and total thoughtsexpressed. For brevity, only the analyses onthe derived indices are reported. Relativelymore message-oriented than communicator-oriented thoughts were expressed by high(vs. low) consequences subjects (p = .06),and also by subjects exposed to six (vs. two)arguments (p < .005) and the likable (vs.unlikable) communicator (p < .005). A lik-ability effect on positive minus negativethoughts (p < .001) indicated that the likablecommunicator elicited more positive thannegative thoughts, whereas the unlikable com-municator elicited more negative thoughts. Noeffects obtained on total thoughts expressed.

Source Perception

A factor analysis (varimax rotation) of thesource ratings yielded two rotated factors,labeled attractiveness (e.g., warm, likable,friendly) and expertise (e.g., knowledgable,intelligent, competent). These factors ac-counted for 55.8% and 10.2% of the vari- •ance, respectively. Factor scores were com-puted for each subject and treated by analy-sis of variance. The trustworthy, sincere, andunbiased scales, which loaded highly onneither factor, were analyzed separately.These analyses revealed that the likable (vs.unlikable) communicator was viewed as moreattractive, expert, trustworthy, sincere, andunbiased (ps < .005).

Other Dependent Variables

High (vs. low) consequences subjects spentmore time reading the persuasive message(p < .005) and reported spending more timethinking about the communicator's argumentsthan his personal characteristics (/><.01).Subjects exposed to six (vs. two) persuasivearguments also took longer to read the mes-sage (/><.001), spent more time thinkingabout the communicator's arguments (p <.001), expressed a greater desire to be wellinformed on the topic (p < .005), reportedexerting greater effort reading the message

(p < .05), and rated the topic as being moreimportant (p < .001).

Correlational Analyses

Correlational analyses provided further in-formation regarding the cognitive mediationof initial opinion change. Correlations be-tween opinion change and perceptions of thecommunicator (source ratings, positive andnegative communicator-oriented thoughts)were generally low and nonsignificant, withthe exception of perceived expertise, whichcorrelated positively with opinion changewithin both low and high consequences con-ditions (r = .29, p < .005, and r = .25, p <.05, respectively). As anticipated by the sys-tematic versus heuristic analysis of persua-sion, argument recall and both positive andnegative message-oriented thoughts were sig-nificantly correlated with initial opinionchange within high consequences conditions(recall: r = .26, p < .05; M + : r = .41, p <.001; M-: r= -.32, p< .001), but onlynegligibly related within low consequencesconditions (recall: r = .06; M+: r = .20,p < .10; M-: r = -.16, p < .15).

Multiple regression analyses using initialopinion change as the criterion variable werealso performed.5 Within high consequencesconditions, greater initial opinion change waspredicted primarily by a greater number ofboth positive message-oriented thoughts (p <.05) and positive communicator-orientedthoughts (p = .07), fewer negative message-oriented thoughts (p = .10), and greater ar-gument recall (p = .11). Within low conse-quences conditions, variables that signifi-cantly or marginally predicted greater initialopinion change included heightened percep-

5 Two (regular) multiple regression analyses wereperformed, one employing data from high conse-quence subjects (n = 89) and one using data fromlow consequences subjects (n = 94). Both analysesemployed the following predictor variables: age,time spent reading the message, self-reported timespent thinking about the communicator's argumen-tation (vs. personal characteristics), attractivenessfactor scores, expertise factor scores, arguments re-called, positive and negative communicator-orientedthoughts, and positive and negative message-orientedthoughts.

760 SHELLY CHAIKEN

tions of communicator expertise (p < .05)and a greater number of positive communi-cator-oriented thoughts (p — .06).

Experiment 2

A conceptual replication of the first experi-ment was conducted in order to extend thegenerality of the systematic versus heuristicanalysis. This study included only the mosttheoretically interesting combinations of thearguments and likability variables and ma-nipulated issue involvement ("personal rele-vance") rather than response involvement.In the 2 x 2 design of Experiment 2, sub-jects read a message consisting of either fivepersuasive arguments from an unlikable com-municator or one persuasive argument from alikable communicator under conditions ofeither high or low personal relevance. It waspredicted that subjects in high personal rele-vance conditions would be more persuaded byfive arguments from an unlikable communi-cator than by one argument from a likablecommunicator. Conversely, low personalrelevance subjects were expected to showgreater opinion change in the one-argument/likable-communicator condition than in thefive-arguments/unlikable-communicator con-dition. Experiment 2 did not examine opinionpersistence and did not include the full set ofdependent measures employed in Experiment 1.

MethodSubjects

Subjects were 80 University of Toronto under-graduates enrolled in the author's introductory psy-chology course who volunteered to remain afterclass one evening to participate in an "impressionformation" experiment.

Procedure

Each subject received one of four versions (repre-senting the four experimental conditions) of an ex-perimental booklet. The instructions printed on thecover page informed subjects that they would recordtheir impressions of a "target speaker" after firstreading a "partial transcript of an interview withthe target" and a "transcript of a speech made bythe target."

Communicator likability. The first page of thebooklet presented the "interview transcript." Thetarget speaker (communicator) was portrayed as a

University of Toronto (U of T) administrator whohad previously held a similar position at the Uni-versity of British Columbia (UBC). Similar to thedevice employed in Experiment 1, the communi-cator's likability or unlikability was conveyed viahis response to the interviewer's question, "By theway, how do you like U of T compared to UBC?"In essence, the likable communicator responded bypraising U of T students and faculty and extollingthe virtues of living in Toronto (at the expense ofUBC and Vancouver), whereas the unlikable com-municator responded by derogating U of T studentsand faculty and the city of Toronto (to the benefitof UBC and Vancouver).

Personal relevance of message topic. The nextpage of the booklet informed subjects that theywould read a transcript of the "target's speech onthe issue of switching from a two-semester to a tri-mester system at the university." It was stated thatthe speech had been presented to a university com-mittee charged with studying the issue. High (vs.low) personal relevance was manipulated by statingthat, if the Committee approved a switch, the uni-versity would adopt the trimester system in the1981 (vs. 1985) academic year.

Persuasive message. The last page of the book-let contained the persuasive message, whose overallposition stated that "the University of Torontoshould switch from its current two-term system toa trimester system." Opinion data from an inde-pendent group of control subjects (N = 125) indi-cated that U of T undergraduates disagreed moder-ately with this position (M = 10.33 on a 15-pointscale on which 15 signified extreme disagreement).In the message, the communicator first stated hisoverall position and, if portrayed earlier as unlik-able, provided five arguments supporting this posi-tion. If the communicator had been portrayed aslikable, he presented only one supportive argument(drawn randomly from the pool of five arguments).Both messages ended with the communicator re-stating his overall position on the message topic.

Dependent measures. After examining their ex-perimental booklets, subjects were given a one-pagequestionnaire. Subjects first wrote down the com-municator's overall position (all subjects satisfac-torily recalled the position) and then indicated theiragreement with this position by marking a 15-pointscale anchored by strongly disagree and stronglyagree. Next, subjects rated the communicator on thefollowing 15-point bipolar adjective scales: warm-cold, likable-unlikable, friendly-unfriendly, and sin-cere-insincere. On 15-point scales, subjects nextindicated how personally relevant they consideredthe message topic to be, the relative amount of timethey had spent thinking about the communicator'sarguments (vs. personal characteristics), and theextent to which they typically agreed (disagreed)with people they liked (disliked). Subjects also indi-cated their sex and wrote down their interpretationsof the experiment. Three subjects (retained in theanalyses) were assessed as being suspicious of aninfluence attempt (two) or the cover story (one).

SOURCE VERSUS MESSAGE CUES 761

Results

The design included two levels each of sub-ject sex, personal relevance of message topic,and source-argument pairing (unlikable com-municator - five arguments vs. likable com-municator-one argument). Since preliminaryanalyses revealed no sex differences, subjectsex was ignored in the reported analyses.

Check on Experimental Conditions

Subjects informed that the universityplanned to adopt the trimester (if approved)in 1981 (vs. 1985) judged the trimester issueto be significantly more personally relevant:Ms = 5.58 versus 7.42, F(l, 76) = 5.62,p< .025). No other effects were significanton this measure. The only significant effecton subjects' likableness ratings was due tothe source-argument pairing variable, F(l,76) = 10.11, p < .005: The communicatorwho praised (vs. insulted) U of T studentsand faculty and complimented (vs. derogated)Toronto was regarded as more likable (Ms =5.82 vs. 7.75).

Opinion Change

Opinion change scores (see Table 2) wereformed by subtracting the mean opinion ofthe control group from each subject's opinion.As expected, the Relevance X Source-Argu-ment Pairing interaction proved significanton these scores, F(l, 76) = 4.30, p < .OS.Within low personal relevance conditions, sig-nificantly greater opinion change occurredamong subjects in the likable-communicator/one-argument condition than in the unlikable-communicator/five-arguments condition, F(l,76) = 4.52, p < .05. Within high relevanceconditions, nonsignificantly greater opinionchange occurred among subjects in the unlik-able-communicator/five-arguments (vs. lik-able/one-argument) condition, F < 1.0. Noother effects were significant on opinionchange.

Other measures. The likable communi-cator (who presented one argument), com-pared to the unlikable communicator (whopresented five arguments), was viewed asmore likable (p < .005), warmer (p < .005),

Table 2Mean Opinion Change and Use of theLiking/Agreement Rule as a Function ofPersonal Relevance and Source-ArgumentPairing: Experiment 2

Dependentvariable

Unlikable/five

arguments

Likable/one

argument

High relevance

Opinion change 3.73 2.78Use of liking/

agreement rule 7.25 10.30

Opinion changeUse of liking/

agreement rule

Low relevance

2.18

8.75

4.68

8.95

Note. Higher numbers indicate greater opinionchange and greater use of the liking/agreement rule.Cell ns = 20.

and friendlier (p < .001), but as less sincere(p < .05). No other effects were significanton these ratings.

It had been expected that low relevancesubjects would report greater use of the lik-ing/agreement rule. Although the means werein the expected direction, the relevance effectproved nonsignificant (F < 1.0) on subjects'reports of the extent to which they typicallyagreed (disagreed) with people they liked(disliked)." Analysis of this measure (seeTable 2 for means) did yield a source-argu-ment pairing effect (p<.025), and more

0 Because this item appears to ask for self-reporteduse of the liking/agreement heuristic over an ex-tended period of time, it is quite reasonable to arguethat subjects' responses would not necessarily beaffected by their present level of involvement (in-deed, the results on this item conformed only par-tially to predictions). The rationale for includingthis item and thinking that it might reflect highand low involvement subjects' differential use of theliking/agreement rule is similar to the logic under-lying Bern's (1972) self-perception hypothesis. Justas we might expect a person's presumably stableself-description of how honest he or she is typicallyto be influenced by a recent and salient honest (ordishonest) behavior on his or her part, it was ex-pected that subjects' self-reports of their typical useof the liking/agreement heuristic would be influencedby whether or not they had just employed this rulein making their opinion judgments.

762 SHELLY CHAIKEN

interestingly, a Relevance X Source-Argu-ment Pairing interaction (p < .05). The firsteffect indicates greater reported use of theliking/agreement rule among subjects in lik-able-communicator/one-argument (vs. unlik-able/five-arguments) conditions. The inter-action indicates that among subjects exposedto five arguments from an unlikable com-municator, greater use of the liking/agreementrule was, as expected, reported by subjectswithin low relevance conditions. However,among subjects receiving only one argumentfrom a likable communicator, those withinhigh personal relevance conditions reportedgreater use of this heuristic.

No significant effects were obtained on sub-jects' ratings of the amount of time they hadspent thinking about the communicator'sarguments versus his personal characteristics.

Discussion

The systematic versus heuristic analysis ofpersuasion suggests that high levels of re-sponse or issue involvement lead message re-cipients to employ a systematic informationprocessing strategy in forming their opinionjudgments, whereas low levels of involvementlead recipients to employ a more economicheuristic strategy. Based on the assumptionthat systematic processing maximizes the per-suasive impact of message cues and mini-mizes the impact of noncontent cues and thatheuristic processing minimizes the persuasiveimpact of message cues and maximizes theimpact of noncontent cues, it was hypothe-sized that the initial opinion judgments ofhigh involvement subjects would be morestrongly influenced by the amount of argu-mentation provided in the message than bythe communicator's likability, whereas thereverse would be true for low involvementsubjects.

The two experiments provided convergingevidence for this hypothesis. In the firststudy, high consequences subjects exhibitedsignificantly greater initial opinion change inresponse to messages containing six argu-ments but were unaffected by the communi-cator's likability. Conversely, low conse-quences subjects exhibited significantly greateropinion change in response to the likable

communicator but were unaffected by theamount of argumentation provided. In Experi-ment 2, subjects for whom the message topicwas high in personal relevance showed slightlygreater opinion change when receiving fivearguments from an unlikable communicatorthan when receiving one argument from a lik-able source. In contrast, subjects for whomthe topic was low in personal relevance ex-hibited significantly greater opinion changewhen they received one argument from a lik-able (vs. five arguments from an unlikable)communicator.

Analyses on the supplementary measuresincluded in Experiment 1 yielded additionalsupport for the assumptions underlying themajor hypothesis. The fact that high conse-quences subjects expressed a greater desire tobe well informed on the message topic cor-roborated the assumption that the conse-quences manipulation would engender differ-ing levels of motivation for in depth informa-tion processing. The data were also generallyconsistent with the assumption that thesediffering motivational sets would lead highconsequences subjects to employ a systematicprocessing strategy in which message-basedcognitions would primarily mediate opinionchange and would lead low consequences sub-jects to employ a heuristic strategy in whichtheir perceptions of the communicator wouldprimarily mediate persuasion. High, comparedto low, consequences subjects spent more timereading the persuasive message, reportedspending more time thinking about the com-municator's argumentation (vs. his personalcharacteristics), recalled more arguments, andgenerated relatively more message-oriented(vs. communicator-oriented) thoughts. Fur-ther, correlational analyses revealed that ar-gument recall, positive message-orientedthoughts, and negative message-orientedthoughts were significantly correlated with andrelatively good predictors of initial opinionchange (ps < .11) for high consequences sub-jects. In contrast, these variables were negli-gibly correlated with and poor predictors ofinitial opinion change (ps > .28) for lowconsequences subjects. Finally, although per-ceptions of communicator expertise were sig-nificantly correlated with opinion changewithin both consequences conditions, per-

SOURCE VERSUS MESSAGE CUES 763

ceived expertise proved to be a significantpredictor of initial opinion change for lowconsequences (p < .05), but not high conse-quences (p > .31), subjects.

Communicator Likability and Opinion Change

Although the opinion data of Experiment 1were consistent with the suggestion that thepersuasive impact of communicator likabilitymight reflect recipients' use of a simpleschema such as "people generally agree withpersons they like," the study provided nodirect support for this simple informationprocessing mechanism. Further, because thelikable communicator was regarded not onlyas more attractive but also as more expert,sincere, trustworthy, and unbiased than hisunlikable counterpart, it is probably unwiseto interpret the impact of the likability ma-nipulation on opinion change and other mea-sures as reflecting the unique effect of com-municator likability, unconfounded by othersource dimensions. The fact that subjects' per-ceptions of communicator expertise relatedmore strongly to opinion change than theirperceptions of communicator attractivenessunderscores this point. Indeed, the findingthat expertise (and not attractiveness) pre-dicted initial opinion change for low conse-quences subjects suggests that if, as pro-posed here, these subjects did employ a simpledecision rule in forming their opinion judg-ments, it was most likely one based on theirperceptions of the communicator's expertise.

It is important to note that the likabilitymanipulation of Experiment 1 had little im-pact on measures presumed to reflect sub-jects' processing of message content (e.g.,argument recall, time spent reading the mes-sage, time spent thinking about the communi-cator's argumentation vs. time spent thinkingabout his personal characteristics). Thesefindings are consistent with previous research(e.g., Norman, 1976; Snyder & Rothbart,1971) and compatible with a heuristic viewof persuasion: Communicator characteristicssuch as likability may often directly affectrecipients' tendencies to accept or reject amessage's overall conclusion without neces-sarily influencing their processing of per-suasive argumentation.

To more directly explore recipients' possibleuse of a liking/agreement rule, subjects inExperiment 2 were asked to indicate theextent to which they typically agreed (dis-agreed) with people they liked (disliked).The findings obtained on this measure con-formed only partially to predictions. Amongsubjects receiving five arguments from anunlikable communicator, low relevance sub-jects reported greater use of the liking/agree-ment rule, as expected. However, among sub-jects receiving one argument from a likablecommunicator, high relevance subjects re-ported greater use of this heuristic. It is pos-sible that although high relevance subjects inthis condition may have been primarilyoriented toward assessing message content,they may have considered a message contain-ing only one argument a meager basis forassessing message validity. Consequently,these subjects may have felt it necessary torely also on information regarding the com-municator's likability in forming their opin-ion judgments. This interpretation, althoughtentative, suggests that aside from low levelsof involvement, recipients may rely heavily onsource (or other noncontent) cues in judgingmessage acceptability when the message'sargumentation provides an insufficient basison which to predicate an opinion judgment.In support of this conjecture, McCroskey(1969) found that communicator credibilitysignificantly affected opinion change onlywhen the persuasive message contained mini-mal (vs. strong) evidence in support of itsoverall conclusion (for an elaborated discus-sion of factors other than involvement thatmay enhance the persuasive impact of sourcecues, see Chaiken, 1978).

Persistence oj Opinion Change

As noted previously, Experiment 1 pro-vided adequate support for the assumptionthat message-based cognitions would pri-marily mediate initial opinion change withinhigh consequences conditions, whereas source-oriented cognitions would primarily mediateopinion change within low consequences con-ditions. This assumption led to the hypothesisthat greater opinion persistence would bemanifested by high consequences subjects.

764 SHELLY CHAIKEN

In accord with this hypothesis, opinionchange remained relatively stable over timewithin high consequences conditions but dissi-pated within low consequences conditions,although the decrease was significant only forlow consequences subjects exposed to thelikable communicator. The failure to detect asignificant decrement within the low conse-quences/unlikable communicator condition isunderstandable, since initial change in thiscell was not reliably different from zero, pre-sumably because subjects used the communi-cator's unlikability as a discounting cue.

The present persistence findings and analy-sis of persuasion are relevant to the sleepereffect phenomenon. First, the data showed nohint of a sleeper effect: There was no trend,within or across consequences conditions, foragreement with an unlikable communicator toincrease with the passage of time. The tradi-tional "discounting-cue" explanation for thesleeper effect (Hovland, Lumsdaine, & Shef-field, 1949) holds that over time, the dis-counting cue (e.g., communicator unlikabil-ity) becomes dissociated from the message sothat the persuasive impact of message contentgradually becomes manifest (McGuire, 1969).Cook and Cruder (Note 1) have suggestedthat the sleeper effect is likely to be observedonly if (among other criteria) (a) the mes-sage is convincing and has a strong initialimpact on opinions and (b) the discountingcue significantly inhibits the persuasive im-pact the message would typically have. Animplicit assumption of their analysis as wellas earlier analyses (Hovland et al., 1949;McGuire, 1968) is that recipients have ade-quately processed message content-—other-wise it is unlikely that over time, the per-suasive impact of message content couldpossibly manifest itself.

The systematic versus heuristic analysis ofpersuasion suggests that Cook and Gruder's(Note 1) two criteria may be met and, con-sequently, that the sleeper effect may beobserved only when involvement is at somemoderate level. When involvement is high, thepresent analysis suggests that the persuasiveimpact of potential discounting cues (e.g.,low levels of credibility, likability) will beminimal. Thus, unless perhaps a discountingcue is made salient after recipients have pro-

cessed message content, it is unlikely thatCook and Gruder's second criterion typicallywould be met. Conversely, when involvementis low, the present analysis suggests thatmessage content has a minimal impact onopinions and that recipients generally predi-cate their opinions on the basis of their reac-tions to noncontent cues. Given low levels ofinvolvement, then, it is unlikely that Cookand Gruder's first criterion would often bemet.

Involvement and Persuasion

Finally, the current research has implica-tions for understanding the impact of in-volvement on opinion change and opinionpersistence. The fact that involvement ex-erted no main effect on initial opinion changein either experiment, as well as some recentdemonstrations that greater involvementsometimes facilitates persuasion (e.g., Pallak,Mueller, Dollar, & Pallak, 1972; Petty &Cacioppo, 1979), contradicts earlier theoriz-ing (e.g., Sherif, Sherif, & Nebergall, 1965;Triandis, 1971) and research (Miller, 1965;Sherif & Hovland, 1961) suggesting thatgreater involvement typically lowers messagepersuasiveness. The systematic versus heu-ristic analysis, which argues that greater in-volvement heightens recipients' tendencies toscrutinize message content, suggests that highinvolvement can both facilitate and inhibitpersuasion depending on the quality of per-suasive argumentation. Indeed, Petty andCacioppo (1979) recently found that in-volvement heightened persuasion when a mes-sage contained strong arguments but inhib-ited persuasion when the message presentedweak arguments. On the other hand, the find-ing in Experiment 1 that high consequencessubjects showed relatively greater opinionstability suggests, in accord with previouscorrelational evidence (Watts, 1967), thatopinion change induced under high involve-ment conditions results in greater persistence.This greater persistence presumably occursbecause high involvement fosters in depthprocessing of message content, which leadsrecipients to possess more topic-relevant cog-nitive supports for their adopted opinions.

SOURCE VERSUS MESSAGE CUES 765

Summary and Conclusions

The present research generally supports thedistinction made between a systematic view ofpersuasion, with its emphasis on detailedinformation processing and the mediationalrole of content-based cognitions, and a heu-ristic view of persuasion, with its focus on therole of simple rules or cognitive heuristics inmediating opinion change. Admittedly, theresearch was more successful in documentingthe cognitive mediation of opinion changespecified by a systematic view of persuasionthan it was in documenting the cognitivemediation specified by a heuristic conceptual-ization. Yet the overall pattern of findingssupports the utility of the latter conceptuali-zation and suggests that it may be worth-while for persuasion researchers to furtherexplore the conditions under which recipientsmay engage in heuristic information process-ing and to explore more systematically thekinds of simple information processing mech-anisms that may often underlie persuasioneffects.

The research also provides some support forthe hypothesis that content-mediated opinionchange should persist longer than source-mediated persuasion. Finally, the relevanceof the systematic versus heuristic analysis ofpersuasion to the sleeper effect phenomenonand the role of involvement in persuasion arediscussed. With regard to the former, it issuggested that the preconditions thought nec-essary for obtaining the sleeper effect (Cook& Cruder, Note 1) may not often occur wheninvolvement levels are set very high or low.With respect to involvement itself, it is ar-gued that although heightened involvementcan both facilitate and inhibit the immediateimpact of a persuasive message, change in-duced under conditions of high (vs. low) in-volvement should show greater persistence.

Reference Note

1. Cook, T. D., & Cruder, C. L. The sleeper effectpredicted from the discounting cue hypothesis: Aliving phenomenon. Unpublished manuscript,Northwestern University, 1979.

References

Abelson, R. P. Script processing in attitude forma-tion and decision making. In J. S. Carroll & J. W.

Payne, Cognition and social behavior. Hillsdale,N.J.: Erlbaum, 1976.

Bern, D. J. Self-perception theory. In L. Berkowitz(Ed.), Advances in experimental social psychology(Vol. 6). New York: Academic Press, 1972.

Chaiken, S. The use of source versus message cues inpersuasion: An information processing analysis.Unpublished doctoral dissertation, University ofMassachusetts—Amherst, 1978.

Chaiken, S., & Eagly, A. H. Communication modal-ity as a determinant of message persuasivenessand message comprehension. Journal of Person-ality and Social Psychology, 1976, 34, 605-614.

Cialdini, R. B., Levy, A., Herman, P., Kozlowski, L.T., & Petty, R. E. Elastic shifts of opinion: De-terminants of direction and durability. Journal ofPersonality and Social Psychology, 1976, 34, 663-672.

Eagly, A. H. Comprehensibility of persuasive argu-ments as a determinant of opinion change. Journalof Personality and Social Psychology, 1974, 29,758-773.

Eagly, A. H., & Chaiken, S. An attribution analysisof the effects of communicator characteristics onopinion change: The case of communicator attrac-tiveness. Journal of Personality and Social Psy-chology, 1975, 32, 136-144.

Eagly, A. H., & Warren, R. Intelligence, comprehen-sion, and opinion change. Journal of Personality,1976, 44, 226-242.

Gicsen, M., & Hendrick, C. Effects of false positiveand negative arousal feedback on persuasion.Journal of Personality and Social Psychology, 1974,30, 449-457.

Hendrick, C., & Giesen, M. Self-attribution of atti-tude as a function of belief feedback. Memory andCognition, 1976, 4, 150-155.

Hovland, C. I., Janis, I. L., & Kelley, H. H. Com-munication and persuasion. New Haven, Conn.:Yale University Press, 1953.

Hovland, C. I., Lumsdaine, A. A., & Sheffield, F. D.Experiments on mass communication. Princeton,N.J.: Princeton University Press, 1949.

Johnson, H. H., & Scileppi, J. A. Effects of ego-involvement conditions on attitude change to highand low credibility communicators. Journal ofPersonality and Social Psychology, 1969, 13, 31-36.

Jones, R. A., & Brehm, J. W. Attitudinal effects ofcommunicator attractiveness when one chooses tolisten. Journal of Personality and Social Psychol-ogy, 1967, 6, 64-70.

Kiesler, C., Collins, B., & Miller, N. Attitude change.New York: Wiley, 1969.

Landy, D. The effects of an overheard audience'sreaction and attractiveness on opinion change.Journal of Experimental Social Psychology, 1972,8, 276-288.

McCroskey, J. C. A summary of experimental re-search on the effects of evidence in persuasivecommunication. Quarterly Journal of Speech, 1969,55, 169-176.

McGuire, W. J. Personality and susceptibility tosocial influence. In E. F. Borgatta & W. W. Lam-

766 SHELLY CHAIKEN

bert (Eds.), Handbook of personality theory andresearch. Chicago: Rand McNally, 1968.

McGuire, W. J. The nature of attitudes and atti-tude change. In G. Lindzey & E. Aronson (Eds.),The handbook of social psychology (2nd ed., Vol.3). Reading, Mass.: Addison-Wesley, 1969.

Miller, N. Involvement and dogmatism as inhibitorsof attitude change. Journal of Experimental SocialPsychology, 1965,1, 121-132.

Miller, N., Maruyama, G., Beaber, R. J., & Valone,K. Speed of speech and persuasion. Journal ofPersonality and Social Psychology, 1976, 34, 615-624.

Minium, E. W. Statistical reasoning in psychologyand education.. New York: Wiley, 1970.

Mintz, P. M., & Mills, J. Effects of arousal and in-formation about its source upon attitude change.Journal of Experimental Social Psychology, 1971,7, 561-570.

Norman, R. When what is said is important: Acomparison of expert and attractive sources.Journal of Experimental Social Psychology, 1976,12, 294-300.

Osterhouse, R. A., & Brock, T. C. Distraction in-creases yielding to propaganda by inhibitingcounterarguing. Journal of Personality and SocialPsychology, 1970,15, 344-358.

Pallak, M. S., Mueller, M., Dollar, K., & Pallak, J.Effect of commitment on responsiveness to anextreme consonant communication. Journal ofPersonality and Social Psychology, 1972, 23, 429-436.

Petty, R. E., & Cacioppo, J. T. Issue involvementcan increase or decrease persuasion by enhancingmessage-relevant cognitive responses. Journal ofPersonality and Social Psychology, 1979, 37, 1915-1926.

Petty, R. E., Wells, G. L., & Brock, T. C. Distractioncan enhance or reduce yielding to propaganda:Thought disruption versus justification. Journal ofPersonality and Social Psychology, 1976, 34, 874-884.

Rhine, R. J., & Severence, L. J. Ego-involvement,discrepancy, source credibility, and attitudechange. Journal of Personality and Social Psychol-ogy, 1970, 16, 175-190.

Sherif, M., & Hovland, C. I. Social judgment. NewHaven, Conn.: Yale University Press, 1961.

Sherif, C. W., Sherif, M., & Nebergall, R. E. Atti-tude and attitude change. Philadelphia, Pa.:Saunders, 1965.

Silverthorne, C. P., & Mazmanian, L. The effects ofheckling and media of presentation on the impactof a persuasive communication. Journal of SocialPsychology, 1975, 96, 229-236.

Snyder, M., & Rothbart, M. Communicator attrac-tiveness and opinion change. Canadian Journal ofBehavioral Science, 1971, 3, 377-387.

Stotland, E., & Canon, L. K. Social psychology: Acognitive approach. Philadelphia, Pa.: Saunders,1972.

Triandis, H. Attitude and attitude change. NewYork: Wiley, 1971.

Watts, W. A. Relative persistence of opinion changeinduced by active compared to passive participa-tion. Journal of Personality and Social Psychology,1967, 5, 4-15.

Zimbardo, P. G. Involvement and communicationdiscrepancy as determinants of opinion conformity.Journal of Abnormal and Social Psychology, 1960,60, 86-94.

Received June 18, 1979 •