Journal of Consumer Research, Inc.

Message Order Effects in Persuasion: An Attitude Strength Perspective Author(s): Curtis P. Haugtvedt and Duane T. Wegener Reviewed work(s): Source: Journal of Consumer Research, Vol. 21, No. 1 (Jun., 1994), pp. 205-218 Published by: The University of Chicago Press Stable URL: http://www.jstor.org/stable/2489751 . Accessed: 17/01/2012 19:23 Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

The University of Chicago Press and Journal of Consumer Research, Inc. are collaborating with JSTOR to digitize, preserve and extend access to Journal of Consumer Research.

http://www.jstor.org

Message

Order

Attitude

Strength

in

Effects

Persuasion:

An

Perspective

CURTIS P. HAUGTVEDT DUANE T. WEGENER* Contemporary research on attitude change processes is reviewed for implications regardingthe relative influence of successive opposing messages on finaljudgments. Based on this review, extent of message relevant elaboration is offered as a moderator of primacy versus recency effects in prior research. Support for this view is derived from the ability to explain the results of previous studies and from two experiments in which message presentation order and personal relevance of the topic are manipulated in a factorial design. We find that situations that foster high levels of message elaboration lead to greater influence of an initialmessage on final judgments (a primacy effect) whereas situations that foster low levels of message relevant elaboration lead to greater influence of a second message on finaljudgments (a recency effect).

The

overridinggoalof much consumerresearchover the past20 yearshasbeento gain an understanding of how persuasionvariablesinteractin the development of attitudes toward brands, services, and persons. Besides sharingthe goal of understandingwhen a particular combination of variableswill lead to the greatest amount of persuasion, most consumer research has employed a single-messagemethodology. That is, in most studies,participantsread or listen to a single persuasive message supportingone product or side of an issue. In the most common multimessage paradigm, individualsare exposedto severalmessagessupporting the same product or side of an issue (e.g., Gorn and Goldberg 1980; McCullough and Ostrom 1974). Although such studies provide important information about the factorsinvolved in persuasionsettings,these single-message, single-side methods cannot address questions about some very real consumer situations: those in whicha consumerfacescontradictorymessages from different sources on a single topic, product, or service. *Curtis P. Haugtvedt is assistant professor of marketing and Duane T. Wegener is a doctoral student in the Department of Psychology, Ohio State University, Columbus, OH 43210. Portions of this research were presented in a symposium at the Centennial Meeting of the American Psychological Association, August 1992, Washington, DC. The authors thank symposium participants Arie Kruglanski, Robert Lana, and Norman Miller for their helpful comments on this work. This work also benefited from comments provided by Timothy Brock, Frank Kardes, Thomas Ostrom, Richard Petty, Wendy Warren, David W. Schumann, and Stephen Smith. Finally, the authors thank the reviewers and the editor for their helpful comments and suggestions. Conduct of this research was facilitated by an Ohio State University College of Business Seed Grant awarded to C.P.H. and an NIMH Traineeship T32 MH19728 to D.T.W.

Consider,for example, an individual exposed to the mass media during the week of February6, 1989. By chance, he or she might have first encounteredinformation providedby consumer groups that a chemical used on apples and other fruits was unsafe and should be banned. Soon afterward,he or she might have encountered a message by other sources arguing that chemical treatmentsto apples and other producewere quite safe and were necessaryto insure the continued availabilityof insect- and disease-freefruits(see Evans 1989; Roberts 1989). Another person might have encountered the messages in exactly the opposite order. Sometimes,sourcesof messagesmight even have some control over whether their message is heard before or after their opponents. For instance, a company might know that a competingcompany has purchasedadvertising time in a given time slot (e.g., on television or radio). If so, that company can decide to purchaseadvertising for their product either before or after the competitor. One important question about such situations is the extent to which presentationorderof conflictingmessagesinfluencesthe position adoptedby the consumer. That is, although people might encounter the informationin one orderratherthan the other (either because of the strategyof the marketeror simply by chance), the orderin which that informationis presented might lead people to be more supportiveof one side of the issue or the other. Two effects of order have received attention (e.g., Hovland, Campbell, and Brock 1957). If people who encountertwo opposingmessagesformjudgmentsmore consistent with the first message, a primacy effect has occurred.If the judgment is more consistent with the second (opposing)message,however,a-recencyeffectis present.To the extent that simple orderof presentation

205 ? 1994by JOURNALOF CONSUMERRESEARCH,Inc.* Vol. 21 0 June 1994 All rightsreserved.0093-5301/95/2101-0014$2.00

206

JOURNAL OF CONSUMER RESEARCH

can affecta consumer'sfinaljudgmentsor preferences, it would be especiallyimportantto understandconditions under which primacy versus recency effects are likely to occur.

THE EXISTING LITERATURE Althoughconsumerresearchershave investigatedissues such as effectsof order of presentationof pricing information (Gray 1992; Monroe 1992) and product attributepresentation(Kardesand Herr 1990; see also McGuire 1957), no consumerresearchhas focused on the effectsof opposingmessageorder.Effectsof message orderhavebeen demonstratedin a varietyof persuasion settings, however.In this work, strong evidence exists for the occurrenceof both primacyand recencyeffects, although the conditions fostering each effect and the mechanisms underlyingeach outcome have not been established.' In the earliestresearchin the area, Lund (1925) presented subjects with a written communication either supportingor opposing implementation of protective tariffsand then presentedsubjectswith a communication on the other side of the issue. Lund observed a "disproportionate influence of the first discussion (message)in determiningsubjects' final position" (p. 187). The data, he suggested,supportedwhat he called a "law of primacy in persuasion" (see also Knower 1936). Some yearslater, Hovland and Mandell (1957) conducted a series of studies in the primacy-recency paradigm.One study using the tariff topic showed a significantprimacyeffect and another using the topic of atomicsubmarinesshoweda significantrecencyeffect (see also Cromwell 1950). Based on these findings, Hovlandand Mandell(1957) concludedthat they could discernno universallaw of either primacy or recency. Summarizingthe work on message order, Hovland (1957) stated that studies using well-known issues on which subjects had a great deal of prior knowledge would have a difficulttime in producing any kind of attitudechange,but hypothesizedthat "the nearerone comes to achieving primacy in the sense of the first presentationof unfamiliar material, the more one is apt to obtain primacyeffects"(p. 139). Lana conducted a series of studies that investigated Hovland's (1957) ideas. Unexpectedly, Lana (1961) found primacyeffectsunder conditions of high familiarity and recency effects under conditions of low familiaritywith an issue. In researchinterpretedas consistent with his 1961 findings, Lana (1963b) found a significantprimacy effect for a high-controversyissue (nuclearweapons),but not for a low-controversyissue 'Historically,researchon ordereffectsproceededalong two lines. One line of researchfocused on whetherthe first or second of two opposingmessageswould have a greaterpersuasiveeffect(e.g., Hovland and Mandell 1957). A second line of researchfocusedon positioningof argumentswithina singlepersuasivemessage(e.g.,McGuire 1957).The presentresearchfocuseson the formerratherthan latter line of research.

(Picasso).Finally,Lana(1963a) foundthat subjectswho reportedhigh interestin a topic exhibiteda significant primacy effect, whereas subjects who reportedhaving low interest exhibited a significantrecency effect. Althoughhis resultsconflictedwith Hovland's(1957) position, Lana presentedno frameworkwithin which to organizehis research.Lana hoped that similaritiesbetween the findingsfrom other studies and his findings could be used as a basis for empiricalgeneralizations. That is, he consideredhis studies as "buildinga body of empirical information . . . about which a theory of

communication might be developed" (Lana 1963a, p. 164). Otherresearchersat the time echoed Lana's(1963a) acknowledgment that no theoretical position had proven capableof organizingthe literatureon message order effects (e.g., Rosnow and Robinson 1967). Perhaps because of the lack of an organizingtheoretical framework,researchon the influencesof two opposing communications faded from view in the mid-1960s (Rosnow 1966). Some studies had obtained primacy effects; others using different message topics had obtained primacy,recency, or null effectsin variousconditions. An additionalcomplicatingfactorwas that levels of the factors under investigation had been confounded with message topic (e.g., controversy)or were identified through subject self-report(e.g., selfrated interest in the topic) rather than experimental manipulation. Thus, one is left unsure whether past message ordereffectswere even due to the conceptual factorsnoted by the researchersat the time. While most of the above studies were guided by a goal of understandingconditions underwhich primacy effectsmight be expected, Miller and Campbell(1959) focused on understandingconditions under which recency effectswere most likely to appear.2Drawing on the Ebbinghaus (1913) forgetting curve, Miller and Campbell hypothesizedthat recency effects would be due to differentialmemory for the messagearguments (i.e., the argumentsof the messagepresentedimmediately precedingthe attitudejudgment should be more likely to come to mind than the argumentspresented earlier, especially when some time separatedthe two messages).In examining this issue, Miller and Campbell's (1959) researchused a simulated court trial and includeda condition in which messageswerepresented 2During this period,a numberof additionalexplanationsof some primacyeffectswere also proposed.For example, primacywas hypothesizedto occur because of proactiveinhibition (e.g., Hovland 1951),decreasedattentionto laterinformation(e.g., Anderson1965), greatercriticalitytoward the second message (e.g., Hovland et al. 1957;Lund 1925),or changeof meaningof the secondmessage(e.g., Insko 1967).Most of the studiesfromwhich these explanationswere derived,however,do not fit the criteriafor the studies reviewedin this article-that is, two opposingmessagesfromtwo differentsources presentedclosein time. In addition,theseexplanationsdid not address when a particularprocesswould or would not be likely to occur nor wereany of these perspectivesable to organizethe literatureon messageordereffects(see Eaglyand Chaiken[1993] and Lana[1964] for reviews).

MESSAGE ORDER EFFECTS IN PERSUASION

close in time and attitude measurementoccurredimmediatelyand anothercondition in which the message presentationswere separatedby one week and attitude measurementoccurred immediately after the second message.In supportof theirhypothesis,a recencyeffect was obtained when messages were separatedby one week and attitudemeasurementimmediatelyfollowed the secondmessage.Interestingly,in the condition similar to the studies reviewed earlier(i.e., messagespresented close in time with immediate attitudemeasurement following the second message), Miller and Campbellfound no differencesas a result of message presentationorder. Unfortunately, the proposed mediator of order effects(i.e., memory for message arguments) did not predict final attitudesof participants(r = -.10, NS across all conditions). This lack of correlation betweenmemory and opinion has also been replicated by Insko (1964). Thus, although a variety of perspectiveshave been used to account for particular message order effect studies,n-oone frameworkhas appearedcapableof predicting when the various results of message order are likelyto occur(McGuire1966).If one linksthe variables studied in the message order literaturewith contemporarywork on attitude strength,however, consistent predictionsconcerningordereffectscan be made.

AN ATTITUDE STRENGTH EXPLANATION OF THE EXISTING LITERATURE One potentiallyuseful way to conceptualizeresearch on message orderis as a test of attitude strength-operationalizedhere as the degree of attitude resistance in the face of attack.3That is, if a first message is successfulin changingthe attitudeof the messagerecipient (or in creating an attitude if none exists prior to the firstmessage),the amount of changein responseto the secondmessageis an index of the strengthof the attitude following the first message (see Haugtvedt and Petty 1992).4Ourattitudestrengthperspectiveis derivedfrom the Elaboration Likelihood Model's (Petty and Cacioppo 1986) explicit recognition of the role of elabo'Although there are other potential consequences of attitude strength(e.g., attitudepersistenceor predictionof behaviorfrom attitudes;see Petty, Haugtvedt,and Smith 1994), resistanceappears mostpertinentto the studyof messageorderin thatopposingmessages constitutea directattackon one'sattitudefollowingan initialmessage. 4We use the term "resistance"to referto the amount of attitude changeevidencedin the face of an attack(i.e., high resistancerefers to little changein face of an attack,whereaslow resistancerefersto much change;see McGuire 1964). Importantly,we do not take this term to involve any necessaryassumptionsregardingthe process leadingto attitudechange.For example,one could have an attitude that is highlyresistantbecausethe personignoressubsequentinformation(perhapsbecausethe personthinks s/he alreadyhas enough informationon the topic and has high confidencein the opinion) or becausethe personactivelycounterarguesclaimsthat are againsthis or her currentattitude.

207

ration in creatingstrong attitudes (see Haugtvedtand Petty 1992; Petty et al. 1994). Thoughtful scrutiny (elaboration)of an initialmessagelikelyincludesgreater integrationof new informationwith existingknowledge structures(Pettyand Cacioppo 1986;see also Craikand Lockhart1972) and/or allows for more personaltransformations of information into the underlyingbelief structure(or schema)for an attitudeobject(Greenwald 1968). The process of elaborationmay also allow new information and feelings to come into contact with more points of existing information, enhancing the likelihood that the idiosyncraticelaborationsand information supportingthe attitude will become more availableand accessible(see Haugtvedtand Petty 1992; Pettyand Cacioppo1986).To the extentthat supportive cognitiveresponsesand otherpieces of informationare availablefrom prior elaboration,knowledgestructures supportiveof one's attitude may then be used in scrutinizing informationrelevantto the well-formedopinion. The use of supportiveinformation in processing an opposing message should then lead to resistanceto attitudechangein faceof the attack.This is whatseemed to occur in the Haugtvedtand Petty (1992) resistance study when subjects high in need for cognition (and thus most likely to scrutinize message content; Cacioppo and Petty 1982) were found to counterarguea weakattackingmessageto a greaterextentthan subjects low in need for cognition. Importantly,any variable that increases the amount of elaboration in the formation or change of an attitude should lead to similar increasesin attitudestrength,which can manifestitself as increasedresistanceto change in the face of attack (Haugtvedtand Petty 1992;Petty and Cacioppo 1986).5 One mightconceptualizeprimacyeffectsas involving high levels of attitude strength(and thus resistanceto change)for attitudesformedor changedduringthe first messageand recencyas involving low levels of attitude strength and resistance. If so, then primacy effects should be most likely when motivation and ability to elaborate the initial message are high. That is, when facing an attack, people will likely defend the strong position developed from processingthe firstmessage. On the other hand, recency effects should be most likely when motivation and ability to elaboratethe initial message are low (when low levels of processingof the first message may result in a relatively weak attitude). This might allow processesthat involve the second message to have an impact-for example, recall and use of recentlypresentedinformationas a basis for a final attitude. That is, althoughmemory for message 'Unfortunately,althoughexistingattituderesistancestudiessupportthe viewthat highlevels of elaborationof an initial messagelead to greaterresistance(lesschangein the faceof an attack),thesestudies cannotdirectlyaddressthe role of resistancein messageordereffects. Thisis becauseresistancestudieshaveemployedonly a singlemessage order,with an attackmessagethat is weakerthan the initial message. Becausemessageorderexperimentsuse pro and con messagesthat areof equalstrength,the messageorderparadigmconstitutesa strong test of attitudesstrength/resistancenotions.

208

argumentshas not proven to be a general explanation of all ordereffects(see earlierdiscussionof Insko [1964]; Millerand Campbell[1959]), recentresearchhas found that memory for messageargumentsshould play a role in attitude change processesprimarilyin low-elaboration settings (when recency effects appear most likely to occur). For example,the correlationbetween the favorabilityof postmessageattitude (measuredimmediately after message presentation) and the amount of information recalled is generally higher under conditions that encouragelow ratherthan high elaboration of messagecontent (Mackie and Asuncion [1990]; for similar results see Haugtvedt and Petty [1992]). The generallyacceptedexplanationis that participantsoperatingunderlow-elaborationconditions had not fully formedan opinion until the attitudinalinquirywhereas high-elaborationparticipantshadevaluatedthe message content and formedtheir opinion on-line (e.g., see Hastie and Park [1986], and Lichtensteinand Srull [1985, 1987], regardingon-line vs. memory-basedimpression formation).Thus, if recency effects occur when elaboration is low, one reason for this might be reliance on message content when formulating a response to the attitudinalinquiry.If so, memory for messagecontent (especially memory for recently presented material [Wyerand Srull 1980], that is, argumentsin the second message) should be positively related to attitudes for subjectslow in elaborationlikelihood. There should be little or no relation between memory for recently presented argumentsand final attitude for subjectsunder high-elaborationconditions. By categorizingpast studies of message order accordingto variablesthat influencethe extent of message processing,patternsof findings consistent with an attitude strengthperspectiveare revealed. For example, confound with messagetopic and self-selectionof subjects notwithstanding,it appearsthat highlevels of topic familiarity,controversy,and interestweregenerallyassociatedwith primacyeffects.That is, topics for which subjectswere either made knowledgeable(e.g., vivisection; Lana 1961)or topics that werelikely to be familiar and controversial(e.g., protectivetariffs[Hovlandand Mandell 1957;Lund 1925];prohibition[Knower1936]; nuclearweapons [Lana 1963b]) producedprimacy effects. Similarly,subjectswho reportedhigh interest in a topic showed primacyeffects(Lana 1963a). Recency effects, on the other hand, appearedmore likely when individualsknew little about the topic (Lana 1961) or showed little interestin the topic (Lana 1963a). Results of contemporaryresearchsuggestthat variables like familiarity,controversy,or interestare likely to influencethe extent of messageelaboration.For example, researchhas shown that messagesare processed more thoroughly when participants have high prior knowledge as opposed to low prior knowledge (e.g., Wood, Kallgren,and Priesler 1985). When topics are made more interesting or involving (controversial), messagesarealso processedmorethoroughlythan when the topics are made less interestingor involving (e.g.,

JOURNAL OF CONSUMER RESEARCH

Petty and Cacioppo 1979; Petty, Cacioppo, and Schumann 1983). It appearsthen, that, when organizedby factorsthat have been found to influencemessageelaboration,the existingliteraturecan be viewed as consistent with an attitude strengthperspectiveon message ordereffects. The ability to conceptually organize the studies reviewedhereprovidesinitial supportfor our perspective. Importantly,this theoreticalposition has not been previously offered as a way to organize the literatureon opposing persuasive messages from distinct sources. Support for our conceptualization,however, is based on effectsobservedacross studies and on assumptions madeaboutthe effectsof variablespresentin the studies. Strongersupport for our conceptualization could be obtained throughmanipulation of variablesknown to influence the amount of elaborativeprocessinggiven to the messagesand through the examination of data that measure possible processes underlying effects of messageorder. For instance,manipulationsthat influencethe extent of elaborationgiven to messagecontent should also influence recipients'cognitive responses to the message (e.g., Pettyand Cacioppo1979).Thus, if primacyeffects are due to effortful evaluation of an initial message, then message order should not only influence final attitude but should also influence measuresof cognitive responses.If recencyeffectsgenerallyoccur when elaboration likelihood is low, then message order should not have much influenceon cognitiveresponses.In addition, if high-elaborationsubjectsare using the informationalbase supportingtheir attitudeconsistentwith the firstmessagewhenthey evaluatethe secondmessage, then cognitive responsesof these people might show a larger number of direct counterarguments toward statementsin the second messagethan low-elaboration subjects(i.e., directquestioningor refutationof the validity of argumentsin the second message should be higher for high- than for low-elaboration subjects). Thus, our processing/attitudestrengthperspectiveleads to the following hypotheses about attitudes following the presentationof two conflictingmessages: Hla: Under conditions fostering high message elaboration,finalattitudeswill showprimacy effects(more favorableattitudesfollowing a pro/con than a con/pro messagesequence). Hlb: Under conditions fostering low message elaboration,finalattitudeswill show recency effects(more favorableattitudesfollowing a con/pro than a pro/con messagesequence). A number of auxiliarypredictionsare also made. H2a: Message sequence will influence cognitive responses more under conditions fostering high rather than low message elaboration. That is, pro/con message order will lead to morefavorablecognitiveresponsesthancon/ pro messageorderto a greaterextent under

MESSAGE ORDER EFFECTS IN PERSUASION

high- as opposed to low-elaborationconditions. H2b: Message recipients are more likely to counterargue the second message under high- as opposed to low-elaborationconditions. H2c: Final attitudes under low-elaborationconditions will be positively associatedwith recall of second messageargumentsto a greater extent than final attitudes under high-elaboration conditions.

EXPERIMENT 1 For our first experiment examining message order effectswe directlymanipulatedlikelihoodof elaboration (via personalrelevance of the messagetopic; see Petty and Cacioppo 1979) alongwith messageorder.In order to do this, we chose a topic that has been commonly employed in initial persuasionresearchand for which manipulationsof level of processingarewell established: the topic of institutingsenior comprehensiveexams as a graduationrequirement(see Pettyand Cacioppo1986; Petty, Cacioppo,and Goldman 1981). We manipulated messageorderand issuerelevancein a 2 (messageorder: pro/con, con/pro) X 2 (relevance:high, low) factorial design. Subjectswere first exposed to a message containing eitherpro or con argumentson the exam issue. Students were then presented with a second message that alwaysarguedin the opposite directionof the first. The topic was introducedas either being consideredat the students' university for implementation the next year (high personalrelevance)or being consideredat a distantuniversityfor some time in the future (low personal relevance).

Method Subjectsand Design. Fifty undergraduatestudents participatedfor extra credit in their introductorymarketing course. Subjectswere run in one classroomsession. A random orderingof experimentalpackets assigned studentsto the 2 (messageorder:pro/con, con/ pro) X 2 (personalrelevance:high, low) between-subjects design. Procedure. An introductory sheet informed students that they would be readingexcerptsfrom various publicationsand that, after reading the articles, they would be askedto evaluateeach on a varietyof dimensions, includingwritingstyle, use of grammar,and ease of reading.Studentswere presentedwith two opposing messageson the issue of institutingsenior comprehensive examsas a prerequisitefor graduationfromcollege. The pro and con senior comprehensiveexam messages were developed from messages used in past research (e.g., Petty et al. 1981). Pretests showed that the pro and con messagespresentedindividuallywere equally strong(i.e., producedapproximatelythe same propor-

209

tion of thoughts in support of the advocated position when pretestsubjectswereexplicitlyinstructedto think carefully about the argumentscontained in the messages;see Petty and Cacioppo 1986). The pro message contained five strong argumentsin support of senior comprehensiveexams and the con message contained five strongargumentsagainstthe exams. The messages were equal in length (approximately420 words each).6 IndependentVariables. Half the subjects received messagesin a pro/con orderand half in a con/pro order. An introductorystatement at the top of the page immediatelyprecedingthe firstmessageinformed half of the subjectsthatthe institutionof comprehensiveexams was being discussedby the new presidentof their university for possible implementationbeginning the following spring(high personalrelevance).The other half of the subjectswereinformedthat seniorcomprehensive exams were being consideredat a distantuniversityfor possible implementation beginning the spring of two school years in the future (low personalrelevance). DependentMeasures. Followingthe two messages, subjects completed measures assessing their attitudes towardthe implementationof the exams.Attitudeswere measuredon four 11-point semantic differentialscales assessinghow bad/good, foolish/wise, harmful/beneficial, and negative/positivethe students believed institution of the exams would be. Subjectswere also asked if institutingthe examswouldbe helpfulon a scalefrom 1 = "not very helpful" to 11 = "very helpful." Three questions assessedthe perceivedrelevanceof the exam issue on scales anchoredat 1 = "not very importantto me," "not personally relevant," and "will not affect me" and 11 = "very important to me," "personally relevant,"and "will affectme," respectively. Cognitive Responses. Subjects next encountered printedinstructionsdirectingthem to turn to the next page. Subjectswere instructedto write down all of the thoughtsthey could recall going throughtheir mind at the time they read the first message about senior comprehensiveexams;seven double-spacedlines were provided, and subjects were asked to take no more than two minutes for the task (thoughthe actual time taken was not monitored). On the middle of the same page, subjectswere instructedto performthe same thought listing for the second message. Recall. Instructionson the bottom thirdof the page asked subjectsto write down all of the argumentsthey 6For example, strong pro exam arguments included that (a) students graduating from schools with comprehensive exams were more likely to be accepted into very good graduate schools, (b) students from institutions with comprehensive exams find better jobs, and (c) starting salaries of students who had taken comprehensive exams were $2,000$3,000 higher than those of students who had not taken the exams. Strong arguments against the exams included that (a) capturing the benefits of a four-year program on a single standardized exam would be difficult, (b) the expectation of having to take such an exam may stop some students from entering college, and (c) taking an exam at the end of each course should be sufficient.

210

could recallthat werepresentedin favorof and opposed to the institutionof seniorcomprehensiveexams.When all subjects had finishedthe packet, they returnedthe materialsand were thoroughlydebriefedand thanked for their participation.

JOURNAL OF CONSUMER RESEARCH FIGURE 1 FINALATTITUDEAS A FUNCTIONOF PERSONAL RELEVANCE AND MESSAGE ORDER: EXPERIMENT1 Attitude (11 = Pro)

Results RelevanceManipulationCheck. A composite measure of perceivedpersonal relevance was constructed by averagingresponsesto the three perceivedpersonal relevancequestions(a = .89). This composite was submitted to a 2 (messageorder) X 2 (personalrelevance) ANOVA. Therewas only a main effectfor personalrelevance, F(1,46) = 4.38, p < .04, whereby students in the low personalrelevancecondition viewed the exam issue as less relevant (X = 5.51) than students in the high-relevancecondition (X = 7.36). Attitudes. A composite measure of attitude was constructedby averagingthe responsesto the four semantic differentialsalong with the question askingsubjects how helpful the exams would be (a = .97). This compositemeasurewas submittedto a 2 (messageorder) X 2 (personalrelevance)ANOVA.The analysisrevealed only the predictedmessageorder X personalrelevance interaction,F(1,46) = 8.89, p < .004; see Figure 1. That is, subjectstold that the exams were being considered by their universityfor the followingyear (highpersonal relevance)weremore favorabletowardthe exams after receiving the pro/con message order (X = 6.01) than after receiving the con/pro order (X = 3.95; F(1,46) = 4.58, p < .02). We thus find supportfor Hypothesis la. In contrast,subjectswho were told that the exams were being consideredby a distant university for two years in the future (low personal relevance) were less favorabletowardthe exams afterreceivingthe pro/con messageorder(X = 3.49) than after receivingthe con/ pro order (X = 5.58; F(1,46) = 4.33, p < .025).7 We thus find supportfor Hypothesis lb. CognitiveResponses. Cognitiveresponsemeasures were included in an attempt to discern patterns of thoughtsthat might characterizemessageordereffects. Thoughts were categorizedas favorable,unfavorable, or neutralwith regardto the exam topic by two judges unawareof subjects'relevancecondition. In addition, if thoughts in response to the second message were judged to be explicitly attackingthe validity of an argument in the message, these thoughts were classified as direct counterarguments.The judges agreedon over 80 percentof the thoughts;disagreementswereresolved by discussion. The favorabilityof thoughts toward the institution of seniorcomprehensiveexamswas indexedby dividing 7Because we have strong a priori expectations regarding when primacy as opposed to recency effects should occur, probability levels for the simple effects of message order within relevance conditions are one-tailed for both studies. This is the case for cognitive responses and counterarguments as well.

6.0

5.0

4.0 -a---

Pro/Con Con/Pro

3.I

Low

High Level of Relevance

the number of thoughts favorableto the exams by the total number of thoughts generated(collapsed across both messages).Whereasfavorabilityof thoughts was no differentfor low-relevancesubjectsreceivingthe pro/ con and con/pro message orders (X = .180 and .182, respectively),therewasa tendencyin the high-relevance conditions for message order to affect thoughts in the expecteddirection.That is, high-relevancesubjectswho received the pro/con order reported proportionally more favorablethoughtstowardsenior comprehensive exams (X = .246) than high-relevancesubjectswho received the con/pro order (X = .165). Unfortunately, neitherthe interactionnorthe simpleeffectsapproached statisticalsignificance.8 The second potential indicator of active processes fosteredby high personal relevance fared a bit better. That is, subjects in the high-relevancecondition produced a higher number of direct counterargumentsof informationin the second message(X = .65) than subjects in the low-relevancecondition (X = .33; F(1,46) =

2.05, p < .08; supportive of Hypothesis 2b). Consis-

tent with the use of strong argumentsin our pro and con messages,therewas no differencein counterarguing 8Although we clearly expected differences in counterarguing of the second message (see Hypothesis 2b), we did not have any firm expectations regarding favorable evaluations of the content of the second message. That is, although we would expect relatively few support arguments in response to the second message for high-relevance participants, this does not necessitate any difference in support arguments between high- and low-relevance conditions. This is because we do not expect low-relevance participants to actively process message content and thus do not expect them to generate many support arguments. Indeed, the mean number of favorable thoughts generated in response to the second message did not differ for subjects in high(X = 1.19) and low- (X = 1.46) relevance conditions.

MESSAGE ORDER EFFECTS IN PERSUASION

of the firstmessageby high- versus low-relevancesubjects.9 ArgumentRecall. No main effects or interactions were observedon the number of argumentsrecalled. On average,participantsin the low-relevancecondition recalledan averageof 2.97 argumentsin favor of the exams and 2.54 argumentsagainst the exams. Participants in the high-relevanceconditions recalledan averageof 2.81 argumentsin favor of the exams and 2.90 argumentsagainst the exams. Within each relevance condition, there was no effect of orderon recall. Relationship of Recall and Attitude. Research on the relationshipof verbatim recall and judgment has consistentlyshown that, when individualsengagein little on-line evaluation, the correlation between recall and attitudeis significantand positive (Haugtvedtand Petty 1992;Mackie and Asuncion 1990). Basedon our conceptualizationof the processesunderlyingrecency effectsand previousworkon recall-judgmentsrelationships, in Hypothesis 2c we hypothesizedthat final attitudesunderlow-elaborationconditionswould be positively associated with recall of second message argumentsto a greaterextent than final attitudesunder high-elaborationconditions.To test this hypothesis,recall of the second message was correlated with final judgmentsaftercollapsingacrosslevels of messagepresentation order. The number of arguments recalled againstthe exams was coded as a negative number (to ensurethat relianceon argumentrecallis indexed by a positive recall-attitudecorrelation). Favorabilityof attitudes under low-relevance conditionswaspositivelyrelatedto argumentrecall(r = .48, p < .01), whereasfinaljudgmentsunderhigh-relevance conditions were not significantlyrelated to final judgments (r = -.33, p < .10). If anything, it appearsthat greaterrecallof second messageargumentsunderhighrelevanceconditions is negativelycorrelatedwith final judgments-consistent with the greater tendency of high-relevance subjects to counterargue the second message. Such processeswould make the information 9Usingdifferentindices of the favorabilityof cognitive responses showedsimilarpatternsof results.That and of counterargumentation is, the numberof favorablecognitive responsestends to be greater followingpro/con as opposedto con/pro messageorder(Xs = 1.79 and .75, respectively)for high-relevanceparticipants.For low-relevanceparticipants,thereis little differencein the numberof favorable cognitive thoughtsin responseto pro/con versus con/pro message orders(Xs = .92 and .82, respectively).The number of valenced thoughtsgeneratedduring the thought-listingperiod tended to be greaterunderhigh- (X = 4.85) as opposedto low- (X = 3.87) elaborationconditions.Also, the percentageof message-relatedthoughts in responseto the secondmessagethatweredirectcounterarguments tended to be higherfor high- as opposed to low-relevancesubjects (Xs = 23 percentand 17 percent,respectively).None of the patterns reachedsignificance.However,consistentwith the idea that finalattitudeswouldbe basedmoreon elaborationsof initialmessagecontent underhigh-than underlow-relevanceconditions,the numberof favorablethoughtsgeneratedin responseto the firstmessagecorrelated significantlywith attitude under high- (r = .501, p < .01) but not underlow-relevanceconditions(r = .086, NS).

211

memorable, but the fact that it was successfully counterarguedwould lead to its negative relationship with the finaljudgment.Consistentwith this possibility, when a partialcorrelationbetween attitude and recall was computed(controllingfor counterargumentsto the second message), the negative correlation becomes nonsignificant(r = -.05, NS), and the positive correlation in the low-relevance condition remains unchangedand significant.We thus find supportfor Hypothesis 2c.

Discussion Results of experiment 1 provide the first direct evidence in supportof the attitudestrengthinterpretation of past researchon messageordereffectsin persuasion. A manipulationof participantmotivation determined whetherthe firstor secondmessagehad a greaterimpact on attitudes. When personal relevance of the message topic was high (i.e., motivation to processwas high), a significantprimacyeffectwas obtained.Whenrelevance of the messagetopic was low (i.e., motivationto process was low), a significant recency effect was obtained. There was also a nonsignificanttendency for message order to influence thought content more under highthan under low-relevance conditions. Unfortunately, conditionsin this studywerenot particularlyconducive to adequate collection of cognitive responses. For instance, because subjectswere all seatedin a largeclassroom and were likely finishingat differenttimes, subjects who finished early may have distractedsubjects who were still reportingtheir cognitive responses. In addition, the small number of subjects in the current studymade it less likelythat differencescould be found. Although the senior exam messagehas proven to be an extremelyuseful topic in persuasionresearch,there area numberof importantdifferencesbetweenthis topic and more typical consumer-relevantmessages.One of these differencesis the fact that the message topic is fairly unique, and students may not have given much thought to the topic prior to the experimentalstudy. Thus, it is somewhat similar to a new product introduction. Becauseof this, the comprehensiveexam issue might be one that is particularlysensitive to manipulations of personalrelevance. In orderto insure that these resultsare not only obtained with topics that might be characterizedas low in familiarity,we conducteda second experimentusing a topic more familiar to our subjects:nuclear power. Becausethere are a numberof nuclearpowerplants in the subjects'home state and becauseboth pro- and antinuclearpowerpositionshave receivedattentionin the local and national press (e.g., Columbus Dispatch 1990a, 1990b) and in advertisements(Greenwald1991), and because ordereffectshave been hypothesizedto be more likely for unfamiliartopics (Hovland 1957), use of the nuclearpowertopic was seen as a relativelystrong test of our hypotheses.

JOURNAL OF CONSUMER RESEARCH

212

EXPERIMENT 2 Undergraduatemarketingstudentswerefirstexposed to a message presentingeither pro or con arguments and then encountereda second messagethat alwaysargued in the opposite direction. In order to provide a more controlledenvironmentfor experiment2, participation took place in individualcubicles, providingvisual and acousticalisolation.

Method Subjectsand Design. One hundredthirty-sevenintroductorymarketingstudents participatedin the experiment for extra course credit. Subjectswere run in groups of two to four. Each subject was randomly assigned to the 2 (messageorder:pro/con, con/pro) X 2 (personalrelevance:high, low) between-subjectsdesign. Messages. Pro and con versions of the nuclear power messages were developed by combining arguments generatedby marketingstudents with information from the popular press (e.g., Time). Subsequent pretestswith additionalsubjectswho receivedonly the pro or con messageshowed that the pro and con messages were equally strong (that is, thinking about the materialin the messagesproducedsimilar numbersof thoughts in support of the advocated positions). The pro messagecontained seven strongargumentsin support of nuclearpowerplants, and the con messagecontained seven strong argumentsagainst nuclear power plants.10The pro and con nuclearpower messagescontained approximately300 words each.

All dependent measureswere presentedafter subjects had read all three messages. IndependentVariables. Half of the subjectsreceived the nuclear power messagesin the pro/con order and halfin the con/pro order.Priorto each message,subjects were informedof the upcomingtopic; for the recycling message and the first nuclear power message subjects weregiven backgroundinformation.For example,they were informed that the first message was on the topic of recycling-"a topic of currentinterestin many parts of the country."Low personalrelevanceof the nuclear energytopic was createdby informing half of the subjects that recently released Federal Energy Program documents proposedthat new nuclearpowerplants be built in distant states (i.e., Texas, Arizona, and New Mexico). In the high-relevancecondition, the subjects' own and nearbystates(i.e., Ohio, Michigan,and Pennsylvania) were listed instead. This manipulation was validated in a pretest with students from the same course."1Studentsin the pretestwere ineligible for the final study. Dependent Measures. The first few questions focusedon recycling.The remainderof the questionswere attitude questions regardingthe nuclear power issue. Questions were answeredby clicking the "mouse" inside a long narrowbox (approximately5 inches by .25 inch) at the bottom of the computerscreen.Scaleswere anchoredby descriptionstyped below the box at each end. Althoughthe box was not labeledwith numerical values, the computerrecordedeach responseas having occurredwithin one of 11 equally spaced sections of the box. Thus, each responsewas assigneda scale value by the computer. Each of the attitude questions began with the statement "In general,I think building more nuclearpower

Procedure. Studentswereseatedin privatecubicles containingan Apple MacintoshIlsi computer and 13inch color monitor. All instructions,messages,and dependent measureswerepresentedusing the application softwareHyperCard.An introductoryscreeninformed studentsthat they would be readingexcerptsfrom various publicationsand that, after readinga series of articles, they would be askedto evaluateeach on a variety of dimensions,includingwritingstyle, use of grammar, and ease of reading.They were asked to read the excerptsas they wouldany newspaperor magazinearticle. Subjectsfirstread a message on the topic of recycling; the next two messagesregardednuclear power plants.

with the anchorsbad/good, unwise/wise,harmful/beneficial,and negative/positive.Afterthe computer-based questionnaire,an experimenterhanded each subject a sheetcontainingcognitiveresponseand recallprotocols. The thought-listingand recall instructions were identical to those used in experiment 1. Aftersubjectscompleted the dependent measures, they were debriefed, thanked, and dismissed.

10 Pro argumentsincluded(a) nuclearplants would replacefossil fuel plantsthatproducecarbondioxide,thuslesseningthe greenhouse effect, (b) nuclear plants are prevalentin.other countries, such as France,whichproduces70 percentof its energyfromnuclearplants, (c) U.S. nuclearplants have operatedfor 20 years without serious incidents,and (d) currentstandardsfor safety are very high. Arguments againstadditionalnuclearplantsincluded:(a) problemswith the technologyand the lack of safe disposalof wastesmake nuclear an unwise choice, (b) there is no guaranteethat a disposal site will remainunchangedor that groundwaterwill not be contaminated,(c) the currentsafetytolerances,thoughthey soundimpressive,areonly arbitrary,and (d) nuclearproponentsoften fail to consideran alternative to continuallyincreasingpower supplies-the wise use and conservationof existingresources.

"Subjectsin this pretestweregiven informationaboutthe nuclear power topic that was identical to that presentedto subjectsin the study to follow. Immediatelyfollowing receiptof this information, subjectswere asked to what extent they believed that the issue of building more nuclear power plants could affect them personally, wasimportantto them personally,and was relevantto them personally, on scales anchoredat 1 = "not at all," "not at all important," and "not at all relevant"and 5 = "very much," "veryimportant," and "veryrelevant,"respectively.Theseitemswerecombinedto form a relevanceindex (a = .87). Subjectsexposed to the high-relevance manipulationfelt the issue was more relevant to them (X = 4.20) than subjectsexposedto the low-relevancemanipulation(X = 3.24); F(1,57) = 24.27, p < .000 1.

plants is. . ." Four semantic differentials were formed

MESSAGE ORDER EFFECTS IN PERSUASION

213

Results

FIGURE2

Attitudes. A composite measure of attitude was constructedby averagingresponsesto the four attitude items (a = .94). The attitude index was submittedto a 2 (message order) X 2 (personal relevance) ANOVA. Replicatingexperiment1, the only significanteffectwas the predictedmessage order X personal relevance in-

FINALATTITUDEAS A FUNCTIONOF PERSONAL RELEVANCE AND MESSAGE ORDER: EXPERIMENT2 Attitude (Pro -11) 8

teraction, F(1,133) = 6.36, p < .001; see Figure 2. As

expected,subjectsin the high-relevancecondition exposed to the pro/con message order were significantly more favorabletoward nuclearpower (X = 7.94) than were high-relevancesubjects exposed to the con/pro message order (X = 5.96; F(1, 133) = 12.21, p < .001). Also as expected,subjectsexposedto the messagesunder conditions of low relevancewere less favorabletoward nuclearpowerafterreceivingthe pro/con messageorder (X = 5.80) than after receiving the con/pro message

7

6 0

Pro/Con

;

Con/Pro

order (X = 7.11; F(1,133) = 5.49, p < .01). 5

Cognitive Responses.

As in experiment 1, thoughts

were categorizedas favorable,unfavorable,or neutral with regardto the nuclear power topic by two judges unaware of subjects' relevance conditions. Also, thoughtsthat werejudgedto be explicitlyattackingthe validity of an argument in the second message were classifiedas directcounterarguments. Thejudgesagreed on over 86 percent of the cases. Disagreementswere resolvedby discussion. The favorabilityof thoughts toward nuclear power plants was indexed by dividing the total number of thoughts favorableto the issue by the total number of thoughts(collapsedacrossboth messages).The pattern of thought favorabilitywas nearly identical to that of experiment1. The only overalleffecton favorabilityof thoughtswas a messageorder X personalrelevanceinteraction,F(1,133) = 4.07, p < .05. The proportionof favorablethoughts for subjects receiving the pro/con (X = .22) and the con/pro message orders (X = .254)

was nearly identical for the low-relevanceconditions (F < 1). In contrast,high-relevancesubjectsin the pro/ con ordercondition were significantlymore favorable toward nuclear power (X = .332) than high-relevance subjects in the con/pro order condition (X = .209; F(1,133) = 5.03, p

<

.02). 12

In addition, analysesof the extent to which subjects arguedagainststatementsin the secondmessageshowed only the predictedeffectof personalrelevance,F( 1,133) = 6.18, p < .007. That is, regardlessof order, subjects in the high-relevance condition exhibited a higher numberof explicitcounterargumentsto informationin the second message(X = .418) than did subjectsin the low-relevancecondition(X = .1 14). Consistentwith the 12Asin experiment 1, we did not have any firm expectationsregardingfavorableevaluationsof the content of the second message (see n. 8 above). Similarto experiment 1, the mean number of favorablethoughtsgeneratedin responseto the second message did not differforsubjectsin high-(X = 1.15)andlow-(X = 1.34)relevance conditions.

,

I

Low

High Level of Relevance

use of strong argumentsin our pro and con messages, there was no differencein counterarguingof the first messageby high- versuslow-relevanceparticipants."3 ArgumentRecall. No main effects or interactions were observed on the number of argumentsrecalled. Participantsin the low-relevancecondition recalledan average of 1.93 argumentsin favor of nuclear power and 1.65 argumentsagainstnuclearpower.Participants in the high-relevanceconditions recalledan averageof 1.96 argumentsin favor of nuclearpower and 1.53 ar"3Usingdifferentindicesof the favorabilityof cognitiveresponses and of counterargumentation showedsimilarpatternsof resultsand statisticalsignificance.That is, the number of favorablecognitive responsesweregreaterfollowingpro/con as opposedto con/pro message order(X = 2.67 and 1.47, respectively)for high-relevanceparticipants.For low-relevanceparticipants,there is little differencein the numberof favorablecognitive thoughtsin responseto pro/con versus con/pro message orders(Xs = 1.94 and 1.66, respectively). The percentageof message-relatedthoughtsin responseto the second messagethat were direct counterargumentswere higherfor high- as opposedto low-relevancesubjects(Xs = 13.2percentand4.67 percent, respectively).As in experiment1, consistentwith the idea that finai attitudes would be based more on elaborationsof initial message contentunderhigh-thanunderlow-relevanceconditions,the number of favorablethoughtsgeneratedin responseto the firstmessagecorrelatedsignificantlyunder high- (r = .40, p < .001) but not under low-relevanceconditions (r = -.007, NS). High-relevancesubjects also tended to generatea greaternumber of valenced thoughts (X = 4.38) than low-relevancesubjects(X = 3.81), although this was not significant.To the extent thatthe manipulationof messageorder influencesnot only attitudesbut also cognitive responses,however, a thoughtfulprocess becomes a more plausible mediator of those ordereffects.This would be the case regardlessof the total number of valencedthoughtsgeneratedin a thought-listingprotocol(see Petty and Cacioppo[1986, pp. 38-39] for a more thoroughdiscussionof inferencesregardinguse of thought-listingsas assessmentsof amount of messageprocessing).

JOURNAL OF CONSUMER RESEARCH

214

guments againstnuclearpower.Within each relevance condition, there was no effect of orderon recall. Relationship of Argument Recall and Attitude. Correlationsbetween favorabilityof attitudes and the numberof argumentsrecalledfrom the second message were computed. Replicatingexperiment 1, under conditions of low relevance,recall of second message argumentswaspositivelyrelatedto finalattitudes(r = .34, p < .004). In contrast, a negative correlationwas observed in the high-relevance condition (r

= -.29,

p <

.02). As in experiment1, this mightbe the case because subjects in the high-relevanceconditions engaged in counterargumentationof the second message. Consistent with this possibility,when a partialcorrelationbetween attitudeand recallwas computed(controllingfor counterargumentsof the second message),the negative correlationbecomes nonsignificant(r = -.19, NS) and the positive correlationin the low-relevancecondition remainsunchangedand significant.

Discussion The resultsof experiment2 providedstrongconvergent supportfor our hypothesizedexplanation of past studiesof messageordereffects.Replicatingexperiment 1, we found a significantprimacyeffect when message recipientsencounteredthe materialsunder conditions of high personalrelevanceand a significantrecencyeffect whenmessagerecipientsencounteredthe materials under conditions of low personalrelevance. In addition,resultsprovidedconvergentevidence regardingthe natureof processesthat led to primacyand recencyeffectsin our experiments.In both experiments, message order tended to influence the favorabilityof cognitiveresponsesonly for subjectswho read the materialsunder conditions of high personalrelevance.In addition, high-relevancesubjects tended to counterargue the second message more than low-relevance subjectsin both studies. This differencein counterarguing did not occur in response to the first message, however, suggestingthat counterargumentswere elicited in defense of relativelystrongattitudes.Finally, in both experiments,the number of arguments recalled from the second messagewas positively relatedto the positivity of the final attitude for subjectsin low-elaboration conditions, but not for subjectsin high-elaborationconditions. Thus, supportfor each of the formal hypotheseswas obtained.

GENERAL DISCUSSION Reviewsof the messageorderliterature(e.g., Rosnow 1966; Rosnow and Robinson 1967; our introduction section) have noted that past studies often did not have cleara prioripredictions.In addition,most past studies findingordereffectsconfoundedeithersubjector topic factorswith the conceptualfactorsof interest(e.g., classifyingtopic interestthroughparticipantself-report,or varying controversythrough presentationof different

topics) and typically did not find primacyand recency effectsin the same study.A coherentpictureof the message order literatureemerged when past order effects wereconsideredwithin an attitudestrengthframework, however.By relatingconditions in past studies of messageorderto recentworkon messageprocessingin persuasion, we found that past messageordereffectscould be usefully organizedby our model. Importantly,two new experiments in which message recipients' motivation to elaborate message content was directly manipulatedsupportedour views. When participantswere motivated to elaborate on messagecontent, primacyeffects occurred.When participants were unmotivated, recency effects occurred. Thus, we found and replicated primacy and recency effects when participantswere randomly assigned to high- versuslow-elaborationconditions. In addition to providinga theoreticalposition that appearscapableof organizingpast studieson opposingmessagespresented close in time, resultsof the presentstudiesalso provide empiricalsupportfor the attitudestrengthperspective's a priori predictions of when first messages or second messagesshould have the most impact.

Future Research Persistenceof OrderEffects. Althoughthe research reportedhereis an importantfirststep in understanding effects of message order, additional researchcan help us betterunderstandthe processesunderlyingthese effects. For example, because strong attitudes are more likely to be the result of effortful as opposed to noneffortful processing of topic-relevant information (Haugtvedtand Petty 1992;Petty et al. 1994), message ordereffectsthat occur via effortfulmeans should tend to be more persistentthan order effects that are relatively noneffortful.Thus, from the currentperspective, primacy effects due to effortfulprocessingof early information (and counterarguingof later information) should tend to persistlongerthan recencyeffectswhich result from noneffortfulreliance on the number of arguments that come easily to mind on a topic. An important next step in order effects research,therefore, will be examination of the temporal persistenceof attitudesformedor changedundervariousmessageorder conditions. DifferentProcessesLeadingto OrderEffects. Future researchshould also address conditions under which effortfulas opposed to noneffortfulprocesses may be responsiblefor primacyand recency effects.Although we have arguedthat primacyeffectsin currentand past messageorderstudies are likely the resultof high levels of messageelaboration,we do not mean to imply that other (yet unexplored)factorsmight not also affectthe existence of messageordereffectsand the types of processes that might be responsiblefor those effects. For example, some conditions (such as those that foster "6epistemicfreezing"; Kruglanski and Freund 1983) may exist where noneffortfulprocessescould result in

MESSAGE ORDER EFFECTS IN PERSUASION

primacy. Thus, if opposing messages were presented underspecificconditionssuch as time pressure(Mayseless and Kruglanski1987) or were given to individuals who score high in dispositional "need for closure" (Websterand Kruglanski1992), primacyeffectsdue to ignoranceof laterinformationmightoccur.Thus, some effects of message order may be due to factors other than attitudestrengthper se. In such a case, however, the attitude strength perspective guided by the ElaborationLikelihood Model (Petty and Cacioppo 1986) would predictthat the primacyeffectbroughtaboutby ignoringlaterinformation shoulddecay fasterthan primacyeffectsbroughtabout by a high level of elaboration.Similarly,there may be conditions under which recency effectscould be made to endure.For instance,the low-elaborationsubjectsin our studiescould be inducedto evaluatethe arguments they recall. If these subjects engage in effortfulevaluation of these arguments, then the observed recency effect might persistto a relativelyhigh degree. If these subjectswere disruptedfrom doing such evaluation of recalledarguments,however,the recencyeffectswould likelynot persist.Thus,anotherimportantgoal of future researchshould be to delineate the conditions under which effortfulversus noneffortfulprocessesare more likely and to explore the differentialconsequences of these processes. Such researchmay offerunique insightsinto the creation of especially durableattitudes through the strategic use of variablesused to increaseprocessingof initial and counterpersuasivemessages. Such processing may serve to strengthenand reinvigorateexisting attitudes (see Haugtvedt,Leavitt,and Schneier 1993). Future research might also profitably apply alternative manipulationsand classificationsof motivationto process information such as individual versus group responsibilityfor messageevaluation (Petty et al. 1977), self-referencingof message content (Burnkrant and Unnava 1989), and identifying message recipients as beinghighversuslow in need for cognition (Haugtvedt, Petty,and Cacioppo1992).In addition,whilewe suggest that an understandingof the extent of elaborationmay be a usefulway to gain understandingof some message ordereffects,it is importantto note that, for practical and theoreticalreasons, the influence of various individual persuasionfactors should be systematicallyexplored in future research.That is, understandingthat elaborationplays an important role does not obviate the usefulnessor importanceof identifyinga wide range of factorsand processes. Order Effects in Two-sided Advertising Settings. Anotherdirection for future researchis to understand the extent to which the processesidentifiedin the current studies relate to processes operatingin common advertisingcontexts. For example, consumerresearchers have investigatedthe effectivenessof one-sidedversus two-sidedadvertisementsin attemptsto understand when a particularformatwould be most effective(e.g.,

215

Etgarand Goodwin 1982; Pechmann 1992; Swinyard 1981). In this research,two-sidedads have been shown to be more effective to the extent that the "negative" (i.e., con-product)information allows one to perceive the productas having more of the "positive"attribute (i.e., pro-product) qualities (according to "correlational" inferences;see Pechmann 1992) and to the extent that the two-sided nature of the ad increasesperceptions of source credibility (Pechmann 1992). Althoughthis researchhas not varied the order of pro versuscon information,the processesfound to underlie primacyversus recency effectsin messageorder might also operatein two-sidedads. That is, if conditions encouragehigh levels of elaborationof the ad, enhanced of informationlate in the ad may counterargumentation lead to more favorableviews of the product when the informationis presentedin a pro/con ratherthan a con/ pro order.If conditions encouragelow levels of elaboration,however,then informationpresentedlate in the ad might influenceproductattitudesthroughmemorybased processes that encourage recency effects (i.e., more favorable product views following a con/pro si ratherthan a pro/con order). One potentiallymajordifferencebetweenthe pro and con informationin the two-sidedads, studiedby Pechmann (1992) and others, and the opposing messages used in message order research,is that the opposing messages are from different sources. A two-sided ad typically contains information from a single source. Therefore, if two distinct sources of opposing information are necessary for the primacy/recencyeffects we observed,then these effects may not operatein the same way for two-sided advertisingappeals. Even if two distinct sources of informationare necessary for the order effects observedin the currentresearch,advertisersmightbe able to createads that benefit fromboth sets of processes(i.e., processesassociated with two-sided ad effectiveness[Pechmann 1992] and processes associated with primacy/recency effects of opposing messages).For example, an advertisermight utilize differentpersons(e.g., two differentperspectives providedin a "personon the street"scenario)in order to make informationthat forms the two sides of an ad appearto be from distinct message sources. By doing this, the advertisermight capitalize on primacy processes that enhance perceptionsof the product (e.g., if the ad is presentedin a way that encourageshigh levels of initial elaboration,increasedcounterarguingof the opposing information might maximize favorabilityof productperceptions).If the advertiserwere then to reveal that both sets of information were sponsored by the same company at the end of the ad (i.e., at the end of the "negative"information), enhanced perceptions of source credibility (i.e., the "correspondentinferences" studied by Pechmann [1992]) may furtherenhance the effectivenessof the two-sidedad.

Limitations It is importantto point out some limiting conditions for the kind of phenomenadescribedin the presentar-

JOURNAL OF CONSUMER RESEARCH

216

ticle. First, our perspectivesuggeststhat effectsof message order are most likely when some attitude change can be achieved from the exposure to the respective persuasivemessages.Thus, for issues on which individuals hold extremelystrong initial attitudes (and thus, for which attitude change is unlikely), message order may not affectfinaljudgment. The fact that we found changesfor an issue that has receivedconsiderablemedia attention(i.e., nuclearpower)suggeststhat message ordereffectscan occurfor a varietyof issuesthat people face in their daily lives. Second, in our research,attitudes and process data were collected only after exposure to both messages. Althoughcollectionof attitudedataand/or processdata immediatelyafterexposureto an initial messagewould be interestingand useful, experiencesdescribedin related researchsuggest that collection of such data artificiallyincreasesthe level of elaborationbeyond what would naturallyoccur(see Haugtvedtand Petty 1992), making it very difficultto test the influence of the intendedprocessingmanipulationalone.Thus, a potential limitation of our currentresearchis a lack of measures on the availabilityof informationsupportiveof the attitude after exposure to the initial message alone. A challengefor futureresearch,therefore,will be the developmentof methodsto obtainthese measureswithout artificiallyinfluencingthe natureof the attitude.From a practicalperspective,however,it should also be noted that assessmentof attitudes or thoughts only after exposure to both messages is not an atypical consumer situation. That is, if both pro and con information is available,consumersmay often encounter both kinds of informationbeforebeing askedto give their opinion or make a choice. Finally, in the currentresearch,cognitive responses were collected in the orderin which the messageswere presented.Because our overall indices of cognitive responseswerecollapsedacrossmessages,this should not have markedlyinfluencedthose results.Even so, future research might benefit from having participants list thoughtswithout havingthem attemptto decide which messageengenderedeach thought. Importantly,our predictionsmay not hold in every settinginvolving opposingmessagesfrom two different sources. For example, political candidates may often want to have the last word in a debate.We suspectthat one of the reasonsfor this is that they can adapt their comments to fit what has occurredduring the debate. That is, they can remind potential voters of their strengthsand of their opponent's weaknesses. In addition, speakinglast keeps one's opponent from being able to rebut what one says duringthat portion of the debate.Sometimes,however,one cannot engagein online changeof the text of one's messageto directlyrebut an opponent's statements (e.g., if the text must be canned aheadof time-as is the case in print, radio, or television advertising).In such cases candidatesmight want audiencesthat are highly involved to think carefully and form thoughtfulopinions based on a presen-

tation of their views before ratherthan after the views of their opponents(consistentwith the attitudestrength perspective).Of course,if candidatesview the electorate as largelyunable or unmotivated to process the information in their speeches, then our perspectivewould also predictthat they would want to presenttheir views last in a debate.

Importanceof Multiple-MessageResearch This line of researchmay also have important implications for marketingpractitioners,public relations firms,and even courtroomlawyers.Thatmarketersmay have control over the sequencein which consumersare exposed to information is a fact that is recently being recognized.For example, companies might make special effortsto purchasedisplayspace in exhibitionhalls that is likely to be seen beforecompetitors'spaceif they anticipate high-elaboration conditions. Publications like AdvertisingAge often contain discussions about how one firm hopes to attractconsumersby being the firstto introducenew products,attributes,or features. Knowledgeof futurecompetitoractions is also becoming more common. Researchand theory such as that outlined in the present article could be used to help marketersdecide how to best allocate their resources under certainconditions-or even when it may be important to create certain kinds of conditions prior to otheractions. For example,if a firmanticipatesattacks fromcompetitorsor the media,it can focus on attempts to get consumersto think extensivelyabout the strong argumentsin supportof theirposition beforethe attack appears. Even if exposuresequence is beyond the immediate control of the marketer,the amount and kind of processing a consumer can be induced to engage in may be underthe control of the marketerin many instances. This may mean raisingthe involvementof the consumer at the time of message exposure or using other techniques that increase message elaboration.It may also mean increasingthe amount of time a consumeris exposed to a message or is given to "consolidate"their attitudesafterexposureto the message(e.g., the use of "reflection time," as in Haugtvedt and Strathman [1990]).

Conclusion In a recent comment on the messageordereffectliterature, Eagly and Chaiken (1993, p. 300) stated: "Given the complexity of the two-opposing-messages paradigm,and the numerous variables and processes that have been shownto influencepersuasionin a standard one-message paradigm,a firm understandingof order effects in persuasion seems unlikely in the near future."Althoughmany questionsremainsurrounding messageordereffects,the researchand perspectivesoffered in this article provide a first step in organizing and understandingthis importantareaof research.We

MESSAGE ORDER EFFECTS IN PERSUASION

hope that the currentwork also encouragesmarketers and psychologiststo exploremultiple messagecontexts and the relative durability of attitudes formed or changedthroughsuchexposures.Althoughthis research presents unique challenges, attempts to increase understandingof these contexts may lead to important theoreticaland practicaladvances. [Received November 1992. Revised September 1993.]

REFERENCES Anderson, Norman H. (1965), "Primacy Effects in Personality Impression Formation Using a Generalized Order Effect Paradigm," Journal of Personality and Social Psychology, 2 (January), 1-9. Burnkrant, Robert E. and H. Rao Unnava (1989), "Self-Referencing: A Strategy for Increasing Processing of Message Content," Personality and Social Psychology Bulletin, 15 (December), 628-638. Cacioppo, John T. and Richard E. Petty (1982), "The Need for Cognition," Journal of Personality and Social Psychology, 42 (January), 116-131. Columbus Dispatch (1 990a), "Policy Change at Nuclear Regulatory Commission Clears Way for Low Level Waste to Be Dumped in Landfills," Columbus, OH, September 2. (1 990b), "Not in My Back Yard: Ohio and Michigan Residents Unite against Low Level Nuclear Waste Dump," Columbus, OH, September 12. Craik, Fergus I. M. and Robert S. Lockhart (1972), "Levels of Processing: A Framework for Memory Research," Journal of Verbal Learning and Verbal Behavior, 11 (October), 671-684. Cromwell, Harvey (1950), "The Relative Effect on Audience Attitude of the First versus the Second Argumentative Speech of a Series," Speech Monographs, 17 (January), 105-122. Eagly, Alice H. and Shelley Chaiken (1993), The Psychology of Attitudes, Fort Worth: Harcourt Brace Jovanovich. Ebbinghaus, H. Hermann (1913), Memory, trans. H. A. Roger and C. E. Busenius, New York: Teachers College, Columbia University. Evans, M. Stanton (1989), "Apple Scare," Consumers' Research Magazine, 72 (January), 1. Etgar, Michael and Steven M. Goodwin (1982), "One-sided versus Two-sided Comparative Message Appeals for New Brand Introductions," Journal of Consumer Research, 8 (March), 460-465. Gorn, Gerald J. and Marvin E. Goldberg (1980), "Children's Responses to Repetitive TV Commercials," Journal of Consumer Research, 6 (March), 421-425. Gray, James (1992), "Effect of Order of Presentation on the Formation of Internal Reference Prices," paper presented at the American Psychological Association meeting, Washington, DC. Greenwald, Anthony G. (1968), "Cognitive Learning, Cognitive Response to Persuasion, and Attitude Change," in Psychological Foundations of Attitudes, ed. Anthony G. Greenwald et al., New York: Academic Press, 147-170. Greenwald, John (1991), "Time to Choose," Time (April 29), 54-61. Hastie, Reid and Bernadette Park (1986), "The Relationship between Memory and Judgment Depends on Whether

217 the Judgment Task Is Memory-based or On-line," Psychological Review, 93 (July), 258-268. Haugtvedt, Curtis P., Clark Leavitt, and Wendy L. Schneier (1993), "Cognitive Strength of Established Brands: Memory, Attitudinal, and Structural Approaches," in Brand Equity and Advertising, ed. David A. Aaker and Alexander Biel, Hillsdale, NJ: Erlbaum, 247-261. and Richard E. Petty (1992), "Personality and Persuasion: Need for Cognition Moderates the Persistence and Resistance of Attitude Changes," Journal of Personality and Social Psychology, 62 (August), 308-319. , Richard E. Petty, and John T. Cacioppo (1992), "Need for Cognition and Advertising: Understanding the Role of Personality Variables in Consumer Behavior," Journal of Consumer Psychology, 1 (July), 239-260. and Alan J. Strathman (1990), "Situational Personal Relevance and Attitude Persistence," in Advances in Consumer Research, Vol. 17, ed. Marvin E. Goldberg et al. Provo, UT: Association for Consumer Research, 766769. Hovland, Carl I. (1951), "Human Learning and Retention," in Handbook of Experimental Psychology, ed. Stanley S. Stevens, New York: Wiley, 613-689. (1957), "Summary and Explications," in The Order of Presentation in Persuasion, ed. Carl I. Hovland et al., New Haven, CT: Yale University Press, 129-157. , Enid H. Campbell, and Timothy C. Brock (1957), "The Effects of 'Commitment' on Opinion Change Following Communication," in The Order of Presentation in Persuasion, ed. Carl I. Hovland et al., New Haven, CT: Yale University Press, 23-32. and Wallace Mandell (1957), "Is There a 'Law of Primacy' in Persuasion?" in The Order of Presentation in Persuasion, Carl I. Hovland et al., New Haven, CT: Yale University Press, 1-22. Insko, Chester H. (1964), "Primacy vs. Recency in Persuasion as a Function of the Timing of Arguments and Measures," Journal of Abnormal and Social Psychology, 69 (May), 621-623. (1967), Theories of Attitude Change, New York: Appleton-Century-Crofts. Kardes, Frank R. and Paul M. Herr (1990), "Order Effects in Consumer Judgement, Choice, and Memory: The Role of Initial Processing Goals," in Advances in Consumer Research, Vol. 17, ed. Marvin E. Goldberg et al., Provo, UT: Association for Consumer Research, 541-545. Knower, Franklin H. (1936), "Experimental Studies of Changes in Attitude: A Study of the Effect of Printed Argument on Changes in Attitude," Journal ofAbnormal and Social Psychology, 17, 315-347. Kruglanski, Arie W. and Tallie Freund (1983), "The Freezing and Unfreezing of Lay Inferences: Effects on Impressional Primacy, Ethnic Stereotyping, and Numerical Anchoring," Journal of Experimental Social Psychology, 19 (October), 448-468. Lana, Robert E. (1961), "Familiarity and the Order of Presentation of Persuasive Communications," Journal of Abnormal and Social Psychology, 62 (April), 573-577. (1963a), "Controversy of the Topic and the Order of Presentation of Persuasive Communications," Psychological Reports, 12 (February), 163-170. (1963b), "Interest, Media, and Order Effects in Persuasive Communications," Journal of Psychology, 56 (January), 9-13.

218 (1964), "Three Theoretical Interpretations of Order Effects in Persuasive Communications," Psychological Review, 61 (July), 314-320. Lichtenstein, Meryl and Thomas K. Srull (1985), "Conceptual and Methodological Issues in Examining the Relationship between Consumer Memory and Judgment," in Psychological Processes and Advertising Effects: Theory, Research, and Application, ed. Linda F. Alwitt and Andrew A. Mitchell, Hillsdale, NJ: Erlbaum, 113-128. and Thomas K. Srull (1987), "Processing Objectives as a Determinant of the Relationship between Recall and Judgment," Journal of Experimental and Social Psychology, 23 (March), 93-118. Lund, Frederick (1925), "The Psychology of Belief. IV. The Law of Primacy in Persuasion," Journal of Abnormal and Social Psychology, 20 (February), 183-191. Mackie, Diane M. and Arlene G. Asuncion (1990), "On-line and Memory-based Modifications of Attitudes: Determinants of Message Recall-Attitude Change Correspondence," Journal of Personality and Social Psychology, 59 (January), 5-16. Mayseless, Ofra and Arie W. Kruglanski (1987), "What Makes You So Sure? Effects of Epistemic Motivations on Judgmental Confidence," Organizational Behavior and Human Decision Processes, 39 (February), 162-183. McCullough, John L. and Thomas M. Ostrom (1974), "Repetition of Highly Similar Messages and Attitude Change," Journal of Applied Psychology, 59 (March), 395-397. McGuire, William J. (1957), ".Orderof Presentation as a Factor in 'Conditioning' Persuasiveness," in The Order of Presentation in Persuasion, ed. Carl I. Hovland et al., New Haven, CT: Yale University Press, 98-114. (1964), "Inducing Resistance to Persuasion: Some Contemporary Approaches," in Advances in Experimental Social Psychology, Vol. 1, ed. Leonard L. Berkowitz, New York: Academic Press, 191-229. (1966), "Attitudes and Opinions," Annual Review of Psychology, 17, 475-514. Miller, Norman and Donald Campbell (1959), "Recency and Primacy in Persuasion as a Function of the Timing of Speeches and Measurements," Journal ofAbnormal and Social Psychology, 59 (January), 1-9. Monroe, Kent B. (1992), "Effect of Order of Presentation on Buyer's Reference Prices," paper presented at the American Psychological Association meeting, Washington, DC. Pechmann, Cornelia (1992), "Predicting When Two-sided Ads Will Be More Effective than One-sided Ads: The Role of

JOURNAL OF CONSUMER RESEARCH Correlational and Correspondent Inferences," Journal of Marketing Research, 29 (November), 441-453. Petty, Richard E. and John T. Cacioppo (1979), "Issue-Involvement Can Increase or Decrease Persuasion by Enhancing Message-relevant Cognitive Responses," Journal of Personality and Social Psychology, 37 (October), 19151926. and John T. Cacioppo (1986), Communication and Persuasion: Central and Peripheral Routes to Attitude Change, New York: Springer-Verlag. , John T. Cacioppo, and Rachel Goldman (1981), "Personal Involvement as a Determinant of Argument Based Persuasion," Journal of Personality and Social Psychology, 41 (May), 847-855. , John T. Cacioppo, and David W. Schumann (1983), "Central and Peripheral Routes to Advertising Effectiveness: The Moderating Role of Involvement," Jourhial of Consumer Research, 10 (September), 134-148. , Stephen G. Harkins, Kipling D. Williams, and Bibb Latane (1977), "The Effects of Group Size on Cognitive Effort and Evaluation," Personality and Social Psychology Bulletin, 3 (December), 579-582. , Curtis P. Haugtvedt, and Stephen M. Smith (1994), "Elaboration as a Determinant of Attitude Strength: Creating Attitudes That Are Persistent, Resistant, and Predictive of Behavior," in Attitude Strength:Antecedents and Consequences, ed. Richard E. Petty and Jon A. Krosnick, Hillsdale, NJ: Erlbaum (in press). Roberts, Leslie (1989). "Alar: The Numbers Game," Science, 234 (March), 1430. Rosnow, Ralph (1966), "Whatever Happened to the 'Law of Primacy'," Journal of Communication, 16 (January), 1031. and Edward Robinson (1967), Experiments in Persuasion, New York: Academic Press. Swinyard, William R. (1981), "The Interaction between Comparative Advertising and Copy Claim Variation," Journal of Marketing Research, 18 (May), 175-186. Webster, Donna L. and Arie Kruglanski (1992), "IndividualDifferences in Need for Cognitive Closure," unpublished manuscript, Department of Psychology, University of Maryland. Wood, Wendy, Carl A. Kallgren, and Rebecca M. Priesler (1985), "Access to Attitude-relevant Information in Memory as a Determinant of Persuasion," Journal of Experimental Social Psychology, 21 (January), 73-85. Wyer, Robert S. and Thomas K. Srull (1980), "The Processing of Social Stimulus Information: A Conceptual Integration," in Person Memory: The Cognitive Basis of Social Perception, ed. Reid Hastie et al., Hillsdale, NJ: Erlbaum, 227-300.