Effects of WOM and Product- Attribute Information on Persuasion an Accessibility Diagnosticity Perspective

Embed Size (px)

Citation preview

  • 8/3/2019 Effects of WOM and Product- Attribute Information on Persuasion an Accessibility Diagnosticity Perspective

    1/10

    Effects of Word-of-Mouth and Product-Attribute Information on Persuasion: AnAccessibility-Diagnosticity PerspectivePAUL M. HERRFRANK R. KARDESJOHN KIM*

    The effects of word-of-mouth (WOM) communications and specific attribute infor-mation on product evaluations were investigated. A face-to-face WOM communi-cation was m ore persuasive than a printed format (experiment 1). Although a strongWOM effect was found, this effect was reduced or eliminated whe n a prior impressionof the target brand was available from mem ory or when extremely negative attributeinformation was p resented (experiment 2). The results suggest that diverse, seem-ingly unrelated judgmental phenomenasuch as the vividness effect, the perse-verance effect, and the negativity effectcan be explained through the accessibility-diagnosticity model.

    S everal studies have shown that word-of-mouth(WOM) communications often exert a strong in-fluence on judgments of products. For example, con-sumers frequently rely on WO M when selecting an au-tomotive diagnostic center (Engel, Blackwell, andKegerreis 1969), when ch oosing a physician (Feldm anand Spencer 1965), or when considering the purchaseofa new product or service (Arndt 1967; Brown andReingen 1987; Reingen and K ernan 1986; Richins1983).Prior research has focused primarily on the effectsof communicator characteristics (e.g., similarity,credibility, trustworthiness) on interpersonal influence(Bearden and Etzel 1982; Bearden, Netemeyer, andTeel 1989; Price, Feick, and Higie 1989). Our researchfocuses on the manner in which information is pre-sented (vividly vs. pallidly) and on the type of evidenceavailable for processing (anecdotal vs. detailed attri-bute information). Our goal is to investigate the me-

    'Paul M. Herr is assistant professor of marketing. GraduateSchool of Business, Indiana University, Indianapolis, IN 46223.Frank R. Kardes is associate professor of marketing. College ofBusiness Administration, University ofCincin nati , Cincinnati , OH45221. John Kim was a doctoral student at the University ofCin-cinnati and is currently assistant professor of marketing. School ofBusiness Administration, Oakland University, Rochester Hills, Ml483 09. The authors wish to thank Russell Fazio, Richard Olshavsky,Martha Powell, and three anonymous reviewers for their helpfulcomments on an earlier version ofthis article.

    diation of WOM effects on persuasion and to identifyadditional moderating variables.One factor that is likely to mediate the effects ofWOM on judgm ent is information vividness. Vividlypresented information is inherently interesting, atten-tion drawing, and thought provoking. Consequently,vividly (as opposed to pallidly) presented informa tiontends to have a stronger influence on product judg-ments (for reviews, see Kisielius and Sternthal 1986;Nisbett and R oss 1980). However, the disp roportiona teeffect of vividly presented in forma tion on judg me nt isunwarranted because manner of presentation does notinfluence the diagnostic or probative value of infor-mation.Taylor and Thompson's (1982) well-known reviewrevealed that the vast majority of empirical investi-gations of the vividness hypothesis found effects ofvividness on mem ory but not on judgm ent. Moreover,the few judgm enta l vividness effects that were observedare difficult to interpret because vividness was con-founded with other informa tional variables.' Finally,they argue that judgm ental vividness effects should befound only when the information context provides anopportunity for differential encoding of vivid versuspallid information.

    Consider the Borgida and Nisbett (1977) experi-ment. College students received either extensive anddetailed course evaluations based on ratings from alarge sample of students or brief, face-to-face, coursecomments from a single individual. Vivid WOM in-454

  • 8/3/2019 Effects of WOM and Product- Attribute Information on Persuasion an Accessibility Diagnosticity Perspective

    2/10

    EFFECTS OF WORD OF MOUTH 455

    formation had a greater impact on course choices rel-ative to the extensive but pallid course evaluations.However, this study has been criticized because theWOM c omm unications and the course evaluations dif-fered on many relevant dimensions (i.e., amount ofinformation, information content, base rate vs. caseinformation).We avoided these confounds in the present experi-ment by developing a control condition in whichamoun t ofinformation, information content, and typeof information were held constant. The only factorinfluenced by our vividness manipulation was themanner in which information was presented: half ofthe subjects were exposed to anecdo tal informationpresented in a face-to-face manner (the vivid WOMcondition), and half were exposed to the exact sameinformation presented in a printed mode (the pallidcondition). In addition, extensive and detailed Con-sumer Reports information about product features wasprovided to allow selective encoding of evidence.

    H I : Word-of-mouth communications should have agreater impact on product judgments relative toless vivid printed information.E X PE R I ME N T 1

    OverviewEighty-four undergraduates were randomly assignedto one of four conditions in a 2 (WOM or printedanecdote) X 2 (positive or negative anecd ote) factorial

    design. Anecdotal information was presented either ina face-to-face (vivid) manner or in a printed (pallid)format. Anecdotal information and detailed ConsumerReports product attribute information were evalua-tively inconsistent to permit assessment ofthe relativeimpact of anecdotal versus attribute information.Attribute Information

    Subjects received information, "condensed fromConsumer Reports," about a new personal computer(PC). The description contained a brief summary ofstandard features (this summary was held constantacross conditions), a ranking (the brand was rankedeither third or seventeenth best of the 20 modelstested), and information about seven attributes (mem -ory, monitor, keyboard, hard-drive system, printerport, graphics, and clock speed). In positive-valenceconditions, all attributes were described favorably (e.g.,640-KB memory, high-quality color monitor, separatecursor control and function keys on the keyboard),whereas in negative-valence co ndition s, attribute s weredescribed less favorably (e.g., 512-KB memory, low-quality monochrome monitor, inconveniently locatedcursor control and function keys on the keyboard).

    Anecdotal InformationSubjects participated in small groups of four to six.One member ofthe group was a confederate who de-livered the face-to-face, WOM communication after

    subjects had read the product description. None ofthesubjects knew the confederate. The confederate statedthat she owned a computer similar to the describedbrand , and, in positive-anecdote conditio ns, she added,"I t's th e best com pute r I've ever owned . It's really easyto use, and I haven't had a single problem with it." Incontrast, in negative-anecdote conditions she said, "It 'sthe worst computer I've ever owned. It's really hardto use, and I've had nothing but problems with it."The experimenter asked the confederate to write thebrand name of her computer on a slip of paper, saidit was the described brand, and thanked and dismissedher. In pallid conditions, the same anecdotes were pre-sented verbatim in a printed format transcribed froma ' 'Consumer Reports telephone interview with a col-lege student."Attitude Measures

    After the confederate's departu re, subjects evaluatedthe target product on three 11-point scales anchoredby 0 and 10 (bad/go od, favo rable/unfavorable, desir-able/unde sirable). These ratings were averaged to forma single brand-attitude index (Cronb ach's alpha = .95,

    Attribute Relevance MeasuresSubjects may focus more on vivid WOM informa-tion than on attribute information because vivid in-formation is more attention drawing or because theattributes may be perceived as irrelevant (Borgida andNisbett 1977). To rule out the second possibility, weadministered attribute-relevance measures. For eachof the seven attributes, subjects estimated the per-centage of high- versus low-quality PCs possessing theattribute in question on scales from 0 (0 percent) to10 (100 percent). Low quality was defined as averageor low quality to ensure that mutually exclusive andexhaustive categories would be employed. Bayesianlikelihood ratios were computed from these subjectiveprobabilities by dividing the probability that a low-quality PC possesses a given attribute from the prob-ability that a high-quality PC possesses the attribute.Att ribut e relevance or diagnosticity is expressed as themagnitude of the likelihood ratio or its reciprocal,whichever is larger (Fischhoff and Beyth-Marom 1983).Results

    Brand attitudes as a function of anecdote formatand valence are presented in the following tabulation.

  • 8/3/2019 Effects of WOM and Product- Attribute Information on Persuasion an Accessibility Diagnosticity Perspective

    3/10

    456 JOURNAL OF CONSUMER RESEARCH

    Anecdote formalAnecdote valencePosit iveNegat ive

    WOM6.753.53Printed3.706.87

    Higher scores indicate more favorable overall evaluations.Cell size is 18 for the negative printed anecdote format(lower right entry), and 19 otherwise.A 2 X 2 between-subjects analysis of variance performedon brand attitudes yielded the predicted vividness-by-va-lence interaction (F(l,71) = 111.73, p < .0 01 , co = .59).The main effects were not significant (F's < 1).Simple-effect tests were performed to interpret the in-teraction while controlling for the compounding of alpha.In positive-anecdote conditions, more favorable brandattitudes were formed in WOM t han in print condit ions( X = 6.75 vs. 3.70; F(l ,71) = 49.50, p < . 001 , oP- = .28).Conversely, in negative-anecdote conditions, less favor-able brand attitudes were formed in WOM than in printcondit ions (X = 3.53 vs. 6.87; F(l ,7 1) = 60.24,/? < . 001 ,oj = .32). Thus, anecdotal information had a strongerimpact on b rand attitudes w hen it was presented in a vivid,face-to-face m ann er as opp osed to a pallid, printed format.The observed vividness effect is even more impressivewhen one considers the diagnost ic value ofthe at tributeinformation that was provided. If an at tribute is equallylikely to be associated w ith a high- or low-quality product,that attribute does not enable the consumer to determinewhether a product is high or low in quali ty. Thus, theattribute is nondiagnost ic, and the likelihood ratio willequal 1. To the extent that the likelihood ratio differsfrom unity, a given piece of information is considered tobe informative or di agnos t i c '

    Likelihood ratios were com pute d and averaged acrossthe seven target attributes, and this index was subjected

    'Bayes's theorem prescribes how beliefs should be updated inlight of new evidence (for an excellent review, see Fischhoff andBeyth-Marom 1983). The theorem states thatP(HIA) P(AIH) ^^ P(H)P(H/A) P(AIH) P(H)'in which, reading from right to left, P(H)IP(H) refers to the priorodds that the target product is a high- (as opposed to low) qualityproduct; P(A/H)/P{A/H) is the likelihood ratio, which representsthe degree to which attribute A ini^plies a high- (as opposed to low)quality product; and P(H/A)/PiH/A) refers to the posterior oddsthat H is true given ^. If the probability of observing A given thatH is true is different from the probability of observing A when H isnot true, the likelihood ratio will differ from 1, and the posteriorodds will differ from the prior odds. The values of th e numeratorand denominator ofthe l ikelihood ratio are independent. Thus, Amay imply H even if .4 is unlikely given H. provided that A is evenmore unlikely given H. Similarly, A may be nondiagnostic even ifA is likely given H. provided that A is equally likely given H. Con-sumers tend to overlook these possibilities because they tend tofocus on the numerator and neglect the denominator ofthe l ikeli-hood ratio (the pseudodiagnosticity efTect; see Fischhoff and Beyth-Marom 1983; Hoch and Deighton 1989).

    to a 2 (anecdote vividness) X 2 (attribu te valence) analysisof variance. The attributes were perceived as more diag-nostic in negative {X = 4.63) than in positive {X = 2.61)valence condit ions (F(l ,71) = 6.78, p < .02). Moreover,likelih ood ra tios differed significan tly from I in both neg-ative (F(l ,36 ) = 39.05,/7 < .001) and posit ive (F (l ,37) =26.04, /7 < .001) conditions. Anecdote vividness did notinfluence the perceived diagnosticity of the attributes (F< 1), and the vividness-by-valence interaction also wasnot significant (F < 1).Thus , the results provide direct support for the prop-osition that negative infonn ation is more informative thanpositive information, in the sense that it helps consumersdiscriminate between low- and high-quality products(Mizerski 1982; Wright 1974). More important, the resultsindicate that robust anecdote vividness effects on judg-ment were obtained even though highly relevant attribute

    information was available.DiscussionThe results of experiment 1 indicate that WOMcommunications have a strong impact on productjudgments, relative to less vivid printed communica-

    tions. Moreover, favorable brand attitudes are formedwhen a single, favorable WOM communication ispresented, even when extensive, diagnostic attributeinformation is available. In contrast, the judgmentaleffects of printed anecdotal information are over-whelmed by detailed attribute information. Hence, themanner in which information is presented can producestrong judgmental effects, even when the diagnosticvalue ofthis information is controlled.These findings suggest that vividly (as opposed topallidly) presented information is more accessible frommemory and is weighed more heavily in judgment (seealso Kisielius and Sternthal 1984, 1986; McGill andAnand 1989). Several studies have shown that, as in-formation accessibility increases, the likelihood withwhich this information is used as an input for judgmentand choice also increases (e.g., Biehal and Chakravarti1986; Higgins and King 1981; Wyer and Srull 1986).Ease of retrieval is an important determinant of in-formation utilization.

    To summarize, the results of experiment 1 suggestthat information accessibility mediates the effects ofWOM on persuasion. Experiment 2 was designed toextend these findings by investigating variables thoughtto moderate the effects of WOM on judgment (for anexcellent discussion of the moderator-mediator dis-tinction, see Baron and Kenny 1986). The accessi-bility-diagnosticity model was used to identify possiblemoderators.

    ^Moderator variables influence the direction or strength oftherelationship between two variables, whereas mediator variables ex-plain how one variable influences another (Baron and Kenny 1986).For a phenomenon to be understood, both enabling (mediating)and l imiting (moderating) conditions must be specified.

  • 8/3/2019 Effects of WOM and Product- Attribute Information on Persuasion an Accessibility Diagnosticity Perspective

    4/10

    EFFECTS OF WORD OF MOUTH 457

    THE ACCESSIBILITY-DIAGNOSTICITYM O D E LAccording to the accessibility-diagnosticity model,accessible information is not used as an input for judg-

    ment and choice when more diagnostic or probativeinformation is available (Feldman and Lynch 1988;Lynch, M armo rstein, and Weigold 1988). A piece ofinformation is perceived as diagnostic if it helps theconsumer assign a product to one (and only one) cog-nitive category. In contrast, information that is am-biguous (i.e., information that has multiple interpre-tations; see Hoch and Deighton 1989) or that impliesmultiple possible categorizations is nondiagnostic.Hence, diagnosticity refers to the extent to which agiven piece ofin form atio n discrimin ates between al-ternative hypotheses, interpretations, or categoriza-tions.To briefly summarize the Feldman and Lynchmodel, the probability that any piece of informationwill be used as an input for judgment or choice dependson (1) the accessibility of the inpu t, (2) the accessibilityof alternative inputs, and (3) the diagnosticity or per-ceived relevance of the inputs. Any factor that in-creases the accessibility of an input (such as a vividnessmanipulation) should also increase the likelihood withwhich that input will be used. Further, any factor thatincreases the accessibility of one input should decreasethe accessibility and use of alternative inputs. Finally,diagnostic inputs that strongly suggest one categori-zation over alternative possibilities are more likely to

    be used than inputs that are ambiguous with respectto category membership (i.e., multiple categorizationsare possible).The model emphasizes that perceived (as opposedto objective) diagnosticity determines the likelihoodofinformation utilization. Therefore, inferential biasesare possible when consumers overestimate the diag-nostic value of a given piece of information. For ex-ample, consumers are likely to overestimate the valid-ity of their prior im pressions for several reasons. P riorimpressions are remarkably persistent and resistant tochange because (1) ambiguous information is inter-preted as consistent with the impression, (2) impres-sion-consistent, nonambiguous information supportsand increases one's confidence in the impression, and(3) impression-inconsistent, nonam biguous informa-tion is discounted or ignored (Hoch and Deighton1989; Lord, R oss, and L epper 1979, 1984). Hen ce, ex-posure to any type ofinformation , even contradictoryinformation, will increase confidence in the validityof prior impressions.Consumers also tend to hold expectations about thediagnosticity of positive, neutral, and negative cues.Recent research on the negativity effect in impressionformation has shown that extremely negative attributesgenerally have strong implications for category mem-

    bership, whereas less negative features are commonlypossessed by high-, average-, and low-quality produ cts(Skowronski and Carlston 1987, 1989). For example,poor hand ling (an extremely negative attribu te) is per-ceived as characteristic of only low-quality automo-biles, but average handling (a neutral feature) can befound in many high-, average-, and low-quality au-tomobiles. Extremely negative information is usefulfor categorizing a product as low in quality, whereasless negative information is less useful. In general, ex-tremely negative cues are less ambiguo us than positiveor neutral cues (for exceptions, see Skowronski andCarlston 1987, 1989), especially in produ ct-judgm entcontexts (as suggested by the results of experiment 1and by Mizerski [1982]; Wright 1974).

    Hence, the impact of vivid and accessible WOM in-formation should be reduced when more diagnosticinformation is available. Prior imp ressions are high inperceived diagnosticity because they suggest one in-terpretation of evidence to the exclusion of others. Ex-tremely negative attribute information is diagnosticbecause it suggests one categorization (i.e., low quality)over other possibilities.

    H2: Vivid WOM communications should have areduced effect on product judgments when aprior impression of the product is available(vs. not available) from memory.H3: Vivid WOM communications should have areduced effect on product judgments whenextremely negative attribute information is

    available (vs. not available).E XP E R I M E NT 2

    OverviewOne hundred and twenty undergraduates were ran-domly assigned to one of 24 conditions in a 3 (positive,neutral, or negative attributes) X 2 (impression ormemory set) X 2 (positive or negative WOM) X 2(counterbalancing order: judgment before recall or viceversa) factorial design. Attribute valence and WOMvalence were manipulated orthogonally to permit as-sessment ofthe relative contribution ofthese two types

    ofinformation to overall judg men ts.ProcedureWhen subjects arrived at the laboratory, processinggoals were manipulated through experimental in-structions, attribute information about the targetproduct (a new automobile) was provided, a WOMcommunication was delivered by a confederate, andjudgment and recall measures were administered.Subjects were told either to form a coherent, overallimpression ofthe described brand (impression set) orto memorize the attributes of the described brand

  • 8/3/2019 Effects of WOM and Product- Attribute Information on Persuasion an Accessibility Diagnosticity Perspective

    5/10

    458 JOURNAL OF CONSUMER RESEARCHTABLE 1

    OVERALL EVALUATIONS AS A FUNCTION OF PROCESSINGOBJECTIVES. ATTRIBUTE VALENCE, ANO ANECDOTEVALENCE (EXPERIMENT 2)

    AttributevalencePositiveNeutralNegative

    Memory setPositiveWOM

    9.10,6.44.1.45.

    NegativeWOM8.20^5.49b1.85.

    Impression setPositiveWOM8.33.65.41b1.89.

    NegativeWOM7.38c4.71b2 .21 .

    NOTE. Higher scores indicate more favorable overall evaluations. Means inthe same row not sharing a common subscript differ at p < .05. n = 10 per cell(except in the lower right entry, for w hich n = 9).

    (memory set). The description contained a brief sum-mary of standard features, a ranking (the brand wasranked sixth, twentieth, or thirty-fourth best ofthe 40brands tested), and information about nine im portantattributes (engine performance, transmission, han-dling, braking, ride, noise, displays, seating, and reli-ability). The attributes were described favorably (e.g.,braking was smooth), neutrally (e.g., braking was usu-ally smo oth), or unfavorably (e.g., braking was abrup t).Subjects participated in small groups of four to six.After subjects had read the description, the confederatesaid that the car seemed similar to a car her fatherowns. In positive WOM conditions she added, "It'sthe best car he's ever had. He hasn't spent a dime onrepairs since he bought it. He says if it ever wears outhe'll get another just like it." In negative WOM con-ditions she said, "It's the worst car he's ever had. Itseems like it's always in the shop being repaired. I thinkhe's spent more to keep it running than it originallycost him." The experimenter asked her to write thebrand name of the car on a slip of paper, said it wasthe described brand, and thanked and dismissed her.

    Judgment and Recall MeasuresAfter dismissal of the confederate, subjects wereasked to recall as much attribute information as pos-sible within a three-minute period, rate each attributeon scales ranging from 0 (extremely low quality) to 10(extremely high quality), and judge the overall qualityof the target product. Finally, subjects were thankedand debriefed.Two judges coded the recall protocols using a gistcriterion, and high interrater reliability was attained(97 percent). Attribute ratings were averaged to forma single brand attitude index (Cronbach's alpha = .98,p < .001). This index correlated highly with the overallquality measure (r = .94, p < .001).

    ResultsEffects of Prior Impressions on Judgmental Vivid-ness Effects. Overall product evaluations as a func-tion of processing goals, attribute valence, and WOM

    valence are presented in Table 1. To test the hypothesisthat vivid WOM communications should have a re-duced effect on product judgments when a priorimpression ofthe product is available (vs. not avail-able) from mem ory (Hy pothesis 2), a series of planne dcontrasts was performed on overall product judg-ments.^ Because prior im pressions are less likely to beavailable in memory set than in impression set con-ditions (Hastie and Park 1986; Kardes 1986; Lichten-stein and Srull 1985, 1987), WOM effects on judgmentshould be weaker in impression set than in memoryset conditions. Consistent with this prediction, in pos-itive-attribute conditions, a significant WOM effect wasobtained in the memory set condition (X = 9.10 vs.8.20, p < .05). However, unexpectedly, a significantWOM effect was also found in the imp ression set con-dition {X = 8.33 vs. 7.38, p < .05). Thus, p artial sup-port for Hypothesis 2 was found in positive-attributeconditions.A different pattern of results was found when neutralattributes were presented. In the mem ory set cond ition,more favorable judgments were formed in positive-than in negative-WOM conditions (X = 6.44 vs. 5.49,p < .05). Although a robust WOM effect on judgmentwas obtained in the memory set condition, no WOMeffect was found in the impression set condition {X

    = 5.41 vs. 4.71, NS). Thus, consistent supp ort for Hy-pothesis 2 was found in neutral attribute conditions.EffTects of Negative-Attribute Information on Judg-mental Vividness Effects. To test the hypothesis thatthe presence of extremely negative attribute infor-mation reduces the effect of anecdotal WOM com-munications on judgment (Hypothesis 3), plannedcomparisons w ere performed separately in m emory setand in impression set conditions. Although robustWOM effects on judgment were found when positive

    'Planned comparisons should be performed instead of overall F-tests when interactions involving differences between specific cellsare predicted (Hays 1981; Keppel 1982; Kirk 1982). Nevertheless,the reader may be interested in the results of omnibus F-tests. A 2(impression or memory set) X 3 (positive, neutral, or negative at-tributes) X 2 (positive or negative WOM) X 2 (counterbalancingorder) between-subjects analysis of variance was performed onoverall evaluations. Counterbalancing order produced no significantmain effects or interactions. Main effects were found for processinggoal (F( l ,9 5) = 6.9 7,/> < .01), attribute valence (F(2,95) = 47 4.44 .p < .001), and WOM valence (F(l ,95) = 7.90, p < .01). More fa-vorable evaluations were formed in memory set than in impressionset conditions, and more favorable evaluations were formed as at-tribute and WOM information increased in favorability. Significantprocessing goal by attribute valence (f(2,95) = 5.92, p < .01) andattribute valence by WOM valence (/=^(2,95) = 5.8l,p < .01) inter-actions were also found.

  • 8/3/2019 Effects of WOM and Product- Attribute Information on Persuasion an Accessibility Diagnosticity Perspective

    6/10

    EFFECTS OF WORD OF MOUTH 459

    or neutral attributes were presented to memory setsubjects, the WOM effect was eliminated when ex-tremely negative attributes were presented to memoryset subjects (X = 1.45 vs. 1.85, NS). Similarly, inimpression set conditions, no WOM effect was foundwhen extremely negative attributes were presented(X = 1.89 vs. 2.21, NS). Thus, consistent supportwas found for Hypothesis 3 across processing-goal(impression vs. memory set) conditions.Judgment-Recall Correlations/Positive-AttributeConditions. Judgment-recall correlations were ex-amined to facilitate interpretation of the unexpectedWOM effect on judgment obtained in the positive at-tribute-impression set cell. Prior research has shownthat an impression set induces subjects to engage inon-line processing and to form separate evaluation-based and attribute-based representations in memory

    (Hastie and Park 1986; Kardes 1986; Lichtenstein andSrull 1985, 1987). Because independent representa-tions are formed, a lack of correspondence betweenjudgment and recall is likely to be observed. In con-trast, a memory set prompts subjects to form an attri-bute-based representation in memory (no evaluation-based representation is formed). When memory setsubjects are subsequently asked to evaluate a product,a judgment is .computed on the basis ofinformationthat is retrieved directly from the attribute-based rep-resentation. Consequently, a strong judgment-recallcorrelation is likely to be observed.Consistent with this model, significant judgment-

    recall correlations were found in memory set condi-tions (r = .51 and .59 for positive- and negative-WOMconditions, respectively, p < .05). Moreover, inimpression set conditions, a nonsignificant relation-ship between judgment and recall was found in thepositive-WOM cell (r = .26, NS). However, in theimpression set-negative WOM cell, a significant cor-relation between judgment and recall was obtained (r= .64,> < .05). This unexpected finding implies thatjudgments formed in this cell were influenced signif-icantly by memory for attributes. However, the un-certainty created by the inconsistency between the at-tribute versus the WOM information may haveprompted subjects to adjust their judgm ents toward amore mod erate position. This type ofjudgmental shiftis observed frequently under co nditions of uncertaintybecause a moderate stance is justified easily and is al-tered readily as additional information becomes avail-able (Cialdini et al. 1973; Jaccard an d W ood 1988).Judgment-Recall Correlations/Neutral-AttributeConditions. As expected, nonsignificant judgment-recall correlations were found in impression set con-ditions {p > .20). Unexpectedly, in memory set con-ditions, judgment-recall correlations differed as afunction of WOM valence. In the positive-WOM cell,an inverse relationship between judgment and recall

    was obtained (r = - . 7 0 , p < .05), whereas, in the neg-ative-WOM cell, the judgment-recall correlation wasnot significant (p > .20). Thus, in the positive-WOMcell, as the number of neutral attributes that could berecalled increased, less favorable overall judgmentswere formed. In the negative-WOM cell, relativelyunfavorable overall judgments were formed regardlessof the number of neutral attributes available frommemory.Judgment-Recall Correlations/Negative-AttributeConditions. Finally, in negative-attribute conditions,nonsignificant judgm ent-recall correlations were foundacross impression/memory set conditions {p > .20)and across WO M-valence cond itions {p > .20). Theseresults imply that on-line judgments were formed inboth impression and memory set conditions. Thisfinding is consistent with previous research demon-

    strating that extremely negative information canprom pt spontaneous judgm ent formation (Bettmanand W eitz 1983; Weiner 1985).Discussion

    Although WOM communications were found tohave a strong impact on product judgments in exper-iment 1, this effect was reduced when a well-definedprior impression was available from memory or whenextremely negative attribute information was encoun -tered. Unexpectedly, a WOM effect on judgment wasfound when positive attributes were presented toimpression set subjects. Analyses of the relationshipbetween judgment and recall, however, suggest thatthese subjects failed to form a clear, well-definedimpression ofthe target product in the negative-WOMcondition, presumably because the uncertainty createdby the attribute valence-WOM valence inconsistencyprompted a judgm ental shift toward m oderation(Cialdini et al. 1973; Jaccard an d W ood 1988).When extremely negative attributes were presented,vivid WOM co mm unications had no effect on productjudgments. Because extremely negative information isperceived as highly diagnostic (experim ent 1) and be-cause diagnostic information can reduce the judg-mental impact of less diagnostic but more accessibleinpu ts, judg me ntal vividness effects were elimin atedcompletely when extremely negative attribute infor-mation was provided.

    GENERAL DISCUSSIONConsistent with the accessibility-diagnosticity model(Feldman and Lynch 1988; Lynch et al. 1988), the re-sults suggest that information accessibility mediatesthe effects of WOM information on product jud gm ents(experiment 1). How ever, information-accessibility ef-fects on judgment are reduced, when more diagnosticinformation, such as prior impressions or extremely

  • 8/3/2019 Effects of WOM and Product- Attribute Information on Persuasion an Accessibility Diagnosticity Perspective

    7/10

    460 JOURNAL OF CONSUMER RESEARCH

    negative attribute information, is available (experi-ment 2). Thus, information accessibility mediates theeffects of vividly presented information on judgment,and the perceived diagnosticity of other available in-puts moderates these effects.The results also imply that judg m ental vividness ef-fects are not as elusive as implied by previous research(Taylor and Thom pson 1982). Rob ust vividness effectson judgment were obtained in two experiments em-ploying two different produ ct classes. Consistent w iththe accessibility-diagnosticity model, the results sug-gest that judgmental vividness effects are not elusivewhen (1) the judgment context provides the oppor-tunity for differential encoding of vivid versus pallidinformation, (2) prior impressions are unavailablefrom memory, and (3) information having extremelynegative judgmental implications is unavailable forprocessing. Many ofthe studies included in Taylor and

    Thompson's (1982) highly cited review failed to meetthese criteria, and, consequently, judgm ental vividnesseffects were unlikely to be observed.In addition to shedding light on the vividness con-troversy, the accessibility-diagnosticity model has thecapacity to synthesize and integrate a wide variety ofseemingly disparate judgmental phenomena. For ex-ample, the results of our experiments suggest that thevividness effect (Kisielius and Sternthal 1984, 1986),the perseverance effect (Lord et al. 1979, 1984), andthe negativity effect (Skowronski and Carlston 1987,1989) can be explained through the accessibility-di-agnosticity model.Experiment 1 demonstrated that WOM communi-cations have a greater impact on product judgmentsthan less vivid printed infbrmation even when infor-mation con tent is held constant. A lthough the m annerin which information is presented does not affect itsprobative value, vividly presented information has adisproportionate effect on judgm ent. Furtherm ore,consumers are likely to overestimate the diagnosticityof accessible information because alternative inputsare likely to be overlooked.Using the same methodology but a different productclass, experim ent 2 showed tha t the judgmental effectsof accessible WOM information are reduced when

    other more diagnostic pieces ofinform ation are avail-able. Consumers trust their own opinions more thanthey trust the opinions of others (Hoch and Deighton1989). Other individuals are likely to exhibit knowl-edge or reporting biases (see Wood and Eagly 1981),and, consequently, information provided by othersources is often ambiguous (i.e., multiple interpreta-tions are possible). In con trast, self-generated infor-mation is less likely to be con tam inated by knowledgeor reporting biases (although, under certain circum-stances, self-deception has been observed; see, e.g.,Berglas and Jones 1978; Gur and Sackeim 1979).Moreover, selective encoding and selective retrieval

    processes induce consumers to overestimate the valid-ity of their prior impressions.In consumer settings, negative information tends tobe more diagnostic or informative than positive orneutral information. Negative attributes strongly im-ply membership in one category (i.e., low quality) tothe exclusion of others, whereas positive or neutralattributes are m ore ambiguous with respect to categorymembership. Positive and neutral features are asso-ciated with many high-, medium-, and low-qualityproducts. Negative features, on the other hand, havestronger implications for categorization. Even whenmany positive features are exhibited (e.g., the soup hasfresh meat, fresh Grade-A potatoes, and fresh vita-min-rich vegetables), a single extremely negative fea-ture (e.g., the broth is rancid) can be highly informa-tive. Consequently, negative-attribute information isweighed heavily in judgment.

    In conclusion, WOM com mu nications often have astrong impact on product judgments because infor-mation received in a face-to-face manner is more ac-cessible than information presented in a less vividmanner. Although WOM information is highly acces-sible from memory, its itnpact on judgment is reducedwhen more diagnostic information is available. Priorimpressions are easy to interpret, and they influenceencoding and retrieval processes in a manner that re-duces the correspondence between objective and per-ceived cue diagnosticity. Consequently, prior beliefs(as opposed to other possible beliefs) are held with ahigh degree of confidence. Similarly, negative infor-mation tends to suggest one categorization over others.Information that strongly implies one hypothesis, in-terpretation, or categorization over other possibilitiesis weighed heavily in judgment, regardless of whetheralternative possibilities are improbable or are simplyoverlooked.

    [Received June 1990. Revised August 1990.]REFERENCES

    Arndt, Johan (1967), "Role of Product-related Conversa-tions in the Diffusion of a New Pr od uct," Journal ofMarketing Research, 4 (August), 291-295 .Baron, Reuben M. and David A. Kenny (1986), "The Mod-erator-Mediator Variable Distinction in Social Psycho-logical Research: Co nceptual, Strategic, and StatisticalConsiderations," Journal of Personality and SocialPsychology, 51 (December), 1 173-1182.Bearden, W illiam O. and M ichael J. Etzel (1982), "ReferenceGroup Influence on Product and Brand Purchase De-cisions," Journal of Consumer Research, 9 (September),183-194., Richard G. Netemeyer, and Jesse E. Teel (1989),"Measurement of Consumer Susceptibility to Interper-sonal Influence," Journal of Consumer Research, 15(March), 473-481.

  • 8/3/2019 Effects of WOM and Product- Attribute Information on Persuasion an Accessibility Diagnosticity Perspective

    8/10

    EFFECTS OF WORD OF MOUTH 46 1Berglas, Steven and Edward E. Jones (19 78), "Dru g Ch oiceas a Self-handicapping Strategy in Response to Non-contingent Success," Journal of Personality and SocialPsychology, 36 (April). 405 -417 .Bettman, James R. and Barton A. Weitz (1983), "Attribu-

    tions in the Boardroom: Causal Reasoning in CorporateAnnual Reports," Administrative Science Quarterly, 28(June), 165-183.Biehal, Gabriel and Dipanker Chakravarti (1986), "Con-sumers' Use of Memory and External Information inChoice: Macro and Micro Perspectives," Journal ofConsumer Research, 12 (March), 382-405 .Borgida, Eugene and Richard E. Nisbett (1977), "The Dif-ferential Impact of Abstract vs. Concrete Informationon D ecisions," Journal of Applied Social Psychology, 7(July-August), 258-271.Brown, Jacqueline Johnson and Peter H. Reingen (1987),"Social Ties and Word-of-Mouth Referral Behavior,"Journal of Consumer Research, 14 (December), 350-362.Cialdini, Robert B., Alan Levy, C. Peter He rman , and ScottEvenbeck (197 3), "Attitud inal Politics: The Strategy ofModerat ion," Journal of Personality and Social Psy-chology, 25 (January), 100-108.Engel, James E., Roger D. Blackwell, and Robert J. Kegerreis(1969), "How Information Is Used to Adopt an Inno-vation," Journal of Advertising Research, 9 (December),3-8.Feldman, Jack M. and John G. Lynch (1988), "Self-gener-ated Validity and Other Effects of Measurement on Be-lief, Attitude, Intention, and B ehavior," Journal ofAp-plied Psychology, 73 (August), 421-435.Feldman, Sidney P. and Merlin C. Spencer (1965), "TheEffect of Personal Influence in the Selection of Con-sumer Services," in Proceedings ofthe Fall Conferenceof the American Marketing Association, ed. Peter D.Bennett, Chicago: American Marketing Association,440-452.

    Fischhoff, Baruch and Ruth Beyth-Marom (1983), "Hy-pothesis Evaluation from a Bayesian Perspective," Psy-chological Review, 90 (July), 239-260.Gur, Ruben C. and Harold A. Sackeim (1979), "Self-De-ception: A Concept in Search ofa Phen om enon ," Jour-nal of Personality and Social Psychology. 37 (February),147-169.Hastie, Reid and Bernadette Park (1986), "The Relationshipbetween Memory and Judgment Depends on Whetherthe Judgm ent Task Is Memory-based or On-Lin e," Psy-chological Review, 93 (July), 258-268.Hays, William L. (1981), Statistics, New York: Holt, Rine-hart & Winston.Higgins, E. Tory and Gillian King(1981), "Accessibility ofSocial Constructs: Information-processing Conse-quences of Individual and Contextual Variability," inPersonality, Cognition, and Social Interaction, ed .Nancy Cantor and John F. Kihistrom, Hillsdale, NJ:Erlbaum, 69-121.Hoch, Stephen J. and John Deighton (1989), "ManagingWhat Consumers Learn from Experience," Journal ofMarketing, 53 (April), 1-20.Jaccard, James and Gregory Wood (1988), "The Effects ofIncomplete Information on the Formation of Attitudes

    toward Behavioral Alternatives," Journal of Personalityand Social Psychology. 54 (April), 580-591.Kardes, Frank R. (1986), "Effects of Initial Product Judg-ments on Subsequent Memory-based Judgments,"Journal of Consumer Research, 13 ( J u n e ) , I - l l .Keppel, Geoffry (1982), Design and Analysis: A Researcher'sHandbook, Englewood Cliffs, NJ: Prentice-Hall.Kirk, Roger E. (1982), Experimental Design: Proceduresfor the Behavioral Sciences, Monterey, CA: Brooks/Cole.Kisielius, Jolita and Brian Sternthal (1984), "Detecting andExplaining Vividness Effects in Attitudinal Judgments,"Journal of Marketing Research, 21 (February), 54-64.and Brian Sternthal (1986), "Examining the VividnessControversy: An Availability-Valence Interpretation,"Journal of Consumer Research, 12 (M arch), 418-431.Lichtenstein, Meryl and Thomas K. Srull (1985), "Conceptualand Methodological Issues in Examining the Relationshipbetween Consumer Memory and Judgment," in Psycho-

    logical Processes and Advertising Effects: Theory, Reseaand A pplication, ed. Linda F. Alwitt and Andrew A. Mitch-ell, Hillsdale, NJ: Erlbaum, 113-128.and Thomas K. Srull (1987), "Processing Objectives asa Determinant of the Relationship between Recall andJudgment," Journal of Experimental Social Psychology, 23(March), 93-11 8.Lord, Charles G.. Mark R. Lepper, and Elizabeth Preston(1984), "Considering the Oppo site: A Corrective Strategyfor Social Ju dgm ent," Journal of Personality and SocialPsychology. 47 (December), 1231-1243., Lee Ross, and Mark R. Lepper (1979), "Biased Assim-ilation and Attitude Polarization: The Effects of P rior The-ories on Subsequently Considered Evidence," Journal ofPersonality an d Social Psychology, 37 (November), 20982109.Lynch, John G ., Howard Marmo rstein, and Michael F. Weigold(1988), "Choices from Sets Including Remembered Brands:Use of Recalled Attributes and Prior Overall Evaluations,"Journal of Consumer Research, 15 (September), 169-184.McGill, Ann L. and Punam Anand (1989), "The Effect of VividAttributes on the Evaluation of Alternatives: The Role ofDifferential Attention and C ognitive Elaboration," Journalof Consumer Research, 16 (September), 188-196.Mizerski, Richard W. (1982), "An Attribution Explanation ofthe Disproportionate Influence of Unfavorable Informa-tion," yoi/wfl/o/Co/jra/Mer/{eiearc/i, 9 (December), 3 0 1 -310.Nisbett, Richard and Lee Ross (1980), Human Inference: Strat-egies and Shortcomings of Social Judgment, EnglewoodCliffs, NJ: Prentice-Hall.Price, Linda L., Lawrence F. Feick, and Robin A. Higie (1989),"Preference Heterogeneity and Coorientation as Determi-nants of Perceived Informational Inttuence" Journal ofBusiness Research, 19 (November), 227-242.Reingen, Peter H. and Jerome B. Kernan (1986), "Analysisof Referral Networks in Marketing: Methods and Illustra-tion," Journal of Marketing Research, 23 (November),370-378.Richins, Marsha L. (1983), "Negative Word-of-Mouth by Dis-satisfied Consumers: A Pilot Study," Journal ofMarketing.47 (W inter), 68-7 8.

  • 8/3/2019 Effects of WOM and Product- Attribute Information on Persuasion an Accessibility Diagnosticity Perspective

    9/10

    462 JOURNAL OF CONSUMER RESEARCHSkowronski, John J. and Donal E. Carlston (1987), "SocialJudgment and Social Memory: The Role of Cue Diagnos-ticity in Negativity, Positivity, and Ex tremity Biases," Jour-nal of Personality and Social Psychology, 52 (April), 689-699.

    and Donal E. Carlston (1989), "Negat ivi ty and Ex-tremity Biases in Impression Formation: A Review ofExplana t ions , " Psychological Bulletin, 105 (January) ,131-142 .Taylor, Shelley E. and S uzanne C. Thom pson (1982), "Stalkingthe Elusive 'Vividness" Effect," Psychological Review, 89(March), 155-181.

    Weiner , Bernard (1985), " 'Spontaneous ' Causal Thinking,"Psychological Bulletin, 97 (January) , 74-84.Woo d, Wen dy and Alice H. Eagly (1981), "Stages in the Analysisof Persuasive Messages: The Role of Causal Attributionsand Message Comprehension," Journal o f Personality andSocial Psychology, 40 (February) , 246-259 .Wright, Peter (1974), "The Harassed Decision Maker: TimePressures, Distractions, and the Use of Evidence," Journalof Applied Psychology, 59 (October), 5 5 5 -5 6 1 .Wyer, Robert S. and Thomas K. Srull (1986), "Human Cog-nition in Its Social Context," Psychological Review, 93(July), 322-359.

  • 8/3/2019 Effects of WOM and Product- Attribute Information on Persuasion an Accessibility Diagnosticity Perspective

    10/10