Sie sind auf Seite 1von 19

Journal of Consumer Research Inc.

Antecedents and Consequences of Attitude Toward the Ad: A Meta-Analysis


Author(s): Steven P. Brown and Douglas M. Stayman
Reviewed work(s):
Source: Journal of Consumer Research, Vol. 19, No. 1 (Jun., 1992), pp. 34-51
Published by: The University of Chicago Press
Stable URL: http://www.jstor.org/stable/2489186 .
Accessed: 21/11/2011 15:55

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 support@jstor.org.

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
Antecedents and Consequences of Attitude
toward the Ad: A Meta-analysis

STEVEN P. BROWN
DOUGLAS M. STAYMAN*

A meta-analysis of pairwise relationships involving attitude toward the ad was con-


ducted. Analyses of correlations across studies are first analyzed and reported.
Because significant variance across studies was found, moderator analyses were
conducted to account for interstudy variance. The results suggest a number of
methodological variables that moderate the strengths of relationships found in studies
of ad attitudes. Analyses were also conducted to assess the robustness of the
dual-mediation path model of the effects of ad attitudes. Results indicate support
for the model as well as a more importantrole for the indirectinfluence of ad attitudes
on brand attitudes (via brand cognitions) than that found in previous model tests.

Consumers' affectiveresponsesto ads have been of from before 1985, 21 between 1985 and 1987, and 30
increasinginterestto both academicsand practi- after 1987. In addition, this topic has been sufficiently
tioners. In particular,one researchstream has investi- importantto merit its own section in the two most re-
gated the influence of attitude toward the ad (Aad)on cent reviewsof consumerresearchin the AnnualReview
brand attitudes. Articles in the 1970s (e.g., Holbrook of Psychology(Bettman 1986; Cohen and Chakravarti
1978) first suggestedthe importance of understanding 1990).
viewers'global evaluationsof ads. Articlesby Mitchell Despite this interest in ad attitudes, there has not
and Olson (1981) and Shimp (1981) introduced and been a comprehensive attempt to assess the general
suggestedthe importanceof the Aadconstruct. findingsacrossstudies. Such an attemptappearsuseful
Recently,appliedstudies,particularlythe Advertising for two reasons.First,researchon ad attitudeshas been
ResearchFoundation copy testing project (Haley and conductedin a numberof methodologicalcontexts, yet
Baldinger 1991), have suggestedthat liking of an ad no attempt has been made to assess the robustnessof
may be the best indicator of advertisingeffectiveness. effects across conditions. As in other researchstreams
Academic studies have focused on such issues as con- (e.g., Rao and Monroe 1989), such an assessmentcan
ditions in which ad attitudeshave relativelystrong ef- be useful in understandingthe general strength and
fects (e.g., Gardner 1985; Park and Young 1986), the variabilityof the relationshipsand the studyconditions
determinantsof ad attitudes(e.g., MacKenzieand Lutz that moderatethose relationships.Second, although a
1989), the role of feeling responses(e.g., Staymanand numberof researchershave found that the dual-media-
Aaker 1988),and tests of causalmodels focusingon the tion model of the effectsof ad attitudesbest fitsthe data
role of ad attitudes in determiningad outcomes (e.g., in individualstudies, more specificfindingsconcerning
Burkeand Edell 1989;Homer 1990;MacKenzie,Lutz, pathswithin the model have been mixed. For example,
and Belch 1986;Miniard,Bhatla,and Rose 1990). Fur- while some researchershave found no evidence of a
ther, researchers'interest in ad attitudesappearsto be significantindirectpathbetweenad attitudesand brand
increasing.Ourliteraturereviewfound 60 articleswith attitudes via brand cognitions (e.g., MacKenzie and
ad attitudesas a partof the research,with only 9 articles Lutz 1989),othershave found a significantindirectpath
(e.g., Homer 1990). Similarly,differentstudieshave re-
*Steven P. Brown is assistantprofessorof marketingand distri- ported widely varying strengthsof the direct effect of
butionat the TerryCollegeof Business,Universityof Georgia,Athens, ad attitudeson brandattitudes.The relativeimpact of
GA 30602. DouglasM. Staymanis assistantprofessorat the Johnson brandcognitions and ad attitudeson brandattitudesis
GraduateSchool of Management,Cornell University, Ithaca, NY of centralconcernin the researchstream,and assessing
14853. The authors,who contributedequally to this article, thank
Bob Peterson,TerryShimp,VithalaRao, and especiallythe reviewers these effectsacrossall availablestudies is an important
and editor for their helpful comments on an earlierversion of the objectiveof our research.Thus, it appearsuseful to dis-
manuscript.Thanksare also expressedto the many researcherswho tinguish between competing models and to assess the
generouslyprovidedcorrelationmatricesnot availablein published strengthof specific model paths on the basis of effects
studiesfor use in the analysesreportedhere.
aggregatedacross all availableresearch.
34
? 1992by JOURNALOF CONSUMERRESEARCH,Inc.* Vol. 19 . June 1992
All rightsreserved.0093-5301/93/1901-0004$2.00
EFFECTS OF AD ATTITUDE 35

Ourresearchfocuses on three areasin which a quan- ological literatureon meta-analysisand of the substan-
titative review is likely to provide useful insights. The tive literatureon Aad. Eleven study characteristicswere
firstobjectiveis to provide a generalassessmentof the identifiedas potential moderatorvariables.
strengthof pairwiserelationshipsbetween ad attitudes
and antecedent and outcome constructs. The second Coded Study Characteristics
objective is to assess the variability of pairwise rela- Number of Scale Items. In all studies included in
tionshipsinvolvingad attitudesand the extent to which the analysis,ad attitudewas conceptualizedas a global
they are moderatedby differencesin the contexts in evaluationof a specificadvertisingexecution.However,
which these effectshave been studied. The third objec- there was substantial variety in the number of scale
tive is to use the aggregatedstudy effectsto investigate items used, with most studies (36) using multiple-item
the robustnessof the dual-mediationmodel and to as- scales, but 11 studies used only a one-item scale. Anal-
sess the strength of specific paths in the model. The ysis of single- versusmultiple-itemscales has been sug-
next sectiondescribesthe researchdomain and specifies gested for meta-analysisbecause multiple-item scales
a numberof characteristicsof existing studiesthat may are expected to be more reliableand sensitive (and re-
influencethe strengthof relationshipsinvolving ad at- liability of multi-item scales can be assessed).If, as ex-
titudes. pected, multi-item scales are more reliable, less atten-
tuation from measurementerrorshouldresultin greater
THE RESEARCH DOMAIN effectsizes (Houston, Peter,and Sawyer1983;Johnson
and Eagly 1989).
As Brinbergand Jaccard(1986) and othershave em-
phasized,the primaryquestions of interest in a meta- Sample. The use of student subjectshas often been
analysis concern the robustness of the relationships debated in consumer research (Calder, Phillips, and
studied and the specification of conditions that limit Tybout 1981; Lynch 1982). In meta-analyses,subject
these relationships'generalizability.In this study, we type is often coded as a moderatorbecause the use of
assessthe strengthof pairwiserelationshipsbetweenad a homogeneousstudentpopulationmay resultin a bias
attitudes and various antecedent and outcome con- towardstrongereffectsthan would be found in the gen-
structsacrossthe diversestudies that constitute the re- eral population(e.g., Houston et al. 1983;Johnson and
searchstream.Althougha few studieshave investigated Eagly 1989; Monroe and Krishnan 1983). In addition,
effectsof variablesthat moderaterelationshipsbetween for researchon ad attitudes, the importance of liking
ad attitudes and other constructs (e.g., Droge 1989; for an ad may differbetween student and nonstudent
Homer 1990;Staymanand Aaker 1988), no systematic samplesbecausemediapreferencesin generalmay differ
attempt has been made to investigatestudy character- between college-studentand adult populations.
isticsthat mightaccountfor the total variationin results Operationalizationof Feelings. In reviewing the
acrossall availablestudies. Thus, in a manner similar literature,two characteristicsrelatedto feelingresponse
to Rao and Monroe's(1989) assessmentof variationin variedacrossstudies and were coded as potential mod-
results for the price-qualityrelationship,we pose and eratorvariables.First, some studies manipulatedfeel-
investigatemore specificquestionsconcerninghow the ings experimentally,whereas others measured differ-
methodologicaldecisions made by researchersmight ences in feelings. Use of manipulatedversus measured
affect the strengthof the relationshipsbetween ad at- independentvariableshas been suggestedas a potential
titudes and antecedentand outcome constructs. moderatorvariable (Brinbergand Jaccard 1986). The
Severaldiscussionsof how to code study featuresfor distinction between measured and manipulated vari-
moderatoranalysesoffereduseful guidelinesfor our re- ables is importantbecause measuredvariablesmay re-
search (e.g., Brinbergand Jaccard 1986; Farley and flectthe resultof experiencewith the relevantmeasures
Lehmann 1986; Hedges and Olkin 1985; Monroe and in a varietyof contexts, whereasmanipulatedvariables
Krishnan 1983;Rosenthal 1984). As is typical in meta- may reflectonly a single situationalvariablein one spe-
analysis, the coded study characteristicscan be cate- cific context. For a variablehavingsuch strongindivid-
gorized generallyas variablesrelated to measurement ual differencesas feeling responses to advertisingdo
and operationalizationof constructsor to other study (Harrisand Moore 1989), experimentalmanipulation
designfeatures(e.g., Brinbergand Jaccard1986;Farley of feelings may not be as effective as measurementof
and Lehmann 1986). The impact that these potential the intensity of subjects'feelings.
moderatorsexert on effect sizes can be furthercatego- Second, a number of studies assessedthe effectsof a
rized as substantiveor methodological. For example, single feeling on ad attitudes whereas others assessed
differentprocessing goals may affect study outcomes the effectsof multiple feelings. Becausethe strengthof
throughsubstantiveprocesses,whereasdifferentoper- relationships involving ad attitudes may vary across
ationalizations of variables may affect them through feelings (Burke and Edell 1989; Stayman and Aaker
methodological processes, such as attenuation from 1988), averagingeffects across feelings may yield dif-
measurementerror.The potential moderatorvariables ferentresultsthan assessingthe effectsof a singlefeeling
were selected on the basis of a review of the method- would.
36 JOURNAL OF CONSUMER RESEARCH

Products. Two differentproduct-relatedstudychar- Medium. The medium in which a persuasivemes-


acteristicshave variedacrossstudiesof ad attitudes.The sage is presented has been found to be an important
firstis whetherthe stimulusbrandwas novel or familiar. moderator of message elaboration and of persuasion
Edell and Burke(1987) found effectsof ad attitudesfor effects (Chaikenand Eagly 1983; Krugman 1965). Al-
bothnovelandfamiliarbrands,but theydid not explicitly thoughresearchersstudyingad attitudeeffectshaveused
testfordifferencesin the strengthof relationships
fornovel differentmedia (primarilyprint or television),variation
versusfamiliarbrands.There are two reasonsto expect in effects across media generallyhas not been assessed
strongereffectsof ad attitudesfor novel brands.First, (for an exception, see Liu and Stout [1987]). Chaiken
Mooreand Hutchinson(1983) have shownthat priorat- and Eagly(1983) found that messageinformation(and
titudestowarda brandcan affectad attitudesand related productcognitions)was moreimportantin ad outcomes
constructs.Thus,forfamiliarbrands,priorbrandattitudes for print media, whereasa more peripheralevaluative
may accountfor variationin relationshipsinvolvingad cue, communicatorlikability,was more importantfor
attitudes.Second,establishedbrandattitudestowardfa- audio and videotape.Therefore,it may be expectedthat
miliarbrandsare not likelyto be as stronglyaffectedby the impact of ad attitudes, as a more peripheraleval-
ad attitudesas are brandattitudestowardnovel brands uativecue, would be greaterfor studiesusingtelevision,
about which consumers have little other information whereasbrandcognitioneffectsmaybe relativelygreater
(Johnsonand Eagly 1989). for studies using print media.
The second product-relatedcharacteristicis the type
of productused in stimulus ads. Approximatelyhalf of Imbeddedness. In some studies, ads were shown
the studiesin researchon ad attitudeshave assessedthe underconditions in whichthey wereimbeddedin other
effects of ads for common repeat-purchaseproducts, material, often to mask experimenterinterest in the
whereasothershave used ads for a wide varietyof other ads. Thus, imbeddingadvertisingin otherprogramming
types of productsand services. Differencesin the type or print content is likely to reduceboth subjects'focus
of productmay promptdifferencesin subjects'involve- on ads and their elaborationas comparedwith the sit-
ment with, cognitive elaborationof, and affecttoward uation in which subjects are exposed to ads as stand-
alone messages.
advertisedinformation(Hoyer 1984).
Experimental Instructions. In some studies, sub-
jects were explicitly told that their reactionsto the ad-
ProcessingGoals vertisingthey would watch were of interest,whereasin
A number of individual studies have suggestedthat others this interest was masked by experimentalpro-
relationshipsinvolving ad attitudesare affectedby sub- cedures.We expectthat elaborationwill be greaterwhen
jects' processinggoals at the time of ad exposure(e.g., subjectsknow of experimenterinterestin ad reactions.
Gardner1985;Keller 199la; Madden,Allen, and Twi- Design. Brinbergand Jaccard(1986) suggestusing
ble 1988; Park and Young 1986). Gardner(1985) in- type of designto partitiondata in meta-analyses.Many
dicated that the effectsof ad attitudesdifferdepending studies in our sample utilized within-subjectsdesigns
on whether subjectsviewed ads with an ad- or brand- to increase power or to control for individual differ-
evaluationset. Maddenet al. (1988) found that subjects ences, whereasothersused completelybetween-subjects
who wereinstructedto evaluatead executionshad more designs. Thus, we expect that within-subjectsdesigns
subduedaffectiveresponsesto adsthan did subjectswho may yield largereffects.
were not given any indication that their reactions to
program-imbeddedads were of interest. The Madden Article Focus. Finally, one influence on effects re-
et al. study also found that processingset affectedcog- ported in the literaturemay be a publication bias fa-
nitive and affectiveresponsesdifferently,with sets that voringstudiesreportinglargereffectsizes (Brinbergand
favorgreatercognitiveelaborationenhancingcognitive Jaccard 1986; Monroe and Krishnan 1983; Rosenthal
responsesand inhibiting affectiveresponses.Addition- 1984). In the literaturereviewedhere, studies variedin
ally, Greenwaldand Leavitt(1984) have suggestedthat, terms of the centrality of focus on ad attitudes. If a
when viewersare highly involved, messageelaboration publicationbias exists, it is likely that studies in which
is increasedand cognitive evaluation of ads predomi- ad attitudeis the centralconstructof interestwill report
nates over affectiveprocessing. largereffectsthan studiesin whichad attitudeis of more
In light of these observations,coding study charac- peripheralinterest.Moreover,the significanceof effects
teristicsto permitassessmentof similarquestionsacross involving ad attitudes in studies where Aad is not the
all availablestudiesappearedworthwhile.Threepoten- centralconstructof interestwould providea strongad-
tial moderatorsrelated to motivation or direction of ditional indication of the generalimportanceof ads in
messageprocessingwere codable:advertisingmedium, determiningadvertisingoutcomes.'
whether ads were imbedded in other materials, and
whetherexperimentalinstructionsto subjectsinformed
them that their reactionsto ads were of interest. 'We thank an anonymousreviewerfor this suggestion.
EFFECTS OF AD ATTITUDE 37

DATA SET AND PROCEDURES each pairwise relationship analyzed, the individual
study effects included in our analysis representedin-
Studies for inclusion in the analysis were identified dependentobservations,and no single sampleprovided
througha searchof the Journalof ConsumerResearch, more than a singleobservationin the analysis(Bangert-
Advancesin ConsumerResearch,Journalof Marketing Downs 1986; Brinbergand Jaccard 1986).3
Research,JournalofAdvertising,JournalofAdvertising After establishingthe relevantarticlesfor inclusion,
Research,JournalofMarketing,Journalofthe Academy each studywas independentlycoded by the two authors
of MarketingScience, and CurrentIssues and Research on the 11 variablesof interest identifiedabove. Agree-
in Advertising.In addition,the ABI Informand Psychlit ment averaged95 percentacrossthe 11 variables,with
computerized data bases were searched to ensure a 100 percent agreementon 7 variables.A complete list
completereview(but no additionalstudieswerefound). of studieswith the relevantcodes for each studyis given
These sources were searchedfrom 1981, when the ar- in Table L.'
ticles by Mitchell and Olson (1981) and Shimp (1981)
introducingthe Aadconstructappeared,to issuescurrent ANALYSIS OF PAIRWISE
at the time the analysiswas conducted(July 1991). Ad-
ditional articleswere identifiedthroughbibliographies RELATIONSHIPS
and personalcontacts.In line with the recommendation To present as broad and precise a synthesis as pos-
of Cooper (1989), all publishedpaperswith effects in- sible, we calculatedmean and median correlationsand
volving ad attitudeswere included in the analysis. confidence intervalsfor all pairwiserelationshipswith
The effect-size metric selected for the analysis was ad attitudes that were representedby multiple usable
the zero-orderproduct-momentcorrelationcoefficient. study effects (as reported in Table 2). However, rela-
This is an easily interpretable,scale-free, effect-size tively few studies (five or less) reportedthe effects for
metric (Hedges and Olkin 1985; Hunter and Schmidt summated Fishbein scales, repetition, message sided-
1990). However, few articles reported pairwise rela- ness, comparative advertising, prior brand attitudes,
tionships in a usable form. For example, results from sex-role portrayal,and recall;a greaternumber inves-
multivariatemodels could not be used because "par- tigated effects such as feelings and emotions (17), ad
tialed" statistics are not directly comparable across cognitions (12), and the outcome variables of brand
studies unless the same variables are included in the cognitions (17), brandattitudes (33), and purchasein-
model in each study (Hunter and Schmidt 1990). tentions (14). Thus, becauseof sample-sizelimitations,
To overcomethis problem,correlationmatriceswere we conductedsubgroupanalysesonly for the more fre-
solicited from the authors of relevant articles that did quently studied relationships.
not report statistics in a usable form. For studies for The analysis plan for each pairwiserelationshipfol-
which we did not receivecorrelations,effectsthat could lows the procedurefor correlationcoefficientsdescribed
be included in the analyses were taken from the pub- in Hedges and Olkin (1985). This is a statisticallyrig-
lished reports.In these cases, Student's t and F-ratios orous and commonly used meta-analyticmethod that
with one df in the numeratorwere converted to r by emphasizes assessment of variance in study effects by
means of formulasgiven in Hunterand Schmidt(1990, means of homogeneity tests (Bangert-Downs 1986).
p. 272): These proceduresfacilitate identification of inconsis-
tencies across studies.
r = tlVt2 + N- 2, We reportthe resultsin threesections.First,we report
or resultsfor the mean strengthof each pairwiserelation-
ship and test for homogeneity of effectsacrossall stud-
r= 1/FlVF+N-2. ies. If the results are homogeneous, then the existence
In all, effectsfrom 47 independentsamples(reported of moderatorvariablesis not indicated. When effects
in 43 articles)were included in the analysis.One effect across studies are not homogeneous and the existence
per pairwiserelationshipper sample was used. If more of moderatorsis indicated, we next report the results
than one effect per relationshipwas given for any one
sample (e.g., when ads was a within-subjectsfactorand 3Becauseour unit of analysisis the "independentsample"rather
resultswere reportedby ad), the mean of the reported than the "singlestudy,"some studiesthat reportresultsfrom more
effects(e.g., acrossads) was used so that only one study than one independentsamplemay receivegreaterweightin the anal-
effect per relationshipper sample resulted.2Thus, for ysisthanstudiesreportingresultsfroma singlesample.The alternative
wouldbe to averageeffectsacrossindependentsampleswithinstudies
and weightby the inverse of the varianceof the sum of the sample
sizes. This would not changethe weightingthat we used and would
2Acomparisonwas made of the effectof using mean (vs. median) resultin lesspreciseestimatesof the variancein pairwiserelationships.
within-studyeffectswhen multipleeffectswithin a single studywere 4Correlationmatriceswere unavailablefor 14 studies,and in three
reportedfor a pairwiserelationship.Most instancesof this situation studies a unique pairwiserelationship(i.e., one not assessedin any
involvedthe feelingsand ad attituderelationship.Substitutingme- otherstudy)was examined.Thus, these studieswerenot includedin
dians for means did not change the overall averagecorrelationfor the analysisor in Table 1. A list of these studiesis availablefrom the
this relationship. authors.
TABLE I
CODED CHARACTERISTICSOF INCLUDEDSTUDIES

Consumer
Novel nondurable
Single or Student or Feelings Single or or (CND) or Instructed Within-or
multiple nonstudent measured or multiple familiar "other" Advertising to attend between-subject Focus on
Study scales sample manipulateda feelingsa product product type medium Imbedded to ads design Aadb

Alpert, Golden,
and Hoyer
1982 Single Student Manipulated TV No Yes Between DV
Park and Young
1983 Single Nonstudent Novel CND TV No Yes Between DV
Belch and Belch
1984 Multiple Student Manipulated Novel Service TV Yes Yes Between DV
Golden and
Johnson
1984 Single Student Manipulated Familiar Mixed TV No Yes Within DV
Duncan and
Nelson
1985 Multiple Student Manipulated Single Novel CND Radio Yes No Between DV
Gresham and
Shimp 1985 Multiple Student Familiar Mixed TV No Yes Within DV
Reddy and
LaBarbera
1985 Multiple Student Novel CND Print No Yes No
Aaker,
Stayman,
and
Hagerty
1986 Single Student Measured Single Familiar Mixed TV No Yes Within DV
Batra and Ray
1986 Single Nonstudent Measured Multiple Familiar Mixed TV No Yes Within DV
Kilbourne 1986 Multiple Nonstudent Novel Durable Print No Yes Between Aad
MacKenzie et al.
1986:
Study 1 Multiple Nonstudent Novel CND TV Yes Yes Aad
Study 2 Multiple Student Novel CND TV/radio Yes/No Yes/No Aad
Mitchell 1986 Multiple Student Novel CND Print No Yes Within Aad
Park and Young
1986 Single Nonstudent Novel CND TV No Yes Between DV
Zinkhan et al.
1986 Multiple Nonstudent Novel Mixed Print No Yes Between DV
Edell and Burke
1987:
Study 1 Single Student Measured* Multiple* Familiar Mixed TV Yes/No Yes Within DV
Study 2 Single Student Measured* Multiple* Novel Mixed TV Yes No Within DV
Goldberg and
Gorn 1987 Single Student Manipulated Single Familiar CND TV Yes No Within No
Holbrook and
Batra 1987 Multiple Nonstudent Measured Multiple Familiar Mixed TV No Yes Within DV
Keller 1987 Multiple Nonstudent Novel CND Print No Yes Within DV
Leigh, Rethans,
and
Whitney
1987 Multiple Nonstudent Familiar CND Print No Yes Between DV
Muehling 1987 Multiple Student Novel Durable Print No Yes Between Aad
Cox and Cox
1988 Multiple Student Novel CND Print Yes/No Yes/No Between Aad
Hill1988 Multiple Student Manipulated Single Novel CND Print No Yes Within DV
Machleit and
Wilson
1988 Multiple Student Measured Multiple Both Mixed TV Yes No Between Aad
Madden et al.
1988 Multiple Student Manipulated* Single* Both CND Radio Yes/No Yes/No Between Aad
Muehling and
Laczniak
1988 Multiple Student Novel Durable Print Yes No Between A-
EFFECTS OF AD ATTITUDE 39

TABLE 1 (Continued)

Consumer
Novel nondurable
Single or Student or Feelings Single or or (CND) or Instructed Within-or
multiple nonstudent measured or multiple familiar "other" Advertising to attend between-subject Focus on
Study scales sample manipulateda feelingsa product product type medium Imbedded to ads design Aadb

Stayman and
Aaker 1988 Single Student Measured Multiple Familiar Mixed TV Yes Yes Within Aad
Burke and Edell
1989 Single Student Measured* Multiple* Novel Mixed TV Yes No Within Aad
Droge 1989 Multiple Student Novel CND Print Yes No Between Aad
Hill1989 Multiple Student Familiar Political Print No Yes Between DV
Kamins et al.
1989 Multiple Nonstudent Novel Service Print No Yes Between No
Laczniak and
Carlson
1989 Multiple Student Novel Durable Print Yes No Between Aad
MacKenzie and
Lutz 1989:
Study 1 Multiple Student Novel Durable Print No Yes Aad
Study 2 Multiple Student Novel Durable Print No Yes Aad
Chattopadhyay
and Basu
1990 Multiple Student Manipulated Single Novel CND TV Yes No Between DV
Homer 1990:
Study 1 Multiple Student Novel CND TV Yes Yes Between Aad
Study 2 Multiple Student Novel Service Print No Yes Between Aad
Kamins 1990 Multiple Student Novel Durable Print No Yes Between No
Miniardet al.
1990 Multiple Student Novel CND Print No Yes Between Aad
Russo and
Stephens
1990 Multiple Student Measured Multiple Novel Mixed TV No Yes Within DV
Yi 1990a Multiple Student Manipulated Single Novel Durable Print Yes Yes Between DV
Yi 1990b Multiple Student Novel Durable Print No Yes Between No
Keller 1991b Multiple Student Novel CND Print No Yes Within DV
Keller 1991 a Multiple Nonstudent Novel CND Print No Yes Within DV
Maclnnis and
Park 1991 Multiple Student Measured Multiple Novel CND TV Yes No Between DV
Olney et al.
1991 Multiple Student Measured Multiple Familiar Mixed TV No Yes Within DV

NOTE.-Empty cells in the table indicate characteristics that were either absent from or not codable in the studies.
aAsterisks indicate that, although the relationship between feelings and Aadwas studied, the study effects were not available in this analysis.
bDVmeans Aadwas used as one of several dependent variables; "No" indicates consideration of Aadwas only incidental to the study; Aadindicates the study focus
centered on Aad-

of moderatoranalyses.Finally, we use regressionanal- tervals. In addition, Figure 1 provides stem-and-leaf


yses to assesswhetherthe moderatorsidentifiedin this plots of the distributionof effectsacrossstudies.As can
study can account for the preponderanceof variance be seen in Table 2 and Figure 1, significantvariation
in the relationshipsstudied. Significantresidual vari- in effectsacross studies is apparentfor each frequently
ance resultingfrom the regressionswould indicate that studied pairwiserelationship.
the model as specifieddoes not totally account for the Tests of effecthomogeneitywere conductedon Fish-
variancein study effects. er's z-transformsof the correlationcoefficientsaccord-
ing to the formula
OverallHomogeneity Q(n - 3)(Zi-Z
For each pairwise relationship, Table 2 reports the wherez is the mean of the weightedz-transformedcor-
numberof studiesfor each effect,the cumulativesample relations(each study effectwas weightedby the inverse
size from those studies, mean and median observed of its variance to give greaterweight to more precise
correlations,mean and median correlationscorrected estimates).This statistichas an approximatechi-square
for measurementerror,and 95 percent confidence in- distributionwith k - 1 degreesof freedom (wherek is
40 JOURNAL OF CONSUMER RESEARCH

TABLE 2
DESCRIPTIVESTATISTICS FOR PAIRWISERELATIONSHIPS

95 Percent
Mean Median Mean Median confidence
Cumulative observed observed corrected corrected interval for
Relationship N r F r r corrected f

Feelings and ad attitude (n = 17) 1,882 .49 .50 .54 .55 .00 < p < .84
Ad cognitions and ad attitude (n = 12) 1,700 .49 .42 .55 .48 .26 < p < .75
Ad attitude and brand cognitions (n = 17) 2,535 .28 .28 .33 .32 .08 <p< .57
Ad attitude and brand attitude (n = 33) 4,613 .60 .60 .67 .68 .26 <p < .85
Ad attitude and purchase intention (n = 14) 1,765 .38 .33 .43 .36 .03 < p< .71
Ad attitude and le1b1(n = 5) 561 .25 .30 .28 .34 -.02 < p< .58
Repetition and ad attitude (n = 4) 905 .15 ... .17 ... .09 < p< .25
Sidedness and ad attitude (n = 3) 537 .36 .37 .41 .42 .35 < p< .39
Comparative versus noncomparative and
ad attitude (n = 3) 600 .26 .27 .31 .33 .22 < p< .40
Prior brand attitude and ad attitude (n = 2) 260 .31 ... .34 ... .24 < p< .44
Sex-role portrayal and ad attitude (n = 2) 189 .17 ... .20 ... .16 < p< .24
Ad attitude and recall (n = 2) 389 .17 ... .20 ... .17 < p< .23

the numberof studies included in the analysis;Hedges uncorrected correlations. The first conclusion drawn
and Olkin 1985, p. 235). The analyses supportedthe from Table 4 is that the results of the analyses of the
conclusion of significant variation for all of the fre- correctedand uncorrectedcorrelationsare very similar.
quently studied relationships(p < .01). Significanceof the moderatordifferedbetweenthe two
Further, for each pairwise relationship, the homo- analysesfor only one of 47 tests (i.e., the effectof study
geneity statistic, Q, was computed with study effects focus on the ad attitude and purchaseintentions rela-
successively deleted to identify outliers (a procedure tionship, with only the correctedresult significantat p
recommended by Hedges and Olkin [1985], p. 256). < .05). Thus, the discussionbelow does not distinguish
Table 3 reportsthe resultsof this procedure,identifying betweenthe correctedand uncorrectedresults(statistics
the studies that were deleted from the overall analyses for the subgroupanalyses are for the correctedcorre-
to achieve homogeneoussets of study effectsand com- lations).
paring the mean corrected correlation for all studies
reportingthe relationshipwith the mean correctedcor- Numberof Scale Items and Sample Type. Two po-
relation for the reduced homogeneous set of study ef- tential moderators,number of scale items and type of
fects. The table reveals that the number of deletions sample, were coded to assess the impact of reliability
necessaryto achievea homogeneousset of effectsranged and sample homogeneity. Use of multiple-item scales
from 1 of 12 (8 percent) for the ad cognitions and ad to measure ad attitude, because they should be more
attituderelationshipto 6 of 17 (35 percent)for the feel- reliable(Peterand Churchill 1986), was expectedto re-
ings and ad attituderelationship.The proportionof ob- sult in strongerrelationshipsthan the use of single-item
servations deleted to achieve homogeneity of effects, measures. This expectation was supported for the ad
althoughhighin some cases,is consistentwith previous attitude and brand attitude relationship.The number
meta-analysesand the recommendationsof meta-an- of scaleitems had a smallbut not statisticallysignificant
alytic methodologists(e.g., Hedges and Olkin 1985, p. effect on the feelings and ad attitude relationship.No
250; Johnsonand Eagly 1989).Also, althoughrelatively otherrelationshipswereaffectedby the numberof scale
few outliers needed to be eliminated to achieve ho- items. Thus, it appearsthat the effects of using multi-
mogeneity for the ad- and brand-cognitionsvariables, item scales are relativelysmall with respect to the va-
analysisof the varianceacrossstudiesis still useful(Rao lidity of relationships.
and Monroe 1989). Thus, in summary, these data do The type of sample appearedto have a more consis-
indicate significantvariationin the frequentlystudied tent effect than did the number of scale items across
pairwiserelationshipsand suggestthe usefulnessof an- the five relationships studied. As expected, student
alyzing the effectsof moderatorvariables. samples (expectedto be more homogeneous;Calderet
al. 1981) tended to yield higher correlationsthan did
Tests of Moderators nonstudent samples. This result was particularlyap-
parentfor the relationshipsbetween ad attitudeand its
Table4 reportsthe resultsof the moderatoranalyses, antecedents (i.e., feelings and ad-relatedcognitions).
includingthe meansand significanceof each moderator The effect of sample type on the relationshipsbetween
for both the corrected(i.e., for measurementerror)and ad attitudeand outcome variables(i.e., brandattitudes
EFFECTS OF AD ATTITUDE 41

FIGURE1
STEM AND LEAF PLOTS OF PAIRWISERELATIONSHIPS

Feelings Ad Cognitions- Ad Atfitude- Ad Attitude- Ad Atfitude-


Ad Attitude Ad Attitude BrandCognitions BrandAtfitude PurchaseIntention

Stem Leaf Stem Leaf Stem Leaf Stem Leaf Stem Leaf

.9 .9 .9 .9 .9
.8 1 .8 .8 .8 00 113666 .8
.7 068 .7 1 13 .7 .7 013578 .7 5
.6 14 .6 .6 .6 28999 .6 2
.5 23378 .5 778 .5 12 .5 11225569 .5 35
.4 3 .4 06778 .4 00236 .4 346 .4 1 18
.3 2 .3 1 .3 0 12 7 9 .3 1 .3 06
.2 69 .2 .2 0 44 .2 58 .2 2 67 9
.1 79 .1 .1 6 .1 .1 3
.0 .0 .0 9 .0 .0
NOTE.-Plots are of corrected correlation coefficients.

and purchaseintentions) was directionallyconsistent chase intentions. In contrast, product type appearsto
with expectationsbut was not statisticallysignificant. influenceonly the more intermediaterelationships(i.e.,
The use of student subjectsin researchon ad attitudes in the causal sequence;see Homer 1990;MacKenzieet
thus appearsto have an upward-biasingeffect on the al. 1986) between ad attitudesand feelings and brand-
strengthof some relationships,and this effect should relatedcognitions.
be noted as a limiting condition on the generalizability Cognitive Processing Goals. Three moderators (ad-
of resultsgeneratedfrom student samples. vertising medium, imbeddedness of ads in other ma-
Product-relatedCharacteristics. Two product-re- terials, and experimental instructions) were coded in
lated characteristicswere examined as potential mod- an attempt to account for the influence of processing
erators. We expected relationships to be weaker for goals on the relationships studied. These processing-
studiesthatusedfamiliarbrandsin stimulusadsbecause goal variables influenced feelings and brand-related
strongpriorbrandattitudesmight inhibit ad effective- cognitionsin distinctlydifferentways. As expected,un-
ness. It is interestingthat this expectationwas supported der conditions favoring greater message elaboration
for the outcome variablesof brand attitudes and pur- (i.e., for studies using print media, nonimbeddedads,
chase intentions but was not significantfor effects in- and those instructingsubjectsto attend to ads),the cor-
volving feelings, ad-relatedcognitions, and brand-re- relationbetweenfeelingsand ad attitudewas lowerthan
lated cognitions. Thus, it appears that the effects of underconditionsnotfavoringmessageelaboration.This
antecedent variables, such as feelings and ad-related resultis consistentwith expectationsand with the results
cognitions, and of ad attitudes on process constructs, of Madden et al. (1988). However, under conditions
such as brand-related cognitions, are not strongly favoring greater elaboration, the correlation between
influenced by prior brand knowledge. However, ad attitudeand brandcognitionswas greaterthan under
the effect of ad attitudes on relevant outcomes does conditions not favoring message elaboration (the dif-
appear to be influenced by the existence of brand ferencewas significantfor advertisingmedium but not
familiarity. for imbeddedness and experimental instructions, al-
The secondproductcharacteristicexaminedwas type though the mean differencesfor the latter two were di-
of product (i.e., consumer nondurable vs. "other" rectionally consistent). Thus, processinggoal appears
producttypes).It is interestingthat this analysisyielded to have an importantbut inconsistentinfluenceon pro-
effectsoppositeto those resultingfrom the novel versus cesses relatedto ad attitudes.
familiar brand analysis. Product type significantly Processinggoal appearedto have no influenceon ei-
moderatedthe feelingsand ad attitudeand ad attitude ther the ad cognitions and ad attitude relationshipor
and brand cognitions relationshipsbut had no signifi- on the ad attitude and brand attitude relationship.
cant effect on the ad cognitions and ad attitude, ad at- However, both imbeddednessand instructionssignifi-
titude and brandattitude,and ad attitudeand purchase cantly moderatedthe ad attitude and purchaseinten-
intentions relationships. Thus, in summary, product tions relationship, with greater elaboration yielding
novelty does not appearto influence the generationof highercorrelations.Why processinggoal appearsto in-
ad attitudesbut does affectthe ability of ad attitudesto fluence purchase intentions but not brand attitude is
influence outcomes such as brand attitudes and pur- not clear and may warrantfurtherresearch.
42 JOURNAL OF CONSUMER RESEARCH

TABLE3
COMPARISON OF COMPLETESETS OF STUDIES AND HOMOGENEOUSSUBSETS

Complete set Homogeneous subset

No. of Cumulative Corrected No. of Cumulative Corrected


Relationship studies N r studies N r Studies deleted

Feelings and ad
attitude 17 1,882 .54 11 1,163 .48 Machleit and Wilson (1988)
Madden et al. (1988)
Stayman and Aaker (1988)
Chattopadhyay and Basu (1990)
Golden and Johnson (1984)
Batra and Ray (1986)
Ad cognitions and ad
attitude 12 1,700 .55 11 1,497 .53 MacKenzie and Lutz (1989), study 1
Ad attitude and
brand cognitions 17 2,535 .33 15 2,277 .35 Maclnnis and Park (1991)
Chattopadhyay and Basu (1990)
Ad attitude and
brand attitude 33 4,613 .67 23 2,583 .68 Keller (1987)
Keller (1991 a)
Keller (1991 b)
Zinkhan et al. (1986)
MacKenzie and Lutz (1989), study 1
MacKenzie et al. (1986), study 1
MacKenzie et al. (1986), study 2
Droge (1989)
Stayman and Aaker (1988)
Reddy and LaBarbera (1985)
Ad attitude and
purchase intention 14 1,765 .43 10 1,224 .42 Muehling (1987)
Miniardet al. (1990)
Duncan and Nelson (1985)
Muehling and Laczniak (1988)

StudyDesign and Focus. Studydesign appearedto on both the feelingsand ad attitudeand ad attitudeand
have an inconsistentbut ratherpervasiveinfluence.Be- brandattitude relationships,as well as a significantin-
tween-subjectsdesigns resulted in higher correlations fluence on the ad attitude and purchaseintentions re-
for the feelingsand ad attituderelationship(significant lationship(for the correctedcorrelations).As expected,
difference)and for the ad attitude and purchaseinten- resultswere strongerwhen ad attitudewas a more cen-
tions relationship(differencesof .11 and .13 for uncor- tralfocus of the article.These resultssuggesta potential
rectedand correctedcorrelations,respectively;however, publicationbias favoringlargereffectsinvolving ad at-
neitherwas significant,perhapsbecausetherewereonly titude when it is a centralfocus of the article.However,
two usable within-subjectsstudies). This higher corre- it should also be noted that even for those studies in
lation is unexpectedsince within-subjectsdesignstend whichad attitudewas not the centralfocus of the article,
to reduce errorvarianceby making each experimental the mean correlationsare still quite high (rangingfrom
unit, in essence, its own control (Winer 1971). One ex- .26 to .61), indicating that it is the magnitude of the
planation for the feelings and ad attitude result is that relationships,ratherthan their significance,that is at
virtuallyall of these studiesused ads as the within-sub- issue. In supportof this conclusion, an analysis of the
jects factor, and feeling responses across ads may be "file drawer"problem(Rosenthal 1984) indicatedthat
quite variable(Holbrook and Batra 1987; Olney, Hol- at least 453 studieswith null resultswould be necessary
brook, and Batra 1991). Hence, it is possible that as- to reduce the ad attitude and purchase intentions re-
sessing the impact of feelings on ad attitude across an lationshipto nonsignificance.Eachof the otherpairwise
assortmentof ads mixes qualitativelydifferenteffects relationshipswould requirea greaternumber of zero-
and diminishesthe overallstrengthof the relationship. effect studies to lead to conclusions of nonsignificance
Between-subjects designsdid resultin significantlylower (e.g., at least 758 studiesfor the feelingsand ad attitude
correlationsfor the ad attitude and brand attitude re- relationship).
lationship.
It is noteworthythat the extent to which ad attitude OtherResults. Our expectationthat measuringfeel-
was the focus of the article had a significantinfluence ings wouldyield highercorrelationsbetweenfeelingsand
EFFECTS OF AD ATTITUDE 43

TABLE 4
SUBGROUP MEANS BY MODERATORVARIABLES

Ad attitude and Ad attitude and


Feelings and Ad cognitions and brand Ad attitude and purchase
ad attitude ad attitude cognitions brand attitude intention

Single versus multi-item scale


(ad attitude):
Uncorrected K .43 vs. .52 .50 vs. .49 .30 vs. .28 .53 vs. .61** .37 vs. .42
Corrected r .48 vs. .58 .57 vs. .55 .35 vs. .33 .61 vs. .68** .43 vs. .46
N 7vs. 10 1 vs. 11 3vs. 14 7vs. 26 3vs. 11
Student versus nonstudent
subjects:
Uncorrected r .50 vs. .25a .52 vs. .41* .29 vs. .26 .60 vs. .59 .41 vs. .32*
Corrected r .56 vs. .29a .58 vs. 47* .34 vs. .31 .68 vs. .65 .46 vs. .36*
N 16vs. 1 9vs. 3 14vs. 3 26vs. 7 12vs. 2
Novel versus familiarproduct:
Uncorrected K .43 vs. .41 .49 vs. .50 .28 vs. .32 .62 vs. .38** .45 vs. .29**
Corrected r .48 vs. .45 .55 vs. .57 .33 vs. .37 .69 vs. .43** .50 vs. .32**
N 6vs.9 11 vs. 1 16vs. 1 28vs.4 10vs.3
Consumer nondurable versus
' other" product type:
Uncorrected K .30 vs. .50a .47 vs. .52 .25 vs. .36** .62 vs. .65 .45 vs. .41
Corrected r .32 vs. 57a .52 vs. .59 .28 vs. .42** .70 vs. .74 .44 vs. .46
N 6vs.1 6vs.5 9vs.5 16vs.8 2vs.11
TV versus print advertisements:
Uncorrected r .47 vs. .25a .51 vs. .49 .23 vs. 35** .58 vs. .60 .38 vs. .46
Corrected r .51 vs. .26a .55 vs. .55 .26 vs. .41** .65 vs. .67 .44 vs. .50
N 14vs. 1 3vs. 9 7vs. 8 14vs. 17 5vs. 8
Imbedded versus nonimbedded:
Uncorrected K .54 vs. .31** .46 vs. .50 .26 vs. .34 .58 vs. .60 .32 vs. .50
Corrected r .59 vs. 35** .50 vs. .57 .31 vs. .38 .65 vs. .67 .36 vs. .55*5
N 8vs.7 4vs.8 9vs.7 13vs. 17 7vs.7
Instructed to attend to ads versus
not instructed to attend:
Uncorrected K .40 vs. .58** .49 vs. .49 .30 vs. .28 .60 vs. .55 .46 vs. .24**
Corrected r .44 vs. .64** .56 vs. .54 .35 vs. .33 .68 vs. .62 .52 vs. .27**
N 11 vs.5 9vs.3 9vs.7 22vs.9 10vs.4
"Ad attitude study" versus ad
attitude as dependent variable:
Uncorrected r .72 vs. .37** .49 vs. .50 .29 vs. .26 .65 vs. .54** .43 vs. .34
Corrected r .79 vs. .42** .54 vs. .57 .34 vs. .30 .73 vs. .61** .48 vs. 39*
N 3vs. 13 7vs.5 10vs.7 15vs. 17 8vs.6
Withinversus between-subjects
design:
Uncorrected K .42 vs. .54** .47 vs. .47 .30 vs. .29 .65 vs. .53** .31 vs. .42
Corrected K .47 vs. .61** .54 vs ..51 .35 vs. .34 .73 vs. .59** .34 vs. .47
N 11 vs.6 4vs.6 5vs.8 11 Vs. 17 2vs. 12
Single versus multiple feelings:
Uncorrected K .50 vs. .57
Corrected r .57 vs. .63
N 6vs. 9
Measure versus manipulate
feelings:
Uncorrected K .57 vs. 49**
Corrected r .63 vs. 45**
N 10vs. 7

NOTE.-All data are means.


aWhileresults for this test are significant at p < .01, it is noted that only one observation is available for one level of the moderator.
*Significant at p < .05.
**Significantat p < .01.

ad attitudethan would manipulatingfeelingswas sup- Finally,we note that, for severalpotentialmoderator


ported.However,ourexpectationthatassessingsinglever- variables(e.g.,studentvs. nonstudentsubjects,consumer
sus multiplefeelingswouldaffectthe relationshipbetween nondurablesvs. other types of stimulus products,and
feelingsand ad attitudewas not supported. televisionvs. print ads), one of the subgroupscontained
44 JOURNAL OF CONSUMER RESEARCH

so few observations(one or two) that meaningfulconclu- For both the feelings and ad attitude and the brand
sions were difficultto draw,althoughsizable and often cognitionsand ad attituderelationshipsthe test statistic
statisticallysignificantdifferencesresultedbetweena single was not significant (X2(4) = 6.65, p > .10; and X2(6)
correlationrepresentingone of the subgroupsand the = 10.40, p > .25, respectively).Thus, the moderators
mean of multiple study effects representingthe other appearedto account for the variationin these relation-
subgroup.Thus,futureresearchprovidingadditionaldata ships.Forthe ad cognitionsand ad attituderelationship,
pointswouldbe veryuseful.Forexample,researchrelating the test statisticwas significant(X2(8)= 18.54,p < .05).
feelingsto ad attitudeis homogeneouswith respectto However, as noted above, elimination of one outlier
employingstudentsamplesand using televisionads for (MacKenzie and Lutz 1989, study 1) yielded overall
consumernondurableproducts,but it wouldbe usefulto homogeneity for this relationship.
includestudiesusing nonstudentsamplesand print and For both outcome relationships, ad attitude and
radioads for productsotherthan consumernondurables brandattitudeand ad attitudeand purchaseintentions,
to furtherassessthe moderatingeffectsof thesevariables. the test statistic was significant (X2(19) = 129.94, p
In summary,the 11 moderatorsanalyzedhad varied < .01; x2(3) = 21.56, p < .01, respectively),indicating
effectson the five pairwiserelationshipsstudied.It is ev- that the model did not account well for the variancein
ident that differencesin methodsacrossstudieshave led study effects. Thus, for both of these relationshipsthe
to much of the variancein resultsin articlesinvolvingad clustering procedure recommended by Hedges and
attitude,and understandingthese methodologicaleffects Olkin (1985, p. 280) was implemented to determine
leadsto a clearerpictureof the robustnessof the effects which observationsdifferedsignificantlyfrom one an-
of ad attitudes.To determinehow completelythe coded other (in a manner analogousto the Scheff6procedure
studycharacteristics wereableto accountfor variancein for testing post hoc contrasts).
the study effects,furtheranalyseswere undertaken.The For the ad attitude and brand attitude relationship,
next sectionreportsthese analyses. the procedureindicated that the five smallest correla-
tions (r = .25-.44; Droge 1989; Reddy and LaBarbera
1985; Stayman and Aaker 1988; Yi 1990a; Zinkhan,
RegressionAnalyses Locander,and Leigh 1986) differedsignificantlyfrom
Hedgesand Olkin (1985, pp. 240-241) suggesta two- the threelargestcorrelations(all r's = .86; Keller 199Ib;
step procedurefor analyzingunexplainedvarianceafter MacKenzie et al. 1986 study 2; MacKenzie and Lutz
the significance of moderators for correlation coeffi- 1989, study 1). The procedurealso indicated that the
cients is assessed.First, homogeneitywithin each level five largest correlations (the three above plus Keller
of a significant moderator is assessed because, if ho- 1991a and MacKenzieet al. 1986 study 1; r = .83 and
mogeneityis found, it can be concluded that the mod- .80, respectively)differedsignificantlyfrom the smallest
eratoradequatelyaccounts for the variationin the cor- (Droge 1989; r = .25).
relationsacrossthe studies analyzed.If homogeneityis For the ad attitudeand purchaseintentions relation-
not found, individual study correlations(convertedto ship, the clustering procedure indicated that the six
Fisher'sz-transforms)are regressedon the coded study lowest correlations(r's = .13-.29; Duncan and Nelson
characteristics(the moderators)to assesswhetherthey, 1985; Kamins et al. 1989; Machleit and Wilson 1988;
taken together, can account for the variation. The Muehling and Laczniak 1988; Stayman and Aaker
weighted sums of squares of the residuals from the 1988; Yi 1990a) differedsignificantlyfrom the highest
regressionhave an approximatechi-squaredistribution correlation(Muehling 1987; r = .75) and that the two
with k - p degreesof freedom (wherek is the number highest correlations(Muehling 1987; Miniard,Bhatla,
of study effects and p is the number of predictorvari- and Rose 1990; r = .62) differedsignificantlyfrom the
ablesin the model).Significantvaluesof the test statistic lowest correlation (Muehling and Laczniak 1988; r
suggestthat the model is not well specifiedto account = .13). Post hoc scrutinyof these significantdifferences
for variancein the study effects.5 did not suggestany moderatorvariablesin addition to
In our research,the subgroupswere not completely those originallycoded and describedabove.
homogeneous for any of the moderator variables. In summary, these analyses indicate that the 11
Therefore,the regressionanalysis was run for each of moderators studied in this research not only had a
the five pairwiserelationshipsstudied. number of significant influences on the five pairwise
relationshipsstudiedbut also could adequatelyaccount
for the total variationin the study effects for two rela-
5Onlythe test statisticis reportedbecausebetasfromthe regressions tionships involving ad attitude and the intermediate
are not interpretablefor two reasons.First, significantcollinearity variablesof feelingsand brandcognitions. Elimination
was found. Second, missing or nonclassifiableeffects for any one of one outlier would account for variation in a third
moderatorexcluded a study from the regressionanalysis. Because relationship(ad cognitions and ad attitude). However,
this occurredfor a numberof studies,the two-grouptest reportedin
the moderatorsectiongives a more accuratepictureof the influence furtherresearchis necessaryto accountfor the variation
of each moderatorbecauseit makes more complete use of the data between ad attitude and the two outcome variablesof
(i.e., all studiescodableon each moderatorareusedin each analysis). brandattitudesand purchaseintentions.Althoughfour
EFFECTS OF AD ATTITUDE 45

of our moderatorssignificantlyinfluencedeach of these FIGURE2


relationships(see Table 4), togetherthey could not ac- MODELSOFAD ATTITUDE
FOURALTERNATIVE
count forall of the variationfound, suggestingthat other A. Affect TransferHypothesis
potentialmoderatorvariablesnot includedin the mod-
Ad Cognitions - Ad Attitude
els exist.
BrandCognitions-0-Brand Attitude-p- PurchaseIntention
CAUSAL MODEL ANALYSIS
One of the most prevalent recent directions in re- B. DualMediationHypothesis
search on ad attitudes has been testing alternative
structuralmodels of the process through which ad at- Ad Cognitions - Ad Attitude

titudes are influenced by antecedent variablesand, in


BrandCognitions- -Brand Attitude- PurchaseIntention
turn, influenceadvertisingoutcomes. The four models
most often tested are those first proposed by Lutz,
MacKenzie,and Belch(1983) and aredepictedin Figure C. ReciprocalMediationHypothesis
2: the affect transfer,dual mediation, reciprocal me- Ad Cognitions -- Ad Attitude
diation, and independentinfluenceshypotheses.
Initial work on the four alternativemodels suggests BrandCognitions--*-Brand Attitude- PurchaseIntention
supportfor the dual mediation hypothesis(Lutz et al.
1983;MacKenzieet al. 1986). Supportfor the dual me- D. IndependentInfluencesHypothesis
diation model has provedrobustto such effectsas level
of involvement and brand considerationset (Gardner Ad Cognitions 0 Ad Attitude
1985; Homer 1990) and the decomposition of ad atti-
BrandCognitions _- Brand Attitude- PurchaseIntention
tude into claim and nonclaim dimensions (Miniardet
al. 1990).
However,the studiestestingthe model representonly
a small portion of the stimuli, situations, and other (Joreskogand S6rbom 1986), and specifyingthe model
contextual factors under which the role of ad attitude so that correlationsamong the latent factorswereequal
has been testedthroughoutthe researchstream.Hunter to the correlationsamong the observedvariables(Werts
and Schmidt(1990, p. 40) suggestthat the role of meta- et al. 1976). The values of the intercorrelationsamong
analysisin theorydevelopmentcan be enhancedby us- the factorswerethen constrainedto be equal acrossthe
ing weighted mean correlations based on aggregated two groups.The significanceof the chi-squarefit statistic
empirical study effects in path-analyticmodels speci- resultingfrom maximum likelihood estimation consti-
fying hypothesizedcausal relationships.Because path- tutesa test of the equalityof the two correlationmatrices
analyticmodels have played such an importantrole in (Werts et al. 1976). Estimating the two-group model
researchon ad attitudes,we estimatedthe variouscausal with all parametersconstrainedto be equal across the
models proposed as explanations for the effects of ad two groupsresultedin a fit of X2(15) = 14.42 (p = .49),
attitudes using the mean correlationsfrom the meta- which indicates that no significant overall difference
analysis.We performedthis analysisto assess the gen- betweenthe two matricesof averagecorrelationsexists.6
eralizabilityof previouspath-analyticresults.Thus, our On the basis of this result, the matrix of averagecor-
primarypurposein conducting this meta-analytictest relations calculated from all available pairwise corre-
was to evaluate whether the support for the dual me- lations was used in subsequentanalyses.
diation model was consistent with data generalized Causal Model Results. Tests were conducted as-
acrossthe researchstream.In addition, we assessedthe sessing the fit of the four causal models proposed by
correspondencebetween the path values found in pre- MacKenzie et al. (1986) to the aggregateddata from
vious researchwith those based on the meta-analysis. the availablestudies.Again, the fourcompetingmodels
Causal Model Estimation. Before estimating the are depicted in Figure2.
causal model, we assessedthe effects of using a matrix Estimationof the affect-transfermodel yielded X2(6)
of averagecorrelationsconsistingof effectsfrom differ- = 171.00, indicatingthat the model fit the data poorly.
ent studies as input data for causal model estimation. The reciprocalmediation model produceda fit to the
A test of overall equality of correlation matrices, de- data of X2(5) = 156.27, also indicatinga ratherpoor fit.
scribedby Wertset al. (1976), was conductedcomparing Fitting the independent influences model resulted in
the matrix computed from all available studies with a x2(6) = 693.70, indicating a worse fit to the data. Fi-
matrixof averagecorrelationscomputed from only the nally, fittingthe dual mediationmodel to the aggregated
studies that made up homogeneous subsets for each
pairwiserelationship. 6In this and the followinganalyses,the median sample size from
The test consisted of estimatinga two-groupconfir- the meta-analysisof relationshipsamong constructsin the model (N
matory-factor-analysismodel by means of LISREL = 1,252) was used.
46 JOURNAL OF CONSUMER RESEARCH

data producedx2(5) = 17.32. Althoughthis is still sta- FIGURE3


tisticallysignificant(p < .01), the Joreskogand Sorbom STRUCTURALPATH ESTIMATESAND T-VALUES
goodness-of-fitindex (.995) and root mean square re- MODEL
FOR THE DUAL-MEDIATION
sidual(.018) indicatethe dual-mediationmodel'sability .52
to reproducethe originalmatrixof averagecorrelations. Ad Cognitions - - Ad Attitude
(21.53)
These statisticswereless supportiveof the othermodels .57
(the goodness of fit = .952, .955, and .853 and the root (25.58)

mean square residual = .129, .115, and .258 for the


affect-transfer,reciprocalmediation, and independent .20 .73
BrandCognitions - -Brand Attitude IN PurchaseIntention
influences models, respectively).The dual mediation (8.59) (37.78)
model's dramaticimprovementin fit over the compet-
NOTE.-Structural path estimates are standardized coefficients; t-values are
ing models is evidence of its superiorityas a represen- in parentheses, and t-values > 2.0 are statistically significant (Joreskog and
tation of the aggregateddata from the availablestudies. Sorbom 1986, chap. 3, p. 12).
Thus, these results support the conclusions of Mac-
Kenzie et al. (1986), Miniard et al. (1990), Homer
(1990), and otherswho found supportfor the dual me- any meta-analysis,the dataprovidea quantitativesum-
diation model. mary of the 47 samples reportedin the 43 articles for
Analysis of ParameterValues. Parameterestimates which we could obtain usable data. As shown in Table
and t-valuesfor the dual mediationmodel arepresented 2 and discussedabove, the aggregatedstudy effectssug-
in Figure3. Pathcoefficientswereestimatedon weighted gest a substantialand significantrelationshipbetween
averagecorrelationscorrectedfor measurementunre- ad attitudesand a numberof importantconstructs,in-
liability.Therefore,pathsfromlatentconstructsto their cluding both antecedents (feelings, r = .54; ad-related
single observedindicatorswere set to one. Each struc- cognitions, r = .55) and "downstream"consequences
tural path estimated was statistically significant (p (brand-relatedcognitions, r = .33; brand attitude, r
< .01). In general,the resultsare consistent with those
= .67; purchase intentions, r = .43). The 95 percent
of past studies.However,two points areworthyof note. confidence interval around the mean correlationsfor
First, results of previous empirical estimates of the the five focal relationshipsincluded zero for only one
dual-mediationmodel havedifferedas to whetherbrand of the relationships(feelings and ad attitude).The fact
cognitionssignificantlyaffectbrandattitudes(cf. Homer that relationships involving ad attitudes were strong
1990; MacKenzie and Lutz 1989; MacKenzie et al. even in studies in which the ad attitude constructwas
1986). The aggregateddata analyzed here suggestthat not a primaryfocus (r = .26 and .61, respectively)sug-
a significantbrand cognition and brand attitude rela- gests its importancein explainingadvertisingeffects.
tionship exists, which is important in understanding An importantgeneralfindingof this study is its sup-
the indirecteffectof ad attitude.It may be worthnoting, port for the dual mediation model that was first pro-
however, that the brand cognition and brand attitude posed by Lutz et al. (1983; MacKenzie et al. 1986).
relationshipis, as in most studies, the weakest in the This model, which posits a direct effect of ad attitude
model. on brandattitudeas well as an indirecteffectvia brand
Second, while the meta-analyticmodel supports a cognitions, has been supportedby a number of studies
significantindirecteffectof ad attitudeon brandattitude (e.g., Homer 1990;Miniardet al. 1990)and thus appears
(via brand cognitions), the direct effect of ad attitude fairlyrobust. However, the meta-analysissuggeststhat
on brandattitudeappearsto be weakerthan that found the indirect route may be relatively more important
in past tests of the model. For example, of the nine than previous researchhas suggestedfor two reasons.
studiesin our literaturereviewthat tested the dual me- First, the meta-analysissuggestsa substantialand sig-
diation model, only two had a parameterestimate less nificantindirectpathfromad attitudesto brandattitude
than .70, and only one had a parameterestimateequiv- via brandcognitions. Previouspath-analyticstudies of
alent to the .57 in our model (Homer 1990, experiment ad attitudes have been divided on whetherbrand cog-
2; estimate= .55). Giventhatthesecoefficientsarebased nitions have a significanteffecton brandattitudes.The
on correlationscorrectedfor measurementerror, this path-analyticstudy by MacKenzieand Lutz ( 1989), for
result is surprisingbecause our path coefficientshould example, did not find a significant effect, whereas
be increasedby the correction.Thus, the meta-analysis Homer (1990) concludedthat a significantpath did ex-
appearsto suggesta relativelystrongerindirect versus ist. Our findings,which are based on aggregatedstudy
direct influence of ad attitudethan would be indicated effects, suggestthat brand cognitions do have a signif-
by previoustests of the model. icant effect on brandattitudes.This point is significant
in view of the fact that researchon ad attitudes grew
out of the Fishbein tradition of attitude research(e.g.,
DISCUSSION Mitchell and Olson 1981; Shimp 1981) and modified
The results of this meta-analysisprovide a number the Fishbeinian view that only brand beliefs affected
of insightsinto the effectsof ad attitudes.First, as with brand attitudes.
EFFECTS OF AD ATTITUDE 47

Indeed,the Fishbeinianview was even more extreme Moderatoranalysesindicatedthat usingmulti-versus


in suggestingthat only utilitarian beliefs affect brand single-itemscales to measuread attitudehad relatively
attitudes. Recently, some studies on Aad (e.g., Mac- minor effectson averagecorrelations,making a signif-
Kenzie and Lutz 1989) have gone so farin the opposite icant differenceonly in the case of the ad attitude and
direction as to suggestthat brand cognitions may not brandattituderelationship.The use of studentsamples,
have a significanteffect on brandattitudes.The results a pervasivepracticein researchon ad attitudes,tended
of our meta-analyticstudysuggestthat brandcognitions to produce larger effects than the use of nonstudent
do significantly,albeit modestly,affectbrandattitudes. samplesdid, and it significantlymoderatedrelationships
Theseresultsareconsistentwith Zannaand Rempel's between ad attitude and the antecedent constructs of
(1988; Zanna 1980) attitude model in suggestingthat feelingsand ad-relatedcognitions.These resultssuggest
brand attitudes can be based on a variety of sources that the possibility of upwardbias in effect sizes gen-
(e.g., brandcognitions, brand affect, ad cognitions, ad eratedfrom student samples should be recognizedand
affect, etc.). In a related study, Brecklerand Wiggins that caution should be used in attemptingto generalize
(1991) demonstratedthat cognitive responsesto a per- such resultsto other populations.
suasive messageare relatedto both affectiveand eval- The product-relatedmoderatorvariables(i.e., novel
uative dimensions of precommunicationattitude. The vs. familiarbrandand consumernondurablevs. "other"
meta-analysis supports the results of the study by product type) each affected differentad attitude rela-
Brecklerand Wiggins (1991) in suggestingthat both tionships. Studies using novel brands reportedsignifi-
affectiveand cognitive influences play a role in deter- cantly strongerrelationshipsbetween ad attitude and
mining brandattitudes.It appearsthat researchon de- the outcome constructsof brandattitudeand purchase
terminantsof brand attitudeshas shifted from one ex- intentions, suggestingthat the existence of priorbrand
treme to another but that a more moderate position, attitudes reduces the impact of ad attitudes on these
incorporatingboth affective and cognitive influences, outcomes. Producttype (i.e., consumernondurablevs.
is more accurate. "other")significantlyaffectedthe feelings and ad atti-
Second, and perhapsmore interesting,the direct in- tude and ad attitudeand brandcognitionsrelationships
fluence of ad attitude on brand attitude in our meta- (both relationships were stronger for products other
analysis was much weaker than that found in almost than consumernondurables).These resultsmay reflect
all previous path-analyticinvestigations.Inasmuch as greatersubject involvement with services and higher-
previous investigations of the dual mediation model priced durable products. The results also suggest that
have focused centrally on the ad attitude construct, it the types of productsrepresentedin stimulusads should
is possible that the strongerad attitude and brand at- be selectedcarefullybecausethey can significantlyaffect
titude effects reportedin those studies reflect a publi- study outcomes. By using ads for a mixture of product
cation bias against studies reporting smaller effects. types as stimuli, a number of studies in the research
Collectively,these findingssuggestthat caution is war- stream (e.g., Holbrook and Batra 1987; Olney et al.
ranted not to overemphasizethe direct effect of ad at- 1991; Stayman and Aaker 1988) have generalizedthe
titudes on brandattitudes, nor to underemphasizethe effects of ad attitudes across product types, and this
significantrole played by an indirect effect via brand practiceis likelyto be effectivein avoidingany potential
cognitions in the determinationof brand attitudes. biasing effects of product type.
Tests of effect homogeneity suggestedthat the effect Threecoded study characteristicsrelatedto subjects'
of ad-relatedcognitions on ad attitudeand the effectof processinggoalsduringexposureto ads (i.e., advertising
ad attitudeson brand-relatedcognitionswererelatively medium, whetherads wereimbeddedin othermaterial,
robust across study contexts and that overall homoge- and whethersubjectswere instructedto attend to ads)
neity of effects could be obtained by deleting one or affected ad attitude relationships. As expected, these
two outliers.The homogeneitytests, however,indicated variables had different effects on the feelings and ad
substantial variation in the relationshipsbetween ad attituderelationshipthan they had on the cognitiveand
attitudeand feelings,brandattitudes,and purchasein- behavioral outcomes of ad attitudes. Conditions that
tentions. These findingssuggestthat these latter effects favored greater cognitive elaboration (i.e., print ads,
were not robustacross study contexts. nonimbedded ads, and instructions to attend to ads)
Of 11 coded study characteristicsused as potential producedsmallereffectsof feelingson ad attitude.These
moderatorvariables,all but one had significantmod- same conditions, however, tended to yield higher cor-
eratingeffectson at least one pairwiserelationship.Of relations between ad attitudes and the downstream
the nine moderatorsthat were tested with respectto all constructs of brand cognitions, brand attitudes, and
five pairwise relationships,eight significantlyaffected purchase intentions. These results, generalizedacross
two differentrelationships.Thus, it appearsthat many the entireresearchstream,corroborateearlierempirical
methodologicaldecisions regardingstudy context have findingsby Madden et al. (1988) and strongly suggest
significanteffectson the resultsof studies involving ad that processinggoals that enhance cognitive responses
attitudes. may inhibit affective responses. Furtherresearchinto
48 JOURNAL OF CONSUMER RESEARCH

these phenomena is likely to produce useful insights remained after accounting for the joint effects of the
into advertisingeffectiveness. coded moderators,suggestingthat additional sources
Within-subjectsstudydesignshad the expectedeffect of variationexist besides those identifiedhere.
of producingsignificantlylargereffect sizes for the ad Collectively, these findings suggest that, although
attitude and brand attitude relationshipand substan- some ad attitude relationships are relatively robust
tially (thoughnot significantly)largereffectsizes for the across study contexts (e.g., ad- and brand-relatedcog-
ad attitudeand purchaseintentions relationship.These nitions), othersare more significantlyaffectedby meth-
results are consistent with those of Rao and Monroe odological choices. Useful insights are gained by ana-
(1989),who founda strongerpriceand perceivedquality lyzing the effects of these methodologicalchoices, and
relationshipfor studies using within-subjectsdesigns. they should be carefully considered in the process of
Study design, however, had the opposite effect on the designing (and interpreting)individual studies. Such
feelings and ad attitude relationship.Feelings (or ads basic methodologicalchoices as advertisingmedium or
elicitingdifferentfeelings)werethe within-subjectsfac- whetherto imbed ads in other materials,for example,
tor in virtuallyall studies included in the feelings and are seldom well justified with respect to the research
ad attitude analysis. In light of previous evidence that questions posed, but these meta-analyticfindingssug-
feelingresponsescan be quite variableacrossads (Hol- gest that such decisions may have importanteffectson
brook and Batra 1987;Olney et al. 1991), it is possible the resultsobtained.
that attemptingto generalizethe effects of ad attitude The partitioningof study effects by design features
across feelings and feeling executions mixes qualita- indicates the presenceof some systematicbiases in the
tively different effects and diminishes the overall methodological choices made by researchers.For ex-
strengthof the relationship.In futureresearch,it would ample, of studies relating feelings to ad attitudes, 14
be useful to carefullyidentify the nature and strength used television ads, whereas only one used print ads.
of effectsthat differentfeelingsand differentfeeling ex- Likewise,for all relationshipsinvestigated,the prepon-
ecutions have on ad attitudes and related advertising deranceof studiesemployedstudentsubjects.Although
outcomes. moderatoranalyses suggest that some of these design
The moderatoranalysessuggestthat the coded study features may not significantly influence effect sizes,
characteristicsinfluencedifferentportionsof the system others suggest that repeated study of a relationship
of relationships represented in the dual mediation within the same researchcontext may produce results
model. The novel versus familiar brand variable, for that do not generalizeto other contexts. For example,
example, primarilyaffectedthe downstreamrelation- use of student samples produced largeraveragecorre-
ships in the nomological network(i.e., ad attitude and lations in two of the five focal relationships,but only a
brandattitudeand ad attitudeand purchaseintentions
relationships). Study focus also affected these same small number of studies used nonstudent subjects in
downstreamrelationships,althoughit also had a strong both cases. A greaternumber of studies using nonstu-
effect on the feelings and ad attitude relationship. dent subjects would permit greaterconfidence in the
In contrast,for example,producttype and advertising validityof conclusionsregardingthe magnitudeof effect
medium both significantlyinfluenced the ad attitude sizes.
and brand cognitions relationship. It is possible that For similarreasons,studiesusing producttypes other
the strongereffect for products other than consumer than consumer nondurablesto assess the effects of ad
nondurables reflects higher involvement levels and attitudes on ad- and brand-relatedcognitions, brand
greatercognitiveelaborationof advertisedmaterialsfor attitudes,and purchaseintentions could provideuseful
more involving products. It is also possible that the insights regardingthe effects of prior familiarity and
strongereffect for print ads reflectsgreateropportunity knowledgeon these relationships.They could also pro-
to cognitively elaborate printed information versus vide greater confidence in conclusions regardingthe
televised information(Chaikenand Eagly 1983). generalizabilityof these relationships.Suchbroadening
The ability of moderatorvariablesto accountjointly of the contexts in which ad attitudesare studiedshould
for variancein study effectsdifferedacrossthe relation- increaseconfidencein the effectsizes found and provide
ships investigated. After deletion of only one or two avenues for assessmentof new moderatorvariables.
study effects, there was no significantvariance in the Finally, in integratingresearchfindings on Aad, we
relationshipsbetween ad attitude and ad- and brand- have been guided by Cooper's(1989) discussion of va-
relatedcognitions left to be explained.For the feelings lidity issues. Cooperhighlightsthe importanceof care-
and ad attitude and ad attitude and brand cognitions fully definingthe scope and objectivesof the integrative
relationships,the coded moderatorvariableseffectively review,thoroughlysearchingfor relevantstudies, com-
accounted for the preponderanceof variancein study plete reporting of coding and analytical procedures,
effects,suggestingthat there was no need to specify ad- rigorousand systematicanalysis, and carefulinterpre-
ditional potentialmoderators.For the relationshipsbe- tation of results. Like any meta-analysis,the findings
tween ad attitude and brand attitude and purchase and conclusions of our study should be evaluated in
intentions, however, significant residual variance terms of these criteriafor safeguardingvalidity.
EFFECTS OF AD ATTITUDE 49

The scope of this reviewassessinga numberof issues Bangert-Downs, Robert L. (1986), "Review of Developments
relatedto the effectsof ad attitudeswas broad. A thor- in Meta-analytic Method," Psychological Bulletin, 99 (3),
ough search for relevant studies produced 43 articles 388-399.
forwhich usabledatacould be obtained.However,data Batra, Rajeev and Michael L. Ray (1986), "Affective Re-
could not be obtained for 14 additional studies that sponses Mediating Acceptance of Advertising," Journal
were identified as relevant to the topic. Although no of Consumer Research, 13 (September), 234-249.
Belch, George and'Michael A. Belch (1984), "An Investigation
differencesbetween those studies for which data were of the Effects of Repetition on Cognitive and Affective
or were not available seemed apparent, it is possible Reactions to Humorous and Serious Television Com-
that some differencesmight have resultedif it had been mercials," in Advances in Consumer Research, Vol. 11,
possibleto include all studies.Thereappearsto be little ed. Thomas C. Kinnear, Ann Arbor, MI: Association for
if any basis, however,for positing the existence of sys- Consumer Research, 4-10.
tematic differencesin effects for the relatively many Bettman, James R. (1986), "Consumer Psychology," in An-
studies included in the analysis and the relativelyfew nual Review of Psychology, Vol. 37, ed. Mark R. Rosen-
for which effectswere not available. zweig and Lyman W. Porter, Palo Alto, CA: Annual Re-
In addition, as noted above, some cells in the mod- views, 257-290.
eratoranalyseshad a verysmall numberof studyeffects Breckler, Steven J. and Elizabeth C. Wiggins (1991), "Cog-
(as few as a singleobservation).Thus, particularcaution nitive Response in Persuasion: Affective and Evaluative
must be used in interpretingthese results. Finally, as Determinants," Journal of Experimental Social Psy-
with all meta-analyses,the results are constrained by chology, 27 (March), 180-200.
the methods used in the studies. Althoughresearchon Brinberg, David and James Jaccard (1986), "Meta-analysis:
Techniques for the Quantitative Integration of Research
ad attitudeshas been conductedin a numberof settings Findings,"in Perspectiveson Methodologyin Consumer
and no one data set was used twice, power to assess Research, ed. David Brinberg and Richard J. Lutz, New
potential moderatingeffectswas underminedby small York: Springer, 155-180.
subgroupsample sizes. These problems,along with in- Burke, Marian Chapman and Julie A. Edell (1989), "The Im-
terpretationaldifficultiesin the coding of studies, pre- pact of Feelings on Ad-based Affect and Cognition,"
cluded specific investigation of such potentially inter- Journal of Marketing Research, 26 (February), 69-83.
esting moderator variables as involvement (although Calder, Bobby J., Lynn W. Phillips, and Alice M. Tybout
the effects of processing-goalvariablesare likely to be (1981), "Designing Research for Application," Journal
partiallydue to task involvement). of Consumer Research, 8 (September), 197-207.
In summary,effortsto understandthe effects of ad Chaiken, Shelly and Alice H. Eagly (1983), "Communication
attitudes appear to represent an important ongoing Modality as a Determinant of Persuasion: The Role of
stream of research on advertising effects. Our meta- Communicator Salience," Journal of Personality and
analysisof findingsrelatedto ad attitude suggeststhat Social Psychology, 45 (20), 241-256.
the relationshipsbetweenad attitudeand othervariables Chattopadhyay, Amitava and Kunal Basu (1990), "Humor
in Advertising: The Moderating Role of Prior Brand
are strongbut vary accordingto a number of method- Evaluation,"Journal of MarketingResearch, 27 (No-
ologicalfactors.In addition,the influenceof ad attitude vember), 466-476.
appearsto occur via paths specifiedin the dual media- Cohen, Joel B. and Dipankar Chakravarti (1990), "Consumer
tion model, with a relativelystrongerindirect path via Psychology," in Annual Review of Psychology, Vol. 41,
brand-relatedcognitionsthan has been reportedin pre- ed. Mark R. Rosenzweig and Lyman W. Porter, Palo
vious empiricalresearch.However,futureresearchthat Alto, CA: Annual Reviews, 243-288.
identifiesadditionalsources of variationappearsto be Cooper, Harris M. (1989), Integrating Research: A Guide for
necessary to completely account for variation in the Literature Reviews, Newbury Park, CA: Sage.
relationshipsinvolving ad attitudes. Cox, Dena S. and Anthony D. Cox (1988), "What Does Fa-
miliarity Breed? Complexity as a Moderator of Repetition
Effects in Advertisement Evaluation," Journal of Con-
[ReceivedMarch 1991. Revised November1991.] sumer Research, 15 (June), 111- 116.
Droge, Cornelia (1989), "Shaping the Route to Attitude
Change: Central versus Peripheral Processing through
REFERENCES Comparative versus Noncomparative Advertising,"
Aaker, David A., Douglas M. Stayman, and Michael R. Hag- Journalof MarketingResearch,26 (May), 193-204.
erty (1986), "Warmth in Advertising: Measurement, Im- Duncan, Calvin P. and James E. Nelson (1985), "Effects of
pact, and Sequence Effects," Journal of Consumer Re- Humor in a Radio Advertising Experiment," Journal of
search, 12 (March), 365-381. Advertising, 14 (2), 33-40.
Alpert, Mark I., Linda L. Golden, and Wayne D. Hoyer Edell, Julie A. and Marian Chapman Burke (1987), "The
(1982), "The Impact of Repetition on Advertisement Power of Feelings in Understanding Advertising Effects,"
Miscomprehension and Effectiveness," in Advances in Journal of Consumer Research, 14 (December), 421-433.
Consumer Research, Vol. 10, ed. Richard P. Bagozzi and Farley, John U. and Donald Lehmann (1986), Meta-analysis
Alice M. Tybout, Ann Arbor, MI: Association for Con- in Marketing:Generalizationof ResponseModels, Lex-
sumer Research, 130-135. ington, MA: Lexington.
50 JOURNAL OF CONSUMER RESEARCH

Gardner, Meryl (1985), "Does Attitude toward the Ad Affect Joreskog, Karl G. and Dag Sorbom (1986), LISREL: Analysis
Brand Attitude under a Brand Evaluation Set?" Journal of Linear StructuralRelationships by the Method of
of Marketing Research, 22 (May), 281-300. Maximum Likelihood, 4th ed., Mooresville, IN: Scientific
Goldberg, Marvin E. and Gerald J. Gorn (1987), "Happy and Software.
Sad TV Programs: How They Affect Reactions to Com- Kamins, Michael A. (1990), "An Investigation into the
mercials," Journal of Consumer Research, 14 (Decem- 'Match-Up' Hypothesis in Celebrity Advertising: When
ber), 387-403. Beauty Is Only Skin Deep," Journal of Advertising, 19
Golden, Linda L. and Keren A. Johnson (1984), "The Impact (1), 4-13.
of Sensory Preference and Thinking versus Feeling Ap- , Meribeth J. Brand, Stuart A. Hoeke, and John C.
peals on Advertising Effectiveness," in Advances in Con- Moe (1989), "Two-sided versus One-sided Celebrity En-
sumer Research, Vol. 10, ed. Richard P. Bagozzi and dorsements: The Impact on Advertising Effectiveness and
Alice M. Tybout, Ann Arbor, MI: Association for Con- Credibility," Journal of Advertising, 18 (2), 4-10.
sumer Research, 203-208. Keller, Kevin Lane (1987), "Memory Factors in Advertising:
Greenwald, Anthony G. and Clark Leavitt (1984), "Audience The Effect of Advertising Retrieval Cues on Brand Eval-
Involvement in Advertising: Four Levels," Journal of uations," Journal of Consumer Research, 14 (December),
Consumer Research, 11 (June), 581-592. 316-333.
Gresham, Larry G. and Terence A. Shimp (1985), "Attitude (1991 a), "Cue Compatibility and Framing in Adver-
toward the Advertisement and Brand Attitudes: A Clas- tising," Journal of Marketing Research, 28 (February),
sical Conditioning Approach," Journal of Advertising, 42-57.
14 (1), 10-17. (1991b), "Memory and Evaluative Effects in Com-
Haley, Russel I. and Allan L. Baldinger (1991), "The ARF petitive Advertising Environments," Journal of Con-
Copy Research Validity Project," Journal ofAdvertising sumer Research, 17 (March), 463-476.
Research, 31 (April/May), 11-32. Kilbourne, William E. (1986), "An Exploratory Study of the
Harris, William and David J. Moore (1989), "Affect Intensity Effect of Sex Role Stereotyping on Attitudes toward
as an Individual Difference Variable in Consumer Re- Magazine Advertisements," Journal of the Academy of
sponse to Advertising Appeals," paper presented at the Marketing Science, 14 (Winter), 43-46.
Association for Consumer Research, New Orleans. Krugman, Herbert (1965), "The Impact of Television Ad-
Hedges, Lawrence V. and Ingram Olkin (1985), Statistical vertising: Learning without Involvement," Public Opin-
Methods for Meta-analysis, Orlando, FL: Academic ion Quarterly,29 (Fall), 349-356.
Press. Laczniak, Russell N. and Les Carlson (1989), "Examining
Hill, Ronald Paul (1988), "An Exploration of the Relationship the Influence of Attitude toward the Ad on Brand Atti-
between AIDS-related Anxiety and the Evaluation of tudes," Journal of Business Research, 19 (December),
Condom Advertisements," Journal ofAdvertising, 17 (4), 303-311.
35-42. Leigh, Thomas W., Arno J. Rethans, and Tamatha Reichen-
(1989), "An Exploration of Voter Responses to Po- bach Whitney (1987), "Role Portrayals of Women in
litical Advertising," Journal of Advertising, 18 (4), 14- Advertising,"Journal of AdvertisingResearch, 27 (Oc-
22. tober/November), 54-63.
Holbrook, Morris B. (1978), "Beyond Attitude Structure: Liu, Scott S. and Patricia A. Stout (1987), "Effects of Message
Toward the Information Determinants of Attitude," Modality and Appeal on Advertising Acceptance," Psy-
Journal of Marketing Research, 15 (November), 545- chology and Marketing, 4 (Fall), 167-188.
556. Lutz, Richard J., Scott B. MacKenzie, and George E. Belch
and Rajeev Batra (1987), "Assessing the Role of (1983), "Attitude toward the Ad as a Mediator of Ad-
Emotions as Mediators of Consumer Responses to Ad- vertising Effectiveness: Determinants and Conse-
vertising," Journal of Consumer Research, 14 (Decem- quences," in Advances in Consumer Research, Vol. 10,
ber), 404-420. ed. Richard P. Bagozzi and Alice M. Tybout, Ann Arbor,
Homer, Pamela M. (1990), "The Mediating Role of Attitude MI: Association for Consumer Research, 532-539.
toward the Ad: Some Additional Evidence," Journal of Lynch, John G., Jr. (1982), "On the External Validity of Ex-
Marketing Research, 27 (February), 78-86. periments in Consumer Research," Journal of Consumer
Houston, Michael J., J. Paul Peter, and Alan G. Sawyer Research, 9 (December), 225-239.
(1983), "The Role of Meta-analysis in Consumer Re- Machleit, Karen A. and R. Dale Wilson (1988), "Emotional
search," in Advances in Consumer Research, Vol. 10, ed. Feelings and Attitude toward the Advertisement: The
Richard P. Bagozzi and Alice M. Tybout, Ann Arbor, Roles of Brand Familiarity and Repetition," Journal of
MI: Association for Consumer Research, 503-508. Advertising,17 (3), 27-35.
Hoyer, Wayne D. (1984), "An Examination of Consumer Maclnnis, Deborah J. and C. Whan Park (1991), "The Dif-
Decision Making for a Common Repeat Purchase Prod- ferential Role of Characteristics of Music in High- and
uct," Journal of Consumer Research, 11 (December), Low-Involvement Consumers' Processing of Ads," Jour-
822-829. nal of Consumer Research, 18 (September), 16 1-173.
Hunter, John E. and Frank L. Schmidt (1990), Methods of MacKenzie, Scott B. and Richard J. Lutz (1989), "An Em-
Meta-analysis: Correcting Error and Bias in Research pirical Examination of the Structural Antecedents of At-
Findings, Newbury Park, CA: Sage. titude toward the Ad in an Advertising Pretesting Con-
Johnson, Blair T. and Alice H. Eagly (1989), "Effects of In- text," Journal of Marketing, 53 (April), 48-65.
volvement on Persuasion: A Meta-analysis," Psycholog- , Richard J. Lutz, and George E. Belch (1986), "The
ical Bulletin, 106 (2), 290-314. Role of Attitude toward the Ad as a Mediator of Adver-
EFFECTS OF AD ATTITUDE 51

tising Effectiveness: A Test of Competing Explanations," Background Music on Brand Attitude Formation,"
Journal of Marketing Research, 23 (May), 130-143. Journal of Marketing Research, 23 (February), 11-24.
Madden, Thomas J., Chris T. Allen, and Jacquelyn L. Twible Peter, J. Paul and Gilbert A. Churchill, Jr. (1986), "Rela-
(1988), "Attitude toward the Ad: An Assessment of Di- tionships among Research Design Choices and Psycho-
verse Measurement Indices under Different Processing metric Properties of Rating Scales: A Meta-analysis,"
'Sets,'" Journal of Marketing Research, 25 (August), Journalof MarketingResearch,23 (February),1-10.
242-252. Rao, Akshay R. and Kent B. Monroe (1989), "The Effect of
Miniard, Paul W., Sunil Bhatla, and Randall L. Rose (1990), Price, Brand Name, and Store Name on Buyers' Percep-
"On the Formation and Relationship of Ad and Brand tions of Product Quality: An Integrative Review," Journal
Attitudes: An Experimental and Causal Analysis," Jour- of Marketing Research, 26 (August), 351-357.
nal of Marketing Research, 27 (August), 290-303. Reddy, Srinivas K. and Priscilla A. LaBarbera (1985), "Hi-
Mitchell, Andrew A. (1986), "The Effect of Verbal and Visual erarchical Models of Attitude," Multivariate Behavioral
Research, 20 (October), 451-471.
Components of Advertisements on Brand Attitudes and
Rosenthal, Robert (1984), Meta-analytic Proceduresfor Social
Attitude toward the Advertisement," Journal of Con-
Research, Beverly Hills, CA: Sage.
sumer Research, 13 (June), 12-24. Russo, J. Edward and Debra L. Stephens (1990), "Ad-specific
and Jerry C. Olson (1981), "Are Product Attribute Emotional Responses to Advertising," in Emotion in
Beliefs the Only Mediator of Advertising Effects on Brand Advertising:Theoreticaland PracticalExplorations,ed.
Attitudes?" Journal ofMarketing Research, 18 (August), Stuart Agres et al., Westport, CT: Quorom, 113-124.
318-322. Shimp, Terence A. (1981), "Attitude toward the Ad as a Me-
Monroe, Kent B. and R. Krishnan (1983), "A Procedure for diator of Consumer Brand Choice," Journal ofAdvertis-
Integrating the Outcomes across Studies," in Advances ing, 10 (2), 9-15.
in Consumer Research, Vol. 10, ed. Richard P. Bagozzi Stayman, Douglas M. and David A. Aaker (1988), "Are All
and Alice M. Tybout, Ann Arbor, MI: Association for Effects of Ad-induced Feelings Mediated by Aad?"Journal
Consumer Research, 503-508. of Consumer Research, 15 (December), 368-373.
Moore, Danny L. and J. Wesley Hutchinson (1983), "The Werts, C. E., D. A. Rock, R. L. Linn, and K. G. Joreskog
Effects of Ad Affect on Advertising Effectiveness," in Ad- (1976), "Comparison of Correlations, Variances, Co-
vances in Consumer Research, Vol. 10, ed. Richard P. variances, and Regression Weights with or without Mea-
Bagozzi and Alice M. Tybout, Ann Arbor, MI: Associ- surement Error," Psychological Bulletin, 83 (6), 1007-
ation for Consumer Research, 526-531. 1013.
Muehling, Darrel D. (1987), "Comparative Advertising: The Winer, B. J. (1971), Statistical Principles in Experimental
Influence of Attitude toward the Ad on Brand Evalua- Design, New York: McGraw-Hill.
tion," Journal ofAdvertising, 16 (4), 43-49. Yi, Youjae (1 990a), "Cognitive and Affective Priming Effects
and Russell N. Laczniak (1988), "Advertising's Im- of the Context for Print Advertisements," Journal ofAd-
mediate and Delayed Influence on Brand Attitudes: vertising, 19 (2), 40-48.
Considerations across Message-Involvement Levels," (1 990b), "The Effects of Contextual Priming in Print
Journal ofAdvertising, 17 (4), 23-34. Advertisements,"Journal of Consumer Research, 17
(September), 215-222.
Olney, Thomas J., Morris B. Holbrook, and Rajeev Batra
Zanna, Mark P. (1990), "Attitude Function: Is It Related to
(1991), "Consumer Responses to Advertising: The Effects
Attitude Structure?" in Advances in Consumer Research,
of Ad Content, Emotions, and Attitude toward the Ad Vol. 12, ed. Marvin Goldberg et al., Ann Arbor, MI: As-
on Viewing Time," Journal of Consumer Research, 17 sociation for Consumer Research, 98-100.
(March), 440-453. and J. K. Rempel (1988), "Attitudes: A New Look at
Park, C. Whan and S. Mark Young (1983), "Types and Levels an Old Concept,"in TheSocial Psychologyof Knowledge,
of Involvement and Brand Attitude Formation," in Ad- ed. D. Bar Tal and Arie Kruglanski, New York: Cam-
vances in Consumer Research, Vol. 10, ed. Richard P. bridge University Press, 315-334.
Bagozzi and Alice M. Tybout, Ann Arbor, MI: Associ- Zinkhan, George M., William R. Locander, and James H.
ation for Consumer Research, 320-324. Leigh (1986), "Dimensional Relationships of Aided Re-
and S. Mark Young (1986), "Consumer Response to call and Recognition," Journal ofAdvertising, 15 (1), 38-
Television Commercials: The Impact of Involvement and 46.

Das könnte Ihnen auch gefallen