Professional Documents
Culture Documents
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.
American Marketing Association is collaborating with JSTOR to digitize, preserve and extend access to
Journal of Marketing Research.
http://www.jstor.org
Traditional models of consumer behavior in the mar- glected (for noteworthy exceptions, see Holbrook et al.
keting literature implicitly assume that consumer deci- 1984; Westbrook 1980). This neglect is surprising for
sion making is principally the result of cognitive pro- several reasons. First, the postpurchaseperiod comprises
cesses involving the semantic meaning of product product ownership and usage, both of which might be
attributes(e.g., Bettman 1979; Howard and Sheth 1969). expected to provide opportunities for varied affective re-
Affective processes, broadly described as those involv- sponse of considerable personal significance to con-
ing subjective feelings, generally are relegated to a sec- sumers (Holbrook and Hirschman 1982). Second, affec-
ondary role (Holbrook and Hirschman 1982; Peterson, tive variablesmight be expected to contributesubstantially
Hoyer, and Wilson 1986). However, recent advances in to the explanation and prediction of postpurchase be-
social cognition, cognitive psychology, and social psy- havior, which in addition to product usage includes sat-
chology suggest that affective processes may constitute isfactionappraisal,seller-directedcomplaintactions, word-
not only a powerful source of human motivation, but of-mouth transmission, disposition behavior, and repur-
also a major influence on information processing and chase planning. Typically, accounts of these processes
choice (Hoffman 1986; Isen 1984; Zajonc 1980). As a have been limited to cognitive/semantic belief variables
result, marketers are increasingly interested in under- (e.g., Beardenand Teel 1983; Oliver 1980; Richins 1983).
standing the nature of affect and its contribution to con- The inclusion of product/consumption affective re-
sumer decision making and response to marketing vari- sponses therefore might be expected to extend the ex-
ables (Aaker, Stayman, and Hagerty 1985; Batra and Ray planatory ability of current theoretical postpurchase
1986; Donovan and Rossiter 1982; Gardner 1985; Wein- models.
berg and Gottwald 1982). Accordingly, the role of product/consumption-based
Despite a growing number of conceptual and empiri- affective responses is examined in three central forms of
cal studies of consumer affect, most attention to date has postpurchase behavior: satisfaction appraisal, seller-di-
been on its role in consumer prepurchaseprocesses. Af- rected complaint actions, and word-of-mouth transmis-
fect in postpurchase processes has been relatively ne- sions. Satisfaction appraisal usually is regarded as the
central mediator of postpurchase behavior, linking pre-
choice product beliefs to postchoice cognitive structure,
*Robert A. Westbrook is Professor of Marketing, College of Busi- consumer communications, and repurchase behavior
ness and Public Administration, The University of Arizona.
The author gratefully acknowledges the many constructive criti-
(Beardenand Teel 1983; LaBarberaand Mazursky 1983;
cisms provided by three anonymous JMR reviewers.
Oliver 1980). Complaint behavior is of interest as an
important form of market feedback to organizations
258
Table 2
STATISTICSFOR AFFECTMEASURESAND REGRESSIONMODELVARIABLE$
DESCRIPTIVE
Automobiles CATV
DES-II (N = 200) (N = 151) DES-IIsubscale correlations
measures M S M S 1 2 3 4 5 6
1. Joy 3.43 .91 2.78 .83 (.79, .69) .66 -.08 .10 .04 .60
2. Interest 3.25 .91 2.85 .88 .76 (.78, .74) -.02 .07 .09 .58
3. Anger 1.48 .77 2.00 .91 -.43 -.22 (.89, .80) .62 .66 .22
4. Disgust 1.45 .70 2.14 .86 -.28 -.27 .66 (.89, .76) .57 .30
5. Contempt 1.33 .69 1.56 .78 -.25 -.15 .67 .82 (.76, .86) .30
6. Surprise 2.30 1.07 2.28 .84 .38 .42 -.07 .14 .15 (.76, .77)
Regression
Regeson
model Regressionmodelvariablecorrelations
variables' M S M S 1 2 3 4 5 6 7 8 9
1. POS AFF 3.51 .86 3.12 .88 .02 .01 .40 .36 -.04 .29
2. NEG AFF 2.02 .91 2.34 .88 -.37 -.05 -.24 -.32 .32 .32
3. EXP BEN 8.09 1.62 .32 -.21
4. EXP PRB 4.25 2.70 -.23 .09 -.18
5. EXP TOT 11.14 2.59 -.23 .02 -.01 .02
6. DISCONF 4.59 1.13 5.51 1.84 .47 -.30 .22 -.18 .53 -.28 -.11
7. SAT 7.18 1.33 5.04 1.16 .54 .45 .46 -.32 .61 -.47 -.20
8. COMPLN 1.85 2.47 .31
9. WOM 6.60 4.43
Note:Entriesin parenthesesareCronbachalphaestimatesof DES-IIreliabilityfor automobilesandCATV, respectively.Valuesbelow diagonal
areproduct-moment correlationsfor automobiles;those above arefor CATV. Valuesgreaterthan .12 are statisticallyat the .05 level andbeyond
for autos;for CATVthe correspondingvalue is .14. Blanksindicatedatanot availablebecauseof study design.
'See Table 3 for constructs.
Table 3
MEASUREMENT
OF POSTPURCHASEVARIABLES
ministered during a single interview, a check was made isfaction measures. The multiple correlations between
on the seriousness of potential common methods vari- satisfaction and the two sets of affect measures do not
ance. In the CATV pilot study the affective measures differ significantly (F73,73= 1.072; p > .05), indicating
were assessed on two separate occasions, the first time that the inflation in relationships due to this source of
alone and the second time in conjunction with the sat- methods variance is minimal.
Finally, a test for order effects in the sequence of mea- 8.30, 7 d.f., p = .305; GFI = .981. Though in the case
sure administrationwas made among subsamples (N = of automobiles the chi squaretest suggests the two-factor
20) of respondents in both final field studies, who were model leaves unexplaineda statisticallysignificant amount
given the DES-II subscales prior to other measures in of the covariationamong the DES-II subscales, the model
lieu of the opposite ordering as used among the rest of GFI of .960 indicates that relatively little variation of
the respondents. No significant differences were found practical import remains to be explained. Therefore, the
between the two order conditions in affect, satisfaction, two-factor model can be considered a plausible repre-
complaining, or word-of-mouth measures (all F-statis- sentation of the interrelations among the DES-II affect
tics p > .10). measures in the product/consumption contexts studied.
Parameterestimates for the two-factor model are re-
RESULTS ported in Table 4, as well as estimates of construct re-
liability and shared indicator variance. All factor load-
DES-II Subscales ings are statistically significant and in the expected
Table 2 indicates the relative frequency with which direction in both product categories. The positive and
each type of postpurchase affect was reported for auto- negative affect factors are uncorrelated for CATV, but
mobiles and CATV. Joy and interest are more frequent inversely related for automobiles (p < .001). Thus H1
for automobiles than for CATV (t = 6.12 and 2.81, can be accepted for both automobiles and CATV, though
p < .01), whereas anger and disgust are more frequent the independence of the factors appears to vary by prod-
for CATV than for autos (t = 4.83 and 6.78, p < .01). uct category.
Despite these differences, the relationships between the
six classes of affect appear broadly similar within the Relationships to Postpurchase Behavior
two product categories. An exception is surprise, which Hypothesized relationships between affective response
correlatesmore highly with the affect measuresfor CATV dimensions and satisfaction appraisal, complaint behav-
than for automobiles. ior, and word-of-mouth activity were evaluated by hi-
erarchicalOLS regressions. Factor scores (computed by
Factor Structure of Affective Response the regression method) from the two-factor confirmatory
To examine the validity of the hypothesized dimen- models served as indicators of positive and negative af-
sional structure of product- and consumption-based af-
fective response, confirmatory factor analysis was per-
formed on the six DES-II subscales. Tests were made of Table 4
both one- and two-factor models corresponding to the DES-IISUBSCALE
FACTOR TWO-FACTOR
LOADINGS:
unidimensional and bidimensional conceptualizations of MODELSa
affective response. Because the covariance structures of
the affect measures varied significantly by product cat- Automobiles CATV
egory (X2 = 124.59, 21 d.f., p < .000), separate con- DES-II Positive Negative Positive Negative
firmatory analyses were conducted. subscale affect affect affct affect
aect
The single-factor model of affective response does not Joy .888 .000 .710 .000
have a suitable fit to the original covariance matrix in t 13.483b 10.758b
each product category (for automobiles, x2 = 315.71, 9 Intearest .761 .000 .700 .000
t 11.377b 10.040b
d.f., p < .000; for CATV, X2 = 182.74, 9 d.f., p < Antger .000 .622 .000 .774
.000). In both instances the Joreskog-Sorbom (1986) t 11.629b 11.003b
goodness-of-fit index indicates that substantial unex- Dis gust .000 .578 .000 .608
plained variation remains. The null hypothesis of a un- t 13.579b 9.243b
idimensional model of affective response structurethere- Contempt .000 .614 .000 .569
t 15.102b 10.137"
fore is rejected. Surprise .573 .313 .592 .260
Next, a simple two-factor model was tested in which t 6.614b 3.859" 9.213b 4.432b
joy and interestloaded on a positive affect factor whereas
Factor
anger, disgust, and contempt loaded on an independent correlation
(possibly correlated) negative affect factor. Surprise was -.379 .002
assumed to load on both factors because of its theoretical t 4.952b .016
properties. To identify the model, the variances of the
common factors were fixed to 1.0. This model achieved Construct
a large and significant improvement in goodness of fit reliability .784 .769 .725 .762
Shared indicator
over the one-factor model (X2iff = 295.61, 2 d.f., p < variance .555 .468 .467 .472
.000 for automobiles; X2ff = 174.40, 2 d.f., p < .001
'All parameters based on LISREL VI maximum likelihood esti-
for CATV). For automobiles, test statistics on the two- mation (Joreskog and Sorbom 1986); only standardized values are
factor model are X2 = 20.10, 7 d.f., p = .005; GFI = shown; zero values fixed.
.960. For CATV, the corresponding statistics are X2 = bp < .001.
fective response. Table 5 gives the estimated regression tion-based affect is linked to all three forms of postpur-
coefficients and Table 2 indicates the simple zero-order chase behavior as hypothesized: positive affect relates
correlations among the regression model variables. directly to satisfaction appraisal (H2) and WOM activity
Note first the effects of affective responses alone. Ta- (H8) but not to complaint behavior (H6), whereas nega-
ble 5 indicates that the multiple correlations in equations tive affect relates inversely to satisfaction (H3) and di-
la, lb, 4, and 10 are all statistically significant and at rectly to complaint behavior (H5) and WOM (H8).
least moderate in size. The magnitude of the multiple To what extent do positive and negative dimensions
R's is greatest for the prediction of satisfaction and least of affective response add to the explanatory ability of
for complaint behavior. Moreover, the multiple R's also cognitive satisfaction determinants? Comparing equa-
exceed the largest of the correlations between individual tions 2a, 2b, 3a, and 3b in Table 5, one sees that the
affect factor scores and each criterion. Both positive and affective variablesadd significantlyto the predictivepower
negative affect regression coefficients in satisfaction of the cognitive expectation and disconfirmation belief
equations la and lb and WOM equation 10 are statis- variables alone (F2,194 = 16.855, p < .001 for auto-
tically significant (p < .001). In complaint behavior mobiles; F2,46 = 8.922, p < .001 for CATV). The af-
equation 4, only the negative affect factor has a signif- fective variables alone (equations la and lb) explain al-
icant coefficient, as expected. Thus, product/consump- most as much variance in satisfaction judgments as do
Table 5
OLSREGRESSIONS
OF SATISFACTION, COMPLAINT BEHAVIOR, AND WORD-OF-MOUTHTRANSMISSION
(beta with
coefficients significancelevels in parentheses)
the cognitive/semantic belief variables(equations2a and littleto the explanationof WOM;neitherof the multiple
2b). R values is significant(p > .05). By the ANOCOVA
The full regressionmodelof equations3a and 3b also approach,neithersatisfactionalone (equation 13) nor
affordsa morerigoroustest of the affect-satisfaction hy- satisfactionin conjunctionwith its cognitive belief an-
pothesesH2, H3, and H4by holdingconstantthe effects tecedents(equation14) eliminatesthe significanteffects
of the cognitive satisfactionantecedents.Both positive of the affectiveresponsevariables.Likewise, the intro-
andnegativeaffectdimensionsmaintaintheirsignificant ductionof the complaintbehaviorvariable,whetherwith
positive and negative coefficients, respectively,of ap- satisfactionalone (equation15) or togetherwith its cog-
proximately equalmagnitude.Disconllinationbeliefsalso nitive antecedents(equation16), has no impacton the
continueto relatedirectlyto satisfaction(p < .001). Ex- significance of the affectivevariablesin explainingWOM.
pectationbeliefs are relatedsignificantlyto satisfaction Thusthe affect-WOMlinkagescan be considereddirect
in the case of automobiles(expectedbenefits directly, andunmediatedby the otherpostpurchasevariables.
expectedproblemsinversely).ForCATV, the single ex- Finally,the regressionmodelof equations14, 15, and
pectationbelief measurefails to attainconventionalsta- 16 indicatesthat aftercontrollingfor the effects of af-
tisticalsignificance,despitethe correctsign. This find- fective responses,satisfactionhas a modest inverse re-
ing may reflectthe marginalreliabilityof the measure, lationshipand complaintbehaviora modestpositive re-
as noted previously.Overall, these findings strengthen lationshipto extent of WOM transmissions.Hence the
supportfor H2 and H3 and affirmthat the influenceof relationshipsbetween satisfactionand WOM and be-
affectivevariableson satisfactionis unmediated(H4). tween complaintbehaviorand WOM do not appearto
As the hypothesizedrelationshipof negativeaffect to be the spuriousresultof theirjoint dependenceon post-
complaintbehavior(H5)is supported,the next issue is purchaseaffectiveresponses.
whetherthis linkageis director wholly mediatedvia sat-
isfactionjudgments.To test the possible mediatingrole DISCUSSION
of satisfaction,the analysisof covariance(ANOCOVA)
approachwas employed (Batraand Ray 1986; Insko, Themajorresultsof thisresearchon product-andcon-
Turnbull,and Yandell 1974). This approachcompares sumption-based affectin two separateproductcategories
directand mediatedcausal models by enteringthe hy- arethreefold.First,good andbad feelings representrel-
pothesizedmediatorinto the regressionequationas a ativelyindependentdimensionsof affective responseto
covariate.When satisfactionis introducedin this way, productsin use. Second, reportsof such feelings relate
the regressioncoefficientof negativeaffectdoes not drop directly to product satisfactionjudgments, extent of
to nonsignificance,indicatingthat satisfactiondoes not complaint behavior, and volume of word-of-mouth
completelymediatethe affect-complainingrelationship. transmissionin the postpurchaseperiod.Third,affective
Theseresultsare shownin Table 5, equation5. Hence, responsesappearto accountfor significantincremental
negativeaffectappearsto influencecomplainingactivity variancebeyondextanttheoreticalexplanationsof these
directly,independently of its indirecteffects throughsat- postpurchase phenomena.
isfaction,supportingH7.
As a furtherexplorationof the role of satisfactionin Dimensionality of Affective Response
mediatingcomplaintbehavior, equations6 and 7 ex- Especiallynoteworthyis the relativeindependenceof
aminethe relationship betweencognitiveexpectationand positiveandnegativeaffectiveresponsein postpurchase
disconfirmationbeliefs and complaintbehavior.Equa- processes.The two dimensionsof affect are completely
tion 6 reveals that disconfirmationbeliefs (but not ex- orthogonalin the case of CATVandonly modestlyneg-
pectationbeliefs) are negatively (p < .01) related to atively relatedfor automobiles.Evidentlypleasant af-
complaining.However, its regressioncoefficient drops fects in relationto product/consumption outcomes,such
to nonsignificancewhen satisfactionis enteredas a cov- asjoy, interest,andexcitement,do not necessarilyimply
ariatein equation7. Thussatisfactionfully mediatesthe the absenceof unpleasantaffectssuch as anger,disgust,
effectsof disconfirmation beliefs on complaintbehavior. andcontempt.The resultsconfirmthe findingsof Abel-
This mediationalrole is uniqueto satisfactionand is not son et al. (1982) for politicalcandidatesandquestionthe
sharedwith the affectivevariables.When the latterare assumptionof a single bipolar affective dimensionof
enteredinto the regressionof complainingon expecta- good-bad(e.g. Davitz 1969; Russell 1983). Dual uni-
tion and disconfirmation beliefs (equation8), disconfir- polaraffectivedimensionsareparticularlyappropriate in
mationretainsits significant(p < .01) effect. product/consumption contextsbecause they make pos-
Next, it is of interestto see whetherthe simple rela- sible a distinctionbetween ambivalenceand indiffer-
tionshipsbetweenproduct/consumption affectandWOM ence, which may not be infrequentreactionsin this do-
transmissionsnoted before occur directly, as hypothe- main.
sized in H8, or are mediatedthroughsatisfactionjudg- Thoughaffecttaxonomies(e.g., Izard1977)assertthat
mentsand/or its cognitiveantecedents.First, however, subjectivefeelings are organizedinto a finite set of dis-
note fromequations11 and 12 thatthe lattercontribute tinctiveaffect categories,the resultshere indicatethata
World
names
Becausemostwesternlanguagesderivefroma singlesource
(Indo-European), it is possibleto construct'worldnames"'..
productand companynameswhich are properlymeaningfulin
mostwesternnationsandJapan.(Roughly10%of urbanJapanese
is derivedfromEnglish.)
At NameLab,weVemade such worldnamesasAcura,
CognosandSentraby constructionallinguistics.
Theresultof a NameLabprojectis a reportpresentingand
analyzingtrademark-cleared nameswhichexpressyourmarketing
identitypreciselyand powerfully. Wequotecosts accuratelyin
advanceand completemostprojectswithinfourweeks.
Foran informationpacket,contactNameLabInc.,
711MarinaBlvd.,SanFrancisco,CA94123,415-563-1639.
NAMELAB"