Sunteți pe pagina 1din 19

Alternative Models of Categorization: Toward a Contingent Processing Framework

Author(s): Joel B. Cohen and Kunal Basu


Source: The Journal of Consumer Research, Vol. 13, No. 4 (Mar., 1987), pp. 455-472
Published by: The University of Chicago Press
Stable URL: http://www.jstor.org/stable/2489368
Accessed: 14/08/2010 11:22

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=ucpress.

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.

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 is collaborating with JSTOR to digitize, preserve and extend access to The
Journal of Consumer Research.

http://www.jstor.org
Alternative Models of Categorization:
Toward a Contingent Processing Framework

JOEL B. COHEN
KUNALBASU*

Widely different accounts of how people categorize new instances have been ad-
vanced in recent years. This article reviews these alternative formulations with a
particularfocus on the use of concrete category exemplars (from priorexperience)
as an alternative to category-defining rules and prototypes. Itadvances a contingent
processing formulation that emphasizes the flexibilityof the information processing
system in its response to important contextual factors, and describes empirical
procedures useful in identifying categorization processes.

A significant amount of money is spent creating information processingassumptions of our most fre-
and communicating specialized product "posi- quently used attitude models apt descriptions of the
tionings" to appeal to the wants and situational re- underlyingcognitive process?
quirementsof particularmarketsegments.At the con- This article takes the view that identification and
sumerlevel this translatesinto a judgmentof a product's evaluationare fundamentallyintertwinedand are out-
categorymembership.The outcome of this process is comes of a process designed both to provide meaning
not only a particularidentification of a product, but and to facilitate a readinessto respond. Accordingly,
the increasedsalience of information relevant to that those interestedin either or both of these topics might
category(and the correspondingsuppressionof infor- benefit from giving added considerationto alternative
mation relevantto other categories)together with the viewsof the categorizationprocessper se. The objectives
category-basedinferencesthat result. How consumers of this articleare,then to (1) reviewthe most frequently
go aboutdecidingwhat an objectis (andby implication advancedmodels of the categorizationprocess,empha-
whether-it is seen as differentfrom other products in sizing the distinctions between feature-basedand ex-
the class) should, therefore,be of particularconcern to emplar-basedaccounts of the process, (2) present a
consumerresearchers. contingency-based"mixed model" incorporatingthe
Whilethe perceivedsimilarityof one productto oth- effectsof categorylearningandtask-relatedfactorslikely
ers can be studied directlyusing a variety of measure- to be important in categorizationepisodes similar to
ment procedures,these methodsby themselvesprovide those faced by consumers, and (3) introduce research
very little insight into how consumers actually make paradigmsdesignedto examinethe effectsof contingent
categorizationjudgments. Identificationof a subset of processingfactors on the categorizationprocessesuti-
attributesas particularlyimportant to a judgment of lized by the individual. A number of researchpropo-
similarityis useful, but this only speaksto the "what" sitions will be presentedthat offer specific predictions
and not the "how"of the process.Is categorizationtyp- of when feature-basedand exemplar-basedprocessesare
ically an informationintegrationprocess involving an most likely to be observed.
attribute-basedrule, or are there other accounts of the
processthat may fit many consumerjudgmentcontexts CATEGORIZATION: CONCEPTS
better?What is the link between object identification
and categorization,on the one hand, and object eval- AND MODELS
uation on the other-and if there is such a link, are the From the categorization perspective, consumers'
knowledge about products/brandsforms at least par-
tially integrated(i.e., schema-like)structuresin mem-
*Joel B. Cohen is Professor, Marketing Department and Director
of the Center for Consumer Research, University of Florida, Gainesville,
ory. Such memory structuresare composed, at a min-
FL 32611. Kunal Basu is Assistant Professor at the Faculty of Man- imum, of similarlyperceived/judgedobjects(whichwe
agement, McGill University, Montreal, Quebec H3A lG5. The authors referto as a category)and associatedobject-basedand
acknowledge the helpful comments provided by John Lynch and category-basedknowledge. In evaluating alternative
financial assistance from the Center for Consumer Research, University categorizationmodels it may be usefulto recallthat the
of Florida.
rationaleunderlyingthis orientationis primarilyfunc-
455
?3 JOURNAL OF CONSUMER RESEARCH0 Vol. 13 0 March 1987
456 THE JOURNAL OF CONSUMER RESEARCH

tional rather than structural or representational:by InformationProcessingIssues


groupingobjects/eventstogether that are alike in im- in Categorization
portant respects, we enhance information processing
efficiencyas well as cognitive stability (Bruner,Good- Categorization,in essence, involves comparisonbe-
now, andAustin 1956;Lingle,Altom,and Medin 1984). tween a target and categoricalknowledge. Some fun-
Also, such organizedknowledgestructuresallow us to damental information processingissues that differen-
identify (give meaningto) novel items/events, respond tiate alternativemodels are: (1) the nature of category
to them in terms of their class membershipratherthan knowledgethat forms the basis/standardfor the com-
theiroften irrelevantuniqueness,drawinferencesabout parison,(2) the natureof the comparisonprocessitself,
featuresand interactionoutcomes, and make causal or and (3) the degree of automaticityassociatedwith the
evaluativejudgments. process. Possible conceptualizationsof the categoriza-
Indeed, categorizationis considereda fundamental tion process, then, may be arrangedin a 3 X 2 X 2
cognitive activity encompassingall forms of stimulus frameworkdependingon a model'sposition on each of
situations (Mervis and Rosch 1981). Initially investi- the three key factors:type of categoryrepresentation,
gated in the cognitive literatureon object perception analyticvs. non-analyticcomparisonprocesses,and the
and conceptidentification,this perspectivehas recently degreeto which the mechanismsare seen as being au-
receivedmuch attentionin social information-process- tomatic ratherthan deliberative.'(See Exhibit 1.)
ing research.Cantor and Mischel (1979) develop the Whilenot all cells representwell-definedor even well-
view that a categorizationscheme allows one to give understoodprocesses,this frameworkis usefulboth for
coherence to one's general store of knowledge about comparingalternativemodels and for pinpointingkey
people:"applyingour categoriesabout otherpeople of- (and often unstated)assumptionsand unresolvedcon-
ten allows us to feel an almost instant general under- ceptual issues of each. The most frequentlyadvanced
standingof someone we hardlyknow." Workingwith models, often discussedas the "classical,""prototype,"
trait categoriesand social roles, Wyerand Srull (1981) and "exemplar"views of the categorization process
posit that people categorizesocial experiencesand as- (Smith and Medin 1981), will be discussed in the fol-
sociatedexpectationsand activelyor passivelyuse such lowing sections. They are distinguishable,first, on the
integratedknowledgestructuresto attend to informa- issue of categoryrepresentation(although as we shall
tion, interpretand infer meanings,solve problems,set see, prototype conceptualizationsoften do not take
goals, or select a behavior. unambiguouspositionson this topic);this issue will be
The relationshipbetween categorizationand evalu- given particularattentionbecauseof the importanceof
ation has only begunto be discussedand explored(e.g., the basis against which a target (e.g., new product
Barsalou 1983; Fazio 1986; Fiske 1982). Yet a key offering)is to be compared.The "how" of the process
premise underlyingthe hypotheticalexistence of cate- is treated in two ways: first, in terms of whether the
goriesis thatthey arefunctionaland may thus be shaped processis presumedto buildup piecemealfromindivid-
by personal goals, values, or the need to respond in ual features(i.e., analytical)or is somehowmore "holis-
specific ways. In the context of a general product cat- tic" (i.e., nonanalytical);and second, in terms of how
egory, then, subcategoriesmay form not only around automaticor deliberatethe specificmechanismis likely
discriminably different subsets but also (or instead) to be.
around evaluatively different subsets (Cohen 1982). While positions on these issues imply alternative
Since consumers' interaction with various products views of categorization,they also reflectthe breadthof
usually occurs in an evaluativecontext, categorization the topic and the fact that researchershave a somewhat
and evaluation may be intertwinedfrom the very for- differentfocus.Most cognitivescientistsagree,however,
mation of the category(i.e., evaluativediscriminations that the types ofjudgments ultimatelyinvolved in pro-
serving the same function as purely descriptive attri- cessessuch as categorizationare the outcome of various
butes).Subsequentaffectiveresponsesto products,then, computationsacrossunderlyingdimensions(Smithand
may be derived from their identificationas a member
of a particularcategory(Cohen 1982;Hutchinson and 'To our knowledge,the earliestattemptto drawsimilartypes of
Alba 1985;Sujan 1985).In addition,productcategories conceptualdistinctionsamong categorizationmodels can be found
may be defined particularly.well in terms of clear in- in Alba and Hutchinson 1985. Alba and Hutchinson(1985, 1987)
stancesof the categoryas well as (or instead)of feature- use the terms"analytic"and "nonanalytic"in a somewhatdifferent
based rules. These may be conceived of as alternative sense,conformingmoreto Brooks(1978). In that view, analyticpro-
cesseshavethe directeffectof identifyingcriterial(e.g., rule-defining)
routes to product identificationand evaluation. Since aspectsof a categorydefinition.Nonanalyticprocesses,in contrast,
the bases for such consumerjudgments may be quite wouldimply thatjudgmentof categorymembershipis basedon sim-
differentdependingon which route is taken, processes ilarityto one or severalcategorymemberswhetheror not the aspects
of category formation and subsequent product cate- being reliedupon in such a comparisonare necessarilycategory-de-
fining.Whilethis is an importantdistinction,feature-basedcompar-
gorizationshould be of particularimportanceto those isonsareat the heartof both.Ourconceptualization hereseeksinstead
with both theoreticaland appliedinterestsin consumer to isolatethe differencesbetweenfeature-by-feature comparisonsand
behavior. the use of overallsimilarityin makingthe categorizationjudgment.
ALTERNATIVEMODELS OF CATEGORIZATION 457

EXHIBIT1
INFORMATIONPROCESSING ISSUES IN CATEGORIZATION

Category representation

Comparison processes Rule defined Prototype defineda Exemplar defined

Analytic
Automatic mechanisms Preconscious dimensional analysis Target is compared to either a Category is represented by a number of
of target (e.g., brightness, shape, probabilistic rule or an ideal "good" examples. These are specific
location for visual inputs) category member possibly via concrete instances. Comparison to a
instigates serial comparisons to automatic (and exhaustive) target is carried out on a feature-to-
hierarchicallyorganized memory retrieval of features associated to feature basis following initialretrieval of
representations until a some degree with a category. The "plausible" or available exemplars as in
"satisfactory" match is found. larger the number of "successful" (1). Mechanisms tailored to such a multi-
Illustrated in models of stimulus comparisons, the closer the target stage process have not been specified.
detection/identification and word is to the centroid of the category.
recognition. Not treated explicitly Possible mechanisms are
in models of categorization. insufficientlyexplored.
(1) (2) (3)
Deliberative mechanisms Emphasis is on more conscious Same as (2) except mechanism is Exemplar retrieval is probably guided by
and top-down mechanisms. Key more similar to hypothesis testing top-down or contextual factors.
features of target are singled out wherein either a prescribed Closeness of match is determined by
for explicit comparison against a number of critical features tests individualfeatures tests based most
category-defining feature-based must be confirmed or the matches likely on some combination of number of
rule (which itself may be must be "sufficiently close" on a feature matches and closeness of
abstracted from prior learning requisite number of features to individualmatches. Once exemplar(s)
experiences). Illustrated by much establish family resemblance. is (are) identified, idiosyncratic features
of the work described under the may play an important role in all
"classical view." exemplar comparisons.

(4) (5) (6)

Nonanalytic
Automatic mechanisms Same as (1) except configural or Though both representational Preconscious processing of target
template representation of target versions of a prototype model are feature-level information produces an
and feature-based rule are possible, a more "holistic" overall representation that is compared
compared directly using some comparison to some hypothetical directly against an overall representation
overall "automatic" response to "ideal" category member is of an exemplar. Significant dissimilarity/
similarity/dissimilarity(possibly suggested. The mechanism is incongruity leads to search for better
activation level following an unspecified, possibly similar to (7) fitting exemplars. No well-specified
unsuccessful comparison). No wherein a significant departure model exists for this cell.
clearly specified model is available from a "good fit" leads to
for this cell. continued memory search.

(7) (8) (9)


Deliberative mechanisms Same as (4) except a configural or Differs from (8) in that a more "top Similarto (1 1) except that the matching
template representation of target down" comparison process is process and criterion center on some
and feature-based rule are suggested. The criterion, direct determination of degree of overall
compared directly, and criterion therefore, may be similar to a similarityto one or several exemplars.
centers on degree of overall fit "nearest neighbor" rule Configural properties (at a minimum,
(e.g., pattern matching). No clearly proceeding from the most correlations among features) are
specified model is available for "plausible" or available category. possible as a basis for comparison;
this cell. Nonfeature-based comparisons however, no more global model has
are not well specified and may be been clearly specified.
difficultto distinguish from well-
learned and rapid feature-based
automatic mechanisms.
(10) (11) (12)

a Two somewhat different representations of a prototype have been discussed in the literature:(1) a category-defining rule specified in terms of characteristic features of a prototypical category
member (permitting a probabilistic rather than "all or nothing" appraisal of a target), and (2) some overall representation of a hypothetical "ideal" category member (possessing the average
value for the category on all relevant features).
458 THE JOURNAL OF CONSUMER RESEARCH

KemlerNelson 1984).However,accessto such feature- Significant developments in cognitive psychology


level informationduring initial stages of perceptionis overthe lastdecadehavechallengedthe "allor nothing"
almostcertainlyoutsideof awareness.Suchinformation perspectiveof the classicaltheory,however.It has been
may alreadyhave been compiled in a more unifiedrep- arguedthat in realitythereare few categorieswith clear-
resentationof the stimulus at the stage at which it is of cut boundariesjustifyingan exclusivenessof properties
interestto those studyingcategorization.In additionto for all membersof one category.It thus becomes con-
focusingon differentlevelsof analysis,manyresearchers ceptually and operationallydifficultto achieve a con-
who are more interested in judgment and decision sensus on a unique set of defining propertiesfor most
making (e.g., category accessibilityeffects, the use of objects and events. Also, empirical researchhas dem-
concreteinstances in analogicalreasoning)are willing onstrated that subjects' ratings of category typicality
to defer some of the "how" questionspertainingto in- differsystematicallyacrosscategorymembers,as do the
formationprocessingoperations.Our focus will be on speedand accuracyof theirclassification(Rips, Shoben,
processingissues beyond the level of perceptualinput, and Smith 1973; Rosch 1973). The reality of such a
at which point there is likely to have already been a gradedcategorystructureor variabilityin representa-
rapidand automaticread-inof featureinformationand tivenessamong memberscame to be appreciatedas re-
initial representationof the stimulus. searchmoved from the learningof artificialstimuli to
the learningof naturalobjectscharacterizedby greater
The ClassicalView dimensional variability.Moreover, to the extent that
attributelistings reflect categoryrepresentation,some
The classical view of categorizationdeparted from generatedfeaturesare neither necessarynor sufficient,
an S-R responsegeneralizationaccount of behaviorto- but are merely characteristicof particular category
wards "like" objects in its emphasis on formalization members.Further,we categorizethe same item differ-
of informationin memory based on some concept ac- ently in differentcontexts (e.g., orangejuice may be
quisition principle.A guiding premisewas that people viewed as a memberof the class of breakfastnutrients
(consciouslyor not) adopt concept identificationstrat- as well as a refreshmentoption along with carbonated
egies that maximize the informationgained from each beverages).
stimulus presentation,ensure that the concept will be The classicalview, in summary,betteraddresseseas-
attainedwith certainty,minimize errorson the way to ily definable and unambiguous concepts and related
solution, and minimize cognitive strain (Bruneret al. learning contexts; as such, it may not provide a very
1956; Dominowski 1974). Such identification"strate- accuratedescriptiveaccount of the categorizationpro-
gies" would presumablylead to formal structuresof cess in most consumersettings.
knowledgein memory,and such categoricalknowledge
would act as input in a problem-solvingcontext to de- The Prototype/ProbabilisticView
termine behavior.The classical theory has viewed the
concept acquisition process as a search for necessary Categorizationresearchershave advanced a signifi-
and sufficientattributesthatjointly determinecategory cant variety of models that incorporatea "prototype"
membership.Any entity possessing the set of critical construct.The characterizationof naturalcategoriesas
featuresis a member of the category,while any entity "fuzzysets"and the recognitionof a continuum of cat-
lackingeven one of the attributesis not a member.This egory membership(high to low in representativeness)
impliesthat all membersof a categoryshouldbe equally led Rosch and Mervis (1975) to define a prototype in
representativeand that learninga categoryconsists of terms of a set of features commonly associated with
discoveringits definingattributes. members of a category (hence, abstractedfrom expe-
For example, if an individual'sconcept of an econ- rience), with each featureassigneda weight according
omy car were to include a definingattribute"cramped to its degree of association with the category. Such a
seatingspace"and s/he should ever encountera single view de-emphasizesthe search for necessary and sig-
economycarwith ampleleg room, the classicalposition nificant critical features and emphasizes instead a
requires either a change of concept, since "cramped "goodnessof membership,"where degreeof prototyp-
seating space" would no longer be true of all category icality is operationallydefined by people's judgments
members,or assignmentof this carto a different(maybe of how well variousobjectsfit with their concept of the
new) category,e.g., "comfortableeconomy cars." meaning implied by a category label. Though Rosch
In Exhibit 1, the classical model seems to fit best in and Mervis do not preclude memory for specific ex-
the rule-analyticsection of the framework.While no emplars,they seem to imply that the latterwill undergo
specificprocessingmechanismis explicit in its formu- some type of abstractiveprocess.It is unclearwhether
lation, comparisonagainst a feature-basedrule repre- this abstractiveprocess occurs entirely at the level of
sentation seems to imply an attribute-by-attributean- features(i.e., formationof a probabilisticfeature-based
alyticprocess.It also appearsthat such an analyticpro- rule) or whetherthere might even be a representation
cess may be deliberateas well as automatic depending of such a hypotheticalentity that could be used more
on task and other contextual aspects. directly in categorizing target instances. Some have
ALTERNATIVEMODELS OF CATEGORIZATION 459

simply assumed the latter. The major thrust of this of a categoryand utilize the former in making subse-
stream of research is toward development of taxonomies quent categorizationjudgments for new instances. A
or hierarchies of categories and delineating intra-cate- relatedpossibilityis that prototypesresultfromthe loss
gory relationships. There is less attention to the specific of individuatinginformation pertainingto previously
mechanisms people are likely to employ in the cate- categorized instances, such that one is left with a
gorization process. schema-likerepresentationthat resemblesthe "typical"
Other discussions of a prototype continue to stress categorymember.
that categorization is best seen in terms of some overall In comparingthe differentperspectiveson the "pro-
criterion of fit rather than the application of a fixed set totype view" to the frameworkdescribedin Exhibit 1,
of defining features. But the nature of a prototype be- thereappearsto be hardlyany consensusvis-a-visboth
yond that of a probabilistic rule remains vague: "it is representationalform and processingmethod or mech-
probably the degree of family resemblance, not the anism.Both abstractentities(including"ideal"category
continuous surpassing of a few critical properties tests, representatives)and probabilisticrules have been sug-
that determines category membership in everyday per- gestedas categorycomparisonstandards,and the nature
son perception" (Cantor and Mischel 1979). Cantor of the processingitself has receivedless than adequate
(1982) makes a nice case for a prototype model over a attention.The traditionalemphasiswas on the structure
more classical model since it can better represent one's of prototypesin terms of "fundamentalunits" of cog-
response to the variety of possible subsets of features nition (i.e., features),and this seems more consistent
encountered in the social domain. Yet there is some with a piecemeal. analytic categorization process.
question about whether these abstracted features (i.e., However,recenttrendsin socialinformationprocessing
of typical category members) are thought to combine researchreflecta tendency to assume the possibilityof
into a constructed representation of a hypothetical pro- alternativemechanisms (e.g., based on family resem-
totype or to merely comprise a probabilistic rule: blance or the overall configurationof attributes)that
while incompletely specified,are certainlyless mecha-
This prototype-matchingmodel assumesthat in typing nistic (Cantorand Mischel 1979;Wyerand Srull 1981).
an object(person,situation,self) the perceiverestimates As Exhibit 1 shows, level of conscious deliberation/au-
the degree of similaritybetween the object and a pro-
totype for each category into which the object might tomaticityhas not been adequatelydiscussedin special
plausibly fit. Categorizationis a probabilisticprocess relationto prototypemodels (or any other model of the
based on degreeof overlapin featuresbetween the new categorizationprocess).Such operationsmay be similar
object and the prototypefor each differentcategory. to alertingmechanismsin signalingshifts of attention
and in directingresponsesto incongruity. These pro-
A similar process of categorization is advanced by Reed cesses may also not be very differentfrom those un-
(1972). derlyingdeliberative(or at leastconscious)mechanisms,
Though primarily concerned with category accessi- but they may operatewith minimal thought and effort
bility using a memory bin model, Wyer and Srull ( 198 1) due to prior learningand repetition.2
describe a process in which specific aspects of infor- In summary,there is considerabledivergencein the
mation drawn from the target stimulus are compared meaningand implicationsof the prototypeconcept.On
to features of the most accessible schema, and where one hand the term is used simply to referto a category-
interpretations are drawn on the basis of the outcome definingrule specifiedin termsof characteristicfeatures
of such a comparison. However, they also mention the of a prototypicalcategorymember. Other uses of the
possibility that an instance may be categorized on the term, however, suggestsome overall representationof
basis of a comparison of the configuration of its features a hypotheticalcategory member. In this event a pro-
with configurations contained in additional bins with- totype could be an abstractedrepresentationof either
out prior encoding of the individual features comprising individualinstances(i.e., a composite) or an ideal cat-
the configuration. This is clearly a different (and cer- egorymemberbased on a computationalprocessto ar-
tainly more nonanalytical) view of the process from the rive at the centroid of feature-basedvalues. Or, finally,
traditional perspective in which one views the prototype a prototypemay simply resultfrom the loss of individ-
as the central tendency, defined as the mean value of
the set of stimulus objects on each relevant feature di-
mension (Posner and Keele 1968). Such a central ten- 2Furtherdiscussionof the natureof possible (and unresearched)
dency prototype would not necessarily be descriptive automaticmechanismsin categorizationwouldbe speculative.There
is a considerableliteraturedealing with preconsciousprocessesin
of any actual or imagined member of a category. Cat- perceptionthat may offersome suggestionsto the interestedreader
egory members may be viewed as varying in represen- (e.g., Posner and Snyder 1975). The literaturedescribingmultidi-
tativeness in relation to their departure from the cen- mensional stimulus comparisons has incorporatedvery fast and
troid of mean values. Homa and Vosburgh (1976) have probablypreattentiveholisticcomparators(Krueger1973;Smithand
KemlerNelson 1984) that appearto supportoverallsimilaritycom-
also departed from this interpretation in seeing the Pos- parisons.A few sourcesfrom other domainsinclude Bargh's(1984)
ner and Keele findings as providing evidence that people discussion of automatic and conscious processingand Mandler's
abstract a prototype during interaction with exemplars (1982) ideas relatingincongruityto arousal.
460 THE JOURNAL OF CONSUMER RESEARCH

uating information associated with particular category target object is matched against-a stored or con-
members (hence, a by-product of memory decay). structedprobabilisticrule or hypotheticalcategorypro-
In addition to (and somewhat as a result of) confusion totype versus salient "good instances"of the category
resulting from the variety of uses of the term, the pro- (see exemplar-analyticsection of Exhibit 1). Such a
totype views have been critiqued in a number of aspects: version of an exemplarmodel would differfrom a pro-
1. Categoriesare frequently modified when complex totype model in that "rule-irrelevant"or idiosyncratic
concepts are learnedover time. Yet if an abstracted aspects of salient exemplarscould prove to be critical
representationhas been formed, it is not clear how in the categorizationjudgment, and these featuresare
the new informationis to be combined with it (e.g., likely to differdependingon which exemplarsbecome
some complex type of weightingscheme) depending salient. Since categorymembersalience is likely to de-
on variouscharacteristicsof the alreadyexistingpro- pend substantiallyon contextualfactorsthat influence
totype. category member retrieval, exemplar models would
2. Systematicbiasingeffectsin categorizationare found predict variable categorizationoutcomes better than
for memorableinstances(Lingleet al. 1984). That is, would prototypemodels, unless one assumesthat sep-
there is a propensityto categorizesomewhatambig- aratecategories(each with a prototype)are createdfor
uous targetitems consistentwith "similar"concrete each situation or task the individualfaces. This would
instancesexperiencedin the past when the latter are clearly imply the existence of a large set of categories
used as primes prior to the judgment. Central ten- and would seem to undermine the usefulness of pro-
dency notions by themselvesdo not seem to handle totype constructionrelativeto more parsimoniousex-
this. emplaraccounts.3
3. People appear to use information beyond just the A second possible exemplar comparison process is
characteristicpropertiesof a concept. One such ad- more holistic. Its premiseis that "psychologicalmean-
ditional type of informationis correlationsbetween ing" exists at the level of the entitywhetheror not more
attributes.In fact, some authors(Medin et al. 1982) elemental (feature-level)information is also stored in
have claimed that most of the complex entities we memory. The rapid assembly of feature-level infor-
encountermay not be representableby differentcom- mationinto a moreintegralrepresentationof a stimulus
binationsof a basic set of orthogonalattributes. may also imply that access to entity-level representa-
4. 'Aprototypeviewimpliesthatcategorylearningshould tions is more immediate than access to feature-level
be easiest when membersof two categoriesare sepa- information (Smith and Kemler Nelson 1984). This
rableby a linearfunctioncorrespondingto a summing suggeststhat nonanalyticoverallsimilaritycomparisons
of feature information. However, there is evidence may indeed be primary,with attribute-basedanalytic
that this is unnecessaryand that a more important processes"available"when necessaryto meet task re-
criterion is the frequency of within-group close quirements(Cohen 1982). One parallelhere is with the
matches to the target (Medin and Schwanenflugel
1981). Gestalt view of perception,whereinobject recognition
seems almost to bypass accumulation of "elemental"
sensory impressions (i.e., building up to a "whole"
The Exemplar View which is subsequentlyrecognized)but instead reflects
"fitting"entire configurationsof sensory impressions
In the exemplar view, target instances are categorized to immediate prior expectations (Bruneret al. 1956).
by cuing the retrieval of specific category exemplars. Further,items are often recognizedthrougha top-down
The more similar the target instance is to concrete ex- processguidedby prioranticipationand consideration
emplars of a category, the more likely it will be placed of the entire pattern of representation.In object cate-
in that category. This view emphasizes the importance gorization,prior mental states and/or situationalcues
of the particular category exemplars that are accessible may evoke certainexpected items (e.g., brands),and a
during the categorization process. So, factors that in- configuralrepresentationof the targetobjectcould then
fluence exemplar retrieval (e.g., frequency, recency, sa- be comparedto the salientexemplar(see the exemplar-
lient goals, distinctiveness) rather than a more general nonanalyticsection of Exhibit 1).
rule (even a probabilistic one) are likely to be important Brooks(1978) has probablyreportedthe most com-
in object classification. The exemplar view, then, differs pelling evidence for the use of an exemplar-matching
from the others initially in a "quasi representational"
sense-namely, that it assumes that a molar represen- 3A similarargumentis made by Lingleet al. (1984). In addition,
tation of specific category members is brought to mind normtheory(Kahnemanand Miller1986)assignscategoryexemplars
(regardless of how such information in fact may be a pivotal role-not only in the representationof categories,but in
stored in memory). establishinga normativebaselineafterthefactas a functionof memory
Two distinct types of comparative processes may then accessibility.Such a derivednorm guides interpretationand evalu-
ationmuchas a precomputedexpectationor schemaprovidesa frame
follow. The first is a type of aspect-by-aspect matching of reference.However,normtheoryquestionsthe flexibilityand gen-
process. In this event the major difference between the eral usefulnessof such precomputedknowledgestructuresmuch as
three models would lie in their treatment of what the we have questionedthe ubiquityof storedprototypes.
ALTERNATIVEMODELS OF CATEGORIZATION 461

strategyfor acquisition of complex concepts. His sub- formationin diagnosis,even if it meant violatingsome
jects learnedto pair structuredletter stringsgenerated independentfeature-basedrule learnedpreviously.
from two Markov grammarswith the names of cities Medin and Schaffer( 1978) sought to account for the
and animals in a paired associate task. The two types Posnerand Keele (1968) findings(suggestedas provid-
of labels were not correlated with the grammatical ing evidence for use of prototypes)by a model based
membershipof the letter strings. Also, subjects were solely on exemplar information. Their cue-context
unawarethat the letter strings could be classifiedinto model proposes that a probe stimulus functions as a
two categories.Subjectswere presentedwith new letter retrievalcue to access informationstored with stimuli
stringsand were asked to classify them into three cat- similar to the probe. They claim that the various cue
egories (i.e., the two grammar-generatedletter strings dimensions comprisingstimuli are combined in an in-
and those belongingto neithercategory).Subjectsper- teractive (multiplicative) manner to determine the
formed the task with 60-65 percent accuracy,but ap- overall similarity of two stimuli. In a similar model
parentlywithoutbeing able to specifythe categoryrules buildingeffort,Hintzmanand Ludlum(1980) wereable
and defining attributes.Subjects, therefore, appeared to simulatethe Posner and Keele findingswith a com-
to havelearnedthe basisof categorizationin some other puter model (MINERVA)that used the exemplarsas
way duringthe pairedassociatetask. Brooksconcluded the sole basis for generalization.While both the cue-
that given a sufficientlycomplex concept or difficult context model and MINERVA demonstratethat the
task environment,subjectsappearto fall back on clas- "pureexemplar"and the "pureprototype"basedmod-
sificationthroughmemorizationof the exemplarsand els are often equivalent in terms of predicting target
matchingnew instancesto these. classification(since categoryexemplarswill themselves
Workingin a differentresearchdomain,Smith(1981) often be similarto an idealizedprototype),this research
investigatedthe perceptualprocessesthat childrenuse does not provide sufficientinsight into the underlying
across differentdevelopmentallevels. The traditional mental process that distinguishes comparisons to an
view in the developmental literaturewas that young exemplar model from comparisonsto the product or
children perceive objects as wholes and classify them elements of a cue-combinatorialmodel.
by holistic similarityrelationswhile older childrenand Followinga differentline of inquiry,Read (1983) and
adultsperceiveobjectsas conjunctionsof attributesand Gilovich(1981) demonstratedthat as a particularcause-
classify objects by those attributes. Smith's studies effectrelationshipbecomes complex, or in the absence
showedthat this was true only in the case of extremely of a known causal relationship,people tend to make
simple stimulus variations.As the variationsbetween predictionsbased on the overall similarityof the new
objects became more complex, overall similarityrela- instance to a similar instance experiencedin the past.
tions were used by children as well as adults in classi- In Read's studies, subjectslearned about members of
fying objects. Smith and Kemler Nelson (1984) argue a foreignculture,some of whom performeda rule-gov-
furtherthat holistic similarity may be the most basic erned behavior. Subjectsthen predictedthe behavior
categorizationprocessin that access to components of of othermembersof the culture.One memberwas sim-
a successfullyunitized representation(from an earlier ilar to an individualwho performedthe behavior,but
stage in the perceptualprocess) may deteriorateat a the rule predictedthat this new memberwould not per-
faster rate. The advantagesof overall similarity seem form the behavior. The studies showed that subjects
even more pronounced as the speed or complexity of reliedon the similarityof this memberto the previously
the classificationtask is increasedand unless people are encountered individual in making these predictions.
given instructions to process in a more analytical Further, it was found that people were increasingly
fashion. likely to use a single similar instance as the rule gov-
In another study, Kossan (1981) examined whether erning the behavior became more complex, and that
concept-acquisitionstrategiesvaried as a function of suchjudgmentswere made quite confidently.Gilovich
the complexity of the internal structureof the concept demonstratedthe same type of reliance on presumed
to be learned,and whetherchildren at differentdevel- correlatedaspectsof highly similarcategoryexemplars
opmental levels preferredone form of strategyto an- in two separatedomains:ratingof college footballplay-
other.In her experiment,conceptsdefinedby necessary ers in terms of their potential to play professionally,
and sufficientfeatures(a two-featureconjunctive rule) and recommendationsof how to solve a hypothetical
wereacquiredthrougha featureabstractionstrategyby international crisis. Neither Read nor Gilovich at-
both secondand fifthgradesubjects.However,for more tempted to specify a formal categorizationmodel be-
complex concepts, the common feature-abstraction yond relianceon analogousreasoning.
strategywas found not to be the optimal approach,and The use of exemplarsas a basisfor comparativejudg-
an exemplar-matchingstrategyseemed to be used by ments seems to be consistent with introspective ac-
both second and fifth graders.Workingwith hypothet- counts of the process and of the importanceof easily-
ical medical case studiesas stimuli, Medin et al. (1982) brought-to-mindinstances in various domains of hu-
found that subjects were sensitive to correlationsbe- man judgment.As indicatedin Exhibit 1, however,the
tween symptoms and actively used such configuralin- lack of well-specifiedinformation processing mecha-
462 THE JOURNAL OF CONSUMER RESEARCH

nismsto accountfor exemplar-basedcategorization(as priate and useful in the study of consumer behavior?"
well as the use of concrete instances in analogousrea- Some reduction in scope and complexity can be
soning)remainsan obstacleto the acceptanceof often achieved by concentrating on the underlying infor-
closely linked nonanalyticexplanationsof the process. mation processingdimensions identified in Exhibit 1
Analyticand more computationallydriven models are that are deservingof the most emphasis.While each is
better developed, and there is a sense at least that we importantin its own right,the "level of automaticity"
aremakingprogressin understandingtheirinnerwork- issue is probablyless directlyrelevantto consumerbe-
ings and relatingthem to broadly based work in cog- havior and the most difficult to investigate in a con-
nitive science. sumer setting. By contrast,alternativerepresentations
Exemplar-basedconceptualizationsemploying ana- of product-definedcategoriesand comparisonprocesses
lytic bases for comparison(see Exhibit 1) may steer a used to assess product similarity fall more directly
middle coursebetweenthe importancein everydayex- within the domain of consumer behavior. These may
periencesof concrete instancesand episodic playback, be shaped by specific learning histories of individuals
on one hand, and the wide acceptanceof feature-based (including marketingand advertisinginfluences) and
processeson the other (see Alba and Hutchinson 1985 the contextual influences of choice and consumption
for a related ctiscussion).Nevertheless, there is now environments.
considerableevidence for the kinds of preattentivefea- As evidencedin Exhibit 1, even after such a dimen-
ture-basedprocessingnecessaryto createa moreintegral sional reductionthere remain severaldiverse views of
representation.Having created such a holistic repre- the categorizationprocess. Most of these are at least
sentation, it would seem pointless to immediately at- plausible(e.g., examplesof each can easilybe generated
tempt to retrieve the underlying features in order to if not recalled).5It may be most productive,therefore,
carryout fairlycommonplacecategorizationtasks.The to assumean extremelyflexibleinformationprocessing
latterwould appearto be a slowerand less efficientpro- system and to carry out researchdesigned to explore
cess. the conditions under which various routes to catego-
We think it is reasonableto begin the study of cate- rizationarefollowedalongwiththe implicationsof each
gorizationwith the assumptionthat preconsciousfea- (e.g., for subsequentevaluation and inference as well
ture extraction/analysismechanisms are undoubtedly as susceptibilityto bias in initial categorization).So,
at work at an earlierstagein the perceptualprocess.At ratherthan ask, "which model is correct?"one could
the level at whichwe begin to focus attentionon a stim- assumethat people have or could constructdescriptive
ulus, however, information processing operations are or hypotheticalfeature-levelas well as entity-levelrep-
likely to reflect the natural co-occurrenceof features resentations of category members and, in addition,
and overallcharacteristicsthat enable one to deal with could process information in an analytical or nonan-
the objectas a concreteentity.Garner(1978) has argued alyticalfashion(i.e., that people have the requisiteabil-
that as objects become complex and heterogeneous, ity and cognitiveflexibilityto do so). The questionthen
separabledimensionsareperceivedto be more integral, becomes, "underwhat conditions do people engagein
which may even tend to inhibit analyzabilityby parts. each type of informationprocessing?"
In part,this may be a resultof the very automaticityof Within a consumer context, for example, an evalu-
initial featureextractionprocesses.A mechanism that ative category might be "departmentstores I love to
providesa rapidand exhaustive"reading-in"of feature- shop in." Knowledge about such a category may be
based information can set the stage for more holistic representedin the form of concrete stores the person
processingsimplyby makingthe necessaryinformation has frequented in the past as well as more abstract
available. It then becomes easier to discover overall featuresextractedfrom priorexperiences(such as, "de-
patternsamong the more elemental information, and partment stores I love to shop in usually have high-
this becomes a plausiblealternativeto a serial,self-ter- quality merchandisein contemporarystyles, a distinc-
minating analysis based on individual features.4This tive environment in terms of lighting, aisle space, ex-
suggeststhat categorizationjudgments are likely to be pensive decorativefixtures,etc."). On encounteringan
particularlysensitiveto correlatedfeaturesas well as to unfamiliardepartmentstore, the person could adopt
overall,unique configuralpropertiesof the targetstim- (deliberatelyor as a quasi-automaticresponse to con-
ulus. textual stimuli) severalcategorization"strategies"(ig-
noringthe detailed processingmechanismsof each):
A CONTINGENCY-BASED, MIXED 1. Bringto mind one or severalexemplarsof the cate-
MODEL APPROACH gory,select focal featuresand comparefeaturesof the
Given the diversity of categorizationmodels, a key
question is, "which ones are likely to be most appro-
'While the representationalaspectsimplied in cells 7 and 10 are
logicallyseparablefromthe others,when the individualfeaturesare
4Thisissue has been discussedin relationto the retrievalof infor- integratedinto some type of composite the process so stronglyre-
mation from short-termmemory.For a summaryof particularrel- semblesa prototypeview (i.e., cells 8 and 11) that we are merging
evanceto consumerbehaviorsee Lynchand Srull 1982. these togetherin the interestof parsimony.
ALTERNATIVEMODELS OF CATEGORIZATION 463

unfamiliardepartmentstore to them; make the cat- malevolently engineered(i.e., nonexperimental)envi-


egorizationjudgmenton the basisof surpassinga crit- ronmentsin which they live.
ical number of individual features tests (Exhibit 1, When we consider typical consumer settings, there
cells 3 and 6). is at least an intuitive sense that the type of evidence
2. Bringto mind a criterialset of featuresthat definethe producedby studies such as those of Read (1983) and
categoryor the categoryprototypeand comparefea- Gilovich (1981) is quite pertinentto product categori-
tures of the unfamiliar department store to them; zation.Thatconsumerswould have recourseto familiar
makethe categorizationjudgmenton the basisof sur- categoryexemplarseither apartfrom or instead of cat-
passinga critical numberof individualfeaturestests egory-definingrules seems to squarewith conventional
(Exhibit 1, cells 1, 2, 4, and 5). thoughtand practice.Brandsare often viewedas having
3. Bringto mind one or severalexemplarsof the cate- an "identity"of their own. At the veryleast this implies
gory;make the categorizationjudgment on the basis that consumers recognize patterns of relationships
of overallsimilarityof the unfamiliardepartmentstore among featuresand possibly configuralpropertiesbe-
eitherto a focal exemplaror exemplarsor on the basis yond merelycorrelatedfeatures.With productsas with
of an acceptablematch againsta criterialnumberof people, necessaryand sufficientfeaturesand the idio-
exemplars'(Exhibit1, cells 9 and 12). syncratic combine to create a distinct identity, and
4. Bringto mind a categoryprototype(or constructone consumers seem able and willing to think about and
based on representativeperformanceon individual decide among products as intact entities. In a super-
features);make the categorizationjudgment on the marketsetting,for example, it seems likely that an un-
basisof the degreeof overallsimilarityto the prototype familiarproductofferingwill be categorizedand eval-
(Exhibit 1, cells 8 and 11). uated on the basis of a high degreeof overallsimilarity
to an exemplarof a particularcategory.This notion has
There are two senses in which these models may be been exploitedin packagedesignand sometimesresults
"mixed" (see related discussions in Cohen 1982; Elio in the purchaseof a productthat, on closer inspection
and Anderson 1981; Lingle et al. 1984). First, they may at home, is differentfrom the product the person was
be mixed in the sense of each being included in the intending to buy. Whetherthis happens because con-
individual's repertoire, so that depending on contextual sumersdo not haveor do not use category-defining rules
variables, one or the other might be used. Second, we under certain conditions, and what those conditions
advance the argument that in any given categorization may be, would seem to be interestingquestionsfor con-
instance there may be a "first cut," using the most ef- sumerbehaviorresearch.
ficient means of making a quick judgment, and that Analytic and nonanalytic operations at the level of
this could be followed by processes that would give assessing the category membershipof a target object
greater precision, ifthey are deemed necessary (by some requiremore precise specification.Feature-basedpro-
deliberate or automatic mechanism). cessingimplies that either all or a necessarynumberof
In the latter sense, a process similar to item 3, or individualfeaturecomparisonsneed to be successfully
possibly item 4 if a prototype were already available in made for categorizationto occur. Clearlythis is not as
memory, would seem advantageous either when there cut and dried as it might first appear, since different
is relatively low motivation to make a careful catego- criteria could be used both at the individual feature
rization judgment or, conversely, when the costs asso- level and to determine a successfulfit across a feature
ciated with delay are very high, even when the judgment set. Such criteriaare probablynot fixed, but are likely
is a difficult one to make. Suppose, for example, that to vary as a function of involvement and the salience
an object needs to be categorized as one that a person of contrastcategQries(to the extent that the fit of other
should either immediately approach or one that should potential"matches"aretaken into account).Thereare
be immediately avoided, and no isolated high validity clear links here to researchon decisions rules.
cue(s) is (are) available. In this instance, one rationale A simply stated criterion for an exemplar-based,
for using a category exemplar is that it may be brought nonanalytic processingmodel is simply the degree of
to mind quickly, and yet it incorporates considerable "overall fit" to one or several salient exemplars. But
feature-level information. So, if an exemplar of each of this also raises a great many issues, such as specifying
two opposing action-directed categories is compared how any type of template-matchingprocess would be
against the target instance, with the best "match" al- carriedout (i.e., how the degree of fit is determined),
lowed to determine the outcome, this is likely to be as well as how the numberof exemplarsinvolved would
optimal under the circumstances. We frequently ob- be determined. Lingle et al. (1984) have proposed a
serve that people seem almost predisposed to utilize criterion that essentially involves the number of ex-
concrete "good examples" in making judgments even emplarrepresentationsthat a targetobjectis ableto cue
though they have access to more precise feature-based from within a given category. The probability of re-
information known or assumed to have greater cue va- trievingany given exemplarshouldbe a function of the
lidity (e.g., Kahneman, Slovic, and Tversky 1982). Per- similarity of the encoded representationof the target
haps people are relying on a generally useful exemplar- stimulusto the memoryrepresentationof the exemplar.
based strategy that tends to be successful in the less Perhapssome "automatic"responsemechanismto the
464 THE JOURNAL OF CONSUMER RESEARCH

lack of a close fit (see Mandler 1982, for a discussion they may differin initialaccessibilityand responsiveness
of responses to incongruity)-possibly involving af- to contextual cues. This suggeststhat attribute-based
fect-would be worth investigating. rules are more likely to exist and to prove to be valued
Demonstratingthe operation of nonanalytic versus functionally for categories learned through a defini-
analytic processingin a categorizationtask is a formi- tional process or for categoriesinvolving formal rela-
dable undertaking.Categoryassignmentitself will not tionshipswith otherconcepts.In such casesthe objects/
normallybe a sensitiveenoughmeasureunlessdifferent conceptsthemselvesare initially thoughtof in terms of
outcomesfollow directlyfrom individualfeature-based collectionsof properties.For example,generallyspeak-
comparisonsratherthan from more complex patterns ing, consumersmay have and use formalrulesforjudg-
of featureassociation(e.g., KemlerNelson 1984;Medin ing members of categories such as "financial instru-
and Schwanenflugel1981). Even here one needs to be ments" ratherthan for categorieswhose members are
able to demonstrate that some multiple feature rule accumulatedsequentially over time as a result of in-
(e.g., having conjunctive and disjunctive properties) dividual experiences (e.g., preferredfast food Trestau-
would not provide the same results. Process-oriented rants, country music).
measures have the potential to tap or at least reflect Also, learningthat accompaniesexposure to exem-
alternativeinformation processingoperations.To the plarsfrom contrastingcategoriesof products(e.g., Bur-
extent that memorytracesare made more accessibleby gundy versusBordeauxwines) may focus attention on
the use of the correspondinginformation,for example, characteristicsthat separatethe two groups and may
we may be able to infer the use (or nonuse) of various encouragesubsequentuse of simple heuristicsbasedon
units of information (though determiningappropriate such features(e.g., bottle shape).On the other hand, in
baselines against which comparisonsshould be made the naturalecology we typically experienceindividual
is not alwayseasy when differentlevels of information entities in some depth and richness(e.g., we meet "Joe
areat issue).We will look furtherat some of theseissues Jones,"we read a particularbook, we taste a hot fudge
in a subsequentsection of the paperdevotedto research sundae),and we groupitems into categoriesby observ-
paradigms. ing similaritiesamong individual entities ratherthan
In summary,our contingency-basedmixed model is by seeking to identify differencesbetween contending
derivedfrom the followingassumptions:(1) that a rea- sets. Then upon encounteringa targetinstance, a com-
sonableamountof feature-basedinformationaboutany parison can be made to those previously experienced
given stimulushas been automaticallyreadin upon ex- instancesof a categorythat appearto offera good match
posureto that stimulus, (2) that both analytic(i.e., fea- quite apart from any explicit or even implicit consid-
ture-by-feature)processing of this information and erationof alternativecategories.One reasonfor the so-
nonanalytic(i.e., basedon correlatedfeaturesand con- called "confirmatorybias" in reasoningmay be that it
figuralproperties)processingare possible, and (3) that simply does not occurto people to think about counter
the information processorhas the flexibility to adopt examples (e.g., Holyoak and Glass 1975; Tverskyand
one or the other mode, or both sequentially,as a func- Kahneman 1973), which are in fact exemplarsof plau-
tion of contextual factors. These factors, then, should sible contrastcategories.Thus, the impact of category
be expected to have a major role in selecting an indi- learningconditions on subsequentcategorizationpro-
vidual'scategorizationresponse,whetherin the natural cesses may be investigatedfocusing on the following
consumerenvironmentor in a studyexaminingaspects researchpropositions:
of the process.We will discuss these contextual factors
in terms of initial categorylearning,task involvement, P1: Conditions that favor category acquisition
rule complexity, and the judgment setting, with each througha definitionalprocess (i.e., involving
followed by some summaryresearchpropositions. learning the formal relationships between
more abstract/nonperceptualaspects of an
entity or entities) lead to greaterreliance on
The CategoryLearningContext rule-based and analytic processes in subse-
Beginning in childhood, natural ongoing concept quent categorizations.
learningis focused around a name-picture paired as- P2: Conditionsthat favor "in-depth"exposureto
sociateratherthan a definitionalrule (e.g., pointingand exemplarsof a category,in a contextfreefrom
saying "cat"). We typically acquire a considerable any explicit or implicit consideration of al-
amount of informationabout a particularobject or in- ternative(contrast)categories,lead to greater
dividual present in each situation, and this rich store reliance on exemplar-basedand nonanalytic
of instance/episodicinformationis availablein memory processesin subsequentcategorizations.
for subsequent information processing activities. At-
tribute-basedrules for category membership, on the
other hand, are more easily developedthroughformal Task Involvement/Motivation
training and exposure to more abstract(i.e., nonper-
ceptual)concepts. Such rules may also coexist with ex- By its very nature,a mix of categorizationheuristics
emplars as part of category-basedknowledge, though is likely to be particularlysensitiveto task-relatedcon-
ALTERNATIVE MODELS OF CATEGORIZATION 465

textual cues. Among the most important of these are Rule Complexity
cues that signalthe natureand importanceof the prob-
lem for which categorizationof stimuli is a necessary In our exposure to the objects in our environment
step. thereis no shortageof clearinstancesof most categories.
Severalauthors(Brooks 1978;KemlerNelson 1984) We generallyhave no difficultybringingthese category
have arguedthat when categoryinformationis acquired exemplarsto mind; indeed, we can hardly keep from
from experiencesthat are incidental ratherthan goal- doing so. On the other hand, some analytic rules are
directedin nature, subsequentclassificationsare more both easy to discover and easy to apply;some are dif-
likely to be characterizedby exemplar-based(and pos- ficult in one of these respects,and some are difficultin
sibly nonanalytic)processing.On the otherhand,when both. Discoveringa ruleimplies,first,a desireto engage
informationis acquiredvia a task explicitly involving in some abstractiveprocess and, second, being able to
a categorizationdecision, subsequentanalyticprocess- do so. Conditionsof exposureto categorymembersmay
ing is more likely (Martinand Caramazza1980). simply not favor isolation of features,especiallywhen
If consumersfindthemselvesin a tasksettingin which we regardinstances for their own sake (ratherthan as
they are confrontedby a problemto solve, thereshould membersof categories)and do not seek to differentiate
firstof all be heightenedattention to a narrowerrange membershipin one categoryfrom membershipin an-
of cues relevantto the problem.A key issue, then, is to other. As has been pointed out in connection with the
understand(or correctly manipulate) people's assess- classicalview of categorization,many categoriesdo not
ment of the nature of the problem. If the problem is appearto have straightforward definingrules,but rather
perceivedto be one of object identificationand assign- involve disjunctivecombinations of featuresand cor-
ment, then aspects of the stimuli that are pertinentto related attribute patterns. Such rules are not easy to
differentiatinginstances from each other yet that are acquire, at least in the natural course of things, and
perceived to be systematic (i.e., homogeneous within appearto demandmore ideal acquisitionopportunities
the set of similar instances and heterogeneousacross (e.g., simultaneouspresentationof multiple instances
dissimilarinstances)are probablybroughtinto sharper that encourageshypothesistesting).Even if such a rule
focus. Such settings,then, promotethe constructionof wereacquired,its applicationwould depend on the op-
a categorizationrule (if one had not already existed) portunitiespresentin the categorizationsetting (as dis-
and prompt its use. This tendency is heightened con- cussed in the following section).
siderablyby a task structurethat stresseschoice among For the ad hoc and evaluative categoriesimportant
alternativecategories.It is lowered to the extent that in product categorization,some types of advertising
the problem focus is not explicitly linked to categori- may be instrumentalin focusing attention on features
zation (e.g., how much money to spend on a record that implicitlydifferentiatebrands,while othertypes of
album ratherthan which of severalcategoriesof music advertisingmay simply provide a richerportrayalof a
a particularalbum best fits in), and it is loweredwhen particular brand. Similarly, some distribution and
contrastingcategoriesare not explicitly suggested. productpresentationsystems either encourageor mit-
In the consumer context, task involvement may co- igate against the development of rules to discriminate
vary with product type (Lastovicka 1979), with rele- among product offerings(e.g., one may have to visit
vance and centralityto the need system governingthe several stores to compare products).These factors, in
level of attention and effortdevoted to the product or additionto inherentdifferencesin products(e.g., com-
message.Moregenerally,the wide rangeof well-learned plexity of rule structure;consumers' desire to "see"
activities that people carry out with little thought at certainproducts-cigarettes, beer, etc.-in a more ho-
normal levels of activation may be carried out more listic fashion,possiblyto the degreethat they arelinked
deliberatelyat moderatelyhighlevelsof activation.Both to the self-concept),provide ample opportunityto ex-
general and task-specific increases in involvement, plore the full range of categorizationmechanisms im-
therefore,will likelylead to greaterattentionbeinggiven plied by the following proposition:
to analyticalbases for categorization(unless attention P4: Increasingthe structuralcomplexityof an an-
is directed elsewhere). Much of consumer behavior alytic rule describing category membership
seems to involve incidental exposure to products and leads to greater reliance on exemplar-based
marketingactivities, and many decisions are either re- representationsand nonanalyticclassification
petitiveor of minor importance.Under such conditions processes.
nonanalyticalcategorizationprocessesmay well be the
norm.Adequatelysimulatingsuch importantfactorsin
consumer categorization research would provide an The JudgmentSetting
opportunityto test the following researchproposition: The environmentin which a categorizationis made
may be characterizedby a numberof aspectsthat work
P3: Conditions that increase an individual's in- for or againstthe use of an analyticalor nonanalytical
volvement (i.e., level of processing activity) strategy.Time constraintsrelativeto the amount of in-
with an entity lead to greaterrelianceon rule- formationa personwould need to processusing a par-
based and analytic categorizationprocesses. ticularrule, and the opportunityto engagein a careful
466 THE JOURNAL OF CONSUMER RESEARCH

analysisof individual featuresmay be key factors de- by no means exhaustive of those factors likely to
terminingthe use of alternativeprocesses.Conditions heightenthe chances of people using one type of cate-
of viewingmay be less than ideal, and distractionsmay gorization process or another (others have been de-
be present-both potentiallyleadingto greaterreliance scribedin Homa 1984 and Alba and Hutchinson 1987).
on nonanalytic processing. On the other hand, cues As a startingpoint, however, researchersinterestedin
presentin the judgmentsetting(e.g., point-of-purchase studyingconsumers'categorizationprocessesmay want
displays)can serveto prime eitherspecificaspects/rules to considerthe choice of "level"of such variables,since
or storedexemplarsand increasethe likelihood of one these are likely to have a strong impact on a respon-
type of comparisonover another. dent's processingstrategy.If, for example, one is inter-
Of particularimportanceis the accessibilityof alter- ested in exploringthe use of exemplarprocessing,cat-
native categories.If two or more categoriesare salient, egory learningshould certainlyoccur in a context that
categorizationper se becomes a focal problem, and a allows in-depth exposureto individual instances, and
more deliberativecomparativeprocessis likely (rather thejudgmentcontext shouldavoid a stresson precision
than a more automatic assignment as soon as a good in assigninginstancesto contrastcategories.If the ob-
fit is foundto a categoryexemplar).This type of activity jective is to understandwhen various categorization
is more likely to involve the use of a rule. Increasing processesare likely to be adoptedby consumers,it may
the salience of multiple categoriesalso contributesto be usefulto developa betterunderstandingof potential
the constructionof "rationalsettings"in that multiply purchasesituationsand otherjudgmentsettings(as well
salient categoriesseem to evoke a rational, problem- as conditions of knowledgeacquisition)in terms of the
solvingorientationas well as the use of analyticalstrat- types of factorsjust discussed.
egies (especiallywhen these are well learned and nor-
matively appropriate-typical laboratoryexperiments RESEARCH PARADIGMS
being a case in point). The categorizationprocess is
likely to be quite differentdepending on whether the In orderto choose among alternativeinterpretations
settingemphasizesa rangeof behaviors/judgmentsper- of categorizationprocesses it is necessaryto develop
taining to a focal targetobject or instead increasesthe researchparadigmsthat are sensitive to the retrievalof
salienceof the categoryitself or categoryinstances(e.g., exemplar versus merely feature-levelinformation. In
exemplars,an ideal prototype).In the formercase, the addition, such researchparadigmsshould be appropri-
distinctiveattributesof the targetwill tend to be given ate to the types of phenomenaand levels of psycholog-
more weightsince it (ratherthan the category)becomes ical functioningcharacteristicof consumerbehaviorif,
the point of reference(see Tversky1977)in any retrieval indeed,we wish to examinethe conditionsunderwhich
or matchingprocess-culminating perhapsin the iden- these presumablyflexibleresponsemodes operate.Ac-
tificationof a comparablecategoryexemplarthat shares cordingly,this section emphasizes methodologicalas-
the target'smost prominentfeatures.Increasingthe sa- pects of researchto distinguishstrictlyrule-basedpro-
lienceof a categorymay haveverydifferenteffects(since cessing (i.e., analytic)from exemplar-basedprocessing
the category becomes the referent) depending on (i.e., initialaccessof storedexemplarsfollowedby either
whethera feature-basedrule is readilyaccessible.Cat- an overall or feature-basedcomparison process). The
egorizationexperiments(and settings) can thus easily two major research paradigmsand their operational
tip the balance in the direction of a particularmodel predictions(i.e., ways of identifyingparticularmodels)
as follows: will be discussednext.
P5: Conditionsthat reducethe availabilityof cog-
nitive resourcesin a given context (e.g., time The CategoryJudgmentParadigm
pressure,distractions,etc.) lead to greaterre-
liance on exemplar-basedand nonanalytic The identification of categorization strategies has
processing. been typically undertaken using a concept acquisition
research design (Bruner et al. 1956; Kemler Nelson
P6: Increasingthe accessibility of previously ex- 1984; Posner and Keele 1968). In these experiments,
periencedexemplars(prototypes)in memory subjects categorizestimuli and receive feedback (e.g.,
leads to greater reliance on exemplar-based typicallywhetherthey are "correct"or not). Following
(prototype) processing. Similarly, rule-ana- this, they are tested (without feedback) on another
lytic processing is encouragedby priming a stimulusset comprisingboth new and old (i.e., learning
storedrule or retrievalof features. set) instances. Differentaspectsof performanceon the
P7: Increasingthe salience of multiple categories new instancesformthe basisfor inferenceregardingthe
(i.e., those that are potential "fits to a target categorizationprocessesadopted.The generalpremise
item") leadsto greaterrelianceon rule-based, of this paradigmis that exemplar-based,nonanalytic
analytic processing. processingand rule-based,analytic processingshould
be differentiallysensitive to manipulations of overall
The precedingset of factorsis intended to illustrate similarity between a test item and specific category
the contingent nature of categorizationprocesses.It is learningitems. Since an exemplar-basedprocessentails
ALTERNATIVEMODELS OF CATEGORIZATION 467

a comparison between the categorization object and in- EXHIBIT2


dividual instances stored in memory, its adoption THE CATEGORY-JUDGMENTPARADIGM
should lead to responses that are more sensitive to vari-
ations in overall similarity than responses arising out Features
of a process that involves the use of an attribute-based Category
rule. Likewise, rule-based processing should exhibit members Fl F2 F3 F4
more sensitivity (in specific categorization tasks) when
Learning Stimuli
test items differ in terms of their "match" to a specific Category X
rule, even if they are equivalent in terms of overall sim- 1 0 0 0 0
ilarity to a specific set of exemplars. Thus, one may 2 0 1 0 0
examine categorization performance on specially se- 3 0 0 1 0
lected test items, chosen to reflect different patterns of 4 0 0 0 1
rule and/or exemplar similarity characteristics, to infer Category Y
whether the predicted processing differences were in- 1 1 1 1 1
deed operative. 2 1 0 1 1
3 1 1 0 1
Individual studies vary in stimulus construction and 4 1 1 1 0
other details, depending on their respective goals. Ex-
hibit 2 is the sketch of a typical use of such a paradigm. Test (Transfer) Stimuli
As depicted, the learning stimuli consist of two cate- a 0 1 1 1
gories (X and Y) with four members in each. Each b 1 0 0 0
member in both of the categories is described in terms Predictions
of four category-relevant attributes (F1 to F4), with 0
and 1 representing two different values on each of the Category Y - -
features. Thus, while F 1 always has a value 0 for all
four members of category X, its value is 1 for all four Category X - -
members of category Y. Also, each member of X has
more 0's than l's in its profile. The reverse is true for (a) (b)
members of Y. The test stimuli are also described in
terms of profiles consisting of the same four attributes. NOTE: Analytic processing.
---- Nonanalytic processing.
Given such configurations of learning and test stimuli,
the specific predictions of each model are as follows. If
respondents categorize test item a as belonging in cat-
egory X and b as belonging in Y, then the basis of their categorization process. First, respondents could have
judgments is probably the defining value of Fl. Such induced or discovered the feature-based rule used to
an outcome is believed to reflect an analytic categori- create the similar/dissimilar configurations and em-
zation process. If, on the other hand, respondents clas- ployed it in an analytic fashion in making their judg-
sify a as belonging in Y and b as belonging in X, then ments. Alternatively, respondents might have relied
the process is termed nonanalytic because it reflects upon the autonomous item-level impressions of the
sensitivity to the entire profile rather than to a critical category members and used a nonanalytic matching
value on a single attribute. Obviously, the paradigm mechanism to make their judgments. Since judgment
can be extended to allow for the use of a multiple-feature outcome patterns in almost all cases may be made to
rule through the use of a larger number of stimuli. "fit" potential feature-based rules, data from this par-
It does not, however, directly address the question of adigm are not likely to be entirely convincing for pur-
what type of category representation was retrieved from poses of identifying specific categorization processes.
memory (i.e., rule versus exemplar/prototype) and then Put another way, very different processes may yield the
used in the categorization judgment. In addition, while same categorization result (though related outcomes
allowing one to choose with some confidence between and inferences may differ in important respects).
some alternative analytic models (especially specific Among other issues related to this paradigm is the
forms of single-cue and/or multi-cue models), the par- question of ecological validity. Stimulus sets are quite
adigm is less successful in differentiating between a artificial and are created specifically to meet a relatively
nonanalytic process and a number of competing ana- narrow range of theoretical considerations. Hence, ex-
lytic models. To illustrate, note (from Exhibit 2) that ploring the full range of the categorization phenomenon
overall similarity is operationalized in terms of certain and important contextual factors may be problematic.
configurations of features that are used to describe cat-
egory members. Subsequently, if subjects in certain The VerificationAccuracyParadigm
conditions demonstrate sensitivity towards the pres-
ence/absence of such configurations (say the "nonan- In the verification accuracy paradigm, subjects ini-
alytic pattern" in Exhibit 2 is found), two alternative tially learn about a category of products and are sub-
explanations are confounded in any description of the sequently asked to determine whether or not new in-
468 THE JOURNAL OF CONSUMER RESEARCH

stances would fit into this learned category. The initial Accordingly, the memory advantage in feature ver-
learning conditions vary the form of categorical knowl- ification that results from further reliance on a (learning
edge acquired by subjects. Some subjects are only ex- set) exemplar when categorizing new instances is pre-
posed to the set of learning stimuli while others are also dicted to lead to differential accuracy over rule-based
provided with a feature-based rule. Equivalent learning processing in the following cases: (1) when a to-be-ver-
of the category exemplars is maintained for all subjects. ified aspect of a category exemplar is inconsistentwith
The major difference between the two research para- its category designation, but the feature hadindeed been
digms is that the verification accuracy paradigm differ- associated with the exemplar, and (2) when a to-be-
entiates between categorization mechanisms in terms verified aspect is consistentwith its category designation
of the strength of memory traces that are implied by but it had not, in fact, been associated with the exem-
the underlying information processing requirements of plar. For the remaining aspects, both memory-based
each. and purely categorical inference would lead to similar
An exemplar-based categorization process involves "correct" verification responses. This, then, provides
a comparison between the to-be-categorized new in- one fundamental basis for demonstrating the use of ex-
stance and members of a previously categorized (i.e., emplars in categorization. A second prediction is that
learning) set. The key, then, is that in rule-based cate- idiosyncratic (i.e., non-rule-relevant) features of ex-
gorization no retrieval of these stored exemplars is as- emplars should display considerable improvement in
sumed. Thus, when compared to rule-based processing, verification accuracy as a function of their (automatic)
exemplar-based processing should lead to a memory rehearsal if exemplars are used during subsequent cat-
advantage for subsequent recall of previously learned egorization of new instances.
instances of a category (based on greater recency of ac- Enhanced memorability (resulting from activation
tivation). This advantage should extend to all salient of an exemplar or exemplars during categorization of
features of an exemplar when it is used in a subsequent a test item) should result in faster retrieval in addition
effort to categorize a target object. Say, however, that to higher accuracy in verification of the retrieved ex-
an exemplar was not used, and the person had instead emplars' predicted aspects. As such, reaction time pre-
used a feature-based rule to categorize the target object. dictions should not only parallel verification accuracy
Then there should have been no retrieval of particular patterns, but potentially provide an even more sensitive
exemplars and no beneficial effect on subsequent recall measure of exemplar retrieval as a way of identifying
of information uniquely linked to any particular ex- categorization processes.7
emplar. Let us work through an example to see what The major advantage of the verification accuracy
memory differences should result. paradigm is that it allows one to differentiate among a
Say an individual (following categorization of a new larger set of potential processes. By systematically de-
instance) was asked to verify whether a certain "luxury" lineating the processing assumptions underlying each
feature had been included in a particular model (i.e., model and creating appropriate verification items, one
exemplar) of a previouslyencountered appliance. S/he can rely on memory measures for processing activity
could either scan memory in an attempt to recall such to infer particular strategies. Categorization judgment
a feature or make an inference based on the "category data, per se, are not as revealing, since different pro-
label" s/he had earlier assigned to the model.6 So if the cessing operations may lead to a similar categorization
earlier model (i.e., exemplar) had been included in a outcome.
category of "economy" appliances, the person might Unfortunately, however, the systematic variance in
infer "no," based on the aspect-category incongruence. verification accuracy data may be small, making it dif-
Which of the two processes-memory-based feature re- ficult to detect a difference in processing operations. If
call or category-based inference-would dominate in past exemplars of the category are well learned (through
verifying a feature's presence in a previously seen cat- repeated usage or memorization in the context of a
egory member? The answer would seem to depend on study), then memory advantages arising out of a single
the strength of the memory trace for the category mem- instance of exemplar retrieval (that during categoriza-
ber and feature, on one hand, and the confidence one tion of a test item) may not be substantial enough to
has in the placement of the object in a category plus lead to significant accuracy or time differences. Such
the "cue validity" or prototypicality of the feature on differences may reflect a nonequivalence of prior
the other. knowledge of category exemplars in studies that do not
utilize artificial (or "new") instances.
One potential source of experimental demand may
6Category-based inferenceis a potentiallypowerfulheuristicin this be unwanted special attention to rule-relevant aspects
or any otherjudgmentsetting.Indeed,muchof the functionalbenefit
accruingto a personas a resultof categorizingobjects,people, and
eventslies in his/herresultingabilityto respondquicklyandefficiently 7A lengthy discussionof alternativetechniquesfor assessingthe
when "educatedguesses"are necessary(i.e., the "correctanswer" strengthof memorytracesis beyond the scope of this paper.An ex-
cannotbe knownor is not available).Unfortunately,relianceon such tremely helpfuldiscussionof recall, recognition,and reaction-time
inferencescan produceerrorsin judgments(e.g., halo effects,stereo- proceduresas appliedto the domainof socialcognitioncan be found
typing). in Srull 1984.
ALTERNATIVEMODELS OF CATEGORIZATION 469

(and "nonattention"to backgroundaspects)on the part be seen by recallingthe example discussedin conjunc-
of subjectsif they are providedwith a category-defining tion with Exhibit 2. We noted that a categoryassign-
rulepriorto being exposedto exemplarsfromthe target ment that might appearto result from a nonanalytic
category or categories. Precautions for this include processmightin fact resultfromthe use of a more com-
postponement of rule information until exemplar plex analyticrule (say, one involving all four features).
learninghas been completed, since it is essential that In the latter case, however, respondentsusing such a
initial acquisition of categoryinformation be equiva- strategyshould fare no betterthan those employingan
lent. analyticrulebasedsolelyon F I in verifyingthe presence
Several other measures can be used to supplement or absence of the category-relevantfeatures (i.e., Fl-
the key indices of both the categorization-and verifi- F4) or the category-irrelevantaspects of the set of ex-
cation-basedparadigms.In makinga categoryjudgment emplarsencounteredin the learning phase. Yet if ex-
for a new instance,one's perceivedsuccessin beingable emplar-basedprocessingmechanisms had been oper-
to execute a chosen strategy(i.e., accessingappropriate ative, then there should be enhancedmemorabilityfor
memoryrepresentationsand performingnecessaryop-_ these featuresof specificlearningset exemplars,result-
erations) should be reflectedin his/her level of confi- ing in higheraccuracy,fasterretrieval,and greatercon-
dence in that judgment. For a rule-analyticprocessor, fidence.Thoughthis is more speculative,one difference
confidence should be a function of prior expertise in between exemplar analytic mechanisms (i.e., a test
rule usage and the relative ease with which the latter stimulusis comparedto a storedexemplaron a feature-
may be applied vis-a-vis a given stimulus set and in a by-featurebasis) and exemplar nonanalytic processes
givenjudgment setting.The extent of ambiguityof the maybe seen in the somewhatgreatermemoryadvantage
target stimulus, perhapsrelating to the proportion of the former would give to shared, ratherthan unique,
featuresit shareswith the exemplars/prototypesfrom featuresof the exemplar.This patternshould be most
two or more alternativecategories,should also lead to evident in the more subtle responsetime measures.
differentfeelingsof confidence.For an exemplar-based, The researchstrategyjust introducedalso appearsto
nonanalyticprocessor,confidenceshouldbe a function be useful in accommodatingdifferentresearchobjec-
of the relative availability of past exemplarsthat are tives. If the objective is to understandwhen each cat-
similarin an overallsense to the test item/s. To explore egorizationstrategyis likely to be adoptedby consum-
processingdifferences,stimulus sets (e.g., product as- ers, it may be helpfulto develop a generaltaxonomy of
sortments)may be created that reflect differentlevels potentialcategorylearningand productevaluationsit-
of ease of featureidentificationand ambiguityand/or uationsin termsof contingentfactorsdiscussedearlier.
overallsimilaritybetween to-be-categorizeditems and If, for example,categoryformationoccursin a discrim-
trainingexemplars. ination learningcontext (such as throughcataloguead-
Confidencein verifyingthe presence or absence of vertisementsof brands grouped by specific classifica-
specific aspects of past exemplars,on the other hand, tions or contiguous displays of two or more different
shouldbe a functionof the memorabilityand respective brandlines), and an unfamiliarbrand is subsequently
cue validities of each specific aspect. Communication encountered,both exemplarand rule-basedprocessing
aspects (e.g., prominence, attentional aspects) may, are possible;thus judgment task and involvement fac-
therefore,be quite important.Theoretically,the overall tors are likely to be critical. Such factors could be in-
pattern of verification confidence ratings in a given corporatedinto the researchdesign. If the researcher's
study should parallelthat of verificationaccuracies. objectiveis ratherto documentdifferentcategorization
Following categorizationof new instances, respon- strategies,then "levels"of the contingentfactorsshould
dents may be asked to describetheir mental processes. be chosen so that it is reasonablefor subjectsto adopt
Eventhoughthe level of consciousawarenessassociated each strategy.
with suchjudgmentsremainan issue, to the extent that
theseare somewhatavailableto consciousintrospection IMPLICATIONS AND FUTURE
they may provide supplementaryevidence, and at the DIRECTIONS
very least, expose proceduralproblems(e.g., unantici-
pated rule inductions). Collecting concurrent self-re- In this final section, we returnto a broaderconsid-
portsduringcategoryjudgmentsis stronglydiscouraged eration of the categorizationperspectiveto summarize
on grounds of interferencewith the mental processes some of its majorimplicationsfor consumerbehavior.
of interest. Either free-format elicitation or forced It is difficultto talk meaningfullyabout product class
choices between severalpotential process descriptions boundaries,evoked sets, product substitutability,and
may be collected at little additional cost following the relatedways of describingbetween-product(or brand)
judgment/s. homogeneityof responsewithoutrecognizingthat these
Whileno studyreportedin the literaturehas yet done ideas are intimately linked to categorization.
so, the two researchparadigmsmay be usefully com- One failingof manycontemporaryviewsof consumer
bined within a single experimentalframeworkby fol- behavioris that they focus to such a degree on the re-
lowing a categorizationtask by the verificationproce- lationship (and often the process underlyingthe rela-
duresdiscussedabove. The advantagesof doing so may tionship) between a consumer and a single, specific
470 THE JOURNAL OF CONSUMER RESEARCH

consumptionobject that they tend to ignore the struc- birds). Social psychologists as well as consumer re-
turalcontextwithinwhich this relationshipexists. Such searchers have long recognized the importance of per-
a "one-on-one"perspective-whetherin termsof brand ceiving an instance as a member of a general class in
beliefs, evaluation, advertisingresponse, or overt be- terms of subsequent judgments and inferences (e.g.,
havior (e.g., brand loyalty)-is artificial. Consumers stereotyping, price-quality inferences).8 Categorization
rarelyconceive of a product in isolation, much to the research in both person perception (e.g., Cantor 1982;
dismayof wishful-thinkingbrandmanagerswho would Fazio 1986; Fiske 1982; Fiske and Pavelchak 1986) and
preferthat a givenbrandoccupycenterstageall to itself. product perception has the potential to greatly sharpen
In contrast, the categorizationperspectiveprovides a our understanding of the interplay between "purely
way of thinking about the relationshipbetween a con- cognitive" and affective processes. In consumer behav-
sumerand an individuallydefinedassortmentof prod- ior, for example, affectively neutral categories may be
ucts. In doing so it recognizesthe importanceof both the exception: many categories may be goal-related.
instrumentalaspects(i.e., relatedprimarilyto product Further, if the assumption of a rule-like judgment
use and function) and expressivefactors,whethermar- process (e.g., multiattribute model) is in fact incorrect
keter-initiatedor basedon generalor personalizedgoals in many situations, our views of how attitudes are
and desires. At the individual level, then, the catego- formed or changed may need to be modified. While an
rizationperspectiveemphasizesthe viewthatconsumers analytically based categorization model would also
tend to organize their belief systems about products suggest that favorable evaluation follows from the per-
around category-relevantfactors (e.g., consumption ceived presence of features that place an object in a
goals, productfunctions, common properties,clear or positively evaluated category, an exemplar-based model
idealinstances)ratherthana set of singlebrandattitudes might stress perceived similarity to an exemplar or pro-
or beliefs. totype to accomplish a similar purpose. Since catego-
The researchagendafollowingfrom this orientation rization produces a reduction in uncertainty, positive
is quite varied. We have very little systematic knowl- affect may result from a successful fit and negative affect
edge, for example, about the respectiveinfluences on from an inability to categorize an item-particularly if
productcategoryformationand change.There is not a the resulting judgments or inferences are important.
grevatdeal of researchon how productcategoriesdevelop When a consumer has little experience with a product,
in consumers'minds despitethe fact that this may well for example, being able to categorize it with products
be a more fundamentallevel at which to look at the that are familiar may permit a set of important infer-
organizationof productknowledge.For example,brand ences to be made. How such categorization will be ac-
or product salience or accessibilityin memory, while complished may depend importantly on the degree of
interesting in its own right, may largely be a conse- product complexity and the consumer's background
quence of categorystructureonce we escape the tran- knowledge. If preattentive, automatic processes can
sitoryeffectsof experimentalmanipulations.It may thus create a unitized representation, overall similarity-based
be more importantto take a carefullook at these more processes become more likely. If, however, increasing
fundamentalissuesand processes.Thereis the realpos- complexity or a lack of product knowledge make it dif-
sibility, for example, that the consumer environment ficult to combine feature-level information using au-
favorscategorydefinitionin termsof specificexemplars tomatic processes, more analytical treatment of indi-
ratherthan category-definingfeatures(at least for cer- vidual attributes becomes necessary. Since, however,
tain types of products),and this has considerableim- this may also be more effortful (particularly under time
plicationsfor evoked-setformationand change.For the constraints) there may be a tendency to focus on a re-
latter,productpresentationstrategiesthat have a direct duced number of features, perhaps those that are most
impact at early stagesof productperception(e.g., that prominent, which may lead to errors of categorization
increase the likelihood of perceived similarityto par- and suboptimal product choice.
ticular exemplars)are likely to be of considerableim- Many comparison ads seem to be based on the prem-
portance.Conditionsunderwhich consumersare likely
to createdifferentiatedsubcategoriesof productsalmost
8Aconsiderableamountof researchhasexaminedthe consequences
certainly would benefit from closer scrutiny. This is of makingvariouscategoriesmoreaccessibleto people.Situationally
likely to have considerablepracticalsignificancesince inducedprimingeffects,for example,arewell knownto increasecat-
so much marketingeffortis directedat producingthis egoryaccessibility.Whenattemptingto generalizefromthesestudies,
type of effect. however,it is importantto recall that such effectstend to be short
One of the most interestingfeatures of future con- lived.Also, as Herr,Sherman,and Fazio(1983)point out, sucheffects
arenot robustin the absenceof ambiguousstimuli.Indeed,marketers
sumer categorizationresearchis likely to be its focus may wish to note with caution that increasingcategoryaccessibility
on both evaluativecategoriesand the link between ob- may even producea contrasteffect when the product that follows
ject identificationand evaluation. The traditionallit- enhancedcategoryaccessibilityis not ambiguousand also seems to
eratureon categorization,much of it springingdirectly be a poor match with the category.In that event, some judgments
may actuallybe displacedfrom the category.Finally, we ignore at
from researchon concept learning, has tended to use considerablerisk the probabilitythat initial accessibility-likeeffects
affectivelyneutral stimuli (e.g., geometric forms, arti- on categorizationwill give way to more deliberativecategorization
ficiallanguage)or generalclassesof objects(e.g., tables, judgmentsas informationprocessingprogresses.
ALTERNATIVEMODELS OF CATEGORIZATION 471

ise that simply presentinga less well-knownbrandto- Bargh, John A. (1984), "Automatic and Conscious Processing
getherwith a brand that is favorablyevaluated(i.e., a of Social Information," in Handbook of Social Cognition,
memberof a desiredsubcategory)leads to not only fa- Vol. 3, eds. Robert S. Wyer, Jr. and Thomas K. Srull,
vorableattitudes toward the former but similar infer- Hillsdale, NJ: Erlbaum, 1-43.
ences on unmentioned features.The implication here Barsalou, Lawrence W. (1983), "Ad Hoc Categories," Memory
& Cognition, 11 (May), 211-227.
is thatthe assignmentof affectto a new instance,which Brooks, Lee (1978), "Nonanalytic Concept Formation and
has been viewed as a function of some type of feature- Memory for Instances," in Cognition and Categorization,
basedcognitive algebraby traditionalattitude models, eds. Eleanor Rosch and Barbara B. Lloyd, Hillsdale, NJ:
may alternativelyresultfrom exemplar-basedcompar- Erlbaum, 169-211.
ison processes.By relatingthe assignmentof affect to Bruner, Jerome S., Jacqueline J. Goodnow, and George A.
morefundamentalcategorizationprocesseswe are also Austin (1956), A Study of Thinking, New York: Wiley.
more likely to anticipate the type of inferences that Cantor, Nancy (1982), "A Cognitive-Social Approach to Per-
consumersuse to fill in "informationgaps"when direct sonality," in Personality, Cognition and Social Interac-
product knowledge is absent or when the motivation tion, eds. Nancy Cantor and John F. Kihlstrom, Hillsdale,
to seek such information is low. Obviously, this may NJ: Erlbaum, 23-44.
not always result in positive outcomes for consumers and Walter Mischel (1979), "Prototypes in Person
Perception," in Advances in Experimental Social Psy-
to the extent that such inferences are incorrect. Ex- chology, Vol. 12, ed. Leonard Berkowitz, New York: Ac-
emplar-based(and possibly nonanalytical) processes ademic Press, 3-52.
may also provide some insight into affect-transferphe- Cohen, Joel B. (1982), "The Role of Affect in Categorization:
nomena that have previouslybeen discussed in terms Towards a Reconsideration of the Concept of Attitude,"
of classicalconditioningand generalizationor, simply, in Advances in Consumer Research, Vol. 9, ed. Andrew
"affectreferral"(Wright1975), particularlyas they oc- A. Mitchell, Ann Arbor, MI: Association for Consumer
cur in low involvement settings. Research, 94-100.
Finally, while describingand attempting to under- Dominowski, Roger L. (1974), "How Do People Discover
stand consumer decision making in terms of funda- Concepts?" in Theories in Cognitive Psychology: The
Loyola Symposium, New York: Wiley, 257-288.
mental information processingconcepts has attracted Elio, Renee and John R. Anderson (1981), "The Effects of
numerousresearchers,there appearsto be growingdis- Category Generalizations and Instance Similarity on
enchantmentwith what some would describeas "overly Schema Abstraction," Journal of Experimental Psy-
cognitiveapproaches."Some of this criticismseems to chology: Human Learning and Memory, 7 (November),
be -directedat what is seen as mechanistic and "com- 397-417.
puter-like"assumptions that underlie many of these Fazio, Russell H. (1986), "How do Attitudes Guide Behav-
approaches.While it would clearly be unwise to turn ior?" in Handbook of Motivation and Cognition, eds.
away from the analyticaland networktypes of models Richard M. Sorrentino and E. Tory Higgins, New York:
that have proven so useful, it might be equally unwise Guilford, 204-243.
to ignorethe possibilityof developingmore integrated Fiske, Susan T. (1982), "Schema-Triggered Affect: Applica-
tions to Social Perception," in Affect and Cognition: The
and holistic accounts of consumers'judgmentsand be- 17th Annual Carnegie Symposium on Cognition, eds.
havior.Categorizationmay be a good meeting ground Margaret S. Clark and Susan T. Fiske, Hillsdale, NJ: Erl-
for researcherswho are seriouslyinterestedin consumer baum, 55-78.
decision making and in exploringthe nature and role and Mark A. Pavelchak (1986), "Category-Based vs.
of self-definedproduct categories (including category Piecemeal-Based Affective Responses: Developments in
formation, assignment of instances to categories,and Schema-Triggered Affect," in Handbook of Motivation
category-based inferenceprocesses)fromboth analytical and Cognition, eds. Richard M. Sorrentino and E. Tory
and nonanalyticalperspectives.We believe that cate- Higgins, New York: Guilford, 167-203.
gorization is a topic of considerable conceptual and Garner, Wendell R. (1978), "Aspects of a Stimulus: Features,
practicalimportancewithin the field of consumer be- Dimensions, and Configurations," in Cognition and
Categorization, eds. Eleanor Rosch and Barbara B.
havior, and we hope that this article will be a spur to Lloyd, Hillsdale, NJ: Erlbaum, 99-133.
researchin this area. Gilovich, Thomas (1981), "Seeing the Past in the Present:
The Effect of Associations to Familiar Events on Judg-
[ReceivedMay 1985. Revised October1986.] ments and Decisions," Journal of Personality and Social
Psychology, 40 (May), 797-808.
Herr, Paul M., Steven J. Sherman, and Russell H. Fazio
REFERENCES (1983), "On the Consequences of Priming: Assimilation
Alba, Joseph W. and J. Wesley Hutchinson (1985), "A and Contrast Effects," Journal of Experimental Social
Framework for Understanding Consumer Knowledge: Psychology, 19, 323-340.
II. Comparison and Inference Processes," working paper, Hintzman, Douglas L. and Genevieve Ludlum (1980), "Dif-
Center for Consumer Research, University of Florida, ferential Forgetting of Prototypes and Old Instances:
Gainesville, FL 3261 1. Simulation by an Exemplar-Based Classification Model,"
and J. Wesley Hutchinson (1987), "Dimensions of Memory and Cognition, 8 (July), 378-382.
Consumer Expertise," Journal of Consumer Research, Holyoak, Keith J. and Arnold L. Glass (1975), "The Role of
13 (March), 411-454. Contradictions and Counter-examples in the Rejection
472 THE JOURNAL OF CONSUMER RESEARCH

of False Sentences," Journal of Verbal Learning and rability in Classification Learning," Journal of Experi-
Verbal Behavior, 14 (April), 215-239. mental Psychology: Human Learning and Memory, 7
Homa, Donald (1984), "On The Nature of Categories," in (September), 355-368.
The Psychology ofLearning and Motivation, Vol. 18, ed. Mervis, Carolyn B. and Eleanor Rosch (1981), "Categoriza-
Gordon H. Bower, New York: Academic Press, 49-94. tion of Natural Objects," in Annual Review of Psychol-
and Richard Vosburgh (1976), "Category Breadth and ogy, Vol. 32, eds. Mark R. Rosenzweig and Lyman W.
the Abstraction of Prototypical Information," Journal Porter, Palo Alto, CA: Annual Reviews, Inc., 89-115.
of Experimental Psychology: Human Learning and Posner, Michael I. and Steven W. Keele (1968), "On the Gen-
Memory, 2 (May), 322-330. esis of Abstract Ideas," Journal of Experimental Psy-
Hutchinson, J. Wesley and Joseph W. Alba (1985), "A chology, 77 (July), 353-363.
Framework for Understanding Consumer Knowledge: I. and Charles R.R. Snyder (1975), "Attention and Cog-
Structural Aspects and Information Acquisition," work- nitive Control," in Information Processing and Cogni-
ing paper, Center for Consumer Research, University of tion: The Loyola Symposium, ed. Robert L. Solso, Hills-
Florida, Gainesville, FL 32611. dale, NJ: Erlbaum, 55-85.
Kahneman, Daniel and Dale T. Miller (1986), "Norm Theory: Read, Stephen J. (1983), "Once is Enough: Causal Reasoning
Comparing Reality to its Alternatives," Psychological from a Single Instance," Journal ofPersonality and Social
Review, 93 (2), 136-153. Psychology, 45 (August), 323-334.
, Paul Slovic, and Amos Tversky, eds. (1982), Judg- Reed, Stephen K. (1972), "Pattern Recognition and Cate-
ment Under Uncertainty. Heuristics and Biases, Cam- gorization," Cognitive Psychology, 3 (July), 382-407.
bridge: Cambridge University Press. Rips, Lance J., Edward J. Shoben, and Edward E. Smith
Kemler Nelson, Deborah G. (1984), "The Effect of Intention (1973), "Semantic Distance and the Verification of Se-
on What Concepts are Acquired," Journal of Verbal mantic Relations," Journal of Verbal Learning and Ver-
Learning and Verbal Behavior, 23 (December), 734-759. bal Behavior, 12 (February), 1-20.
Kossan, Nancy E. (1981), "Developmental Differences in Rosch, Eleanor (1973), "On the Internal Structure of Per-
Concept Acquisition Strategies," Child Development, 52 ceptual and Semantic Categories," in Cognitive Devel-
(March), 290-298. opment and the Acquisition of Language, ed. Timothy
Krueger, Lester E. (1973), "Effect of Irrelevant Surrounding E. Moore, New York: Academic Press, 111-144.
Material on Speed of Same-Different Judgment of Two and Carolyn B. Mervis (1975), "Family Resemblances:
Adjacent Letters," Journal of Experimental Psychology, Studies in the Internal Structure of Categories," Cognitive
98,.252-259. Psychology, 7 (October), 573-605.
Lastovicka, John L. (1979), "Questioning the Concept of In- Smith, Edward E. and Douglas L. Medin (1981), Categories
volvement Defined Product Classes" in Advances in and Concepts, Cambridge, MA: Harvard University
Consumer Research, Vol. 6, ed. William L. Wilkie, Ann Press.
Arbor, MI: Association for Consumer Research, 174- Smith, J. David and Deborah G. Kemler Nelson (1984),
179. "Overall Similarity in Adults' Classification: The Child
Lingle, John H., Mark W. Altom, and Douglas L. Medin in All of Us," Journal of Experimental Psychology: Gen-
(1984), "Of Cabbages and Kings: Assessing the Extendi- eral, 113 (1), 137-139.
bility of Natural Object Concept Models to Social Smith, Linda B. (1981), "Importance of the Overall Similarity
Things," in Handbook of Social Cognition, Vol. 1, eds. of Objects for Adults' and Children's Classifications,"
Robert S. Wyer and Thomas K. Srull, Hillsdale, NJ: Erl- Journal of Experimental Psychology: Human Perception
baum, 71-118. and Performance, 7 (August) 811-824.
Lynch, John G. and Thomas K. Srull (1982), "Memory and Srull, Thomas K. (1984), "Methodological Techniques for
Attentional Factors in Consumer Choice: Concepts and the Study of Person Memory and Social Cognition," in
Research Methods," Journal of Consumer Research, 9 Handbook of Social Cognition Vol. 2, eds. Robert S.
(June), 18-37. Wyer, Jr. and Thomas K. Srull, Hillsdale, NJ: Erlbaum,
Mandler, George (1982), "The Structure of Value: Accounting 1-72.
for Taste," in Affect and Cognition: The 17th Annual Sujan, Mita (1985), "Consumer Knowledge: Effects on Eval-
Carnegie Symposium on Cognition, eds. Margaret S. uation Strategies Mediating Consumer Judgments,"
Clark and Susan T. Fiske, Hillsdale, NJ: Erlbaum, 3-36. Journal of Consumer Research, 12 (June), 31-46.
Martin, Randi C. and Alfonso Caramazza (1980), "Classifi- Tversky, Amos (1977), "Features of Similarity," Psychological
cation in Well-Defined and Ill-Defined Categories: Evi- Review, 84 (July), 327-352.
dence for Common Processing Strategies," Journal of and Daniel Kahneman (1973), "Availability: A Heu-
Experimental Psychology: General, 109 (September), ristic for Judging Frequency and Availability," Cognitive
320-353. Psychology, 5 (September), 207-232.
Medin, Douglas L., Mark W. Altom, Stephen M. Edelson, Wright, Peter L. (1975), "Consumer Choice Strategies: Sim-
and Deborah Freko (1982), "Correlated Symptoms and plifying Versus Optimizing," Journal of Marketing Re-
Simulated Medical Classification," Journal of Experi- search, 11, (February), 60-67.
mental Psychology: Learning, Memory and Cognition, 8 Wyer, Robert S. and Thomas K. Srull (1981), "Category Ac-
(January), 37-50. cessibility: Some Theoretical and Empirical Issues Con-
and Marguerite M. Schaffer (1978), "Context Theory cerning the Processing of Social Stimulus Information,"
of Classification Learning," Psychological Review, 85 in Social Cognition: The Ontario Symposium (Vol. 1),
(May), 207-238. eds. E. Tory Higgins, C. Peter Herman, and Mark P.
and Paula J. Schwanenflugel (1981), "Linear Sepa- Zanna, Hillsdale, NJ: Erlbaum, 161-197.

S-ar putea să vă placă și