An investigation of extension and feedback effects

C. WHAN PARK, SUNG YOUL JUN, and ALLAN D. SHOCKER*The authors report two studies investigating the effectiveness of a com-posite brand in a brand extension context. In composite brand extension,a combination of two existing brand names in different positions as head-er and modifier is used as the brand name for a new product (e.g., Slim-Fast chocolate cakemix by Godiva).The results of both studies reveal thatby combining two brands with complementary attribute levels, a compos-ite brand extension appears to have a better attribute profile than a directextension of the header brand (Study 1) and has a better attribute profilewhen it consists of two complementary brands than when it consists oftwo highly favorable but not complementary brands (Study 2). Theimproved attribute profile seems to enhance a composite's effectivenessin influencing consumer choice and preference (Study 2). In addition, thepositions of the constituent brand names in the composite brand nameare found to be important in the formation of the composite's attribute pro-file and its feedback effects on the constituent brands. A composite brandextension has different attribute profiles and feedback effects, dependingon the positions of the constituent brand ite Branding Alliances:An Investigation of Extensionand Feedback EffectsA brand name is a basic element of a firm's market offer-able evaluations of extension products (Aaker and Kellering that aids consumers' understanding of the offering's1990; Bousch and Loken 1991; Broniarczyk and Albacharacteristics. It serves as an encoding and retrieval cue for1994a; Dacin and Smith 1994; Park, Milberg, and Lawsonbrand-related infonnation (Cohen and Basu 1987). It signals1991) and whether extensions result in negative feedback ef-the often intangible product properties that must otherwisefects for the original brand (Keller and Aaker 1992; Lokenbe leamed through experience or taken on faith (Erdemand Roedder John 1993; Park, McCarthy, and Milberg 1993;1993). It also serves as a powerful heuristic cue for evalua-Romeo 1991). Such research has centered primarily on is-tion and choice decisions (Park and Lessig 1981). The brandsues of fit between the brand and extension product cate-name thus can have a significant strategic impact on long-gories (e.g., a Honda lawn mower). Although additionalterm brand performance, going far beyond its role as abranding ideas have been suggested (Farquhar 1989; Park,subelement of the marketing mix (Park, Jaworski, andMcCarthy, and Milberg 1993; Schmitt and Dube 1992), rel-Maclnnis 1986).atively little attention has been paid to other possible brand-Strategic implications of branding strategies have increas-ing strategies that may afford cost efficiency and competi-ingly been addressed in the broader context of brand us research has examined line and brand extensionWe examine one such branding strategy, which involvesdecisions to identify when such decisions will lead to favor-combining two existing brand names to create a compositebrand name for a new product. For instance, in the fictitious*C. Whan Park is the Albert Wesley Frey Distinguished Professor ofexample Slim-Fast chocolate cakemix by Godiva, the twoMarketing and Sung Youl Jun is a visiting assistant professor. University ofPittsburgh. Allan D. Shocker is the Curtis L. Carlson Professor offirms (brands) ally themselves to enter a new product-mar-Marketing. University of Minnesota. The authors thank Deborah Roedderket by sharing manufacturing and marketing expertise. WeJohn (University of Minnesota), Deborah Maclnnis (University of Southernterm that approach a composite brand extension (CBE) strat-California). S. Ratneshwar (University of Connecticut). Dan Smith, Timegy. Current examples include Healthy Choice cereal byHeath (University of Pittsburgh), and three anonymous JMR reviewers forKellogg's and Special K frozen waffles by many helpful comments and suggestions on drafts of this article. Theauthors also acknowledge Cheil Communications (Seoul. Korea) for itsThere are many reasons why firms or brands might formfinancial support of their alliance to market a new product with a CBE er the fictitious Slim-Fast chocolate cakemix by Go-Journal of Marketing Research453Vol. XXXIII (November 1996), 453-466

454JOURNAL OF MARKETING RESEARCH, NOVEMBER 1996diva. Slim-Fast diet food products are currently targeted to amass market and have extensive distribution. Because itsproducts are associated mainly with dieting and weight loss,Slim-Fast may consider cakemix a suitable product for itsextension. It may find, however, that general perceptions ofbland taste in its products are the main obstacle to marketpenetration in the cakemix category. Godiva chocolates'strength in delivering superior taste may highly complementSlim-Fast's products. Association with Godiva may enableSlim-Fast to extend its name to other products with greatersuccess than would be possible by a direct extension or sub-branding. By indirect association of the Slim-Fast namewith Godiva, the original image of Slim-Fast also may beimproved. Godiva, however, because of its appeal to a selectupscale market, limited distribution outlets, and strong iden-tification with boxed chocolates, may not want to market anew cakemix with mass appeal. It may nevertheless want toconvey a healthier connotation (e.g., rich yet healthy) by re-ducing the perception of high fat and calorie content throughassociation with a brand that has the opposite image. In ad-dition, it may seek to generate additional revenue by licens-ing its name to other products. An alliance with Slim-Fastmay help Godiva meet those objectives. Finally, viewing abrand from a signaling perspective (Wemerfelt 1988) sug-gests that two brand names may provide greater assuranceabout product quality than one alone. The presence of thesecond brand name on a product may signal potential buy-ers that another firm is willing to place its reputation on theline. If both firms gain in such a relationship, an alliancemakes an alliance may raise questions about Godiva's brandimage, however, through possible negative feedback effectsdue to the combination of its high status brand with the (low-er status) Slim-Fast name or the association, albeit indirect,with a product catering to a mass market. Customers may beconfused about the combination of seemingly inconsistentcharacteristics of the two brands (Godiva primarily associat-ed with expensive, high-calorie boxed chocolates and Slim-Fast associated with inexpensive, low-calorie diet food).Such confusion could damage the images of both alliances may be the ultimate form of cooperationbetween two firms in the sense that they make the relationhighly visible and the firms possibly staJce their reputationson the outcome. The potential benefits and risks of brand al-liances therefore must be identified and examined both the-oretically and empirically. Our limited knowledge of thebenefits and risks may be leading to inappropriate er, though favorable consumer evaluations can beencouraged by advertisements explaining the bases for fitbetween a parent brand and its extension (Chakravarti,Maclnnis, and Nakamoto 1990), the effectiveness may beenhanced when the brand name itself also conveys informa-tion about the fit. In view of the powerful role of consumers'intuitive beliefs in attribute inference making (Broniarczykand Alba 1994b), composite branding may serve such pur-poses study the potential impact of brand alliances on con-sumers, we conduct a series of pretests and two main stud-ies. In Study 1, we examine the CBE in terms of (1) whetherrecent theoretical developments (e.g., concept combinationtheory) could be useful in discovering how consumers formthe concept of a composite brand from the concepts of itsconstituents, (2) the respective roles of each constituent(partner) brand in consumers' formation of a composite con-cept, and (3) the effectiveness of the CBE strategy relative toa direct extension in fostering favorable extension evalua-tion and in generating positive feedback for the constituentbrands. In Study 2, we explore the relative contributions oftwo potentially important variables for the effectiveness ofthe CBE strategy: attitudinal favorability toward the modifi-er brand and attribute-level 1TheoryA brand can be understood in terms of a set of attributes,each at particular performance levels, but how does a con-sumer form a composite concept from two or more brands?The composite concept literature suggests two models toaccount for the composition process: the selective modifica-tion model (Smith et al. 1988) and the concept specializationmodel (Cohen and Murphy 1984; Hampton 1987; Murphy1988). The former model was developed in the context ofadjective-noun conjunctions (e.g., red apple), and the latterwas developed in the context of noun-noun , apartment dog). We use the concept specializationmodel because we focus on noun-noun composites andbecause it also incorporates adjective-noun composites (seeMurphy 1988). The relationship between two independentconcepts in the concept specialization model can bedescribed in terms of modifying concept and modified con-cept. The last concept in a noun-noun composite is general-ly the header and the preceding concept acts to modify it(Murphy 1988). For example, in the composite concept"apartment dog," "apartment" is the modifier and "dog" isthe e brands are purposefully created by firms, theirmeaning is affected by the marketing actions of those firmsrather than arising more "naturally." The CBE applicationtherefore requires an adaptation of the composite conceptmodels. First, though determination of header and modifierconcepts is reasonably straightforward in the compositeconcept examples used in the literature, it is not always asclear in the case of a CBE. For the CBE, whether consumerswill interpret the noun preceding a preposition such as by orfrom as a header or as a modifier may not be clear. The rea-son is that the header and modifier in the CBE descriptionmay be determined partly by such criteria as linguistic usageprecedence, the relative salience of the two brands, or thebrand's relative operative control of the product. Moreover,through advertising and packaging, a firm can (at some cost)suggest the relationship it intends, even if that would con-tradict the conclusions a consumer may reach without suchinfonnation. While acknowledging multiple detenninantsfor header and modifier roles, we determine from the pretestresults (see the Method section) that in the context of Study1 the relative ordering determines the header and modifierpositions; the header is the noun preceding and the modifieris the noun following the preposition , an issue can be raised about the degree of pro-cessing equivalence between composite concepts and CBEs,that is, how well the process of interpreting a CBE can beexplained by composite concept theory. The two differsomewhat because a composite concept typically involvestwo concepts (e.g., apartment and dog), whereas a CBE in-

Composite Branding Alliancesvolves three concepts (e.g., Slim-Fast, chocolate cakemix,and Godiva). We propose that interpreting a CBE involvesan intermediate step before the composite concept fonnationprocess: a nested concept formation process ("idiomatic"concept formation, according to Murphy 1988). Schmitt andDube (1992) examine the first stage of nesting concepts inthe brand extension context. That process occurs when asalient attribute of the nested concept assumes the value ofthe same attribute of the nesting concept, because the nest-ing concept has less variability on the attribute in questionthan the nested concept. In the example of Slim-Fast choco-late cakemix by Godiva, we can think of chocolate cakemixas being nested under the Slim-Fast concept, because thechocolate cakemix concept as a general product categoryhas greater variability on the calorie attribute than does thespecific Slim-Fast brand concept (i.e., a specific brand with-in a product category). When Slim-Fast and chocolatecakemix are combined to form the concept of Slim-Fastchocolate cakemix, the value of the calorie attribute in Slim-Fast (i.e., low) thus serves to identify the specific value ofthat attribute for the chocolate cakemix concept. According-ly, the nested concept formation is characterized by a one-way value transfer from the nesting concept to the nestedconcept for salient attributes, but not the a nested concept is formed, it can be combined withanother independent concept by the process of compositeconcept formation. In the composite concept, the header andmodifier concepts have their own attributes that differ fromeach other in salience. The values of those salient attributeshave similar variability and sometimes conflict with eachother (e.g., ocean land). A person may not be able to definethe attributes of the header concept by simply using the val-ues of the modifier concept, because certain attributes in theheader already have some fixed values. For example, sup-pose Slim-Fast chocolate cakemix, a nested concept, is com-bined with Godiva to create Slim-Fast chocolate cakemix byGodiva. A person still may consider the calorie attributesalient. The values of that attribute for Slim-Fast and Godi-va are relatively fixed in opposing directions (low and high,respectively). A person must find a reasonable way to re-solve the conflict and combine the different attribute infor-mation from the header and modifier to understand the com-posite preceding discussion suggests that when nested andindependent concepts are present in a combination, a personis likely to process the nested concept before combining itwith the independent concept. The reason is that the nestedconcept formation process, in general, requires less cognitiveeffort than the composite concept fonnation process. Ac-cordingly, in evaluating Slim-Fast chocolate cakemix by Go-diva, consumers may follow a sequence in which Slim-Fastand chocolate cakemix are processed first to form the nestedconcept, Slim-Fast chocolate cakemix, and then that nestedconcept is modified by Godiva to form a composite , the nature of the relationships between individualconstituent brands and between those brands and the exten-sion product must be operationalized specifically in the caseof the CBE. Although the concept specialization model as-sumes a fit between individual concepts (i.e., there is a basisfor relating the two), it does not address the strategic objec-tives that often underlie brand alliances. The two brandsshould have a high degree of product-level fit with the ex-455tension product to maximize transfer of relevant informationfrom individual constituent brands to the extension product(Aaker and Keller 1990; Park, Milberg, and Lawson 1991).Using a CBE strategy would not make sense if consumersdid not understand the relationship between the characteris-tics of the two brand names and the characteristics of the ex-tension product (i.e., the relationship between the brandnames and the extension product). The two brands alsoshould be complementary to each other in their attributesalience and performance levels. Complementarity is judgedto be present when (1) two brands have a common set of rel-evant (but not necessarily salient) attributes, (2) two brandsdiffer in attribute salience such that attributes not salient toone are salient to the other, and (3) the brand for which theattribute is salient has a higher performance rating on that at-tribute than the brand for which the attribute is not alliance would make sense when two brands are com-plementary in the sense that performance-level strengths andweaknesses of their relevant attributes mesh esis DevelopmentConcept fonnation for the composite brand gh composite concept fonnation is based on the com-bination of header and modifier concepts, the compositemay be different from the sum of its parts (Hampton 1987;Lakoff 1985; Murphy 1988; Murphy and Medin 1985).Hampton (1987) and Murphy (1988) share similar viewsabout nonadditive fonnation of attributes for the composite,but Hampton (1987) specifies the process by which attributelevels for a composite concept are formed: attribute-leveldetermination for a composite concept is based on a unionrule, subject to necessity and impossibility ing to that rule, if an attribute is (not) necessary toboth constituent concepts, it is also (not) necesseuy to thecomposite concept. However, if an attribute (e.g., low calo-rie) is necessary for one concept (e.g., Slim-Fast cakemix)but not the other (e.g., Godiva), a maximum rule applies:The attribute becomes necessary to the composite , Slim-Fast cakemix by Godiva).' If an attribute isjudged logically impossible for one constituent, a minimumrule applies: The value of that attribute of the other con-stituent concept will have no effect on the composite con-cept. For attribute levels in the midrange, the combinatorialformula approximates an averaging rule. The minimum andaveraging rules are straightforward, but the maximum rulerequires further discussion and has important implicationsfor evaluation of the composite ically, research on the nature and categorization ofconcepts suggests that a set of core attributes in a conceptconstitutes its definition and is the most essential and salientset of attributes for understanding that concept (see Eysenckand Keanne 1990; Miller and Johnson-Laird 1976).^ Coreattributes of a concept are thus difficult to change and arehighly influential in changing the meaning of other conceptswhen they are combined. Therefore, attributes highly salientfor the header concept (i.e., its core attributes) shouldremain unaffected by the low salience of those attributes for'To avoid redundancy, we use the term cakemix hereafter to refer tochocolate cakemix.^Although Hampton (1987) uses the term necessity, we use the termsalience hereafter. We assume that an attribute that serves as a necessarycharacteristic for a concept is also salient for the concept.

456JOURNAL OF MARKETING RESEARCH, NOVEMBER 1996the modifier (Murphy 1988; Smith et al. 1988). However, ifan attribute is not salient for the header but is for the modi-fier concept, the latter's attribute salience should influencethe former's upward in the composite concept (Finin 1980;Murphy 1988). That rationale leads to Hjg. Moreover, weexpect the same outcome for perceived attribute perfor-mance levels (Hampton 1987) as for attribute a composite concept takes on salient attributes, eitherfrom the header or the modifier concept, we expect the com-posite concept to approximate the performance levels ofthose attributes (the maximum rule). Specifically, if anattribute is highly salient for the header concept, its per-ceived performance level should not be affected by the per-formance level of the same attribute in the modifier , the performance levels of nonsalient attributes in theheader should be influenced by the performance levels ofthose same attributes when they are salient in the modifierconcept (Thagard 1984). That rationale leads to Hn,.Hi^: When an attribute is highly salient to at least one of the con-H|(,: When either one of the constituent brands is perceived tostituent brands, it is also highly salient to the m well on a highly salient attribute, the CBE also isperceived to perform well on that ential effect of header and modifier brands. Whenthe salience and performance levels of an attribute aregreater for one constituent brand than for the other, theyinfluence the formation of the attribute salience and perfor-mance-level perceptions of the CBE (Hampton 1987).Therefore, it could be argued that the pereeived attributesalience and performance levels of the CBE will be inde-pendent of which brand serves as the header. However, weexpect differential influences on the CBE from header andmodifier brands (i.e., an asymmetric structure; Murphy1988; Smith et al. 1988). According to Hampton (1988) andStorms and colleagues (1993), formation of attribute infor-mation for a composite concept is based more on the domi-nant concept than on the dominated one through what iscalled dominance effect. Because the extension product isidentified with the header brand in the CBE, we expect theheader to be more dominant in influencing understanding ofthe CBE than the modifier brand. The CBE would still beunderstood as a member of the header category. Thus, theformation of attribute information for the CBE would bebased more on the header than on the modifier brandthrough the asymmetric our example of Slim-Fast cakemix by Godiva, the com-posite takes on the most salient attributes of Slim-Fast, theheader brand. It becomes a low-calorie, low-fat, dietcakemix. The taste attribute from Godiva modifies themeaning of Slim-Fast cakemix to produce a meaning of theCBE that is coherent for both constituent brands. Althoughthe taste attribute is salient in Slim-Fast cakemix by Godiva,the CBE's most salient attribute is still its low-calorie, dietattribute. We can think of this CBE as a Godiva-like Slim-Fast cakemix (i.e., still Slim-Fast). Similarly, Godivacakemix by Slim-Fast may be understood initially as aGodiva cakemix, rich in taste and luxury. The low-calorieattribute from the Slim-Fast concept augments the charac-teristics of this cakemix, but it is still a Godiva cakemix withtaste as its most salient attribute (i.e., a Slim-Fast-likeGodiva cakemix). We therefore expect the taste attribute tobe more salient in Godiva cakemix by Slim-Fast than inSlim-Fast cakemix by Godiva, and the low-calorie attributeto be more salient in Slim-Fast cakemix by Godiva than inGodiva cakemix by Slim-Fast. Consequently, we expect theattributes salient to a constituent brand to be associated morestrongly with the composite brand when the constituentbrand is the header than when it is the modifier.H2: An attribute regarded as highly salient and/or high perform-ing in a constituent brand has a greater effect on the corre-sponding characteristic of the composite brand when thatconstituent brand is the header than when it is the iveness of the composite brand extension ing to Hj^ and Hn,, the salience and the performancelevel of an attribute of a CBE match those of the moresalient and higher-performing constituent brand. Hence,when a modifier brand is highly complementary to the head-er brand in attribute salience and performance level, it canmake relevant but nonsalient attributes of the header brandsalient and replace the weak attribute performance levelswith its strong ones. Consumers' reactions to the CBEshould therefore be more favorable than their reactions to adirect extension of the original header brand (i.e., Slim-Fastcakemix).H3: When the modifier brand is highly complementary to theheader brand in attribute salience and perfonnance levels, aCBE strategy leads to more favorable reactions to the exten-sion product than does a direct brand ite brand extension and feedback ing to the logic underlying, H2, we can assume thatthe CBE is associated more directly with the header thanwith the modifier brand and is understood as a member ofthe header brand category. Accordingly, when the CBE dif-fers from the header brand in the salience of attributes, aswell as in their performance levels, the attribute characteris-tics of the header brand may change in the direction of thosecharacteristics of the CBE through the process of accommo-dation (Crocker, Fiske, and Taylor 1984). If the CBE hashigher performance levels than the header brand, acconuno-dation should lead to positive changes in the originalattribute characteristics of the header (i.e., positive feedbackeffects on the header brand). In contrast, a direct extensionof the header (e.g., Slim-Fast cakemix) does not receive anysuch benefits from a modifier brand because there is nomodifier brand in a direct extension. In addition, though ahighly complementary modifier brand would provide posi-tive feedback effects on the header brand through the CBE,the CBE would have little effect on the modifier. The modi-fier brand just alters some aspects of the header (Hampton1987, 1988; Murphy 1988; Smith et al. 1988). Because it isnot associated directly with the CBE, the modifier is rela-tively independent of the CBE. Hence, we do not expect theCBE to have the same degree of feedback effects on themodifier brand as it has on the header brand.H4: When a modifier brand is highly complementary to a head-er brand in terms of attribute salience and performance lev-els, the header brand in the CBE receives more positivefeedback than (a) that brand in a direct extension and (b) themodifier brand in the CBE.

Composite Branding AlliancesTable 1STUDY DESIGNNumber ofA. Control GroupsSubjects1. Slim-Fast diet food242. Godiva chocolates25B. Experimental Groups3. Slim-Fast cakemix with its feedbackeffect on Slim-Fast diet food31 •4. Godiva cakemix with its feedback effecton Godiva chocolates31Slim-FasC cakemix by Godiva5. With its feedback effect on Slim-Fast diet food346. With its feedback effect on Godiva chocolates29Subtotal63Godiva cakemix by Slim-Fast7. With iu feedback effect on Slim-Fast diet food298. With its feedback effect on Godiva chocolates32Subtotal61Total235MethodDevelopment and pretesting of stimulus materials. Tofind a reasonable extension product class and appropriatebrands from which to form CBEs, we conducted a series ofpretests. To test the hypotheses, we created six between-subjects experimental groups and two control groups (seeTable 1). AH subjects were graduate business studentsenrolled in a large northeastern university. Two stages ofpretesting were necessary to develop appropriate stimulusmaterials and t 1. The objective of the first pretest was to identifyappropriate brands and an extension product for the CBEand generate a set of product attributes relevant to the threeproducts involved in the study—the extension and two con-stituent products. On the basis of informal interviews withstudent subjects (N = 5), we chose Slim-Fast and Godiva astwo constituent brands and cakemix as an extension then conducted a pretest to confirm the appropriatenessof those primary stimuli. First, one group of subjects (N =20) was asked to indicate on a seven-point scale (scale val-ues in an ascending order) their degree of familiarity withthe Slim-Fast and Godiva brands. They exhibited high andequivalent familiarity with the Slim-Fast (x = 5.28) andGodiva (x = 5.21) , the same subjects were asked to indicate theirattitudes toward Slim-Fast and Godiva on two seven-pointscales: likability and favorability. Attitude measures wereobtained because effectiveness of the CBE may be limitedwith highly unfavorable brands. Because the two itemsused were highly correlated (r = .94), they were results indicated that subjects had a nearly neutral atti-tude toward Slim-Fast and a highly favorable attitudetoward Godiva (x = 3.94 and x = 5.41, respectively; p <.01).Third, to examine the level of complementarity betweenthe two brands, the same subjects were asked to provide457salience and performance-level ratings for each brand oneach of several attributes.^ Salience and performance weremeasured on seven-point scales (the importance of eachattribute in brand evaluations; the effectiveness of a brand inperformance of each attribute).'* The ratings showed that alllisted attributes were salient for at least one of the twobrands, Slim-Fast and Godiva, and that those brands highlycomplemented each other in terms of performance levels ofsalient , to check the perceived fit between each brand andthe cakemix product (i.e., to determine whether cakemixwould readily accommodate the primary characteristics ofthe Slim-Fast and Godiva brands), the same group of sub-jects was asked to rate the extendability of Slim-Fast andGodiva to cakemix products on a seven-point scale (1 = doesnot make sense at all, 7 = makes sense a lot). The results in-dicated a high degree of perceived fit between the extensionproduct class and each brand, Slim-Fast and Godiva (x =6.27 and x = 5.89, respectively; both greater than 4.0 atp <.05). That finding also suggests Godiva's good fit with Slim-Fast cakemix as a nested concept, because Godiva is com-plementary to Slim-Fast and fits well with t 2. The objective of the second pretest was to iden-tify the role of the word by as used in the study to createcomposite brands. A group of subjects (N = 34) was askedto judge the degree of association of the composite brandswith the constituent brands on two seven-point scales (1 =still Godiva, 7 = still Slim-Fast; 1 = associated more withGodiva, 7 = associated more with Slim-Fast), which werehighly correlated (r = .95). Those who were exposed to theSlim-Fast cakemix by Godiva composite associated itstrongly with Slim-Fast (x = 5.76) and those exposed to theGodiva cakemix by Slim-Fast composite associated itstrongly with Godiva (x = 2.46). In both cases the compos-ite brand was associated more strongly with the brand posi-tioned before than with the brand positioned after the wordby (t = 10.45, p < .01). Those results suggested that the com-posite brand was associated more closely with the headerthan with the modifier ts. The 235 graduate business students who partic-ipated in Study 1 were placed in small groups and random-ly assigned to one of the eight conditions (two controlgroups and six experimental groups) described in Table preclude any confounding effect, subjects in direct exten-sion or CBE conditions received only one proposed exten-sion. Subjects in control groups provided responses for the'From informal interviews and expert judgments, we obtained a list ofeight attributes relevant to the three product classes (diet food, chocolatecandy, and cakemix): good taste, richness, low calorie, low fat, good value,attractive package, convenience, and luxury. To support selection of thoseattributes, we asked a group of student subjects (N = 18) to rate the per-ceived relevance of the attributes to each product class on two seven-pointscales (ability to judge the product based on this attribute and relevance tothe product). Results showed all the attributes were relevant for the threeproducts."•We operationalized attribute salience to refer to the importance of anattribute with a specific value (level) in brand evaluation (e.g., low calorie).That approach is consistent with Murphy's (1988) definition of attributesalience, which refers to both attributes and their values (e.g., red apple).Hence, attribute salience, as it is measured in our study, is not the same asattribute importance, which is typically measured at a product level withoutany specific value (e.g., /ow-calorie attribute for Slim-Fast cakemix versuscalorie attribute for cakemix products).

458JOURNAL OF MARKETING RESEARCH, NOVEMBER 1996Slim-Fast and Godiva brands in the absence of exposure toany direct extension or mental procedure and measurement. Subjects com-pleted questionnaires that included a brief cover story ex-plaining that the survey was designed to determine con-sumers' reactions to different types of products and brands(for control groups) or to different types of branding strate-gies (for experimental groups). To assess the attributesaliences of all constituent brands and both extensions (thedirect extensions and the CBEs), subjects in all conditionswere asked to indicate on a series of seven-point scales,wherever possible, the degree to which the eight attributeswith specific values (good taste, richness, low calorie, goodvalue, low fat, attractive package, convenience, and luxury)were or would be important in evaluating brands (for justifi-cation, see footnote 4). Subjects also were asked to indicateon a series of seven-point scales how well the brand (Slim-Fast diet food, Slim-Fast chocolate cakemix by Godiva, etc.)performed or would perform on the same set of were then asked to rate the favorableness of their atti-tudinal and behavioral predispositions toward the direct andcomposite brands on three seven-point scales: liking, favor-ableness, and likelihood of purchase. The three measuresproduced statistically indistinguishable evaluations andhence were averaged to form the attitudinal and behavioralreaction (Cronbach's alpha = .92) reported in Tables 2through addition, groups exposed to the CBEs received severalother questions on their thoughts about the composite ically, at the beginning of the questionnaire they wereasked to write down thoughts that had come to mind whenthey first leamed of the composite brand. To examine feed-back effects of the various brand extension strategies, each ofthe groups exposed to either direct extensions or the CBEswere asked to respond to the same attribute salience, perfor-mance level, and attitudinal and behavioral favorablenessquestions about one of the constituent brands (i.e., Slim-Fastdiet food or Godiva chocolates) after their exposure to thedirect extensions or CBEs. On completion of the question-naires, subjects were debriefed with an explanation of thepurpose of the study and sData relevant to H| through H4 are reported in Tables 2,3, and 4. Mean ratings on the attribute salience and perfor-mance scales are reported in Table 2, Parts B and C, for bothdirect extensions of the constituent brands and the x products are used as referents (rather than dietfood and candy, respectively) to hold effects due to productcategory constant.^ Note that the attributes are listed inTables 2, 3, and 4 to facilitate comparisons between theSlim-Fast and Godiva columns. The first three attributes areeither more salient or higher in performance ratings (orboth) for Slim-Fast than for Godiva, and the opposite is truefor the last four r to the results of Pretest 1, those of Study 1 con-firm attribute-level complementarity between the two con-'The results remained the same for all hypotheses, including H,, whenthe original constituent brands, Slim-Fast diet food and Godiva chocolates,were compared with the CBE instead of Slim-Fast cakemix and Godivacakemix. That finding is likely to be a consequence of high similarity inattribute information between the original brands and their 2DIRECT AND COMPOSITEBRAND EXTENSIONSA. EvaluationSlirn-FastGodivaSlim-FastGodivacakemixcakemixcakemixcakemixby Godivaby Slim-FastAverage4.085.11*''5.32*^5.13*B. Attribute SalienceAttributesLow calorie5.97*''2.906.24*<:5.34*"Low fat6.00*'>3.006.19*'5.49*"Good value5.16*"4.48*5.30*"=4.69*Convenience5.45*5.55*5.40*5.03*Good taste6.42*6.58*6.13*6.46*Package3.455.03*"4.73*»5.08*Richness4.90*6.03*"5.51*»6.33*<:Luxury3.105.29*''4.90*"5.64*<:ute PerformanceAttributesLow calorie6.23*'>2.006.14*':4.92*»Low fat5.8 l*b2.526.14*':4.85*»Good value3.553.803.683.41Convenience5.23*4.94*5.02*4.85*Good taste4.58*6.39*"5.76*=6.39*':Package4.48*6.00*"5.11*»5.77*':Richness3.776.32*"5.35*»6.33*':Luxury4.295.39*"4.97*=5.87*':Note: Standard deviation information is suppressed to enhance readabil-ity of the table. The typical standard deviation of mean was approximately.24.*The rating is significantly greater than the midpoint (= 4.0) at p < .05(one-tailed test)."Significant difference in mean attribute rating between a compositebrand (Slim-Fast cakemix by Godiva and Godiva cakemix by Slim-Fast)and the header brand (Slim-Fast cakemix and Godiva cakemix) on that at-tribute (p < .05, one-tailed test)."Significant difference in mean rating between Slim-Fast cakemix andGodiva cakemix (p < .05, one-tailed test).':Significant difference in mean rating between Slim-Fast cakemix byGodiva and Godiva cakemix by Slim-Fast (p < .05, one-tailed test).stituent brands, Slim-Fast and Godiva. Specifically, Table 2,Parts B and C, provide evidence that Godiva cakemix iscomplementary to Slim-Fast cakemix in ratings of attributesalience and performance levels (i.e., one brand is generallyrated well, whereas the other is weak, and vice versa for sev-eral salient attributes, which thus provides a manipulationcheck for attribute-level complementarity). For example,Godiva is rated well on the good taste, richness, attractivepackage, and luxury attributes and poorly on the low-calorieand low-fat attributes.H[a states that when an attribute is salient to at least oneconstituent brand, it is also salient to the CBE. Results arereported in Table 2, Part B. To test the hypothesis, the con-stituent brands were compared with the CBE in attributesalience. First, in the case of Slim-Fast cakemix by Godiva,the comparison between Slim-Fast cakemix and the CBEshows that Godiva, as a modifier, influences the CBE in at-tributes whose mean saliences are greater for Godiva thanfor Slim-Fast cakemix (e.g., attractive package, richness,luxury; indicated by the "b" superscript in the Godivacakemix column). That influence results in increased at-tribute salience ratings for those particular attributes in theSlim-Fast cakemix by Godiva CBE. Similarly, the compari-

Composite Branding Alliancesson between Godiva cakemix and the CBE shows that the at-tributes whose salience ratings are lower for Godiva than forSlim-Fast cakemix (e.g., low calorie, low fat, good value; in-dicated by the "b" superscript in the Slim-Fast cakemix col-umn) have significantly higher ratings in the CBE whenSlim-Fast is the header. As a result, the influences of Godi-va as a modifier and Slim-Fast as a header are combined inthe CBE, for which an attribute is salient whenever it issalient to at least one of the constituent brands. The same re-sults are obtained when the two constituent brands are com-pared in a similar way for another CBE, Godiva cakemix bySlim-Fast. For example, the low-fat attribute, which is notsalient for Godiva cakemix (x = 3.00), becomes highlysalient for the CBE (x = 5.49), presumably because of itshigh salience for Slim-Fast cakemix (x = 6.00). In each in-stance when an attribute is salient to at least one of the con-stituent brands, it is also salient to the CBE. Thus, H|a isstrongly predicts the same results for attribute performanceratings as for attribute salience ratings. Table 2, Part C, pro-vides strong support for that prediction for the Slim-Fastcakemix by Godiva CBE. In every instance when at leastone constituent brand is perceived to perform significantlybetter on a salient attribute (by the same logic discussed forHja, adapted to the performance level data in Table 2, PartC), the composite brand also performs well on that attribute(see results for the low-calorie, low-fat, good taste, attractivepackage, richness, and luxury attributes). A similar pattemof the results is found for the Godiva cakemix by Slim-FastCBE. Thus, H||, also is supported.H2 predicts a greater effect of the header brand than of themodifier on the composite brand. In Table 2, we contrast themean attribute salience and performance levels of the con-stituent brands, Slim-Fast and Godiva, with those of the twoCBEs formed from the constituents. Consistent with the hy-pothesis, attributes significantly better performing in oneconstituent brand than the other are better performing in theCBE when that brand is the header than when it is the mod-ifier (indicated by the "c" superscript in the CBE columns).For example, the low-calorie attribute is high performing forSlim-Fast cakemix (x = 6.23), but not for Godiva cakemix (x= 2.00). The low-calorie attribute is a better performingcharacteristic of Slim-Fast by Godiva (Slim-Fast is the head-er) than of Godiva by Slim-Fast (Slim-Fast is the modifier):X = 6.14 and 4.92, respectively; p < .05. The same pattem ofresults is found for other high-performing attributes (e.g.,richness, good value, low fat, luxury) and also holds for at-tribute salience (see Table 2, Part B).In addition, subjects' thought responses for the two CBEsprovide further evidence about the role of header and modi-fier brands in the processing of the CBE.^ In the Slim-Fastcakemix by Godiva condition, 26 statements identified thecomposite in terms of the header, Slim-Fast (e.g., "low calo-rie cakemix," "diet cakemix with good taste"), whereas on-*Two coders independently classified the thought statements, wheneverpossible, into the two categories strictly on the basis of the sequence andstructure of words contained in the thought statements. For example, whenthe extension product was described only with the Slim-Fast characteristicsor with the Slim-Fast characteristics followed by the Godiva characteristics,it was operationalized as being identified with Slim-Fast. Responses weresimilarly coded when Godiva characteristics dominated. Thought state-ments on which the two coders disagreed were excluded from the analyses(a total of 14 statements).459ly 8 identified it in terms of the modifier, Godiva (e.g.,"sweet and expensive cakemix," "good chocolate productwith low calories"). Similarly, in the Godiva cakemix bySlim-Fast condition, 25 statements identified the compositein terms of the header, Godiva, whereas only 6 identified itin terms of the modifier, Slim-Fast (x^ = 4.60, p < .05).Those responses support the dominance of the header in thecomposite processing. Examination of subjects' thought re-sponses shows no specific concems about the fit betweenthe header brand and the extension product. In both com-posites, they processed the extension product as a part of theheader brand without much difficulty or elaboration.^ Sub-jects apparently readily processed the header brand togetherwith the extension product to form a nested concept.H3 predicts more positive effects on subjects' reactions tothe extension product from a CBE strategy than from a di-rect extension strategy. Table 2, Part A, provides evidencethat the composite Slim-Fast cakemix by Godiva wasviewed significantly more favorably (indicated by an "a" su-perscript) than the direct extension, Slim-Fast cakemix (x =5.32 versus 4.08, respectively). However, the composite Go-diva cakemix by Slim-Fast (x = 5.13) was not viewed sig-nificantly more favorably than the direct extension, Godivacakemix (x = 5.11). Although Table 2, Parts B and C, showhigh complementarity between Slim-Fast and Godiva, thegreater effectiveness predicted for the CBE is supported forSlim-Fast cakemix by Godiva but not for Godiva cakemixby Slim-Fast. This result suggests that even when the twobrands are complementary in attribute salience and perfor-mance, the modifier may not necessarily foster favorable re-actions to the CBE. Thus, H3 receives mixed e its lack of greater overall effectiveness, the Godi-va cakemix by Slim-Fast CBE has interesting results for at-tribute-level information in comparison with Godivacakemix. For example, the modifier brand, Slim-Fast, influ-ences the attribute performance ratings of Godiva cakemixby Slim-Fast, as is shown in Table 2, Part C (see low calorieand low fat). Although such improvements did not extend tothe subjects' overall reactions, the Slim-Fast modifier appar-ently causes attribute-level changes.H4 states that a header brand in a CBE receives more pos-itive feedback than (1) that brand in a direct extension con-dition and (2) a modifier brand in a CBE. Table 3, Part A,provides information about subjects' reactions to each of theconstituent brands in their original product categories frommeasures taken both before and after subjects were exposedto the CBE. In the Slim-Fast cakemix by Godiva condition,subjects' reactions to the header brand, which were mea-sured after exposure to the CBE, improved significantlyfrom those measured before exposure (x= 4.92 versus 3.74,respectively; p < .05). In contrast, there was no positive'A question can be raised about the generalizability of the header andmodifier positions in other forms of the composite brand extension such asGodiva's Slim-Fast cakemix. We examined that issue by comparing sub-jects' reactions to the Slim-Fast cakemix by Godiva CBE with those to theGodiva's Slim-Fast cakemix CBE, using two groups (N = 29 and 30,respectively). The results, not reported because of space limitations, revealhighly similar pattems in terms of attribute salience and attribute perfor-mance levels between Godiva's Slim-Fast cakemix and Slim-Fast cakemixby Godiva, but show greater effectiveness for the latter than for the formerin the attitudinal and/or behavioral reactions. Thus, though the generaliz-ability of the header and modifier positions holds well in an altemativebranding form, the relative effectiveness of the composite brand extensionseems to depend on the specific form of composite branding.

460JOURNAL OF MARKETING RESEARCH, NOVEMBER 1996Table 3CONSTITUENT BRANDS AND FEEDBACK tionSlim-Fast diet foodGodiva chocolateSlim-Fast diet foodSlim-Fastafter Slim-FastGodivaafter Slim-Fastafter Slim-Fastdiet foodby Godivachocolateby GodivacakemixAverage score3.744.92*ab5.47*5.37*3.66B. Attribute SalienceAttributesLx>w calorie6.38*6.47*2.282.036.19*Low fat6.46*6.00*2.922.796.13*Good value5.42*5.15*4.124.245.10*Convenience5.58*5.35*4.68*4.315.26*Good taste6.04*6.15*6.76*6.62*6.13*Package3.544.03"5.28*5.03*3.68Richness4.085.50*^6.44*6.38*4.52*Luxury2.173.68"'>5.60*5.72*2.52C. Attribute PerformanceAttributesLow calorie6.08*6.47*1.482.006.19*Low fat6.25*6.00*1.522.006.13*Good value4.334.383.243.244.00Convenience5.08*5.18*4.56*5.07*5.23*Good taste4.42547«ab6.44*6.31*4.58*Package4.214.62*5.36*5.48*4.29Richness4.175.00*»b6.28*6.52*4.00Luxury3.294.18^''5.72*5.86*3.23Note: Standard deviation information is suppressed to enhance readability of the table. The typical standard deviation of mean was approximately .24.*Rating is significantly greater than the midpoint (= 4.0) at />< .05 (one-tailed test)."Significant difference in mean rating of a header brand measured after exposure to a composite brand (Slim-Fast diet food after Slim-Fast by Godiva) ver-sus before exposure (Slim-Fast diet food) (p < .05, one-tailed test).''Significant difference in mean rating between Slim-Fast diet food after Slim-Fast by Godiva and Slim-Fast diet food after Slim-Fast cakemix (p < .05, one-tailed test).feedback effect for either Slim-Fast in the direct extension3.64 versus 3.74 before and after exposure to the CBE, re-condition (x = 3.74 versus 3.66 before and after exposure tospectively; thus d = .10)—shows no differential positivethe direct extension, respectively) or for Godiva when itfeedback effects of the CBE on the header, Godiva. Thus, H4served as the modifier brand in the CBE (x = 5.47 versusis not supported in the Godiva cakemix by Slim-Fast condi-5.37 before and after exposure to the CBE, respectively).tion. Examination of the attribute performance ratingsComparisons of the differences in overall reactions between(Table 3, Part C) shows that the header brand, Godivabefore and after exposure to brand extensions show that thechocolate (Tables 3, Part C, and 4, Part C), improved on at-header, Slim-Fast, received more positive feedback than ei-tributes on which the modifier was stronger than the headerther that in the direct extension (d = 1.18 versus .08, , low calorie, low fat, convenience; indicated by the "a"tively; p < .05) or the modifier, Godiva (d = 1.18 versus -. 10,superscript in the Godiva after Godiva by Slim-Fast column)respectively; p < .05), which thus supports H4. A similar pat-after exposure to the CBE. Therefore, when the modifiertem of results was found for the attribute performance rat-was Slim-Fast rather than Godiva, the header brand (i.e.,ings. That is, after exposure to the CBE, the header brand,Godiva) did not improve in subjects' overall reactions butSlim-Fast (Tables 3, Part C, and 4, Part C), improved on thedid improve in attribute performance levels. However, as inattributes on which the modifier was stronger than the head-the Slim-Fast cakemix by Godiva condition, no such im-er (e.g., good taste, richness, luxury), whereas similar im-provements in attribute performance ratings were found forprovements failed to occur either for the modifier brandthe modifier brand (Slim-Fast) or for Godiva in the direct(Godiva) in the CBE condition or for Slim-Fast in the directextension condition, namely, Godiva ion the Godiva cakemix by Slim-Fast condition, however,Discussionno such positive feedback effects occurred for the headerThe results of Study 1 reveal that under some conditionsbrand in subjects' overall reactions. The Godiva chocolateof high fit of the constituent brands with the extension prod-measure taken after the CBE exposure (Table 4, Part A) isuct and high attribute-level complementarity between thenot significantly different from the before measure (Table 3,two constituent brands, a composite brand name can favor-Part A; X = 5.15 versus 5.47, respectively). Comparison ofably influence subjects' reactions to the extension productthis difference (d = .32) with those in the direct extensionby allowing salient attributes and performance levels of thecondition—Godiva cakemix (x - 5.47 versus 5.23 beforemodifier brand to be assimilated into the characteristics ofand after exposure to the direct extension, respectively; thusthe header brand. The Slim-Fast cakemix by Godiva CBE isd = .24) and the modifier brand (Slim-Fast) in the CBE (x =an example. In addition, that CBE had positive feedback

Composite Branding AlliancesTable 4FEEDBACK EFFECTS IN OTHER CONDITIONSA. EvaluationSlim-FastGodivaGodivadiet foodchocolatechocolateafter Godivaafter Godivaafter Godivaby Slim-Fastby Slim-FastcakemixAverage score3.645.15*5.23*B. Attribute SalienceAttributesLow calorie6.55*4.06=2.06Low fat6.20*3.91^2.35Good value5.10*4.344.65*'>Convenience5.48*5.09*4.52*Good taste5.83*6.56*6.71*Package3.345.06*5.39*Richness4.246.03*6.52*Luxury2.485.53*5.32*C. Attribute PerformanceAttributesLow calorie6.17*3.47"1.45Low fat6.00*3.47a1.68Good value3.933.443.29Convenience5.45*5.38*"4.84*Good taste4.076.56*6.42*Package3.975.38*6.00*1'Richness4.076.44*6.52*Luxury3.215.81*5.29*Note: Standard deviation information is suppressed to enhance read-ability of the table. The typical standard deviation of mean was approxi-mately .24.*Rating is significantly greater than the midpoint (= 4.0) at p < .05 (one-tailed test)."Significant difference in mean rating of a constituent brand measuredafter exposure to a composite brand (Godiva by Slim-Fast) versus beforeexposure (p < .05, one-tailed test).''Significant difference in mean rating of a constituent brand measuredafter exposure to a direct extension (Godiva cakemix) versus before expo-sure (p < .05, one-tailed test).effects on the header brand but no particular effect on themodifier brand. Because chocolate is seen as an ingredientof a chocolate cakemix and Godiva is associated withchocolate, the Slim-Fast cakemix by Godiva CBE is likelyto be seen by subjects as Godiva serving as an ingredient ofthe Slim-Fast cakemix. Hence, the high effectiveness of theSlim-Fast cakemix by Godiva CBE may be explained by theso-called ingredient cobranding phenomenon (e.g., Ben &Jerry's Heath Bar Crunch Ice Cream and Nabisco's Fat-FreeCranberry Newtons with Ocean Spray Cranberries).Moreover, Godiva as a modifier brand received little feed-back from the CBE, perhaps because it is not the primarycomponent of the same, however, could not be said for the Godivacakemix by Slim-Fast CBE. Although the two brands werecomplementary, the CBE was not more effective than a di-rect extension (i.e., Godiva cakemix). In addition, no posi-tive feedback effects on the header brand (i.e., Godiva) fromthe CBE were apparent. That finding is interesting because,despite an absence of improvement in the overall reactionsto the CBE, salience and performance levels changed up-ward for some attributes of the sts in the results from the two CBEs raise two pos-sibilities: (1) a ceiling effect and (2) a favorability effect of461the modifier brand in the CBE. Although the modifier brandwas complementary to the header brand in the Godivacakemix by Slim-Fast CBE, the complementarity effectmight not have been strong enough to overcome the ceilingeffect of the header brand, which was already highly favor-able. Specifically, note that Godiva chocolate's mean ratingwas 5.47 (Table 3, Part A), the highest of all ratings in thestudy in terms of overall reactions. Because Godiva is per-ceived to be one of the best brands in its product category, aceiling may have formed in subjects' ratings at the level ofthe low 5's. Alternatively, independent of the attribute-levelcomplementarity, attitudinal favorability of the modifierbrand might have influenced the subjects' reactions to theCBE. The moderate attitudinal favorability of Slim-Fastmight not have been sufficient to increase the effectivenessof the Godiva cakemix by Slim-Fast CBE. Consumers' atti-tudes toward a CBE may improve and a header brand mayreceive positive feedback effects from the CBE when theheader is linked to a highly favorable modifier brand. Hence,the Study 1 results raise an interesting issue: the role of at-tribute-level complementarity and attitudinal favorability ofthe modifier brand in determining the overall effectivenessof the 2Study 2 was designed to explore the relative contributionsof attribute-level complementarity and attitudinal favorabil-ity of the modifier brand to the effectiveness of the CBEstrategy while controlling for any potential ceiling effects byincorporating comparative evaluation measures (i.e., choiceand preference). No formal hypothesis was stated, however,because of the difficulty of conceptually justifying a priorithe difference in the respective Study 1 results show the attribute complementarity of amodifier to be an important factor for the effectiveness of theCBE strategy. A complementary modifier enables the CBE tohave a better attribute profile than do direct extensions of itsconstituent brands, which may subsequently lead to more fa-vorable reactions to the CBE. Alternatively, attitudinal favora-bility of a modifier brand may explain the effectiveness of theCBE strategy. Previous research (see Cohen and Areni 1991)found strong effects of overall affect toward one object on theevaluation of another object when the two were arranged to-gether temporally or spatially (e.g., affect transfer). Such asso-ciation-b£ised affect transfer may explain the effectiveness ofthe CBE. Furthermore, by being paired with the header brand,the modifier brand may reduce consumers' perceived uncer-tainty in evaluating and purchasing a new product. For exam-ple, when a header brand is combined with a modifier brandthat evokes highly favorable attitudes, consumers' evaluationsof the CBE may increase, because having a favorable brandname associated with the' product reduces uncertainty. Thus,the attitudinal favorability of a modifier brand may be an im-portant contributor to the effectiveness of the CBE ment of the relative effects of attitudinal favorabil-ity of a modifier brand and attribute-level complementarityon the effectiveness of the CBE can be performed at two dif-ferent levels of attitudes toward the header brand: a moder-ate level and a high level. However, identifying appropriateCBEs for the comparison is difficult when the header brandis only moderately (not favorably) evaluated. The challengeis to find modifier brands that meet two orthogonal condi-tions of attribute complementarity and attitudinal favorabil-

462JOURNAL OF MARKETING RESEARCH, NOVEMBER 1996ity: (1) being high in attribute complementarity and low inattitudinal favorability and (2) being low in attribute com-plementarity and high in attitudinal favorability. For exam-ple, suppose Slim-Fast (diet food), which was not favorablyevaluated in Study 1, is a header brand and we need to findtwo modifier brands that are orthogonal to each other in thetwo factors, complementarity and favorability. First, if amodifier brand is favorably evaluated (e.g., Godiva), thatbrand must have some strong attributes (e.g., good taste).Then, for the header brand (e.g., Slim-Fast) not to be com-plementary with that brand, the header must be perceived tobe strong on those same attributes (good taste); otherwisethe strong attributes of the modifier can compensate for theweak ones of the header. It is empirically difficult to identi-fy a moderately favorable header brand and a highly favor-able modifier brand that are similarly strong or weak in at-tribute contrast, when the header brand is evaluated favorably,it becomes possible to identify the two orthogonal condi-tions: (1) both header and modifier brands are evaluated fa-vorably but are not complementary to each other (i.e., Godi-va cakemix by Haagen-Dazs) and (2) the modifier is com-plementary to but not as favorably evaluated as the , Godiva cakemix by Slim-Fast). By comparing the ef-fectiveness of the Godiva cakemix by Haagen-Dazs CBEwith that of the Godiva cakemix by Slim-Fast CBE, we canexplore the relative contributions of attribute-level comple-mentarity and attitudinal favorability of the modifier brandto the overall effectiveness of the CBE. The two conditionsdiffer only in the characteristics of the modifier brand; theheader brand (i.e., Godiva) remains the comparing the two CBEs, we must control for a poten-tial ceiling effect in the judgment (evaluation) task. In suchtasks, the goal is often to "categorize" objects, with no re-quirement to respond to objects differentially if they arelargely similar (Alba, Hutchinson, and Lynch 1991; Billingsand Scherer 1988; Tversky, Slovic, and Sattath 1988). In ad-dition, the judgment task produces "holistic" and "by-alter-native" evaluations without much attention to the intradi-mensional differences (Rosen and Rosenkoetter 1976). Ac-cordingly, overall judgment or evaluation measures may notfully reflect the finer differences between the CBE and thedirect extension (see Alba, Hutchinson, and Lynch 1991).In Study 2, we therefore incorporate choice and preferencemeasures to control the ceiling effect resulting from the high-ly favorable header brand (i.e., Godiva) in both the effective-ness of the CBE and its feedback effect on the header. Choic-es are more sensitive than judgments when overall judgmentsare fairly similar because choices force discrimination(Heath, McCarthy, and Mothersbaugh 1994; Payne 1982). Inother words, comparative evaluations such as choice andpreference involve looking for dissimilarities, that is, featuresthat distinguish between aitematives (Payne 1982), and there-fore may result in increased sensitivity to the differences, ifany, between the CBE and the direct Pretest of stimulus material. To find a brand to serve as amodifier in the condition "favorably evaluated but not com-plementary with the header brand," we performed a selected the Haagen-Dazs (ice cream) brand on the basisof informal interviews (N = 6), because it was thought to bewell-known, a good fit with cakemix products, evaluated asfavorably as Godiva, and less able to compensate for theweaknesses of Godiva than the Slim-Fast brand. We verifiedthose conditions with a group of students (N = 27), using thesame measures as were used in the pretests of Study I. Theresults showed that the subjects were familiar with theHaagen-Dazs brand (x = 5.60, significantly greater than thescale midpoint, p < .05) and evaluated it as favorably as theGodiva brand (x = 5.34 and 5.47, respectively; no significantdifference, p < .05). Subjects also perceived cakemix prod-ucts to be a reasonable extension for Haagen-Dazs (x = 5.49,significantly greater than the midpoint, p < .05). Finally, theHaagen-Dazs brand has attribute associations that are simi-lar to those of Godiva and thus is less compensating forGodiva's weaknesses (i.e., high calorie and high fat) than theSlim-Fast ts. Ninety business students participated in the mainstudy. They were assigned randonfily to one of three condi-tions (one control group and two experimental groups).Subjects in the experimental conditions received just one pro-posed CBE (either Godiva cakemix by Slim-Fast or Godivacakemix by Haagen-Dazs) to preclude any confoundingeffect. Those in the control group responded to the Haagen-Dazs brand because subjects' responses to the Godiva andSlim-Fast brands had already been obtained in Study mental procedure and measurement. Subjects com-pleted questionnaires identical to the ones used in Study in the experimental conditions of Study 2 providedresponses to two additional sets of measures about tfie first set was for choice between (1) the CBE (e.g.,Godiva cakemix by Slim-Fast) and Duncan Hines cakemixand (2) the CBE and Godiva cakemix. The second set was forpreference between (1) the CBE and Duncan Hines cakemix(seven-point scales: 1 = I like the CBE less than DuncanHines .cakemix, 7 = 1 like the CBE more than Duncan Hinescakemix) and (2) the CBE and Godiva cakemix (seven-pointscales: 1=1 like the CBE less than Godiva cakemix, 7 = 1 likethe CBE more than Godiva cakemix). Those measures wereintended to reduce any potential ceiling effects in responsesand detect any finer differences in the relative impact of theattribute complementarity and the attitudinal favorability ofthe modifier brand. Similarly, an additional measure ofchoice between Godiva and Hershey was included in themeasures of feedback effects on the header brand, GodivaDuncan Hines cakemix was compared with the CBEs inaddition to the Godiva cakemix, because Duncan Hines, as aleading brand in the cakemix product category, was a readilyavailable reference point to which the CBEs would be com-pared. Hershey, as a leading brand in a chocolate product cat-egory, also was included for the comparison with Godiva todetect any differences in feedback effects on the header sIn Table 5, we contrast the mean ratings of overall reac-tions (Part A) and the attribute salience and performancelevels (Parts B and C) of the two CBEs (Godiva cakemix bySlim-Fast and Godiva cakemix by Haagen-Dazs), therebysatisfying the orthogonal condition (i.e., complementary butnot favorable versus favorable but not complementary).Specifically, Haagen-Dazs is not complementary to Godivabecause Haagen-Dazs does not compensate for the weakattributes (i.e., low calorie and low fat) of Godiva in terms ofsalience and performance. Feedback measures on the head-

Composite Branding Alliances463Table 5GODIVA CAKEMIX BY HAAGEN-DAZS VERSUS GODIVA CAKEMIX BY SLIM-FASTA. Overall EvaluationsGodivaGodivaGodivaGodivatiaagen-DatsGodivacakemix bycakemix bychocolatechocolateice creamchocolateHaagen-DazsSlim-Fastafter G-by-Hafter G-by-S{N = 27)(N = 25)(N = 30)(N = 33)(N = 30)^(N = 33)''Average5.35*5,47*5,11*5,23*5,08*5,21*B. Means of Attribute SalienceAttributesLow calorie3,212,282,775,61 *<=2,203,73"Low fat3,082,922,935,73*<^2,333,85''Good value5,52*4,124,67*4,61*3,873,79Convenience4,154,68*4,87*5,09*4,334,39Good taste6,59*6,76*6,57*6,39*6,60*6,36*Package3,635,28*5,17*4,76*5,13*5,27*Richness5,33*6,44*6.03*5,91*6,57*^5,97*Luxury4,51*5.60*5,20*5,06*5,43*5,27*C. Means of Attribute PerformanceAttributesLow calorie2,111,481,834,88*c1,772,91''Low fat2.171,521,674,76*"^1,803,24''Good value4,303,243,573,703,233,55Convenience4,78*4,56*4,50*5.06*<:4,80*5,12*Good taste6,19*6,44*6,33*5,97*6,60*6,45*Package4,265,36*5,87*5,76*5,91*5,76*Richness5,74*6.28*6,37*5,93*6.43*6,09*Luxury4,98*5,72*5,93*5,50*6,03*5,85*Note: Standard deviation information is suppressed to enhance readability of the table. The typical standard deviation of mean was approximately .25,*Rating is significantly greater than the midpoint (= 4,0) at p < ,05 (one-tailed test),"G-by-H refers to Godiva cakemix by Haagen-Dazs,•KJ-by-S refers to Godiva cakemix by Slim-Fast.'^Significant difference in mean rating between Godiva cakemix by Haagen-Dazs and Godiva cakemix by Slim-Fast {p < ,05, one-tailed test),''Significant difference in mean rating between header brand measured after exposure to Godiva cakemix by Haagen-Dazs and header brand after exposureto Godiva cakemix by Slim-Fast (p < ,05, one-tailed test).er brand, Godiva chocolates, also are summarized in Tablein the Godiva cakemix by Haagen-Dazs condition (72% ver-5. The results show that though Godiva cakemix by Slitn-sus 55% of the choices, respectively; x = 5.12 and x = 4,28,Fast led to a better attribute profile in terms of attributerespectively; significant at p < .05).salietice and performance (see Table 5, Parts B and C), thereIn Table 5, Part A, we show that the feedback effects of theis no significant difference between Godiva cakemix byCBEs on the header brand made no significant difference be-Haagen-Dazs and Godiva cakemix by Slim-Fast in terms oftween the two CBE conditions (Godiva cakemix by Slim-Fastoverall reactions (x = 5.11 and x = 5.23, respectively; seeand Godiva cakemix by Haagen-Dazs) in terms of overall re-Table 5, Part A), as is consistent with the results of Study s (x = 5.21 and x = 5.08, respectively). In addition, theIn addition, the comparison of overall reactions between theresults in Table 6 show no significant difference between thetwo CBEs and the direct extension of the header, Godivatwo conditions when subjects were asked to make a choice orcakemix (x = 5.11; see Table 2, Part A), shows no significantexpress their preference between the header brand and y brand after their exposure to the CBEs. Yet, the threeHowever, other measures, such as choices and prefer-measures—attitudinal and behavioral reactions, choice, andences, show significant differences between the two CBEspreference—show the same pattems favoring Godiva choco-(see Table 6). When subjects were asked to choose betweenlate after exposure to the Godiva caketnix by Slim-Fast ratherthe CBEs and the Duncan Hines cakemix, 70% (23 of 33)than to the Godiva cakemix by Haagen-Dazs (see Table 5, Partchose the CBE in the Godiva cakemix by Slim-Fast condi-A, and Table 6), They provide at least consistent directionaltion, whereas 57% (17 of 30) chose the CBE in the Godivasupport for a more positive role of attribute-level complemen-cakemix by Haagen-Dazs condition (significantly differenttarity than of attitudinal favorability in feedback p < .05). In terms of preference, subjects' preference ofthe Godiva cakemix by Slim-Fast to the Duncan HinesDiscussioncakemix was greater than their preference of the GodivaIn Study 2, we explore the relative contributions of theCEikemix by Haagen-Dazs to the Duncan Hines cakemix (xtwo major factors in successful CBE strategies—the= 5.24 and x = 4.40, respectively; significant at /? < .05), Inattribute-level complementarity and the attitudinal favorabil-addition, when the CBEs were compared with the Godivaity of the modifier brand—while controlling for a potentialcakemix, more choices and greater preference were given toceiling effect. The choice and preference measures show athe CBE in the Godiva cakemix by Slim-Fast condition thangreater contribution by attribute-level complementarity than

464JOURNAL OF MARKETING RESEARCH, NOVEMBER 1.996Table 6EFFECTS OF COMPOSITE BRAND EXTENSIONS ON CHOICES AND PREFERENCESGodivaGodivaGodivacakemix byGodivacakemix bychocolatechocolateHaagen-DazsSUm-Fastafter G-by-H"after 0-by-S''(N = 30)(N = 33)(N = 30)(N = 33)Choices1, CBE'= versus Duncan Hines cakemix17-13(57%)23-10 (70%)''2. CBE versus Godiva cakemix16-13 (55%)21-28 (72%)''3. Godiva versus Hershey chocolates17-12 (59%)19-10 (66%)Preferences4, CBE versus Duncan Hines cakemix4,405,24*''5, CBE versus Godiva cakemix4,285,12*''6, Godiva versus Hershey chocolates4,76*4.94*Note; Standard deviation information is suppressed to enhance readability of the table. The typical standard deviation of mean was approximately ,25,*Rating is significantly greater than the midpoint (= 4,0) at p < ,05 (one-tailed test),

Composite Branding Alliancestionships. The stimuli we use thus fit well with ingredientcobranding. However, complementarity also may be basedon an overall image (e.g.. Jaguar sedan by Sony) or usagesituations (Little Caesar's breakfast pizza by Eggo).Whether consumers process those types of composite brand-ing in the same way as composite branding with attribute-level complementarity is an interesting issue for further re-search. In addition, the perception of complementarity maydepend on the nature of the extension. For example, in Slim-Fast chocolate cakemix by Godiva, Godiva's complementar-ity to Slim-Fast may be perceived as greater than Slim-Fast's complementarity to Godiva, whereas in Godiva dietcakemix by Slim-Fast the opposite may be the , research is needed to assess the importance of thefit between the constituent brands and the extension e the name of the CBE, in general, looks novel and in-teresting, it may draw consumers' attention, motivating themto rely on their own idiosyncratic "theories" (Murphy andMedin 1985) in interpreting the meaning of the CBE. Hence,the issue of fit may be less critical in a CBE context than ina direct brand extension context. Fit with the extension prod-uct may be even less critical for the modifier brand, becausethe CBE seems primarily associated with the , though the results of Study 2 show less contribu-tion to CBE effectiveness by the attitudinal favorability ofthe modifier brand than by the attribute-level complemen-tarity, they do not necessarily mean the absence of such con-tribution by the former in an absolute sense. Conceivably,for some technological or high-risk purchases, brand al-liances among highly favorable brands may reduce con-sumers' perceived risk (e.g.. Power personal computer byApple, Motorola, and IBM). Moreover, that we found nosubstantial contribution by Haagen Dazs to the Godivacakemix by Haagen Dazs CBE may not necessarily meanthat such a result would occur for a header brand of onlymoderate attitudinal favorability. To identify precisely theeffect of attitudinal favorability of the modifier brand on theCBE, additional research may need to employ a full factor-ial design in which different levels of attitudinal favorabili-ty and attribute-level complementarity are manipulated ex-perimentally, rather than simply measured, because of thedifficulty of identifying real-world examples for each of themultiple experimental treatment conditions. Such researchwould enable researchers to assess the main effect of attitu-dinal favorability of the modifier brand, as well as its poten-tial interaction effect with the attribute-level complementar-ity between two constituent , our study examines only customer reactions to theCBE, not the evaluations that occur when expectations areconfronted by the reality of actual product experience. Inthat sense, our results are suggestive of what may be desir-able for the producer to try to achieve, not what may be fea-sible for it to accomplish through actual product perfor-mance (Ratneshwar, Shocker, and Srivastava 1995). Howev-er, because not all attributes important to the consumer maybe readily observed or experienced (Darby and Kami 1974;Nelson 1970), in certain instances, the CBE strategy stillmight be effective if favorable beliefs about such attributesare formed and persist on the basis of brand name alone. Forexample, examining branding strategy in the context ofsearch and experience goods (Nelson 1970) and credencegoods (Darby and Kami 1974) may afford useful conditions465for studying the contingency effects of a CBE strategy. Re-searchers could investigate whether the CBE strategy ismore effective for experience and/or credence goods thanfor search NCESAaker, David A. and Kevin Lane Keller (1990), "Consumer Evalu-ations of Brand Extensions," Journal of Marketing. 54 (Janu-ary), 27-+, Joseph W., J. Wesley Hutchinson, and John G. Lynch, Jr,(1991), "Memory and Decision Making," in Handbook of Con-sumer Behavior. Thomas S. Robertson and Harold H. Kassar-jian, eds, Englewood Ciiffs, NJ: Prentice Hall, gs, Robert S, and Lisa L, Scherer (1988), "The Effects of Re-sponse Mode and Importatice on Decision-Making Strategies:Judgment Versus Choice," Organizational Behavior and HumanDecision Processes. 41 (1), 1-19,Bousch, David M. and Barbara Loken (1991), "A Process-TracingStudy of Brand Extension Evaluation," Journal of Marketing Re-search. 28 (February), rczyk, Susan M. and Joseph W. Alba (1994a), "The Impor-tance of the Brand in Brand Extension," Joumat of MarketingResearch. 31 (May), (1994b), "The Role of Consumers' Intuitionsin Inferetice Making," Journal of Consumer Research. 21 (De-cember), varti, Dipankar, Deborah Maclnnis, and Kent Nakamoto(1990), "Product Category Perceptions, Elaborative Processing,and Brand Name Extension Strategies," in Advances in Con-sumer Research. Vol. 17, Richard Pollay, Marvin Goldberg, andGerald Gom, eds. Provo, UT: Association for Consumer Re-search, , B. and G. L. Murphy (1984),"Models of Concepts,"tive Science. 8(1), 27-58, Cogni-Cohen, Joel B. and Charles S, Areni (1991), "Affect and ConsumerBehavior," in Handbook of Consumer Behavior. Thomas son and Harold H. Kassarjian, eds, Englewood Cliffs, NJ:Prentice Hall, Kunal Basu (1987), "Altemative Models of Catego-rization: Toward a Contingent Processing Framework," Journalof Consumer Research. 13 (March), r, Jennifer, Susan T, Fiske, and Shelly E, Taylor (1984),"Schematic Bases of Belief Change," in Attitudinal Judgment. J,Richard Eiser, ed. New York: Springer-Verlag, , Peter A. and Daniel C. Smith (1994), "The Effect of BrandPortfolio Characteristics on Consumer Evaluations of Brand Ex-tensions," Joumat of Marketing Research. 31 (May), , Michael R. and Edi Kami (1974), "Free Competition andthe Optimal Amount of Fraud," Journal of Law and Economics.16 (April), , Tulin (1993), "Brand Equity as a Signaling Phenomenon,"working paper, Haas Graduate School of Business, University ofCalifomia, k, Michael W. and Mark T, Keanne (1990), Cognitive Psy-chology. London: Lawrence Eribaum ar, Peter H, (1989), "Managing Brand Equity," MarketingResearch. (). , T, (1980), "The Semantic Interpretation of Nominal Com-pounds," in Proceedings of the First Annual National Confer-ence on Artificial Intelligence. Vol, 1, Robert Balzer, eds. Stan-ford University, CA: American Association for Artificial Intelli-gence, n, James A. (1987), "Inheritance of Attributes in NaturalConcept Conjunctions," Memory & Cognition, 15(1), 55-71.(1988), "Overextension of Conjunctive Concepts: Evi-dence for a Unitary Model of Concept Typicality and Class In-clusion," Journal of Experimental Psychology: Leaming, Mem-ory and Cognition. 14(1), 12-32,

466JOURNAL OF MARKETING RESEARCH, NOVEMBER 1996Heath, Timothy B., Michael S. McCarthy, and David L. Mothers-baugh (1994), "Spokesperson Fame and Vividness Effects in theContext of Issue-Relevant Thinking: The Moderating Role ofCompetitive Setting," Journal of Consumer Research, 20(March), , Kevin Lane and David A. Aaker (1992), "The Effects of Se-quential Introduction of Brand Extensions," Joumal of Market-ing Research, 29 (February), , George (1985), Women, Fire, and Dangerous o: University of Chicago , Barbara and Deborah Roedder John (1993), "DilutingBrand Beliefs: When Do Brand Extensions Have a Negative Im-pact?" Joumal of Mariceting, 57 (July), , G. A. and P. N. Johnson-Laird (1976), Language and Per-ception. Cambridge: Cambridge University , Gregory L. (1988), "Comprehending Complex Concepts,"Cognitive Science, 12 (4), Douglas L. Medin (1985), "The Role of Theories inConceptual Coherence," Psychological Review, 92 (3),, Phillip (1970), "Information and Consumer Behavior,"Joumai of Political Economy, 78 (March/April), , C. Whan, Bernard J. Jaworski, and Deborah J. Maclnnis(1986), "Strategic Brand Concept-Image Management," Joumalof Marketing, 50 (October), V. Parker Lessig (1981), "Familiarity and Its Impact onConsumer Biases and Heuristics," Joumai of Consumer Re-search, 8 (September), 223-30.-, Michael S. McCarthy, and Sandra J. Milberg (1993), 'TheEffects of Direct and Associative Brand Extension Strategies onConsumer Response to Brand Extensions," in Advances in Con-sumer Research, Vol. 20, Leigh McAlister and Michael L. Roth-schild, eds. Provo, UT: Association for Consumer Research,28-33.-, Sandra Milberg, and Robert Lawson (1991), "Evaluationof Brand Extensions: The Role of Product Feature Similarity andBrand Concept Consistency," Joumal of Consumer Research, 18(September), . J. W. (1982), "Contingent Decision Behavior," Psychologi-cal Bulletin, 92 (2), hwar, S., Allan D. Shocker, and Rajendra Srivastava (1995),'The Managerial Relevance of an Adaptive View of Market Be-havior," working paper. College of Business Administration,University of , Jean B. (1991), "The Effect of Negative Infonnation onthe Evaluation of Brand Extensions and the Family Brand," inAdvances in Consumer Research, Vol. 18, Rebecca H. Holmanand Michael R. Solomon, eds. Provo, UT: Association for Con-sumer Research, 399^, L. D. and P. Rosenkoetter (1976), "An Eye Fixation Analy-sis of Choice of Judgment With Multiattribute Stimuli," Memo-ry & Cognition, 4 (6), t, Bemd H. and Laurette Dube (1992), "ContexUializedRepresentations of Brand Extensions: Are Feature Lists orFrames the Basic Components of Consumer Cognition?"keting Utters, 3 (2), 115-26. Mar-Smith, Edward E., Daniel N. Osherson, Lance J. Rips, and Mar-garet Keane (1988), "Combining Prototypes: A Selective Modi-fication Model," Cognitive Science, 12 (4), , Gert, Paul De Boeck, Iven Van Mechelen, and Dirk Geer-aerts (1993), "Dominance and Noncommutativity Effects inConcept Conjunctions: Extensional or Intensional Basis?" Mem-ory & Cognition, 21 (6), d, P. (1984), "Conceptual Combination and Scientific Dis-covery," in PSA, Vol. 1, P. Asquith and P. Kitcher, eds. East Lans-ing, MI: Philosophy of Science Association, y, Amos, Paul Slovic, and Shmuel Sattath (1988), "Contin-gent Weighting in Judgment and Choice," Psychological Review,95 (3), elt, Birger (1988), "Umbrella Branding as a Signal of NewProduct Quality: An Example of Signaling by Posting a Bond,"Rand Joumai of Economics, 19 (Autumn), 458-66.