The Impact of Brand Extension Introduction on Choice

7/29/2019 The Impact of Brand Extension Introduction on Choice

1/16

Vanitha Swaminathan, RicharcJ J. Fox, & Srinivas K. RecJdy

The Impact of Brand ExtensionIntroduction on ChoiceThis article focuses on the impact of a new brand extension introduction on choice in a behavioral context usingnational household scanner data involving multiple brand extensions. Particularly, the authors investigate the rec-iprocal impact of trial of successful and unsuccessful brand extensions on parent brand choice. In addition, theauthors examine the effects of experience with the parent brand on consumers' trial and repeat of a brand exten-sion using household scanner data on six brand extensions from a national panel. In the case of successful brandextensions, the results show positive reciprocal effects of extension trial on parent brand choice, particularly amongprior non-users of the parent brand, and consequently on market share. The authors find evidence for potentialnegative reciprocal effects of unsuccessful extensions. In addition, the study shows that experience with the par-ent brand has a significant impact on extension trial, but not on extension repeat.

A s competitive pressures mount, brand marketers seekways to achieve growth while reducing the cost ofnew product introduction and the risk of new prod-uct failu re. One pop ular brand strategy is to attach an exist-ing brand name to a new product introduced in a differentproduct category, that is, brand extension. Such a strategy isoften seen as beneficial because of the reduced new productintroduction cost and the increased chance of success(Kapferer 1994). In add ition, a brand extension can producereciprocal effects that enhance or dim inish the equity of theparent brand.

Reciprocal effects research has focused on examiningattitudinal changes toward the parent brand. From a man-agerial perspective, it is interesting toexamine the impact oftr ial of an extension onchoice and market share of the par-ent brand. Extension trial should strengthen consumers'propensities toward buying the parent brand unless theextension experience is negative. This e ffect should be mostpronounced among consumers who have low levels of loy-alty toward the parent bra nd, because parent b rand sales arealready m aximized among highly loyal consumers. The roleof category similarity in moderating reciprocal effects hasalso been examined in an attitudinal context, but not in anactual purchase context. In addit ion, the reciprocal effect of

Vanitha Swaminathan Is Assistant Professor of Marketing, IsenbergSchool of Management, University of Massachusetts, Amherst. Richard J.Fox is Associate Professor of Marketing, Terry College of Business, Uni-versity of Georgia. Srinivas K. Reddy is Professor of Marketing, Terry Col-lege of Business, University of Georgia. The first author thanks Procter &Gamble's Innovation Research Fund for providing funding for this study.The authors also acknowledge the help of ACNielsen in providing the dataused in this study and the Coca-Cola Center for Marketing Studies at Uni-versity of Georgia. The authors also thank Chris Allen, Deborah Roedder-John, Gary Russell, Kevin Keller, and Seenu Srinivasan for their input inthe earlier stages of this research and the three anonymous JM reviewersfor their insightful comments on previous drafts of this article. The thirdauthor acknowledges research support through the Terry and Sanford Fel-lowship, Terry College of Business, The University of Georgia.

extension purchase across prior users and prior nonusers ofthe parent brand has not been studied.In addit ion to the potential benefits associated with pos-itive reciprocal effects, the use of brand extensions provideseconomies in securing trial in the marketplace, as noted pre-viously. In the words of Allan Maccusker, president of abrand consultancy group, "A lot of marketers are going this[the brand extensions] route because an established brandname will generate, hopefully, quicker trial by consumersand then heavier repurchase" (Advertising Age, p. 12). Theassumption underlying the use of the brand extension strat-egy is that extensions induce trial due to brand awarenessamong existing consumers. However, little research empiri-cally tests the role of brand extensions in inducing trial inthe marketplace. Therefore, another goal of this research isto examine the relationship between prior experience withthe parent brand and extension trial and repeat.

In summary, this research has three major objectives.First, we investigate the reciprocal effects of extension trialon parent brand choice among users and nonusers of the par-ent brand. Second, we examine the role of category similar-ity as a moderator of reciprocal effects. Third, we investigatethe impacts of experience with the parent brand on tr ial andrepeat of a brand extension.We address the objectives outlined through a series ofthree studies. In Study 1, we demonstrate positive reciprocaleffects of extension trial among prior nonloyal users andnonusers of the parent brand. In addit ion, we demonstratethat parent brand experience has a significant effect on exten-sion trial but not on repeat. In Study 2, we examine the roleof category similarity as a moderator of positive reciprocaleffects. In Study 3, we show the existence of negative recip-rocal effects associated w ith an extension product failure.

Researchers in the brand extensions area have relied p ri-mari ly on marketing experiments conducted in lab settingsin which consumers are typically provided with descriptionsof hypothetical brand extensions and are asked to providetheir instantaneous reactions (e.g., Aaker and Keller 1990;Keller and Aaker 1992). One l imitation of this type of


2/16

research is that effects may be overstated (Dacin and Smith1994). In addition, reciprocal effects develop over time andfrequently cannot be captured using this approach. Further-more, the extensions are hypothetical and not necessarilyindicative of what a firm would really consider doing.Finally, lab experiments such as those just described cannotcapture the impact of actual experience with a brand exten-sion on future purchasing in the parent category. Thus,although previous research offers valuable insights regard-ing brand extension strategies from a theoretical perspective,the use of these highly controlled lab experiments has, tosome extent, limited the usefulness of this research from amanagerial perspective. One exception is the research byErdem (1998), who examined household purchase data col-lected after the brand extensions had been introduced anddemonstrated that quality perceptions transfer betweenumbrella-branded products in the case of the companion cat-egories of toothpaste and toothbrushes. Another exception isthe research by Kim and Sullivan (1998), who model trialand repeat of a new brand introduction in the context of lineextensions. However, neither of these studies explicitlyfocuses on testing a framework of positive and negative rec-iprocal effects.

We examine the phenomena of both direct and recipro-cal effects across multiple brand extension cases by usingACNielsen scanner panel data to monitor household pur-chasing immediately before and after extension introduc-tion. In addition, we provide a conceptual framework toexamine factors that moderate reciprocal effects. An in-mar-ket study involving brand extension strategies and realbrands provides rich insights for managers interested inevaluating the risks and merits of extension strategies.All the brand extensions examined in this study repre-

sent national extensions of well-known brands. Althoughspecific brand names and categories are not revealedbecause of the proprietary nature of the data, we provideexamples and details of the categories to help interpret theresults. We examine reciprocal effects by modeling parentbrand choice, using extension brand experience as an inde-pendent variable. We develop logit models of extension trialand repeat using parent brand experience variables and otherrelevant purchase factors as independent variables todemonstrate direct effects. The article is organized as fol-lows: First, we present a conceptual framework andhypotheses. Second, we describe the data. Third, we presentan overview of the models along with the associated mea-sures; describe Studies I, 2, and 3; and discuss the key find-ings. In conclusion, we discuss the implications of thisresearch and directions for further research.

Conceptual FrameworkAs previously noted, there has been mixed support for theexistence of positive and negative reciprocal effects in theliterature. Keller and Aaker (1992) find that positive recip-rocal effects exist only when an average-quality parentbrand introduces a successful extension, Keller and Sood(2000) posit that evaluations of parent brands that arealready well regarded will not change significantly as aresult of favorable extension experience. Gurhan-Canii and

Maheswaran (1998) show that enhancement effects exist forbrand extensions that are similar to the parent brand.In general, previous research finds support for the mod-erating role played by category similarity in influencing bothpositive and negative reciprocal effects, Milberg, Park, andMcCarthy (1997) show that negative reciprocal effects canoccur when e xtension sim ilarity is extremely low. Keller andSood (2000) demonstrate that negative reciprocal effects canalso occur when the extension is highly similar to the parentbrand, Gurhan-Canli and Maheswaran (1998) show that dilu-tion of a family brand name occurs in response to incongru-ent and negative information, particularly w hen the extensionis similar to the parent brand. The evidence regarding theexistence of negative reciprocal effects at the brand attributelevel is also considerable (e,g,, Loken and Roedder-John1993; Roedder-John, Loken, and Joiner 1998), However, it isless clear whether negative reciprocal effects exist at theoverall attitude level (Keller and Aaker 1992),To illustrate the process of reciprocal effect formation inan actual choice setting, we present a hypothetical example.Suppose Nivea, known for its skin-care products, introduces a

cosmetic product under the Nivea brand nam e. The new prod-uct is tried by a group of consum ers w ho are heterogeneous intheir prior experience with Nivea skin care products. Assum-ing that the new entry is a success, the trial of the product isgenerally a favorable experience. The trial of the Nivea cos-metic furnishes new infonnation regarding the Nivea brandname to both prior users and prior nonusers. Consistent withprevious work in the context of product experience, such asHoch and Deighton's (1989) and Kempf and Smith's (1998),the learning provided by the product experience will lead tostrongly held beliefs regarding the Nivea cosmetic. Previousbrand literature has viewed brand knowledge as a network ofbeliefs and associations (Roedder-John, Loken, and Joiner1998), Therefore, the beliefs regarding the Nivea cosmeticproduct are transferable to the Nivea skin care brand.However, two conditions must be present for the transferto occur. First, the extension information must be deemedrelevant in the parent category. Previous research has identi-fied category similarity between the extension and parentcategories as necessary for the extension infomiation to beconsidered relevant, a condition satisfied in our Nivea exam-ple (Keller and Aaker 1992; Loken and Roedder-John 1993).Second, for this transfer to occur, the beliefs about the par-ent brand must undergo a change, Roedder-John, Loken,and Joiner (1998), in the context of flagship products, sug-gest that the network of beliefs linked to the flagship prod-uct tends to be extreme, strongly held, and resistant tochange because of the accumulated exposure and experiencewith the flagship product. The discussion regarding flagshipproducts is relevant to our analysis of the core parent brand.The beliefs associated with the core parent brand are likelyto be of varying strength across different segments of con-sumers. Among segments of consumers that already have ahigh level of loyalty toward the Nivea skin care brand andhave a well-developed set of associations regarding thisbrand, the provision of new information regarding Niveacosmetics is unlikely to produce a significant change. Thisis particularly true if the new information does not signifi-cantly contradict the existing knowledge structure.


3/16

The consumers with a high degree of loyalty toward theNivea skin care brand are likely to have beliefs and associa-tions that are resistant to change. According to Roedder-John, Loken, and Joiner (1998), these may be viewed asbrands that have a well-developed memory structure as aresult of frequent exposure and direct experiences. Again,consistent with Roedder-John, Loken, and Joiner's (1998)arguments, the accumulated exposure and direct experiencesamong prior users of the parent brand make the parent brandbeliefs of highly loyal prior users less resistant to change.Conversely, among segments of consum ers with moderate tolow loyalty toward the parent brand, the less frequent expo-sure to the parent brand suggests that these consu mers' par-ent brand beliefs and associations are likely to be moreamenable to change.' Therefore, the potential for a positivereciprocal effect is strongest among segments of consumerswith low to moderate loyalty toward the parent brand.Although th is scenario applies to a successful brandextension, another possibility is negative reciprocal effectsdue to unsuccessful brand extension. Sup pose that the Niveacosmetic is discontinued in the marketplace because of low

market share. Therefore, it is likely that extension trierswere generally not favorably disposed toward the new prod-uct and that trial furnishes negative or at least neutral infor-mation regarding the brand. Among prior users, the provi-sion of new negative information regarding an extensionproduct is likely to contradict existing knowledge structures.

'This is not to discount the possibility that there may be negativeassociations among prior nonusers. Because prior nonusers aredefined as people who have not tried the parent brand in a one-yeartime period prior to extension introduction, any negative experi-ence with the parent brand occurred at least one year before exten-sion introduction. Following Feidman and Lynch (1988), we notethat the passage of time should weaken information accessibility,which in turn may weaken the potential negative associationsamong prior nonusers or highly infrequent users.

particularly among consumers with high levels of prior loy-alty toward the parent brand. However, because the propen-sity to purchase the parent brand among prior nonusers isalready zero, the provision of new negative informationcannot result in a negative reciprocal (behavioral) effectamong these consumers. Therefore, negative reciprocaleffects of extension failure can be observed only amongprior users, especially those with high loyalty toward theparent brand.In summary, the introduction of a brand extension that issuccessful is likely to result in positive reciprocal effects.These positive reciprocal effects are likely to be moderated bycategory similarity; there will be stronger reciprocal effectsunder conditions of high category similarity. Conversely, neg-ative reciprocal effects may be associated with a failed exten-sion product. These negative reciprocal effects are likely to bestrongest under conditions of high category similarity. Wetherefore propose that positive and negative reciprocal effectsof extension trial on parent brand c hoice exist and are moder-ated by category similarity. Further positive (negative) recip-rocal effects in the context of purch ase behavior are limited to

consumers with low to moderate loyalty toward the parentbrand (users of the parent brand). A summary of the proposedeffects is presented in Figure 1. In addition, experience withthe parent brand is likely to increase the propen sity to try theextension but not to repeat. In the next section, we outline thefirst of three studies designed to test this framework.

Study 1Reciprocal Effects of Brand ExtensionIntroduction on the Parent BrandAs noted previously, research regarding both positive andnegative reciprocal effects has been somewhat mixed. Ceil-ing effects associated with well-regarded brand names, dif-ficulties associated witb examining attitudinal sbifts in

FIGURE 1A Framework of Positive and Negative Reciprocal EffectsCategory Similarity

High Low

SuccessfulBrandExtension

Unsuccessful

Prior NonlovalUsers of ParentBrandPositive

PriorUsers of ParentBrandNegative

Prior Nonusersof Parent Brand

PositivePrior Nonusersof Parent Brand

No effect

Prior NonlovalUsers of ParentBrandNo effect



No effectPrior Nonusersof Parent Brand

No effect


4/16

experimental settings, and the omission of" prior usage as amoderating variable may account for some of these mixedfmdings in past research. Consumers are likely to be hetero-geneous in their purchasing of a brand. It is not possible toenhance purchase probability of a brand among highly loyalusers of the brand, simply because propensities to purchasethe brand are already so high. This "ceiling effect" argumentis consistent with the prior research fmdings that point outthat evaluations of well-regarded brand names generally donot change on account of exposure to favorable extensioninformation (Keller and Aaker 1992), However, amongusers with moderate to low levels of loyalty, the exposure tothe extension brand may induce positive reciprocal effectsby enhancing brand familiarity, strengthening brand atti-tude, and ultimately increasing the likelihood of purchasingthe parent brand.For prior nonusers, extension trial provides new infor-mation regarding a brand. Information generated from prod-uct trial typically results in increased brand recall andstronger brand attitudes, which in turn have a powerfulimpact on parent brand evaluation and purchase (Kempf and

Smith 1998; Smith and Swinyard 1982; Wright and Lynch1995), Thus, increases in memory and familiarity and attitu-dinal shifts resulting from the extension trial experienceshould induce reciprocal effects for the parent brand amongprior nonusers. Consistent with these arguments, we expectthat positive reciprocal effects of extension trial will beobserved among prior nonloyal users and nonusers.Impact of Parent Brand E xperience on ExtensionTrialConsumer behavior research suggests that information orlearning gathered from product usage is often granted a specialstatus by consumers (Hoch and D eighton 1989; Kempf andSmith 1998; Smith and Swinyard 1982), Information gatheredfrom personal experience is more vivid and therefore morememorable (Kempf and Smith 1998), Because infomiationfrom product experience is self-generated, it is deemed moretrustworthy than information gathered from advertising orcom mu nications, w hich results in strongly held beliefs (Smithand Sw inyard 1982), Thu s, consumers with parent brand expe-rience have greater parent brand knowledge, better recall of theparent brand, and greater confidence in their beliefs about theparent brand than consumers with no parent brand experience.It has also been suggested that an existing brand name providesan assurance of quality, thereby reducing the risks involved inpurchasing a new product (Erdem 1998; Wemerfelt 1988),Although both the direct effects of parent brand on theextension and the reciprocal effects are cross-categoryeffects, brand extension researchers typically examine theseas distinct processes. Whereas the former refers to formationof brand evaluations for a new brand introduction, the latterrefers to changes in the evaluations of an existing brand. Onthe basis of these arguments, we hypothesize that parentbrand experience increases the likelihood of extension trial.Impact of Parent Brand Experience on RepeatPurchase BehaviorInformation economics theory suggests that the quality ofthe product is unambiguously revealed during product use

(e,g,. Nelson 1970, 197 4), This is especially true when theproduct quality can be gauged accurately even with a singleexposure to the product,^ Thus, when the extension has beentried, the repeat purchase decision should depend on the evi-dence furnished by the trial experience rather than parentbrand experience. In addition, the familiarity with the exten-sion among both past users and nonusers of the parent brandis likely to be similar after the extension has been tried.Therefore, we expect that parent brand experience has noimpact on repeat purchase of the brand extension.DataThe data for this study, obtained from ACNielsen, providehousehold p urchase histories for selected product categoriesfor a national panel in the time period 1990-94, We usedpurchase data for three brand extensions introduced duringthis time period to test the hypotheses. Household purchasehistories in the parent categories are available for approxi-mately one year preceding the extension introduction andfor one year following extension introduction. For each ofthe extension categories, data are available for one year fol-lowing extension introduction. Given the nature of the cate-gories examined, such as foods and personal care items, webelieve that one year is sufficient to capture variations inpurchase behavior d ue to the extension. The fictitious namesAlpha, Gamma, and Zeta are introduced to label the threeparent brands and their respective extensions. Alpha refersto a large personal care brand extended into a related per-sonal care category. An example of this type of extensionmight be Nivea introducing Nivea beauty soap. Gammarefers to a food product extended into a related food cate-gory. An example of this type of extension is Hershey'sintroducing Hershey's chocolate milk, Zeta is a personalcare brand extended into another personal care category(different from A lpha), An exam ple of this type of extensionis Dial bar soap introducing Dial deodorant. All three exten-sions leveraged the parent brand positioning in the extensioncategory.

Criteria for qualifying hou seholds. Households in thepanel enter and leave continuously. To ensure comparabilityin preextension and postextension introduction purchases,consistencies in time available for trial, and so forth, weconstructed a "static" panel of participating households foreach study. For each of the three cases, we included a house-hold in the panel if it recorded at least one purchase in eitherof two frequently purchased categories during the first sixmonths and during the last six m onths of the particular two-

^There may be conditions in which the intluence of parent brandexperience goes beyond the initial trial of the extension. This maybe the case when products contain experience attributes whosequahties can only be gauged with repeated exposures to the prod-uct. This may also be true in the case of credence goods whoseproduct quality is impossible to evaluate even after consumers haveexperienced the product. Other conditions in which a confirmatorybias may apply may include highly technical products that requirea great deal of expert knowledge and image products (Hoch andDeighton 1989), This study is restricted to frequently purchasedpackaged goods products for which these conditions are not likelyto apply.


5/16

year study period. To eliminate households that are infre-quent users, we further required households to have made atleast two purchases in both the parent and the extension cat-egories during the one-year period following extensionintroduction and two purchases in the parent category in theone-year period before extension introduction.Sometimes, an extension rollout takes place over a rela-tively long time period, so an important consideration is the

date of extension introduction in a market. We identified thedate on which the first purchase of the extension brand wasmade in a market, and we used the week prior to this date asthe introduction date for the extension brand in that market.To balance the time available for purchasing before and afterextension introduction, we excluded markets where theextension introduction took place either very early or verylate."*Descriptive DataTable 1 contains infonnation regarding the characteristics ofboth the parent and extension categories for the brandsAlpha, Gamma, and Zeta. Also provided are data regardingmarket share and share of voice based on percentage ofadvertising expenditure for the parent brand categoriesbefore extension introduction and for the extension brandsfollowing extension introduction. The data provided inTable 1 give an overview of the types of product categoriesand the nature of competition in the categories used in thisstudy. In addition, we use the information from Table I togain insights into our fmdings.

^For brand Alpha, the dates of introduction varied over a 46-week period. In balancing the time period before and after exten-sion introduction, we were able to retain 50% of the markets. Forthe brands Gamma and Zeta, we retained approximately 80% ofthe markets.

Reciprocal Effects of Extension Trial on ParentBrand ChoiceModel development atid measures. A model of house-hold choices in the parent category demonstrating that sig-nificant changes in the likelihood of parent brand purchas-ing are observed after the trial of the extension in theextension category would provide strong evidence of theexistence of reciprocal effects. We examine reciprocal

effects of extension trial using a binary logit model in whichthe dependent variable is parent brand choice.'' The unit ofanalysis is an individual choice occasion. Therefore, thereare multiple observations for each household. The modelincludes household heterogeneity, marketing-mix effects,and the effects of competition as independent variables. Weestimate the model using purchase data for the parent cate-gory for the static panel in the time period following exten-sion introduction. We describe the variables used in the logitchoice model next.

"The multinomial logit model has been widely used in previousresearch (Guadagni and Little 1983; Kamakura and Russell 1989)to model brand choice behavior and capture price elasticities andchanges in market structure. The multinomial logit model allowsfor the capture ot competitive marketing-mix effects. However, inour context, because the focus is one brand introducing a brandextension, the binary logit approach with the choice of the parentbrand as the dependent measure is a parsimonious approach tomodeling the choice behavior in the parent category. The binarylogit is also preferred becau se the loyalty coefficient in the multi-nomial logit does not enable u s to estimate the incremental effectof the brand extension on choice of the parent brand over andabove the impact of parent brand loyalty. This is because the loy-alty coefficient is a general co eftlcient for all brands in the categoryand not specific to the parent brand. Because it is important toassess this incremental effect in the context of this research, thebinary logit with the choice of the parent brand as the dependentvariable is better suited to the purposes of this study.

TABLE 1Key Characteristics of Parent and Extension Ca tegories: Study 1

VariablesCategory description

Number of brandsAverage interpurchasetime (in weeks)Strength of the brand^

Market sharedShare of voiced

ParentCategoryPersonal

caret 8

19.871

22%14 %

AlphaExtensionCategoryPersonal

care25

11.301

5%1 5 %

GammaParentCategoryFood

45

6.546

12 %3%

ExtensionCategoryFood

7

10.594

4%4 %

ParentCategoryPersonal

care25

11.301

6%3%

ZetaExtensionCategoryPersonal

care22

13.119

1 %2 %

^Figures for market sh are and sh are of voice for the parent brand refer to the time pe riod prior to extension introduction. For the extension brand ,the figures reflect the market share and share of voice one year after the extension was introduced,b obtained from various issues of Market Share Reporter (19 90-95).of voice figures were obtained from LNA/Mediawatch Multimedia S ervice (1990-95).


6/16

Parent brand experience (EXP). Prior experience withthe parent brand is operationalized as a loyalty measure, thatis , the frequency of purchasing the parent brand comparedwith the other purchases made in the category. This relativefrequency measure is consistent with those in previous stud-ies that measure brand loyalty (Russell and Kamakura1994). This variable is constant for all choices made duringthe time period following extension introduction.-"*

Relative price (RELPRI). We calculated the relativeprice per gram of the parent brand by indexing the parentbrand's price per gram to a weighted average ofthe previous14-day average price per gram or ounce for the majorbrands; weights were determined by market shares.^Because data on competitors' prices are not available withevery purchase, this is an approximate measure of competi-tive pricing.

Reciprocal effects indicator variable (IND). The recip-rocal effects indicator is a dummy variable that indicateswhether the extension was purchased on or before the datewhen a purchase in the parent category is made. This indi-cator variable is a "switch" that goes "on" (takes the value Ias opposed to 0) when the extension trial takes place.

Displays (DISP) and advertisements (AD). The displayand feature advertisement variables were also dummy (0/1)variables tbat indicated the presence of displays or featureadvertisements associated with the parent brand.

Reciprocal effects logit model. Let X = 1 or 0 dependingon whether or not the parent brand is chosen on a categorypurchase occasion. The probability that the parent brand ischosen is given by(1)

3|(EXP ) + |3,( RELPRI) + P,(IND)P(X = I) = -whicb can be rewritten as fol lows:(la) ln[P(X =

|3,( AD )

where a and Pi t h rougb ^5 are pa rame te r s to be est imated.Overal l model signif icance and predictive validity are

judged, respect ively, by m e a n s of the l ikelihood ratio test

^Heterogeneity has received increased attention in the modelingliterature (Chintagunta, Jain, and Vilcassim 1991) because esti-mates of parameters of discrete choice models that ignore hetero-geneity are likely to be biased and inconsistent (Hsiao 1986). Het-erogeneity am ong households is incorporated in our model throughparent brand experience (EXP).^On eof the limitations of using market shares of the major com-petitors as weights is that the heterogeneity in consideration setsacross households is not captured. Given the dynamic nature ofconsideration sets across time and over purchase occasions(Andrews and Srinivasan 1995) and the need for information onmarketing-mix variables (e.g., exposure to television advertising)to capture heterogeneity in consideration sets accurately, we didnot incorporate this heterogeneity into the assessment of relative

price. We acknowledge this as a limitation in the measurement ofrelative price. We thank the reviewer who pointed this out.

statistic (}(2) and the classification accuracy (as j udged bythe percentage correct ly c lassi f ied) . ' ' Because of unequalgroup s i ze s , the percentage correct ly c lassi f ied is comparedwith a benchmark based on a proport ional chance cr i ter ion(Morrison 1969).^

The e s t ima te s for the rec iprocal effec ts of ex tens iontrial for b r a n d s A l p h a , G a m m a , and Ze ta are presen ted inTable 2. Reciprocal effec ts cannot exist at very high loy-alty levels (e.g., for househo lds wi th a 100% l oya l ty ) .The re fore , the ana lys i s was rest r ic ted to househo lds wi th aparent brand loyal ty of less than 80% but greater thanze ro .9 Pr io r nonuse rs were ana lyzed sepa ra t e ly . Amongextension t r iers , only households that have had the o p p o r -tunity to p u r c h a s e in the parent ca tegory at least once fol-l owing the trial of the ex tens ion were inc luded in theana lys i s . Because the impac t of the ex tens ion in t roduc t ionwas expec ted to vary on the bas i s of pr io r usage , theana lyses for prior users and pr io r nonuse rs are presen tedsepa ra t e ly .

In the case o f the b rand Alpha , d i sp l ays and advert isinginformat ion were not inc luded, because these fac tors wereact ive in less than 1% of the purchases made in this cate-gory. In the case of the brand Zeta , the display informat ionis not inc luded for the same reason . The reciprocal effectvariable (IND) is significant at the 1% risk level for exten-s ions Alpha and Zeta , but it is not significant for brandG a m m a . The odds ra t io for tbe reciprocal effect variable is16 in the case of the brand Alpha and 52 in the case of thebrand Zeta . 'O Tbis suggests tbat in the case of the brandAlpha, extension t r ia l enhances the o d d s of purchas ing theparent brand as opposed to some other brand by 16 t imes .Similar ly, in the case of the brand Zeta, extension trialenhances the odds of purchas ing the parent brand by 52t imes. Across all three extensions, the parent brand experi -

''Typically, a holdout sample is used to avoid bias in estimatingthe predictive accuracy ofa model. The logistic regression programin SAS uses a "jackknitlng" approach that omits the observationson e at a time and classifies them as "events" or "nonevents" on thebasis of the model, which is estimated without the observationsbeing classified. Because theSAS procedure provides an unbiasedmeasure of predictive validity, we do not use a holdout sample toassess predictive validity (see SAS Institute Inc. 1983, p. 45).^The proportional chance criterion is based on the hit rate, a- +(I - a)2, where a is the percentage of times the parent brand waspurchased.'The argument for truncating the loyal consumers at those withless than 80% loyalty is a conceptual one. Among perfectly loyalconsumers, it is not possible to observe increases in the probabilityof purchasing theparent brand, because loyalty is already so high(a ceiling effect). An analysis of the distribution of frequenciesacross the various loyalty levels showed that across all categories,an 80% level of loyaity cutoff appeared todiscriminate reasonablybetween the perfectly loyal group and the less than loyal group,hecause of a concentration of households around the 80% cutoffpoint across categories. We also did sensitivity analyses to ensurethat choosing alternative cutoffs, such as 75%, 85%, and 90%, didnot change the results significantly.lOThe odd s ra tio for the brand Zeta is a large number because thebrand has a relatively small market share in the parent category.Therefore, theprior probabilities of purchasing this brand are low


7/16

TABLE 2Reciprocal impact of Extension Trial on Parent Brand Choice

VariableConstant

Parent brand experience (EXP)

Relative price (RELPRI)

Reciprocal effects indicator (IND)

Advertisements (AD)

Displays (DISP)

Sample sizeChoice-2 Log L y}

Percentage correctly classifiedProportional chance

Alpha-.111(.485)1.749*5.751(.762)- .798*.450(.240)2.795*16.359(1.054)_

1769740.158(p = .000)

67%5 0 %

Prior UsersGamma

- .743*(.328)4.563*95.938(.236)- .710*.492(.312)

.1421.152(.085)

.730*2.074(.280)

- .308*.735(.167)45461726

554.331(p = .000)69%52%

Zeta-2.516*(.485)

4 2 2 4 *88.340(.253).2141.239(.538)

3.955*52.180(.399).0921.097

(.538)

2840769581.295(p = .000)

79 %6 0 %

PriorAlpha

-2.732*(.409)

- .374.688(.319)1.451*4.269(.532)

26 11946.118(p = .000)

93%87%

NonusersGamma

- .459(.165)

"- .709*.492(.159)- .139.091(.869)

.424*1.528(.127).032.085(1.032)

19,415454030.838(p = .000)

77%64%

*Significant at the p < .01 level.Notes; Figures in boldface represent odds ratios; figures in parentheses represent standard errors.

ence variable is signiticant at the 1% risk level." In the caseof Gamma and Zeta extensions, the parent brand experiencevariable is the most infiuential of all the variables. In thecase of the Alpha extension, the impact of the reciprocaleffects indicator variable appears to be more infiuential thanparent experience or relative price. Across all three exten-sions, the overall model is significant at the 1% level. Theincremental percentages correctly classified using the modelover and above the proportional chance criterion for thebrands Alpha, Gamm a, and Zeta are 17%, 17%, and 19%,respectively.Reciprocal effects were also examined among priornonusers of the parent brand, (i.e., EXP = 0). The results forextension trial among prior nonusers of the parent brand arealso presented in Table 2. The limitations imposed on house-

holds to be included in the reciprocal effects model, that is,at least one purchase in the parent category following exten-sion trial, resulted in a limited sample in the case of thebrand Zeta, which precluded analysis of prior nonusers inthis case. The impact of extension trial on parent choice issignificant in the case of brand Alpha but is not signiticantfor brand Gamma. Reciprocal effects of brand extensionintroduction were examined in terms of market share for amacro perspective. The parent brand market shares werecompared before and after extension introduction amongextension triers (see Table 3).'2Significant increases in market share were observedamong prior nonusers for all three cases and among priornonloyal users for brand Alpha. In the case of brand Zeta,there were indications of share increases among parentbrand users, though this difference was not significant.

"Note that the parent experience variable is expressed in termsof the frequency of purchasing the parent brand relative to all pur-chases made in the parent category. EXP is therefore a fraction thatassumes a value between 0 and 1. The reason that the odds ratiosare so large is that they represent changes in parent brand experi-ence from 0 to I, that is, I unit of the independent variable.

12 We conducted a t-test to examine differences in market sharebefore and after extension introduction. Because the shares weredrawn from the same sample, a test for differences must accountfor correlations in the shares. This procedure is outlined byCochran(1977).


8/16

TABLE 3Percentage Change in Parent Brand MarketShares Before and After Extension Introduction:Means and Standard Deviations of Differences

ExtensionAlpha

Gamma

Zeta

Prior Nonusersand ExtensionTriers(% Change)+ 8 . 1%* *(.040)

+9.7%**(.020)

+7%**(.023)

Prior Usersand ExtensionTriers{% Change)+13%*(.090)

-4.5%ns.(.043)

+2.5%"s.(.028)

* p < . 0 1 ."p < .05.Notes: Figures in parentheses represent standard deviations of thedifferences, n.s. = not significant.

These results are consistent with the previous findingsregarding the relatively weak reciprocal effects in the case ofprior nonloyal users of brand Gamma.Impact of P arent Brand Experience on ExtensionTrial and Repeat

Model devetopment and measures. In this section, theimpact of parent brand experience on extension trial andrepeat is quantified. Dichotomous dependent variables areintroduced on the basis of classifying each household in thestatic panel as an extension trier or a nontrier (for the trialmodel) and classifying triers as either repeaters or nonre-peaters (for the repeat m odel). Thus, each household consti-tutes one observation for the trial model, each trier house-hold constitutes one observation for the repeat model, andthe information for each of the independent variables is cap-tured at the household level, as opposed to the transactionlevel as in the previous analysis. The independent variablesin the models are discussed next. Values for these variableswere obtained through household purchase histories beforeextension introduction.

Parent brand experience (EXP). As discussed previ-ously, parent brand experience is operationalized in terms ofrelative frequency of buying the parent brand in the parentcategory.

Deat proneness (DPRO NE). Deal proneness is definedas the degree to which a consumer is influenced by salespromotion. We measured active deal proneness using Web-ster's (1965) deal proneness measure. For each householdand for each brand in a given household's purchase history,we calculated the difference between the percentage oftimes a given household used a coupon when pu rchasing theparticular brand and the average percentage of times a brandwas bought using a coupon across all households. The over-all deal proneness measure is a weighted average of this dif-ference for the brands bought by a household; the weights

are market shares of the brands for the given household.Because information regarding deal proneness in the exten-sion category before extension introduction was not avail-able, we used the household's deal proneness in the parentcategory as a proxy for deal proneness in the extension cat-egory. This is consistent with prior research that suggeststhat there is a generalized deal proneness construct thatresults in a correlation in deal proneness across categories(Bawa and Shoemaker 1987).Category experience (TOTCAT). Prior research suggeststhat the effect of brand knowledge may be different if the con-sumer is regarded as an expert rather than a no vice in the con-text of the extension category (Broniarczyk and Alha 1994;Smith and Park 1992). The frequency of purchasing in a cate-gory is an indicator of the knowledge or expertise in a category(Alba and Hutchinson 1987). The num ber of purchases in theextension category after extension introduction was introducedas a third variahle that influenced extension trial and repeat.

Triat modet. Let T = 1 or 0 depending on whether or noa household purchases the extension. The equation for theprobability of trial is(2) P(T = I) = P| (EXP) + PjCDPRONE) + P3(T0TCAT)a + P I (EXP)+ (),(DPRONE)+ P3(T0TCAT)which can be rewri t ten as fol lows:(2a) ln[P(T = I )/P(T = 0)] = a + Pi(EXP ) + PjCDPRONE)

where the left-hand side represents the log odds ratio. If acoefficient, P |, (J , or p j, is significant, the co rrespo ndingvariable has an impact on the log odds ratio.Table 4 presents results of the logistic regression analysesfor the trial and repeat models. The overall trial model is sig-nificant across all three extensions (as shown in Table 4). Thepercentage co rrectly classified is 77% for the brand Alpha,and the incremental classification over and above a propor-tional chance criterion is 12%. In the case of the brandGamma, the percentage correctly classified is 75%, and theincremental percentage correctly classified is 13%. In the caseof the brand Zeta, the percentage correctly classified is 90%,and the incremental percentage correctly classified is 8%.As Table 4 shows, the parent brand experience variable(EXP) is significant across all three extensions at the 1%risk level, which is consistent with expectations. The oddsratio for this loyalty variable is 1.932 for the brand Alphaextension,'* 1.800 for the brand Gamma, and 6.752 for thebrand Zeta. The other two variables, deal proneness and

'3|n addition to loyalty, another aspect of parent brand experi-ence is the time-varying nature of parent brand preference. Consis-tent with Bucklin and Lattin (1991), we created the last brandbought in the parent category before extension introduction as analternative parent brand experience measure. We created a dummyvariable labeled LP, which takes on a value of I if a household'slast purchase in the parent category before extension introductionwas the parent brand and a value of 0 otherwise. The trial andrepeat models, which we estimated incorporating the LP (last pur-chase dummy variable) instead of the EXP variable, yielded simi-lar results.


9/16

TABLE 4Impact of Parent Brand Experience on Extension Triai and Repeat

VariabieINTERCEPT

Parent brand experience(EXP)

Deal proneness(DPRONE)

Total category experience(TOTCAT)

Sample sizeTotalTriers

Percentage correctProportional chance-2 Log L x2

Alpha-1 .650*(,141)

,659*1.932(,294)1,078*2.938(,241)

,034*1.034(,017)

99523 37 7 %6 5 %

27,861(p = ,000)

Extension TrialGamma-1.910*(.094)

,588*1.800(.205),420*1.522(,099),108*1.114(,010)

242863 5

75%6 2 %

149,09(p = ,000)

ModeiZeta

-3 ,69 4(,113)1,910*6.752(,267)

2,465*11.762(,204),138*1.148(011)

449644 7

90%82%

338,829(p = ,000)

Alptia- ,376(,263).5241.688(.555)

- .211.810(-348).0321.033(,030)

218101

5 2 %5 0 %

2.489(p = ,477)

Extension RepeatGamma-1 .362(.204),3951.484(.204),1941.215(,422),092*1.097(017)

57 321 76 5 %52 %

36,675(p = ,000)

ModeiZeta

-1 ,029(.169)

- .491.612(,818).0481.049(,389),183*1,201(.065)

38 2125

6 9 %6 6 %

8.737(p = ,033)*Significant at the p < ,01 level.Notes: Figures in boldface represent the odds ratio; figures in parentheses represent standard errors.

total extension category experience, are also significantacross all three extensions at the 1% level.

Repeat model. Our second hypothesis examines theimpact of parent brand experience on extension repeat. Totest this hypothesis, using only extension triers, we fit alogistic regression model to the data, which used the sameset of independent variables as in the trial model but hadextension repeat as the dependent variable. In addition, weconstrained the sample to ensure that only households thatpurchased at least once in the extension category after trialof the extension were included in the sample. The results ofthe mod els are also presented in Table 4, The overall modelis not significant in the case of the brand Alp ha, and no indi-vidual variable is significant. For the Gamma and Zetaextension s, overall category experience (TOTCAT) is signif-icant at the 1% level, and the corresponding overall modelsare significant. Most important, in all three cases, the parentbrand experience variable is not statistically significant. Inother words, as expected, parent brand experience does notaffect repeat purchasing of a brand extension.

Summary and discussion. For brand Alpha, extensiontrial positively affected the propensity to buy the parentbrand among both prior nonloyal users and prior nonusers.For brand Zeta, the analysis was restricted to prior users andsupports a positive reciprocal effect of extension trial. In thecase of the brand Gamma, no support for the hypothesizedreciprocal effect of extension trial is observed among either

prior users or nonusers. The analyses also strongly supportthe positive impact of parent brand experience on extensiontrial but not on extension repeat across all three extensions,Tberefore, from Study I, we conclude the following:

There can be positive reciprocal effects of extension trial onparent brand choice among both prior users and prior nonusers,Reciprocal effects of extension trial can result in market shareincreases for the parent brand, especially among priornonusers of a brand,Parent brand experience has a significant impact on extensiontrial,Parent brand experience does not have an impact on extensionrepeat.The lack of reciprocal effects in the case of the Gammaextension indicates the existence of factors that moderatereciprocal effects. Previous research suggests that perceivedfit, category similarity between the parent and extensionbrands (e,g,, Aaker and Keller 1990), and relevance of theparent brand associations in the extension category (e,g,,Broniarczyk and Alba 1994) moderate cross-categoryeffects. To investigate the role of category similarity andbrand association relevance in this study, a survey of studentconsumers was undertaken for the three extensions used inthis study. The measures used in the survey included (1) per-ceived similarity between parent and extension categories,(2) the relevance of parent brand associations in the exten-sion category, and (3) overall perceived fit. Subjects (n = 54)


10/16

were asked to provide ratings of category similarity, overallperceived fit, and relevance of parent brand associations forthe extensions used in this study.The results reported in Table 5 show that tbe average rat-ings of overall perceived fit and extension association rele-vance for the three brands vary little (between 6,06 and 6,34and between 2,32 and 2,63, respectively). However, as Table5 shows, category similarity for the Gamma extension is farlower than the category similarity ratings for the other twoextensions. The lower category similarity may account forthe lack of significant reciprocal effects in the Gammaextension.

Another possible explanation for the lack of significantreciprocal effects in the case of the Gamma extension maybe the nature of the parent category. This category is char-acterized by low interpurchase times, a relatively large num-ber of brands (see Table 1), and a great deal of impulse buy-ing. Therefore, it is possible that, in this category, marketshare increases are obtained by short-term promotionalactivities rather than by long-term investments in brand-building activities, so that brand extensions are unlikely toproduce reciprocal effects for the parent brand. We examineeach of tbese alternative explanations in Study 2.

Study 2Two factors may con tribute to the lack of reciprocal effectsin the case of the Gamma extension: relatively low categorysimilarity between the parent and extension categories andthe unique nature of the parent category. One way to rule outthe category itself as the explanation for the lack of recipro-cal effects would be to demonstrate that other extensions ofthe same parent brand generate reciprocal effects. Also,greater evidence for the role of category similarity in mod-erating reciprocal effects is necessary to substantiate the

argument that category similarity may have contributed totbe previous result regarding the Gamma extension.The need to control for differences in parent brand andcategory characteristics made it important for us to focus onother extensions introduced by the same parent brand at var-ious levels of category similarity to the parent brand. Thisalso enables us to gain further insights into the role of cate-gory similarity as a moderator of reciprocal effects. TheGamma brand introduced two additional brand extensionswithin three years of introducing the first extension,''' Bothextensions, introduced within one year of each other, werereasonably successful, in that they had gained noticeablemarket share in their extension categories by the end of thefirst year after introduction,

A second survey of student subjects (n - 55), incorpo-rating the same measures of fit, relevance, and similarity asdescribed in Table 5, was conducted. Subjects were asked torate the original Gamma extension (Gamma 1) and the twolater extensions (Gamma 2 and Gamma 3) on these mea-sures. As shown in Table 6, the Gamm a 2 extension categoryis perceived as more similar to the parent category than theoriginal Gamma extension but does not reach the similaritylevels of Alpba and Zeta, In contrast, the Gamma 3 exten-sion category is perceived as less similar to the parent brandthan the original Gamma extension. Therefore, if categorysimilarity is a moderator of reciprocal effects. Gamma 2 ismore likely than Gamma 3 to produce positive reciprocaleffects. The ratings for the Gamma 1 extension were not sig-nificantly different from the ratings provided by subjects in

''No other brand extensions of the Gamma brand were intro-duced in the three-year time period separating the introduction ofthese extensions.

TABLE 5Survey of Perceived Fit: Study 1Alpha Gamma Zeta

Category similarity

Relevance of key parent brandassociation to extension^Overall perceived fit

4.75a(1.25)b2,39(1,81)6,34(,98)

Aipha VersusGammap.10p< .05

2,81(1.36)2,63(2,07)6,06(1.06)

t-TestsGamma VersusZeta

p , 1 0

4,47(1.64)2,32(2,04)6,06(.97)

Aipha VersusZetap = .1 Op > . 1 0p < . 1 0

Category similarityAssociation relevanceOverall perceived fit^Numbers represent means,bNumbers represent standard deviations,^Subjects were asked to name the first parent brand associations that came to mind and rate the relevance of the first three associations in theextension category. The most commonly mentioned association across consumers was used in the calculation of extension association rele-vance. E xtension association relevance is reverse-cod ed, where 1 = "very relevant" and 7 = "very irrelevant,"


11/16

Category similarity

Relevance of key parent brandassociation to extension^Overall perceived fit

TABLE 6Survey of Perceived Fit: Study 2

Gamma 12.9ia(1.33)b2.51(2.18)6.14(.84)

Gamma 23.77(1.51)3.66(1-65)4.92(1.49)

Gamma 32.08(1.34)4.53(2.01)4.38(1.66)

t-Test(Gamma 2 VersusGamma 3)p


12/16

TABLE 7Reciprocal Impact of Extension Trial on Parent Brand Choice: Study 2

VariableConstant




Advertisements (AD)

Displays (DISP)

Sample sizeChoice-2 Log L y}

Percentage correctly c lassifiedProportional chance

PriorGamma 2High Similarity

-1 .092*(051)2.775*16.051(.050)- .883*.439(.034)

.150*1.163(.024)

.082*1.086(.025)

.097*.907(.024)120,61619,822

3612.540(p = .000)86%72%

UsersGamma 3Low Similarity

-2 .050(.096)2.396*10.983(.068)- .025.975(.066)

.0521.053(.032)- .102.902(.060)

.117*1.124(.046)45,7479169

1231.822(p = .000)80%68%

PriorGamma 2High Similarity

-1 .152*(.099)

- .862*.422(.068).571*1.771(.063).0731.076(.049)

- .009.991(.046)52,7324541

249.783(p = .000)9 1 %84%

NonusersGamma 3Low Similarity

-2 .619*(.153)

- .273*1.314(.108).309*1.362(.060).1001.106(.081).0361.036(.074)

29,611296533.123(p = .000)

90%82%

"Significant at the p < .01 ievel.Notes: Figures in boidface represent the odds ratio; figures in parentheses represent standard errors.

brand and has been in existence for several years. The par-ent brand is a market leader with a 53 % market share in thetime period prior to the introduction of a food extension in acategory that was perceived as dissimilar to the parent brandcategory and as low in overall fit (ratings of 1.92 and 3.11,respectively).The data w ere analyzed in a manner similar to the analy-ses of Studies 1 and 2. The results are presented in Table 8.As can be seen, the reciprocal effects indicator (IND) has asignificant, n egative coefficient, and this coefficient is notsignificant among nonusers. This suggests that there arenegative reciprocal effects of extension trial on the parentbrand among prior users. This study shows that a failedextension (where failure is defined as the extreme situationwhen a brand is eventually withdrawn from the market) maycause negative reciprocal effects among prior users of theparent brand. In summary. Study 3 suggests the following:

An unsuccessful extension can produce negative reciprocaleftects among prior users of the parent brand, even when theextension category has relatively low similarity to the parentcategory.

DiscussionSummaryOur fmdings indicate that positive reciprocal effects ofextension trial exist, particularly among nonloyal users andamong prior nonusers of the parent brand. These positivereciprocal effects also appear to translate into market shareincreases. In our research, we show that category similarityappears to moderate the existence and m agnitude of po sitivereciprocal effects. In addition, negative reciprocal effects ofunsuccessful extensions exist among prior users of the par-ent brand. A summary of the findings from the various stud-ies is presented in Figure 2.

The fmdings from this research are generally consistentwith previous findings based on experimental data (e.g.,Gurhan-Canli and M aheswaran 1998; Keller and Aaker 1992;Roedder-John, Lo ken, and Joiner 1998). Building on previousresearch by Keller and Aaker (1992), our research shows theexistence of positive reciprocal effects in terms of parent brandchoice associated with successful extensions and exam ines therole of prior parent brand experience as a moderator of recip-


13/16

TABLE 8Reciprocal Impact of Extension Failure on Parent Brand Choice: Study 3Eta Failed Extension

Variable Prior Users Prior NonusersConstant




Advertisements (AD)

Displays (DISP)

Sample sizeChoice-2 Log L x2Percentage correctly classifiedProportional chance

-1.362*(.338)2.207*9.085(.194)- .078.926(.090)

- .453*.636(.099).435*1.546(.152).998*

2.713(.217)36041303

216.240(p < .000)65%54%

-2.287*(.775)

.2381.269(.214)- .171.842(.241)

.4441.560(.377).6081.836(.589)

97 91894.402(p = .354)82 %69%

'Significant at the p < .01 level.Notes: Figures in boldface represent the odds ratio; figures in parentheses represent standard errors.

rocal effects. The previously overlooked role of parent brandexperience as a moderator of reciprocal effects sugg ests a pos-sible explanation for mixed findings regarding positive recip-rocal effects in prior research that has used attitudinal data.Previous studies, such as Gurhan-Canli and Mah-eswaran's (1998), Loken and Roedder-John's (1993), andRoedder-John, Loken, and Joiner's (1998), provide evi-dence of the existence of negative reciprocal effects at theattribute level. We find evidence of negative reciprocaleffects of extension failure on parent brand choice amongprior users of the parent brand. Among prior nonusers, nonegative reciprocal effects were apparent because of theseconsumers' low prior probability of purchasing the parentbrand. Establishing the potential for negative reciprocaleffects in a behavioral setting contributes to our knowledgeregarding the impact of a failed extension on consumerchoice behavior.This research contributes to the extant knowledgeregarding category similarity as a mo derator of positive rec-iprocal effects by examining its role in a real-world setting.Consistent with previous findings in lab settings (Gurhan-

Canli and M aheswaran 1998; Keller and Aaker 1992), wefind evidence of a positive association between the magni-tude of a positive reciprocal effect and the degree of simi-larity between the extension and parent categories. How-ever, category similarity did not appear to matter as much inthe context of negative reciprocal effects. Further researchshould focus on the differential role of category similarity inthe case of successful versus unsuccessful extensions.Previous research has shown that buyers who lack infor-mation regarding product quality tend to use brands as indi-cators of product quality (Rao, Qu, and Ruekert 1999).Therefore, our findings regarding the impact of experiencewith the parent brand on extension trial are not unexpected.However, scant research exists that examines the role playedby parent brand experience on extension repeat purchases.Our findings provide evidence that the role of parent brandexperience in the evaluation of a brand extension diminishesafter trial. This finding has implications for researchers whosupport the existence of confirmatory biases that may oper-ate and prevent a product from being judged on its ownmerit. However, our findings may be confined to frequently


14/16

FIGURE 2Testing a Framework of Positive and Negative Reciprocal Effects: A Summary of FindingsCategory Similarity

Successful

BrandExtension

Unsuccessful

High LowPrior NonlovalUsers of ParentBrand

PositiveConfirmed inStudies 1 and 2



PositiveConfirmed inStudies 1 and 2Prior Nonusersof Parent Brand

No effect

Prior NonlovalUsers of ParentBrandNo effectConfirmed inStudies I and 2PriorUsers of ParentBrandNegativeConfirmed in Study 3


No effectConfirmed in Study 1


No effectConfirmed in Study 3

purchased packaged goods for which product trial may besufficient to gain complete infonnation regarding quality.Further research incorporating experience or credencegoods will enhance the understanding of the role of parentbrand experience in influencing repeat purchases.Managerial ImplicationsBrand managers need to consider potential reciprocal effectsin assessing the benefits of extension introduction. The roleof brand extensions in enhancing the appeal of the parentbrand among prior nonusers of the parent brand has beenoverlooked as an important added benefit of the extensionstrategy.The introduction of a brand extension also has associ-ated risks. The failure of a brand extension can harm brandequity by producing negative reciprocal effects. Contrary towhat was believed previously, this appears to be the caseeven when the extension is introduced in a category with rel-atively little similarity to the parent category.That parent brand experience had an impact on exten-sion trial but not on repeat purchases suggests that the exten-sion strategy may be used primarily to reduce the initialexpenditures associated with product introduction. How-ever, parent brand experience appears to have little impacton long-term repeat purchasing of an extension across arange of cases in which perceived similarity between theparent and extension categories varied considerably.

Limitations an d Further R esearchReciprocal effects have been investigated in this articleentirely in the context of purchase behavior. One way tostrengthen the findings would be to support the purchasebehavior data with attitudinal data. In addition, no datainvolving a highly similar extension that failed in the marketwere available. Cases of this nature will strengthen ourframework.Another issue that we do not address is the similarity ofthe target market for the parent and extension brands and therole of target market similarity versus category fit in influ-encing the transfer of associations from parent to extensioncategories. In other words, it is possible that reciprocaleffects may exist, even in the absence of category similarity,if the target audience for the parent and extension productsis similar. Although we have not addressed this in the cur-rent study, this is a promising avenue for further research,'*The potentially moderating roles played by factors suchas extension category loyalty in the direct transfer of parentbrand experience to the extension category are not exam-ined in this research. Further research should examine thefactors that may moderate the impact of parent brand expe-rience on extension trial and should assess the validity of thefindings across various categories, such as services or tech-nical products.

'8We thank a reviewer for this insight.

REFERENCESAaker, David A, and Kevin Lane Keller (l990),"Consumer Evalu-ations of Brand Extensions," Journal of Marketing, 54 (Janu-

ary), 2 7 ^ 1 ,Advertising Age (1993), "Brand E xtensions Take Center Stage," 64(March 8), 12,


15/16

Alba, Joseph and J. Wesley Hutchinson (1987), "Dimensions ofConsumer Expertise," Journal of Consumer Research, 13(March), 411-54 .Andrews, Rick L. andT.C . Srinivasan (1995), "Studying Consider-ation Effects in Empirical Choice Models Using Scanner PanelData," Journal of Marketing Research, 32 (February), 30-41.Bawa, Kapil and Robert W. Shoemaker (1987), "The Coupon-Prone Consumer: Some Findings Based on Purchase BehaviorAcross Product Classes," Journal of Marketing, 51 (October),99-110 .Broniarczyk, Susan M. and Joseph W. Alba (1994), "The Impor-tance of Brand in Brand Extension," Journal of MarketingResearch,},[ May), 214-28.Bucklin, Randolph and James Lattin (1991), "A Two-State Modelof Purchase Incidence and Brand Choice," Marketing Science,1 0 ( 0 , 2 4 - 3 9 .Chintagunta, Pradeep, Dipak C. Jain, and Naufel J. Vilcassim(1991), "Investigating Heterogeneity in Brand Preference inLogit Models for Panel Data," Journal of Marketing Research,28 (November), 417-2 8.Cochran, William G. (1977), Sampling Techniques. New York:John Wiley & Sons.Dacin, Peter A. and Daniel C. Smith (1994), 'The Effect of BrandPortfolio Characteristics on Consumer Evaluations of BrandExtensions," Journal of Marketing Research, 31 (May),229-42.Erdem, Tulin (1998), "An Empirical Analysis of Umbrella Brand-ing," Journal of Marketing Research, 35 (August), 339-51.Feldman, Jack M. and John G. Lynch Jr. (1988), "Self-GeneratedValidity and Other Effects of Measurement on Belief, Attitude,Intention, and Behavior," Journal of Applied Psychology, 73(August), 421-35.Guadagni, Peter M. and John D.C. Little (1983), "A Logit Modelof Brand Choice Calibrated on Scanner Data," Marketing Sci-ence, 1 (3), 203-38.Gurhan-Canli, Zeynep and Durairaj Maheswaran (1998), 'TheEffects of Extensions on Brand Name Dilution and Enhance-ment," Journal of Marketing Research, 35 (No vem ber), 46474.

Hoch, Stephen J. and John Deighton (1989), "Managing WhatConsumers Learn from Experience," Journal of Marketing, 53(April), 1-20.Hsiao, C. (1986), Analysis of Panel Data. Cambridge, UK: Cam-bridge University Press.Kamakura, Wagner and Gary J. Russell (1989), "A ProbabilisticChoice Model for Market Segmentation and Elasticity Struc-ture," Journal of Marketing Research, 26 (November), 379-90 .Kapferer, Jean-Noel (1994), Strategic Brand Management: Creat-ing and Sustaining Brand Equity Long Term. New York: TheFree Press.Keller, Kevin Lane and David A. Aaker (1992), 'The Effects ofSequential Introduction of Brand Extensions," Journal of Mar-keting Research, 24 (February), 35-50.and Sanjay Sood (2000), "The Effects of Product Experi-ence and Branding Strategies on Brand Evaluations," workingpaper. Tuck School of Business, Dartmouth University.

Kempf, Deanna S. and Robert E. Smith (1998), "Consumer Pro-cessing of Product Trial and the Influence of Prior Advertising:A Structural Modeling Approach," Journal of MarketingResearch, 35 (3), 325-3 9.Kim, Byung-Do and Mary Sullivan (1998), 'The Effect of ParentBrand Experience on Line Extension Trial and Repeat Pur-chase," Marketing Letters, 9 (2), 181-93.LNA/Mediawatch Multi-Media Service: 1990-1995. New York:Leading National Advertisers, Inc.Loken, Barbara and Deborah Roedder-John (1993), "DilutingBrand Beliefs: When Do Brand Extensions Have a NegativeImpact?" Journal of Ma rketing, 57 (July), 71-84.Market Share Repo rter, 1990-1995. Detroit: Gale Research Inc.Milberg, Sandra J., C. Whan Park, and Michael S. McCarthy(1997), "Managing Negative Feedback Effects Associated withBrand Extensions: The Impact of Alternative Branding Strate-gies," Journal of Consumer Psychology, 6 (2), 119-40.Morrison, Donald G. (1969), "On the Interpretation of Discrimi-nant Analysis," Journal of Marketing Research, 6 (May),156-63.Nelson, Phillip (1970), "Information and Consumer Behavior,"Journal of PoliticalEconomy, 11 (March/April), 311-29.(1974), "Advertising and Information," Journal of PoliticalEconomy, &\, 729-54.

Rao, Akshay R., Lu Qu , and Robert W. Ruekert (1999), "S ignalingUnobservable Product Quality Through a Brand Ally," Journalof Marketing Research, 36 (2), 258-6 8.Roedder-John, Deborah, Barbara Loken, and Christopher Joiner(1998), 'The Negative Impact of Extensions: Can FlagshipProducts Be Diluted?" Journal of Ma rketing, 62 (1), 19-32.Russell, Gary J. and Wagner Kamakura (1994), "UnderstandingBrand Competition Using Micro and Macro Scanner Data,"Journal of Marketing R esearch, 3\ (May), 289-303.SAS Institute Inc. (1995), Logistic Regression Examp les Using theSAS System, Version 6, 1st ed. Cary, NC: SAS Institute Inc.Smith, Daniel C. and C. Whan Park (1992), "The Effects ofBrand Extensions on Market Share and Advertising Effi-ciency," Journal of Marketing Research, 29 (August) ,296-313.Smith, Robert E. and William R. Swinyard (1982), "InformationResponse Models: An Integrated Approach," Journal of Mar-keting, 46 (Winter), &\-93.Sullivan, Mary (1990), "Measuring Image Spillovers in Um-brella Branded Products," The Journal of Business, 63 (3),309-30 .Webster, Frederick E., Jr. (1965), 'The 'Deal-Prone' Consumer,"Journal of Marketing Research, 2 (2), 186-89.Wernerfelt, Birger (1988), "Umbrella Branding as a Signal of NewProduct Quality: An Example of Signaling by Posting a Bond,"RAND Journal of Economics, 19 (3), 458-66 .Wright, Alice and John G. Lynch Jr. (1995), "CommunicationEffects of Advertising Versus Direct Experience When BothSearch and Experience Attributes Are Present," Journal of Con-sumer Research, 21 (March), 708-1 8.


16/16

The Impact of Brand Extension Introduction on Choice

Documents

Transcript of The Impact of Brand Extension Introduction on Choice