Endorsers in advertising: The case of negative celebrity information
Journal of Advertising, Spring 1998
Authors: Brian D Till & Terence A Shimp


       Three studies examining how negative information about a celebrity can
       affect the brand the celebrity endorses are presented. Using an associative
       network model of memory as a theoretical framework, 4 moderating variables
       were considered: the size of the association set for the brand, the size
       of the association set for the celebrity, the timing of the negative celebrity
       information and the strength of the associative link between the brand
       and the celebrity. In the first 2 studies, a fictitious but realistic
       celebrity endorser was used and in the 3rd an actual celebrity was used.
       Negative information about a celebrity resulted in a decline in attitude
       toward the endorsed brand only for the fictitious celebrity. That general
       relationship was moderated in varying degrees by association set size,
       timing of the negative information and the strength of the link between
       brand and celebrity.


       The use of celebrity endorsers is a popular executional device, but it
       is not without risk. The authors report three studies examining how negative
       information about a celebrity can affect the brand the celebrity endorses.
       Using an associative network model of memory as a theoretical framework,
       they considered four moderating variables: the size of the association
       set for the brand, the size of the association set for the celebrity, the
       timing of the negative celebrity information, and the strength of the associative
       link between the brand and the celebrity. In the first two studies, they
       used a fictitious but realistic celebrity endorser and in the third they
       used an actual celebrity. Negative information about a celebrity resulted
       in a decline in attitude toward the endorsed brand only for the fictitious
       celebrity. That general relationship was moderated in varying degrees by
       association set size, timing of the negative information, and the strength
       of the link between brand and celebrity.

       The use of celebrity endorsers is prevalent in advertising. In addition
       to the intuitive arguments that rationalize the practice, academic researchers
       have mounted empirical evidence to demonstrate the benefits of product
       endorsements. Atkin and Block (1983), for example, found that the use of
       a celebrity in beer advertising led to more favorable ad ratings and more
       positive product evaluations. Freiden (1984) tested four types of endorsers
       (celebrity, CEO, expert, and typical consumer) and determined that in comparison
       with other endorser types, the celebrity endorser scored particularly well
       on dimensions such as trustworthiness, believability, persuasiveness, and
       likeability. In a test of the matchup hypothesis, Kamins (1990) demonstrated
       that the positive impact of a celebrity endorser depends in part on proper
       fit between the celebrity and the product. Some evidence even suggests
       that Wall Street values the use of celebrity endorsers-Agrawal and Kamakura's
       (1995) analysis of stock price movements showed that press releases announcing
       celebrity endorsement contracts resulted, on average, in a .44% excess
       return.

       Our research differs from prior research by examining the impact that negative
       information about a celebrity might have on consumer evaluations of endorsed
       brands. Practitioners hope their target audience's positive feelings toward
       a chosen celebrity will transfer to the endorsed brand or will otherwise
       enhance the brand's standing. What happens, however, if either during or
       after an advertising campaign negative information about the celebrity
       becomes public?

       Widely publicized incidents (e.g., O.J. Simpson's indictment and later
       acquittal on murder charges, and Pepsi Cola's series of debacles with three
       tarnished celebrities-Mike Tyson, Madonna, and Michael Jackson) suggest
       that celebrity endorsers may at times become liabilities to the brands
       they endorse. The fear of potential celebrity scandals has given rise to
       a minitrend toward using deceased celebrities-individuals who posthumously
       can no longer engage in behaviors that might bring embarrassment and injury
       to the brands with which they are linked (Goldman 1994; Lefton 1994; Miller
       1993). Observation also reveals an increase in the use of animated characters
       as product endorsers. Callcott and Lee's (1994) content analysis determined
       that 28% of animated advertisements contained cartoon-character celebrities
       such as Bugs Bunny and Mickey Mouse. Like deceased celebrities, those characters
       are generally immune to negative publicity.

       No research with which we are familiar has directly tested the possible
       effects of negative information about celebrities on consumers' evaluations
       of endorsed brands, though two recent studies have touched on the issue.
       Langmeyer and Shank (1993) demonstrated a positive relationship between
       people's perceptions of a celebrity (Madonna) and perception of a nonprofit
       agency (Mothers Against Drunk Driving). For subjects who had a positive
       (negative) image of Madonna, perceptions of MADD became more positive (negative)
       after it was paired with Madonna. Tripp, Jensen, and Carlson (1994) found,
       among other results, that the effect of a celebrity endorsing multiple
       products is to reduce the celebrity's credibility and likeability and to
       lower attitudes toward the ad itself. Though both sets of findings are
       interesting, neither study examined the specific issues addressed in our
       research. We report three studies examining the conditions under which
       negative press about a celebrity may affect the endorsed brand.

       Marketers obviously eschew nonpositive information about their brands and
       anything or anyone (such as a celebrity endorser) that is associated negatively
       with their brands. They have good cause for their concern inasmuch as literature
       in various psychological traditions has theorized and/or shown empirically
       that negative information has disproportionate influence on consumers'
       beliefs and evaluative judgments. Mizerski (1982) has done a particularly
       good job of describing the underlying reasons and demonstrating from an
       attribution theory perspective how negative information operates.

       Using an associative network framing, we next explain how negative information
       about a celebrity can operate to lower evaluations of the advertised brand
       with which the celebrity has been associated. Three studies in which we
       applied associative learning principles are then reported. In the first
       two we used a fictitious, albeit realistic, celebrity endorser and in the
       third we used an actual celebrity.

       Associative Learning and Celebrity Endorsers

       Associative learning principles are based on a conception of memory as
       a network consisting of various nodes connected by associative links (Anderson
       1976; Collins and Loftus 1975; Rumelhart, Hinton, and McClelland 1986).
       In our research context, celebrities and brands both represent nodes, which
       initially are unconnected but become linked over time through the endorsement
       process.

       Feelings toward a celebrity and/or meanings in the celebrity are expected
       to transfer to the endorsed brand through their recurring association.
       The repeated exposure to two stimuli results in simultaneous activation
       of memory nodes representing those stimuli, building an associative link
       between the two nodes (Domjan and Burkhard 1986; Klein 1991; Martindale
       1991; Rumelhart, Hinton, and McClelland 1986). After an associative link
       has been forged between an advertised brand and its celebrity endorser,
       subsequent negative information about the endorser may result in a lower
       evaluation of the celebrity, which in turn may reflect back to the endorsed
       brand through the associative link established between the two entities.
 

       The Impact of Negative Information

       Repeated advertised pairings of a brand and celebrity establish/modify
       the pattern of connectivity by strengthening the associative link between
       them (Anderson 1976, 1983b; Berger and Mitchell 1989; Domjan and Burkhard
       1986; Fazio, Powell, and Williams 1989; Furstenberg, Sebrechts, and Seamon
       1987; Klein 1991; Martindale 1991). The concept of association sets is
       especially insightful for understanding that process. An association set
       represents the preexisting associates, or group of concepts, that are related
       meaningfully to an object (Nelson, Schreiber, and McEvoy 1992). Both the
       celebrity and the brand represent nodes connected to other nodes based
       on experiences with the brand and the celebrity. Those connections represent
       the association set for the celebrity and brand. For example, for some
       people, the association set for Michael Jordan might consist of "Chicago
       Bulls," "University of North Carolina," "basketball," "baseball," "Nike,"
       "Dream Team," "Gatorade," and so forth. Association sets also include individuals'
       attitudes toward objects such as brands and celebrities (cf. Berger and
       Mitchell 1989; Fazio, Powell, and Williams 1989; Fazio et al. 1986; Judd
       et al. 1991; Noffsinger, Pellegrini, and Burnell 1983).

       When a consumer thinks about a brand, the link with the celebrity node
       is animated to a certain level through spreading activation (Anderson 1983a).
       The joint activation of brand and celebrity provides a path over which
       one's evaluation of the celebrity has an opportunity to transfer to the
       brand. The key to the process is the simultaneous activation of the brand
       and celebrity nodes. Negative information about the celebrity activates
       the celebrity node, which then activates the brand node to some degree
       and allows reduced evaluation of the celebrity to transfer to the brand.
       Studies by Noffsinger et al. (1983) and Judd et al. (1991) provide empirical
       evidence demonstrating that attitudes can be affected in such a way. The
       preceding discussion suggests the following general hypothesis.

       H1: Given a sufficiently strong associative link between a celebrity and
       a brand, subsequent negative information about the celebrity results in
       lowered evaluations of the brand.

       Moderating Role of Association Set Size

       Repeated pairings of celebrity endorser and brand should facilitate celebrity
       and brand becoming part of each other's association set. (For example,
       what celebrities come to mind when you see the brand names Jello and Nike?)
       Thinking of the brand will increase activation of the celebrity node, and
       thinking of the celebrity will increase activation of the brand node. However,
       thinking of a celebrity may not be as likely to activate an associated
       brand node as thinking of a brand is to activate an associated celebrity
       node. The level of activation of any one particular node depends in part
       on the number of competing nodes associated with that node. In the marketplace,
       consumers may have a larger association set for a certain celebrity than
       they have for a particular brand. In that case, thinking of the brand is
       more likely to activate the celebrity node than thinking about the celebrity
       is to activate the brand node.

       As the size of an association set for a given concept increases, the likelihood
       of any given associated node also being activated is reduced; the greater
       the number of concepts activated, the less intensively each will be activated
       (Anderson 1983a, b; Collins and Loftus 1975; Nelson, Bajo, and Casanueva
       1985). The learning of additional facts about a concept creates competition
       to take strength away from already known facts (Anderson 1983a), and activation
       of a set of nodes can inhibit the activation of related nodes (Martindale
       1991). For example, consider a person about whom you know two things: he
       plays the violin and drives a sports car. Because only those two concepts
       are associated with this person in your mind, when thoughts of that individual
       come to mind, the two associated concepts (plays violin, drives sports
       car) are fairly likely to be activated spontaneously to a reasonably high
       degree. If over time you learn more about the person (has two children,
       works at a hospital, served in the Navy, etc.), then all else being equal,
       the initial two nodes (plays violin and drives sports car) become less
       likely to be as highly activated. The basic principle is known as the fan
       effect (Wang, Seidman, and Reggia 1988).

       Though competing nodes may reduce activation of any one target node, the
       target node may still activate strongly enough to come into working memoryprovided
       the association between the originating node and the associated target
       node is strong (Martindale 1991). In other words, "plays violin" may be
       associated so strongly with a friend that learning additional facts, though
       reducing the level of activation of "plays violin," does not reduce it
       significantly.

       Consider now celebrities and brands. A given brand evokes a particular
       association set consisting in part of information about the brand as well
       as association with the celebrity endorser. As the size of the association
       set for the brand increases, the level of activation of the associated
       celebrity node is likely to decrease, potentially tempering the impact
       of negative celebrity information on the brand. By similar logic, as the
       size of the celebrity association set increases, the level of activation
       of the associated brand node is likely to decrease. Hence, we propose two
       moderating hypotheses.

       H2a: As the size of the brand association set increases, the effect on
       brand evaluations of negative celebrity information decreases.

       H2b: As the size of the celebrity association set increases, the effect
       on brand evaluations of negative celebrity information decreases.

       Study 1

       In the first experiment we manipulated brand and celebrity association
       set sizes at two levels (relatively small or large set sizes) and whether
       or not subjects received negative information about the advertised brand's
       celebrity endorser. Of particular interest was the change in subjects'
       evaluations of the advertised brand based on the negative information about
       the endorser. Subjects' brand evaluations were measured at the beginning
       of the study (before introduction of negative information about the celebrity)
       as well as later in the study (after introduction of negative information
       about the celebrity). In sum, the study had a 2^sup 4^ mixed factorial
       design consisting of three between-subjects factors (brand set size, celebrity
       set size, presence/absence of negative information) and one within-subjects
       factor (time of dependent variable measurement).

       Choice of Brand and Celebrity

       To minimize preexisting knowledge and affect due to prior exposure and
       familiarity, we used a fictitious brand from a product category relevant
       to our student subject pool. On the basis of pretesting, the racing bicycles
       product category and the brand name Avenix were selected. (In the context
       of the experiment, Avenix was a French brand name and was pronounced phonetically
       to subjects as Ah-ven-ee.) A fictional celebrity was created to overcome
       potential problems with using a real celebrity. Though celebrities are
       by definition well known and generally popular, liking of a celebrity can
       vary considerably even in a relatively homogeneous population such as students.
       High variation in subjects' evaluation of a celebrity, even though randomly
       distributed across experimental conditions, could cause an error variance
       that swamps treatment effects. A second problem overcome by the use of
       a fictional celebrity relates to the association set size manipulation.
       Ideally, one might use two real celebrities who differ only in the size
       of their association sets. Practically, however, that is impossible. Any
       two celebrities may indeed differ in the size of their association sets,
       but, regardless of how closely matched, the two celebrities would also
       vary in other uncontrollable respects.

       On the basis of pretesting, French Olympic cyclist Pierre Varnay was selected
       as the celebrity endorser for Avenix racing bicycles. Varnay is fictional,
       but subjects were not aware of that fact and not a single subject questioned
       the authenticity of Varnay or the study ruse that incarnated him. Note
       that, to overcome the potential liabilities associated with using a created
       celebrity (such as reduced ecological validity), we used a real celebrity
       in study 3.

       Size of the Association Sets

       Subjects in the small association set condition received two pieces of
       information each about the celebrity and about the brand, and subjects
       in the large association set condition received six items of information
       each for the celebrity and the brand. For the celebrity manipulation, the
       following six items were presented in the large set size condition.

       1. Pierre Varnay was the winner of the 1991 European Invitational.

       2. Varnay has competed in the prestigious Tour de France three times.

       3. Varnay is a vegetarian.

       4. Varnay is one of the few left-handed cyclists.

       5. Varnay's family has always worked on a farm.

       6. Varnay was born near Normandy, France.

       The following six information items were used in the large brand association
       set size condition.

       1. Avenix tires self-seal if punctured.

       2. Avenix has quick-release hubs.

       3. Avenix comes with a one-year warranty.

       4. Avenix has 10 speeds (gears).

       5. Avenix is made just outside Paris.

       6. Avenix was designed by a graduate of the Sorbonne, a French university.
 

       Only the first two pieces of listed information for celebrity and brand
       were given to subjects in the small set size conditions. Pretesting determined
       that attitudes toward the brand and endorser were intentionally equivalent
       across the two set sizes.

       Subjects

       After 15 student subjects were dropped because of incomplete responses
       (6) or demand awareness (9), responses from 283 subjects remained for analysis.
       Three to 14 subjects participated in each session, with seven or eight
       as the modal attendance.

       Procedures

       Background and Study Ruse. Subjects were recruited primarily from introductory
       marketing courses and offered extra credit for their participation. The
       experimenter informed them that a major bicycle manufacturer was interested
       in their opinions about an advertising campaign for the new Avenix. That
       ruse served to rationalize subjects' (1) viewing the ad campaign that was
       designed to build an association between Avenix and Pierre Varnay, (2)
       reading articles (including the negative information) about Pierre Varnay,
       and (3) completing both pre and post measures of attitude toward Avenix.
 

       Association Sets. Prior to viewing the ad campaign, subjects were told
       that they may not be familiar with Pierre Varnay, a French Olympic cyclist
       who endorses Avenix. (One might conjecture that such an announcement encouraged
       subjects to focus unduly on the ad campaign and the subsequent negative
       information. However, as is apparent from our subsequent presentation of
       findings, no demand-artifact explanation could possibly account for the
       pattern of results obtained.) Subjects then were given either two or six
       pieces of background information about Varnay and either two or six pieces
       of information about Avenix. At the end of the study, written instructions
       asked subjects to recall each fact through prompting questions such as
       "How long is the Avenix warranty?" and "What race did Varnay win?" The
       subjects recalled 91.5 to 100% of the facts, with an average of 96.4% across
       conditions, thus indicating that they had learned the necessary association
       set information.

       Initial Evaluation of Avenix. The initial evaluation was completed before
       subjects were exposed to the ad campaign and before they read articles
       that included the negative information about endorser Varnay. An initial
       lead-in question ("My overall feeling about Avenix bicycles is...") was
       followed by three 9-point bipolar scales (favorable/unfavorable, positive/negative,
       strongly like/strongly dislike). A second lead-in question ("I think that
       the Avenix bicycle is...") was followed by four additional items (good/bad,
       high quality/low quality, superior/inferior, and fast/slow). The seven
       items were summed to represent an overall attitude toward Avenix (Cronbach
       alpha=.93).

       Forging the Association. A link between Pierre Varnay and Avenix bicycles
       was established by having subjects view a series of print ads. After being
       informed that the bicycle company's ad agency wanted to get college students'
       reactions to the campaign, subjects received an ad folder containing 10
       black and white print ads. Each ad featured the same picture of Pierre
       Varnay riding an Avenix bicycle. (An actual photograph of an unknown cyclist
       representing the fictitious Pierre Varnay was selected through pretesting
       that identified it as the most attractive and athletic of several tested.)
       The headlines, though different in each ad, all linked Pierre Varnay with
       Avenix-for example, "Pierre Varnay Wins with Avenix." Each ad had minimal
       body copy that conveyed in different terms Pierre Varnay's endorsement
       of Avenix. The experimenter gave subjects 10 seconds for viewing/reading
       each ad. Subjects then evaluated the campaign by responding to nine 7-point
       bipolar scale items. Those measures were taken merely to maintain consistency
       with the study ruse and are not pertinent to any hypothesis tests.

       Presenting the Negative Information. Subjects were told that there has
       been some recent publicity about Varnay and that the ad agency sponsoring
       the research was interested in determining students' awareness of the publicity.
       Subjects then read three short magazine articles. The first two-one ostensibly
       from Billboard and the other supposedly from Peoplewere about Varnay's
       music and reading preferences and were included to reduce subjects' demand
       awareness. The third, allegedly from Sports Illustrated, contained the
       negative information. (Control group subjects received the Billboard and
       People articles but not the Sports Illustrated article.) After reading
       each of the three articles, subjects were asked if they had seen the article
       previously, if they subscribe to the particular magazine, if they read
       the magazine at least once a month, and if the article influenced how they
       felt about Varnay. The purpose of those questions was to reinforce the
       ruse used to present the articles.

       The Sports Illustrated article gave subjects either of two types of negative
       information about Pierre Varnay: that he had engaged in steroid use or
       that he had multiple driving-under-the-influence (DUI) infractions on his
       record. Because the two negative information instantiations yielded virtually
       identical results, we pooled the data. Obtaining similar results with two
       types of negative information simply adds to study generalizability.

       Memory-Clearing Task. To help ensure that any observed effect was due to
       a restructuring of longterm memory, a memory-clearing task intervened between
       the negative information manipulation and measurement of the key dependent
       variables. Subjects read a newspaper article on commuting and carpooling
       and then were asked a variety of closedand open-ended questions about their
       attitudes toward commuting, carpooling, and the use of bicycles as an alternative
       to car use.

       Post Evaluation of Avenix. Subjects were told that now that they had seen
       the proposed ad campaign for Avenix, the ad agency was interested in their
       evaluations of Avenix. They completed the same set of measurements (i.e.,
       seven 9-point bipolar scale items) as in the initial evaluation of Avenix.
 

       Manipulation Check and Demand Awareness Measure. As a manipulation check
       on the negative information, subjects evaluated Pierre Varnay on a variety
       of 9-point bipolar scales: three global evaluation measures (good/bad,
       favorable/unfavorable, and positive/negative), two trust measures (trustworthy/not
       trustworthy and believable/not believable), two expertise measures (knowledgeable/not
       knowledgeable and qualified/not qualified), and two endorser appropriateness
       measures (appropriate/inappropriate and effective/ineffective). The three
       global items were summed to represent an overall attitude toward the endorser
       (Cronbach alpha=.98), as were the two trustworthiness items (r=.78), the
       two expertise items (r=.74), and the two appropriateness items (r=.86).
 

       Several checks were performed to establish that the negative information
       from the Sports Illustrated article affected subjects' evaluation of Pierre
       Varnay. Immediately after reading each of the articles on Varnay, subjects
       were asked, among other things, whether the article had influenced their
       feelings toward Varnay. Slightly more than 88% indicated that the Sports
       Illustrated article had affected their feelings. A stronger manipulation
       check was the evaluation subjects made of Varnay at the end of the study.
       Table 1 shows that subjects who received the negative information about
       Varnay had significantly less favorable evaluations of Varnay than subjects
       in the control groups, who did not read the negative Sports Illustrated
       article.

       To identify potential demand awareness, subjects responded to an open-ended
       question about the purpose of the study. Responses were categorized as
       demand-aware if the comments identified the purpose of the study as determining
       how the negative information about the celebrity would influence people's
       feelings about the brand. Of the 196 subjects in the four negative information
       conditions, nine (4.6%) were classified as demand-aware and hence dropped
       from subsequent analysis.

       Results

       Effect of Negative Information. H1 suggests that given a sufficiently strong
       associative link between the celebrity and the brand, negative information
       about the celebrity will lower brand evaluations. As subjects had evaluated
       Avenix twice (before and after reading articles about Avenix), repeated
       measures MANOVA was used to examine whether the change in Avenix evaluations
       was greater in the negative information conditions than in the control
       conditions. Table 2 reports descriptive statistics and focused contrasts
       between experimental conditions and their corresponding controls.

       The pre and post evaluations of Avenix averaged across the four negative
       information conditions were 46.9 and 45.9, respectively, indicating that
       attitudes toward Avenix declined after the negative information about the
       endorser. Comparatively, a moderately positive effect of the advertising
       information is evident in the control conditions with average pre and post
       evaluations of 46.5 and 47.3, respectively. The within-subjects effect
       for the interaction between time of measurement (pre vs. post) and condition
       (negative information vs. control) is significant (F1,276= 10.1;
       p<.01), supporting the expectation in H1 that negative information about
       the celebrity lowers evaluations of the advertised brand.

       Moderating Role of Brand and Celebrity Association Set Sizes. H2a and H2b
       propose that the impact of the negative celebrity information on evaluations
       of the endorsed brand should be greater for smaller (vs. larger) association
       sets. H2a speaks to the moderating role of brand set size and predicts
       a time of evaluation by brand set size (with collapsing across celebrity
       set size) by negative information interaction; however, the test for that
       interaction is not significant (F1,279 = 1.0, p=.31). The test of H2b
       on the moderating role of celebrity set size (with collapsing across brand
       set size) is also nonsignificant (F1,279=2.3, p=. 13).

       Though H2a and H2b are not supported, separate contrast tests for the four celebrity/brand set size
       conditions show a statistically significant attitude change toward Avenix in the small
       brand set/small celebrity set/negative information condition versus its
       matching control group (F =14.96; p< .01).The finding demonstrates that
       the effect of negative celebrity information on the endorsed brand is greatest
       when both the brand and the celebrity set sizes are relatively small; comparatively,
       no other contrasts between the respective treatment and control conditions
       are statistically significant.

       Discussion

       As supported by the logic of associative network models, our results indicate
       that activation of negative information about a celebrity can have an adverse
       effect-through lowered brand evaluations-on the endorsed brand with which
       that celebrity is associated. Negative information had a strong effect
       when the association set sizes were small for both the brand and the celebrity.
       However, when either the brand or the celebrity had larger association
       sets, negative celebrity publicity did not have a significant detrimental
       effect on the endorsed brand. That crucial finding suggests that negative
       information about a celebrity may be problematic for the associated brand
       only when consumers have scant association sets, or knowledge structures,
       for both brand and celebrity. When knowledge structures for brand and/or
       celebrity are more fully developed, a brand may be somewhat insulated from
       negative press about the endorsing celebrity.

       

 
       Study 3

       The third study had two objectives: (1) to examine the generalizability
       to a real celebrity of the effect of negative celebrity information on
       the endorsed brand and (2) to examine whether the order in which the endorser
       and the brand are evaluated moderates the effect of the negative celebrity
       information. In studies 1 and 2 subjects evaluated the celebrity after
       evaluating the brand. The evaluation of the brand in study 3 was manipulated
       as either before or after the celebrity evaluation. Study 3 had a 2^sup
       3^ mixed factorial design. The between-subjects factors were order of the
       brand and celebrity evaluations and presence/absence of negative information.
       The within-subjects factor, as in the first two studies, was the time of
       the dependent variable measurement. Use of a Real Celebrity

       The use of a fictitious celebrity in studies 1 and 2 allowed for considerable
       experimental control. Control was particularly important in study 1, which
       manipulated size of the celebrity association set. The association sets
       for real celebrities naturally vary from consumer to consumer, rendering
       any experimental manipulation of association set problematic. Though the
       subjects found the experimental ruse, the background information on Varnay,
       and the overall experimental environment to be realistic, we wondered whether
       negative celebrity information would affect the endorsed product in the
       case of a real and wellknown endorser.

       The rationale for expecting negative celebrity information to transfer
       to the endorsed brand was developed previously, but would the use of a
       real celebrity yield stronger or weaker effects? An argument in favor of
       stronger effects is that the negative information might have greater salience
       or memorability when it is about someone people already know. An offsetting
       argument is that people have larger association sets for real than for
       experimentally created celebrities and, as shown in study 1, association
       set size can influence the impact of negative information on the endorsed
       brand. Hence, the effect on brand evaluations may be weaker for a real
       celebrity than for the created celebrity used in studies 1 and 2. We hypothesized,
       as in Hi involving a fictional celebrity, that negative information about
       a real celebrity would lead to lowered brand evaluations. We formed no
       hypothesis for whether the effect of negative information about a real
       celebrity would be stronger or weaker than the effect found in studies
       1 and 2.

       H5: Given a sufficiently strong associative link between a real celebrity
       and a brand, subsequent negative information about the celebrity lowers
       brand evaluations.

       Order of Brand/Endorser Evaluation

       In studies 1 and 2 subjects evaluated the brand before evaluating the celebrity.
       That ordering is ecologically valid inasmuch as consumers typically make
       choices at the point of purchase without celebrity information, either
       positive or negative. Any impact of celebrity information on brand evaluation
       must derive from long-term memory. In study 3 we manipulated the ordering
       of celebrity and brand evaluations. Our reasoning was that having subjects
       evaluate the celebrity first would effectively serve to prime the negative
       celebrity information and its linkage with the endorsed brand immediately
       prior to subjects' evaluation of the endorsed brand. We therefore sacrificed
       ecological validity of the ordering for the offsetting advantage of potentially
       greater impact.

       H6: Negative celebrity information has greater impact on brand evaluations
       when the celebrity is evaluated before the brand than it does when the
       celebrity is evaluated after the brand.

       Method

       The Avenix bicycle was the endorsed product. A well-known American cyclist
       (Greg LeMond) was used as the endorser for Avenix. The picture in the ads
       in studies 1 and 2 portrayed a very athletic cyclist riding a racing bicycle.
       Because the cyclist was wearing a helmet and his face was not clearly visible,
       we used the same ads in study 3 and merely substituted the name Greg LeMond
       for the name Pierre Varnay in the headline and short copy. The fictitious
       Sports Illustrated article provided the negative information induction
       (steroid usage). Size of the brand and celebrity association sets was held
       constant at three items of information each. Three facts about Avenix (drawn
       from study 1) were presented to the subjects. Three true facts provided
       the association set for Greg LeMond, namely that Greg is an avid outdoorsman,
       who grew up in California, and has won the Tour de France three times.
 

       Subjects and Procedures. Ninety-seven undergraduate students completed
       the study, but two were later classified as demand-aware and dropped. Procedures
       were very similar to those used in the first two studies; specifically,
       subjects (1) read background information about the alleged purpose of the
       study, (2) learned the association sets for Avenix and LeMond, (3) completed
       an initial (pre) evaluation of Avenix, (4) viewed the Avenix ad campaign,
       (5) read the fictitious LeMond magazine articles, (6) performed a memory-clearing
       task, (7a) completed a post evaluation of Avenix or (7b) completed an evaluation
       of Greg LeMond, (8a) completed an evaluation of Greg LeMond or (8b) completed
       a post evaluation of Avenix, (9) answered diagnostic questions, and (10)
       were debriefed at the end of the session and told that none of the materials
       used in the study should be taken to imply that Greg LeMond, in actuality,
       has ever used steroids.

       Results

       We performed several checks to establish that the negative information
       from the Sports Illustrated article adversely affected subjects' evaluation
       of Greg LeMond. As in the previous studies, after reading the articles
       about LeMond, subjects were asked whether their feelings toward him had
       changed. Approximately 81% indicated that the Sports Illustrated article
       had affected their feelings about LeMond. Table 5 shows that subjects who
       received the negative information about LeMond had significantly less favorable
       evaluations of LeMond than subjects in the control groups who did not read
       the negative Sports Illustrated article.

       Tests of Hypothesized Effects

       Effect of Negative Celebrity Information. H5 examines whether the deleterious
       effect of negative celebrity information on the endorsed brand, as found
       in studies 1 and 2, holds in the case of a real celebrity. The average
       pre and post evaluations of Avenix for the negative information condition
       are 47.2 and 46.7, for a decrease of .5. For the control condition, the
       pre and post evaluation averages are 46.4 and 47.1, for an increase of
       .7. The pattern of results is consistent with expectations, but the interaction
       between type of information (negative or not) and time of evaluation (pre
       vs. post) is not significant (F1,91=1.97; p=.16).

       Effect of Order of Brand/Endorser Evaluation. H6 posits that the effect
       of negative celebrity information on an endorsed brand is stronger when
       subjects evaluate the celebrity before evaluating the brand. Table 6 shows
       that the change in evaluation of Avenix is -.4 for subjects in the negative
       information group who evaluated LeMond first, whereas the change is 1.0
       for the corresponding control group (overall difference 1.4). The change
       in Avenix evaluations is -.7 for subjects in the negative information group
       who evaluated Avenix first, but is .5 for the control group (overall difference
       1.2). The information (negative or control) by order of evaluation (celebrity
       or brand first) by time of evaluation (pre or post) interaction is not
       significant (F1,91 = 0.16; p=.70).

       Discussion

      

       Negative celebrity information had a somewhat weaker effect than it did
       in the two studies with the fictitious Pierre Varnay. One plausible explanation
       is that subjects had richer association sets for LeMond than for the fictitious
       Varnay, which may have diluted the impact of the negative information.
       A comparison of Table 5 with Tables 1 and 3 adds support for such an interpretation
       as the negative Sports Illustrated article had a greater effect on subjects'
       evaluations of Varnay than on their evaluations of LeMond.

       We also examined the possible effect of the order in which subjects evaluate
       the brand and the celebrity. One might expect a greater effect of negative
       celebrity information on the endorsed brand when the celebrity is evaluated
       just before the brand. However, as order of brand/celebrity evaluation
       is not significant, negative celebrity information appears to have been
       sufficiently accessible, without artifactual priming, to influence evaluations
       of the brand with which the celebrity was linked.

       General Discussion

       Our research broadens the theoretical domains used in understanding the
       processes involved in celebrity endorsements by considering celebrity issues
       within the context of an associative memory framework. Indeed, the desired
       outcome from the use of a celebrity endorser is a strong and positive associative
       link between the brand and the celebrity as a means of enhancing the brand's
       equity (Keller 1993). Though advertisers expect the associative link between
       the brand and the celebrity to transfer positive feelings about the celebrity
       to the endorsed brand, our research shows that a lowered evaluation of
       the celebrity can lower brand evaluations.

       Study 1 shows that in the small brand/small celebrity association set size
       condition-albeit in no other condition-the negative information about Varnay
       attenuated subjects' attitudes toward Avenix. That finding is consistent
       with the results of fan effect studies (e.g., Anderson and Reder 1987)
       demonstrating how activation of a particular node increases as the number
       of competing nodes decreases.

       The results from study 2 replicate those from study 1. Additionally, the
       data show that negative celebrity information presented prior to the brand/celebrity
       pairing had a greater effect on the endorsed brand than negative celebrity
       information presented after the pairing. However, the findings fail to
       support the prediction that the effect of negative celebrity information
       on the endorsed brand would be greater when there is a strong (rather than
       weak) associative link between the brand and the celebrity.

       The results of study 3 do not support the prior studies' findings about
       the deleterious effects of negative celebrity information on the endorsed
       brand. We speculate that subjects had a fuller association set for Greg
       LeMond than for the fictitious Pierre Varnay, which may have blunted the
       effect of the negative celebrity information.

       Implications and Limits

       Use of celebrity endorsers has been an enduring practice in advertising,
       but our studies illustrate its potential risk. The advertiser who chooses
       to use a celebrity has no control over the celebrity's future behavior.
       Any negative news about a celebrity may reduce the celebrity's allure,
       and therefore the appeal of the brand the celebrity has endorsed. The risk
       is potentially great for new or unfamiliar brands for which the association
       set is relatively scant and for which the celebrity is essentially the
       primary attribute on which consumers form evaluations of the brand. Negative
       celebrity information may have a much greater effect on such brands than
       it does on familiar, established brands. Also at great risk are brands
       closely tied to a specific celebrity, as study 2 provides suggestive directional
       evidence that the effect of negative celebrity information is greater when
       the associative link between the brand and celebrity is strong.

       Our research has several limitations. One is the compressed time in which
       the phenomenon was examined. Each complete experiment-which involved learning
       about the brand and the celebrity, developing the association between the
       brand and the celebrity, and reacting to exposure to the negative celebrity
       information-was done in 45 to 50 minutes. In the marketplace, the process
       would occur over weeks, if not years.

       Another limitation is the laboratory setting. Because the marketplace can
       be both more intense and less intense than the laboratory, the results
       from our laboratory studies may not generalize to the marketplace. Celebrities
       in the marketplace often evoke a much richer set of feelings and responses
       than a fictitious celebrity used in laboratory research. Strong associations
       between a celebrity and a brand are established over years of advertising
       exposure, and individual celebrities often endorse multiple brands. Because
       of the higher salience of an actual celebrity, negative celebrity information
       might have more impact for an actual than for a created celebrity. Negative
       information about Michael Jordan, for example, may be more startling, interesting,
       and meaningful than negative information about either the fictitious Pierre
       Varnay or the less famous Greg LeMond. In many ways, the marketplace can
       be more intense than the laboratory, and any effect found in the laboratory
       also would be expected to occur in the marketplace. However, certain characteristics
       make the marketplace seem less intense than the laboratory. In the marketplace
       consumers often face communication clutter, celebrities (such as Michael
       Jordan and Tiger Woods) endorsing multiple brands, brands employing multiple
       celebrities (such as the well-known milk mustache campaign), and sources
       of information about celebrities that vary widely in credibility. All of
       those factors may mitigate the impact of negative celebrity information
       and limit the generalizability of a laboratory study.

       A third limitation of our research is the use of a fictitious brand (Avenix).
       Ecologically, results from our experiments could at the limit be generalized
       to the subset of new or relatively unknown brands for which consumers'
       knowledge structures are scant. Our use of Avenix structured a situation
       in which the information about the celebrity endorser was essentially the
       most important ground on which subjects could form evaluations of the brand.
       Further, because negative information is notable in its tendency to gain
       attention and evoke cognitive effort (Taylor 1991), our experiments created
       a best-case environment for the negative celebrity information to be integrated
       with initial attitudes formed toward Avenix and therefore to result in
       less favorable attitudes.

       Though our findings are limited in their range of applicability, the purpose
       of laboratory experimentation is not to generalize findings to business
       practice, but rather to test practitioners' "theories in use" and to generalize
       processes to actual practice (cf. Mook 1983). Marketing practitioners apparently
       operate under the assumption that negative celebrity news holds strong
       potential for sullying their brands' reputations, as celebrity endorsers
       are commonly discharged when negative information about them surfaces (Miciak
       and Shanklin 1994).

       Our findings support the practitioner's reasons for concern, but only when
       an unknown brand is advertised. Predictions derived from associative memory
       models suggest that marketers have much less cause for concern when an
       established brand is advertised. In the latter situation, consumers' rich
       and varied cognitive structures should insulate the brand from negative
       press. Understandably, advertising executives and brand managers who are
       responsible for selection of a subsequently besmirched celebrity endorser
       sense great urgency to dump the celebrity quickly to save face or for fear
       of consumer retribution. Their actions might be fully justified for the
       aforementioned reasons, but theoretical predictions based on associative
       network models of memory do not necessarily indicate that negative information
       about a celebrity causes serious harm to an established brand. Indeed,
       Hertz apparently has not suffered any significant reduction in market share
       since news emerged about O. J. Simpson's alleged connection with the murders
       of his former wife and her friend, nor did Pepsi lose share after Michael
       Jackson's child-molestation charge or Mike Tyson's rape conviction. Executives
       undoubtedly were embarrassed in all those instances, but their well-established
       brands remained relatively unscathed.

     


       Brian D. Till is Assistant Professor of Marketing, Saint Louis University.

       Terence A. Shimp is Professor of Marketing, University of South Carolina.