Literature Review

 Empirical findings on negative celebrity information

  Advertisers have long tried to transfer celebrity to advertised brands. By nassociating representative characteristics of a celebrity with a brand, the hope is that consumers will identify celebrity endorsement with the object,nthus enhancing advertising effectiveness and positive reactions to the en-dorsed brand. Indeed, research confirms the strategy is successful when consumers feel celebrity endorsers and endorsed brands are well-matched (Kahle & Homer, 1985; Kamins, 1990; Till & Busler, 2000; Miciak & Shanklin, 1994). However, advertising campaigns using celebrity endorsersnare problematic because the behavior of celebrity endorsers is uncontrollable (Money, Shimp, & Sakano, 2006; Miciak & Shanklin, 1994; Loulie, Kulik,n& Johnson, 2001). The unlawful or erratic behaviors of celebrities havenchronically threatened the reputation and sales of endorsed brands (see Money et al., 2006).

  Despite the uncontrollable risk of celebrity misbehavior, as mentioned earlier, studies examining the impact of negative celebrity information on endorsed brands are less common than studies about positive celebrity information. A study by Till and Shimp (1998) was the first to examine the effects of negative celebrity information on consumer response. Using an associative network model of memory (Anderson, 1976, 1983) as their conceptual foundation, they reported three important results: (a) as the size of the associative sets of celebrity endorser/brand decreases, the impact of negative celebrity information on the endorsed brand increases; (b) more negative impact on an endorsed brand is exhibited when negative celebrity information is provided prior to the ad cam-paign rather than after the campaign;n(c) when there is a strong association between celebrity endorser and endorsed brand, negative celebrity information is likely to result in a more negative brand evaluation.

  A later study by Money et al. (2006) used cultural difference as their theoretical framework to investigate the effects of negative information in celebrity endorser advertising. Their experiment compared U.S. and Japanese subjects and investigated whether the form of negative celebrity information (other- vs. self-oriented) resulted in differential evaluations of the celebrity endorsed brand. They reported that both American and Japanese subjects evaluated endorsed products more positively in the prese cenof self-oriented negative celebrity information (i.e., where the behavior is problematic but does not harm other people) than in the presence of other-oriented negative information (i.e., where the behavior is both problematic and does harm to other people) (Money et al., 2006).

  Although good contributions to the literature, the two studies did not examine two important factors regarding negative celebrity information research. The first factor is the degree of negativity in the endorser information (i.e., how the impact of negative celebrity information can differ depending on the level of seriousness). Gorn, Jiang and Johar (2008) found that the impact of negative information on the attitude toward company depends on the severity of the crime that a CEO commits. Accordingly, a CEO featured as committing intentional harm more negatively influenced the attitude toward company more than a counterpart depicted as committing unintentional harm (Gorn,Jiang, & Johar, 2008). Similarly, Louie and his colleagues (2001) investigated how a firm’'s stock returns are influenced when celebrity endorsers becomeninvolved in undesirable events. The study found stock market reactions werennegatively related when celebrity endorsers’' had higher culpability for badnbehavior, but positively related when the culpability was lower. For advertisers,nthe degree of negativity in the information is relevant to how companiesnrespond to a celebrity endorsement crisis.

  The second factor is the passage of time after negative information is aired about a celebrity. As noted in the previously mentioned examples (i.e., Tiger Woods, Michael Phelps, Alex Rodriguez, Adrian Peterson, and Ray Rice), advertisers are often inclined to terminate a contract of an endorser who has engaged in undesirable behaviors. However, celebrities, once asso-ciated with an undesirable event, often regain their reputation. By showing steady and extraordinary performance in their respective fields, celebrities counter negative associations with positive outcomes, and finally regain their popularity and fame. Indeed, Michael Phelps regained popularity after the marijuana incident (Crouse, 2009), and Kobe Bryant was awarded his first NBA Finals MVP trophy after the trial for his rape charge in 2003 (“"Mannon a mission,”" 2009).

  Taking these two neglected factors into account, this experiment tests (a)whether different degrees of negativity in celebrity information differentially im-pacts response to celebrity endorser advertising, and (b) whether the impactnof negative celebrity information decreases over time relative to rehabilitationnof celebrity reputation (i.e., the celebrity continues to performnon a high level after the undesirable event).

 Associative learning theory and consumer contamination theory

  Associative learning theory is a theoretical framework that has been usednto explain the effect of celebrity endorsers in advertising on endorsed brandsn(see Till & Shimp, 1998; Till & Busler, 2000). According to this theory,nindividuals hold related subjects in their memory as an interconnected structure.nThe structure represents an associative network consisting of patterns of nodesn(concepts) that are linked together (Anderson, 1976, 1983). In endorser advertising,nthe goal is to establish a link between celebrity and advertised brandnthrough endorsement (Till & Shimp, 1998). The effect of celebrity endorsementnis assumed to depend on how much the associative link between endorsernand brand produces a match-up effect in terms of congruence, fittingness,nappropriateness, and consistency (Till & Busler, 2000).

  An unstudied question is how does negative information that surfaces about an endorser influence consumer response to celebrity en-dorser adver-tising after a relationship between endorser and brand has been associatively established. Till and Shimp (1998) explained that negative information acti-vates celebrity-node and, to some degree, the endorsed brand node when such information is encountered. The result is that reduced evaluation of celebrity transfers to the endorsed brand. Till and Shimp (1998) tested the link between negative information and associative learning in a study that varied effect of one piece of negative information on the association sets between the celebrity endorser and the endorsed brand. The study found that negative celebrity information was related to deterioration in the asso-ciativenlink between celebrity endorser and brand.

  The experiment reported here builds on Till and Shimp (1998) to examine the impact of varying degrees of negativity and the timing of exposure to negative news (i.e., passage of time) on consumer response to celebrity endorsernadvertising. Rather than applying associative learning theory alone,however, it also adopts consumer contamination theory (Argo, Dahl, & Morales, 2006) as its conceptual framework.

  Consumer contamination theory posits that consumers evaluate contaminated products less favorably (Argo et al., 2006). The central principle of contamination is the law of contagion: “when a source (person/object) and a recipient (another perso/object) come into direct or indirect contact, the source influences the recipient, . . . so the source’s essence remains a part of the recipient even after contact has been broken” (as cited in Argo et al., 2006, p.82; Rozin & Nemeroff, 1990). Based on this principle, research has found that consumers tend not to buy a product that someone else has already touched. Furthermore, the researchers have demonstrated that the negative effect of physical contamination(a) increases with the proximity of a product to a location of contact and the number of contact sources and (b) decreases with the amount of time that elapses after last contact.

Figure 1. THE Mechanism of Physical Mental Contaminations under Associative Network

A.THE mechanism of physical mental contaminations under associative network

Contamination cue nodes

B.The mechanism of mental contamination on celebrity endoresr under associative network

Contamination cue nodes

  Applying principle of consumer contamination theory in accordance with associative learning theory, it may be surmised that a contamination cuenode can activate a product node in the consumer’'s memory set so that the activated link between the two nodes sequentially decreases the evaluation of the contaminated product (see Figure 1). The varied effects of con-tamination,nwhich are dependent on proximity to contact, the number ofncontact sources, and lapsed time since last contact, can be explained by the tenets of associative learning theory.

  Proximity to contact and number of contact sources are concepts derived from social impact theory (SIT) (Latané; 1981). SIT proposes that the im-pactnof social presence affects a target and the power of its impact results from immediacy (i.e., proximity) and number (i.e., how many) (Argo et al.,2006, p.83; Latané, 1981). When this mechanism is considered relative to associative learning theory (i.e., the mechanism happens in human memory network), the expectation is intensity of a contamination effect (social presence) impacts the strength of the associative link between product node and contamination cue node. That is, the strength of the link grows (or lessens) as proximity to contact is closer (or farther) and as the number of contact sources increases (or decreases). In addition, the increased number of contact sources enhances the contamination effect because the number of associative sets between contamination cues and a product also increases.

  Time elapsed since contact is a concept derived from construal-level theory(Liberman & Trope, 1998; Trope & Liberman, 2003). According to this theory, when people consider distant-past events, they form higher-level,less concrete construals about the event than when they consider near-past events (Trope & Liberman, 2003). In their study of consumer contamination,Argo et al. (2006) found that the salience of contamination is mitigated as time passes: the shorter the time lapse in consumer contact with a product, the stronger the contamination effect (p.83). From the perspective of associative learning theory, the concept can be also explained in this way: as time passes, the strength of the associative link between a contamination cue and a product is attenuated.

 Associative learning theory, consumer contamination theory, and negative information

  If consumer contamination theory can be explained in accordance with associative learning theory, the application sheds light not only on the ef-fects of physical contamination (Argo et al., 2006), but also on the “mental contamination” of an endorsed brand caused by negative celebrity information. For example, exposure to news about crime committed by a celebrity endorser might activate a new negative information node called “crime,” which might be activated and linked to consumer memory node for that celebrity endorser. The activated link could then sequentially contaminate the linked nodes for an endorsed brand. Because the structure of an associative network in memory is the same for either physical or mental contamination, consumer attitude toward the celebrity endorser could be mentally contaminated by negative information. As a result, like the se-quencenof physical contamination, consumer evaluation of an endorsed brand could be negatively contaminated. The expectation is that such proc-ess could take place either in short- or long-term memory situations (i.e., when consumers have pre-existing, longstanding nodes for celebrity and brand or newly established celebrity and brand nodes created by immediate exposure to negative information).

  If the mechanism of mental contamination is similar to the mechanism of physical contamination, various possible mental contamination cues could also moderate the impact of negative celebrity information on evaluations of celebrity endorser and endorsed brand. Therefore, in this study, we identified and examined two mental contamination cues: seriousness of negative information and time lapse in exposure to negative information (see Figure 1). These two mental cues share the same theoretical backgrounds as the three physical contamination cues identified in the work of Argo et al.

(2006), namely proximity to contact, number of contact sources, and time elapsed since last contact.

 First, the moderating impact of the seriousness of negative celebrity in-formation is similar to that of proximity to contact and number of contact sources on the physical contamination effect. As previously noted, proximity to contact and number of contact sources correspond to the immediacy and number factors in SIT (Latané, 1981), and these factors (i.e., closer location and higher number of contact sources) have been shown to enhance the impact of social presence (i.e., contamination). SIT also suggests one more factor that enhances the impact of social presence: the strength or importance of source (Latané, 1981). Therefore, as with proximity to contact and number of contact sources, it is assumed in this study that the strength (i.e., importance) of social presence can also moderate contamination effect.

 The assumption here is that the strength (i.e., importance) of social presence, defined as the perceived seriousness of negative celebrity information, moderates the impact of negative celebrity information. We predict that different levels of negative celebrity information will have differential influence on the seriousness that consumers might perceive. Thus, we expect that the associative link between the negative celebrity information and the celebrity endorser will strengthen as perceived seriousness heightens. Sequentially, the strengthened link is expected to influence the evaluation of the celebrity endorsed brand more negatively. Based on this assumption, we hypothesize:

  H1: Exposure to more serious negative information about a celebrity endorser in an advertisement will result in lower (a) celebrity endorser evaluations, (b) ad evaluations, (c) brand evaluations, and (d) purchase intention than exposure to less serious information about the celebrity.

  Second, the moderating impact of time elapsed regarding exposure to negative celebrity information is similar to the moderating impact of time elapsed since last physical contamination contact. As previously mentioned, Argo et al. (2006) posited that the salience of consumer contamination would be mitigated as time passes because people tend to form higher-level, less concrete construals about distant-past events than they do about near-past events (Trope & Liberman, 2003). According to associative learn ing theory, the strength of an associative link between a contamination cue and a product is attenuated as time passes. In the case of mental contamination effect, the strength of an associative link between negative information and the celebrity endorser should be expected to lessen as time elapses. As time passes, the implication is that consumers might construe negative celebrity information in a celebrity endorser ad less concretely and on a higher-level. In addition, celebrities can rehabilitate their reputations through positive behavior after negative information has become public. In light of Till and Shimp’'s (1998) finding that the impact of negative celebrity information on an endorsed brand increases as the size of the associative sets for celebrity endorser and brand decreases, we hypothesize that the increasing size of positive associative sets following disclosure of undesirable event will attenuate the impact of negative celebrity information:

H2: Recent exposure to negative information about a celebrity endorser in an advertisement will result in lower (a) endorser evaluations, (b) ad evaluations, (c) brand evaluations, and (d) purchase intentions than earlier exposure to negative information.

 Underlying mechanism of negative celebrity information

  In their consumer contamination effect study, Argo and her colleagues (2006) found disgust to be the underlying mechanism that causes consumers to respond unfavorably to products that have been touched by other consumers. As the degree of contamination increased, the feeling of disgust was found to lower the evaluation of the contaminated product. According to associative learning theory, the mediating mechanism can be explained differently: a contamination cue node can be viewed as activating a link between a brand node and a feeling of disgust. Thus, emotions may become attached to each event in an associative network so that the event can be remembered more easily (Bowner, 1981). In the context of celebrity endorser advertising, the mediating factors that influence the memory mechanism for negative celebrity information might include the endorser’'s expertise, trustworthiness, and attractiveness (Ohanian, 1990), for research indicates that feelings toward a celebrity and/or the meaning invested in a celebrity are expected to transfer to the endorsed brand through recurring association (Till & Shimp, 1998). In other words, a decrease in perceived celebrity expertise, trustworthiness, or attractiveness might lower evaluations of an endorsed brand. Thus, we ask the research question:

R1: How is the link between contamination cues and brand evaluations mediated by perceived expertise, trustworthiness, and attractiveness of the celebrity endorser depicted in celebrity endorser advertising?

Method

 The two hypotheses and the research question were addressed by a 2 (seriousness of negative celebrity information: minor vs. major offense) x2 (passage of time since offense: recent vs. past offense) + 1(control) between- subject experimental design. Prior to the main experiment, two pre-tests were conducted.

 Pretest 1: Selection of Contamination Cue

 The first pretest was conducted to select pairs of major/minor offenses for the main experiment. Based on the work of Gorn and his colleagues (2008), offense cases were identified and selected to build a plentiful pool of choices. Eight cases were tested and evaluated using five items on 7-point bipolar scales: unintentional/intentional; not serious/serious; minor offense/ major offense; not purposeful/purposeful; trivial matter/significant matter. Twenty undergraduate students at a Southeastern U.S. university participated in the first pretest (30% male, 70% female) in exchange for extra credit.

Table 1. Comparison of the Level of Negative Information between Minorand Major Offenses.

 Items asking about level of negativity for each offense had a significant level of internal consistency (Cronbach’s Alpha >.74). Thus, the items were averaged into one category called “level of negative information.” Then, four pairs of “major” offenses and “minor” offenses were compared using a paired-samples t-test. All of the pairs exhibited significantly different levels of seriousness (p < .05) between major and minor offenses. From these pairs, the incident of drug use was chosen for further investigation because it has been a chronic issue in U.S. sports industry since 2000s (see “Baseball commissioner,” 2009).

 Product, brand, and celebrity selection

  Before conducting the second pretest and the main experiment with the “drug use” pair, we selected a product, brand, and celebrity endorser for use in news articles and advertising stimuli. For product selection, we referred to the Simmons Choice 3 data and selected sports shoes, because college students regularly use this product category (over 60%). An actual shoe brand, “"Asics”", was selected because this company produces a wide range of sports shoe-ware (e.g., basketball, hockey, boxing, soccer, and so on).

  Previous studies indicate that using fictitious celebrities in experimental studies enables researchers to minimize prior exposure to and perceptions about endorsement relations (Till & Shimp, 1998), while using real celebri-ties enables the experimental situation to be more realistic (see Money et al., 2006; Kamins, 1990). To this end, an unknown, but real non-U.S. celeb-rity was selected as an endorser in the stimuli for the two reasons. First, because the goal of this experiment is to examine how the associative link between endorser and endorsed product are influenced by different levels of negative information, the establishment of associative link must be con-trolled internally in the experiment. If the researchers provided manipulated negative information about well-known celebrity, respondents may not be-lieve the information. Second, even though using unknown celebrity limits the external validity of experiment, the researchers intended to maximize it by providing the celebrity information based on real facts. Therefore, Dutch soccer player Edgar Davids was selected as an endorser in this experiment. Davids was once suspended due to taking a performance en-hancing drug (“Davids,” 2001).

 Pretest 2: Manipulation Check of News Articles

  The second pretest was conducted as a manipulation check for the fabricated news articles used in the experiment. News articles about “drug use” weredeveloped from the first pretest following the procedure of Gorn et al. (2008) (see Appendix A for details). The stimuli were created based on real news articles about the celebrity offense, but the degree of negative information was manipulated in accordance with minor or major offense condition.

  Sixty-one college students participated in the second pretest. These students did not participate in the first pretest or the main experiment. The participants were asked to evaluate the level of seriousness of negative information using the same measurement employed in the first pretest. As result, respondents evaluated the news article about the athlete who used a banned drug on purpose as more serious than the article about the athlete who used the drug by mistake (M’'s = 4.94 vs. 3.62; F(1, 59)=16.44, p <.01). Therefore, the manipulation check was successful.

 Main Experiment

  For the main experiment, 291 undergraduate students from the Southeastern U.S. university were randomly assigned to one of the experimental conditions. Participants were recruited from classes in exchange for extra credit and asked to go to a web page which randomly directed them to one of the five online survey sets: recent major offence, recent minor offence, past major offence, past minor offense, and no negative information. The procedure of the main experiment followed Money et al. (2006), which used different types of negative celebrity information for different experimental condition.

  The procedure also partially used the methodological technique of Till and Shimp (1998), which provided an association set between celebrity en-dorser and endorsed brand prior to the main experiment. Accordingly, partic-ipants were exposed to four pieces of product information and ten pieces of celebrity information in order to establish an association set in their memory. Celebrity information was provided in chronological order, from debut through recent activities. In the past offense condition, a news article reporting the celebrity’s minor or major offense was placed in the middle of the celebrity information. The article stated that the news had originally been reported in 2000. In the recent offense condition, the news article was placed at the end of the celebrity information: the article stated that the news had been reported a week ago. In the control condition, no negative information was exposed. After the treatments, participants were directed to look at a print advertisement portraying the celebrity’s endorsement of the brand (see Appendix B) and then instructed to complete the ques-tionnaire regarding the ad stimuli and celebrity endorser. All participants were debriefed after completing the survey tasks. 

  Measures. Consistent with the first and second pretest, level of negative information for each condition was averaged (Cronbach’s α >.75) and used for the manipulation check. Five, 7-point, bipolar items (i.e., bad/good, dis-like/like, uninteresting/interesting, irritating/not irritating, and unfavorable/ favorable) were used to assess attitude toward ad (Aad) (Mitchell & Olson, 1981; Cronbach’s α =.92). Four bipolar question items (i.e., bad/good, dis-like/like, unfavorable/favorable, and negative/positive) were used to measure Attitude toward brand (Ab) (modified from Andrews et al., 2000; Cronbach’s α = .97). PI was measured using an item from a previous neg-ative celebrity information study: “If the (endorsed product) were available at a local retail outlet and you had the income to afford it, what are the chances that you would purchase it?” Response options ranged in 10 percent increments from “No chance at all” (0%) to “Definitely would” (100%) (Money et al., 2006). Additionally, Ohannian’s (1990) 7-point bipolar, vali-dated scales were used to measure perceived expertise (Cronbach’s α = .92), trustworthiness (Cronbach’s α = .92), and attractiveness (Cronbach’s α = .91) of the portrayed celebrity endorsers. 

Results

  Thirty-three of the 291 subjects knew the celebrity endorser and thus were omitted from analysis. The following results are based on data analysis of responses from 258 respondents: 40.2% male and 59.8% female.

 Manipulation and Assumption checks

   The manipulation check for offense seriousness was successful in the main experiment. Respondents perceived intentional drug use as more serious than unintentional drug use regardless of the time of offense (for recent offense, M’s = 4.89 vs. 3.10; for past offense, M’s = 4.88 vs. 3.11; F(1, 204) = 106.49, p < .01). There was no significant difference between the time conditions for the offenses (F(1, 204) = .00, p >.05). Regarding the assumption check of equality of covariance matrices, a Box’s M-test re-vealed that the condition did not exist in either experimental set (Box’ M = 106.18, F = 1.60, p < .01). Thus, multiple sets of ANOVA were used to test the hypotheses.

 Hypothesis testing

Table 2. Evaluations of Endorser in Ad

  Table 3 displays the mean values for the subjects’ reactions to the Asics sports shoe ads endorsed by Edgar Davids in different negative celebrity information conditions. Multiple 2 X 2 ANOVAs resulted in significant main effects of offense level on celebrity trustworthiness (F (1, 201) = 36.51, p < .01), attractiveness (F (1, 201) = 6.01, p < .05), Aad (F (1, 201) = 22.93, p <.01), Ab (F (1, 201) = 21.05, p <.01), and PI (F (1, 201) = 17.76, p < .01). The main effect for celebrity expertise was not significant at the .05 (F (1, 201) = 2.68, p = .10).

  To further probe the significant main effects, follow-up comparison tests were conducted with the control group and two negative information conditions. For celebrity endorser and ad evaluations, a significant difference was found for the control group for major offense. However,  there was no significant effect of minor offence on trustworthiness (for control versus major offense: t(255) = -2.27, p <. 01; for control versus minor offense: t(255) = -1.94, p > .05), attractiveness (for control versus major offense: t(255) = -2.97, p < .01; for control versus minor offense: t(255) = -1.21, p > .05), expertise (for control versus major offense: t(255) = -2.24, p < .05; for control versus minor offense: t(255) = -.85, p > .05), Aad (for con-trol versus major offense: t(255) = -5.08, p = .00; for control versus minor offense: t(255) = -1.10, p > .05), Ab (for control versus major offense: t(255) = -4.37, p = .00; for control versus minor offense: t(255) = -1.15, p > .05), and PI (for control versus major offense: t(255) = -4.37, p < .01; for control versus minor offense: t(255) = -1.15, p > .05). Major offense produced a more negative effect on the celebrity trustworthiness, attractive-ness, Aad, Ab, and PI compared to minor offense, and minor offense did not show a statistical difference compared to the control group.

  The main effect of timing of negative information exposure was only found for celebrity trustworthiness (F (1, 201) = 4.82, p < .05). Although the mean difference for PI between recent and past offense timing appeared to meet our expectations, the statistical significance did not reach a sufficient level (F (1, 201) = 2.42, p = .12). Post hoc analysis on celebrity trust-worthiness indicated that both recent and past offense conditions were neg-atively perceived by the control group (for control versus recent offense: t(111.82) = -5.68, p <.01; for control versus past offense: t(132.13) = -3.44, p < .01). Thus, the main effect of negative information timing is more lim-ited than the main effect of negative information level. Additionally, sig-nificant interaction effects did not appear for the experimental set. 

  Based on the results of two experimental sets, H1 was fully supported by the data. H2 was supported for celebrity trustworthiness only.

 Mediation analysis

  To answer the research question (R1) regarding how the link between contamination cues and brand evaluations are mediated by feelings toward endorser, analysis identified celebrity endorser trustworthiness as the most sensitive variable relative to both independent variables (crime seriousness and time of exposure to negative news).  

 The method proposed by Zhao, Lynch and Chen (2010) was used to test the mediation effect of trustworthiness. Applying this new typology, which is developed from Baron and Kenny’s mediation procedure (1986), to our data set, we found indirect-only mediation (full mediation) of celebrity trustworthiness on Aad and Ab (see Table 3). On the other hand, celebrity trustworthiness was found to work in a complementary mediating role only between offense level and PI. That is, although the direct effect between offense level and PI was still significant even after adding trustworthiness mediator in the analysis, the mediator was found to exhibit complementary mediation between offense and PI. According to bootstrap analysis, indirect effect did not include 0 within the 95% confidence interval (-6.7768 to -2.0446). However, trustworthiness was not a significant mediator between time of offense and PI because time of offense did not have a significant direct effect on PI.

Discussion

  Using the frameworks of consumer contamination theory and associative learning theory, this experiment was designed to examine the effects of the level of negative celebrity information (major vs. minor offense) and the timing of exposure to the negative information (recent vs. past) on responses to celebrity endorser advertising. In addition to analysis of main and interaction effects, a mediation analysis was also performed to identify the prime mediating variable between the independent and dependent variables. As predicted, the more negative the information about a celebrity endorser, the more negatively consumers respond to advertising featuring the endorser. In the experiment, major offenses produced significantly more negative impact on dependent variables than did minor offenses. These findings indicate that the potential effectiveness of celebrity endorser advertising is contaminated by negative endorser information. The time lapse between negative information exposure (past and recent news) was found to have limited impact in celebrity endorser advertising. The factor was significant only for perceived endorser trustworthiness. According to mediation analysis, endorser trustworthiness was found to be a significant mediating factor. The mediation analysis found that perceived trustworthiness fully mediated (indirect-only mediation) between the independent variables and Aad/Ab and complementarily mediated between the level of offense and PI.

  The experimental findings have a number of implications for research on celebrity endorser advertising. Theoretically speaking, the findings reveal that the mechanism of mental contamination provides a viable explanatory framework for considering endorser effects in celebrity endorser advertising. In this study, exposure to major offenses resulted in greater negative impact on consumer responsiveness than exposure to minor offenses across both stimuli sets, regardless of timing of negative information exposure. Although the result is somewhat different from the physical contamination effect of Argo et al. (2006), the finding is plausible when considering the unique influence of negative celebrity information on human memory.

  It has been shown that negative celebrity behavior receives more attention, is better encoded, and is more easily recalled than positive information (Folkes, 1988; Ybarra & Stepban, 1996). Therefore, unlike physical contamination cues, the salience of mental contamination caused by negative celebrity information might not be easily mitigated over time, even if consumers form higher-level, less concrete construals about distant-past events. From this perspective, the result of the mediation analysis regarding PI is also plausible. Although the passage of time helps to rehabilitate endorser trustworthiness and attitudes toward the endorsed ad and brand, consumers tend to focus more on the content of negative information, especially when they are formulating purchase intention.

  In particular, it should be noted that level of negative information did not make any significant difference between the minor offense condition and the no offense condition across all dependent variables. This finding is plausible because it is consistent with the findings of Money et al. (2006): when a celebrity endorser committed a self-oriented offense (the offense not harming other people) in his/her youth, the researchers found purchase intentions increased more compared to the no offense condition and other- oriented offense condition (the offense harming other people). Money and his colleagues reasoned that sympathy and empathy for the celebrity led to a more positive evaluation. However, it is reasonable to believe that the concept of self- and other-oriented negative information could be included in the low and high levels of negative information in the theoretical framework for the current study. Accordingly, consumers might perceive other- oriented negative information and intentional offenses as more serious violations than self-oriented negative information and unintentional offenses. As a result, consumers might exhibit more negative feelings toward celebrity endorsers’major offenses than their minor offenses.

  The mediating role of endorser perceived trustworthiness is also an important finding. Similar to the feeling of disgust in physical contamination theory (Argo et al., 2006), a celebrity endorser’ perceived trustworthiness was found to indirectly or complementarily mediate the study’ independent and dependent variables. Theoretically, this finding confirms not only that consumer contamination theory can be applied to mental contamination caused by negative celebrity information, but also that consumer contamination theory falls within the boundaries of associative learning theory.

  According to Till and Shimp (1998), negative information activates celebrity nodes and endorsed brand nodes to some degree; therefore, such unfavorable evaluations of celebrity endorsers might transfer to endorsed brands. This study demonstrates this theoretical mechanism by finding that perceived trustworthiness of a celebrity endorser is located between the celebrity node and the endorsed brand. Due to the level of negative information, however, the trustworthiness in negative celebrity information likely works in indirect or complementary mediating roles depending on dependent variables. Accordingly, the implication is that purchase intention is more influenced by the level of negative information than by the attitudes toward ad and brand.

  In the view of practical contribution, the findings suggest that advertisers should be especially concerned about level of negative celebrity information, but they would be wise not to ignore the time factor associated with negative endorser information. In this experiment, time lapse had a significant indirect impact on the dependent variables via perceived endorser trustworthiness, although its influence was weaker than level of negative information. For advertisers, the results have practical implications for managing celebrity endorser advertising. A damaged celebrity endorser can endanger the equity health of an endorsed brand. First, if a contracted celebrity endorser commits a minor offense, the findings suggest that the advertiser might consider postponing or withdrawing endorser ads until the celebrity’s perceived trustworthiness recovers with the passage of time. Second, even if a celebrity has committed a relatively minor offense in the past, an en-dorsement contract with that celebrity might be profitable, as long as the advertiser is convinced that the celebrity will not commit such an offense again or that the celebrity’s reputation can be rehabilitated over time. However, when the celebrity endorser’s offense is serious, the advertiser might be wise to terminate or avoid an endorsement contract with a problem celebrity, regardless of the timing dynamics of negative behavior. Whether committed recently or in the distant past, a serious offense might follow a celebrity and diminish the effectiveness of advertising featuring the problem offender.

Limitations and Suggestions for Future Research

  Like all studies, this experiment has limitations. First, the selected brand does not represent a full array of product categories. Thus, care should be exercised in generalizing the results to other products in the marketplace. Also, the current study did not consider service products (e.g., credit card and insurance), for which a celebrity endorsers’trustworthiness is probably more important. Further research is recommended to examine the impact of negative celebrity information on responses to celebrity endorser advertising for other products and services. Moreover, internal validity (i.e., use of online experiment) and external validity (i.e., unknown celebrity endorser) limitations make interpretation of the results problematic. Even though we used a real brand, real celebrity endorser, and real offenses, the external validity might have been compromised because the participants were relatively unfamiliar with the endorsed brand and celebrity endorser. Future research might consider using more well-known brands and celebrities. It is also recommended that additional investigations be conducted under more controlled lab conditions to enhance internal validity as well as in field situations to improve external validity.

Conclusion

 Consumer contamination theory in conjunction with associative networks in human memory is a useful approach to studying how negative information affects celebrity endorser advertising. This experiment found that (1) seriousness of negative information has a greater impact on how consumers respond to celebrity endorser ads than timing of such information (2) the timing of negative celebrity information only influences celebrity endorser perceived trustworthiness, an indication that the timing of negative information cannot be ignored due to its indirect effect on endorser trust; and (3) perceived trustworthiness for a celebrity endorser works in indirect or complementary mediating roles for negative endorser information processing. The lesson for advertisers is that the seriousness of negative information about celebrity endorsers and their trustworthiness must be fully considered when planning and managing celebrity endorser advertising.

APPENDIX

A: Created News Stimuli of Major and Minor Offense>

B: Created Advertising Stimulus>