About 25% of US advertisements employ celebrity endorsers (Shimp 2000). These actions suggest many US firms have bought into the premise that celebrity endorsers positively impact consumer attitudes towards an advertisement and the associated brand, consumers' purchase intention, as well as other measures of effectiveness (Kaikati 1987; Ohanian 1991; Tripp et al. 1994; Goldsmith et al. 2000; Erdogan et al. 2001). There is substantial research on the topic, suggesting celebrity endorsement may materially improve financial returns for companies that employ them in their advertising campaigns (Farrell et al. 2000; Erdogan et al. 2001).
The importance of celebrity endorsers, however, does not lie in the fact that they are used by firms who wish to increase revenue, but in how these celebrities add value to a company, brand or product. The literature implies that celebrities add value through the process of meaning transfer (McCracken 1986, 1989). The meaning transfer model posits that celebrities develop a persona through the types of roles they play in society as well as how they are portrayed in the media. Collectively, the culturally constituted society then assigns meaning to celebrities. When celebrities endorse a product, the meaning developed around a particular celebrity will - or at least it is hoped for by advertisers - transfer to a company, brand, or product (Erdogan & Baker 2000). Thus, when a consumer identifies with a celebrity (identification occurs when a person is willing to accept influence from another person) (Kelman 2006, p. 3), he/she purchases the product in the hope of claiming some of these transferred meanings for their own lives (McCracken 1989).
Researchers have examined this transfer of celebrity meaning to the product and its resultant effectiveness by examining the source effects of celebrity endorsers. Two narrative reviews have attempted to amalgamate the literature pertaining to the effectiveness of celebrity endorsers in advertising (Kaikati 1987; Erdogan 1999). Kaikati (1987) identified the prevalent types of celebrity endorsers, advantages of using celebrity endorsers, hazards associated with using celebrity endorsers, and Federal Trade Commission guidelines. Erdogan (1999) reviewed the effectiveness of celebrities, advantages and disadvantages associated with celebrity endorsement, and the application of the source credibility and source attractiveness models. More recently, researchers have produced several empirical studies addressing specific subject areas within the broader realm of celebrity endorsement effectiveness. However, to our knowledge, no systematic effort to quantitatively integrate this literature has been undertaken.
By reporting results derived from a meta-analysis of the effectiveness of celebrity endorsement literature, this study addresses this gap. This study investigates the source effects of celebrity endorsers in the advertising and communication literature to determine whether the existing body of literature - considered collectively - can yield any theoretically relevant and managerially relevant insights. More specifically, this paper reviews quantitative studies in the literature and documents the relationships between use of a celebrity endorser and the resulting effectiveness of that endorsement.
An examination of the relevant literature reveals that numerous celebrity endorsement source effects have been investigated. Naturally, if researchers were better able to identify the most influential source effect(s), they could focus investigations on models composed of superior explanatory variables. And were such an ideal to eventuate, more accurate empirical results would surely follow. In pursuit of such a model, the first research question addressed in this study was:
Q1: What are the most important source effects of celebrity endorsement?
Researchers have occasionally investigated whether matching the dominant traits of specific celebrities to specific products' attributes improves advertising effectiveness (Batra & Homer 2004). However, many studies have failed to investigate this particular moderator effect. This oversight may be critical. Certainly, it appears important to examine the product/celebrity endorsement fit moderator across the extant body of findings to ascertain whether this moderator might account for some portion of the differences in findings observed to date.
In pursuit of these outcomes, this study's second research question was:
Q2: What effect does the celebrity endorser/product fit have on celebrity endorsement effectiveness?
The studies that comprise the extant celebrity endorsement literature vary widely in their sampling units, data analytic approaches, experimental settings, and statistical findings. When the resulting findings are clear, such variation can increase the generalisability of the results. However, since most findings in any meta-analysis require interpretation, varying theoretical and methodological approaches can prove confusing and unclear (Rosenthal & DiMatteo 2001). Therefore, it is always important to determine which research techniques yield the most accurate results. The associated insights may help prevent the onset of problems in future studies (Wilson & Sherrell 1993). The third research question addressed in this study was:
Q3: Within the relevant research domain, what methodological variables produced the most variation in terms of significant findings?
Whenever a meta-analysis is conducted, source effects as well as interaction of those source effects with other substantive variables should be explored. This is important because the exploration invalidates one of the major criticisms associated with the technique. That criticism derives from the suggestion that meta-analyses are themselves limited because often only main effects are examined (Cooper & Hedges 1994). It is therefore important to include interaction effects as a potential moderator. If main effects and interaction effects significantly differ, researchers might reason that further exploration of celebrity endorser effectiveness is required in order to determine whether other moderators critically influence source effects (Wilson & Sherrell 1993). In pursuit of this objective, a fourth research question was examined:
Q4: How do interaction effects differ from main effects in celebrity endorsement source effects literature and what implications do these differences have for researchers?
Finally, as the existing literature associated with the effectiveness of celebrity endorsement is explored empirically, significant gaps in the literature that are not currently apparent may be revealed. The results of this meta-analysis should bring at least partial closure to what has been discovered in prior studies - in effect, tying extant findings together while providing a platform from which fruitful future research endeavours may be launched. The final research question driving this study was:
Q5: What potentially rewarding topic areas remain relatively under-explored in the current celebrity endorser literature?
A review of the literature revealed the following key predictors: (1) celebrity performance, (2) negative information, (3) celebrity credibility, (4) celebrity expertise, (5) celebrity trustworthiness, (6) celebrity attractiveness, (7) celebrity familiarity, (8) celebrity likeability, and (9) celebrity/product fit. Each predictor is examined below.
In the context of this study, celebrity performance refers to the level of achievement a celebrity attains at any given time in their chosen profession. Performance could refer to the level of athletic performance, acting success, musical success, etc. of any given celebrity. This perceived level of performance may be enduring, as in the case of the Rolling Stones who have maintained a high level of rock and roll notoriety for decades, or fleeting, as in the case of one-hit-wonder Right-Said-Fred.
However, no guarantee exists that any celebrity can continuously produce popular music, act in financially successful movies, or win sports championships. In fact, depending on their level of performance, celebrities do rise and fall in popularity throughout their entire career (Agrawal & Kamakura 1995). When a celebrity fails to perform acceptably, as defined by consumers, a celebrity endorser's effectiveness tends to decline (Agrawal & Kamakura 1995).
Negative Celebrity Information
Since repeated pairings of a brand and celebrity strengthen the associative link consumers establish between brand and celebrity, negative information about the celebrity may negatively impact the endorsed brand (Erdogan & Baker 2000). Till and Shimp (1998) observed that a strong associative link between celebrity and product must be present before negative celebrity information lowers brand evaluations. Regardless of the strength of association consumers perceive between the celebrity endorser and the product, negative information about celebrity endorsers can put a firm's products and image at risk. Pop star Michael Jackson's child molestation indictment produced negative connotations. The resultant celebrity image transformation would have bottomed out Jackson's endorsement effectiveness for Pepsi - had the company not long since opted out of Jackson as an endorser due to his prior alleged indiscretions.
Celebrities are generally viewed by consumers as credible sources of information about the product or firm they endorse (Goldsmith et al. 2000). The literature exploring celebrity endorsements has generally employed one of two foundational source models: (1) the source-credibility model, and (2) the source-attractiveness model (Erdogan 1999).
Source credibility can be defined as 'a communicator's positive characteristics that affect the receiver's acceptance of a message' (Ohanian 1990, p. 41). The source-credibility model analyses the factors leading to the perceived credibility of the communicator (Hovland et al. 1953). The model contends that the effectiveness of a message depends upon the perceived level of expertise and trustworthiness associated with an endorser or communicator (Erdogan 1999). When considered jointly, expertise and trustworthiness are presumed to embody the source credibility construct (Ohanian 1990).
The source-attractiveness model posits that the attractiveness of any source is determined by the communication receiver's perceptions of the source's similarity, familiarity, and likeability. Essentially, if consumers perceive a celebrity endorser as similar to them and they are familiar with and like the celebrity, they will tend to find the celebrity more attractive.
Erdogan (1999, p. 298) defines celebrity endorsers' expertise as 'the extent to which a communicator is perceived to be a source of valid assertions'. The literature investigating source credibility in settings involving persuasive communication generally indicates that a receiver's perception of the source's expertise positively influences source effectiveness (Ohanian 1990). Respondents' actions in response to the source's recommendations seem to vary directly with the source's perceived level of expertise and the target person's level of agreement with those recommendations. Subjects exposed to a source perceived as highly expert exhibit more agreement with the source's recommendation than did those exposed to a source with low expertise (Ohanian 1990). The level of perceived celebrity expertise should predict celebrity endorser effectiveness.
Trustworthiness is the degree of confidence consumers place in a communicator's intent to convey the assertions s/he considers most valid (Ohanian 1990). Giffin (1967) describes favourable disposition, acceptance, psychological safety, and perceived supportive climate as favourable consequences of trust. Much of the literature supports the positive effect of trustworthiness on effectiveness (Chao et al. 2005). Miller and Baseheart (1969) found that a highly opinionated message from a highly trustworthy communicator produces an effective attitude change, while non-trusted communicators' impact proved immaterial. Perceived communicator trustworthiness has also been shown to produce a greater attitude change than perceived expertise (McGinnies & Ward 1980). The extant literature on celebrity endorsers suggests that trustworthiness is an important predictor of celebrity endorsement effectiveness.
Celebrity endorsement literature has indicated that attractiveness is an important indicator of effectiveness (Chao et al. 2005); however, the attractiveness construct is multi-dimensional in nature. Far from just encompassing aspects of physical attractiveness, which themselves are rather arbitrary, attractiveness also entails other characteristics such as personality and athletic ability (Erdogan 1999).
Some authors suggest that physically attractive celebrities are a predictor of advertising effectiveness (Till & Busler 2000). Certainly, physically attractive celebrities are generally viewed more favourably on various personality traits than their less attractive counterparts (Kahle & Homer 1985; Eagly et al. 1991). Joseph (1982) studied endorsers' attractiveness beyond the level of personality traits. Specifically, he examined the impact of endorser attractiveness on opinion change, product evaluation, and other measures of effectiveness. The study concluded that attractive endorsers have a more positive impact on the products they endorse than less attractive endorsers. Baker and Churchill (1977), however, found that while attractiveness was effective in increasing positive advertisement evaluations, it was not effective in producing stronger purchase intentions. Similarly, Caballero et al. (1989) observed that endorser attractiveness had no effect on advertising effectiveness. Within the broader context of celebrity endorsement, endorser attractiveness is certainly a relevant construct. However, the nature and scope of the attractiveness construct remains uncertain, and therefore appears worthy of additional attention.
Celebrity Familiarity and Likeability
In some studies, celebrity familiarity and likeability are treated as if each were analogous to attractiveness (Kahle & Homer 1985). Each celebrity attribute may, in fact, be subsumed within the attractiveness construct. But other studies address familiarity and likeability separately, investigating each construct's effect on effectiveness as if each were distinct from endorser attractiveness (O'Mahoney & Meenaghan 1998).
In the celebrity endorsement context, familiarity has been defined as 'knowledge of the source through exposure' (Erdogan 1999, p. 299). Likeability is defined as 'affection for the source as a result of the source's physical appearance and behaviour' (Erdogan 1999, p. 299). On this basis, in this study the two constructs are treated as if each were distinct from attractiveness. This path was followed in an attempt to determine each construct's value as a possible predictor of celebrity endorsement effectiveness.
The celebrity/product fit, also called the 'match-up hypothesis', refers to the harmony of the match between the celebrity endorser and the product being endorsed (Till & Busler 2000). Celebrity/product fit is thought to function as a key determinant of endorsement effectiveness (e.g. Friedman et al. 1978; Friedman & Friedman 1979; Kahle & Homer 1985; Kamins 1989, 1990; Kamins & Gupta 1994; Erdogan & Baker 2000; Till & Busler 2000; Erdogan et al. 2001; Batra & Homer 2004). Celebrity effectiveness does vary across different product types. Friedman and Friedman (1979) concluded that the better the celebrity/product fit, as perceived by consumers, the higher the level of endorsement effectiveness. Till and Busler (2000) found that celebrity/product fit was effective for only certain measures of effectiveness such as brand attitude, but not for other measures such as purchase intention. Regardless of the impact celebrity/product fit has on effectiveness, the absolute weight of the existing literature suggests that the phenomenon should play an important role in celebrity endorser effectiveness (Till & Busler 2000).
Meta-analysis is a quantitative review of a research domain that illustrates the typical strength or effect of a phenomenon, its variability, its statistical significance, and the nature of the moderator variables from which one can predict the comparative strength of the effect or phenomenon (Rosenthal 1995). Many advantages result from employing meta-analysis as a research method. The primary advantage clearly derives from the method's ability to scrutinise any literature as a meaningful whole. At that point, similarities and differences among methodologies and the results of many studies can be uncovered more easily. Meta-analysis also permits small and non-significant effects to contribute to a more complete picture of the results of a stream of literature (Cooper & Hedges 1994). Finally, meta-analysis identifies moderators by identifying and exploring potentially meaningful patterns in quantitative studies.
When meta-analysis is employed, exploration, as opposed to simple confirmation, of the relevant literature is emphasised. This emphasis is practically and theoretically significant. Exploration provides a more effective means of formulating causal influences and understanding, at least inferentially, why various results occurred (Cooper & Hedges 1994).
In accordance with Rosenthal (1995), we focused on providing a meta-analysis that would provide a succinct look into celebrity endorser source effects and effect size. Hence, five experts were consulted to aid in evaluating studies for inclusion in this meta-analysis. The majority of studies in this meta-analysis measured celebrity endorsement effectiveness via the foremost categories of constructs: (1) purchase intention, (2) brand attitude, and (3) attitude towards advertisement, (4) believability, (5) recall, and (6) recognition. Other studies included: measured effectiveness as actual purchase behaviour, expected excess returns, or other measures of behavioural intention and attitude. Some studies used only one measure celebrity endorsement effectiveness (e.g. Goldsmith et al. 2000; Erdogan et al. 2001; Forehand & Perkins 2002). But others examined celebrity endorsement source effects across several measures, using two or more constructs categories to assess perceived celebrity endorsement effectiveness (Till & Shimp 1998; Silvera & Austad 2004).
While effectiveness has been measured in the included studies by various constructs, the purpose of meta-analysis is to focus on the effect size of chief explanatory variables (Cooper & Hedges 1994). Therefore, when conducting a meta-analysis, effect size is the essential component, not the individual dependent variables used in the studies under consideration (Rosenthal 1995). From the studies used in this meta-analysis, the relationship between the independent source effect variables and the measure of effectiveness was converted into a weighted effect size that was, in turn, used in the subsequent analysis as the dependent variable.
Selecting the Relevant Literature
A comprehensive literature review identified all relevant empirical studies that dealt directly or indirectly with celebrity endorsement. The ABI Inform, EbscoHost, Google Scholar, Digital Dissertations and Science Direct databases were all searched. Peer-reviewed academic journals, as well as trade journals, were searched in the areas of marketing, advertising, business, psychology, and communication.
To qualify for inclusion, studies must have specifically evaluated celebrity characteristics. Advertising studies that examined source effects without specifically addressing celebrities or provided an evaluation of celebrities in comparison to other types of spokespersons were too general for inclusion. Eighty-seven studies were originally identified as worthy of further evaluation. Of this sample, 12 articles were conceptual and another 43 studies failed to report information required to conduct the analysis due to their focus on aspects of celebrity endorsers outside the realm of source effects. In total, then, 32 studies were retained for analysis in this meta-analysis. Total effect size was 266. In all, 27 journal articles, two unpublished studies, and three unpublished dissertations were included.
While an extensive search was conducted, the articles included in this meta-analysis probably represent a sample, as opposed to the complete population, of the celebrity endorsement literature. But given the wide diversity of journals as well as the unpublished works (See Table 1) contained in the sample, it should be representative of the relevant advertising literature.
Table 1: Journal articles collected for meta-analysis
Correlations were obtained from the sample studies using formulas from Cooper and Hedges (1994). As suggested by Hunter and Schmidt (1990), weighted correlations were used in the analysis to account for sample size and also as the dependent variable in the analysis. For the independent variables, initial coding consisted of coding the independent variables exactly as the authors articulated them. These independent variables were then reinterpreted and grouped into redefined variables.
These redefined variables were then presented to five experts in the area of advertising for confirmation. A content analysis of the studies yielded six substantive and methodologically meaningful dimensions on which all studies could be compared. The dimensions were source manipulation, celebrity/product fit, experimental effect, study setting, sample type, and origin. Coding was performed by two trained coders. Any disagreements about potential inconsistencies in the coding were resolved through discussions involving the authors and consulting experts.
Table 2 summarises the effects from the sample studies. Among the original 266 total effects, 185 were statistically significant (p < 0.05). Forty-four per cent of the retained studies examined celebrity/product fit. With respect to methods characteristics, 86% of studies sampled reported main effects, 62% used a survey instrument, and 52% used a student sample. Exactly 60% of the studies used a US-based sample.
Table 2: Percentage distribution of source effects
These data were skewed. This was expected, given the small effects characteristic associated with behavioural research (Sawyer & Ball 1981; Wilson & Sherrell 1993). To ensure that interpretation of the results was not influenced by transformation of the data, a nonparametric procedure was performed on the weighted correlation coefficients. The Kruskal-Wallis test was used to analyse the data from the non-normal population (Iman & Conover 1983; Wilson & Sherrell 1993). Kruskal-Wallis provides a powerful alternative to the t-test for the equality of means (Wilson & Sherrell 1993). Compared with the F-test, the Kruskal-Wallis test has an asymptotic efficiency of 95.5% when used with non-normal populations (Siegal 1956). The mean correlation of each variable is provided to permit comparison of effect sizes between variables.
Publication bias was analysed using Rosenthal's (1991) fail-safe N. To render Znormal for z-transformed correlation coefficients non-significant, Znormal would have to be less than 1.645 (p > 0.05). Znormal for this study is 23.02. Thus the results of this meta-analysis are statistically significant. In fact, the results suggest this study would have to find 14,454 non-significant effects before the 185 significant results could be considered due to chance. This robust result infers the use of published studies does not threaten the integrity of this study's findings.
Results for Predictors of Celebrity Endorsement Effectiveness
The average ranking of the weighted effect size is shown in Table 3. For source manipulation, the various levels were significantly different. The mean ranking of the source manipulation variables indicated that 'negative information' (MR = 189.94) exercised the most influence on celebrity endorser effectiveness. Celebrity 'trustworthiness' exercised the second most influence (MR = 151.59), followed closely by celebrity 'expertise' (MR = 150.96), and 'attractiveness' (MR = 140.94). Celebrity 'credibility' (MR = 114.11), 'familiarity' (MR = 100.36), and 'likeability' (MR = 94.25) subsequently followed. Celebrity performance (MR = 24.04) was the least influential endorser source effect. The source manipulation variables that significantly differed from each other (p < 0.05) are also shown in Table 3.
Table 3: Average ranking of weighted effect size
Studies that did not examine celebrity/product fit had a statistically significant higher average ranking (Ï‡2 = 77.01, 1 df, p < 0.00) than studies that examined celebrity/product fit.
Results for Methodological Variables
The average rankings of effects were statistically significant for all four method dimensions examined (See Table 3). In studies reporting interaction effects, the average ranking of the weighted correlation was significantly greater than in studies reporting main effects (Ï‡2 = 14.23, 1 df, p < 0.00). For experimental studies, the average ranking of the weighted correlation was statistically greater (Ï‡2 = 58.61, 1 df, p < 0.00) than for cross-sectional studies, where a survey was administered. For studies using college students as subjects, the average ranking of the weighted correlation was statistically greater (Ï‡2 = 100.34, 1 df, p < 0.00) than for studies using non-student subjects. Finally, studies using subjects from the US had a significantly greater average ranking (Ï‡2 = 49.78, 1 df, p < 0.00) of the weighted correlation than studies using non-US subjects. The practical and theoretical implications associated with each observed effect are substantial. These are discussed in the next section.
This meta-analysis confirms a good portion of the 30+ years of celebrity endorsement literature. More meaningfully, it also ties together much of that literature - resulting in a more conclusive picture.
Negative information about the celebrity exercised the largest impact (R = âˆ’0.62) on celebrity endorsement effectiveness in advertising. This result underscored the high risk associated with using celebrity endorsers as well as the huge impact negative information about that celebrity can have on consumer perceptions. This finding also suggests that when negative information about a celebrity endorser emerges, the revelation can dilute the equity of the product/brand associated with the celebrity. The necessity, whenever possible, to pre-empt the arrival of negative information about celebrity endorsers is clearly confirmed.
The results of this meta-analysis likewise infer that positive celebrity information and image can also transfer to the product/brand. But of equal importance, these results infer negative information transfers to the product/brands, as well. This implies firms should exercise extreme caution when choosing celebrity endorsers. Firms should also develop quick response contingency plans to countermand any possible negative information/events. The likely payoff associated with adopting the right celebrity endorser, as well as the high costs associated with tapping the wrong option, again are emphasised.
Source expertise and trustworthiness invariably contribute to source credibility. But prior research implies source expertise and trustworthiness may make independent contributions to source effectiveness (Mowen & Minor 2006). In a confirmatory manner, this meta-analysis revealed that celebrity 'trustworthiness', and 'expertise' along with 'attractiveness' each exercised more influence on effectiveness than did the celebrity 'credibility' source effect. This suggests that a source credibility construct composed of trustworthiness, expertise, and attractiveness dimensions should be employed in future studies but credibility may also encompass other dimensions.
Trustworthiness and expertise have each traditionally been associated with source credibility, with expertise generally identified as the more important dimension (Homer & Kahle 1990). But this quantitative synthesis revealed trustworthiness was the second most important predictive construct. Trustworthiness typically includes the items trustworthy, dependable, honest, reliable and sincere (Ohanian 1991). Moreover, in an advertising context, trustworthiness refers to the honesty, integrity and believability of a celebrity endorser (Erdogan 1999). Morgan and Hunt (1994, p. 23) conceptualised trust as 'confidence in an exchange partner's reliability and integrity'. These conceptualisations imply that the trustworthiness construct should proxy the confidence consumers have in the reliability and integrity of a given source. Celebrity trustworthiness certainly represents a critical issue for advertisers.
One viable, and managerially relevant, explanation for the significance of trustworthiness may ensue from the continuing shift in the emphasis of marketing practice away from a sales orientation and towards a customer relationship management orientation. Among the studies investigated in this meta-analysis, 75% were published between 1990 and 2005 - essentially the height of the CRM movement. In the emerging marketing environment, few should be surprised that celebrity endorser trustworthiness was revealed to play so important a role.