Celebrity vs. Influencer endorsements in advertising: the role of identification, credibility, and Product-Endorser fit

Abstract In their marketing efforts, companies increasingly abandon traditional celebrity endorsers in favor of social media influencers, such as vloggers and Instafamous personalities. The effectiveness of using influencer endorsements as compared to traditional celebrity endorsements is not well understood. Therefore, the present research investigated the impact of celebrity vs. influencer endorsements on advertising effectiveness (attitudes toward the advertisement and product, and purchase intention), moderated by product-endorser fit. Moreover, this research investigated two potential mediators underlying this relationship: identification (perceived similarity and wishful identification) and credibility (trustworthiness and expertise). Two experiments (N = 131, N = 446) investigated celebrity vs. influencer endorsers with good vs. poor fit with a beauty and a fitness product (Study 1), or a food and a fashion product (Study 2). Overall, our results showed that participants identify more with influencers than celebrities, feel more similar to influencers than celebrities, and trust influencers more than celebrities. In terms of advertising effectiveness, similarity, wishful identification, and trust mediate the relationship between type of endorser and advertising effectiveness. Product-endorser did not explain the relationship between type of endorser and any of the mediating and dependent variables. In all, our results show the added value of using influencer endorsers over celebrity endorsers and the importance of similarity, identification and trust in this process.

effectiveness (e.g., Atkin and Block 1983;Erdogan 1999;Amos, Holmes, and Strutton 2008;Bergkvist and Zhou 2016). However, in addition to using 'traditional' celebrities such as actors, supermodels, and athletes to add value to their brand, companies increasingly turn to social media influencers (also called 'micro-celebrities'), such as vloggers and 'Instafamous' personalities, to endorse their brands (Marwick 2015). In contrast to traditional celebrities who have gained public recognition because of their professional talent, social media influencers (from now on called 'influencers'), have gained fame by successfully branding themselves as experts on social media platforms (Khamis, Ang, and Welling 2017). By enthusiastically sharing self-generated content on topics like beauty, fitness, food, and fashion, these (mostly female) social media users have gained a large follower base, turning their online social presence into a primary profession such as 'fashionblogger', or 'fitgirl' (Lin, Bruning, and Swarna 2018).
Despite the increasing deployment of influencers for brand endorsement, scientific knowledge on their marketing value is limited. Whereas the effectiveness of traditional celebrity endorsement has been widely investigated (Bergkvist and Zhou 2016), the impact of influencer endorsement on brand responses is relatively understudied. Although recent qualitative research suggests that influencers, compared to celebrities, have a stronger impact on brand attitudes and purchase behaviors of young consumers (Djafarova and Rushworth 2017), no study to date has directly compared the two endorsement types. Scholars as well as practitioners would benefit from more insight into the effects of celebrity versus influencer endorsements on advertising effectiveness. Therefore, in two experiments, the present research will investigate the effectiveness of endorsement advertising by celebrities versus influencers in terms of attitude toward the advertisement, attitude toward the advertised product, and purchase intention. In addition, we investigate two important mediating processes that may explain potential differences in effectiveness between these two types of endorsements: identification (in particular, perceived similarity and wishful identification; cf. Hoffner and Buchanan 2005) and credibility (consisting of trustworthiness and expertise; cf. Sternthal, Phillips, and Dholakia 1978). Both identification and credibility have been shown to play a key role in explaining celebrity endorsement effects, and may play a significant role in influencer endorsement as well (Chapple and Cownie 2017;Djafarova and Rushworth 2017).
Finally, since the effectiveness of an endorsed advertisement is inseparably tied to the degree in which the endorser fits the advertised product (e.g., Kamins and Gupta 1994;Till and Busler 2000;Fink, Cunningham, and Kensicki 2004), we will take product-endorser fit into account as a moderator in the relationship between type of endorser and marketing outcomes. We will argue that the effect of product-endorser fit on advertising effectiveness may be even more pronounced for influencers than for celebrity endorsers.

Theoretical background
Influencers attract millions of followers by sharing content curated from their daily lives on platforms like Instagram and YouTube, evolving around one particular domain of interest (Chapple and Cownie 2017;MediaKix 2017). Although 'traditional' celebrities have also found their way to social media, influencers built their careers online and were unknown to a general public before. Marketers have eagerly embraced these influencers as spokespersons for their brands, and advertisers are investing large budgets on influencer endorsements (WFA 2018). Influencers do not only have the power to directly influence the purchase decisions of a large audience, but their followers also judge them to be reliable information sources (De Veirman, Cauberghe, and Hudders 2017;Djafarova and Rushworth 2017).
A growing body of academic research is investigating the merits of influencer marketing, and the processes playing a role in influencer effects on brand responses. For example, Lee and Watkins (2016) showed that vloggers positively affect consumer purchase intentions for (luxury) brands promoted in their vlogs. Likewise, in an interview study by Chapple and Cownie (2017), consumers stated to regularly follow lifestyle vloggers' product recommendations, either by buying a product themselves or recommending it to others. In this study, participants considered vloggers as credible sources of information, mainly driven by perceptions of trust and similarity. In another study, Colliander and Dahl en (2011) found that a blog post about a fashion brand resulted in higher brand attitude and increased purchase intent compared to an online magazine article on the same topic, because readers felt closer to the blogger.
Although no study has directly compared the impact of influencer endorsements to more traditional forms of endorsement advertising, interviews with female Instagram users (Djafarova and Rushworth 2017) suggest that influencers are perceived as more credible and relatable than traditional female celebrities, and their product reviews have a significant impact on young females' purchasing behavior. In an earlier study comparing the effectiveness of celebrity endorsements with product reviews of an unknown 'ordinary' customer, female participants were more positive about a promoted experience product when it was promoted by a relatable consumer (Wei and Lu 2013).
Although still in its infancy, scientific research on influencer endorsement seems to corroborate the acclaimed success of influencer endorsements as a marketing tool and suggests that influencers may nowadays have a more significant impact on brand attitudes and purchase behaviors than traditional celebrities. We therefore expect that influencer endorsements will result in higher advertising effectiveness than celebrity endorsements. In this study, we will measure advertising effectiveness in terms of attitude toward the ad, attitude toward the advertised product, and purchase intention (cf. Karson and Fisher 2005). In sum, we hypothesize the following: H1: Compared to celebrity endorsements, influencer endorsements will lead to higher attitude toward the ad, attitude toward the product, and purchase intention.

The role of identification and credibility
Existing research on endorsement marketing has identified two major processes that may underlie the effect of brand endorsement on advertising effectiveness: identification with the endorser (e.g., Basil 1996) and perceived endorser credibility (e.g., Ohanian 1991). Since both processes have been suggested to play a significant role in influencer endorsements as well (Chapple and Cownie 2017;Djafarova and Rushworth 2017), we will investigate to what extent identification and credibility can explain influencer versus celebrity endorsement effects.
When consumers believe that they share certain interests, values, or characteristics with an endorser, they are more likely to adopt their beliefs, attitudes, and behaviors (Cialdini 1993;Kelman 2006). Identification derives from both actual and perceived similarity, or the degree to which one perceives to have things in common with another person, as well as wishful identification, which is the desire to be like the other person (Hoffner and Buchanan 2005). In the case of celebrity endorsements, identification mostly arises from wishful identification, or an individual's aspiration to be like that celebrity (Kamins et al. 1989). In contrast, we expect identification with influencers to be more strongly determined by perceived similarity (Gr€ ave 2017). In comparison to celebrities, influencers are perceived as more relatable and approachable, like having a long-distance friend (Djafarova and Rushworth 2017). Influencers tend to directly address their followers in their posts, which connotes a certain closeness, and makes followers see them as peers (Erz and Christensen 2018;Gannon and Prothero 2018). The ability to comment on influencers' posts and the possibility for interaction may strengthen the feeling that the influencer is similar to oneself (Schmidt 2007).
In sum, unlike celebrities, influencers present themselves like 'ordinary', approachable, and authentic personalities (Chapple and Cownie 2017), which could make people feel more similar to them. However, when it comes to wishful identification, consumers may be more attracted to the glitter and glamourous fame of traditional celebrities. Since people are more likely to accept product claims communicated by endorsers they can identify with (Basil, 1996), we expect both types of identification to positively affect advertising effectiveness. In this study, we will measure advertising effectiveness in terms of attitude toward the ad, attitude toward the advertised product, and purchase intention (cf. Karson and Fisher 2005). In sum, we hypothesize the following: H2a: Influencer endorsements lead to a higher perceived similarity with the endorser than celebrity endorsements.
H2b: Celebrity endorsements lead to more wishful identification with the endorser than influencer endorsements.
H2c: Perceived similarity and wishful identification mediate the relationship between influencer vs. celebrity endorsements and attitude toward the ad, attitude toward the product, and purchase intention.
A second process that has been shown to play a role in the relationship between endorser and advertising effectiveness is perceived endorser credibility (Sternthal, Phillips, and Dholakia 1978;Ohanian 1991). In general, research on endorsement effects has shown that consumers are more likely to positively evaluate brands and products endorsed by people that they perceive to be credible (Erdogan 1999;Bergkvist and Zhou 2016). Credibility consists of two components: trustworthiness and expertise (Sternthal et al. 1978). Trustworthiness refers to perceptions of honesty, integrity, and believability of an endorser, whereas expertise refers to the relevant knowledge, skills, or experience the endorser is perceived to be possessing (Erdogan 1999).
For traditional celebrities, trustworthy endorsers seem to have more persuasive power than untrustworthy endorsers (e.g., Priester and Petty 2003), and perceived endorser expertise has been found to positively affect product attitudes and purchase intentions (e.g., Ohanian, 1991;Eisend and Langner 2010). Regarding influencers, studies in the more general domain of electronic word-of-mouth (eWOM) have shown that the effectiveness of eWOM on consumers' product attitudes and purchase intentions is determined by endorser credibility (Reichelt, Sievert and Jacob 2014;Erkan and Evans 2016). For influencers in particular, findings from Chapple and Cownie (2017) and Djafarova and Rushworth (2017) suggest that influencer credibility plays an important role in affecting purchase behavior.
In the present study, we argue that influencers may be perceived as more credible product endorsers than celebrities. First, influencers are known to share user-oriented product reviews, recommendations, and personal experiences on their social media channels, such as beauty influencers demonstrating make-up articles in their vlogs 1 . Although a growing amount of this content is company-sponsored and designed to persuade, the majority of influencer-generated content reflects honest opinions and does not have promotional goals (Evans et al. 2017). In contrast, consumers are generally well aware that celebrities get paid for their endorsements (Friestad and Wright, 1994). Second, influencers generally promote products in authentic, real-life settings, which may increase perceptions of trustworthiness as compared to celebrities. Uzuno glu and Kip (2014) found that bloggers seem trustworthy because by posting about certain brands they demonstrate that they have tried the products themselves. Russell and Rasolofoarison (2017) found that when celebrities do endorse a product in a more authentic way (being associated with the product in a real-life setting), they are perceived as more credible compared to more commercial forms of endorsement. Hence, when influencers endorse a product, consumers may be more likely to attribute this to the endorser genuinely believing in the positive characteristics of the product than when a celebrity endorser is used (cf. Zhu and Tan 2007). Third, an inherent part of influencers' success is that they have been able to establish a career by devoting themselves to a particular domain of interest and create their own expert profession (Balog, Rijke, and Weerkamp 2008;Erz and Christensen, 2018). Their selfacclaimed expertise could make it more likely for influencers to be perceived as knowledgeable on products and services in their domain of interest than traditional celebrities. Based on these propositions, we state the following hypotheses: H3a: Influencer endorsements lead to a higher perceived trustworthiness and expertise than celebrity endorsements.
H3b: Perceived trustworthiness and expertise mediate the relationship between influencer vs. celebrity endorsements and attitude toward the ad, attitude toward the product, and purchase intention.

Product-Endorser fit
In addition to identification and credibility as potential explanatory processes of influencer vs. celebrity endorsement effects, we investigate the moderating role of product-endorser fit. Using an endorser in an advertisement does not automatically guarantee a successful advertisement. Previous research has demonstrated that the effectiveness of an endorsed advertisement is inseparably tied to the degree in which the image, personality, or expertise of the endorser fits the advertised product (e.g., Kamins 1990;Kamins and Gupta 1994). Multiple studies have confirmed that when an endorser's perceived expertise matches the product endorsed, this enhances product evaluation and purchase intention (e.g., Till and Busler 2000;Fink et al. 2004). Moreover, endorsers that advertise products that do not fit their expertise are deemed less credible (Dwivedi and Johnson 2013;Lee and Koo 2015).
For influencers, endorser relevance to the product was mentioned by consumers as important to endorsement success as well (Djafarova and Rushworth 2017). Indeed, the effect of product-endorser fit on credibility may be even more pronounced for influencers than for celebrity endorsers. Since influencers have successfully branded themselves as representative of a particular domain of interest, such as 'beautyvlogger', and regularly share product information with their followers (Balog et al. 2008), an associative link between product and endorser may be more easily established. Therefore, as compared to celebrities, influencers may be more likely to be frowned upon when they endorse products that do not fit their specific 'niche' specializations and may only be perceived as credible information sources in endorsements that fit well with their domain of interest. A good fit between the endorser and the product could therefore be of higher importance for influencers than for celebrities, which results in the following hypothesis: H4: The effect of product-endorser fit on trustworthiness and expertise is more pronounced for influencer endorsers than for celebrity endorsers.

Study overview
We conducted two studies to test our hypotheses, using different endorsers and products in each study. All hypotheses are visualized in the conceptual model in Figure 1. In both studies, we manipulated type of endorsement by creating advertisements with either a celebrity or influencer endorser. Fit was manipulated by pairing the endorser with a product that fit or did not fit with the endorser's profession. The products in both studies were selected to be experience goods as their product characteristics (such as quality) are difficult to determine pre-purchase, so people tend to rely more on the evaluations of others (Park and Lee 2009).
The first study was based on a convenience sample of 131 Dutch adults, consisting mostly of students. The second study used a random sample of 446 US adults collected through Amazon MTurk. Both studies focused on women only as the majority of influencers tailor to a female audience (Gannon and Prothero 2018;IndaHash Labs 2017). Moreover, both studies focused on women between the ages of 18 and 40 as this constitutes the most common Instagram demographic, with 64.1% of women between the ages of 18 and 40 using Instagram compared to only 22.6% of women over 40 (Pew Research Center 2018). Descriptions of both samples can be found in Table 1.

Method
Design & manipulation Study 1 employed a 2 (type of endorsement: celebrity vs. influencer endorsement) X 2 (product-endorser fit: good vs. poor fit) between-subjects design. Moreover, as a  within-subject factor, participants were shown both an advertisement for a beauty product (tinted moisturizer) and an advertisement for a fitness product (protein shake). The two manipulated advertisements used in the main experiment consisted of existing advertisements (but unknown to our participants) of the beauty and fitness products, combined with the picture of the endorser. For both product categories, two influencers and two celebrities were selected based on a pretest (see below). The celebrity good-fit conditions consisted of either an advertisement of a tinted moisturizer (beauty product) combined with the picture of Kate Moss, a famous supermodel, or an advertisement of a protein shake (fitness product) combined with the picture of Fatima Moreira de Melo, a famous Dutch field hockey player. In the influencer good-fit conditions, the advertisement of the tinted moisturizer was shown with a picture of Serena Verbon, a beauty influencer, and the advertisement of the protein shake was combined with a picture of Kayla Itsines, a fitness influencer. The beauty blogger is well known in the Netherlands with 119.000 followers on Instagram and 30.000 unique daily website visitors. The fitness influencer has over 10 million Instagram followers world-wide.
In the poor-fit conditions, we swapped the products that the celebrities or influencers advertised, so that, for example, the beauty influencer was shown advertising the fitness product (see Figure 2 for an example and see the appendix for all stimuli) .

Pretest
A pretest (n ¼ 33) was conducted to examine which influencers and celebrities could serve as good and poor fits with beauty and fitness products. Participants were presented with names and pictures of six influencer endorsers and six celebrity endorsers (three for each product category), which were selected based on our own knowledge of famous celebrities and well-known influencers in the beauty and fitness industry and a Google search for famous influencers and celebrities in both industries. For each endorser, participants indicated the perceived fit of the presented celebrity or influencer with beauty products and the perceived fit with fitness products on 7-point scales (1 ¼ not at all, 7 ¼ very well). For the main experiment, for each product category we chose the celebrity and influencer with the highest fit score.

Procedure
All participants were presented with two stimuli, after each of which they filled in a questionnaire. First, participants were randomly presented with either the manipulated advertisement for the fitness product or the advertisement for the beauty product. Thereafter, they were shown the advertisement for the other product category. For each product, participants were randomly assigned to one of the four conditions. Both name and profession of the endorser were shown before presenting the respective advertisement. This way, participants could still evaluate the advertisements even when they did not recognize the endorser.

Measures
Wishful identification and perceived similarity were measured with two subscales from Hoffner and Buchanan (2005), on a scale from 1 (totally disagree) to 7 (totally agree).
Endorser credibility was assessed by using the trustworthiness and expertise subscales of the credibility scale by Ohanian (1990). Participants rated the endorser's trustworthiness on five 7-point semantic differential scales: undependabledependable, dishonest -honest, unreliablereliable, insinceresincere, and untrustworthytrustworthy. Expertise was also measured with five 7-point semantic differential scales: not an expertexpert, inexperiencedexperienced, unknowledgeableknowledgeable, unqualifiedqualified, and unskilledskilled. All items loaded on their respective factors, for both the fitness ad (trust: Attitudes toward the advertisement and product were assessed with a scale based on Spears and Singh (2004). Both were rated with five items on 7-point semantic differential scales: unappealingappealing, badgood, unpleasantpleasant, unfavorablefavorable, and unlikablelikable. Factor analyses yielded two factors corresponding with both subscales, for both the fitness ad (Attitude ad: EV ¼ 4.07, R 2 ¼ 40.69%, a ¼ .94; Attitude product: EV ¼ 4.32, R 2 ¼ 43.16%, a ¼ .96) and beauty ad (Attitude ad: Finally, purchase intention was measured with one item asking 'The next time you are looking to purchase a tinted moisturizer/protein shake, how likely are you to purchase this product?', rated on a 7-point scale from 1 (very unlikely) to 7 (very likely).
To ensure the validity of our manipulation of product-endorser fit, we asked participants to indicate perceived fit between the presented product and endorser on a 7point scale ranging from 1 (totally disagree) to 7 (totally agree). In addition, familiarity with the endorser was measured on a similar scale: 'I know [name endorser].'

Manipulation checks
Of our participants, 54.7% of respondents were at least somewhat familiar (4 or higher on the scale) with the Dutch hockey player, and 68.0% were familiar with the supermodel. In contrast, 37.5% of respondents were familiar with the fitness influencer, and 39.3% were familiar with the beauty influencer. For both the fitness product (M good fit ¼ 4.65, SD ¼ 1.54; M poor fit ¼ 3.18, SD ¼ 1.64) and the beauty product (M good fit ¼ 5.86, SD ¼ 1.24; M poor fit ¼ 4.29, SD ¼ 1.61), the data confirmed that the advertisements with good product-endorser fit were indeed considered a better fit than the advertisements with poor product-endorser fit, t fitnessproduct (129) ¼ -5.30, p < .001; t beautyproduct (129) ¼ -6.24, p < .001).

Overview of analysis
First, we tested the main and interaction effects of type of endorser and product-endorser fit on our dependent variables (H1) and our mediating variables (H2a, H2b, H3a, H4). Since product type is a within-subject factor as all participants were presented with both a beauty and fitness advertisement, we conducted a repeated-measures ANOVA on each of our mediating variables, with type of endorsement (celebrity endorsement vs. influencer endorsement) and product-endorser fit (good fit vs. poor fit) as between-subjects variables, and product type (beauty product vs. fitness product) as within-subjects variable. Means and standard deviations are presented in Table 2. Correlations between mediating and dependent variables are presented in Table 3.
In order to test the mediation hypotheses put forward in H2c and H3b, we conducted mediation analysis using the Process v2.15 macro in SPSS (Hayes 2013; model 4, 10,000 samples). Because of our repeated measures design, we had to test the effects of beauty and fitness advertisements with separate analyses. We therefore conducted six mediation analyses, for each of the three dependent variables (attitude toward the ad, attitude toward the product, and purchase intention), for both the fitness and beauty product advertisements separately. For clarity reasons, we only report significant Table 2. Study 1 means and standard deviations (in parentheses) for all mediating and dependent variables for the fitness product and the beauty product, as a function of type of endorsement and product-endorser fit. mediation effects. In our results below, we first discuss direct effects of endorser type on the mediating (H2a, H2b, H3a) and dependent variables (H1). We then discuss the mediation analyses (H2c, H3b), and effects of product-endorser fit (H4).
Direct effects of endorser type on advertising effectiveness H1 posed that influencer endorsements would results in a higher effectiveness than celebrity endorsements. This was confirmed for product intention only, F(1, 127) ¼ 11.42, p ¼ .001, g 2 ¼ .083. No effects were found for attitude toward the ad, and attitude toward the product, F's < 1.50, p's > .25.

Mediation analysis
For the advertised fitness product, only expertise was significantly related to attitude toward the ad, b ¼ .388, t(129) ¼ 3.24, p ¼ .002. However, neither expertise nor any other variable (wishful identification, perceived similarity or trustworthiness) significantly mediated the relationship between endorser type and attitude toward the ad. Expertise was also significantly related to attitude toward the product, b ¼ .339, t(129) ¼ 2.61, p ¼ .010, as well as similarity, b ¼ .350, t(129) ¼ 3.14, p ¼ .002. Similarity, though, was the only variable that mediated the relationship between type of For the advertised beauty product, only trustworthiness was significantly related to attitude toward the ad, b ¼ .534, t(129) ¼ 4.13, p < .001. Trustworthiness also significantly mediated the relationship between endorser type and attitude toward the ad, b ¼ .226, 95% CI [.050, .469]. Similar results were obtained for attitude toward the product. Trustworthiness was the only variable significantly related to attitude toward the product, b ¼ .730, t(129) ¼ 5.73, p < .001, and mediated the relationship between endorser type and attitude toward the product, b ¼ .309, 95% CI [.067, .609]. Finally, trustworthiness also was related to purchase intention, b ¼ .568, t(129) ¼ 2.86, p ¼ .005, and mediated the relationship between endorser type and intention, b ¼ .240, 95% CI [.039, .594]. In sum, we observe different effects for the fitness and beauty product. For the fitness product, H2c is confirmed for similarity, but we find no support for the mediating effect of trustworthiness and expertise (H3b). For the beauty product, H2c is not confirmed, but we do find that trust is a significant mediator, partly confirming H3b. For a visualization of our mediation analyses, see Figure 3.

Product-Endorser fit
Product-endorser fit was not related to any of the mediating variables, F's < 3.20, p's > .07. A good fit between endorser and product did not result in higher ratings of trustworthiness and expertise than a poor fit. However, additional analyses showed that product-endorser fit did directly affect attitude toward the ad, F(1, 127) ¼ 5.53, p ¼ .020, g 2 ¼ .042, and purchase intention, F(1, 127) ¼ 14.82, p < .001, g 2 ¼ .105, with well-fitting endorsements resulting in a more positive attitude toward the ad and a higher purchase intention than poor-fitting endorsements. Disconfirming H4, no interaction effects between type of endorser and product-endorser fit were observed for any of the mediating and dependent variables, F's < 3.20, p's > .070.

Method
In Study 2, we focused on two other product categories that influencers regularly endorse: food and fashion. The design and procedure for Study 2 were similar to that of Study 1, except that each participant received one stimulus only, resulting in a 2 (type of endorsement: celebrity endorsement vs. influencer endorsement) X 2 (product-endorser fit: good fit vs. poor fit) X 2 (product type: food vs. fashion product) between-subjects design.
A pretest (n ¼ 46) was conducted using Amazon MTurk to select the endorsers and products for our stimuli. First, participants were asked to list up to five female endorsers in each of these four categories: celebrities in the food industry, food influencers, celebrities in the fashion industry, and fashion influencers. The four endorsers that were mentioned most often in these four categories were selected for our main experiment, being Rachael Ray (food celebrity, mentioned 32 times), Gaby Dalkin (food influencer, mentioned 10 times), Kendall Jenner (fashion celebrity, mentioned 9 times), and Danielle Bernstein (fashion influencer, mentioned 10 times).
Next, participants were asked to indicate for three food products and three fashion products how well they felt that each product would fit a food/fashion endorser on a five-point scale from 1 (Terrible) to 5 (Excellent). For our main experiment, we selected the two products with the highest perceived product-endorser fit, which was a stand mixer for the food endorsers (M ¼ 4.59, SD ¼ 0.62), and a watch for the fashion endorsers (M ¼ 4.37, SD ¼ 0.90).
In the main experiment, in the good fit conditions the food celebrity or food influencer were paired with the stand mixer and the fashion celebrity or fashion influencer were paired with the watch. In the poor fit conditions, this was the other way around (see the appendix for all stimuli).

Manipulation checks
Of our participants, 91% of respondents were at least somewhat familiar (4 or higher on the scale) with the food celebrity, while 84.5% were familiar with the fashion celebrity. In contrast, only 29% of participants were familiar with the fitness influencer, and 38.1% were familiar with the fashion influencer. The advertisements with a good product-endorser fit (M ¼ 5.15, SD ¼ 1.51) were indeed considered a better fit than the poor-fitting advertisements (M ¼ 3.84, SD ¼ 1.86), t(444) ¼ -8.16, p < .001.

Overview of analysis
Similar to Study 1, we first discuss direct effects of endorser type on the mediating (H2a, H2b, H3a) and dependent variables (H1). We then discuss the mediation analyses (H2c, H3b), and effects of product-endorser fit (H4). To test the main and interaction effects of type of endorser and product-endorser fit on our mediating and dependent variables, we used a between-subjects factorial ANOVA. Next, we tested the mediation hypotheses using the Process v2.15 macro (Hayes 2013; model 4, 10,000 samples) for each of the three dependent variables separately (attitude toward the ad, attitude toward the product, and purchase intention). For clarity reasons, we only report significant mediation effects. Means and standard deviations are presented in Table 4. Correlations between mediating and dependent variables are presented in Table 5.
Direct effects of endorser type on advertising effectiveness H1 posed that influencer endorsements would results in a higher effectiveness than celebrity endorsements. This was confirmed for product intention only, F(1, 438) ¼ 7.56, p ¼ .006, g 2 ¼ .017. No effects were found for attitude toward the ad, and attitude toward the product, F's < 1.00, p's > .40.

Product-endorser fit
Product-endorser fit positively affected both trustworthiness, F(1, 438) ¼ 7.63, p ¼ .006, g 2 ¼ .017, and expertise, F(1, 438) ¼ 8.98, p ¼ .003, g 2 ¼ .020. A good fit between endorser and product lead to higher ratings of trust and expertise than poor fit. Product-endorser fit also directly affected attitude toward the ad, F(1, 438) ¼ 5.89, p ¼ .016, g 2 ¼ .013, with well-fitting endorsements resulting in a more positive attitude toward the ad than poor-fitting endorsements. No interaction effects between type of endorser and product-endorser fit were observed for any of the mediating and dependent variables, disconfirming H4, F's < 1.00, p's > .400.

Discussion
The goal of our study was to compare the effectiveness of endorsement advertising by social media influencers to traditional celebrity endorsements. We have investigated our hypotheses in two experiments, using different celebrity and influencer endorsers across four popular influencer marketing domains (beauty, fitness, food, and fashion) using different measures of advertising effectiveness. Overall, our results showed that participants identify more with influencers than with celebrities, feel more similar to influencers than celebrities, and trust influencers more than celebrities. In terms of advertising effectiveness, similarity, wishful identification, and trustworthiness mediate the relationship between type of endorser and advertising effectiveness, albeit not consistently. In contrast to expectations, product-endorser fit did not moderate the relationship between type of endorser and expertise or trust.
Hypothesis 1 was confirmed only for purchase intention. Participants were more willing to buy a product endorsed by an influencer than by a celebrity. No direct effects were observed for attitude toward the advertisement or product. Therefore, although participants are more likely to buy products from an influencer, they do not particularly like advertisements of influencers and the products advertised by influencers more than those of celebrities.
In line with H2a, influencer endorsements led to higher perceived similarity than celebrity endorsements. However, contrary to H2b, influencer endorsements also led to higher wishful identification with the endorser than celebrity endorsements. In hindsight, this may not come as a surprise, as many people now aspire to become a social influencer (Chae 2018). Because influencers are seen as more similar to 'ordinary people' than celebrities, perceived similarity may even make it easier for people to wishfully identify with them, because it is easier to believe they could be like them. Moreover, Uzuno glu and Kip (2014) previously found that a main reason for following Instagram bloggers was a combination of admiration on the one hand, and the feeling of connectedness to the influencer on the other.
H2c posed that perceived similarity and wishful identification would mediate the relationship between type of endorser and attitude toward the ad, product, and purchase intention. Overall, we found support for this hypothesis. In Study 1, perceived similarity mediated the relationship between type of endorser and attitude toward the product and purchase intention, albeit only for the fitness product. An explanation for why similarity did not mediate the relationship between endorser type and purchase intention for the beauty product may be because people feel they may be able to achieve a similar fitness or shape as the influencer and are therefore more inclined to try the protein shake, whereas obtaining similar beauty by simply using a tinted moisturizer is more difficult.
In Study 2, perceived similarity mediated the relationship between type of endorser and attitude toward the ad and purchase intention. People feel more similar to influencers than celebrities and as a result like their endorsements more and are more inclined to purchase the endorsed product. Similar to the fitness product in study 1, the products endorsed in study 2 (food and fashion) may actually allow the consumer to feel more similar to the endorser, in contract to beauty products, therefore resulting in higher purchase intention. In all, perceived similarity is an important explanation for the success of influencer endorsements, but its role depends on the type of products endorsed. Most likely, when the product endorsed will not lead the consumer to feel more similar, endorsing products will not be as effective as when the product endorsed may actually make the consumer feel more similar to the influencer (e.g., I may feel as fashionable as the influencer by buying a watch, but I'll never become as pretty by wearing make-up).
Wishful identification also mediated the relationship between endorser type and attitude toward the ad and purchase intention, albeit only in Study 2. A possible explanation may be that the products we used in Study 2 (a watch and a stand mixer) are more generally appealing to a larger group of people than the products in Study 1 (a tinted moisturizer and protein shake), and purchase intention is therefore more likely to be affected when people aspire to be like an endorser. This mirrors the explanation given above: It is likely that wishful identification explains the effectiveness of influencers endorsements only when consumers actually feel they may become like the endorser (i.e., genuine aspiration), which is most likely easier in the areas of food and fashion, than it is becoming fit or pretty.
Our third hypothesis (H3a) posed that influencer endorsements would lead to higher perceived trustworthiness and expertise than celebrity endorsements. In both studies, we found that influencer endorsers were indeed seen as more trustworthy, whereas no differences were found on expertise. An explanation for not finding differences in expertise could be that in both studies, we selected influencers and celebrities that are well-known experts in their specific fields. In addition, the expertise subscale used in our experiments assessed generic expertise (e.g., 'this person is knowledgeable') and did not measure specific expertise on the product endorsed. Future research is therefore advised to measure specific expertise, as to gain better insight in the role of expertise in influencer versus celebrity endorsements.
In H3b, we posed that trustworthiness and expertise would mediate the relationship between type of endorser and advertising effectiveness. Expertise was not a significant mediator in any of the tested relationships. However, in both studies, trustworthiness was an important variable explaining why influencers are more effective endorsers than celebrities. Trustworthiness mediated the relationship between type of endorser and attitude toward the advertisement (in Studies 1 and 2), attitude toward the product (in Studies 1 and 2) and purchase intention (in Study 1). Overall, our results are in line with earlier studies that stress the role of authenticity and trust in influencer endorsements (Uzuno glu and Kip 2014). A reason that trust, not expertise mediated the relationship between type of endorser and advertising effectiveness may be that trust is as much of a social-affective construct than a cognitive construct, in that trust depends on a feeling of integrity, believability, and mutual understanding (Uzuno glu and Kip 2014). Thus, consumers are most likely to trust an influencer more and are therefore more persuaded by influencer advertisements because an influencer is more like them and hence trustworthy.
H4 was not confirmed. We did not find any interaction effects of product-endorser fit and endorser type on trustworthiness or expertise. We also did not find any main effects of product fit in Study 1. An explanation may be that a fitness influencer endorsing a beauty product may not have been considered a very poor fit, since the fitness influencer was an attractive woman and beauty products are generally endorsed by pretty endorsers. In study 2, we do find main effects of product-endorser fit on expertise and trustworthiness, as the difference between food and fashion products is like to be more marked than the difference between beauty and fitness. Yet, we find no evidence that fit is more important for influencers than celebrities.

Conclusion
The present research is one of the first to directly compare celebrity and influencer endorsements in terms of their advertising effectiveness. We have shown that influencers are deemed more trustworthy than celebrities, and that people feel more similar to influencers and identify more with them than celebrities. These processes, in turn, affect advertising effectiveness. Therefore, a first implication of our study is that influencers may be more effective product endorsers than traditional celebrities and a practical recommendation is therefore to continue to use influencer endorsers in marketing campaigns.
A second implication is the importance of investigating the processes underlying effective product endorsements. Wishful identification, similarity, and trust are important explanations for why product endorsements work and also explain why influencer endorsements are more effective than celebrity endorsements. However, our results also show that the underlying processes explaining advertising effectiveness may depend on specific product-endorser combinations. Specifically, for an influencer endorsement to be more effective than a celebrity endorsement, an endorsed product must be able to enhance feelings of similarity and wishful identification. In all, our results show that when endorsing products, it is important for influencers to be perceived as similar to their audience and that identification with an influencer needs to be based on true aspiration rather than only wishful thinking.
Third, finding no direct effects of endorser type on advertisement and product attitude, while we do find several mediation effects via trust, similarity, and identification may indicate that there may be several other mediators that may explain the relationship between endorser type and advertising effectiveness. Possibly, the positive effect of influencer over celebrity endorsements on advertisement and product attitude via the mediators may be mitigated by other variables, such as likeability of the influencer (De Veirman et al. 2017; Gr€ ave 2017).
Although (or because) our study is one of the first to investigate the effects of influencer endorsements on advertising effectiveness, it is not without drawbacks. First, a potential improvement may be the way we presented the endorsements to our respondents. In all experimental conditions, our stimuli consisted of images of a product next to an endorser, resembling a basic advertising format. Although this makes for an experimentally valid comparison between endorser types, this is not how influencers on social media normally engage with a product. Usually, the product that is endorsed is part of a larger message and is integrated in a social media post, such as a vlog or an Instagram post (Kapitan and Silvera 2016). Theory on product placement teaches us that integration of a product in a storyline in a visually appealing way is positively related to endorsement effectiveness (Russell 1998). We therefore recommend investigating different types of social media endorsements with different levels of product engagement in future studies. On the other hand, that a relatively simple endorsement as ours yielded such effects shows the potential power of influencer endorsements over celebrity endorsements.
Furthermore, future research could examine other moderators than product-endorser fit influencing the relationship between endorser type and advertising effectiveness. For example, as this study only included experience goods, future research could compare the effects of influencer vs. celebrity endorsements on other types of products, such as search goods. Findings of Wei and Lu (2013) suggest that in contrast to experience products, search products like a pair of boots are more effectively endorsed by celebrities than by 'ordinary' consumers. Future research could also distinguish between informational and transformational or utilitarian and hedonic goods.
Finally, in our study we compared endorsements of traditional celebrities with endorsements by social media influencers. However, in reality, this distinction is not always so clear-cut. Numerous cases are known of successful social media influencers transgressing into more 'traditional' celebrities, pursuing a career as talk show presenter or fashion model and making their way to the general public and mass media. On the other hand, many traditional celebrities have become popular influencers on social media. This raises the question which type of influencers are the most successful endorsers, and to what extent popularity of the endorser is an important variable in explaining endorser effectiveness. In our studies we used well-known influencers with a large follower base, so-called 'micro-celebrities', but influencers who are relatively less popular may be even more effective endorsers. As compared to more popular influencers, 'micro-influencers' may be experts in a relatively small field and engage with their audience more, and may therefore be seen as more similar to their followers (De Veirman et al. 2017; Gr€ ave 2017). Nevertheless, even the 'micro-celebrities' we used in our studies appear to be more aspirable, relatable, and trustworthy than traditional celebrities, and are therefore more effective brand endorsers.
In conclusion, influencer endorsements are more effective than celebrity endorsements, which can be explained by processes of wishful identification, similarity, and identification. For an influencer endorsement to be more effective than a celebrity endorsement, an endorsed product must be able to enhance feelings of similarity and wishful identification. Moreover, an influencer must elicit trust in order to be effective. Influencers are not deemed more knowledgeable than celebrity endorsers, and expertise does not explain why influencer endorsements may be more effective than celebrity endorsements. Product-endorser fit has no effect on the relationship between type of endorsement and trust, expertise, or advertising effectiveness, although the effectiveness of influencer vs. celebrity endorsements hinges upon specific influencer-product combinations that elicit feelings of similarity, identification, and trust. Note 1. e.g., http://www.nikkietutorials.com

Disclosure statement
No potential conflict of interest was reported by the authors.

Notes on contributors
Loes Janssen is an assistant professor of Business Communication and Digital Media at Tilburg University. Her expertise lies in the domains of social psychology and communication science to understand consumer responses to different types of persuasive communication. She specifically focuses on the dynamics behind the (in)effectiveness of persuasive messages in the domains of marketing communication and health communication.
Alexander P. Schouten is an assistant professor of Business Communication and Digital Media at Tilburg University. His research interests include social media use, online collaboration, and online impression management. Within these areas, he is specifically interested in how different media capabilities affect the way in which people and organizations can effectively use new media technologies to communicate, to market, to work together, and to present themselves.
Maegan Verspaget is a master student Business Communication and Digital Media. She currently works as an online advertising consultant at Orangedotcom Online Marketing.