Amsterdam Business School Master Thesis
Friends or fame? Enhancing credibility of sponsored influencer
content: the moderating role of influencer-identification
Author: Anne van der Klugt – 10610936Program: MSc in Business Administration – Marketing Track Supervisor: Mr. Marco Mossinkoff
Date: 22/06/18
ABSTRACT
Nowadays, influencer marketing is booming. Influencer-generated content is found to be more credible than traditional advertisements, since persuasive intention is less obvious. However, new legislations require influencers to disclose sponsorship. An online experiment was done to examine if the level of sponsorship disclosure of influencer-generated content affects consumers’ perceived source credibility. Moreover, it proposes three concepts to boost credibility namely: social identification, wishful identification and brand-influencer personality fit. Although the direct effect of sponsorship disclosure on perceived source credibility was not found, the results indicate that both types of influencer-identification improve credibility. Additionally, highly identified people indicated posts in the low sponsorship disclosure condition as more credible than posts of which sponsorship was not disclosed. This could be due to the fact that those people perceive the disclosure of sponsorship as honesty. However, influencers should not overdo disclosure. In this case their persuasive intention becomes too obvious. Finally, brand-influencer personality fit does not significantly influence perceived source credibility.
STATEMENT OF ORIGINALITY
This document is written by Anne van der Klugt, who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
TABLE OF CONTENTS
1. INTRODUCTION ... 5 2. THEORETICAL FRAMEWORK ... 8 2.1 Influencer marketing ... 8 2.2 The effect of sponsorship disclosure on perceived source credibility ... 9 2.3 The moderating role of influencer-identification ... 12 2.3.1 Social identification ... 12 2.3.2 Wishful identification ... 13 2.3.3 Distinguishing social and wishful identification ... 15 2.4 The moderating role of brand-influencer personality fit ... 16 2.5 Conclusion: literature gap & conceptual model ... 17 3. METHODLOGY ... 18 3.1 Platform and brand selection ... 18 3.1.1 Platform selection ... 19 3.1.2 Brand selection ... 19 3.2 Sample ... 20 3.3 Research Design ... 21 3.4 Procedure ... 22 3.5 Stimulus material ... 23 3.6 Pretest ... 25 3.7 Measures ... 29 3.7.1 Back-translation procedure ... 29 3.7.2 Observation of participants’ behavior ... 30 3.7.3 Measurement of the variables ... 30 4. RESULTS ... 35 4.1 Preliminary analysis ... 35 4.2 Randomization checks ... 36 4.3 Manipulation checks ... 37 4.4 Descriptive statistics ... 39 4.5 Hypothesis testing ... 43 5. DISCUSSION AND CONCLUSION ... 55 REFERENCES ... 64 APPENDICES ... 72LIST OF TABLES AND FIGURES TABLES
Table 1 Experimental groups
Table 2 Perceived sponsorship disclosure pretest Table 3 Classification influencers in stimuli pretest Table 4 Classification additional influencers pretest
Table 5 Rotated factor loadings wishful and social identification Table 6 Frequency of participants across conditions
Table 7 Descriptive statistics
Table 8 Correlations between the variables and Cronbach’s Alphas
Table 9 Results effect of sponsorship disclosure on perceived source credibility Table 10 Descriptive statistics perceived source credibly across the conditions Table 11 Regression results moderating role of social identification
Table 12 Regression results moderating role of wishful identification Table 13 The moderating role of brand-influencer fit
Table 14 Summary of results FIGURES
Figure 1 Conceptual model Figure 2 Stimuli main experiment
Figure 3 Conceptual and statistical diagram
Figure 4 Interaction effect social identification and low sponsorship disclosure Figure 5 Interaction effect social identification and high sponsorship disclosure Figure 6 Interaction effect wishful identification and low sponsorship disclosure Figure 7 Interaction effect wishful identification and high sponsorship disclosure
1. INTRODUCTION
Product placement is everywhere! It is defined as “a paid message of branded products via audio, visual or both in media entertainment in an attempt to influence an audience” (Shoenberger & Kim, 2017). For a long time now, brands are promoted within media. Formerly, characters in movies and television programs where used to promote brands. However, since the number of social media users worldwide keeps increasing (Statista, 2018), social influencers are becoming an attractive marketing tool. Social Influencers are social media users that have a great amount of followers and high persuasion power on other users (Li, Lai & Chen, 2011). If brands use these people for advertisements it is called influencer marketing (Colliander & Erlandsson, 2015; Lu, Chang & Chang, 2014). Previous research found that influencer-generated content is more credible and interesting than traditional marketing sources (Johansen & Guldvik, 2017; Shareef et al., 2018; Zhu & Tan, 2007). Moreover, people will display less critical behavior since the persuasive intention of influencers is less obvious (Boerman et al., 2012).
Since the 1st of January 2014, influencers are forced to mention if their posts contain sponsored content due to a new law: Reclamecode Social Media (Stichting Reclame Code, 2014). Social influencers are catching up by this rule by including the hash tag ‘spon’ or ‘ad’ in their caption. However, how will customers perceive this change? If the degree of transparency on social media is increasing, will influencer marketing lose its power? Furthermore, if this is truly the case, what could companies do to minimize potentially negative consequences?
This research examines if the degree of sponsorship disclosure will influence consumers’ perceived source credibility on social media. As influencer marketing may lose its effectiveness, one may wonder what could be a solution for this negative
effect. Former research already found that the number of followers, the type of good and brand awareness could positively moderate the relationship between sponsorship disclosure and perceived credibility (Lu et al., 2014; Jin & Phua, 2014). The current study focuses on the moderating role of influencer identification namely: social-identification and wishful-social-identification. Furthermore, brand-influencer personality fit is analyzed as potential moderator of successful product placements on social media. Although the last one is not the main question of interest, data allowed the researcher to explore this additional hypothesis. The main research question is formulated as follows:
RQ: “What is the effect of the degree of sponsorship disclosure on perceived source credibility and how does influencer-identification moderate this relationship?”
To answer this research question, an online experiment was conducted in which 209 Dutch consumers participated. Results showed that there was no direct effect of the level of sponsorship disclosure on perceived source credibly. Although this effect was not found, both social and wishful identification positively affected the credibility of posts. Moreover, people that highly identified with the influencer even indicated to find posts in the low sponsorship condition more credible than posts in which sponsorship was not disclosed. Furthermore, brand-influencer personality fit did not significantly affect perceived source credibility.
Theoretical relevance
In terms of theoretical relevance, this research contributes to the limited literature about social influencer marketing (Lee & Watkins, 2016). First of all, up till now the
effects of sponsorship disclosure are mainly researched for traditional media. Boerman et al. (2012) already showed that when sponsorship is disclosed on TV, this results in a more negative brand attitude than when sponsorship was not disclosed. Furthermore, Wirth et al. (2009) found that product-placement causes a loss of credibility in films or television program. It would be interesting to see if this result holds in the context of social media since there are many differences with traditional media. In fact, social media is easier to update, people can give more commentary, sharing is more encouraged and there is much more freedom (Stokes, 2013). Recent research already focuses more on the effects of sponsorship disclosure on credibility for new media but these studies are mainly about bloggers (Carr & Hayes, 2014; Colliander & Erlandsson, 2015; Hwang & Jeong, 2016). Therefore, the current study could give interesting new insights in the effects of sponsorship disclosure on this relationship. Secondly, this study is the first that makes a distinction in the nature of influencer-identification and tries to compare the moderating role of wishful and social identification. Also testing these moderations in the context of social media instead of TV is new. Third, former literature is mainly focused on the effects of product-spokesman fit (Kamins & Gupta, 1994). However, little is known about the effects of the fit between the personality of the brand and the spokesman. This research will also examine this moderating role on the relationship between sponsorship disclosure and credibility. Therefore, the current research provides insights in new ways to improve credibility of sponsored content.
Managerial relevance
In terms of managerial relevance, the findings will enable brands to forecast what effect collaborating with an influencer will have on perceived source credibility. It
will provide a brand guidelines in the search for new collaborations with social influencers. Since influencers are now obligated to be transparent about sponsoring (Adformatie, 2014; Stichting Reclame Code, 2014), it is crucial to know how collaborations could still be effective. Besides this, brands and companies are increasingly using online platforms like Instagram for advertising (Awareness, 2008). Especially in the beauty and fashion industry Instagram is a popular marketing tool (Risepro, 2018). Furthermore, product placement is excessively used in this industry (Joshi, 2009; Risepro, 2018). Therefore, the outcomes will be very useful in practice.
This thesis will proceed as follows. First of all, this thesis will provide an explanation of the concepts and explain how the hypotheses are developed. Next, the methodology provides information about how this study is done. After that, the results are summarized. Lastly, the discussion and conclusion are presented.
2. THEORETICAL FRAMEWORK
In the first paragraph of this section, the concept influencer marketing is discussed. After that, the effect of sponsorship disclosure on perceived source credibility is explained. Additionally, the concepts social identification, wishful identification and brand-influencer personality fit are explained including their potential moderating role. Lastly, the conceptual model is provided.
2.1 Influencer marketing
Besides advertising via a brand’s own social media channels, social influencers are often part of the advertising strategy to camouflage advertisements (Johansen & Guldvik, 2017). Social media influencers are defined as “normal” social media users with great persuasion power and large networks of followers (Li, Lai & Chen, 2011).
Due to the enormous popularity and reach, social influencers are a very popular advertising medium (DeMers, 2017). The success of influencers provides marketers with a new tool to get in touch with the consumer (Lee & Watkins, 2016). The use of social influencers for advertisements is called influencer marketing. In this case, influencers collaborate with companies who are paying them to create posts about a product or service of the brand (Colliander & Erlandsson, 2015; Lu, Chang & Chang, 2014). Social influencers will generate positive electronic word-of-mouth for a brand, which is defined as information about a product or service provided by a potential consumer via the Internet (Daugherty & Hoffman, 2013).
As a result, the positive electronic word-of-mouth of the influencers will have favorable effects on brand attitude and purchase intentions (Bergkvist, Hjalmarson & Magi, 2016) Additionally, consumers do not immediately recognize a promotion of products as advertising because the persuasive intention of influencers is less obvious (Johansen & Guldvik, 2017; Zhu & Tan, 2007). Consequently, consumer attitude and behavior will be more favorable when confronted with influencer-generated content compared to brand-generated content (Becker-Olsen, 2003; Boerman et al., 2012; Johansen & Guldvik, 2017; Shareef et al., 2018; Zhu & Tan, 2007).
2.2 The effect of sponsorship disclosure on perceived source credibility
As stated in the former paragraph, the persuasive intention of influencer-generated messages is less obvious and therefore perceived as more credible than brand-generated posts (Becker-Olsen, 2003; Boerman et al., 2012; Johansen & Guldvik, 2017; Shareef et al., 2018; Zhu & Tan, 2007). Consequently, when credibility is high, people easier believe the content compared to when credibility is low (Aguirre et al., 2015). Furthermore, high source credibility has a positive effect on brand attitude
Nonetheless, at the beginning of 2014, the Reclamecode Social Media was adopted (Stichting Reclame Code, 2014). This law is designed to provide guidelines for social media marketing. The new rules force people, and thus influencers, to mention if posts contain sponsored content. The law distinguishes three actors: the marketer, the company, and the one that shares the message, which is the one that post the message online. In case the company pays the one that shared the message (i.e. the influencer), this should be mentioned in the post (Adformatie, 2014; Stichting Reclame Code, 2014). Due to this new law, influencer marketing becomes more transparent, which could have potential negative effects on the perceived credibility of the influencer and thus of the advertorial message.
Colliander and Erlandsson (2015) empirically study the relationship between sponsorship disclosure and credibility. Participants are randomly assigned to an experimental group. In the first group, a sponsored blog post that contained a positive review is presented. The participants receive a note that the blog post is sponsored and that the blogger receives money in return for the review. The other group views the same blog post, but sponsorship is hidden. Comparison of the attitudes of both experimental groups reveals that sponsorship disclosure negatively affects credibility. However brand attitude and customers’ purchase intentions are not affected by sponsorship disclosure (Colliander & Erlandsson, 2015).
Additionally, Carr and Hayes (2014) examine the same relationship but, in contrast to the above study, they have four experimental groups in which the third-party influence disclaimer varies. The four experimental groups include: explicit disclosure, implied disclosure, impartial disclosure and no disclosure. Explicit disclosure means the blogger fully disclosures the influence of a third party. In the implied sponsorship disclosure condition, third party influence is minimally disclosed
and nothing is said about the degree or nature. Impartial disclosure means that the blogger is mentioning that the review is not influenced by an outside party. The no disclosure condition refers to making no reference of third party involvement at all. Results show that there is no significant difference in credibility between a blogger that makes no mention about sponsorship and one that explicitly discloses third party influence. In addition to this, implied sponsorship makes a blogger less credible than bloggers who explicitly disclose sponsorship. Furthermore, bloggers are least credible when implicitly mentioning third-party influence. The researchers explain this effect by cognitive dissonance: since nothing is explicitly said about sponsorship, the reader starts doubting neutrality of the blog post whereas full sponsorship disclosure increases credibility by reducing uncertainty about the quality of the blogger. Furthermore, disclosing sponsorship is perceived as honesty, which in turn increases credibility (Carr & Hayes, 2014).
Unlike the former study, Hwang and Jeong (2016) found that when bloggers mention all opinions are their own, it increases credibility compared to explicit sponsorship. The results showed that attitudes towards sponsored blog posts are more negative compared to attitudes towards posts in which there was no sponsorship disclosure. However, when the influencer mentions all opinions are honest opinions that are not induced by the company, this negative effect disappears (Hwang & Jeong, 2016).
In brief, former literature reveals mixed results and therefore potential moderators of this relationship should be studied. The first study found that sponsored posts have lower credibility compared to post were sponsorship is not mentioned. However, the second study reveals that there is no significant difference in credibility between the no sponsorship condition and the explicit sponsorship condition because
honesty is positively correlated with credibility. They state that mentioning a post is not sponsored even as implicit sponsorship gain less credibility than posts that were not or were explicitly sponsored. However, the last study found that mentioning neutrality should increase credibility compared to explicit sponsorship. The first study was done twice: online and in a more controlled classroom setting. These results were robust and thus seem plausible. The other two studies are both online experiments. Taken together, the above discussion leads to the following hypothesis:
H1: There is a direct negative relationship between sponsorship disclosure and source credibility
2.3 The moderating role of influencer-identification
This study proposes influencer-identification as a potential moderator to the effect of sponsorship disclosure on perceived source credibility. In this paragraph, the two types of identification that will be tested are discussed i.e. social and wishful identification. Moreover, the antecedents of both types are summarized. In the last section, a comparison between the two types of influencer-identification is made.
2.3.1 Social identification
Social identification refers to “the process of people adopting the identity of a group that they are categorized to” (McLeoad, 2018). According to Hall (1954), we “always tend to identify with people who have the same characteristics as we have” Consequently, social identification occurs when a person likes or perceives similarity with another person or group (Hoffner & Cantor, 1991).
Lewicki, Tomlinson and Gillespie (2006) reviewed earlier research to summarize influential trust-generating factors. According to them, social
identification is the most important factor of all. Consequently, people perceive similar sources as more credible than dissimilar sources (Simons, Berkowitz & Moyer, 1970). Simons, Berkowitz and Moyer (1970) found that differences in similarity occur due to different feelings in the extent of group membership. In-group identification facilitates credibility of other members. Credibility increased by trust, respect and attraction. Furthermore, the researchers state that an opinion leader should be a “super-representative” of a group, which means that an opinion leader should be similar but also more competent, interested and informed than average. In this way, a kind of expert-status is gained, which in turn increases credibility. Therefore, sponsorship disclosure should be a weaker driver for perceived source credibility when people highly socially identify with the influencer.
H2: The negative relationship between the degree of sponsorship disclosure and perceived source credibility is moderated by the level of social identification with the influencer, so that this relationship is weaker for higher values of social identification
2.3.2 Wishful identification
In 1957, Maccoby and Wilson (1957) introduced the concept of wishful identification. This refers to the “desire to be like somebody”. In fact, “the viewer puts him- or herself in fantasy, in the place of one of the characters and views the action of the drama through this character’s eyes, sharing vicariously the experiences the chosen character undergoes” (Maccoby and Wilson, 1957). The character with which people identify differs but is predictable because wishful identification is based on need relevance and similarity. Need relevance refers to the fact that actions of the character are relevant for the viewer. Similarity simply means similarity between the character
Hoffner (1996) examined which particular features of TV characters increase wishful identification. The researcher asked children to pick their favorite character. Results showed that 91.1% of the boys, but only half of the girls, chose characters of the same sex. Besides this, both boys and girls perceived the behavior of female characters as more positive than that of male characters. Furthermore, the kids had higher wishful identification with characters that were perceived as more intelligent. Additionally, girls indicated female characters as more attractive an intelligent than male characters. Humor and attractiveness were important predictors of wishful identification, especially for girls.
A few years later, Hoffner and Buchanan (2005) examined wishful identification of adults instead of children. Just like in the former study, participants were asked to pick their favorite character. Overall most respondents chose favorite characters of their own race and gender. Besides this, almost 90% of their favorite characters were older. Additionally, similarity in attitude positively affects wishful identification. Furthermore, identification with characters of the other gender was greater for characters perceived as more successful and admired by other characters. Moreover, men identified more strongly with male characters that were perceived as smarter and more successful. Women identified more with female characters that were smarter, more successful, admired by other characters and more physically attractive.
More recently, Shoenberger and Kim (2017) researched the effect of specific character attributes of TV-stars on wishful identification. Again, participants were asked to choose a favorite character. The results showed that the attributes admired, successful and attractive increased wishful identification, which resulted in more product buying. Interestingly, the researchers found that brand trust is higher for
brands used by favorite characters (Shoenberger & Kim, 2017). As mentioned before, high source credibility could increase brand trust (Colliander & Dahlén, 2011). The fact that higher brand trust is found when wishful identification is high could be the result of higher source credibility. Consequently, high levels of wishful identification should reduce the role of sponsorship disclosure in driving perceived source credibility.
H3. The negative relationship between the degree of sponsorship disclosure and perceived source credibility is moderated by the level of wishful identification with the influencer, so that this relationship is weaker for higher values of wishful identification
2.3.3 Distinguishing social and wishful identification
None of the former researches explicitly makes a distinction between social and wishful identification. According to the former definitions, there is some overlap between the two concepts. Similarity is important for both types of identification. In addition to that, liking the other person increases social as well as wishful identification with a person or group. However, according to the definition of social identification, this is “the process of people adopting the identity of a group that they are categorized to”. Consequently, people socially identify with others in their associative reference group, defined as the group people currently belong to (McLeod, 2008; Weihrauch, 2017) In contrast, wishful identification refers to “the desire to be like somebody”. These people belong to the aspirational reference group of a person defined as people we admire and desire (Shoenberger and Kim; Weihrauch, 2017).
group. Consumers perceive some irritation when they receive promotional content about a product from a formal reference who is not a peer in their network. Therefore, social identification is expected to be a “stronger” moderator than wishful identification. In this case, stronger is defined as the higher predictive power (Cassidy, 2014).
H4. Social identification is the “strongest” moderator
2.4 The moderating role of brand-influencer personality fit
Brand fit or congruence is defined as the “perceived link between a cause, and the firm’s product line, brand image, position or target market” (Becker-Olsen & Simmons, 2002). Research on sponsorship has shown that higher perceived brand-influencer fit will generate more positive consumer responses towards sponsored content compared to low perceived fit (Kamins & Gupta, 1994). The likability of the influencer is positively influencing the purchase intentions of consumers and consequently their loyalty towards the brand (Kamins & Gupta, 1994). Furthermore, high fit increases recall of the brand (Misra & Beatty, 1990).
Moreover, Kamins and Gupta (1994) argue that the believability of advertising via a celebrity is influenced by the fit between the celebrity spokesman and the product. Although celebrity spokesperson are more likable and attractive than non-celebrities, they are often less believable (Atkin & Block, 1983). Low congruence between the image of the spokesman and the product could be an explanation for this (Kamins & Gupta, 1994). Research showed that advertising a product via a spokesperson whose image fits well with the product is significantly more believable than advertising via a spokesman whose image does not fit the product. This means that good fit between the image of the spokesman and the promoted product lead to
higher believability of the spokesperson as well as the brand (Kamins & Gupta, 1994).
According to Pruppers (2018), brand personality is part of a brand image and defined as a concept about how human characteristics are associated with a brand (Aaker, 1997). Pruppers (2018) states that the personality of an influencer can facilitate consumers’ perception of brand-influencer fit. According to Aaker (1997) there are five main dimensions of brand personality namely: sincerity, excitement, competence, sophistication and ruggedness. This paper argues that besides the importance of product-influencer fit, brand-influencer personality fit should be taken into account. Therefore, higher brand-influencer fit should positively influence the relationship between sponsorship disclosure and credibility.
H5: The negative relationship between the degree of sponsorship disclosure and perceived credibility is moderated by the level of brand-influencer fit, so that this relationship is weaker for higher values of wishful identification
2.5 Conclusion: literature gap & conceptual model
This thesis contributes to the limited literature about social influencer marketing (Lee & Watkins, 2016). First of all, the effects of sponsorship disclosure are mainly researched for traditional media (Boerman et al., 2012; Wirth et al., 2009). It would be interesting to see if the results hold in the context of social media since there are many differences with traditional media (Stokes, 2013). Additionally, this study is the first that makes a distinction in the nature of influencer identification and tries to compare the moderating roles of wishful and social identification on the relationship between sponsorship disclosure and credibility. Moreover, the concepts have not yet
focused on product-spokesman fit (Kamins & Gupta, 1994) but still little is known about the effect of the fit between the personality of the brand and the spokesman on the relationship between sponsorship disclosure and credibility. The research question is formulated as follows:
RQ: “What is the effect of the degree of sponsorship disclosure on perceived source credibility and how does influencer-identification moderate this relationship?”
The following model illustrates the assumptions of this research
Figure 1 Conceptual model
3. METHODLOGY
This chapter describes the research method of the study, including a description of the platform and brand, the sample, the research design, the procedure, the stimuli, the pretest and a description of the measures.
3.1 Platform and brand selection
In this paragraph the choice of the platform and brand used in this study is justified.
Level of sponsorship disclosure No/low/high Social identification Perceived source credibility Wishful identification H1 (-) H2 (+), H4 (+) H3 (+) Brand-influencer personality fit H5 (+)
3.1.1 Platform selection
The platform that was chosen for this study is Instagram. Instagram is described as “a social network where pictures and short movies can be shared” (Belch & Belch, 2015). Currently, Instagram is the third most popular social networking site over the world (Kallas, 2018) and one of the most fast growing channels (Statista, 2018). This increasing popularity and enormous reach is the reason marketers are falling in love with the platform. Nowadays, brands and companies are increasingly using Instagram for advertising (Awareness, 2008; DeMers, 2017). Especially in the beauty and fashion industry Instagram is a popular marketing tool (Risepro, 2018). Furthermore, product-placement is extensively used in the fashion industry (Joshi, 2009) Recently, Instagram introduced a new tool: the ”paid partnership feature” (Chacon, 2017). When the feature is used consumers will immediately see the text “Paid partnership with [business partner]” above the posts in the location section. This will help influencers to easily disclose sponsorship as well as businesses to see how sponsored content campaigns perform (Chacon, 2017).
3.1.2 Brand selection
Nowadays, especially the fashion industry has recognized the power of influencer marketing (Woods, 2016). Therefore, the brand used for this study is a fashion and accessory brand named Cluse. To gather additional insights, an interview was held with a spokesman of Cluse: Eirini Papaioannou. The brand mainly sells watches that have a quite neutral and minimalistic style. Therefore, the products of Cluse are very suitable for this study, since it minimizes product attitudes. The brand targets people between the age of 18 and 25. Furthermore, Cluse is a brand that intensively uses social influencers as a marketing tool (Cluse, n.d.). Reason for this is that the target
influencers because customers can easily relate to them. The brand is using a mix of micro and macro influencers, which means a mix of influencers with a number of followers between 3 thousand and 5 million. Currently, Cluse is working on its brand image and trying to create a specific Cluse-character: a strong independent graceful young woman. Consequently, the brand takes personality of the influencers into account when searching for new collaborations by looking at their photos and captions (E. Papaioannou, personal communication, 2018).
3.2 Sample
The population for this study is Dutch consumers, which is a quite large and diverse group. The rationale for selecting all Dutch consumers, instead of only women, is that according to the former literature identification does not necessarily have to be gender matched, especially for women. A non-probability convenience sample is used to generate a reliable sample. The online experiment was conducted in the period of May 4 till May 31. Participants are recruited through and online open survey link, which was distributed through Instagram, Facebook and Whatsapp. Consequently, the sample mainly consists of students and acquaintances.
A total of 260 participants started the survey and 209 of them completed it. Before downloading the data to SPSS, cases were excluded listwise, meaning that only cases without any missing data in any variable are analyzed. Because all questions were ticked with the box ‘force response’, participants were forced to answer every question that was presented to them. In addition to this, participants could not go back to the question before. Consequently, every finished survey is useful. In addition to this, it was checked if there were participants that rushed through the survey. There was nobody who finished the survey within 60 seconds, so
this seems not to be the case. For this research design, a minimum of 30 participants per condition is required (A. Weihrauch, personal communication, 2018), which means a total of 180 participants is needed. The total of 209 participants met this requirement. The average age of the participants is 33 years old. Furthermore, 78.9% of the participants are female.
3.3 Research Design
This research made use of a mixed-method approach.
An interview with a spokesman of Cluse was conducted to identify the most relevant brand personality characteristics. The interview resulted in the following 12 personality characteristics presented in the final survey: grace, drive, curiosity, togetherness, honest, wholesome, cheerful, daring, up-to-date, intelligent, charming and outdoorsy (E. Papaioannou, personal communication, 2018). Besides this, general questions regarding the target group, style and marketing techniques of Cluse were asked to get a better overview of the brand. The interview questions and corresponding answers can be found in appendix A.
To further test the hypotheses, an online experiment was done. A survey method was used to collect cross-sectional data. A 3 (degree of sponsorship disclosure: high versus low versus no/control) x 2 (type of influencer-identification: social identification versus wishful identification) between-subjects design was used. Respondents have been randomly assigned to one of the six conditions presented in table 1.
Table 1 Experimental groups
Manipulation level of sponsorship disclosure
High Low No Manipulation type of influencer-identification Social Condition 1: High sponsorship, socially identifiable influencer Condition 2: Low sponsorship, socially identifiable influencer Condition 3: No sponsorship, socially identifiable influencer Wishful Condition 4: High sponsorship, wishfully identifiable influencer Condition 5: Low sponsorship, wishfully identifiable influencer Condition 6: No sponsorship, wishfully identifiable influencer 3.4 Procedure
First of all, participants are provided with a short introduction. Before being exposed to the stimulus material, participants were asked to answer a couple of questions. The survey starts with questions regarding their social media and Instagram usage. After that, participants should indicate if they know the brand Cluse and if this is the case, they were provided with questions regarding their brand attitude. Furthermore, a set of characteristics is provided to them. They should indicate how well each characteristic fits both the influencer on the presented picture and the brand. Afterwards, they are asked to indicate if they knew the shown influencers and to sort both separately into one of the four reference groups of Englis & Solomon (1995). This is done to check if the manipulation of the type of influencer-identification was successful. Thereafter, a short explanation about the post that will be displayed is given. After being exposed to the photo of one of the six conditions, respondents were presented with questions regarding the credibility. Besides this, questions were asked to measure both the participant’s level of social and wishful identification with the
influencer. In the end of the survey, participants were asked if they had seen any sponsorship disclosure and indicated their perceived third-party influence. Subsequently, participants were asked for their gender and age. The complete survey can be found in appendix B.
3.5 Stimulus material
Manipulation varied in sponsorship disclosure and type of influencer-identification. Sponsorship disclosure was manipulated to see if the extent of disclosure influences perceived source credibility. The type of influencer-identification is manipulated to create enough variation in the levels of identification to later analyze their moderating role. The stimuli of the main experiment are presented in figure 2.
The level of sponsorship disclosure was manipulated by adding tags, mentions, hash tags and the ‘Payed Partnership’ tool. Consequently, in the high sponsorship setting, the post included a tag with ‘Cluse’, a mention @cluse, the hash tags #fallforcluse and #sponsored and a text saying ‘Paid partnership with Cluse’. The low sponsored post only included the mention and the hash tags. Lastly, the control setting does not reveal any sponsored links.
The type of influencer-identification is not easy to manipulate, since this is highly personally dependent. Of course this research controlled for factors that could have a potential influence (age, gender etc.). Additionally, to manipulate the type of influencer-identification as good as possible, two different types of influencers are selected. This study selected one influencer that is expected to belong to the participants’ aspirational reference group and therefore evoke high wishful identification. Secondly, a second influencer is selected who is expected to belong to a participants’ associative reference group and therefore evoke high social
types of identification found and discussed in the theoretical framework. For a socially identifiable influencer, especially similarity is very important. A wishfully identifiable influencer is often successful, admired, smart and physically attractive. In order to find the right types, six influencers were selected for a pretest according to these requirements. Participants were asked to sort them into one of the four categories of Englis & Solomon (1995) namely: aspirational, associative, avoidance or irrelevant reference group. The results of the pretest show that Negin Mirsalehi is perceived most aspirational. She is one of the most popular Dutch fashion influencers (Le Guide Noir, 2015), has her own millions company (Vogue, 2017), has 4.7 million followers (Instagram, 2018) and earns approximately €15.000 for one post (Imee, 2017). She is also known for her collaborations with luxury brands (Time, 2017). Consequently, she is expected to evoke high wishful identification. Additionally Isa Obiols was indicated as most associative influencer. This girl is more a girl-next-door type, has only 2.7k followers (Instagram, 2018) and is not very well known in the Netherlands. Nobody in the sample knew who she was (N=0). As a result, she was selected for the social influencer-identification condition.
The same text and number of likes are used in both conditions in order to minimize external effects. The post has 157.899 likes and is accompanied by the text: ‘A little bit of every thing I love’. The location of the photo is also similar: both influencers are sitting in a restaurant.
Figure 2 Stimuli main experiment
3.6 Pretest
In order to ensure successful manipulation of the level of sponsorship disclosure and type of influencer-identification, a pretest was conducted by means of a survey via Qualtrics. This was done between April 26 and May 1. In total, 24 people participated of which 87.5% was female. The mean age of the participants was 26 years old. All
participants finished the questionnaire. They were approached via Whastapp or in real life. The pretest survey can be found in appendix C.
Before the participants were randomly assigned to one of the six conditions, questions regarding their social media and Instagram usage were asked. Furthermore, the survey contained questions regarding participants’ influencer familiarity and brand attitude. Afterwards, participants were randomly assigned to one of the six experimental conditions presented in appendix D. When the participants click further, the survey continues with questions about perceived credibility, social identification and wishful identification. At the end of the survey, manipulations were checked. Furthermore, participants were asked to sort six different influencers, presented in appendix E into their perceived reference group (Englis & Solomon, 1995). Lastly, participants were asked for their demographic characteristics namely age and gender.
To check if manipulation was successful, two questions were asked at the end of the survey. Manipulation of sponsorship disclosure was checked by the question: “Was the Instagram picture by the influencer sponsored?” Participants could choose between the options yes, no or don’t know (Colliander & Erlandsson, 2015). Table 2 shows that in the high sponsorship setting, 83.3% of the participants correctly noticed that the post was sponsored. In the low sponsorship setting, this was only 33.3% and 44.4% falsely remembered that the post was not sponsored. In the no sponsorship control condition, 44.4% of the people indicated that they thought the post was not sponsored while 33.3% incorrectly indicated that the post was sponsored. Chi-squared tests were done to check if the answers significantly differ between the three conditions, this was not the case (𝒳2 (4) = 5.188, p = .269). More specifically, the no and low sponsorship disclosure condition (𝒳2 (2) = 0 p = 1), the low and high sponsorship disclosure condition (𝒳2 (2) = 4.410, p = .110) and the no and high
sponsorship disclosure condition (𝒳2 (2) = 4.410, p = .110) do not significantly differ in answers.
Table 2 Perceived sponsorship disclosure pretest
Condition N Yes No I don’t know
High
(Paid partnership, tag, hash tags and mention) 6 5 (83.3%) 0 (0%) 1 (16.7%) Low
(hash tags and mention) 9 3 (33.3%) 4 (44.4%) 2 (22.2%)
No 9 3 (33.3%) 4 (44.4%) 2 (22.2%)
Furthermore, to check manipulation of the type of influencer-identification, participants were asked to indicate in which reference group the influencer fits best: aspirational, associative, avoidance or irrelevant influencer (Englis & Solomon, 1995). Table 3 indicates that 23.1% of the people that viewed a picture of Rachelle Ruwiel perceived her as somebody in their associative reference group whereas 15.4% indicated her as somebody in their aspirational reference group. Besides that, 27.2% of the participants who were presented with a picture of Negin Mirsalehi, indicated that she belongs to the aspirational reference group. However, the answers do not significantly differ across the two conditions (𝒳2 (3) = 5.356, p = 0.147). To check if there was a significant difference in the level of social and wishful identification between the two conditions, an independent sample t-test was done. This showed that there was a significant difference (t (22) = 1.322, p < .05) in the level of social identification between the social influencer-identification condition (M=2.785, SD=1.170) and the wishful influencer-identification condition (M=2.255, SD=.682). However, there was no significant difference (t (22) = -.282, p = .778) in the level of wishful identification between the social influencer-identification
condition (M=3.462, SD=1.6133) and the wishful influencer-identification condition (M=3.667, SD=1.903).
Table 3 Classification influencers in stimuli pretest
Condition N Aspirational Associative Avoidance Irrelevant
Social influencer-identification (Rachelle Ruwiel)
13 2 (15.4%) 3 (23.1%) 2 (15.4%) 6 (46.2%)
Wishful influencer-identification
(Negin Mirsalehi) 11 3 (27.3%) 0 (0%) 0 (0%) 8 (72.7%
Based on these findings, manipulation of both sponsorship disclosure and type of influencer-identification were not (completely) successful. To make third-party influence in the sponsored conditions more clear, #sponsored was added to the low and high sponsorship disclosure conditions. Besides this, an extra measure of perceived sponsorship was added to check if there is a difference in perceived sponsorship between the low and high sponsorship disclosure condition. To improve the type of influencer-identification manipulation, the results of the categorization of the six influencers were analyzed. The results in table 4 show that Negin Mirsalehi is clearly an aspirational influencer and therefore is expected to evoke wishful identification. 20.8% of the total number of participants indicated that she belongs to their aspirational reference group. None of the participants indicated here as associative. Isa Obiols is clearly perceived as an associative influencer and therefore is expected to evoke social identification. 16.7% indicated she belongs to the associative reference group whereas only 4.2% classified her as aspirational.
Table 4 Classification additional influencers pretest
Influencer Aspirational Associative Avoidance Irrelevant 1. Rachelle Ruwiel 3 (12.5%) 5 (20.8%) 2 (8.3%) 14 (58.3%) 2. Negin Mirsalehi 5 (20.8%) 0 (0%) 3 (12.5%) 16 (66.7%) 3. Kiara King 7 (29.2%) 7 (29.2%) 4 (16.7%) 6 (25%) 4. Isa Obiols 1 (4.2%) 4 (16.7%) 3 (12.5%) 16 (66.7%) 5. Zoë Pastelle 7 (29.2%) 8 (33.3%) 0 (0%) 9 (37.5%) 6. Sam Schouten 5 (20.8%) 4 (16.7%) 2 (8.3%) 13 (54.2%)
Besides manipulation checks, reliability checks were done for social identification (𝛼 = .821) , wishful identification (𝛼 = 0.944) , source credibility 𝛼 = .465 and brand attitude 𝛼 = .900 to see the extent to which data collection techniques yield consistent findings. The measures all had a Cronbach’s Alpha that exceeds the threshold of 0.7, except for credibility, measured by the three-dimensional variable of Mackenzie and Lutz (1989). Therefore, it was decided to include a more frequently used credibility measure of Ohanian (1990) to increase reliability in the final online experiment.
3.7 Measures
3.7.1 Back-translation procedure
All construct items in the pretest and in the online experiment were adopted from articles published in English. Since the target population is Dutch, the items were translated. To make sure this will not affect the results, a back-translation procedure was used whereby three participants translated the items back to English. The items were successfully translated back and all discrepancies were corrected.
3.7.2 Observation of participants’ behaviour
Two participants were observed during the process of filling in the survey. They acted concentrated and according to them, questions were completely clear.
3.7.3 Measurement of the variables
In this section, the measures used in the survey are discussed. A complete overview of all measures, adaptions and Dutch translations can be found in appendix F.
Sponsorship disclosure Sponsorship disclosure is the independent variable manipulated by the researcher. Participants were randomly assigned to the no (=0), low (=1) or high (=2) sponsorship disclosure condition.
Perceived source credibility Perceived source credibility is the dependent variable and measured by the validated scale of Ohanian (1990). The construct counts 15 items measuring three main components of credibility namely: physical attraction, trustworthiness and competence. Participants indicated their agreement on a 7-point Likert scale (1= Completely disagree to 7= Completely agree). The explanatory factor analysis of credibility showed that the Kaiser-Meyer-Olkin statistic is greater than 0.6 (KMO = .849), Bartlett’s Test of Sphericity is significant (χ2 (105) = 1713.078, p < .001) and that one factor is reliable (Cronbach’s Alpha attraction 𝛼 = .846 , trustworthiness (𝛼 = .882) and competence (𝛼 = .772)).
Social and wishful identification Social and wishful identification are both moderating variables. Social identification is measured by the inclusion of other in the self-scale of Aron et al. (2008) in which participants indicate their relationship with the influencer. Respondents select the picture that best describes their relationship from a set of seven Venn-like diagrams each representing different degrees of overlap of two circles. In addition to that, four questions of Jin and Phua (2014) were used.
influencer (3) I feel close to the influencer and (4) I like the influencer. Participants indicated their agreement on a 7-point Likert scale (1= Completely disagree to 7= Completely agree). Wishful Identification was measured according to Hoffner (1996). An adapted three-item variable of the researcher was used. The items presented are: (1) I’d like to do the kinds of things she does on Instagram (2) She is the sort of person I want to be like myself and (3) I wish I could be more like her. All the items were measured on a 7-point Likert scale (1=Completely disagree to 7 = Completely agree). A factor analysis is done to evaluate the goodness of the scales. Because the concepts are quite close to each other, we want to make sure SPSS can identify them separately. The Kiaser-Meyer-Olkin measure verified the sampling adequacy for the analysis, KMO = .863. Bartlett’s test of sphericity 𝒳2 (28) = 1023.103, p < .001, indicated that the correlations between items were large enough for the factor analysis. An initial analysis was done to compute eigenvalues for the two components. Both eigenvalues exceed Kaiser’s criterion of 1 and explained in total 69.841% of the variance. The scree plot revealed a leveling off at the second factor. Therefore, two factors were retained and rotated with an Oblimin with Kaiser normalization rotation. Table 5 shows the factor loadings after rotation. The items that belong to the same factors suggest that factor 1 represents wishful identification and factor 2 represents social identification. The results suggest that the fourth item measuring social identification has high loadings on wishful identification but not on social identification. This item asked participants if they like the influencer. According to literature, liking the influencer could indeed also measure wishful identification (Hoffner, 1996). Therefore, it was decided to leave this item out (Cronbrach’s Alpha social identification (𝛼 = .835) and wishful identification (𝛼 = .881)). The mean of the four items measuring social identification that are left
will form the variable. For wishful identification the mean of the first three items create the variable.
Table 5 Rotated factor loadings wishful and social identification
Wishful identification Social identification
Wish1 .890 -.137 Wish2 .854 .140 Wish3 .894 .032 Soc1 .317 .712 Soc2 .327 .669 Soc3 .272 .715 Soc4 .454 .277 Soc5 -.219 .793 Eigenvalues 4.505 1.082 % of variance 56.316 13.525
Note: factor loadings over 0.40 appear in bold.
Brand-influencer personality fit Brand-influencer personality fit is the last moderating variable examined in this study. It is measured by an adapted version of Moisescu (2009). Participants were asked to indicate their agreement to the following statement: [characteristic] is applicable to [brand or influencer]. A 5-point Likert scale (1=low fit to 5 = high fit) was used to measure agreement. Consequently, fit was measured as the absolute differences between the values of the perceived influencer and brand personality. To make sure high values display high congruence, the absolute differences were subtracted from the maximum value of 5. This means 5 points indicate high congruence and a score of 1 means low congruence. These questions were only viewed by participants who indicated they knew the brand Cluse.
(KMO = .479), Bartlett’s Test of Sphericity was not significant (χ2 (66) = 72.403, p = .275) and the scree plot showed no clear cut of point of the number of variables (Cronbach’s Alpha 𝛼 = (.415)) However, the increase in Cronbach’s Alpha if item is deleted was not higher than .10 for all items. Therefore, the variable brand-influencer personality fit is measured by the initial 12 items. Only for participants who indicated they knew the brand Cluse, perceived brand-influencer personality fit was measured (N=59).
Manipulation checks To check if the sponsorship disclosure manipulation was successful, participants were asked to indicate their opinion on a 7-point Likert-scale (1=Completely disagree to 2=Completely agree) on four statements: (1) The content of the Instagram photo is influenced by Cluse (2) The person on the Instagram photo received financial compensation for posting this photo (3) The person on the Instagram photo received goods or services in return for posting this photo and (4) This Instagram photo reflects a fair and unbiased product review (Carr & Hayes, 2014). The final item is a reversed measure. In addition to that, to check manipulation of the type of influencer-identification, participants are asked to sort the pictures of the presented influencers into one of the four reference group categories of Englis & Solomon (1995) namely: (1) aspirational ‘This person is very similar to how I would like to be’ (2) associative ‘This person is very similar to how I currently see myself’ (3) avoidance ‘This person is very similar to ho I would not like to be and (4) irrelevant reference group ‘This person has no meaning to me’.
Control variables This research used six control variables in total namely social media usage, Instagram usage, brand attitude, influencer similarity, gender and age.
Social media usage Social media usage was measured by the question ‘How many hours do you spend on average on social media at a day? This measure was retrieved from Gutierrez & Cooper (2016). The question was only shown to participants who indicated that they are familiar with social media. However, every participant indicated to be familiar (N=209).
Instagram usage Instagram usage is often just measured by the number of hours spend on the channel (Gutierrez & Cooper, 2016; Ahadzadeh & Sharif, 2017). However, many people do not exactly know how addicted they actually are. Therefore, an adapted version of the Internet Addiction Test (Young, 1998) was used. The measure was reduced to five items namely (1) How often do you prefer the excitement of Instagram instead of being with your close friends? (2) How often do your grades or schoolwork suffer because of the amount of time you spend on Instagram? (3) How often do you check your Instagram before you do something else that you need to do? (4) How often do you lose sleep due to late night log-ins to Instagram? and (5) How often do you find yourself saying “just a few more minutes” when on Instagram? Participants indicated their behavior on a 6-point Likert scale (1=Never to 6=Always). The explanatory factor analysis showed that the Kaiser-Meyer-Olkin statistic is greater than .6 (KMO = .807), Bartlett’s Test of Sphericity is significant (χ2 (10) = 467.385, p < .001) and that one factor is reliable (Cronbach’s alpha 𝛼 = .839). The mean of those five items measures the variable Instagram usage. The question was only shown to participants who indicated that they are familiar with Instagram (N=197). All other participants were assumed to have the lowest rate on the addiction test, since they do not use Instagram (N=12).
Brand attitude Brand attitude is measured according to Mackenzie and Lutz (1989). The variable consists of three-items formulated as: (1) My impression of
[brand] is good (2) My impression of [brand] is pleasant and (3) My impression of [brand] is favorable. The items are measured on a 7-point Likert scale (1=Completely disagree to 7= Completely agree). The explanatory factor analysis of brand attitude indicates a Kaiser-Meyer-Olkin statistic is greater than .6 (KMO = .689), a significant Bartlett’s Test of Sphericity (χ2 (3) = 137.633, p < .001) and a clear cut off point at one variable (Cronbach’s Alpha 𝛼 = .902). The mean of those three items measures a participant’s brand attitude. Only if participants indicate they know the brand Cluse, brand attitude is measured (N=59). Brand attitude is assumed to be neutral when a participant is not familiar with Cluse (N=150).
Influencer familiarity If participants already know the influencers displayed, they might already have an attitude towards them. Therefore, the question “Do you know [name influencer]?” was asked. A dummy was created that indicated 1 if participants answered ‘yes’ and zero if their answer was ‘no’.
Demographics This research controls for gender and age in years. Questions regarding these variables were asked at the end of the survey
4. RESULTS
This chapter begins with a description of the preliminary analysis of the data. Afterwards, randomization and manipulation checks are discussed. Additionally, descriptive statistics of the data are provided. In the last subsection, the developed hypotheses are tested.
4.1 Preliminary analysis
First of all, data was checked for errors. A frequency table of all variables was created to examine errors in data entry, no errors were found. Besides this, descriptive
media was above 24, which is impossible. Therefore, answers of people who indicated to spend over 10 hours on social media (N=9) were checked in Qualtrics. It was found that SPSS did not recognize the ‘.’ as a ‘,’. These answers were corrected. Secondly, items were recoded. This dataset had one counter-indicative item in the manipulation check variable for perceived sponsorship. RevManSp4 was recoded into ManSp4. Furthermore, the variable that indicates in which condition people are, is recoded. A new variable called ConSpon was created. People that were in condition 3 or 6 were in the control condition (=0). Additionally, people in condition 2 and 5 were in the low sponsorship disclosure condition (=1) and participants in condition 1 or 4 were allocated to the high sponsorship disclosure condition (=2). Besides this, a dummy variable called ConType was created for the type of influencer-identification. If people were in condition 1, 2 or 3 they were in the social influencer-identification condition (=0). Additionally, all participants in condition 4, 5 or 6 (=1) were assigned to the wishful influencer-identification condition. The third step is to compute scale means. These were computed for measures consisting of different items, i.e. perceived source credibility, social identification, wishful identification, brand-influencer personality fit, Instagram usage, brand attitude and perceived sponsorship.
4.2 Randomization checks
Participants were randomly assigned to one of the six conditions of this study. A frequency table was created to check how many participants were assigned to each condition. The results are shown in table 6.
Table 6 Frequency of participants across conditions
Manipulation level of sponsorship disclosure
High Low No
Manipulation type of influencer-identification
Social Condition 1: N=31 Condition 2: N=35 Condition 3: N=36
Wishful Condition 4: N=36 Condition 5: N=36 Condition 6: N=36
Randomizations checks were done to check if the participants across conditions are significantly different from each other. Chi-squared tests revealed that there were no significant differences in age (𝒳2 (210) = 229.106, p = .174), gender (𝒳2 (5) = 3.663, p = .599), social media usage (𝒳2 (70) = 70.839, p = .449) and Instagram usage (𝒳2 (90) = 85.047, p = .628). Influencer familiarity differs across conditions (𝒳2 (5) = 13.167, p < .05). Nobody in this sample knows the presented influencer in condition 1, 2 and 3 (Isa Obiols) while some people do recognize the influencer in condition 4, 5, and 6 (Negin Mirsalehi). Besides this, brand attitude significantly differs across the six conditions (𝒳2 (55) = 76.468, p < .05). In condition 2, brand attitude is somewhat higher while in condition 4 it is somewhat lower. Except for brand attitude and influencer familiarity, the randomization was successful.
4.3 Manipulation checks
Of all participants in the no sponsorship disclosure condition, 68.8% indicated that the post was not sponsored. In the low sponsorship disclosure condition 25.0% remembered that the posts was sponsored whereas in the high sponsorship disclosure condition, this was 57.6%. According to the Chi-squared test, the answers to the question “Was the Instagram post by the influencer sponsored?” significantly differ
sponsorship condition differ significantly (𝒳2 (2) = 34.844, p < .001). This is also true for the no and high sponsorship condition (𝒳2 (2) = 73.676, p < .001) and the low and high conditions (𝒳2 (2) = 12.852, p < .01). In order to find out if there is also a difference in level of perceived sponsorship, a one-way ANOVA analysis was done, using perceived sponsorship disclosure as dependent variable and the level of sponsorship disclosure as independent variable. There was a statistically significant effect of the level of sponsorship disclosure on perceived sponsorship (F(2, 206) = 7.330, p < .001). The turkey post-hoc test revealed that the perceived level of sponsorship was significantly higher in the high sponsorship disclosure condition than in the no sponsorship disclosure condition (p < .001). Furthermore, the perceived sponsorship is also higher in the low sponsorship disclosure condition than in the no sponsorship disclosure condition, however the difference is not significant (p = .058). Additionally, perceived sponsorship is higher in the high sponsorship disclosure condition than in the low sponsorship disclosure condition but again, this difference is not significant (p = .279). Therefore, manipulation of the level of sponsorship disclosure is only partially successful.
To check if manipulation of the type of influencer-identification was successful, participants were asked to “indicate in which group the Instagram user fits best”. According to the Chi-squared test, the answers significantly differ across the conditions (𝒳2 (3) = 15.059, p < .01). In the wishful influencer-identification condition, 19.6% indicated that they saw an aspirational person whereas only 0.9% indicated that the influencer belonged to their associative reference group. In the social influencer-identification condition, 12.7% of the participants assigned the influencer to their aspirational reference group and only 8.8% indicated that they saw an influencer that belonged to their associative reference group. This already indicates
that the manipulation was not perfect. Additionally, an independent sample t-test is done to check if there is a difference in the level of social and wishful identification with the influencer between the conditions. The level of social identification is higher in the social influencer-identification condition (M=2.123, SD=1.118) compared to the wishful influencer-identification condition (M=1.792, SD=1.792). This difference was significant (t (207) = 2.294, p < .05). However, the level of wishful identification does not significantly differ between the social influencer-identification condition (M=2.601, SD=2.683) and the wishful influencer-identification condition (M=2.683; SD=1.341) (t (207) = .415, p =. 679). Again, manipulation was only partially successful.
4.4 Descriptive statistics
According to table 7, the youngest participant in the sample is 14 while the oldest one is 65. Almost 80% of the participants in this sample are female. The mean hours spend on social media is more than two, with a maximum of eight hours a day. Furthermore, participants do not score very high on the Instagram addiction test. The mean of this measure is only 2.079. Moreover, the attitude towards Cluse is quite neutral amongst the participants in this sample (M=4.287, SD=.751). Additionally, influencer familiarity is low (M=.057, SD=0.233). None of the participants indicated they recognized Isa Obiols, and only a few (N=12) indicated they were familiar with Negin Mirsalehi.
The mean social identification (M=1.953, SD=1.052) is somewhat lower than the mean wishful identification (M=2.641, SD=2.641), which could be due to the fact that people do wishfully identify with both influencers but not really socially identify with Negin Mirsalehi. The mean fit between the influencers and the brand is 4.319
4.190 (M=4.190, SD=.775). The maximum as well as the minimum level of perceived source credibility are not reached.
Table 7 Descriptive statistics
N Minimum Maximum Mean Std.
Deviation Scale A. Background characteristics
Age 209 14 65 32.665 13.790 Years
Gender 209 0 1 .79 .409 0-1
Social media usage 209 0 8 2.111 1.433 Hours
Instagram usage 209 1 5.02 2.079 1.070 1-6 Brand attitude 209 2 7 4.287 .751 1-7 Influencer familiarity 209 0 1 .057 .233 0-1 B. Identification Social identification 209 1 5.75 1.953 1.052 1-7 Wishful identification 209 1 7 2.641 1.422 1-7 C. Other
Brand-Influencer personality fit 59 3.58 4.92 4.319 0.279 1-5 Perceived source credibility 209 1.93 6.13 4.19 0.775 1-7
Besides the summary statistics, correlations between all variables were calculated to quantify the intensity of the relationship between the variables of this study. Table 6 shows the results. For the record, Cronbach’s Alpha is presented in the diagonal. The results indicate the following. Age is positively correlated with gender (r = .205, p < .01) and social media usage (r = .216, p < .01), which means that if a participant is female, they are often older and spend more time on social media than male participants. Both social media (r = -.318, p < .01) and Instagram usage (r = -.545, p < .01) are decreasing when age is increasing. This confirms the fact that millennials spend more time on social media. The lower level of social media usage of older people could possibly declare the negative correlation with influencer familiarity (r = -.176, p < .05). In fact, Cluse is mainly advertising via social media and the presented ambassadors are mainly active on these channels. Moreover, age is negatively
