<<

Bachelor Thesis

What makes your message credible?

A descriptive study on the effect of source credibility on message credibility.

Authors: Alexander Andersson Eliisa Kreegimäe Nicole Niiranen Supervisor: Michaela Sandell Examiner: Åsa Devine Term: Spring 19 – Semester 6 Course code: 2FE21E Date: 28-05-2019

Acknowledgement

We would like to take this opportunity to show our gratitude to the people who have supported and helped us during the writing process.

Firstly, we would like to say our utmost thank you to our tutor Michaela Sandell, who has been supportive, always been available and has given us useful advice throughout this journey. We appreciate all of the encouraging comments and that you never stopped believing in us.

Secondly, we would like to thank Setayesh Sattari, for helping us with her knowledge on quantitative , that guided us through the statistical phase of this study.

Thirdly, we would like to say thank you to Åsa Devine, for the support and inputs during the seminars.

Additionally, we would like to send a thank you to all of the participants within this research for their contribution to the findings, whom without this thesis would have not been possible to complete.

Lastly, we would all like to thank our parents for the encouragement and unwavering support throughout this academic journey. We would not be here without you.

Linnaeus University, Växjö, Sweden 28th of May 2019

Alexander Andersson Eliisa Kreegimäe Nicole Niiranen

Abstract

Message credibility is a commonly used term to indicate how well the recipients approve the message. Source credibility, on the other hand is a “communicator's positive characteristics that affect the receiver's acceptance of a message” (Ohanian, 1990, pp 41). It has been stated that an effect is apparent between the two, therefore the purpose of this study is to describe the effect of source credibility characteristics; trustworthiness, expertise and attractiveness on message credibility. The study utilized a quantitative research approach in order to describe the effect between the concepts. The researchers used a Likert-scale online questionnaire as the data collection method. The findings of this research suggest that all of the characteristics within the source credibility model are proven to have a significant and positive effect on the message credibility.

Keywords; Message credibility, source credibility, trustworthiness, expertise, attractiveness

Table of Contents

1.Introduction ...... 1 1.1 Background ...... 1 1.2 Problem Discussion ...... 2 1.3 Purpose ...... 3

2. Literature review ...... 4 2.1 Source credibility ...... 4 2.1.1 Trustworthiness ...... 4 2.1.2 Expertise ...... 4 2.1.3 Attractiveness ...... 5 2.2 Message credibility ...... 5

3. Conceptual Framework ...... 7 3.1 Trustworthiness ...... 7 3.2 Expertise ...... 7 3.3 Attractiveness ...... 8 3.3 Research model ...... 8

4. Methodology ...... 9 4.1 Research approach ...... 9 4.1.1 Deductive Research ...... 9 4.1.2 Quantitative research ...... 9 4.2 Data sources ...... 10 4.3 Data collection method ...... 10 4.5 Data collection instrument ...... 11 4.5.1 Operationalization and measurement of variables ...... 11 4.5.2 Operationalization table ...... 12 4.5 Sampling ...... 15 4.6 Data Analysis Method ...... 16 4.6.1 Data Entry, Coding and Cleaning ...... 16 4.6.2 Descriptive Statistics ...... 17 4.6.3 Simple and Multiple linear regression ...... 18 4.7 Quality Criteria ...... 19 4.7.1 Validity ...... 19 4.7.3 Pre-Test ...... 21 4.7.4 Reliability ...... 21 4.7.6 Replication ...... 22 4.8 Ethical Considerations ...... 22 4.9 Societal considerations ...... 23

5. Results ...... 25 5.1 Demographics - Covariance questions ...... 25 5.2 Descriptive Statistics ...... 26 5.3 Reliability and Cronbach’s Alpha ...... 28

5.4 Validity and Correlation Coefficient ...... 29 5.5 Hypothesis Testing ...... 29

6. Discussion ...... 32 6.1 Discussion of the model ...... 32 6.2 Hypothesis 1: Trustworthiness ...... 33 6.3 Hypothesis 2: Expertise ...... 34 Hypothesis 3: Attractiveness ...... 35

7. Conclusion ...... 37 7.1 Accepted Model ...... 37

8. Research Implications ...... 38 8.1 Theoretical implications ...... 38 8.2 Managerial implications ...... 38

9. Limitations and Future Research ...... 39 9.1 Limitations ...... 39 9.2 Future Research ...... 39

Reference list: ...... 40

Appendix ...... 47

1. Introduction

This chapter will present the overall subject of this research, along with problems related to the subject which will lead to the purpose.

1.1 Background Credibility is a subjective measure and it is a perceived quality, that other people determine based on the interaction of numerous factors (Budzowski, 2019). To be more precise, there are two distinct concepts within credibility, which are source credibility and message credibility (Metzger, Flanagin, Eyal, Lemus, & Mccann, 2003, Appelman & Sundar 2016, Li & Suh, 2015). Message credibility is a commonly used term to indicate how well the recipients approve the message. Message credibility is a crucial feature to keep up public relations and to maintain a favourable relationship between two actors; the subject and the public (Ledingham, 2003). Furthermore, it is the receiver of the message who in the end determines how credible the message is (Roberts, 2010). Moreover, the second concept of credibility is as mentioned source credibility. Source credibility is a “communicator's positive characteristics that affect the receiver's acceptance of a message” (Ohanian, 1990, pp 41). A commonly used model within the concept of source credibility is the Ohanian model (1990), which describes characteristics of a person and is often used to measure how credible the source is perceived. The model suggests that there are key characteristics of an individual, which are trustworthiness, expertise and attractiveness. Appelman and Sundar (2016) further state that source credibility could have an effect on the message credibility and suggest the concepts should be investigated if an effect is apparent between the two.

Due to the eruption of social media, it has created a wave of new types of celebrities called social media influencers (Khamis, Ang & Welling, 2017). Social media influencers are online personalities and content creators found on for instance on platforms like Twitter, Instagram or YouTube (Varsamis, 2018). McQuarrie, Miller and Philips, (2013) states that social media influencers are “regular” people who have become well-known due to their presence on social media and they usually have some sort of theme, whether it be fashion or a political stand, that makes them stand out (Varsamis, 2018). The influencers’ content is regulated by the influencer’s themselves, who can choose to portray only beneficial messages and images (Kutthakaphan & Chokesamritpol, 2013). Due to the that every single individual with an internet connection can post information and

1 content online, meaning that anybody can state anything (Tuten & Solomon, 2018). This creates an issue of the message not being perceived credible, therefore losing the possibility to gain a strong social media presence, which results in losing job opportunities.

1.2 Problem Discussion The credibility of a message is important to study in this time and age, since in today’s society every individual has the ability to create content on the Internet (Tuten & Solomon, 2018, Burbules, 1998). There is a problem of being exposed to misinformation online (European Commission, 2018) and misinformation online and especially in social media has become one of the biggest dangers in society (Del Vicario, Bessi, Zollo, Petroni, Scala, Caldarelli, Stanley & Quattrociocci, 2016). According to Burbules (1998), the internet and technology may be set up so that all information is equally accessible to everybody, which consequently leads to that all authors online have similar level of credibility. In any kind of , it is crucial that the receiver perceives the individual and as well as the message credible (Adams Consulting, 2008). Credibility, both source and message credibility, have an impact on influencers ability to influence on social media, and those who are not perceived as credible jeopardize their place in the social media influencer world (West, 2019). Therefore, it is of importance for an influencer to understand what credibility characteristics have an effect on message credibility in order to gain more influence, since according to Influencer Marketing Hub (2019) an influencer is an individual who has the power to influence other individuals. With this being said, an influencer’s job is to more or less to persuade other individuals.

Harkins and Petty (1987) found that the source who communicate the message can have some effect on message credibility, particularly if the source is considered trustworthy and an expert. Furthermore, researchers Strømsø, Bråten, Britt, and Ferguson (2013) found that the respondents often referred back to the sources when they were asked about the information they had read online. This shows that source credibility is present and utilized online by receivers, indicating that source credibility could have an effect on the perceived credibility of the messages put out. In addition, if the information comes from a credible source, for instance if the influencer is seen as credible source, it can influence the receivers beliefs, , attitudes and even behaviour for example (Erdogan, 1999). Additionally, influencers who are considered highly knowledgeable, are respected and socially connected are considered credible sources (Lin, Bruning & Swarna, 2018). Moreover, other studies (e.g. Arndt, 1967; Bickart, 2001; Berkman and Gilson, 1986; Lee and Youn, 2009; cited in Martensen, Brockenhuus-Schack & Zahid, (2018) have found that in order for an individual to be considered

2 influential or someone whose opinions one should listen to, source credibility was found to be of high importance to persuade that individual. Furthermore, research within the subject of persuasion expresses that a highly credible source can increase the persuasiveness of the message (Pornpitakpan, 2004). As stated earlier, the two concepts within credibility, source and the message credibility are considered as two separate concepts (Metzger et al., 2003, Appelman & Sundar 2016, Li & Suh, 2015). However, some of the measures in message credibility can be related to some source credibility dimensions but are not connected in practice and have theoretically two different meanings. The measure constructed by Appelman & Sundar (2016) displays the difference between the two concepts. They further suggest that there is a need for a research measuring the effect of one credibility on the other. Therefore this research tries to study the source credibility effect on message credibility.

1.3 Purpose The purpose of this study is to describe the effect of source credibility on message credibility.

3 2. Literature review

This chapter presents the theories, source credibility and message credibility, which will be used as a foundation for this research.

2.1 Source credibility 2.1.1 Trustworthiness Trust, according to Ohanian (1991), is the receivers confidence in and acceptance of the speaker. Meanwhile Chen (2004) suggest that being authentic, being true to oneself, indicates trustworthiness. The importance of authenticity is further expressed by Banks (2008), who expresses that speakers themselves consider being authentic is important for them. Cheung et al., (2009) argues that consumers are more easily convinced if the source is seen as trustworthy and comes across as dependable and genuine. According to Erdogan (1999), trustworthiness is recognized as honesty and believability. This is further supported by Copeland, Gunawan and Hernandez (2011) who suggest that honesty and believability of a source are considered essential dimensions when determining one’s trustworthiness. Ohanian (1990) studied which different dimensions correlate with source-credibility and found that honesty, dependable and being reliable having a significant impact on perceived trustworthiness. Additionally, Morgan and Hunt (1994) state that trustworthiness exists when one has certainty in other’s reliability and honesty.

2.1.2 Expertise Expertise is defined by Erdogan (1999, pp. 298) as ‘the extent to which a communicator is perceived to be a source of valid assertions and by McCracken (1989) as the ability of the person to provide valid information. It refers to the knowledge or experience by an endorser (Erdogan, 1999), additionally Gilly, Graham, Wolfinbarger and Yale (1998) argue that expertise is the knowledge the source has. Tedeschi, Schlenker and Bonoma (1973) argue that expertise develops from for example experience. Study conducted by Martensen et al., (2018) states that respondents perceived influencers as having expertise if they had worked within a certain area for a longer period of time, meaning that they possess experience within a field. A study made by Zhao, Leo, He, Lin & Wen, (2016) found that there is a relationship between reputation and expertise. Influencers with bigger reputations are perceived as more skilled than influencers with smaller reputations. The relationship is also stronger if the influencer is in his/hers expertise category, but nonetheless there is a correlation between the amount of influence and a category that the influencer has no expertise in.

4

2.1.3 Attractiveness The attractiveness of an influencer often refers to their physical appearance and studies have shown that it is a factor that seemingly has a positive affect on how brands are perceived (e.g. Eisend and Langner 2010; Till and Busler 2000, cited in De Veirman, Cauberghe, & Hudders, 2017). However, the significance of the effect of attractiveness may depend highly on the product being advertised. It is important to mention that attractiveness is more than the physical attractiveness and outer looks of the person, other factors relate to it too, such as, perceived familiarity, similarity and likeability (McGuire 1985, Ohanian, 1991, Nunes, Ferreira, De Freitas & Ramos, 2018). It has been found that individuals tend to follow influencers with whom they identify with and therefore, the followers’ perceived similarity to the influencers have a positive effect to their trust in influencer-generated branded content (Lou & Yuan 2019). A study made by Martensen et al., (2018), discovered similar findings when examining influencers’ perceived similarity. More specifically, they found physical appearance and lifestyle to be particularly important factors in the context of perceived similarity. The results showed that the individuals are able to relate and see similarities with their favorite influencer because they have a “normal body” and the influencer is a “normal person”. Additionally, the findings of Martensen et al., (2018) indicated that perceived similarity relates to the amount of personal content the influencer posts. The more personal content, the more it humanizes the influencers profile and this makes them more approachable and therefore seem more familiar (Martensen et al., 2018).

2.2 Message credibility Message credibility refers to the characteristics of a message that influence the believability of the message (Roberts, 2010). Newell & Goldsmith, (2001) state that message credibility is said to be the impressions and judgements the receiver has in relation to a message. Ideally, the perception of the message is positive (Ohanian, 1991). It refers to the characteristics of messages that could make them more or less credible (Metzger et al., 2003). This could be further supported by, Roberts (2010) who states that that the individual that receives the message decides on the credibility of the message. Appelman and Sundar (2016) found that to be able to measure message credibility, the adjectives accurate, authentic and believable should be used together to demonstrate message credibility. O’Keefe (2002) suggests that credibility could be characterized as an individual’s perception of how believable the communicator’s content is. According to Appelman and Sundar (2016) it is the receivers of how true the communicated content is which could be seen as an attribute of message credibility.

5

Message credibility may be affected by which channel it is distributed via or even how the message is structured (Metzger et al., 2003). Furthermore, Sisco and McCorkindale (2013) and Appelman and Sundar (2016), state that in order for the message to be perceived as credible, the content needs to be coherent across different platforms and regularly updated. Additionally, Sisco and McCorkindale (2013) state that the subject should join in on the conversation on the platform. Appelman and Sundar (2016) have also recognized as a factor of credibility, meaning that the message should be unbiased. Furthermore, Roberts (2010) has also recognized unbiasedness as an item within message credibility to consider.

6 3. Conceptual Framework

In this chapter the theoretical concepts will be conceptualized and hypotheses will be developed and presented along with a proposed model.

3.1 Trustworthiness McGinnies and Ward (1980) found that a trustworthy source has an effect on the source’s persuasiveness and opinion change. This is further strengthened by Martensen et al., (2018) since they found a correlation between trustworthiness and persuasiveness. Moreover, they suggest that an influencer is more trusted than other sources, mainly due to their ability to be personal and real, which the respondents expressed was important when absorbing recommendations. Meaning that influencers have the ability to construct credible messages as well, since Banks (2008) states that in order for a message to be credible one needs to be personal. This expresses that a highly trusted source are most likely to possess message credibility as well. Therefore the hypothesis is as follows:

H1+: Trustworthiness of the influencer has a positive effect on perceived message credibility.

3.2 Expertise The second attribute in this study is expertise. Crisci and Kassinove (1973) found that the perceived level of expertise of an individual had an impact on how well receivers took their advice. When the individual was an expert, the receivers took a more positive stance towards the advice, compared to a non-expert source (Crisci and Kassinove, 1973). This can be connected to message credibility, which is characterized as how believable (O’Keefe, 2002) and positive (Ohanian, 1991) the message is perceived by an individual. Furthermore, a source with expertise has the possibility to produce valid information (McCracken, 1989) meaning that the source has the ability to construct messages that are seen “as being true and acceptable” (Cambridge dictionary, 2019). Meaning that expertise source has the ability to communicate credible messages, since Appelman and Sundar (2016) who state that message credibility is the receivers of how true, ie. valid, the content is. Thus the following hypothesis:

H2+: Expertise of the influencer has a positive effect on perceived message credibility.

7 3.3 Attractiveness Attractiveness is the third characteristic of an influencer in this study and identified as a part of source credibility. Someone’s physical attractiveness can have a great effect on people, since being attractive is commonly perceived as positive (Burgoon, Guerrero and Floyd, 2010) and it could even change individuals attitude towards being more positive (Kahle and Homer, 1985). Additionally, Castellow, Wuensch, Moore (1990) found that attractive people are seen as more believable than unattractive people, indicating that attractive people can create more credible messages, since Appelman and Sundar (2016) found that attributes such as accurate, authentic and believable are dimensions that determine message credibility. This is also supported by O’Keefe (2002), who stated that the believability of the communicators content characterises message credibility. This implies that attractiveness could have a positive effect on message credibility, thus, the hypothesis is as follows:

H3+: Attractiveness of the influencer has a positive effect on perceived message credibility.

3.3 Research model The research model developed shows the different independent variables in the form of the characteristics of social media influencers and the hypotheses are constructed in order to see the effect on the dependent variable, message credibility.

Figure 1. Proposed model.

8 4. Methodology

This chapter justifies the research approach and strategy for the study and further presents the operationalization table with the theories, which then are transferred into a questionnaire.

4.1 Research approach 4.1.1 Deductive Research Choosing the nature of the study is one of the first steps to take when conducting a research. Deductive nature is characterized as a nature where the theory directs the study, meaning that the researcher tries to answer questions presented by the theoretical discussion. In deductive research, based on what is already established with theoretical in certain area, the researcher can then conclude a hypothesis or hypotheses. The hypotheses contain concepts that require translation into researchable, independent existence. The researcher should carefully propose accurate hypotheses and then convert them into operational and working terms. This means that the researcher should designate the data collection methods related to the concepts that build up the hypotheses. These hypotheses are then tested on empirical investigation (Bryman & Bell, 2011). The researchers have studied the existing theories and studies regarding message and source credibility, basing the study on existing theories and creating hypotheses out of them, therefore making it a deductive research. Data was collected and analyzed, and the created hypotheses were either accepted or rejected.

4.1.2 Quantitative research Quantitative research often starts with an deductive approach and is defined as a strategy that accentuates on quantification of data collection and data analysis and it can be therefore generalized, due to its large data. Quantitative research is thought to be objective, since the researcher is not in close contact with the respondents and is considered to have lower risk of being biased. This is due to the fact that often in quantitative research the Quantitative research tests the hypotheses that are proposed by the theory and investigates the relationship between the variables. Variables are aspects that can be measured and is divided usually into dependent variables and independent variables. An dependent variable is an concept which is thought to be affected by the independent variable. Quantitative research therefore tries to demonstrate considerable connection between variables (Ingham-Broomfield, 2014). In this study, source credibility; trustworthiness, expertise and attractiveness are the independent variables and message credibility is the dependent variable. Quantitative research makes it possible to analyze the different variables and describe what effect

9 source credibility has on message credibility. Additionally, this strategy enables to keep the study objective, unbiased and generalisable.

4.2 Data sources Research can be based on data collected from primary or secondary sources. This research has gathered primary data, to collect relevant information to gain an interpretation of the effect that the independent variables can have on the dependent variable (Bryman & Bell, 2011). Primary data enables to get answers to the particular subject the researcher studies (Malhotra, 2010). Furthermore, Bryman and Bell (2011) state that the primary data collected should be relevant and impartial in order to fulfil its purpose. In order to ensure a collection of only relevant information, the data collection instrument must be carefully decided upon, which we be described more in chapter 4.4.

4.3 Data collection method A survey is a suitable method to collect large amounts of data in quantitative study and Bryman and Bell (2011) state that using a survey is a time- and cost-efficient way of gathering data. Since this study had little resources to use, a survey was chosen. More specifically, a survey in form of a online self-completion questionnaire with closed questions was created in order to gather data from the respondents. Bryman and Bell (2011) explain closed questions as offering prepared answers to the questions, which makes it easier to measure the outcome and to make sure the intention for the study stays clear. Since the respondents are given options to answer from, it makes the process of answering more effortless. Saunders, Lewis and Thornhill (2009) point out that online questionnaires are an effective way to gather data and collect answers. A major benefit of an online questionnaire is that it often reaches a large audience.

The survey should be easy to follow with as little distractions as possible and the questions should be stated in a sensible matter to avoid confusion. Before the participant was able to fill in the survey, concepts such as “social media” and “influencer” were explained in the landing page as well as concepts such as “authentic”,“dependable”, “honest”, “genuine” and “familiar” in order for the respondents to understand the difference. Furthermore, the respondents were instructed on what or who to think of when doing a certain section of the survey, an example of one instruction was “Think of one influencer that you follow on social media and keep that influencer in mind when answering the statements”. The survey also included instructions on how to fill out the survey (See Appendix). The questionnaire was posted on Facebook, Instagram and LinkedIn, meaning that it was an online

10 survey. Facebook, Instagram and LinkedIn were chosen because of the large audience and because all of the platforms are used by social media influencers, therefore it was reasonable to post the survey on those platforms.

4.5 Data collection instrument 4.5.1 Operationalization and measurement of variables Once the theoretical framework has been gathered it is possible to conduct an operationalization table. Operationalization is a concept of measurement and is accustomed to build the items in order to measure the concept (Bryman & Bell, 2011). Before the respondents could answer the statements developed, it was made sure that they answer the control questions. The control questions were “Are you active on social media?” and “Do you follow at least one influencer on social media?” in order to gather relevant data. It was further disclosed that being active means that one regularly checks their social media, but it does not mean one has to actively post content. If the respondents answer “No” to one of the two questions, they will be they will be exited from the survey. If they answer “Yes” to both questions, they could continue to fill out the survey. The questionnaire in its full form can be found in the appendix. In addition, questions regarding age, gender and educational levels were asked, since those factors could end up to be moderators.

The survey used Likert-scale enabling to measure the answers. In this research a Likert-scale from 1 to 5 was used. The questionnaire had answers from 1 as Strongly disagree, 2 as Disagree, 3 as Neutral, 4 as Agree and 5 as Strongly agree. This allows to move the gathered data for example into SPSS and makes the coding and analysis effortless (Bryman & Bell, 2011). However, four questions from the survey did not use Likert-scale, but these questions were not used to measure the relationship among the variables. These questions were regarding gender, age, educational level and country one lives in. Gender had three options to choose from; Female, Male and Non-binary, meaning it used nominal scale. Regarding age, the research used ordinal scale starting from 15-19, 20-24, 25-29, 30-35 and 36 and above. Due to the legal age in Sweden to participate in without parental consent is 15, this was seen as an appropriate age (European Union Agency for Fundamental Rights, 2019). The 5-year-span was chosen because of not wanting to miss any important data and possibly not identify age as moderating variable. The reasoning behind stopping at 36 and above, was that it could be seen at that age one already has a fully developed adult brain (Wallis, 2013). An education level was additionally asked, to see if that could have effect on how respondents reply, if that could be moderating variable. The educational levels were “High school”, “Vocational college”, “University of applied sciences”, “Bachelor’s degree”, “Master’s degree”, “Doctorate degree” and “I do not wish

11 to say”, since in some cases respondent do not wish to say. Lastly, the survey asked which country the respondent lives in, which was an open question, meaning that the respondents could answer any country they wanted.

4.5.2 Operationalization table The operationalization table shows how the statements are connected to the theory and therefore gives direction when creating the questions for the questionnaire. The table shows which statements belong to which theoretical concept and dimension. The last theoretical concept is the dependent variable message credibility.

Table 1: Operationalization table.

Theoretical Item Reference Dimension Statement concept number

Trustworthiness Trust1 Chen, 2004 & Banks, Authenticity. I consider the influencer (H1) 2008. authentic.

Trust2 Cheung et al., 2008. Dependability. I consider the influencer dependable.

Trust3 Cheung et al., 2008. Genuineness. I consider the influencer genuine.

Trust4 Erdogan, 1999, Honesty. I consider the influencer Copeland, Gunawan being completely honest. & Hernandez, 2011 & Morgan & Hunt, 1994.

Trust5 Erdogan, 1999, Believability. I consider the influencer Copeland, Gunawan believable. & Hernandez, 2011.

12

Trust6 Morgan & Hunt, Reliability. I consider the 1994 & Ohanian, influencer reliable. 1990.

Expertise (H2) Expert1 McCracken 1989 & Validity. I consider the influencer to Erdogan, 1999. post valid information.

Expert2 Erdogan, 1999, Knowledge. I consider the Tedeschi, influencer knowledgeable. Schlenker, Bonoma, 1973 & Gilly, Graham, Wolfinbarger and Yale, 1998.

Expert3 Erdogan, Experience. I consider the influencer 1999, Tedeschi, experienced in her/his field. Schlenker and Bonoma, 1973 & Martensen et al., 2018.

Expert4 Zhao, Leo, He, Lin & Reputation. I consider that the Wen, 2016. influencer has a good reputation.

Attractiveness Attract1 De Veirman, Physical I consider the influencer (H3) Cauberghe, & appearance. physically attractive. Hudders, 2017.

Attract2 Lou & Yuan, 2019. Similarity. I consider it important to have similarities between myself and the influencer.

13

Attract3 Martensen et al., Similarity. I appreciate if the 2018. influencers lifestyle is similar to mine.

Attract4 Lou & Yuan, 2019. Familiarity. I consider it important to be familiar with influencer.

Attract5 McGuire 1985 & Likeability. I consider the influencer Ohanian, 1991. likeable.

Message MCred1 Appelman and Perception of I consider the messages put Credibility Sundar, 2016. the message out by the influencer (accuracy, accurate, authentic and authenticity believable. and believability).

MCred2 O’Keefe, 2002. Believable I believe everything the communicator influencer

says.

MCred3 Newell & Goldsmith, Positive My impressions are 2001 & Ohanian, impressions in positive when I see the 1991. relation to the influencers’ messages on message. my social media platform(s).

MCred4 Sisco and Coherence. I appreciate it if the McCorkindale, 2013, influencer is coherent with Metzger et al., 2003 his/her posts across & Appelman and different social media Sundar 2016. platforms.

MCred5 Sisco & Updated. I appreciate if the McCorkindale, 2013. influencer regularly

14 updates their social media platforms.

MCred6 Sisco & Engagement. I appreciate it if the McCorkindale, 2013. influencer engages with their followers.

MCred7 Appelman and Objectivity. I appreciate if the Sundar, 2016. influencer’s message is unbiased.

4.5 Sampling In order to describe sampling, one needs to understand what a population and a sample is. A population defines all of the individuals relevant to the study and a sample is the portion of the population that will represent the population. The usage of the whole population is rarely emphasised due to the large size of the population and due to budget. This research emphasised the convenience sampling technique, meaning simply gathering the sample based on the researchers’ accessibility. More specifically, the questionnaire was distributed where thought it would be most suitable and convenient, in this case on social media. When utilizing the convenience sampling method the researcher is most likely to receive a higher response rate. However, the problem with this method is that the generalizability of the findings are relatively low, since it is not known what population the sample is representative of. Despite the issue of generalizability of convenience sampling, it does provide a foundation to the area of research (Bryman & Bell, 2011). The questionnaire was posted on the researchers Facebook, Instagram and LinkedIn pages, as well as other relevant Facebook groups, which the researchers thought where suitable. The researchers chose to not have any control over who specifically participated, however in addition to the researchers personal social media circle, the questionnaire was posted on forums and groups in order to gain results outside of the circle and therefore making the results to some extent more generalizable (Bryman & Bell, 2011).

When sending out a survey, the sample size is one factor that should be considered. Time and cost are two big aspects that can have an effect on the sample size. A larger sample is more time and cost consuming, however too small of a sample can have bigger sampling errors and if the sample is insufficient, it cannot represent the population (Green, 1991; cited in Morgan & Van Voorhis, 2007).

15 Therefore, a formula was used in order to understand how big of a sample size the researcher should reach. Green (1991; cited in Morgan & Van Voorhis, 2007) suggest that in multiple correlation testing, the amount of respondents (N) should be at least 50 + 8m, “m” meaning the number of independent variables. This formula is considered suitable for studies that have less than 7 independent variables. Thus, the sample size can be calculated with the formula N> 5 + 8m. This calculation would be performed as following: 50 + 8 x 3 = 50 + 24 = 74. Therefore the sample size should be at least 74. However, Maxwell (2000) cited in Henning and Cooper (2011), suggest that samples that are smaller than 140 are too small to identify correlations, therefore the minimum amount of respondents for this study was decided to be 140.

4.6 Data Analysis Method 4.6.1 Data Entry, Coding and Cleaning According to Bryman and Bell (2011), in order to analyze the data collected, the most suitable approach is to use a computer software program. Nowadays the most popular package of statistical computer software is called SPSS, which stands for Statistical Package for the Social Sciences and was used for this research as well, in order to simplify the data analysing process. Furthermore, Bryman and Bell (2011), state that once the data has been entered into SPSS, it should be coded. Coding meaning that the variables are defined and given codes that enable the variables to be categorized into different groups. Once the different categories are defined, they are numbered so that SPSS can process the given information. Data cleaning is done as the the researcher assigns the numbers/value of each variable, by doing so SPSS can indicate a missing values through inconsistent or missing answers (Bryman & Bell, 2011). No data cleaning was required for this research, due to all of the questions being mandatory to be answered and it was impossible to move forward without answering all of the questions asked. Furthermore, SPSS did not detect any inconsistencies in the values, confirming that data cleaning was not necessary.

For this research, since most of the questions in the questionnaire had a Likert scale 1-5, the responses were already coded, which helped the transferring process into SPSS. Moreover, the variables were named after the hypothesis it belonged to, so for instance, the statements regarding trustworthiness were coded as Trust1-6, expertise as Expert1-4, attractiveness as Attract1-5. The statements regarding credibility were coded as MCred1-7 and the statements regarding the respondents themselves were coded as gender, age, education and country. For example, gender was coded as 1 = Female, 2 = Male, 3 = Non-binary. Ages were coded as 1 = 15-19, 2 = 20-24, 3 = 25-29 , 4 = 30-35 and 5 = 36 and above. The educational levels were coded as 1 = “High school”, 2 = “Vocational college”, 3 =

16 “University of applied sciences”, 4 = “Bachelor’s degree”, 5 = “Master’s degree”, 6 = “Doctorate degree” and 7 = “I do not wish to say”. Lastly, country was coded as 1 = Nordic countries and 2 = Other.

4.6.2 Descriptive Statistics In order to get an overall understanding of the generated data, descriptive statistics is commonly used. Descriptive statistics is described as numerical technique to present and analyse data. It allows the researcher to cut down all of the collected data from for instance a questionnaire into a brief summary of numbers, which enables the researcher to generate meaning from the results (Fisher & Marshall, 2009). This technique was therefore chosen by the researchers of this study because it enabled to summarize the large amount of data collected from the questionnaires into a more simplified form. According to Bryman and Bell (2011), the two most common measures of describing variables are central tendency and dispersion. Saunders et al., (2009) suggest that there are three main measurements for central tendency: mode, median and mean. Mode is described as the value that occurs in the data most frequently. Median is defined as the middle point of all the data, once it has been ranked. Lastly, the arithmetic mean, also known as simply the mean, is the most common measure of central tendency. The mean is the average of all the values and is calculated by adding all of the values in the collected data set and dividing them by the number of values (Bryman & Bell, 2011). Moreover, Fisher & Marshall (2009), state that when calculating central tendency, in order to get a normal distribution of the variables, the values of mode, median and mean should be more or less equal.

Dispersion on the other hand was measured by standard deviation in this research, which is explained by Bryman and Bell (2011), as the average amount of variation to the mean. More specifically, it is calculated by taking the difference between each value in the data and the mean and dividing the total of the differences by the number of values. This was done in order to find out to what extent the measure differs from the mean. According to Fisher and Marshall (2009), the value of standard deviation can be defined as, the slimmer the standard deviation, the closer the majority of the measures will be to the mean. Furthermore, they state that the researchers should aim to have a similar value when it comes to mode, median and mean, if the values differ it is said to be “skewed”. Skewness and kurtosis are used to gain a better understanding of the distribution, more specifically they are used to measure the shape of the distribution (Malhotra, 2011). According to Nolan and Heinzen (2014), skewness is a term for data that is off-centered and non-symmetric. Moreover, skewed data has a tail more in one direction or the other meaning that depending on the value it can

17 be either positive or negative. For instance a positively skewed distribution has a tail more to the right and a negatively skewed distribution has a tail more to the left. Kurtosis, according to Malhotra (2011), is a measure of a distributions peakedness or flatness. The shape of the distribution is considered normal if the kurtosis is closer to zero, whereas a value between ±2.0 is in many cases acceptable. As for skewness, a value between ±1.0 is considered excellent, however as with kurtosis, a value between ±2.0 is in many cases also acceptable (George & Mallery, 2003).

4.6.3 Simple and Multiple linear regression After coding the data and receiving the descriptive statistics, the relationship between the dependent and the independent variables were investigated. In order to do so, techniques such as correlation analysis and multiple linear regression analysis was used (Hair, Black, Babin & Andersson, 2014). Moreover, these technique were seen as appropriate for this research, since the purpose was to test the independent variables trustworthiness, expertise and attractiveness on the dependent variable message credibility.

Simple linear regression allows the researchers to investigate relationships between two quantitative variables (The Pennsylvania State University, 2018). Meanwhile, multiple regression is a similar statistical technique that predicts scores of a single dependent variable from the scores of more than one independent variable. It was beneficial especially for this research, since it allowed the researchers to see the dependent variables’ relationship with the independent variables. More specifically, the regression analysis enabled the researchers to see the effect that the independent variables had on the dependent variable and therefore better predict a given result. When interpreting the results, the researchers can look at the coefficient of multiple determination (R²), which shows how strong the relationship is between the numerical dependent variable and two or more independent ones (Nolan and Heinzen, 2014). The value of coefficient determination is usually between 0 and 1 (Saunders et al., 2009). Moreover, the researchers can look into the adjusted coefficient of determination (adjusted R²), which takes into consideration the number of independent variables included in the regression equation and the sample size. More specifically, the adjusted R² only measures the independent variables that have enough over the dependent variable (Hair et al., 2014). Furthermore, when conducting the multiple linear regression on SPSS, the researchers can look at the results of standardized regression coefficient (β), also called the beta coefficient. Standardization is defined as the process of raw data being transformed into new variables and the term β coefficient is used to express the standardized regression coefficient (Malhotra, 2011).

18 Moreover, Hair et al., (2014), define the β-value as the value determining which independent variable holds the most explanatory power over the dependent variable, so which one of them has the most impact. Additionally, the β-value should be high, the closer it is to 1 the better. The independent variable with the β-value closest to 1 signifies that it has the largest impact on the dependent variable (Malhotra, 2011). Nolan and Heinzen, (2014), define standardized regression coefficient, β-value, as a standardized version of a slope obtained by the degree of change in the dependent variable when changing the independent variable(s). Furthermore, Nolan and Heinzen, (2014), state that coefficients standard error of the estimate is the number signifying the average distance between a regression line and the actual data points or in other words, it indicates the average amount that a data falls from the regression line.

Significance testing is done through hypothesis testing, where the relationship between variables are examined. This is also determines whether or not the hypotheses are accepted or rejected (Malhotra, 2011). Moreover, according to Malhotra (2011), when entering the data on SPSS, it allows the researcher to test the probability of the test-results (p-value). Furthermore, if the p-value of the test- results are low, below 0.05, this indicates that there is a statistically significant relationship between the independent variable(s) and the dependent variable. This indicates that the hypotheses can be accepted (Malhotra, 2011). However, if the p-value is higher than 0.05, there is no significant relationship between the independent variable(s) and the dependent variable, meaning that the hypotheses are to be rejected (Saunders et al., 2009). Furthermore, the researchers should look at the F-value, which results from using the F-test. If the F-value is high and significant, this indicates that the variables contribute to the model. Therefore a high and significant F-value is most favorable (Hair et al., 2014).

4.7 Quality Criteria 4.7.1 Validity When expressing validity in a research context it is often implied that an indicator or a set of indicators is measuring what it supposed to. In other words, the extend of which the method correctly measures the concepts that was meant to be measured (Saunders et al., 2009). Moreover, content validity is often used to refer to the degree of which the questionnaire questions in reality measure the item or concept that it was intended to measure (Saunders et al., 2009). It is subjective yet efficient and organized interpretation how adequately the content displays the measurement. The researcher investigates if the items actually comprehend the whole area of being studied. Bryman and Bell (2011) explain that doing that is the least of what the researcher can do to determine if they gather

19 appropriate material. Additionally, the researcher can further examine if the measures demonstrate the topic by asking other people, eg. experts on the subject (Saunders et al., 2009; Bryman and Bell, 2011). However, content validity is not enough to measure validity of the research, however it presents reasonable understanding of the measurements (Malhotra, 2011). The researchers of this study have done a precise and comprehensive background review of the existing theories and studies about source credibility and message credibility. Based on the existing literature that was reviewed, the characteristics which are considered as different variables, were then turned into measurable items and each item was exploring certain element of the variable. These connections between theory and variables were then examined by two experts in the field and after gaining the feedback and suggestions, tweaks were made. After that the statements were reviewed by the pre-testers.

The construct validity focuses on if the experiment or the measures are constructed in a way that it measures what it is alleged to measure. Meaning that the investigated measures could have smaller impact than claimed, due to that other measures that were not measured could have an impact on the results. Perhaps even the measure could be an invalid measure of the concept in focus (Bryman & Bell, 2011). In this research Pearson’s correlation or perhaps better known as Pearson’s r, was used to measure how much variables correlate, i.e if the variables have some sort of a relationship. The calculation will yield a coefficient, which will always appear between -1 to +1. However, the coefficient is almost always between 0 and +1. When the coefficient between two variables is zero it indicates that it is no correlation at all and when the coefficient are +1 it is implying that it is a perfect relationship. If the coefficient is -1, it is still a perfect relationship, however it indicates that the relationship is in the opposite direction. When the coefficient is getting nearer 1, it expresses a stronger relationship (Bryman & Bell, 2011). Furthermore, a value higher than 0.9 will cause problem (O’Brien and Scott, 2012), due to the fact that the variables are more or less measuring the same thing. Variables should be deleted if they have a tolerated value lower than 0.1. A tolerance value is calculated as 1- R2 (O’Brien & Scott, 2012).

A positive/negative perfect relationship expresses that when one of the variable increases/decreases the other correlating variable increases/decreases by the same amount, meaning that no other variable is interfering or connected to them. This further implies that if the coefficient appears to be less than 1, indicating that other variables are related or have a connection to the original variable, however the variables still have a correlation even though other variables have an impact. Otherwise the results of the calculations would be zero indicating no relationship. Moreover, if there would not appear any apparent patterns on the diagram it would mean no correlation, i.e zero coefficient. This implies that

20 the variation of each variable is connected to other variables not present in the conducted analysis (Bryman & Bell, 2011). Additionally, according to Saunders et al., (2016), if r = 0, the variables are perfectly independent, however this is extremely unusual in business research.

4.7.3 Pre-Test When developing a new measure one need to consider face validity. Meaning that the measure should mirror the concept that one is investigating. In order to ensure that one obtain face validity one could ask other individuals if the measure in the research is focusing on concepts. Therefore, one is often using some type of expert or someone with previous knowledge within the field of the research. Thereby, this process is often referred to as a intuitive process (Bryman & Bell, 2011).

Therefore, pre-test, often referred to pilot test, is important when conducting a questionnaire and it assures that the survey operates and the research instruments function well. Pre-test is done by allowing other individuals to through the questions or statements critically and letting them comment on the structure, wording and understanding in general. Pre-testing has the ability to stop any unnecessary time and effort spent on finishing problems after the questionnaire is sent out. Bryman and Bell (2011) state that pre-testing and the received feedback can guide the researchers in formulating the statements and questions and is beneficial since pre-testing can help avoiding possible problems when individuals start answering. Because some problems may not be apparent until individuals start answering, it is important to eliminate or change the questions that the individuals may have problems with, before sending it out (Bryman & Bell, 2011).

This questionnaire was pre-tested by 14 individuals, one of them being a university lecturer and others individuals who use social media and follow influencers. Those people were chosen because they are familiar with the context and those where the criteria for answering the questionnaire. Furthermore, it was asked from the pre-test participants if the questions made sense, if the order of the statements was relevant, is the text and the statements understandable, and any suggestions for improvements. After the collected information was gathered by the participant some minor tweaks to the questionnaire was made in order to erase some confusion. The used and final version of the questionnaire is attached in Appendix 1.

4.7.4 Reliability If a study and the findings of a research is repeatable it is often referred to as reliable. Meaning that the measures used should be consistent. Quantitative studies is often affected by the problem of using

21 variables that measure the intended purpose (Bryman & Bell, 2011). In order to assure or determine if one possess internal reliability one can utilize the split-half method, meaning that one divides all of the measures randomly into two different groups. Thereafter a correlation between the scores of each groups could be calculated and that calculation will construct a coefficient. Meaning that one could investigate for example a measure that scores very good at one of the groups do that in the other as well, if so, it is indicating what it suppose to and it does not lack coherence with other measures. Furthermore, to test internal reliability Cronbach’s alpha is commonly used and was used in this study to provide reliability, since Cronbach alpha calculates the average of every split-half coefficients. The calculated Cronbach’s alpha value is something between 0, meaning no internal reliability, and 1, meaning perfect reliability (Bryman & Bell, 2011). Bryman and Bell (2011) suggest that the result should be at least 0.80 and is used as a rule of thumb, however, Cortina (1993) has argued that no additional improvement is needed if the alpha is 0.70 and the number indicates sufficient reliability. Taber (2016, pp. 1278) describes the alphas as following “excellent (0.93–0.94), strong (0.91–0.93), reliable (0.84–0.90), robust (0.81), fairly high (0.76–0.95), high (0.73–0.95), good (0.71–0.91), relatively high (0.70–0.77), slightly low (0.68), reasonable (0.67–0.87), adequate (0.64–0.85), moderate (0.61–0.65), satisfactory (0.58–0.97), acceptable (0.45–0.98), sufficient (0.45–0.96), not satisfactory (0.4–0.55) and low (0.11)”.

4.7.6 Replication The topic of replication is of great importance when for example determining the reliability of a research. Meaning that, without the possibility to replicate the study one can not determine if the findings are reliable or not. Therefore, the researchers must express their process and how they conducted their research clearly and with great precision, in order for others to be able to replicate the study. Since, occasionally researchers try to replicate a study, perhaps due to new contradictory findings within the field or that the original findings do not match other results (Bryman & Bell, 2011). The ability for others to replicate the study is ensured if one carefully follows the guidelines made clear in this chapter.

4.8 Ethical Considerations According to Bryman and Bell (2011), when conducting a study that has people involved, some potential ethical issues might occur. Therefore, when collecting primary data from respondents, there are few ethical considerations to examine. The factors that should be considered are potential harm, lack of informed consent, invasion of privacy and action of withholding the main for the conduction. These factors are developed by the Institutional Review Board (IRB) which is an

22 administrative board to provide safety for humans when they are taking part of researches (Oregon State University, 2019). Due to the fact that this study used human participants, it was important to consider these guidelines developed by IRB.

Bryman and Bell (2011) explain that researchers must ensure that there is no potential harm to the respondents when they participate in the study. Mainly, potential harm meaning physical harm, stress, harm to self-esteem and career wise. Furthermore, the participants should voluntarily and knowingly take part and not be pressured by the researchers to take part of the study. Additionally, especially for this study, the participants were informed that there was no compensation from participation. Since this study used an online questionnaire to gain primary data, the researchers were able to make the participation anonymous, meaning that the answers cannot be traced back to the participant. The respondents filled them out voluntarily and therefore the issue with lack of consent was not a concern since the questionnaire was posted online. There was no time limit to fill out the survey nor did the questions provoke any disturbing thoughts thus not creating stressful environment. The questionnaire was formed in a way that it ensured anonymity to the respondents of the study, there were no questions asked that would have made it possible to identify who has answered what. Anonymity was also ensured and stated on the cover letter of the questionnaire, so it was guaranteed that there is no harm to the participant if they take part. It was made sure to the respondents the purpose of why they are filling the questionnaire, since Bryman and Bell (2011) explain that withholding the true reason for the research should not take place and its considered unethical. No deception took place since the questionnaire clearly stated the purpose of the research and who the authors of the thesis were. Also the participants had the opportunity to email to the researchers, if they would need any further information regarding the questionnaire or the study. These factors can be seen in the Appendix.

4.9 Societal considerations When executing a study the researchers need to consider that the findings may cause different societal issues (Malhotra, 2011). Since this research is more or less giving individuals or influencers the ability and knowledge to construct more persuasive messages i.e be more influential, this research is could potentially create a problem of individuals misusing their platform and their influence. Influencers need to be held responsible for the messages they put out on social media, especially with the large audience they have, also come large responsibilities. Influencers should know that the messages they put out can affect for example the audience’s beliefs and actions. Additionally, the influencers should keep in mind that they might have a younger audience as well following them, who are more impressionable and can be more easily persuaded (Ruder, 2008). However, due to that

23 the researchers cannot control who obtains the knowledge within this research, this could create a problem of people misusing this knowledge in order to convey messages that are harmful for society and the world.

24 5. Results

This chapter presents the results gained from the questionnaire together with the data from SPSS.

5.1 Demographics - Covariance questions The number of respondents in this study was 243, however 25 respondents did not fill the criteria and did not fit within the sampling frame and were removed from the results. The valid number of responses thus was 218. 73.4 % (160) of the respondents were female, 26.1% (57) was male and 0.5 % (1) was non-binary.

Table 2: Gender.

Gender Percentage

Female 73.4 %

Male 26.1 %

Non-binary 0.5 %

Out of the 218 respondents, 14.7 % (32) was in the age range of 15-19, 58.3 % (127) was in the range of 20-24, 17.9 % (39) in the range of 25-29, 3.7 % (8) in the range of 30-35 and 5.5% (12) of the respondents were 36 and above.

Table 3: Age.

Age Percentage

15-19 14.7 %

20-24 58.3 %

25-29 17.9 %

30-35 3.7 %

36- above 5.5 %

25 Regarding educational level 34.4 % (75) of the respondents reported their highest educational level to be high school, 2.8 % (6) as vocational college, 6.4 % (14) as university of applied sciences, 44 % (96) as bachelor’s degree, 9.2 % (20) with master’s degree, 1% (2) with doctorate degrees and 2.3 % (5) of the respondents did not wish to disclose.

Table 4: Educational level

Educational level Percentage

High School 34.4 %

Vocational College 2.8 %

University of applied sciences 6.4 %

Bachelor Degree 44 %

Master’s Degree 9.2 %

Doctorate Degree 1 %

I do not wish to say 2.3 %

Out of 218 respondents 64.7 % (141) lived in Nordic countries and 35.3 % (77) lived outside Nordic countries.

Table 5: Country

Country Percentage

Nordic 64.7 %

Other 35.3%

5.2 Descriptive Statistics Descriptive statistic was the first step done in order to gain an overall view of the data gathered. The table below shows each variables’ Mean, Median, Mode, Standard Deviation, Skewness and Kurtosis. Since the questionnaire used Likert-scale from 1 to 5, the minimum and maximum numbers of the

26 data are 1 and 5. Then the central tendency, which are mean median and mode (Saunders et al., 2009) and dispersion, which is the standard deviation (Bryman and Bell (2011), was run. The highest mean of 4.0413 was for Expert4 and followed by MCred7 with mean of 4.0229. The lowest was MCred2 2.1927 followed by Attract4 with the mean of 2.8440. For the median, most of the items had 3 or 4, except for one item which was MCred2 with the median of 2. When examining the mode of the items MCred2 and Trust4 were the outliers, which had a mode of 1 and 2 respectively, otherwise the other items had the mode of 3, 4 and 5. For the dispersion, the highest standard deviation was Trust4 with the value of 1.26577 and the lowest value on the standard deviation was on Expert4 with a value of 0.89192.

As for the skewness, all of the items had a value between ±1.0, which is considered excellent, except for one item Attract5 which indicates a value of -1.073. However, value between ±2.0 acceptable, therefore it was included in further analysis. Furthermore, the value for kurtosis for all of the items was considered acceptable, since they indicated a value between ±2.0. The highest number for kurtosis was for Attract5 with the value of 1.463 and the lowest with a number of -1.111 was Trust4. Additionally, the value which where closest to zero was Mcred5 with a value of 0.024.

Table 6: Descriptive statistics for the independent and dependent variables.

Item Mean Median Mode Std. deviation Skewness Kurtosis

Trust1 3.5275 4 4 1.11615 -0.310 -0.812

Trust2 3.3761 3 4 1.09703 -0.262 -0.616

Trust3 3.3349 3 4 1.23795 -0.189 -1.069

Trust4 3.0780 3 2 1.26577 0.072 -1.111

Trust5 3.3303 3 4 1.18014 -0.173 -0.889

Trust6 3.1789 3 3 1.12373 0.035 -0.889

Expert1 3.4358 4 4 1.12697 -0.316 -0.690

Expert2 3.2752 3 4 1.18650 -0.130 -1.021

Expert3 3.8211 4 5 1.11136 -0.738 -0.150

27 Expert4 4.0413 4 4 0.89192 -0.711 0.171

Attract1 3.9587 4 5 1.07470 -0.794 -0.196

Attract2 2.9817 3 4 1.25893 -0.063 -1.063

Attract3 2.9083 3 3 1.22975 -0.004 -1.002

Attract4 2.8440 3 3 1.22403 0.042 -1.034

Attract5 3.8073 4 4 0.90682 -1.073 1.463

MCred1 3.3578 3 4 1.16019 -0.267 -0.772

MCred2 2.1927 2 1 1.17561 0.667 -0.598

MCred3 3.7064 4 4 0.98189 -0.768 -0.479

MCred4 3.6101 4 4 1.08585 -0.439 -0.485

MCred5 3.6560 4 4 1.10131 -0.726 0.024

MCred6 3.9862 4 5 1.03166 -0.938 0.381

MCred7 4.0229 4 5 0.98581 -0.745 -0.110

5.3 Reliability and Cronbach’s Alpha Cronbach’s Alpha for both independent and dependent variables; trustworthiness, expertise, and message credibility are reliable since they all surpass 0.7 limit. Trustworthiness is considered as an excellent reliable value, 0.942. Expertise is considered a fairly high reliable value, 0.796. Message credibility is considered fairly high as well, 0.764. Attractiveness, however, indicated Cronbach’s Alpha of 0.666, however the value can be considered as adequate.

28 Table 7: Cronbach’s Alpha

Variable Cronbach’s Alpha

Trustworthiness 0.942

Expertise 0.796

Attractiveness 0.666

Message credibility 0.764

5.4 Validity and Correlation Coefficient The dependent variable message credibility shows value of 0.668 with trustworthiness, 0.706 with expertise and value of 0.474 with attractiveness. Trustworthiness shows value of 0.730 with expertise, 0.341 with attractiveness. Furthermore, expertise displays the value of 0.405 with attractiveness. The Pearson’s correlation indicates that the values are accepted since they all show a value smaller than 1.0.

Table 8: Pearson’s correlation

Trustworthiness Expertise Attractiveness Message Credibility

Trustworthiness 1 0.730** 0.341** 0.668**

Expertise 0.730** 1 0.405** 0.706**

Attractiveness 0.341** 0.405** 1 0.474**

**Correlation is significant at the 0.01 level (2-tailed).

5.5 Hypothesis Testing The final stage in SPSS was to test hypotheses which were generated for this study. This was done with multiple linear regression analysis.

Multiple linear regression analysis in SPSS was conducted in order to see if the hypotheses are accepted or rejected. Model 1 shows the dependent variable, message credibility, in connection to gender, age, education and country. Trustworthiness is added in Model 2, expertise in Model 3,

29 attractiveness in Model 4 and Model 5 tests all the independent variables and age, gender, education and country (control variables) regarding message credibility. The table presents unstandardized β for the Constant, coefficient standard error which is expressed in the brackets. The last category of variables that were tested contained variables such as R2, Adjusted R2, Standard Error of Estimates, F-value and Degree of Freedom.

Table 9: Multiple and simple Linear Regression

Model 1 Model 2 Model 3 Model 4 Model 5 Control All

Intercept 3.732 1.996 1.602 2.018 (0.309) 0.962*** (0.247) (0.231) (0.229) (0.242)

Gender -0.058 0.026 -0.024 0.023 0.036 (0.105) (0.080) (0.074) (0.094) (0.070)

Age 0.129 0.052 0.067 0.128* 0.059 (0.054) (0.041) (0.038) (0.047) (0.035)

Education -0.124 -0.039 -0.103* -0.090 -0.059 (0.031) (0.024) (0.022) (0.028) (0.021)

Country -0.086 -0.023 -0.045 -0.059 -0.022 (0.100) (0.075) (0.071) (0.088) (0.065)

Trustworthiness

H1+ - 0.661 - - 0.296*** (0.035) (0.044)

Expertise

H2+ - 0.698 - 0.399*** (0.039) (0.054)

Attractiveness

30 H3+ - - - 0.473 0.214*** (0.056) (0.045)

R2 0.031 0.451 0.512 0.248 0.592

Adjusted R2 0.013 0.438 0.500 0.230 0.578

F-Value 1.690 34.793*** 44.456*** 13.953*** 43.498***

Std. Error of the Estimates 0.68822 0.51931 0.48957 0.60779 0.44981

Degree of Freedom 4 5 5 5 7

* p< 0.05, ** p<0.01, *** p<0.001, N=218 S.E. (standard error) is presented in the brackets.

Table 10: Hypotheses results.

Variables Significant P-Value

Trustworthiness Significant P = 0.000

Expertise Significant P = 0.000

Attractiveness Significant P = 0.000

31 6. Discussion

In this chapter the findings from the research and the outcomes of the hypotheses will be discussed. The chapter will begin with an overall discussion of the model, moving on to discuss each hypothesis.

6.1 Discussion of the model When the dependent variable Message Credibility was tested with all of the independent variables, i.e multiple regression analysis, and the demographic variables one can understand by examining table 10, that the independent variables all had a significant value of 0.000. Meaning that all independent variables had a positive impact on the dependent, confirming the proposed model and approving the hypotheses. However it can be understood that the different dimensions within each category (such as trustworthiness) had a different impact and importance, which will be discussed in more detail in following subchapters. When examining the descriptives statistics of the dependent variable on can see that the variable have to some extent a normal distribution, except for the MCred2, which could be seen as an outlier. This indicates that the sample did not consider the believability to be of importance within message credibility, which is directly contradictory towards of what O’Keefe (2002) suggest. This is further strengthened by the results of median and mode, which is 2 and 1 respectively. However, the dependent variable does still measure what it is suppose to, since the value of Cronbach’s alpha is 0.764. Which is according to Cortina (1993) is sufficient reliability and according to Taber (2016, pp. 1278) is considered fairly high. Furthermore, by examining the Pearson’s R one can understand that the variance between all of the variables are within the tolerated level (O’Brien and Scott, 2012), meaning that all of the variables measures different things and is answered differently, which is good. Moreover, when all of the demographic variables, gender, age, education and country, were measured alongside the independent variables against the dependent variable i.e multiple regression analysis, the demographics did not have a significant value under 0.05, meaning that demographics had no impact on the dependent variable when measured together. In addition, when analyzing the result of the combined regression analysis, i.e model 5, 59.2% of message credibility is explained by the independent variables, since the R² is 0.592. However, the adjusted R² expressed a value of 0.578, meaning that 57.8 % of message credibility could be measured by the variables that appeared significant, i.e variables that actually affected the dependent variable. Which in this case means that the demographics lower the value of the other variables. This can be further supported by the F-value, since the F-value shows a very low value on Model 1 and on all of the other models the F-value is high and significant.

32

6.2 Hypothesis 1: Trustworthiness This study investigated trustworthiness and its attributes such as authenticity, dependable, genuineness, honesty, believability, reliability as something that an influencer should possess in order to make the message credible. As already mentioned, the hypothesis was approved, meaning that the hypothesis affect the message credibility in a positive and sufficient way. This can be supported by the β-value, due to the value being 0.296 and is significant, indicating that trustworthiness has a positive effect on message credibility. Meaning that authentic, dependable, genuine, honesty, believable and reliability have an effect on message credibility and thereby generally confirming the theories Chen (2004), Banks (2008), Cheung et al., (2009), Erdogan (1999), Copeland, Gunawan and Hernandez (2011), Ohanian (1990), Morgan and Hunt (1994). This can further be strengthened by Cronbach Alpha, since the value was considered an excellent reliable value of 0.942 (Taber, 2016, pp. 1278). This almost expresses perfect reliability of the variable, meaning that the measure indicates what it suppose to (Bryman & Bell, 2011).

Moreover, when examining the descriptive statistics for the trustworthiness and its indicators on its own, some differences can be seen. The trustworthiness and the indicators expressed similar means with a value in between 3 and 4. This means that when investigating the central tendency Trust1, Trust2, Trust3, Trust4, Trust5 and Trust6, which according to Fisher and Marshall (2009) shows a normal distribution of the variables. However, some minor differences can be found between the values. Variables such as Trust1 which had a higher value, indicating that the sample considered influencer’s authenticity as the most important attribute within trustworthiness, confirming studies by Chen (2004) and Banks (2008). This is further supported by the higher value in the median, Trust1 is the only item possessing a value of 4, which indicates that the item is more important than the items within trustworthiness. Meanwhile the variable that have the least value on the mean was Trust4 displaying that the sample considered influencer’s honesty least important. This is contradictory to what Copeland, Gunawan and Hernandez (2011) and Erdogan (1999) expresses, since the result of this study does not express that honesty is an essential dimension within trustworthiness. This is further backed up by the results on the mode, where the results convey that Trust4 is the only variable that possess a value of 2, which in comparison to the other variables is considered low.

33 6.3 Hypothesis 2: Expertise The second hypothesis examined in this thesis is expertise and its attributes validity, knowledge, experience and reputation. As previously mentioned, the hypothesis was approved, confirming that the expertise has a positive effect on message credibility. This can be confirmed by the β-value, which expresses that expertise has the highest value of 0.399 and is significant. This means that expertise has the most significant positive effect on message credibility. Therefore, all of the attributes (validity, knowledge, experience and reputation) were found to have an effect on message credibility, generally confirming the theories by McCracken (1989), Erdogan, (1999), Tedeschi, Schlenker, Bonoma (1973), Gilly, Graham, Wolfinbarger and Yale (1998), Martensen et al, (2018), Zhao, Leo, He, Lin & Wen, (2016). This claim can be enhanced by the Cronbach alpha, since the Cronbach alpha was considered fairly high, since the value is 0.796 (Taber, 2016, pp. 1278). Which according to Cortina (1993) is considered sufficient enough and thereby it is indicating what it suppose to.

Proceeding to the descriptive statistics of the independent variable expertise of which the results indicate that the expertise does have a tendency to have a normal distribution, because of the small differences in the mean, median and mode (Fisher and Marshall, 2009). However, our results conveyed that the variable that had the highest mean was Expert4 (4.0413), which had a slight advantage over the others. This result expresses that the sample values the influencer’s reputation the most, indicating that reputation has an important role when determining an influencer’s expertise and thereby the result support the suggested findings by Zhao, Leo, He, Lin & Wen, (2016). However, the median and the mode have a similar value likewise the other variables, which indicates that the variables are closer to each other regarding to the importance like the mean is suggesting. The variables Expert1/2/3 were all quite similar in their mean values, indicating that all of the variables have similar importance by the sample. Which in turn means that the findings of McCracken 1989, Erdogan, 1999, Tedeschi, Schlenker, Bonoma (1973) and Gilly, Graham, Wolfinbarger and Yale (1998), Martensen et al., (2018) is not as perhaps important as they suggest, since the mean values are lower. Moreover, when examining the independent variable on its own, i.e simple regression analysis, when expertise and all of the demographics were measured against the dependent variable (Model 3), only education was found to be significant. Meaning that the level of education the respondents possessed could sufficiently explain some of the relationship between expertise and message credibility.

34 Hypothesis 3: Attractiveness Lastly, this study investigated independent variable attractiveness and its attributes physical appearance, similarity, familiarity and likeability. This hypothesis was also approved, meaning that attractiveness has a positive effect on message credibility. This can be reinforced by the β-value, due to that the value is significant and is of a positive nature (0.214). This value is indicating that attractiveness has an effect on message credibility, however it displays that it has the lowest effect out the three independent variables. Since the hypothesis was approved the attributes found to have some effect on the message credibility, generally confirming the theories; De Veirman, Cauberghe, & Hudders, 2017, Lou & Yuan, 2019, Martensen et al., 2018, McGuire 1985; Ohanian, 1991, however all of them are perhaps not as important as they suggest. The strength of this relationship could perhaps be explained by the Cronbach alpha. The value of Cronbach alpha is 0.666, which displays that this measure is perhaps not indicating what it suppose to (Bryman & Bell, 2011), or at least not as strongly as the other variables. However, Taber (2016, pp. 1278) refers the value as adequate and could be seen as reliable.

The descriptive statistics of the independent variable attractiveness expresses a clear difference between Attract1/5 and Attract2/3/4. Attract1, physical appearance, is considered the most important by the sample, since it has the highest value of the mean, 3.9587. This indicates that physical appearance is considered as the most important and the results confirm the findings by Eisend and Langner 2010; Till and Busler 2000, cited in De Veirman, Cauberghe, & Hudders, (2017). This can be further supported with the median and the mode, which are 4 and 5 respectively, which are high compared to Attract2/3/4. Moreover, the results of Attract5, shows that likeability is also considered important, since the difference between the Attrack5 and Attrack1 are marginable. The mean of Attract5 is 3.8073 which is very close to Attract1 and the median is likewise 4. However, there is a difference in the mode. Attract5 has a mode of 4 instead of 5, which was the mode for Attract1. This indicates that there is difference in the importance of the values, however 4 is still considered a high value. This confirms that other factors other than just physical appearance is also of importance for the sample (McGuire 1985, Ohanian, 1991, Nunes, Ferreira, De Freitas & Ramos, 2018). The results reveal to some extent that it is only physical appearance and how likeable the influencer is that express one’s attractiveness, since the values of Attract2/3/4, have a low mean, median and mode when compared to Attract1 and Attract5. The lowest mean is as low as 2.8440, which is under 3.0, the half point of the 1-5 likert scale. Therefore, the values of familiarity and similarity showed no importance for the sample. This is to some extent contradictory towards the findings that McGuire 1985, Ohanian (1991), Nunes, Ferreira, De Freitas & Ramos (2018), Lou & Yuan (2019), Martensen et al., (2018)

35 suggested. Additionally, the results of the simple regression analysis, which examines attractiveness on its own with the demographics against the dependent variable (Model4), expressed that participants age had an impact on the relationship between the attractiveness and message credibility, since age was the only dimensions found significant.

36 7. Conclusion

This chapter presents the conclusion and the accepted model of this study.

When investigating the results of this research it was apparent that all of the hypotheses were approved. Meaning that all of the characteristics within the source credibility model are proven to have a significant effect on the message credibility. This can be strengthened by that 57.8 % of the significant independent variables could explain the dependent variance, which is a fairly high number. Thus the model is actually describing the effect on message credibility. Moreover, when examining the β-value one can understand that the different characteristics had different levels of effect on the dependent variable. The β-value conveys that the source expertise has the largest effect on message credibility, meanwhile the attractiveness of the source was found to have the least effect. However, since all of the hypotheses were approved, social media influencers should aim to focus on all of the characteristics if they wish to heighten their message credibility.

7.1 Accepted Model Figure 2. Accepted model

37 8. Research Implications

This chapter presents theoretical implication and managerial implications.

8.1 Theoretical implications The research literature is thoroughly researched, however any relationship or any effects on either concepts lacked knowledge, therefore the aim of this research was to fulfill previous gaps within the field of credibility. This research has accomplished the aim of filling that gap, since all of the hypotheses were approved, proving that the tested model applies. Therefore, the research has given useful understanding of which characteristics an influencer should focus on in order to convey credible messages and that source credibility characteristics have an effect on message credibility. Additionally, it was proven that source credibility can explain as much as 57.8 % of message credibility, confirming yet again that the constructed model within this study has a contribution to the research field. Furthermore, the theoretical findings of this research can be utilized in order to generate new knowledge or theories in the field of credibility. Because of the rapid growth of the social media world, therefore this research could contribute with new theoretical findings in the future.

8.2 Managerial implications When it comes to managerial implications of this study, influencers can consider making use of the findings to strengthen their message credibility on social media. Based on the findings, influencers should consider heightening their perceived expertise on social media, since it was shown to have the most effect on message credibility, however all of the characteristics should be noted. Within source expertise, reputation and education are considered most important. As for source trustworthiness, the focus should be put on being believable and as for attractiveness, physical appearance and likeability would improve message credibility. However, those factors can be hard to change, since physical appearance is solely subjective, and as for likeability, the influencers simply cannot please everyone. Concluding, by focusing on the implications, they can help the influencers make their messages on social media more credible and gain stronger social media presence.

38 9. Limitations and Future Research

In this final chapter limitations regarding the study will be presented, that could have potentially affected the outcome of the study. Additionally, this chapter will provide suggestions for future research.

9.1 Limitations It should not come as a surprise that as for most studies, some limitations occur that should be discussed. The limitation of this study is concerning the data collection, a non-probability convenience sample method was used due to time and resource restrictions. Using convenience sampling resulted in biased findings, since most of the respondents were the same gender and age as the researchers; 73.4 % were female and 58.3 % of all the respondents were in the age group of 20- 24. The problem with this type of sampling method is that it is a weak representation of the population since not everybody had an equal chance of being included in the sample, therefore the results are not as generalizable.

9.2 Future Research Based on the results of this study, the researchers are able to give some suggestions for future research. Mainly, the researchers would suggest that more dimensions within the two concepts to be added in order to give more breadth. If more dimensions were to be added it would be interesting to see how the results would be impacted and how the outcome could possibly be different. Especially adding dimensions on attractiveness, since attractiveness was found to have the least importance. One could perhaps find that other dimensions within this variable have an effect, thus perhaps changing the difference in importance. Therefore it should be further researched and tested in the future.

As for the sampling method, further research should be conducted by using another sampling technique, for instance a probability sampling. Since the sample is selected by using a random selection, would give everybody in the population an equal chance of getting selected. This would result in a more diverse sample and the results would be generalizable. Therefore, a suggestion for future research could be to utilize a more generalizable sampling method in the same context as in this study and see if the outcome is changed.

39 Reference list:

Appelman, A. & Sundar, S. S, 2016, Measuring Message Credibility: Construction and Validation of an Exclusive Scale. Journalism & Mass Communication Quarterly, 93(1), 59–79.

Adams Communications Consulting, 2008. Communicating Credibility. [online] Available at: [Accessed 20 May 2019]

Banks, M., 2008. Blogging Heroes: Interviews with 30 of the World's Top Bloggers. Indianapolis: Wiley Publishing, Inc.

Burgoon, J. K., Floyd, K. & Guerrero, L. K., 2010, Nonverbal communication theories of interpersonal adaptation. Boston: Allyn & Bacon.

Bryman, A. & Bell, E., 2011. Business Research Methods. 3rd ed. New York: Oxford University Press Inc.

Budzowski, B., 2019. What is Credibility and Why Do You NEED to Care. InCredibleMessage. [online], available at: [Accessed 9 March 2019]

Burbules, N.C., 1998. of the Web: Hyperreading and critical literacy. In I. Snyder (Ed.), Page to screen: Taking literacy into the electronic era, 102 - 122. London: Routledge.

Cambridge dictionary, 2019. Meaning of valid in English. [online] Available at:

Castellow, W. A., Wuensch, K. L. & Moore, C. H., 1990 Effects of physical attractiveness of the plaintiff and defendant in sexual harassment judgements. Journal of social behaviour and Personality, 5(6), 547-562.

40 Chen, X., 2004. Being and Authenticity. Amsterdam: Editions Rodopi B.V.

Cheung, M.Y., Luo, C., Sia, C.L. & Chen, H., 2009. Credibility of Electronic Word-of- Mouth: Informational and normative determinants of online consumer recommendations. International Journal of Electronic Commerce, 13(4), 9-38.

Copeland, D., Gunawan, E. & Bies-Hernandez, K., 2011. Source credibility and syllogistic reasoning. Memory & Cognition, 39(1), 117–127.

Cortina, J. M., 1993. What is coefficient alpha? An examination of theory and applications. Journal of Applied Psychology, 78(1), 98–104.

Crisci, R. & Kassinove, H., 1973. Effect of Perceived Expertise, Strength of Advice, and Environmental Setting on Parental Compliance, The Journal of Social Psychology, 89(2), 245-250.

De Veirman, M, Cauberghe, V. & Hudders, L., 2017, “Marketing through Instagram Influencers: The Impact of Number of Followers and Product Divergence on Brand Attitude,” International Journal of Advertising, 36 (5), 798–28.

Del Vicario, M., Bessi, A., Zollo, F., Petroni, F., Scala, A., Caldarelli, Stanley, H. E. & Quattrociocci W., 2016. PNAS, 113 (3), 554-559.

Eisend M. & Langner, T., 2010, “Immediate and delayed advertising effects of celebrity endorsers’ attractiveness and expertise”, International Journal of Advertising, 29(4), 527-546.

Erdogan, Z.B., 1999. Celebrity Endorsement: A Literature Review, Journal of Marketing Management, 15(4), 291-314.

European Commission, 2018. Fake news and online disinformation. [online] Available at: [Accessed 9 May 2019]

41 European Union Agency for Fundamental Rights, 2019. Child participation in research. [online], available at: [Accessed 8 May 2019]

Fisher, M.J. & Marshall, A.P., 2009. Understanding descriptive statistics. Australian Critical Care, 22(2), 93–97.

George, D. & Mallery, P., 2003. SPSS for Windows step by step: a simple guide and reference 11.0 update. 4. ed., Boston: Allyn and Bacon.

Gilly, M.C., Graham, J.L., Wolfinbarger, M.F. & Yale, L.J., 1998. “A dyadic study of interpersonal information search”, Journal of the Academy of Marketing Science, 26(2), 83-100.

Hair, J. F., Black, W. C., Babin, B. J. & Anderson, R. E., 2014. Multivariate Data Analysis: Pearson New International Edition, vol Seventh edition, Always Learning, Pearson, Harlow, Essex, [online] Available at: [Accessed 12 May 2019]

Harkins, S. G. & Petty, R. E. 1987. Information utility and the multiple source effect. Journal of Personality and Social Psychology, 52(2), 260-268.

Hennig, C. & Cooper, D., 2011. Brief communication: the relation between standard error of the estimate and sample size of histomorphometric aging methods. American journal of physical anthropology, 145(4), pp.658–64.

Influencer Marketing Hub, 2019. What is an influencer?, [online], available at: https://influencermarketinghub.com/what-is-an-influencer/ [Accessed 21 May 2019]

Ingham-Broomfield, R., 2014. A nurses' guide to quantitative research. Australian Journal of Advanced Nursing, 32(2), 32.

42

Kahle, L. R. & Homer P. M. 1985. Physical Attractiveness of the Celebrity Endorser: A Social Adaptation Perspective. Journal of Consumer Research, 11, 954-961.

Khamis, S., Ang, L. & Welling R., 2017. Self-branding, ‘micro-celebrity’ and the rise of Social Media., Influencers. Celebrity Studies, 8 (2), 191-208.

Kutthakaphan, R., & Chokesamritpol, W., 2013. The use of celebrity endorsement with the help of electronic communication channel (Instagram): Case study of Magnum ice-cream in Thailand. Malardalen University, School of Business, Society and Engineering.

Li, R. & Suh, A., 2015. Factors Influencing Information credibility on Social Media Platforms: Evidence from Facebook Pages. Procedia Computer Science, 72(C), pp.314–328.

Lin, H-S., Bruning, P. F. & Swarna, H., 2018. "Using online opinion leaders to promote the hedonic and utilitarian value of products and services," Business Horizons, Elsevier, 61(3), 431-442.

Ledingham, J. A., 2003. Explicating Relationship Management as a General Theory of Public Relations, Journal of Public Relations Research, 15(2), 181-198.

Lou, C., & Yuan, S., 2019. Influencer Marketing: How Message Value and Credibility Affect Consumer Trust of Branded Content on Social Media, Journal of Interactive Advertising, 19(1), 58- 73.

Malhotra, N.K., 2011. Marketing research : an applied orientation 6th ed., Upper Saddle River, N.J. ; London: Pearson Education.

Martensen A., Brockenhuus-Schack S. & Zahid. A, 2018. How citizen influencers persuade their followers, Journal of Fashion Marketing and Management: An International Journal, 22(3), 335- 353.

Metzger, M. J, Flanagin, A. J., Keren Eyal, Daisy R. Lemus & Robert M. Mccann., 2003. Credibility for the 21st Century: Integrating Perspectives on Source, Message, and Media Credibility in the

43 Contemporary Media Environment. Annals of the International Communication Association, 27(1), 293-335.

McGinnies, E. & Ward, C. D., 1980. Better Liked than Right: Trustworthiness and Expertise as Factors in Credibility, Personality and Social Psychology Bulletin, 6(3), 467–472

McGuire, W.J., 1985. Attitudes and Attitude Change. In: Lindzey, G. and Aronson, E., Eds., Handbook of Social Psychology, 3rd Edition, Vol. 2, Random House, New York, 233-346.

McCracken, G., 1989, Who is the celebrity endorser? Cultural foundations of the endorsement process. The Journal of Consumer Research, 16(3), 310-321.

McQuarrie, E., Miller, J. & Phillips, B., 2013. The megaphone effect taste an audience in fashion blogging. Journal of consumer research, 40(1), 136–158.

Morgan, B.L. & Van Voorhis, W, C.R., 2007. Understanding Power and Rules of Thumb for Determining Sample Sizes. Tutorials in Quantitative Methods for Psychology, 3(2), 43–50.

Morgan, R., & Hunt, S., 1994. The Commitment-Trust Theory of Relationship Marketing. Journal of Marketing, 58(3), 20-38.

Newell & Goldsmith, 2001. The development of a scale to measure perceived corporate credibility. Journal of Business Research, 52(3), 235–247.

Nolan, S. A. & Heinzen, T. E. 2014, Statistics for the Behavioral Sciences, 3rd ed. New York: Worth Publishers.

Nunes, R.H., Ferreira, J.B., De Freitas, A.S. & Ramos, F.L., 2018. The effects of social media opinion leaders' recommendations on followers' intention to buy. Brazilian Journal of Business Management, 20(1), 57–73.

O’Brien, D., & Sharkey Scott, P., 2012, Approaches to Quantitative Research – A Guide for Dissertation Students, Ed, Chen, H, Oak Tree Press.

44

Pornpitakpan, C., 2004. The persuasiveness of source credibility: A critical review of five decades’ evidence. Journal of Applied Social Psychology, 34, 243–281

Ohanian, R., 1991. The impact of celebrity spokespersons' perceived image on consumers' intention to purchase. Journal of Advertising Research, 31(1), 46-54.

Ohanian, R., 1990. Construction and Validation of a Scale to Measure Celebrity Endorsers' Perceived Expertise, Trustworthiness, and Attractiveness. Journal of Advertising, 19(3), 39-52.

O’Keefe, D. J., 2002. Persuasion: Theory and research. 2nd ed. Thousand Oaks, CA: Sage

Oregon State University, 2019. What is the Institutional Review Board (IRB)? [online], available at: [Accessed 4 May 2019]

Roberts, C., 2010. Correlations Among Variables in Message and Messenger Credibility Scales. American Behavioral Scientist, 54(1), 43–56.

Ruder, D. B, 2008. The teen brain. Harvard Magazine, [online], Available at: [Accessed 19 May 2019]

Saunders, M., Lewis, P. & Thornhill, A., 2009. Research Methods for Business Students. Fifth Edition. Prentice Hall.

Sisco, H. F. & Mccorkindale, T., 2013. Communicating “Pink”: An Analysis of the Communication Strategies, Transparency, and Credibility of Breast Cancer Social Media Sites.

Strømsø, H. I., Bråten, I., Britt, M. A., & Ferguson, L. E., 2013. Spontaneous sourcing among students reading multiple documents. Cognition and Instruction, 31(2), 176–203.

45 Taber, K. S., 2016. The Use of Cronbach’s Alpha When Developing and Reporting Research Instruments in Science Education. Research in Science Education, 48(6), 1273–1296.

Tedeschi, J. T., Schlenker, B. R. & Bonoma T. V, 1973. Conflict, power and games : the experimental study of interpersonal relation. Chicago: Aldine.

The Pennsylvania State University, 2018. What is Simple Linear Regression? [online], available at: [Accessed 26 May 2019]

Tuten, T., & Solomon, M., 2018. Social media marketing. 3rd ed. London:Sage.

Varsamis, E., 2018. “Are Social Media Influencers the Next-Generation Brand Ambassadors?,” Forbes, [online], available at: [Accessed 23 March 2019]

Wallis, L., 2013. Is 25 the new cut-off point for adulthood?, BBC, [online], available at: [Accessed 8 May 2019]

West, T., 2019. Credibility, Trust and Authenticity in Influencer Marketing, Scrunch, [online] Available at [Accessed 20 2019]

Zhao, WX., Liu, J., He, Y., Lin, CY. & Wen, J-R., 2016. A computational approach to measuring the correlation between expertise and social media influence for celebrities on microblogs. World Wide Web, 19(5), 865-886.

46 Appendix

Questionnaire as presented to the respondents.

47 If the respondents answered “No” to one of the control questions they were exited from the questionnaire, with the following message.

48 When the respondent passed through the control questions by answering “Yes” they were allowed to continue with the questionnaire.

49

50

51

52

53 54

55

56