The power of the number

(1)

University of Amsterdam, Amsterdam, The Netherlands Master in Business Administration – Marketing Track

MASTER THESIS

“The power of the number”

Author: Ms R.A. van Baarsen MSc Student Number: 10908420 Thesis Supervisor: Mr dr. F.H. Mattison Thompson Date: 23rd of June, 2017

(2)

Statement of originality

This document is written by Romy van Baarsen, who declares to take full responsibility for the content of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the content.

(3)

Acknowledgements

This Master thesis is the final assignment for my MSc Business Administration – Marketing at the University of Amsterdam. Accomplishing this theses has been a rollercoaster, with a lot of ups and some downs. Completing my study, including this thesis would not have been possible without the help of other people. I would like to thank them in this section.

First of all Dr. Mattison Thompson from the University of Amsterdam, for all her guidance and patience. When I had my down, she stayed calm and this gave me confidence that everything would work out at the end.

Moreover, I would like to express my gratitude to Janneke Strebus and Pepijn Ernst two colleague student who helped and supported me in the process of writing this thesis. I could not have done this without our discussions, lunch and dinner breaks.

Accordingly my parents who are always there for me and who I could always call. One time when I was alone in the library because my friends had already left I had to eat diner alone. I called them and they put me on speakers, so I could join their dinner conversations over the phone, as if I was there.

Finally I should also thank all the participants in the experiment for collectively providing me with the data I needed to empirically test my research question. The survey was quite long and still a lot of people took the time to fill it in.

I learned a lot from writing this thesis. Since I absolutely love this topic I didn’t mind reading and writing about it. Sometimes I couldn’t stop reading which gave a bit of an information overload at times, but I do feel it made me an expert on this topic and gives me the confidence to say that I did a good research.

Best regards,

(4)

Abstract

Despite the increasing popularity of the social network site Instagram, no study to date has investigated the influence of number of Instagram followers on self-evaluation. An online experiment (N= 552) was conducted to examine this. The picture-oriented focus of Instagram creates new opportunities for social comparison. This establishes new changes for businesses to create brand awareness and engagement through Instagram influencers that will reach a wide consumer network, and thus have a high number of followers. Therefor it is important to examine whether observing either an Instagram profile with a high number, low or no number of followers shown would result into reporting different levels of self-evaluation. Two dimensions of self-evaluation were analyzed: a broad and a specific measure. Neither of these two dimensions showed significant differences between the three conditions. Not finding differences between the control group and both experimental conditions are a contribution to existing social comparison literature on social media (Utz, 2010; Vogel et al., 2014; Jin & Phua, 2014; Roble, Albrecht, LaCount & Hanks, 2015), which only analyzed differences between two experimental conditions, whereby no comparisons with a neutral reference point could be made. The number of followers was expected to create a halo effect on how the participants would evaluate the Instagram user. The evaluations differed significantly between the low number and the control condition and also between the low number and the high number condition. No discrepancy between the high number and control condition was found. Also no relation was found between the overall evaluation of the target and the participants’ self-esteem. Therefore it can be concluded that this overall evaluation of the target did not function as a mediator.

(7)

1) Introduction

“Apart from economic payoffs, social status seems to be the most important incentive and motivating force of social behavior”- John Harsanyi (Nobel laureate economist).

Last year a Dutch campaign ‘The Good Life Agency’ gave people the opportunity to buy a better image on social media. “We will get you an image to make others jealous”- visitors of the website could choose from different options ranging from “the better you Photoshop service” to packages that offered more followers and services that made celebrities appear to be your friend online or offline. Prices ranged from 10 up to 500 euros. In less than 3 days the website reached 16 million people, 31000 people visited the site and 170 orders where placed. Despite the success, the website was not for real use. It was part of a social experiment to see how far people would go to appear to be successful. The message of the campaign was: “Voor wie doe jij het eigenlijk?” which is Dutch for: “For who are you doing it?” The report showed that 1 out of 5 Dutch people are comparing themselves on a daily basis to others. In the age group between 16-35 years this is 36% and below 25 years it is even 48%.

The reason for this can be retrieved from the combination of global capitalism combined with the rising of communication technologies like social media. This brought significant cultural, economic and political disruption. Individuals use self-branding1 as a way to maintain and affirm personal agency and stay independent within an overwhelming context of uncertainty and change.

People use social comparison to reduce uncertainty and to learn how to define themselves. Human have the tendency to evaluate their own opinions and abilities by comparing themselves to others (Festinger, 1954). Social networking sites (SNSs), provide abundant social comparison opportunities (Vogel, Rose & Roberts, 2014). Likewise, the

1

Self-branding involves individuals developing a distinctive public image for commercial gain and/or cultural capital.

(8)

impact of numbers has grown immensely. Numbers are thrown at us everywhere: the number of visitors of a blog, amount of views on a Youtube video, friends on Facebook or followers on Instagram. A high number equals social proof, since many people show interest in the account.

The number of followers or friends one has on social media seems to be of importance, as people tend to use this to make assumptions of the other. This can trigger a halo effect on how the perceiver evaluates the target. It is useful to discover whether the number of followers or friends on SNSs will have an influence on the evaluation of other attributes because this will be an evidence of the prevalence and appeal of self-branding, which persists through the rise of Social Media Influencers (SMIs) 2.

Individuals learn and understand information by looking at others (Moschis and Churchill, 1978). Social media gives endless possibilities to exploit this, as such more and more brands make use of SMIs to benefit from “the intimate, more ‘trustworthy’ relationships SMIs have ostensibly created” (Gormly, 2016, p.203). Especially Instagram influencers get more power (Veirman, 2016). Instagram is the fastest growing social media platform nowadays (Appendix A). Nevertheless, to the best of my knowledge no existing literature has examined the influence of the number of followers of an Instagram profile on the perceivers’ self-esteem in an experimental setting. Previous social media research on social comparison and the relation with self-esteem mostly look at Facebook and Twitter (e.g. Vogel, Rose, Roberts & Eckles, 2014; Zuo, 2014; Jin & Phua, 2014). On Facebook, social capital is measured in number of friends while on Instagram this is measured by the number of followers. Accounts can have many more followers than followees3. The number of followers on Instagram therefore serves more as social proof than the number of friends on Facebook.

2_{Social Media Influencers are people that have a wide social network which is used by companies and} advertisers for consumer outreach (Hearn and Schonhoff, 2016, p. 194).

(9)

Another thing that distinguishes Instagram from Twitter and Facebook is that Instagram is more visually based. The main objective of Instagram users is to post photos for all their followers to see and to look at photos of other Instagram users. Images attract more attention than written or spoken words (Aubrey, Henson, Hopper, & Smith, 2009) and most individuals describe themselves as visual learners (between 65 and 85 percent)(Vong, 2012). Faber (in McNely, p.1, 2012) calls this image-power: “the organization’s conscious, self-reflective management of public perception and the concomitant shaping of patron and audience identities”. Therefore it is expected that social comparison will be even more likely to happen than on the other SNSs.

To fill the gap in the literature, the current research sought to construe under which conditions the number of observed Instagram followers have an effect on the perceivers’ self-esteem. This will be done by conducting an online experiment in which participants will be randomly assigned to one of three conditions. All participants will see an Instagram account of a same-sex target. The number of Instagram followers is manipulated in each condition. Next to the control condition, which doesn’t show any number of followers, there are two experimental conditions, a high and a low number condition. In the high number condition the number of followers shown will be extremely away from the standard (21.000). The other condition will show a moderate number of followers (210), which will be called the low number condition from now on to express the difference and avoid confusion. These number of followers were manipulated based on actual real Instagram statistics (Appendix B).

The research question of the current study is: How does the number of followers on an Instagram account affect a person’s self-evaluation and is this effect mediated by how this Instagram user is evaluated?

To answer this question in the best of reach, two dimensions of self-evaluation will be taken into account. In the first part of the study, a broad and abstract dimension will be

(10)

analyzed for, measuring the participants’ self-esteem. Differences in the subcomponents (performance, social and appearance) will also be taken into account. The second part of this study will analyze a more specific and concrete dimension, that will be measured by the participants’ aggregated evaluations of specific attributes4.

By adding a control group with no number of followers shown, this thesis differentiates itself from other social comparison research, that usually just compares the differences between two groups (e.g. Mussweiler, Rüter & Epstude 2004; Utz, 2010; Vogel et al., 2014; Jin & Phua, 2014; Roble, Albrecht, LaCount & Hanks, 2015). These researches will be analyzed into further detail in the Conceptual framework.

Investigating this will result in gaining a deeper understanding of the dynamics of the number of Instagram followers, the overall evaluation of the target and both self-evaluations. This is essential for both scholars and practitioners. Although many studies have showed that individuals evaluate social media users with a big social network as more desirable than social media users with a small social network (e.g. Utz, 2010; Antheunis & Schouten, 2011; Jin& Phua 2014) no study to date has examined whether this evaluation about the target 5 will mediate the relation between the social network size and self-evaluation. It is expected that if the observed Instagram account has a lot of followers this will trigger a positive halo effect, which will lead to an overall more positive evaluation of the target. This evokes upward comparison and will result into a lower self-esteem of the perceiver. A low number of Instagram followers on the other hand will trigger a negative halo effect, which will lead to an overall lower evaluation of the target and evokes downward comparison that will lead to a higher self-esteem of the perceiver. Next to self-esteem, also more visible self-evaluation will

4

Specific self-evaluation: Participants will also rate themselves on these same 6 attributes. These aggregated ratings will function as a dependent variable in the second analysis only.

5

Overall target-evaluation: Participants will rate the target on 6 attributes. These aggregated ratings will function as a mediator in both analyses.

(11)

be measured. The same analysis will be done to see if there is an effect on the overall self-evaluation 6 Both analyses will generate new scientific knowledge.

The thesis also contributes to marketing research and management in several ways. If a high amount of Instagram followers creates a positive halo effect this can be of interest for marketers, because a positive halo effect creates a high likability and admiration for this Instagram influencer. Just like with celebrity endorsement this can create a more positive brand image. Furthermore it is useful to examine the effect this has on the self-esteem of the customer because many studies have shown that a low self-esteem will lead to more compulsive, more expensive or more luxury buying (e.g. O’Guinn & Faber, 1989; Yurchisin & Johnson, 2004; Sivanathan & Pettit, 2010).

Academically this research will add insight into the effect of Instagram on self-esteem and more visible self-evaluations. If the results will be significant it will confirm the importance of the social proof theory for Instagram. And it will show something simple, like a number can influence our psychological well-being. Also, the halo effect has not yet been used to study the influence of an number of friends or followers on self-esteem. 2) Conceptual framework

2. Conceptual framework

2.1 Theoretical background 2.1.1 Social comparison

Before getting specific about the number of followers, the broad concept of social comparison and its underlying mechanisms are to be defined and explained. It is important to grasp the wide concept of social comparison, as there are many factors that affect social comparison and many situations in which social comparison occurs. The purpose of social

(12)

comparison is to make sense of our self and to determine our own level of abilities and success. We tend to do this, consciously and unconsciously, to others who we believe are like us (Festinger, 1954). The dimensions that are compared can be subjective or objective. This study will look at the number of Instagram followers, which is an objective measure that is most likely to set a reference point for comparison. But it will also look at more subjective dimensions of social comparison, since participants will be asked to rate the target and themselves.

2.1.1.1 Social comparisons regarding to gender and age

The repeated media images of skinny females and athletic males are setting a tone for standards of attractiveness. The media sends out the belief that women need to be thin and attractive to be accepted in our society. Women magazines contain more of these messages than men’s magazines (Malkin, Wonian, & Chrisler, 1999). Self-comparison with these idealized images seems to create an unreasonable evaluation of attractiveness of the self, leading to the more negative evaluation of self.

Jones (2001) found that weight comparisons happened for both boys and girls to either peers, models or celebrities of the same-sex. Nevertheless most studies on social comparison and the media measure the pressure of media on body image of women (Fallon, 1990; Wolf, 1991; Kilbourne, 1994). Research has shown that women define themselves as higher in relational interdependence than men, while men define themselves as higher in independence than women (Guimond, 2006). As such it is expected that women will attend in more social comparison. The effects of experimental conditions in this study should be stronger for female than for male participants. On the other hand, Roble, Albrecht, LaCount and Hanks (2015) did not find a significant effect of gender on self-esteem in their study on Instagram, perceived attractiveness and self-esteem.

Prior studies on SNSs that reported gender differences also reported the age differences in using SNSs (Muscanell & Guadagno, 2012; Wang, Wang, Gaskin, & Hawk,

(13)

2017). Callan, Kim, and Matthews (2015) conducted an experiment on age differences in social comparison tendency. They found that older adults reported lower levels of social comparison tendency than younger people, which also resulted into lower level of personal relative deprivation. As such results will be controlled by age.

2.1.1.3 Social comparison on Social Media

Besides the standard media, social media also induces social comparison. Websites like Facebook give users the opportunity to create a profile of themselves that can be seen by billions of other people (Fox & Rooney, 2015). On these profiles lots of information are shared, like personal details, interests, friends, and pictures that can be used for social comparison and as mechanisms of evaluating self-worth (Lee, 2014).

Pinterest is an online pin board that is, just like Instagram, entirely driven by visuals. Lewallen & Behm-Morawitz (2016) researched how Pinterest contributed to social comparison as well as intentions to engage in extreme weight-loss behaviors. Individuals that followed more fitness boards on Pinterest were more likely to report intentions to engage in extreme weight-loss behaviors. Also, endorsement of an ideal female body type was positively related to both social comparison and intentions to engage in extreme weight-loss behaviors.

A recent study of Chae (2017) found that exposure to influencers on social media is positively associated with comparison of one’s life to that of influencers. Just like her first part of the study this thesis will test cross-sectionally, which means that comparisons will be made at a single point in time, because social comparison almost automatically occurs (Gilbert, Giesler, & Morris, 1995).

2.1.1.4 Social comparison, self- evaluation and self-esteem

The above findings indicate that social comparison is used not only to make sense of our self, but also to define and evaluate the self (Stapel & Marx, 2007). Most values in life, such as happiness, health and success, are intangible assets. There are no unbiased and

(14)

reliable measures to evaluate how successful you are. In the absence of objectiveness for self-evaluation, we compare ourselves to others to find out how we are doing. Social comparison can affect self-evaluations in a positive or negative way.

Self-evaluation is a very broad concept. The core self-evaluations measure (Judge, Bono & Thoresen, 2003) for example exists of a lot of dimensions like: Generalized Self-Efficiency, Locus of Control, Neuroticism and self-esteem. The first part of this study will look specifically at self-esteem, since this is one of the most fundamental core dimensions of self-evaluation, as it measures a person’s overall subjective emotional self-evaluation of his or her own worth (Judge, Locke, Durham & Kluger, 1988). As well as a judgement of oneself it is an attitude toward the self (Coopersmith, 1967). However self-esteem is the evaluative emotional component of the broader self-concept, it is still rather broad and abstract in its emphasis (Vogel et al., 2014).

Considering that individuals often seem to make social comparison to others who are similar to themselves, specifically in distinguishable dimensions (Festinger, 1954), the second part of this thesis will measure more visible self-evaluations, like fitness, health likeability, popularity, trustworthiness and attractiveness. These aggregated evaluations will be referred to as the specific evaluation. The effect social comparison has on esteem and self-evaluation depends on many factors. This thesis will try to embody as many as possible. 2.1.1.5 Social comparison- the ups and the downs

The social comparison theory proposes two types of social comparison that have a different effect on self-esteem. Upward comparison is when people compare themselves to someone who seems better than themselves and downward comparison happens when someone compares itself to a person that seems not as good as themselves (Schachter, 1959; Stapel & Marx, 2007). The theory assumes that the comparison of an individual can have two effects,

(15)

assimilation and contrast. Assimilation has a positive effect on self-esteem. Contrast and has a negative effect on self-esteem.

2.1.1.5 Contrast and assimilation

According to Festinger (1954) downward social comparison can be a way of self-enhancement which leads to assimilation. The individual will look at a person that is worse off than themselves, in order to feel better about themselves. The opposite happens with upward comparisons, when the person compares himself with someone who is better off or superior. This leads to contrast and has a negative impact on an individual’s self-esteem.

Contrariwise, Aspinall (1977) found that downward comparison can sometimes also cause a lower self-esteem, because it informs the individual of how things could be worse (Aspinwall, 1997). Nonetheless more usually it leads to a higher self-esteem (Wills, 1981). The same accounts for upward comparison. Lockwood and Kunda (1997) found that upward comparison can also be beneficial for self-esteem when it inspires people to become more like their comparison target. However it more regularly makes people feel incapable, upset and have poorer self-esteem (e.g., Marsh& Parker, 1984; Pyszczynski, Greenberg & LaPrelle, 1985; Vogel et al., 2014).

2.2 Theoretical Framework 2.2.1 The power of the number

Social capital is a resource that is created by an individuals’ social relationships (online or offline) can be utilized to accomplish positive social outcomes. Traditional research has shown two types of social networks. Strong ties, which are people close to the individual that provide them with emotional support and weak ties that can give the individual new information which can lead to innovative insights and ideas. The weak tie works as a bridging capital with which different groups of people can be reached. According to Granovetter

(16)

(1973) weak ties are therefore more important than strong ties. Because of SNSs people nowadays have more weak ties than ever before, which create more heterogeneous networks and be as beneficial for the individual as financial capital (Utz, 2015).

Social proof is known as informational social influence. It is a psychological phenomenon where people assumed the actions in attempt for reflect correct behavior for given situations (Kelman, 1958; Aronson, Wilson, & Akert, 2005). The 42nd Street experiment (Milgram, Bickman & Berkowitz, 1969) showed that when a single person stopped in the middle of a street and looked at the sky for a whole minute, others walked past but ignored him. However, when fifteen people did the same, 40% of the people in this same busy street, also paused their way to look up, bringing the entire street to a complete halt within a minute.

Social proof is used in marketing campaigns a lot because it helps to increase confidence at customers. The evidence of social proof comes from displaying social engagement including the number of comments, likes, shares et cetera. A high number of followers or friends on SNSs can also show as social proof, as it implies that many people have showed their interest in a certain account, as they subscribed to its updates. Therefore, the number of friends on SNSs is one of the most frequently used benchmarks to indicate his or her online popularity (Utz, 2010).

Devumi.com, a Social Media Marketing website states that “a user on Twitter with a million followers is perceived as more trustworthy and reputable than a similar user with a thousand followers, resulting in faster growth of followers and higher engagement and click-through-rates.” This statement is supported by Jin & Phua (2014) who found that the number of followers of the celebrity on Twitter influenced the consumer’s perceived credibility of that celebrity. In their study, the celebrity was an endorser for a product. Besides receiving higher ratings on credibility, the celebrity with a high number of followers scored higher on product

(17)

involvement buying intention and the pass along electronic word-of-mouth (eWoM). Their findings are in alignment with the findings of Utz (2010) who found that Hyves7 users with a high number of friends were judged as more popular then Hyves users with a low number of friends.

These articles show that SNSs stimulate the growth of social capital, especially by implementing new communication techniques that allow users to make use of more explicit relationship management strategies, which give the possibilities to maintain more social capital from weak tie (Utz & Muscanell, 2015). Furthermore SNSs have raised the impact of social proof tremendously. Just like on Twitter the number of “followers” one has on Instagram function as an evidence of social proof. Studies have already examined the effects of social proof on Facebook and Twitter, but despites the recent popularity the social media, Instagram has not been researched for that matter.

According to Mussweiler et al. (2004), similarity does not only affect if social comparison will happen (Festinger, 1954), but also influences the direction of comparison consequences to it. Specifically this implies that, the self-evaluation is likely to be more positive if the target and standard are close, while on the other hand self-evaluation seems to be more negative when they are not.

In particular, they investigated whether comparing oneself with an extremity from the standard (dissimilar) would affect self-esteem in a different way than comparing oneself with a moderate standard (similar). In their study professional water polo players were asked to evaluate their own athletic abilities after reading a paragraph about either a reasonable or an extremely good sportsman. The water polo players who were confronted with the extreme sportsman evaluated their own athletic competences as being worse than the water polo players that were confronted with the reasonable good sportsman. Since this experiment

(18)

compared the reported athletic self-evaluations between an extreme condition and a moderate condition, which serves as a benchmark, these results show that when someone is confronted with an extreme standard this will devaluate the evaluation about the self. This was conform to the expectations of the researchers.

Just like the above mentioned study, this thesis will confront the participants with either an extreme standard (high number condition) or a moderate standard (low number condition), that are based on real Instagram statistics (Appendix B). However, this thesis also adds a control condition (no number shown) to the experiment, that will function as a reference point instead. This way, the thesis differentiates itself from all of the other studies described so far. This will add new insights regarding to the theoretical mechanism that may underlie in the power that the number of SNSs followers or friends have on self-esteem.

All of the above mentioned articles only compared two groups and most of these articles measured differences between a low and high condition only (Utz, 2010; Vogel et al., 2014; Jin & Phua, 2014). By comparing a low to a high condition, it is never completely sure which direction the effect is coming from, since there is no neutral benchmark. Mussweiler et al. (2014) also measured differences between two conditions only, but they did compared a high (extreme) condition to a moderate condition, that can act as some sort of reference point, in contrast to a low condition. Nevertheless, they don’t account for a truly neutral condition either, since they explain that their moderate condition, is more like a moderately high standard as they describe: “In particular, he was introduced as a professional athlete who was among the better members of his team and typically made a good impression during competition” (Mussweiler et al., 2014, p. 835).

Therefore the Instagram account in the control condition of this thesis will be completely the same as the two experimental conditions, except for not showing any number of followers at all. It could also have been decided to measure a pre-evaluation of self-esteem

(19)

before the manipulation, but this could lead to an overspill, since the pre- and post-evaluation would be measured in a very short period of time. Moreover, the condition with no number of followers shown was chosen as a neutral reference point for this thesis to really measure the impact of the number of followers an sich.

Recall that if individuals compare themselves to a target that is close to the standard, this will have a positive effect on self-evaluation, while on the other hand comparing oneself with a target that is not close to the standard seems to have a negative effect on self-evaluation (Mussweiler et al., 2014). Therefore it is expected that participants assigned to the high number condition will report a lower self –evaluation, while participants in the low number condition on the other hand will report a higher self-evaluation compared to the control group. Therefore hypothesis 1 will be:

H1: The perceiver’s state self-esteem is lower after observing an Instagram account with high number of followers than after observing Instagram account with low number of followers.

2.2.2 The halo effect

The way in which we process information can be explained by the Elaboration Likelihood Model (ELM) (Petty & Cacioppo, 1981). It has two routes, central and peripheral. The central route creates and adjust attitudes by accurate examination and observation of information. Because of the conscious consideration the attitude changes it provokes are of long term and a high involvement is needed.

In the peripheral route, on the contrary, attitudes are constructed and adjusted without accurate examination, but rather by associating the element and its dimensions with positive or negative cues. People are unaware of the incongruence of these associations, since it happens on an automatic pilot. The attitude change this provokes, is of temporary use, because the information was not processed carefully, but used “short cuts” instead.

(20)

Peripheral cues are especially used by consumers for information processing under low-involvement, low-knowledge or low-ability conditions (Petty & Cacioppo, 1981). Although the ELM made a distinction between the two routes, a more resent study that examined the online persuasion process found that people with high involvement use both central and peripheral processing routes together for information processing (San José-Cabezudo, Gutiérrez-Arranz & Gutiérrez-Cillán, 2009).

Social proof is identified as one of the six types of peripheral cues (Cialdini, 1993). In accordance, Jin and Phua (2014) demonstrate that the number of followers can also be seen as a peripheral cue. Their results confirm this by showing that the value of the number of followers, is correlated with other evaluations of the target: when a celebrity had a high number of followers, he or she was associated with more positive associations (attractive, trustworthy and component) then when the celebrity had a low number of followers. An explanation they give for this is that a high number of followers is a cue a larger bridging social capital and therefore people will be more eager to join their network and tap into their social capital resources. Likewise, other social media studies have also demonstrated that social media users with many friends are evaluated as overall more positive, e.g. as being more physically attractive, popular and extravert than users with fewer friends (Utz, 2010; Antheunis & Schouten, 2011).

The reason why users with many friends are evaluated to be in possession of more socially desirable characteristics than users with fewer friends can be explained by the halo effect. Thorndlike first invented this term in 1920. In his study, he asked commanders of the army to evaluate their soldiers a variety of qualities. He found that the ratings of one quality bled over onto the assessment of other characteristics. The halo effect is a cognitive bias in which our overall evaluation of a persons’ character can be influenced by the one single attribute. Kahneman (2011, p.81) describes the halo effect as an “exaggerated emotional

(21)

coherence”. The halo effect can be powerful and although we know of its existence, it is not easy to avoid its impact on our perceptions. It affects the way we truly feel about another person or thing.

The peripheral route, forms attitudes by associating the object and its attributes with positive or negative cues. Social proof and therefore the number of Instagram followers is such a peripheral cue, which is expected to create a halo effect. Therefore to test the “exaggerated emotional coherence”, inter-correlation between different attributes will be measured. Furthermore it will be measured if this causes the perceivers’ overall evaluation of the target to be more positive in the high number than in the control condition and more negative in the low number than in the control condition. This then automatically means that the overall evaluation will also be more positive in the high number condition than the low number condition. No prior research has examined the relationship between the number of followers on Instagram and the overall evaluation (popularity, likeability, health, fitness, trustworthiness) in the context of the halo effect. To address this gap the following hypotheses are proposed:

H2: The evaluated characteristics of the target (popularity, likeability, health, fitness, trustworthiness) have a high average inter-correlation (above r = 0.60), which means there is a halo effect.

H3.1: An observed Instagram user with high number of followers will be evaluated more positively than the control condition.

H3.2: An observed Instagram user with low number of followers will be evaluated overall more negatively than the control condition.

H3.3: An observed Instagram user with high number of followers will be evaluated overall more positively than the low number condition.

(22)

2.2.1 The six attributes to measure target-evaluation

There is not one universal way to measure the strength of the halo effect or to decide on its direction. Huber and James (1978) studied the diverse definitions of the halo effect and recommend to use attributes that have a clear negative and positive rating and measure inter-correlation between them. In contrast to other definitions of halo their operationalization is simple and logically pursues from the concept of halo as perceptual bias due to preference.

The study of Huber and James (1978) did not solve the problem of the direction of causality between preference and assumptions. This issue can be ironed out in this thesis, since participant will see only one of the three conditions. They will see the Instagram account before they evaluate the attributes of the target. Since the three conditions will look exactly the same, except for the manipulated number of Instagram followers shown on the accounts, significant differences in overall evaluation between the three conditions can explain that the halo effect was caused by this number of Instagram followers. This thesis will look at both the magnitude of the hallo effect and the overall rating level of the target. A high number of followers coupled with a superior rating of overall evaluation is clearly suggestive of a positive halo, whereas a low number of followers coupled with an inferior rating of overall evaluation is clearly suggestive of negative halo.

2.2.1.1 Fitness, health, likeability and popularity

In this thesis, five attributes (fitness, health, likeability, popularity, trustworthiness) were chosen to measure the overall evaluation of the target and see if there is the halo effect. Four of these dimensions (fitness, health, likeability and popularity) were taken directly from the study done by Vogel et al. (2015), who used these dimensions to measure the overall evaluation of the target too in their second part of study 2. They chose these attributes because they were most relevant to their manipulations. Their research consisted of four conditions

(23)

that made a distinguishing between user content and social network content to create social comparison. To manipulate user content a Facebook profile with positive healthy- and one with negative healthy information was created. To manipulate social network content, the target profile either had a high network activity, with a large number of likes and comments, or a low activity, with a small number of likes and comments. Resulting in a negative and positive condition for both variables. Their results indicated that the aggregated evaluations of the target were rated as overall more negative in the negative healthy condition than in the positive healthy condition. The same direction was found for the social network, but these findings were insignificant. Nevertheless, in the first part of their study 2 they demonstrated that social network content had a bigger impact on state esteem than user content.

In this research, the target-evaluation is used as a mediator, while Vogel et al. (2014) do not. Other differences are that they examined the SNS Facebook, while this thesis examines Instagram and they did not measure for a control condition, which means they would not even be able to measure a mediating halo effect of target-evaluation, since they could not compare their findings with a neutral condition.

2.2.2.2 Trustworthiness

Trustworthiness will be added to the scale of Vogel et al. (2014). A reason for this is that Annie and Phua (2014) found that source credibility was rated as higher for a celebrity with a high number of followers than for a celebrity with low number of followers. Trustworthiness is one of the three dimensions dimensions of source credibility together with attractiveness (which is already in the Vogel et al. (2014) scale and competence. However competence was evaluated as too broad compared to the other attributes, this would be likely to confuse the participants.

H4: The overall evaluation of observed account is negatively correlated with the perceiver’s state self-esteem.

H5: The overall target evaluation mediates the relationship between the number of Instagram followers and the state self-esteem.

(24)

2.2.3 The relation between target evaluation and self-evaluation

The expected relationship between overall target evaluation and self-evaluation can be

explained by the relative deprivation theory. The relative deprivation theory of Merton

derives from 1938 and is caused by the fact that we life in a consumer society with a lot of incentives, created by marketing techniques and advertisements. The consumer society and the rise of self-branding puts a lot of pressure on individuals to achieve the socially accepted goals and to be successful. Being relatively deprived means that you feel disadvantaged in

comparison to a reference group or target. Feelings relatively deprived can involve money,

justice, status or privilege. This causes a stressful situation, which affects self-rated health and self-evaluation (Yngwe et al. 2003).

Social media magnifies this pressure, since individuals on social media only present their “best self”. It is claimed that on dating sites, for example, people inflate their height by two inches and their income by 20%, not even mentioning the attractive, out of date photos they pick for their profile (Ungar, 2015). This results in social-comparison to people that seem better than the self.

Crosby (1976) found that not only economic disadvantaged people, also people who are ‘well off’ can feel unfairness about their lot in life. It is therefore expected that if an individual looks at an Instagram account with a high number of followers, which is expected to generate a an overall more positive evaluation of the target, this will results into a lower self-evaluation, since the participant will feel of less worth in comparison to the other.

On the other hand, the theory of Mussweiler et al. (2004) can be applied again to explain the relationship between the overall target evaluation and the self-evaluation for all conditions, because of the expected halo effect, the target evaluations in the high conditions might be evaluated as more extreme from the self, while in the low condition, that showed a moderate number of followers, the target evaluation will be closer to the self to, which results

Formatted: Normal, Justified, Level 3, Space Before: 10 pt, Keep with next, Keep lines together

(25)

into reporting a higher self-evaluation. Taking both of these theories into account, Hypothesis 4 will be:

H4: The target evaluation is negatively correlated with the perceiver’s state self-esteem.

Figure 2.2.3 The conceptual framework visually represents the combined set of hypotheses 2.3 Part II

2.3.1 The specific self-evaluation dimension

As mentioned in the beginning of this thesis, the analysis will be done for two different dimensions of self-evaluation. A broad one (self-esteem) and a more specific one (specific self-evaluation). The following section will provide information why doing so will be an addition to the broad concept of self-evaluation, and will provide the reader with some expected differences.

Another article of Mussweiler (2003) states that social comparison has an influence on our self-evaluation because it influences explicitly which knowledge about our self will be accessible. The results of Mussweiler et al. (2004), also measured specific self-evaluations, since the participants had to evaluate their athletic competences. In addition, before making

(26)

the self-evaluations, the participants were presented specific information about the target too, because they were confronted with information about either an average or an extremely good sportsman. In this thesis however, the participants have to make the evaluations about the targets themselves, but the 6 dimensions on which these evaluations are made have already been chosen for them.

In their article Kahneman and Miller (1986) discuss that specific objects or events generate their own norms, which is call the norm theory. They look at stimulus norms that result in comparative judgement. Their central idea is that norms are gauged after the event rather than in advance. Thus, it is expected that if participants are asked to evaluate the target, on specific attributes, these attributes will act as a norm to evaluate the self, on these same attributes. Since this self-evaluation is more comparable to the target evaluation, social comparison is more likely to happen.

The only study on self-esteem and Instagram that was found was from Roble et al. (2015). They also measured the difference in the effect of a popular and unpopular Instagram account on self-esteem. They did not use the number of followers as a criterion, but manipulated the number of likes and the comments. Their targets’ photos did not include average pictures, but pictures of the target in bikini. Their results did not show a significant effect of the popularity on self-esteem, however this may be affected by methodological limitations this article faced. The title of their article for example talks about the effects of perceived attractiveness, while this is not measured at all.

Nevertheless in their discussion they do stress out a point of importance. As a consideration for not finding an effect on self-esteem they explain that a reason for this might be because of the fact that the targets’ profiles highlighted physical characteristics. They discuss that significant results may have been recorded if something they would have measured something that was more closely related or specific to physical attractiveness, like

(27)

body image. In their case, self-esteem removed too far away from the concept that they should have measured. To avoid making the same mistake, it was decided to analyze a second dimension of self-evaluation to measure for more specific self-evaluation. The same six attributes (health, fitness, trustworthiness, attractiveness, likeability and popularity) as for the target evaluation will be used to measure the specific self-evaluation of the participants. The relations between the variables are expected to be the same as in the first part of the study. The same theories can be used to justify the relations. The analysis of this second part of the study will look at the differences in the target and specific self-evaluation specifically, but also two hypotheses were draught. Since this part of the thesis will function as an addition on the first part that measured self-esteem- the primary construct of interest in this thesis- a lower load should be given to this part.

H6: The perceiver’s specific self-evaluation is lower after observing an Instagram account with high number of followers than after observing Instagram account with low number of followers.

(28)

3) Methodology

3.1 Choosing the sample

The participants in the study were reached using convenience sampling and purpose

sampling. Convenience sampling is a method of sampling in which participants are collected based on their relative reachability. With convenience sampling the external validity could fall because only easy to reach participants will be selected. However, this method of sampling enables the investigator to reach a lot of participants in a limited time (Bryman, 2008). Participants were mostly reached the participants via Facebook, but also used face-to-face invitations and some international schools participated. Facebook was used as a main medium to reach participants, since that is an easy, quick and inexpensive way.

Purpose sampling was also used in order to gather the participants. Respondents are approached using forums, direct mail and media channels. Also, some international school were mailed if they could have their students participate, since the questionnaire is written in English and Instagram has young users. Allocation of the participants to the control group and the treatments is entirely random. The Qualtric’s survey flow ensures that each respondent

(29)

only evaluates an Instagram account of the same gender, and allocates to each of the six categories a similar number of participants

3.2 Selection of the Respondents

Participants were asked to take an online survey in Qualtrics. They could do this on a computer, laptop, tablet, or a cell phone. The questionnaire was the same for all experimental conditions. The complete set of items is shown in the Appendix C. Participants were given a paragraph of informed consent before beginning the survey that assured that the answers would be treated with an absolute confidentiality and that only aggregated results would be used.

Participants were asked to fill in some demographic questions about their age, gender, country of origin, country of residence and education.Allocation of the participants to the control group and the treatments is entirely random.They were shown one of the manipulation assigned to their own gender.

After that the scales were presented in a randomized order to prevent a spillover effect. For the selected scales, a five-point Likert scale is used for the questions. Weijter, Cabooter & Schillewaert (2010) recommend a 5 point Likert scale when the questions include a normal population. Another study of Revilla, Saris, and Krosnick (2013) agrees with that. Furthermore, they found that there is not much difference in the reliability of a five (α = 0.717) - and seven-point scales (α = 0.716), adding that a 5-point scale provides better quality of data compared to a 7 or 11-point scale. Therefore a 5 point Likert scale was used adding a neutral point.

3.2 Experimental design

This experiment consisted of three conditions. This design is also known as vignette research. The vignette consists of stimuli presented to research participants. The purpose of the vignette is to selectively portray aspects of reality to which the participants are asked to respond. This is done through a survey approach, which is distributed online via the use of Qualtrics. Looking at the timeframe, a pre-test of all the treatments is too ambitious. Therefore, this is a post-test-only control group design. During this experiment, participants will only enter one treatment, and will not be aware of any other treatments. In total, there are three conditions, which have different levels of shown followers (none, low and high) and gender (boy or girl).

(30)

Participants would see a category with their similar gender. All participants would see the same survey questions (Appendix..)

3.3.2 Manipulations

This thesis will contain of three conditions. All conditions will see an Instagram account with a different number of Instagram followers. Next to the control condition, which doesn’t show any number of followers, there are two experimental conditions, a high and a low number condition. In the high number condition the number of followers shown will be extremely away from the standard (21.000). The other condition will show a moderate number of followers (210), which will be called the low number condition to express the difference. This would manipulate the user content to convey upward comparative information, that would presumably affect the participants’ state self-esteem.

These number of followers were manipulated based on actual real Instagram statistics (Appendix B). Participants were shown a profile of their same sex and were then randomly assigned to either of the three conditions (Appendix D). It was chosen to show the participants a profile of their same sex because an important condition for social comparison to happen is similarity (Festinger,1954).

Participants were told to remember details about the target person. They would first see an overview of an Instagram account from the person of their own sex and after that they would see a picture of this person that he or she posted. This was done because in the overview all the photos are quite small and this bigger picture made the person more real. After that the manipulations were checked by asking participants the question if the Instagram user had a very small (=1) vs. very large (=5) number of followers.

For the overview of the Instagram account of the selected boy and the girl, all characteristics should be relatively similar. Dissimilarities between the two could be a source for alternative explanations. First, the names should sound alike as a name can also be an indicator of status and can lead to different evaluations. Second, both persons should have similar sort of interests. Therefore both profiles include a photo of food, a full body photo, a sportive photo, a photo of a watch, a photo of landscape, and a photo while being busy with food. These interests include both hedonic (watch, travel, fashion) and utilitarian (food, health, sport) motivation that a lot of people share on social media (Voss & Spangenberg, 2003). Third, the description of both profiles should be similar. Furthermore, their profile

(31)

picture should look alike. Both pictures are presented in black and white, their posture is the same.

The photo that the boy and the girl posted should also be similar. A photo was chosen in which both the girl and the boy are looking at the same direction. Both had something on their head and are smiling so that you can see their teeth. Both are sitting outside with something to eat/ drink. They both posted the picture five hours ago and both pictures are in color. Both are sitting in the shadow but you can still see some sunlight on the side. These pictures where chosen because they show an average boy and girl, doing an everyday thing.

For the low number of followers, it was chosen not to look at the average number of Instagram followers, but at the median number of followers. The median number of followers is more representative of the average user then the average number of followers, because all it would take is one account with one million of followers in the same sample to influence the average number of followers drastically. The median number of followers is 194, but given the two accounts the same characteristics, it was chosen to use 210 instead. This would manipulate the user content to convey downward comparative information, that would presumably affect the target person’s post esteem to be higher than the pretest self-esteem.

The experiment thus compares these evaluations of the control Instagram account where no number of followers are shown, to the Instagram account with a high or low number of followers shown.

3.3.3 Target evaluation scale and the self-evaluation scale

The three different conditions are expected to have different overall evaluation of the target. The six chosen dimensions to evaluate this were popularity, likeability, attractiveness, health, fitness and trustworthiness. Participants judged the extent to which the target person and themselves scored on the five dimensions using a 5 point scale (1=not at all; 5= extremely). Expected was that the overall evaluation would be rated superior after seeing an Instagram account with a high number of followers, whereas it would be rated inferior after seeing an Instagram account with a low number of followers.

Participants judged the extent to which the target person and themselves scored on the five dimensions using a 5 point scale (1=not at all; 5= extremely). These target and self-evaluations dimensions were chosen because they are of great significance of value in life (Vogel et al. 2014).

(32)

3.3.4 State Self-Esteem Scale

For the first dimension of self-evaluation, the state self-esteem scale was used (Heatherton & Polivy, 1991). This is a 20-item scale, designed to measure a participant’s self-esteem at the given point in time, i.e. short-lived (or so-called state) changes in the self-esteem. The scale is sensitive to manipulations designed to temporarily alter self-esteem. The 20 items dive into three factors of self-esteem: performance-, social- and appearance self-esteem, that correlate with each other. The original scale has a 5-point Likert scale (1= not at all, 2= a little bit, 3= somewhat, 4= very much, 5= extremely). Since the difference between a little bit and somewhat is intangible and therefore arguable, the same 5-point Likert scale as from the pretest (strongly agree to strongly disagree) will be used instead (Appendix E)

4. Results and conclusion

This chapter is organized accordingly: In the first section the study’s responses and sample characteristics are analyzed. These sample characteristics are used to verify whether a significant difference between the condition exists. This is of great relevance, due to the experimental setting. If the groups are found to be the same, this can be used to exclude alternative explanations for the given results. Any diagnosed limitations will be considered within the analysis of the experiment and will be explained in the discussion chapter.

The first part of this chapter will describe the sample, check for equality of the groups, check the manipulation, tests the reliability of the scales, check the normality distributions on the scales and will present a correlation matrix. The second part of this chapter will report the analyses made for answering the research question. This second part will first reports the effect of the number Instagram followers on the state self-esteem and if there is a moderator effect of the overall evaluation of the target. The second part of the section will dive deeper into self-esteem by looking at different attributes for examining self-evaluation and will compare this to the target evaluations of those some attributes.

(33)

A confidence level of 95% was chosen to execute the analyses. Which means that all effects are reported as significant with a a p-value smaller than 0.05 (Field, 2009). Effect sizes are analyzed according to Cohen’s directions (1988). According to him a partial eta-squared (ƞ²) of .01 indicatese a small effect, partial ƞ² = .06 a moderate effect, and partial ƞ² = .14 a large effect. Thereby, Cohen (1988) advised as a criterion, a Cohens d of .20 as small, a d of .50 as moderate, and a d of .80 as large. Additionally, a correlational coefficient (r) of .10 is considered as small, a r of .30 as moderate, and a r of .50 as large (Cohen, 1988). Results will be presented in tables or in the text between brackets. Common abbreviations will be used for reporting, a total list of abbreviation is included in Appendix F.

4.1 Sample Response analysis

On May 21, a message was placed on my own Facebook page and on multiple Facebook groups. Since Instagram has broad range of users, many people fitted the target group. Nevertheless at first, too many girls compared to boys replied. Therefor the survey was posted in many male community pages on Facebook, like Nike, Football and sport car groups. The survey was also posted on Facebook groups that promoted Instagram, for example influencer pages. Furthermore a lot of nationalities were reached through multiple international student platforms and direct. I have studied half a year in Australia and friends from all over the world were directly approached through the social media channel Facebook.

Instagram users are fairly young compared to other Social media users. Therefore, also some international schools were emailed. The main language on these school is English, which is of importance since the survey was in English. One international school, the International School of Hilversum replied and the assistant principal posted the survey twice to its students. They have 750 students with 43 different nationalities. Since Instagram is used

(34)

by people all over the world these different nationalities will give a representation of the real world.

4.1.1 Demographic characteristics

The initial number of participants who entered the survey in Qualtrics was 1134. By using the option Select Cases all the participants who did not finish the survey were deleted. 558 participants finished the survey. Out of them, 6 people who took the survey in Preview mode were also excluded. Finally, 552 participants were kept for further analyses.

Of those 552 participants who completed the survey, 366 were women (66.3%) and 186 were men (33.7%). This is a good presentation of real Instagram use since, according to eMarketer (2016), female users account for two-thirds of Instagram’s US daily audience (Appendix G). In Table 1. It is seen that the number of participants per condition (low number of followers, high number of followers, and control group) is almost equal.

The average age of the participants was 28.25 years old (SD = 13.269). This is also a good representation for Instagram users since the same research of eMarketer (2016) found that 55% of all Instagram users fall within the 18-to 29-year old range and over 90% of people on Instagram are under the age of 35. The youngest participant was 13 years old in the moment of taking the survey and the oldest participant was 71 years old. 13 years old was chosen as the lowest age, because this is the age minimum of most social media sites, because of the Children's Online Privacy Protection Act (COPPA).23 participants didn’t report about their age. The average age of female participants was 28.65 (SD = 13.659), and the average age of men was 27.43 years old (SD = 12.441). Independent Samples T Test shows insignificant difference in age between genders.

Most of the participants had a bachelor’s (181, i.e. 32.8%) and a high school degree (111, i.e. 24.4%). 84 (15.2%) participants had a master’s degree, 53 (9.6%) a college degree, and 26 (4.7%) had no degree.

(35)

Table 1. Number of participants per condition Low number of followers

High number of followers

Control group Total

Male 60 (10.9)* 61 (11.1) 65 (11.8) 186 (33.7)

Female 120 (21.7) 122 (22.1) 124 (22.5) 366 (66.3)

Total 180 (32.6) 183 (33.2) 189 (34.2) 552

*_{Percentages are shown in parentheses.}

4.1.2 Checking the equality of the treatment groups

Pearson’s chi-square tests were conducted to verify if significant differences exist between the treatment groups regarding the control variables education and gender. Each observation of the various treatments is independent of all others for both control variables. Moreover, no more than 20% of the expected counts is less than five and all expected counts are higher than 1 (Field, 2009).

In fact, the minimum expected count is 8,2 for education and 61,10 for gender which means that the assumptions for the Chi-Square tests are met. There is no significant difference between the three conditions regarding gender, Chi2 (2) = 0.09, p = 0.96. There is also no significant association between the type of condition and the level of education, Chi2 (8) = 4.36, p = 0.07. Cramer’s V is according to Field (2009), the preferred effect size for reporting effect sizes with variables with more than 2 categories. Cramer’s V is only 0.01 for gender and 0.07 for education, which indicates a small effect size.

Finally, in order to verify whether significant differences exist as a result of a difference in age, One-Way ANOVA is conducted. Group sizes of the conditions (Neutral N=178, Low=176, High=176),) seem appropriate for the analysis. Kolmorogov-Smirnov tests indicate that the distribution of neutral number of followers 𝐷𝐷(178) = 0.21, 𝑝𝑝 < 0.01, low number of followers 𝐷𝐷(178) = 0.213, 𝑝𝑝 < 0.01, and high number of followers 𝐷𝐷(178) =

(36)

0.21, 𝑝𝑝 < 0.01 are significant and thus not normally distributed. However, since these group sizes are following the central limit theorem (N > 50) generally large, they should be suitable for parametric tests for each condition. Levene’s test is not significant 𝐹𝐹(2.53) = 1.47, 𝑝𝑝 = 0.23, which indicates that the variance between the treatment groups are homogenous for age, and that all assumptions for the one-way ANOVA are met. The results of this test, presented in Table 2, indicate that no significant differences exist between the treatment groups 𝐹𝐹(2.53) = 0.22, 𝑝𝑝 = 0.80

Table 2. Average age of participants for each condition

Age

Nr of followers N mean SD*

Not shown (control) 178 27.71 12.696

Low 176 28.43 14.282

High 176 28.61 12.843

*

SD, standard deviation

In conclusion, Chi-square tests and an One-Way anova ANOVAs indicate no unpredicted significant differences exist in treatment groups regarding gender, education and age.

4.2 manipulation check

Before checking the hypotheses, a manipulation check was done in order to see if participants noticed the number of followers well in each condition. A One-Way ANOVA was used to do so. The model is significant, F (2) = 472.431, p = 0.00. Table 3 and Figure 1 show how the participants evaluated the number of followers of observed account in each condition on average. It was measured on 1 – 5 scale (1 = „very small“, 5 = „very large“). We see that the participants estimated the highest number of followers in the high number of followers condition (M = 4.26), and the lowerst number of followers in condition with low number of followers (M = 1.69). In control condition, with no information about the number of followers shown on the observed account, the participants estimated the number of

(37)

followers somewhere in between the rest two conditions (M = 3.40). LSD Post-Hoc test shows that all the differences between each pair of conditions are significant. These results mean that the experimental manipulation was successful.

Table 3. Estimation of the number of followers per condition

Estimation of the number of followers**

Nr of

followers N mean SD*

Not shown (control) 189 3.40 0.734

Low 180 1.69 0.922

High 183 4.26 0.759

*

SD, standard deviation **

On scale from 1 (very small) to 5 (very large)

Figure 1. Estimation of the number of followers per condition

1,5 2 2,5 3 3,5 4 4,5

low not shown high

E st im a ti o n o f t he num be r o f fo ll o w er s ( 1 -5 s ca le)

(38)

4.3 Reliability of the scales

In this section, the internal consistency of the different constructs were assessed by computing the Cronbach’s alpha. The required threshold is 0.7, which indicates that the scales are internally consistent at a confidence interval of 95%. The Cronbach’s alphas for the different constructs are shown in table 4 This reliability analysis is especially crucial for the state self-esteem scale, as the average of these scales is used to determine the evaluation of self-self-esteem after being assigned to one of the conditions with low, high or no number of followers shown. Internal consistency of this scale is very good, Cronbach’s Alpha = 0.90

Evaluation scale. The reliability of evaluation scale is checked by including all target evaluation and self-evaluation items, at the start of this reliability analysis the scale includes 12 items. The Evaluation scale consists of two subscales, the Target evaluation subscale and the Self-evaluation subscale. In the first subscale evaluations of the target were made, while in the second subscale evaluations of the perceiver self were made. Both scales rated 6 dimensions: Healthiness, fit, likeable, popularity, trustability and attractiveness.

The evaluation scale has an Cronbach’s Alpha equal to 0.69, which is just below the minimum required threshold of 0.7. In the table Item-Total Statistics, in column Corrected Item-Total Correlation it was seen that values for perceiver’s (self) attractiveness, trustability, popularity and fitness are a bit lower than 0.3. However, when any of these items is deleted, Cronbach’s Alpha would become lower. However the value for target’s attractiveness is the lowest of all (r =0.23) and if this item is deleted the Cronbach’s Alpha increase, Cronbach’s Alpha if Deleted=0.70, which passes the threshold.

For a more detailed check, the internal consistency of the Target evaluation subscale and Perceiver evaluation subscale (self) was also calculated. For the Target evaluation subscale Cronbach’s Alpha is equal to 0.73. In the table Item-Total Statistics it was seen that the lowest Corrected Item-Total Correlation are for trustability (r = 0.31) and attractiveness (r

(39)

= 0.33). The Cronbach’s Alpha would stay the same if that trustability is deleted (Cronbach’s Alpha if Item Deleted = 0.73), but the Cronbach’s Alpha would increase if attractiveness would be deleted (Cronbach’s Alpha if Item Deleted=0.75). For these reasons it was decided to exclude the attractiveness item.

For more detailed check, we again calculated the internal consistency of the Target evaluation subscale without the attractiveness item. Remember that the Cronbach’s Alpha for this scale after deleting attractiveness was equal to 0.75. In the table Item-Total Statistics it was now seen that the lowest Corrected Item-Total Correlation is for trustability (r = 0.32), and that the Cronbach’s Alpha would increase if that item was deleted (Cronbach’s Alpha if Item Deleted = 0.77).

For the Perceivers evaluation subscale Cronbach’s Alpha is equal to 0.66, which is below the acceptable. In the table Total Statistics we see that the lowest Corrected Item-Total Correlation is for trustability (r = 0.22), and that the Cronbach’s Alpha would increase if that item was deleted (Cronbach’s Alpha if Item Deleted = 0.69). For these reasons, we decided to exclude the dimension trustabiliy from the Evaluation scale.

Table 4. Reliability of scales

Measure (variable) Cronbach’s α (>0.7)

Rosenberg Self-Esteem (CV) 0.87 State Self-esteem (DV) 0.90 Target (IV)* Self (IV)* 0.69 0.77

*Note: healthy, fit, likeable, popular

The Cronbach’s alpha of perceiver evaluation subscale is just below 0.7, but it is measured in addition to the primary measure of state self-esteem, to give a deeper insight by evaluating themselves on five dimensions (health, fitness, likeability, trust) and not used to answer the core research question. Additionally, 0.69 is very close to 0.7, so therefor it is not considered a problem.

The power of the number