#Influencer_Marketing: Essential choices to consider

(1)

#Influencer_Marketing:

Essential choices to consider

An analysis of the impact of disclosure labels, type of celebrity and

brand presence on advertising effectiveness on Instagram

University of Groningen MSc Marketing Management

(2)

2

#Influencer_Marketing:

Essential choices to consider

An analysis of the impact of disclosure, type of celebrity and brand

presence on advertising effectiveness on Instagram

by Kiki Wesselink k.wesselink@student.rug.nl S2352702 Dr. C. Hofstede de Grootkade 11-74 9718 KA, Groningen +31 6 29 600 863 11th_{of March 2019} University of Groningen Faculty of Economics and Business

First Supervisor: dr. J. Berger Second supervisor: dr. J. P. R. Thomassen

(3)

3

ABSTRACT

Over the past few years, social media have become a meaningful advertising platform for companies. It offers new opportunities to reach the target audience in a fast and cheap way. An effective strategy to use is Influencer Marketing. With influencer marketing celebrities are used as endorsers to promote specific products or services of a particular brand. There are essential factors a marketer needs to consider when he wants to implement this strategy to achieve high mobile advertising effectiveness. Prior research found that the use of disclosure labels, the choice for type of influencer and the variety of brand presence could influence the effectiveness of an advertisement. This research aims to investigate the effects of disclosure labels, type of influencer and brand presence and their interplay on the advertising effectiveness on Instagram. Brand attitude, purchase intention and the intention to spread e-WOM are indicators for advertising effectiveness and represent together the mobile advertising effectiveness. An online experiment (N = 296) was conducted, employing a 2 (yes disclosure label vs. no disclosure label) x 2 (traditional celebrity vs. non-traditional celebrity) x 2 (high brand presence vs. low brand presence) between-group design. All participants were Dutch females between the age of 18-30 years. A three-way MANOVA analysis was conducted to examine the main and interaction effects on advertising effectiveness. The results show significant positive main effects for disclosure and type of influencer. Also, two significant interactions effects were found who suggest that a combination of non-traditional celebrities without disclosure and the combination of low brand presence with disclosure had the highest positive impact on mobile advertising effectiveness. Overall, this research provides meaningful insights for brands and marketers about the essential choices concerning influencer marketing on Instagram.

(4)

4

PREFACE

This study aims to investigate the essential factors that could influence the effectiveness of influencer marketing advertising on Instagram. Influencer marketing on Instagram is a strategy of high personal interest since I am an active Instagram user who is fascinated by the celebrities and their rapidly growing influence. Therefore, it was a great pleasure to conduct this research and read much literature about this topic. During this journey, I have learned a lot about online advertising and hopefully it is possible to bring all this knowledge in practice soon. Since influencer marketing is the new marketing strategy for brands, I already received questions of companies who were interested in the outcomes of this research.

I would like to thank my supervisor Dr. Hans Berger for his support, guidance and motivation during the process of my research. His help and patience were most valuable for me since it has been a long journey to complete this research. Also, I would like to thank study advisor Juliette Kars, for her kindness, help and support during this difficult process. Furthermore, I would like to thank my friends and all the female participants who filled in the survey, without them I would not have been able to conduct this research. Lastly, I would like to thank my family and Bas for their great support and motivation during the journey.

I hope you enjoy reading!

Kiki Wesselink

(5)

5

1. INTRODUCTION

(8)

8 transparency among advertising posts, which means that all the sponsored posts need to include disclosures characteristics (FTC, 2015). A disclosure contains information that indicates the relationship between the influencer and the sponsor. To date, little is known about the effect of these disclosing language characteristics on consumers attitude towards a brand (Evans et al., 2017). Since it is obligated to include disclosure characteristics in social media posts (FTC, 2015), it is vital for marketers to know more about the effects of disclosure if they want to use influencer marketing as a strategy. In the end, their goal is to achieve positive attitudes towards their brand. Another factor which could influence the effectiveness of the advertisement on Instagram is the choice of the endorser also called the type of celebrity. Prior research has shown that in case of celebrity endorsement, the purchase intention of the consumers is positively influenced even when they are aware of the fact that the Instagram post is a paid advertisement (Djafarova and Rushworth, 2017). Nowadays a new type of celebrity endorsers came up which are called influencers (Weinswig, 2016). These influencers are bloggers, YouTubers or society people who owe their celebrity to their popularity on Instagram. Recently, differences found between the effect of traditional celebrity endorsers and these new form of influencer endorsers on the consumer responses (Djafarova and Rushworth, 2017). Therefore, the choice of the type of influencer is essential to take into consideration when setting up an influencer marketing strategy. Another key element of native advertising is brand presence. The level in which a brand refers to itself in a native advertisement is called brand presence. Prior research has shown that increased brand presence will result in negative advertisement evaluations (Krouwer, Paulussen & Poels, 2017). When the advertiser often positively refers to itself, the information can be perceived as subjective, and consumers become more suspicious (Eisend, 2006). According to Krouwer et al. (2017), the influence of brand presence has not been investigated yet on social media networks.

(9)

9 achieve high advertising effectiveness (Campell & Mark, 2015). Therefore, this study aims to examine the effects of disclosure, brand presence and type of influencer on mobile advertising effectiveness. Brand attitude, purchase intention and intention to spread e-WOM are key factors that indicate the effectiveness of an advertisement (Evans et al., 2017) and are therefore the dependent variables in this study. Together they represent the overall construct of mobile advertising effectiveness. Separately, the three independent variables have been studied before (Evans et al.,2017; Krouwer et al., 2017; Djafarova & Rushworth, 2017; Van Reijmersdal et al., 2015). Together these three factors have never been the focus of a study before and certainly not in interaction with each other. It is likely to expect that these three independent variables are related to each other since they are all linked to influencer marketing (Jin and Phua, 2014; Evans et al., 2017; Krouwer et al., 2017). The individual effects of disclosure, brand presence and type of influencer on mobile advertising effectiveness are known based on prior studies, but the interactions between these effects are not known yet. These interaction effects are interesting to look at since these effects can strengthen or weaken each other on the effect of mobile advertising effectiveness. Subsequently, this results in higher or lower advertising effectiveness. It is plausible to suggest that an Instagram post with the combination of a disclosure label and high brand presence will intensify the negative impact on mobile advertising effectiveness since the commercial intent becomes too clear for the consumer (Krouwer et al., 2017). Therefore, this study is, a contribution to existing literature while it looks at the different interactions between the independent variables on advertising effectiveness. The findings of this study could reveal useful information for marketers and brands because it shows all the possible effects of the combinations between brand presence, disclosure and type of influencer on mobile advertising effectiveness. It is important to be aware of the positive or negative influences when these factors are combined in a sponsored Instagram post, as it can result in higher or lower advertising effectiveness. Therefore, the research question this study aims to answer is:

RQ: what are the main and interaction effects of disclosure characteristics, type of celebrity

and brand presence on advertising effectiveness on Instagram?

(10)

10

2. THEORETICAL FRAMEWORK

2.1 Concepts

In this section, the concepts of persuasion knowledge model and message source credibility will be explained briefly to get a better understanding behind the relations between the independent variables on advertising effectiveness.

2.1.1 Persuasion knowledge model

According to Friestad and Wright (1994), consumers learn to be more aware of persuasive messages, through their experiences with persuasive intent. The persuasion knowledge model developed by Friestad and Wright (1994) provides an understanding of the conceptual context of how consumers react to persuasive information. It presumes that consumers process persuasive messages differently when they are aware of the persuasive intent. As a result, their persuasion knowledge is activated, which can affect their behaviour and attitudes towards a brand (Boerman, Reijmersdal & Neijens, 2012). When the persuasion knowledge is activated, consumers are likely to have an increased scepticism, resistance, and counter-arguing towards the persuasive message of a particular brand. Subsequently, this can have a negative impact on their attitude and behavioural intent towards the brand (Nelson, Wood, and Paek, 2009).

2.1.2 Source credibility theory

(11)

11 2.2 Influencer Marketing

Influencer marketing is a popular tool among marketers, since the use of celebrities (Jin and Phua, 2014), bloggers (Lee and Watkins, 2016) and other influencers positively increase consumers purchase intentions, consumer engagement and attitude towards the brand (Phua, Jin and Kim, 2017). Compared to traditional advertisements, social media-based influencer marketing is a relatively cheap marketing strategy with an ability to reach a large target audience in a relatively short period (Phua, Jin & Kim, 2017). According to Burgess (2016), a primary reason for the high effectiveness of influencer marketing is that consumers perceive the influencer endorsements as more trustworthy than traditional advertisements. On social media, influencers can be seen as individuals who have a certain impact on people. Companies select them in order to promote the brand to that specific audience (Sammis, Lincoln & Pomponi, 2016). Influencers can be seen as opinion leaders for the mass, where the mass can be seen as their followers (Katz & Lazarsfeld, 1966).

With influencer marketing on social media, the e-WOM of the product or brand is more controlled compared to traditional advertising, as the brand can choose its endorser. Brands can collaborate with influencers in many ways: influencers can share sponsored content, promote brands or product placements, hosting an event or reviewing about it afterward (Mediakix, 2016). These collaboration activities found to be positively affecting consumer engagement for the brand (Evans et al., 2017). The credibility of influencers plays an important role here (Gotlieb and Sarel, 1991). Kutthakaphan and Chokesamriptol (2013) suggest that when consumers formed a positive perception among the endorser on Instagram, the endorser is likely to generate more positive feedback from his or her followers. Besides, Townsend (2015) found that when celebrities are liking a particular brand or product, it is plausible that their followers copy the same opinion. Influencer marketing on social media is, therefore, a powerful mechanism for marketers and brands. According to Djafarova and Rushworth (2017), research findings on influencer marketing within social networks are limited since it is rather a new issue to the digital world. It is interesting to look at this concept on Instagram, as it is a rapidly growing social network with great potential for brands and marketers (Djafarova & Rushworth, 2017).

2.3 Dependent variables: mobile advertising effectiveness

(12)

12 2.3.1. Brand attitude

Spears & Singh (2004) suggest that brand attitude can be seen as an evaluation of the brand. According to Aaker (1991), the brand image of a consumer is identical to the brand associations a consumer. Also, Keller (1993) suggests that associations with a brand are benefits, attributes, and attitudes perceived by the consumer. Prior research has found that the credibility of the endorsers impacts the attitude towards the brand (Ohanian, 1990; Goldberg and Hartwick; 1990). Also, endorsements by celebrities and other influencers can directly impact the consumers brand attitude (Amos, Holmes & Strutton, 2008; Jin and Phua, 2014). Also, the use of disclosure labels and the intensity of the level of brand presence in an advertisement can affect the attitude towards the brand (Evans et al., 2017, Wojdynski 2016, Krouwer et al., 2017).

2.3.2. Purchase Intention

“Purchase intentions are an individual’s conscious plan to make an effort to purchase a brand” (Spears and Sing, 2004, p.56). It is of great interest for brands to measure consumer’s purchase intention since it is a widely-used indicator for the effectiveness of a marketing strategy (Morwitz, 2014). Consumers search online for reviews of products and brands to reduce the risk of lousy purchase decisions (Moore, 2010). Especially the use of social media can lead to an increased purchase intention for celebrity-endorsed products (Wilcox and Stephen, 2013). According to Hwa (2017), the consumers' intention to purchase depends on the effectiveness of the endorsers used for influencer marketing. In addition, Pornpitakpan (2004) found that celebrity endorsers are likely to affect the purchase intention of consumers. Especially females are sensitive to the concept of celebrity endorsements, which reflects their buying behaviour (Wilcox and Stephen, 2013). Also, disclosure labels might have an impact on consumers behaviour regarding their purchase intentions (Tessitore & Geuens 2013).

2.3.3. Intention to spread e-WOM

(13)

13 Furthermore, Lee, Kim & Ham, (2016) found that disclosure labels can influence the online sharing intention.

2.4. Independent variables: Main effects

In the following, the independent variables and their main effects on mobile advertising effectiveness are briefly described in order to show their relation to influencer marketing and advertising effectiveness.

2.4.1. Disclosure labels

Influencer marketing is a strategy that makes use of the concept of native advertising. It is a paid form of advertising with the purpose to make readers feel like they are reading editorial content instead of commercial content (Krouwer et al., 2017). The native content aims to be informative and entertaining (Campell and Marks, 2015). However, there are concerns about the fact that the effectiveness of native advertisement depends on the consumers’ lack of awareness of the commercial intent of the content (Wojdysnki and Evans, 2016; FTC, 2015). Therefore, The Federal Trade Commission set up strict guidelines about native advertising for brands and influencers in order to protect the consumers from being misled (FTC, 2015). Native advertisement content needs to contain labels; these labels are called ‘disclosures’ and identify the persuasive intent. Disclosure labels have to be efficient and clear in order to manifest the commercial intent behind the message to the consumer (Rozendaal, Matthew, Van Reijmersdal & Buijzen, 2011). Evans et al. (2017) found that #PaidAd is the most recognizable disclosure label for advertising recognition. According to the FTC (2015), every influencer who is cooperating with a brand needs to include disclosure labels when they endorse sponsored content. Influencer marketing on social media has not been identified as covert advertising yet, but it is crucial for brands and the influencers that they stick to these guidelines and be transparent to the consumer.

(14)

14 context of influencer marketing on social media (Evans et al., 2017). Since disclosure labels reveal the commercial intent of the message, is it plausible to suggest that when consumers recognize the content as advertising on Instagram, it can evoke resistance which will increase negative attitudes and behavioural intention (Evans et al., 2017). Therefore, the following hypothesis will be tested:

H1: Exposure to disclosure labels in an Instagram post will have a negative influence on mobile

advertising effectiveness.

2.4.2 Brand presence

Not only disclosure labels but also the content of the advertisement could reveal the commercial intent of the message, which is the variety of brand presence in the text (Krouwer et al., 2017). Brand presence can be defined as the number of times the brand is mentioned in a native advertisement. The level of brand presence can vary from one time to multiple times. When the persuasive intent of the editorial content is prominent, the consumers' persuasion knowledge can be activated (Van Reijmersdal, Neijens, and Smit, 2005). In addition, when the brand often positively refers to itself, which means that brand presence is high, readers can become more suspicious, and their persuasion knowledge may be activated (Eisend, 2006; Krouwer et al., 2017). Prior research showed that even when there is no disclosure label included in the advertorial, most readers still recognize this information as commercial content which is due to the commercial tone of the advertisement (Kim, Pasadeos and Barban, 2011).

Krouwer et al. (2017) found that high brand presence in native advertisements on news websites increased readers attitudinal persuasion knowledge which leads to more negative evaluations of the brand. Therefore, it is likely to suggest that content characteristics, like brand presence, could in consumers attitudes and behaviour. Still, this has not been investigated on social media networks (Krouwer et al., 2017). Bang and Lee (2016) found that social media users are less patient with advertising when the persuasive intent of the advertisement is perceived. Therefore, we can posit that when a brand is mentioned multiple times in an Instagram post, the brand presence of that content is high which can increase consumers persuasion knowledge and thus could negatively impact consumers responses. According to the just mentioned literature the following hypothesis will be tested:

H2: High brand presence in an Instagram post will have a negative influence on mobile

(15)

15 2.4.3 Type of Influencer

The use of celebrity endorsements is a popular strategy among marketers since it useful in building brand equity (Keller, 2005). Prior research found that endorsements by celebrities and influencers can positively impact consumers’ brand awareness, loyalty and intention to purchase (Miller & Laczniak, 2011; Jin and Phua, 2014). Also, celebrity endorsements are found to be useful in generating positive eWOM regarding certain products and services, since celebrities are perceived as credible (Kutthakaphan & Chokesamritpol, 2013). The credibility of the celebrity could have a positive influence on the credibility of the endorsed brand (Spry et al., 2011; Djafarova & Rushworth, 2017). Wiley (2014) found that celebrities appear to be effective in online communication. With the rise of social media, a lot of new opportunities were created for the strategy of celebrity endorsements. Traditional celebrities, like film stars, musicians, tv personalities are famous because of their achievements in their career. They are therefore already popular before they go on social media. On average traditional celebrities have many followers on their social media accounts (Jin & Phua, 2014; Djafarova & Rushworth, 2017). Although, with the rise of social media a new type of celebrity was created, which are called Influencers or micro-celebrities. These non-traditional celebrities are famous due to their online social media activities (Chahal, 2016; Djafarova and Rushworth, 2017). They are seen as ‘Instafamous’ personalities (Scott, 2015). Wiley (2014) suggest that traditional celebrities are not as influential as they once were since non-traditional celebrities have a greater impact on consumers due to their perceived credibility as more authentic and accessible.

(16)

non-16 traditional celebrity as they do from their peers ( Smith, Menon & Sivakumar, 2005). Therefore, the following hypothesis will be:

H3: Non-traditional celebrity endorsements will have a more positive influence on mobile

advertisement effectiveness compared to traditional celebrity endorsements.

2.5 Interaction effects

In the previous section, the expected main effects of disclosure labels, brand presence, and the type of influencer were described. Subsequently, the following section will explain how these main effects are expected to interact with each other and therefore could impact the mobile advertising effectiveness on Instagram.

2.5.1 Disclosure and type of celebrity

(17)

17 H4: The effect of type of influencer on mobile advertising effectiveness depends on whether

disclosure is included in an Instagram post or not and vice versa.

2.5.2 Type of influencer and brand presence

As stated before, the type of influencer is an essential indicator for the effectiveness of the advertisement, since the credibility of the influencer overrules the disclosure characteristics (Boerman & Reijmersdal, 2013). Does the celebrity endorsers’ credibility also overrule the commercial intent of the message when multiple brand references are included? To this date, little research is done about this topic (Krouwer et al., 2017). Gageler and van der Schee (2016) found that high brand presence can positively impact consumers intention to purchase, whereby this relation depends on the perceived credibility of the endorser. Prior research showed that when consumers feel persuaded by the endorser because it uses unfair manipulative techniques (brand presence), this could negatively influence their attitude towards the advertisement and brand (Cotte, Coulter & Moore, 2005; Wentzel, Tomczak & Herrmann, 2010). Therefore, we assume that when the commercial intent of a message is prominent because it contains a high intensity of brand presence, this might impact the credibility of the endorser and therefore have a negative influence on consumer responses. Also, we can assume that the effect of brand presence on advertising effectiveness depends on the credibility of the type of influencer. Hence, the following hypotheses will be tested:

H5: The effect of brand presence on mobile advertising effectiveness depends on the type of

influencer used in the Instagram post and vice versa.

2.5.3 Disclosure and brand presence

(18)

18 Krouwer et al., (2017) suggest that the differences in brand presence might be an explanation why some studies find negative effects of disclosure labels on consumer responses and other studies did not find any significant effect (Becker-Olsen 2003, Colliander and Erlandsson, 2015; Wojdynski and Evans, 2016). In line with this, Boerman and Van Reijmersdal et al., (2016) suggest that content characteristics like variations in brand presence might be an influence on the disclosure effects. Therefore, Krouwer et al., (2017) investigated the relation of brand presence and disclosure on consumers attitudes and behaviour on news websites. They indeed find interaction effects, which show that when participants did not recognize a disclosure label and brand presence was high, this had the most impact on activating their persuasion knowledge. It is plausible to suggest that the participants could have perceived the text as an advertisement, even when there was no disclosure label included, the participants could have felt manipulated and being misled (Ferrer Conill, 2016). Kirmani & Campell (2004) found that consumers could react positively to an advertisement when recognizing a persuasion attempt when the information is beneficial to them. According to the mentioned literature, we can assume that the intensity of brand presence and disclosure labels in an Instagram post can be dependent on each other concerning advertising effectiveness. Therefore, the following hypothesis will be tested:

H6: The effect of disclosure on mobile advertising effectiveness depends on whether brand

presence in an Instagram post is high or low and vice versa.

2.5.4 Type of influencer, disclosure and brand presence

Based on prior literature it is expected that there are interaction effects between disclosure and type of influencer, type of influencer and brand presence and disclosure and brand presence. Since the two-way interaction effects are likely to exist, it can be expected that these interaction effects together can result in an interaction effect on mobile advertising effectiveness. For the sake of completeness of this analysis, we will look at the interaction between all three independent variables. The current study is partly explorative because little research is done on influencer marketing (Djafarova & Rushworth, 2017). Therefore, it is interesting to look at what happens when these three factors interact with each other. Next, the following hypothesis will be tested:

H7: The interaction effect between disclosure and type of influencer on mobile advertising

(19)

19 2.6 Conceptual Model

The hypotheses mentioned above resulted in the following conceptual framework:

(20)

20

3. METHODOLOGY

3.1 Research Design

In this study, an experimental methodology with a 2x2x2 in-between-subjects factorial design has been chosen to examine the main and different interaction effects of disclosure, brand presence and type of influencer on brand attitude, purchase intention and intention to spread e-WOM. In this research context this factorial design is the best approach since each independent variable has two categorial levels: disclosure (no vs. yes), brand presence (high vs. low) and type of influencer (traditional vs. non traditional). The three independent variables were manipulated and resulted in eight different conditions which will be tested on the dependent variables (See Table 1). The three dependent variables (brand attitude, purchase intention and intention to spread e-WOM) are all three separate indicators of mobile advertising effectiveness (Evans et al., 2017).

Condition Disclosure Type of celebrity Brand presence

1 yes Traditional high

2 yes Traditional low

3 yes Non-traditional high

4 yes Non-traditional low

5 no Traditional high

6 no Traditional low

7 no Non-traditional high

8 no Non-traditional low

Table 1: Overview of the experimental conditions

3.1.1 Data Collection

(21)

21 with an Instagram account. According to prior research young adult females between 18-30 years old are the most active users on Instagram (Duggan, 2015; Djafarova & Rushworth, 2017; Statista, 2018). Males and females happen to react differently to online advertising and celebrity endorsers (Stephen & Wilcox, 2013). To exclude these differences and enhance a valid data set the focus will be only on women. This also benefits the choice for the brand used in the manipulation material, since it is focused on one gender a product aimed at females was chosen. If the respondents did not meet the requirements to attend the survey, they were kindly asked not to participate. To check if the participants meet these conditions, demographic questions about their age, gender, and Instagram account were asked. After these questions, they were randomly assigned to one of the eight manipulations (the Instagram posts). Above the photoshopped Instagram post, the participants saw a short description about the type of influencer since the pre-test showed that it was difficult to give answers if they did not know the source of the Instagram post. In addition, it was important to make a distinction between traditional and non-traditional celebrity to enhance the construct validity. Subsequently, manipulation check questions and questions about the dependent variables were shown which were the same for each respondent. It was chosen to present the survey in Dutch, since the respondents are all Dutch it would be easier for them to interpret the questions and have more reliable results. The construct questions based on prior literature were translated via the backward-translation method to enhance internal validity. This resulted in a few adjustments of the Dutch translation for the final version of the survey. The text in the Instagram posts was displayed in English to enhance reality as much as possible since most bloggers and celebrities post on Instagram in English.

3.1.2 Pre-test and Manipulation Check

(22)

22 influencer has to be added since it is difficult to give answers when the respondents do not know the type of influencer. To ensure the quality of the dataset, even more, the manipulations have to be checked to see if they have been successfully received by all respondents in the database to determine that the results are reliable for the analysis. Therefore, manipulation checks were conducted by adding several manipulation questions to the survey. The manipulation check questions were all measured on a 5-point Likert scale to ensure validity (Preston & Colman, 2000). The manipulations of the material per independent variable and their manipulation check questions are explained below. The final stimulus material of the eight conditions can be found in Appendix B. The results of the manipulation checks are explained in Chapter 4.

Brand used for the manipulation

This study uses a fictitious brand name “matte.me.cosmetics” which represents a make-up brand for women. In the Instagram post the celebrity promotes a lipstick from matte.me.cosmetics, the product is not shown in the picture, but the result of the lipstick can be seen on the lips of the celebrity endorser. Since a fictitious brand is used the participants' previous experiences or associations with a brand that could impact the brand image of awareness can be excluded (Davis, Golicic & Marquadt, 2008). According to Keller (1993), a brand name has to be familiar, characteristic and straightforward where it is recommended to focus on the important attributes or refer to the service category. The pre-test has shown that the name “matte.me.cosmetics” confirms to these assumptions.

Type of celebrity

(23)

23 these two types as traditional and non-traditional celebrities, several manipulation questions were asked in the survey: “Beyoncé/ Yara Michels is famous” – “Beyoncé / Yara Michels is known for her career” - “Beyoncé / Yara Michels is a celebrity”. The three-item scale questions were measured with a five-point Likert scale ranging from “strongly disagree” (1) to “strongly agree” (5).

Disclosure label

Evans et al. (2017) found that the words “Paid Ad” as a disclosure label were most effective in activating advertising recognition (persuasion knowledge). Hence, this study used #PaidAd as disclosure label, since this was found as the most clear and transparent language for sponsorship disclosure (Evans et al., 2017). Also, the pre-test showed that #PaidAd was interpreted correctly as sponsorship disclosure. This #PaidAd is placed at the end of the caption of the Instagram post since this is found to be most visible for consumers (Wojdynski and Evans, 2016) and is commonly used by influencers. The interpretation of the presence of a disclosure label in the Instagram post cannot be tested. Since the categorical level of disclosure is yes (included) or no (not included) there is no need to test whether the interpretation of the participants is right. The disclosure label is applied in the manipulation, or it is not applied. Therefore, no manipulation questions were asked about the disclosure labels since this could only raise more awareness of the fact that the Instagram post is an advertisement. To make sure that participants can recognize the #PaidAd better they were asked to look closely to the Instagram post and read the caption carefully.

Brand presence

(24)

24 #brandname. The manipulation for low brand presence is an Instagram post were the brand is called one time in the caption as @brandname. In order to make sure that the distinction between high and low brand presence is perceived correctly by the participants the following manipulation questions were asked: “the brand matte.me.cosmetics was high on visibility in the Instagram post” and “the brand matte.me.cosmetics was mentioned many times in the Instagram post.” These two-item scale questions were measured on a five-point Likert scale with ranging from to “strongly disagree” (1) to “strongly agree” (5).

3.2 Measures of the dependent variables

The dependent variables of the conceptual model form the constructs which are the focus of this research. These constructs will be measured via multiple-item scales which are adapted from prior studies were the scales are validated. In order to draw equal conclusions on the dependent variables all items will be measured on a seven-point Likert scale, ranging from “strongly disagree” (1) to “strongly agree (7). The list of scale items used in this research can be found in Appendix B.

Purchase intention

Baker and Churchill (1977) developed a scale to measure purchase intention which consists of four questions about the intention of the consumer to purchase a product Y of brand X. In this study, they were asked about their intention to purchase a ‘lipstick’ from ‘matte.me.cosmetics.’

Attitude towards the brand

To measure the brand attitude of the respondents the scale developed and validated by Bruner and Kumar (2000) was adapted in this study. This scale consists of six items which were measured on a 7-point semantic differential scale. In the study of Evans et al. (2017) the same scale was used and confirmed valid when measuring the effect of disclosure labels on the brand attitudes of the consumers.

Intention to spread e-WOM

(25)

25 3.3 Plan of Analysis

This study aims to find out whether the independent variables (disclosure and brand presence and type of influencer) affect the advertising effectiveness individually. Furthermore, it aims to find out if there are any interaction effects between the independent variables on advertising effectiveness. The three dependent variables (brand attitude, purchase intention and intention to spread e-WOM) are separate indicators of mobile advertising effectiveness, but for the specific statistical technique used in this study, they are combined into one construct. In order to test all these possible effects on the dependent variables the following effects need to be tested: the three main effects of the independent variables on advertising effectiveness, three two-way interaction effects on advertising effectiveness (disclosure * brand presence, disclosure * type of influencer and brand presence * type of influencer) and a three-way interaction effect (disclosure * brand presence * type of influencer). As the conceptual model includes three independent variables and three dependent variables a multivariate analysis of variance (MANOVA) is appropriate (Stevens, 2009; Laerd Statistics, 2016).

The primary purpose of a factorial MANOVA is to understand if there are interactions between multiple independent variables on multiple combined dependent variables. The depended variables are combined by taking the unweighted average score of each dependent variable, which is, in this case, the unweighted average score of three dependent variables for the construct of mobile advertising effectiveness. This statistical technique can be considered as an extension of the two-way ANOVA (which includes multiple independent variables), but then it measures for multiples dependent variables instead of just one dependent variable. It has several advantages over ANOVA; first, it protects against Type I Errors that might occur when multiple ANOVA’s were run separately. Second, it finds differences in factors which cannot be discovered by running separate ANOVA’s. Another reason why the MANOVA procedure is appropriate for this conceptual model is that the values of the dependent variables in a MANOVA are considering together as a linear composite so that comparisons can be made between groups (Laerd Statistics, 2016).

(26)

26 In order to find out which dependent variable is significant when a (multivariate) interaction effect is detected, the three-way MANOVA were followed up with univariate statistical tests (Pituch & Stevens, 2016). These tests look at the significant interaction effects for each dependent variable separately. Therefore, running multiple two-way ANOVA’s are appropriate for this conceptual model, since they measure all three independent variables on one dependent variable.

Besides the interaction effects, this study also looks at the main effects of the three independent variables on the underlying construct of the dependent variables. These multivariate main effects can be detected in the same three-way MANOVA which is run for this study. As stated above, the interpretation of these main effects can be misleading (Malhotra, 2010). In order to ensure the quality of the interpretation of the main effects, three separate one-way MANOVA’s were conducted as a security check. The results were compared with the results of the three-way MANOVA. When there is no significant difference found in this comparison, the interpretation of the main effects of the three-way MANOVA is reliable for the analysis. Then, when there are significant multivariate main effects, two-way ANOVA’s followed up to look at the univariate effects for each dependent variable separately.

(27)

27

4. RESULTS

Firstly, the sample statistics will be discussed in this chapter followed up by the manipulation checks and reliability check for the constructs of the survey. Afterward, the assumptions in order to run a multivariate test of variances (MANOVA) are specified and tested. Subsequently, when the obtained dataset satisfies these assumptions, the MANOVA can be conducted and will be analysed. The multivariate main and interaction effects will be followed up by two-way ANOVA’s to determine the effects for each dependent variable separately, which are called univariate main and interaction effects (Pituch & Stevens, 2016). Subsequently, the hypotheses will be tested.

4.1 Sample Statistics

The survey on Qualtrics was active for only four days since the minimum of respondents was achieved very fast. Within these four days, an amount of 366 respondents participated in the survey. In order to prepare the dataset for further analyses of missing values and not completed surveys were deleted. After cleaning these data records, 301 respondents remained. As it turned out five respondents need to be deleted from the dataset since there were four male participants and one 17-year-old participant. This resulted in the final amount of 296 suitable respondents for this study (N=296). These 296 respondents were equally assigned to one of the eight conditions which resulted in an average of 37 respondents per condition. This sample is large enough for the results in order to be reliable for a factorial design (Mooi & Sarstedt, 2011). This means that each condition of the research design must have a minimum of respondents which is similar to the number of dependent variables used in the design, which is, in this case, three respondents per condition. A summary of the sample size characteristics can be found in table 2 below. Since the requirements were that only females with a Dutch nationality between 18 and 30 years could attend the survey, this study looked at their education level, differences in age and their activity on Instagram.

Table 2. Age groups, education level and activity on Instagram of the sample size

Age group Education level Activity on Instagram

18-20 30% 21-25 54,3% 26-30 15,7%

Secondary education 19,8% Community college (MBO) 15,5% Polytechnic degree (HBO) 31,3% Bachelor’s degree (WO) 16,8% Master’s degree (WO) 16,5%

(28)

28 The average age within this sample is 22 years (13,3%). When looking at table 2, it is shown that the participants between 18 years and 25 years represent 84,3% of the sample size. This demographic information about the age group on this sample size can be explained by the fact that the survey was distributed within the researchers’ network.

4.2 Reliability

In this study, the dependent variables were measured on validated constructs based on prior literature. To enhance the consistent internal reliability of this research, it needs to find out whether these question items of a construct are strictly related as a group and measure the same construct. To test the internal consistent reliability, the constructs were measured on Cronbach’s Alpha.

Table. 3 The internal consistency reliability measured on Cronbach’s Alpha

For a construct to have a good level of internal consistency, the recommended values for Cronbach’s Alpha should be 0.7 or higher (DeVillis 2003; Kline, 2005). Looking at the values of Cronbach’s Alpha in the table above, it can be concluded that all constructs are found to be internally consistent.

4.3 Manipulation checks

For this study, the manipulations need to be tested to enhance validity and ensure that the manipulations are perceived as they intended to. The manipulations for the type of influencer and brand presence were tested since the manipulation for disclosure label cannot be tested.

Type of influencer

In order to find out whether the respondents perceived the traditional celebrity (Beyoncé) and the non-traditional celebrity (Yara) as intended, manipulation check questions were asked. In order to check the reliability of the construct, Cronbach’s Alpha was checked for the three items. The scale for type of influencer proved to be highly reliable (a=.817). Subsequently, to check whether manipulation was successful an independent t-test was performed on the means between the two conditions (traditional vs. non-traditional). The results show significant

Construct Items Cronbach’s Alpha

Brand attitude 6 0.934

(29)

29 differences between the traditional celebrity manipulation (Mtraditional = 4,68, SD= 0,61) and the

non-traditional celebrity manipulation (Mnon-traditional = 2,81, SD = 0,74) with p < .001. (See

Appendix C). This means that the manipulation was successful, participants perceived Beyoncé as a traditional celebrity and Yara Michels as a non-traditional celebrity.

Brand presence

In order to find out whether the respondents perceived the level of brand presence high (intensity = 5 times) and low (intensity = 1 time) as intended two manipulation check question was asked in the survey. To check if the construct is reliable, Cronbach’s Alpha was checked for the internal consistency of the items. The scale for level of brand presence proved to be highly reliable (a=.817). Next, an independent t-test was performed to check if the manipulation worked. This test showed that this manipulation was successful, since the high brand presence manipulation scored significantly higher on value of means then the low brand presence manipulation (Mhigh = 4,32, SD = 0,34; Mlow = 2,97, SD = 0,58; p < .001) (See

Appendix C). It can be concluded that the participants perceived the manipulations of brand presence as intended.

4.4 Assumptions for MANOVA

In order to run a MANOVA, there are several assumptions that need to be considered (Stevens, 2009). These assumptions are: 1) the research design consists of two or more dependent variables measured at a continuous level; 2) the research design consists of two or more independent variables were each variable consists of two or more categories; 3) the observations of the participants must be independent, there cannot be a relation between the participants in any of the groups, which is considered to be the most important assumption (Hair, Black, Babin & Anderson, 2014; Pituch & Stevens, 2016); 4) there should be no univariate or multivariate outliers; 5) the dependent variables must have a linear relation with each group of the independent variable; 6) there should be no multicollinearity between the dependent variables; 7) each of the group combinations of the independent variables must be normally distributed for all the dependent variables (multivariate normality); 8) the sample size of the dataset must be adequate; 9) there must be homogeneity of variance-covariance matrices; 10) there should be homogeneity of variances which means that the variances are equal in the conditions for each dependent variable (Laerd Statistics, 2016; Stevens, 2009).

(30)

30 Likert scale, the independent variables are categorical with two or more independent groups, and the observations are independent (in-between-subject design). The other seven assumptions need to be checked in SPSS-Statistics; these assumptions check whether the data fits the MANOVA so the statistical test can be conducted. The results of the tests which are explained below can be found in Appendix D.

First, the univariate and multivariate outliers need to be detected since they can negatively affect the results of the MANOVA (Stevens, 2009). The difference between univariate and multivariate outliers is that the multivariate outliers represent an unusual combination of scores on the dependent variables and the univariate outliers are the unusual individual scores for the dependent variables (Laerd Statistics, 2016). There were three univariate outliers in the data, as assessed by inspection of a boxplot. When testing for the multivariate outliers, the Mahalanobis distance measure is used with a critical value of 16.27 (Tabachnick & Fidell, 2014). Results show that there were two respondents above this critical value, which were the same respondents for the univariate outliers (participant: 78 and 188). Since this study has a great number of participants (N = 296), it is expected that the outliers would not significantly impact the results of the MANOVA. In order to test if these outliers do impact the results significantly, another test of MANOVA without the outliers is run. Afterwards the results were compared, and it can be concluded that the outliers have no significant influence on the data set since the conclusions for these two models are essentially the same. Therefore, it is chosen to include the outliers in the data set since it is an ethical violation to remove outliers from a data set (Stevens, 2009).

Next, the dependent variables need to be linearly related to each group of the independent variables. Otherwise, the power of the MANOVA is reduced (Tabachnick & Fidell, 2014). This relationship is tested by scatterplots for each cell in the design and showed linear results. Furthermore, there should be no multicollinearity of the dependent variables in the MANOVA. Ideally, the dependent variables should correlate with each other, but not too much otherwise there is multicollinearity (value of correlation greater than 0.9). For this dataset, there was no evidence found of multicollinearity, as assessed by Pearson correlation (|r| < 0.9).

(31)

31 eight Shapiro Wilk tests witch each significant result for each dependent variable. Therefore, a Bonferroni correction is applied to the level at which statistical significance is accepted (Pituch & Stevens, 2016). The new level at which statistical significance is accepted is 0.05 divided by the number of tests run. This means that for this study the new statistical significance would be 0.05/8 = .00625 (i.e., accept statistical significance if p < .00625). In order to have normally distributed data the values of the Shapiro-Wilk test must not be significant, then H0 will not be rejected, and no differences can be found (Stevens, 2009). The Shapiro-Wilk tests showed that brand attitude, purchase intention and e-WOM scores were normally distributed, all 24 values were greater than .00625 and therefore not significant. The results of the Shapiro-Wilk tests can be found in Appendix D.

Then, the data for a MANOVA needs to have an adequate sample size, which means that each condition must have at least as many cases as there are dependent variables (Stevens, 2009). In this study, there are three independent variables and eight conditions. To satisfy this assumption, there must be at least three respondents for each condition. Since this study has on average 37 respondents for each condition, this assumption is satisfied.

Finally, an important assumption for the MANOVA is that the variances and covariances of the dependent variables in each group combinations are similar (Stevens, 2009). In order to find out whether the group combinations of the dependent variables have the same covariance, the Box’s M test is run. The results of the Box’s Test of Equality of Covariance Matrices show that p = .029 which is higher than the significance level of p = .009. In contrast to the other statistic tests, the results must not be significant in order to satisfy the assumption for the MANOVA. Therefore, it can be concluded that there was the homogeneity of covariance matrices for the population of the current study.

(32)

32 4.5 Multivariate test of variances

A factorial MANOVA was conducted with three independent variables (disclosure, brand presence and type of influencer) and three dependent variables (brand attitude, purchase intention and intention to spread e-WOM). The dependent variables were combined by taking the unweighted average score of each dependent variable, together these scores were used to assess the construct of mobile advertising effectiveness. The MANOVA tests the main and interaction effects of the three independent variables on the combined dependent variables. Therefore, it is important to note that when a significant effect was found in the MANOVA, it is an effect on the combined dependent variables. It can be the case that this effect is only significant for one dependent variable when testing the dependent variables separately in the two-way ANOVA which is automatically run by the MANOVA. When the effect is significant in the MANOVA results, it means that the effect great enough to impact the overall construct of mobile advertising effectiveness.

The factorial MANOVA revealed two main effects on the combined dependent variables for type of influencer F (3.283) = 6,472, p < .001 and disclosure F (3.283) = 6,829, p < .001 when looking at the commonly used score the Wilk’s Lambda (Bray & Maxwell, 1985) (see table 4 on the next page). Furthermore, two interaction effects were found on the combined dependent variables for type of influencer*disclosure F (3.283) = 3,757, p = .011, and for brand presence*disclosure F (3.283) = 3,681, p = .013. These results show the effects on the combined dependent variables. This means that these effects significantly impact the underlying construct of mobile advertising effectiveness. The main effects will be further explained, and the hypotheses will be tested, followed up by an explanation of the interaction effects.

Table 4: Multivariate tests results

Effect F Sig. Partial Eta Squared Type of influencer (TI) 6,472 0,001*** ,250

Disclosure (D) _6,829 0,001*** ,068 Brand presence (BP) 1,127 0,338 ,012 TI X D 3,757 0,011* ,038 TI X BP 1,403 0,242 ,015 D X BP 3,681 0,013* ,038 TI X D X BP 1,392 0,245 ,015

(33)

33 4.6 Main effects

The MANVOA test revealed two significant main effects - type of influencer and disclosure - on the combined dependent variables (see table 4). Three separate one-way MANOVA’s were conducted as a security check for the reliability of the interpretation of the main effects (Jaccard, 1998). The results of these tests showed no differences in the statistically significance for the main effects, also here type of influencer (p < .001) and disclosure (p <.001) showed significant results (See Appendix E). Therefore, it can be said that the interpretations of the main effects for the three-way MANOVA are reliable. Since brand presence is found to be not significant in the three-way MANOVA (p = .338) and the one-way MANOVA (p = .566) no conclusions can be further made about this main effect on the dependent variables, which also means that hypotheses H2 is not supported.

In order to make more specific assumptions about the significant main effects and test the hypotheses H1 and H3, we look at univariate the effects of disclosure and of type of influencer on each dependent variable separately (Pituch & Stevens, 2016) (see table 5).

Table 5: Test of Between-Subject Effects

Variables Brand attitude

Purchase Intention

Intention to spread e-WOM Type of influencer (TI) 0.000*** 0.001*** 0.000***

Disclosure (D) 0.498 0.701 0.000*** Brand presence (BP) 0.075 0.354 0.584 TI X D 0.183 0.823 0.004** TI X BP 0.872 0.090 0.472 D X BP 0.001** 0.055 0.848 TI X D X BP 0.095 0.143 0.560

Note. Statistically Significant at: * p < 0.05 , ** p < 0.01, ***p < 0,001

4.6.1. Disclosure

The univariate test results (see table 5, also see appendix E) shows that there was a statistically significant main effect of disclosure on e-WOM score, F(1,288) = 4.846, p < .01, partial η2 = .062. Consequently, the main effect on the intention to spread e-WOM differs between whether disclosure is included (Myes = 2,12) and not included (Mno = 2,54). The mean for no disclosure

(34)

34 effect size is low, as only 6,2% of the variance (see Partial Eta Squared, table 12 in Appendix E) in intention to spread e-WOM can attributed to this main effect. But since this effect is significant in the MANOVA test, it is great enough to have an impact the combined dependent variables. Based on these results we can conclude that H1 is supported.

Table 6: Means of each condition for each of the dependent variables

Type of influencer Disclosure label Brand presence

Traditional

Non-traditional Yes No High Low Brand attitude 4,59 (0.09)* 4,15 (0.09)* 4,42 (0.09) 4,33 (0.09) 4.28 (0.09) 4,46 (0.09) Purchase intention 3,46 (0.09)* 3,01 (0.09)* 3,22 (0.10) 3,26 (0.09) 3,17 (0.09) 3,32 (0.09) Intention to spread e-WOM 1,93 (0.07)* 2,73 (0.07)* 2,12 (0.07)* 2,54 (0.7)* 2,32 (0.07) 2,35 (0.07)

Note. Statistically Significant difference between groups at: * p < 0,001 Note. Mean (Standard Error Mean)

4.6.2. Type of influencer

In order to make more assumptions about the significant effect of type of influencer on mobile advertising effectiveness, we look at the main effect for each of the dependent variables separately. The Between-Subjects Effects test (see table 5, also see Appendix E) shows that the main effect was significant for all three dependent variables. There was a statistically significant main effect of type of influencer for brand attitude score, F(1.285) = 3,745 p < .001, partial η2 = .043. The mean of the groups had to be compared in order to make more assumptions about which categorial level scored higher. This comparison as seen in table 6 show that traditional celebrity (Mtraditional = 4,59) scored higher on brand attitude compared to non-traditional

celebrity (Mnon-traditional = 4,15) with a significant mean difference of 0,439 ( p < .0001). As for

purchase intention there was a significant effect as well F(1.285) = 6,663, p = .001, partial η2 = .039. In addition, the traditional celebrity (Mtraditional = 3,46) scores higher on purchase

intention compared to the non-traditional celebrity (Mnon-traditional = 3,01) with a significant mean

difference of 0.443 (p < .0001) (See table 6).

(35)

35 the intention to spread e-WOM was more positively affected by non-traditional celebrity (M non-traditional = 2,73) compared to traditional celebrity (Mtraditional = 1,93). The results from the t-test

show a significant mean difference of .803 (p < .0001). Interesting to see is that the effect size for e-WOM (21,2 %) is much greater compared to brand attitude (4,3 %) and purchase intention (3,9 %). Accordingly, this means that H3 is partly supported. Since non-traditional celebrity has a more positive impact on e-WOM compared to non-traditional celebrities whereas 21,2% of the variance in intention to spread e-WOM can be attributed to this main effect.

4.7 Interaction Effects

The MANOVA test revealed two interaction effects (type of influencer X disclosure) and (brand presence X disclosure) which were significant for the combined dependent variables of mobile advertising effectiveness. No significant interaction effect was found for (type of influencer X brand presence) which means that H5 is not supported (p = .242). Also, no significant effect was found for the three-way interaction (disclosure X brand presence X type of influencer) which means that H7 is not supported (p = .245). The results of the MANOVA can be found in table 5 and Appendix E. Both significant two-way interaction effects will be explained further below in order to test the hypotheses of H4 and H6.

4.7.1. Type of influencer and Disclosure

To further investigate the interaction of type of influencer and disclosure on mobile advertising effectiveness the univariate main effects have to be analysed (see table 5). These showed a statistically interaction effect between type of influencer and disclosure for eWOM score

F(1.285)= 8,653 p = .006, partial η2_{= .026.}

(36)

36 main effects analysis was conducted for e-WOM score to find out what the simple main effect is of disclosure at each type of influencer.

The univariate test of the disclosure show that there was a significant difference between the levels (yes vs. no) for the non-traditional celebrity condition F(1.285) = 5,528, p < .001, but not for the traditional celebrity condition, F(1.285) = 1,333, p = .249. This means that disclosure depends on type of influencer when type of influencer is a non-traditional celebrity. We also see that there was a statistically significant difference between traditional and non-traditional celebrities for both disclosure levels (yes and no). This means that type of influencer depends on both levels of disclosure (yes = p < .001 and for no = p < .001) and the effects differ significantly when a disclosure is included or not for both types of celebrities. In order to find out which level has the highest impact and the lowest, the pairwise comparisons need to be analysed.

Note. Significant difference in means at * p < .001

Table 7 shows, that when an Instagram post contains a non-traditional celebrity and no disclosure label the effect is highest for intention to spread e-WOM. We also see here that when a disclosure is included, still the non-traditional celebrity (M=2.40) scores higher on the intention to spread e-WOM compared to the traditional celebrity (M=1.85). This can be explained by the fact that disclosure only depends on type of influencer when it is a non-traditional celebrity. An Instagram post with a non-traditional celebrity and a disclosure label included has the most negative impact on the intention to spread e-WOM (M=1.85). Since this interaction effect is great enough to represent the combined dependent variables, we can conclude that H3 is supported.

4.7.2. Disclosure and Brand presence

In order to make more assumptions about the interaction of disclosure and brand the univariate main effects have to be analysed. Table 5 shows, that there was a statistically interaction effect between disclosure and brand presence for brand attitude score F(1.288) = 9,621 p = .002. As

Table 7: Pairwise comparisons on e-WOM

Non-traditional Traditional Yes disclosure 2,40* 1,85*

(37)

37 such, a simple main effects analysis was

conducted for brand attitude score to find out what the simple main effect is of disclosure at brand presence (see appendix E for univariate tests)

The univariate tests show that disclosure is dependent on brand presence when brand presence is low (p < .008) and that brand presence is dependent on disclosure when disclosure is included (p < .002). In order to find out more about the mean differences in these interactions and their relationship the pairwise comparison table was analysed (see table 8).

Note. Significant difference in means at * p < .001

Table 8 shows, that when disclosure is included the level of low brand presence (M = 4.70) scores higher on brand attitude compared to the level of high brand presence (M =4.13) with a statistically significant mean difference of .570, (p =.002). Vice versa, we see that when brand presence is low the and disclosure is included (M =4.70) scores higher on brand attitude compared to when disclosure is not included (M = 4.23) with a significant mean difference of M = .483 (p < .006). Furthermore, we see that when a disclosure is included, and brand presence is high this has significantly the most negative effect on brand attitude (M = 4.13). When disclosure is not included, no significant differences are found between traditional and non-traditional celebrity. The highest effect on brand attitude can be achieved when an Instagram post has low brand presence and a disclosure label is included. Based on these results hypotheses H4 is supported.

Table 8: Pairwise comparisons on brand attitude

Yes disclosure No disclosure High brand presence 4,13* 4,44

(38)

38 4.8 Hypotheses

Based on the results the hypotheses were tested. Table 10 shows, that three hypotheses were supported (H1, H4, and H6). Then, H3 is partly supported since non-traditional is only significant for intention to spread e-WOM whether traditional celebrities were significant for brand attitude and purchase intention, but with relatively low effect size. At last three

hypotheses were rejected (H2, H5, H7).

Hypotheses Supported / Rejected

H1: Exposure to disclosure labels will have a negative influence on advertising effectiveness.

Supported

H2: High brand presence in an advertisement will have a negative influence on advertising effectiveness.

Rejected

H3: Non-traditional celebrity endorsements will have a more positive influence on advertising effectiveness than traditional celebrity endorsements.

Partly supported

H4. The effect of type of influencer on mobile advertising

effectiveness depends on whether disclosure is included in an Instagram post or not and vice versa.

Supported

H5: The effect of brand presence on mobile advertising effectiveness

depends on the type of influencer used in the Instagram post and vice versa. Rejected

H6: The effect of disclosure on mobile advertising effectiveness depends on

whether brand presence in an Instagram post is high or low and vice versa. Supported

H7: The interaction effect between disclosure and type of influencer on mobile advertising effectiveness depends on whether brand presence in an Instagram post is high or low and vice versa.

Rejected

(39)

39

5. Discussion

The goal of the current study was to investigate the effect of different key aspects of influencer marketing on the advertising effectiveness on Instagram. In the following section, the findings of this study will be discussed.

Main effects

The assumption that disclosure labels will negatively impact the advertising effectiveness on Instagram is supported. The finding is in line with previous research (Evans et al., 2017; Wojdynski & Evans, 2016) who showed that clear disclosure labels increase disclosure recognition and could, therefore, activate the persuasion knowledge of the consumers. Activated persuasion knowledge could raise sceptic and negative feelings towards the brand (Van Reijmersdal et al., 2015). The current study shows a significant positive effect for the ‘no’ disclosure condition on intention to spread e-WOM. Subsequently, the effect was great enough to impact the overall construct over mobile advertising effectiveness. This finding could be explained by the fact that the intention to spread e-WOM is positively affected when consumers think that the endorser shares its own opinion (Djafarova & Rushworth, 2017). Since no disclosure label is included, it can be assumed that is it not clear enough for the consumer whether the Instagram post is an advertisement or not.

Furthermore, no significant effects for the variety in brand presence on mobile advertising effectiveness were found. This gives a reason to believe that the variety in brand presence in an Instagram post, does not affect mobile advertising. This finding is in line with the findings of Krouwer et al. (2017) and could be explained due to the fact that influencers who do not cooperate with brands as endorsers on Instagram, also refer to brands in the caption of the Instagram post. Therefore, referring to a brand in native advertisement may not activate the persuasion knowledge of the consumers, since no differences were found. Also, in contrast with the findings of van Reijmersdal et al., (2015), it could be that consumers do not become more suspicious when the brand refers multiple times to itself.

#Influencer_Marketing: Essential choices to consider

#Influencer_Marketing:

Essential choices to consider

An analysis of the impact of disclosure labels, type of celebrity and

brand presence on advertising effectiveness on Instagram

#Influencer_Marketing:

Essential choices to consider

An analysis of the impact of disclosure, type of celebrity and brand

presence on advertising effectiveness on Instagram

ABSTRACT

PREFACE

TABLE OF CONTENTS

1. INTRODUCTION

2. THEORETICAL FRAMEWORK

3. METHODOLOGY

4. RESULTS

5. Discussion