UNIVERSITY OF AMSTERDAM Amsterdam Business School
Master Business Administration Track: Digital Business
How Information Transparency in Personalized advertisements improves Consumers’ Perceived Trust
An experimental study to optimize personalized advertisements for retailers given different characteristics
MSc Thesis by
Nick van der Werf 10641467
Supervisor: Prof. Feray Adigüzel Amsterdam, 21 June 2018
1
Statement of Originality
This document is written by Student Nick van der Werf, who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
2 Table of Content Abstract 5 1. Introduction 6 2. Literature Review 9 2.1 Theoretical Background 9 2.1.1 Retailers 9
2.1.2 Consumers’ Perceived Trust 10
2.1.3 Personalized Advertisements 10
2.1.4 Privacy Concerns 13
2.1.5 Information Transparency 15
2.2 Information Transparency in Online Advertisements 16
2.3 Consumers’ Perceived Trust in Retailers with Different Characteristics 17
3. Conceptual framework 19
4. Methodology 20
4.1 Research Type 20
4.2 Data Collection Method 22
4.3 Experimental Survey 22
4.4 Sampling method and sample size 31
4.5 Reliability and Validity 32
4.6 Testing Hypothesis 34 5. Results 36 5.1 Hypothesis Testing 44 6. Conclusion 49 7. Discussion 53 References 57 Appendix A 72 Appendix B 75 Appendix C 76 Appendix D 91
3
List of Tables and Figures
Table Page
1. Literature Review Summary 72
2. Conditions Experiment 23
3. Scales and Questions 29
4. Sample Statistics 39
5. Skewness and kurtosis Trust 76
6. Skewness and kurtosis Advertisement Perception 76
7. Skewness and kurtosis Privacy Concern 76
8. Normality Trust 76
9. Normality Advertisement Perception 77
10. Normality Privacy Concern 77
11. KMO and Bartlett 77
12. Factor Analysis: Variance 77
13. Factor Analysis: Trust and Advertisement Perception 78
14. Reliability Trust 78
15. Reliability Advertisement Perception 79
16. Reliability Privacy Concern 79
17. Mean, Standard Deviation and Correlation 41
18. Independent Samples T-test: Information Transparency 79
19. Independent Samples T-test: Type of Retailer 80
20. Independent Samples T-test: Novelty of Retailer 80
21. Two-Way ANOVA: Information and Type 81
22. Two-Way ANOVA: Information and Novelty 81
23. Descriptive Analysis 2: Independent T-tests and ANOVA’s 43 24. PROCESS: The effect of Information Transparency, Type
4
and Novelty on Trust 45
25. PROCESS: Model 2 84
26. The conditional effect of Information Transparency on perceived trust 48
27. PROCESS: Model 3 87 Figures Page 1. Conceptual Model 20 2. Advertisement 1 75 3. Advertisement 2 75 4. Information Transparency 75
5. Homoscedasticity and Linearity for Trust 83
6. Result Conceptual Model 47
5
Abstract
With the amount of personalized advertising continuously growing, display banners are expected to be the future of retailers’ online advertising strategy. As privacy concerns from
consumers are increasing, this can lead to a loss in the effectiveness of personalized advertisements for retailers. These retailers, given specific characteristics such as the type and novelty of the retailer, want to improve consumers’ perceived trust with information transparency provided with
personalized advertisements to increase the effectiveness of the ads. To understand the effect of information transparency, i.e. how individual data is collected and processed, the author conducted a scenario-based online experiment. To uncover the underlying mechanisms of this phenomenon, the effect of information transparency on consumers’ perceived trust was moderated by the type of a retailer, whether it had a physical store or not, and the novelty of the retailer, whether it was novel or established. The experimental results show that information transparency improves consumers’ perceived trust. Furthermore, the type and novelty of a retailer affect perceived trust. A novel retailer with both types should use information transparency to improve consumer trust. An established retailer with an online and physical store should also use information transparency to increase trust. Interestingly, for an established retailer with only an online store information transparency is not required, although it has no negative effects. From legal and ethical point of view, it is recommended for all types and novelties to provide information transparency in personalized advertisements to increase consumer trust.
6
1. Introduction
With the growing usage of the internet in daily life, retailers have more and more
opportunities to track all movements of their customers which resulted in increased understanding of consumers’ preferences and behavior (McAfee and Brynjolfsson, 2012). When retailers know more about their customers, they can offer more personalized advertisements in an online environment. In 2017, 97.1% of the Dutch population had access to the internet (CBS, 2017a) to whom 2.5 billion advertisements were displayed in 2016. Companies have been increasingly spending more of their advertising budget for online advertisements instead of traditional media. For instance, in 2017 Dutch companies spent €954 million for online advertisements (Peters, De Jager and La Verge, 2017), which nearly doubled from €570 million in advertisement spending the year before (Wiegman and Punt, 2017). Retailers have participated in this trend and invest in online advertising. For
instance, Coolblue spent most of all Dutch retailers with €13.5 million for online advertisements. According to Boerman, Kruikemeier and Zuiderveen Borgesius (2017) personalized
advertisements are the future of online advertising, which leads to providing more relevant content and greater click-through intentions for consumers (Aguirre, Mahr, Grewal, De Ruyter and Wetzels, 2015). On the contrary, personalized advertisements also evoke a feeling of dislike, loss of control and advertisement avoidance with the consumer because of privacy concerns (Boerman et al., 2017). The positive and negative effects of personalized advertisements can be referred to as the
personalization paradox (Aguirre et al., 2015). Consumers respond differently to personalized advertisements dependent on the level of consumers’ trust with the retailers (Bleier and Eisenbeiss, 2015b; Martin and Murphy, 2017), but according to Kim and Kim (2011) and Aguirre et al. (2015) information transparency improves consumers’ perceived trust. Providing transparency of
information occurs through informational cues, which can be placed in display banners to inform the consumer how the collection of data took place.
7
A previous study by Bleier and Eisenbeiss (2015b) focused on the importance of trust for personalized online advertising. They described advertisement personalization along two
dimensions, a banners depth and breadth, with consumers’ perceived trust as a moderator. This research builds further on their results regarding consumers’ perceived trust. Aguirre et al. (2015) researched the effect of information collection techniques and trust-building strategies on online advertisement effectiveness. In their research they concluded that information transparency
positively increases the effectiveness of online personalized advertisements when published on social network sites. This research adds information to the effectiveness of online personalized
advertisements when they are not published on social network sites. Shankar, Urban and Sultan (2002) and Darke, Brady, Benedicktus and Wilson (2016) studied how trust is affected by the interrelationship between online and multichannel firms and mention that future research should address online strategies that improve consumers’ perceived trust. Subsequently, Pavlou and Fygenson (2006) and Vos, Marinagi, Trivellas, Eberhagen, Skourlas and Giannakopoulos (2014) concluded that trust-building strategies are long-term goals. A gap originates here to address what the effect is of information transparency published by established retailers and novel retailers.
So, based on these discussions, the research question is formulated as: How does information transparency improve consumers’ trust towards personalized advertisements given by different retailers?
Although the concept of transparency has gained popularity in the research field, about the relationship between information transparency in personalized advertisements and consumers’ perceived trust is still much to discover (Aguirre et al., 2015; Bleier and Eisenbeiss, 2015b). This research builds on the personalized advertising literature and theory on trust and online marketing is added. More concretely, this study focuses on the effect of personalized advertisements including information transparency on consumers’ perceived trust moderated by several characteristics of the retailer.
8
With an increase of 21% of retailers and 500% increase of online retailers in the past ten years, Ogonowski, Montandon, Botha and Reyneke (2014) argue that consumers’ trust in a retailer is essential for the continuity of that retailer. Furthermore, with the introduction of the General Data Protection Regulation (GDPR) companies are required to inform consumers how their data is collected and why (Tankard, 2016; Autoriteitspersoonsgegevens, 2017). One way for retailers is to inform the consumers through information transparency forms attached to personalized
advertisements. Therefore, it is essential for retailers to understand the effect of such forms on consumers’ perceived trust towards the retailer. And as consumers’ perceived trust is dependent on the type of a retailer (online vs multichannel) and the novelty (established vs novel), the effect of information transparency will be studied retailers depending on different characteristics.
To conclude, this study will contribute to managerial practice how consumers’ perceived trust can be improved through information transparency in personalized advertisements. Retailers that publish personalized display banners will understand if it will be beneficial to inform consumers how their data is collected and used, which is required by law since May 25, 2018. If information
transparency will be beneficial is measured through consumers’ perceived trust. Furthermore, this information will be applicable to a wide variety of retailers, where this study focuses on online versus multichannel retailers and novel and established retailers.
Next, there are some delimitations to this study. First, the goal of this research is to understand how information transparency in personalized display banners affects consumers’ perceived trust. This form of online advertising is displayed through the Google Display Network (Google Partners) and therefore not displayed on social networks such as Facebook. The Display Network provides four possibilities to advertise, namely text ads, image ads, rich media ads and video ads. In this research there will only be a focus on image ads. As consumers can respond to online advertising with a wide variety of behavioral aspects, there will only be a focus on consumers’ perceived trust. Finally, the advertisements will be displayed by retailers, where the retailers are
9
characterized by the type and the novelty. No other characteristics will be focused on in this research. This research will provide a relevant overview of the literature in the next chapter. Second, after formulating the hypotheses a conceptual model is presented. Next, the research design and methodology of the study are described. The results of the experiments are introduced in the fourth chapter. Fifth, the conclusions are explained in further depth and the final part of this thesis is the discussion, which contains the managerial and scientific implications, the limitations and further research suggestions on this topic.
2. Literature Review
This paper studies how personalized advertisements affect consumers’ perceived trust in retailers and how information transparency in these personalized advertisements can improve the effectiveness of personalized advertisements dependent on several characteristics of a retailer. To study this matter, the most important concepts must be researched in more depth. Based on earlier literature research and current market research this study will concentrate on: (1) An overview of different retailers; (2) Consumers’ perceived trust; (3) personalized advertisements; (4) Privacy concerns; and (5) Information Transparency. See table 1., Appendix A for all sources and constructs). Finally, the hypotheses development is explained.
2.1 Theoretical Background
2.1.1 Retailers
A retailer can be defined as a firm that sells products directly to the consumer (Business dictionary, 2018). Herhausen, Binder, Schoegel and Herrmann (2015) define three types of retailers: first the retailer with only physical stores such as Action; the second type is the purely online retailer like Wehkamp and Bol.com; and finally, the third type of retailer is the multi-channel retailer, who combines brick-and-mortar stores with web shops, such as Coolblue and Fietsenwinkel.nl. In 2016, there were 95,630 physical stores and 32,170 web shops in the Netherlands (CBS, 2017b). The
10
number of web shops grew with more than 500% in the past ten years, although 7,000 web shops were liquidated in 2016 (Creditsafe, 2016). According to Ogonowski, Montandon, Botha and Reyneke (2014), the main reason for failure and this high number of liquidations of stores is due to consumers’ lack of trust.
2.1.2 Consumers’ Perceived Trust
Trust is key in the relationship between retailer and consumer (Morgan and Hunt, 1994) and is the belief that the retailer will perform actions with positive outcomes as result and will not perform unexpected actions with negative outcomes as result (Anderson and Narus, 1990). Gefen, Karahanna and Straub (2003) explain trust as the consumer’s expectations that the retailer will take no advantage of the situation and does not take an opportunistic attitude, but will behave ethically, dependable and socially accepted to accomplish the commitments despite the vulnerability and dependence of the consumer. According to Verma, Sharma and Sheth (2016) the most important effect of consumers’ perceived trust is building and maintaining a relationship between consumer and retailer. Trust in the retailer leads to purchase loyalty and attitude loyalty, which in turn has a positive effect on market share (Chaudhuri and Holbrook, 2001). And Sung and Kim (2010) state that trust creates brand loyalty. According to Suh and Han (2003), Pavlou and Fygenson (2006) and Vos et al. (2014) building trust is the long-term goal for retailers, but advertising can change
consumers’ attitudes towards retailers and can have both positive and negative effects on consumers’ perceived trust (Petty, Cacioppo and Schumann, 1983; Kopalle and Lehmann, 2006; and Pappas, 2016).
2.1.3 Personalized Advertisements
In general, marketing can be divided into two types, traditional and online marketing. Retailers must manage these two types simultaneously and make strategic decisions in allocating budgets across these different types to optimize the effectiveness of advertisements (Dekimpe and
11
Hanssens, 2007; Lehmann, 2004). Traditional advertising can be explained as mass marketing through traditional sources such as radio and television (De Haan, Wiesel and Pauwels, 2016). For online marketing, multiple forms of advertising have been given rise since the introduction of the internet. Generally, advertising can be explained through the concept of firm-initiated advertising and consumer-initiated advertising. Consumer-initiated advertising is triggered by consumers and examples are organic and paid search, price comparison sites, referrals and retargeting. This form of advertising is considered more effective and less intrusive than firm-initiated advertising, where the retailer uses a push strategy to advertise. Examples of firm-initiated advertising are the offline forms such as television and radio and the online forms, such as email and display advertisements (Li and Kannan, 2014). An advantage of online advertising in comparison with offline advertising is the availability of individual data from consumers.
Consumers share their data through social media, when they access information or ask personal questions via search engines, when they share photos in the cloud and when they
communicate through email and texts. Nowadays, the internet is accessible by everyone and through smartphones even 24 hours on-the-go. Through all these activities consumers share personal
information, wittingly or unwittingly, with third parties (Acquisti, Brandimarte and Loewenstein, 2015). Retailers use this information to optimize the customer journey but also to provide tailored, location-sensitive and time-sensitive advertisements (Lemon and Verhoef, 2016). This tailored form of advertising leads to lower costs for online marketers because they can deliver more relevant messages to more specific consumers. Targeting consumers results in a more efficient way of advertising and provides the consumer with advertisements in the right context (Malthouse and Li, 2017). For instance, targeted advertising leads to less advertisement irritation and advertisement avoidance (Ham, 2017) and enhances positive responses like improving advertisement credibility and attitude (Tran, 2017).
12
with personalized advertisements is called programmatic advertising, where the buying, placement and optimization of advertising is automated (Solomon, 2015). Market research conducted by
Nielsen (Wiegman and Punt, 2017) revealed that of the sold advertising space in 2016 71% was used for programmatic advertising. The individual data and corresponding consumer demographics are gathered through overt and covert collection techniques by online publishers such as Google and Yahoo according to Aguirre et al. (2015). When consumers are aware that their information is gathered, for example through cookies, is an overt collection technique. Covert collection techniques are used to gather unbiased data, when consumers are not aware of the data collection. Both these forms lead to better customer understanding and the possibility to refine and tailor advertisements, which results in personalized advertising. Advertisements can be delivered as banner advertisements and videos (Taylor, Lewin and Strutton, 2011) on social network sites and on other websites such as news, entertainment and retailer sites.
Statista, one of the leading providers of market and consumer data worldwide, researched the global banner advertising market. For 2017, they found that the market size is $45.5 billion in 2017 and display banners have a share of 19% in the online advertising market. And according to Google (2018) display banners reach over 90% of the internet population. The placement of banners that consumers are exposed to, can reveal different preferences of the consumers (Bleier and Eisenbeiss, 2015a). As explained earlier through Tucker’s (2014) research, social network sites are data-rich because consumers entered their personal information and consumers have control over their privacy settings. Therefore, a distinction is made between social network sites and the internet without social network sites.
Display banners that are published on the internet are personalized through retailers engaging with advertising agencies (Bleier and Eisenbeiss, 2015a). For each consumer a unique profile is created and linked to a cookie that is placed on the consumer’s device. Through these cookies consumers can be tracked online and his or her online behavior is stored in their profile. Advertising
13
agencies tap the profile of consumers and determine if the profile of the consumer matches the content or goal of the retailer’s advertisement. When the advertisement and the profile match, the banner will be delivered to the consumer’s device.
According to Goldfarb and Tucker (2011b) advertisements with general content are less effective than advertisements tailored for specific consumers. Personalized advertisements not only improve consumers’ purchase intent but also cause more noticeability (Malheiros, Jennett, Patel, Brostoff and Sasse, 2012) and from a consumer perspective it requires minimum effort because retailers will identify and satisfy the consumers’ needs (Montgomery and Smith, 2009). Malthouse and Li (2017) clarify the negative effects of personalized advertisements and state that a fine line exists between impressing the consumer with cautiously targeted and relevant advertisements and violating consumers’ perceived trust with too personalized information. The negative effects are increasing privacy concerns, decreasing purchase intentions and lower click-through rates, feelings of ad avoidance, dislike and loss of control (Boerman et al., 2017; Aguirre et al., 2015; and Pappas, 2016) To these positive and negative effects of personalized advertising can be referred to as the personalization paradox. With a better understanding of this personalization paradox, Goldfarb and Tucker (2011a) explain the decline in advertising effectiveness of display banners. The reason for the failure of banners is a combination of obtrusiveness and targeting. They state that privacy concerns drive the negative performance due to a loss in trust in the retailer (Eastlick, Lotz and Warrington, 2006). According to Aguirre et al. (2015) transparency about data collection can restore the loss in trust and provide the consumer with a feeling of more control over their privacy.
2.1.4 Privacy Concerns
Malhotra, Kim and Agarwal (2004) explain privacy concern as an individuals’ subjective view of fairness to determine for themselves when, how and to what extent information about themselves is collected, used and transferred to others (Westin, 1967; Campbell, 1997). Privacy
14
concerns have been a major concern of the “information age”, where consumers are constantly exposed to tradeoffs between all sort of online benefits and privacy intrusion (Norberg, Horne and Horne, 2007). Awad and Krishnan (2006) studied such tradeoffs in further depth and concluded that privacy invasion is accepted when the potential benefits outweigh the potential risk of privacy invasion. For personalized services, the privacy invasion is well accepted, for personalized
advertisements less. This is due to the case that privacy invasion is more apparent and the benefits less for those forms of advertisements. According to Eastlick et al. (2006) privacy concerns have a strong and negative effect on consumers’ perceived trust regarding online advertising. In the
contrary, Acquisti et al. (2015) explain that privacy concerns by individuals are usually not lived up to in their daily life. This discrepancy between attitude and behavior is referred to as the privacy paradox.
Certain behavior towards privacy concerns can most easily be studied through an individual’s privacy settings on social networks such as Facebook. Facebook is probably one of the best-known examples where consumers outweigh the benefit against the privacy risks (Debatin, Lovejoy, Horn and Hughes, 2009). A high vulnerability exists on social networks, not only is a lot of demographic information visible for others, also a lot of personal information is posted on these social networks. Consequently, this causes high risks for someone’s privacy (Ibrahim, 2008). Research from Dwyer, Hiltz and Passerini (2007) found a very weak relationship between one’s privacy concerns and the privacy settings at social networks.
As the privacy paradox refers to the attitude and subsequent behavior regarding general privacy concerns, Li, Edwards and Lee (2002) argue that it is not only important to study the general privacy concern, but also the perception of individual advertisements. Ying, Korneliussen and
Gronhaug (2015) add here, that there will always be a negative perception of advertisements, but that an acceptable and optimal middle ground can be found regarding the consumer’s experience.
15
influence consumers’ perceived trust, which in turn has a strong effect on behavioral intentions towards a retailer (Liu, Marchewka, Lu and Yu, 2004).
2.1.5 Information Transparency
Westin (1967), Stone, Gueutal, Gardner and McClure (1983) and Awad and Krishnan (2006) describe privacy as an individual’s ability to control information about one’s self and how this information is acquired and used. Concerns about privacy control has been one of the most important ethical issues since the rise of the internet (Smith, Milberg and Burke, 1996; and Awad and
Krishnan, 2006). Using personal information is a critical success factor for firms in an online
environment, firms must create online experiences where they handle the privacy of consumers very careful and minimize consumers’ perceived feeling of privacy invasion. Hajli and Lin (2016) state that perceived control over one’s privacy is more important for consumers than actual control and can be understood as the amount of control consumers have over a situation. According to Xu (2007), perceived control is the key factor explaining privacy concerns, and when firms fail in this mission to protect consumers’ perceived control, this leads to lower trust. Miyazaki (2008) and Martin and Murphy (2017) state that lower trust due to privacy concerns result in a decrease of behavioral intentions and negative word of mouth.
Consumers lower trust caused by a perceived loss in control over information results in negative responses to firms and the firms collection and use of data. As firms increasingly emphasize and use personal information to increase the effectiveness of personalized advertisements, Martin, Borah and Palmatier (2017) demonstrate that transparency in data collection can counteract the negative effects of a perceived loss in control of information. Information transparency in personalized advertisements occurs through showing informational cues in display banners that openly inform the consumer how the data collection took place (Aguirre et al., 2015). These social responsible messages are ethical and inform the consumer by providing information about the
16
collection that is often difficult to obtain (Turilli and Floridi, 2009). 2.2 Information Transparency in Online Advertisements
Bleier and Eisenbeiss (2015b) concluded that trust is the most important factor affecting consumer behavior and intentions for online advertisements. The level of consumers’ perceived trust is even more important when online advertisements are personalized banner displays. Personalized display banners cause that consumers experience a decline in perceived control of their privacy (Awad and Krishnan, 2006). This decline leads to lower perceived trust and therefore negative behavior from the consumer. Malthouse and Li (2017) also state that personalized advertisements cause negative effects in consumers’ perceived trust and Milne, Bahl and Rohm (2008) suggest that uninformed exposure result in negative reactions for consumers.
To improve this negative behavior, the use of information transparency positively affects consumers’ perceived trust in personalized advertisements (Diallo and Lambey-Checchin, 2017). Information transparency in personalized display banners in an online environment explains how the personal information from an individual consumer is collected. By informing consumers how the collection of data took place through informational cues improves consumers’ perceived trust (Kim and Kim, 2011). Schnackenberg and Tomlinson (2014) state that transparency has a positive effect on consumers’ perceived trust in an organization and explain transparency as a perception of the quality of intentionally shared information from the sender. Albu and Flyverbom (2015) add that the disclosure of information towards consumers, when the information is qualitative, increases
transparency. The increase in transparency leads to clarity and understandability for the consumer, which in turn increases perceived trust. Besides that, the practice of information transparency is ethical (Martin and Murphy, 2017), to provide information transparency in display banners that are personalized is recommended for all retailers because it increases consumers’ perceived trust (Bleier and Eisenbeiss, 2015b; Aguirre et al., 2015; Morey, Forbath and Schoop, 2015). This results in the following hypothesis:
17
H1: Information transparency in personalized advertisements leads to an increase in consumers’ perceived trust compared to personalized advertisements without information transparency. 2.3 Consumers’ Perceived Trust in Retailers with Different Characteristics
Of all the sectors in the economy, retailers have the highest spending on personalized display banners (Hajli, 2015). To benefit from this advertising strategy, retailers are dependent on the
perceived trust from consumers. According to Aguirre et al. (2015) personalization of online advertising is not always an effective strategy and retailers must carefully analyze if this form of advertising is the best option. Herhausen, Binder, Schoegel and Herrmann (2015) defined that there are three types of retailers. First and second, the brick-and-mortar retailer and the multi-channel retailer who both possess physical stores that consumers can visit. And third, the purely online retailer that does not possess a physical store. Li, Jiang and Wu (2014) emphasize on the importance of consumers’ perceived trust in retailers, where they state that consumers perceive retailers with a physical store with more trust compared to retailers without a physical store. Darke et al. (2016) support the argument that the lack of a physical store increases perceived risk. Martin and Murphy (2017) conclude that consumers perceive a retailer with a physical store with more trust because consumers feel that they are fewer steps removed from the retailer as they can visit the store.
Besides the importance of the possibility to visit a retailer with a physical store, consumers’ perceived trust is influenced by the level of novelty of a retailer. Darke et al. (2016) state that novel retailers face more distrust from consumers than retailers that exist for a longer period. Trusov, Ma and Jamal (2016) explain that every retailer can access consumers’ individual data and target
consumers with personalized banner displays. Novel retailers that expose consumers to personalized display banners with exceeding personal information are perceived with less trust by retailers
compared to longer existing retailers. Consumers experience personalized advertising by novel retailers as unethical and perceive the retailer as less trusted (Diallo and Lambey-Checchin, 2017). Furthermore, creating trust through data privacy practices is not a short-term approach. Retailers
18
must prove that they protect the personal information from individuals and to build trust takes time (Martin and Murphy, 217). Not only is there fewer information available about how novel retailers protect privacy, there is less information available about novel retailers in general (Jang, Chang and Chen, 2015). Consumers that have limited knowledge about a retailer and are unable to find much information about a retailer online results into declining perceived trust towards the retailer when exposed to personalized display banners. This results in the following hypotheses:
H2a: Personalized advertisements lead to a higher level of perceived trust for the consumer when a retailer has a physical store compared to a retailer without a physical store.
H2b: Personalized advertisements lead to a higher level of perceived trust for the consumer when a retailer exists for a long period compared to a retailer that is novel.
Lower concerns about individuals’ personal data increases one’s perceived trust, but Laroche, Yang, McDougall and Bergeron (2005) conclude that although informational cues increase
consumers’ perceived trust towards a retailer without a physical store, the level of perceived trust will still be higher when a retailer has a physical store. Furthermore, Martin and Murphy (2017) explain that retailers who involve their customers in the information privacy dialogue through open and transparent communication, will result in a positive performance. Although they conclude that this sharing practice will only have a positive effect on the long-term. Therefore, longer existing retailers are expected to experience more perceived trust by consumers compared to consumers’ perceived trust when the retailers are novel. This results in the following hypotheses:
H3a: With information transparency, personalized advertisements lead to a higher level of perceived trust for the consumer when a retailer has a physical store compared to a retailer without a physical store.
H3b: With information transparency, personalized advertisements lead to a higher level of perceived trust for the consumer when a retailer exists for a long period compared to a novel retailer.
19
3. Conceptual framework
In the retail industry three types of retailers exist, namely offline, multichannel and online retailers. The type of a retailer, where there is a possibility to visit a physical store or not, and the novelty of a retailer explain the level of trust in the retailer through personalized advertisements, which is important to understand before the effect of exposure to personalized advertisements can be studied, as can be seen in figure 1. Retailers with physical stores, offline and multichannel, are expected to have a higher level of trust than retailers without physical stores. Retailers that exist longer are expected to have a higher level of trust than new online retailers. Apart from the characteristics of a retailer, personalized advertisements cause a decline in consumers’ perceived trust in general. Therefore, the effect of information transparency in display banners is studied. Informational cues have a positive effect on consumers’ perceived trust in personalized online advertisements.
When advertising online, privacy concerns are a major issue. When retailers advertise online through personalized display banners consumers can perceive lower trust towards the retailer.
Therefore, it is important to understand the effect of information transparency in personalized display banners on the level of trust perceived by customers, as can be seen in figure 1. Next, there are retailers that differ from each other for specified characteristics. First, the type of retailer is
important. In the retail industry a distinction can be made between three different types, namely the purely offline, purely online and multichannel retailer. Retailers with a physical store, thus offline and multichannel retailers, are expected to be perceived with more trust by consumers. Second, the novelty of a retailer is important. Long existing retailers are expected to be perceived with more trust by consumers than novel retailers. For both the type and the novelty of a retailer is studied how significant the influence of information transparency in personalized display banners is.
20
Figure 1. Conceptual Model
4. Methodology
To test the hypothesis about the effect of information transparency in personalized display banners on consumers’ perceived trust and the moderating effect of a retailer’s characteristics, namely type and novelty, an online experiment is conducted. First the type of research is explained and why this type best fits this research. The method used for the data collection is discussed in the next section. Third, the experiment design and a detailed description of the experiment are given. Next, the details about the sample are provided, including the sampling method and sample size. The evaluation regarding the reliability and validity of the experiments are discussed next. And finally, the testing of the experiments is discussed.
4.1 Research Type
To study consumers’ perceived trust when exposed to personalized advertisements, it is common to conduct scenario-based online experiments (Bleier and Eisenbeiss, 2015b; White, Zahay,
21
Thorbjørnsen and Shavitt, 2008; Auschaitrakul and Mukherjee, 2017). To make inferences about trust, this study is quantified. First, open questions about trust are vague, abstract and hard to interpret and according to Van Tulder (2007) it fits best to study psychological effects, such as the effect of information transparency. Second, a primary point of this study is that quantitative measurements of trust can be analyzed into a hierarchical variation between groups of consumers exposed to different conditions (Glaeser, Laibson, Scheinkman and Soutter, 2000). Furthermore, to prove a causal relationship between the use of information transparency and consumers’ perceived trust when exposed to personalized banner displays, experiments fit this research best (Druckman and Kam, 2011).
In an online environment, especially when consumers are exposed to personalized display banners, trust in the retailer is essential for the retailer to be successful (Vos et al., 2014; Pappas, 2016; Verma et al., 2016). The dependent variable of this study was consumers’ perceived trust in retailers. To test consumers’ perceived trust, the presence of information transparency in
personalized display banners was manipulated. Information transparency was the independent variable. Personalized advertisement banners use individual data that are expected to negatively affect consumers’ perceived trust. By including information transparency about how data was collected with the advertisements, consumers’ perceived trust can be improved (Bleier and Eisenbeiss, 2015b; Aguirre et al., 2015). The type of a retailer, whether it has physical and online stores or only online stores, and the novelty of a retailer are important characteristics of a retailer which might affect consumers’ perceived trust (Darke et al., 2016; Martin and Murphy, 2017). These two variables are moderated as they are expected to affect the nature of the relationship between the independent and dependent variable (Saunders, Lewis and Thornhill, 2012). Furthermore, individual privacy concerns (Liu, Marchewka, Lu and Yu, 2005) and personalized advertisement perception (Li, Edwards and Lee, 2002) are controlled for. These variables might influence the effect of the independent variable on the dependent variable and should be controlled for (Saunders et al., 2012)
22
and improve the precision of the experiment (Lewis and Rao, 2015). Finally, participants might have had predetermined opinions towards retailers and to avoid natural skepticism (Ichikawa and Jarvis, 2007). For this reason, fictitious retailers and advertisements were used.
4.2 Data Collection Method
The data collection method of this study is conducted with an experimental survey where information transparency in personalized display banners is manipulated. The manipulation of information transparency occurs through the absence or presence of a transparency form attached to the advertisement. Furthermore, the characteristics of the retailer are moderated. The use of an experimental survey approach is common in the domain of ad personalization where effects for different combinations are investigated (Fisher and Dubé, 2005; Goldfarb and Tucker, 2011a; Bleier and Eisenbeiss, 2015b). According to Koschate-Fischer and Schandelmeier (2014) this method is especially suitable when there is only one independent variable which can be systematically varied, which is in this study information transparency. The reason for an online experiment is that
participants will remain in their natural environment which is most familiar to the participants (Reips, 2002). Consumers normally are not browsing the internet in a lab and should therefore be in a location that is familiar for them to fill in the survey.
4.3 Experimental Survey
To answer whether information transparency effects consumers’ perceived trust in
personalized advertisements moderated by type and novelty of a retailer, the design of this scenario-based experiment is a 2 (no information transparency vs information transparency) x 2 (physical and online store vs online store) x 2 (established vs novel) mixed factorial design. The conditions of this design are depicted in table 2. First, a factorial design provides the possibility to manipulate more than one variable at a time, which are in this study the transparency, type and novelty (Blumberg, Cooper and Schindler, 2011). The participants were presented a brief and explicit description of the situation which contained the independent variable and moderators. Next, the participants had to
23
provide information on how they perceived trust towards the retailer, which was the independent variable. To quickly reach a sample size with sufficient statistical power, a mixed factorial design was developed. According to Blumberg et al. (2011) for the results of a factorial design, where the same participant responds to multiple situations, needs to be taken into account that responses from the same participant are not independent from each other. Furthermore, another critique to a mixed-factorial design is that they do not respond to reality. To preserve the results from being affected by these critiques is chosen to investigate only two situations per respondent and for an online survey. This was considered a reflection of reality, where consumers see multiple online display banners per day.
Table 2. Conditions Experiment 2x2 Design No information
transparency
Information transparency
Physical store 1. Long existing retailer 2. Novel retailer
1. Long existing retailer 2. Novel retailer
No physical store
1. Long existing retailer 2. Novel retailer
1. Long existing retailer 2. Novel retailer
This mixed factorial design was developed with between-subject factors and within-subject factors. A mixed design explains that each group of participants receive different versions of the advertisement but within groups each participant receives the same version (Aguinis and Bradley, 2014). The between-subjects factors were information transparency and the type of retailer. Which means that each of the four groups was presented with a different combination of the presence of information transparency or not in an advertisement and the type of retailer, if the retailer has a physical store or a physical and online store. The within-subjects factor was the novelty of a retailer. Within each group, the participant saw an advertisement of a retailer that is novel and an
24
and Aguinis and Bradley (2014) conclude that if the participants are provided with enough background and contextual information responses accurately reflect the true judgements of the participants.
To measure the main effects and interaction effects, block randomization was used for this experiment. First, to conduct a true experiment, randomization guarantees that the characteristics of participants assigned to the control group are equal to the characteristics of the participants assigned to the experimental groups (Blumberg et al., 2011). The randomization of blocks was used to assure that an equal number of participants were assigned to each group (Efird, 2010) and to isolate and remove sources of information that might vary substantially between participants. According to Kirk (1982) the use of block randomization is considered a far more effective method to control for the heterogeneity between participants than complete randomization.
This experiment was conducted as an online scenario-based experiment. First, the experimental setting had to be controlled and manipulation of the independent variable and moderators was necessary. This is a common approach in this field of research (Bleier and Eisenbeiss, 2015b; Walrave, Poels, Antheunis, Van den Broeck and Van Noort, 2016), especially when there is no access to data from retailers who fit the characteristics. Controlling and being able to manipulate the variables was very important for this study to understand how consumers perceive personalized advertisements dependent of the manipulations, although the participants being aware that they were part of an experiment could be interpreted as a disadvantage (Blumberg et al., 2011). Therefore, an online experiment was chosen where participants took part in their natural
environment.
Before carrying out the survey a pretest was conducted. First, a pretest was conducted with 19 participants (male n = 11, female n = 9, age mean = 31.95) to obtain which kind of display banner was perceived as most personalized using a five-point Likert scale (lower values indicate higher
25
personalized ad perception) and if the participants believed that retailers can collect the presented personal information to create advertisements. Advertisement 1 was personalized for occupation (mean = 2.50, st.dev = 0.94), advertisement 2 was personalized for occupation and geographical location (mean = 2.06, st.dev = 0.56), advertisement 3 was not personalized (mean = 3.81, st. dev = 0.53) and advertisement 4 was personalized for occupation and the location the consumer was interested in (mean = 1.63, st.dev = 0.93). As a result of this process, a display banner advertisement was developed based upon affinity information retrieved from a personal conversation at a social media network including location of interest and occupation (e.g. advertisement 4). In this pretest, a second objective was tested to obtain information about consumers’ concerns towards the novelty of a retailer. As a result of this process, we concluded that a retailer is not novel when it is 10 years old (mean = 5.47 years old, max = 10 years old and min = 2 years old). A one-year old retailer was interpreted as novel by every participant.
The experimental survey consisted of 8 sections and was previewed with several operation systems to emphasize a clear design of the survey. The first part of the survey was an introduction. This introduction included a covering letter and summary of the main message. For online surveys it is required to stress that participation is voluntary, and confidentiality is assured (Saunders et al., 2012). In addition, the main message included the objective stated as: “Consumers’ reactions to personalized advertisements” to not give away the precise goal to prevent participants from getting biased (Dillman, Smyth and Christian, 2009).
In the second section the privacy settings of the participants were asked, which was the first control variable of this experiment. All the variables and accompanying scales and sources can be found in Table 3. The Privacy setting was measured with one item adapted from Hofstra, Corten and Van Tubergen (2016). Based on the social network they visited most frequently, the participants’ private information could be accessible to: “friends only”, to “friends and their friends”, to “public” or they “don’t know”. The subject of the personalized advertisement is a camera. For this reason, the
26
participants were also asked if they did search for a camera in the past five years.
In the next section, a cover story informed the participants that they had to imagine
themselves as a 23 years old student from Amsterdam. The previous week, they had a conversation with a friend through Facebook Messenger on their phone. They were discussing a trip planned to Paris, but they were arguing if they had to take a camera with them. Therefore, the participant explored the internet about some new camera models. During the experiment, the participant allegedly visited a news website on a laptop. There, the participant encountered a display banner. Depending on the treatment group, this banner featured an advertisement from CameraNow with or without data collection information. Furthermore, the participant was explained what the type and novelty of CameraNow was. For a screenshot of the advertisement, see Appendix B (figure 2.). For a screenshot of the informational cue, see Appendix B (figure 4.).
Furthermore, in this section participants also had to fill out a questionnaire. Its measures included consumers’ perceived trust towards the retailer and the perceived ad perception of the advertisement. Trust was measured with six items: “I think that this retailer can be relied on to keep its promises,” adapted from Verhoef, Franses and Hoekstra (2002) and Buttner and Gortitz (2008); “I thinks that this retailer will provide a good service,” adapted from Verhoef et al. (2002); “I think that this retailer will treat me fairly if problems arise,” adapted from Buttner and Goritz (2008); “I think that this retailer has high standards by which it advertises,” adapted from Buttner and Goritz (2008; “I think that this retailer will not misuse my personal information,” adapted from Bansal and Gefen (2010) and; “My overall confidence in the promise of this advertisement is high,” adapted from Bart, Shankar, Sultan and Urban (2005). The corresponding scale was a seven-point rating scale that ranged from 1 (strongly disagree) to 7 (strongly agree). Perceived trust has been measured in multiple studies, but scales for perceived trust towards retailers in online advertisements that fit this research could not be found. Therefore, this new scale was developed based on collecting items from multiple previous scales used to measure perceived trust.
27
Ad perception was measured with 7 items from, and validated by Li et al. (2002), Goodrich, Schiller and Galletta (2015) and Ying, Korneliussen and Grønhaug (2015): “I find this advertisement disturbing,” “I find this advertisement obtrusive,” “I find this advertisement interfering,” “I find this advertisement invasive,” “I find this advertisement forced,” “I find this advertisement intrusive,” and “I find this advertisement distracting.” The corresponding scale was a seven-point rating scale that ranged from 1 (strongly disagree) to 7 (strongly agree).
To check for the understanding of the manipulation by the participants and to control for attention two attention check questions were asked. These questions were used to make sure that participants were paying enough attention to these tasks which are highly valued (Hauser and Schwarz, 2015). Furthermore, by asking these questions in a new block was to make sure that participants could not override their original response (Goodman, Cryder and Cheema, 2013). First, participants were asked if they could replicate the age of the retailer and second, if they could replicate the type of the retailer.
In the fifth section, a distraction question was asked to change the participants mind: “Can you state how many times you went on holiday last year?” This was used to avoid earning effects, which could occur because similar questions were asked (Jimenez and Mendez, 1999). Taking a break from the survey was not possible since it was an online survey. For this reason, a distraction question was asked. Furthermore, to not decrease the quality of response for the remaining questions, the length of this question was short (Galesic and Bosjnak, 2009; Auspurg, Hinz and Liebig, 20).
Next, the second advertisement was displayed. For a screenshot of the advertisement, see Appendix B (Figure 3.). The participant allegedly visited the same news website the next day, where the participant was exposed to an advertisement from CameraToday. This retailer had the same characteristics and a comparable advertisement except that it was novel in comparison with the previous retailer. In this section the participant had to fill out the same questionnaire as in section
28
three. Trust and advertisement perception were measured.
Seventh, demographic questions, privacy concern questions, the manipulation check question and debrief questions were asked. All these questions were asked at the end to minimize order effects (Downes-Le Guin, Baker, Mechling and Ruyle, 2012). Demographic questions were measured with 2 items (Shavitt, Lowrey and Haefner, 1998): “What is your age,” and “What is your gender.” Finally, the participant was asked to provide his thoughts about the goal of this experiment. This was measured with one item: “What do you think that the goal of this survey is?” Participants could answer: “To understand if consumers trust retailers if they show personalized advertisements,” “To understand which kind of personalized advertisement consumers see as most personalized,” “To understand if the characteristics of retailers affect consumers’ trust when exposed to personalized advertisements,” “To understand how information transparency in personalized advertisements affects consumers’ trust,” or the participants could answer “other” and fill in their own opinion about the goal of the research.
The manipulation check was added to control if all the participants perceived the advertisement as personalized. According to Oppenheimer, Meyvis and Davidenko (2009),
manipulation checks increase statistical power and the reliability of data and are a tool for detecting participants who are not following instructions. The question that was asked to check for
manipulation was: “I think that the advertisement I was exposed to is personalized for:”. The
possible answers, where multiple could be selected, where: “Demographics, such as age, gender and occupation; The location you are interested in; The context of the website you visit; A product you are interested in; A product you searched for through a search engine (such as Google) and;
nothing”. If the participant would answer “nothing”, the participant would be excluded from the dataset.
29
(2004), and validated by Angst and Agarwal (2009) (Cronbach’s alpha between 0.72 and 0.93): “It usually bothers me when retailers ask for personal information,” “It bothers me to give personal information to so many retailers,” “I’m concerned that retailers are collecting too much personal information about me,” “Consumer privacy is a matter of consumers’ right to exercise control and autonomy over decisions about how their information is collected, used and shared,” “Retailers using personal information in online advertising should disclose the way data are collected, processed and used,” and “Retailers should not use personal information for any purpose unless it has been
authorized by the individuals who provided the information.” A seven-point rating scale was used.
Last, the participants were debriefed. First, the participant was thanked for completing the survey. Then, the real objective of the experiment was explained: “This study explores how
consumers react to personalized advertisements when they are browsing the internet and the effect of information transparency on their reaction”. Finally, important information was restated
accompanied with contact information. Table 3. Scales and Questions
Construct / Variable
Source Scale Scale type
Perceived trust Bansal and Gefen, 2010 Bart et al., 2005 Buttner and Goritz, 2008 Verhoef et al., 2002
I think that this retailer: Q1: Can be relied on to keep its promises. Q2: Will provide a good service. Q3: Will treat me fairly if problems arise.
Q4: Has high standards by which it advertises.
Q5: Will not misuse my personal information.
Q6: My overall confidence in the promise of this advertisement is high.
7 points rating scale (ordinal but mean of multi-item assumed interval so metric) Independent variable: Information transparency Bleier and Eisenbeiss, 2015b Nominal: - No information transparency 0) - Information transparency (1)
30 Moderating variables: Type of a retailer Martin and Murphy, 2017 Nominal: - Only online (1) - Both online and physical (0) Novelty of a retailer Darke et al., 2016 Nominal: - <1-year old (1) - >10 years old (0) Control variables: Privacy settings Hofstra, Corten and Van Tubergen, 2016
Q1: Based on the social network visited most frequently, my private information is accessible to:
Ordinal: - Friends only (1) - Friends and their friends (2) - Public (3) - Don’t know (4) Advertisement Perception Li, Edwards and Lee, 2002 Goodrich, Schiller and Galletta, 2015 Ying, Korneliussen and Grønhaug, 2015
I find this advertisement: Q1: Disturbing Q2: Obtrusive Q3: Interfering Q4: Invasive Q5: Forced Q6: Intrusive Q7: Distracting
7 points rating scale (ordinal but mean of multi-item assumed interval so metric) Privacy Concerns Malhotra et al., 2004 Kim and Agarwal, 2009
Here are some statements about the use of your individual information in an online environment. From the standpoint of personal privacy, please indicate to which you, as an
individual, agree or disagree with each statement:
Q1: It usually bothers me when retailers ask for personal information. Q2: It bothers me to give personal information to so many retailers. Q3: I’m concerned that retailers are collecting too much personal information about me.
Q4: Consumer privacy is a matter of consumers’ right to exercise control and autonomy over decisions about how their information is collected, used and shared.
7 points rating scale (ordinal but mean of multi-item assumed interval so metric)
31
Q5: Retailers using personal information in online advertising should disclose the way data are collected, processed and used.
Q6: Retailers should not use personal information for any purpose unless it has been authorized by the individuals who provided the information.
Age Shavitt,
Lowrey and Heafner, 1998
Q1: What is your age? Metric
Gender Shavitt, Lowrey and Heafner, 1998
Q1: What is your gender? Nominal
- Male (1) - Female (0)
4.4 Sampling method and sample size
The experimental survey was developed with Qualtrics, a tool to easily develop online surveys. To generalize the findings of this study for the entire population, a sample was selected. The population can be interpreted as everyone who has access to the internet and is exposed to personalized
advertisements from retailers. In general, their might be differences between variables that affect consumers trust for people from different cultures, countries and even age (Bjørnskov, 2007), although the most important variables that might have an effect on perceived trust once exposed to online advertisements, is controlled for (Liu et al., 2005; Li, Edwards et al., 2002; Tucker, 2014).
The sampling strategy that was chosen for in this study was non-probability sampling. The sampling frame could be anyone who browses the internet and uses Social Media, as some of the information required to personalize the advertisement came from a conversation on Facebook. To confirm the latter requirement, the survey was distributed on Facebook. Non-probability sampling is in general no appropriate sampling method to make statistical inferences about a population
(Saunders et al., 2012). Brickman Bhutta (2011) states that online snowball sampling is an adequate method to reach a large group of people that are heterogenous and representative of the population. Moreover, Gosling, Vazire, Srivastata and John (2004) and Greene, Speizer and Wiitala (2008) agree that samples from social media networks become increasingly representative of the population
32
because of the increased penetration rate of the internet. Samples do not only become more
representative but online surveys also reduce socially-desirable responding and therefore increase the possibility on truthful answers (Chang and Krosnick, 2009). Besides snowball sampling, the survey was distributed throughout several Facebook groups from a wide range of topics. Everyone
throughout the social media network could have filled in the survey and therefore it is defined as a non-probability sampling method. Furthermore, the survey was distributed between the period May 6, 2018 and May 24, 2018.
To decide on the sample size, the central limit theorem was used as a rule of thumb.
According to Saunders et al. (2012) and Stutely (2003), at least a sample size of thirty participants is required for each condition to get the mean of the sampling distribution close a normal distribution. But according to the law of large numbers, larger sample sizes are likely to be more representative of the population, therefore is chosen to set the sample size per condition at 40 participants. A total sample size of 160 was required to collect enough participants.
4.5 Reliability and Validity
First, the reliability of this study will be discussed. The measurements that were used to test the different constructs were evaluated. The items that were used to measure the privacy setting, advertisement perception and privacy concern were validated in previous research. When analyzing the data another reliability test was executed. Cronbach’ alpha had to exceed 0.70 to be reliable (Litwin and Fink, 1995; DeVellis, 2003). For trust, only separate items were validated and not the entire set. Here, also a reliability test was executed, and Cronbach’s Alpha had to exceed 0.70 to be reliable. Furthermore, the Corrected Item-Total Correlation had to exceed 0.40 (Gliem and Gliem, 2003). If the Corrected Item-Total Correlation did not exceed 0.40, a qualitative consideration had to be made to delete an item. Furthermore, the Kolmogorov-Smirnov test was conducted to explore the distribution and the descriptive Kurtosis and Skewness were measured.
33
Second, the validity of this study will be discussed. The internal validity of the experiment was high since the independent variable and moderators could be manipulated and the most
important control variables were controlled. As these variables were abstracted and controlled for in the experiment, this increased possibility to provide causal interpretations of the results (Roe and Just, 2009) and can be easily replicated (Kessler and Vesterlund, 2015).
The external validity of experiments is low according to Levitt and List (2009), suggesting that experiments fail to generalize the results of this kind of studies. The main goal of this research was to identify the general principle if information transparency in personalized advertisements increases consumers’ perceived trust. Therefore, the promise of this experiment was to generalize constructs, and the quantitative results were externally valid (Kessler and Vesterlund, 2015). Furthermore, a situation was created where the scenarios were as close as possible to the real world to increase the external validity (Heslin, Vandewalle and Latham, 2006). No monetary stakes were included to participate in the experiment, or survey panels were used to reach participants for the experiment, which both could decrease the internal validity due to “survey-fatigue” (Schoenherr, Ellram and Tate, 2015).
The ecological value of this experiment was considered low since it was an online scenario-based experiment. Participants realized that they were a part of an experiment. First, from an ethical point of view it was necessary to inform the participants that they were part of an experiment. Second, there was no opportunity to access data that satisfied the different conditions. To improve the ecological validity an online experiment was conducted. In this situation, participants took part in an environment where they would browse the internet in real life as well. On the contrary, there was no opportunity to provide an environment without noise (Black, 1999). Furthermore, there was also noise created in the experiment with the distraction question. This form of noise was created without compromising the internal validity (Koschate-Fischer and Schandelmeier, 2014).
34
Next, the construct validity of this research was tested with a factor analysis and a reliability test. This can be explained as the degree to which an instrument measures the construct it is intended to measure (Cronbach and Meehl, 1955). The construct of trust has been evaluated by scholars for many years and a wide variety of definitions had been evaluated (Anderson and Narus, 1990;
Morgan and Hunt, 1994; Gefen et al., 2003). Although there are some differences, a widely accepted consent was reached that one’s actions should have positive results and will not perform unexpected actions with negative outcomes as a result. Furthermore, information transparency, privacy concerns and advertisement perception are widely accepted constructs and compared to the scales that were used during this research.
4.6 Testing Hypothesis
To study the relationship between information transparency in personalized advertisements and consumers’ perceived trust, statistical significance tests were performed (Saunders et al., 2012). The results of these tests are clarified in chapter 5. Next, the normality test, Spearman’s correlation and an ANOVA and independent t-test are clarified. Then, the analyses that were chosen to examine the relationships for each hypothesis will be explored. The results were analyzed with SPSS.
To test for normality, several methods were used. First, the skewness and kurtosis for each item was measured to see if the distribution was bell-shaped with a symmetrical pattern. Another way to test for normality was established with the Kolmogorov-Smirnov test. To be normally distributed the significance had to be p>0.05. Regardless of the previous results, normality was proven according to Saunders et al. (2012) and Stutely (2003) with a sample size higher than 30, therefore parametric statistics were used.
Next, the linear relationships between the important variables were quantified. This linear relationship was measured with Spearman’s rho procedure, which bridges between parametric and nonparametric statistics and treat the statistics as rank transformation procedures (Conover and Iman, 1981). The correlations between the variables age, gender, information transparency, total perceived
35
trust, total advertisement perception, privacy concerns and privacy settings were measured.
According to Cohen (1988) the variables have a small effect when the correlation coefficient is 0.1, a medium effect when the correlation coefficient is 0.3 and a large effect when the correlation
coefficient is higher than 0.5.
This experiment was conducted with a mixed factorial design with between and within
factors. To test the means for the effect of information transparency on trust and the effect of the type of retailer on trust an ANOVA test was conducted. The ANOVA can be used when the population consists of multiple groups, the independent variable is categorical and the dependent variable at least at the interval level (Muijs, 2010). In this research, the between factors were (1) No information transparency and both an online and physical store; (2) No information transparency and only an online store; (3) information transparency and both an online and physical store; and (4) information transparency and only an online store. Trust was the dependent variable and because the mean of TotalTrust was measured it did not violate the requirements of the ANOVA and the variance could be analyzed. Finally, the tests that were used to analyze the strength of the relationships between information transparency and consumers’ perceived trust moderated by type and novelty of the retailer are explained. First, the analysis was controlled for with Privacy Concerns, Privacy Settings, Advertisement Perception and Gender. SPSS’ program PROCESS (Hayes, 2012) was used. A linear regression-based path analytical framework is used by PROCESS to estimate direct and indirect effects, since the dependent variable is not dichotomous. Furthermore, PROCESS uses the ordinary least squares regression to estimate the variables on the left side of the model. According to Montoya and Hayes (2017), a multiple regression is the standard method to analyze within-factors.
Furthermore, they advise to use PROCESS because the fewer tests that are used, the fewer chances there are to make errors.
In this study, PROCESS was used to explore significant interactions and conditional indirect effects with multiple moderators. Hayes (2012) estimated the effect of X on Y, in this case
36
Information Transparency on Perceived Trust, in a model that included multiple moderators, in this case Type and Novelty. Model 2 was used which measures: (1) The effect of X on Y; (2) The effect of M on Y; (3) The effect of W on Y; (4) The effect of XM on Y; and (5) The effect of XW on Y. X and Y have been stated before, M is the type of retailer and W the novelty of the retailer. This model resulted in the following:
PT = β₀ + β₁IT + β₂Type + β₃Novelty + β₄(ITxType) + β₅(ITxNovelty) + β6PrivacyConcern + β7AdPerception + β8Gender + error.
The conditional effect of Information Transparency on Perceived trust was here: IT = β₀ + β₁ + β₄Type + β₅Novelty
Furthermore, model 3 of PROCESS was conducted additionally. The interaction effect between information transparency, the type of retailer and the novelty of the retailer was measured. This measure was stated as: (6) the effect of XMW on Y. The interaction effect of information transparency on perceived trust was here:
IT = β₀ + β₁ + β₄Type + β₅Novelty + β6Type*Novelty
5. Results
After the data collection was finished, several steps had to be performed to make statistical inferences about the population. First, all the data had to be coded before it was analyzed to discover any discrepancies in the data and to identify careless respondents (Meade and Craig, 2012). Next, the basic statistics were analyzed. In the consecutive steps the descriptive statistics were analyzed by measuring the skewness and kurtosis, conducting normality tests, a factor analysis and measuring the reliability of the scales. Subsequently, a correlation table was developed before conducting the independent t-tests and ANOVA tests. Finally, the analysis of the results was completed by examining the strength of the relationships between the variables with PROCESS’ regression.
