Covert strategies for personalized social media advertising: Useful instrument or a reason for concern?
Voogt, B.Q., MSc 10280766
Master’s Thesis – Final Draft
MSc. Business Administration – Marketing The Amsterdam Business School (ABS)
University of Amsterdam Dr. A. Zerres 22 – 03 - 2018
ABSTRACT
While marketers increasingly recognize the beneficial effects of personalized advertising on social media platforms, little is known about the underlying mechanisms that create (favorable) attitudes towards these promotional strategies. This cross-sectional online between-subjects experiment (N = 329) studies the impact of covert behavioral and locational retargeted tools for social media advertisement purposes on general consumer responses (advertisement attitude, brand attitude, and purchase intention). Furthermore, perceived advertisement relevance and perceived advertisement intrusiveness are considered to be two important mediators that maximize attitudinal effects. Results indicate that personalized advertisements create less favorable advertisement attitudes than generic advertisements. Brand attitudes only decrease when covert locational retargeting strategies are used. No significantly different effects are found for purchase intention. Also, perceived relevance and intrusiveness significantly impact attitudes in both experimental conditions, especially in the locational retargeted group. Herein, privacy concerns appear to be an important moderating variable.
Key words: Personalized Social Media Advertising, Behavioral and Locational Retargeting, Consumer Responses, Perceived Intrusiveness, Perceived Relevance, Privacy Concerns.
STATEMENT OF ORIGINALITY
This document is written by B.Q. Voogt (student number: 10280766) 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 other 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 the contents.
INDEX
Abstract 2
Chapter I: Introduction 5
Research question 7
Theoretical and managerial relevance 7
Chapter II: Literature review 7
Personalized advertising 7
Personalized advertising on social network sites 9 Covered personalization strategies: behavioral and locational retargeting 10 Consumer responses towards behavioral and locational retargeting 12 Mediating effects: perceived relevance and intrusiveness 15
Privacy concerns 17
Conceptual model 18
Chapter III: Methodology 19
Sample 19
Experimental design 20
Procedure 20
Stimulus materials: 22
Generic advertisement 22
Behavioral retargeted advertisement 22
Locational retargeted advertisement 22
Measurement instruments: 23
Dependent variable: Advertisement attitude 23
Dependent variable: Brand attitude 24
Dependent variable: Purchase intention 24
Mediator: perceived advertisement relevance 24 Mediator: perceived advertisement intrusiveness 25
Moderator: perceived privacy concerns 25
Factor analysis 25
Chapter IV: Results 27
Manipulation check 27
Randomization check 27
Correlation analysis 28
Hypothesis testing: 30
Mancova analysis 31
Mediation analysis: advertisement relevance 33 Moderated-mediation analysis: advertisement intrusiveness 38
Chapter V: General discussion 46
Discussion 46
Managerial implications 49
CHAPTER I: INTRODUCTION
Traditional marketing techniques seem to lose their effectiveness due to perceived advertisement clutter (Rotfield, 2006), rising consumer demands (Fournier & Avery, 2011), and an increase in persuasion knowledge (Ham & Nelson 2016). In this regard, many markets are doubting whether mass marketing efforts are still sufficient in generating positive consumer responses. Therefore, all over the world, companies are investing in a new solution that is, according to many business strategists, able to capture the eye of the consumer, namely: personalized advertising (Wirtz, Göttel, & Daiser, 2017).
Although many personalized advertisement strategies can be distinguished, all share the same basic principle: they use online algorithms to collect real-time personal information to provide individualistically tailored content (Luna-Nevarez & Torres, 2015; Boerman, Kruikemeier, & Zuiderveen Borgesius, 2017). This allows marketers to not only reach new target audiences, but the practice also has a powerful impact on pre-defined organizational goals. For instance, personalized advertisements on social media platforms alone contribute to an average annual revenue of thirty-six billion dollars. Furthermore, with an expected growth of seventeen percent in the upcoming five years, the industry claims that personalized advertisements are an ideal tool to improve profit-generating effects (Business Intelligence, 2016; Interactive Advertising Bureau, 2016).
However, this marketing strategy finds itself in a paradoxical situation. Since the first for-profit brands adopted the approach to deliver personally relevant content on a large-scale, consumer perception varied considerably (Kim & Han, 2014; De Keyzer, Dens, & De Pelsmacker, 2015). Extensive academic research on this subject indicates that personalized advertisements need to be perceived as truly unique and relevant in order to increase favorable consumer sentiment, such as ad attitude, brand attitude, and purchase intention (Anand &
Shachar, 2009; Wegert, 2015). Yet, personalized content that is not able to touch upon the interests of the consumer will eventually backfire to the business organization. This lack of perceived personal relevance will then not only lead to growing concerns about privacy and organizational intrusiveness, but can also negatively affect a brand intention, which in its turn can lead to diminished consumer attitudes (Xu et al., 2011; Bang & Wojdynski, 2016).
Given this situation, the question arises to what extent this advertisement personalization process can be stretched without damaging business performance. Because marketers can easily exceed the balance between perceived personal relevance and feelings of intrusiveness, one might expect them to be hesitant when implementing even more personal characteristics into their marketing tactics. However, recent years paint a different picture. In fact, in addition to increasingly integrating social media platforms in their marketing strategies (Barbu, 2014), organizations are heavily investing in covert retargeting solutions that primarily focus on consumer behavior, such as behavioral and location-based retargeting (Smit, Van Noort, & Voorveld, 2014; Morsello, 2017), allowing marketers to form personal relevant content based on consumers’ online browsing behavior and locational patterns (Morsello, 2017).
Therefore, it is important for marketers and researchers to understand how consumers react to these specialized marketing strategies and how they affect organizational goals. Although extensive research is performed on different degrees of personalization (Dijkstra, 2008; Aguirre et al., 2015), this study is one of the first that focus on covert retargeting techniques. While behavioral retargeting is commonly used by marketers, locational retargeting is relatively new (Johnson, Lewis, & Reiley, 2016). Furthermore, effects of these technological tools are examined within the context of social networking sites; media platforms that have become omnipresent within our society. Finally, special attention is devoted to the mediating roles of perceived advertisement relevance and perceived intrusiveness as two separate paths
that potentially affect consumer attitudes. To examine these issues the following research question will be the point of focus:
RQ: “To what extent do different personalized social media advertisements (behavioral and locational retargeted ads) affect consumer attitudinal responses, and how are they mediated by perceived advertisement relevance and intrusiveness?”
The results of this study will not only broaden the (academic) insights on two important mediating antecedents (perceived advertisement relevance and perceived advertisement intrusiveness), but by creating a distinction between behavioral and locational retargeted social media advertising strategies, this study is also able to give cleared insights into which kinds of personalization metrics can be helpful in improving advertisement attitude, brand attitude, and purchase intention. To achieve this goal and to perform an adequate study, a thorough theoretical framework that captures relevant academic literature about aforementioned characteristics will be set forth. These theoretical insights will then be applied to form the methodological structure of the experiment. Thereafter, the hypothesis will be tested and discussed within the results chapter. Lastly, a comprehensive discussion that touches upon pre-expounded academic literature, accompanied with managerial implications and research limitations, will be provided.
CHAPTER II: LITERATURE REVIEW
Personalized advertising
Although advertisement personalization has received extensive thought among academic scholars, its practice is hard to define (Boerman, Kruikemeier, & Zuiderveen Borgesius, 2017).
In general, personalization can be conceptualized as the approach to deliver individually informative and relevant content at the right time to increase immediate and future business transactions (Tam & Ho, 2006). In order to execute this process, customer information – that is either previously obtained or provided in real-time – is used to tailor online interactions between the individual and the firm (Boerman, Kruikemeier, & Zuiderveen Borgesius, 2017). In this definition, personalized advertising distinguishes itself from other marketing activities that focus on individualistic consumer targeting. For example, its practice is commonly mistaken for customization, which is instead based on consumer-controlled individualization processes (e.g. self-reported advertisement preferences) (Arora et al., 2008).
This is why many academic researchers argue that it is important for consumers to perceive personalization as such (De Keyzer et al., 2015; Walrave et al., 2016). According to Tucker (2014), the use of private characteristics from the recipient is needed to tailor advertisements to a person’s situational context in order to perform true marketing personalization. By using demographics, personally identifying information (e.g. name, location or job type), and shopping-related data characteristics (e.g. purchase history or previously indicated consumer brand preferences), marketers are able to create distinctive content (Bang & Wojdynski, 2016). In doing so, (digital) marketers go through three subsequent phases. According to Murthi and Sarkas (2003), marketers need to learn to analyze consumer data in order to understand current dynamics and to generate proper personalized content. These data insights then have to be matched to an individual offer (i.e. a personalized social media advertisement), after which a firm will need to evaluate its current tactics and adapt its marketing strategies when necessary.
While aforementioned academic researchers perceive personalized advertising as one coherent process with different phases (Murthi & Sarkas, 2003), Dijkstra (2008) argues the existence of three distinct tailoring techniques that can be used for persuasive messages. First,
marketers could choose to adapt information to fit personal characteristics. In her conceptual research, Dijkstra illustrates this by referring to the adaptation of a generic behavioral message that focuses on the negative consequences of smoking to a more general, but personal cue of one’s gender (“Smoking tobacco is responsible for 69% of deaths due to lung cancer in the Netherlands among women.”, p. 767). The second approach is what she calls personalization and goes one step further by incorporating the recipient’s personal characteristics, such as a first name. Lastly, the feedback approach refers to “(…) the provision of information to the target person about one aspect of his or her assessed psychological or behavioral state” (p. 768). As an example, she describes a persuasive message which says that based on a questionnaire it appeared that the tailored person underestimates the negative effects of smoking tobacco.
Based on these illustrations, there seems to be an overall consensus among academic scholars about the definition and interpretation of advertisement personalization. Although differing views about the implementation of this persuasive tactic are present, all assert great importance to the existence of the individual consumer, who must be approached on an individual level. Here, one’s personal needs form the key to generate distinctive marketing success (Imhoff, Loftis, & Geiger, 2001).
Personalized advertisements on social network sites
Regardless of integrating personal contextual cues within marketing communication, marketers are able to choose all sorts of media platforms to deliver personalized advertisements to their target audiences. Marketers’ first attempt to distribute personalized trading offers dates back to the 19th century, in which businessmen used direct mailing techniques to sell their goods and services to their existing consumer base (Ross, 1992). Nowadays, however, personalized marketing focuses more on the adoption of digital media outlets in reaching existing and new
consumers. These outlets include: search engines (e.g. Google, Bing, and Yahoo!), mobile devices, and social media platforms (e.g. Facebook, Twitter, Instagram, and Snapchat) (Ledford, 2015; Zhu et al., 2015; Tuten & Solomon, 2017).
Barbu (2014) argues that out of these new media platforms, social media websites are especially suitable for marketing personalization. From the moment consumers create a private account on a social network site, they are asked to provide basic demographical information, such as their name, gender, date of birth and current home town, which in turn can be used to create personal profiles for advertising purposes. This enormous database of demographical and behavioral characteristics creates an interesting potential for digital marketers to target their audiences on a one-to-one level (Luna-Nevarez & Torres, 2015). Also, by creating personalized advertisements within the context of a social media platform which consumers are generally in control about the information they provide, social exchange between the advertising company and the consumer may occur (Kelly, Kerr, & Drennan, 2009). This provides the opportunity for both parties to equally express themselves through a two-way conversation, which in its turn can embolden business-consumer relationships (Fournier & Avery, 2012).
Covert personalization strategies: behavioral and locational retargeting
Keeping the beneficial opportunities for personalized advertising on social media platforms in mind, Facebook and other social networking sites have increasingly invested in new metrics that are able to provide third parties with a great amount of consumer data (Lambrecht & Tucker, 2013; Stadd, 2014; Kannan, 2017). By collaborating with major technological enterprises, social networking sites developed covert, individually-tailored advertising techniques based on online behavioral search patterns and consumer proximity, also known as behavioral and location (re)targeting (Aguirre et al., 2012; Viglia, 2014). Here, consumer data gathering is purposely undisclosed or hidden from consumers (Milne, Bahl, & Rohm, 2008).
Behavioral retargeting (sometimes referred to as dynamic retargeting) can be defined as a specialized type of covert marketing personalization in which consumers are actively tracked based on their browsing behavior (Smit, Van Noort, & Voorveld, 2014). Here, the consumer, a (third-party) website and the social networking site form a three-way relationship (Zuiderveen Borgius, 2016). By means of so-called ‘cookies’, unique identifiable tags that are able to record every action a person performs on a specific website, third-parties are able to construct online advertisements that are relevant for the consumer. These ads are then offered through various digital platforms, such as sponsored content on Facebook (Aziz & Telang, 2017). A good example could be when a consumer is searching the Internet for a new pair of sunglasses. When a website has installed tracking mechanisms, the consumer will be tracked from the moment he or she enters the website and the site will be notified which specific products capture the attention of that specific person. However, when the consumer decides not to buy a product right away, because, for example, the consumer wishes to search further for alternatives, the stored information can be used to create a personalized advertisement on third-party websites, like Facebook, in order to entice the consumer in making a purchase after all (Smit, Van Noort, & Voorveld, 2014).
Location-based targeting is a less actively used covert data-collection strategy, but offers a good opportunity to reach consumers in an individual way as well (Morsello, 2017). The term location-based marketing is generally understood to mean a marketing strategy that uses consumers’ current proximity in order to provide direct or indirect business offers. It therefore also meets the criteria of personalized advertising (Xu, Oh, & Teo, 2009). This technique allows marketers to not only create commercial content tailored to the individual, but they are also able to use this unique opportunity to present advertisements, promotions, and coupons that can directly be redeemed by the consumer (Molitor et al., 2016). The goal of subtracting personal characteristics by means of consumers’ proximity can be reached in
various ways, by using QR codes (Hopkins & Turner, 2012), Near Field Communication systems (NFC) (Coskun, Ok, & Ozdenizci, 2011), mobile apps (Dhar & Varshney, 2011), or through Wi-Fi connections (Namiot & Schneps-Schneppe, 2011). Furthermore, technological advancements that focus on proximity tools have grown substantially. Beacons, which are physical devices that continuously intercept mobile signals and are able to collect real time data and push commercial content to these Bluetooth-connected devices, are considered by Google as the “next big thing” in retail marketing (Google Inc, 2016). Although location-based marketing is mainly being used in having a direct influence on consumer behavior and their current proximity, more and more marketing professionals recognize the opportunity to use this real-time collected geographical data as a tool for retargeting proposes (GeoMarketing, 2015; Harvard Business Review, 2017). Here, ‘delayed’ personalized social media advertisements are offered based on a consumer’s previous proximity to a certain location. However, academics have only just started paying attention to this topic (Johnson, Lewis, & Reiley, 2016).
Consumer responses towards behavioral and locational retargeting
As covert strategies, such as behavioral and locational retargeting, are adopted by marketers in generating tailor-made social media advertisements, the question arises how these personalization techniques are perceived by the consumer. General studies on marketing personalization, for example, show that consumers benefit from covert personalization tools, because they limit interruption during (online) shopping experiences and are able to keep consumers in their consumer journey (Milne, Bahl, & Rohm, 2008). Besides, Dijkstra (2008) argues that cognitive activity increases when consumers are cued with personal characteristics, such as one’s behavioral patterns. Here, self-referencing, which is the process in which content is processed by relating provided information to one’s self, contributes to the perception of more meaningful and personally relevant content (Dijkstra, 2008).
Nevertheless, annoyance and discomfort may arise when consumers discover that their personal data is being used for commercial purposes afterwards. Especially the absence of an explicit agreement for data collection violates the foundation of a social contract between the marketing organization and the consumer (Miyazaki, 2008). The social contract theory describes the phenomenon in which consumers are willing to disclose personal information in exchange for the agreement that their data is handled with care. However, when personal characteristics are collected without consent, as is being done during behavioral and locational retargeting, this social contract comes at risk. As a result, discomfort may occur when consumers realize that their data was illegitimately used (Boerman, Kruikemeier, & Zuiderveen Borgsius, 2017).
This affective reaction is a result of psychological ownership (Avery et al., 2009). According to Pierce, Kostova, and Dirks (2001); people have a deep attachment towards and are psychologically tied to ideas, objects, information, and other people. Therefore, a person feels personally responsible for the protection of their personal property (Avery et al. 2009). However, when their ownership is used without them knowing, actual feelings of loss arise (James, 1981; Belk, 1988). This decline in control further enhances vulnerability (Baker, Gentry, & Rittenberg, 2005) and psychological reactance (Brehm, 1966). The latter describes an event in which consumers will act contradictory towards the intentions of the opposite party. This is an immediate response to consumers feeling heavily pressured to adapt a certain point-of-view, as is done in personalized social media advertisements (Sundar & Marathe, 2010).
Based on previous research, it is therefore expected that consumers who will receive a personalized social media advertisement based on behavioral and location-based retargeting strategies, will experience a sense of loss of personal information. Since a covert data-collecting method is used, no social contract could be established and psychological reactance and feelings of vulnerability will occur, which in its turn negatively impact marketers’ (direct)
organizational goals, such as advertisement attitude, brand attitude, and purchase intention (Aguirre et al., 2015). Based on aforementioned literature, the following is hypothized:
H1. Behavioral and locational retargeted social media advertisements will evoke more negative consumer responses, a) advertisement attitude, b) brand attitude, and c) purchase intention, than a generic social media advertisement.
Although these three consumer responses are expected to be more negative towards personalization compared to generic advertisements, attitudinal differences between behavioral and locational retargeted social media advertisements are assumed to occur as well. According to White et al. (2008), consumers are more likely to approve personalized advertisements when organizations are able to justify their reasoning behind their (covert) data collection. When organizations are not able to provide consumers with a clear match between their personal characteristics and the displayed advertisement, consumer attitude will more likely decrease. As Aguirre et al. (2016) state, consumers do not perceive justification for advertisements that combine personal data from different sources. For example, Aguirre et al. (2016) argue that when collected data of offline purchases is matched with online social network information to form personalized advertisements, consumer attitudes decline because the two sources are not in congruence with each other. The same could be true for locational retargeting. While behavioral retargeting focuses on data retrieval of behavioral patterns within an online context (congruence), locational retargeted marketing uses non-congruential sources whereby physical consumer locations are used to create online personalized advertisements on social media. Therefore, it is hypothesized that:
H2. Locational retargeted social media advertisements evoke more negative consumer responses, a) advertisement attitude, b) brand attitude, and c) purchase intention, than behavioral retargeted social media advertisements.
Mediating roles of perceived advertisement relevance and perceived intrusiveness
In spite of the above, two important mediating variables are expected to maximize the influence of the relationship between personalized social media advertisements and consumer responses, namely perceived advertisement relevance and perceived intrusiveness.
According to Clunan and Armstrong (1999), covert data collection procedures will less likely provoke discomfort when (a) personal information is gathered by an organization that has an existing relationship with the consumer, (b), consumers feel like they are able to control future use of this data, (c) consumers believe that personal characteristics will only be used to draw valid inferences about them, and (d) consumers perceive displayed personalized advertisements to be truly relevant to them. The latter determinant is especially in line with personalized advertising, since this marketing tactic wishes to generate positive effects by seamlessly connecting consumer needs with offered goods and services (Boerman, Kruikemeier, & Zuiderveen Borgesius, 2017).
Other academic scholars also argue that personalized advertisements do not by definition have negative consequences for beneficial consumer responses (Chellappa & Sin, 2005; James et al., 2015). For example, as Dijkstra (2008) states, when marketers are able to capture actual consumer needs and translate them to personalized advertisements, general consumer responses, such as advertisement attitude, brand attitude, and purchase intention, may increase. This reasoning can be explained by the cost-benefit analysis theory (Robinson, 1997; Brent 2007). This theory describes an existing trade-off between beneficial outcomes and potential costs. When the outcomes outweigh the potential costs, consumers are more inclined
to behave according to the favorable results. The same applies to personalized social media advertising. Consumers are more willing to disclose personal information against the cost of receiving personalized advertisements when they perceive the commercial content to be truly relevant. As a result, consumers will take a more positive view towards the advertisement and the brand behind it (White et al., 2008).
However, when the benefits do not exceed the potential loss of private information, a negative trade-off will emerge (Robinson, 1997). The use of personal characteristics that are used for commercial purposes without consumer consent may appear intrusive. This affective response is defined by Lim, Kim, and Sundar (2002) as the “psychological reaction to ads that interfere with a consumer’s ongoing processes” (p. 39). Perceived intrusiveness especially acts as an important mediator when personal data is considered too personal. As Tucker (2014) states, when a personalized advertisement is designed around the information that a specific person likes cooking, the advertisement will come off as less intrusive, since cooking is a more general personal characteristic. However, when targeting technologies have indicates that a person likes the Korean delicacy Kimchi and build a personalized advertisement around this information, the ad will be considered as highly intrusive due to this relatively rare preference (Tucker, 2014). Subsequently, reactance will be provoked and consumer responses will further decline.
With both mediating variables in mind, it is expected that consumers who perceive behavioral and locational retargeted social media advertisements to fit with existing needs and are considered as personally relevant, potential benefits will outweigh the costs for personal data usage. As a result, these consumers will have higher consumer responses. The opposite is true for consumers who feel that used data is too personal. Here, the costs do not exceed the potential benefits, since they do not fit with personal needs. The personalized advertisements
will then be considered as intrusive, resulting in more negative perceptions towards personalized social media advertisements. Therefore, the following is hypothized:
H3. Perceived advertisement relevance will mediate the relationship between personalized social media advertisements and consumer responses (ad attitude, brand attitude, and purchase intention), such that personalization will lead to more perceived relevance, which in its turn will lead to higher positive responses.
H4. Perceived advertisement intrusiveness will mediate the relationship between personalized social media advertisements and consumer responses (ad attitude, brand attitude, and purchase intentions), such that personalization will lead to more perceived intrusiveness, which in its turn will lead to more negative responses.
Privacy concerns
Finally, perceived intrusiveness is also linked by different academics to existing privacy concerns (Aguirre et al., 2015; Li, 2012). In the most general sense, privacy is defined as the “individual’s ability to control the flow of information concerning of describing him” (p.1107) and involves the psychological skill to control which personal information may be disclosed to others (Miller, 1969). Moreover, Goodwin (1991) adjusted this definition to an interpretation that is more fitting within a commercial context. Here, he describes two important privacy control mechanisms. Besides attributing the individual control over personal data collection techniques by commercial parties, he also argues that privacy affects the control over unwanted persuasive means, such as direct and personalized marketing (Goodwin, 1991).
Privacy is a highly valued concept by many. Therefore, many people are concerned that their privacy will be violated when commercial organizations use personal identifying
information. As a result, highly concerned consumers will develop feelings of frustration, discomfort and intrusiveness when organizations illegitimately use these data (Arora et al., 2008). According to Fransen and colleagues (2015), personalized advertising is able to trigger such feelings among this specific group of consumers. Since marketing personalization uses privacy sensitive information, such as online search behavior and one’s location, control over personal privacy comes at stake. Subsequently, feelings of intrusiveness arise. With this in mind, the following hypothesis is formulated:
H5. Privacy concerns will moderate the relationship between personalized social media advertisements and perceived intrusiveness. For those with high privacy concerns, personalization will lead to higher feelings of intrusiveness.
Figure 1 – Conceptual model.
Perceived relevance Advertisement type 1. Generic Ad 2. Behavioral Ad 3. Locational Ad Advertisement attitude Brand attitude Purchase intention Perceived intrusiveness Privacy concerns
CHAPTER III: METHODOLODY
The present study used an online experimental cross-sectional between-subjects design in order to study the direct and mediating effects between the independent categorical variable (personalized) advertisement type and the different consumer attitudinal responses as expounded in the literature review. Experimental designs are a preferred manner to study causal effects and lay bare the relationships between predicting and outcomes variables (Field, 2009). In order to ensure a valid and reliable study design, the following methodological paragraphs will be discussed below. First of all, the study’s sample characteristics and experimental design will be set forth. Thereafter, the experimental procedure with corresponding stimulus materials and previously validated quantitative measurement instruments will be expounded.
Sample
In total, 478 individuals gave consent and started an online experiment by means of the digital survey tool Qualtrics. From this group, 329 respondents finalized the survey completely and were therefore used to form the study’s sample. This amount is above the recommended sample size of N = 300, which ensures stable factors and generalizability (Worthington & Whittaker, 2006). The demographical characteristics are divided as follows: 50.5% (n = 166) of the individuals described themselves as female, 49.2% (n = 162) as male, and one person (.3%) did not indicate a specific gender. The average age of all participating persons is 33.37 years (SD = 12.24), where the youngest person indicated the age of 18 and the oldest person the age of 70. With regard to educational level, 30.1% of the participants have obtained their academic Master’s degree, 22.2% received their higher vocational program diploma (e.g. Dutch hbo), and 19.5% finalized an academic Bachelor’s program. All other participants are divided over several other educational programs, such as primary school (.3%), high school (15.5%) or
intermediate vocational programs (e.g. Dutch mbo) (6.4%). Further, more than half (58.7%) of the individuals is employed for wages, 22.2% of the participants are fulltime students and 10.9% of the participating persons are self-employed. Others indicated that they either were out of, but currently looking for work (4.0%), out of work and not looking for work (.9%), unable to work (1.5%), retired (1.2%) or a homemaker (.6%). As previously described, this study devotes special attention to social media users. Therefore, social media usage rates are described as well. Based on this study’s sample it can be concluded that Facebook is the most used social media network site (88.4%), followed by Twitter (84.2%), Instagram (76.0%) and LinkedIn (70.8%). On average, participants indicated to login at least once a day (87.8%). No person indicated that he or she did not use social media in the past year.
Experimental design
To investigate whether significantly different consumer responses towards generic and personalized social media advertisements are present, a factorial between-subjects design with three conditions is used. Participants were presented with a hypothetical narrative (vignette approach) and randomly divided over either the control condition (i.e. a generic, non-personalized social media advertisement) or one of the two experimental conditions (i.e. a personalized social media advertisement based on behavioral retargeting or a personalized social media advertisement based on locational retargeting).
Procedure
Online questionnaire distribution took place between 13 November 2017 and 22 November 2017. In order to execute the study properly, a number of criteria were formulated before the participants were approached. First of all, for ethical reasons, all participating individuals needed to be older than 18 years to join this study. Further, since the present study wishes to
examine consumer responses to different (personalized) advertisements on social media sites, all participants must have at least one active social media account (e.g. Facebook, Twitter, Instagram or YouTube) on which they login at least once a month. Persons who indicated that they indeed have a social media account on one of these websites, but have not logged in the past year, were redirected to the end of the online survey and were excluded from the study’s sample. In the meantime, no restrictions were defined based on, for example, nationality or job type.
All participants are selected through convenience sampling. Primarily, the online questionnaire was sent out on different social media channels in order to reach as many persons at the same time. Next, through direct e-mailing, friends, family, and other acquaintances were approached. In order to boost the number of participating persons (n = 87), the online crowdsourcing website Amazon Mechanical Turk (Amazon MTurk) was further used. This internet marketplace allows people to upload Human Intelligence Tasks (HIT’s), such as online questionnaires, that will be filled-out by individuals in exchange for a small monetary payment by the employer (€0.20 per questionnaire; Amazon Mechanical Turk, 2017). Each person that was willing to participate and opened the online questionnaire was first informed by the main goals of this study and notified that the study is part of a Master’s thesis from the program Business Administration – Marketing Track of the University of Amsterdam. In this informed consent, that is approved by the ethical board of the Amsterdam Business School, that belongs to the University of Amsterdam, participants were ensured of complete anonymity during and after the study and it was made clear that discontinuity of participation was allowed at any moment. Further, contact details from the master’s student were provided so that the participating persons could ask questions or provide remarks about the study. Finally, after filling out the questionnaire, in which a randomized stimulus treatment was displayed plus corresponding questions, a debriefing protocol was provided. In this debriefing statement, the
participants were thanked for their participation and the main purpose and ethical reasoning of this study were provided once more.
Stimulus Materials
Behavioral retargeting advertisement. Both experimental conditions consisted of a narrative that was adapted in order to fit as a personalized advertisement. In the behavioral retargeted condition, participants were provided with a story in which it was asked to imagine that the participant was searching on various websites related to Ray-Ban sunglasses earlier that day. A few hours later, an advertisement of the same brand was displayed on his or her personal Facebook timeline. Below the image the following personalized caption could be read: “Searching online for new sunglasses? - Discover them exclusively @ Ray-Ban.com”, (see Appendix A).
Location retargeting advertisement. In this experimental condition, participants were again encountered with a personalized narrative. This time the narrative emphasized on a story where the participants had to imagine that they visited a physical store earlier that day where Ray-Ban sunglasses are sold. The participants further had to visualize that they discussed buying the brand’s sunglasses with an accompanying friend. A few hours later, a personalized advertisement of the same brand was displayed on their personal Facebook timeline. The caption beneath the image read: “Couldn’t find what you were looking for in our brand store? Explore our newest models here! – Discover them exclusively @ Ray-Ban.com”, (see Appendix A).
Control condition. The control condition contained a narrative in which participants were asked to imagine that a Ray-Ban advertisement is displayed on their personal Facebook account
while they are scrolling through their timeline. The advertisement showed an image with a pair of Ray-Ban sunglasses. This time a non-personalized caption could be read: “Find Your Favorites. - Discover them exclusively @ Ray-Ban.com”, (see Appendix A).
In respect of the aforementioned conditions, all stimuli are formatted in a way that they represent an actual webpage from the social media network site Facebook. As respondents were asked to imagine that the used screenshots were taken from their personal Facebook accounts, all information that could be linked to a different person’s social media account (e.g. name, number of friends or social invitations) were made blank. Further, all aspects other the experimental treatment were left untouched, so that direct causal effects could be measured (Field, 2009).
Measurement instruments
After the respondents were treated with one of the three stimulus materials, questions indicating different consumer responses and personality treats were presented. The individual variables were operationalized as follows:
Dependent variable: advertisement attitude. The first dependent variable, advertisement attitude, is measured by means of a pre-validated measurement instrument by Dahlén et al. (2008). Participants could indicate their personal attitude towards the displayed ad by means of a five-item semantic differential scale. The used scale had a high internal validity, with a Cronbach’s Alpha of .942. The corrected item-total correlations indicate that all items have a good correlation with the sum score of the scale (all above .35). Therefore, no items had to be deleted (see Table 1).
Dependent variable: brand attitude. For the variable brand attitude, a pre-validated measurement instrument to examine brand opinion differences between the different conditions is used. This measurement is created by Spears and Singh (2004) and adapted to fit this study. The participants were asked to answer a total of five items on a five-point semantic differential scale. The created scale has a high internal reliability, with a Cronbach’s Alpha of .942. Moreover, the corrected item-total correlations indicate that all items have a good correlation with the sum score of the scale too (all above .30; see Table 1).
Dependent variable: Purchase intention. The measurement instrument purchase intention is based on the pre-validated scale variable of Dodds, Monroe, and Grewal (1991) and again adapted for this study. The instrument uses a five-point Likert scale, ranging from 1 = totally disagree to 5 = totally agree. Internal reliability tests show a high reliability, with a Cronbach’s Alpha of .952. No items had to be deleted, since all single corrected items correlated above .35 (see Table 1).
Mediator: Advertisement relevance. In order to measure to what extent the different advertisements were perceived as personally relevant by the study’s sample, the previously validated measurement instrument by Srinivasan et al. (2002) is used and adapted to fit this study. The participants were asked to indicate perceived advertisement relevance by answering a total of five question, that each had to be answered on a five-point Likert scale, ranging from 1 = strongly disagree to 5 = strongly agree. The advertisement relevance scale had a high internal reliability with a Cronbach’s Alpha of .869. The corrected item-total correlations indicate that all items have a good correlation with the sum score of the scale (all above .30). Therefore, no items had to be deleted (see Table 1).
Mediator: Advertisement intrusiveness. For the second mediator, perceived advertisement intrusiveness, a pre-validated measurement instrument is also used and adapted to fit the present study. This measurement is created by Edwards et al. (2002). A total of nine items are used to measure this latent construct and each item could be answered on a five-point Likert scale, ranging from 1 = totally disagree to 5 = totally agree. Internal reliability tests showed a good reliability, with a Cronbach’s Alpha of .942. Further, no items have to be deleted, since the corrected item-totals correlated above .30 (see Table 1).
Moderator: privacy concerns. As previously mentioned, the variable privacy concerns are treated as existing personality treats within this study. Therefore, the measurement instrument of Baek and Morimoto (2012) is used. This instrument consists of a total of four items that each could be answered on a five-point Likert scale, ranging from 1 = totally disagree to 5 = totally agree. The reliability test showed a good internal validity, with a Cronbach’s Alpha of .853. No items had to be deleted, since all corrected item-totals correlated above .35 (see Table 1).
Factor analysis. A principal axis factoring analysis (PAF) was conducted for all items from aforementioned measurement instruments. The Kaiser-Meyer-Olkin measure verified sampling adequacy for the used analysis, KMO = .926. Further, Bartlett’s test of sphericity,
!
2 (496) = 9482.93, p < .001, indicated that the correlations between each item is sufficiently large for PAF. An initial analysis is performed to obtain Eigenvalues for each factor in the data. Six factors had Eigenvalues over Kaiser’s criterion of 1 and together had an explained variance of 71.35%. Besides, a scree plot analysis also revealed a total of six factors. Therefore, six factors are used and rotated by means of a Oblimin with Kaiser normalization rotation. Table 1 shows each factor with corresponding factor loadings. Additionally, no high cross-loadings between the different factors are found.Table 1 – Factors with corresponding item loadings, Eigenvalues, Eigenvalues and Cronbach’s Alphas. Factor & indicators Loadings Eigenvalue EV* "
Advertisement attitude 11.830 36.10% .941
“My overall impression of this advertisement is…”
Negative - Positive .84 Dislike – Like .83 Unfavorable – Favorable .83 Bad – Good .82 Unpleasant – Pleasant .78 Brand attitude 1.025 2.34% .942
“I think Ray-Ban is...”
Unlikeable – Likeable .92 Unfavorable – Favorable .90 Unpleasant – Pleasant .85 Bad – Good .76 Unappealing – Appealing .72 Purchase intention 1.974 5.20% .952
“The probability that I would consider buying the products of Ray-Ban is high.”.
.94 “My willingness to buy the products of Ray-Ban is high.”. .88 “I would consider buying the products of Ray-Ban.”. .86 “I would consider buying the products of Ray-Ban.”. .85
Advertisement relevance 2.861 7.97% .869
“Overall, this ad on Facebook is tailored to my situation”. .95 “This advertisement on Facebook makes purchase
recommendations that match my needs.”.
.76 “I think that this advertisement on Facebook enables
me to order products that are tailor-made for me.”.
.72 “I believe that this advertisement on Facebook is
customized to my needs.”.
.71 “This advertisement on Facebook makes me feel that I am a
unique customer.”.
.40
Advertisement intrusiveness 1.384 3.69% .942
“I think this advertisement is alarming.”. .89 “I think this advertisement is uncomfortable.”. .87 “I think this advertisement is disturbing.”. .83 “I think this advertisement is obtrusive.”. .79 “The advertisement gives me an uneasy feeling.”. .79 “The advertisement gives me an unsafe feeling.”. .74 “I think it is uncomfortable that personal information is
used in this offer.”.
.61 “I think this advertisement is irritating.”. .55 “I think this advertisement is annoying.”. .43
Perceived privacy concerns 5.308 15.78% .853
“I am concerned about the potential misuse of my online personal data.”.
.86 “I fear that information is not being stored online safely.”. .81 “I believe that personal data are being misused online too
often.”.
.71 “I feel uncomfortable when online data are shared without
permission.”.
.61 Note. EV*: Explained variance.
CHAPTER IV: RESULTS
Manipulation check.
In order to investigate whether the displayed social media advertisements (i.e. generic, behavioral and locational retargeted advertisements) were perceived as intended and enough attention was paid by the participants towards these stimulus materials, a manipulation check is performed. At the end of the experiment, a last question was provided where the participants had to indicate which advertisement type they had seen. Based on this test it can be concluded that there is significant difference between the three different conditions, χ2(4) = 236.95, p < .001. In total, 71.3% of the participants within the generic advertisement condition reported the correct displayed advertisement. However, in the personalized behavioral retargeted advertisement group, ‘only’ 62.23% of the participants were able to choose the right condition type. A great amount of them reported to have seen a personalized locational retargeted advertisement instead (24.33%). Furthermore, 78.0% of the participants within the last experimental condition recognized the locational retargeted advertisement correctly. However, a Chi-square test based on the two different experimental treatments alone still revealed a significant difference, χ2(3) = 91.53, p < .001. Thus, although a high percentage of the participants within the first experimental condition perceived a different personalized stimulus, it could be stated that the manipulation was successful.
Randomization checks
Different randomization checks are performed in order to investigate whether the participants were equally distributed over the control and the two experimental conditions with respect to gender, χ2 (4) = 2.72, p = .605, age, F(2, 326) = 1.28, p = .280, education, F(2, 326) = .38, p = .681, employment, F(2, 326) = 1.42, p = .244, and social media use, F(2, 326) = .23,
p = .795, Based on these non-significant statistical analyses it can be concluded that randomization is successfully executed and these parameters do not negatively affect the accuracy of this study.
Correlation analyses
Table 2 presents a correlation matrix in which the overall means, standard deviations and bivariate correlations between each measured variable is given. Besides the demographical variables gender, age, educational level, work-status and average social media use, the latent variables perceived advertisement relevance, perceived advertisement intrusiveness, ad attitude, brand attitude, purchase intention, and privacy concerns are included as well.
A first investigation of the bivariate correlations indicates a moderately significant, symmetrical relationship between age and all other variables. For example, a moderately, negative significant relationship between age and perceived advertisement relevance is found, r = -.26, p < .001. This indicates that the older a consumer is, the less likely he or she perceives one of the three displayed advertisements to be personally relevant. On the other hand, correlation coefficients between age and perceived advertisement intrusiveness suggests a moderately, positive relationship between these variables as well, r = .22, p < .001, indicating that the older a person is, the more likely that person perceives a certain displayed advertisement to be personally intrusive. Furthermore, (moderately) strong relationships can be distinguished between several latent variables. The strongest (negative) relationship can be found between perceived advertisement intrusiveness and advertisement attitude, r = -.72, p < .001. Subsequently, bivariate coefficients between purchase intention and brand attitude, r = .60, p < .001, perceived privacy concerns and advertisement intrusiveness, r = .48, p < .001, perceived advertisement relevance and purchase intention, r = .46, p < .001, and brand attitude and advertisement attitude, r = .44, p < .001, suggest positively strong relationships. The correlations between perceived advertisement relevance and perceived advertisement
Table 2 – Means, Standard Deviations, Bivariate Correlations, and Reliability.
N = 329 / Cronbach’s Alpha values are reported between parentheses. ** , Bivariate correlation is significant at .01 level (two-tailed)
*, Bivariate correlation is significant at .05 level (two-tailed)
Variables M SD 1 2 3 4 5 6 7 8 9 10 11 12
1 Gender 1.50 .51 -
2 Age 33.75 12.42 -.01 -
3 Educational level 5.47 1.58 .08 .21** -
4 Work-status 2.53 2.20 -.14* -.30** -.30** -
5 Social media use 6.32 1.01 -.03 -.20** -.09 .15** -
6 Ad relevance 2.72 .91 .05 -.26** .02 -.02 .04 -! (.869) 7 Ad intrusiveness 3.17 1.01 -.17** .22** .01 -.09 -.02 -.12* - (.942) 8 Ad attitude 2.59 .96 .08 -.24** .08 .08 -.05 .37** -.72** - (.941) 9 Brand attitude 3.53 .90 .07 -.19** .08 .04 .05 .27** -.31** .44** - (.942) 10 Purchase intention 2.93 1.11 .07 -.18** .13* -.02 .00 .46** -.20** .36** .60** - (.952) 11 Privacy concerns 4.01 .79 -.10 .25** .04 .07 -.07 -.25** .48** -.40** -.19** -.12* - (.853)
intrusiveness is considered rather weak, r = -.12, p < .001, which provides preliminary evidence that consumers consider an advertisement to be either personally relevant or intrusive.
Hypothesis testing
To test the first hypothesis, in which a significant difference between (personalized) advertisement type and the thee consumer affective responses (ad attitude, brand attitude, and purchase intention) is predicted, a multivariate analysis of covariance (MANCOVA) is performed. Based on aforementioned bivariate correlations it is expected that age has an important effect on the outcome effects and is therefore controlled for.
Furthermore, in order to investigate whether any (moderated) mediation is present, Hayes’s (2012) SPSS Macro PROCESS is used. This model allows us to test full (in)direct mediating effects rather than calculating individual effect sizes between variables to estimate mediation. Besides, due to its ability to use bootstrapping as a resampling method, normal distribution is guaranteed (Hayes, 2012). However, just like any other statistical model, PROCESS has its limitations. Since the independent variable social media advertisement type is measured on a categorical scale with three levels, indicator coding (dummy variables) is used to carry out the analysis. This coding system splits a single multicategorical variable into two separate binary variables, each with the generic advertisement as reference point. By doing so, the behavioral retargeted and the locational retargeted advertisement group are individually compared to the generic advertisement group. To test hypothesis 3, a simple mediation model (PROCESS Model 4) with perceived advertisement relevance was performed for each dependent variable separately. To investigate hypothesis 4 and 5, moderated mediation models (PROCESS Model 8) are used in which the mediating variable is considered to be perceived advertisement intrusiveness and the moderating variable perceived privacy concerns. When
moderated mediation did not occur within one or both experimental groups, an additional simple mediation model with the corresponding mediating variable is performed.
Main effects of social media advertisement type on advertisement attitude, brand attitude and purchase intention
According to Wilks’ Lambda, a significant effect of social media advertisement type on advertisement attitude, brand attitude, and purchase intention is found, Λ = .83, F (6, 646) = 10.67, p < .001. Furthermore, Box’s Test of equality of covariance metrics indicates sufficient heterogeneity within covariance, p = .097.
Advertisement attitude. A separate univariate ANOVA revealed a significant effect between social media advertisement type and ad attitude when controlled for age, F (2, 325) = 8.67, p < .001, !2 = .05. The variance in advertisement type explains for 5% the variance in advertisement attitude, indicating a weak effect. Moreover, consumers who saw the generic advertisement expressed the highest average mean score: they indicated an average of 2.95 on a five-point Likert scale (SD = .09). Participants who were treated with either the behavioral (M = 2.55, SD = .09) or locational retargeted advertisement (M = 2.45, SD = .09) scored on average lower than the generic advertisement group (see Figure 2). Bonferroni post-hoc tests further indicated a significant difference between the generic advertisement and the behavioral retargeted advertisement (Mdifference = .40, p = .001) and the generic and the locational retargeted advertisement (Mdifference = .49, p < .001). However, no significant difference is found between the behavioral and the locational retargeted advertisement (Mdifference = .10, p = 1.00).
Brand attitude. A second univariate ANOVA also revealed a significant effect between social media advertisement type and brand attitude, F(2, 325) = 8.18, p < .001, !2 = .05 (controlled for age). Here too, a weak overall effect is found. An almost equal average score is identified between the generic advertisement (M = 3,60, SD = .08) and the behavioral retargeted
advertisement group (M = 3.62, SD = .08), while the participants in the locational retargeted group indicated lower scores on average (M = 3.19, SD = .08) (See Figure 2). Further, Bonferroni post-hoc tests revealed a significant difference between the generic and the locational retargeted advertisement, (Mdifference = .41, p = .002) and between the behavioral and location retargeted advertisement, (Mdifference = .43, p < .001). No significant difference is found between the generic and behavioral retargeted advertisement for brand attitude.
Purchase intention. To test hypothesis 1c, another univariate ANOVA is analyzed. According to this test, a marginally significant model is found when controlled for age, F(2, 325) = 2.84, p = .060, !2 = .02. This marginal effect has to be interpreted extra carefully. Average mean scores indicated that consumers within the generic advertisement condition scored on average lower on purchase intention (M = 2.73, SD = .10) than participants within the two experimental conditions (behavioral retargeted ad: M = 3.03, SD = .10; locational retargeted ad: M = 3.03, SD = .10; see Figure 1). However, Bonferroni post-hoc tests revealed no significant differences between the different conditions on purchase intention.
Based on these findings, hypothesis one is partly accepted. The expectation that both personalized advertisements evoke significantly lower attitudes compared to generic advertisements is only true for advertisement attitude. Next, hypothesis two is also partially accepted, since a significant difference on brand attitude is found between the behavorial and locational retargeted condition.
Figure 2 – Mean scores between the conditions on the three dependent variables. ! 1 1.52 2.53 3.54 4.55
Ad attitude Brand attitude Purchase intention
L ik er t sc al e m eas u re m en t Generic ad Behavioral ad Location ad
Mediating effects of perceived advertisement relevance on the relationship between advertisement type and advertisement attitude
An OLS regression model is performed in order to investigate whether advertisement relevance significantly mediates the relationship between social media advertisement type and ad attitude. The model proved to be significant with a variance of 33.1% in advertisement attitude explained by the variation of the dependent, mediating and control variables, F(3, 324) = 40.149, p < .001. Further, based on the indirect effects of the two experimental groups, a mediation by perceived relevance on the relationship between advertisement type and advertisement attitude is present. For the behavioral retargeted group, an indirect effect of a1b1 = .46 is found, which indicates that the participants within this group, compared to the generic advertisement group, are estimated to score .46 units higher on advertisement attitude as a result of perceived advertisement relevance. For the locational retargeted group this mediating effect is slightly higher with a2b1 =.56 units. Both effects are considered significant based on their 95% BC Bootstrap confidence which are entirely above zero (.31; .64) and (.39; .74) (see Table 4). When the direct relationships between the individual variables are studied, we see a significant direct effect for both experimental groups on advertisement relevance. The respondents within the behavioral retargeted advertisement group indicated a higher advertisement relevance of a1 = .78 units than those within the generic advertisement group, t =
7.38, p <. 001, CI[.58; .99]. As for the locational retargeted group the perceived relevance is considered to be slightly higher with a2 = .94 units, t(325)= 8.79, p < .001, CI[.73; .1.15].
Moreover, a significant direct relationship is found between perceived advertisement relevance and advertisement attitude. The effect of b1 = .59further indicates that consumers who differ one unit on advertisement relevance score on average .59 higher on advertisement attitude, t(324) = 6.31, p <.001, CI[.24; .46] (see Table 3). With respectively c1’ = -.86 for the behavioral
Table 3 - Regression Coefficients, Standard Errors, and Model Summary
NB: Generic advertisement as dummy reference point.
Consequent
Ad relevance
(M) Ad attitude (Y1) Brand attitude (Y2) Purchase intention (Y3)
Antecedent Coeff. SE p Coeff. SE p Coeff. SE p Coeff. SE p
Behavioral retargeting (X1) a1 .784 .106 <.001 c 1’ -.862 .117 <.001 c1’ .278 .120 .022 c1’ -.166 .144 .248 Locational retargeting (X2) a2 .940 .107 <.001 c 2’ -.1.050 .121 <.001 c2’ -.485 .118 <.001 c2’ -.234 .149 .109 Ad relevance (M) - - - b 1 .592 .057 <.001 b1 .579 .058 <.001 b1 .588 .069 <.001 Constant i1 2.719 .146 <.001 i 2 2.021 .214 <.001 i2 3.054 .220 <.001 i2 1.659 .263 <.001 R2 = .267 R2 = .331 R2 = .190 R2 = .222 F(3,325) = 39.473, p <.001 F(4,324) = 40.149, p <.001 F(4,324) = -19.007 p <.001 F(4,324) = 23.123, p <.001 Control Age -.017 .004 <.001 -.010 .004 .007 -.008 .004 .038 -.006 .005 .217
Table 4 – Direct, Indirect and Total effects per dependent variable.
Ad Attitude (Y1) Brand Attitude (Y2) Purchase intention (Y3)
Behavioral retargeting
Effect SE p LLCI ULCI Effect SE p LLCI ULCI Effect SE p LLCI ULCI
Direct Effect -.862 .117 <.001 -1.092 -.632 -.277 .120 .022 -.515 -.041 -.166 .149 .248 -.449 .116 Total Effect -.398 .125 .002 -.644 -.153 .020 .118 .867 -.213 .253 .295 .147 .045 .006 .584 Effect Boot SE LLCI Boot ULCI Boot Effect Boot SE LLCI Boot ULCI Boot Effect Boot SE LLCI Boot ULCI Boot
Indirect effect .464 .084 .314 .640 .297 .066 .175 .433 .461 .089 .302 .647
Locational retargeting
Effect SE p LLCI ULCI Effect SE p LLCI ULCI Effect SE p LLCI ULCI
Direct Effect -1.050 .121 <.001 -1.288 -.811 -.762 .125 <.001 -1.008 -.517 -.240 .149 .108 -.533 .053 Total Effect -.493 .126 <.001 -.741 -.246 -.406 .125 <.001 -.640 -.172 .313 .148 .035 .023 .601 Effect Boot SE LLCI Boot ULCI Boot Effect Boot SE LLCI Boot ULCI Boot Effect Boot SE LLCI Boot ULCI Boot
retargeted advertisement group and c2’= -1.01 for the locational retargeted group, a significant direct relationship is present between these advertisement types and advertisement attitude, t (324) = -3.92, p <.001, CI[-.74; .-25]. These direct relationships are in line with aforementioned MANCOVA analysis.
Finally, the total effect of the behavioral and locational retargeted groups on advertisement attitude, compared to the generic advertisement, are c1 = -.40 and c2 = -.49. This means that the respondents within one of these two experimental groups are estimated to respectively differ .40 and .49 units on their reported advertisement attitude. The negative sign indicates that a person who sees one these two advertisements perceived a lower advertisement attitude than the generic advertisement group. The effects are significant from zero, t(324)= 7.37, p <.001, CI[-.1.09; -.63] and t(324) = -8.66, p <.001, CI[-1.29; -.81] (see Table 3).
Mediating effects of perceived advertisement relevance on the relationship between advertisement type and brand attitude
A second linear regression model is performed by means of Hayes’ PROCESS Macro in order to investigate whether the relationship between advertisement type and brand attitude is mediated by perceived relevance. Here too, a significant model is found, F(4, 324) = 19.01, p < .001. This time the variance in the dependent variable brand attitude is explained by 19% of the variance of the other variables within the model. Furthermore, a significant mediating effect is found for both the experimental conditions. The behavioral retargeted group scored on average a1b1 = .30 higher on brand attitude as a results of advertisement relevance than the generic advertisement group. Just like the dependent variable advertisement attitude, the locational retargeted group indicated with a2b1 = .36 units a marginal higher brand attitude as a result of advertisement relevance than the other experimental group. Both indirect effects are considered to be significant, 95% Bootstrap confidence (.18; .43) and (.22; .50) (see Table 4).
Moreover, as previously indicated, a direct effect is found between both personalized advertisements and advertisement relevance. The relationship between advertisement relevance and brand attitude also proved to be significant with b1 = .38 units, t (324) = 4.35, p <.001, CI [.13; .34] (Table 3). Further, as in line with previously performed MANCOVA analysis, a significant direct effect of respectively c1’ = .28, t(324) = -3.41, p =.001, CI[-.64; -.17], and c
2’ = -.48, t(324) = -2.31, p =.022, CI[-.52; -.04], is found for both experimental groups. Based on this result, it can be concluded that a negative relationship between the locational retargeted ad on brand attitude is present, but do to perceived advertisement relevance, this effect is reversed into a positive relationship. The total effect for the locational retargeted group is significant, with c2 = -.41 units, indicating that a consumer who sees this type of personalized advertisement indicated a lower brand attitude than consumers who see a generic advertisement, t(324)= -6.109, p <.001, CI[-.1.01; -.52]. The total effect for the behavioral advertisement group is not significant (see Table 3).
Mediating effects of perceived advertisement relevance on the relationship between advertisement type and purchase intention
A third OLS regression model with the potentially moderating variable perceived advertisement relevance on the relationship between advertisement type and purchase intention is performed. The full model is considered to be significant, F(4, 324) = 23.12, p < .002, and with 22.2% of the variation in purchase intention explained by the other variables within the model, the predication is moderately strong. Furthermore, in both conditions, a significant, indirect effect of respectively .46 for the behavioral retargeted advertisement and .58 for the locational retargeted advertisement is found. This indicates that consumers within the latter condition score somewhat higher on purchase intention as a result of advertisement relevance than the respondents who saw the behavioral retargeted advertisement (each individually compared to
