When personalization becomes too personal:
Personalization of online advertising and resistance
Lisa Wadle – 10186336
Master Thesis – Graduate School of Communication MSc: Persuasive Communication
Supervisor: Marieke Fransen 24/06/2016
Abstract
The use of personalization as a marketing strategy is not only increasing, but is even regarded as the future of advertising. However, personalization is also seen as a rather delicate issue as it thrives on consumer data. Understanding how consumers may respond to a personalized advertisement is thus crucial in order to design successful personalization approaches. Therefore, this study examined the effects of
personalization of online advertisements on the use of resistance strategies by
receivers of these advertisements. In an experiment, 204 participants were assigned to one out of three advertising conditions: non-personalized, slightly personalized and highly personalized. It was investigated whether these three levels of personalization induced the use of different types of resistance strategies, namely avoidance,
empowering and contesting. Despite partially unsuccessful manipulations, results indicated that this was not the case. Results did show, however, that participants used avoidance strategies the most, followed by empowering strategies. Privacy concerns and persuasion knowledge were investigated as potential mediators. Although no mediating effects were found, persuasion knowledge showed to be a predictor of using empowering and contesting strategies, and privacy concerns proved to be a predictor of using avoidance strategies. Finally, exploratory analyses showed a
positive impact of personalization on attitude towards the brand as well as on noticing the advertisement.
Introduction
“Can we figure out if a customer is pregnant, even if she doesn’t want us to know?” A few years ago, this somewhat unusual question from Target’s marketing
department to a statistician was the starting point of Target’s new personalization strategy (Duhigg, 2012). The marketers explained to the statistician that they wanted to send specially designed advertisements to women in their second trimester, as this is the moment most expectant mothers start to buy necessities. Target, the second largest discount retailer in the United States, started to keep tabs on their customers’ behavior and everything they bought in order to answer the marketers’ question. Through data collection and examination, the marketers and the statistician seized the opportunity to be ahead of the competition. As birth records are public, Target
understood that timing is everything. New parents will be overloaded with incentives and offers by Target’s competition as soon as the baby is born. Through
personalization, Target grabbed the opportunity to build a relationship with the mother-to-be by reaching out to her long before other retailers even knew a baby was on the way.
With their personalization strategy, Target made sure specific advertisements were shown to the right person at exactly the right time. The example shows that when executed correctly, personalization can be an extremely powerful tool. It is a concept that is perceived as a shift in the entire business landscape and by some even regarded as the future ruler of marketing success (Edwards, 2016). In today’s business world, consumers are expecting a personal and unique experience (Newman, 2015). Offering consumers a tailored experience is necessary to differentiate from
competitors and build lasting consumer relationships. The overwhelming majority of participants (87 percent) in a study by BloomReach even indicated that they are influenced to buy more when retailers use personalization (O’Brien, 2015). It can take advertising to the next level and be an answer to problems marketers face daily, such as banner blindness (Resnik & Albert 2014). Personalized advertising is believed necessary in order to break through today’s media clutter and catch the consumer’s attention (Köster, Rüth, Hamborg, & Kaspar, 2015; Wegert, 2015). Simply put, personalization can be viewed as excellent customer service – tailoring the experience and thus making it better for the customer (Singer, 2012).
There is, however, also a less positive side to personalization. Done wrong, personalization can backfire. The example stressed in the beginning shows the fine
line between personalization and the main issue that goes hand in hand with it – privacy. Target’s personalization strategy aimed at expectant mothers proved to be successful for the company. However, a small misstep that occurred illustrates the ‘creepy side’ of personalization (Duhigg, 2012; Hill, 2012). As Target sent out coupons for baby clothes and cribs to expectant mothers, a teenage girl received them as well. The girl’s dad contacted Target, furious that his underage daughter was receiving maternity advertisements. “My daughter is still in high school and you are sending her this! Are you trying to encourage her to get pregnant?” It turned out the girl was actually expecting a child; Target knew the girl was pregnant before her own father did, based on her shopping behavior (Hill, 2012). Hence, Target’s example does not only highlight the positive side, but also shows the unwanted yet prevalent side effect of personalization: consumers may feel their privacy is invaded and therefore experience discomfort (Arora et al., 2008). If online retailers become too familiar with their customers, they risk alienating them, according to several marketing professionals (Singer, 2012). This could result in consumers resisting a personalized message; hence personalization also could generate less desirable effects for advertisers (Baek & Morimoto, 2012; Tucker, 2014).
As the use of personalization is not only increasing (Maslowska, Van den Putte, & Smit, 2011), but as personalization is even regarded as the future of
advertising (Entis, 2015; Wegert, 2015), it is crucial for marketers to understand how and when consumers might resist it. Research so far however has not focused on how and when people use resistance in reaction to online personalization. The aim of this research therefore will be to identify which resistance strategies described by Jacks and Cameron (2003) and Fransen, Verlegh, Kirmani and Smit (2015) consumers use to resist personalized online advertising. These resistance strategies will be defined later on in this study. For the purpose of this study, three levels of online
personalization will be assessed based on the advertising continuum of De Keyzer, Dens and De Pelsmacker (2015); non-personalized, slightly personalized and highly personalized online advertising. Connecting these concepts, the following research question will be investigated:
RQ: Do consumers use different resistance strategies for different levels of personalization of online advertisements?
In this study the different types of resistance strategies will be divided into three different clusters; avoiding, empowering, and contesting strategies. It will be argued
that with non-personalized online advertising avoiding strategies will dominate; with slightly personalized online advertising empowering strategies; and with highly personalized online advertising contesting strategies will be most prevalent. An experiment will be conducted in order to answer the research question, as well as to shed light on the potential mediating effects of persuasion knowledge and privacy concerns. Finally, two additional concepts, attitude towards the brand and noticing the advertisement, will be investigated in order to enhance the results. Implications and suggestions for future research will be discussed.
Theoretical Background Defining personalization
Personalization is a very broad construct with many different definitions (Versanen, 2007). It is often used as a synonym to customization although they are not the same concept (Arora et al., 2008). Personalization differs from customization in such that with customization, the consumer defines what s/he wants and needs (Arora et al., 2008). Individuals customize their experience themselves by stating their preferences and interests. Personalization however is provided by a company based on a consumer’s previous behavior (Arora et al., 2008). Tam and Ho (2006) offered the definition of personalization as a customer-oriented marketing strategy, aimed to deliver the right content to the right person at the right time in order to maximize immediate and future business opportunities. A simple example of
personalization would be tracking a consumer’s online behavior via cookies. Cookies are small text files that are put on users’ devices to facilitate the functionality of a website (Smit, Van Noort, & Voorveld, 2014). During a second visit, this information will be used to optimize the consumer’s experience. Thus with personalization, the marketer identifies the consumer’s needs and acts upon them, and the consumer is not necessarily aware of it (Versanen, 2007; Arora et al., 2008; Aguirre, Mahr, Grewal, De Ruyter, & Grewal, 2015). In computer mediated advertising regarding
personalization, three so called tailoring ingredients can be identified: personalization, feedback and adaptation (Dijkstra, 2008; Hawkins et al., 2008). Personalization, the first ingredient, refers to incorporating recognizable aspects of an individual in the content, such as a first name. It explicitly claims that the message is meant for the receiver. The second ingredient, feedback, always identifies itself as saying something about the individual, such as “about you”, or “this information is especially for you.”
With feedback, the message is aimed at one aspect of an individual’s assessed psychological or behavioral state. Dijkstra (2008) provides an example of this with the following advertising message: “It appears from your responses that you underestimate the dangers of smoking.” The last tailoring ingredient, adaptation, refers to the adjustment of a message to individual characteristics without explicitly mentioning that the message is meant for the receiver. The message is presented as if it is directed at a general audience; consequently, the receiver may not be aware that the message is adapted for him or her. Advertising based on gender provides a great example. On the same website, a male website visitor would be shown an
advertisement made for men; however, a female visitor would be shown an
advertisement made for women. Adaptation primarily provides content information and is commonly used for target group segmentation (Dijkstra, 2008). This makes adaptation very suitable for designing personalized online advertising such as banners meant for several receivers, rather than one particular individual. Out of the three aspects, adaptation is most easily applied to a larger scale. Therefore this research will focus on the adaptation aspect of personalization.
Personalization focused on the adaptation aspect can be combined naturally with the advertising continuum proposed by De Keyzer et al. (2015). In their continuum, three levels of online personalization are regarded, ranging from no personalization, to general personalization to full personalization (De Keyzer et al., 2015). Non-targeted, generic online advertising of a random product or service in the form of a banner can be regarded as non-personalized online advertising. With slightly personalized online advertising, one can think of a generic advertisement derived from the website an individual has just visited, shown on a different website. De Keyzer et al. (2015) also provided the example of sending local bridal shop ads to women whose relationship status is ‘engaged’. Lastly, an example of highly
personalized online advertising would be a banner of the exact same product an individual has just viewed, placed on a different website than where s/he viewed the product. Here, the message is tailored to the individual based on website visits, viewed content, any other behavior such as clicks, or even personal characteristics (De Keyzer et al., 2015).
Now that personalization is defined, various resistance strategies will be briefly explained and put into a framework. After, these resistance strategies will be linked to different levels of personalization.
Types of Resistance Strategies
In their research on persuasive communication and resistance, Jacks and Cameron (2003) identified seven different resistance strategies: selective exposure, attitude bolstering, self-assertion, social validation, source derogation, counter arguing and negative affect.
Selective exposure, the first strategy, involves selectively exposing oneself to messages that are in line with an individual’s original attitude. The second strategy, attitude bolstering, involves generating thoughts that are consistent with an
individual’s original attitude, without directly refuting message arguments. Self-assertion, the third strategy, includes asserting that nothing or no one could ever change an individual’s opinion. The fourth strategy, social validation, is about bringing in mind important others who share one’s original attitude; individuals validate their attitude with the attitude of significant others (Fransen et al., 2015). The fifth strategy, source derogation, involves insulting the source or rejecting his or her validity. Counter arguing, the sixth strategy, includes direct rebuttal of message arguments. The seventh strategy, negative affect, involves responding to the message by getting angry, irritated or upset.
Next to these seven strategies, three more resistance strategies can be derived from the research of Fransen et al. (2015) and Fransen, Smit and Verlegh (2015): avoidance, distraction and message derogation. Avoidance in the context of online advertising primarily simply involves trying not to look at a banner. With distraction, the receiver tries to distract him or herself from the advertisement. Finally, message derogation aims at downplaying the message of the advertisement. Next, these ten strategies will be categorized into three clusters in the following section.
Framework of Resistance Strategies
Fransen et al. (2015) offered the ACE typology, which clusters resistance strategies into three different groups: avoidance, empowering and contesting. This framework will be slightly adapted for the purpose of this study and serve as the base of this research.
The avoiding strategies cluster contains physical, mechanical and cognitive avoidance strategies. Physical avoidance strategies are aimed at not seeing or hearing an advertisement, which for example includes leaving the room when a commercial is aired. Mechanical avoidance involves muting the television when a commercial is aired, or clicking on a different website when encountering an advertisement online.
Lastly, cognitive avoidance involves not paying attention to specific advertisements and thus is synonymous with selective exposure. Avoidance, distraction and selective exposure are the resistance strategies clustered in this group.
The empowering strategies cluster contains strategies that are related to the recipients; not the message. Instead of challenging the persuasive message, recipients assert their own existing views. Three strategies are included in the empowering strategies cluster: attitude bolstering, self-assertion and social validation. Individuals may validate their previous attitudes, thoughts or behaviors through generating complementing thoughts, or by reminding themselves that they are confident about them (Fransen et al., 2015). Individuals may also seek external support from their environment, validating their views with peers (Jacks & Cameron, 2003).
Finally, the contesting strategies cluster contains strategies that involve actively refuting an advertisement by challenging it. Strategies found in this cluster are source derogation, message derogation, counter arguing, as well as negative affect. With the first three strategies, recipients dismiss the validity of the source, the message or the arguments presented in the message. The fourth strategy, negative affect, is a strategy where recipients get emotionally upset and evoke negative feelings (Jacks and
Cameron, 2003). It is chosen to include negative affect in the contesting cluster as experiencing negative emotions can be viewed as a way of contesting the persuasive attempt.
Out of the three clusters, avoidance strategies can be considered to be the most passive. They involve the mere avoidance of persuasive attempts and are adopted before exposure to the persuasion attempt (Fransen et al., 2015). Empowering and contesting strategies, however, are adopted during or after the persuasive attempt, and are used to cope with it (Fransen et al., 2015). The strategies found in the contesting cluster can be regarded as the most active resistance strategies, as they are aimed against the persuasive attempt and challenge it. An overview of these clusters and their strategies is provided in Appendix 1, Table 1.
Motives for Resistance
Now that the resistance strategies have been defined and clustered, it is valuable to discuss why and when one is likely to use one of these strategies. According to Fransen et al. (2015), people have three motives to resist a message. These motives will be briefly discussed in the context of personalization; a more thorough review per resistance strategy cluster will follow in the section thereafter.
The first motive, threats to freedom, derives from Brehm’s psychological reactance theory (Brehm, 1966). The theory assumes that humans have an innate desire for autonomy and independence. As they sense that their freedom is threatened, they experience psychological reactance (Burgoon, Alvaro, Grandpre, & Voulodakis, 2002; Baek & Morimoto, 2012). When people experience this feeling, they are motivated to maintain and restore the opinion or behavior that has been threatened (Brehm & Brehm, 1981). In the context of advertising and marketing, persuasive attempts are often perceived as external threats to freedom. This could even result in a “boomerang-effect” in which the receivers distance themselves from the persuasive message, as well as become more motivated to engage less in or even contest the message (Fransen et al., 2015). Personalization could be a message factor triggering the feeling that one’s freedom is being threatened, as the individual could feel that his or her privacy has been invaded (Maslowska et al., 2011). This could result in the use of a resistance strategy, the type of strategy depending on the severity on the threat.
The second motive, reluctance to change, could be caused by either an
unwillingness to change, or a desire to stay the same (Fransen et al., 2015). The desire to not lose something of value; believing that the proposed change does not make sense; perceiving greater risks than benefits; and being satisfied with the current situation, are the identified reasons that could make people reluctant to change (Hultman, 1995; Kotter & Schlesinger, 2008). An enhancing factor involves the strength of existing attitudes and held beliefs; the stronger the attitudes, the less willing one might be to change; the same applies for the importance of one’s beliefs (Fransen et al., 2015). In the context of personalized online advertising, the desire not to lose something of value could be privacy, as privacy is valuable to most people (Arora et al., 2008). Also one might perceive the risks of personalization greater than the benefits of it. Therefore, one’s unwillingness to change could result in the use of a resistance strategy against the advertisement.
Concerns about deception is pointed out as the third and final motive by Fransen et al. (2015). This motive involves people’s desire to hold attitudes and opinions that are accurate. An example in the context of personalization might be the privacy statements or the terms and conditions websites often provide, for example alerting visitors that cookies are being collected. Website visitors often get confused by such privacy policies, which could result in concerns about being deceived (Smith, 2014). Persuasion knowledge is a factor that might increase concerns about deception
(Friestad & Wright, 1994). When an individual is exposed to a personalized online advertisement, s/he might be more aware that it is targeted directly at him or her when s/he has high persuasion knowledge; as a consequence, persuasion knowledge might contribute to the motive concerns about deception.
Different Resistance Strategies for Different Levels of Personalization
According to the avoidance, empowering and contesting typology used in this study, different types of resistance can be regarded. In the following section, these will be linked to the previously discussed levels of personalization.
As stressed before, this study will focus on the adaptation aspect of personalization. To explain the effects of adaptation, Dijkstra used (2008) the Elaboration Likelihood Model (Petty & Cacioppo, 1986). The model argues that people process information via one of two routes: the peripheral route or the central route. With the peripheral route, the message is not being evaluated in an elaborate way and persuasion may occur through peripheral cues. With central processing however, people are generally more involved, process information more thoroughly and evaluate relevant arguments. Adaptation could lead to central processing for several reasons (Dijkstra, 2008). The most significant reason is that the information the individual receives is relevant to him or her. The message may fulfill a need, or the information may match a psychological or behavioral style. In the case of such a match, an individual may process the message in a more elaborate way due to the similarity with him or herself (Dijkstra, 2008). That adaptation can lead to a more central way of processing is important when taking the level of personalization into account. In the case of highly personalized online advertisements, matching has the potential to produce less desired effects (Dijkstra, 2008). It could for example lead to an overload of self-relevant information which lowers persuasion effects (Burnkrant & Unnava, 1989; Na, 1999; Dijkstra, 2008). Central processing could also lead to a more critical evaluation of arguments presented in the message (Petty & Cacioppo, 1986). As a result, it can be argued that a high level of personalization could lead to a more critical evaluation of an advertisement. Hence this could provoke a more
negative response from a receiver; using a more active type of resistance strategy. This is in line with what research on health-related advertising shows, where central processing mostly induces defensive processing of information (Reed & Aspinwall, 1998).
A more critical evaluation is not the only effect a high level of personalization may achieve. When tied to the motives discussed earlier, it seems that a high level of personalization risks a stronger, more negative reaction from the receiver than lower levels of personalization. As an advertisement becomes more personal, all three motives for resistance proposed by Fransen et al. (2015) may grow stronger. A high level of personalization could trigger the first motive, the feeling of one’s freedom being threatened, as the receiver may feel that his or her privacy has been invaded (Malowska et al., 2011). For resistance to occur, the threatened freedom must be relevant to the individual (Chandler, 1990). As a high level of personalization means more relevancy to the receiver, the perception of threat may be greater. Also
according to Brehm’s psychological reactance theory, people experience reactance when they sense that their freedom is threatened (Brehm, 1966). A highly
personalized online advertisement could very well be seen as more of an intrusion of privacy than a non- or slightly personalized one; consequently, one may feel more threatened. Hence a higher level of personalization could provoke a more active response. In turn, this can be tied to the second motive. Reluctance to change is triggered by the desire to not lose something of value. As privacy is something most people value (Arora et al., 2008), and as a higher level of personalization might be viewed more as an intrusion of privacy, the second motive might be strongest as well in the case of a high level of personalization. Finally, the realization that a source has gathered specific information about him or her could result in a feeling of loss of control. The feeling of no control may in turn cause a stronger worry about being deceived, the third and final motive. As a way to neutralize an individual’s motivation for resistance is to provide him or her with control over the situation, the feeling of no control might evoke stronger motivations to resist the persuasion attempt (Fransen et al., 2015). Also the realization of the receiver itself that the advertisement is meant specifically for him or her, could increase his or her concerns about deception (Friestad & Wright, 1994; Fransen et al., 2015). Because this realization is more likely to happen in the case of a high level of personalization, the motive might be strongest in the case of high personalization as compared with lower levels of personalization.
A high level of personalization may thus provoke a more critical way of processing a message. Furthermore, as the level of personalization increases, the motivations of resistance may grow stronger; hence, when compared to slightly and
non-personalized online advertising, a high level of personalization may have the strongest effect on all three motives of resistance. As a result, a receiver may use a more active strategy when exposed to a highly personalized online advertisement than to a non- or slightly personalized one. As contesting strategies were identified as the most active resistance strategies, the following hypothesis is proposed.
H1: Highly personalized online advertisements (vs. non- or slightly
personalized online advertisements) are more likely to induce the use of contesting strategies (vs. avoidance or empowering strategies).
As a personalized online advertisement can lead to more elaborate processing, one can argue that a non-personalized online advertisement could result in the opposite. Because the advertisement is being processed in a less critical way, the receiver might be more likely to use a more passive resistance strategy. Also a non-personalized online advertisement could generate less attention by the receiver, as the advertisement is less relevant for him or her (Dijkstra, 2008; Maslowska et al., 2011). This is also in line with the phenomenon that is suggested to be countered by
personalization; banner blindness. Banner blindness can be described as the tendency for users to avoid attending to a banner or advertisement or anything that resembles banner ads (Resnick & Albert, 2014). In their eye-tracking study, Resnick and Albert (2014) found that participants paid less attention to locations on a web page that were less likely to contain relevant content. As a non-personalized online advertisement is likely to be less relevant for an individual than a personalized one, it could thus lead to less attention (De Keyzer et al., 2015). Another eye-tracking study dedicated to personalized banners (Köster et al., 2015) found exactly that; personalized banners received more attention than less personalized banners. Avoidance may occur naturally when encountering a non-personalized online advertisement, because one simply pays less attention.
That avoidance is more likely to happen when encountering a non-personalized online advertisement than a personalized one is what can also be derived from the motives of resistance proposed by Fransen et al. (2015). When compared to any level of personalization, a non-personalized online advertisement is probably least likely to pose a threat to someone’s freedom.. Also, a non-personalized online advertisement is probably least likely to be seen as a risk. Therefore, reluctance to change is probably least strong for a non-personalized online advertisement. Finally, a non-personalized online advertisement is probably least likely to raise concerns about deception as it is
least likely to be seen as an intrusion of privacy when compared to different levels of personalization (Maslowska et al., 2011).
In sum, one may process a non-personalized online advertisement less critically, and may be less likely to notice the advertisement because it is not personally relevant per se. A non-personalized online advertisement probably poses least of a threat, is thus seen as least of a risk, and is probably least likely to raise concerns about deception. Because of this, one may use the least active resistance strategies as an answer to the advertisement. Hence the following hypothesis is proposed.
H2: Non-personalized online advertisements (vs. slightly or highly personalized online advertisements) are more likely to induce the use of avoidance strategies (vs. empowering or contesting strategies).
A slightly personalized online advertisement should be not as relevant to the consumer as highly personalized one, yet more relevant than a non-personalized one. As a result, a slightly personalized online advertisement is likely to be processed less critically than a highly personalized one, but more critically than a non-personalized one. Also all three motives may be less strong for a slightly personalized online advertisement than for a highly personalized one; yet the motives for a slightly personalized online advertisement may be stronger than for a non-personalized one. As empowering strategies are not as passive as avoidance strategies and not as active as contesting strategies, the empowering strategies may dominate for a slightly personalized online advertisement. Therefore the third hypothesis is proposed.
H3: Slightly personalized online advertisements (vs. non- or highly personalized online advertisements) are more likely to induce the use of empowering strategies (vs. avoidance or contesting strategies).
A visualization of these hypotheses is found in Appendix 2, Image 1.
Persuasion Knowledge and Privacy Concerns
Two concepts seem to take a central place in the discussion of personalization, namely persuasion knowledge and privacy concerns (Debatin, Lovejoy, Horn, & Highes, 2009; Stanaland, Lwin, & Miyazaki, 2011; Baek & Morimoto, 2012; Duhigg, 2012; Hill, 2012; Singer, 2012; Singer, 2012; Fransen, Ter Hoeven, & Verlegh, 2013; Tucker, 2014; Nu.nl, 2015). For this reason, these concepts will be investigated as potential mediators in the relationship between personalization and the use of resistance strategies. The concepts will be briefly discussed and hypotheses will be proposed.
According to the persuasion knowledge model, people develop knowledge about the strategies marketers use in their persuasive attempts (Friestad & Wright, 1994). This knowledge can then be used as protection against the attempt and such control may result in resistance (Fransen & Fennis, 2014; Voorveld & Van Noort, 2014). Put in other words, when consumers become suspicious of the manipulative intent of certain content, they are more likely to resist the message. They will feel restricted in their freedom of thinking and feeling what they want, which leads them to become actively motivated to restore this freedom (Fransen & Fennis, 2014). When a message is not perceived as persuasive, it conveys more favorable affective
responses as well as having a positive effect on purchase intentions than when it is not perceived as such (Voorveld & Van Noort, 2014). Research already found that the persuasive impact of a message is undermined when the message or the source is perceived as using manipulative tactics (Sagarin, Caldini, Rice, & Serna, 2002). Fransen et al. (2013) even consider contesting the persuasive tactics used in an advertisement as yet another resistance strategy, namely invoking persuasion
knowledge. This strategy seems to be related to the motive concerns of deception. As stressed before, a receiver’s realization that the advertisement is meant specifically for him or her could increase concerns of deceptions (Friestad & Wright, 1994; Fransen et al., 2015). Hence one’s persuasion knowledge may contribute to one’s concerns of deception, and thus be a motivation for the use of a resistance strategy.
When an advertisement is personalized, the receiver might pick up quicker that someone is trying to influence him or her as it is directed to him or herself. Therefore, in the case of personalization, the manipulative intent might be clearer as the
advertisement is made relevant for the receiver. Personalization, regardless of what type or level, might be viewed as a manipulative tactic, which may result in the receiver becoming more concerned about deception. Hence, the receiver might be more likely to use any type of resistance strategy due to his or her persuasion knowledge. Put differently, personalization might heighten one’s persuasion
knowledge, which in turn might increase the use of resistance strategies. This leads to the fourth, fifth and sixth hypothesis:
H4: Persuasion knowledge mediates the relation between personalization and the use of avoidance strategies.
H5: Persuasion knowledge mediates the relation between personalization and the use of empowering strategies.
H6: Persuasion knowledge mediates the relation between personalization and the use of contesting strategies.
A second mediating effect may occur due to privacy concerns. The concept privacy can be defined as the claim of individuals to determine for themselves when, how and to what extent information about them is communicated to others (Westin, 1967). More fitting with the current discussion, Goodwin (1991) already added the consideration of the specific content of information (data) that is stored in a database, and the likelihood that this information will be used to harm the individual. This is where personalization comes in. Data, which serves as the base of personalization, is regarded as essential in order to design modern and effective business strategies (Leeflang, Verhoef, Dahlström, & Freundt, 2014). However, it is often unclear to consumers what happens with their data, and this consequently spurs privacy
concerns (Debatin et al., 2009; Stanaland et al., 2011; Nu.nl, 2015). To illustrate this problem, researchers Youyou, Kosinki and Stingwell (2015) demonstrated how data can be used in a privacy sensitive manner. They showed that by using data retrieved from social media giant Facebook, rather accurate profiles of individuals could be created. In fact, an individual’s personality traits were predicted more accurately by a computer than by most of the particular individual’s friends and family. These
predictions were derived from the individual’s likes on Facebook, which the researchers then linked to certain personality traits. One’s likes on Facebook thus already serve as a simple example of data, which can be used in a great way; creating accurate profiles of individuals. Because personalization thrives on data,
personalization of advertising is viewed as a rather delicate issue (Arora et al., 2008). As consumers remain fearful of the usage of their personal data (Baek & Morimoto, 2012; Morrison, 2016) and personalization is based on personal information,
personalized advertising has the potential to raise consumer’s privacy concerns (Baek & Morimoto, 2012). This is because personalized content highlights the fact that consumer information is being used, which in turn heightens privacy concerns (Baek & Morimoto, 2012; Aguirre, Roggeveen, Grewal, & Wetzels, 2016).
Privacy concerns are perceived to induce several unwanted effects of
personalization. They may increase skepticism towards a personalized advertisement (Baek & Morimoto, 2012). Also they heighten the receiver’s risk perceptions,
decrease his or her trust and thus prompt negative reactions (Aguirre et al., 2016). These negative implications could have something to do with the feeling of loss of
control, as Milne and Boza (1999) suggest that when consumer’s control is high, privacy concerns are lowest. The feeling of no control might contribute to the risk perceived. This is in line with the findings of Ziesak (2013); privacy concerns seem to decrease the value of personalization, and increase people’s risk beliefs. The more receivers are concerned, the more risk is perceived, the less value they see in the advantages of personalization.
Exposure to a personalized online advertisement could lead to the realization of one’s personal data being used. This realization could raise privacy concerns,
regardless of the level of personalization. Due to the negative effects of privacy concerns, a receiver may become more motivated to use a resistance strategy. Thus the relationship between personalization and the use of resistance strategies might become stronger. In other words, personalization might heighten one’s privacy concerns, which in turn may increase the use of resistance strategies.
This leads to the seventh, eighth and ninth hypothesis:
H7: Privacy concerns mediate the relation between personalization and the use of avoidance strategies.
H8: Privacy concerns mediate the relation between personalization and the use of empowering strategies.
H9: Privacy concerns mediate the relation between personalization and the use of contesting strategies.
A visualization of these hypotheses is found in Appendix 2, Image 2.
Attitude and Noticing the Advertisement
This study revolves around personalization and resistance strategies. Therefore until now, mainly negative effects of personalization have been discussed. However, as already pointed out in the introduction, personalization is seen as the future of advertising and can be a powerful tool when executed correctly (Edwards, 2016). Personalization aims to maximize immediate as well as future business opportunities (Tam & Ho, 2006). By making advertising relevant for the consumer it has the potential to be more effective than non-personalized advertising (Maslowska et al., 2011). Tucker (2014) even states that personalized advertisements could be twice as effective as similar, yet non-personalized advertisements. Previous studies indeed found several positive effects of personalization. Through personalization, consumers experience an improvement in services, a better preference match, more convenience, as well as reduced cognitive overload (Ansari & Mela, 2003; Versanen, 2007; Aguirre
et al., 2015). Personalization showed to increase recall, content evaluations, as well as purchases in an online advertising context (Tam & Ho, 2006). Additionally, Baek and Morimoto (2012) found skepticism towards an advertisement to be lower in the case of personalization. Also it has been argued that because a personalized advertisement is made more relevant to the receiver, it has the potential to increase attention and thus get noticed more than a non-personalized advertisement (De Keyzer et al., 2015; Köster et al., 2015).
The theoretical framework of this study argued that different levels of personalization induce the use of different types of resistance strategies. However, previous research has also found positive effects of personalization. Therefore, a set of exploratory analyses will be conducted to investigate two additional concepts: attitude towards the brand and noticing the advertisement. It is chosen to investigate potential effects of personalization on attitude towards the brand, because it is useful in predicting consumer behavior (Spears & Singh, 2004). Next, as banner blindness was proposed as a problem that could be solved through personalization (Resnick & Albert, 2014; Köster et al., 2015), noticing the advertisement is chosen to investigate as a second possible positive outcome of personalization. Therefore, two more hypotheses are proposed.
H10: Participants exposed to a personalized online advertisement have a more positive attitude towards the brand, than participants exposed to a non-personalized online advertisement.
H11: Participants exposed to a personalized online advertisement are more likely to notice the advertisement, than participants exposed to a non-personalized online advertisement.
A visualization of these hypotheses is found in Appendix 2, Image 3.
Method Participants and Design
To test the effects of different levels of personalization on the use of different types of resistance strategies, a single factor between-subjects design was used. An online experiment was created in Qualtrics for the purpose of this study. By posting the survey-link on Facebook and Instagram, spreading the link via e-mail to
colleagues, asking friends and family to spread the survey, as well as handing out the link of the survey on paper to participants in various fitness classes, a convenience
sample of 363 participants was gathered. Out of these participants, 118 did not finish the study (32.5%) and two participants indicated they did not agree to the terms, and were thus removed from the study. Another 33 participants (13.6%) indicated to not have seen the advertisement and were therefore not included in further analyses. Out of the remaining, five participants incorrectly indicated the brand of the advertisement they were exposed to and hence were excluded from the study. Finally, one more participant was removed from the study as he answered every statement with five, and seemed to have not taken the experiment seriously.
This resulted in a total of 204 participants, of which 143 were female (70.1%) and 61 were male (29.9%) with an average age of 28 years (M = 28.38, SD = 10.87). The sample consisted mainly of Dutch participants (51.5%), followed by German participants (26.5%). The majority of the participants indicated to be highly educated, with 38.2% having completed a Bachelor’s Degree, and 24.5% a Master’s Degree.
The participants were randomly assigned to one of the three advertising conditions: non-personalized, slightly personalized or highly personalized. This resulted in 58 participants in the non-personalized condition (28.4%). For the slightly personalized condition, this resulted in 85 participants (41.7%). Finally, 61
participants (29.9%) were assigned to the highly personalized condition.
Procedure
To manipulate different levels of personalization, three different quizzes were designed to gather personal data of the participants. This data was used to
subsequently personalize the advertisement. Before starting the quiz, participants were asked for their gender. A brief description per condition will now follow.
For the highly personalized condition, the quiz the participants were exposed to was based on gender. Participants got to choose between several paired images, among them H&M and Zara. The images of the products shown in these quizzed were extremely similar to the products in the advertisement participants were exposed to later. To which advertisement the participant would be exposed to was based on the participant’s gender and choice between H&M and Zara. For example, when a female participant chose Zara, she would be exposed to the Zara advertisement for women. When a male participant chose H&M, he would be exposed to the H&M
advertisement for men. Therefore, the advertisement participants would be exposed to was dependent on their personal preferences indicated in the quiz.
In the slightly personalized condition, the quiz was the same for males and females. Participants were asked to choose between several paired images, among them H&M and Zara again. Participants were exposed to an advertisement based on their gender and their choice between H&M and Zara, just as in the highly
personalized condition. However, instead of having to choose between products that were shown later in the advertisement, participants were only exposed to one element that was relevant to the advertisement they would be exposed to later; the brand shown in the advertisement (H&M or Zara).
The non-personalized quiz was the same for males and females as well. This time however, exposure was not based on gender, and the paired images participants had to choose between had nothing to do with the advertisement they would be exposed to later. For example, participants were asked to choose between Microsoft and Apple. The H&M or Zara question thus did not return. After choosing between the paired options, participants were randomly exposed to one of the four
advertisements. As a result, the advertisement participants were exposed to was not dependent on their personal preferences in the non-personalized condition. Now that the conditions have been explained, a description of the advertisements will follow.
Four fashion advertisements were designed specifically for this study: two advertisements for men, and two for women. The brands Zara and H&M were selected for the advertisements because they are widely known, each other’s competition, and their logos easily recognizable. The imagery used for the
advertisement was derived from old lookbooks of H&M and Zara. In general, H&M’s focus is more on casual wear with a lower price point than Zara; Zara’s focus is more on business and formal wear with a slightly higher price point than H&M (Smith, 2014). Based on this perception, two types of advertisements were designed: one of casual wear and one of formal wear. These advertisements were designed for males and females. Thus for the advertisements for men, this resulted in a H&M
advertisement of a male model wearing casual clothing and a Zara advertisement of a male model wearing formal wear. For the advertisements for women, the same principle applied. For the highly and slightly personalized conditions, the
advertisement participants would be exposed to was based on gender and their choice between H&M and Zara. In the non-personalized condition, the advertisements were randomly assigned to the participants. In the experiment, each advertisement was placed next to a website article. As a result, participants were exposed to what looked
like a general webpage. The article originated from the existing website Health.com and evolved around tips against stress (Greene, n.d.). Images of the stimulus material are found in Appendix 3.
Pretest. Prior to the actual experiment, a pretest among 39 participants (74.4%
female, M = 1.74, SD = 0.44) was conducted in order to make sure the stimulus material achieved the desired effects. The advertisement in the non-personalized condition should be considered by the participants as non-personalized; the slightly personalized online advertisement as slightly personalized; and the highly
personalized online advertisement as highly personalized. Five participants were removed for not completing the survey, and one participant was removed out for indicating not seeing the advertisement, leaving 33 participants for the analyses. After conducting the quiz, participants were asked to rate four statements on a 5-point Likert scale (1 = Not true at all, 5 = Very true), in order to determine the degree of perceived personalization of the advertisements by the participants, e.g. “I felt like the advertisement was personalized for me.” An overview of these statements are found in Appendix 4. These items were averaged into a perceived personalization scale, α = .87 (M = 2.66, SD = 1.01). The participants (N = 14) in the non-personalized
condition showed that they indeed did not view the advertisement as personalized (M = 2.02, SD = 0.73). However, the difference between the slightly personalized and the highly personalized condition was not as desired: the participants in the slightly personalized condition (N = 12) scored M = 3.21 (SD = 0.98) on perceived
personalization, while the participants in the highly personalized condition (N = 7) scored M = 3.00 (SD = 0.90) on perceived personalization. An analysis of variance showed that these results were significant, F (2, 30) = 6.82, p = .004, which thus indicates that the manipulation was not successful. Therefore, an adaptation was made in the actual experiment. In the pretest, participants in the highly personalized
condition were exposed to an advertisement based on choice of clothing (jeans or suit, jeans or dress). Initially it was chosen to assign the participants based on the choice of clothing because the pieces of clothing they could choose in the quiz were extremely similar to the pieces of clothing shown in the advertisement. However, the pretest indicated that the choice of brand might evoke stronger feelings of personalization; therefore in the experiment, participants were exposed to an advertisement based on choice of brand (H&M or Zara).
Dependent variables. A total of 25 items was used to measure the ten
identified resistance strategies, based on the research of Fransen et al. (2013). Ten more filler statements were added in order to minimize biased answers. All 35 items are found in Appendix 5.
A principal component factor analysis with varimax rotation was conducted over the 25 items for the ten different resistance strategies. The analysis was set for three factors, in order to test if the three proposed clusters (i.e., avoidance,
empowering and contesting) could be formed in line with the argued framework. The Kaiser-Meyer-Olkin test showed that the size of the sample was large enough to conduct the factor analysis, KMO = .826. The three set factors showed eigenvalues above Kaiser’s criterion of 1, and together explained 48.78% of the total variance. Counter arguing, message derogation, source derogation and negative affect formed the first factor (contesting); attitude bolstering and self-assertion formed the second factor (empowering); and finally, avoidance and distraction formed the third factor (avoidance). The factor analysis accounted the two items for selective exposure to the second factor (empowering), yet the factor loadings were very close to the loadings in the third factor. Therefore, based on theory, it is chosen to include them in the
avoidance strategies cluster. A similar effect was shown for the two items for social validation. The items were accounted to the first factor (contesting), yet the loadings were very close to the loadings in the second factor (empowering). Based on theory, these items were included in the empowering strategies cluster.
A reliability analysis indicated that the eight items for avoidance, selective exposure and distraction formed a reliable scale (α = .74); therefore these items were averaged into a scale for the avoidance strategies (M = 2.90, SD = 0.66). The seven items for attitude bolstering, self-assertion and social validation showed a reliable scale as well (α = .82). Hence, these items were averaged into a scale for the empowering strategies (M = 2.26, SD = 0.75) was formed. Finally, the items for counter arguing, source derogation, message derogation and negative affect also proved a good fit (α = .87) which resulted in averaging these items into a scale for the contesting strategies (M = 2.13, SD = 0.71).
Mediator variables. Persuasion knowledge was measured through five items
on a 5-point Likert scale (1 = Not true at all, 5 = Very true). An example of an item would be “I thought how the ad tries to persuade me.” These items were derived from the research of Fransen et al. (2013); an overview of these statements can be found in
Appendix 6. A reliability analysis showed that these five items formed a reliable fit, α = .78. Therefore these items were averaged into one measure for persuasion
knowledge (M = 2.92, SD = 0.88).
Privacy concerns was measured through four items on a 5-point Likert scale (1 = Not true at all, 5 = Very true). An example of an item would be “I believe that personal data are being misused too often.” These items were derived from the research of Baek and Morimoto (2012); an overview of these items can be found in Appendix 6. A reliability analysis showed that these four items formed a reliable scale, α = .85, therefore the items were averaged into one measure for privacy concerns (M = 3.78, SD = 0.82).
Attitude and noticing the advertisement. Attitude towards the brand was
measured for both H&M and Zara through five items on a 5-point semantic scale, for example “I think H&M is … Unappealing – Appealing.” The items were derived from the research of Spears and Singh (2004); an overview can be found in Appendix 7. Two potential positive results of personalization were investigated: attitude and noticing the advertisement. After conducting a reliability analysis, the items were averaged into two reliable scales: attitude for H&M (α = .93, M = 3.69 , SD = 0.73) and attitude for Zara (α = .92, M = 3.78, SD = 0.75). Finally, noticing the
advertisement was measured by simply asking the participants: “Did you notice the ad?”
Results
Manipulation check. A one-way ANOVA showed that participants in the
actual experiment indeed viewed the advertisement different across conditions based on personalization, F (2, 201) = 11.49, p < .001, η 2 = .10. Unfortunately a post hoc test indicated that the manipulation was again not entirely successful. Participants who were in the non-personalized condition perceived the advertisement as less personalized (M = 2.21, SD = 1.24) than the participants in the slightly (M = 3.11, SD = 1.33) and highly personalized conditions (M = 3.28, SD = 1.39). The
non-personalized condition significantly differed from the slightly non-personalized condition with a mean difference of 0.90 (p < .001), as well with the highly personalized condition with a mean difference of 1.07 (p < .001). However, the slightly personalized condition did not differ significantly from the highly personalized condition, with a mean difference of 0.17 (p = 1.000). For this reason it was chosen to
make one personalized condition out of the slightly and the highly personalized condition. This resulted in 58 participants in the non-personalized condition (28.4%), and 146 participants in the personalized condition (71.6%). In the non-personalized condition, 28 participants (48.3%) were exposed to a H&M advertisement; 30 participants to a Zara advertisement (51.7%). In the personalized condition, 79 participants (54.1%) were exposed to a H&M advertisement, and 67 (45.9%) to a Zara advertisement. With a mean difference of 0.06, an independent samples t-test showed that there was no difference between being exposed to H&M or Zara in the non-personalized and the personalized condition, t (202) = 0.75, p = .454.
Randomization check. To ensure that the participants were randomly assigned
to the conditions, a MANOVA was conducted. The MANOVA indicated no
significant effects across the conditions for age, nationality and education. Therefore the randomization was successful and thus no control variables were taken into account. An overview of these results can be found in Appendix 8, Table 2.
Main hypotheses. Unfortunately, Hypotheses 1 and 3 cannot be investigated in
the way they were formulated because of the failed manipulation. Therefore, two other effects will be taken into account. First, it will be investigated which resistance strategies are used the most by the participants (avoidance, empowering or contesting strategies). Second, Hypothesis 2 will be answered and it will be investigated if personalization is more likely to induce the use of resistance strategies (vs. no personalization).
A GLM repeated measures was conducted in order to see if participants used different resistance strategies in the different conditions. The assumption of sphericity was met, χ2 = 0.86, p < .001. This indicates that despite the unequal sizes of the two groups, the variances of the differences between the levels of the independent variables (the two conditions) are equal. First, the results show that there was a significant difference between the use of resistance strategies, F (2, 404) = 92.80, p < .001. A test of within subject contrast showed that the use of avoidance strategies (M = 2.89, SD = 0.65) significantly differed from the use of empowering strategies (M = 2.26, SD = 0.75), F (1, 202) = 77.12, p < .001. Also the use of empowering strategies significantly differed from the use of contesting strategies (M = 2.11, SD = 0.68), F (1, 202) = 0.59, p = .001. Therefore it can be said that participants used avoidance strategies the most in resisting the advertisement of this study; followed by empowering strategies. The contesting strategies were used the least. Second, the
GLM repeated measures showed no main effect of personalization, F (1, 202) = 0.00,
p = .974. Also there was no interaction effect found for the resistance strategy clusters
with the non-personalized and personalized condition, F (2, 404) = 0.48, p = .617. As a result, Hypothesis 2 was rejected: non-personalized online advertisements (vs. slightly or highly personalized online advertisements) are not more likely to induce the use of avoidance strategies (vs. empowering or contesting strategies).
Although no significant effects were found, independent samples t-tests were conducted as an enhancement to these results. Participants in the non-personalized condition scored 2.86 on the avoidance strategies (SD = 0.60); in the personalized condition they scored 2.90 (SD = 0.67). This small difference was not significant, t (202) = -0.36, p = .722, 95% CI = [-0.24, 0.16]. The same effects were found for the empowering and contesting strategies. Empowering strategies were not used more often in the non-personalized (M = 2.31, SD = 0.83) than in the personalized (M = 2.24, SD = 0.71) condition, t (202) = 0.61, p = .546, 95% CI = [-0.16, 0.30]. Lastly, participants did not use contesting strategies more often in the non-personalized condition (M = 2.10, SD = 0.68) than in the personalized condition (M = 2.12, SD = 0.67), t (202) = -0.25, p = .807, 95% CI = [-0.24, 0.18]. Therefore, participants do not seem to use resistance strategies more often in the case of personalization (vs. no personalization).
An overview and a visualization of the means of the resistance strategies per condition is found in Appendix 9, Table 3 and Image 3.
Mediation hypotheses. The PROCESS tool (Hayes, 2012) makes it possible to
test the direct effects and mediation effects of the independent variables on the dependent variables. Therefore, this tool was used to examine the two possible mediators on the relationship between personalization and the use of resistance strategies. The analyses were conducted per resistance strategy cluster. First, the role of persuasion knowledge will be investigated; second, the role of privacy concerns.
Hypothesis 4 proposed that persuasion knowledge mediates the relation between personalization and the use of avoidance strategies. The explained variance of the model was not significant, R2= .01, F (2, 201) = 1.23, p = .290. The analysis again showed no direct effect, indicating that personalization did not predict the use of avoidance strategies, β = 0.03,t = 0.33, p = .742. Moreover, it revealed that
persuasion knowledge did not predict the use of avoidance strategies, β = 0.09, t = 1.55, p = .123. Also no indirect effect was found, β = 0.00, 95% CI [-0.01, 0.06].
Hypothesis 4 was not supported. A visualization of these results is found in Appendix 10, Image 4.
Hypothesis 5 proposed that persuasion knowledge mediates the relation between personalization and the use of empowering strategies. The explained variance of the model was significant, R2= .27, F (2, 201) = 38.52, p < .001. The analysis again showed no direct effect, indicating that personalization did not predict the use of empowering strategies, β= -0.09,t = -0.86, p = .393. The results reveal a
significant effect of persuasion knowledge on the use of empowering strategies, β = 0.44, t = 8.71, p < .001. However, no indirect effect was found, β = -0.09, 95% CI [-0.08, 0.14]. The results thus indicate a relationship between persuasion knowledge and the use of empowering strategies, and showed no mediation effect; therefore, Hypothesis 5 was not supported. A visualization of these results is found in Appendix 10, Image 5.
Hypothesis 6 proposed that persuasion knowledge mediates the relation between personalization and the use of contesting strategies. The explained variance of the model was significant, R2= .19, F (2, 201) = 28.04, p < .001. The analysis again showed no direct effect, indicating that personalization did not predict the use of contesting strategies, β= 0.01 ,t = 0.10, p = .924. The results reveal a significant
effect of persuasion knowledge on the use of contesting strategies, β = 0.34, t = 7.48, p < .001. However, no indirect effect was found, β = 0.02, 95% CI [-0.08, 0.10]. The results thus indicate a relationship between persuasion knowledge and the use of contesting strategies, and showed no mediation effect; therefore, Hypothesis 6 was not supported. A visualization of these results is found in Appendix 10, Image 6.
Hypothesis 7 proposed that privacy concerns mediate the relation between personalization and the use of avoidance strategies. The explained variance of the model was significant, R2= .05, F (2, 201) = 3.71, p = .026. The analysis again showed no direct effect, indicating that personalization did not predict the use of avoidance strategies, β= 0.05,t = 0.49, p = .628. The results reveal a significant
effect of privacy concerns on the use of avoidance strategies, β = 0.17, t = 2.72, p = .007. However, no indirect effect was found, β = -0.10, 95% CI [-0.06, 0.03]. The results thus indicate a relationship between privacy concerns and the use of avoidance strategies, and showed no mediation effect; therefore, Hypothesis 7 was not
Hypothesis 8 proposed that privacy concerns mediate the relation between personalization and the use of empowering strategies. The explained variance of the model was not significant, R2= .01, F (2, 201) = 0.43, p = .651. The analysis again showed no direct effect, indicating that personalization did not predict the use of empowering strategies, β = -0.07,t = -0.59, p = .558. Moreover, it revealed that
privacy concerns did not predict the use of empowering strategies, β = 0.07, t = -0.59, p = .558. Also no indirect effect was found, β = 0.00, 95% CI [-0.01,
0.05]. Hypothesis 8 was not supported. A visualization of these results is found in Appendix 10, Image 8.
Hypothesis 9 proposed that privacy concerns mediate the relation between personalization and the use of contesting strategies. The explained variance of the model was not significant, R2= .01, F (2, 201) = 0.79, p = .454. The analysis again showed no direct effect, indicating that personalization did not predict the use of contesting strategies, β = 0.03,t = 0.29, p = .773. Moreover, it revealed that privacy
concerns did not predict the use of contesting strategies, β = 0.07, t = 1.24, p = .217. Also no indirect effect was found, β = -0.01, 95% CI [-0.04, 0.01]. Hypothesis 9 was not supported. A visualization of these results is found in Appendix 10, Image 9.
Attitude and noticing the advertisement. Because the mean difference of the
attitude for H&M and attitude for Zara was only 0.10, one measure for attitude was formed based on the averages (M = 3.73, SD = 0.74). A one-way ANOVA showed that the non-personalized (M = 3.52, SD = 0.75) and the personalized condition (M = 3.82, SD = 0.73) differed from each other in attitude, F (1, 202) = 6.68, p = .010, η 2 = .03. This small effect showed that participants hold a slightly higher attitude for a personalized online advertisement than for a non-personalized online advertisement. Therefore, Hypothesis 10 was accepted.
In order to test Hypothesis 11, the filter that excluded the 33 participants that did not notice the advertisement was turned off. Crosstabs showed that 17 out of 75 participants (22.7%) did not notice the advertisement in the non-personalized
condition and 16 out of 162 participants (9.9%) did not notice the advertisement in the personalized condition. Chi square indicated that this difference between the
conditions was significant, χ2 (1) = 7.00, p = .008. This seems to indicate that participants noticed the advertisement less in the non-personalized condition than in the personalized condition. An independent samples t-test was conducted to take a closer look at this result.. It indeed showed a significant effect of personalization on
noticing the advertisement, t (110) = 2.34, p = .020, 95% CI = [-0.02, 0.24]. Therefore participants exposed to a personalized online advertisement were more likely to notice the advertisement, than participants exposed to a non-personalized online
advertisement. Hypothesis 11 was accepted.
General discussion
The aim of this study was to investigate if consumers use different resistance
strategies for different levels of personalization of online advertisements. Through an experiment it was investigated whether highly personalized online advertisements are more likely to induce the use of contesting strategies (H1); whether non-personalized online advertisements are more likely to induce the use of avoidance strategies (H2); and whether slightly personalized advertisements are more likely to induce the use of empowering strategies (H3). Next, two potential mediators of these relationships were investigated, namely persuasion knowledge and privacy concerns. Finally, attitude towards the brand and noticing the advertisement were investigated as as two potential positive effects of personalization.
Unfortunately because of partially unsuccessful manipulations, two main hypotheses (H1 and H3) could not be investigated in the way they were phrased. The hypotheses were based on three different levels of personalization; non-personalized, slightly personalized and highly personalized. A manipulation check showed that the participants in the experiment did not perceive the slightly personalized condition differently from the highly personalized condition. Therefore, it was decided to combine the slightly and the highly personalized condition into one condition, namely the personalized condition. This way, differences between non-personalized online advertising and personalized online advertising could still be investigated, although not in different levels. Despite the unequal sizes of the two groups, the variances of the differences between the levels of the two conditions showed to be equal.
Results indicated that personalization did not influence the receiver’s choice of resistance strategy. For each of the three types of resistance strategies, no significant differences were found between the non-personalized and the personalized condition. Hypothesis 2, which proposed that non-personalized online advertisements are more likely to induce the use of avoidance strategies, was rejected. Avoidance strategies however proved to be the most dominant out of the three types of strategies, followed by empowering strategies. Contesting strategies were used the least. As avoidance has
been regarded as the most prevalent type of resistance traditionally (Van ‘t Riet & De Ruiter, 2013), this outcome is in line with previous research.
There were no mediating effects found; hence, H4, H5, H6, H7, H8 and H9 were rejected. However, other effects were found in conducting mediation analyses. Persuasion knowledge showed to be a predictor of the use of empowering strategies and the use of contesting strategies; the more active resistance strategies. The least active resistance strategies, avoidance strategies, showed to be predicted by privacy concerns.
In line with the present findings, Baek and Morimoto (2012) demonstrated privacy concerns to have a positive impact on avoidance. However, they did not take other resistance strategies into account. Why privacy concerns only seem to induce the use of avoidance strategies, and persuasion knowledge the use of empowering and contesting strategies, may be explained by the study of Smit et al. (2014). The
researchers showed an important interaction between persuasion knowledge and privacy concerns, namely that participants with the least knowledge held the strongest concerns. Milne and Boza (1999) already suggested that when consumer’s control is high, privacy concerns are lowest, because persuasion knowledge might offer a sense of control. A receiver having more privacy concerns may thus be explained by simply having less knowledge. Furthermore, as people tend to resist persuasion attempts when they recognize them as such (Quinn & Wood, 2004; Van Reijmersdal et al., accepted), it is plausible that the receiver then has a more active response when s/he encounters a personalized advertisement. Recognizing a personalized advertisement as a way to persuade him or her could lead to more critical processing (Petty & Cacioppo, 1986), and as previously argued, also to increased motives for resistance. Thus it may explain the use of more active strategies. When a receiver simply does not have the knowledge however, s/he may avoid the advertisement naturally as s/he does not know how to cope with it in a different way. Hence, the advertisement may be processed less critically, motives for resistance may be less strong, and as a result one may be less likely to respond with a more active resistance strategy.
After conducting main and mediation analyses, two potential positive effects of personalization were investigated. Participants showed to hold a higher attitude for the brand when exposed to a personalized online advertisement than when exposed to a non-personalized one, supporting Hypothesis 10. This positive impact of
Grier (2000), who found that targeted advertising can lead to more favorable attitudes. Finally, Hypothesis 11, which stated that participants exposed to a personalized online advertisement are more likely to notice the advertisement (vs. a non-personalized one), was supported as well. Consequently, this study showed that personalization contributes to receivers noticing an advertisement. This result is supported by
research on banner blindness which suggests that by making advertising relevant, it is more likely to get noticed (Resnick & Albert, 2014; Köster et al., 2015). The present positive findings enhance earlier results as they may offer an explanation of why resistance strategies are not used more often when an advertisement is personalized. A positive attitude may counter possible negative effects of personalization, and thus one may not be more likely to use a resistance strategy in the case of personalization. However, more research is needed in order to make any definite conclusions. Another possible explanation might be the fact that the level of personalization in this study was not high enough. This may also explain why personalization did not contribute to the use of a resistance strategy. Because the advertisement was not seen as too
intrusive, it might have resulted in a more positive attitude and thus a less negative response. This points out a major shortcoming in this research, which is found in the partially unsuccessful manipulations. As a result, different levels of personalization could not be taken into account. Future research could therefore investigate the effects of different types of personalization through better manipulations. In the case of this experiment, participants had to choose between logos of competing, well-known fashion brands, H&M and Zara. This might have caused a too harsh priming effect, as it could have raised participants’ expectations of getting exposed to an advertisement from one these brands. Therefore it is suggested to use different brands, less known brands or no brands at all as this may have been another factor affecting the results. Previous attitudes towards H&M and Zara might have had an influence on the
participants’ perceptions of the advertisements. Additionally, the brands may have not been relevant enough to participants, which may have also affected the receiver’s perceived personalization of the advertisements.
A further shortcoming of this research is found in certain operationalizations. Measuring resistance strategies remains difficult, as it can be argued that using a resistance strategy is a subconscious process. Participants may not have recognized their actions in the statements that were asked in the survey. The other way around, their answers might actually be biased through asking explicitly. As a result it can be
