A comparison between the effects of generic and smart
advertisements on consumer response in online and
offline advertising environments
Master ThesisDewi Simon 10528091 22-09-2018 Final version
MSc. In Business Administration – Marketing track Amsterdam Business School, UvA
Statement of originality
This document is written by Student, Dewi Simon, who declares to take full
responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and
that no sources other than those mentioned in the text and its references have
been used in creating it.
The Faculty of Economics and Business is responsible solely for the
supervision of completion of the work, not for the contents.
Abstract
The Internet of Things (IoT) makes it possible to make advertisements more personalized than ever, the so-called smart advertisements. This study is investigated what the effect of smart advertisements was on diverse consumer responses (Advertising attitude, privacy concerns, level of intrusiveness) and if this differs if the ad is seen in a different media environment (online/offline). An online experiment was conducted. Results show that there for smart advertisements the advertisement attitude is generally higher than for generic advertisements. This effect is fully mediated through perceived relevance. The level of intrusiveness is for smart advertisements also generally higher than for generic ones. There are no significant results found for privacy concerns and the media environment has no effect on advertising attitude. However, to keep intrusiveness and privacy concerns low smart advertisements are better to show in an online environment and generic advertisements work better in an offline environment. This research contributes to existing research in multiple ways: 1) it closes the existing literature gap about smart advertisements and divers consumer responses, 2) they give marketers insights in how consumers perceive smart advertisements and 3) it sets the future research agenda.
Key words: Advertising personalization, smart advertisement, privacy concerns, level of intrusiveness, advertising attitude, The Internet of Things.
Table of contents
1. Introduction ... 6
2. Literature review ... 8
2.1 The Internet of Things (IoT) ... 8
2.2. Smart cities ... 10
2.3 Smart advertising ... 11
2.4. Advertisements in the online environment ... 12
2.5 Literature gap ... 13
3. Conceptual framework and hypotheses ... 15
3.1 Advertisement attitude ... 15
3.1.1 Different kinds of advertisements ... 16
3.1.2 Perceived relevance ... 17
3.1.3 Advertising environment and attitude ... 19
3.2 Privacy Concerns ... 20
3.3. Level of intrusiveness ... 22
3.4 Advertisements and different media environments ... 23
3.5 Conceptual models ... 24 4. Research method ... 26 4.1 Sample ... 26 4.2 Design ... 26 4.3 Procedure ... 27 4.4 Stimulus materials ... 28 4.5 Measures ... 29 4.6 Randomisation check ... 34 5. Analysis and results ... 35
5.1 Testing the hypotheses ... 35
5.1.1 Hypothesis 1 and 2 ... 35
5.1.2 Hypothesis 3 ... 37
5.1.4 Hypothesis 5 ... 38
5.1.5 Hypothesis 6 ... 38
5.1.6 Hypothesis 7 ... 39
6. Discussion and conclusions ... 41
6.1 General discussion ... 41 6.1.1 Attitude ... 41 6.1.2 Privacy concerns ... 42 6.1.3 Level of intrusiveness ... 43 6.2 Theoretical implications ... 44 6.3 Managerial implications ... 45
6.4 Limitations and future research suggestions ... 46
6.4.1 Other future research suggestions ... 48
6.5 Conclusions ... 49 7. References ... 51 8. Appendices ... 59 8.1 Appendix 1 ... 59 8.2 Appendix 2 ... 61 8.3 Appendix 3 ... 70 8.4 Appendix 4 ... 71
1. Introduction
One of the buzzwords of the past years is definitely the Internet of Things (IoT) (Groopman, 2015). However, even an interested reader might experience some kind of difficulty in understanding what IoT really means, which basic ideas stand behind this concept, which social, economical and technological implications the full deployment of IoT will have (Atzori, Iera & Morabito, 2010). They are not alone. A recent study found that 87% of consumers were unfamiliar with the term Internet of Things (Acquity group, 2014). The authors define it as follows: “Internet of Things” means “a world-wide network of
interconnected objects uniquely addressable, based on standard communication protocols”. How did this start? Groopman’s (2015) study takes us through the tremendous changes in the marketing world of the past years. The pre-digital technology phase was one where media relied on a one-way, one-to-many communications model, which they call a broadcast model. After the internet came around, social media and mobile brought the impact of dialog: two-way communications between brand and consumer and consumer to consumer. Now with the rise of the IoT there are opportunities for any element of the brand experience to have a voice. There are sensors added to beings, places, objects and environments, as a result we grant these things a voice through the data they generate. IoT enables multi-way communication between brand and consumer, brand and object, consumer and object and object to object. This is called the connectivity model, where multi-way communications are taking place between object and brand, object and consumer, and object and object. This is all happening real time.
The IoT is a phenomenon where the digital world is converging with the physical world. This represents the next era of computing. It is an era in which just about anything can be connected, through sensors and open data, to other objects, environments, people and of course, the internet (Groopman, 2015). He predicted there will be nearly 20 billion devices on
the IoT and that the number of smartphones, tablets and PC’s in use will reach about 7.3 billion units by 2020. Considering that it is 2018 already, that movement is coming faster than we can imagine. Those devices and interfaces represent windows into a new world of
capabilities that can be mutually beneficial, empowering and contextually opportune for brand and consumer simultaneously (Groopman, 2015).
On the one hand this IoT era is going to benefit consumers greatly. ‘Smart’ products like surveillance apps to lock your front door or turn your heating on before you get home are time efficient and make their lives easier. On the other hand also companies benefit greatly, because they have more and more connected consumer data (Finn & Wadhwa, 2014). Companies want to identify, categorise and profile consumers as much as possible to make their marketing strategies more effective and to offer products that are most relevant for their consumers’ interests. With the fast development of IoT, not only smart products are rapidly produced. Also cities are becoming smarter (e.g. Martínez-Ballesté, Pérez-Martínez & Solanas, 2013; Li, Nucciarelli, Roden & Graham, 2016). With cities becoming smarter the first personalized advertisements 2.0 are already live. Before we know it, digital billboards will pick up data from our devices to switch advertisements depending on who is walking by (eMarketer, 2017).
There are only a few the marketing studies, which explore IoT (Nuygen & Simkin, 2017). Therefore research is needed in order to investigate consumer responses towards these new advertising possibilities. Motivated by the current technological developments that make it possible to go to the next step with personalized advertising, together with the gap that is obviously present, this study investigates consumers’ responses towards personalized advertising. This study’s findings shed light on how consumers’ perceive those new advertisements. Shortly said, the online experiment points out that IoT advertisements are received be more positive, because they are more relevant. There are no differences in privacy
concerns. There is however a difference between the level of intrusiveness for the IoT advertisement and a generic one. This research also highlights which media environment works for the kind of advertisement.
2. Literature review 2.1 Internet of Things (IoT)
A revolution is happening via the Internet of things (IoT) (Nuygen & Simkin, 2017). The IoT involves the interconnected devices, systems and services that rely on the
autonomous communication of physical objects within the existing Internet infrastructure (Atzori, Iera & Morabito (2010). Nuygen and Simkin (2017) state that this development brings internet intelligence to physical products and with that making all products more connected and “smart”. The IoT span numerous areas: wearables, smart homes, smart cities, industrial automation and many more (Chuah et al., 2016). Unlike traditional Internet that connects people in the exchange of information, the IoT integrates machines and objects with embedded sensors and allows them to communicate autonomously over the Internet (Hsu & Lin, 2016) With the rising popularity of the IoT, more and more customers are enjoying personalised, autonomous and optimised services provided by smart and connected objects (Wu, Chen & Dou, 2016).
A lot of previous marketing research in an IoT context is about smart products. For example Wu, Chen and Dou (2016) researched the effectiveness of the interaction style of a smart device. They investigated whether the style of smart interaction improves consumer’s perceptions of brand warmth and competence increases their emotional attachment to the focal brand. The results show that a friend-like style of smart interaction leads to more positive brand warmth and brand attachment in comparison with the engineer-like style. Also
brand competence is found to be a more important variable than brand warmth as a mediator of the relationship between interaction style and brand attachment. On top of that, the authors found that the friend-like style of smart interaction is superior to the engineer-like style in constructing valuable brand equity. Mani and Chouck (2017) looked at resistance to smart products. Their findings show that perceived usefulness, perceived novelty, perceived price, intrusiveness, privacy concerns and self-efficacy have an impact on consumer resistance to smart products. Hsu & Lin (2016) also look at IoT services. They define them as “services that are identified as application layer functions which allows users to interact directly with hardware (i.e. smartphone), self-aware things and smart environments to improve life quality and work productivity. Examples are: The nest thermostat which allows users to easily control the smart home environment, video surveillance cameras, motion- detection security systems, door locks, garage door openers etc. BMW ConnectedDrive (driver assistance) is another example, which offers services such as emergency calls, real time traffic information and concierge service through the driver’s smartphone.
The IoT does not only develop products to help consumers. It also creates lots of
opportunities for brands (Groopman, 2015). For example by leveraging sensors and customer insights, IoT brings brands closer than ever to the ultimate marketing objective: delivering the right content or experience in the right context. In other words, delivering the right
information or service to the right person, at the right place and time via the right platform (Groopman, 2015). Through this data collection brands are able to listen better, to observe, use data generated insights to inform how they engage with customers and evolve their products and services.
2.2 Smart cities
Nuygen and Simkin (2017) state that many of the rules of marketing are changing and many new approaches will be introduced in this new IoT era. Not only smart products like Google home, apple’s homePod and smart watches are being developed quickly. Entire cities are becoming smart. Due to factors related to economies of scale, many services are more easily provided in highly populated areas. People are moving from the country to the cities and an urbanization trend is happening (Martínez-Ballesté, Pérez-Martínez & Solanas, 2013). As a result the authors state that cities realize they need to use technology in order to improve their living experiences. The effective use of IoT data will make cities ‘smart’ and further enrich the notion of smart cities in the coming years (Li, Nucciarelli, Roden & Graham, 2016). For much of the 20th century, the idea of a smart city was a science fiction that was pictured in the popular media (Batty et al. 2012). Cities are becoming smart not only in ways we can
automate routine functions serving individual persons, buildings and traffic systems (Batty et al. 2012). The authors state that they are becoming smart in ways that enable us to monitor, understand, analyse and plan the city to improve efficiency, equity and quality of life for its citizens.
There are major shifts happening with the rise of smart cities (Batty et al. 2012). The authors say it starts with the development of information infrastructure that underpins the city through distributed computing and networks that are available to everyone with devices that can access such infrastructure. The access of that information of course depends on the security. The fact is that such infrastructure is going to be available in those cities. Routine data sensed in real time is yielding big data (Batty et al. 2012). This is going to lead to a new development of human behaviours with a focus on location and mobility.
2.3 Smart advertising
What do all of these new developments mean for the advertising industry? The key in marketing is to provide the right content, in the right format, to the right person, at the right time (Groopman, 2015). From outdoor, radio, print and TV to the Internet, advertisements have become more interactive, more relevant and more contextual in their design to attract consumers and to raise attention of the audience (Bauer & Strauss, 2015). With the
development of IoT and high presence of mobile devices, bringing advertisements to these devices is the next wave of advertising. Consumers may be addressed individually, wherever they are (Bauer & Strauss, 2015). Smart cities provide the ideal environment for a range of new innovations at the interface between online and offline activities (Li et al. 2015). For example a retailer can link their online channel to the store channel by using the mobile app in combination with Beacons. When the customers that downloaded the app walk past the shop, they receive offers through the app related to their online shopping behaviour (Holdowsky, Mahto, Raynor & Cotteleer, 2015). In the store the app provides in-store navigation to the exact location where the items are stored. There have also been examples of digital out of home that becomes ‘smarter’. For example Unilever in Turkey used temperature sensors implanted in digital billboards to dynamically serve ads for their soup or ice cream brands (Holdowsky et al., 2015). The online behavioural data, combined with the data that smart cities will have, will change the marketing world like we have not seen before (Holdowsky et al., 2015) Another form of advertising that is made possible by the IoT development is location-based advertising (LBA). This advertising method adds important opportunities for businesses: it targets consumers individually, based on their current location and in real-time (Bauer & Strauss, 2015).
2.4 Advertisements in the online environment
There has been a lot of research focused on the data usage by companies that gathered this data from online environments. Companies use all possible channels to develop personal information databases about consumers. These databases make it possible to create
personalized advertisements with customized messages for each individual consumer (Kim et al., 2001).
For example, sending local bridal shop ads to women whose relationship status is “engaged” on Facebook. Or sending a completely tailored advertisement to a particular individual based on his or her name, previous searches, web page visits, viewed content, or friends with connections to specific pages, groups, or application (Arora et al. 2008; Hoy & Milne 2010). This is not new anymore. In order to break out of cluttered ad environment, advertisers provide customized ad messages for an individual consumer based on personal information (Jung, 2017). De Keyzer, Dens and de Pelsmacker (2015) did a study on the impact of perceived personalization on consumer responses to advertising on Facebook. In line with previous research (Arora et al 2008; Tam & Ho, 2005) personalization has a positive effect on consumer responses. The results of the mediation analysis: the effect of
personalization on consumer responses occurs mostly as indirect only through perceived relevance. Another example is Bleier and Eisenbeis (2015) who researched the effectiveness in display banners by personalizing ads based on individual consumers’ recent online
shopping behaviours using a method called retargeting. They looked at the effectiveness of ad personalization through retargeting by taking into account its interplay with relevant timing and placement factors. Their results show that personalization strongly increases click through rates and that banners of relatively higher personalization intensity achieve the highest CTRs. They also found that over-personalization is possible. If that is the case the ads that fail to
meet initially revealed preferences since time has passed since the last store visit. So the effectiveness of the ad also depends on a consumer’s current online browsing mode.
All the examples above are based on either the online environment or traditional media such as telephone and letters in the mail. However, IoT makes it possible to personalize the ads in the offline world. IoT technologies give opportunities to collect consumer data in limitless amount and quality. This can be age, gender, interests, social status, income, tastes, purchase and other types of habits, location and many more. Since interaction between a consumer, IoT devices and physical world can be rather intense, marketers are able to track even the most detailed traits of consumer behaviour (Abashidze & Dabrowski, 2016).
Because of these IoT developments, ad personalization can also happen at media that are not known for collecting data. For example: digital billboards, sensors in physical shops that pick up information from your device and who knows what is going to be possible in the offline world next (Groopman, 2015).
2.5 Literature gap
As stated before only a few marketing studies which explore IoT have been conducted (Nuygen &Simkin 2017). Research that did look into this field mostly looked at consumers’ adoption of smart services (Hsu and Lin, 2016; Chuah et al 2016), at interaction style with a smart device (Wu, Chen & Dou 2017), to smart cities that focused on urban planning (Li et al, 2015) and not about marketing perspectives. With the developments in the IoT world
advertisements are becoming more personalized in a whole new technological way. With the world around us becoming ‘smarter’, so are advertisements. There are limited numbers of studies exploring smart advertisements. The ones that do research this topic look at the ethics and regulatory initiatives (Finn & Wadhwa, 2014). The literature review of Bauer and Strauss (2016) shows that of the 32 publications they looked at, only two publications also included
user perspective. You can conclude from that, that smart advertising research is not focused on user-centric research. According to the authors it is necessary to investigate user
acceptance from a non-technical point of view as well. Abashidze and Dabrowski (2016) add that the more comprehensive the scientific research the more opportunities for the elimination of risks and threats.
To conclude, there has been a lot of research on the so-called personalized advertising in the online world (e.g. Yu & Cude, 2009; De Keyzer, 2015; Bleier & Eisenbeis 2015), but to my knowledge there has not been a research that looked at the new smart advertisements. Most of the research is about some kind of personalization of the ad based on online
generated data. Little is known about the kinds of advertisements that the Internet of Things makes possible. Despite the increase in the amount of personalized advertising as well as the development of diverse new technologies that can be used to deliver it, few academic
researchers have examined consumer responses to it (Yu & Cude, 2009). Van Doorn and Hoekstra (2013) stated already in their research that technology is increasingly facilitating sophisticated ways to customize ads, enhancing knowledge in this field is essential.
Considering it is 2018 now, this technology has increased only further and therefore research is needed. Future studies about personalized advertising should include new and popular styles of personalized advertising (Yu & Cude, 2009).
The current technological developments make it possible to go to the next step with personalized advertising, together with the gap that is obviously present, this study
investigated consumers’ responses towards personalized advertising. The focus is on attitude towards the advertisement (ad), privacy concerns and level of intrusiveness. More
specifically, this study examined whether consumer’s perceptions of smart advertising differ depending on the type of media used to deliver it: Online (through banner ads), offline
(through digital billboards) based on the research by Yude and Cude (2009) that took that into account too. The research question of this study therefore is:
3. Conceptual framework and hypotheses 3.1 Advertisement attitude
To start of, an explanation about how attitudes in persuasion are formed in the human mind. Cacioppo and Petty (1981) build on their Elaboration Likelihood Model (ELM) to explain those attitude formations. They suggest that there are ‘central’ and ‘peripheral’ routes to persuasion. The central route represents the process involved when the elaboration likelihood is high and the peripheral route occurs when elaboration likelihood is low. This central route encompasses a systematic view of persuasion while the peripheral route embraces the heuristic view of persuasion. These two routes lead both to attitude change (Petty & Cacioppo, 1981). To conclude, the central route views attitude change as resulting from a person’s diligent consideration of information that an individual feels is central to the true merits of a particular attitudinal position (Petty, Cacioppo & Schumann, 1983). The authors state that the peripheral route however says that attitude changes do not occur because an individual has personally considered the pros and cons of the issue, but because the attitude issue or object is associated with positive or negative cues.
‘What is the effect of ‘smart advertising’ (vs. generic) on the advertising attitude, the privacy concerns, level of intrusiveness and differs this for media environments
Advertising attitude is defined as: “Predisposition to respond in a favourable or unfavourable manner to a particular advertising stimulus during a particular exposure occasion” (Lutz, 1985). Advertising attitude because it is an antecedent of brand attitudes (MacKenzie, Lutz & Belch, 1986). Those authors also state that the idea that consumers may have affective reactions to commercial stimuli is hardly new. Mehta (2000) adds that
consumers’ attitudes toward advertising are important indicators of advertising effectiveness. She states that people with a more favourable attitude toward advertising recalled a higher number of advertisements the day after exposure and were more persuaded by them. Attitude towards the advertisements is therefore an important variable to look at.
3.1.1 Different kinds of advertisements
In the case of this research it is important to look at the different kinds of advertisements there nowadays are. There are generic advertisements that are the same for every one (Nuygen & Simkin, 2017). For example a billboard down the road. Smart advertising however is
advertising that uses data from the participant receiving the ad, made possible by the Internet of Things (Nuygen & Simkin, 2017). Marketers try to segment audiences into tailored clusters utilizing big data in the form of demographics, geographic location and previous online behaviours to offer a more personalized way of advertising (Brinson & Eastin, 2016). Previous research shows that there is a more positive effect of personalized advertising on brand attitude, click intention and advertising attitude (De Keyzer et al., 2015; Arora et al., 2008; Tam & Ho, 2005). Generally speaking it is the case that the more the advertising generated self-related thoughts, the more positive responses towards advertising are produced in terms of ad message recall, attitude toward the ad/brand, click intention and purchase intention (e.g. Ahn & Bailenson, 2011; De Keyzer et al., 2015). Two meta analyses (Noar, Benac & Harris, 2007; Sohl & Moyer, 2007) conclude that personalized messages are
generally more effective than non personalized messages in terms of being more memorable, more likeable and sparking behavioural change. Petty Barden and Wheeler (2002) build on the ELM to explain how tailoring or personalization can change attitudes and subsequent behavior. In the logic of ELM, personalization should benefit attitudes and behaviour both under high and low elaboration. Under conditions of high elaboration personalized arguments could be perceived as stronger than non-personalized arguments. At the same time, under conditions of low elaboration, perceived personalization can serve as a heuristic cue that leads to a positive attitude change. This is in line with self-referencing: the extent to which a
consumer relates information to himself or herself. Self-referencing can have a positive effect on attitudes under both central processing and peripheral processing. Due to self-referencing, readers could also be more motivated to process personalized messages (and thus follow a more central route) when a message is perceived as more personally relevant, for example because it is personalized, it leads not only to greater attention but also to greater elaboration, message processing and ultimately persuasion (Bright & Daugherty, 2012; Noar, Harrington & Aldrich, 2009; Rimer & Kreuter, 2006). Since smart advertisements also use personal data the following hypothesis is formed:
H1. Smart advertising (vs. generic advertising) will have a positive effect on the attitude towards the ad.
3.1.2 Perceived relevance
Relevance is the degree to which consumers perceive an object to be self-related (Celsi & Olson, 1988). The authors defined two types of sources from which people feel personal relevance: Situational or intrinsic sources. They state that situational sources come from the physical and sociological immediate environment and intrinsic sources are based on personal
experience and knowledge. To conclude: regardless of which types of sources, when sources are related to personal needs and values people perceive relevance. According to the
Elaboration Likelihood Model, personal relevance the only determinant of the route to
persuasion (Petty, Cacioppo & Schumann, 1983). They state that personal relevance increases a person’s motivation for engaging in an active mind-set regarding the issue or product relevant information presented in order to form a correct opinion.
In the advertising context Zhu and Chang (2016) have defined relevance as: “the degree to which consumers perceive a personalized advertisement to be self-related or in some way instrumental in achieving their personal goals and values”. In other words:
Perceived ad relevance means that advertising products or the situations in which a product is located are related to personal needs and values (Jung, 2017). Pechmann and Steward (1990) elaborate on the fact that higher relevance attracts more attention. Next to that, people are also more likely to show a positive attitude toward advertising when it includes personally relevant products compared to less relevant products (Trampe et al., 2010). The relationship between ad relevance and positive consumer responses can be explained by self-referencing theory (Roger, Kuiper & Kirker, 1977). This theory explains that self-referencing refers to a
cognitive process in which people are more likely to be persuaded by self-relevant messages. In advertising context, self-referencing has been applied to explain how ad contents affect consumer persuasion. According to existing literature (Petty, Barden & Wheeler, 2002; De Keyzer et al., 2015) personalized advertising seems to increase message processing by enhancing the perceived relevance of the message. This result signals that personalized advertising can be effective, but only if the ads are perceived as more relevant (Kramer, 2007). Other research has showed that an advertisement’s relevance influences consumers’ reactions, including paying closer attention towards the ad (Pechmann & Stewart, 1990),
displaying higher purchase intensions (Pavlou & Stewart, 2000) and providing better advertising effectiveness (Drossos & Giaglis, 2005).
As said before, previous research showed positive effects of personalized
advertisements will result in positive consumer responses such as ad message recall, attitude toward the ad, brand attitude, click intention and purchase intention (e.g. Ahn & Bailenson, 2011; De Keyzer et al., 2015). De Keyzer et al. (2015) also found that there is a positive effect of personalized advertising on brand attitude and click intention. These positive effects of perceived personalization were fully attributed to the mediating role of personal relevance. They state that if a personalized message is received as personally relevant, the responses to the message are more positive.
Therefore the following hypothesis is posed:
H2. Consumers that see a smart advertisement (vs. generic ad) will have a more positive advertising attitude. This effect is mediated by perceived relevance.
3.1.3 Advertising environment and attitude
There are a lot of studies that have examined consumer’s attitudes toward advertising in the traditional media environment (e.g. Gong & Madox, 2003). Besides, a lot of research investigated the impact of consumers’ general beliefs and attitudes toward advertising
effectiveness (e.g. MacKenzie, Lutz & Belch, 1986; Muehling, 1987). An important outcome was that consumer behaviour such as advertisement avoidance might be a result of
consumers’ general negative attitudes toward advertising (e.g. Li et al., 2002). With the rise of the Internet many companies turned to web advertising, because the web is a distinctive advertising medium with characteristics such as constant message delivery (Wolin,
Korgaonkar & Lund, 2002). At the same time a new branch of research looked into
advertising attitudes in this online world. For example Ducoffe (1996) showed that consumers saw online advertising as valuable, informative, not very entertaining and not particularly irritating. Other studies however showed way more negative results and showed that online advertising was viewed as intrusive (Li et al., 2002). According to Wolin, Korgaonkar and Lund (2002) web users display less positive attitudes towards web advertising. To my
knowledge there are limited studies that looked at both the online environment and the offline environment in combination with advertisement attitude, creating a gap in scientific literature. Therefore it is important to look at. The following hypothesis is posed:
H3. Online advertising (vs. offline advertising) will have a more negative effect on the attitude towards the ad.
3.2 Privacy concerns
While the new developments in the Internet of Things era seem to make it possible to take this personalization level even further, this also means that consumers who are concerned about their privacy might reject this. The personalisation trend as Turban, Lee, Liang, King and Chung (2015) describe also come with concerns about invasion of individual privacy. This could be been seen as a paradox for advertisers (Zhu & Chang, 2016). They state that on the one hand personalization is associated with higher customer loyalty and satisfaction as well as higher influence and conversion. On the other hand, the personalized and targeted advertising is closely connected to privacy concerns. This is a popular research topic, so there are multiple articles that agree with Zhu and Chang (2016). Jung (2017) points out that the ad personalization technique is a successful way for advertisers to increase advertising
positive effect on ad effectiveness (e.g. Pechmann & Stewart, 1990; De Keyzer et al., 2015). On the other hand multiple articles also showed that people perceive this as an invasion of their privacy (e.g. Zhu & Chang, 2016; Bleier & Eisenbeiss, 2015; Tucker, 2014; Van Doorn & Hoekstra, 2013)
Privacy invasion perceptions and self-awareness affect consumer’s continuous use intentions. Prior research (Baek & Morimoto, 2012; Sheehan & Hoy, 1999) shows that privacy invasion perception is negatively related to continuous use intentions; self-awareness, on the contrary, contributes to continuous use intentions. The research outcomes of Zhu and Chang (2016) provide evidence that providing highly relevant personalized advertisements are conducive in reducing user’s privacy concerns. Perceived relevance is significantly negatively associated with user’s perceptions of privacy invasion. However, Jung (2017) found that when people are exposed to highly personalized ads on social media, they think that marketers track their information and use it for marketing purposes. In turn, consumers have increased privacy concerns. Bleier and Eisenbeiss (2015) add that permission for using consumers’ data is often not asked and therefore the consumers are not aware of firms’ undertakings until they receive an individualized communication. They state that ad personalization causes privacy concerns, primarily because it confronts consumers with their vulnerability to the data collecting of advertising retailer. Together with the new privacy law that has been live since 28th of May in the EU, it is a also a really actual and relevant topic nowadays.
Based on the theory above the following hypothesis is posed:
H4. Smart advertising (vs. generic advertisements) will have a positive effect on privacy concerns.
3.3 Level of intrusiveness
Lavy et al. (2009) define intrusiveness as: “Creating an imbalance between closeness and autonomy”. Closeness in this case refers to the degree of relatedness and interdependence between two parties. Autonomy is the degree to which personal identity can be perceived (Bleier & Eisenbeiss, 2015). In the advertising context however intrusiveness can be seen as a psychological reaction to ads that interfere with a consumer’s on-going cognitive processing (Li et al., 2002). The authors therefore state that ads within programming or editorial units are not themselves intrusive, but rather the ads must be perceived as interrupting the goals of the viewers to be considered intrusive. Although consumers generally perceive personalized messages as more relevant and informative than non-personalized communication, they also view them as more intrusive (Bleier & Eisenbeiss, 2015; Tucker, 2011). Li et al. (2002) also state that although the fit of the customized ad may increase the ad’s relevance and thus result in positive behavioural effects, the use of more personal information may induce feelings of intrusiveness that interfere with the consumer’s cognitive processing and interrupt goal pursuit. The consequence of this is that it even prevents the consumer from taking notice to the ad contents (Morimoto & Chang, 2006). Intrusive ads may also be perceived as annoying and result in reactance in such a way that consumers behave in the opposite way of what is intended by the advertiser (Clee & Wicklund, 1980). Van Doorn and Hoekstra (2013) found that higher degrees of personalization, such as adding personal identification or transaction information to an ad, beyond just using browsing data, increase feelings of intrusiveness and negatively affect purchase intentions. They also found that high fit did lead to higher purchase intention, but also to higher levels of perceived intrusiveness. So trying to serve consumers better by providing them messages that are relevant to them can backfire for the advertiser. On top of that, Bleier and Eisenbeiss (2015) state that even though personalized ads attract attention more easily, they are likely to be perceived as more intrusive.
Therefore the following hypothesis is posed:
H5. Smart advertising (vs. generic advertising) will have a negative effect on the level of intrusiveness.
3.4 Advertisements and different media environments
Prior research also investigated different media environments when consumers were exposed to advertisements. For example, Chaudhuri and Buck’s (1995) study was about the
relationship of different media channels to psychological outcomes. They indicated that media differences are one of the key predictors of consumers’ emotional and rational responses to advertising. Other research looked at the relationship between media type and consumer responses as well (e.g. Cutler & Thomas, 2000; Chachko, 2004; Yu & Cude, 2009). The most recent research is from Yu and Cude (2009), they looked at consumers’ perceptions of personalized advertising delivered online (e-mail) and offline (letter and telephone call). Nowadays personalized advertising can be delivered by using more advanced techniques than traditional e-mails, including personalized web pages that use cookies to capture an
individual’s history of web surfing, personalized interactive television advertising, smart banners and mobile advertising (Yuan & Tsao, 2003; Pramataris et al., 2001). Yu and Cude (2009) wondered if there was a possible difference between the media used to deliver personalized advertising. They found that consumers’ general perceptions of personalized advertising were negative. However, they were less likely to take personalized advertising seriously if it was delivered online, rather than offline. Participants were less likely to reject the letter (mail) immediately, more likely to take it seriously, less threatened by the
personalized advertisements as a violation of their privacy and somewhat more likely to view it as personal attention. However, the research of Yu and Cude (2009) was based on direct
mail, direct e-mail or direct phone calls. The IoT makes it possible to reach people with personalized advertising while they are walking passed a billboard (eMarketer, 2017). A billboard is a static piece of marketing what the crowd of people can see (Hendricks, 2017). With the IoT, targeted Digital Out of Home is possible. This digital billboard is always adapting to the people that walk by (Hendricks, 2017). The expectation is that people are less used to personalized advertisements in the offline world than in the online world. Therefore there might be a difference in whether smart advertisements are delivered by online or offline media and their effects on privacy concerns and level of intrusiveness. Therefore the
following two hypotheses are formed:
H6. Smart advertising (vs. generic) online results in less privacy concerns than smart advertising (vs. generic) offline.
H7. Smart advertising (vs. generic) online environment is less intrusive than smart advertising (vs. generic) in the offline environment
3.5 Conceptual models
Figure 1: Conceptual model 1.1
Figure 2: Conceptual model 1.2
* = In order to show the hypotheses in a conceptual model there has been chosen to split it in two visuals. Otherwise the conceptual model was not clear.
4. Research method
4.1 Sample
An online experiment was conducted which was spread on MTurk and through social media channels (Facebook and Whatsapp) with a qualtrics link. It was live for a week (7th until 14th of May). The sample was a so-called convenience sample. In total 213 participants started the experiment, all of them finished. The participants from MTurk (n=200) received $0.10 to take part in the online experiment. The other participants (n=13) got no financial reward. The participants have different nationalities, but most participants are from the Americas (61%). 57.3% of the participants was male (SD=.50) and 50,2% is in the age group of 25-34
(M=3.55, SD=1.32). Participants were randomly assigned to one of the four conditions.
4.2 Design
According to Hammarberg, Kirkman and de Lacey (2016) quantitative research methods are appropriate when general or probability information is desired on opinions, attitudes, views, beliefs or preferences. They state that it is important that variables can be linked to form hypotheses before data collection and that the research question is known, clear and
unambiguous. Considering there are multiple quantitative research methods, an experiment was in this case the most appropriate. To quote Vargas, Duff and Faber (2017): “The best tool researchers have for determining relationships is experimental research”. They state that an experiment involves a researcher who is manipulating and controlling one or more potentially causal variables (independent variables) and then observing the corresponding differences in the results. Therefore, experiments are typically the preferred research method for
demonstrating causation. Gravetter and Forzano (2012) agree with that. An online experiment is chosen to answer the research question because limited amount of time and money was
available. The experiment was a 1 factorial between subjects design with 2 levels (smart vs. generic advertising) x 2 (offline vs. online environment). The online experiment was written in English and was conducted through the online tool Qualtrics.
4.3 Procedure
The online experiment starts off with a short introduction. In this introduction the participants were informed about the voluntary participation and that they were free to withdraw at any point throughout the duration of the experiment. Thereafter, the goal of the experiment was explained, timing was stated and they were ensured that everything was confidential. After that they had to check a box that they understood their rights and agreed to participate (See appendix 1 for the full text).
The first page of the experiment kicked off with the demographic questions. Because this research made the ads ‘smart’ based on gender, this was an important question to fill in. Also, it is good to know some basic demographics about your sample. Therefore, there were forced responses added in Qualtrics for every question in this part of the experiment.
There were four conditions in total. The only thing that was different was the visual and the scenario text the participants saw after the demographic questions. In chapter 4.4 the stimulus materials will be discussed in detail. For every condition there was a short
introduction text about what would be seen next. This was a different text for the smart conditions and the generic conditions (See appendix 1). Participants were randomly assigned to one of the four conditions. The two smart conditions were based on gender, so if you were assigned to one of those conditions the answer to the question “what is your gender” decided if you got to see a visual with a male or a visual with a woman.
Then the visual and the scenario sketch were shown. Because it is important that the participants actually take time to look at the visual and the text, a minimum of 10 seconds was added before participants could go to the next questions.
After being exposed to one of the four conditions there were questions about the dependent variables – advertising atittude, privacy concerns and level of intrusiveness. There were also questions about the perceived relevance. These questions were the same for every condition (See appendix 2 for the questionnaire)
4.4 Stimulus materials
As mentioned before, in order to make the advertisements ‘smart’ this research had to
manipulate based on some kind of characteristics of the participants. Previous research shows that when people have identical ethnicity and gender they are more likely to think that the products are self-related in advertising models. This affects favourable responses to the ad, product and models (Lee, Fernandez & Martin, 2002). Keyzer et al. (2015) personalized the ads based on gender. Dijkstra (2005) and Webb, Hendricks and Brandon (2007) also had a minimal degree of personalization in their stimulus materials. They state that the mere prime of personalization is still sufficient. De Keyzer et al (2015) used a perfume brand, because they thought it was a popular affordable product for the target group. They used a fictitious brand to avoid potential confounds of prior brand attitudes. Depending on the condition, the advertisement was either personalised (“For men with confidence” or “for women with confidence”) depending on the gender of the respondent or generic (not personalized; “confidence”). In this study there will be a personalization based on gender as well. The fictitious clothing brand R&R was used with different visuals: ads for male/female or a generic ad with both a male and female (See appendix for examples). Clothing was chosen because similar to the perfume example, it is an affordable product category that everyone has
to buy. Therefore, it is relevant and accessible for everyone. For the four different visuals see appendix 1.
In order to explain the participants how the advertisements knew certain information about them to make the advertisements smart, a scenario was sketched for every condition. This was based on Bleier and Eisenbeiss’ (2015) study. They used a scenario sketch in order to explain whether the retailer was trusted or not. Van Doorn and Hoekstra’s (2013) research also used scenarios. They asked their participants to imagine themselves as a certain person, who was a customer of a fictitious bank, used internet banking and has just gotten a raise in pay. The text explains a couple of other things about the person’s browsing behaviour. Therefore, this study used three different scenario sketches to make the advertisements smart or generic (see appendix 1).
4.5 Measures
In total there were four dependent variables (Intrusiveness, privacy concern, attitude towards the ad and relevance). For all four variables there has been a check on counter indicative items first. After that, reliability and validity are checked. A principal axis factoring analysis (PAF) checks the validity of the variables. The results of those analyses are shown in the table below. The four variables will be described first (also see correlation table in table 1)
Level of intrusiveness
This variable is measured based on an existing scale used in Edwards et al (2002) and
Mooradian (1996). The scale has ten items (e.g. “I think this offer is disturbing”, “I think this offer is alarming”) that were answered on a seven-point Likert scale ranging from 1 (“strongly disagree”) to 7 (“strongly agree”). See table 2 for a description of the then items.
There were no counter-indicative items in this scale. The level of intrusiveness scale has a high reliability, with a Cronbach’s Alpha of .97. The corrected item-total correlations indicate that all the items have a good correlation with the total score of the scale (all above .69). None of the items would substantially improve the reliability score if they were deleted.
Intrusiveness has a mean of 1.43 and a SD of .50.
Privacy concern
This variable is measured based on an existing scale used in Dinev and Hart (2004). The scale has four items (e.g. “It bothers me that the firm is able to track information about me”) that were answered on a seven-point Likert scale ranging from 1 (“strongly disagree) to 7 (“strongly agree”). For all four items see table 2. The four items ended up being recoded because the descriptive analysis pointed out that the items were negatively keyed items. The privacy concern scale has a high reliability, with a Cronbach’s Alpha of .95. The corrected item-total correlations indicate that all the items have a good correlation with the total score of the scale (all above .83). None of the items would improve the reliability score if they were deleted. Privacy concern has a mean of 4.56 and a SD of 1.69.
Advertising attitude
This variable is measured based on an existing scale used in Mitchell and Olson (1981). The scale has four items (good-bad, like-dislike, interesting-uninteresting) measured on a five-point bipolar scale. At first there was a Cronbach’s Alpha of only .50. Also, a corrected item-total correlation of Q7_3 indicated a negative loading. After this item was deleted the
Cronbach’s Alpha rose to .89. Therefore Q7_3 (“irritating-not irritating”) was removed from the scale. Q7_3 was recoded, but the loading stayed negative and the Cronbach’s alpha remained low. A possible reason for this is that the participants did not read the question
correctly. All the other items in Q7 where formulated on a positive to negative scale (E.g. good-bad, like-dislike, interesting-uninteresting). Q7_3 was formulated in the opposite way (irritating-not irritating) so there could be a chance that the participants did not notice that this question was formulated differently from the others in Q7. After Q7_3 was deleted the
corrected item-total correlations indicate that all the items have a good correlation with the total score of the scale (all above .74). Attitude towards the ad has a mean of 2.28 and a SD of 1.18.
Relevance
This variable is measured with two items (“The information in this advertisement was relevant, the information in the advertisement was useful”) based on an existing scale of Ahluwalia, Unnava and Burnkrant (2001). Perceived relevance is measured on a seven-point Likert scale ranging from 1 (“strongly disagree”) to 7 (“strongly disagree”). The two items were recoded first, because they were negatively keyed. The perceived relevance scale has a strong reliability with a Cronbach’s Alpha of .91. The corrected item-total correlations indicate that all the items have a good correlation with the total score of the scale (all .83). Relevance has a mean of 4.16 and a SD of 1.72.
Factor analysis
A principal axis factoring analysis was conducted on the scales. This analysis has been done for all the variables at the same time so that the analysis not only looks at validity within the construct, but also at the validity between the different constructs. The Kaiser-Meyer-Olkin measure verified the sampling adequacy for the analysis, KMO=.90. Bartlett’s test of
sphericity χ2(171) = 4345.21, p<.001. This indicates that the correlations between items were sufficiently large enough to do a principal axis factoring analysis. The analysis was initially run to obtain eigenvalues for each component in the data. Three components had eigenvalues Kaiser’s criterion of 1 and in combination explained 78.32% of the variance. After examining the screeplot the conclusion was that there were three different constructs. But there was one construct that had an eigenvalue of .91 with high loadings (.92 and .86). There has been decided to also construct this variable because of the high eigenvalue and the high loadings within the construct. To conclude, there were four factors retained and rotated with an Oblimin with Kaiser normalization rotation. Table 2 shows the factor loadings after rotation. All loadings between the other constructs were < .35. This indicates that there were no problematic items with high loadings (> .50) in more than one factor. Therefore those loadings are not showed in the table on the next page.
Table 1: Means, Standard Deviations, correlations and Cronbach’s alpha
Variable M SD 1 2 3 4 5 6 7 8 9 1 Gender 1.43 .50 - 2 Age 3.55 1.32 .15* - 3 Education 4.26 1.30 -.09 -.02 - 4 Attitude 2.28 1.18 .06 .01 -.02 (89) 5 Intrusiveness 3.59 1.70 -.02 .02 -.05 .32** (97) 6 Relevance 4.16 1.72 -.12 -.01 .11 -.55** -.07 (91) 7 Privacy concern 4.56 1.69 .07 .02 .06 .11 .16* -.08 (95) 8 Advertising .51 .50 .07 .03 .04 .02 .28** .16* .02 - 9 Environment .50 .50 .04 -.06 -.10 .05 -.01 -.10 -.02 -.02 -
*. Correlation is significant at the 0.05 level **. Correlation is significant at the 0.01 level
Table 2: Factor analysis
Item Rotated Factor Loadings
Intrusive Privacy Attitude Relevance How would you evaluate the advertisement you just saw?
Good-bad
.88
Like-Dislike .88
Interesting-uninteresting .83
The information in the advertisement was
Relevant .91
Useful -.38 .86
I think this offer is
disturbing .90 alarming .86 obtrusive .88 irritating .86 annoying .86 uncomfortable .88
I think it is uncomfortable that personal information is used in this
offer .86
The supplier knows a lot about me .75 This offer gives me an uneasy feeling .91 This offer gives me an unsafe feeling .91
It bothers me that the firm is able to track information about me .95 I am concerned that the firm has too much information about me .93 It bothers me that the firm is able to access information about me .93 I am concerned that my information could be used in ways I could
not forsee .90
Eigenvalues 8.73 3.43 3.09 0.91
% of Variance 43.66 17.15 15.43 4.53
To conclude: all four variables are found valid and reliable and therefore the scales are formed.
4.6 Randomisation check
There has been a check on whether the randomization of qualtrics has succeeded. For gender there has been a chi square test: χ2(3)=1.60, p=.66. This means there are no significant differences between male and females in the different conditions (generic offline, generic online, smart offline and smart online). From this can be concluded that in every condition the male/female ratio is quite similar.
Nationality has been recoded into four continents (Americas, Asia, Africa and
Europe). The χ2(9)=6.25, p=0.71. This means that the ratio of people from a certain continent is quite similar for all conditions.
Also the variables age and education have been tested. The kruskal wallis test shows that there were no significant differences for age: χ2 (3)=2.35, p=.50 and for level of education χ2 (3)=3.13, p=.37.
The overall conclusion from this analysis is that there are no differences in conditions. This means the randomisation has succeeded.
5. Analysis and results
Several ANOVA analyses have been conducted in order to test some of the hypotheses. There are different assumptions underlying an ANOVA model and therefore this research looked at the normality, homoscedasticity and independence of the variables. The latter assumption is easily met because there were different participants in every condition. The normality check has been conducted. The results (see appendix 3) point out that there is a normal distribution in the dependent variables and in all subgroups. The skewness and kurtosis is between -2 and +2 (Gravetter & Wallnau, 2014). This means that there are no other factors influencing the results and it can therefore be concluded that it is appropriate to conduct ANOVA analyses. The homoscedasticity test will be checked in the ANOVA analyses through a Levene’s test.
5.1 Testing the hypotheses 5.1.1 Hypotheses 1 and 2
In order to reject/accept hypotheses 1 and 2 a process model analysis has been conducted. For hypotheses 1 the process model was not necessary but the direct effect of advertising on attitude of the ad is also visual in the process model. In figure 1 the visual overview of the statistical model is shown
The regression model that was conducted to test the first hypothesis is displayed in Table 5 under total effect c1. From the table can be observed that c1=.05, p=.77. Because the
coefficient is not significant it can be concluded that advertisement has no direct effect on the attitude of the ad. Therefore hypothesis 1 is rejected.
However, there is still a chance that hypothesis 2 could be accepted. Possibly there is a relationship between advertisement and attitude through a third variable, the so-called
mediator. As is shown in Figure 1, that mediator is relevance. In order to accept hypothesis 2 there are other criteria that play a part. One of them is that the coefficient of the mediator has to be significant (b1=-.39, p<.001). The low p value shows that his criteria is met. The second criteria is that the absolute value of c1’ has to be smaller than the absolute value of a1. This is also the case: c1’=.258 < a1=.540. So despite the fact that c1’ is not significant (p=.058) it is lower than c1 (p=.764). This means the two criteria have been met. From a simple mediation analysis conducted using ordinary least squares path analysis, advertising indirectly
influenced attitude through its effect on relevance. Therefore hypothesis 2 is accepted. As can be seen in Table 4 whether participants see a smart advertisement instead of a generic one are estimated to have a higher relevance level (a1=.54, p=.02), and participants that have a higher relevance level are estimated to have a more positive advertising attitude (b1=-.39, p<.001). Important here is that b is negative, but the higher you score on advertising attitude scale the more positive your attitude is.
Table 3: Mediation analysis through process model
Consequent
Relevance (M) Attitude (Y)
Antecedent Coeff. SE p Coeff. SE p
Advertising (X) a1 .540 .022 .021 c1’ .258 .136 .058 Relevance (M) --- --- --- b1 -.389 .040 <.001 constant iM 3.891 .233 .022 iY 4.213 .181 <.001 R2 = .025 R2 = .316 F (1,211) = 5.356, p = .022 F (2,210) = 48.446, p < .001
Table 4: Other results of process model
5.1.2 Hypothesis 3
In order to reject or accept hypothesis 3 an one-way ANOVA was conducted. Levene’s test showed that there are equal variances: F(1,211) = 2.08, p=.15. This means the conditions for conducting an ANOVA are met. A non-significant effect of environment on advertising attitude can be concluded from table: F(1,211)=.60, p=.44, η2=.003. See appendix 4 for the full ANOVA table.
5.1.3 Hypothesis 4
In order to reject or to accept hypothesis 4 an one-way ANOVA analysis has been conducted. Levene’s test points out that there are equal variances for the dependent variable:
Effect SE p LLCI ULCI
Direct effect c1’ .258 .136 .058 -.009 .525
Total effect c1 .048 .161 .765 -.270 .367
Boot SE Boot LLCI Boot ULCI
effect of advertising on privacy concern: F (1,211)=.07, p=.80. See appendix 4 for the ANOVA table.
5.1.4 Hypothesis 5
To reject or accept this hypothesis another one-way ANOVA was conducted. The conditions to conduct the ANOVA are met: Levene’s test pointed out that there are equal variances: F(1,211) =.26, p=.61. There was a statistically significant effect of advertising on level of intrusiveness F(1, 211)= 17.74, p<.001, η2= .08. See appendix 2 for the ANOVA table. This means that the participants in the generic advertising condition have a significantly lower level of intrusiveness (M=3.11; SD=1.61) than participants in the smart advertising condition (M=4.05, SD=1.66). Therefore hypothesis 5 is accepted.
5.1.5 Hypothesis 6
To reject or accept hypothesis 6 a two way ANOVA is conducted. Levene’s test shows that there are unequal variances: F(3,209)=3.00, p=.03. The homoscedasticity assumption is violated, but this heteroscedasticity has a weak effect on the F test (F is robust against this violation). Group sizes are in this case equal (see table 4 in appendix 4), therefore this
ANOVA analysis can still be conducted. There is a non-significant main effect of advertising on privacy concern: F(1,209) = .08, p=.78, η2<.001. The same conclusion can be drawn regarding the main effect of environment on privacy concern: F(1,209)= .03, p=.86, η2<.001. However, there was a statistically significant interaction effect of advertising and environment on privacy concern: F(1,209)=17.32, p<.001, η2=.07. For the full ANOVA table see table 5 in appendix 4. This means that when the generic advertisements are showed in an offline
environment the privacy concerns are significantly lower (M=3.97, SD=1.85) than when you show them in an online environment (M=4.86, SD=1.35). For smart advertisements that are shown in an offline environment the privacy concerns are significantly higher (M=4.96,
SD=1.73) than when they are shown in an online environment (M=3.99, SD=1.59). This means that for smart advertisements it is better to show them in an online environment while for generic advertisements it is better to show them in an offline environment. The graph below will visualize this interaction effect. This means that hypothesis 6 is accepted.
Graph 1: Interaction effect of advertisement and environment on privacy concern
5.1.6 Hypothesis 7
To reject or accept this hypothesis a two-way ANOVA has been conducted. There are two independent variables: Advertising (smart/generic) and environment (online/offline). The conditions to conduct the ANOVA are met: Levene’s test shows that there are equal variances for the dependent variable ( F(3,209)=.36, p=.78. There was a significant main effect of advertising on intrusiveness, F(1,209) = 18.61, p<.001, η2=.08. There was a non- significant main effect of environment on level of intrusiveness, F(1,209)=.003, p=.96, η2<.001. There was a significant interaction effect between advertising and environment on the level of
3,8 4 4,2 4,4 4,6 4,8 5 Of/line Online
Privacy concern
Generic Smartappendix 4. This means that the level of intrusiveness for a smart advertisement in an offline environment (M4.40, SD=1.63) is significantly higher than a smart advertisement in an online environment (M=3.69, SD=1.64). For generic advertisements there are significant differences too. In the offline environment the level of intrusiveness is significantly lower (M=2.73, SD=1.63) than in the online environment (M=3.46, SD=1.52). The following graph will give a visual overview of the results. Hypothesis 7 is therefore accepted.
Graph 2. Interaction effect of advertisement and environment on intrusiveness
2 2,5 3 3,5 4 4,5 Generic Smart M ea n s
Intrusiveness
Of/line Online6. Discussion and conclusion
6.1 General discussion
The purpose of this study was to investigate the effects of different kinds of advertising (smart or generic) on diverse consumer responses (advertising attitude, intrusiveness and privacy concerns, while taking into account the mediating role of perceived relevance. Besides, the assumption was made that the environment in which the ad is shown has influence on some of the responses. The following chapter is going to discuss the research outcomes in a more detailed manner than has been done in the results chapter.
6.1.1 Advertising attitude
Based on previous research (e.g. Ahn & Bailenson, 2011; Noar, Benac & Harris, 2007; Sohl & Moyer, 2007; De Keyzer et al., 2015) there was an expected positive relationship between smart advertising and attitude towards the ad. However the analysis points out that there is no direct effect. This does not mean that consumers do not have a more positive attitude towards smart advertising. It means that that effect is caused by something else, namely relevance. There was a mediating effect of relevance hypothesized. The analysis points out that when participants saw the smart advertisement (instead of generic ads) they scored higher on relevance and therefore also had a more positive attitude towards the advertisement. This result suggests an indirect-only mediation. This means that the direct effect of advertising attitude is no longer significant when accounting for perceived relevance. In terms of ELM, this would suggest central processing: perceived personalization works by increasing the motivation to process, because the advertisement is considered more relevant. This is in line with previous research (De Keyzer et al. 2015).
Another relationship that was researched tested was whether there is a relationship between the environments where the ad was shown (online or offline) and attitude. This relationship has been studied before, their results showed a more negative attitude towards online advertising (e.g. Ducoffe, 1996; Li et al., 2002). However, this paper does not show a direct effect between environment and attitude towards the ad. This can be possibly explained based on the kinds of advertisements that were used. Previous studies used strictly stimulus materials that had characteristics that were known for that particular media. So, for example offline ads were print advertisements and online ads were banners. The lines between what is typical for offline media and for online media are becoming blurrier (Groopman, 2015). Therefore, it could be the case that participants did not see the environment where the advertisement was shown as more likeable.
6.1.2 Privacy concerns
Privacy concerns is one of the consumer responses this research investigated. Based on the analysis no direct effect between the different advertisements and privacy concern is found. This means that it does not matter if the ad is personalized based on personal data for consumer’s privacy concerns. This is not in line with previous research (e.g. Zhu & Chang, 2016; Bleier & Eisenbeiss, 2015; Tucker, 2011; Van Doorn & Hoekstra, 2013). A possible explanation is that the smart conditions in the experiment where not specifically explained based on the use of personal data. The text stated that the advertisement was being
personalized based on online browsing behaviour and gender (see full text in appendix 1). It could be the case that that text did not raise privacy concerns, because it is still a hypothetical situation and not happening in real life. Maybe the participants could not identify themselves with the person on the advertisement. It is also possible that they have privacy concerns for every kind of advertisements, so therefore there are no significant differences between the
four conditions. Another possible explanation is that personalization based on gender is not seen as something they find private. Gender is something that is generally known for
everyone, and therefore it might be the case that participants just accept that gender is known. Other detailed information about an individual might be seen as more private.
This research also looked whether there are differences in privacy concern in the different kinds of advertisements in different environments. The analysis pointed out that there was a significant interaction effect between advertising and environment on privacy concerns. This means that when the generic advertisements are shown in an offline
environment, the privacy concerns are significantly lower than when you show them in an online environment. For smart advertisements that are shown in an offline environment the privacy concerns are significantly higher than when they are shown in an online environment. This means that for generic advertisements it is better to show them in an offline environment while for smart advertisements it is better to show them in an online environment. The
analysis shows that there are effects on privacy concerns, but only in certain cases, when the relationship is studied between different kinds of advertising and privacy concerns or between ad environment and privacy concerns. You have to look at the effect of advertising and environment together. This means that the personalized advertising works better in the offline environment, which is confirmed in the study by Yu and Cude (2009). Their research shows that consumer’s were less threatened by personalized advertisements as a violation of their privacy when it was delivered offline. The research and Yu and Cude (2009) used other offline environments (letters and phone calls) than the current research (digital billboards). The outcomes are thus in line with Yu and Cude’s research (2009).
6.1.3 Level of intrusiveness
Even though there was no effect found between privacy concerns and advertising, there was a direct effect found between level of intrusiveness and advertising. The analysis points out that the participants in the generic advertising condition have a lower level of intrusiveness than participants in the smart advertising condition. This is in line with previous research (e.g. Bleier & Eisenbeiss, 2015; Tucker, 2011, Van Doorn & Hoekstra, 2013).
The interaction effects of advertising and environment on the level of intrusiveness are also investigated. A significant interaction effect between advertising and environment on the level of intrusiveness is found. This means that the level of intrusiveness for a smart advertisement in an offline environment is significantly higher than a smart advertisement in an online environment. For generic advertisements this means that in the offline environment the level of intrusiveness is significantly lower than in the online environment. It can be concluded from the results that it is better to show generic advertisements in an offline environment, than in an online environment.
6.2 Theoretical implications
This research aimed to fill the literature gap of the effect of smart advertisements on diverse consumer responses. Because there were only a few academic studies that looked at the new advertisements that are made possible by the Internet of Things, the findings of the present study offer several theoretical contributions. This study experimentally tested how smart advertising is processed and whether environment influences consumer response. Using ELM as a theory to investigate advertising attitude, the proposed mediation model was tested. A positive effect of smart advertising on advertising attitude was found. This positive effect could be fully attributed to the mediating role of perceived relevance. If the advertisement is smart, this is perceived as more relevant and therefore the advertising attitude is more
