Which shopping motive is the best to target in mobile
advertising? A study of location-based advertising
effects on consumers’ response.
by
Which shopping motive is the best to target in mobile
advertising? A study of location-based advertising
effects on consumers’ response.
by
Claudi Oostewechel
Master thesis June 20, 2016
University of Groningen Faculty of Economics and Business
Department of Marketing
ABSTRACT
The use of smartphones and the time spent on mobile devices continuously to grows the coming years, which brings many opportunities for location-based advertising. Location-based advertising is a form of targeting Location-based on the geographical location of a consumer, which makes it possible for companies to track consumers and to offer tailor-based advertisements (ads) at the right time and on the right place. A disadvantage of location-based ads is that privacy concerns can occur. However, when offering a certain benefit, consumers’ privacy concerns often decrease. Furthermore, consumers have different shopping motivations: they shop for fun and enjoyment (hedonic motives) or for functional needs (utilitarian motives). These shopping motivations are essential in understanding consumers’ shopping behaviour. This study examines the impact of shopping behaviour on consumers’ response when exposed to a based ad. It further investigates the effect of location-based ads with and without discount and incorporates privacy concerns by conducting an online experiment. The results indicate that general ads do not differ from location-based ads regarding consumers’ response towards the company and their privacy concerns. However, including a discount in the location-based ad does lead to a more favourable response and to fewer privacy concerns. Furthermore, in five out of six cases there is no difference between hedonic shoppers compared to utilitarian shoppers regarding the location-based ads and their response, which indicates that retailers could use the same strategies most of the time when targeting consumers with location-based ads.
MANAGERIAL SUMMARY
The number of smartphone users in 2014 was 2.7 billion. This number is expected to be more than 6.1 billion in 2020. Besides, the time people spend on their mobile phone continues to grow in the coming years. This brings many opportunities for marketers when it comes to mobile advertising. This research is about location-based advertising, which is part of mobile advertising. Location-based advertising is a form of targeting based on the geographical location of a consumer, which makes it possible for companies to track consumers and to offer tailor-based advertisements (ads) at the right time on the right place. Location-based revenues are expected to increase from €10.3 billion in 2014 to €34.8 billion in 2020. However, a disadvantage of location-based ads is that privacy concerns can occur. Furthermore, consumers have different shopping motivations: they shop for fun and enjoyment (hedonic motives) or for functional needs (utilitarian motives). These shopping motivations are essential in understanding consumers’ shopping behaviour. It is expected that these motivations have a different effect on consumers’ response when they are exposed to a location-based ad. Therefore, this study investigated the effects of location-based ads with and without a discount and it incorporates privacy concerns and consumers’ shopping motivations by conducting an online experiment.
The results of this study found that a general ad does not differ from a location-based ad regarding consumers’ response and consumers’ privacy concerns. A reason can be that consumers do not feel there is a differences between a general ad and a location-based ad, because both are pushed to the consumer on their mobile phone. Therefore, retailers can choose to use both types of ads in order to reach the consumer on different locations such as in front of the store or at home. Furthermore, when sending location-based ads, it is highly recommended to include a discount message rather than leaving a discount message out. Results in this study show that including a discount message in the location-based ad leads to a more favourable attitude towards the company and subsequently leads to higher visit probabilities and purchase intentions compared to a message that includes only information about a new collection in store.
PREFACE
This master thesis is my final project of the master Marketing Intelligence. After I finished my bachelor of applied science in communications, it was clear that I wanted to increase my knowledge about the various facets of marketing. Due to the high level of the master Marketing at the University of Groningen, I decided to start the pre-master Marketing in Groningen followed by the master Marketing Intelligence. This thesis completes the two amazing years of studying at the University of Groningen. Although these two years and this research period was intense, I learned a lot and it was a lot of fun.
This thesis could not have been accomplished without the help of others. Firstly, I would like to thank my first supervisor dr. L. Lobschat for the help, support and feedback throughout the whole process. Additionally, I would like to thank my fellow students for helpful discussions and insights. Finally, I would like to thank my family and friends for their support during this period as well as all the people who filled in my questionnaire.
CHAPTER 1
INTRODUCTION
In the past two years, there has been a noticeable rise in the percentages of people in the world that own a smartphone: in 2013 45 percent of the people in emerging and developed countries reported that they own a smartphone whereas the percentage is grown to 54 percent in 2015 (Pew Research Center 2016). It is expected that there are more than 6.1 billion smartphone users in 2020, whereas in 2014 the number of smartphone users was 2.7 billion (Ericsson 2014). This means there will be an increase of approximately 126 percent. Hence, the use of smartphones and other mobile devices is enormous (Kelley et al. 2011). The time people spend on their mobile devices continues to grow in the coming years (Emarketer 2015) and continues to grow faster than the use of all other media (Emarketer 2014). This gives opportunities for marketers to reach consumers through their mobile phones, namely with mobile messages (Tsang, Ho, and Liang 2004). These messages can include texts, images and voices (Barwise and Strong 2002). Since a mobile phone is a portable device that consumers can carry with them, companies can use the location of consumers to reach them at the right place and time for advertising purposes (Unni and Harmon 2007). This way of using consumers’ location to reach them is called ‘location-based advertising’. “Location-based advertising makes use of user location information and pushes location related advertisements to users’ mobile devices” (Wang, Yang, and Zhang 2015, p. 213). Moreover, to target the location of consumers, location-tracking technologies like the Global Positioning System (GPS) are used to precisely determine a users’ location (Kölmel and Alexakis 2002). GPS makes it possible for companies to track where consumers are (Pavlou and Stewart 2000) to send ads to the mobile devices of consumers related to that particular location registered (Wang, Yang, and Zhang 2015). Berg Insight (2015) expects a growth in the coming years for location-based advertising in the global market. They predicted that location-based revenues increase from €10.3 billion in 2014 with a growth rate of 22.5 percent to €34.8 billion in 2020. Furthermore, Berg Insight (2014) estimated that in 2018, Asia-Pacific is the largest location-based advertising market, followed by North America and Europe.
location (Unni and Harmon 2007) which might attract consumers into a store they are passing by (Ketelaar et al. 2015). PC users are normally not nearby the store when they receive the ad that is send to them, which might be a reason why the geographically proximate brands are less attractive to them.
In conclusion, location-based advertising is an interesting way to reach consumers and is expected to grow in the coming years (Berg Insight 2015). Besides, location-based ads can provide great opportunities for brick and mortar stores to, for example, fight competition with online shops. For instance, Zara, H&M, and WEfashion have both brick and mortar stores and an online store. These companies can send a location-based ad with a promotional message towards a consumer passing by their store in order to get consumers’ attention and to get them into the store. For pure players like Zalando, Wehkamp.nl and Amazon this is not possible because they do not have a physical store and are purely online.
Location-based ads make it possible to offer timeliness ads (Lee, Kim, and Sundar 2015) which might attract consumers into a store they are passing by and might persuade them to buy the advertised product (Ketelaar et al. 2015). Hence, the location-based ads are highly relevant in shopping districts because this is the place where many brick and mortar stores that can send out location-based ads are located. When consumers are in a shopping district, they often have either a hedonic or a utilitarian shopping motivation (Babin, Darden, and Griffin 1994). The impact of these two shopping motivations on location-based ads might be mixed because the hedonic shoppers shop for enjoyment, fun and entertainment (Hirschman and Holbrook 1982) and the utilitarian shoppers shop to fulfil a task (Jones, Reynolds, and Arnold 2006). This can subsequently lead to different responses towards the company after receiving a location-based ad. Companies owners can use this information to investigate whether they should send different types of consumers a location-based ad or not in order to increase the response of consumers towards the company. Therefore, this research captures the question to what extent consumers with different shopping motivations weaken or strengthen the relationship between exposure to location-based ads and consumers’ responses.
In sum, this research will examine the following research questions: (1) What is the effect of exposure to a location-based ad (relative to a general ad) on the consumers’ attitude towards the company, on the visit probability and on the consumers’ purchase intention? (2) What is the effect of adding a discount message in a based ad relative to a location-based ad with no discount message? (3) What is the effect of a location-location-based ad on privacy concerns? (4) Is this relationship between location-based ads and the consumers’ response mediated by privacy concerns? And (5) does the type of shopping motivation weaken or strengthen the relationship between location-based ads, consumers’ attitude towards the company, the visit probability and consumers purchase intentions? To answer these research questions, an online experiment is conducted including six different scenarios.
ads, like consumers’ evaluations regarding privacy concerns and perceived benefits (Unni and Harmon 2007) and the effectiveness of permission based ads (Barwise and Strong 2002; Gazley, Hunt, and McLaren 2015). Another interface in the location-based literature is the content of the message. Xu, Oh and Teo (2009) focussed on the differences between text- and multimedia messages whereas Gazley, Hunt and McLaren (2015) investigated the customization of location-based services. Furthermore, the acceptance of location-based ads and the attitude towards the ads is examined extensively and this acceptance often leads to higher purchase intentions (Bruner and Kumar 2007; Tsang, Ho, and Liang 2004; Wells, Kleshinski, and Lau 2012; Xu, Liao, and Li 2008; Xu, Oh, and Teo 2009). However, the attitude of consumers towards the company that sends out the message is not investigated yet. Therefore, this research will bridge the gap between the response of consumers towards the ad and their response towards the company. Secondly, how to target a consumer online in a website setting is already examined. In fact, mobile users that receive location-based ads are often constantly online and accessible since the mobile phone is a portable device (Junglas and Watson 2006). However, it is questionable whether these targeting techniques of website settings can be used in location-based settings because a mobile phone is often location-aware (Dhar and Varshney 2011) and the location-based targeting techniques want to serve the specific needs at the right time and the right place (Tam and Ho 2006). For online targeting at a website it is more difficult to reach the consumer at the right time and right place. Therefore, mobile users have a stronger interest for proximate brands compared to PC users (Ghose, Goldfarb, and Han 2012) as explained in the beginning of this introduction. This shows that targeting in a location-based setting is not always comparable to targeting in an online website setting. Wherefore the location-based topic is an interesting topic to study more extensively. Finally, the impact of consumers’ shopping motivations is not taken into account yet. However, consumers shop with a certain motivation and these hedonic or utilitarian motivations are important indicators of the behaviour of a consumer (Babin, Darden, and Griffin 1994). In order to improve the effectiveness of the location-based ads, it is valuable to examine whether hedonic or utilitarian shoppers have different responses towards location-based ads.
ads they send. For example, they can use the insights to investigate whether they should send consumers nearby a supermarket a location-based ad or not in order to increase the attitude of consumers towards the company, the visit probability and the purchase intentions. The same can be applied for stores that mainly attract hedonic shoppers. When being aware of this given fact, a company can or cannot send a location-based ad to the hedonic shoppers walking by the store in order to increase sales. Furthermore, since a discount decreases privacy concerns of consumers and often leads to more acceptance of the location-based ad, this research investigates whether a company should use a discount by default to attract consumers via a location-based ad or if it is also considerable to leave the discount out and to use a promotional message about for example a new collection in store.
CHAPTER 2
LITERATURE REVIEW
The following section gives an overview of the existing literature and introduces the hypotheses based on the gaps found in the literature. First the literature with respect to location-based ads and privacy concerns will be discussed, followed by the different shopping motivations. At the end of this section, an overview will be presented in the form of a conceptual model.
2.1 LOCATION-BASED ADS AND CONSUMER’S RESPONSE
As said in the introduction, the rise of smartphone owners (Pew Research Center 2016) and the time people spend online (Emarketer 2015) give companies the opportunity to reach consumers via mobile messages. Moreover, companies can use the location of consumers for advertising purposes. This is called: location-based advertising. Unni and Harmon (2007, p. 28) defined location-based advertising as “targeted advertising initiatives delivered to a mobile device from an identified sponsor that is specific to the location of the consumer”. In other words, “location-based advertising makes use of user location information and pushes location related advertisements to users’ mobile devices” (Wang, Yang, and Zhang 2015, p. 213). With location-based ads, companies can reach consumers when and where they are most likely to buy something, all based on geographical targeting (Kölmel and Alexakis 2002). This geographical targeting is done with location-tracking technologies like the Global Positioning System (GPS) to precisely determine a users’ location and to deliver location-specific ads to mobile devices of consumers (Lee, Kim, and Sundar 2015). This technology makes it possible to offer timely personalized services that are based on a certain location (Unni and Harmon 2007) which might persuade consumers to go into the store they are passing by and convince them to buy the advertised product (Ketelaar et al. 2015).
promotions and coupons of retailers located in that district without requested for it. When it is about the push approach, marketers have more control over the message flow (Unni and Harmon 2007) and they can trigger impulsive buying (Molitor et al. 2016) and contemporaneous sales (Fang et al. 2015). Most common reason for this impulsive buying is that consumers are exposed to the message at the right time and the right place (Fang et al. 2015). For instance, when a consumer is walking nearby a store and receives a location-based ad, he could be remembered of that store. The ad is subsequently trigger him to walk inside and perhaps to buy something, whereas he otherwise might not even think of the store in the first place.
clutter. Hence, the targeted location-based ads can encourage the consumer to purchase a certain product (Dhar and Varshney 2011).
In addition, Luo et al. (2013) explore two types of targeting in mobile advertising that are both features of location-based advertising, namely temporal targeting and geographical targeting. Temporal targeting is sending timely messages to connect with consumers at the right time whereas geographical targeting is targeting consumers by location. The authors found that temporal and geographical targeting individually increases sales because shorter temporal and geographical distances let consumers think and interpret the promotional offer more concretely (Luo et al. 2013).
Concisely, location-based ads serve the specific needs at the right time and the right place (Tam and Ho 2006) which could create more satisfaction for consumers towards the ad (Xu, Liao, and Li 2008). Satisfaction subsequently has a positive effect on attitude (Oliver 1980). Moreover, Xu, Oh and Teo (2009) found that when a consumer has a positive attitude towards a location-based ad, he/she is more likely to use the location-based ad technology. Since the attitude towards a brand is influenced by the consumer’s attitude towards the ad (Shimp 1981), it is expected that a positive attitude towards a location-based ad leads to a positive attitude towards the particular company. Fisbein and Ajzen (1975) developed the theory of reasoned action, which states that actual behaviour is determined by the intentions to perform that behaviour and this intention is determined by the attitude towards the behaviour and the subjective norms regarding that behaviour. Based on this theory it can be concluded that the positive attitude consumers have towards a company or brand subsequently leads to a higher visit probability and purchase intention. This higher purchase intention lies in the extension with the findings of Xu, Oh and Teo (2009) who found that both SMS and MMS messages in a location-based setting leads to more spontaneous purchase intentions, probably because the ad is time bound and provides local information. Based on these previous literature, there is expected that location-based ads will increase consumers’ favourable attitude towards the company, and therefore it will increase their visit probability and their purchase intentions. Hence, hypothesis 1 is stated as follows:
H1: Exposure to location-based advertisements leads (a) to a favourable attitude towards the company, (b) to an increasing visit probability and (c) to an increasing purchase intention.
direct order and sales (Barwise and Strong 2002). An example of a brand building message is a text message, informing consumers about the availability of a new brand in stores whereas the direct response message offers an explicit promotional offer (Unni and Harmon 2007). During this research there will be examined whether a promotional offer in the location-based ad, e.g. a discount of 20%, increases or decreases the attitude of consumers towards a company, their visit probability and their purchase intentions. Unni and Harmon (2007) found that when the location-based ad is pushed, promotional messages have a greater perceived value than brand building messages (Unni and Harmon 2007). This indicates for example the response on discount content in location-based ads that will increase the likelihood of a more favourable response towards the company, in comparison to for example informative information about a new brand. In addition to this example, when there is a promotion, a higher discount leads to a more positive perception of the offer (Alford and Biswas 2002). Consumers have a higher likelihood to purchase the advertised product with a higher discount compared to a lower discount and a higher discount lowers the search intentions of consumers towards other retailers (Alford and Biswas 2002). Hence, based on the findings of Unni and Harmon (2007) and Alford and Biswas (2002), it could be assumed that a discount induce a positive attitude because it leads to less search intentions, a higher likelihood to purchase and it leads to a better perceived value. Furthermore, a monetary incentive could be an added value of mobile marketing (Pousttchi and Wiedemann 2006). When taking price discount into account as monetary incentive, Grewal et al. (1998) find out that price discount, compared to brand name and store name, is the most important variable for predicting purchase intention. Following from this, research shows price discounts have a positive influence on purchase intention (Peinkofer et al. 2015) and price cuts are the most effective tool to accelerate purchases (Neslin, Henderson, and Quelch 1985). Therefore, the following hypothesis is stated:
H2: Exposure to location-based advertisements with a discount message leads (a) to a more favourable attitude towards the company, (b) to a higher visit probability and (c) to a higher purchase intention than a location-based advertisement without a discount message.
2.2 THE EFFECT OF PRIVACY CONCERNS 2.2.1 Location-based ads and privacy concerns
demographics (Kumar and Reinartz 2012). As discussed in the previous chapter, a company can also collect information about a consumers’ location. With all this information, companies are in the position to build a database and try to understand the preferences and desires of consumers in order to improve sales or to build a better relationship with the consumer (Kumar and Reinartz 2012). In de the past few years, companies were able to collect this information without the consumer being aware (Chung and Paynter 2002). Moreover, companies have collected a lot of information about consumers without reporting how they were going to use it and with whom they were sharing the information (Wang and Zhang 2015). This subsequently raises concerns about the privacy rights of the consumers (Chung and Paynter 2002). In this subsection, the privacy concerns of consumers regarding location-based ads are discussed.
information and a consumers’ location, companies can reveal consumers’ behaviour, preferences and beliefs which could be a threat for the consumer and its safety (Wicker 2012).
In many cases, people say one thing (they are intended to limit disclosure) and then do another (they actually provide their personal information) (Norberg, Horne, and Horne 2007). This is called the privacy paradox. Consumers are cautious when it comes to providing personal information and this cautioun can be weakened when offering a benefit in a message. Consumers are willing to give personal information when an incentive or discount is offered, even if they have the intention to limit disclosure of personal information (Norberg, Horne, and Horne 2007). In the same scope, Hann, Lee, Hui and Png (2002) found out that individuals made trade-offs between the costs and benefits when providing their personal information and that an economic benefit like a monetary reward helps in making the trade-off. Moreover, Wells, Kleshinksi and Lau (2012) found that consumers are willing to receive location-based ads when a certain benefit is offered. The fact that location-based ads could lead to privacy concerns might harm the growth of future location-based advertising (Kelley et al. 2011) because consumers become reluctant to exchange information (Kumar and Reinartz 2012). Xu and Teo (2005) and Unni and Harmon (2007) suggests that companies should take the importance of privacy protection for consumers into account when working with location-based ads. Especially in case of push ads, because these ads can be perceived as more privacy intrusive since consumers are not asking for an offer, but these offers are pushed to them by companies (Molitor et al. 2016). Based on the literature described in this subsection, hypothesis 3a and 3b are stated as follows:
H3a: Exposure to location-based advertisements leads to privacy concerns.
H3b: Exposure to location-based advertisements with a discount message leads to lesser privacy concerns than location-based advertisements without a discount message.
2.2.2 The mediation effect of privacy concerns
The previous subsection discusses the relationship between location-based ads and privacy concerns, this subsection describes the potential meditation effect of privacy concerns in the relationship between location-based ads, attitude towards a company, visit probability and purchase intention.
permission indeed influences the attitude towards a mobile ad. They found that consumers only have a favourable attitude towards a mobile ad when a company has asked for their permission. When they did not ask for permission, the attitude towards the ad was less favourable. They conclude that asking for permission is important to overcome privacy concerns. Hence, these privacy concerns influence the relationship between the ad and the attitude towards the ad. In addition to this and as already mentioned before, Shimp (1981) found evidence that the attitude towards a brand and the purchase intention is influenced by the attitude a consumer has towards the ad. Therefore, a negative attitude towards an ad often contributes to a negative attitude toward the company and vice versa. When taking into account the theory of reasoned action developed by Fishbein and Ajzen (1975), this means also that a negative attitude towards the company due to privacy concerns leads to a negative visit probability and a negative purchase intention.
Whereas Tsang, Ho and Liang (2004) focused on the SMS function of mobile ads, Gazley, Hunt and McLaren (2015) focused on the location aspects of mobile ads. These latter authors show that asking for permission has an impact on preventing privacy concerns. They found that if permission is obtained, consumers see the location-based ad as less interruptive which lead to more favourable attitudes towards the message and subsequently to an increasing purchase intention. Since consumers’ privacy is stated as “the unwanted intrusion by others” (Foxman and Kilcoyne 1993, p. 107), it seems reasonable that attitudes becomes less favourable and subsequently the visit probability as well as the purchase intentions lowers when a consumer feels a form of intrusion into their privacy. In addition to the negative effect of privacy concerns on attitude, Phelps, D’Souza and Nowak (2001) found that privacy concerns are also negatively related to the purchase behaviour of consumers. In the same scope, Phelps, Nowak and Ferrel (2000) found that the amount of informational control has a big impact on consumers’ purchase intentions. A high control means that consumers have a say in how information about them is used, whereas a low control means that consumers cannot influence this. Therefore, the purchase intentions of consumers and the profit of a company might increase substantially when consumers have a higher level of control over their personal information (Phelps, Nowak, and Ferrel 2000).
H4: Privacy concerns lead to (a) a less favourable attitude towards a company, (b) a lower visit probability and (c) lower purchase intentions.
H5: Privacy concerns mediate the effects of exposure to location-based advertisements on (a) the attitude towards the company, (b) the visit probability and (c) the purchase intentions.
As mentioned before, consumers are willing to give personal information when they can take advantage of it, even if they have the intention to limit disclosure of personal information (Norberg, Horne, and Horne 2007). They make a trade-off between the costs and benefits of providing their personal information and are willing to give their personal information in exchange for a reward (Acquisti and Grossklags 2005; Norberg, Horne, and Horne 2007). Furthermore, privacy concerns can raise when a consumer receives a location-based ad. However, the use of coupons with promotions and discounts can reduce these privacy concerns (Hsu, Wang, and Wen 2006). Hui, Teo and Lee (2007) found that offering a monetary reward triggers consumers to provide personal information and makes consumers willing to risk privacy invasions. Since price discounts have a positive influence on purchase intention (Grewal et al. 1998; Neslin, Henderson, and Quelch 1985; Peinkofer et al. 2015) and consumers give their information easier when an incentive is offered (Acquisti and Grossklags 2005; Hui, Teo and Lee 2007; Norberg, Horne, and Horne 2007), it is expected that privacy concerns do not mediate the relationship between the location-based ads with a discount message and the attitude towards a company, visit probability and purchase intentions.
2.3 THE IMPACT OF DIFFERENT SHOPPING MOTIVATIONS
Hedonic shoppers are sensory and experiential (Batra and Ahtola 1991). These shoppers are looking for time distortion, arousal (Bridges and Florsheim 2008), enjoyment, fun, and entertainment when shopping (Hirschman and Holbrook 1982). Hedonic shoppers seek for multisensory images like taste, music, scent, fantasy and emotional arousal during their shopping task or when using products (Hirschman and Holbrook 1982). Besides, the hedonic motivation can be affected by adventure, authority and status (To, Liao, and Lin 2007). Contrary to hedonic shoppers, utilitarian shoppers want to shop as efficient as possible. They want to achieve their shopping goals with the minimum feelings of being annoyed or irritated (Childers et al. 2001). They want to accomplish the shopping task where they are in (Jones, Reynolds, and Arnold 2006) and they are affected by convenience, cost saving, information availability and selection (To, Liao, and Lin 2007). In addition, utilitarian shoppers are goal-directed and they fully consider and evaluate information before purchasing (Babin, Darden, and Griffin 1994).
Banerjee and Dholakia (2008) examine location-based advertising as most useful in leisure time because consumers than have time to engage and process the ad better. The main reason for consumers to use location-based services is because they want to enjoy it (Ho 2012). This might cause that hedonic shopper types are relatively more attracted to location-based ads than utilitarian shoppers, because in leisure time the hedonic shoppers can enjoy and seek for adventure. Xu, Oh and Teo (2009) studied the effect of more entertaining ads compared to informative ads. They found that more entertaining ads are perceived as more attractive and effective which could also be a reason why consumers enjoy location-based ads. This entertaining message fits with the hedonic shopping motivations like fun and entertainment.
towards a particular product or service”. Thus, when consumers are loyal to a company, it could be presumed they have a favourable attitude towards the company, because attitudinal loyalty is about consumer’s perceptions and attitudes. It is also expected that consumers have a higher visit probability and a higher purchase intention when they are loyal, because behavioural loyalty is stated as the observed actions consumers take.
Jones, Reynolds and Arnold (2006) state that emotional experiences and affective components, which are part of the hedonic shopping motivations, are more important when it comes to loyalty towards a company. This could also be a reason why hedonic shoppers are more likely to create a more favourable attitude towards the company, are more probable to visit the store and have higher purchase intentions after being confronted with location-based ads compared to utilitarian shoppers. Although consumers with utilitarian shopping motives put the company they visit in their consideration set, they do not build loyalty towards that company (Jones, Reynolds, and Arnold 2006). Based on previous information it can be assumed that the positive attitude for utilitarian shoppers is not that strong as the positive attitudes hedonic shoppers have. This might also explain the actions of utilitarian shoppers in a particular store. Furthermore, utilitarian shoppers might be distracted by the location-based ads they receive, because they are goal-directed. Therefore, utilitarian shoppers might not be attracted to that kind of ads. This might indicate that utilitarian shoppers do not have a favourable attitude towards a company when sending location-based ads whereas hedonic shoppers do have a favourable attitude because they enjoy the ads. Additionally, the utilitarian shoppers probably do not have a high visit probability and a high purchase intention when exposed to location-based ads whereas the hedonic shoppers perhaps do have these intentions. Based on previous literature, the following hypotheses are formulated:
H6: Consumers with hedonic shopping motives relative to consumers with utilitarian shopping motives strengthen the relationship between location-based ads and (a) attitude towards the company, (b) visit probability and (c) purchase intention. Hence, hedonic shoppers have a more favourable attitude towards the company, have a higher visit probability and a greater purchase intention than utilitarian shoppers after being exposed to a location-based ad.
utilitarian shoppers after being exposed to a location-based ad with a discount message.
2.4 CONCEPTUAL MODEL
In order to get a clear overview of the variables investigated in this study, figure 1 shows a conceptual model. This conceptual model summarizes the hypotheses and the expected relationships of the variables. First, the relationship between exposure to location-based ads with and without a discount message on the attitude of consumers towards the company, their visit probability and their purchase intentions will be measured. Second, this relationship is expected to be mediated by the privacy concerns consumers have. Finally, a moderation effect regarding hedonic and utilitarian shopping motivations of consumers will be measured.
Figure 1. Conceptual Framework
+ Privacy - concerns Exposure to location-based advertisements - Exposure to general, no location-based ads - Exposure to location-based ads - Exposure to location-based ads with discount
Attitude towards company
Purchase intention
Shopping motivations: hedonic motives relative to utilitarian motives Visit probability
- +
CHAPTER 3
METHODOLOGY
This chapter discusses the methodology of this research. It includes the measures, the procedure, the pre-test and the sample. To assess the relationship between location-based ads (with and without discount message) and attitude towards a company, visit probability and purchase intention and to what extent these relationships are mediated by privacy concerns and moderated by shopping motivations, data among people in the age between 16 and 64 years old is collected. This is done randomly by an online experiment including six different scenarios. Malhotra (2010, p. 253) defined an experiment as followed: “an experiment is the process of manipulating one or more independent variables and measuring their effect on one or more dependent variables, while controlling for the extraneous variables.” In this study, the location-based ads and consumers’ shopping motivations are manipulated. The online experiment with scenarios is a well suited method for the purpose of this study, because in this study the effect of either a location-based ad compared to a general, no location-based mobile ad (in the sequel mentioned as: general ad) is measured along with the effect of a location-based ad with a discount message compared to a location-based ad without a discount message. Moreover, the effect of shopping motives is taken into account as well as the privacy concerns of respondents.
3.1 MEASURES
particular shopping motivation and therefore a shopping district is an appropriate setting for the scenarios. Aside from the manipulated variables, there are variables that are measured by using multiple items. These items were chosen from larger and more general questionnaires and are explained below. Appendix 2 gives an extensive overview of these items.
The structure of the online experiment is as follows: first, some pre-questions about the awareness of Zara and the current attitude of consumers regarding to Zara are asked for. Second, one out of six scenarios is presented followed by questions regarding consumers’ attitude towards Zara, their visit probability, their purchase intention and their loyalty towards the company. Third, four items with respect to privacy concerns are asked followed by some recall questions for the manipulation check and three demographic questions (age, gender and educational level). Appendix 3 represents the entire online experiment. In addition, the independent variables, the dependent variables, the moderator and the mediator are discussed below.
Independent variable
Location-based ads. To see whether different messages have a certain influence on consumers’ responses, this variable consists of three levels: (1) general, no location-based ad, no discount (general ad); (2) location-based ad, no discount; (3) location-based ad, yes discount. To manipulate this independent variable, respondents were provided with either a general ad or a location-based ad. When exposed to the location-based ad, respondents receive a 20% discount or they do not receive a discount. To emphasize the general ad in the scenarios, respondents received the following message: ‘Earlier this week, when you were at home, you received a message on your phone from Zara with the text: Check out our new collection’. The location-based setting is emphasized in the scenario by saying: ‘When you walk by Zara, you get a pop up advertisement on your phone with the message: We see that you are passing by Zara, come inside and see our new collection’. The location-based setting with a discount message is as follows: ‘When you walk by Zara, you get a pop up advertisement on your phone with the message: We see that you are passing by Zara, come inside, see our new collection and get a discount of 20%’. The different scenarios are presented in Appendix 1.
Dependent variables
measure this attitude. It includes the following four items, all measured on seven-point semantic differential scale: ‘To get an impression of your attitude towards Zara, please indicate your feelings towards Zara: (1) 1=Bad;…7=Good; (2) 1=Dislike;…7=Like; (3) 1=Unfavourable;…7=Favourable; (4) 1=Negative;…7=Positive.
Visit probability. Consumers’ visit probability is a variable that is measured by using a marketing scale from Chandran and Morwitz (2005). It includes the following three items, all measured on a 7-point Likert scale: (1) How likely are you to visit Zara when receiving the mobile message? 1=Highly unlikely; … 7= Highly likely; (2) How certain is it that you will visit Zara when receiving the mobile message? 1=Highly uncertain; … 7=Highly certain; (3) What chance is there that you will visit Zara when receiving the mobile message? 1=No chance at all; … 7=Very good chance.
Purchase intention. Consumers’ purchase intention is a variable that is measured by using the same marketing scale as for the visit probability, namely the scale from Chandran and Morwitz (2005). It includes the following three items, all measured on a 7-point Likert scale: (1) How likely are you to buy a product at Zara when receiving the mobile message? 1=Highly unlikely; … 7= Highly likely; (2) How certain is it that you will purchase a product Zara when receiving the mobile message? 1=Highly uncertain; … 7=Highly certain; (3) What chance is there that you will buy a product at Zara when receiving the mobile message? 1=No chance at all; … 7=Very good chance.
Mediator
Privacy concerns. The mediator in this study is the variable privacy concerns. A marketing scale from Unni and Harmon (2007) is used to measure privacy concerns. It includes the following four items, all measured on a 7-point Likert scale (1=Strongly disagree;…7=Strongly agree): ‘Please indicate your concerns about location-based activities: (1) I am not comfortable sharing my interests/preferences with my cell phone service provider; (2) I do not like my location to be tracked; (3) I am very concerned how my location is shared with retailers; (4) I am very concerned that my cell phone provider is tracking my location.
Moderator
shopping motive is emphasized as follows: ‘Imagine that you are on a shopping trip, alone, because the next day you have a job interview and you need new clothes. You cannot go home until you found a new outfit’. The hedonic shopping motive is emphasized by saying: ‘Imagine that you are on a shopping trip with your friends, just because you want to look around and enjoy. You do not need anything specific.’
Additional measures
Extraneous variables: extraneous variables are all variables other than the independent variables and they influence the response of the respondents and pose a threat to the internal and external validity unless they are controlled for (Malhotra 2010). The demographics age, gender and education are extraneous variables. Furthermore, the fact that respondents do or do not know Zara, did ever buy or did never buy something at Zara and that they probably have an opinion about Zara before the scenario is showed to them could affect the dependent variable. Therefore, three background questions are included in the beginning of the online experiment to measure the effects: (1) Do you know Zara; (2) Did you ever purchased something at Zara; (3) Give your overall feelings towards Zara (same scale used as for the dependent variable ‘attitude towards Zara’). Furthermore, to measure whether the manipulations for location-based ads and shopping motives work as intended, four recall questions are included: (1) Where did you receive the message? (2) What kind of discount did you receive? (3) What kind of information is given to you? (4). What was the situation in which you were during shopping. Recall question 4 exists of the following three items, all measured on a 7-point semantic differential scale (Wakefield and Inman 2003): ‘Think of the situation in which you were during shopping: 1=Practical purposes;…7=Just for fun; 1=Purely functional;…7=Pure enjoyment; 1=For a need;…7=For pleasure’.
3.2 PROCEDURE
motive is included as interaction effect in all comparisons. Since the independent variables are either nominal or categorical, a multiple regression analysis is conducted with effect coding variables. Furthermore, privacy concerns are expected to mediate the relation between a location-based ad and the attitude towards a company. To see whether there is a full mediation effect, a partial mediation effect or no mediation effect, the ‘MEDIATION macro’ of Hayes and Preacher (2014) will be used in combination with the assumptions introduced by Baron and Kenny (1986) which will be explained in subsection 4.3.5. In this study, different responses of consumers towards location-based ads will be measured, namely attitude towards the company, visit probability and purchase intention. As mentioned in the literature review, Fishbein and Ajzen (1975) developed the theory of reasoned action which states that action behaviour is determined by behavioural intentions and that behavioural intentions are determined by the attitude towards that behaviour. To examine whether this is also the case in this study, an extra mediation analysis is conducted with ‘attitude towards the company’ as mediator between a location-based ad and visit probability and as a mediator between a location-based ad and purchase intentions.
General ad Location-based ad
No discount No discount Discount Hedonic shopping motive A B C Utilitarian shopping motive D E F
Table 1. Fractional factorial design 3.3 PRE-TEST
p=0.097). Since the sample size is small and the means fit well with the corresponding answer options, it can be assumed that the manipulation worked as intended although the effect is marginally significant. The same test is conducted to measure if the manipulation for the discount message is working as intended. Recall question 2 (what kind of discount did you receive?) is used to test this manipulation. The answer options for this recall question are: (1) The message offers a discount of 20%; (2) The message offers a discount of 25%; (3) The message did not offer a discount; (4) Not sure/do not remember. The right answer for the group exposed to a discount message is answer option 1 and the right answer for the group not exposed to a discount message is option 3. An independent t-test is conducted (MDiscount=1, MNoDiscount=3; no F value and p-value are given since the standard deviations of
both groups are 0). It can be assumed that this manipulation worked as intended, since the means of the groups do correspond correctly to the right answers. Furthermore, recall question 3 (what kind of information is given to you?) is introduced to measure whether respondents did read the message carefully. The possible answer for this recall question are: (1) The message gives information about the new collection; (2) The message gives information about a new brand in the store; (3) The message did not give information about something new; (4) Not sure/do not remember. The only right answer for all the groups is answer option 1. Hence, there should be no significant difference between the six different groups. An ANOVA is used to test this (M=1 for all groups; there is no F and p-value given, since the standard deviation for each group is 0). Hence, all participants gave the right answer and it can be concluded that the manipulation worked as intended. Lastly, recall question 4 (what was the situation in which you were during shopping) is used to test whether the manipulation for shopping motivations is working as intended. An independent t-test is conducted to test the difference between the group with utilitarian shopping motivations and the group with hedonic shopping motivations (MUtilitarian=1.33, MHedonic=6; F=1.429, p=0.260).
This test indicates that there is no significant difference between the two groups. However, when looking at the means for the two shopping motives, they differ a lot. The six respondents in the utilitarian group did indicate that they were shopping for practical purposes, for functional goals and for a need (M=1.33) whereas the six respondents in the hedonic group did indicate that they were shopping just for fun, for enjoyment and pleasure (M=6). A possible reason for this insignificant result might be the small sample size of the pre-test.
very well with the corresponding answer options. Additionally, all the respondents answered at least three out of four recall questions right, therefore I can conclude that the manipulation in the scenarios work as intended in this pre-test. However, the evaluation of the pre-test indicated the need to modify some wording to emphasize that consumers do get a location-based message. Therefore, I transformed “you are passing by Zara” into “We see that you are passing by Zara”. Additionally, in the pre-test the four items about privacy concerns were asked before the scenario was introduced, which could prime the respondent too much. Hence, in the final questionnaire these questions were asked after introducing the scenario. 3.4 SAMPLE
In total, 438 people participated in this study. From this 438 people, only 344 people finished the questionnaire. The 94 people who did not finish the questionnaire are removed from the dataset. Respondents were randomly assigned for one of the six scenarios. After reading a brief description of the scenarios, consumers indicate their responses towards privacy concerns, their attitudes towards the company, their visit probability and their purchase intention by filling in the online questionnaire.
The sample descriptives are as follows: the age of the respondents lies between 16 and 64 years. 148 respondents are male whereas 196 respondents are female. From the respondents, 27 did high school, 42 respondents did college, 197 respondents have a bachelor’s degree, 74 respondents have a master’s degree and 4 respondents have their Ph.D or are even higher educated. The sample descriptives are extensively discussed in chapter 4. The following subsections includes manipulation checks, outlier checks, the distribution of the variables and the final sample.
3.4.1 Manipulation check
and 4. An independent t-test is conducted to test if there is a significant difference between the groups. The results show a significant difference (MNoLocation=2.53, MLocation=3.53;
F=69.609, p=0.000). However, the mean for the group exposed to a general ad does not fit well to the corresponding answer option 1. This group should have a mean of 1 when rounding up or down, because they received the message ‘at home’ whereas the group exposed to the location-based ads should have a mean of 3 or 4, which does fit well to the corresponding answer option. Since the mean for the general ad is not correct, it can be concluded that the manipulation did not work well.
To test whether the manipulation for the discount message is working as intended, another independent t-test is conducted. Recall question 2 (what kind of discount did you receive?) is used to test this manipulation. The answer options for this recall question are: (1) The message offers a discount of 20%; (2) The message offers a discount of 25%; (3) The message did not offer a discount; (4) Not sure/do not remember. The right answer for the group exposed to a discount message is answer option 1 and the right answer for the group not exposed to a discount message is 3. An independent t-test is conducted and shows a significant difference between the groups (MDiscount=1.72, MNoDiscount=3.16; F=75.882,
p=0.000). Since the mean for the discount group does not fit well with the corresponding answer option, it can be concluded that the manipulation is not working as intended.
To test whether respondents did read the message carefully, recall question 3 is introduced (what kind of information is given to you?). The answer options for this recall question are: (1) The message gives information about the new collection; (2) The message gives information about a new brand in the store; (3) The message did not give information about something new; (4) Not sure/do not remember. The only right answer for all the groups is answer 1. Hence, there should be no significant difference between the six different groups. An ANOVA is used to measure this and shows a marginally significant difference between the six groups (MScenarioA=2.08, MScenarioB=2.24, MScenarioC=2.65, MScenarioD=1.79,
MScenarioE=2.28, MScenarioF=2.59; F=2.049, p=0.071). Additionally, the mean does not fit well
with the corresponding answer option 1. Therefore, it can be concluded that the manipulation is not working as intended.
p=0.001). Hence, it can be concluded that the manipulation for the different shopping motivations worked as intended.
Respondents who answered two or three recall questions wrong probably did not pay many attention towards the questionnaire and scenario. Therefore, these respondents are seen as outliers. In total 127 respondents did answer two or more recall questions wrong. I decided to remove these respondents from the dataset and 217 respondents remain. To evaluate whether the manipulation results are improved, the same tests are conducted as before. For question 1, there is still a significant difference between the two groups (MNoLocation=1.79,
MLocation=3.42; F=22.592, p=0.00). However, the manipulation for this question is still not
working as intended because the mean of the group exposed to the general ad does not fit well with the corresponding answer option. For question 2, the manipulation did also not work as intended. There is no significant difference between the group exposed to a location-based ad with a discount message and the group exposed to a location-based ad without a discount message (MNoDiscount=2.97, MDiscount=1.03; F=1.557, p=0.214). Although the means are better than
before removing the respondents who answered two or more recall questions wrong. For question 3, the manipulation is still not working as intended. There is a significant difference between the six groups while the only right answer is 1 for all groups (MScenarioA=1.15,
MScenarioB=1.65, MScenarioC=2.35, MScenarioD=1.11, MScenarioE=1.66, MScenarioF=2.06; F=29.425, p=0.00). For the
fourth recall question, the mean indicates that the respondents in the utilitarian condition identified the condition as more utilitarian and the respondents in the hedonic condition identified the condition as more hedonic (MUtilitarian=1.92, MHedonic=5.61; F=2.464, p=0.118).
However, the independent t-test shows no significant difference between these two groups. Hence, it can be concluded that the manipulation for the different shopping motivations did not worked as intended.
In conclusion, the manipulations still not work as intended which is a limitation of this study. Since the means are closer to the right answers after removing the respondents with two or more wrong recall questions and because these participants probably did not pay attention when filling in the survey, I decided to keep them from the dataset in order to improve the data quality.
3.4.2 Sample distribution and outliers
about the mean (Malhotra 2010). Distributions are either symmetric or skewed. When they are symmetric, the values on either side of the centre of the distribution are the same. Additionally, the mean, mode and median are equal (Malhotra 2010). When the distribution is skewed, the positive and negative deviations from the mean are unequal (Malhotra 2010). The kurtosis measures the relative peakedness or flatness of the curve defined by the frequency distribution. If a distribution is highly skewed or noticeable peaked or flat, the statistical procedures that assume normality should be used with caution (Malhotra 2010). Histograms in Appendix 4 shows the distributions of the dependent variables and indicate the skewness and kurtosis.
This subsection also indicates whether there are outliers in the variables. Many outliers are already removed from the sample size since 127 respondents answer two or more recall questions wrong. To detect other outliers, the outlier labelling rule of Hoaglin and Iglewick (1987) is used, which is based on multiplying the interquartile range by a factor of 2.2. The rule gives an upper- and lowerbound and there is an outlier when the (extreme) values are outside the range of the upper- or lowerbound. Additionally, the boxplots in Appendix 4 visualize potential outliers.
Regarding to the attitude variable, the distribution is negatively skewed (skewness=-0.563). This means that consumers tend to indicate their feelings about Zara on the left side of the 7-point Likert scale, which is the negative side. In addition, the kurtosis statistic has a value of -0.221, indicating that the distribution is flatter than a normal distribution. Boxplot 1 in Appendix 4 shows that case number 1, 212 and 216 are outliers. Based on the outlier labelling rule, no outliers were found (highest extreme value=1.86, lowest extreme value=-2.45; upperbound=3.26; lowerbound=-3.05). Therefore, it is decided to keep these case numbers in the dataset.
The variable ‘visit probability’ has a negatively skewed distribution (skewness=-0.266), this means that the participants tend to indicate their probability to visit Zara after seeing location-based ad as slightly negative. Additionally, the kurtosis has a value of -1.018 which indicates that the distirubtion is flatter than a normal distribution. As well as boxplot 2 does not show outliers, the outlier labelling rule also did not discover any outliers (highest extreme value=1.74, lowest extreme value=-1.99; upperbound=4.56; lowerbound=-4.40).
Appendix 4 and the outlier labelling rule, no outliers were found (highest extreme value=2.45, lowest extrme value=-1.88; upperbound=3.94; lowerbound=-3.86).
The variable ‘privacy concerns’ has a negatively skewed distribution (skewness=-0.220). This means that the participants tend to indicate their privacy concerns on the left side of the 7-point Likert scale which is the less concerned side. Additionally, the kurtosis statistic has a value of -0.812 which indicates that the distribution is flatter than a normal distribution. Based on boxplot 4 in Appendix 4 and the outlier labelling rule, no outliers were found (highest extreme value=1.54, lowest extreme value=-2.46; upperbound=4.01; lowerbound=-4.11).
3.4.3 Final sample
After excluding the outliers in the dataset, the distribution of the respondents in each experimental group is as demonstrated in table 2. The descriptive statistics of the overall sample are discussed in chapter 4 as well as the descriptives for each experimental group. Hedonic Control (A) Hedonic Location-based (B) Hedonic Location-based Discount (C) Utilitarian Control (D) Utilitarian Location-based (E) Utilitarian Location-based Discount (F) 33 34 40 38 38 34
CHAPTER 4
RESULTS
In this chapter, different analyses are conducted to obtain results. Section 4.1 starts with a description of the dataset followed by sections 4.2 which explains the assumption for conducting a regression analysis. Section 4.3, 4.4 and 4.5 follow with the estimation of the models.
4.1 DESCRIPTIVE STATISTICS 4.1.1 Reliability of the items
Since the items used in this study (Appendix 2) are multi-item scales, the internal consistency between the items is tested by using Cronbach’s Alpha (CA). Internal consistency is present when the CA is higher than 0.7 (Nunnaly 1978). In this study, all the multi-item scales have a CA higher than 0.7. Additionally, to reduce the number of items a factor analysis is conducted. The factor analysis has several criteria which should be met. First, the Kaiser-Meyer-Olkin (KMO) measurement should be higher than 0.5. This measurement gives an index to examine the appropriateness of factor analysis. Values higher than 0.5 indicate that factor analysis is appropriate (Malhotra 2010). Second, Bartletts’s Test of Sphericity should be significant. This test examines whether each variable correlates perfectly with itself but has no correlation with the other variables (Malhotra 2010). Third, the items should have communalities higher than 0.6 (Field 2009). As demonstrated in Appendix 2, all the items have a CA > 0.7 and the criteria for factor analysis are met. Hence, a factor analysis is conducted to reduce the multi-item scales into a new combined variable by calculating a factor score. A factor score composite scores estimates for each respondent on the derived factors (Malhotra 2010). Hence, the multi-item scales are combined into new factors. In other words, the multi-item scales are combined into new variables in order to reduce the the number of items in this study.
4.1.2 Overall sample descriptives
20 and 40; people between 40 and 65; people between 65-80; people older than 80 (CBS, 2015a). In the Netherlands, 22.7% of population is younger than 20, 24.5% of the population is between 20 and 40, 35.5% of the population is between 40 and 65, 13.4% of the population is between 65 and 80 and 4.3% is older than 80 (CBS 2015a). In this study, 1.4% of the respondents is younger than 20, 88.9% is between 20 and 40, 9.7% is between 40 and 65 and no one is older than 65. Hence, it can be concluded that this sample is not generalizable with respect to age for the Dutch population. This is a limitation of this study since there are too many respondents between 20 and 40 years old and too few respondents that are younger than 20 and that are older than 40. The same holds for the educational level. The educational level of the participants is in most of the cases Bachelor’s degree (60.4%), followed by Master’s degree (25.3%), High school (6.9%), College (6.5%), and Ph.D. or higher (0.9%) whereas the educational level of the Dutch population is distributed differently (CBS 2015b).
From all the participants, 99 are male (45.6%) and 118 participants are female (54.4%). Based on gender, this study is generalizable with respect to gender for the Dutch population (49.5% of the Dutch population is male whereas 50.5% of the Dutch population is female) (CBS 2015a).
In total, 212 respondents do know Zara, which means that only 5 respondents do not know the Spanish apparel store. In the experiment, the scenarios explain that Zara is an apparel retailer that offers a wide assortment of clothes and that the assortment of Zara consist of business clothing as well as leisure clothing. Therefore, these 5 respondents can remain in the dataset. This variable will be included as an extraneous variable in the analyses because it can influence peoples’ response towards the location-based ad. This method of including the extraneous variables in the statistical method used to measure and adjust for its effect is called ‘statistical control’ (Malhotra 2010). For the same reason, the variable that asks respondents if they have ever purchased something at Zara will be included as an extraneous variable, because 164 respondents (75.6%) did ever purchased something at Zara and 53 respondents (24.4%) never purchased something at Zara.
included as an extraneous variable because prior attitude can influence a consumer’s response. The average for visit probability is 4.27 and for purchase intention is 3.61. Additionally, participants tend to be privacy concerned since the average is 4.64 on a scale from 1 to 7.
4.1.3 Sample descriptives per group
For an experiment, it is essential to change only one variable at a time and to hold all other variables constant because an experiment is designed to evaluate how the changing factors may lead to change in response (Settlage and Southerland 2012). In this study, there are two variables that are changing: the location-based ad and the shopping motive. Hence, this indicates that all the demographical data of the respondents should be equally distributed over the six different scenarios. When this is not the case, the statistical control method is used to control for the extraneous variables. To test whether there are significant differences in age between the groups, the ANOVA analysis is conducted. For age, the boxplot in figure 2 indicates some differences between the scenarios. The ANOVA measurement confirms this difference (MScenarioA=26.9, MScenarioB=25.9, MScenarioC=27.4, MScenarioD=28.5, MScenarioE=23.6,
MScenarioF=26.8; F=3.662, p=0.003). This means that age is not equally distributed between the
Figure 2. Age respondents per scenario
4.2 ASSUMPTIONS FOR THE LINEAR ADDITITVE REGRESSION
In order to measure the effect of location-based ads on consumers’ response moderated by different shopping motives, various multiple regression analyses are conducted with effect coding variables. A multiple regression is a statistical technique that simultaneously develops a mathematical relationship between a metric dependent variable and two or more independent variables (Malhotra 2010). Moreover, the independent variables explain the variation in the dependent variable (Malhotra 2010). The various regression analyses conducted in this study involves at every turn one single dependent variable and multiple independent, explanatory variables. When conducting the different multiple regression analyses, the following six assumptions should be met: (1) nonzero expectations, (2) heteroscedasticity, (3) correlated disturbances (autocorrelation), (4) nonnormal errors (nonnormality), (5) endogenous predictor variables and (6) multicollinearity (Leeflang et al. 2015). These assumptions are explained in this subsection. In subsection 4.3, 4.4 and 4.5 these assumptions are applied to the different estimated models and in the same subsections the results for the different models are presented.
functional form is accommodated and to see whether there are no omitted variables in the model, the Ramsey RESET-test for misspecification is conducted. The test considers powers (^2, ^3 and ^4) of the ‘original’ predictor variable and is therefore useful for detecting nonlinearities. The predictor variable in the power of 2 (ram1), 3 (ram2) and 4 (ram3) should be included in the overall model as an independent variable. When the variables have a significant effect, there exist nonlinearity and the model has a wrong functional form or there are omitted variables (Leeflang et. al 2015).
The second assumption is about heteroscedasticity. Homoscedasticity identifies whether all observations have an equal amount of variance in the residuals. When this is not the case the data includes heteroscedasticity. A heteroscedasticity test is conducted for every model by using a one-way ANOVA and testing the Homogeneity of Variances of the unstandardized residuals with as factor the dummy for either ‘general ad versus a location-based ad’ or the dummy for ‘location-based ad vs location-based ad with discount’. The output shows a Levene Statistic with a p-value. When p<0.05, the Nullhypothesis that the variances in the experimental groups can be assumed as equal is rejected. Hence, the disturbances in the model are heteroscedastic. This needs to be remedied by doing a re-estimation of the betas by Generalized Least Squares (GLS). The GLS method allows for more general disturbance characteristics and can resolve the heteroscedasticity (Leeflang et al. 2015).
The third assumption is autocorrelation. Since autocorrelation can only be applied to systematic patterns over time, this assumption is not applicable for this dataset.
