1
Spending differences between app, web and
multi-platform customers based on marketing contacts and a
new redirection form
By
Jelle van Iersel
2
Spending differences between app, web and multi-platform
customers based on marketing contacts and a new redirection
form
By
Jelle van Iersel
University of Groningen
Faculty of Economics and Business
Master Thesis Marketing Intelligence and Marketing Management
15 January 2018
Jan Jelle Christiaan van Iersel Diezerplein 22A
8021 CV Zwolle 0683949224
jjcvaniersel@gmail.com
Student number: S2022931 First Supervisor: Dr. F. Eggers Second Supervisor: Dr. J.T. Bouma
University of Groningen Faculty of Economics & Business
3
MANAGEMENT SUMMARY
The app market clearly shows large opportunities for companies to address. Online retailers are among the participants whom are identifying the mobile app market as a large
opportunity. The main focus in this paper will be the difference between web, app and multi-platform customers spending levels and several research areas are addressed to discuss this question.
Research asserts that repeated use of the app increases spending levels and that discontinuing the app reduced future spending. Customers do not only engage in single platform shopping, however, but also in platform shopping. Previous research has shown that multi-platform spending has a positive effect on average spending, spending amount and improved perceptions of the companies’ brand. Thus raising the argument if multi-platform spending is higher than that of a single platform spender. Moreover, marketing contacts can also affect the average spend of customers. Research has indicated that firm initiated contacts (FICs) are becoming increasingly unwanted, where customer initiated contacts (CICs) are becoming much more important. Next, is the point of putting a higher importance on redirecting customers more to app than to web. Where it was argued that customers spend more on average at app, it should also be expected that when a higher importance is placed on sending traffic to app this would increase customer average spending. This leads to the following problem statement:
What are the differences in spending levels between app, web and multi-platform customers?
To observe the effects, propensity score matching (PSM) had to be administered first to ensure that groups before the new redirection and after the redirection where similarly enough to be compared. Before matching, 285.000 cases were present and after the procedure 154,354 cases were used for further analysis. The further analysis is performed by using multiple linear regression.
4 Implications for retailers are that they should try to focus on the app customers where they have a higher average spend than web customers and thus represent larger opportunities. Additionally, they should try to make both their single platform web and app customers buy on the platforms they are not on, where this substantially improves average spend. Moreover, retailers should focus on promoting CICs such as increasing expenditure in Google AdWords to ensure a rise in average spending over FICs. Additionally, online retailers should try to extract value from the finding that redirecting to app over web increases average spending. This can be achieved by making appropriate app landing pages such as that customers stick to the app and will make sales on the more profitable platform.
5
PREFACE
This master thesis is the end result of my time as a Marketing Intelligence and Marketing Management student at the University of Groningen. The last five months were a rollercoaster filled with struggles and successes. In this period I was able to learn a lot about the marketing field and gain insights in the working life of a big online retailing company.
Furthermore, with the utmost gratitude I want to thank the following individuals whom supported and encouraged me during the process of creating this thesis. First, my thesis supervisor dr. Felix Eggers, who has always been able to provide me with positive and critical feedback. Only through the skype sessions and meetings we had I was able to successfully accomplish my thesis. Next I want to thank my supervisor at the online retailer for giving me the opportunity to finish my master at their company. His sharp comments and critical insights helped me tremendously. Furthermore I want to thank my thesis cohort at the online retailer, Hylke Vietmeijer, who provided me with useful insights and feedback.
Lastly, I want to thank my girlfriend, family and friends for their support and endless patience during my thesis project. Anouk, thanks for all the loving support and encouragement you gave me, I could not have done it without you. Mom, thanks for always providing a listening ear and helping me through the tough times. Dad, thanks for the motivating speeches and the interesting discussions we had about this topic.
The support of all the persons mentioned beforehand assured that I was able to finish my thesis and therefore they have my eternal gratitude. Thank you.
Jelle van Iersel
6
Table of Content
1 INTRODUCTION ... 8
2 LITERATURE REVIEW ... 11
2.1 Mobile Application Marketing ... 11
2.2 Multi-platform spending ... 12 2.3 Marketing Contacts ... 14 2.4 Redirecting customers ... 15 2.5 Conceptual Model ... 17 3 RESEARCH DESIGN ... 21 3.1 Research setting ... 21 3.2 Method ... 21 3.3 Data ... 22 3.4 Models ... 23
3.4.1 Propensity score matching ... 23
3.4.2 Multiple linear regression ... 24
4 ANALYSIS ... 27
4.1 Sample ... 27
4.2 Propensity score matching ... 28
4.3 Dataset Descriptives ... 30
4.4 Assumptions before multiple linear regression ... 30
5 RESULTS ... 32
5.1 Direct effect model ... 32
5.2 Moderation model ... 32
5.3 Moderated moderation model ... 33
5.4 Hypothesis testing ... 35
6 DISCUSSION ... 37
6.1 Discussion of main effects ... 37
6.2 Discussion of moderation effects ... 38
6.3 Discussion of control variables ... 39
6.4 Managerial implications ... 39
6.5 Limitations and future research directions ... 40
7 REFERENCES ... 42
8 APPENDIX ... 45
8.1 Appendix 1: Marketing Contact Variable Operationalization ... 45
7
8.3 Appendix 3: Histogram variable Average Spend ... 47
8.4 Appendix 4: R-script ... 48
8.5 Appendix 5: GLM output ... 55
8.6 Appendix 6: Common support graph ... 56
8
1 INTRODUCTION
Mobile device users have installed almost 156 billion mobile apps worldwide in 2015 and these figures are expected to grow to more than 210 billion installs in 2020 (International Data Corporation, 8 May 2016). Even more so, the overall global mobile app revenue in 2016 accounts for $88 billion with a revenue forecast of $189 billion for 2020 (Statista, 2017a). Therefore, the app market clearly shows large opportunities for companies to address. Although the mobile app market is a fast growing one, it is also a relatively new one. In-app purchase monetization was introduced only nine years ago in 2009 (BusinessofApps, 11 September 2017). The recent rise of the mobile apps is showing in an emerging stream of research in this area. Mobile marketing and especially mobile application marketing are very distinct from website marketing. Shankar and Balasubramanian (2009) state that the scope of the audience is different where existing and potential product users owning mobile devices have to opt in to receive marketing communication where web users often do not opt in to receive marketing communications. Moreover, they mention other differences such as a higher ability to deliver message by target location, higher ability to measure and track response and a better customer targetability for mobile marketing over web marketing. These distinctions indicate clear differences between web and app customers which will be the focus of this research. Moreover, this thesis will explicitly divide between research done on mobile and web marketing.
Many online retailers will use apps to sell their products, however they mostly use the app as an extra platform to sell instead of solely using the app as selling platform. Customers who buy on multiple platforms engage in multi-platform spending. Previous research has shown that multi-platform spending has a positive effect on average spending, spending amount and improved perceptions of the companies’ brand (Wallace, Giese and Johnson, 2004, Kumar and Venkatesan, 2005, Kushawa and Shankar, 2005). However, there is no research which analyzes a purely online multi-platform setting with incorporating the app, indicating room for extra research. Moreover, there are many ways for online retailers to reach their
9 embedded in Web content pages)”. Marketing contacts boost positive findings regarding the effect of these contacts on different key performance indicators. However, all of these effects are researched in a web environment only. Therefore, research is needed to find the effect of different online marketing contacts on apps as well as in a multi-platform scenario. Another related research question regarding multi-platform spending by customers is to which platform customers should be redirected. This question is relevant for customers who have access on a device to both the app and the web platform. When a customer decides to click on a specific advertisement in order to make a purchase the question arises whether they should be sent to the website or the app. Where website and app both have different characteristics, it is very possible that marketing contacts have a different effect on customer spending when sent to another platform. It automatically raises the question if using the app or web platform when redirecting is more effective and profitable.
This research will try to determine the difference between app customers and web customers on their spending level on a multi-platform level. Moreover, the marketing contacts who result in customer spending will be carefully examined to obtain insights how effective these distinct contacts are for both platforms spending level. This thesis will also account for different types of redirecting to observe if there are differences in customer spending levels for both the platforms. The main research question will therefore be:
What are the differences in spending levels between app, web and multi-platform customers?
It also results in several sub-questions which are:
How do marketing contact types effect both app, web and multi-platform spending? Is prioritizing app over web when redirecting more effective?
10 effectiveness of different redirecting types. Practical relevance for online retailers is to
observe whether using an app actually improves their overall customer spending. This research also aims to provide evidence that multi-platform customers have a higher average spending customers and that online retailers should aim to target them Moreover, online retailers use many different marketing contacts. It is very valuable for them to see which type of marketing contacts are preferred by their customers. Finally, online retailers are currently redirecting customers in ways they see fit. If proof is provided for a redirecting type to result in higher average spending levels, online retailers can commit to the more profitable type.
The structure of the paper is as follows: first a literature review will be presented which provides all relevant literature resulting in hypotheses and a conceptual model. Next, the research design and the data will be described. Subsequently, the results will be presented and interpreted and the main findings of analysis will be interpreted. Finally, managerial
11
2 LITERATURE REVIEW
2.1 Mobile Application Marketing
Shankar and Balasubramanian (2009) define mobile marketing as ‘’the two-way or multi-way communication and promotion of an offer between a firm and its customers using a mobile medium, device or technology’’. Shankar and Balasubramanian (2009) indicate that a mobile phone makes it easier for customers to interact with firms. This is strengthened by the fact that the mobile phone is always in the proximity of the customer. This indicates a potentially strong relationship between the customer and its phone. There are also substantial differences between mobile and web marketing. Shankar and Balasubramanian (2009) indicate three important differences between mobile and website marketing. Firstly, they explain two connected features of the mobile phone: its ultra-portability and the mobile being wireless. These features increase usage of the mobile phone and allows marketers to convey marketing messages and communicate with customers at any point in time. Secondly, the authors illustrate the advantages of location-sensitivity which provides marketers the option to target customers with individual location based offers. Shankar et al. (2010) also indicate a fourth difference which is the mobile phone being of more personal nature than the desktop resulting in customers using their mobile phones for all kinds of activities. Because the mobile phone is of a more personal nature the customer is often more engaged than desktop customers. This is even more evident for the app where mobile application commerce is a type of mobile
12 customer profitability. Dinner, van Heerde and Neslin (2015) find evidence that app usage result in both online and offline customer purchases. They ensure that it is critical to make customers use the app where it strongly enhances the probability the customer purchases from the firm. Kim, Wang and Malthouse (2015) show that branded app adoption has a positive effect on app spending as well as total customer spending for at least six months after the adoption. They also assert that repeated use of the app increases spending levels even more and that discontinuing the app reduced future spending. Moreover, Gill, Sridhar and Grewal (2017) provide evidence that the app availability in a B2B environment increases annual sales revenues. However, Gu and Kannan (2017) argue that mobile app is not necessarily
associated with higher customer spending where they present findings that suggest that app adoption has a negative impact on average spending per customer. Despite this contradictory research, it is still backed to argue that an app customer is more loyal and thus spends more than a web customer. Therefore, I hypothesize:
H1: App purchases have a higher average spend than web purchases
2.2 Multi-platform spending
It is interesting to observe differences solely between app and web users, however in reality many customers are considered ‘hybrid’ customers. These hybrid customers do not use a single platform to make sales but engage in multi-platform purchasing. There is already a vast amount of research which provide explanations why customers decide to use multiple
13 (2004) discovered that customer who use multiple platforms will also have improved
perceptions of the firm its platform offerings which leads to improved satisfaction and loyalty. Sousa and Voss (2006) explain that increased coordination between platforms result in
increased customer satisfaction which then improves retention rates. Kumar and Venkatesan (2005) proved that multi-platform customers buy more. Moreover, Kushawa and Shankar (2005) observed that multi-platform shoppers spend more than single platform shoppers as well as buying more items and more regularly. Zhang et al. (2010) find that multi-platform selling by retailers increases their share of customer’s wallets. Neslin and Shankar (2009) assert that multi-platform customer have a higher lifetime value than single platform customers. Kumar and Venkatesan (2005) proof that multi-platform customers provide significantly higher revenues, larger share of wallet, higher past customer value and a larger likelihood to stay active than single platform shoppers.
Although the amount of research on multi-platform retailing is vast, almost all research currently done focuses on multi-platform with both off- and online platforms without
involving the app. There is, however, some existing research which incorporates app usage in a multi-platform setting. Einav, Levin, Popov and Sundaresan (2014) discuss that a customer adopting a mobile shopping application results in an immediate and sustained increase in total platform purchasing. Moreover, Huang, Lu and Ba (2016) finds that consumers’ purchases increased overall when adopting a mobile shopping service at expense of a slight negative cannibalization effect for web platform sales.
This cannibalization effect is a negative synergy element of multi-platform retailing. Pauwels and Neslin (2008) also find this cannibalization effect in their multi-platform project, however the increase in purchase frequency more than compensates this effect. Many positive
synergies are also existing. According to Zhang et al. (2010) synergies can be created through cross-platform promotion, communication, information sharing, digitization, and sharing of common assets.
14 new platform for multi-platform retailers will increase average spend. Therefore it is
hypothesized:
H2: Multi-platform customers spend more on average than single platform customers
2.3 Marketing Contacts
15 increasingly unwanted, CICs are becoming much more important. Sarner and Herschel (2008) deduce that traditional FICs response rates are expected to be a lot lower than CIC response rates based on research of Shankar and Malthouse (2007) who propose that a customer’s own search is expected to be less intrusive and possibly higher interest. Moreover, multiple studies indicate that CICs are more effective in generating revenue than FICs are (De Haan et al., 2016, Li and Kannan, 2014, Shankar and Malthouse, 2007, Wiesel, Pauwels and Arts, 2011). This indicates that substantial differences are to be expected regarding the type of contact, where CICs have a much stronger positive effect on average spend than FICs. Moreover, app and web are often use for different objectives, thus differences are expected in the
effectiveness of CICs on app and web spending indicating a possible positive moderation effect. Furthermore, in the previous chapter it was asserted that multi-platform customers were more loyal and were spending more than single platform customers. This would also indicate that CICs positively moderate the effect of multi-platform customers on average spending. According to the literature previously mentioned, the following hypotheses arise:
H3: CICs have a stronger positive effect on average customer spend than FICs H3a: CICs moderate the positive effect such as that app purchases have a higher
average spend than web purchases
H3b: CICs moderate the positive effect such as that multi-platform customers spend more on average than single platform customers
2.4 Redirecting customers
16 Ginty and Page 2012). Moreover, Gofman, Moskowitz and Mets (2009) assert that the first step for any online retailer is to direct its visitors to the planned landing page. They argue that the key objective is to keep the customer on the retailers’ pages where customers are then able to perform the wanted actions. Ailawadi and Farris (2017) affirm that online retailers who have landing pages that are not of sufficient quality will lose sales to more visibly situated and attractive rival offerings. Where app and web platforms are often accessed at different
devices, it seems necessary for online retailers to create distinct landing pages per platform. According to Becker et al. (2009) there are three main classes of landing pages. The first is the homepage which features daily promotions together with outlining the retailers’
assortment. The second is a browse-able sub-catalog of products being presented on the retailers’ platform. For example the search query ‘Pampers’ can result in a landing page dedicated to all kinds of ‘Pampers’ products. The third class is called ‘search transfer’ and leads the customer to the specific item which was triggered by the search query. It is important to stress that these different classes are often triggered by distinct marketing contacts. The homepage is often accessed through direct load, product categories are often found through search engines and distinct items may be found through affiliates. When an online multi-platform retailer has the possibility to redirect customers, he has to make a decision to which platform he wishes to send them e.g. which marketing contacts are better able to stick customers to distinct platforms.
Where a new type of redirecting customers is a large change in the core business strategy of a online retailer, several effects can be expected. It was previously hypothesized that app customers spend more on average than web customers and are better customers overall. This implies a direct effect of prioritizing app redirection over web redirection has a positive effect on average spend. Moreover, the prioritization on app redirection might moderate the
previous hypotheses mentioned as well as moderating the moderated effects because of the large change in business implementation. It is therefore hypothesized:
H4: Prioritizing app redirection over web redirection has a positive effect on average spend
17
H4b: Prioritizing app redirection over web redirection moderates the positive effect such as that multi-platform customers spend more on average than single platform
customers
H4c: Prioritizing app redirection over web redirection moderates the positive effect such as that CICs have a stronger positive effect on average customer spend than FICs H4d: Prioritizing app redirection over web redirection moderates the moderated effect of CICs on platform spending such as that app purchases have a higher average spend
than web purchases
H4e: Prioritizing app redirection over web redirection moderates the moderated effect of CICs on multi-platform spending such as that multi-platform customers spend more
on average than single platform customers
2.5 Conceptual Model
Based on the concepts described in the literature review, several hypotheses were designed (repeated in table 1). As can be seen in the table, there are in total four main effects, five possible moderation effects and two possible moderated moderation effects. The hypotheses are visually represented in three conceptual models, one featuring the direct effects, one featuring the moderation effects and one featuring the moderated moderation effects (figure 1, 2 and 3).
Table 1. Hypotheses Hypothesis
H1 App purchases have a higher average spend than web purchases
H2 Multi-platform customers spend more on average than single platform customers
H3 CICs have a stronger positive effect on average customer spend than FICs
H3a CICs moderate the positive effect such as that app purchases have a higher average spend than web purchases
18
H4 Prioritizing app redirection over web redirection has a positive effect on average spend
H4a Prioritizing app redirection over web redirection moderates the positive effect such as that app purchases have a higher average spend than web purchases
H4b Prioritizing app redirection over web redirection moderates the positive effect such as that multi-platform customers spend more on average than single platform customers
H4c Prioritizing app redirection over web redirection moderates the positive effect such as that CICs have a stronger positive effect on average customer spend than FICs
H4d Prioritizing app redirection over web redirection moderates the moderated effect of CICs on platform spending such as that app purchases have a higher average spend than web purchases
H4e Prioritizing app redirection over web redirection moderates the moderated effect of CICs on multi-platform spending such as that multi-platform customers spend more on average than single platform customers
The direct effects conceptual model in figure 1 shows the expected hypotheses that app has a higher average spending than web and that multi-platform spenders have a higher average spending than either app or web customers. It also shows the difference in effect marketing contacts have on average spending for all platforms. Additionally, it provides the positive effect of the new redirection on average spending.
The moderation effects conceptual model in figure 2 also shows all the direct effects
previously mentioned indicated by the black straight lines. The moderation effects are shown through H3a, H3b, H4a, H4b and H4c indicated by the blue dotted lines.
19
Figure 1: Direct effects conceptual model
20
21
3 RESEARCH DESIGN
In the first part of this chapter the methodology associated with the analysis is given. Next, the data collection itself is presented and discussed as well as providing answers regarding questions about sampling size, sampling method and the population.
3.1 Research setting
This research will be based on data provided by a large online retailer. The online retailer specializes in fashion items, but also sells many other products such as household appliances, home and garden equipment and toys. This online retailer used to sell products through a catalog but now utilizes both app and website to sell products and does not have any brick-and-mortar stores. This online retailer used to sell products mainly through its website, although it is increasingly emphasizing importance towards their branded app. On the 1st of August 2017, the online retailer introduced a new way of redirecting their customers with a larger priority on their branded app. Marketing contacts that previously redirected customers to the branded website are now sent to the app granted that the customer has actually
downloaded the app. This indicates a relative increase in traffic to the app and a relative decrease in traffic to the website. Therefore, this research is based on two distinct time
periods: Pre- and post of introducing the new redirecting type with three different groups: app, web and multi-platform buyers.
3.2 Method
22 other, the change in period can be attributed solely to the treatment . This will remove timing issues from the model. It is then possible to observe the change in redirection by using a dummy in a multiple linear regression. This analysis has a continuous dependent variable which is average spending and multiple explanatory variables such a marketing contacts, events and platform type. By using these techniques, the effect of the new redirecting can be observed as well as differences between web, app and multi-platform users on their spending levels. Therefore, using PSM in combination with multiple linear regression will answer all the drawn up hypotheses and thus this research design will be chosen.
3.3 Data
23
Table 2. Variable Descriptions
Variable Description Code
New Redirection The change in redirecting customers to the branded app/website.
Dummy-coded, where a 0 indicates the period with the new redirection and a 1 indicates the period with the old redirection
Marketing Contact The type of marketing contact which is used before a customer makes a sale (more information in table 2)
Dummy-coded, where a 0 indicates a FIC and a 1 indicates a CIC
Platform The platform a customer uses to make a purchase, which is either web or app
Dummy-coded, where a 0 indicates web and a 1 indicates app
Events A promotional event was launched by the online retailers
Dummy-coded, where a 0 indicates no event and a 1 indicates a promotional event
Multi-platform Variable indicates if a customer has spent at a single or multiple platforms
Dummy-coded, where a 0 represents single platform and a 1 indicates multi-platform
Age The age of the customer who
made the purchase
Continuous discrete
variable, ranging from 16-80
Membership Starting Year (MSY)
The year a customer made its first purchase at the online retailer.
Continuous discrete
variable, ranging from 1964-2017
Gender The gender of the customer who made the purchase
Dummy-coded, where a 0 represents male and a 1 indicates female
Average spending The amount a customer spent on average per unique sale
Continuous variable, where the amount of the sales is divided per amount of orders
3.4 Models
3.4.1 Propensity score matching
A distinction is made between customers who made purchases when using the old redirection in comparison with customers who made purchases using the new redirection. In total 77,395 sales are observed using the old redirection and 207,102 sales are observed using the new redirection. On base of these observations a logistic model is established, accompanied by the propensity scores of the customers. The PSM model will now be clarified.
24 𝑃𝑟(𝑁𝑒𝑤𝑅𝑒𝑑𝑖𝑟𝑒𝑐𝑡𝑖𝑜𝑛𝑖 = 1) =
𝛽0 + 𝛽1(𝑃𝑙𝑎𝑡𝑓𝑜𝑟𝑚𝑖) + 𝛽2(𝑀𝑢𝑙𝑡𝑖 − 𝑃𝑙𝑎𝑡𝑓𝑜𝑟𝑚𝑖) + 𝛽3(𝑀𝑎𝑟𝑘𝑒𝑡𝑖𝑛𝑔𝐶𝑜𝑛𝑡𝑎𝑐𝑡𝑖) + 𝛽4(𝐸𝑣𝑒𝑛𝑡𝑠𝑖) + 𝛽5(𝐺𝑒𝑛𝑑𝑒𝑟𝑖) + 𝛽6(𝑀𝑆𝑌𝑖) + 𝛽7(𝐴𝑔𝑒𝑖) + 𝜀𝑖
where
𝑃𝑟(𝑁𝑒𝑤𝑅𝑒𝑑𝑖𝑟𝑒𝑐𝑡𝑖𝑜𝑛𝑖 = 1) is the probability score of a consumer i being redirected differently, when this is true = 1 otherwise New Redirection = 0;
𝛽0 is the intercept of the logistic model;
𝛽𝑛𝑖 is the coefficient of the variable used for matching per customer which are: marketing contact, events, platform, multi-platform, age, gender and membership starting year;
𝜀𝑖 is the error term.
This model will result in a propensity score per customer. Based on these scores, PSM is performed and a dataset is retained with matchings based on the aforementioned variables. This dataset will then be used to run multiple linear regression.
3.4.2 Multiple linear regression
Three separate models will be estimated, a direct effect model, a model incorporating direct effects and moderation effects and a model which incorporates direct effects, moderation effects and moderating moderated effects. All three models will now be clarified.
Direct Model
𝑌𝑖 = 𝛽0 + 𝛽1(𝑁𝑒𝑤𝑅𝑒𝑑𝑖𝑟𝑒𝑐𝑡𝑖𝑜𝑛𝑖) + 𝛽2(𝑃𝑙𝑎𝑡𝑓𝑜𝑟𝑚𝑖) + 𝛽3(𝑀𝑢𝑙𝑡𝑖𝑃𝑙𝑎𝑡𝑓𝑜𝑟𝑚𝑖) + 𝛽4(𝑀𝑎𝑟𝑘𝑒𝑡𝑖𝑛𝑔𝐶𝑜𝑛𝑡𝑎𝑐𝑡𝑖) + 𝛽5(𝐸𝑣𝑒𝑛𝑡𝑠𝑖) + 𝛽6(𝐺𝑒𝑛𝑑𝑒𝑟𝑖) + 𝛽7(𝑀𝑆𝑌𝑖)+ 𝛽8(𝐴𝑔𝑒𝑖) + 𝜀𝑖 where
𝑌𝑖 is the average spend for a customer of observation i ;
𝛽0 is the intercept of the fixed-effect regression;
𝛽𝑛𝑖 is the coefficient of variable 𝑛 for observation i ;
25 Moderation Model 𝑌𝑖 = 𝛽0 + 𝛽1(𝑁𝑒𝑤𝑅𝑒𝑑𝑖𝑟𝑒𝑐𝑡𝑖𝑜𝑛𝑖) + 𝛽2(𝑃𝑙𝑎𝑡𝑓𝑜𝑟𝑚𝑖) + 𝛽3(𝑀𝑢𝑙𝑡𝑖𝑃𝑙𝑎𝑡𝑓𝑜𝑟𝑚𝑖) + 𝛽4(𝑀𝑎𝑟𝑘𝑒𝑡𝑖𝑛𝑔𝐶𝑜𝑛𝑡𝑎𝑐𝑡𝑖) + 𝛽5(𝐸𝑣𝑒𝑛𝑡𝑠𝑖) + 𝛽6(𝐺𝑒𝑛𝑑𝑒𝑟𝑖) + 𝛽7(𝑀𝑆𝑌𝑖)+ 𝛽8(𝐴𝑔𝑒𝑖) + 𝛽9(𝑀𝑎𝑟𝑘𝑒𝑡𝑖𝑛𝑔𝐶𝑜𝑛𝑡𝑎𝑐𝑡 ∗ 𝑃𝑙𝑎𝑡𝑓𝑜𝑟𝑚𝑖) + 𝛽10(𝑀𝑎𝑟𝑘𝑒𝑡𝑖𝑛𝑔𝐶𝑜𝑛𝑡𝑎𝑐𝑡 ∗ 𝑀𝑢𝑙𝑡𝑖𝑃𝑙𝑎𝑡𝑓𝑜𝑟𝑚𝑖) + 𝛽11(𝑁𝑒𝑤𝑅𝑒𝑑𝑖𝑟𝑒𝑐𝑡𝑖𝑜𝑛 ∗ 𝑃𝑙𝑎𝑡𝑓𝑜𝑟𝑚𝑖) + 𝛽12(𝑁𝑒𝑤𝑅𝑒𝑑𝑖𝑟𝑒𝑐𝑡𝑖𝑜𝑛 ∗ 𝑀𝑢𝑙𝑡𝑖𝑃𝑙𝑎𝑡𝑓𝑜𝑟𝑚𝑖) + 𝛽13(𝑁𝑒𝑤𝑅𝑒𝑑𝑖𝑟𝑒𝑐𝑡𝑖𝑜𝑛 ∗ 𝑀𝑎𝑟𝑘𝑒𝑡𝑖𝑛𝑔𝐶𝑜𝑛𝑡𝑎𝑐𝑡𝑖) + 𝜀𝑖 where
𝑌𝑖 is the average spend for a customer of observation i;
𝛽0 is the intercept of the fixed-effect regression;
𝛽1−8𝑖 is the coefficient of the specific variable of observation i;
𝛽9−13 is the moderation coefficient of the specific variable for observation i;
𝜀𝑖 is the error term.
Moderated moderation model
𝑌𝑖 = 𝛽0 + 𝛽1(𝑁𝑒𝑤𝑅𝑒𝑑𝑖𝑟𝑒𝑐𝑡𝑖𝑜𝑛𝑖) + 𝛽2(𝑃𝑙𝑎𝑡𝑓𝑜𝑟𝑚𝑖) + 𝛽3(𝑀𝑢𝑙𝑡𝑖 − 𝑃𝑙𝑎𝑡𝑓𝑜𝑟𝑚𝑖) + 𝛽4(𝑀𝑎𝑟𝑘𝑒𝑡𝑖𝑛𝑔𝐶𝑜𝑛𝑡𝑎𝑐𝑡𝑖) + 𝛽5(𝐸𝑣𝑒𝑛𝑡𝑠𝑖) + 𝛽6(𝐺𝑒𝑛𝑑𝑒𝑟𝑖) + 𝛽7(𝑀𝑆𝑌𝑖)+ 𝛽8(𝐴𝑔𝑒𝑖) + 𝛽9(𝑀𝑎𝑟𝑘𝑒𝑡𝑖𝑛𝑔𝐶𝑜𝑛𝑡𝑎𝑐𝑡 ∗ 𝑃𝑙𝑎𝑡𝑓𝑜𝑟𝑚𝑖) + 𝛽10(𝑀𝑎𝑟𝑘𝑒𝑡𝑖𝑛𝑔𝐶𝑜𝑛𝑡𝑎𝑐𝑡 ∗ 𝑀𝑢𝑙𝑡𝑖𝑃𝑙𝑎𝑡𝑓𝑜𝑟𝑚𝑖) + 𝛽11(𝑁𝑒𝑤𝑅𝑒𝑑𝑖𝑟𝑒𝑐𝑡𝑖𝑜𝑛 ∗ 𝑃𝑙𝑎𝑡𝑓𝑜𝑟𝑚𝑖) + 𝛽12(𝑁𝑒𝑤𝑅𝑒𝑑𝑖𝑟𝑒𝑐𝑡𝑖𝑜𝑛 ∗ 𝑀𝑢𝑙𝑡𝑖𝑃𝑙𝑎𝑡𝑓𝑜𝑟𝑚𝑖) + 𝛽13(𝑁𝑒𝑤𝑅𝑒𝑑𝑖𝑟𝑒𝑐𝑡𝑖𝑜𝑛 ∗ 𝑀𝑎𝑟𝑘𝑒𝑡𝑖𝑛𝑔𝐶𝑜𝑛𝑡𝑎𝑐𝑡𝑖) + 𝛽14(𝑁𝑒𝑤𝑅𝑒𝑑𝑖𝑟𝑒𝑐𝑡𝑖𝑜𝑛 ∗ 𝑀𝑎𝑟𝑘𝑒𝑡𝑖𝑛𝑔𝐶𝑜𝑛𝑡𝑎𝑐𝑡 ∗ 𝑃𝑙𝑎𝑡𝑓𝑜𝑟𝑚𝑖) + 𝛽15(𝑁𝑒𝑤𝑅𝑒𝑑𝑖𝑟𝑒𝑐𝑡𝑖𝑜𝑛 ∗ 𝑀𝑎𝑟𝑘𝑒𝑡𝑖𝑛𝑔𝐶𝑜𝑛𝑡𝑎𝑐𝑡 ∗ 𝑀𝑢𝑙𝑡𝑖𝑃𝑙𝑎𝑡𝑓𝑜𝑟𝑚𝑖) + 𝜀𝑖 where
𝑌𝑖 is the average spend for a customer of observation i;
𝛽0 is the intercept of the fixed-effect regression;
𝛽1−8𝑖 is the coefficient of the specific variable of observation i;
26 𝛽14−15 is the moderated moderation coefficient of the specific variable for observation i;
27
4 ANALYSIS
4.1 Sample
Firstly, the data has to be cleaned to be able to analyze the data. Age was present in the data set ranging from age 0 to 117. These large outliers were observed where some customers did not want to fill in their date of birth or where one filled in the minimum year of birth of 1900. Online retailers mostly sell to a relative younger customer base, therefore relatively large age values might interfere with the results of the age variable. It was decided that all cased were omitted which fell outside the age range of 16 to 80 years of age. This resulted in 9,777 cases being omitted and made the age variable more cohesive. The membership starting year
(MSY) variable was examined as well. MSY commences at the year 1964 and ranges to 2017. Online retailing has begun to grow expansively mostly in the last two decades. Indeed, the boxplot in appendix 2 proves this by showing that all outliers lie after 1990. It is however very plausible that some customers are lifelong loyal to the online retailer, especially in such a large dataset. It is therefore decided to leave the MSY variable as it is, to also incorporate the effect of these very loyal customers. Finally, the average spending variable was closely
examined. This variable ranges between values of 0 and 8,509.65. The online retailer does sell large household appliances and electronics, which allows for average spending numbers to become substantially larger, especially when customers buy more of these items at the same purchase. The online retailer does also sell smaller and cheaper items which costs only a few euros. The differences in these item costs result in a large disparity in the variable. The mean of the variable is 140.62 and the standard deviation 168.23 indicating that especially the very large values will be an issue. It is therefore decided that roughly the largest and the smallest 1 % will be omitted from the data (60,774 cases) to obtain more cohesiveness. This results in the variable having a range from 13 to 811. The results can also be seen in the histogram in appendix 3. As can be seen, there are still substantial outliers because of the large
28
4.2 Propensity score matching
PSM will be used to determine if there is a significant effect in different redirection effects on average spending. However, where PSM is a very data sensitive type of modelling, it will not be possible to use the entire dataset to match pairs of data. Therefore, a random sub-sample accounting for approximately 10 % of the total dataset of 2,779,480 cases will be taken instead. This marks the sub-sample at 285,000 observations. A substantial difference in effect should be existent between the two periods for the dependent variable Average Spending. The statistics of the sample regarding the two groups can be found in table 3.
Table 3. Descriptives Redirection Groups on Average Spending
Redirection Group Observations Mean Average Spending Std. Error
Old Redirection 77,177 126.82 0.46
New Redirection 207,823 133.53 0.29
A t-test of the redirection type on Average Spending (t = 12.359, Sig = .000, P < 0.05)
indicates a significant effect providing evidence that the two periods have different effects on the average spending variable. It is now important to observe the means of all the other variables which are used in this analysis. These variables should be significantly different from each other before matching. As table 4 shows, all variables are significantly different between both redirection periods.
Table 4. Mean differences and significance levels of the covariates per redirection group (P<0.05)
29
Age 42.0339 41.5898 -8.6453 .0000***
Where all of these variables are significant they can all be used in the GLM model to assess if they have a significant effect on the redirection variable. The GLM output can be seen in appendix 5. All of the variables have a significant effect on the redirection type and therefore all of the variables will be used for PSM.
First the propensity scores are measured for all observations in both periods. Although the t-test had previously shown shows that the variables are indeed significantly different of each other per period, the means are still relatively close to each other. This is also indicated by the common support graph based on the propensity scores of the observations which can be found in Appendix 6. It indicates that the groups are already quite similar of form before matching. Nonetheless, PSM will be proceeded to find an even better fit of the data. In total 154,354 observations remained after matching. PSM strengthened comparability between redirection groups resulting in large balance improvements per variable which can be observed in
Appendix 4. Moreover, the variables should not be different between periods anymore. Table 5 indeed shows that all the variable are not significant and are thus not different anymore between periods.
Table 5. Variable coefficients and significance levels on the redirection dummy after matching (P<0.05)
30 Moreover, a plot with all variables and their distribution can be examined in Appendix 5. It clearly shows matching propensity scores for all variables. It should be noted that the effect on redirection on average spending should still hold a significant effect between periods. The t-test of the redirection dummy on average spending confirms that there is still a difference between periods ((t = 12.172, Sig = .000, P < 0.05). Therefore the new dataset will be used to perform multiple linear regression.
4.3 Dataset Descriptives
The final dataset comprises of 154,354 observations. Of these observations, 121,137 are female (78.48 %) and 33,217 male (21.52 %). The average age of the observation group is 42 years and the average membership starting year is the year 2008. The descriptives clearly show the strategy of the online retailer to target families with children. Moreover, of all observations, 33,155 (21.48 %) were sales made during an event and 111,682 (72.35%) observations were CICs. This indicates that CICs are already much more used by customers. Furthermore, average spend of the sub-sample is €130.89. Finally, the amount of app sales (18,151, 11.76 %) and multi-platform buyers (20,185, 13.08 %) are a lot smaller than web sales and single platform buyers. In the next chapter, the results of the direct and the interaction model will be examined.
4.4 Assumptions before multiple linear regression
Several assumptions first need to be addressed before the data can be analyzed. Firstly, linearity has to be assumed. The scatterplots seen in figure X. do not show a linear
31 possible presence of autocorrelation. The Durbin-Watson test shows no evidence of
autocorrelation (DW = 2.0011, Sig = 0.5893, P < 0.05). The final assumption which has to be checked for is the presence of multicollinearity. Table 6 provides evidence that
multicollinearity is not an issue for the variables in the dataset where the VIF values do not exceed the threshold of 10. In the next chapter, the results of the direct and the interaction model will be examined.
Table 6. VIF-scores for multicollinearity
32
5 RESULTS
This chapter will describe the final data results from the three developed multiple linear regression models. Moreover it will provide tables in a manner that the aforementioned hypotheses will be answered.
5.1 Direct effect model
Multiple linear regression was used for the (main) final analysis of the direct effect model. Table 7 shows the results of this analysis. It can be derived from the table at the last column on p-values that all effects have a significant effect on customer average spending at the .05 significance level. The main hypotheses estimates of New Redirection, platform, multi-platform are all highly significant with p-values of .0000. Marketing contact is also significant, however not as strongly, with a p-value of 0.0203. Moreover, the direct model shows highly significant p-values for the control variables events, gender, membership starting year and age.
Table 7. Variable estimates direct model and significance levels (P<0.05)
5.2 Moderation model
The results in table 8 show relatively higher p-values for the moderation model than for the direct effect model. However all the main effects stay significant with a p-value of .0009 for the new redirection, a p-value of 0.0003 for platform, a p-value of .0000 for multi-platform
33 and a p-value of .0051 for marketing contacts. The control variables of events, gender,
membership starting year and age all stay highly significant with a p-value of .0000. The moderation effects are mostly not significant where only the interaction between marketing contacts and multi-platform is significant with a p-value of .0009.
Table 8. Variable estimates moderation model and significance levels (P<0.05)
5.3 Moderated moderation model
Again, multiple linear regression was used the capture all the moderation effects including the moderated moderation effects. The results in table 9 show higher p-values for the moderated moderation model than for the other two models. However all the main effects stay significant with a p-value of .0186 for the new redirection, a p-value of 0.0197 for platform, a p-value of .0000 for multi-platform and a p-value of .0192 for marketing contacts. The control variables of events, gender, membership starting year and age all stay highly significant with a p-value of .0000. The moderation effects are mostly not significant where only the interaction
between marketing contacts and multi-platform is significant with a p-value of .0111.
34
Table 9. Variable estimates moderation moderation model and significance levels (P<0.05)
Next, the Bayesian Information Criterion (BIC) is used to assess which model should be used to interpret the estimates. The scores can be found in table 10. Where the BIC of the direct effect model is lowest, this model will be used to assess the direct effects. The moderation model outperforms the moderated moderation model so this model will be used to interpret the moderation effects.
Table 10. BIC scores
Variables Estimates Std. Error T-score P-value Intercept 2069.6686 82.5023 25.086 .0000*** NewRedirection -10.3168 4.3839 -2.353 .0186* Platform 6.9650 2.9868 2.332 .0197* MultiPlatform 9.5419 2.7174 3.511 .0000*** Marketing Contact -8.7095 3.7192 -2.342 .0192* Events 3.4331 .8044 4.268 .0000*** Gender -5.3903 .8161 -6.605 .0000*** MSY -.9618 .0407 -23.634 .0000*** Age -.2919 .0303 -9.642 .0000*** MarketingContact*Platform 3.3642 3.4983 .962 .3362 MarketingContact*MultiPlatform 8.2473 3.2471 2.540 .0111* NewRed*Platform 3.1662 4.1438 .764 .4448 NewRed*MultiPlatform .76384 3,7941 .201 .8404 NewRed*MarketingContact 3.5077 5.2221 .672 .5018 NewRed*MarketingContact*Platform -4.5264 4.8785 -.928 .3535 NewRed*MarketingContact*MultiPlatform -1.5153 4.5471 -.333 .7390
Variables Direct effect model Moderation model
Moderated moderation
35
5.4 Hypothesis testing
Here, the described hypotheses will be tested and evaluated based on the aforementioned data. The results will indicate if the suggested hypotheses are supported or not as well as providing a concise explanation. The first hypothesis can be accepted where there is a significant effect between the platform variable and average spending in the direct model of p-value .0000. Multi-platform also shows a significant effect on average spending for the direct model with a p-value of .0000. H3, H3a and H3b cover the construct of marketing contacts its direct effect and its moderation effect on platform and multi-platform. H3 can be accepted where the direct gives a significant p-value for the effect of contact type on average spending of .0203. The moderation effect of marketing contact type of platform of H3a will not be accepted where the p-value indicates a not significant level of .665. The moderation effect of H3b is existent however, shown by the significant p-value of marketing contact type on multi-platform of .0009. Furthermore, H4, H4a, H4b, H4c, H4d and H4e cover the construct of the new
redirection its direct effect, its moderation effect and the moderation on the moderated effect. The direct effect of new redirection on average shows a significant p-value for the direct model of .0000 and therefore support in accepting H4. The moderation effect of the new redirection on platform, multi-platform and contact type on average spending all give insignificant p-values (.9671, . 9134 and .4025). Therefore, H4a, H4b and H4c are clearly rejected. Finally, the moderation effect of new redirection on the moderated effect of marketing contact type on platform and multi-platform spending provides insignificant p-values (.3535 and .7390). Consequently, H4d and H4e have to be rejected. Moreover, an overview of all the hypotheses and their outcome is presented through table 11.
Table 11. Support of hypotheses
Hypothesis Supported
H1 App purchases have a higher average spend than web purchases
YES
H2 Multi-platform customers spend more on average than single platform customers
YES
H3 CICs have a stronger positive effect on average customer spend than FICs
36
H3a CICs moderate the positive effect such as that app purchases have a higher average spend than web purchases
NO
H3b CICs moderate the positive effect such as that multi-platform customers spend more on average than single platform customers
YES
H4 Prioritizing app redirection over web redirection has a positive effect on average spend
YES
H4a Prioritizing app redirection over web redirection moderates the positive effect such as that app purchases have a higher average spend than web purchases
NO
H4b Prioritizing app redirection over web redirection moderates the positive effect such as that multi-platform customers spend more on average than single platform customers
NO
H4c Prioritizing app redirection over web redirection moderates the positive effect such as that CICs have a stronger positive effect on average customer spend than FICs
NO
H4d Prioritizing app redirection over web redirection moderates the moderated effect of CICs on platform spending such as that app purchases have a higher average spend than web purchases
NO
H4e Prioritizing app redirection over web redirection moderates the moderated effect of CICs on multi-platform spending such as that multi-platform customers spend more on average than single platform customers
37
6 DISCUSSION
The next chapter will discuss the main and the moderation effects. Additionally, managerial implications will be provided. The final section will cover both limitations of the study as well as suggestions for future research possibilities.
6.1 Discussion of main effects
The main effects considered in this paragraph are the effects of the independent variables on the outcome variable. The first main effect in this research was that app purchases are of higher average spend than web purchases. The hypothesis was based on the theory that app customers are more loyal which in turn would lead to higher average spending for app over web customers. Outcome of research indeed support the claim that app purchases are of higher spend. This is indicated by the positive estimates in the direct and moderation model and significant p-values. According to the direct model, average spend on app is 9.3277 euro higher than on web.
The second main effect was that customers who buy on multiple platforms have a higher average spend than single platform customers. This was in line with many previous research which supported the relationship between multi-platform spending and a higher average spend (Kumar and Venkatesan, 2005; Kushawa and Shankar 2005). Again, research provided
significant evidence in support of this claim as well as a positive estimate. According to the direct model, average spend on app is 15.0789 euro higher than on web.
The third main effect was that CICs have a larger positive effect on average spending than FICs. This argument was based on Blattberg, Kim and Neslin (2008), who argue that FICs are becoming increasingly unwanted. Moreover, De Haan et al. (2016), Li and Kannan (2014), Shankar and Malthouse (2007), Wiesel, Pauwels and Arts (2011) all find evidence of CICs being more effective in creating revenue than FICs. Research results find substantial significance of difference between FIC and CIC on average spending, however estimates differ largely between the direct and the moderation model. According to the direct model, average spend of CICs is 1.7208 euro higher than that of FICs.
38 that the new redirection will result in higher average spend. Outcome of research indeed supports the claim that the new redirection type increases average spend again provided with positive estimates and significant p-values. According to the direct model, average spend increases by 8.1596 euro after the new redirection was implemented.
6.2 Discussion of moderation effects
The first two moderation effects are based around the marketing contact type. It was
hypothesized that marketing contacts moderate the positive effect such as that app purchases have a higher average spend than web purchases. Because an insignificant p-value was found, the hypothesis was rejected. This indicates that marketing contact type does not moderate the difference in effect of app and web spending. A reason might be that customers on different platforms are not as differently influenced by distinctive marketing contacts. This could also be observed from the direct effect which has a small estimate. The other hypothesis suggested that marketing contacts moderate the positive effect such as that multi-platform customers spend more on average than single platform customers. This hypothesis has a positive estimate and a significant p-value and therefore this hypothesis is confirmed. This indicates that marketing contact type moderates the difference in effect between multi-platform and single platform customers. The positive estimate of 7.5106 illustrates that when multi-platform customers use a CIC, their average spend increases substantially. The reason might be that multi-platform customers are more loyal and thus are more engaged with the online retailer. Hence, they will try to interact with the firm and engage more true voluntary CICs.
39
6.3 Discussion of control variables
Several control variables were also included in this research. All these variables are highly significant and will now be discussed shortly. Events have a positive estimate of 3.4353 indicating that when there is a special promotion average spend increases by the
aforementioned estimate. Gender has a negative estimate of 5.3903 indicating that women have a substantial lower average spend than men. A possible argument for this result might be that men buy less, but more valuable items such as household appliances, where women buy more, however less valuable items such as fashion items. Membership starting year has a negative estimate of -.9633 indicating that the older the starting year, the higher the average spend. This is to be expected where members with an older starting year are most likely more loyal to the online retailer. Age also has a negative estimate of -.2919 indicating that the younger the customer, the higher the average spend is. Where the online retailer is targeting families it is likely that younger customers have a larger average spend.
6.4 Managerial implications
The results of this research provide many interesting insights for retailers and in particular online retailers. Moreover, it provides useful guidelines for successful marketing strategy planning. Firstly, evidence is provided that app customer spend more on average. This
indicates that a focus on app customers is strategically a solid option where they are relatively better customers than web customers. Even more so, retailers should try to focus on customers who buy on multiple platforms instead of a single platform. This study has therefore clearly showed that customer who buy on multiple platforms have a significantly higher average spend than single platform customers. This suggests an interesting remark. Retailers should not try to make their customer shift from web to app purchases. Instead they should try to make both their single platform web and app customers buy on the platforms they are not on. The most interesting group would be the single platform web users, which value could be raised the most by enabling them to buy on the app platform.
40 have a significantly higher average spend than single platform customers when using a CIC. Therefore customers who are already engaging in multi-platform buying should be targeted by CICs even more so than single platform customers.
Finally, this research has clearly shown that a focus on app customers by redirecting more traffic to them has a positive effect on average spending. It proves that a more extensive focus on the app over web is a successful business opportunity for online retailers. Online retailers should try to extract value from this finding by making appropriate app landing pages such as that customers stick to the app and will make sales.
6.5 Limitations and future research directions
Even though there was much data available, the data spanned a quite short period. Where the redirection was only in place for four months only short period conclusions could be drawn. The first limitation of this study is thus that results on the long-term can have very different conclusions also indicating room for future research.
The dataset used in this analysis was very skewed to the right resulting in non-normality. A different way to deal with this issue would be to use a log-scale. This is a large limitation of the study, where changes in the distribution could deliver substantially different results.
Where the independent variable MSY can be seen as an approximation for customer loyalty, there is no such variable in this analysis. It might very well possible that loyalty has such a strong effect that it will negate all other effects found in this research. Therefore this is a limitation of the study as well as an option for future research.
This analysis has focused on all product categories of the online retailer. There are, however, possibly large differences between product categories. This indicates rooms for further research into finding specific spending differences between product categories.
Many retailers do not only make use of an app and a website, however also often have brick-and-mortar stores. Future research might add a brick-brick-and-mortar store to the possible sale platforms to find if there are any difference in average spending levels.
41 focus was on buying customers, however it might also be interesting to see how many
customers did not make sales.
Only average spending was used as dependent variable. There are many other dependent variables which might be interesting for future research such as customer lifetime value and conversion rates. Moreover, as already mentioned in the discussion of the moderation effects, the customer journey of buying a product is an extensive and complicated one. Variables such as app quality, quality of landing pages and website/app ease of usage might all have
42
7 REFERENCES
Ailawadi, K. L., & Farris, P. W. (2017). Managing Multi-and Omni-Channel Distribution: Metrics and Research Directions. Journal of Retailing, 93(1), 120-135.
Appiah-Adu, K., & Singh, S. (1998). Customer orientation and performance: a study of SMEs. Management decision, 36(6), 385-394.
Ash, T., Ginty, M., & Page, R. (2012). Landing page optimization: The definitive guide to
testing and tuning for conversions. John Wiley & Sons.
Blattberg, R. C., Kim, B. D., & Neslin, S. A. (2008). Why database marketing? Database
Marketing, 13-46 Springer New York.
Breuer, R., Brettel, M., & Engelen, A. (2011). Incorporating long-term effects in determining the effectiveness of different types of online advertising. Marketing Letters, 22(4), 327-340. BusinessofApps (2017, September 11), ‘’App Revenues 2017,’’ Retrieved from:
http://www.businessofapps.com/data/app-revenues/.
De Haan, E., Wiesel, T., & Pauwels, K. (2016). The effectiveness of different forms of online advertising for purchase conversion in a multiple-channel attribution
framework. International Journal of Research in Marketing, 33(3), 491-507.
Dinner, I. M., H. J. Van Heerde, and S. Neslin (2015). “Creating customer engagement via mobile apps: How app usage drives purchase behavior.” Working Paper, University of North Carolina at Chapel Hill.
Einav, L., Levin, J., Popov, I., & Sundaresan, N. (2014). Growth, adoption, and use of mobile E-commerce. The American economic review, 104(5), 489-494.
Ghose, A., & Todri, V. (2016). Towards a Digital Attribution Model: Measuring the Impact of Display Advertising on Online Consumer Behavior. MIS Quarterly, 40(4), 1-40.
Ghose, A., & Yang, S. (2009). An empirical analysis of search engine advertising: Sponsored search in electronic markets. Management Science, 55(10), 1605-1622.
43 Gu, X., & Kannan, P. K. (2017, March). THE IMPACT OF MOBILE APP ADOPTION ON CONSUMER PURCHASE. In ROBERT MITTELSTAEDT DOCTORAL SYMPOSIUM, 95-124.
Helgesen, Ø. (2006). Are loyal customers profitable? Customer satisfaction, customer (action) loyalty and customer profitability at the individual level. Journal of Marketing
Management, 22(3-4), 245-266.
Huang, L., X. Lu, and S. Ba (2016). “An empirical study of the cross-channel effects between web and mobile shopping channels.” Information & Management, 53 (2), 265–278.
International Data Corporation (2016, May 08), ‘’Mobile App Revenue Outlook Remains Healthy Despite Slowing Download Volumes and Smartphone Growth, According to IDC,’’ Retrieved from: https://www.idc.com/getdoc.jsp?containerId=prUS41240816.
Kanagal, N. B. (2015). Innovation and product innovation in marketing strategy. Journal of
Management and marketing research, 18, 1-25.
Kim, J., & Ah Yu, E. (2016). The holistic brand experience of branded mobile applications affects brand loyalty. Social Behavior and Personality: an international journal, 44(1), 77-87. Kim, S. J., R. J.-H. Wang, and E. C. Malthouse (2015). “The effects of adopting and using a brand’s mobile application on customers’ subsequent purchase behavior.” Journal of
Interactive Marketing 31, 28–41.
Konuş, U., Verhoef, P. C., & Neslin, S. A. (2008). Multichannel shopper segments and their covariates. Journal of Retailing, 84(4), 398-413.
Ktoridou, D., Epaminonda, E., & Kaufmann, H. R. (2008). Technological challenges and consumer perceptions of the use of mobile marketing: Evidence from Cyprus. International
Journal of Mobile Marketing, 3(2), 34-43.
Kumar, V., & Venkatesan, R. (2005). Who are the multichannel shoppers and how do they perform?: Correlates of multichannel shopping behavior. Journal of Interactive
marketing, 19(2), 44-62.
44 Neslin, S. A., & Shankar, V. (2009). Key issues in multichannel customer management: current knowledge and future directions. Journal of interactive marketing, 23(1), 70-81. Pauwels, K., & Neslin, S. A. (2015). Building with bricks and mortar: The revenue impact of opening physical stores in a multichannel environment. Journal of Retailing, 91(2), 182-197. Peng, K. F., Chen, Y., & Wen, K. W. (2014). Brand relationship, consumption values and branded app adoption. Industrial Management & Data Systems, 114(8), 1131-1143. Sarner, A., & Herschel, G. (2008). A checklist for evaluating an inbound and outbound multichannel campaign management application. Report G00160776, Gartner, Stamford, CT. Shankar, V. and S. Balasubramanian (2009). “Mobile marketing: a synthesis and prognosis.”
Journal of Interactive Marketing 23 (2), 118–129.
Shankar, V., & Malthouse, E. C. (2007). The growth of interactions and dialogs in interactive marketing. Journal of interactive Marketing, 21(2), 2-4.
Shankar, V., A. Venkatesh, C. Hofacker, and P. Naik (2010). “Mobile marketing in the retailing environment: Current insights and future research avenues.” Journal of interactive
marketing 24 (2), 111–120.
Statista (2017), “Worldwide mobile app revenues in 2015, 2016 and 2020 (in billion U.S. dollars),” Retrieved from: https://www.statista.com/statistics/269025/worldwide-mobile-app-revenue-forecast/.
Van Baal, S., & Dach, C. (2005). Free riding and customer retention across retailers’ channels. Journal of Interactive Marketing, 19(2), 75-85.
Wallace, D. W., Giese, J. L., & Johnson, J. L. (2004). Customer retailer loyalty in the context of multiple channel strategies. Journal of retailing, 80(4), 249-263.
45
8 APPENDIX
8.1 Appendix 1: Marketing Contact Variable Operationalization
Online contact
Description Contact type
Direct load When a customer accesses the website or app directly either by entering it through typing the URL in the address bar or by opening the app.
Customer-initiated
Paid Search (SEA)
When a customer searches for a keyword in a search engine and goes to the website or app through sponsored/paid search results (SEA)
Customer-initiated
Natural Search (SEO)
When a customer searches for a keyword in a search engine and goes to the website or app through results ranked by search algorithms (SEO)
Customer-initiated
Price comparison
Price comparison websites are search engines that customers can use to compare products.
Customer-initiated Display Also named banner advertising. This enables companies to
pose graphical images on websites with an advertising message.
Firm-initiated
Email E-mail marketing is about sending marketing messages to customers using e-mail.
Firm-initiated Affiliate The affiliate is rewarded for referring a user to the website
based on a commission.
Customer- initiated/firm-initiated Referral Referral covers all traffic that is forwarded by external
content.
46
47
48
8.4 Appendix 4: R-script
###Libraries and Data reading###
library(haven) library(Hmisc) library(dplyr) library(ggplot2) library(MatchIt) library(cem) library(car) library(stats) library(lmtest) library(gvlma)
WORK_Final <- read_sav("C:/Users/Jelle van Iersel/Downloads/Final.sav") View(WORK_Final) ##Pre Data### ##Cleaning Age### WORK_Final$LEEFTIJD_JAREN_AANTAL[WORK_Final$LEEFTIJD_JAREN_AANTA L > 80] <- NA WORK_Final$LEEFTIJD_JAREN_AANTAL[WORK_Final$LEEFTIJD_JAREN_AANTA L < 16] <- NA
###Create AVG Spend variable###
WORK_Final2$AVG_Spending <-
WORK_Final2$SUM_of_SUM_of_BRUTO_VRAAG_BEDRAG / WORK_Final2$COUNT_DISTINCT_of_ORDER_NR
###First analysis to see some pre-effects
49 summary(fit)
##Sub-Sample
CR_Thesis <- WORK_Final[sample(1:nrow(WORK_Final),285000, replace=FALSE),]
CR_Thesis %>%
group_by(DummyR) %>%
summarise (KLANT_RELATIE_ID = n(), mean_math = mean(AVG_Spending),
std_error = sd(AVG_Spending) / sqrt(KLANT_RELATIE_ID))
with(CR_Thesis, t.test(AVG_Spending ~ DummyR))
##PRE DATA COVARIATES###
My_Thesis_cov2 <- c('Display', 'Aantal', 'FIC_CIC', 'Events', 'Gender', 'START_JAAR', 'LEEFTIJD_JAREN_AANTAL')
CR_Thesis %>%
group_by(DummyR) %>%
select(one_of(My_Thesis_cov2)) %>% summarise_all(funs(mean(., na.rm = T)))
### T-test to evaluate if they are statiscally distinguishable###
lapply(My_Thesis_cov2, function(v){
t.test(My_Thesis[, v] ~ My_Thesis$Dummy_Datum) })
50 with(CR_Thesis, t.test(Gender ~ DummyR))
with(CR_Thesis, t.test(START_JAAR ~ DummyR))
with(CR_Thesis, t.test(LEEFTIJD_JAREN_AANTAL ~ DummyR))
###Estimating Model###
Thesis_PSM2 <- glm (DummyR ~ Display + Aantal + FIC_CIC + Events + Gender + START_JAAR + LEEFTIJD_JAREN_AANTAL, family = binomial(), data = CR_Thesis) summary(Thesis_PSM2)
###Propensity score###
Score <- data.frame(pr_score = predict(Thesis_PSM2, type = "response"), DummyR = Thesis_PSM2$model$DummyR)
###Histogram PSM###
labs <- paste("Dummy:", c("before", "after")) Score %>%
mutate(DummyR = ifelse(DummyR == 1, labs [1], labs [2])) %>% ggplot(aes(x = pr_score)) +
geom_histogram(color = "white") + facet_wrap(~DummyR) +
xlab("Probability of observing the treatment") + theme_bw()
###Matching Algorithm###
Thesis_Match2 <- CR_Thesis %>%
