• No results found

The Online Consumer Purchase Journey:

N/A
N/A
Protected

Academic year: 2021

Share "The Online Consumer Purchase Journey:"

Copied!
117
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

1

The Online Consumer Purchase Journey:

The Importance of Consumers’ Online Information Search and the Influential

Power of a Company’s Online Marketing Activities

Ires Diane Verdoorn

(2)
(3)

3

The Online Consumer Purchase Journey:

The Importance of Consumers’ Online Information Search and the Influential

Power of a Company’s Online Marketing Activities

Ires Diane Verdoorn University of Groningen Faculty of Economics and Business

MSc Marketing Intelligence and Marketing Management Master Thesis

14-01-2019

Ires Diane Verdoorn Raadhuisplein 170 9203EC Drachten The Netherlands (+31) 6 53 50 27 55 i.d.verdoorn@student.rug.nl S2538881 First Supervisor dr. Peter S. van Eck

p.s.van.eck@rug.nl

Second Supervisor dr. Abhi Bhattacharya abhi.bhattacharya@rug.nl

University of Groningen Faculty of Economics and Business

Department of Marketing PO Box 800

(4)
(5)

5 Management Summary

Given the easy and constant access to the internet, a cognitive change has occurred. Consumers no longer rely on the information they have stored in their own minds. Rather, consumers rely on the information they can retrieve from search engines to make a purchase decision. As a result, consumers’ online information search is becoming an important predictor of consumers’ online conversion, which creates the need to get a better understanding of the influential power of consumers’ information search in consumers’ purchase journeys.

Consumers are commonly engaged in either a goal-directed or an exploratory information search. A goal-directed information search is concerned with acquiring relevant information about a specific product a consumer is already considering to buy. An exploratory information search is concerned with acquiring information about a less well-defined product category. This search is known as a knowledge building process, as consumers are not necessarily concerned with immediately using the information to end the search with a purchase. Most of the research concerned with consumers’ information search has failed to distinguish between consumers’ divergent online informational needs. Hence, little is known about the impact of these different types of information search on consumer conversion, and, more importantly, how a company’s marketing activities should be utilized to reach these consumers effectively. This study will fill these research gaps by providing an answer to the following research questions: (1) What is the effect of consumers’ online information search on online conversion?, (2) What is the effect of a company’s online marketing activities (i.e. firm-initiated contacts) on consumers’ online information search?, (3) What is the effect of a company’s online marketing activities (i.e. firm-initiated contacts) on the relationship between consumers’ online information search and online conversion?, and (4) Do the aforementioned effects differ depending on the type of online information search?

(6)

6 consumers’ information search (both goal-directed and exploratory). The effect of display and affiliate advertising is expected to be stronger for consumers’ exploratory search, and email advertising is hypothesized to have a stronger effect on consumers’ goal-directed search. The initial focus of this research is on the direct effect of company’s marketing activities as indicated by the formulated hypotheses. However, to thoroughly study the effect of a company’s marketing activities, both the short-term and long-term advertisement effects were studied. Event-based, online purchase journey data of a Dutch travel agency was analyzed in order to provide an answer to the research questions.

The research findings showed that consumers’ purchase decision is positively influenced by consumers’ information search (goal-directed and exploratory), with the effect being stronger for consumers’ who perform exploratory information searches. Furthermore, only positive long-term advertisement effects were found between display advertisement and exploratory information search, and between affiliate advertisement and goal-directed search. No statistical results were found for the other hypothesized relations.

The study results indicate that consumers’ information search significantly influences consumers’ decision to convert or not. However, companies need to be aware of the different types of information search as each advertisement activity (display, affiliate, email) has a different effect depending on the type of information search. Marketing managers are recommended to devote more of the marketing budget to optimize its display and affiliate advertising practices as these two marketing activities are proven to be most effective. Given the research findings, marketing managers are also advised to retarget consumers engaged in an exploratory search with display advertisement, and use affiliate advertisement to retarget consumers with a goal-directed search intent to increase the effectiveness of its retargeting practices and subsequently increase consumer conversion. Marketing managers need to be aware of the fact that their marketing activities might not immediately provide the desired results, but will only lead to positive effects in the long-run as shown by the study findings. By optimizing its marketing practices as suggested by this research, a company will be more likely to increase its return on investments and increase customer conversion.

(7)

7 Preface

This master thesis is the final step in order to graduate as Master of Science in Marketing Intelligence and Marketing Management at the University of Groningen. The completion of my Master of Marketing also marks the end of my student time at the University of Groningen, and signals the beginning of my new, professional life, with hopefully a lot of new experiences and adventures along the way.

Upon completion of my Bachelor of International Business in 2016, I still had no idea in what field of business I wanted to pursuit a career in. Given my broad interest and eagerness to learn new things, I wondered in what area I could gain additional knowledge that would be useful for my professional career. During my gap year, in which I worked full-time at an interesting and fast growing organization, I discovered my enthusiasm for the field of (online) marketing and especially the data-driven, analytical side and I knew that the master programs of Marketing Intelligence and Marketing Management at the University of Groningen would be a perfect fit for me.

Over the course of the last one and a half year, I participated in both programs and it has brought me a lot of new insights and knowledge for which I want to thank the professors of the faculty of Economic and Business and the University of Groningen. Writing my master thesis was a great and challenging experience, where my skills and knowledge were tested, but where I also got the opportunity to extend my data analysis and academic working skills even further. The writing of my master thesis was a long and intensive process, and I’m proud to present you my final work.

During this long and intensive process of writing my master thesis, I was assisted by my first supervisor dr. Peter van Eck. I want to thank him for his support, suggestions, and contributions which have helped me tremendously in the completion of my master thesis. Furthermore, I want to extend my gratitude towards dr. Abhi Bhattacharya. I want to thank him for his time and effort for being my second supervisor.

Most importantly, I want to thank my partner Richard Beukema, my parents, sister and brother, family and friends, for always believing in me and providing me with the support and encouraging words I needed during my entire time studying at the University of Groningen. I could not have done it without them. Thank you.

Ires Verdoorn

(8)
(9)

9 Table of Content

1. Introduction... p.14

2. Theoretical Framework... p.17 2.1 The Online Consumer Purchase Journey………... ……... p.17 2.2 Consumers’ Online Information Search……… p.18 2.3 The Relation Between Consumers’ Online Information Search & Conversion……… p.18 2.3.1 Goal-Directed Information Search………. p.18 2.3.2 Exploratory Information Search………. p.19 2.4 Firm-Initiated Contacts………... p.20 2.4.1 Display Advertising……… p.20 2.4.2 Affiliate Advertising……… p.22 2.4.3 Email Advertising………... p.23 2.5 The Moderating Role of Retargeting……… p.24 2.6 Conceptual Model……… p.25

3. Methodology……….. p.26 3.1 Data……… p.26 3.2 Variables……… p.26

(10)

10 3.4.1 Logistic Regression Models……… p.31 3.5 Plan of Analysis………. p.33

4. Results………. p.34 4.1 Preliminary Checks………... p.34 4.2 Preparing the Datasets………... p.35 4.3 The Relation Between Consumers’ Online Information Search & Conversion……… p.36 4.3.1 Descriptive Statistics & Data Exploration………. p.36 4.3.2 Multicollinearity……….. p.37 4.3.3 Model Selection………..………. p.38 4.3.4 Hypotheses Testing…………..………. p.41 4.4 The Relation Between Firm-Initiated Contacts & Goal-Directed Information Search. p.45 4.4.1 Descriptive Statistics & Data Exploration………. p.45 4.4.2 Multicollinearity……….. p.46 4.4.3 Model Selection………..………. p.46 4.4.4 Hypotheses Testing…………..……… p.47 4.5 The Relation Between Firm-Initiated Contacts & Exploratory Information Search…. p.48 4.5.1 Descriptive Statistics & Data Exploration………. p.48 4.5.2 Multicollinearity……….. p.49 4.5.3 Model Selection………..………. p.50 4.5.4 Hypotheses Testing…………..………. p.51 4.6 Validation of the Research Results……… p.52

(11)

11 5.2 Firm-Initiated Contacts……….. p.56 5.3 Theoretical & Managerial Implications………. p.59

6. Limitations & Future Research………... p.60

(12)
(13)
(14)

14 1. Introduction

‘’The digital revolution has launched a new era of human empowerment and engagement across business, society, and in every aspect of our lives. Never before has there been a more powerful influence on human behavior.’’

- Mervyn Eyre (Leading in a Digital World, 2017) Digitalization, and especially the rise of the Internet, has caused a fundamental change in the way consumers experience and create their purchase journeys (Ghose & Yang, 2009; Verhoef, Kannan, & Inman, 2015). As a result, online touchpoints are becoming more important in shaping consumers path to purchase (Verhoef, Kannan, & Inman, 2015). The Internet enables consumers to be more in control of their purchase journeys (Lemon, 2016; Pires, Stanton, & Rita, 2006), especially since the Internet provides consumers with the opportunity to instantly search for and acquire the desired information without having to wait for any firm-initiated contact (Klein & Ford, 2003; Li & Kannan, 2014). As a result, consumers are increasingly starting their purchase decision process online (Lecinski, 2011). The increased control of consumers over the kind of channels they use or media they consume, has greatly diminished a company’s ability to ‘’push’’ any marketing activity to consumers (Lemon, 2016). Consumers own initiated contacts are becoming more important and influential in a consumer’s path to purchase (Anderl, Schumann, & Kunz, 2016; Li & Kannan, 2014).

(15)

15 the purchase journey, and how a company’s marketing activities are able to influence consumers’ information search in order to increase company sales.

Previous research provides support for the importance of consumers’ online information search in guiding a consumer towards making a purchase (Rose & Samouel, 2009; Shim et al., 2001). However, consumers’ information search is not neutral, as consumers vary with regard to the type of information search they employ online (Klapdor, 2013). Consumers are usually engaged in either a goal-directed or an exploratory online information search (Moe, 2003). However, most research relating to consumers’ information search failed to distinguish between consumers’ divergent online informational needs. Hence, little is known about the impact of these different types of information search on consumer conversion (Shim et al., 2001). Neglecting to account for these different types of online information search results in an inaccurate representation of the effect of consumers information search as predictor of conversion, especially since the way consumers search for information impacts their purchase propensity (Roscoe et al., 2016). Additionally, it does not contribute much to companies’ understanding of consumers’ information search and how their marketing activities need to be utilized in order to effectively reach consumers. This information is important, as consumers decision to search for information and hence their likelihood of converting may be influenced by a company’s marketing activities (Roscoe et al., 2016). To the best of the authors knowledge, current research is unable to provide meaningful insights on these topics. Given the growing importance of consumers’ information search as a key predictor of conversion (Shim et al., 2001), there is a need for additional insights in the aforementioned research areas. The present research will fill these research gaps by assessing whether consumers’ purchase probability varies depending on the type of information search employed online, and to what extent a company’s marketing activites are able to influence consumers’ information search in order to increase company sales. The aim of this research is to provide an answer to the following research questions:

1. What is the effect of consumers’ online information search on online conversion?

2. What is the effect of a company’s online marketing activities (i.e. firm-initiated contacts) on consumers’ online information search?

(16)

16 Answering these research questions should result in a deeper understanding of how consumers’ online information search affects consumers online conversion propensity. Additionally, a deeper understanding on how a company’s online marketing activities are able to influence consumers’ online information search result, and whether a company’s online marketing efforts are able to influence consumers’ online conversion. Event based, online purchase journey data of a Dutch travel agency is analyzed to reach this goal. This research offers a unique contribution to existing literature as it will provide new quantitative insights on the drivers of consumers’ information search, and offers a more in-depth understanding of the influence of consumers’ information search on online conversion. Hence, this paper distinguishes itself from prior work as this research is one of the first to distinguish between different types of online information search in order to assess the influence of consumers’ information search on conversion. Furthermore, a more in-depth understanding of the workings of a company’s marketing activities in relation to consumers’ divergent types of information search will result, which is currently missing in contemporary research (Moe, 2013). Overall, this research will be an enrichment for current research related to consumers’ purchase journeys.

The research findings also have significant practical relevance. The resulting insights provide marketing managers with an in-depth understanding of the workings of their marketing activities. As a result, more efficient marketing strategies can be created to ensure that consumers are effectively reached and targeted in the early stages of their purchase decision making process (Hoffman & Novak, 2000). Additionally, based on the research results, the allocation of the marketing budget can be optimized which creates a greater return on investment for the company (Anderl, Schumann, et al., 2016).

(17)

17 2. Theoretical Framework

2.1 The Online Consumer Purchase Journey

Creating consumer conversion is an essential business objective for many commercial firms (Jasmand, Blazevic, & de Ruyter, 2012), which results in the need for a better understanding of the path consumers take to a purchase (Lemon & Verhoef, 2016). A common understanding of consumers’ path to purchase concerns the idea that consumers proceed through a series of stages during their purchase journey in which they interact with a firm through a myriad of touchpoints and channels (Følstad & Kvale, 2018; Lemon & Verhoef, 2016; Srinivasan, Rutz, & Pauwels, 2016). A(n) (online) consumer purchase journey thus includes all contacts an individual consumer has with a firm over all (online) marketing channels, prior to a potential purchase (Anderl, et al., 2016). These contacts can be considered the building blocks of the purchase journey, especially since each touchpoint has an important influence on a consumer’s decision whether to convert or not (Følstad & Kvale, 2018).

Academic literature distinguishes between firm-initiated contacts and customer-initiated contacts (Anderl, Becker, et al., 2016; Li & Kannan, 2014). Firm-customer-initiated contacts (FICs) are firm-initiated marketing communications towards the consumer (Li & Kannan, 2014). These contacts are a firm initiated attempt to push a marketing message to consumers, and are usually considered intrusive and unwanted (Anderl, Becker, et al., 2016; Haan, Wiesel, & Pauwels, 2016). On the contrary, customer-initiated contacts (CICs) are triggered by the consumers’ own initiative, and are thus considered any communication with a company that is initiated by consumers themselves (Bowman & Narayandas, 2001; Haan et al., 2016). Since CICs are stemming from consumers own initiative, CICs are considered less intrusive and hence more effective in leading a consumer to a conversion (Haan et al., 2016; Li & Kannan, 2014).

(18)

18 2.2 Consumers’ Online Information Search

Today’s digital economy is driven by information technology and e-commerce (Zhao, Wallis, & Singh, 2015), which makes the Internet an important aspect of consumers’ everyday lives (Sparrow & Chatman, 2013). Research results show that the seamless access to information on the Internet caused a significant cognitive change with regard to the way consumers use and rely on their own memory (Sparrow et al., 2011; Sparrow & Chatman, 2013). Consumers no longer have to make costly efforts to store needed information, as they can simply ‘’Google’’ it (Lecinski, 2011; Sparrow et al., 2011). Consumers are treating the internet as a primary form of external memory (Sparrow et al., 2011). As a result, consumers are in a decreasing fashion using and trusting their own memory and are increasingly relying on the Internet to assist them in their informational needs (Sparrow & Chatman, 2013). This cognitive change explains why information search as part of the consumer purchase journey is becoming more important in guiding a consumer towards a purchase (Shim et al., 2001).

The way consumers search for information online differs with regard to consumers’ intentions of the information search (Klapdor, 2013). Academic literature distinguishes between two types of online information search, which this research also adheres to. On the one hand, consumers’ online information search can be classified as a goal-directed search and on the other hand consumers can be engaged in an exploratory information search (Moe, 2003; Novak, Hoffman, & Duhachek, 2003; Wolfinbarger & Gilly, 2000). An elaborated discussion on the two search types and their relation to conversion will be provided subsequently.

2.3 The Relation Between Consumers’ Online Information Search & Conversion

Previous research showed that a consumer’s online information search positively enhances ones online purchase intentions, which makes information search an important element in leading a consumer to convert online (Shim et al., 2001). However, consumers vary in the type of information search they employ online, which makes it plausible to assume that consumers conversion propensity differs with regard to the type of information search employed online (Kireyev et al., 2016; Moe, 2003; Shim et al., 2001).

2.3.1 Goal-Directed Information Search

(19)

19 about a specific product or product category a consumer is already considering or planning to buy (Moe, 2003; Wolfinbarger & Gilly, 2001). Consumers reach out to the internet in order to acquire relevant information that will help them to make a more optimal product choice (Moe, 2003; Wolfinbarger & Gilly, 2001). As a result, consumers’ online information search is highly deliberate and extremely focused as the search is goal-driven (Moe, 2003; Novak et al., 2003; Wolfinbarger & Gilly, 2000, 2001). Since consumers are already actively searching for information about a particular object, a goal-directed information search can be interpreted as an indicator of consumer interest (Hu et al., 2014), and hence of conversion (Shim et al., 2001; Zigmond & Stipp, 2010). Furthermore, research results of Moe (2003) show that goal-directed information seekers have the highest conversion rate because these consumers already specifically know what they want. Additionally, Agarwal, Hosanagar, and Smith (2011) reveal that consumers who use specific search queries in their online information search, have a higher chance of converting. Thus, consumers’ deliberate and focused information search has a positive influence on consumers’ purchase probabilities (Agarwal, Hosanagar, & Smith, 2011). Based on the aforementioned research findings, the following hypothesis is developed:

H1a: Consumers’ goal-directed online information search has a positive influence on online conversion.

2.3.2 Exploratory Information Search

(20)

20 Even though the motivation of consumers engaged in an exploratory information search is not necessarily to convert, previous research findings reveal that an exploratory search can also result in a conversion (Dickinger & Stangl, 2011; Moe, 2003). Since an exploratory information search is considered stimulus-driven (Moe, 2003; D. N. Smith & Sivakumar, 2004), the right stimulus encountered during the information search, such as a positive consumer review, can direct a consumer to a(n) (impulse) purchase (Ludwig et al., 2013; Mandel & Johnson, 2002; Moe, 2003). Furthermore, the acquired product information can be found useful for future purchase decisions (Moe, 2003). This explains why a purchase may not immediately result, but may happen sometime in the future (Moe, 2003). Previous research thus indicates that consumers’ exploratory information search has a positive influence on consumers’ conversion propensity (Moe, 2003; D. N. Smith & Sivakumar, 2004). However, the chance of converting can be considered lower for consumers engaged in an exploratory information search compared to consumers engaged in a goal-directed information search (Moe, 2003; D. N. Smith & Sivakumar, 2004). As such, the following hypotheses can be formulated:

H1b: Consumers’ exploratory online information search has a positive influence on online conversion.

H1c: The positive influence on online conversion will be stronger for consumers’ goal-directed online information search than for consumers’ exploratory online information search.

2.4 Firm-Initiated Contacts

In the era where pre-purchase online information search is becoming the norm, firms’ online advertising activities have to be sculpted in such a way that consumers’ need for information search is effectively being triggered (Hu, Du, & Damangir, 2014). Several online advertising strategies can be used to stimulate consumers’ informational needs such that consumers will immediately search for information, or might perform a search query later in the purchase journey based on a previously encountered advertisement (Li & Kannan, 2014).

2.4.1 Display Advertising

(21)

21 positively influence a consumer’s search applications (Kireyev, Pauwels, & Gupta, 2016; Li & Kannan, 2014; Papadimitriou et al., 2011), with the effect being stronger for an information search relating to the advertised product or product category (Lewis & Nguyen, 2014; Papadimitriou et al., 2011). Viewing the display ad draws a consumer’s attention to the displayed product or product category, which increases the overall interest in that particular product category (Lewis & Nguyen, 2014). As a result, a consumer’s need to search for information with regard to the displayed product category is activated (Lewis & Nguyen, 2014). Previous research findings provide evidence for a positive influence of display advertising on consumers’ information search activities (Kireyev, Pauwels, & Gupta, 2016; Li & Kannan, 2014; Lewis & Nguyen, 2014; Papadimitriou et al., 2011). However, no research specifically considers the influence of display advertising on either a goal-directed information search or exploratory information search. Yet, based on the aforementioned research findings, it can be argued that display advertising will also have a positive influence on consumers’ goal-directed information search and consumers’ exploratory information search.

A consumer engaged in a goal-directed information search is concerned with acquiring information with regard to a predetermined product category (Dickinger & Stangl, 2011; Moe, 2003). If the ad displays a product that fits within this predetermined product category, consumers will be influenced by the ad and hence subsequently search for information (Lewis & Nguyen, 2014). However, the displayed product needs to fit exactly with the predetermined product category, otherwise consumers will not even notice let alone be influenced by the display ad (Moe, 2003). On the contrary, consumers engaged in an exploratory information search are concerned with acquiring relevant information and building their knowledge base with regard to a less well-defined product group (Dickinger & Stangl, 2011; Moe, 2003). As such, the information being searched for encompasses a broader product segment which makes it more likely that a displayed ad will fit within the interest of a consumer (Dickinger & Stangl, 2011; Moe, 2003). Based on the above, the following hypotheses are formulated:

H2a: Display advertising has a positive influence on consumers’ goal-directed online information search.

H2b: Display advertising has a positive influence on consumers’ exploratory online information search.

(22)

22

2.4.2 Affiliate Advertising

Affiliate advertising consists of a partnership between two organizations in which companies are allowed to place an ad, blog post, or link of their own website on their partner’s website (Anderl, Schumann, et al., 2016; Edelman & Brandi, 2015). The success of affiliate marketing, or any marketing activity for that matter, mainly depends on a consumer’s attitude towards the activity’s perceived usefulness (Andeleeb, 1996; Pauwels, 2004; Ul Haq, 2012). Previous research reveals that the perceived usefulness of an affiliate link is the strongest factor that influences consumers to click on the link (Ul Haq, 2012).

Consumers’ click-through rate can be perceived as an indicator of consumers’ interest about the advertised product(s) (Gauzente, 2010; Parsons, 2002). Consumers clicking on an affiliate link can be said to have an interest to search for additional information about the advertised product. No previous research has focused on the effect of affiliate advertising on consumers’ information search (Dwivedi, Rana, & Alryalat, 2017). Therefore, the positive effect of affiliate advertising on consumers’ click-through rates can be used to argue that affiliate advertising might also have a positive influence on consumers’ information search.

Consumers with a goal-directed search intent are searching for specific information with regard to one predetermined product category (Moe, 2003; Wolfinbarger & Gilly, 2001). Given a consumers narrow product interest, the chance that the affiliate link fits within the predetermined product category and hence a consumer clicking on the link is small, but not unlikely. On the contrary, consumers performing an exploratory information search are interested in acquiring relevant information with regard to a broader product range (Moe, 2003; D. N. Smith & Sivakumar, 2004). As a result, the affiliate link has a higher chance to be perceived as interesting, resulting in a higher chance for consumers to click on the link and continue their information search (Dickinger & Stangl, 2011; Moe, 2003). Based on the aforementioned results, the following hypotheses are formulated:

H3a: Affiliate advertising has a positive influence on consumers’ goal-directed online information search.

H3b: Affiliate advertising has a positive influence on consumers’ exploratory online information search.

(23)

23

2.4.3 Email Advertising

Email advertising is one of the most inexpensive marketing activities with a high response rate (Bucklin & Sismeiro, 2009; Dreze, 2006). It consists of a firm sending a promotional email to consumers including text or a link to the company’s website (Anderl, Schumann, et al., 2016). Consumers have provided the company with their email address and hence have given permission to the company to send promotional emails (Bucklin & Sismeiro, 2009). A major incentive for consumers to sign up for a firm’s emailing list, is the prospect of receiving information that is in line with their interests (Ellis-Chadwick & Doherty, 2012).

Previous research shows that email advertising positively stimulates consumers’ information search (Haan, Wiesel, & Pauwels, 2016; Li & Kannan, 2014), especially since the content of the email usually matches with the consumers’ interest. The content of the email hence triggers curiosity which stimulates consumers to search for information with regard to the content of the email (Ellis-Chadwick & Doherty, 2012). However, there is a lack of research that studies the relationship between email advertising and consumers’ online information search (Haan, Wiesel, & Pauwels, 2016). Moreover, no research exists that specifically investigates the relationship between email marketing and either type of information search. Since previous research reports positive results of the relationship between email advertising and consumers’ information search (Haan, Wiesel, & Pauwels, 2016; Hartemo, 2016; Li & Kannan, 2014), this research will therefore use these study results to build the theoretical foundation of the effect of email advertising on both types of information searches. Furthermore, according to the findings of Haan, Wiesel, and Pauwels (2016), most consumers only react to an email when they have already decided what to buy and are reminded about these purchase intentions by the email. This indicates that email advertising will have a stronger effect on consumers’ goal-directed information search, as consumers of this type of search already have a predetermined product category in mind (Haan et al., 2016; Lambrecht & Tucker, 2013). Based on the above, this research proposes the following hypotheses.

H4a: Email advertising has a positive influence on consumers’ goal-directed online information search.

H4b: Email advertising has a positive influence on consumers’ exploratory online information search.

(24)

24 2.5 The Moderating Role of Retargeting

During a consumers’ online information search, a variety of websites and products are being viewed which all contributes to a consumers’ digital footprint (Nicholas et al., 2008). Retargeting is a marketing method in which organizations use consumers’ recent online browsing data to personalize their ads and hopefully increase sales (Lambrecht & Tucker, 2013). Previous research shows that personalized retargeted ads are more effective in activating a consumer response, as these ads are highly relevant and in accordance with consumers’ current needs (Bleier & Eisenbeiss, 2015). Since not all online purchase journeys immediately end with a conversion, research shows that retargeting can serve as an effective means to stimulate consumers to make the purchase after all (Ansari & Mela, 2003).

According to Lambrecht & Tucker (2013), the effectiveness of a retargeted ad depends on whether a consumer has a detailed viewpoint of the kind of products he or she is wanting to purchase. The authors argue that consumers with a well-defined product preference have a higher chance of positively responding to such highly specific advertisement (Lambrecht & Tucker, 2013). Furthermore, study results reveal that the effectiveness of a retargeted activity is stronger for consumers who are already closer to the decision of making a purchase (Kumar, Venkatesan, & Reinartz, 2008). These research findings suggest that retargeting has a stronger effect on consumers who are engaged in a goal-directed information search as these consumers already have a purchase intention in mind (Moe, 2003). The specific ads of retargeting are assumed to be less effective in increasing the purchase probability of consumers engaged in an exploratory search since these consumers are still in a knowledge building phase and have a less well-defined product category in which they are interested in. However, the content or type of retargeting activity used by a company might sometimes trigger an impulse purchase which still renders retargeting effective for consumers’ exploratory information search (Kumar et al., 2008). Considering the above, the following moderating influence of retargeting is expected:

H5a: Retargeting strengthens the relation between consumers’ goal-directed online information search and online conversion.

H5b: Retargeting strengthens the relation between consumers’ exploratory online information search and online conversion.

(25)

25 2.6 Conceptual Model

The previous derived hypotheses form the basis of the conceptual frameworks. Note that this research is concerned with analyzing the relationship between consumers’ information search and conversion, and the relationship between firm-initiated contacts and consumers’ information search. Therefore, in order to analyze these relationships and provide and answer to the above stated research questions, this research will develop separate models which will be elaborated on in subsequent sections. A graphical illustration of the conceptual frameworks used in this research can be seen in Figure 1 and 2.

Figure 1. Conceptual Framework Model 1.

Figure 2. Conceptual Framework Model 2.

H1c: The effect is stronger for H1a than for H1b H5c: The effect is stronger for H5a than for H5b

(26)

26 3. Methodology

3.1 Data

To test the hypotheses, event-based online purchase journey data obtained from a Dutch travel agency will be analyzed. The data is provided by Gfk, which is the fourth largest market research institute in the world (Gfk, 2018). Given Gfk’s well-known reputation and its high-quality performance, the data can be assumed to be of high-high-quality which is an important requirement for any data analysis (Leeflang et al., 2015). The data was collected by Gfk through GfK Crossmedia Link. By means of a browser plug-in, a panelist’s online browsing behavior was passively measured and registered on a personal level. Passive measurement made it possible to measure and collect data with regard to a panelist’s complete information chain (i.e. purchase journey). The dataset includes individual level data and spans 68 weeks from June 1, 2015 to September 9, 2016. The dataset consists of behavioral data and data on panelists’ demographics. For each panelist, the type of touchpoints (i.e. events) encountered during an online browse were measured and registered. The sequence of touchpoints represent a panelist’s individual purchase journey. A panelist can have several different purchase journeys, which may either end with a conversion or not. The touchpoints encountered were either customer-initiated or firm-customer-initiated. An overview of all the types of touchpoints included in the data can be seen in Appendix A. The dataset consists of 2456414 observations and includes information on 29012 purchase journeys of 9678 participants. Of the 29012 purchase journeys, 3674 end with a conversion of which 192 purchases were done at the focal travel agency. To test the hypotheses, information at the level of consumers’ individual purchase journeys is needed. The data needs to be restructured such that this kind of information can be extracted. Since this research incorporates three dependent variables, multiple datasets will be created. The way by which the dataset is restructured will be elaborated on in section 4.2.

3.2 Variables

3.2.1 Variables Conceptual Model 1

(27)

27 journey ends with a booking at the focal travel agency and ‘0’ if otherwise. By focusing on focal conversion, it can be examined if a company’s retargeting activities are able to increase consumers’ conversion probability and hence increase company sales. However, Focal Conversion is a rare event in the data as only 192 of the 29012 purchase journeys end with a conversion at the focal company. Consequently, the analysis outcomes may be insignificant (King & Langche, 2001). This is most likely due to the large deviation between focal company conversion and non-conversion instead of incorrect hypothesized relationships. If insignificant results are found, further testing will be done by including all purchase journeys that end with a conversion. The new dependent variable Conversion represents whether a purchase journey ends with a conversion (i.e. ‘1’) or not (i.e. ‘0’). The chance of acquiring significant results will increase as more observations to conversion will be available (Leeflang et al., 2015).

The two independent variables are consumers’ goal-directed information search and consumers’ exploratory information search. This study follows Moe (2003) by operationalizing both variables. According to Moe (2003), consumers performing a goal-directed information search mostly view product- and category level pages. Therefore, the customer-initiated touchpoints related to specific product pages (i.e. Accommodation Search, Touroperator/Travel Agent Search Focus Brand, and Flight Tickets Search) are used as a measure to represent the variable Goal-Directed Information Search. The variable represent the total number of goal-directed information searches performed within a consumer purchase journey. Furthermore, Moe (2003) states that consumers with an exploratory search intent mostly view informational pages. Therefore, the customer-initiated touchpoint related to informational web pages (i.e. Information/Comparison Search) is used to measure the variable Exploratory Information Search. This variable represent the total number of exploratory information searches performed within a consumer purchase journey. Note that the touchpoint Information/Comparison Search does not only include informational pages but also includes comparison web pages. Comparison web pages may serve a different function compared to purely informational pages. However, the effects cannot be separated thus the research results should be interpreted with care.

(28)

28 higher chance of persuading a consumer to convert (Montgomery et al., 2004). Therefore, this research will only include the retargeting touchpoints that have been encountered after either information search. As a result, a more accurate effect of retargeting can be analyzed. The variable Retargeting thus represents the total number of retargeting advertisement within a purchase journey that have been encountered after either information search. The Gfk dataset needs to be restructured in such a way that it only includes the aforementioned observations. The way the dataset is restructured will be explained in section 4.2. An overview of the main variables included in conceptual model 1 can be seen in Table 1.

Table 1. Operationalization of the variables as included in conceptual model 1.

Variables Operationalization Dependent Variable Focal Conversion

Conversion

Whether or not a consumer purchase journey ends with a booking at the focal travel agency

Whether or not a consumer purchase journey ends with a conversion

Independent Variables Goal-Directed Information Search

Exploratory Information Search

Total number of goal-directed

information searches performed within a consumer purchase journey

Total number of exploratory

information searches performed within a consumer purchase journey

Moderator Variable Retargeting Total number of retargeting advertisement within a purchase journey that are encountered after either information search.

3.2.2 Control Variables Conceptual Model 1

(29)

29 Table 2. Control variables for conceptual model 1 and academic reasoning.

Control Variable Academic Reasoning

Gender Gender differences exist with regard to online buying behavior: males value the utility of online shopping more than females and hence shop and purchase more online compared to females (Hasan, 2010; Rodgers & Harris, 2003).

Kind of Work ‘’Students spend more money online than any other demographic group. Students

spend hours surfing the Internet each day, and are among the most eager consumers to make online purchases’’ (Seock & Bailey, 2008, p. 113).

Age Education Income

Consumers who are younger, more educated, and who are wealthier have a higher change of buying online (Bellman, Lohse, & Johnson, 1999; Naseri & Elliott, 2011).

Region

Size of Municipality

The use of the Internet and online buying is still an urban phenomenon (Farag et al., 2006; Ren & Kwan, 2009).

3.2.3 Variables Conceptual Model 2

The second part of this research is concerned with analyzing the relationship between a company’s marketing activities and consumers’ information search (i.e. goal-directed and exploratory). The variables Goal-Directed Information Search and Exploratory Information Search are both treated as dependent variables in this model. Both variables are still measured by means of the touchpoints as explained in section 3.2.1. However, the variables are operationalized differently and are considered binary in this model. The reason for operationalizing the variables differently is because this model is interested in whether someone searches rather than how often. More specifically, this research is interested in finding out whether a consumer will perform a goal-directed or exploratory information search as a result of a company’s marketing activities. Therefore, a binary variable is needed to investigate this relationship. The dependent variables represent a ‘1’ if a goal-directed information search (exploratory information search) is performed in a purchase journey, and ‘0’ if otherwise.

(30)

30 research. Otherwise, the effect of each marketing activity on a subsequent goal-directed or exploratory search cannot be established. The initial focus of this study is on the direct or short-term effect of a company’s marketing activities on either information search as indicated by the formulated hypotheses. However, the researcher is aware of the fact that positive long-term advertising effect are also possible. A company’s advertising practices (display, affiliate, and email) can also trigger a search query later in a consumers’ purchase journey (Anderl et al., 2016; Li & Kannan, 2014). Therefore, to ensure that the effect of a company’s marketing activities on either information search is thoroughly studied, both the short-term and long-term effect of display, affiliate, and email advertisement will be analyzed. Two different datasets are needed, one to analyze the short-term advertisement effect and one for the long-term effect. The way the datasets are created will be explained in section 4.2. Table 3 provides an overview of the operationalization of the variables included in the model.

Table 3. Operationalization of the variables as included in conceptual model 2.

Variables Operationalization Dependent Variables Goal-Directed Information Search

Exploratory Information Search

Whether or not a goal-directed information search (exploratory information search) is performed within a purchase journey.

Independent Variables Display Advertising Affiliate Advertising Email Advertising

Total number of display advertisement (affiliate; email) within a purchase journey

encountered before a goal-directed or exploratory information search.

3.2.4 Control Variables Conceptual Model 2

(31)

31 Table 4. Control variables for conceptual model 2 and academic reasoning

Control Variable Academic Reasoning

Gender Gender differences exist with regard to online search behavior e.g.

 Men and women differ in web navigation behavior, with men being less engaged in exploratory search behavior (Richard et al., 2010).

 Females have a higher need for online information search and are more involved in online information search (Kim, Lehto, & Morrison, 2007).

Education Higher educated consumers are more inclined to search the web for product/service information (Bhatnagar & Ghose, 2004).

Region The use of the Internet and online information search is still an urban phenomenon (Farag et al., 2006; Ren & Kwan, 2009).

3.3 Choice of Technique

This research is considered to be both causal and quantitative (Malhotra, 2009), as it is concerned with statistically testing and analyzing the relationship between firm-initiated contacts and consumers’ information search, and the relationship between consumers’ information search and conversion. All dependent variables in this research are binary, which means that the same data analysis technique can be employed for all models. According to Leeflang et al., (2015, 2017), marketing problems with a binary response require either a binomial logit or probit model. The focus of such models is to estimate probabilities instead of observed values (Leeflang et al., 2015; Leeflang, Wieringa, Bijmolt, & Pauwels, 2017). Both logit and probit models provide similar research results (Leeflang et al., 2015, 2017). However, based on mathematical convenience a logit model or logisitic regression analysis is often preferred, which this research therefore also applies (Leeflang et al., 2015, 2017).

3.4 Model Specification

Separate models based on each dependent variable will be estimated and analyzed in order to test the hypotheses. The following section will specify the basic logistic regression models applied in this research.

3.4.1 Logistic Regression Models

(32)

32 Equation (3.5.1.1) displays the basic structure of a binary logit model, which serves as the basis for the specification of the logistic regression models used in this research.

𝜋𝑖 = 1

1 + exp (−{𝑥𝑖′𝛽})

The logistic regression model specification with dependent variable Focal Conversion can be seen in equation (3.5.1.2). To account for the interaction effect of Retargeting, the product of the variables that have an interaction effect need to be added to the model (Malhotra, 2009). Therefore, the interaction between the variables Goal-Directed Information Search and Retargeting, and Exploratory Information Search and Retargeting are included in the model in equation (3.5.1.2). Additionally, in order for the interaction effect to be effective, the main effect of Retargeting also need to be included in the model (Malhotra, 2009). The logistic regression model specifications with dependent variables Goal-Directed Information Search and Exploratory Information Search can be seen in equations (3.5.1.3) and (3.5.1.4) respectively. 𝐹𝐶𝑖= 1 1 + exp (− ( 𝛽0+ 𝛽1𝐺𝐷𝑆𝑖+ 𝛽2𝐸𝑋𝑆𝑖+ 𝛽3𝑅𝐸𝑇𝑖+ 𝛽4(𝐺𝐷𝑆 ∗ 𝑅𝐸𝑇)𝑖 +𝛽5(𝐸𝑋𝑆 ∗ 𝑅𝐸𝑇)𝑖+ 𝛽6𝐺𝑖 + 𝛽7𝐾𝑂𝑊𝑖+ 𝛽8𝐴𝐺𝐸𝑖 +𝛽9𝐸𝐷𝑈𝑖+ 𝛽10𝐼𝑁𝐶𝑖+ 𝛽11𝑅𝑖+ 𝛽12𝑆𝑀𝑖 )) Where:

𝐹𝐶𝒊 = Probability that a consumer converts at the focal company in purchase journey i;

𝐺𝐷𝑆𝑖 = Consumers’ goal-directed information search in purchase journey i;

𝐸𝑋𝑆𝑖 = Consumers’ exploratory information search in purchase journey i;

𝑅𝐸𝑇𝑖 = Retargeting in purchase journey i;

(𝐺𝐷𝑆 ∗ 𝑅𝐸𝑇)𝑖 = Interaction term of consumers’ goal-directed information search and retargeting in purchase journey i;

(𝐸𝑋𝑆 ∗ 𝑅𝐸𝑇)𝑖 = Interaction term of consumers’ exploratory information search and retargeting in purchase journey i;

𝐺𝑖 = Gender of user of purchase journey i;

𝐾𝑂𝑊𝑖 = Kind of work of user of purchase journey i;

𝐴𝐺𝐸𝑖 = Age of user of purchase journey i;

𝐸𝐷𝑈𝑖 = Completed education of user of purchase journey i;

𝐼𝑁𝐶𝑖 = Gross income of user of purchase journey i;

𝑅𝑖 = Region of user of purchase journey i;

𝑆𝑀𝑖 = Size of municipality of user of purchase journey i;

(33)

33 𝐺𝐷𝑆𝑖= 1 1 + exp (− (𝛽0+ 𝛽+ 𝛽1𝐷𝐴𝑖+ 𝛽2𝐴𝐴𝑖+ 𝛽3𝐸𝐴𝑖 4𝐺𝑖+ 𝛽5𝐸𝐷𝑈𝑖+ 𝛽6𝑅𝑖 )) Where:

𝐺𝐷𝑆𝑖 = Probability that a consumer performs a goal-directed information search in

purchase journey i;

𝐷𝐴𝑖 = Display advertising in purchase journey i;

𝐴𝐴𝑖 = Affiliate advertising in purchase journey i;

𝐸𝐴𝑖 = Email advertising in purchase journey i;

𝐺𝑖 = Gender of user of purchase journey i;

𝐸𝐷𝑈𝑖 = Completed education of user of purchase journey i;

𝑅𝑖 = Region of user of purchase journey i;

𝐸𝑋𝑆𝑖=

1

1 + exp (− (𝛽0+𝛽+ 𝛽1𝐷𝐴𝑖+ 𝛽2𝐴𝐴𝑖+ 𝛽3𝐸𝐴𝑖

4𝐺𝑖+ 𝛽5𝐸𝐷𝑈𝑖+ 𝛽6𝑅𝑖 )) Where:

𝐸𝑋𝑆𝑖 = Probability that a consumer performs an exploratory information

search in purchase journey i;

𝐷𝐴𝑖 = Display advertising in purchase journey i;

𝐴𝐴𝑖 = Affiliate advertising in purchase journey i;

𝐸𝐴𝑖 = Email advertising in purchase journey i;

𝐺𝑖 = Gender of user of purchase journey i;

𝐸𝐷𝑈𝑖 = Completed education of user of purchase journey i;

𝑅𝑖 = Region of user of purchase journey i;

3.5 Plan of Analysis

In order to analyze the data and estimate the models as presented above, the statistical program RStudio will be used. As a first step, the data will be cleaned by means of checking for missing values and outliers (Malhotra, 2009). If missing values or outliers are detected, appropriate treatment will be applied to assure the validity and reliability of the research outcomes (Malhotra, 2009). Subsequently, the dataset will be prepared by restructuring the data in such a way that the information required from the data is on the level of consumers individual purchase journey. In order to get preliminary insights into the data and the variables, descriptive statistics will be provided. Furthermore, the presence of multicollinearity between the independent variables will be analyzed and treated accordingly. Subsequently, several models will be estimated and compared on appropriate model fit measures. The best performing model will be used for hypotheses testing and the significant individual parameters of this model will

(3.5.1.3)

(34)

34 be analyzed and assessed. The binary logit models will be estimated by Maximum Likehood Estimation, and the Generalized Linear Model (GLM) function will be used to run the estimations (Malhotra, 2009; Leeflang et al., 2015). To finalize the data analysis, the estimation results of the best models are validated by means of bootstrapping. The upcoming section will discuss the results of the data analysis.

4. Results

4.1 Preliminary Checks

First, the data was checked for missing values. For 2031 of the 9678 users who participated in the study, missing data was found. The missing values originated all in the demographic data of the participants. The reason behind the missing values can be attributed to the fact that the users participated in different panels in the study and GfK was not able to retrieve the demographics for all participants. Since the missing values were only found for the control variables, the users related to the missing values were not deleted as this will lead to a loss of valuable information with regard to the variables of interest (Schafer & Graham, 2002). Missing values in continuous control variables were treated by mean substitution and missing observations in categorical control variables were replaced by the mode. This is common practice in scientific research (Schafer & Graham, 2002). To ensure that the implementation techniques will not influence the research outcomes, the models as specified in equations (3.5.12), (3.5.1.3), and (3.5.1.4) were estimated with imputed missings and with deleted cases. The estimation results of all models can be seen in Appendix B. The estimation results do not substantially differ between the models, thus imputing the missing values is appropriate and will ensure that no information with regard to the variables of interest will be lost.

(35)

35 were inspected to get a first idea about the distribution of the observations for each variable. Based on these results, in combination with a thorough look at the boxplots of the main variables of interest, it was concluded that the variables Exploratory Information Search and Display Advertising included one outlier. Additionally, Affiliate and Email Advertising included two outliers, Goal-Directed Information Search four, and Retargeting included five extreme values. These extreme observations deviate too much from the normal distribution of the observations and cannot be considered a true representation of the population (Obsborne & Overbay, 2013). Therefore, these extreme values were deleted. The boxplots of the main variables of interest before and after outlier treatment can be seen in Appendix D and Appendix E.

4.2 Preparing the Datasets

(36)

36 Furthermore, multiple datasets were created in order to be able to analyze the relation between a firm’s advertising practices (i.e. display, affiliate, email) and consumers’ information search (i.e. goal-directed and exploratory) (equations (3.5.1.3), (3.5.1.4)). The loop assured that the new created datasets report information at the level of consumers’ individual purchase journeys and only include the touchpoints related to display, affiliate, and email advertising that consumers have encountered before either information search. Since consumers can perform multiple information searches within the same purchase journey, the loop made sure that this information is also included in the new dataset by means of recording multiple lines of information for the same purchase journey. Separate datasets for goal-directed information search and for exploratory information search were created: one to account for the short-term advertisement effect and one to account for the long-term advertisement effect. With regard to the short-term advertising effect, only the advertising touchpoints (i.e. display, affiliate, email) that are directly seen before either information search were registered. Meaning, if a purchase journey includes two goal-directed information searches, the advertising touchpoints seen before the first search would be registered with the first search. With regard to the second search, only the advertising touchpoints seen after the first search and up until the second search will be registered. To account for the long-term advertisement effect, all advertisement touchpoints before an information search are included. This means that a second information search within a purchase journey would register the advertisement touchpoints seen before the first search in addition to the advertisement touchpoints seen between the first and second search. This relation assumes that an advertisement activity seen earlier in the purchase journey can still trigger an information search later in the journey. The datasets include all relevant dependent, independent, and control variables as explained in sections 3.2.3 and 3.2.4. A complete description of the loop(s) can be found in Appendix G.

4.3 The Relation Between Consumers’ Online Information Search & Conversion

4.3.1 Descriptive Statistics & Data Exploration

(37)

37 searches. Of these 9 purchase journeys, 7 have also been retargeted after having performed an exploratory search. With regard to demographics, 34.74% of the population are males and 65.25% are females. Their average age is 51.8. A complete overview of the variables including their operationalization and descriptive statistics is provided in Appendix H and Appendix I. An excerpt of the full overview as provided in Appendix H can be found in Table 5.

Table 5. Descriptive statistics.

Variable Min Max Mean SD.

Focal Conversion 0 1 0.007 0.081

Goal-Directed Search 0 80 0.513 2.542

Exploratory Search 0 25 0.059 0.597

Retargeting 0 256 0.225 4.941

(Goal-Directed Search * Retargeting) 0 10105 1.775 73.604

(Exploratory Search * Retargeting) 0 3500 0.348 23.085

To get a first idea of the relation between consumers’ goal-directed information search and conversion, and exploratory search and conversion, two Welch Two Sample t-test were performed. The results show that the purchase journeys who end with a conversion at the focal company include significantly more goal-directed information searches (M = 0.010) than the purchase journeys who did not end with a conversion at the focal company (M = 0.006), t(4425.6) = 2.38, p = 0.008. Additionally, the results show that the purchase journeys who end with a conversion at the focal company include marginally significantly more exploratory information searches (M = 0.013) than the purchase journeys who did not end with a conversion at the focal company (M = 0.006), t(670.5) = 1.59, p = 0.056. These preliminary results show that consumers who converted also performed significantly more information searches (goal-directed and exploratory).

4.3.2 Multicollinearity

(38)

38 Table 6. VIF scores of the predictor variables..

Variables VIF Score

Goal-Directed Search 1.165

Exploratory Search 1.603

Retargeting 3.350

(Goal-Directed Search * Retargeting) 2.906 (Exploratory Search * Retargeting) 1.826

Gender 1.246 Kind of Work 3.777 Age 2.683 Education 1.891 Income 1.991 Size of Municipality 4.700 Region 4.356 4.3.3 Model Selection

A common practice in scientific research is to mean-center the predictor variables when estimating a (logistic) regression model that includes an interaction effect (Dalal & Zickar, 2012; Echambadi & Hess, 2007). It is argued that mean-centering does not change the outcomes but will only aid in the interpretation of the estimation results in situations when the scales of the predictor variables do not include a meaningful zero-point (Dalal & Zickar, 2012; Echambadi & Hess, 2007). Since the independent and moderating variables included in this analysis are measured on a ratio scale (i.e. include a meaningful zero-point), and because using mean-centered predictor variables indeed does not change the estimation outcomes (see Appendix J for the estimation results), this research will not use mean-centered variables.

(39)

39 outcomes can be detected. Therefore, this research will continue with Maximum Likelihood Estimation as this method is easier to use for model comparison practices (Wang, 2014).

A base model was created which only included the independent variables (Model 1). Subsequently, control variables were added or left out based on an assessment of the significant impact of the control variables (Lani, 2014). 8 different models were estimated and compared based on five commonly used model comparison criteria, as can be seen in Table 7. The decision to include five criteria is because each criteria focuses on a different aspect when assessing model fit. A model that performs well on all five criteria is assumed to be of better quality and hence will produce more reliable research results. The complete estimation results of all 8 models is presented in Appendix L.

(40)

40 measures are not able to choose a best model that is able to effectively select market segments (Greene & Milne, 2010). The top-decile lift is a metric that shows how the incidence of the 10% consumers with the highest model predictions is related to the incidence of the overall sample (Greene & Milne, 2010). A higher decile lift score is considered to be better, and a top-decile value of 1 indicates that a model classifies equally as a random model (Greene & Milne, 2010). High risk consumers can easily be classified with this practice, which makes it therefore a valuable metric used by practitioners (Greene & Milne, 2010). Including a practically relevant model comparison next to the more traditional (scientific) measures is assumed to be helpful in deciding on the best performing model for this research. Model 7 has the highest TDL value. Overall, Model 8 is considered the best as it performs quite well on all model comparison criteria. A likelihood ratio rest showed that Model 8 fits the data significantly better than a null-model (p = 0.000).

Table 7. Outcomes of model comparison criteria.

Model AIC Hitrate

(41)

41

4.3.4 Hypotheses Testing

Model 8 was estimated and used for hypotheses testing. The outcomes of a logistic regression analyses are somewhat difficult to understand and directly interpreting the parameter estimates should be avoided (Leeflang et al., 2017; Peng, Lee, & Ingersoll, 2002). Instead, research suggest to use three means of interpretations when assessing the outcomes of a logistic regression, which this research also adheres to (Leeflang et al., 2017). First, one can assess the sign of the coefficient estimates and see whether there is a positive or negative relation between the predictors and the dependent variable. A positive (negative) β-value indicates that when the value of the relevant predictor increases (decreases), the possibility of observing Y=1 (or in this case a consumer converting) increases (decreases). Secondly, the odds-ratio can be assessed which is merely the exponent of the parameter estimates. Values larger than one indicate a positive relationship and values below 1 a negative relationship. Lastly, to get an idea of the size of the effect, the parameter estimates need to be transformed to marginal effects coefficients. The marginal effects for each variable assumes that all other predictor variables are at their average value. Note that conclusions can only be drawn from these three measures if the outcomes are statistically significant. Table 8 provides an excerpt of the estimation outcomes of Model 8. A complete overview of the outcomes can be seen in Appendix L.

(42)

42 effect on the relation between consumer’s goal-directed information search and conversion. Given the insignificant results of both interaction terms, hypothesis 5c is rejected.

Table 8. Excerpt of the estimation results of model 8 (dependent variable Focal Conversion).

β S.E. Z-Value P-Value Odds Ratio Marginal Effects (Intercept) -8.030 1.154 -6.959 0.003*** 0.000 Goal-Directed Search 0.021 0.020 1.020 0.302 1.021 0.000090 Exploratory Search 0.018 0.097 0.180 0.857 1.018 0.000070 Retargeting 0.028 0.006 4.856 0.000*** 1.028 0.000120 (Goal-Directed Search*Retargeting) -0.001 0.001 -0.937 0.349 0.993 -0.000003 (Exploratory Search*Retargeting) -0.001 0.001 -1.113 0.266 0.999 -0.000003 Note. p < 0.1 (.), p < 0.05 (*), p < 0.01 (**), p < 0.001 (***)

The aforementioned insignificant results are not a surprise, since Focal Conversion is a rare event in the data. As explained in section 3.2.1, the hypotheses will be retested by replacing the dependent variable Focal Conversion with the variable Conversion. Since a different dependent variable is used, the procedures as explained in sections 4.3.2 and 4.3.3 have to be redone. First, no multicollinearity problems exists since the VIF scores as presented in Appendix M are all below the threshold of 5. Secondly, 8 different models were again estimated and compared of which the results can be seen in Appendix N. Model 8 performs quite well on all model comparison criteria and is overall considered the best model (see Appendix O for model comparison results). Additionally, the results of a likelihood ratio test showed that Model 8 fits the data significantly better than a null-model (p = 0.000). An excerpt of the complete model estimations as expressed in Appendix N can be seen in Table 9. The full model included the variables as expressed in Table 9, and control variables Gender, Kind of Work, Education, Income, and Region.

(43)

43 used to statistically test the difference of the effects because these tests can only be applied when comparing regression coefficients across independent samples (Cohen, West & Aiken, 2014; Paternoster et al., 1998). Both variables are measured on the same scale thus the size of the coefficient effects can be compared to see which type of information search has a stronger effect on conversion (Malhorta, 2009). Both the β-value (0.094) and the marginal effect size (0.0100) for exploratory search is stronger compared to goal-directed information search (β = 0.057 and marginal effect size = 0.0060). Hence, the positive effect of exploratory information search on conversion is stronger compared to goal-directed information search. Thus, hypothesis 1c cannot be supported. Furthermore, the main effect of retargeting is again positive and highly significant (β = 0.018, p = 0.000). However, both interaction effects are still insignificant (goal-directed search: β = -0.0005, p = 0.195; exploratory search: β = 0.001, p = 0.574) Hence, hypotheses 5a, 5b, and 5c are again not supported.

Table 9. Excerpt of the estimation results of model 8 (dependent variable Conversion).

β S.E. Z-Value P-Value Odds Ratio Marginal Effects (Intercept) -2.360 0.2140 -11.024 0.000*** 0.0094 Goal-Directed Search 0.057 0.0058 9.827 0.000*** 1.0587 0.0060 Exploratory Search 0.094 0.0236 3.983 0.000*** 1.0987 0.0100 Retargeting 0.018 0.0040 4.664 0.000*** 1.0188 0.0020 (Goal-Directed Search*Retargeting -0.0005 0.0001 -1.334 0.195 0.9995 -0.00005 (Exploratory Search*Retargeting) 0.001 0.0017 0.562 0.574 1.0010 0.000010 Note. p < 0.1 (.), p < 0.05 (*), p < 0.01 (**), p < 0.001 (***)

(44)

44 The values of the z-statistics range from -7.555 to 3.100. The results show that most values of the true interaction effect are more positive than the marginal effect of the interaction effect as provided in Table 9 and is even significant for some observations. The same holds for Figure 4. The true interaction effect varies widely per observation and is even significant for some observations in contrast to the results in Table 9. The values of the true interaction effect range from -0.00033 to 0.00037 with a mean value of 0.0002 and z-statistics ranging from -3.262 to 1.766. Figures 3 and 4 show that the interaction effects are indeed not as straightforward as indicated by the model outcomes as the interaction effects are positive and even significant for some observations. However, the positive and significant true interaction effects are considered to be such rare events that the graphs still do not provide conclusive evidence to support hypotheses 5a, 5b, and 5c.

Figure 3. True Marginal Effects and Z-Statistics (Goal-Directed Search*Retargeting).

Referenties

GERELATEERDE DOCUMENTEN

The most intriguing difference between the different additives is the reduction in protein score and protein coverage for BSA in the eluted fraction, in which L-glutamic acid has

The importance for more private sector agricultural development in rural areas is clear but there are still some challenges when using agricultural PPPs for

In episode three, the editor/author utilises bodies and spaces such as the king, the Babylonians, Daniel, the lions’ den, the prophet Habakkuk and food to demonstrate the

By modeling different states in the purchase process based on touchpoint data, and by carefully examining the effect of different types of touchpoints in

5.5.1 The use of online channels in different stages of the customer purchase journey In order to test the first hypothesis multiple logit models are tested with a channel as a

Such information includes type of touchpoints visited (customer- or firm- initiated), whether a journey has led to conversions, type of device used to assess a contact and

If this is the case, it is important to ascertain which combination of cross-media marketing activities might have the greatest influence on the purchase behavior of

Three mayor conclusions were drawn: (1) review quantity has a positive effect on sales, (2) review variance has a negative effect on sales and (3) review valence has a positive