• No results found

Patrik Nowak

N/A
N/A
Protected

Academic year: 2021

Share "Patrik Nowak"

Copied!
52
0
0

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

Hele tekst

(1)

WHAT’S A THANK YOU WORTH IN SOCIAL MEDIA? Moderation effects on Gratitude expression in a SocialCRM context

Patrik Nowak

Submitted on 12th January 2015

(2)

WHAT’S A THANK YOU WORTH IN SOCIAL MEDIA? Moderation effects on Gratitude expression in a SocialCRM context

Patrik Nowak

Marketing Department Faculty of Economics and Business

University of Groningen Master Thesis Submitted on 12th January 2015 Friesestraatweg 24-2c, 9718NJ Groningen +31(0)681 60 2200 patriknowak89@gmail.com Student Number: 1859692 1st supervisor: Dr. H. Risselada 2nd

supervisor: Dr. ir. M.J. Gijsenberg

External supervisor: Bob Christiaanse, Media Injection B.V.

Word count: 12.950 words

(3)

ABSTRACT

This study focuses on the connection between social media and customer relationship management (SocialCRM), which is seen as a solution for the ongoing lack of accountability in marketing. By focusing on an underexplored concept, the so called gratitude, found to have a positive impact on the relationship between customer and businesses, this study investigates the effect of customer behavior characteristics on the change of engagement. Furthermore, a first step towards missing accountability is taken because it is the first study that connects real-life social media and conversion data. The datasets in this research, analyzed through the use of ordinary least squares regression, were provided by three Dutch companies, actively engaging with their customers on social media and selling products online.

In the first part of the study, the assumed positive effect of pre-gratitude interaction as well as relationship quality could not be confirmed. Nonetheless, the relationship duration before gratitude has been found to impact change in engagement around gratitude positively. Consequently, it confirms gratitude literature arguing that longer relationships have an impact on the gratitude effect. In the second part of this research, a relationship between change in engagement on social media as well as conversion behavior was confirmed. Thus, evidence is added to the ongoing discussion whether a connection regarding SocialCRM would be valuable.

(4)

PREFACE

Now, that this project is moving toward its end, I take my turn to thank a few people who supported this project in general or just helped me during the last 5 months. First of all, I thank Media Injection for their trust and support for this research project, especially Bob Christiaanse, for your ideas and commitment, as well as Mariano D’Arcangelo for your data support all the way from beautiful Argentina. Next, I thank Dr. Hans Risselada for your support, open mind, and the always fruitful and emboldening sessions. Of course, I also have to thank Jeroen Sitskoorn from bol.com, Rob Abspoel and Jochem Meijer from Transavia, and Marieke van der Heijden and Bouke Plat from Greetz.nl for their willingness to support the project, even when deadlines got close.

(5)

MANAGEMENT SUMMARY

Accountability has been a key issue for marketing since the first day. Already a century ago pioneers like John Wanamaker, were certain that one half of their marketing spending was wasted; they just needed to find out which one (Hoffman and Novak, 2000). Today, seeing marketing budgets for new channels of social media and online advertising rising over $4,3 billion, this question seems to be more relevant than ever. However, one variable has been added to the hunt for accountability, the traceability inherent to the online environment.

Focusing on social media, new possibilities come into play regarding the examination of previously mostly conceptually investigated concepts. Gratitude, a concept that has long been underestimated in the field, but found to be critical in a successful relationship, in both a private and a business context, is the core of this research in order to explore its varying effects more closely. Following reciprocal behavior makes gratitude particularly interesting for businesses that invest in and engage interactively with their customers on new social media channels. A crucial aspect of this concept, when looking at real-life data, is the fact that gratitude can actually be observed through clear expression. Focusing on the connection of behavior on a new social media channel and the actual conversion, I expand the field of gratitude research with a real-life case and simultaneously analyze the underexplored connection between social media and customer relationship management (SocialCRM) (Malthouse et al., 2013). This latter concept refers to the next leap forward regarding the accountability of marketing, potentially enabling businesses in the field to finally assess the return on marketing spending going into new marketing channels. Accordingly, the central research aim of this study is to investigate the moderation effects of customer behavior variables and the impact gratitude expression has on customer engagement on social media, as well as conversion behavior in a SocialCRM context.

(6)

media cases. Greetz.nl was the only company able to provide the data, and their dataset has been used for the second step, connecting social media and customer relationship data. This dataset included a total number of 48 cases, observed on social media and in the customer relationship software of the company.

In the first step of the analysis, the effects of moderation variables, describing the behavioral characteristics of customers on the change in engagement on social media, are explored. In the second step, the connection between effects on social media and conversion is further investigated by taking the change in conversion behavior as the dependent variable.

Accordingly, the core findings of the study can also be divided along these two steps. Regarding the effect of the moderating customer behavior characteristic, the analysis does not confirm the hypothesis that assumes a positive effect of pre-gratitude interaction frequency on the degree of change in engagement around gratitude. Equally, the positive expected effect of relationship quality could not be confirmed. However, the relationship duration, expected to positively effect the change in engagement on social media, could be confirmed. Thereby supporting conclusions made in previous researches in the field of gratitude, as well as relationships in general. This demonstrates that the longer the relationship duration is, the greater effect gratitude expressed has in a relationship between customer and company. In the second step of the research, the assumed positive relation between change in engagement on social media and change in conversion behavior in CRM is confirmed. Another interesting finding regarding SocialCRM, is the negative correlation between number of interaction on social media and average purchase value of a customer. These findings add greatly to the discussions around the necessity of inclusion of behavioral measurement and advocate strongly for further investigation of SocialCRM and movement towards a customer valuation that go beyond the scope of purchase. Moreover, supporting descriptive findings have been included concerning the concept of window of opportunity, describing the short period of time after gratitude during which reciprocal behavior is observed in previous research.

(7)

TABLE OF CONTENTS

1. INTRODUCTION ... 8 2. THEORETICAL FRAMEWORK ... 12 2.1 SocialCRM ... 12 2.2 Gratitude ... 13 2.2.1 Gratitude Concept ... 14 2.2.2 Expression of Gratitude ... 15 2.2.3 Effects of Gratitude ... 16 2.3 Hypotheses ... 18 2.3.1 Control Variables ... 20 3. DATA ... 21 3.1 Data source ... 21

3.1.1 Social Media Data – Media Injection ... 21

3.1.2 Purchase Data – Bol.com, Transavia, Greetz.nl ... 21

3.2 Datasets ... 22

3.2.1 Gratitude Coding ... 22

3.3 Two datasets and two steps for the research ... 24

3.3.1 Step 1 – Gratitude Social Media Sample (n = 349) ... 24

3.3.2 Step 2 – Gratitude SocialCRM Sample (n = 48) ... 26

4. METHODOLOGY ... 27

4.1 Step 1 – Social Media ... 27

4.2 Step 2 – SocialCRM ... 28

5. RESULTS ... 29

5.1 General Effect of Gratitude ... 29

5.2 Step 1 – Social Media ... 31

(8)

9. APPENDIX ... 48

APPENDIX 1: Overall trends in average interaction per day. ... 48

APPENDIX 2: Additional information for residual analysis in step 1. ... 49

APPENDIX 3: Additional information for residual analysis in step 2. ... 50

(9)

1. INTRODUCTION

Gratitude, thankfulness, gratefulness, or appreciation is a feeling or attitude in acknowledgment of a benefit that one has received or will receive. (Wikipedia, 2014)

Accountability of relationship marketing actions, in their traditional, as well as recent digital forms, has been one of the most visible and important drivers of research in the field of marketing. This problem is not new to marketing, as is shown by the quote of John Wanamaker, a pioneer in marketing, who became famous for his statement that “half of the money [he] spends on advertising is wasted; but [he] can never find out which half” (Hoffman and Novak, 2000, p. 6). Today, more than a century later, multiple steps have already been taken. However, limitations of published literature and continuing need for further research show that sufficient accountability has not yet been reached (Gu and Ye, 2014; Yadav, De Valck, Hennig-Thurau, Hoffman, and Spann, 2013). This is especially the case as a new form of marketing through social media channels becomes a more accepted part of the marketing mix, accounting for $4.3 billion in marketing spending in 2014 (Marketingfacts, 2014a; Hennig-Thurau, Hofacker, and Bloching, 2013; de Vries, Gensler, and Leeflang, 2012). Necessary investments to facilitate appropriate interaction on those platforms lead to a new increased need for tools and measurements to assess their effectiveness and return on investment. These new modes of communication have one advantage in contrast to all channels before – their traceability. This ability to analyze user behavior in detail opens up new opportunities comparable to the introduction of scanner data in retail in the 1970s and 1980s.

(10)

(2009) more practically as “the emotional appreciation for benefits received, accompanied by a desire to reciprocate” (Palmatier et al., 2009, p. 1). First of all, the simple positivity of gratitude itself generates a positive impact on the relationship. Moreover, its visibility to other users of the social media channel transfers this positive connection to the business and others connected users (Adjei, Noble, and Noble, 2010). Although the latter part of the definition could be expected to be of great interest to businesses referring to a network effect, the former part is central to this research in order to explore the core process of gratitude. Attaching a value to expressed gratitude, due to measurable impact following on social media, as well as increasing customer value for the business, can be seen as one piece of the puzzle towards the creation of accountability for relationship marketing actions in general. Hence, a simple “thank you” posted on social media, functioning as an indicator for gratitude (Palmatier et al., 2009; Raggio and Folse, 2009) and as a starting point for ongoing reciprocal behavior, enables researchers and practitioners to delve deeper into the effects gratitude can have on networks, relationship marketing effectiveness, and customer behavior in general.

(11)

highly underexplored field of SocialCRM1

, discussed by Malthouse and colleagues (2013) and Yadav et al.(2013). While explaining the potential advantages of the integration of social media behavior into calculations of customer lifetime value, the authors still conclude that there is a significant gap in the literature regarding the valuation of investments in SocialCRM (Malthouse et al., 2013). By focusing on the underexplored concept of gratitude, this research intends to make a start in closing the gap between current knowledge and full accountability of relationship marketing activities in the context of SocialCRM.

One of the reasons for scarcity of research in the field regarding the integration of social media and CRM relates to the persistent difficulties to link data from one source to the other. In most cases, both systems are managed independently by different departments in the same company, thereby the involved departments miss opportunities to learn from each others’ insights, conclusions and the potential of generating new metrics as described in the literature (e.g. Farris, 2009; Peters, Chen, Kaplan, Ognibeni, and Pauwels, 2013; Hennig-Thurau et al., 2013; Malthouse et al., 2013). The aim of this research however, as has been touched upon above, is to tab into that gap and combine data gathered by social media on the one, and CRM data on the other hand. In order to facilitate this, social media data and customer data is matched over a period of 12 months in order to follow individual customers in either database. With this longitudinal approach, this research is unique in its ability to follow the customer behavior regarding social media engagement and relevant customer behavior, and therefore compensating for limitations in previous research (e.g. Raggio et al., 2009).

In order to generate valuable insights regarding the effect of customer characteristic on the effect of expressed gratitude I use a two-step approach, using two different but related samples. The first step focuses only on users expressing gratitude on social media, whereas in the second step, a dataset including social media and CRM data is used. Due to different amounts of variables and cases available on different aggregation levels, I, on the one hand, use a large sample size for more generic social media analysis and on the other, scale down to smaller sample sizes, where observations are only available for certain individual cases. Starting with a large sample in which gratitude-expressers are analyzed regarding moderation

                                                                                                               

1  SocialCRM  also  referred  to  as  CRM  2.0  (Greenberg  et  al.,  2010)  refers  to  the  combination  of  Social  Media  

(12)

effects of third variables on social media, detailed behavior of those expressing gratitude is taken into account. In the second sample, social media and purchase behavior data is combined and analyzed. Accordingly, it is explored how pre-gratitude customer behavior effects the impact of expressed gratitude, both focusing on social media data and on purchase related effects.

Through this novel approach, findings in previous literature regarding an increase in customer engagement on social media after compared to before the moment expressed gratitude are confirmed (Young, 2006; Raggio, Rouge, Walz, Bose, and Folse, 2008). Moreover, regarding the moderation effects of customer behavior characteristics, the analysis demonstrates a positive effect of relationship duration on the change in engagement around gratitude, confirming the hypothesis that a longer relationship leads to a stronger effect of gratitude. The positive effect of interaction frequency and relationship quality, however, could not be confirmed in this study. Especially relevant regarding the discussed connection of SocialCRM, an increase in engagement on social media has been found to positively influence the effect gratitude has regarding conversion behavior. Supported by multiple other correlations found between social media and CRM data, these results enhance the ongoing discussion about the sensibility of a SocialCRM connection in the first place and help practitioners to back their efforts towards a permanent and more detailed connection between both systems. In addition, the controversial concept of window of opportunity, describing the effect that directly after the expression of gratitude customers are most likely to perform reciprocal behavior, has been supported by new descriptive data (Raggio and Folse, 2014). It shows that in our dataset a large share of customers’ purchases occur in a short period of time after gratitude. Nevertheless, this could not be confirmed statistically and has only been left in to serve the ongoing discussion and aid future researches as an anecdotal starting point.

As has been elaborated above, this novel research, focusing on the concept of gratitude and combining social media and CRM, intends to extend the frontier of knowledge in the field of marketing research. The results found both contribute to academia and practitioners, generating new starting points for future research as well as useful tools to make marketing activities more accountable in the field.

(13)

experiment and field-survey based literature with a real-life case. Accordingly, this research forms a next step, not only concerning the academic understanding of the concept of gratitude, but also due to the utilization of the impact of reciprocity in order to advance academia towards understanding of the concept of SocialCRM. I am confident that future research can benefit from this new approach as a starting point for deeper research on the effects of gratitude and the SocialCRM context, and greater willingness for companies to share data and collaborate in grounded research.

Furthermore, this research has implications for multiple actors in businesses in terms of their social media strategy. First of all, in combination with the ongoing discussion regarding the accountability of social media efforts, expressed gratitude has shown to be a valuable part of a metric, evaluating the overall social media effectiveness and therefore the return on a company’s investment. Additionally, this research can serve as a starting point for exploration of different strategies, evoking the expression of gratitude on social media, especially for customers with a long relationship, in order to increase reciprocal behavior and return on investment. Overall, the most important conclusion for practitioners from this research is that the connection between social media on the one and CRM on the other side can lead to valuable conclusions, new opportunities and competitive advantages in business.

2. THEORETICAL FRAMEWORK

In the following section I discuss the core concepts of this research and subsequently present the central research question and the conceptual model. Next, each hypothesis will be presented and explained, based on previous literature. Finally, the control variables included in this research are discussed.

2.1 SocialCRM

(14)

In the business environment, the lacking incorporation of social media metrics into the CRM is still making for a lack of appreciation of these new networks, referring back to the aforementioned lack of accountability in marketing in general (see also Libai, Muller, and Peres, 2013; Langley, Hoeve, Ortt, Pals, and van der Vecht, 2014). For social media, this lack might become a serious threat undermining its credibility and leading to the movement of investments from social media to other channels (Luo, Zhang, and Duan, 2013). Possible approaches around the generation of insights and enabling companies to better assess their relationship marketing actions on social media are well-discussed topics (see also: Mathouse et al., 2013; Peters et al., 2013; Hennig-Thureau et al., 2013; Yadav et al., 2013).

In this discussion, academics and practitioners in the field call for more research exploring the “impact of social commerce” on multiple platforms used, in order to move towards the missing solution (Yadav et al., 2013, p. 320). Social media, as a tool in relationship marketing, continuously develops into a platform on which customers not only engage in discussion with fellow customers (Adjej, 2010), but also directly with the company. In contrast, other marketing channels mostly allow one-way communication. Gu and Ye (2014), state that purchase decisions are influenced by the high degree of activity on those new channels, leading back to the call for research of assessment of investments regarding this new means of activity. When referring to these means of interactions during the study, this concerns active engagement of a social media user towards the business on that medium, either through likes, comments or other messages.

2.2 Gratitude

(15)

performance (Palmatier et al., 2009; Raggio et al., 2014). Another advantage of gratitude is the opportunity to observe the feeling as it is expressed either directly after the beneficiary receives a perceived benefit or somewhat later when she or he intends to share her or his positive experience (Raggio et al., 2014). This observable expression makes it especially attractive for this research. Moving towards research that uses actual social media data instead of survey and experiment based approaches opens up new opportunities that will be used in this paper.

2.2.1 Gratitude Concept

As mentioned, gratitude has been found to have a positive effect on the relationship between customers and businesses. The underlying assumption relates to a study by Houston and Gassenheimer (1987), who note that “reciprocity turns transactions into [an] exchange relationship” (quoted in Palmatier et al., 2009, p. 3). Those exchange relationships have multiple and well known advantages for companies like loyalty, customer satisfaction and decreasing rates of churn, leading to a maximized customer lifetime value. However, using the concept of reciprocity, defined as the act of giving in return, might refer to “mindless tit-for-tat behavior” (Fredrickson, 2004, p. 150), excluding many effects inherent to the concept of gratitude. Accordingly, gratitude is not just referring to returning a favor after the reception of one; instead, it also includes a positive feeling connected to the benefactor, which eventually leads to a positive long-term relationship and can therefore be referred to as the “emotional core of reciprocity” (Raggio et al., 2014, p. 4).

The “emotional core of reciprocity” (Raggio et al,. 2014, p. 4) links to the first conceptual aspect of gratitude, the recognition of gratitude by the beneficiary, which is then followed by an emotional appreciation and a behavioral reaction. Perceiving a benefit resulting from good intentions of the benefactor already leads to feelings of gratitude, meaning that an actual physical or financial benefit is not necessary for the gratitude cycle to begin (Raggio et al., 2014). This has an important implication for the foundation of this research. Although customers did not receive any concrete benefit from the interaction with the company via social media, the expression of gratitude as “thank you” can still be seen as an identifier for gratitude and therefore leads to positive effects in the future (Raggio et al., 2009).

(16)

which increase the positivity of the feeling regarding the benefactor, and furthermore have a positive effect on a person’s evaluation of the benefactor’s trustworthiness (Algoe, Haidt, and Gable, 2008; Dunn and Schweitzer, 2005). The second component is behavior, referring to activities performed by the beneficiary in order to repay for the perceived benefits received (Palmatier et al., 2009). This component represents the highly interesting act of giving in return for a perceived favor received. Regarding the context in which this research takes place, this act can either be identified in the social media network itself as increased engagement, composed of future likes, shares and comments in connection to the company, or in conversion behavior related to the company. One additional interesting concept, frequently discussed in a business to consumer context, is the concept of “window of opportunity”, described by Palmatier et al., (2009, p. 13) (Raggio et al., 2014). It refers to a short period of time after gratitude in which reciprocal behavior, referring to company-relevant behavior, is significantly higher and decreases quickly afterwards. The concept has implications for businesses concerning the use of this subconscious customer behavior and has lead to discussion concerning the ethical tenability of its use. Hence, observing if the window of opportunity is present or not, could aid the discussion whether further investigation is sensible.

2.2.2 Expression of Gratitude

(17)

verbally, are much more likely to engage in reciprocal behavior than those who just perceive the feeling of gratitude. Furthermore, the expression itself led to a more positive long-term relationship between benefactor and beneficiary (Lambert et al., 2010). Applying these findings to this research and the conceptual framework illustrated in figure 1, I expect that moments of expressed gratitude are a usable indicator for the feeling of gratitude and the according reciprocal behavior.

2.2.3 Effects of Gratitude

Looking at long-term effects of gratitude, Schwarz (1967) described the so-called reciprocity cycle being started after an initial expression of gratitude. In his view, the first perceived benefit would lead to a misbalance of debt, fluctuating subsequently between beneficiary and benefactor, leading to an ongoing relationship in which gratitude becomes an important part (Schwarz, 1967 in Palmatier et al., 2009). Translating this to our social media environment, a relationship in which gratitude becomes part of the interaction and is enhanced through multiple touch points leads to an ongoing increasingly beneficial relationship for both customer and business. Additionally, the fact that perceived costs of reciprocation are relatively unimportant with easy opportunities to reciprocate on social media, especially increasing social media engagement relating to spreading of information to friends can be expected. Moreover, more ‘difficult’ actions as relevant conversion can expected to be performed in order to bring the relationship back in balance, since users engaging on the fan page of the company are presumably interested in the brand (Raggio et al., 2014).

(18)

when focusing on interaction on social media platforms (Van Doorn et al., 2010). Following the concept of Van Doorn et al. (2010), conversion behavior in this case refers to the traditional and still today most valuable behavior of the customer, their purchase behavior. Due to the utmost importance of this metric for practitioners, directly linked to the turnover of, for example, an online store (Ludwig, de Ruyter, Friedman, Brüggen, Wetzels, and Pfann, 2013), this metric is still taken into account in multiple researches and central to many debates. In this research, it is once again linked to the discussion about the development of a SocialCRM including all aspects of CEB (Malthouse et al., 2013).

Based on the assumptions above, I now move to the focus of this research, investigating on the one hand the moderation effects customer behavior characteristics have on the effect gratitude has on social media. On the other hand, I generate the link of SocialCRM, making the next step towards the conceptual integration of both concepts. This focus is summarized in the following central research question:

What are the moderation effects of customer behavior characteristics on the impact of gratitude expressed in a customer to business interaction, on social media engagement, as well as conversion behavior in a SocialCRM context?

Figure 1: Conceptual Model

(19)

research steps briefly discussed in the introduction, are indicated. After discussing the hypotheses proposed below, the data section will further explain the two-step approach.

2.3 Hypotheses

First, through the analysis of customer engagement previous to the gratitude event, the research explores whether users with an either high or low degree of pre-gratitude interaction exhibit different degrees of behavioral reactions on their change in engagement after the expression of gratitude. Previous research has shown that degree of engagement in general can have an effect on company relevant behavior. The data not only provides insights whether high interaction per se leads to more positive conversion, but also to discuss its effect on reciprocal reactions of users. One research, in the field of engagement, has already indicated a positive effect of engagement on advertising effectiveness (Calder, Malthouse, and Schaedel, 2009). Furthermore, Van Doorn et al. (2010) state that besides the lack of understanding why certain groups of satisfied people interact more in certain aspects of the CEB, it has been shown that those customers engage more in general. In a related research, Dagger, Danaher, and Gibbs (2008), find a positive effect of interaction frequency on relationship strength, which is reported by the customer, adding to the assumptions made above. Consequently, the effect expressed gratitude has on engagement can be assumed to be influenced by the overall degree of pre-gratitude interaction. Due to the differing degrees of interaction shown by different users, the received benefit might either be perceived as more special or not. Following the literature above, the first hypothesis reads:

H1: Customers with a higher average interaction before an expression of gratitude, display a more positive change in engagement.

(20)

relationship duration, is expected to increase the change in engagement on social media following an event of expressed gratitude (Raggio et al., 2014).

H2: Customers with a longer relationship duration are expected to display a more positive change in engagement.

Although both above presented hypotheses do already refer to what has been used as a proxy for relationship quality, the mere number of different kinds of interactions omits an additional aspect of relationship quality. The social media focus of this research allows further differentiation between means of interaction such as likes, comments and inferring a qualitative distinction between users with a higher share of either or the other. In a social media study, De Vries et al. (2012) explore how companies can influence their social media activities to generate either likes or comments, by placing posts with specific characteristics. They find out that likes and comments are partly achieved through different characteristics regarding vivid- and inter-activeness. The same distinction is made in this research, since interaction that takes place through different means that requires varying degrees of effort. Referring again to the meta-analysis of Palmatier et al. (2006), and the central criteria of relationship quality phrased as the “overall assessment of a relationship” (Wulf, Odekerken-Schröder and Iacobucci, 2001, p. 36), I add this variable to this research in order to better assess the quality of individual relationships. Again, an increased relationship quality is expected to increase the effect generated by expressed gratitude. Together with the relationship duration and interaction frequency, I am confident that a clear picture of the moderation effects customer specific characteristics have, on the change in engagement, can be given.

H3: Customers with a higher relationship quality are expected to display a more positive change in engagement.

(21)

purchase frequency compared to negative information. Equally important, following the reasoning of literature in the growing field of SocialCRM, the connection between social media and the CRM extends the former limited view by incorporation of actual customer behavior (Malthouse et al., 2013). Therefore, I assume that through the conceptual interconnectedness of social media and CRM, the effect of gratitude on social media is positively related to the change in conversion behavior. Accordingly, the following hypothesis is proposed:

H4: The change in engagement stimulated by gratitude on social media is positively related to the change in conversion behavior.

Finally, the “window of opportunity”, a special characteristic of gratitude will be analyzed as well. Due to the opportunity to reciprocate easily in an online setting, the effect is expected to be found in our research setting (Palmatier et al., 2009). As a side note, Raggio et al. (2014) discuss the implication of this concept and while acknowledging its potential, they stress the ethical implication it has regarding the attempt of potential misuse of the window of opportunity in order to force customers to return a favor (Raggio et al., 2014). To further explore the existence of the “window of opportunity” and its potential, I formulate the following fifth hypothesis:

H5: Customers are more likely to exert conversion behavior, directly after an event of expressed gratitude took place.

2.3.1 Control Variables

(22)

3. DATA

As has been discussed in the theoretical framework and the conceptual model, the core interest of this research is the effect of customer characteristics on the effect gratitude expressed on social media has on the engagement of the expresser on that medium, as well as on conversion behavior. Therefore, the changes in social media engagement and in conversion behavior around the moment of expressed gratitude are the two central dependent variables. To be able to assess the effects mentioned, two sources of data have been used. On the one hand, customer data on social media, provided by Media Injection B.V., and on the other hand, online conversion data that was provided by companies using the platform of Media Injection and matched individually with the social media data. Due to differences in data availability, namely the number of variables and cases, two datasets have been created.

3.1 Data source

3.1.1 Social Media Data – Media Injection

The social media data has been provided by the software company Media Injection B.V. in Amsterdam (MI). MI has developed a platform enabling companies to interact with users via different online platforms. This interaction consists of proactive posting of content, responding to activities of users, and analyzing overall brand related online behavior. Looking at the data on social media, I only focus on data generated on the company fan pages on the social media platform Facebook for four reasons. First of all, interaction on Facebook can be more diverse because of different means of interaction e.g. to like, comment and share, instead of a more narrow functionality as provided by Twitter. Secondly, with 1,23 billion users globally and 9 million users in the Netherlands, the platform is the by far most popular social media platform for both customers and businesses (Marketingfacts, 2014b). Thirdly, the data available about users is best suited to connect with other internal company databases due to the availability of names. And finally, interaction histories are easier to follow due to structured conversations and enable a better assessment of expressed gratitude, crucial as the base of this research. Conclusively, focusing on fan pages, describing a created page by a company, all interaction exported can be clearly attributed to the respective company.

3.1.2 Purchase Data – Bol.com, Transavia, Greetz.nl

(23)

consisted of bol.com, the largest online retailer in the Netherlands; Greetz.nl, a large reseller of greeting cards that can be ordered online and send directly to the receiver via mail; and Transavia, a large travel booking agency. All three companies agreed to use their social media data to investigate the proposed research question. Two of the companies, Greetz.nl and Transavia, were able to provide matched purchase data to the gratitude expression cases identified on social media. Due to the unavailability of e-mail addresses in the social media dataset, the matching of both datasets was based on name and surname. Regarding the size of the customer database of bol.com, they were unfortunately not able to match cases validly due to multiple double matches. Although transavia.nl did provide Customer Relationship Management (CRM) data linked to the users on social media, the low number of purchases per person, inherently connected to the product sort, i.e. flights and vacations, did lead to an insufficient number of observations per case in the research period. Therefore, only the matched purchase data of Greetz.nl has been used in further analysis in order to facilitate the connection of SocialCRM.

The step of connecting both databases is the only moment during the research in which the actual user names are used, in order to match both databases. Before the connection of the databases, during coding of gratitude, and afterwards, users were identified only using a new unique number. After the connection, all user names were deleted permanently. Hence, all analyses are performed with an anonymous dataset to ensure the privacy of the users.

3.2 Datasets

3.2.1 Gratitude Coding

As a first step of the research, all individuals expressing gratitude need to be identified and classified in order to create a research sample. This process practically started through the export of all messages available from MI for the observation period of the research (1st

of October 2013 until 31st

of October 2014) consisting of the keywords “thank” and/or “dank” (Table 1). This period has been used for two reasons. First, the end date of the observation period was the date of the research export, exporting all data available up to that moment. Second, at the end of 2013 Facebook introduced new functionality allowing users reply on individual comments, increasing the quality and clarity of interaction histories.

(24)

addressed to the company. It needs to be noted that the chosen keywords were used in different ways in the posts, namely independent from the expression of gratitude. For example, in multiple cases the recipient of gratitude expressed was another user, or the goal of the use of one of the keywords was neither positive nor related to gratitude expression at all. Through this process, between 40% and 60% of all exported cases have been excluded, creating a database with messages of expressed gratitude that can be assumed to be a usable indicator for the feeling of gratitude (McCullough and Tsang, 2004).

Table 1: Words used for Export Search term Words included

Dank Dank je wel, Dankje, Dank, Bedankt.

Thank Thank, Thanks, Thank you.

In a next step, all users that appear more than once are excluded leaving a final dataset of 770 cases from the three companies combined. Furthermore, to be able to assess changes in the overall interaction pattern of individual users, a certain minimum amount of interactions per user is needed. A minimum number of 10 interactions have been chosen, in order to base the results on a sufficiently long relationship between customer and business. At the end of the results section, robustness checks are presented to support this choice and examine whether a further increase of the number of interactions leads to different or more concise results. Naturally, the higher the amount of interaction is, the more information the research’s conclusions are based on. Ten interactions, however, are chosen as a balance between quality and quantity.

Table 2: Database Cleaning Process

Subdivision Export Keywords Coded Gratitude 10+ Interactions

Bol.com 1019 400 (39,3%) 178

Greetz.nl 293 151 (51,5%) 83

(25)

Applying this second filter, again about 50% of all cases needed to be left out from the ensuing analysis, due to an insufficient number of interactions during their relationship with the company fan page (Table 2). Of course, this strict process of coding and excluding cases with insufficient data and characteristics does decrease the total sample size by a great amount. However, to ensure that the effects found in the analysis can be trustworthy, this process is essential.

3.3 Two datasets and two steps for the research

Following the fact that different amounts of data are available from the three participating companies, this research is separated in two steps or case studies. Each of those case studies has a specific database, dependent on the data available and the sub-goal of the step. Nevertheless, all steps do play a role in answering the overall research question and the formulated hypotheses.

3.3.1 Step 1 – Gratitude Social Media Sample (n=349)

The first sample, i.e. the social media sample, created with the goal to assess the effect of the moderation variables, includes a total of 349 cases from all three participating companies. The available variables for those cases are presented in table 3. In the same table, the calculations of the created variables are added. From the fact that the date of gratitude expression is available, the interaction frequency, as well as intensity, can be compared before and after the event of interest. What has to be noted is that the dates of gratitude expression are different for every case in the dataset. Therefore, it is highly unlikely that an unobserved third event, such as a special one-time activity by the company, is the reason for changes in the relationship of all users in the database. Interaction in this case refers to any action the user performs on the Facebook fan page of the company, being either a like, comment, share, vote, reply or message. Thus, the interaction intensity is the average interaction per day a user performed throughout his or her whole relationship. The relationship duration is therefore calculated as the time in days between the point the user liked the fan page and the moment of expressed gratitude. In order to analyze the effect of the overall change regarding those interactions, the change of engagement intensity around expressed gratitude is calculated.

(26)

export, being the 31st

of October 2014. One other variable that needs to be discussed is relationship quality. Interaction on social media, as has been explained before, can manifest itself through multiple means. However for the majority of cases, likes and comments form the by far the largest parts of their interaction history, and those two means are inherently different regarding the effort and personality that can be put into the ‘piece of interaction’. Therefore, the share of one of those means is used for this metric. Throughout the analysis, the share of comments, instead of the share of likes was found to be a useful predictor regarding the effects measured (for details see Robustness Checks at the end of the results section). Consequently, the share of comments is used as a proxy describing quality of the interaction of a specific user with the respective company fan page.

Table 3: Social Media Database (n=349) Variable Name Calculationab

Available Variables

Name n.a.

Gender n.a.

Relationship Start n.a.

Interaction History n.a.

Expressed Gratitude Date n.a.

Brand ID n.a.

Export Date n.a.

Available Variables Interaction (Before/After

Gratitude),

#  𝑜𝑓  𝐼𝑛𝑡𝑒𝑟𝑎𝑐𝑡𝑖𝑜𝑛𝑠 Interaction per day

(Before/After Gratitude), #  𝑜𝑓  𝐼𝑛𝑡𝑒𝑟𝑎𝑐𝑡𝑖𝑜𝑛𝑠  𝑏𝑒𝑓𝑜𝑟𝑒/𝑎𝑓𝑡𝑒𝑟 #  𝑜𝑓  𝐷𝑎𝑦𝑠  𝑏𝑒𝑓𝑜𝑟𝑒/𝑎𝑓𝑡𝑒𝑟 Relationship Duration at Gratitude Event, 𝐷𝑎𝑡𝑒  𝐺! − 𝐷𝑎𝑡𝑒  𝑅𝑒𝑙𝑎𝑡𝑖𝑜𝑛𝑠ℎ𝑖𝑝  𝑆𝑡𝑎𝑟𝑡 Relationship Quality, #  𝑜𝑓  𝐶𝑜𝑚𝑚𝑒𝑛𝑡𝑠 #  𝑜𝑓  𝐼𝑛𝑡𝑒𝑟𝑎𝑐𝑡𝑖𝑜𝑛𝑠 Engagement change around Gratitude = 𝐼𝑛𝑡𝑒𝑟𝑎𝑐𝑡𝑖𝑜𝑛  𝑝𝑒𝑟  𝐷𝑎𝑦  𝐴𝑓𝑡𝑒𝑟 −   𝐼𝑛𝑡𝑒𝑟𝑎𝑐𝑡𝑖𝑜𝑛  𝑝𝑒𝑟  𝐷𝑎𝑦  𝐵𝑒𝑓𝑜𝑟𝑒

n.a. = not available

a Only stated for calculated Variables. b G

(27)

 

After the calculation of the new variables, a few cases have been taken out for the ensuing analysis due to abnormal values. Values of interaction per day exceeding 1 were indications that the relationship duration after gratitude was shorter than a day, leading to insensible outcomes through variable calculations. Moreover, I already noticed during the coding of the database that some messages have been recorded multiply, sometimes even more than 100 times. This indicates recording errors and has an enormous impact on average values of relevant variables.

3.3.2 Step 2 – Gratitude SocialCRM Sample (n = 48)

This second dataset enables the second goal of the research, investigating the effect of social engagement on conversion behavior. Through the connection of social media and CRM data via linking individual users on both platforms, the exploration of effects that one has on the other becomes possible. In order to create this connection, a list of all user names coded as expressing gratitude (Table 2), has been send to the CRM responsible of the respective company. They matched the list of names with their database and only returned purchase information of those customers in their CRM, making sure they could unambiguously link to a person on the list. At the end of this process, Greetz.nl succeeded in providing a dataset linking both databases and all resulting cases have been used anonymously. In fact, a list of 74 cases showing activity both on social media and in the CRM has been generated, of which 48 cases complied with the sufficient social media interaction criterion and have been used to explore the connection of SocialCRM. For this sample, I have the most advanced information, combining information regarding their behavior around the moment of expressed gratitude on social media, presented in table 3, and purchase related data in the web shop of the company. In more detail, next to the purchase history of the customers, I received anonymized information about age, sales volume, the value of each purchase, the duration of the relationship with the company and was able to calculate comparable change variables indicating the effect gratitude has on their relationship in CRM terms (Table 4).

(28)

result into values of up to 1943 interactions per day, are left out. Such extreme outliers can cause problems, especially with a relatively small sample, as it is the case in this second step.

Table 4: Gratitude CRM Database (Additional Variables) (n = 48)

a Only stated for calculated variables.

4. METHODOLOGY

Like the data section before, the methodology section will be split up according to the two steps. The choice for this structure leads to a better understanding of the research done and will therefore cater for a better understanding of the results presented thereafter.

4.1 Step 1 – Social Media

In order to generate answers for the main research question, I now move towards the generation of insights regarding the factors influencing the effect of gratitude on social media.

Variable Name Calculationa Available Variables

Age n.a.

Relationship Start (CRM) n.a.

Sales History n.a.

Value of Orders n.a.

Time till first order after

gratitude n.a.

Available Variables Orders (Before/After

Gratitude),

#  𝑜𝑓  𝑂𝑟𝑑𝑒𝑟𝑠 Orders per Day

(Before/After Gratitude), #  𝑜𝑓  𝑂𝑟𝑑𝑒𝑟𝑠  𝑏𝑒𝑓𝑜𝑟𝑒/𝑎𝑓𝑡𝑒𝑟 #  𝑜𝑓  𝐷𝑎𝑦𝑠  𝑏𝑒𝑓𝑜𝑟𝑒/𝑎𝑓𝑡𝑒𝑟 Value of Orders (Before/After Gratitude), 𝑉𝑎𝑙𝑢𝑒  𝑜𝑓  𝑂𝑟𝑑𝑒𝑟𝑠 Average Value per Order

(Before/After Gratitude)

𝑉𝑎𝑙𝑢𝑒  𝑜𝑓  𝑂𝑟𝑑𝑒𝑟𝑠  𝑏𝑒𝑓𝑜𝑟𝑒/𝑎𝑓𝑡𝑒𝑟 #  𝑜𝑓  𝑂𝑟𝑑𝑒𝑟𝑠  𝑏𝑒𝑓𝑜𝑟𝑒/𝑎𝑓𝑡𝑒𝑟 Conversion Change around

Gratitude

(29)

In step one, the dependent variable consists of the change in engagement on social media after the expression of gratitude. The model used to analyze the effect of third variables on the effect gratitude has on engagement change, can be described as follows:

1      𝑦 =   𝛽!+ 𝛽!𝑅𝑒𝑙𝐷𝑢𝑟 + 𝛽!𝑅𝑒𝑙𝑄𝑢𝑎𝑙 + 𝛽!𝐼𝑛𝑡𝐵𝑒𝑓 + 𝛽!𝐺𝑒𝑛 + 𝛽!𝐺𝑟𝑒𝑒 + 𝛽!𝑇𝑟𝑎𝑛𝑠 + 𝜀 where

𝑦 = Change  in  Engagement  

𝑅𝑒𝑙𝐷𝑢𝑟 = Indicating  the  duration  of  the  relationship  before  gratitude  in  days   𝑅𝑒𝑙𝑄𝑢𝑎𝑙 = Indicating  the  share  of  comments  regarding  the  interaction  history 𝐼𝑛𝑡𝐵𝑒𝑓 = Indicating  the  number  of  interactions  before  expression  of  gratitude 𝐺𝑒𝑛 = Dummy  variable  for  gender  (baseline  category  is  ′male′)

𝐺𝑟𝑒𝑒 = Dummy  variables  for  Greetz. nl   baseline  category  is  ′Bol. com! 𝑇𝑟𝑎𝑛𝑠 = Dummy  variables  for  Transavia   baseline  category  is  ′Bol. com! 𝜀 = Normally  distributed  error  term  for  dependent  variable    

4.2 Step 2 – SocialCRM

In the second step, the dependent variable is the change in purchase behavior. The model I use, in order to explain the differences in effects gratitude on social media has on conversion behavior, can be described as follows:

2      𝑦 =   𝛽!+ 𝛽!𝐶ℎ𝑎𝑛𝑔𝑒𝑆𝑜𝑐𝑖𝑎𝑙 + 𝛽!𝐴𝑣𝑂𝑟𝑑𝐵𝑒𝑓   + 𝛽!𝐼𝑛𝑡𝑒𝑟𝑆𝑜𝑐𝐴𝑓 + 𝛽!𝐺𝑒𝑛   + 𝛽!𝐴𝑔𝑒 + 𝜀 where

𝑦 = Change  in  conversion  behavior  around  gratitude  

𝐶ℎ𝑎𝑛𝑔𝑒𝑆𝑜𝑐𝑖𝑎𝑙 = Long  term  change  of  engagement  on  social  media 𝐴𝑣𝑂𝑟𝑑𝐵𝑒𝑓 = Indicating  the  average  order  value  before  gratitude 𝐼𝑛𝑡𝑒𝑟𝑆𝑜𝑐𝐴𝑓 = Indicating  the  interaction  per  day  after  gratitude   𝐺𝑒𝑛 = Dummy  variable  for  gender  (baseline  category  is  ′male′) 𝐴𝑔𝑒 = Age  of  the  user

(30)

In both steps the degree of the effect generated by gratitude, on social media, as well as in CRM terms, is the variable of interest. In accordance with the theoretical framework and the following hypotheses, the effect of customer behavior characteristics on those changes and the change in behavior are assessed. Additionally to the ordinary least squares (OLS) regression, insights are generated in the second step by the use of general correlation matrixes in order to detect patterns between individual variables (Malhotra, 2010).

All calculations are performed in the IBM software package SPSS Statistics Version 22.0.

5. RESULTS

After having discussed the data collection, the database cleaning process and the methodology, I will now focus on the presentation of the results. Like the sections before, also the results section is divided into subchapters according to the two core research steps. Nevertheless, the added value of the results regarding the overall research goal will be explained conclusively in every part. This will be done by connecting the results to the hypotheses presented in the theoretical framework and by investigating whether the results do confirm these. Following this results section, the conclusion discusses the possible explanations for the results found by referring back to the literature. Finally, all results are integrated into the general perceptive of this research.

5.1 General Effect of Gratitude

As aforementioned, this research is exploring the moderation effects of customer behavior characteristics on the effect gratitude has on customer engagement, on social media and the effect on conversion behavior. In order to support the theoretical assumption regarding the positive effect of gratitude, which is grounded on multiple researches done in the past, I take a look at the overall trends of engagement on social media and also the trend found in the cases available in the dataset.

To get a feeling for the overall trends in social media data and to interpret changes found around the gratitude event in the data, the overall statistics regarding interaction per user for different periods of time are examined. Therefore, Appendix 1 displays the interaction per day for the time period of the 1st

of January 2012 until the date of the data export 31st of October 2014, based on the total number of users and interactions on the

(31)

presented on a monthly, but also on a weekly and even daily basis. Although especially on a daily basis, great variations can be observed, no long-term trend of increase or decrease of average interaction per day can be detected.

Compared to the aggregate data, the dependent variable is analyzed in greater detail. Following from the theoretical discussion, I would expect the expression of gratitude to have a positive effect on social media engagement. As a consequence, the change in engagement during the analysis should be significantly positive, too. With the help of the one-sample t-test, I check the null-hypothesis assuming that there is no measurable change in customer engagement and therefore, the change variable is equal to 0. As can be seen in table 5, the mean difference (0.02092) is positive and at a 95% confidence interval significantly different from 0. Taking the theoretical discussion and the absence of trends in the aggregate data into account, I can state that in my dataset around the moment of expressed gratitude social engagement of users on social media changes positively.

Table 5: One Sample t-test of Change in Customer Engagement One Sample t-test Mean difference

Change in Customer Engagement 0.02092*

a Significance Levels: * = <0.05; ** = <0.01; *** = <0.001.

In order to provide a complete picture regarding the overall changes in the dependent variables used, again, a one-sample t-test has been used to check for a significant difference of the variable describing the change in conversion behavior around the gratitude expression date (Table 6). Although the mean difference is slightly positive (0.0072), it is not significantly different from zero (p = 0.22). This inability to reject the null hypothesis is potentially effected by the relative small size of this sample (n = 48).

Table 6: One Sample t-test of Change in Conversion Behavior One Sample t-test Mean difference

Change in Conversion Behavior 0.0072

(32)

5.2 Step 1 – Social Media

Next, I take a look at the results of the first regression analysis. In table 7, the model summary of the conducted regression is presented, offering evidence for the overall significance of the model (p-value < 0.001) and a reasonable Adjusted R2

of 0.195. Table 7: Model Summary Step 1

R2 Adjusted R2 F Sig. Model 0.210 0.195 14.055 < 0.001

Three variables individually add significance to the model (p-value < 0.001). Overall, those variables have very low betas, relating to the low values of the dependent variable, as well as the relative low scale of the independent variables (Table 8). In order to still compare the predictive power of the betas, the standardized betas have been added. Looking at the whole table 8, relationship duration until gratitude is the only variable with a significant positive effect on the change in social media engagement (β = 0.0001; p-value < 0.001), confirming hypothesis 2, while keeping all other variables equal. The latter aspect is important, since the number of interactions in general has a negative effect on the change due to customer gratitude (β = -0.001; p-value < 0.001), therefore not confirming hypothesis 1.

Table 8: Parameter Estimates Regression Model Step 1

Variable Coefficient Standardized

Beta Sig. VIF Relationship Duration until Gratitude 0.000 0.375 <0.001 1.135 Relationship Quality - 0.140 - 0.192 <0.001 1.038 Number of Interactions before Gratitude - 0.001 - 0.306 <0.001 1.097 Gender -0.013 -0.038 0.476 1.116 Greetz 0.012 0.034 0.543 1.286 Transavia 0.020 0.059 0.263 1.130 a

DV: Change in Customer Engagement.

(33)

quality has a significant negative effect (β = -0.141; p-value < 0.001), again not confirming hypothesis 3. Gender, as well as the brand membership does not significantly affect the change in engagement on social media.

Next, after discussing face validity by checking the general model summary and the parameter estimates, I check whether the assumptions with respect to the use of the OLS regression and regarding the residuals are all met. Through this process, I ensure that the estimation of the parameters is correct and the conclusions based on that can be trusted.

First, multicollinearity can be assessed through the Variance Inflation Factors (VIF), where values of 4 and higher indicate moderate, and higher than 10 strong multicollinearity (Leeflang, Bijmolt, Pauwels, and Wieringa, 2014). High multicollinearity refers to the degree in which 2 or more variables are correlating in their effect on a third variable. Since all VIF scores are close to 1 in table 8, no multicollinearity between the variables is observed. Second, heteroskedasticity can have an impact, referring to inequality of variance of the residuals throughout the sample. Following Leeflang et al. (2014), heteroskedasticity occurs frequently when cross-sectional data, as is the case in this research, is used. In order to test for heteroskedasticity through the Goldfield Quandt Test, the sample is divided into two groups of either above or below median interaction. Interaction is chosen as a dividing variable since it is a sensible division variable describing two kinds of user groups, namely above and below median interacting users. In a next step, for both samples an individual regression is computed to derive the sum of squared residuals, used in the Goldfield Quandt Test to derive an f-statistic used to test the null hypothesis of homoskedasticity. In this case, as can be seen in figure 2, the according p-value is <0.01 and therefore leads to the rejection of the null hypothesis. 𝐹 =  𝑆𝑆𝑅!/(𝑇 − 𝑘) 𝑆𝑆𝑅!/(𝑇 − 𝑘)= 3.742/(159 − 6) 1.258/(165 − 6)= 3.091   𝑝 < 0.01  

Figure 2: Goldfield Quandt Test

(34)

that group. Accordingly, the not described variance available in the residuals should decrease and ideally disappear.

Table 9: Correlation between Residuals and Total Number of Interaction Before GLS After GLS

Pearson Correlation Coefficient 0.166** 0.154**

a Significance Levels: * = <0.05; ** = <0.01; *** = <0.001.

In this analysis, by following the procedure, the correlation between the predictive variable used to split the sample and the overall residuals decreases but stays significant (Table 9). The calculation of regression based on the transformed variables did not lead to any significant changes regarding the model or the direction of the parameter estimates (Appendix 2).

Finally, normality of the residuals is tested. Through a Normality Plot (Appendix 2) and the results of the Kolmogorov and Smirnov tests of normality (Table 10), non-normality has been detected. Again, Leeflang et al. (2014) mention that with increasing sample sizes even small differences from normality can lead to the rejection of the null hypothesis of normal distributed residuals. Nevertheless, in order to remedy this problem, bootstrapping has been applied when computing the OLS regression in SPSS. However, applying bootstrapping did not lead to any significant changes regarding the estimated parameters and does therefore not change the conclusions drawn from the results presented above. The results of the bootstrapping procedure are summarized in Appendix 2.

Table 10: Test of Normality for Unstandardized Residuals Test of Normality Sig.

Kolmogorov-Smirnov <0.001

Shapiro Wilk <0.001

5.3 Step 2 – SocialCRM

(35)

As already described, a regression has been used to investigate the impact different variables have on the change in conversion behavior around the gratitude expression on social media.

Despite the low sample size (n = 42), the regression analysis did create significant as well as relevant outcomes (table 11; table 12). The change in engagement on social media after gratitude has a significant positive effect on the change in conversion behavior (β = 0.316; p-value < 0.05), indicating a connection between a measurable effect on social media and in terms of conversion, confirming hypothesis 4. Since the interaction on social media after gratitude is partly accounted for through the variable of change in engagement, it is not surprising to see no additional significant effect of that variable (β = 0.100; p-value = 0.489). Furthermore, the average order value before gratitude has a significant negative impact on the change in conversion behavior (β = -0.002; p-value < 0.01).

Table 11: Model Summary Step 2

R2 Adjusted R2 F Sig.

Model 0.336 0.280 4.267 0.004

Table 12: Parameter Estimates of the Model in Step 2

Variable Coefficient Standardized

Beta Sig. VIF Change in Engagement 0.316 0.406 0.014 1.464 Average Order Value

before Gratitude - 0.002 - 0.377 0.008 1.038 Interaction Social after

Gratitude 0.100 0.118 0.489 1.663

Age 0.001 0.154 0.301 1.249

Gender -0.003 -0.011 0.934 1.030

a

DV: Change in Conversion Behavior.

As in step 1, validation with regards to the residuals has been conducted in order to ensure the correct estimation of the parameters.

(36)

as above, again using the variable of number of interactions in order to split the sample in two sensible groups. The f-statistic leads again to a rejection of the null hypothesis of homoskedasticity, making a remedy through GLS necessary. The regression, using the transformed variables, too, did not create any significant changes in the estimated outcomes (see Appendix 3 for details). Finally, normality of the residuals has again been tested through the Kolmogorov-Smirnov and the Shapiro Wilk test, indicating indeed a non-normality of the residuals (Table 13). As has been done in step 2, the bootstrapping procedure has been used to remedy this problem, once again not leading to significant changes in the estimated parameters (Appendix 3).

Table 13: Tests of Normality for Unstandardized Residuals Test of Normality Sig.

Kolmogorov-Smirnov <0.01

Shapiro Wilk <0.001

In order to derive as much additional information as possible from the unique connection of social media and the CRM and give the reader a feeling for the connection between the two datasets, additionally, a correlation matrix between all available and sensible variables has been created. Table 14 depicts some of the most striking correlations found in the data (Appendix 4). Especially important to this research is the aspect that change in customer engagement on social media is positively correlated with change measured in conversion behavior (0.403; p-value < 0.05), confirming the result of the regression presented above.  

Table 14: An Excerpt from the Correlation Matrix (Appendix 4)

Correlating Variables Correlation Coefficient Change Engagement – Change Conversion Behavior + 0.403 **

Change Engagement – Average Order Value - 0.524 **

Change Engagement – Orders after Gratitude + 0.374**

Overall Interaction – Average Order Value - 0.263*

Interaction after Gratitude – Number of Orders after Gratitude + 0.492**

(37)

This is again confirmed by the finding that the amount of interaction after gratitude on social media correlates with the number of orders placed after gratitude is expressed (0.492; p-value < 0.05). Moreover, a negative correlation between overall amount of interaction on social media and the average order value, significant at the 90% confidence interval (-0.263; p-value < 0.10), has been found.

Next, the hypothesized effect regarding the concept of window of opportunity is presented. Taking a look at the discussed effect, I check the frequency table of the variable describing the time it took from the moment of gratitude until the next order was registered in the CRM. As can be seen from the frequency table, one quarter of the users did place an order at the exact same day of gratitude expression, and after 3 days more than 40% have placed an order (Table 15).

Table 15: Frequency Table Regarding Time in Days between Gratitude and First Following Purchase

Days after Gratitude Cumulative Amount of Users Having Placed an Order

0 25%

1-3 41.7%

4-10 56.3%

10-40 83.3%

40+ 100%

After the paragraph discussing the robustness of the results presented, all findings will be discussed in perspective to the overall research question.

5.4 Robustness Checks

During the model building process, certain decisions have been made regarding the choice between variables, the format of individual variables and the inclusion of certain cases leading potentially to different outcome. Three of these decisions are discussed here in order to ensure overall robustness of the results presented above.

(38)

the same, although the overall predictive power of the model decreased. In consequence, I continued with the variables as presented above.

Second, the choice for 10 interactions as the minimum number of interactions to be included in this study may seem somewhat arbitrary. Since additional number of interactions might lead to more accurate findings, 12 interactions have been used as a filter variable, excluding an additional 33 cases from the study. Nevertheless, the observed effects and their direction remained the same and only the f-value decreased to 12.924 from 14.055 as presented in table 8. Hence, the choice of 10 interactions has been made and I am confident that the found results are robust.

Third and finally, the decision has been made to use the share of comments in order to measure this additional quality aspect of the relationship between customer and company. As can be seen in table 16, likes and comments together are the largest groups with nearly 83% of total interaction and therefore a choice between those two is most sensible.

Table 16: Share of Means of Interaction on Total Interaction in Social Media Database Mean of Interaction Share of Total Interaction

Comments 38.28%

Likes 44.71%

Messages 11.93%

Other (Share, Reply) 5.08%

Furthermore, conceptually comments are more interesting since they state something about the quality, relating to the fact that more input is necessary for a comment that is knowingly visible to others, than a click which is sufficient for a like. Statistically, first of all both variables correlated significantly (Pearson Correlation Coefficient = -0.505; p < 0.01), confirming that both describe the same effect. Using either one in the model of step 1, relative share of comments does add more to the overall predictive power and individually is significant. Taking this together with the conceptual reasoning, the share of comments has been used instead of the share of likes.

6. CONCLUSION

Referenties

GERELATEERDE DOCUMENTEN

Legal delay can cause additional work, for instance because more preparation time or more time to read up on the case is necessary or due to the communication regarding the delay.

Therefore companies will tend to catch up to the efficient frontier within an industry as well as making ongoing (frontier shift) efficiencies that would be generated throughout

A consumer is closer to the conversion square when visiting the focus brand’s website, than an information/comparison website or app, a generic search or a competitor’s

In addition, in the first part of the questionnaire, respondents were asked to provide the name of a specific retailer they had a personal omni-channel experience with (using both an

Therefore, it might be expected that these more task-oriented customers buy less in terms of quantity (e.g. only the lamp). At the same time, we can expect that

However, the effect is negative indicating that customers with a higher number of PDP visits compared to total pageviews are less likely to make a purchase and spending

To test if depth of processing, cognitive capacity and product involvement are moderating main effects of both admiration on consumer behavior and extremity of the claim

Conceptual model 29-06-15 | 4 Customer touchpoints Physical stores Online stores Catalogues Mobile phones Touchpoint experience - Disconformation - Positivity