WHY CUSTOMERS WANT MARKETING: AN EMPIRICAL STUDY OF THE ANTECEDENTS OF A
CUSTOMER’S OPT-IN CHOICE
by
Mike van der Woude University of Groningen
Faculty of Economics and Business MSc Marketing
June 2018
Wagnersingel 10b 9722 CX Groningen
(06) 81799694
M.R.van.der.Woude.2@student.rug.nl
Student number 2520443
ABSTRACT
Despite new digital ways for companies to connect with their customer base, customers do not always welcome communications from marketeers. With new regulations coming into effect like the General Data Protection Regulation (GDPR) in the EU, it might become even harder for marketeers to connect with their customers. Therefore, there is a need for knowing customers in how they feel, think and act. When companies know how to connect with the right customers it can increase their performance. Customers are not reluctant to receive marketing communications if they are asked for a consent first, but what influences the customer’s choice to opt-in for promotional messages? This research focusses on the antecedents of a customer’s opt-in choice and brings relatively new concepts like customer experience, customer engagement and the fear of missing out together. A moderated mediation logistic regression was performed with data from 162 respondents to find out which variables influence the opt-in probability. This work found that past customer experience is positively related to customer engagement. Also, when people are subscribed to marketing communications from multiple companies, the probability that this person will subscribe to the communications of another company increases. No evidence was found that fear of missing out and customer engagement influence the opt-in probability.
Keywords: customer engagement, customer experience, emotion, permission marketing, fear of missing out, opt-in
Supervisor: L. De Vries 2
ndsupervisor: H. Risselada
Research theme: Customer experiences and emotions
Table of contents
1. INTRODUCTION ... 3
2. LITERATURE REVIEW ... 7
2.1CONCEPTUAL MODEL ... 7
2.2CUSTOMER OPT-IN ... 7
2.3CUSTOMER ENGAGEMENT ... 8
2.4PAST CUSTOMER EXPERIENCE ... 10
2.5FEAR OF MISSING OUT ... 13
2.6CONTROL VARIABLES ... 15
3. RESEARCH METHODOLOGY ... 16
3.1DATA COLLECTION METHOD ... 16
3.2POPULATION & SAMPLING METHOD... 16
3.3OPERATIONAL DEFINITIONS ... 17
3.4RESEARCH DESIGN ... 17
3.5PLAN OF ANALYSIS ... 18
4. RESULTS ... 20
4.1DATA PREPARATION ... 20
4.2DESCRIPTIVE STATISTICS ... 21
4.3CORRELATION ANALYSIS ... 21
4.4RELIABILITY ANALYSIS ... 21
4.5CONSTRUCT VALIDITY ... 22
4.6ASSUMPTION TESTING ... 24
4.7LOGISTIC REGRESSION MODEL ... 25
5. CONCLUSIONS AND RECOMMENDATIONS ... 28
5.1CONCLUSIONS ... 28
5.2SCIENTIFIC AND MANAGERIAL IMPLICATIONS ... 28
5.3LIMITATIONS AND FURTHER RESEARCH ... 30
REFERENCES ... 32
APPENDIX A ... 38
APPENDIX B ... 40
1. INTRODUCTION
There is a need for knowing customers, and to accomplish this you need to know how they feel, think and act (Martin and Murphy, 2017).
Until the early 1990s, managing customers was based on transaction marketing. Managers then used transaction data to design marketing strategies which increased firm performance (Pansari and Kumar, 2018). It was in the years hereafter that managers came to the conclusion that customers wanted more than only transactions. This, together with the rise of availability of customer data, led to a shift in focus from transaction marketing to relationship marketing in the late 1990s (Berry, 1995; Morgan and Hunt, 1994), to customer relationship management in the early 2000s (Lemon and Verhoef, 2016; Reinartz, Krafft, and Hoyer 2004; Verhoef, 2003).
In the 2000s, companies realized that they had to switch from a product-centric view to an customer centric view in order to improve loyalty and in turn firm performance (Sheth, Sisodia, and Sharma, 2000). Developing close and profitable relationships with customers was key in an environment where factors like advances in technology, competition and better informed customers were on the rise (Shah et al., 2006).
Since 2010, developments in digital and social media enabled customers to engage more with firms and empowered these customers to become active producers or destroyers of value for firms (Lemon and Verhoef, 2016). Kumar (2013) stated that the economy nowadays plays a crucial part in decision-making. He also stated that consumers are extremely cautious about their purchases. Because of this, companies have an increasingly tough task to gain and keep customers, which in turn affects firm performance. Thus, selecting the right customers for your marketing efforts is important to improve firm performance.
Selection of customers for marketing efforts is mostly based on traditional customer-focused
measures like customer satisfaction and customer loyalty. However, these measures have been
criticized in recent literature. Satisfaction fails to: measure the depth of customers’ responses
to consumption situations and service performance; differentiate between true loyalty and
repeat purchasing; provide a reliable predictor of loyalty (Amine, 1998; Bennett and Rundle-
Thiele, 2004; Bowden, 2009; Giese and Cote, 2000; Oliver, 2010). Customer loyalty only
measures repeat purchases of the customer but not what customers do on for example social
media. This can lead to undervaluing or overvaluing your loyal customers in terms of value to
the firm (Pansari and Kumar, 2018). For example, a customer might not be very loyal or
profitable in terms of revenue that (s)he brings to the firm, but (s)he might influence people on social media to buy from your firm. There is also evidence that the linkage between customer satisfaction, customer loyalty and profitability is poorly correlated (Kumar, 2013; Reinartz and Kumar, 2002). Thus, these traditional constructs are not sufficient anymore for selecting profitable customers for marketing efforts. That is why since 2010 a new customer focused construct emerged from the literature, namely, customer engagement (Van Doorn et al., 2010;
Kumar et al., 2010; Vivek et al., 2012). This construct goes beyond the traditional customer focussed constructs since it not only focusses on customer purchases but also on customer’s behavioural manifestations towards the firm that generate value (Kumar et al., 2010).
The customer engagement construct can help in understanding customers because of their behaviour towards the firm. However, connecting with engaged customers can be difficult.
Digital and social media help companies to connect with their customer base but Kumar, Zhang and Luo (2014) state that customers do not always welcome communications from marketeers.
With new regulations coming into effect like the General Data Protection Regulation (GDPR) in the EU, which ensures that customers can only receive targeted marketing communications based on personal data if they give voluntarily permission for this (Tankard, 2016), it might become even harder for marketeers to connect with these customers.
However, customers are not reluctant to receive marketing communications if they are asked for a consent first (Kumar, Zhang and Luo, 2014). The idea of seeking customers’ permission before sending them marketing messages is called permission marketing (Godin, 1999).
Connecting with customers will be easier when a customer agrees on receiving promotional messages. When companies know how to connect with the right customers it can increase their performance. But what influences the customer’s choice to give a firm permission so (s)he receives promotional messages? And what factors motivate customers to not give a firm this permission? Pansari and Kumar (2017) proposed that engaged customers are more likely to opt- in for promotional messages. However, this proposition has not been empirically tested yet while the results can hold important insights for marketeers.
Another reason why customers opt-in for promotional messages might be the fear of missing
out (FoMO), which is a relatively new personality construct which involves reluctance to miss
important (social) information (Elhai et al., 2016). The concept of FoMO is a wide accepted
term among marketing practitioners. However, this construct got more interest in the media
than in scientific literature (Oberst et al., 2017). People that have the FoMO might not want to
miss promotional messages from a firm and are thus likely to opt-in. But the relation of this new construct on a customer’s opt-in decision has yet to be examined.
Previous research emphasized the importance of engaging customers (Brodie et al., 2011).
However, it must be noted that engaging with customers is not always successful. For example, an engagement initiative like the McDonald’s hashtag #McDstories turned out to be an avenue for customers to post their negative stories about McDonald’s, thereby spreading a negative word of mouth. Firms need to empower customers for engagement marketing to be successful (Harmeling, Moffett and Palmatier, 2018). But how could this be done? Calder, Isaac and Malthouse (2013) state that engagement flows from experience. And Pansari and Kumar (2018) acknowledge this by stating that customer engagement occurs after the customer’s initial purchase. To date, no research has clearly shown how the customer experience measure differs from other customer-focused constructs, like e.g. customer engagement (Lemon & Verhoef, 2016).
This research will focus on the proposed research gaps and will examine how customer experience and customer engagement are related. Furthermore this research focusses on how customer engagement influences a customer’s opt-in decision and how customer’s FoMO influences that relationship. The research question is: What is the relationship between customer experience and customer engagement, and how does customer engagement influence customer’s opt-in and how does the fear of missing out influence this last relation?
This paper contributes to the scientific literature in various ways. First, by showing that past customer experience is indeed positively related to customer engagement. Second, by examining the relation between customer engagement and customer’s opt-in probability. Third, by proposing a definition and measurement for past customer experience. Fourth, by making a clear distinction between the customer experience and the customer engagement construct.
Fifth, by introducing the FoMO concept to marketing literature and by examining how this
concept influences the customer’s opt-in probability. A managerial contribution is that
marketing practitioners should focus on the customer experience to enhance a customer’s
engagement. Another implication is that marketing practitioners seem to think that the FoMO
is an important predictor of customer opt-in. However, this research could not find such a
relationship. Thus, marketeers that rely solely on this assumption could make wrong decisions.
This report begins with a literature review and a theoretical framework; next will the research
methodology be explained. Then will the results be presented, after which the conclusions,
scientific implications and managerial contributions will be discussed. The report concludes
with limitations and directions for further research.
2. LITERATURE REVIEW
2.1 Conceptual model
The conceptual model that is used in this study is shown in figure 1.
FIGURE 1 Conceptual model
2.2 Customer opt-in
In 1999, Godin proposed the idea of permission marketing (PM). PM is built on the idea that
marketeers should first get a customer’s permission before sending promotional messages. PM
enables a two-way interaction between marketeers and customers which in turn enhances firm
performance (Kumar, Zhang and Luo, 2014). Media channels that are well suited for a two-
way interaction are web, e-mail, social media and mobile (Winer, 2009). Godin (1999) stated
that the main characteristics of PM are that these communications are anticipated, personal and
relevant. Anticipated because customers gave the permission and thus expect that they can get
promotional messages. Personal, because firms can tailor their marketing communications
based on what they know about the customer. And lastly, relevant because these marketing
messages can contain promotional messages that fit customers’ interests (Kumar, Zhang and
Luo, 2014).
There are two types of PM: Opt-in marketing and opt-out marketing. Opt-in marketing is when customers voluntarily sign up to receive marketing communications, for example signing up for a firm’s promotional emails or following firms on social media. Opt-out marketing is when firms send their marketing communications to customers without asking for a consent first, but where customers have the ability to stop this interaction at any time (Pansari and Kumar, 2017).
This research will focus on opt-in marketing since this is the most used form of PM in practice (Kumar, Zhang and Luo, 2014).
An examination of existing literature on PM shows that a customer’s intention to give permission to marketeers is influenced by gender, income, volume of advertising messages, previous experience with mobile ads (Barnes and Scornavacca, 2008), trust, brand image (Jayawardhena et al., 2009), brand equity, and previous relationships (Tezinde, Smith and Murphy, 2002). Also, people are more likely to opt-in when they experience significant benefits of receiving these promotional messages (Blum and McClellan, 2006). Previous research focussed on the reasons why people opt-out of marketing communications has shown that customers tend to opt-out when they receive highly personalized messages (Marinova, Murphy and Massey, 2002). Furthermore, people who have already opted-in for many different marketing communications of different brands are less likely to opt-in to another (Kumar, Zhang and Luo, 2014).
2.3 Customer engagement
The concept of customer engagement has been widely discussed in the past century with different interpretations in numerous contexts (Pansari & Kumar, 2018). In management literature customer engagement has been defined as “an organizational activity with the internal stakeholders” (Pansari & Kumar, 2017, p. 295). In the business context, customer engagement was termed as a contract between the customer and the firm. Since 2010, customer engagement has been widely discussed in marketing academia (Van Doorn et al., 2010; Kumar et al., 2010;
Kumar and Pansari 2016; Vivek et al., 2012). This discussion led to different definitions of
customer engagement. Van Doorn et al. (2010, p. 254) define it as “customer’s behavioral
manifestations that have a brand or firm focus, beyond purchase, resulting from motivational
drivers”. Vivek et al. (2012, p. 133) define customer engagement as “the intensity of an
individual’s participation and connection with the organization’s offerings and activities
initiated by either the customer or organization”. Although these definitions focus on the psychological and behavioural aspects, it still remains vague what kind of behaviours are customer engagement manifestations and what kind are not. Van Doorn et al. (2010) do name some customer behavioural expressions like cross-buying, word of mouth, customer recommendations and referrals. However, Van Doorn et al. (2010) do not include customer transactions as a part of customer engagement while Kumar et al. (2010) argue that excluding customer transactions makes the definition of customer engagement incomplete.
Kumar et al. (2010) propose that customer engagement consists of four kinds of customer behaviour. First, customer purchase behaviour: repeat purchases or additional purchases through upselling and cross-selling. Second, customer referral behaviour: the acquisition of new customers by current customers through a firm-initiated and incentivized formal referral program. Third, customer influencer behaviour: customers’ influence on other acquired customers as well as on prospects via e.g. social media. And lastly, Customer knowledge behaviour: contributing to knowledge development by providing feedback to the firm. The level of how engaged a customer is depends on the extent this customer exercises these behaviours.
The total value generated by these behaviours to the firm, is called the customer engagement value (Kumar et al., 2010).
Thus, for a more practical idea and a better understanding of what customer engagement entails, the following definition will be used in this article: Customer engagement is the mechanic of a customer’s value addition to the firm through direct or/and indirect contribution, whereas direct contribution entails customer purchasing behaviour and indirect contribution is either customer referral behaviour, customer influencer behaviour or/and customer knowledge behaviour (Kumar et al., 2010; Kumar & Pansari, 2016; Pansari and Kumar, 2017).
A firm’s engaged customers tend to have a heightened trust in the firm or brand (Pansari and Kumar, 2017). As shown by Jayawardhena et al. (2009) trust and brand image enhance the probability of customers to give a firm permission to send them marketing communications.
Also, when a customer feels connected to the company, (s)he is more likely to connect with the
company at free will. This means that engaged customers would have a bigger chance to
proactively download mobile apps of the company, follow firms on social media and engage in
e-mail programs (Pansari and Kumar, 2017). Engaged customers also have repeat purchases at
the firm, which enables firms to fit promotional messages to these customers’ wants, needs and
preferences. If customers are aware of this, they see the benefits of giving a consent to the firm.
It has been shown in past literature that when customers experience the benefits, the probability that these customers opt-in is bigger (Blum and McClellan, 2006). It can therefore be expected that a higher degree of customer engagement leads to a higher probability to opt-in. Based on this, the following hypothesis is proposed:
H1: The higher the customer engagement the higher the probability of a customer to opt-in to the firm’s marketing program.
2.4 Past customer experience
Creating strong customer experiences is a leading management objective nowadays (Lemon and Verhoef, 2016). There are multiple definitions of customer experience in existing literature.
Meyer and Schwager (2007, p. 2) stated that customer experience is “the internal and subjective response customers have to any direct or indirect contact with a company”. Definitions proposed in the past state that customer experience consists of cognitive, sensory, emotional, affective, social, physical and spiritual responses from a customer to an action of a company (Bolton et al., 2014; Brakus, Schmitt and Zarantonello, 2009; Gentile, Spiller and Noci, 2007;
De Keyser et al., 2015; Schmitt, 1999; Verhoef et al., 2009). Lemon and Verhoef (2016, p. 71) built on these definitions and came to the conclusion that customer experience is “a multidimensional construct focusing on a customer’s cognitive, emotional, behavioural, sensorial and social responses to a firm’s offerings during the customer’s entire purchase journey”.
The definition of Lemon and Verhoef (2016) is in line with the five types of experiences
proposed by Schmitt (1999): sense (sensorial), feel (emotional), think (cognitive), act
(behavioural) and relate (social). The sense module is the sensory experience through sight,
sound, touch, taste and smell. The feel module is the customer’s inner feelings and emotions
triggered by an experience. Think refers to the cognitive experiences where customers engage
in, like thinking about the experience. Act is the behaviour that an individual exhibits which is
caused by a physical experience. The relate module contains aspects of sense, feel, think and
act but the difference is that an individual relates about these experiences to something outside
his or her private state.
This work focusses on the past customer experience. The difference with past customer experience and customer experience as defined by Lemon and Verhoef (2016) is that the past customer experience is the sum of all experiences thus far with a firm. For example, when going to a restaurant someone might remember how (s)he felt emotionally (happy or disappointed) and what (s)he thought (the food was good) but might not recall the sensorial or physical aspects of the food after a period of time. Also, the experience might be bad at one moment but overall this person can have a positive customer experience. This has implications for the proposed definition of customer experience by Lemon and Verhoef (2016) which focusses on the experience responses to a firm’s offerings at each touchpoint during the entire customer journey. Therefore, this work proposes some differences to the customer experience definition given by Lemon and Verhoef (2016) in order to define past customer experience.
First, the behavioural and sensorial experiences as explained by Schmitt (1999) are considered a unitary dimension in neurophysiological studies, which means that these kind of experiences are likely to measure the same underlying construct: sensorial experience (Gentile, Spiller and Noci, 2007). Also, psychological studies found that there is a special relationship between sensorial experience and emotions (e.g. Herz, 2002). Sensory characteristics of a product or service evoke emotional responses, like the colour red for Coca-Cola (Brakus, Schmitt and Zarantonello, 2009; Spinelli et al., 2014). In fact, Brakus, Schmitt and Zarantonello (2009, p.
55) state that brand experience scales do not assess the (online) experience of the consumer in the here and now, but rather “assess a lasting trace stored in long-term memory based on multiple exposures to brand-related stimuli”. Based on the above, it can be assumed that this lasting trace manifests itself in emotions felt by the customer towards the company. Thus, when assessing the past customer experience, sensorial and behavioural experiences are captured by the emotions felt towards the company.
Second, Pansari and Kumar (2018) state that customer experience is an outcome of the firm’s
actions and does not include customer’s actions toward the firm. They also state that the
behavioural actions of customers to the firm are outcomes of customer engagement as described
in section 2.3. The social response, derived from the relate module of Schmitt (1999) is about
relating your experiences to others. This is an indirect action towards the firm since it might
influence others and can therefore be valuable. This part of customer experience is very similar
to the customer influencer behaviour, which is part of the customer engagement construct. So
for a better distinction between past customer experience and customer engagement it is better
to not consider the social response as a component of past customer experience but rather as a customer engagement behaviour.
Lastly, Lemon and Verhoef (2016) argue that customer experience is broadening the concept of customer satisfaction. They state that customer satisfaction could be one of the components of customer experience which focusses on the cognitive component of the experience. Hence, the cumulative satisfaction construct, which is the sum of evaluations of a good or service based on the total purchase experience (Anderson, Fornell and Lehmann, 1994), could capture the sum of a customer’s cognitive responses to the firm.
Based on the arguments provided, this work defines past customer experience as a two- dimensional construct focusing on the sum of all customer’s emotional and cognitive responses to all the firm’s offerings. This definition thus focusses on a customer’s emotional response to the experience and how satisfied the customer is with the experience thus far.
Customer experience and customer engagement are closely related constructs. Lemon and Verhoef (2016) state that customer engagement is part of the customer experience. However, Pansari and Kumar (2018) state that these constructs are not the same. They clearly describe customer experience as the actions from the firm to the customer, and customer engagement as the actions from the customer to the firm. Calder et al. (2013) state that engagement flows from experiencing a product. Pansari and Kumar (2017) propose that if customers have a positive customer experience in terms of emotions and satisfaction, they are likely to become engaged customers. Specifically, Pansari and Kumar (2018) state that customer satisfaction leads to purchase behaviour (direct contribution), and that emotional attachment results in referrals, influence, and feedback (indirect contribution). Although this reasoning seems obvious, it might not be that straightforward.
There is an extensive amount of literature that found that customer satisfaction evokes word- of-mouth behaviour (e.g. Anderson 1998), which is an influence behaviour. Also, Aksoy et al.
(2018) found that a customer experience with also negative emotions leads to more beneficial
customer influence behaviours than customer experiences with only positive emotions, which
thus might be caused due to higher levels of customer satisfaction. Furthermore, research found
that negative experiences do not lead to unhappiness or dissatisfaction but are rather seen as
lost opportunities (Nicolao, Irwin and Goodman, 2009; Rosenzweig and Gilovich, 2012). These
findings also imply that customer satisfaction and the emotional attachment formed during the
experience do not have to move in the same way. A negative experience might lower the customer satisfaction but a customer might still feel a certain level of positive emotions, which would explain the findings. An example of this is when someone attends a game for a sports team. Someone might feel positive emotions towards the game but the outcome of the match might be dissatisfying which leads to a negative or neutral experience. These findings implicate that emotions and satisfaction during the customer experience are interrelated.
Pansari and Kumar (2017) also noticed these dynamics and proposed the Customer Engagement Matrix. This matrix is based on the intensity of positive emotions (low-high) and the level of customer satisfaction (low-high). They distinguished four groups: Indifference (low emotion–
low satisfaction), Attraction (low emotion–high satisfaction), Passion (high emotion–low satisfaction) and True Love (high emotion–high satisfaction). They proposed that the Attraction, Passion and True Love groups all exhibit engagement behaviours whereas the True Love group engages the most with the firm.
Based on the findings in previous literature it becomes clear that the sum of emotions and the cumulative satisfaction built during the past customer experience are important predictors of customer engagement behaviours. High levels of customer satisfaction are likely to trigger purchases and influence behaviours whereas high levels of positive emotions are likely to trigger influence, feedback and referral behaviours. What past literature also shows is that low levels of positive emotions and high levels of customer satisfaction are eliciting some of the engagement behaviours. Based on these findings the following hypothesis is proposed:
H2: The better the past customer experience, the higher the customer engagement.
2.5 Fear of missing out
The fear of missing out (FoMO) is a relatively new construct that was first introduced in 2011 (Fake, 2011) and operationalized in 2013 (Pryzybylski et al., 2013). While FoMO is a wide accepted term under marketing practitioners, little can be found about this construct in existing literature (Oberst et al., 2017). In psychology literature, FoMO has been defined as a new personality construct which involves reluctance to miss important (social) information (Elhai et al., 2016). The most used definition of FoMO is proposed by Pryzybylski et al. (2013, p.
1841). They define FoMO as “a pervasive apprehension that others might be having rewarding
experiences from which one is absent, FoMO is characterized by the desire to stay continually connected with what others are doing”. These definitions mostly focus on the social aspect of FoMO while it can be argued that FoMO also has an economic aspect (e.g. missing a discount offer from a firm). In this research, the definition proposed by Pryzybylski et al. (2013) will be used with the acknowledgement that rewarding experiences could also have economic aspects.
Previous research investigated the relation of FoMO on social media engagement (Pryzybylski et al., 2013) and on smartphone addiction (Chotpitayasunondh & Douglas, 2016). However, FoMO has not been researched yet in a marketing setting which is surprising since marketing practitioners have merely adopted this concept.
Kumar, Zhang and Luo (2014) found that customer characteristics influence a customer’s decision to opt-in. Since FoMO is a personality construct, it might be that it influences a customer’s opt-in probability. Kahneman and Tversky (1984) found that people’s monetary losses outweigh monetary gains. Specifically, Seymour et al. (2007) found that financial losses activate the same parts of the brain as physical pain. It is in the human nature to always try to avoid physical pain and thus also to avoid financial losses. Missing a discount of a firm can also be seen as a loss, which based on the above, will be avoided by people. Hence, it is expected that people that have the FoMO are more likely to opt-in for promotional messages. Therefore, the following hypothesis is constructed:
H3: The fear of missing out is positively related to the probability of a customer to opt- in to the firm’s marketing program.
Customers that are engaged with a firm are also involved with a firm since customer
involvement is part of the customer engagement construct (Pansari and Kumar, 2018). Since
engaged customers are likely to exhibit influencer behaviours and thus communicate with
others about the firm, it might be that this is how they find out about a firm’s promotional
messages that they have missed. Also, engaged customers are more interested in news or online
posts about the firm. They might find out about promotions or other rewarding experiences that
they have missed, where others did not. When customers have the FoMO, or see these missed
rewarding experiences as a loss, it is likely that the probability increases to opt-in for
promotional messages. Based on the above, the following hypothesis is proposed:
H4: The fear of missing out strengthens the relation between customer engagement and the probability of a customer to opt-in to the firm’s marketing program.
2.6 Control variables
As stated before, previous research on customer opt-in found that gender, income, (Barnes and Scornavacca, 2008) and trust (Jayawardhena et al., 2009) influence the probability that a customer gives an opt-in. Barnes and Scornavacca (2008) did find that gender and income are important in the decision to opt-in but the direction of the relationships of gender and income still has to be determined. Jayawardhena et al. (2009) found a positive effect of trust on a customer’s decision to give permission to companies. Furthermore, people who have already opted-in for many different marketing communications of different brands are less likely to opt- in to another according to Kumar, Zhang and Luo (2014). However, this has not been empirically tested yet. It might also be that customers opted-in for many communications to many different brands easier give opt-in because receiving communications from one more brand does not matter.
Kumar and Pansari (2018) state that trust is also embedded in the customer engagement construct. Thus, controlling for trust is important since it might affect both the opt-in probability and customer engagement. Also, Chathoth et al. (2014) state that customer characteristics like gender, income and age might influence the customer engagement construct since customer expectations are influenced by these factors. Next, customers that interact with many different companies might have lower customer engagement scores for each company compared to someone that interacts with a few companies. The amount of opt-ins a customer has might be an indicator for this phenomenon and is therefore also incorporated as control variable for customer engagement.
It might thus be important to incorporate trust, gender, income and a customer’s amount of opt-
ins because these variables might affect both the customer’s opt-in chance and the level of
customer engagement. It is also important since not all of the effects of these variables on
customer engagement and a customer’s opt-in probability are assessed yet. This research also
considers age as a control variable for both customer engagement and customer opt-in since
this variable was not incorporated in earlier research.
3. RESEARCH METHODOLOGY
3.1 Data collection method
The data for this research was collected via a social-media-based survey. Websites and mobile apps like Facebook, LinkedIn and Snapchat were used to reach the respondents. A post with the survey was put on these sites to attract the direct friends and connections on these websites.
In the Snapchat app it was possible to reach other people by making a so called ‘story’ that would linger in people’s feed in plain sight for a period of time, which seemed to be an effective way of quickly reaching people. The respondents were asked to ‘like’ and share the post so that it would spread across these social media channels, which led to a snowball effect.
This method was chosen because of the lack of resources to attract many respondents, and because the short amount of time in which this research has taken place. The social-media- based survey in combination with the snowball effect led enabled the researcher to easily reach the respondents in a short amount of time. However, a disadvantage of this method is the decreasing control (Blumberg, Cooper and Schindler, 2014). It is not traceable how the survey spread on these social media sites and which respondents filled it in. Hence, control questions were used in the survey that made sure that people opted out of the survey if they were not eligible for this research. This was done to avoid biased results by respondents not in the population group.
3.2 Population & sampling method
The population of this research were people from the Netherlands that have bought something before. This country was chosen to minimize cultural influences in this research since cultural factors were not measured and thus could not be accounted for.
The sampling methods that are used are snowball sampling and convenience sampling.
Convenience sampling is a cheap and easy method to use but not very reliable (Blumberg,
Cooper and Schindler, 2014). At first, the researcher’s direct social connections where reached
which are mostly all around the same age. This could bias the results. However, snowball
sampling was also used in this research because the respondents reached with convenience
sampling shared the survey on social media with their social connections. This led to more
varied responses and made the sampling more reliable.
3.3 Operational definitions
The operationalization of the constructs in these research are based on previous literature and adjusted to fit this research. All the construct related items were measured on a 7 point Likert scale because this scaling gives the most reliable outcome (Symonds, 1924). Using more than 7 categories would make a slight difference but is not worth the effort and using less than 7 categories would make the measure less reliable.
Past customer experience is operationalized by combining the cumulative satisfaction items from the American Customer Satisfaction Index (ACSI) with 24 items that measure the levels of positive and negative emotions (Oliver, 1993, 2010). Customer engagement is measured by using 16 validated items of Kumar and Pansari (2016) that measure purchase, influence, referral and knowledge behaviour. The FoMO items are based on research done by Przybylski et al.
(2013) and are adapted to fit this research.
Customer opt-in was measured by asking the respondent if he or she was subscribed to email communications of the firm. Respondents could only answer yes or no to this question. This research only considered the opt-in for email communications since not all the companies are active on social media. Thus, for the best results it was decided to only consider email opt-ins.
The control variable trust was measured with items derived from research by Garbarino and Johnson (1999). The items can be found in appendix A. These items were translated to Dutch since the population of this research are people from the Netherlands.
3.4 Research design
The survey consisted of 60 questions, divided into different blocks. Each page contained one
block of questions whereas the blocks were based on this research’s variables. To prevent
people from dropping out of the survey, easy questions like gender, income and amount of opt-
ins were asked first. This caused the progression bar to speed up which gave the respondent the
feeling that the survey was progressing quickly. Then, the blocks with the most questions like
emotions and customer engagement were shown since these required the most effort. Showing
these at the end of the survey would have caused higher dropout rates.
To be able to measure the relationships between past customer experience, customer engagement and customer opt-in it was needed that respondents indeed have these feelings and emotional connections with a firm. Therefore, before showing the questions, respondents were asked to think about a firm where they most recently bought something online during answering the questions. Then there was a question which asked which company the respondent has chosen to create a lock-in so people would not change their company choice when they see the questions throughout the survey. This to prevent ambiguous responses. Also, it was explicitly stated to think about an online purchase since online firms have a higher chance to be active in e-mail marketing compared to physical shops (e.g. local grocery store).
All the items in the survey had to be answered in order to continue, this to avoid missing data.
Since some respondents may be reluctant to provide their yearly income, an extra answer option was added which stated “I’d rather not want to answer this question”. This option was added to prevent people from dropping out of the survey at this point.
3.5 Plan of analysis
For the analyses the Statistical Package for Social Sciences (SPSS) version 23 was used. The data was first checked for inconsistencies and outliers and corrected for these (see section 4.1).
As stated in section 3.3 are the items altered to fit this research and translated to Dutch. To check whether these items are still internally reliable and could possibly be taken together, correlation and reliability analyses were conducted. Reliability analyses give the Cronbach’s alpha. A Cronbach’s alpha of 0.7 and above indicates that the items used are internally reliable (Nunally, 1978).
Next, a principal component analysis (PCA) was performed to check the construct validity of the used items. Assumptions of this method are that all the items are coded in the same direction, that the items factor well and that the items are correlated. To test if the items factor well the Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy must be above .6. If the Bartlett’s test of sphericity is significant it means that the items are correlated. Factor rotation was used to prevent that all the items load on one factor and to improve the interpretation of the results.
The type of rotation that was used is the oblique rotation. This type was chosen because
correlation between items of past customer experience, customer engagement and trust is
expected. As already explained in section 2.4 it is likely that a better past customer experience
leads to a higher customer engagement. Furthermore, Kumar and Pansari (2018) stated that trust is embedded in the customer engagement construct. Thus, using orthogonal rotation techniques like varimax and quartimax let the items of past customer experience, customer engagement and trust load on one factor. However, past customer experience and customer engagement are theoretically different constructs. In an attempt overcome this, the Direct Oblimin rotation method was used to allow correlation between factors.
Based on the results of the correlation analyses, results of the reliability analyses, results of the PCA and existing theory was decided how to aggregate the used items and how the variables were constructed. The standardized factor scores could not be obtained since theoretically different constructs loaded on one dimension in the PCA (see section 4.5). So it was decided to calculate the averages of the items that belong to each construct. This method was chosen because the values will still be easy to interpret and because this enables comparison between variables when there are unequal amounts of items per variable (DiStefano, Zhu and Mindrila, 2009).
The outcome variable customer opt-in was measured on a binary scale (yes or no). Performing a logistic regression analysis is then the most suitable in case of a binary outcome variable.
Since the conceptual model of this research contains an interaction variable (FoMO) and a
mediating variable (customer engagement) it was decided to use Hayes’ Process macro for
SPSS (Preacher, Rucker, and Hayes, 2007). The Process macro enables estimation of a
moderated mediation logistic regression (Hayes, 2017). The conceptual model of this research
is specified by Hayes (2017) as model 14 because of its mediating variable and the interaction
variable on the second mediator relation.
4. RESULTS
4.1 Data preparation
The amount of respondents that started the survey was 232, of which 62 (27%) never finished the survey and dropped out. 8 (3%) responds were deleted because these were ambiguous and highly inconsistent. The results are thus based on the answers given by 162 respondents. Before the analyses were done some adjustments were made to the dataset.
First was the income variable corrected. This variable had an answer option that people could select when they would not like to provide their income details. However, this option made the income variable nominal instead of ordinal. To correct for this, all these values were marked as missing values and imputed with the scale mean. This is a fitting solution when it is the case for less than 10% of the total responses because this would not lead to significant changes in the variable itself. This was done for 12 (7%) responses. The reasons that responds with missing values were imputed is because of the small sample size and because the used Hayes’ Process macro in SPSS is unable to deal with missing values.
Next, there were some outliers in the data. The variable that measures the total amount of customer opt-ins a person has, had two outliers. The values that were given by the respondents were higher than a thousand, which is not very realistic in this setting. These values had a high impact on the scale mean and in turn on the analysis. Therefore, it was decided to delete those two values and to impute these with the scale mean. In this way these responds could be preserved.
Another issue was that the items that measure customer referral behaviour had low scores. An explanation for this is that many companies that were chosen by the respondents did not have any referral programs. Therefore, these answers are not reliable and it was decided to ignore the customer referral behaviour subconstruct in further analyses.
Finally, the items that were asked negatively were recoded so that these items were in the same
direction as the other items and could be used in further analyses. This was also done for the
items that measure negative emotions based on the negative loadings in the PCA (see section
4.5 and appendix B).
4.2 Descriptive statistics
After the data preparation some descriptive statistics were obtained. These showed that 39% of the respondents were men and the other 61% were women. The average age was 28 years old (M
age= 27.68, SD = 11.11) whereas the youngest respondent was 15 years old and the oldest 70 years old. 53% of the respondents had an income below €10,000 per year. The respondents are on average subscribed to the promotional e-mails of 11 different companies (M
amt_optin= 10.51, SD = 9.72).
4.3 Correlation analysis
To check whether the used items could be possibly taken together for each construct, a correlation analysis was done first. A p-value above .05 indicates statistical significance. This was done for each construct derived from theory.
This analysis showed that the cumulative satisfaction items, positive emotions and recoded negative emotions all significantly correlated with each other except for some of the negative emotions with some of the positive emotions (e.g. ‘helpless’, ‘embarrassed’ and ‘self- conscious’). The customer engagement items did all significantly correlate with each other within each of the subconstructs (customer purchase behaviour, customer influencer behaviour and customer knowledge behaviour) but did not all correlate with each other outside the subconstructs. Thus combining the items of these subconstructs into one variable like in the research by Kumar and Pansari (2016) might not be possible. The correlation analysis for the items that measure the FoMO showed that one item did not correlate but that the others did correlate with each other. And finally, the analysis showed that the items that measure trust did all correlate with each other.
4.4 Reliability analysis
The reliability analyses were conducted for each subconstruct and for each main construct.
These analyses showed that the Cronbach’s a’s increased when items from the emotions
subconstruct, customer purchase behaviour subconstruct and the customer influencer behaviour
subconstruct were deleted. Based on these findings and the performance of the same items in
the correlation analysis and the principal component analysis (see section 4.5) it was decided to delete 13 items. The Cronbach’s a’s of the constructs after deleting the items can be found in table 1. As can be observed all the a’s are above 0.7 which means that the remaining items are internally reliable.
TABLE 1
Reliability analysis results
Construct Subconstruct a of sub-
construct
a of construct
Past customer experience Customer satisfaction .91
Emotions .94 .96
Customer engagement
Customer purchase behaviour .88
.84
Customer influencer behaviour .76
Customer knowledge behaviour .82
Fear of missing out
.78
Trust
.89
4.5 Construct validity
A PCA was used to check if the items measure the same underlying constructs. The items are coded into the same direction, as stated in section 4.1. The KMO for this research is .92 which is fine. The correlation analyses in section 4.3 showed that many items are correlated. However, Bartlett’s test of sphericity needs to be significant. The PCA showed that Bartlett’s test of sphericity is significant which means that PCA can be used.
First, an explanatory analysis was done to see how many factor could be extracted. Since it was not clear upfront how many factors there would be due to all the subconstructs and the control variable trust. This explanatory analysis showed that all the communalities where above .4 which is good. Next it showed that 5 factors had an eigenvalue bigger than 1. It also showed that three factors each explained more than 5% of the total variance and that these 3 factors together explained more than 60% of the total variance.
Next, PCA’s were performed with a set amount of factors to indicate which amount of factors
led to the best extraction. As explained in section 3.6, the Direct Oblimin rotation method was
used to allow correlation between factors. However, this method did not lead to significantly different results than using a varimax or quartimax rotation.
The PCA with Direct Oblimin rotation also showed that some emotion items were not performing well and loaded together on a different factor. This probably happened because of translation issues which led to interpretation errors by the respondents. These items were then excluded from the principal component analysis and then added one by one. The emotion items that increased the performance of the principal component analysis and enhanced the internal reliability (section 4.4) were kept. The poor performing items were deleted based on factor loadings, internal reliability (section 4.4), the correlation analysis (section 4.3) and face validity.
These analyses showed that extracting 5 factors led to the best results with the highest loadings.
Whereas extracting more factors would lead to low loadings and extracting less factors would result in forcing constructs into factors which in turn also lowers the factor loadings. The results of the PCA with Direct Oblimin rotation and corresponding Cronbach’s alphas can be found in appendix B.
As can be observed in the results in appendix B, past customer experience, customer purchase behaviour and trust were still loading on one factor despite using an oblique technique. The internal reliability of this factor is high (a = 0,97) but the more items, the higher the reliability tends to be (Field, 2013). According to existing literature these three constructs do not measure the same underlying construct, but the constructs are closely related. Pansari and Kumar (2018) propose that trust is embedded in the customer engagement construct. However, trust is also a control variable for customer opt-in. To be able to see what the effects are of trust on customer engagement and on customer opt-in, and based on existing theory on these constructs it was decided to split this extracted factor up. This factor was split up in a past customer experience part (cumulative customer satisfaction and emotions), a customer engagement part (customer purchase behaviour) and trust.
Also, the items that were used to measure FoMO seem to measure two different constructs. As
discussed in section 2.5, FoMO might have two aspects: social and economic. Looking at the
used items it seems that the first two items measured the social aspect and the other five items
measured the economic aspects of this construct.
So based on theory and the factor loadings in appendix B, 7 variables were computed by calculating the averages of the used items by each construct. These variables are past customer experience, customer purchase behaviour, customer knowledge behaviour, customer influencer behaviour, social FoMO, economic FoMO and trust.
Since customer engagement is a second-order construct (Kumar and Pansari, 2015) a one factor PCA was conducted to check whether customer purchase behaviour, customer knowledge behaviour and customer influencer behaviour all loaded on this one factor. This analysis showed that all variables had loadings > 0.5. The same analysis was done for the social and economic FoMO variables since these variables might also be part of a second-order construct.
This analysis showed that these loadings were also >0.5 on one factor. Thus, consistent with common practice the aggregated scale consisting of the averages of the dimensions of customer engagement and an aggregated scale based on the averages of the proposed dimensions of FoMO were used (Kumar and Pansari, 2015). This resulted eventually in the variables PCX (past customer experience), CE (customer engagement), FOMO (FoMO) and TRU (trust).
4.6 Assumption testing
Since the outcome variable customer opt-in is binary (yes or no) a logistic regression will be used to test the hypotheses. This regression method, like linear regression, knows some assumptions that have to be met in order to get good estimates.
For the computation of the variables no factor scores were used but instead the average of the corresponding items. One disadvantage of this method is that the derived variables can still correlate with each other (Field, 2013). Therefore the variables and control variables were tested for multicollinearity. The variance inflation factors were below 10 and the tolerance was above .1 for all variables, which indicates that there is no multicollinearity (Field, 2013).
Another assumption is that logistic regression assumes a linear relationship between continuous predictors and the logit transformation of the dependent variable (Tabachnick and Fidell, 2007).
The continuous variables and their interactions with the natural logarithm were added to a
logistic regression. What was found is that the interactions of FoMO, customer engagement and
amount of opt-ins were significant. This means that these three variables violate the assumption
of linearity of the logit. Transformation of these variables could make the relationship linear
between these variables and its logit transformations. However, it will only manipulate the variable and will not completely solve this violation. Because of this and for interpretation reasons of it was decided not to transform these variables but to acknowledge that this violation might make the analysis less robust which can lead to less generalizable outcomes (Field, 2013).
4.7 Logistic regression model
The logistic regression was estimated with Hayes’ Process macro. This macro first estimates a linear regression of the predictor variables on the mediator variable customer engagement. After that it logistically regresses the predictor and interaction variables on the dependent variable.
The results of the moderated mediation logistic regression can be found in table 2.
TABLE 2
Results moderated mediation logistic regression
* The odds were only calculated for significant coefficients
As can be seen there is a significant positive effect of past customer experience on the customer engagement (B = .569, p < .05). This means that when someone scores one point more on the past customer experience scale, that his or her engagement goes up with 0.569 on that scale.
The control variables trust, gender, age, income and amount of opt-ins do not show significant effects on customer engagement. However, it is worth mentioning that age and the amount of
Consequent
Customer Engagement Customer Opt-In
Antecedent
Coeff. SE p Coeff. SE p Odds*
PCX a 0.569 0.131 < .001 c' 0.506 0.401 .207 ----
CE ---- ---- ---- b1 0.477 0.491 .331 ----
FOMO ---- ---- ---- b2 0.990 0.894 .269 ----
CE*FOMO ---- ---- ---- b3 -0.170 0.213 .425 ----
TRUST -0.111 0.118 .348 -0.161 0.334 .631 ----
GENDER -0.221 0.147 .135 -0.390 0.429 .364 ----
AGE 0.014 0.008 .085 -0.034 0.024 .153 ----
INCOME 0.025 0.060 .678 -0.054 0.170 .753 ----
AMT_OPTIN -0.015 0.008 .052 0.056 0.027 .042 1.058
Constant i1 1.313 0.470 .006 i2 -1.936 2.117 .360 ----
R2 = 0.305 McFadden R2 = .094
F(6,155) = 11.351, p < .001 Cox and Snell R2 = .099 Nagelkerke R2 = .148
opt-ins are close to statistical significance (p < .10) when regressed on customer engagement.
This could mean that when people are older they become more engaged with companies. Also, amount of opt-ins could indicate that when people interact with more companies, they are less engaged with one specific company.
The results of the logistic regression model show that amount of opt-ins has a positive relationship on the probability that someone has an opt-in (B = .056, p < .05). This means that when the amount opt-ins a customer has increases with one unit, the log odds that someone will subscribe to a newsletter compared to not subscribing go up with 5,8%. No other significant effects were found on the opt-in probability.
Since there is no significant direct effect of past customer experience on customer opt-in and thus no partial mediation, it was assessed whether there is a full mediation. The indirect effect was estimated by taking a thousand bootstrap resamples. This gives the bias-corrected bootstrap confidence interval, which can be found in table 3. This table shows the conditional indirect effects of past customer experience on customer opt-in at different values of the moderator. As can be observed, in 95% percent of the cases there is a negative lower level and a positive upper level of the estimates at different levels of the moderator. This means that zero is in the intervals which implies that there are no significant indirect effects (Hayes, 2017). Since there is no significant direct effect and no significant indirect effect it can be concluded that customer engagement does not mediate the relationship between past customer experience and customer opt-in at different levels of FoMO.
TABLE 3
Conditional indirect effects of past customer experience on customer opt-in at values of the moderator
Indirect effect
FOMO (V) ω = a(b1 + b3V) SE 95% Bias-Corrected Bootstrap CI
1.122 0.163 0.208 -0.275 to 0.568
2.065 0.072 0.177 -0.271 to 0.421
3.008 -0.019 0.247 -0.514 to 0.490
As can be seen in table 4, one of the four hypothesized relationships is supported in this model.
As hypothesized, past customer experience is positively related to customer engagement (H2).
Contrary to the expectation, customer engagement and FoMO are not related to customer opt- in (H1 and H3 respectively). Also, FoMO does not influence the relation between customer engagement and the probability of a customer to opt-in (H4).
TABLE 4 Hypothesis and results
Hypothesis Conclusion
H1: The higher the customer engagement the higher the probability of a customer to opt-in to the firm’s marketing program.
Rejected
H2: The better the past customer experience, the higher the customer engagement. Accepted H3: The fear of missing out is positively related to the probability of a customer to
opt-in to the firm’s marketing program.
Rejected
H4: The fear of missing out strengthens the relation between customer engagement and the probability of a customer to opt-in to the firm’s marketing program.
Rejected
