Master Thesis An investigation of switching behaviour between mobile telephone providers The Drivers of Churning Customers in the Dutch Mobile Telephone Industry

The Drivers of Churning Customers in the

Dutch Mobile Telephone Industry

An investigation of switching behaviour between mobile

telephone providers

Master Thesis

Peter Lammers s3021319

First supervisor:

prof. dr. J.E. Wieringa

Second supervisor:

dr. ir. M.J. Gijsenberg

Final date:

The Drivers of Churning Customers in the

Dutch Mobile Telephone Industry

An investigation of switching behaviour between mobile

telephone providers

Peter Lammers s3021319

University of Groningen Faculty of Economics and Business MSc Marketing Intelligence and Management

Master Thesis

Lissabonstraat 18 9718 AZ Groningen (Gn) E-mail: p.lammers@student.rug.nl

First supervisor:

prof. dr. J.E. Wieringa

Second supervisor:

dr. ir. M.J. Gijsenberg

Final date:

Abstract

The push-pull-mooring (PPM) framework is being used in this study to research the switching behaviour of consumers in the Dutch mobile telephone industry. The framework is a useful tool to predict the reason why people churn from their service provider. Push factors indicate variables that push people away from their service carriers, pull factors express the variables that attract customers to other competitors and mooring factors are inhibitors for switching intentions of customers. Furthermore, the mooring factors can moderate the effects of the push and pull factors on the decision to churn.

The sample consists out of two groups, the group of people who actually switched from provider or extended their contract at their current service carrier and a group of people who stated their switch intentions if their contract expires. We discover evidence in our data that the two groups can be pooled for analysis. In this study, we find sufficient evidence that the attractiveness of alternatives pulls people away from their telephone provide, it also indicates that the pull factors have the strongest impact on switching behaviour. Furthermore, we prove that the mooring factors also have an influence on switching behaviour, but it has a negative relationship. People seem to have a lower tendency to churn if they perceive a lot of inconvenience with switching service providers. The push factors do not have any influence on the decision to churn in this study.

The decision to switch from service provider can be considered as a classification issue, hence we have compared our logit model with several other classification techniques. The item values of the methods indicate that our logistic regression model performs well on the predicting customer churn.

The results have been found in earlier research, however the insignificant result of push factors differs from extant literature. Academic and managerial implication are argued which mainly describes the contribution of this research. Furthermore, some limitations are suggested and future research directions are proposed.

Keywords: Customer churn, switching behaviour, PPM framework, push factors, pull

Preface

Dear reader,

In front of you lies my thesis for the Master Marketing Intelligence and Management at the University of Groningen. My thesis is partly written for an online retailer who sells mobile telephones in the Netherlands.

It has been a long way towards the completion of this report. First, I have completed the bachelor Sportmanagement at the University of Applied Science Hanze in Groningen. During my time at the Hanze, I developed a greater interest in the marketing field. After the completion of my bachelor, I enrolled to the premaster track of the MSc Marketing. When I started the premaster, the focus was on completing the management track. However, from the start of the Master program my interest was more guided towards the Intelligence track. Fortunately, the University of Groningen offers the possibility to combine the tracks and this is my final report of my education.

First of all, I would like to thank Jaap Wieringa for his support and feedback during the last months. Furthermore, my thesis group was of great support and helped me towards the completion of my thesis. Second, I would like to thank my fellow students Wisse Smit and Steven Visser for their insights and encouragement to keep on going. Third, I would like to thank the online retailer for the possibility to work with them to set-up this research. We will likely work together for a follow-up study. Last, but certainly not least, a special acknowledgement to my father, mother, sister, friends, and roommates who have always been supportive throughout my studentship.

I hope you will enjoy reading this report,

With kind regards,

Table of Contents

1. Introduction 1

1.1 Background Information 1

1.2 Relevance 2

1.3 Research Purpose 3

1.4 Contribution to the Literature 4

1.5 Structure of this Thesis 5

2. Theoretical Framework 6 2.1 Customer Churn 6 2.2 Customer Switch 6 2.3 Push-Pull-Mooring Framework 8 2.4 Push Factors 9 2.4.1 Customer satisfaction 10

2.4.2 Perceived service quality 10

2.4.3 Perceived value 12

2.4.4 Trust 13

2.4.5 Commitment 14

2.4.6 Price perception 15

2.5 Pull Factors 16

2.5.1 Perceived attractiveness of alternatives 16

2.6 Mooring Factors 17

2.6.1 Subjective norms 17

2.6.2 Switching costs 18

2.6.3 Variety seeking tendency 19

2.6.4 Prior switching behaviour 20

2.6.5 Attitude towards switching 20

2.7 Moderation Mooring Factors 21

2.8 Control Variables 22

2.8.1 Gender 22

2.8.2 Age 22

2.8.3 Geographical location 22

2.8.4 Education level and income 23

3. Methodology 25

3.1 Mobile Telecommunication Industry in the Netherlands 25

3.2 Selection of Variables 26

3.3 Data Collection 27

3.4 Measurement of Variables 27

3.5 Procedure 28

3.6 Recoding of the Variables 29

3.7 Plan of Analysis 29

3.7.1 Reliability and validity 30

3.7.2 Model estimation 31 3.7.3 Model specification 32 3.7.4 Model comparison 33 3.7.5 Model interpretation 33 3.7.6 Classification techniques 34 3.8 Data Description 35 4. Results 37

4.1 Reliability and Validity 37

4.1 Push, pull and mooring variables 37

4.2 Switch intentions 38 4.2 Pooling of Samples 40 4.3 Model Comparison 40 4.4 Model Interpretation 41 4.5 Classification Techniques 42 5. Discussion 43 5.1 Findings 43 5.2 Academic Implications 44 5.3 Managerial Implications 46

5.4 Limitations and Future Research 47

6. References 49

7. Appendices 76

1

1. Introduction

1.1 Background Information

Customer retention continues to be an important topic to researchers and practitioners in the marketing field. Customer churn prevention is part of the program of customer retention of firms (Burez and Van den Poel, 2008). Churn in a contractual setting is defined as “the termination of a contract between the firm and its customer” (Leeflang et al., 2015, p. 321). Many academics state that managing churning customers is a key challenge in customer relationship management (e.g., Blattberg et al., 2008; Reinartz and Kumar, 2000; Reinartz et al., 2004). Ascarza et al. (2016) indicate that a high churn rate and increasing costs of acquiring new customers can have substantial financial impact on a company. Especially in industries such as banking (Ali and Arıtürk, 2014; Glady et al., 2009; Van den Poel and Lariviere, 2004; Xie et al., 2009), telecommunications (Ahn et al., 2006; Huang and Kechadi, 2013; Kim and Yoon, 2004; Yoon et al., 2014; Lee et al., 2017), television providers (Burez and Van den Poel, 2009), credit cards (Kim et al., 2004; Kumar and Ravi, 2008), online services, and health insurances (Guillén et al., 2012; Günther et al., 2014), the importance of customer retention cannot be overstated.

In some cases, the annual churn rate can be up to 63%, which illustrates the extent to which customer attrition can affect a firm’s customer base (Blattberg et al., 2008). Hence, it can also impact the financial performance of the firm. Focusing on retention of customers is more beneficial for an enterprise than increasing profit margins or lowering acquisition costs (Gupta et al., 2004). Forbes Insights (2011) found evidence that top managers recognise these benefits, stating that the marketing spending is more allocated towards customer retention taking it away from customer acquisition.

2

inhabitants (CBS, 2017). One could say that the telephone market is getting saturated in these developed countries (Kim et al., 2014; Verbeke et al., 2012; Yabas et al., 2012). In fact, Europe is the most penetrated region globally of unique telephone subscribers with 85% in 2015 (GSMA, 2016). In the global telecommunications service sector, the topic of churn is of great concern and it is becoming a more serious problem as the market matures (Ahn et al., 2006). The annual churn rate ranges from 20% to 40% in most of the global mobile telecommunications service companies (Berson et al., 2002; Kim et al., 2004; Madden et al., 1999).

Churning customers typically means that these people switch to another service provider, assuming that they still need a mobile subscription. This means that customers who switch from services also is an important research area in the relationship marketing literature (Chiu et al., 2005). In the last two decades, a significant number of studies examining the impact of consumer switching behaviour (e.g., Antón et al., 2007; Bansal et al., 2005; Lopez et al., 2006; Ranganathan et al., 2006).

1.2 Relevance

3

characterised by a high attrition rate (Reinartz and Kumar, 2003). Therefore, it is critical to understand why consumers decide to switch to other service providers.

This is also the case for the focus enterprise of this research, which is one of the biggest online retailer of the Netherlands on mobile telephone subscriptions. The firm has an online website and several offline stores within the country. It specialises itself in selling subscriptions of the biggest Dutch providers, telephones without a contract and accessories for mobile handsets. Within this firm churning customers pose a daily returning problem, and managers of the firm want to gain additional knowledge about the motivations why people switch. They want to help the customer in all possible ways, also when it comes to finding the best fitting service provider. The focus enterprise has a unique position in the market, since it knows which customers defect and to which telephone provider. Service carriers cannot retrieve this kind of information about the churning consumers. So, the enterprise has the opportunity to obtain a unique position in the market if churn arguments can be better understood.

A possible framework that can investigate the switching behaviour is the push-pull-mooring framework (PPM) (Bansal et al., 2005). The PPM model can be useful to explain customer switching behaviours, which can help marketers in mapping the opposing forces that impact switches within their customer base (Chang et al., 2013). The push effects are the factors that motivate people to leave a company, the pull effects are the positive factors that draw prospective migrants to a firm, and the mooring effects consists of the obstacles that prevent switching from occurring (Bansal et al., 2005; Lee, 1966; Moon, 1995). The relevance of this framework is that it can explain which construct have an influence on switching decisions and which of the three constructs has the biggest influence on switching consumers. We elaborate on the PPM framework in chapter two.

1.3 Research Purpose

4

“Which construct of the push-pull-mooring framework has the biggest effect on the switching behaviour of customers between wireless telecommunication carriers in the Netherlands?”

The question of this study is analysed by using transactional data from the biggest Dutch online retailer of mobile telephone subscriptions to gain insights. We will combine this insights with survey data from a sample of the Dutch population who churned and extended their subscription at their service provider in the last six months or people who did not make a decision in the previous six months.

1.4 Contribution to the Literature

The topic of switching behaviour has been discussed for over a decade within the marketing field (Keaveney, 1995), and has remained relevant since that day. Several studies have explored consumer switching behaviour in marketing (Antón et al., 2007; Bansal et al., 2005; Burnham et al., 2003; Ganesh et al., 2000; Jones et al., 2000; Lopez et al., 2006). In particular, this research tries to solve the problem statement through the lens of PPM model. The PPM framework is a dominant paradigm in human migration literature, which is successfully introduced by Bansal et al. (2005) to fully summarize previous findings in consumers' switching behaviour. As stated before, it describes several antecedents to users’ switching intention, consisting of push factors that drive users away from incumbent service, pull factors that attract users to an alternative option, and mooring factors that either hamper or facilitate the switching decision (Bansal et al., 2005). Ever since, there have been studies using this framework, for example Zhang et al. (2012), Ye and Potter (2011), and Hou et al. (2011).

5

contexts. Within the telephone subscription industry, the extensive PPM has never been used before. This gives us an opportunity to further investigate this framework in this setting. Furthermore, the data from the survey includes two groups: people who have shown actual switch behaviour in the last six months and people who have described their intentions after the expiration of their contract. We investigate if these groups illustrate the same behaviour, which also contributes to the extant literature.

1.5 Structure of this Thesis

6

2. Theoretical Framework

This chapter elaborates on the theoretical background of the concepts that are researched. First, we further describe the concept of customer churn and switching behaviour of customers. Second, we extensively explain the PPM framework by extant literature and we present hypotheses at the end of each construct. Lastly, we summarize our conceptual model in a graphical representation.

2.1 Customer Churn

Churn is a marketing-related term indicating that a consumer is terminating the relationship (Glady et al., 2009). Churn can be seen as an act of the customer where the company was unable to retain him/her. Furthermore, churn can be divided into external and internal motivations (Mattison, 2001). External churn is explained as customer end the relationship due to non-voluntary variables (non-motivational/circumstantial factors, such as death or sickness) or voluntary variables (motivational factors, such as change of service provider or moving out of the country) (Braun and Schweidel, 2011; Kaya and Williams, 2005). Internal churn refers to a change of service, such as moving from prepaid to post-paid. For the purpose of this research, we focus on external voluntary churn whereby the customer terminates its existing service agreement between him/her and its cellular phone company. The service provider can influence the decision of voluntary churn of consumers.

Churning customers can be seen as a failure of the retention department of the firm to create loyalty with him/her. The consumer does not want to stay with the company or finds a better opportunity/deal at another firm.

2.2 Customer Switch

7

(Lopez et al., 2006): (1) the outlined process model of switching decisions (Roos, 1999); (2) the factors that drive customers to switch (Keaveney and Parthasarathy, 2001); and (3) the heterogeneous characteristics between stayers and switchers (Ganesh et al., 2000). Regarding the first stream, the work of Roos (1999) distinguishes three antecedents of switching decisions using the Switching Path Analysis Technique (SPAT). The SPAT comprises of pushing determinants which are reasons to switch to another supplier, pulling determinants which are factors that motivate the customers to come back to the original supplier and “swayers” which are reasons that can only mitigate or strengthen the switching decision (Lopez et al., 2006).

The second stream of understanding customer switching behaviour has received the most consideration in existing literature. One of the pioneers is Keaveney (1995) who identifies eight issues behind customers’ switching decisions in service industries, including core service failures, pricing, employee responses to service failures, attraction by competitors, or inconvenience. Moreover, Gerrard and Cunningham (2004) carried out a similar analysis for bank services, incorporating the weight that customers give to each incident that provokes switching. Similar results are found within their study to the research of Keaveney (1995).

The former streams help to identify the processes and factors that motivate the decision to switch from consumers, however they do not provide any information about the differences between the customers who churn and the consumers who stay. Therefore, the third stream has gained additional attention of academics in the last years. The work of Ganesh et al. (2000) in the financial services sector tried to find significant differences in switching behaviour between the two groups. They found that switchers differ from stayers in aspects such as satisfaction, involvement and loyalty. Keaveney and Parthasarathy (2001), examining online services, associate these differences with aspects relating to attitude, behaviour and socio-demographic characteristics. Our study can be classified within the second stream about finding factors that drive the switching decision of consumers.

8

2.3 Push-Pull-Mooring Framework

The PPM model originates from traditional human migration studies (Lee, 1966; Moon, 1995). In 1885, Ravenstein published “Laws” of migration, in which he observed that human migration is influenced by push and pull effects (Hou et al., 2011; Lee, 1966). Cohen (1996) states that the push-pull model is still one of the most commonly used frameworks to understand human migration. The push-pull model refers to migration as the consequence of the interaction between the push effects at the original place and the pull effects of the destination. Push effects refer to the negative influences that enforce to leave their original place, this could be bad climate, lack of jobs, a lack of natural resources or natural disasters. Pull effects are the positive aspects of the destination that attracts people towards it. This could be the opposite of the push effects, for example, better climate, a lot of work opportunities, an abundance of natural resources and no natural disasters (Lee, 1966).

Bansal et al. (2005) adopted the PPM model paradigm from the human migration literature to explain consumers’ switching behaviour. The authors introduced this pioneering research appealing that customers who switch service providers could be associated to migrant theory of Ravenstein because they switch between physically separated, real-life service providers, which can be considered as belonging to “distinct worlds” (Hou et al., 2011). There are three factors that have an influence on the switching intention of a consumers, namely: (1) the push effects, or the factors that motivate people to leave a company; (2) the pull effects, or the positive factors that draw prospective consumers to a firm; and (3) the mooring effects, which consists of the obstacles that prevent switching from occurring (Bansal et al., 2005; Lee, 1966; Moon, 1995).

9

strongest drivers of switching intentions in the study, followed by the pull factors and subsequently the push factors (Bansal et al., 2005).

After the article of Bansal et al. (2005), a considerable number of articles have been released using the PPM framework to explain customer switching behaviour. Zhang et al. (2008) have interpreted predictors in terms of the effect of push, pull and mooring factors on bloggers’ intentions to switch blog services. Their study results indicate that push effects in terms of satisfaction is the strongest factor affecting switching intentions, followed by pull effects (attractive alternatives) and mooring effects (sunk costs). This is distinctly different from the findings of Bansal et al. (2005) being that the order of influence is just the other way around.

Another study of Chang et al. (2014) investigated switching between social network sites (SNS). Their findings show that all the three antecedents were significant and influence the switching intentions of the SNS users. The PPM model recognizes the moderation role of mooring effects (Moon, 1995; Bansal et al., 2005) which they also modelled in their study. Remarkably, Chang et al (2014) describe the role switching cost on switching intention directly is as strong as its indirect effects, meaning that low switching cost makes it less difficult to switch. The PPM model is also applied in the online gaming world, namely massively multiplayer online role-playing games (MMORPGs). The results of Hou et al. (2011) indicate that the PPM model can be extended to explain the switching intentions of online gamers. The mooring factors seem to have a stronger effect on the player’s switching intention than the pull factors, while the push factors appear to have no effect. All in all, it can be concluded that in each setting the results of the PPM framework can differ.

2.4 Push Factors

10

2.4.1 Customer satisfaction

Customer satisfaction is one of the most critical constructs and a core concept in marketing (Garbarino and Johnson, 1999; Mittal and Kamakura, 2001). Han and Ryu (2006) define customer satisfaction as an essential marketing aspect in that satisfying customers’ needs and wants is necessary to a firm’s success to survive. The overall satisfaction is about the complete judgment of the customer which is affected by the service features/attributes, service/sales personnel performances, or other situational factors (Back and Parks, 2003; Han and Ryu, 2006). According to Oliver (1980), when individuals experience a service and compare the encounter with their expectations, the outcome is customer satisfaction. If the experience and the expectation are in line with each other, the customer will be satisfied. In contrast, if the experience does not meet the expectation, they will be displeased with the firm (Han et al., 2010). An experience possesses both cognitive and affective components (Bitner, 1990; Oliver, 1980; Oliver and Swan, 1989). Satisfaction can also be defined as an overall evaluation of performance based on all prior experiences with a firm (Anderson et al., 1994; Bitner and Hubbert, 1994).

In a service environment, customer satisfaction is one of the drivers of the intention of the customer to stay with or leave the service provider. Therefore, it is a powerful predictor of customer switching behaviour (Burnham et al., 2003). Satisfied consumers incline to have a higher usage level of a service than those who are not satisfied (Bolton and Lemon, 1999; Ram and Jung, 1991). Furthermore, they are more likely to possess a stronger repurchase intention and to recommend the product/service to their connections (Zeithaml et al., 1996). Kim et al. (2004) indicate that mobile carriers should maximize customer satisfaction to enhance the customer loyalty, which is linked to lower customer churn. However, anecdotal evidence reveals that many customers who state that they are very satisfied with a service provider nevertheless subsequently defect (Chandrashekaran et al., 2007).

2.4.2 Perceived service quality

11

indicate that the variables are distinctly different. The difference that is most commonly described is that perceived service quality is a long-run overall evaluation, while satisfaction is more a transaction-specific measure (Bitner, 1990; Bolton and Drew, 1991; Parasuraman et al., 1988).

Although there is still no unanimity about the conceptualization and measurement of perceived service quality (Carman, 1990), in this study we describe perceived service quality as: “the consumer’s judgment about the overall excellence or superiority of a service” (Zeithaml, 1988). Service quality differs from the quality of goods, as services are intangible, perishable, produced and consumed simultaneously and heterogeneously (Grönroos, 1990; Zeithaml and Bitner, 2000). The attributes of services make the evaluation of perceived service quality even more difficult than the evaluation of product quality. This also has to do with the fact that the evaluation may be connected with the service delivery process, along with service output (Cody and Hope, 1999). Service quality is essential and important for a telecommunication service provider company to ensure the quality service for establishing and maintaining loyal and profitable customer (Leisen and Vance, 2001; Zeithaml, 2000).

The literature's position is typified by Parasuraman et al. (1988) description of service quality as "…similar in many ways to an attitude" (p. 15). If one considers service quality to be an attitude, Oliver's (1980) research proposes that: (1) in the absence of previous experience with a service provider, expectations initially define the level of perceived service quality; (2) upon the primary experience with the service provider, the disconfirmation process leads to a revision in the initial level of perceived service quality; (3) following experiences with the service provider will lead to further disconfirmation, which again adjusts the level of perceived service quality; and (4) last, the redefined level of perceived service quality similarly modifies a consumer's purchase intentions toward that service provider (Oliver, 1980).

12

intentions, while Boulding et al. (1993) found positive relationships between service quality and repurchase intentions.

2.4.3 Perceived value

Similar to perceived service quality, perceived value has proven to be a difficult concept to define and measure (Woodruff, 1997; Holbrook, 1994; Zeithaml, 1988). Broadly defined, perceived value is the results or benefits customers receive in relation to total costs. In simpler terms, the value is the difference between perceived benefits and perceived costs. However, what constitutes value appears to be highly personal, idiosyncratic, and may vary widely from one customer to another (Zeithaml, 1988; Holbrook, 1994).

According to Parasuraman and Grewal (2000), perceived value is a function of “a ‘get’ component - i.e., the benefits a buyer derives from a company’s offering - and a ‘give’ component - i.e., the consumer’s monetary and non-monetary costs in acquiring the offering.” Other researchers describe perceived value as a basis of customers’ experiences and seen as a trade-off between benefits and sacrifices (Flint et al., 2002; Grönoos, 2000) or between quality and sacrifices (Monroe, 1990; Ravald and Grönroos, 1999), which can be divided into monetary and psychological sacrifices (Dodds et al., 1991). In this study, we define perceived value as the difference between the perceived benefits and perceived costs of the customer.

Holbrook (1994) indicates that customer value is the fundamental basis for all marketing activity. A high perceived value is one primary motivation for customer loyalty. In this regard, Sirdeshmukh et al. (2002) argue that customer value is a superordinate goal and customer loyalty is a subordinate goal, meaning that it is a behavioural intention. According to goal-and-action-identity theories (Yang and Peterson, 2004), a superordinate goal is likely to control subordinate goals. Therefore, customer value regulates “behavioural intentions of loyalty toward the service provider as long as such relational exchanges provide superior value” (Sirdeshmukh et al., 2002, p. 21).

13

al., 1991; Parasuraman and Grewal, 2002). Anderson and Srinivasan (2003) suggested that when the perceived value is low, customers will be more inclined to switch to competing businesses in order to increase perceived value, thus contributing to a decline in loyalty.

2.4.4 Trust

Trust commonly seen as a vital ingredient for successful relationships in marketing (Berry, 1995; Dwyer et al., 1987; Moorman et al., 1993). Morgan and Hunt (1994, p. 23) conceptualize trust as “the existing when one party has confidence in an exchange partner’s reliability and integrity.” This follows the definition by Moorman et al. (1993, p. 82) who stated the following: “Trust is defined as a willingness to rely on an exchange partner in whom one has confidence.” Both build on the classic view of Rotter (1967, p. 651) that interpret trust as “a generalized expectancy held by an individual that the word of another… can be relied upon.” The importance of confidence is highlighted in all the previous definitions and cannot be ignored when one discusses trust. Prior research on trust suggest that confidence of trusting a party results from the consumer belief that the trustworthy party is reliable and has high integrity, which is associated with such qualities as consistent, competent, honest, fair, responsible, helpful, and lastly benevolent (Altman and Taylor, 1973; Dwyer and LaGace, 1986; Larzelere and Huston, 1980; Rotter, 1971).

Trust can be seen as a set of beliefs held by a customer as to certain characteristics of the company, as well as the possible behaviour of the company in the future (Coulter and Coulter, 2002; Ganesan, 1994). As stated before, extant literature has identified several dimensions in trust. Perceived honesty and benevolence have been mostly associated with consumer trust. On the one hand, honesty describes the certainty the consumer has in the business’ sincerity and the fact that it keeps its promises (Gundlach and Murphy, 1993). On the other hand, benevolence is linked to the customer’s belief that the company is interested in his/her welfare, that it does not intend to show opportunist behaviour (Larzelere and Huston, 1980), and that it is motivated by the mission for joint advantage (Doney and Cannon, 1997).

14

they no longer trust the seller, and they are likely to turn to alternatives for the fulfilment of their needs and desires (Lee, 2014). Thus, firms should invest in the perceived trust of their own company. In this study, we focus on the trust of an individual consumer in the service provider.

2.4.5 Commitment

Within the relationship marketing literature, several researchers recognized another potential driver of customer loyalty or an explanation why people switch from service providers (Bendapudi and Berry, 1997; Morgan and Hunt, 1994). Similar to trust, commitment is recognized as an essential ingredient for successful long-term relationships (Dwyer et al., 1987; Morgan and Hunt, 1994). According to Bowen and Shoemaker (1998, p. 15), commitment is “the belief that an ongoing relationship is so important that the partners are willing to work at maintaining the relationship and are willing to make short-term sacrifices to realize long-term benefits.” Fullerton (2005) defined commitment as “an individual's attachment to a particular target, which results in an inclination to maintain a relationship.” Several other marketing scholars have defined commitment, for example, (1) as a desire to maintain a relationship (Moorman et al., 1993; Morgan and Hunt, 1994), (2) as a pledge of continuity between parties (Dwyer et al., 1987), (3) as the sacrifice or potential for sacrifice if a relationship ends (Anderson and Weitz, 1992), and lastly (4) as the absence of competitive offerings (Gundlach et al., 1995). All these definitions describe why people are loyal to their service provider or not, this can even be the case if their satisfaction is low.

Based on the psychological bond between an individual and a service provider, commitment is composed of three dimensions: affective, continuance, and normative commitment (Allen and Meyer, 1990). In many studies, affective commitment has been used as a single dimension for commitment (Hashim and Tan, 2015; Jin et al., 2010). Affective commitment is defined as an emotional attachment to, identification with, and involvement in the service provider and replicates a wish to stay in the relationship with this firm (Allen and Meyer, 1990). We only include this construct, which is in line with many other academics.

15

and are able to simply pledge choices that are meaningful (Pritchard et al., 1999). Pritchard et al. (1999) found commitment to be strongly correlated with customer loyalty. The same result has been found by Hennig-Thurau et al. (2002), that describe that commitment has a significant and strong direct impact on the loyalty of consumers. In the field of information systems, Hashim and Tan (2015) found that with a lack of affective attachment, users may easily switch to an alternative. Therefore, we expect to find the same effect within the telecommunication market.

2.4.6 Price perception

Price consistently appears as an important conceptual determinant of customer (switching) behaviour (Bolton et al., 2004). The research about prices in various contexts, including microeconomics, psychology and marketing, concludes that pricing explains more variance in customer (re)purchase decisions than other marketing mix variables do (Keaveney, 1995; Winer, 1986). In the research on switching behaviour, Keaveney (1995) also found that price was a critical trigger for consumers to switch from their service provider.

Many academics in marketing suggest that the variability in service performance across different experiences increases customer uncertainty (Han and Ryu, 2009). This uncertainty leads to decreased reliance on prior expectations of the service. From the customer’s point of view, price is often used as a cue in their expectations of the service performance (Dodds et al., 1991). Moreover, customers tend to use price as a prompt in evaluating their experiences with a product/service and in shaping their attitude toward a provider (Bolton and Lemon, 1999; Varki and Colgate, 2001).

16

service price, service benefit perceptions as well as lack of number portability have strong effects on customer retention.

To conclude, many of the previous constructs within the push factors have an effect on the switching behaviour of consumers. These push factors are expulsive forces at the origin that motivate people to migrate (Bansal et al., 2005). All the variables have a negative relationship with switching behaviour. Therefore, we hypotheses the following:

H1: The lower the perceived service quality, satisfaction, trust, commitment to the service

provider and the higher the price perceptions, the higher the likelihood consumers will switch to another service carrier.

2.5 Pull Factors

The pull factors in the literature are described as “the positive factors drawing prospective migrants to the destination” (Moon, 1995) or “attributes of distant places that make them appealing” (Dorigo and Tobler, 1983). Similar to the push factors, these attributes are destination features, not characteristics of the consumer. According to the push-pull paradigm, attractive features at the destination pull the customer to this destination. Bansal et al. (2005) state that the only variable from the service switching literature that confirms this is the perceived attractiveness of alternatives.

2.5.1 Perceived attractiveness of alternatives

17

differentiated services that are difficult for a competitor to match or to provide with equivalents, or if few alternative competitors exist in the market, customers tend to remain with the existing enterprise.

According to PPM framework, the attractive attributes of the destination pull the migrants to this place (Moon, 1995). If the perceived attractiveness of alternatives increases, customers are more likely to be involved in solving problems and less likely to remain loyal to the existing service provider (Hirschman, 1970; Ping, 1993; Rusbult et al., 1982), thus the likelihood of switching increases (Bendapudi and Berry, 1997; Jones et al., 2000; Sharma and Patterson, 2000). Rusbult et al. (1982) observed that the perception of high quality alternatives positively influences exit and negatively influences loyalty. Similarly, Jones et al. (2002) and Yim et al. (2007) showed that attractiveness of alternatives had negative effects on commitment and repurchase intention. Thus, this led to the following hypothesis:

H2: The higher the perceived attractiveness of alternatives, the higher the probability

consumers will switch to another service provider.

2.6 Mooring Factors

While the push and pull factors give a valuable overview of the migration decisions, it does not capture the whole complexity of migration decisions (Bansal et al., 2005). Boyle et al. (1998) specified the following statement: “Any simple comparison between push and pull factors is complicated by the presence of intervening opportunities – obstacles such as family obligations at the origin or the high cost of moving, which may prevent migration occurring (p. 64).” This explains when push and pull factors are strong, one will not migrate in the end. This can be due to situational or contextual constraints (Lee, 1966); these constraints are usually person specific, but they can also operate similarly for a large number of people (Gardner, 1981). Following the research of Bansal et al. (2005), the variables that fit with the conceptualization of mooring effects include switching costs, subjective norms (social influences), attitudes toward switching, past behaviours, and variety-seeking tendencies.

2.6.1 Subjective norms

well-18

established construct from theories of behaviour in the psychology field (Ajzen and Fishbein, 1980). Subjective norm refers to a one’s perception of how one should act or not act in the eyes of another (Ajzen, 1991). In other words, that is, subjective norms reflect perceived social pressure. Combining both assertions, one could argue that subjective norms burdens people to conform, thus leading them to feel as they have to conform or not. In the Theory of Reasoned Action (TRA) and Theory of Planned Behaviour (TPB) models, subjective norm is a direct predictor of one’s behavioural motive (Ajzen, 1991).

In prior studies, subjective norms have been examined as one of the motivations of switching intentions between service providers (Bansal and Taylor, 1999). Commonly, the more favourable the subjective norm with respect to the behaviour in question, the stronger an individual’s intention to perform the behaviour should be (Ajzen, 1991; Ajzen and Fishbein, 1980; Fishbein and Ajzen, 1975). Meta-analytic techniques have found support for this relationship (Sheppard et al., 1988). Other academics also have found similar results (Ajzen and Driver, 1992; East, 1993; Taylor and Todd, 1995). Moreover, subjective norm as a main social factor has been found to affect users' willingness to switch to an alternative (Bansal et al., 2005; Ye and Potter, 2011).

2.6.2 Switching costs

Jones et al. (2002) describe switching costs as barriers that “hold customers in the service relationship.” It refers to the cost related with changing service carriers (Dick and Basu, 1994). It includes economic, psychological, physical, and emotional sacrifices that may occur before, during, and after service consumption (Kim et al., 2006). These costs can explain, perceived monetary costs or non-monetary costs, or a mixture of the two costs (Kotler and Andreasen, 1996). Fornell (1992) indicate that satisfaction makes it tougher for rivals to take away customers from the firm, but switching costs could make it costly for customers to switch from their current service provider. This is in line with the research of Dick and Basu (1994), who state that switching costs can be more critical a factor to customer retention than satisfaction. Customers tend to attribute superior weight to the switching costs when making (switching) decisions.

19

(2003) and Bell et al. (2005) found significant direct effects of switching costs on customers’ propensity to continue the relationship with an enterprise. Switching cost makes changing service providers more expensive (Grønhaug and Gilly, 1991; Peter and Tarpey, 1975). As the costs of switching increases, customers are less likely to change service providers (De Ruyter et al., 1998; Jones et al., 2000; Sharma and Patterson, 2000). This explains why service suppliers expend considerable effort in building switching costs into their marketing strategies (Fornell, 1992; Heskett et al., 1994). The higher the switching cost, the higher the likelihood a customer will prefer the same service provider of another firm with a similar product.

2.6.3 Variety seeking tendency

A variety-seeking tendency in purchase behaviour is defined by Kahn (1995) as “the tendency of individuals to seek diversity in their choices of services or goods” (p. 139). The variation can occur over time, for example, if a consumer chooses different restaurant over several diner occasions. It could, however, be very important if a consumer chooses a portfolio option at one time. McAlister and Pessemier (1982) classified varied behaviour as being either derived or direct. On the one hand, derived variety-seeking behaviour is the result of some other motivations that are not directly related to the need for variety. It occurs by cause of multiple needs at the same time. On the other hand, direct variety-seeking behaviour is defined as resulting from intrapersonal motives. This type of variety-seeking occurs as consumers want to change or they want to experience something new (Kahn, 1995).

20

Bansal et al. (2005) have implemented the need for variety as a mooring factor in the PPM framework. They found that it plays an important role in the consumers’ switching intention, which in turn leads to switching behaviour. In the study of Hou et al. (2005) the need for variety of consumers had also a significant influence on mooring effects and positively on the decision to switch.

2.6.4 Prior switching behaviour

Switching behaviour can be defined as “the extent to which the consumer has switched between providers in the past (Burnham et al., 2003, p. 114).” Experience with another service provider leads to increased knowledge with the service, which reduces the hesitation associated switching to a new provider. More expertise is also related with better developed psychological structures (Alba and Hutchinson, 1987). Past switching behaviour reduces mooring effects by increasing the consumer's familiarity with the process of both switching and learning to use new providers (Alba and Hutchinson, 1987; Nilssen, 1992). Furthermore, the duration of the relationship with their current service provider is less if the consumer has switched more in the past, so they have less time to develop brand and personal relationship bonds. These consumers are less likely to distinguish the relationship with their service provider as unique (Bhattacharya et al. 1995).

In the migration theory, prior switching experience also has an important role. On the on hand, if a person has had a successful migratory experience in the past, one will be more likely to switch again in the future compared with one who has no experience at all (Kuznets and Thomas, 1984). On the other hand, if one had an unsuccessful migratory experience this may have the opposite effect. The person will have a bad experience with switching and will be more likely to stay with the current service provider. It has been found earlier that a consumer’s prior switching experience impacts subsequent switching behaviour (Ganesh et al., 2000).

2.6.5 Attitude towards switching

21

H3: The probability that consumers will switch service provider is lower when switching

costs are perceived higher, consumers' need to seek variety is lower, consumers' attitudes and subjective norms of the environment toward switching are less approving, and the consumer

has not switched frequently in the past.

2.7 Moderation Mooring Factors

Bansal et al. (2005) state that: “applying the PPM model to a service context goes beyond its ability to structure a long list of predictor variables into theoretically defined factor categories (p. 103).” It gives the opportunity to research the moderator effect of the migration decision. Prior migration literature shows that mooring factors do not only have a direct effect on the decision to migrate, but can also moderate the effects that the push and pull factors have on the decision to migrate (Lee, 1966). The perceived benefits of switching resulting from dissatisfaction with the current service may be less than the perceived cost of switching when the perceived switching costs increase. On the one hand, if the mooring factors are perceived as high it will enlarge the effects of push and pull factors on the decision to churn. On the other hand, if the mooring variables are perceived as low it weakens the effects of the push and pull factors on the decision to churn.

So, in addition to the direct effect of pull and push factors on switching intentions, mooring factors also moderate the relationship between push and pull factors and switching intentions of the consumer. According to the PPM framework, a consumer has a high probability of remaining with the current service provider when mooring factors are strong, even if push and pull factors are also strong (Bansal et al., 2005; Boyle et al., 1998).

H4: Mooring factors moderate the relationship between push effects and intention to switch

from mobile service carrier. The stronger the mooring variables, the weaker is the relationship between push effects and intentions to switch.

H5: Mooring factors moderate the relationship between pull effects and intention to switch

22

2.8 Control Variables

2.8.1 Gender

Gender has attracted research interest on the purchase behaviour of people (Jasper and Lan, 1992; Slama and Tashlian, 1985; Zeithaml, 1985). Within the topic of customer churn, prior studies found that women are more involved in purchasing activities than their male counterparts (Homburg and Giering, 2001; Slama and Tashlian, 1985). Melnyk et al. (2009) present a contrasting finding and describe that the difference between men and women is highly associated with the object that churning behaviour is concerned with. Men are likely to be more loyal to groups and companies, and women occur to be more loyal to individuals. The findings of this research indicate that women are more likely to churn, because they are less loyal to companies than males (Melnyk et al., 2009). A study by Ranganathan et al. (2006) has also included the variable gender in their research. They suggest that technological facilities are more masculine, and females may show more anxiety toward the usage of mobile technologies (Brosnan and Davidson, 1996; Igbaria and Chakrabarti, 1990). Gilbert et al. (2003) found a significant effect that this anxiety suppresses the switching behaviour for females. Thus, we think it is advisable to control for this variation in the model.

2.8.2 Age

Age is another demographic characteristic that has attracted considerable research attention (Homburg and Giering, 2001). Wong (2011) found that the desire of customers to churn decreases when the customer will increase in age. Research comparing young and older consumers has focussed on the differences in the information-processing abilities needed to evaluate a product (John and Cole, 1986; Smith and Baltes, 1990). Most of these studies conclude that information processing declines with age (Gilly and Zeithaml, 1985). Older people have restricted information-processing capabilities; therefore, their responses to satisfaction alterations might also change (Homburg and Giering, 2001). Thus, we argue that age should be added as a control variable.

2.8.3 Geographical location

23

more loyal than subscribers living in urban areas due to the fact of the limited number of telecom providers in rural areas. However, this study in the Netherlands is not really comparable to the study in Turkey since the country is much smaller and most of the service providers provide coverage above the 90% within the Netherlands (Figure 1). Nonetheless, there could be a difference in customers selecting a service provider. We included the geographical location within our analysis as a covariate.

Figure 1: Geographical coverage of the biggest telephone providers (www.4gdekking.nl)

2.8.4 Education level and income

It is assumed that people with a high education will achieve a higher income from their work. Therefore, we conclude that these two characteristics related to each other (Homburg and Giering, 2001; Zeithaml, 1985). They usually use more of the information processes prior to the purchase decision (Schaninger and Sciglimpaglia, 1981), and their choice is essentially based on the evaluation of the information given to them (Homburg and Giering, 2001). Decision processes are essentially based on the evaluation of information. Hence, with increased cognitive capabilities customers with higher income and higher education levels have a higher chance to switch from service provider. For that reason, we use education and income as control variables.

2.9 Conceptual Model

25

3. Methodology

In this chapter, we elaborate on the methodology of the research. It includes the description of the mobile telecommunication industry of the Netherlands, selection of the variables, data collection, measurement of the variables, procedure, recoding of the variables model estimation, plan of analysis and, lastly, the data description.

3.1 Mobile Telecommunication Industry in the Netherlands

The Dutch mobile telecommunication market is controlled by three leading companies who manage the network for themselves and rent it to several smaller service providers. These alternative providers do not own a telephone network, but pay a premium to make use of their network. The three leading enterprises are KPN, T-Mobile and Vodafone. The Dutch KPN runs the network for Simyo and Telfort, Vodafone sells it right to hollandsnieuwe and T-Mobile does the same for Ben. According to Autoriteit Consument & Markt (ACM), KPN is the biggest mobile provider of subscriptions with 30 - 35% of the market. The Dutch company is followed by Vodafone with 20 - 25%, T-Mobile is on 15 - 20%, Tele2 has 0 - 5% of the market and other service providers without an own network (i.e. Ben, Telfort and hollandsnieuwe) have a combined market share of 20 - 25% within the industry.

KPN and Vodafone can be classified as premium service providers, since they ask the highest price for their subscriptions within the Dutch wireless telecommunication industry. KPN also offers a package where consumers can combine television, internet, telephone at home and mobile subscriptions. Vodafone has combined its forces with Ziggo to offer a similar deal as KPN. The consumers who sign a contract can receive a discount on their subscription and double the number of megabytes for mobile Internet use. It can be defined as Quad Play offers, which is a term that has come to be used in the telecommunications industry to describe a blend of voice, video and data together with mobile services. It offers a range of services in a seamless environment which encourages subscribers to stay with a single service provider, thereby as an intention to reduce churn and opening up new revenue-generation potential (Comverse, 2007).

26

Telfort, Simyo and hollandsnieuwe are more focused towards the low-priced segment, since they all offer subscriptions for a low price with a downgraded network and content of the subscription. These providers differ from the others, since they do not offer the latest models but target more on the economically-priced mobile telephones.

The mobile telecommunication market can be divided into residential and business customers. However, we exclude the demand for business customers who use mobile communications in another way than consumers do. Unlike residential customers, business users often do not themselves make the decision to sign or extend a mobile subscription contract. Within those companies, there is mostly a department who is responsible for purchasing these subscriptions and mobile phones. To reduce complexity, therefore, the analysis only focuses on the residential customer segment.

3.2 Selection of Variables

The focus industry of this study, the mobile telecommunication industry, differs from the study of Bansal et al. (2005) since they have researched switching behaviour within the auto-repair service and hairstyling services industries. So, it is legitimate to state that customers in the telecommunication market can have other reasons concerning their decision to switch. For consumers who can extend their contract or churn from their cellular mobile provider, it could be important to retain their telephone number. All the service providers in the Netherlands offer the possibility to consumers to retain their mobile telephone number. However, the consumer has to indicate that he/she wants to make use of this option.

27

3.3 Data Collection

The data for the present study was obtained via Qualtrics which is offered by the University of Groningen as a tool for questionnaires. On December 7, 2017, the questionnaire started and the last reply was received on December 14, 2017. The respondents for the survey were gathered using the snowball sampling technique, also known as chain referral sampling. The method has been widely used in qualitative sociological research. A study sample comes through referrals made among respondents who share or know other that possess some characteristics which fit the research interest (Biernacki and Waldorf, 1981). It is the best-known approach of chain referral sampling (Goodman, 1961).

3.4 Measurement of Variables

The variables within the survey are based on extant literature, which have measured these constructs before. Multi-item scales are employed to measure most variables in order to adequately capture the domain of the constructs (Churchill, 1979; Nunnally, 1978). The questions for several items are altered slightly from previous studies to fit with the specific research setting. In total, the survey contained 50 questions with fourteen underlying variables. The queries were developed in Dutch for the respondents to answer. Most of the items of the constructs are being converted into a 7-point Likert-scale ranging from “strongly disagree” (1) to “strongly agree” (7), except for the variables satisfaction, attitude towards switching, switching intentions and demographic items of the consumers.

28

The mooring effects considered in this study included perceived switching costs, subjective norms, prior switching behaviour and variety seeking tendencies. Switching costs are assessed using the frequently used scale of Ping (1993). Within this study, the switching costs refer to perceived switching costs (Morgan and Hunt, 1994) as it is not any objective cost that will be measured, but rather the switching costs as perceived by customers (Burnham et al., 2003). The construct of the need for variety is replicated from the research of Bansal et al. (2005), which subtracted their scale from the Acquisition of Products Scale of Van Trijp et al. (1996).

The measure for subjective norms was captured from the study of Taylor and Todd (1995). Past switching behaviour was assessed by adapting the two items from East’s (1993) study. The variable number retention is added to the model for the purpose of the mobile telecommunication market. It has been constructed for this research and has not been used before. Lastly, the pull factor, the measurement of attractiveness of alternatives was applied from the research of Ping (1993) and Jones et al. (2000). All the questions of one track within the survey can be found in Appendix 7.1.

3.5 Procedure

To ensure that the questionnaire is reliable, viable and consistent all the queries were checked by five fellow students and my first thesis supervisor. In that way, we safeguard that the survey was distributed as complete as possible. The validation has been done in the week of November 27, 2017. Based on their critics the questionnaire was slightly altered in the structure, the way some questions were asked and some spelling errors were removed. Furthermore, a more extensive introduction to the survey was added to inform respondents what the subject of the study is. We made sure that the respondents were not informed about the hypotheses of the study, but they could read more general information about the questionnaire.

29

did churn the questions were asked about their previous mobile operator and if they extended their contract they were asked about their present service provider. However, if respondents did not churn or extended their contract they are guided towards the track with more general questions about their current mobile operator.

The difference between the latter and the former is that these respondents were given a question if they are likely to churn if their contract expires. The last part of the survey is about the demographics of the respondent. Lastly, it is appropriate to state that the questionnaire had several incentives to participate for respondents. If a respondent filled in their email address they could be selected for several rewards.

3.6 Recoding of the Variables

In the previous chapters, the factors are described as either positive effect or negative effect on the decision to churn. The push and mooring factors expressed as negative effects on the switching behaviour of consumers, while the pull factors represent a positive effect on the decision to churn. Therefore, the variables are coded in their corresponding effect. Therefore, the valence of questions within the push factors are coded negatively that they resemble the service provider pushing customers away. The variables should be interpreted as low perceived service quality, low satisfaction, low perceived value, low perceived trust, low perceived commitment, high price perception and low content of the subscription.

The mooring factors are also negatively coded since they should reflect an inhibitor for switching. Higher scores of these variables should be interpreted as unfavourable attitude toward switching, thus unfavourable subjective norms toward switching, high switching costs, infrequent prior switching behaviour, low variety-seeking tendencies and high need of number retention. Lastly, the pull factors are positively coded as they should reflect the positive features of the new service provider.

3.7 Plan of Analysis

30

model is compared with other classification techniques. All the techniques for the analysis are described in paragraph 3.7.1 until paragraph 3.7.6.

3.7.1 Reliability and validity

First, a Kaiser-Meyer-Olkin is performed to see whether the questions within the survey can be used for a factor analysis. The minimum score of .50 is considered necessary for the Kaiser-Meyer-Olkin score (Frohlich and Westbrook, 2001). Second, the Bartlett’s Test of Sphericity intends to provide the same information, it tests whether a factor analysis is suitable for the data. The test should show a significant value to assume that a factor analysis is appropriate (p <.05) (Bartlett, 1950). Third, a confirmatory factor analysis (CFA), otherwise referred to as restricted factor analysis (Hattie & Fraser, 1988), structural factor analysis (McArdle, 1996), or the measurement model (Hoyle, 1991) is performed. The CFA is typically used in a deductive mode to test hypotheses regarding unmeasured sources of variability responsible for the commonality among a set of variables. It is contrasting to exploratory factor analysis (EFA) which addresses the same basic question but in an inductive, or discovery-oriented mode, while a CFA is theory driven (Hoyle, 2000; Schreiber et al., 2006).

To assess the CFA analysis, multiple fit indexes are reported. Browne and Cudeck (1993) and March (1994) state that chi-square (χ2_{) may be an unsuitable assessment of the model fit with}

large sample sizes. Thus, we will use four additional, commonly used fit indexes: χ2_/df

(Wheaton et al., 1977), Tucker-Lewis Index (Tucker, 1973), root mean square error of approximation (RMSEA; Steiger, 1990), and Comparative Fit Index (CFI; Bentler, 1990). For a respectable model fit the χ2 would provide an insignificant result at a p < .05 threshold (Barrett, 2007). The index χ2/df has no acceptable ratio, the recommendations for this measure

range from as high as 5.0 (Wheaton et al., 1977) to as low as 2.0 (Tabachnick and Fidell, 2007). A threshold of around .95 is recommended for the Tucker-Lewis Index (Hu and Bentler, 1999).

31

Fourth, to assess the validity of the variables, factor loadings, as well as the squared multiple correlations between the items and the constructs were examined (Bollen, 1989). Bagozzi and Yi (1988) propose a threshold value of .60 for factor loadings as acceptable to infer convergent validity. For the squared multiple correlations, item values above .40 are reflective of a considerable shared variance with their corresponding variables (Taylor and Todd, 1995).

Fifth, we investigated the measure reliability of the variables by using Cronbach’s alpha. The method of Cronbach ensures for the internal consistency reliability of the scale and is a function of the number of items in a test, the average covariance between the pair of items and the variance of the total score (Tavakol and Dennick, 2011). The value of .70 is commonly used as a cut-off criterion for the measure (Nunnaly, 1978).

3.7.2 Model estimation

The logit-model seems to be the most used model within the subject of churning customers (Günther et al., 2014; Leeflang et al., 2017; Lemmens and Croux, 2006). As the model is relatively simple to describe and interpret, and still shows good performance (Leeflang et al., 2017). Furthermore, it is robust and the parameter estimates are interpretable in terms of odds ratios and marginal effects (Günther et al., 2014). Several researchers have used logit models, for example, Ahn et al. (2006), Brockett et al. (2008), Burez and Van den Poel (2007), Kim and Yoon (2004), Lemmens and Croux (2006), Mozer et al. (2000) and Neslin et al. (2006). The logit model describes whether a customer does or does not do something, it is also referred to as the basic choice model. There are various decisions that can be classified as a choice, such as cancelling a contract, buying a product, or switching from service provider. The two most popular models are the logit and probit model. In both models, it is assumed that an unobserved, latent, variable 𝑈_𝑖 drives the buying decision of individual 𝑖. This latent variable can be interpreted as the (indirect) utility of the product (Leeflang et al., 2017). The utility follows a linear description:

𝑈_𝑌𝑖 = 𝛽₀+ 𝛽₁𝑥_𝑖+ 𝜀_𝑖, 𝑖 = 1, … , 𝑁

where xi is a vector of characteristics of the product or customer, α is the intercept, 𝛽 is a vector

32

utility 𝑈_𝑖 and the decision 𝑌_𝑖 are linked to each other by the following rule (Leeflang et al., 2017):

𝑌_𝑖 = {0 if 𝑈𝑌𝑖 ≤ 0

1 if 𝑈𝑌𝑖 > 0

3.7.3 Model specification

The conceptual model of the theoretical framework (Figure 2) is translated into the following equation for the logistic regression technique:

𝑈_𝑌𝑖 = 𝛼_𝑖 + 𝛽₁𝑃𝑈𝑆𝐻_𝑖 + 𝛽₂𝑀𝑂𝑂𝑅_𝑖 + 𝛽₃𝑃𝑈𝐿𝐿_𝑖+ 𝛽₄(𝑀𝑂𝑂𝑅_𝑖𝑃𝑈𝑆𝐻_𝑖) + 𝛽₅(𝑀𝑂𝑂𝑅_𝑖𝑃𝑈𝐿𝐿_𝑖) + 𝛽₆𝐺𝐸𝑁_𝑖 + 𝛽₇𝐴𝐺𝐸_𝑖 + 𝛽₈𝐼𝑁𝐶_𝑖 + 𝛽₉𝐸𝐷𝑈_𝑖+ 𝛽₁₀𝐴𝑅𝐸𝐴_𝑖 + 𝜀_𝑖

𝛼 Intercept

𝛽 Parameter

𝑈_𝑌𝑖 The utility that respondent 𝑖 will churn

𝑃𝑈𝑆𝐻_𝑖 Push factors, including: Perceived service quality,

satisfaction, perceived value, subjective norms, perceived trust, perceived commitment, and content of the subscription for respondent 𝑖

𝑀𝑂𝑂𝑅_𝑖 Mooring factors, including: Perceived switching costs, attitude towards switching, prior switching behaviour, variety seeking, and number retention for respondent 𝑖 𝑃𝑈𝐿𝐿_𝑖 Pull factors, including: Attractiveness of alternatives for

respondent 𝑖

𝐺𝐸𝑁𝑖 Gender for respondent 𝑖

33

3.7.4 Model comparison

Several models are being tested to ensure that the best predictive model is described for further analysis. For the comparison of the models AIC, BIC, loglikelihood values and the loglikelihood ratio-test (LR-test) can be used for logit or probit models. AIC and BIC are predictive measures that use a correction penalty for the number of parameters that are used in the model (Burnham and Anderson, 2004). Furthermore, the BIC includes an even higher penalty for the number of parameters and in addition also a penalty for the number of observations that are used to estimate the model. The lower the AIC and BIC score, the better it performs in terms of predictive validity (Burnham and Anderson, 2004). The loglikelihood values should also be as low as possible for the best predictive power.

The LR-test is a statistical test used for comparing the goodness of fit of two statistical models. Normally, the null model (without explanatory variables) is compared with an alternative model with explanatory variables. It is based on the likelihood ratio, which indicates how many times more likely the data corresponds with the alternative model than the null model (Leeflang et al., 2015). As stated before, we have computed several models to compare the different predictive measures. The following three models have been established:

Model 1 – PPM Framework including interactions and control variables

𝑈_𝑌𝑖 = 𝛼_𝑖 + 𝛽₁𝑃𝑈𝑆𝐻_𝑖 + 𝛽₂𝑀𝑂𝑂𝑅_𝑖 + 𝛽₃𝑃𝑈𝐿𝐿_𝑖+ 𝛽₄(𝑀𝑂𝑂𝑅_𝑖𝑃𝑈𝑆𝐻_𝑖) + 𝛽₅(𝑀𝑂𝑂𝑅_𝑖𝑃𝑈𝐿𝐿_𝑖) + 𝛽₆𝐺𝐸𝑁_𝑖 + 𝛽₇𝐴𝐺𝐸_𝑖 + 𝛽₈𝐼𝑁𝐶_𝑖 + 𝛽₉𝐸𝐷𝑈_𝑖+ 𝛽₁₀𝐴𝑅𝐸𝐴_𝑖 + 𝜀_𝑖

Model 2 – PPM Framework including interactions

𝑈_𝑌𝑖 = 𝛼_𝑖 + 𝛽₁𝑃𝑈𝑆𝐻_𝑖 + 𝛽₂𝑀𝑂𝑂𝑅_𝑖 + 𝛽₃𝑃𝑈𝐿𝐿_𝑖+ 𝛽₄(𝑀𝑂𝑂𝑅_𝑖𝑃𝑈𝑆𝐻_𝑖) + 𝛽₅(𝑀𝑂𝑂𝑅_𝑖𝑃𝑈𝐿𝐿_𝑖) + 𝜀_𝑖

Model 3 – PPM Framework

𝑈_𝑌𝑖 = 𝛼_𝑖 + 𝛽₁𝑃𝑈𝑆𝐻_𝑖+ 𝛽₂𝑀𝑂𝑂𝑅_𝑖+ 𝛽₃𝑃𝑈𝐿𝐿_𝑖 + 𝜀_𝑖

3.7.5 Model interpretation