Determinants of attitudinal and behavioral loyalty in the
banking sector
An empirical study on the investigation of the effects of determinants of attitudinal and behavioral loyalty in the banking sector of the United States of America.
By: Enrico Aarten University of Groningen Faculty of Economics and Business
Department of Marketing MSc Marketing Intelligence
June 2019
Supervisor: A. Bhattacharya Second supervisor: P. S. van Eck
Nieuwe Deventerweg 77 8014 AD Zwolle
06-12574184
2 ABSTRACT
Customer loyalty has been widely known to be an important topic within the marketing literature. For markets in which the competition is strong, customer loyalty can help firms to retain customers. Therefore a strong understanding of the influences that affect customer loyalty is crucial. Previous studies have examined the role of loyalty determinants on customer loyalty but mainly looked at customer intentions instead of customer behavior. However, it is often stated that intention can be used as a proxy for behavior. One may
wonder whether such an assumption applies. Therefore, this study aims to provide insight into the different effects that affect attitudinal loyalty and behavioral loyalty. Furthermore, this study examines the predictive power of several machine learning techniques that are used to predict behavioral loyalty. In order to gain insight into the different effects on attitudinal and behavioral loyalty, data is analyzed from the Bank of the United States of America that consists of 1279 observations, which includes behavioral and attitudinal data. Results show that, within the attitudinal context of loyalty, perceived switching costs, satisfaction, and trust have positive effects on attitudinal loyalty. Two positive interaction effects were found, such that satisfaction positively moderates both perceived switching and trust on attitudinal loyalty. While in the behavioral context of loyalty perceived switching costs and satisfaction only seem to have positive effects, no interaction effects were found. Consequently, for both types of loyalty, no significant mediation effect was found of engagement. Moreover, the findings show that satisfaction has the strongest impact on attitudinal loyalty, while perceived
switching costs has a stronger effect on behavioral loyalty, indicating that a firm should focus on expanding their switching costs cautiously. In addition, decision tree classification models show to have the best predictive power in predicting churners. These models showed superior results over traditional logit models.
PREFACE
In front of you lies my master thesis ‘Determinants of attitudinal and behavioral loyalty in the banking sector’. I have been working on this thesis for one semester as part of the final phase of my master in Marketing Intelligence. I am glad that I was able to implement a lot of my analytical skills which I have developed during several courses of master study. I enjoyed writing this thesis because it gave me the opportunity to dive deeper into an interesting topic regarding customer loyalty and to come up with models that were able to predict churn. But, I am also happy that this journey comes to an end.
I especially want to thank my supervisor Abhi Bhattacharya for his excellent guidance and supervision during the process of my master thesis, but also for making time to discuss the results and to provide feedback. Besides, I also want to thank Yao Zhang, who had kind of the same topic as me, to discuss the use of data and the methods that we could apply.
Furthermore, I want to thank Peter van Eck in advance for reading and evaluating my thesis. Lastly, I want to thank my friends and family for supporting me during this process.
I hope you will enjoy reading my thesis.
4 TABLE OF CONTENT
1 Introduction ...6
2 Theoretical background ...8
2.1 Customer loyalty ...8
2.2 Differences between attitudinal- and behavioral loyalty ...8
2.3 Influence of perceived switching costs on customer loyalty ...9
2.4 Influence of trust on customer loyalty ... 10
2.5 Mediation effects of engagement between PSC and trust on customer loyalty ... 11
2.6 Interaction effects of satisfaction ... 13
2.7 Conceptual model ... 15 3 Methodology ... 16 3.1 Sample ... 16 3.2 Measures ... 17 3.3 Methods ... 17 3.4 Model assumptions ... 18 4 Results of regressions ... 20
4.1 Preacher and Hayes procedure ... 20
4.2 Mediation analysis and conditional indirect effects ... 20
4.3 Direct effects on attitudinal loyalty ... 21
4.4 Direct effects on behavioral loyalty ... 21
4.5 Conditional directs effect on attitudinal loyalty ... 22
4.6 Conditional directs effect on behavioral loyalty ... 23
4.7 Model fit ... 24
5 Results of machine learning techniques ... 25
5.1 Logistic Regression ... 25
5.2 Support Vector Machine ... 25
5.4 Bagging ... 26 5.5 Boosting ... 27 5.6 Random Forest ... 27 5.7 Neural Networks ... 28 5.8 Naïve Bayes ... 28 5.9 Decision Tree ... 29
5.10 Comparison of the machine learning techniques ... 29
6 Discussion, Limitations and further research... 31
6.1 Theoretical implications ... 31
6.2 Managerial implications ... 34
6.3 Limitations and further research ... 35
7 References ... 36
8 Appendix ... 41
6
1 INTRODUCTION
Customer loyalty is an important topic in the field of relationship marketing and has widely been recognized as an important aspect of marketing (Reichfeld and Sasser 1990; Dick and Basu 1994; Jones, Mothersbaugh, & Beatty 2000). It is widely known that loyalty is
important for retaining customers, acquire new customers through a positive worth of mouth (PWOM), to boost profits and to keep customers pleased. In essence, the concept of loyalty is about the importance of retaining current customers. Thus, in order to retain customers, one has to investigate variables that are related to customer loyalty. In this study, a closer look is given to a set of variables such as engagement, trust, satisfaction, and perceived switching costs, which are seen as relational indicators of customer loyalty.
Many studies have tried to uncover the most important determinants of customer loyalty. (Jones et al., 2000; Bansal and Taylor 1999; Matos, Henrique, & Rosa 2009) have investigated the relationship between switching costs and customer loyalty. All authors state that switching costs are seen as a key driver in creating customer loyalty. The relationship between trust and customer loyalty also has been widely recognized in earlier studies (Moorman, Desphande, & Zaltman 1993; Doney and Cannon 1997; Pamies 2012). Trust is seen as a key indicator for customer loyalty in the long term and in times of uncertainty such as an economic downturn. Satisfaction is also seen to be the most important driver for
customer loyalty, the impact of satisfaction, compared to other determinants is the greatest in predicting customer loyalty. In addition, satisfaction as a moderating variable can
significantly influence customer loyalty through determinants such as perceived switching costs and trust (Lee, Lee, & Feick 2001; Yang and Peterson 2004; Lam, Shankar, Erramilli, & Murthy 2004; Aydin, Özer, & Arasil 2005).
Preliminary studies looked at customer loyalty in terms of behavioral customer intentions, while not investigating actual behavior. This research focuses specifically on the effects that influence the intentions of customers and examines whether these effects also influence the actual behavior of customers. Previous literature state that intentions can be used as a proxy for actual behavior (Ranaweera and Prabhu 2003; Kumar and Shah 2004; Auh, Bell, McLeod, & Shih 2007). However, the findings of this study show that there are
about. This study is also interesting for marketing managers and practitioners. It gives practitioners and managers new insights into the impact of different variables related to customer loyalty which can be used to optimize their strategy regarding customer loyalty programs depending on the goal firms want to pursue. No empirical studies have investigated these relational variables in a single framework by investigating both components of loyalty. Understanding the determinants of customer loyalty in a behavioral way is essential for firms to get insights into customers that switch to competitors. The most important contribution of this study emphasizes questioning assumptions made by previous papers regarding factors that influence customer loyalty. Little research is known regarding actual customer behavior, thus such an assumption may lead to misguided resource allocation by firms.
In addition, prior research in terms of customer loyalty in a behavioral way have not used machine learning techniques to predict customer churn in the banking industry. This research makes use of advanced methods to integrate nine machine learning techniques in order to detect when customers switch. Most importantly, variables that show insignificant results in the logit model, seems to be the most important variables in several machine
learning techniques. Also, compared to traditional logit models, several other techniques show to have greater predictive power in predicting customers that churn.
The outline of this thesis starts with a literature review regarding the determinants of customer loyalty, based on that hypotheses and a conceptual model are formulated. Two separate studies have been conducted, the first study aims to get insight into how covariates influence both types of customer loyalty by making use of multiple linear- and logistic regression analyses. Via this way, it can be compared to what extent the covariates influence both types of loyalty. The second study aims to compare several machine learning techniques following a set of criteria in order to assess which classification model is best in predicting actual churners. Consequently, the results of the studies are discussed, followed by providing theoretical and managerial implications as well as directions for future research and
8
2 THEORETICAL BACKGROUND
This chapter consists of a discussion of the literature regarding the topic of this study. The theoretical background is used to provide information in order to formulate a conceptual model. All variables, that are being used in the model, are discussed as well as hypotheses regarding their relationships are explained.
2.1 Customer loyalty
Customer loyalty has received a considerable amount of attention in marketing theories. The concept is positively related to a firm’s probability, such that when an organization is able to retain five percent more of its current customers, profits increase by 25 percent up to 125 percent (Reichfeld and Sasser 1990). Lots of companies invest in acquisition while retaining current customers is less resource intensive and may lead to higher profits. This makes customer loyalty an important topic to conduct research on.
Customer loyalty is difficult to define because it consists of many different approaches. Traditionally, customer loyalty is seen as a form of a repeated purchase of particular services, products or from a specific organization (Newman and Werbel 1973). A more recent study of Tellis (1988) refers to customer loyalty as a dedication towards a particular brand or
organization, which in turn leads to a repeated buying behavior of the same company or brand. Even more recently, Dick and Basu (1994) created a framework for customer loyalty, where the authors divided customer loyalty into two components. These components are subtracted from the definitions given by Newman and Werbel (1973) and Tellis (1988). The first is an attitude component, in which different feelings create the individual connection of an individual to a product, service or organization. Which can be seen as customers intentions towards loyalty. Which is congruent with the definition given by Tellis (1988). The latter is a behavioral component, which manifests itself in making repeat purchases of products,
services, brands or from a specific organization. Which is congruent with the definition given by Newman and Werbel (1973). Within this paper, attitudinal loyalty is referred to the
dedication of a customer towards their bank. While behavioral loyalty is referred to as a stayer of the bank they are currently in.
2.2 Differences between attitudinal- and behavioral loyalty
particular brand or organization, in other words ‘preference loyalty’. Whereas behavioral loyalty is actual repurchasing or simply committing a relationship with an organization in other words ‘purchase loyalty’. The main difference lies in the fact that attitudinal loyalty is about intentions in relation to behavioral loyalty that is about actual behavior. According to previous studies (Ranaweera and Prabhu 2003; Kumar and Shah 2004; Auh, Bell, McLeod, & Shih 2007), attitudinal loyalty is correlated with behavioral loyalty and may be used as a proxy for actual behavior. Such that attitudinal loyalty leads to behavioral loyalty. However, it remains a grey area such that customers state to be loyal to a certain brand or organization, whereas in reality customers may not be actual loyal. Preliminary studies regarding customer loyalty have looked only at the attitudinal aspect of customer loyalty, whereas the behavioral component reflects the actual behavioral aspect of customer loyalty. Hence, this study investigates both components of customer loyalty.
2.3 Influence of perceived switching costs on customer loyalty
Switching costs can be seen as costs in terms of both monetary and non-monetary that a customer perceives and that is associated with switching from one provider to another (Jones et al., 2000). In other words, it encompasses costs that can be measured monetarily as well as non-monetary such as time and effort that is involved in finding a new provider and ending an existing relationship with the current provider (Dick and Basu 1994).
Preliminary research states that switching costs is an important predictor for customer loyalty (Jones et al., 2000; Bansal and Taylor 1999; Matos et al. 2009). A review of the literature suggests that higher switching costs are positively related to customer loyalty, such that when switching costs are perceived as being high, customers are less likely to switch. Switching costs and customer retention is closely related to each other, in a way that the perception of switching costs a customer perceives determines their ability in terms of ease to switch providers. In other words, if customers perceive a high level of switching costs, they are less likely to switch service providers. This can be explained by the fact that high
switching costs discourage customers to change providers due to time, money, and energy one has to invest in finding and setting up a new provider as well as possible cancellation fees one has to take into consideration. This, in turn, makes switching costs an important determinant in relationship marketing.
10
not find a significant direct effect of switching costs on actual switching behavior. The authors conceptualized perceived switching costs as being equivalent to the construct of perceived behavioral control in the theory of planned behavior, which encompasses the facilitating conditions/switching costs component. Showing that the construct of perceived behavior control of switching costs failed to influence switching behavior directly.
The marketing literature lacks empirical evidence for supporting the distinct influence of perceived switching costs on attitudinal and behavioral loyalty. Even though this link has been proposed by Lam et al. (2004), no empirical evidence was found to support their proposition. The study of Matos et al. (2009) examined the relationship between switching costs on customer loyalty in the banking industry and found that switching costs are positively related to both attitudinal loyalty and behavioral loyalty. However, these authors
conceptualized switching costs at three different costs, namely procedural, financial, and relational. Such that switching costs are a multidimensional construct. This study combines switching costs, in a way that switching costs are seen as a unidimensional construct. Based on these arguments the following hypotheses are proposed.
H1a: Perceived switching costs have a positive effect on attitudinal loyalty. H1b: Perceived switching costs have a positive effect on behavioral loyalty.
2.4 Influence of trust on customer loyalty
Trust is conceptualized in considerably different ways. Trust, from a marketing perspective, has been defined as “a willingness to rely on an exchange partner in whom one has
confidence” (Moorman et al., 1993). A breach of trust can lead to adverse effects for a provider. Doney and Cannon (1997) and Moorman et al. (1993) argue that trust is relevant in situations where uncertainty plays a role. Whenever vulnerability arises for customers, trust can help to reduce uncertainty and in situations as such that customers feel safe about making decisions by their current service provider. It makes risks manageable, which simplifies choice. Thus, the concept of trust may be an important determinant of customer loyalty in difficult times, such as times of economic downturn. Furthermore, trust is in the service marketing literature seen as a necessity for long term relationships (Berry, 1995; Grönroos, 1995; Morgan and Hunt, 1994). Trust involves that one believes the other party is able to conform its implicit and explicit promises and that it is willing to do so.
agency sector and found that trust is directly related to customer loyalty. However, the author only looked at the relationship between trust and behavioral loyalty and did not examine the attitudinal component of loyalty. Moreover, the results suggest that trust has a lesser impact on customer loyalty compared to other determinants such as satisfaction. Suggesting that trust is related to customer loyalty, but that it has compared to other loyalty determinants a
relatively low impact. The relationship between trust and customer loyalty has also been widely supported by earlier papers (Garbarino and Johnson 1999; Singh and Sirdeshmukh 2000; Chaudhuri and Holbrook 2001; Sirdeshmukh, Singh, & Sabol 2002; Leninkumar 2017). However, these papers have measured customer loyalty in terms of attitudes, behavioral intentions or a combination of both. This research distinguishes oneself by examining actual switching behavior as well as customers intentions instead of solely examining a customer’s intentions. Based on the findings the following hypotheses are proposed.
H2a: Trust has a positive effect on attitudinal loyalty. H2b: Trust has a positive effect on behavioral loyalty.
2.5 Mediation effects of engagement between PSC and trust on customer loyalty Several engagement concepts have been covered in the literature. For example customer engagement behaviors (Doorn et al. 2010), customer engagement marketing (Harmeling, Moffett, Arnold, & Carlson 2017), customer brand engagement (Hollebeek, Glynn, & Brodie 2014) and the concept of customer engagement (Verhoef, Reinartz, & Krafft 2010; Pansari and Kumar 2017;). For the purpose of this paper, the focus is on the concept of customer engagement, since the definition of engagement is context specific. Kumar et al. (2010) define customer engagement as: “active interactions of a customer with a firm, with prospects, and with other customers, whether they are transactional or non-transactional in nature”. Given this definition, engagement in this study is used as a proxy for all interactions one has with their current service provider.
Limited empirical research has been done in the area where engagement functions as a mediator between determinants of customer loyalty. Consequently, the study of Hussein (2016) investigates the mediated effect of engagement between trust and customer loyalty and found that engagement has a full mediation effect. However, no explanation has been
12
have trust in their current provider they are more willing to make use of those services. Secondly, if customers use services of their provider more frequently this will ultimately lead to a higher level of customer loyalty, because of the interaction between the customer and the provider. This interaction creates a bond between both parties. Suggesting that when an organization aims to build a relationship with its customers in terms of retaining loyal customers it should pay attention to how to engage these customers. However, no firm conclusions can be drawn from a single study investigating this phenomenon. The paper of Hussein (2016) conceptualized engagement as a multidimensional construct, whereas this study examines engagement as a unidimensional construct with the amount of interaction as a proxy for engagement. Moreover, the authors only looked at the attitudinal component of customer loyalty whereas this study extends the relationship by also taking the behavioral component of loyalty into account. In addition, the author did not look into the relation link between satisfaction and loyalty as well as not taking other determinants of loyalty, such as switching costs and trust into account. Based on these arguments the following hypotheses are proposed.
H3a: Engagement mediates the effect of trust on attitudinal loyalty. H3b: Engagement mediates the effect of trust on behavioral loyalty.
The second mediation effect that this study explores is the mediation of engagement such that the effect of perceived switching costs on customer loyalty goes via engagement. Even though engagement is often mentioned in preliminary research to be a covariate for customer loyalty (Banyte and Dovaliene 2014; Thakur 2016; Fernandes and Esteves 2016). No studies have, to date, investigated this mediation phenomenon. It is expected that the higher the engagement of the customer is, the more loyal these customers are. This can be explained by the fact that the level of perceived switching costs directly related is to the level of engagement. If
H4a: Engagement mediates the effect of PSC on attitudinal loyalty. H4b: Engagement mediates the effect of PSC on behavioral loyalty.
2.6 Interaction effects of satisfaction
The influence of satisfaction on loyalty is frequently examined. According to Matos et al. (2009) satisfaction has a direct effect as well as a strong impact on loyalty. Consequently, studies have investigates the interaction effect of switching costs on satisfaction and loyalty (Lee, Lee, & Feick 2001; Yang and Peterson 2004; Lam, Shankar, Erramilli, & Murthy 2004; Aydin, Özer, & Arasil 2005). Lam et al. (2004), proposed that customers stay with their current service provider under high switching costs, regardless of the level of satisfaction. Empirically these authors did not find evidence that supported their hypothesis. It is proposed that customers stay with their current provider when they are satisfied with their provider regardless of their perception of switching costs. The perceived amount of switching costs will be neglected because customers are satisfied with their current provider, thus seeking alternative options will less likely to be taken into consideration. However, when customers are dissatisfied with their current provider then perceived switching costs will play a more important role regarding customer´s loyalty. This can be illustrated with three examples: if the perceived switching costs are low and customers are dissatisfied about their current provider then switching to another provider would be seen as logic behavior since there is no barrier to switch. Furthermore, when perceived switching costs are perceived as high and satisfaction is low, one will retain switching because there is a barrier in terms of monetary and
non-monetary costs. Also, when perceived switching costs are perceived as high and satisfaction is high, one will not be likely to switch because customers are satisfied regarding their current provider, even if customers want to switch it will retain them from switching because of the switching barriers.
There is empirical support for switching costs as a moderator of the relation link
14
relationship between perceived switching costs on both components of loyalty. These studies have an important assumption and state that satisfied customers have a greater tendency to engage in behavioral intentions that are more favorable to them. Satisfied customers tend to have a greater resistance when experiencing a decline in the performance of a particular service than unsatisfied customers. However, the above-mentioned studies investigate perceived switching costs to be a moderator, this study investigates the interaction effect of satisfaction on the perceived switching-customer loyalty link. In essence, the direction of the relationships does not chance, since an interaction effect is simply a combination of two variables without a specific order. Satisfaction as a moderator in the perceived switching costs-customer loyalty link relation is important to investigate because previously mentioned empirical results suggests that this link between may depend on the level of satisfaction which affects the strength and/or the direction of that relationship in a way that it may amplify or weaken the overall effect. Thus, it is important to take these interaction effects into account when explaining both behavioral and attitudinal loyalty. Based on previously mentioned empirical results it is suggested that satisfaction moderates the perceived switching costs and customer loyalty link. Thus, based on these findings the following hypotheses are proposed.
H5a: The positive effect of perceived switching on attitudinal loyalty is higher with decreasing levels of customer satisfaction.
H5b: The positive effect of perceived switching on behavioral loyalty is higher with decreasing levels of customer satisfaction.
rather than exiting the relationship. It is expected that customers who trust their providers to have confidence that their concern(s) will be addressed judiciously by the provider and that the provider will work to resolve the issue(s). Thus, satisfaction may have an important moderating role in the relationship between trust and customer loyalty. Based on this assumption, the following hypotheses are proposed.
H6a: The positive effect of trust on attitudinal loyalty is higher with increasing levels of customer satisfaction.
H6b: The positive effect of trust on behavioral loyalty is higher with increasing levels of customer satisfaction.
2.7 Conceptual model
Based on the literature review the following conceptual model has been proposed. It is assumed that perceived switching costs and trust have a direct effect on attitudinal and behavioral loyalty. Moreover, it is assumed that these relationships are mediated by
engagement, such that a proportion of the direct effects goes via engagement, in other words the indirect effects. It is assumed that satisfaction interacts with the direct effect of perceived switching costs and trust on both loyalty components.
3 METHODOLOGY
This chapter consists of the design that is used for this study. First, the characteristics of both churners and non-churners are described. Secondly, the measures are described which entails how the variables are defined in the dataset. Lastly, an overview of the different methods used in this study is given.
3.1 Sample
The dataset was assembled using customer panel representatives of the entire banking industry of the United States of America and includes survey responses for a set of banks in the United States of America for one year, measured in quarterly waves. In the analysis, only individuals include those who participated in two or more waves in order to observe the attitudinal and behavioral loyalty. After removing respondents who only participated in one wave, a total of 1,280 observations remained. One observation who had no values on all aspects of the dataset has been removed. Thus, a total of 1279 observations remained. Of these 1,279 observations, 148 observation switched banks (churners) in a time range of one year, whereas 1,131 did not switch banks (non-churners).
The average age of the observed churners in the dataset is between 51 till over 70. With a total annual income of household income of $30,000 to less than $35,000 on average. These churners can be qualified as being mass consumers. Most of these churners are females (50.68%), while 49.32% of the observations are men. When looking at education, on average the churners have an associate’s degree to a bachelor’s degree. In terms of race, 91.9% is Caucasian followed by 6.1% to have an Asian race.
The average age of the observed non-churners in the dataset is between 51 till over 70. With a total annual income of household income of $30,000 to less than $35,000 on average. These non-churners can be qualified as being mass consumers. Most of these non-churners are males (51.85%), while 47.00% of the observations are female. Some of the respondents within the non-churners did not indicate their age. When looking at education, on average the non-churners have an associate’s degree to a bachelor’s degree. In terms of race, 89.7% is Caucasian followed by 4.4% to have an Asian race.
3.2 Measures
Although different conceptualizations of customer loyalty has been used in earlier papers, varying in complexity, this study focuses on customer loyalty in two different ways. The first one is the attitudinal component of loyalty, which is conceptualized as the level of dedication a customer has towards their current bank. However, this variable is corrected for biases in a way that first the bias is estimated and secondly the estimator is modified by subtracting the estimated bias from the initial estimation. The second loyalty component is behavioral loyalty, which is conceptualized as customers who switch providers during the observation period of one year. In other words, behavioral loyalty encompasses customers who switch from banks during the observation period.
Perceived switching costs is conceptualized as the costs that customers associate with the process of switching from banks. The higher the value, the higher the perceived switching costs are. This has been calculated by deriving the residuals from the regression between customer loyalty and satisfaction. Another covariate is trust, which is measured as the perception of customers in regard to the trustworthiness of their primary bank in terms of reliability, which is measured on a scale from 1 to 10. The interaction variable and the covariate satisfaction is measured as the satisfaction level of all services a customer uses at their primary bank, which is measured on a scale from 1 to 10. However, a bias-corrected variable has been used in the same way it is used for the conceptualization of attitudinal loyalty. The mediator engagement in this study is used as a proxy for all interactions one has with their current service provider, in other words, the sum of all channels that are being used by the customer. These channels consist of: ‘ATM usage’, ‘branch or banking center usage’, ‘telephone customer service’, ‘website visits to conduct online banking’, ‘website visits to conduct online trading or check accounts’, ‘website visits for reason other than online banking’, ‘received e-mails or bills by the bank’.
[Table 2 about here]
3.3 Methods
First, it was tested whether the dataset contained multicollinearity, by calculating VIF-scores. In order to interpret the interaction effects, the variables that are connected to the interaction effect have been mean centered. Based on this, perceived switching costs, trust, and
18
attitudinal loyalty, two linear regressions were conducted. One for the direct effects and one for the interaction effects. For the dependent variable behavioral loyalty, two logistic
regression have been conducted. One for the direct effects and one for the interaction effects. Lastly, two linear regression have been conducted on the mediator in order to check the direct effect on the mediator as well as interaction effects. To support the regression analyses, the procedure of Preacher and Hayes has been conducted. This procedure is used to retrieve the conditional (in)direct effects. According to Preacher and Hayes, the proposed conceptual model is identical to the paths that are as being discussed in their framework regarding Model 8 (Hayes 2018).
For the machine learning techniques, a total of nine techniques were used. These are described within the chapter ‘results of machine learning’. The dataset is split up into a training (70%) and a testing dataset (30%), no strict requirements within the literature are available for the distribution with regards to the training and testing data set. Also, a seed number is set for the dataset which is set to a random number of 200, which is needed for reproducible results. Moreover, no mediation effects were found in the previous chapter, hence this mediation effect is excluded in the model for machine learning techniques. However, the covariate is included. Besides, all interaction effects and direct effects are included. The parameters chosen are the same for the regressions earlier conducted, which are: ‘interaction effect of satisfaction on PSC’, ‘interaction effect of satisfaction on trust’, ‘satisfaction’, ‘trust’, ‘PSC’, ‘engagement’, ‘USEconomy’, ‘USEconomy1YrAgo’, ‘USEcononmyin1Yr’, ‘age’, and ‘assets’. The USEconomy variables reflect the trust customers have in the US economy. The dependent variable is churn, which reflects when customers churn or not.
3.4 Model assumptions
examined, violation of this assumption states that the covariates do not have equal variances of disturbance terms over time. A Goldfeld-Quandt test is conducted to examine this
assumption. The GQ-test returned a value of 0.806 with a p-value of 0.997, hence one can conclude that this assumption is not violated. Furthermore, it is examined whether the residuals are normally distributed. Three different tests have been conducted. The Shapiro-Wilk normality test, Lilliefors (Kolmogorov-Smirnov) normality test and the Adjusted Jarque-Bera test for normality all show significant results (p < .05) indicating that this assumption is violated. Based on these results bootstrapped results have been used for the interpretation of the linear regressions to accommodate this violation.
Consequently, some assumptions have to be examined for the logistic regression. Multicollinearity is also an assumption for logistic regression. Therefore VIF-scores are calculated. Results show that the VIF-scores are below five, indicating multicollinearity does not exist in the dataset. Secondly, a logistic regression assumes linearity of continuous
20
4 RESULTS OF REGRESSIONS
Three different regression has been conducted without interaction effects to interpret the main effect of the covariates on the dependent variable. The dependent variable, however, can take three different forms. The first one is engagement, which is the mediator in the model. The second one is attitudinal loyalty for the effects on attitudinal loyalty, and the last one is behavioral loyalty for the effects on behavioral loyalty. The procedure of Preacher and Hayes is used to analyze the moderated mediation conceptual model. Bootstrapped p-values have been used in order to accommodate for the non-normality violation in the residuals.
4.1 Preacher and Hayes procedure
The procedure of Preacher and Hayes is used to analyze the proposed conceptual model. This model consists of a mediator and a moderator, more specifically a moderated mediation. Thus, it is used to test the conditional indirect effects of ‘perceived switching costs’ and ‘trust’ on ‘attitudinal-‘ and ‘behavioral loyalty’ through ‘engagement’ depending on levels of ‘satisfaction’. Secondly, it is used to test the conditional direct effects in a way that the direct effect of ‘perceived switching costs’ and ‘trust’ on both types of loyalty is examined
depending on the levels of satisfaction. The different levels of satisfaction are examined by dividing them into three different categories, such that the conditional effects of the focal predictor are examined at -1 standard deviation, the mean and + 1 standard deviation.
4.2 Mediation analysis and conditional indirect effects
between the lower level confidence interval and upper-level confidence interval, indicating that the indirect conditional effects are insignificant.
4.3 Direct effects on attitudinal loyalty
Attitudinal loyalty is taken as the dependent variable. Satisfaction (p <.05; β = 0.846),
perceived switching costs (p <.05; β = 0.216), and trust (p <.05; β = 0.168) all have significant positive effects on attitudinal loyalty. A positive and significant effect of perceived switching costs on attitudinal loyalty supports h1a, while a positive and significant effect of trust on attitudinal loyalty supports h2b. Higher levels of these covariates positively affect attitudinal loyalty. In other words, more satisfied customers and customers that perceive a higher level of switching costs will lead to higher levels of attitudinal loyalty. Whereas satisfaction seems to have to most impact on attitudinal loyalty. The demographic control variable age (p <.05; β = 0.070) shows to be significant and the demographic control variables assets (p <.10; β = -0.047), and trust in the US economy in one year (p <.10; β = 0.067) show to be marginally significant. Meaning that a large part of the variation in attitudinal loyalty can be explained by the control variables. However, when controlling for these variables the main effects still show significant results. Results show that engagement has no significant effect on attitudinal loyalty (p > .05), hence full and partial mediation can be excluded on attitudinal loyalty and therefore not supporting h3a and h4a, which was shown in the previous paragraph.
4.4 Direct effects on behavioral loyalty
Behavioral loyalty is taken as our dependent variable. Since behavioral loyalty is a
dichotomous variable, logistic regression has been conducted. Perceived switching costs (p <.05; β = 0.296) and satisfaction (p <.05; β = 0.179) both show significant effects on
behavioral loyalty. Meaning that a higher level of perceived switching costs and satisfaction decrease the propensity of customers to actual switch. In other words, more satisfied
22
mean. With respect to the effects on attitudinal loyalty, one can state that perceived switching costs play a more important role than satisfaction in behavioral loyalty. Results show that engagement has no significant effect on behavioral loyalty (p > .05), hence full and partial mediation can be excluded on attitudinal loyalty and therefore not supporting h3b and h4b, which was shown in paragraph 4.2. Furthermore one cannot interpret the beta-values as in an OLS. Therefore the marginal effects are calculated. Based on the marginal effects, one can state that a marginally increase in perceived switching costs compared to the mean is associated with an increase of 3.03% in behavioral loyalty. Also, a marginally increase in satisfaction compared to the mean is associated with an increase of 1.8% in behavioral loyalty.
[Table 3 about here]
4.5 Conditional directs effect on attitudinal loyalty
First, the interaction effects on attitudinal loyalty are examined. Results show a significant and positive interaction effect of satisfaction between perceived switching costs and
Furthermore, results show a significant and positive interaction effect of satisfaction between trust and attitudinal loyalty (p < .05; β = 0.039). Since the relationship between trust and attitudinal loyalty was positive and significant (p <.05; β = 0.168), it indicates that the interaction effect of satisfaction enhances the relationship between trust and attitudinal loyalty. In other words, with more satisfied customers, greater trust leads to a higher level of attitudinal loyalty. The procedure of Preacher and Hayes is used to assess the extent to which the conditional direct effects hold. All three levels of satisfaction (at -1 standard deviation, the mean and + 1 standard deviation) show significant results (p < .05 The effect increases when the level of satisfaction increases, such that an effect of 0.0834 is observed for a satisfaction level of one minus the standard deviation. For a mean satisfaction level, an effect of 0.1773 is observed and for a level of satisfaction one above the standard deviation, an effect of 0.2711 is observed. This indicates that if the level of satisfaction of customers increases, the
relationship between trust and attitudinal loyalty also increases. These results support h6a. According to the Johnson-Neyman analysis, the significance region of the moderation is 1.80% below one minus the standard deviation of the level of satisfaction and 98.20% above. Suggesting that the effect holds even for customers that have a low level of trust in the firm. In summary, it is found that higher levels of satisfaction and a higher level of trust and higher levels of perceived switching costs have a significantly greater impact on attitudinal loyalty. Thus, the effects of trust and perceived switching costs on attitudinal loyalty is higher with increasing levels of customer satisfaction.
4.6 Conditional directs effect on behavioral loyalty
24
both interaction effects showed significant results on attitudinal loyalty, it is not the case when looking at the behavioral component of loyalty. Meaning that there is no interaction effect on actual behavioral while it exists when on attitudinal loyalty. These findings have important implications which will be discussed further in the paper.
[Table 4 and 5 about here]
4.7 Model fit
The attitudinal loyalty model, including the interactions, shows to be a relatively good model (R2 = 0.7837), which indicates that the model explains 78,37% of the variance in attitudinal loyalty. Moreover, the adjusted R-square (0.7818) is close to the R-square indicating a model with a few insignificant parameters.
The pseudo-R-square of McFadden is retrieved for the behavioral loyalty model and showed to be .028. A rule of thumb for a good model of the McFadden pseudo-R-square lay in-between .2 and .4, meaning the model shows to have a relatively poor fit. Consequently, if one compares the logit model with a null-model it can be stated that the logit model
5 RESULTS OF MACHINE LEARNING TECHNIQUES Several machine learning techniques have been used in order to predict the churn of
customers in the dataset. In the next paragraphs, the performance and evaluations of several criteria of the machine learning techniques are being discussed. The variable importance is derived from the decision tree modeling classifications.
5.1 Logistic Regression
Logistic regression is a supervised machine learning technique that is used when the
dependent variable has a dichotomous outcome. It is used for classification purposes. In this case, it is used to classify whether a customer churns. This logistic regression is different than the one conducted in chapter four. This due to the fact that the data set has been split up in a training and testing dataset and as a result, generates different values. Results show that the logistic regression has only one significant parameter, which is perceived switching costs. This reflects the earlier statement made that perceived switching costs in a behavioral context of loyalty is more important than satisfaction, which is more important in an attitudinal context of loyalty. The logit model is able to predict 88.62% of the classifications correctly whether customers churn or not. The logit model shows to have a top decile lift of 1.28, indicating that in the 10% highest predictions, 1.28 times more correct predictions are identified by the model than it would be expected for a random selection. The
GINI-coefficient shows that the overall performance of the model is relatively poor since it returns a value of 0.26. The GINI-coefficient can range from 0 to 1. Moreover, the AUC is a
discrimination measure that explains how well one is able to classify churners. The AUC results in a value of 0.606 which is better than a random selection of 0.5. It may have a high hit-rate of predicting the correct classifications, however when predicting churn it shows to have poor results, therefore multiple criteria have been used to examine the predictive power.
5.2 Support Vector Machine
26
expected for a random selection. The GINI-coefficient returns a value of 0.53, meaning that based on the GINI-coefficient only the model is decent in predicting the correct classification. Lastly, the AUC returns a value of 0.769 which can be seen as a good value since the GINI ranges from 0 to 1. The SVM is able to predict better than logistic regression since all criteria except for the hit rate is higher.
5.3 Nearest Neighbor
Nearest neighbor is a form of unsupervised machine learning with its purpose to group
observations that are most similar to each other (Kübler et al., 2017). In this study, the Nearest Neighbor approach aims at grouping observations that are most similar with respect to the IV’s effect on behavioral loyalty. It is chosen to include ten nearest neighbors, including too few, would lead to overfitting, while too many leads to parameters underfitting. The nearest neighbor algorithm is able to predict 88.50% of the classifications correctly. The TDL returns a value of 1.68, indicating that in the 10% highest predictions, 1.68 times more correct
predictions are identified by the model than it would be expected for a random selection. Thus, it can predict the first 10% better than a logit model, however less than the SVM. The GINI-coefficient shows a value of 0.21, which can be seen as a relatively poor model fit. The AUC results in a value of 0.626 that explains how well the model is able to classify churners, this value is better than a random selection on 0.5.
5.4 Bagging
Bagging is a supervised machine learning technique that uses decision tree modeling. It makes use of randomly drawn observation from the dataset with replacement. Hence, observations can appear more than once in the dataset. Bagging is useful when the data contains noise. Since it makes use of decision trees, one is able to retrieve the importance of the variables used. The bagging method is able to predict 88.73% of the classifications correctly. The TDL returns a value of 3.45, indicating that in the 10% highest predictions, 3.45 times more correct predictions are identified by the model than it would be expected for a random selection. Thus, it can predict the first 10% better than a logit model and the nearest neighbor, however less than the SVM. The GINI-coefficient shows a value of 0.39, which can be seen as a relatively decent fit. The AUC results in a value of 0.73 that explains how well the model is able to classify churners, this value is better than a random selection on 0.5.
A higher value implicates that a variable is important based on the number of times a variable is used for splitting. Thus, some variables may have greater importance than others. The most important variables within the bagging technique are ‘the interaction effect of satisfaction on PSC’, ‘engagement’ and ‘the interaction effect of satisfaction on trust’.
5.5 Boosting
Boosting is, like bagging, part of decision tree modeling. However, it does not make use of resampling methods as the bagging method does. The weights of each observation are being recalculated due to misclassifications, in a way that misclassifications receive a higher weight (Kübler et. al, 2017). Results show that the boosting method has a good fit in predicting churners. It is able to predict 93.97% of the classifications correctly. The TDL returns a value of 6.12, indicating that in the 10% highest predictions, 6.12 times more correct predictions are identified by the model than it would be expected for a random selection. The
GINI-coefficient shows a value of 0.64, which can be seen as a relatively decent fit. The AUC results in a value of 0.83, which is a relatively good number since it is close to 1 in classifying churners. The boosting method shows to have relatively good power in predicting churners if all criteria are taken into consideration and to be compared to previous techniques.
The most important variables within the bagging technique are ‘engagement’, ‘the interaction effect of satisfaction on trust’ and ‘the interaction effect of satisfaction on PSC’. Meaning that these variables have higher importance with regards to splitting nodes compared to other variables.
5.6 Random Forest
Random forest is, like bagging and boosting a part of decision tree modeling. The technique is similar to the bagging method but it uses a modified algorithm in which it selects at each node-split a random subset of features, that reduces correlations between estimators. The random forest technique is able to predict 94.42% of the classifications correctly. The TDL returns a value of 5.92, indicating that in the 10% highest predictions, 5.92 times more correct predictions are identified by the model than it would be expected for a random selection. The GINI-coefficient shows a value of 0.7, which can be seen as a good fit. The AUC results in a value of 0.86, which is a relatively good number since it is close to 1 in classifying churners.
28
the GINI criteria, followed by the two interaction effects. The random forest technique shows to have relatively good power in predicting churners. It has a lower TDL than the boosting technique, meaning it is less good at predicting the top 10% highest predictions, but all other criteria have higher values than the boosting method.
[Table 7 about here]
5.7 Neural Networks
A typical neural network consists of three layers, namely inputs, hidden layers, and outputs. The input layers are used to process all input data, within the hidden layers information is processed and patterns are being analyzed which leads to predictions. For this paper, one hidden layer is used. There are currently no theoretical reasons by making use of more than two hidden layers. The neural network technique is able to predict 88.17% of the
classifications correctly. The TDL returns a value of 0.99, indicating that in the 10% highest predictions, 0.01 times less correct predictions are identified by the model than it would be expected for a random selection. Moreover, The GINI-coefficient also shows to have a low value, namely 0.07. Based on these values, one can conclude that the neural network
technique is not useful based on this dataset. The AUC returns a value of 0.52, which is close to a random selection of 0.5. Overall the performance of the neural network shows poor results since its value lay close to a random selection.
5.8 Naïve Bayes
Naïve Bayes is rather a simplistic form of supervised machine learning. The classification is based on the Bayes’ theorem, such that the probability depends on prior knowledge
5.9 Decision Tree
A decision tree is a form of supervised machine learning techniques. It starts with the training data and creates so-called branches in which homogenous subgroups are made in. It contains chance nodes, decision nodes, and end nodes. Similar methods are used in the random forest technique. The decision tree is able to predict 88.72% of the classifications correctly. The TDL returns a value of 1.97, indicating that in the 10% highest predictions, 1.97 times more correct predictions are identified by the model than it would be expected for a random selection. Moreover, The GINI-coefficient also shows to have a low value, namely 0.08. Meaning that the decision tree is decent in predicting the top decile churners, but that the overall model fit is relatively poor. The AUC returns a value of 0.47, which is even worse than a random selection of churners. The overall performance of the decision tree is relatively low.
For the importance of variables, the decision tree only uses the interaction effect of perceived switching costs and satisfaction. Such that other variables are not taken into consideration when classifying churners. This might indicate why the decision tree itself is a relatively poor technique for predicting churners.
5.10 Comparison of the machine learning techniques
To compare the machine learning technique for predicting churners for the banking industry of America, several criteria have to be taken into account. These criteria have been discussed in previous paragraphs. The machine learning techniques that contain decision tree algorithms seem to give the best results for this dataset, besides also support vector machine seems to have an overall good performance. The boosting technique shows to give the best results in terms of predicting the top 10% highest predictions, however, the random forest technique shows to give the best results in terms of the GINI-coefficient, the AUC and the hit-rate. The hit-rate criteria are not adequate to compare different types of machine learning techniques. This study is about predicting actual churners, consequently, the hit-rate is measured by predicting the correct classification, hence also predicting non-churners. It may be the case that a certain technique has a relatively good hit-rate, but that does not mean it has a strong predictive power in predicting churners. Therefore, more weight is assigned towards the GINI-coefficient, the AUC and in particular the TDL. Thus, based on the criteria weights and the results from the machine learning technique, the random forest would be the most
30
Furthermore, the variable importance that was retrieved from the decision tree
classification models shows that in general ‘engagement’, ‘interaction effect of satisfaction on perceived switching costs’ and ‘the interaction effect of satisfaction on trust’ are seen as the most important variables. While no significant effects were found of these variables in the logit model.
6 DISCUSSION, LIMITATIONS AND FURTHER RESEARCH This chapter consists of a discussion of the findings that are linked towards preliminary papers including theoretical and managerial implications. Limitations of the study and further research possibilities are also being addressed in the final section.
6.1 Theoretical implications
Customer loyalty is a widely known concept in marketing literature. Many studies have investigated the antecedents of customer loyalty, yet to date, there are many discrepancies about the operationalization of loyalty. Such that studies have investigated behavioral loyalty or attitudinal loyalty, but not the combination of two. As earlier discussed, (Ranaweera and Prabhu 2003; Kumar and Shah 2004; Auh, Bell, McLeod, & Shih 2007) stated that attitudinal loyalty leads to behavioral loyalty, however results of this study state otherwise. Thus, this paper extends previous studies that examined customer loyalty as a one-dimensional construct (for example Zeithaml et al. 1996). Results show that the attitude or customer’s perceptions towards companies make it appear better than it really is. There is a difference between a customer’s attitude and actual behavioral. This study attempts to shed light on the differences in the effect of antecedents of loyalty in the context of attitudinal and behavioral loyalty, which can help practitioners to get a better understanding of how customers actually behave instead of solely looking at customer’s attitude towards a firm or brand. In terms of theoretical implications, this study advocates a more complex view of customer loyalty by taking more types of loyalty into consideration, the behavioral and attitudinal component of loyalty. Tucker (1964, p. 32) stated that: “no consideration should be given to what the subject thinks nor what goes on in his/her central nervous system, his/her behavior is the full statement of what brand loyalty is.” This is in line with the findings of this study, namely that attitudinal loyalty and behavioral loyalty are different and that a proxy of intention does not reflect actual behavior. This study states that several factors influence attitudinal behavioral while these do not influence behavioral loyalty.
Findings of this study suggest that trust within the attitudinal context is seen as an important covariate which supports previous research (Moorman et al., 1993; Doney and Cannon 1997; Pamies 2012), while within the behavioral context this covariate does not show significant results and thus its impact is insignificant. These findings are in partial
32
However, for explaining behavioral loyalty, this seems not to be the case. It seems that
customers attach little value to the trustworthiness of banks in terms of their reliability since it does not play a significant role in explaining behavioral loyalty.
Consequently, there seems a discrepancy between the impact of satisfaction within the type of loyalty. Satisfaction has been seen as the most important variable for influencing loyalty in previous papers (Lee et al., 2001; Yang and Peterson, 2004; Aydin et al., 2005), however, most of the papers within the concept of customer loyalty solely looked at the attitudinal component of loyalty. For influencing attitudinal loyalty, satisfaction is seen as the most important covariate in terms of its impact. However, within the context of behavioral loyalty, perceived switching costs seem to have the most impact. These findings are in agreement with the findings of Lam et al. (2004). These authors empirically found that switching costs are more important within the patronage dimension and satisfaction in the recommended dimension. Nevertheless, satisfaction is inseparably linked to customer loyalty, even though the impact differs among the type of loyalty. This also reveals that satisfaction is a necessity to create loyal customers, but that satisfaction on itself is not enough to guarantee customer loyalty.
Preliminary research (Lee et al., 2001; Yang and Peterson, 2004) states that the perceived switching costs – satisfaction link is an important interaction effect in explaining attitudinal and behavioral loyalty. Findings from this study show that within the attitudinal context this interaction effect holds, but that it shows contradictive results. It was proposed that perceived switching costs had a stronger positive effect on attitudinal loyalty when satisfaction is low than when satisfaction is high. Consequently, results show that perceived switching costs has a stronger positive effect on attitudinal loyalty when satisfaction is high then when satisfaction is low. This implies that with dissatisfied customers, the relationship between perceived switching costs and attitudinal loyalty has less positive effects compared to satisfied customers. Hence, the effect between perceived switching costs and attitudinal
leads to higher switching barriers, since there is no certainty that customers will also experience that level of satisfaction at another bank. Hence, the perceived switching costs show stronger positive effects in a way that switching banks leads to higher perceived switching costs because customers are already satisfied with their current provider. As a result, customers perceive a higher level of switching costs. Furthermore, results of this study did not found a significant interaction effect of satisfaction of the perceived switching costs – behavioral loyalty link, which is consistent with the results of Lam et al. (2004) and Yang and Peterson (2004), who also did not find a significant effect of the perceived switching costs – satisfaction link on behavioral loyalty.
Another interaction that was investigated is the satisfaction-trust link on both types of loyalty. Findings from this study show that within the attitudinal context this interaction effect holds and that this interaction effect is positive which is consistent with the results of
Ranweera and Prabhu (2003). These findings suggest that customers may be trustworthy towards a firm, despite simultaneously being dissatisfied. But that satisfaction has more impact on attitudinal loyalty than trust has. Also, the combination of satisfied customers and customers who trust their organization even show higher levels of attitudinal loyalty.
Consequently, in terms of analyzing the interaction on behavioral loyalty, no interaction effect was found in the satisfaction-trust link. In other words, the joint effect of trust and satisfaction is not statistically greater than the sum of both effects individually. Meaning that the
combination of trust and satisfaction does not enhance the effect of customers to behave more loyal in a way that these effects have individually. Also, trust in the behavioral context of loyalty plays an insignificant role. Previous papers have not looked at the interaction effect of the satisfaction-trust link on behavioral loyalty.
Theoretically, it was proposed that engagement mediates the effect of the covariates on attitudinal and behavioral loyalty. However, no empirical evidence was found to support these hypotheses. It was suggested that when an organization aims to build a relationship with its customers in terms of retaining loyal customers it should pay attention to how to engage these customers. However, the contact moments that customers have with their organization do not influence the loyalty of customers, in a way that it may not create a bond between the
34
within the banking sector. This indicates that customers are not necessarily more engaged, in terms of frequency usage of the services. It may be that engagement in the banking sector is more utilitarian, as customers do not engage more if they do not have more business with the bank.
These results suggest that customers seem to behave differently compared to what their attitude is towards a firm in the banking sector. The model used for explaining attitudinal loyalty, explains to a lesser extent behavioral loyalty. It is assumed that customers may switch banks because of competitive variables. Such variables may influence the behavior of
customers while they do not affect their intentions at their current bank. It is thought that competitive variables, for example, the impact of competitors’ activities on available alternatives may play an important role in the mind of the customer to decide to actually switch. However, further research is needed to support this assumption.
Lastly, some theoretical implications have to be noted with respect to machine learning techniques. The findings of this study found that decision tree classification models offer the best results in predicting churners. Even though variables that showed insignificant results in the logit model, such as engagement and both interaction effects, it seems to be important predictors for the use of decision tree models. It is demonstrated that the use of decision tree models in predicting churners is superior over traditional logit models. Results show that the decision tree classification models are able to predict 3.45 up to 6.12 times more corrects top 10% predictions compared to a random selection.
6.2 Managerial implications
The confirmation of the moderating role of satisfaction between perceived switching and trust in the dimension of loyalty has some implications. Based on these findings a service provider could retain customers effectively by enhancing perceived switching costs, satisfaction and customer’s trust in their organization. However, a strategy that combines both perceives switching costs and satisfaction would be a less effective strategy for retaining loyal
customers. The same applies to trust and satisfaction. On the contrary, it would be an effective strategy for creating loyal intentions. Such intentions could lead to positive worth-of-mouth and recommendations towards consumers. The strategy a firm takes depends on the goal one wants to accomplish. If the goal is to retain customers then a strategy focusing on perceived switching costs would be an effective one, while the goal is to get recommendations a
switching costs to create a ‘lock-in effect’ in order to create a behavioral loyal customer. Since behavioral loyalty takes into account the action customers take instead of their intentions. Perceived switching costs help organizations to prevent customers from actual switching. However, organizations should not maximize their switching costs for customers since this can backfire in a way that extremely high switching costs can lead to negative worth of mouth and thus lead to unsatisfied customers and harm themselves by suppressing the growth of new customers (Jones, Reynolds, & Mothersbaugh 2007).
Another managerial implication is that satisfaction is an important determinant in
explaining customer loyalty, although several other determinants seem to be important such as perceived switching costs. Managers and policymakers in the banking industry should
evaluate the satisfaction level of their customers in order to avoid situations in which customers feel dissatisfied. The primary focus should be on all services a customer uses at their primary bank since this was the measurement of satisfaction within this study.
Consequently, the findings of this study show that both interaction effects (satisfaction-perceived switching and satisfaction-trust) show significant results on attitudinal loyalty. Managers, therefore, can extend the level of customer loyalty by implementing loyalty programs that enhance both components in the equation, first by building switching costs and secondly by enhancing customer satisfaction and customer trust in their firm if their goal is to create attitudinal loyal customer which can help them to spread a positive worth of mouth.
6.3 Limitations and further research
Like every other study, this study has some limitations that have to be taken into account. Firstly, customer loyalty and its determinants are measured for a single sector, namely
financial service. Which limits the generalizability across other sectors. To extend the level of generalizability future research may replicate our study in other industries.
Furthermore, the loyalty of customers was measured over a period of one year, which limits the study in terms of the number of churners in the dataset. Switching behavior may take place over a longer period of time. Thus, future research should investigate customers over longer periods of time.
From this study, it is clear that there are differences in attitudinal and behavioral loyalty, which might possibly be explained by the activities of competitors. Further research should investigate the impact of activities from competitors such as marketing actions and
36
7 REFERENCES
Auh, S., Bell, S. J., Mcleod, C. S., & Shih, E. 2007. Co-production and customer loyalty in financial services. Journal of Retailing, 83(3): 359–370.
Aydin, S., Özer, G., & Arasil, Ö. 2005. Customer loyalty and the effect of switching costs as a moderator variable. Marketing Intelligence & Planning, 23(1): 89–103.
Bansal, H. S., & Taylor, S. F. 1999. The Service Provider Switching Model (SPSM). Journal of Service Research, 2(2): 200–218.
Banyte, J., & Dovaliene, A. 2014. Relations between Customer Engagement into Value Creation and Customer Loyalty. Procedia - Social and Behavioral Sciences, 156: 484–489.
Berry, L. L. 1995. Relationship Marketing of Services: Growing Interest, Emerging Perspectives. Handbook of Relationship Marketing, 149–170.
Chaudhuri, A., & Holbrook, M. B. 2001. The Chain of Effects from Brand Trust and Brand Affect to Brand Performance: The Role of Brand Loyalty. Journal of Marketing, 65(2): 81– 93.
Chiu, C. M., Chang, C. C., Cheng, H. L., & Fang, Y. H. 2009. Determinants of customer repurchase intention in online shopping. Online Information Review, 33(4): 761–784.
Dick, A. S., & Basu, K. 1994. Customer loyalty: Toward an integrated conceptual framework. Journal of the Academy of Marketing Science, 22(2): 99-113.
Doney, P. M., & Cannon, J. P. 1997. An Examination of the Nature of Trust in Buyer–Seller Relationships. Journal of Marketing, 61(2): 35–51.
Fernandes, T., & Esteves, F. 2016. Customer Engagement and Loyalty: A Comparative Study Between Service Contexts. Services Marketing Quarterly, 37(2): 125–139.
Garbarino, E., & Johnson, M. S. 1999. The Different Roles of Satisfaction, Trust, and Commitment in Customer Relationships. Journal of Marketing, 63(2): 70–87.
Grönroos, C. 1995. Relationship marketing: The strategy continuum. Journal of the Academy of Marketing Science, 23(4): 252–254.
Harmeling, C., Moffett, J., Arnold, M. J., & Carlson, B. D. 2017. Toward a theory of
customer engagement marketing. Journal of the Academy of Marketing Science, 45(3): 312– 335.
Hayes, A. F. 2018. Introduction to mediation, moderation, and conditional process analysis a regression-based approach. New York: The Guilford Press.
Henseler, J., Ringle, C. M., & Sarstedt, M. 2014. A new criterion for assessing discriminant validity in variance-based structural equation modeling. Journal of the Academy of Marketing Science, 43(1): 115–135.
Hollebeek, L. D., Glynn, M. S., & Brodie, R. J. 2014. Consumer Brand Engagement in Social Media: Conceptualization, Scale Development and Validation. Journal of Interactive
Marketing, 28(2): 149–165.
Hsieh, F. Y., Bloch, D. A., & Larsen, M. D. 1998. A simple method of sample size calculation for linear and logistic regression. Statistics in Medicine, 17(14): 1623–1634.
Hussein, A. S. 2016. The effect of trust and brand engagement on mobile telecommunication customer loyalty: the mediating effect of brand engagement. Konferensi Nasional Riset Manajemen.
38
Jones, M. A., Reynolds, K. E., Mothersbaugh, D. L., & Beatty, S. E. 2007. The Positive and Negative Effects of Switching Costs on Relational Outcomes. Journal of Service Research, 9(4): 335–355.
Kim, M. K., Park, M. C., & Jeong, D. H. 2004. The effects of customer satisfaction and switching barrier on customer loyalty in Korean mobile telecommunication services. Telecommunications Policy, 28(2): 145–159.
Kübler, R, Wieringa, J.E. & Pauwels, K.H. 2017. Machine Learning and Big Data. Advanced Methods for Modeling Markets; Springer, New York.
Kumar, V., & Shah, D. 2004. Building and sustaining profitable customer loyalty for the 21st century. Journal of Retailing, 80(4): 317–329.
Kumar, V., Aksoy, L., Donkers, B., Venkatesan, R., Wiesel, T., et al. 2010. Undervalued or Overvalued Customers: Capturing Total Customer Engagement Value. Journal of Service Research, 13(3): 297–310.
Lam, S. Y., Shankar, V., Erramilli, M. K., & Murthy, B. 2004. Customer Value, Satisfaction, Loyalty, and Switching Costs: An Illustration From a Business-to-Business Service Context. Journal of the Academy of Marketing Science, 32(3): 293–311.
Lee, J., Lee, J., & Feick, L. 2001. The impact of switching costs on the customer satisfaction‐ loyalty link: mobile phone service in France. Journal of Services Marketing, 15(1): 35–48.
Leninkumar, V. 2017. The Relationship between Customer Satisfaction and Customer Trust on Customer Loyalty. International Journal of Academic Research in Business and Social Sciences, 7(4): 450-465.
