The Effects of Employee-Customer Interface on Customer Satisfaction
and WOM Intentions in a Customer-Contact Center
Nhat LE Quang
The Effects of Employee-Customer Interface on Customer Satisfaction
and WOM Intentions in a Customer-Contact Center
Completion date: July 10, 2013
Master thesis
University:
RijksUniversiteit Groningen
Faculty:
Economics and Business
Department:
Marketing
Program:
MSc in Marketing, profile Marketing Intelligence
Author:
Nhat LE Quang
Address:
Esdoornlaan 626
9741MG, Groningen, The Netherlands
Phone Number:
+31 (0) 6 447 478 96
Email:
lequangnhatth@gmail.com
Student number:
2483084
First Supervisor:
Prof. Dr. J. E. (Jaap) Wieringa
Management Summary
The goal of this study is to examine the effect of the employee’s customer orientation (CO) and efficiency orientation (EO) – two main aspects of employee-customer interface on customer outcomes including satisfaction and word-of-mouth (WOM) intentions in a customer-contact center. Whereas previous research has made a substantial progress in examining CO, the employee’s EO is a relatively new construct that has not been investigated in this area. EO can be defined as the degree to which contact employees try to maximize the number of customers’ problems solved rather than concerning about the quality of solutions. CO apart, EO is found to be another important aspect of employee-customer interface. This study contributes to the current research stream by simultaneously analyzing the effect of these two independent dimensions of employee-customer interface from the mystery caller’s perspective.
A number of moderators consisting of interaction characteristics (topics of call, length of call) and relationship characteristic (length of relationship) was also included in the estimation model. Two data sets were used, in which the first data set collected from mystery callers was used to analyze all the main effects and the moderating effects of interaction characteristics, whereas the second data set collected from “real” customers was used to analyze the moderating effects of relationship duration. A structural equation modelling approach was used to assess the effects of observed and unobserved (latent) exogenous variables on the observed and unobserved endogenous variables. A full-information maximum likelihood approach and scaled Satorra-Bentler procedure were employed to deal with missing values and multivariate nonnormality.
Preface
The thesis that you are reading is the final assignment that can help me graduating from the two-year MSc program majoring in Marketing both in University of Groningen and in Norwegian Business School. Before being able to write these “happy sentences,” there was a difficult time in February when my thesis writing was started with a different topic. At that time, I chose another topic that I also like it very much although the topic was quite new to me. After struggling for reviewing the related literature, suddenly, a serious breakdown happened. One morning I was informed that some privacy issues of the focal company had prevented them from providing me the necessary data. No data means no further continuing with that topic. Almost two months after it was started, everything seemed to be collapsed. I had lost my direction and started feeling stressful and anxious. Fortunately, for the meantime, Professor Jaap Wieringa received another data from another company, which were collected in order for a dissertation of an external Ph.D. candidate in University of Groningen. Prof. Wieringa encouraged me in starting a new “story” with this data and also allowed me to extend the deadline for my thesis if necessary. Then, in the mid of April, I restarted with lots of documents from that Ph.D. student. It was such a good luck coming from bad times. After spending several weekends on being in front of the laptop, fortunately, I could nearly catch up with my thesis-group mates and then finally finish it before the deadline. After all, I would like to give sincere thanks for my first supervisor – Professor Jaap Wieringa. I fully appreciate his support. I am also thankful to receive data shared by Professor Peter Leeflang, Professor Janny Hoekstra, and Annette Ammeraal. In addition, I would appreciate many comments from my thesis-group members, namely Martine, Elise, Ruud, Leon, and Jan. Then I am also grateful with invaluable feedbacks from Professor Maarten Gijsenberg. Finally, yet importantly, I am deeply indebted to my family for always staying beside me even whenever it is getting dark, cold, and stormy, especially my wife and my future daughter to whom my whole is dedicated.
I hope you will find my thesis interesting. Enjoy reading it!
Table of Contents
Management Summary ... i
Preface ... ii
List of Figures and Tables ... v
List of Appendices ... vi
1. Introduction ... 1
1.1 Introduction of Employee-Customer Interface ... 1
1.2 Two Dimensions of Employee-Customer Interface ... 2
1.3 Customer-Contact Center ... 2
1.4 Problem Statement and Research Questions ... 3
1.5 Theoretical Contribution ... 3
1.6 Thesis Outline ... 4
2. Theoretical Framework ... 5
2.1 Effect of Employee-Customer Interface ... 5
2.1.1 Employee’s Customer Orientation ... 5
2.1.2 Employee’s Efficiency Orientation ... 7
2.2 Moderating Effects ... 8
2.2.1 Characteristics of Service Interaction ... 8
2.2.2 Length of Relationship ... 10
2.3 Customer Outcomes ... 10
2.4 Conceptual Model ... 12
3. Research Design ... 17
3.1 Data Description ... 17
3.1.1 Two Individual Samples ... 17
3.1.2 Sample Characteristics ... 17
3.1.3 Measures ... 17
3.2 Analysis ... 22 3.3 SEM Assumption ... 23 3.3.1 Sample Size ... 23 3.3.2 Missing Data ... 23 3.3.3 Multivariate Normality ... 24 4. Results ... 27
4.1 Direct Effects of Employee-Customer Interface on Customer Outcomes ... 27
4.2 Moderating Effects of Call Duration and Topic of Call... 28
4.2.1 Call Duration ... 28
4.2.2 Topic of Call ... 29
4.3 Partial Mediating Effect of Satisfaction ... 30
4.4 Moderating Effect of Relationship Lengths Using Regular Customer Data ... 31
4.5 Model Validation... 33
4.5.1 Face Validity ... 33
4.5.2 Statistical Validity ... 33
4.5.3 Robustness Check ... 34
5. Conclusions and Recommendations ... 35
5.1 Conclusions ... 35
5.1.1 Dimensionality of Employee-Customer Interface ... 35
5.1.2 CO and EO as Antecedents of Customer Satisfaction and WOM Intentions ... 35
5.1.3 Moderating Effects... 36
5.2 Managerial Implications ... 36
5.3 Limitation and Further Research ... 37
References ... 39
List of Figures and Tables
Figure 2.1: Conceptual Model ... 13
Table 2.1: Literature Review on the Consequences of Employee-Customer Interface ... 13
Table 3.1: Measures Used in Study ... 18
Table 3.2: Measures and Measurement Criteria (MC Data) ... 19
Table 3.3: CO and EO Component Matrix ... 20
Table 3.4: SAT and GRAT Component Matrix... 20
Table 3.5: Model Comparison to Determine the Inclusion of GRAT (MC Data) ... 20
Table 3.6: Means, Standard Deviations, and Correlations (MC Data) ... 21
Table 3.7: Operationalization of Moderators and Control Variables (MC Data) ... 21
Table 3.8: Operationalization of Moderators and Control Variables (RC Data) ... 22
Table 3.9: Assessment of Normality (MC Data) ... 24
Table 4.1: CO and EO as Antecedents of Satisfaction and WOM (MC Data) ... 27
Table 4.2: The Moderating Effects of Call Duration (MC Data) ... 28
Table 4.3: The Moderating Effects of Business-Dummy Variable for Call Topics (MC Data) ... 29
Table 4.4: The Moderating Effects of Financial-Dummy Variable for Call Topics (MC Data) .. 30
Table 4.5: The Mediating Effect of SAT on Relationships between CO, EO, WOM (MC Data) ... 31
List of Appendices
Appendix 1: CO, EO, SAT and GRAT Component Matrix ... 45
Appendix 2: Reliability and Validity of Measures (RC Data) ... 46
Appendix 3: Means, Standard Deviations, and Correlations (RC Data) ... 47
Appendix 4: The First 10 and Final Ranked Observations Farthest from the Centroid Based on Mahalanobis Distance (MC data) ... 47
Appendix 5: Assessment of Normality (RC data) ... 48
Appendix 6: Multiple Correlations of Construct Indicators ... 48
Appendix 7: Computing Pooled Parameters and Variances over Imputed Data sets ... 49
1. Introduction
1.1 Introduction of Employee-Customer Interface
1.2 Two Dimensions of Employee-Customer Interface
As the person who directly interacts with customer, the contact employee is expected to fulfill the needs of the customers and satisfy them. Nevertheless, very frequently, the contact employee has to face the conflict between job performance expectations and performance evaluation criteria, which refers to “role conflict” (Tuten and Neidermeyer 2004). For example, the call center strategy is usually to minimize talk time as the less talk time the greater the number of consumers handled by a call center. Therefore, in a customer-contact center, the expectations of the managers stressing operational efficiency (e.g. short response time, short waiting time) may clash with the desire of customers who want problem resolution or satisfaction (de Ruyter, Wetzels, and Feinberg 2001). The question now is that should the contact employees try to satisfy the customer totally (CO) or should they try to get as many calls answered as quickly as possible (EO).
Previous research has found that this role stress may have negative impact on job satisfaction, organizational commitment, and employee performance (de Ruyter, Wetzels, and Feinberg 2001). Nevertheless, it is unclear that how the employee-customer interface (customer versus efficiency oriented) influence on customer outcomes such as customer satisfaction, gratitude, or WOM intention. Indeed, although a rich body of literature exists concerning the role frontline employees play in shaping customers’ service evaluations (Wieseke, Geigenmüller, and Kraus 2012), previous research focuses primarily on the impacts of service-worker CO that represents only a subset of employee-customer interface. As far as it is concerned, empirical investigations of the simultaneous effects of the employee’s CO and the employee’s EO in service encounters are absent. Furthermore, under which circumstances the employee’s CO and/or EO are proper have not been considered by a research (Grizzle et al. 2009). This research therefore aims to fill these gaps by examining the influence of the employee-customer interface on customer outcomes with the moderating effects of characteristics of the employee-customer interaction (e.g. duration of call, topics of call), and characteristics of the relationship (e.g. length of relationship).
1.3 Customer-Contact Center
2018 (Britt 2012). Despite a huge growth in contact centers to better provide customer services, CO in call centers has been analyzed by very few researchers including Rafaeli, Ziklik, and Doucet (2008), Dean (2007), and de Ruyter, Wetzels, and Feinberg (2001). This study uses both mystery calling and regular customers’ data sets to shed light on and confirm the discussed theoretical impacts of employees’ customer and EO on customer satisfaction and loyalty.
1.4 Problem Statement and Research Questions
From the above perspective, the problem statement can be stated as follows:
“What are the effects of the employee-customer interface on customer satisfaction and WOM intention in a customer-contact center?”
To be in line with the above problem, two main research questions are defined as: 1. How do the employee’s CO and EO affect customer satisfaction and WOM intention? 2. Which characteristics of the employee-customer interaction and relationship moderate the impacts of employee-customer interface on customer outcomes and how large are those moderating effects?
1.5 Theoretical Contribution
customer outcomes. Moreover, the moderating effects of the characteristics of interaction between employees and customers (e.g. call length and topics of call) are also examined. Finally, the data is collected from both regular customers and mystery callers. Although the data collected from mystery shoppers have been used widely in the marketing research context due to its cost effectiveness, only little academic research has addressed the reliability and validity of mystery shopping data (e.g. Finn 2001), and no research exists relating to mystery calling (MC) data (Ammeraal, Hoekstra, and Leeflang 2013). Furthermore, the precise effects of some important aspects of employee-customer interface such as CO from a customer’s perspective are less clear (Dean 2007). As it is hard to capture the customer’s perception of service experience and the size of data set from regular customers is often limited, this study assesses the effects of CO and EO from a mystery caller’ perspective that are expected to bring managers some useful insights into the customer’s thinking.
1.6 Thesis Outline
2. Theoretical Framework
This part reviews the theories used to develop the conceptual model. More specifically, it describes the current marketing literature on employee-customer interface (part 2.1), characteristics of service interaction and relationship (part 2.2), customer outcomes consisting of customer satisfaction and WOM intentions (part 2.3), and the conceptual model will be presented in the last part (2.4).
2.1 Effect of Employee-Customer Interface
The employee-customer interface is one of three perspectives from the employee management process, along with the manager-employee interface, and the employee-role interface. As stated by Hartline and Ferrell (1996), whereas the two latter interfaces deals with customer-contact employees only (i.e. employees’ responses, job satisfaction, self-efficacy…), the employee-customer interface deals with employee-customer interaction during the service encounter such as the relationship between customer perceptions of the service encounter and contact employees’ attitudinal and behavioral responses (Hartline and Ferrell 1996).
In many cases, customer-contact employees are the first and only representation of a service firm (Hartline, Maxham III, and McKee 2000). Therefore, many companies are trying to implement the marketing concept at the employee-customer interface (Homburg, Wieseke, and Bornemann 2009), which implies that the attitudinal and behavioral responses of customer-contact employees towards customers are very important. As mentioned above, two components of the employee-customer interface are the employee’s CO and EO. Whereas the employee’s CO is a familiar construct in service marketing literature, the employee’s EO is a relatively new concept that has not been empirically tested in any academic marketing research. Hereafter, these two concepts will be explained based on existing marketing literature.
2.1.1 Employee’s Customer Orientation
while not excluding those of all other stakeholders such as owners, managers, and employees, in order to develop a long-term profitable enterprise”. More specifically, the concept of CO in salespeople is proposed by Saxe and Weitz (1982) as the practice of the marketing concept at the level of the individual salesperson and customer. Additionally, these authors further defined customer-oriented selling as “the degree to which salespeople practice the marketing concept by trying to help their customers make purchase decisions that will satisfy customer needs.” This definition implies that highly customer-oriented salespeople are mainly concerned with others and they employ a problem-solution approach instead of focusing on sales. Furthermore, they engage in activities targeted at enhancing long-term customer satisfaction (Saxe and Weitz 1982; Franke and Park 2006).
The possible consequence of CO is a main issue that has been investigated by many scholars, especially the positive effect of customer-oriented behavior on customer satisfaction (e.g. Goff et al. 1997; Ramsey and Sohi 1997; Bettencourt and Brown 1997). There seems to be a consensus, which is that customer-oriented employees are more likely to deliver exceptional service quality and create satisfied customers (Stock and Hoyer 2005). The direct effect of employee’s CO on customer satisfaction has been found by some authors (e.g. Goff et al. 1997; Stock and Hoyer 2005). This leads to the first hypothesis:
Hypothesis 1: Employee CO will exert a positive influence on customer satisfaction.
behavioral outcomes such as WOM intentions (e.g. Brady and J. Joseph Cronin 2001; Dean 2007). This leads to the second hypothesis:
Hypothesis 2: Employee CO will exert a positive influence on customers’ WOM intentions. 2.1.2 Employee’s Efficiency Orientation
In order to measure the employee-customer interface in customer-contact centers, apart from the well-known CO metric, Ammeraal, Hoekstra, and Leeflang (2013) has developed the new concept called employee’s EO. Whereas customer-oriented contact employees focus on fulfilling individual needs in order to satisfy customers, efficiency-oriented employees focus on internal efficiency targets such as handling the maximum number of calls in a certain length of time. As mentioned above, in a customer-contact center, the frontline employees often have to deal with conflicting objectives between service quality and quantity (bureaucratic efficiency), which are referred to “a customer-oriented bureaucracy” (Korczynski 2002) or “role conflict” (Tuten and Neidermeyer 2004). The employees who adopt efficiency-oriented approach will not wonder “How can I best solve this person’s problems?” (Saxe and Weitz 1982) but instead ask, “How can I handle the call in the most efficient way?” Therefore, the employee’s EO may be defined as the degree to which contact employees try to maximize the number of customers’ problems solved rather than concerning about the quality of solutions. Because the high level of EO often corresponds to a low level of CO in a salesperson and vice versa, the effects of EO on customer outcomes are expected to be opposite to those effects of CO. One can argue that, in some easy cases, customers may prefer a short and efficient procedure in order resolve their problems quickly. Nevertheless, nowadays, that kind of easy procedure is often designed to allow customers solve their problems by themselves or with the assistance of technology only such as looking at frequently asked questions (FAQ) or using handy forms in the company website. Thus, it is believed that customers only need aid of customer-contact employees when the problems cannot be solved too quickly. As a result, in those cases, customers will perceive the completion of the solution more important than the speed of handling procedure, as no one would like to call or meet again for the old problem. This study concerns the employee’s EO in which the contact employee focuses only on the speed of problem handling rather than a quality of problem solution. Hence, it is reasonable to expect a negative effect of the employee’s EO on customer outcomes. That leads to two following hypotheses:
Hypothesis 4: Employee EO will negatively affects customers’ WOM intentions.
2.2 Moderating Effects
Although the notion that individuals’ behaviors are guided by both personal and situational considerations is not new, prior research has shown little evidence of moderating effects on the relationship between employee-customer interface and performance (Grizzle et al. 2009; Franke and Park 2006). As those results seem very context-dependent, this study would like to confirm the influences of several moderators consisting of service interaction’s duration (duration of call) and length of relationship, which have been also analyzed by previous research (e.g. Rafaeli, Ziklik, and Doucet 2008; Stock and Hoyer 2005). Further, as suggested by Rafaeli, Ziklik, and Doucet (2008), this study also examine other characteristic of the service interaction, apart from the length of the interaction, which is the type of problem or the topic motivating each call.
2.2.1 Characteristics of Service Interaction
Service interaction is the person-to-person interactive process occurs in service encounter when the service is delivered (Surprenant and Solomon 1987). Previous research in this area has examined its characteristics such as topic and length of interaction (e.g. Donavan, Brown, and Mowen 2004).
2.2.1.1 Duration of Call
Brown, and Mowen 2004) and as such, if the employees can perform better, there will be more customers satisfied. This leads to the following hypothesis:
Hypothesis 5: The positive influences of employee CO on (a) customer satisfaction and (b)
WOM intentions are stronger when call length is long than when call length is short.
In contrast, efficiency-oriented employees always try to keep calls as short as they can. Therefore, if the customer perceives that the employee is putting his/her emphasis on efficiency, that customer will feel more dissatisfied when the talk time (the length of the call) is short rather than when it is long. Consequently, it is expected that the shorter the service interaction is, the stronger the negative impacts of EO on customer outcomes will be. This expectation can be stated as follows:
Hypothesis 6: The negative influences of employee EO on (a) customer satisfaction and (b)
WOM intentions are stronger when call length is short than when call length is long.
2.2.1.2 Topic of Call
There is no literature to guide the prediction of the effect of this moderator. One exception is the research of Rafaeli, Ziklik, and Doucet (2008) in which they stated that very complex calls are more likely to give employees more opportunities to perform their CO in comparison with simple calls. In this study, the complexity of the call is not measured. Nevertheless, the calls were classified as new-business versus existing-business calls, or financial versus non-financial calls. Although it is hard to determine which group of calls is more complex, with respect to the new-business call or financial call, the efforts of the employee in trying to find a suitable solutions or products seem to be more highly evaluated by customers. Then, it is expected that the effect of employees’ CO on customer satisfaction and WOM intentions will be stronger when the call is about new business or financial business.
In contrast, if the employee performs the efficiency-oriented behavior, the customer might sympathize because the topic of call is relatively new and complicated (such as financial issues). The customer could think that the employee probably does not know the issues well, causing the quick but incomplete solution. Thus, the negative effects of EO on satisfaction and WOM intentions could be reduced for these calls. That leads to the following hypotheses:
Hypothesis 7: The positive influences of employee CO on (a) customer satisfaction and (b)
Hypothesis 8: The negative influences of employee EO on (a) customer satisfaction and (b)
WOM intentions are weaker when the call is about new (financial) business than when the call is about existing (non-financial) business).
2.2.2 Length of Relationship
The duration of the service provider-customer relationship, which refers to the length of time the relationship between the customer and the firm has existed, has been identified as a key antecedent of customer satisfaction in several studies (Stock and Hoyer 2005). Furthermore, relationship duration was found to have a positive effect on customer perceptions of relationship strength (Dagger, Danaher, and Gibbs 2009). This means that long relationship duration corresponds to greater levels of relationship value. In addition, a strong relationship often leads to greater customer loyalty and retention (Gwinner, Gremler, and Bitner 1998; Bolton 1998). Thus, on the one hand, it is believed that a customer with longer relationship duration will perceive higher satisfaction and loyalty from the employee’s CO than the customer with shorter relationship. On the other hand, Bolton (1998) finds that customers with more experience weigh their prior cumulative satisfaction more heavily and new information (relatively) less heavily. This means that it is easier and quicker to make a customer dissatisfied when he/she is a less-experienced customer rather than when that is a more-experienced customer. Specifically, new customers with less experience will often feel more vulnerable (and more dissatisfied) with respect to the employee’ EO compared to old customers. These predictions are hypothesized as follows:
Hypothesis 9: The positive influences of employee CO on (a) customer satisfaction and (b)
WOM intentions are stronger when relationship duration is long than when relationship duration is short.
Hypothesis 10: The negative influences of employee EO on (a) customer satisfaction and (b)
WOM intentions are smaller when relationship duration is long than when relationship duration is short.
2.3 Customer Outcomes
have also been applied in many earlier studies (e.g. Dean 2007; Stock and Hoyer 2005; Ramsey and Sohi 1997; Goff et al. 1997).
WOM intention, which refers to intention to recommend the product or service to others, is perceived as one metric to measure customer loyalty (Brunner, Stöcklin, and Opwis 2008). Doorn et al. (2010) stated that customer loyalty is an attitudinal antecedent of the engagement behaviors such as WOM activity or recommendations. Previous research also found that WOM metrics like Net Promoter Score are good predictors of firm growth (Keiningham et al. 2007).
Customer satisfaction refers to “the consumer’s fulfillment response” or “a judgment that a product/service feature, or the product or service itself, provided (or is providing) a pleasurable level of consumption-related fulfillment, including levels of under- or over-fulfillment” (Oliver 2009, 8). In the service context, overall satisfaction relates to overall evaluations of service quality (Gustafsson, Johnson, and Roos 2005). By definition, a high level of customer satisfaction can create a high positive attitude toward the object (e.g. brand, service provider) and may consequently cause a true brand loyalty from a customer. Furthermore, satisfied customers who possess greater perceptions of an overall service quality will more likely to engage in behavioral intentions such as the likelihood of repurchasing and recommending (Boulding et al. 1993). As the effect of customer satisfaction on WOM intentions has been investigated extensively in the marketing literature (e.g. Lam et al. 2004; Verhoef, Franses, and Hoekstra 2002; Wangenheim and Bayón 2007; Woisetschläger, Lentz, and Evanschitzky 2011), it is expected that the influence of the employee-customer interface on WOM intentions will be partially mediated through satisfaction. Specifically, the employee-customer interface directly (and significantly) affects both customer satisfaction and WOM intention, but the latter effect will be weakened (but still significant) if the direct effect of customer satisfaction on WOM intention is taken into account. In other words, the indirect effect of customer/EO on WOM (through the mediation of satisfaction) does not equal its total effect (when satisfaction is not considered) but is smaller and of the same sign (Shrout and Bolger 2002). This leads to the following hypothesis:
Hypothesis 11: The influences of employee CO (a) and/or EO (b) on customers’ WOM
In addition, Verhoef, Franses, and Hoekstra (2002) found evidence that supports the moderating effect of relationship age on the association between satisfaction and relationship outcomes such as customer referrals. Thus, it is expected that the length of service provider-customer relationship will strengthen the impacts of satisfaction on WOM intentions. This leads to the hypothesis:
Hypothesis 12: The positive impact of customer satisfaction on WOM intentions is stronger
when relationship duration is long than when relationship duration is short.
2.4 Conceptual Model
Figure 2.1: Conceptual Model
The consequences of the CO have been examined in various research. To sum up, Table 2.1 provides a summary of key impacts of the employee-customer interface (mainly CO), which support the abovementioned rationales behind the variables and the corresponding effects in the proposed model.
Table 2.1: Literature Review on the Consequences of Employee-Customer Interface Author(s),
year
Context of the study Main impacts of
employee-customer interfaces
Other variables
Bagozzi et al. (2012)
Various of firms across multiple industries
CO Contextual knowledge formation, motivation to learn from customers, and buying knowledge formation (+) Bettencourt
and Brown (1997)
Multi-state, western bank Contact employee extra-role customer service Customer satisfaction (+); Contact employee role-prescribed customer service Customer H9a H9b H10b H10a H6b H8b H5a H7a H5b H7b H12 H11a H11b H6a H8a Employee’s CO Employee’s EO Customer satisfaction WOM intentions
* Characteristics of service interaction (call length, topic of call)
* Characteristics of relationship (relationship length)
Moderators
Employee-customer interface Customer outcomes
Relationship length
H4 H1
satisfaction (+) Blocker et al.
(2011)
Global business-to-business firms in information and communication technology services operating in 19 industries across India, Singapore, Sweden, the United Kingdom and the United States
Proactive CO Customer value perceptions (+) (1); Responsive CO Customer value perceptions (+) (2); Proactive CO*Responsive CO Customer value perceptions (+)
M for (1), (2): Customer value change intensity (+); M for (1): Global relationship scope (+), Transnational
relationship structure (+) C: Offer Quality, service support, personal interaction and other offering provider advantages
Brady and J. Joseph Cronin (2001)
Various customers CO Customer perceptions of service employee performance, the service environment, and any physical goods included (+) CO Customer perceptions of overall service quality, value, satisfaction, and behavioral outcomes (I+)
Brown et al. (2002)
Food services industry CO Self- and supervisor ratings of performance (+) Cross et al. (2007) Field business-to-business salespeople Salesperson CO Salesperson performance (+)
Dean (2007) Two call centers: End consumers of insurance and business customers of online banking
Customer focus Perceived service quality, customer loyalty (+); Customer feedback Affective commitment (+) Donavan,
Brown, and Mowen (2004)
Financial institution (a midsize bank) Service-worker CO Job satisfaction (1), Commitment (2), and customer-oriented altruism (+) M for (1), (2): Contact time (+) Franke and Park (2006)
Goff et al. (1997)
Purchasers of new vehicles of all makes
CO Satisfaction with salesperson (+); CO Satisfaction with dealer (P+); Selling Orientation Satisfaction with salesperson (–) Grizzle et al. (2009) A chain of full-service restaurants Employee CO Performance frequency of customer-oriented behaviors (+) M: Unit CO climate (+); C: tenure Harris, Mowen, and Brown (2005)
Real estate industry CO Work Satisfaction (+)
Homburg, Müller, and Klarmann (2011) Various companies operating mainly in B2B markets
Functional CO Customer loyalty (+) M: Customer’s task orientation (+); Product importance (+); Brand strength (–) Jones, Busch, and Dacin (2003)
National consumer goods manufacturer’s sales force and retail trade customers
Salesperson’s CO Role conflict, role ambiguity, and customer’s propensity to switch suppliers (–) Kelley and Hoffman (1997) Regional insurance company Customer-oriented behavior Service quality (+); Sales-oriented behavior Service quality (–) Luo, Hsu, and
Liu (2008)
Chinese firms located in three metropolitan areas of China
CO Customer trust and customer commitment (+) M: Channel networking (+); Governmental networking (P–) Rafaeli, Ziklik, and Doucet (2008)
Service interaction in the call center
Customer-oriented behaviors Customer evaluation of the quality of a service encounter (+)
M: Call Length (+)
Ramsey and Sohi (1997)
Rozell, Pettijohn, and Parker (2004) Medical devices nationwide company
CO Emotional intelligence (+), CO Performance (+)
Stock and Hoyer (2005)
Various industries
Customer-oriented attitudes Customer oriented behaviors (+); Customer oriented behaviors Customer satisfaction (+); Customer-oriented attitudes Customer satisfaction (+)
C: Length of relationship
3. Research Design 3.1 Data Description
3.1.1 Two Individual Samples
This study examines the impacts of employee-customer interface using data obtained from a customer-contact center of a health insurance company in the Netherlands. The data provided by a marketing research agency were collected from two independent rating sources - “real” customers and mystery callers. On the one hand, a total number of 599 mystery callers were recruited via six rounds and they have assessed the performance of the customer-contact center on various aspects such as satisfaction, gratitude, and WOM intention. A number of 80 mystery callers in the first round were not asked about the new construct of EO. Thus, only 519 individuals had answered the questionnaire completely. On the other hand, the data of 208 customers were collected via a telephone survey conducted by a research team. To achieve the most accurate answers, only customers who had contacted the company in the past five months were selected and asked about their last telephone contact. Of 208 customers, only 105 answers had no missing value. Those missing values mainly belong to items of CO, EO, or satisfaction constructs.
Given that the number of non-missing observations in regular customer (RC) data is rather small compared to MC data, the first data set from mystery callers are used to test the proposed model and hypotheses. Nevertheless, as the mystery caller does not have a “real” relationship with the company, the second data set from the regular customers are used to analyze the moderating effect of relationship age.
3.1.2 Sample Characteristics
Of the 599 mystery callers, 69.9% were female, and 56.9% people are older than 40 years old. In addition, 50.3% have done more than 18 projects per year who were referred to more-experienced mystery callers. Whereas among 208 customers, 67.8% were female, 65.4% were in Friesland, and 75% of them had been with the company for five years or more.
3.1.3 Measures
are measured on a 7-point Likert scale (1 = fully disagree, 7 = fully agree). Nevertheless, the scales of last four items composed of “EO” construct were changed into the reverse order (1 = fully agree, 7 = fully disagree) in order to be consistent with the scale of the first item: the higher the score, the lower efficiency-oriented level. Single-item construct of WOM intentions which was measured by 11 point scale ranged from 0 (certainly not recommend) to 10 (certainly recommend) is also reported in Table 3.1.
Table 3.1: Measures Used in Study Variable
Customer orientation (CO) (1 = fully disagree, 7 = fully agree)
CO1 The employee did his/her best to figure out what my needs/wishes were. CO2 The employee did his/her best to figure out which product would help me best. CO3 The employee had my interests in mind.
CO4 The employee looked for the product that helped me solve my problem in the best way. CO5 The employee offered me a product that fit me the best in my situation.
Efficiency orientation (EO) (1 = fully disagree, 7 = fully agree)
EO1 I had the impression that the employee took all their time to assist me.
EO2 The employee tried to finish the call without ensuring that I was satisfied with the service.*
EO3 I felt that the staff wanted to keep the call as short as possible.*
EO4 I had the feeling that the employee was under time pressure.*
EO5 The employee tried to handle the conversation as efficient as possible rather than to offer quality to me.*
Satisfaction (SAT) (1 = fully disagree, 7 = fully agree)
SAT1 I am very satisfied with this call.
SAT2 This call met my expectations completely. SAT3 I am not disappointed with this call. SAT4 My experiences with this call are excellent.
Gratitude (GRAT) (1 = fully disagree, 7 = fully agree)
GRAT1 I feel thankful to the employee I spoke with. GRAT2 I feel appreciative to the employee I spoke with. GRAT3 I feel grateful to the employee I spoke with.
WOM intentions (0 = certainly not recommend, 10 = certainly recommend)
WOM Would you recommend the organization to friends or colleagues? * EO2-EO5 were recoded inversely (1 = fully disagree, 7 = fully agree)
scales provided by Palmatier et al. (2009); and the WOM intentions item was a 11-point scale suggested by Reichheld (2003). In order to test whether measures of these construct are consistent with previous understanding of their nature, confirmatory factor analysis (CFA) was performed. Specifically, CFA was used to examine whether these constructs except for WOM (due to its single-item operationalization) are underlying the corresponding items used.
3.1.4 Reliability and Validity of Measures
Table 3.2: Measures and Measurement Criteria (MC Data)
Variable Standardized Factor Loading Error Variance CRa AVEb
CO .94 .77 CO1 .841 .293 CO2 .896 .198 CO3 .828 .314 CO4 .910 .173 CO5 .908 .175 EO .91 .67 EO1 .782 .388 EO2 .862 .256 EO3 .887 .213 EO4 .855 .268 EO5 .704 .505 Satisfaction .97 .88 SAT1 .960 .077 SAT2 .953 .091 SAT3 .865 .251 SAT4 .967 .065 Gratitude .94 .85 GRAT1 .896 .197 GRAT2 .920 .154 GRAT3 .948 .102
a CR: Composite Reliability; b AVE: Average Variance Extracted. They are computed based on formulas given by
Fornell and Larcker (1981) as follows: CR = (∑𝑛𝑖=1𝜆𝑖)2⁄((∑𝑛𝑖=1𝜆𝑖)2+ (∑𝑖=1𝑛 𝛿𝑖)); AVE = (∑𝑖=1𝑛 𝜆2𝑖)⁄ ∑( 𝑛𝑖=1𝜆𝑖2+ ∑𝑛𝑖=1𝛿𝑖) where λ = standardized factor loading; δ = error variance = 1- λ2; n = number of indicators.
Notes: Item numbering corresponds to sequence of items in Table 3.1.
generated by the model and the observed (actual) covariance matrix. Given the sensitivity to sample size of 𝜒2-value (Jöreskog and Sörbom 1993), this study does not rely on the significance of this Chi-square test. In contrast, the root mean square error of approximation (RMSEA) = .074, which is below the recommended cut-off value of .08 (Bloemer, de Ruyter, and Wetzels 1999), indicates the good model specification. All items significantly loaded on the hypothesized latent variables and all standardized factor loadings were higher than .7. In addition, the CR values were all greater than .9, and the AVE values were more than .65 that indicates a good construct reliability and adequate convergent validity, following Fornell and Larcker (1981). To assess the discriminant validity, the AVE of latent constructs in CFA results is compared with the squared correlation between all construct pairs, following Fornell and Larcker (1981) (Luo, Hsu, and Liu 2008). The results indicated that CO and EO were distinct measures as the AVEs (.77 and .67 respectively) exceeded their squared correlation (.31) (Table 3.6).
In addition, by performing a principal component analysis with Varimax rotation method using SPSS20, Table 3.3 indicates that the five CO and the five EO items loaded highly on separate components. Moreover, all cross-loading were below .4, providing further evidence of convergent and discriminant validity of these scales (Schepers et al. 2012).
Table 3.3: CO and EO Component Matrix
Item Component* 1 2 CO1 .844 .278 CO2 .903 .224 CO3 .808 .350 CO4 .895 .239 CO5 .905 .211 EO1 .393 .737 EO2 .218 .868 EO3 .182 .891 EO4 .183 .870 EO5 .323 .703
Table 3.4: SAT and GRAT Component Matrix
Item Component 1 SAT1 .945 SAT2 .944 SAT3 .901 SAT4 .944 GRAT1 .859 GRAT2 .922 GRAT3 .909
Notes: Item numbering corresponds to sequence of items in Table 3.1; Extraction method: Principal Component Analysis; * Rotation method: Varimax
highly on two different factors, whereas items of SAT and GRAT constructs kept loading highly on the same factor (see Appendix 1). Therefore, SAT, GRAT seem to capture the same construct.
Furthermore, by performing structural equation modeling (SEM) with the full sample, the path model without GRAT fit the data better than the nested model with GRAT (SB scaled 𝜒2
𝑑𝑖𝑓𝑓 =131.084, dfdiff = 49, p-value < .0001) (see Table 3.5). Hence, GRAT is excluded
from the estimation model.
Table 3.5: Model Comparison to Determine the Inclusion of GRAT (MC Data)
Path Model AICa BICb Scaled 𝝌𝟐(df)
Model 1 (without GRAT) 40776.975 41128.596 534.786 (129)
Model 0 (with GRAT) 45163.202 45589.542 667.571 (178)
SB scaled Chi-square difference test*: 𝜒2𝑑𝑖𝑓𝑓 =131.084, dfdiff = 49, p-value < .0001
a AIC: Akaike’s Information Criterion = -2LL + 2k; b Bayesian Information Criterion = -2LL + klnN where LL:
likelihood value; k: number of parameters, N: sample size; * This is computed based on a formula given by (Satorra
and Bentler 2001): 𝜒2
𝑑𝑖𝑓𝑓= (𝑐0𝜒02− 𝑐1𝜒12)/[(𝑐0𝑑𝑓0− 𝑐1𝑑𝑓1)/(𝑑𝑓0− 𝑑𝑓1)] where c0 and c1 are scaling correction
factors associated with the models 0 and 1 (1.2934 and 1.3050 respectively).
Table 3.6: Means, Standard Deviations, and Correlations (MC Data)
M SD 1 2 3 4 5 6 7 8 9 10 11 Focal constructs 1. CO 5.65a 1.55 1 2. EO 5.71a 1.00 .552*** 1 3. SAT 4.92a 1.48 .806*** .610*** 1 4. WOM 7.21 2.02 .670*** .558*** .761*** 1 5. CL 6.01 3.65 .319*** .270*** .242*** .191*** 1 6. NBC N.A. N.A. -.086** -.085* -.108*** -.137*** .074* 1 7. FNC N.A. N.A. -.052 -.058 -.022 -.001 -.147*** -.296*** 1 Control variables 8. Age 41.52 10.87 -.020 .060 .034 -.018 .125*** .036 -.048 1
9. Gender N.A. N.A. .017 .036 .015 -.007 -.011 .009 -.029 -.159*** 1 10. #jobs 61.17 44.69 -.074* .014 -.072 .000 -.092** .026 -.037 .026 -.022 1
11. #months 30.77 9.55 .000 .106** .084** .056 -.038 .069* -.038 .118** -.047 .203*** 1 *p < .1; **p < .05; ***p < .01; Notes: N.A. = not applicable, CL: Call length; a Computed based on factor score weights.
Table 3.7: Operationalization of Moderators and Control Variables (MC Data)
Variable Operationalization
Moderator Call
length
Duration of call in minutes
NBC Topics of call dummy (0 = existing business; 1 = new business) FNC Topics of call dummy (0 = non-financial; 1 = financial) Control variable
Age Age in years
Gender Gender dummy (0 = male; 1 = female)
#jobs Total number of jobs that the mystery caller has performed #months Total number of months since the mystery caller was employed
The results of CFA using RC data are not reported here, as they were similar to those with MC data. Regarding to RC data, one can find the tests of measures reliability and validity in Appendix 2, the descriptive analysis of study variables is in Appendix 3. As mentioned above, the moderator included in the model using RC data is relationship length, whereas gender and region were included as control variables. The operationalization of study variables can be found in Table 3.8.
Table 3.8: Operationalization of Moderators and Control Variables (RC Data)
Variable Operationalization
Moderator Relationship length
Length of relationship between a customer and the company in months
Control variable
Gender Gender dummy (1 = female; 0 = male) Region Region dummy (1 = Friesland; 0 = Others)
3.2 Analysis
reduced compared to the one with interaction terms. Furthermore, another reason is that the main effects of call length and relationship length are not much interesting in this area. Finally, as stated before, three hypotheses H9, H10, and H12 predict the moderating effects of relationship duration. Nevertheless, mystery callers are not “real” customers of the company, thus, relationship length does not exist for the MC data. Therefore, to examine these hypotheses, a multi-group SEM analysis was implemented using the second data set from regular customers.
3.3 SEM Assumption
As with all statistical methodologies, to ensure accurate inferences, structural equation modeling requires that certain underlying assumptions be satisfied such as sufficiently large sample size, completely random missing data, correct model specification, multivariate normality, continuously distributed endogenous variables (Kaplan 2008). Main assumptions will be discussed hereafter.
3.3.1 Sample Size
Jackson (2003) suggested that in ML estimation, an idea sample size-to-parameters ratio would be 20:1. Bentler and Chou (1987) asserted that researchers may go as low as five cases per parameter estimate but only if the data are perfectly well behaved. If the data, for example, are not normally distributed, a larger sample is required. Further, as proposed by Kline (2010, 12), a “typical” sample size is about 200 cases, based on the approximate median sample size in surveys of published articles in which SEM results are reported. Kline (2010) also noted that a sample size of 200 cases seems too small with respect to a complex model. Therefore, it is very difficult to make absolute recommendations as to what sample sizes are required. Nevertheless, in typical analyses about CO, sample sizes of 238 and 300 were used respectively by Luo, Hsu, and Liu (2008) and Schepers et al. (2012). Thus, given the moderate complexity of the structural equation model in this study, a sample size of 599 mystery callers or 208 real customers could lead to a satisfactory statistical power. In other words, they are reasonable sample sizes to use SEM.
3.3.2 Missing Data
value of 5%. In this case, listwise deletion method for dealing with incomplete data could lead to two major problems consisting of: 1) the decrease in statistical power, and 2) the risk of nonconvergent solutions, incorrect standard errors (Byrne 2009). Therefore, other methods should be considered. A number of researchers have asserted that full information maximum likelihood (FIML) outperform traditional methods of handling missing data such as listwise deletion, pairwise deletion, or similar response pattern imputation (e.g. Enders and Bandalos 2001). Specifically, in comparison with other recent and sophisticated methods such as expectation maximization (EM), mean substitution (Mean), and regression imputation (Regression), Olinsky, Chen, and Harlow (2003) indicated that FIML is a superior method in the estimation of most different types of parameters in a SEM format. Summing up, FIML seems to be the most preferred and reliable method to deal with missing data. Hence, this study employed FIML technique to handle incomplete data problems.
3.3.3 Multivariate Normality
A critically important assumption in the conduct of SEM analyses is that the data are multivariate normal (Byrne 2009; Kaplan 2008; Kline 2010). This means that: 1) all the individual univariate distributions are normal; 2) the join distribution of any pair of the variables is bivariate normal or each variable is normally distributed for each value of every other variable; 3) All bivariate scatterplots are linear and the distribution of residuals is homoscedastic (Kline 2010, 60). Hence, it is necessary to check for univariate normality first, before assess the multivariate normality, as the former is a necessary, although not sufficient, condition for multivariate normality.
Table 3.9: Assessment of Normality (MC Data)
Variable Skewness C.R.* Kurtosis C.R.
CO1 -1.058 -9.836 .055 .255 CO2 -.805 -7.487 -.657 -3.055 CO3 -1.434 -13.342 1.454 6.762 CO4 -.976 -9.079 -.226 -1.053 CO5 -.908 -8.445 -.464 -2.156 EO1 -2.339 -21.757 5.355 24.902 EO2 -1.897 -17.640 2.409 11.205 EO3 -1.755 -16.320 1.840 8.554 EO4 -2.013 -18.724 3.004 13.969 EO5 -1.319 -12.268 .489 2.274 SAT1 -1.190 -11.064 .728 3.387 SAT2 -1.051 -9.775 .154 .716 SAT3 -1.230 -11.442 .565 2.627 SAT4 -1.164 -10.826 .467 2.170 WOM -1.040 -9.677 1.062 4.937 Age -.061 -.564 -.227 -1.057 Gender -.921 -8.562 -1.153 -5.360 #jobs 1.406 13.075 1.726 8.026 #months -.720 -6.698 -1.074 -4.993 Multivariate 193.065 77.849
* C.R.: critical ratio. Note: This test was implemented in Amos21 using non-missing data
Regarding to MC data, as shown in Table 3.9, all univariate kurtosis values ranged from .055 to 5.355 are less than 7, indicating no item to be substantially kurtotic. Nevertheless, the critical ratio of multivariate kurtosis (Table 3.9), which implicitly represents normalized estimate of multivariate kurtosis developed by Mardia (1970), is rather large (77.325) indicating nontrivial positive kurtosis. Further, Bentler (2006) has suggested that, in practice, values larger than 5 provide evidence of nonnormality in the sample. Hence, this data are not normally distributed. Moreover, to identify multivariate outliers, which have extreme scores on two or more variables, the squared Mahalanobis distance (D2) for each case was computed. Typically, an outlying case will have a D2 that is substantially different from all
the other D2 values (Byrne 2009). Results showed a list of 100 cases farthest away from the centroid (see Appendix 4 for first 10 and final ranked scores). Given the small gap in Mahalanobis D2 values between any two cases, no evidence of serious multivariate outliers was detected.
or CFI are modestly underestimated (Byrne 2009). One way to the analysis of nonnormal data is to implement asymptotic distribution-free (ADF) estimation. Nevertheless, a problem with the ADF estimator is that it can only perform well when sample sizes are extremely large (1000 to 5000 cases) (West, Finch, and Curran 1995).
4. Results
4.1 Direct Effects of Employee-Customer Interface on Customer Outcomes
To test H1 – H4, a full structural equation model was run with the MC data in which CO, EO, satisfaction, WOM, and other control variables (age, gender, number of jobs, and average number of orders per year) were included, applying the scaled Satorra-Bentler (SB) procedure and FIML approach. Table 4.1 summarizes the findings. The global statistics point to an acceptable fit of the structural model with the empirical data: 𝜒2(129) = 534.786, CFI
= .933, TLI = .914, SRMR = .054 and RMSEA = .072. Overall, the model explained 69.3% of the variance in Satisfaction and 60.5% in WOM.
Table 4.1: CO and EO as Antecedents of Satisfaction and WOM (MC Data)
Hypothesis Full sample β t-value Supported Hypothesized Paths CO Satisfaction H1 .677*** 16.591 Yes CO WOM H2 .131** 2.347 Yes EO Satisfaction H3 .227*** 4.697 Yes EO WOM H4 .139*** 2.965 Yes SAT WOM .578*** 9.432
Control Variable Paths
Age Satisfaction .028 1.108 Age WOM -.048* -1.809 Gender Satisfaction .001 .053 Gender WOM -.030 -1.082 #jobs Satisfaction -.038 -1.288 #jobs WOM .053* 1.913 #months Satisfaction .065** 2.391 #months WOM -.013 -.485 R2 Satisfaction .693 R2 WOM .605
* p < .1; ** p < .05; *** p < .01. Notes: Table reports standardized coefficients.
4.2 Moderating Effects of Call Duration and Topic of Call
To examine the moderating effects of call duration or topics of call on the relationships between employee-customer interface and customer outcomes, the multiple-group analysis was employed with MC data (Schepers et al. 2012), applying scaled SB procedure and FIML approach. First, the sample was split into two (longer versus shorter) call groups at the median call duration value (5 minutes) (Aiken and West 1996). Analogously, the sample splitting was repeated based on the dummy variables of call topics (new business versus existing business, financial versus non-financial). Second, a base (unconstrained) model was calibrated in which the structural path to be moderated was freely estimated. An alternative (constrained) model in which this path was fixed was also calculated. A moderation effect would exist if a significant SB scaled chi-square change in the comparison of these two models was observed (Schepers et al. 2012).
4.2.1 Call Duration
Table 4.2: The Moderating Effects of Call Duration (MC Data)
Longer call Shorter call SB scaled Chi-square difference testa
β t-value β t-value Moderating
hypotheses
∆𝝌𝟐(𝒅𝒇) p-value Supported
Structural Paths
CO SAT .766*** 11.623 .629*** 13.033 H5a 3.788* (1) .0516 Yes
CO WOM .106 1.410 .163** 2.199 H5b .117 (1) .7327 No
EO SAT .119 1.352 .297*** 5.541 H6a 1.128 (1) .2883 No
EO WOM .072 1.017 .185*** 3.066 H6b .495 (1) .4818 No
SAT WOM .672*** 8.456 .488*** 5.729
Control Variable Paths
Age SAT -.001 -.020 .060 1.699 Age WOM -.099*** -2.781 -.011 -.269 Gender SAT -.006 -.154 .001 .031 Gender WOM -.033 -.862 -.031 -.779 #jobs SAT -.031 -.676 -.059 -1.562 #jobs WOM .050 1.205 .043 1.216 #months SAT .090** 2.118 .057 1.468 #months WOM -.064 -1.487 .021 .558
* p < .1; ** p<.05; *** p<.01; a The formula is the same as in Table 3.5. Notes: Table reports standardized coefficients.
Overall fit measures: 𝜒2(280) = 706.645, CFI = .929, TLI = .916, SRMR = .066, RMSEA = .071.
H5 predicts that the association between CO and customer satisfaction/WOM may be stronger with a relatively longer call than a shorter one. As shown in Table 4.2, the path loading from CO to customer satisfaction was .766 in the longer call group, which is larger in magnitude than its path loading of .629 in the shorter call group. Furthermore, results from SB scaled Chi-square difference test between these two groups indicated this difference is significant at 10% significance level (∆𝜒2(𝑑𝑓) = 3.788 (1), p = .051 < .1). Thus, H5a is
In contrast, the difference in size between the path loadings from CO to WOM in the longer call and the shorter call group is not significant (∆𝜒2(𝑑𝑓) = .117 (1), p > .1). Thus, H5b is
not supported by the data. Moreover, although CO significantly affected WOM in the full sample, for the longer call group, its effect was insignificant (β = .106, p > .1). In contrast, this effect was significant for the shorter call group (β = .163, p < .05).
H6 predicts that the relationship between EO and customer satisfaction/WOM may be stronger with a relatively short call than with a long call. As shown in Table 4.2, the effect of EO on satisfaction is insignificant for the longer call group (β = .119, p > .1) but strongly significant for the shorter call group (β = .297, p < .01). Similarly, the effect of EO on WOM is only significant for the shorter call group (β = .185, p < .01), not for the longer call group (β = .072, p > .1). The common finding is that, when comparing the two group, there was no significant change of Chi-square for both the path from EO to satisfaction (∆𝜒2(𝑑𝑓) = 1.128
(1), p > .1) and the path from EO to WOM (∆𝜒2(𝑑𝑓) = .495 (1), p > .1). Hence, H6a and H6b
were not supported by the data.
4.2.2 Topic of Call
Table 4.3: The Moderating Effects of Business-Dummy Variable for Call Topics (MC Data)
New business Existing business SB scaled Chi-square difference testa
β t-value β t-value Moderatin
g hypotheses ∆𝝌𝟐(𝒅𝒇) p-value Supported Structural Paths CO SAT .690*** 9.651 .658*** 12.922 H7ab .023 (1) .8805 No CO WOM .068 .572 .168*** 2.693 H7b .549 (1) .4589 No EO SAT .152 1.563 .261*** 4.633 H8a 1.309 (1) .2526 No EO WOM .082 .998 .165*** 2.848 H8b .742 (1) .3891 No SAT WOM .613*** 5.689 .545*** 7.073
Control Variable Paths
Age SAT -.023 -.387 .038 1.429 Age WOM -.054 -.983 -.039 -1.252 Gender SAT .110* 1.955 -.024 -.820 Gender WOM -.069 -1.181 -.025 -.748 #jobs SAT .074 1.288 -.073** -2.045 #jobs WOM .062 1.061 .052* 1.666 #months SAT -.001 -.020 .087*** 2.637 #months WOM .042 .655 -.022 -.656 * p < .1; ** p<.05; *** p<.01; a
The formula is the same as in Table 3.5; b
This is Wald Test of Parameter Constraints used to replace SB scaled Chi-square difference test as MLR estimator results a negative Chi-square difference values. Notes: Table reports standardized coefficients. Overall fit measures: 𝜒2(280) = 727.059, CFI = .930, TLI = .917, SRMR = .059, RMSEA = .073.
call, all the loading between employee-customer interface and customer outcomes were strongly significant. In addition, when comparing these two groups, there was no significant change of SB scaled Chi-square for all these following paths: CO satisfaction (∆𝜒2 = .023,
df = 1, p > .1), CO WOM (∆𝜒2 = .549, df = 1, p > .1), EO satisfaction (∆𝜒2 = 1.309, df
= 1, p > .1), EO WOM (∆𝜒2 = .742, df = 1, p > .1). Thus, the effects of CO and/or EO on
satisfaction and WOM are equal in magnitude between groups of existing and new calls. Similarly, when comparing financial and non-financial call groups, there was no significant change of Chi-square for all the structural paths: CO satisfaction (∆𝜒2 = 1.162, df = 1, p >
.1), CO WOM (∆𝜒2 = .349, df = 1, p > .1), EO satisfaction (∆𝜒2 = 1.24, df = 1, p > .1),
and EO WOM (∆𝜒2 = .138, df = 1, p > .1) (Table 4.4). Therefore, the effects of CO and/or
EO on satisfaction/WOM are indifferent between financial and non-financial calls. Hence, there was no evidence to support H7a, H7b, H8a, and H8b by the data.
Table 4.4: The Moderating Effects of Financial-Dummy Variable for Call Topics (MC Data)
Financial call Non-financial call SB scaled Chi-square difference testa
β t-value β t-value Moderating
hypotheses ∆𝝌𝟐(𝒅𝒇) p-value Supported Structural Paths CO SAT .765*** 10.998 .645*** 14.255 H7a 1.162 (1) .2811 No CO WOM .064 .461 .144** 2.387 H7b .349 (1) .5548 No EO SAT .172** 2.162 .248*** 4.512 H8a 1.24 (1) .2655 No EO WOM .178* 1.761 .126** 2.334 H8bb .138 (1) .7101 No SAT WOM .647*** 3.783 .570*** 8.684
Control Variable Paths
Age SAT -.019 -.611 .035 1.148 Age WOM -.091 -1.615 -.037 -1.211 Gender SAT -.090* -1.868 .024 .788 Gender WOM .019 .307 -.041 -1.297 #jobs SAT -.050 -.941 -.034 -1.007 #jobs WOM .077 1.386 .049 1.563 #months SAT .138*** 2.691 .051 1.620 #months WOM .012 .199 -.020 -.631
* p < .1; ** p<.05; *** p<.01; a The formula is the same as in Table 3.5; b The Wald Test of Parameter Constraints was used here as MLR estimators using Mplus7 results “negative” SB scaled Chi-square difference statistic. Notes: Table reports standardized coefficients. Overall fit measures: 𝜒2(280) = 810.240, CFI = .920, TLI = .906, SRMR = .061, RMSEA = .080
4.3 Partial Mediating Effect of Satisfaction
FIML approach was still used to deal with missing data problem whereas a scaled SB procedure could not be used together with bootstrapping. Nevertheless, bootstrapping can also be used to estimate non-biased standard errors for nonnormal data (Kline 2010, 42).
Table 4.5: The Mediating Effect of SAT on Relationships between CO, EO, WOM (MC Data)
Relationship
Direct effect Indirect effect Total effect
Hypotheses Supported Β 2.5th pctile Β 97.5th pctile Β CO SAT .677*** --- --- --- .677*** H11a Yes CO WOM .131** .293 .391*** .489 .522*** EO SAT .227*** --- --- --- .227*** H11b Yes EO WOM .139*** .071 .131*** .191 .270*** SAT WOM .578*** .578*** * p < .1; ** p < .05; *** p < .01.
Notes: Table reports standardized coefficients. Overall fit measures: 𝜒2(129) = 697.874, CFI = .940, TLI = .923, RMSEA = .086
As shown in table 4.5, satisfaction significantly affects WOM intentions (β = .578, p < .01). Further, the direct effects of CO on both satisfaction (β = .677, p < .01) and WOM (β = .131, p < .05) were found. In addition, the indirect effect of CO on WOM (β = .391, p < .01) was smaller and of the same sign than its total effect on WOM (β = .522, p < .01), in strong support of H11a. Analogously, the direct effects of EO on both satisfaction (β = .227, p < .01) and WOM (β = .139, p < .01) were also found. The indirect effect of EO on WOM (β = .131, p < .01) was smaller and of the same sign than its total effect on WOM (β = .270, p < .01), in strong support of H11b. Thus, customer satisfaction partially mediated the effect of CO and EO on WOM intention as expected.
4.4 Moderating Effect of Relationship Lengths Using Regular Customer Data
Table 4.6: The Moderating Effects of Relationship Duration
Structural path
Full sample Longer relationship Shorter relationship
β t-value β t-value β t-value
CO SAT .779*** 10.294 .806*** 6.859 .870*** 11.276
CO WOM .503** 1.982 .611* 1.881 .589 1.314
EO SAT .175** 1.994 .153 1.181 .078 .753
EO WOM .017 .148 -.040 -.294 .101 .754
SAT WOM .014 .062 .047 .149 -.183 -.458
Control variable path
Gender SAT .000 -.008 -.048 -.957 .047 .839
Gender WOM -.025 -.436 -.024 -.328 -.042 -.428
Region SAT -.004 -.099 .009 .175 -.014 -.263
Region WOM .069 1.099 -.036 -.370 .061 .734
Chi-square difference test for moderating effects ∆𝝌𝟐(𝒅𝒇) p-value Supported
CO SAT Hypothesis 9a .009 (1)a .9251 No
CO WOM Hypothesis 9b .034 (1) .8534 No
EO SAT Hypothesis 10a .195 (1) .6592 No
EO WOM Hypothesis 10b .849 (1) .3568 No
SAT WOM Hypothesis 12 .253 (1) .6147 No
* p < .1; ** p < .05; *** p < .01; a This is Wald Test of Parameter Constraints used to replace SB scaled Chi-square
difference test as MLR estimator results a negative Chi-square difference values. Notes: Table reports standardized coefficients. Overall fit measures of multi-group SEM model: 𝜒2(232) = 409.535, CFI = .888, TLI = .870, SRMR
= .114, RMSEA = .086
Firstly, the full model was run with the whole sample, applying scaled SB and FIML methods. The global statistics point to an acceptable fit of the structural model with the empirical data: 𝜒2(107) = 223.928, CFI = .912, TLI = .889, SRMR = .057 and RMSEA = .072. Overall, the model explained 80.9% of the variance in satisfaction and 28.7% in WOM. As shown in Table 4.6, in the full sample, only the path loadings from CO and EO to satisfaction, and from CO to WOM are significant (β1 = .779, p1 < .01, β2 = .503, p1 < .05,
and β3 = .175, p3 < .1 respectively).
