A longitudinal study of the impact of service failures and the
recovery process on customer satisfaction and actual loyalty
Bart Hoogstad
10858938
First draft of Master Thesis in Marketing
MSc. in Business Administration – Marketing Track
Amsterdam Business School, part of the University of Amsterdam
Supervised by dhr. dr. F. Situmeang and dhr. dr. U. Konus
Statement of Originality
This document is written by Bart Hoogstad, who declares to take full responsibility for
the contents of this document. I declare that the text and the work presented in this
document is original and that no sources other than those mentioned in the text and its
references have been used in creating it. The Faculty of Economics and Business is
Abstract
The present study investigates the relationship between service failure and intentional and actual loyalty. Furthermore, this study focuses on the nature of the failure (e.g. type of failure and severity) and the recovery attributes (type of solution and recovery speed) as effect on
satisfaction with the service recovery (SSR) and actual loyalty. Also, the direct effect of SSR on actual loyalty is investigated. Few studies in the literature used company-recorded hard data to investigate effects in a real life situation. Moreover, most studies did not use longitudinal study designs to investigate effects on actual loyalty. The relationships in this study have been tested using two datasets with company-recorded data in the automotive industry. Binary regression analysis points out that a service failure has a negative effect on actual loyalty. Furthermore, the results of an ordinal regression show a positive effect of tangible recovery and process-related failures on SSR. In contrast, a psychological recovery and outcome-related failure have a negative effect on SSR. The recovery speed did not have a significant effect on SSR. Although no direct relations were found between the main constructs and actual loyalty, in this study there has been found a positive and significant effect of SSR on actual loyalty.
Keywords: service failure, service recovery, recovery attributes, tangible recovery, psychological recovery, severity, handling speed, recovery speed, customer satisfaction, satisfaction with the service recovery (SSR), intentional loyalty, actual loyalty
Table of Contents
1. Introduction ... 5
2. Conceptual framework ... 9
2.1 Service Failure ... 9
2.2 Customer responses to service failures ... 14
2.3 Customer responses to double deviations ... 15
2.4 Company’s response to service failures ... 16
2.5 Service recovery attributes ... 18
2.6 Nature of service failure ... 20
2.7 Linking service recovery attributes and the nature of the service failure to actual loyalty 22 2.8 Linking SSR to actual loyalty ... 23
3. Methodology ... 24
3.1 Participants/sample ... 24
3.2 Measures ... 25
3.3 Data analysis ... 29
4. Results ... 31
4.1 Descriptive statistics model 1 ... 31
4.2 Hypothesis testing model 1 ... 31
4.2.1 Service failure as effect on intentional and actual loyalty ... 31
4.2.2 Double deviation as effect on intentional loyalty and actual loyalty ... 33
4.3 Descriptive statistics model 2 – Dataset SSR ... 34
4.4 Descriptive statistics model 2 – Dataset Actual Loyalty ... 34
4.4 Hypothesis testing model 2 ... 35
4.4.1 Testing the determinants of SSR: type of solution, recovery speed, type of failure and severity. ... 35
4.4.2 Testing effects of the main constructs of SSR on actual loyalty ... 37
4.4.3 Testing effect of SSR on actual loyalty ... 38
4.5 Hypotheses ... 38
5. Discussion ... 39
5.1 Discussion of hypotheses ... 40
5.2 Limitations and further research ... 45
6. Conclusion ... 48
7. References ... 50
8. Appendices ... 55
8.1 List of words indicating tangible and psuvhological recovery ... 55
8.2 Example of customer satisfaction survey ... 56
1. Introduction
Failures in the service industry are unavoidable (Harrison-Walker, 2012). The failure rate in the service industry (7.7%) compared to the goods industry (1.91%) is 4.03 times higher (Autoblog, 2013). Since production and consumption occur at the same time, there is little or no possibility of control or correction. Also, most services are people-focused and errors are therefore likely to occur. As a result of the fact that people deliver the service, the quality of service often depends on the attitude and behavior of front-line employees (Rio-Lanza et al., 2009). Moreover, the quality of services is subjective and is for a great part based on the expectations of consumers. Thus, although service firms strive for maximum quality, they will not be able to entirely eliminate mistakes during service delivery (Rio-Lanza et al., 2009).
Lots of research has been done among different aspects of service failures. Prior studies have focused on five main aspects: (1) types of service failures (Hoffman, Kelley, & Rotalsky, 1995; Scott W. Kelley, Hoffman, & Davis, 1993); (2) service failure effects on emotional response and
customer satisfaction and intentional loyalty (Choi & Mattila, 2008; McColl-Kennedy, Daus, &
Sparks, 2003); (3) recovery strategies (DeWitt, Nguyen, & Marshall, 2008; Fang, Luo, & Jiang, 2012; Spreng, Harrell, & Mackoy, 1995); (4) impact of severity (Hess Jr, 2008; Maxham III & Netemeyer, 2002; Weun, Beatty, & Jones, 2004) and (5) applications of justice theory (Chebat & Slusarczyk, 2005; del Río-Lanza, Vázquez-Casielles, & Díaz-Martín, 2009; DeWitt et al., 2008; Goodwin & Ross, 1992; Wirtz & Mattila, 2004).
Research found that customers respond differently to the characteristics of the failure and to the recovery efforts of the company. To illustrate, a service failure could be process- or the
outcome-related, which will have different effects on the satisfaction level of the customer (Chuang, Cheng, Chang, & Yang, 2012). Also, there exists some evidence that severity or magnitude (Hess Jr, 2008) of the complaint and prior service experiences (Chuang et al., 2012; Tax, Brown, & Chandrashekaran, 1998) play important roles in how customers evaluate the recovery of the service failure. Looking at the severity of the complaint, a more severe complaint could lead to a lower satisfaction level (Weun et al., 2004). Also, not only the failures
themselves lead to dissatisfaction, but the response to and recovery of the failure is most likely the cause of dissatisfaction, which is a major cause for customers’ exit (Buttle & Burton, 2002). To minimize this negative impact, the response of the companies is of high importance. Previous studies explain the recovery efforts through the lens of justice theory (Chebat & Slusarczyk, 2005; del Río-Lanza et al., 2009; DeWitt et al., 2008; Goodwin & Ross, 1992; Wirtz & Mattila, 2004). Within the justice theory, three types of justice are defined: interactional justice,
distributional justice and procedural justice. Interactional justice refers to how customers perceive the communication between the firm and the customer, for example explanation or excuse. Distributional justice refers to how customers perceive tangible resources to compensate for the failure. And procedural justice refers to how customers perceive methods a firm uses to handle complaints, such as recovery speed. Other studies focus more on the concrete solution and the recovery process (Fang et al., 2012; Spreng et al., 1995). The solution of a service failure could be a tangible (e.g. financial compensation) or a psychological reaction (e.g. an excuse or an explanation) (Fang et al., 2012). Also, there exist some evidence that a faster recovery has a positive impact on the satisfaction level after recovery (Boshoff & Christo, 1997; Wirtz & Mattila, 2004). This study will not only investigate the direct effects of a service failure on satisfaction and loyalty but also focus on the effects of the nature of the service failure (e.g. type of failure, severity and prior experience) and the recovery efforts of the company (e.g. type of recovery and recovery speed).
Current research about service failures and the recovery process as effect on emotions, attitudes and intentions is very useful, but there has to be build upon the existing studies with the effects of a service failure and recovery on actual loyalty. Most of previous studies are experimental and cross-sectional studies and therefore only measure the effects on behavioral intent, not the actual behavior. Subjects in these studies are exposed to a controlled laboratory environment. In these settings, the intent to quit may not reflect the actual consumer response in a real life situation. Only Chebat & Slusarczyk (2005) included actual behavior in their study. In this study the effect of level of justice on actual loyalty is reported. But as already stated, the level of justice is about
how customers perceived the recovery process. The shortcoming of this study is that it is still unknown what the actual response and recovery speed of the company were. Also, this study does not take into account the nature of the service failure (e.g. type of failure and severity).
The end goal of a firm should be to keep a customer satisfied and even more important, to keep the customer loyal. Therefore, it is important to know what the effects are of a service failure on satisfaction and loyalty and, since the recovery of the failure has a greater impact on the failure itself, the most important thing to know is how customers respond to recovery attributes and nature of the failure in terms of satisfaction and loyalty. Summarizing, the gap in current literature is the absence of studies with the direct effect of service failures on actual loyalty as well as the effects of the service recovery and nature of the complaint on actual loyalty. Furthermore, the effect of satisfaction with the recovery on actual loyalty has not been studied yet.
This study will try to link the gaps as described previously and will overcome the
methodological limitations. In the first part, complementary to the effects on intentional loyalty, this study will investigate the effect of a service failure on actual loyalty. In this section this study will also check for the possibly or possible more negative effect of prior service experience. In the second part, the effects of determinants of satisfaction with the service recovery (SSR) and actual loyalty will be tested. The determinants being tested will be different types of recovery, recovery speed, types of service failures and recovery speed. Also, the direct effects of these constructs will be tested on actual loyalty as well as the direct effect of SSR on actual loyalty. In order to reach these goals and also complementary to most prior studies with experiment- or survey-based soft data, this study will make use of company recorded hard data. With this study being a non-experimental design, the results will have a higher external validity. Moreover, by making use of longitudinal data the effects of complaints on actual loyalty are measurable. With these advantages this study will therefore have four main contributions, namely: (1) using a non-experimental study design, (2) using longitudinal data to investigate the effects of a service failure and recovery on actual loyalty, (3) investigating the effect of SSR on
actual loyalty and (4) checking for the possible more negative effect of prior failure experience. This research is set in the automotive industry using data of a big dealership in the Netherlands. It is strongly believed that the results of this study are applicable to other industries since customers experience a typical service setting whenever they bring their car for maintenance or repair. This sector is ranked second in terms of frequency of complaints (right after restaurants and ahead of banks and dental/medical services), as noted by Tax, Brown, & Chandrashekaran (1998).
The research questions are as follows:
RQ 1: What is the effect of a service failure on intentional loyalty, (dis)satisfaction and actual loyalty?
RQ 2: What are the effects of types of solutions, recovery speed, types of failures and severity on SSR and actual loyalty?
RQ 3: What is the effect of SSR on actual loyalty?
By taking a closer look at the relationship between service failures and actual loyalty and the relationship between the recovery attributes and actual loyalty, this study has the potential to contribute to both science and business. From a scientific point of view, this study will be the first study that will research the effects of (1) service failure, (2) recovery attributes, (3) nature of service failure and (4) SSR on actual loyalty. From a managerial point of view this research is important, because for managers and marketers it is valuable to have the right information to generate clear insights in the effects of complaints as well as the recovery process (e.g. different types of solutions). Furthermore, this information could be used to make frontline employees more aware of the possible effects. The outcome of the results could be used to train employees regarding complaint handling in order to generate maximum customer loyalty.
In the first part, an overview of the current literature regarding this subject will be provided. In the same part, the theoretical framework will elaborate further on the expected relationships between the different concepts. Also, a table with the existing research and two research models
are included to clarify the constructs and its relationships. This will give a clear view for the hypotheses being developed. The method section will provide a description of the two databases being used to describe the sample. Subsequently, a description about the measurements and scales of the constructs is provided. The procedure used to examine actual loyalty is also described. The last part of the method section describes the statistical tests that will be used in order to test the hypotheses. The outcomes of statistical tests will be shown in the Result section. In the Discussion section a critical analysis will be provided of the results. After the Discussion the Limitations of this study will be addressed. Finally, the managerial implications and
suggestions for further research will be presented.
2. Conceptual framework
To research and analyze the consequences of a service failure it is prerequisite to define a service failure and to explain under which circumstances a service failure will occur.
2.1 Service Failure
A lot of research has been done among service failures; as a consequence multiple definitions exist. The definitions are substantially the same, most studies define a service failure as a situation where a service provider does not meet customer expectations in terms of its service products or engages in service behaviors that customers evaluate as unsatisfactory (Lin, 2006). In the first period of service failure research, the method of critical incident technique (CIT) was developed to investigate different types of service failures, service recovery strategies,
satisfaction per recovery strategy and assessing subsequent shopping behavior to check the customer retention rates (Scott W. Kelley et al., 1993). Hoffman et al. (1995) describe two forms of service failures: outcome-oriented (core service failure) and process-oriented (service
encounter failure). As the name already indicates, outcome-oriented failures relate to what customers actually receive from the service, whereas process-oriented failures relate to the manner in which the service is delivered. In other words, the complaint of the customer could thus be about the way in which the service is delivered and about the result of the service. The
interaction of the front line employees with the customers is an example of the process and the end result of a visit at the hairdresser (the actual haircut) is an example of the outcome of the service. A service failure will occur when the expectations of the customer are not met by the service process and/or service outcome (Andreassen & Andreassen, 2000; S. W. Kelley & Davis, 1994). The perceived quality of the service by the customer versus their expectations determines if the expectations will be met. Consequently, a service failure will come out earlier among customers with higher expectations. A logical response of a company is to resolve a service failure in order to keep the customer satisfied and loyal. But a prerequisite for being aware of a failure as a company, a company needs to get feedback from a customer whenever he/she is not satisfied and/or has a complaint. So the first step in trying to solve the problem is to identify and contact customers who have experienced a failure (Spreng et al., 1995). The next step is to respond to the failure. In current literature a lot of research has been done in the area of the recovery process of the service failure. A great part of the service failure literature linked the recovery process to justice theory. In this area it is about how consumers evaluate the recovery process in terms of interactional, distributional and procedural fairness (Chebat & Slusarczyk, 2005; del Río-Lanza et al., 2009; DeWitt et al., 2008; Goodwin & Ross, 1992; Wirtz & Mattila, 2004). A different and smaller part of the literature focused more on the recovery options or solutions for a service failure (Boshoff & Christo, 1997; Fang et al., 2012; Wirtz & Mattila, 2004). On top, the effects of service failures and recovery process are measured by emotions, attitudes and intentional behavior and actual behavior. In the next section there will be more elaborated upon the customer response to failures and its recovery process. But first, to give a clear overview of current research this study provides a framework (figure 1) and a table (table 1) of current relevant studies. In the last column (a) number(s) indicate(s) which relationships are studied in the article.
Figure 1
Table 1
Author Year IV DV Findings Arrow
Goodwin, C Ross, I. 1992 Distributional, procedural and interactional fairness
Satisfaction with the service recovery
A tangible solution,
supplemented by an apology has the strongest effect on satisfaction 3 & 4 Kelley, S. W. Davis, M. A. 1994 - Perceived service quality - Customer satisfaction - Customer organizational commitment - Customer satisfaction - Service recovery expectations
Predictive service recovery expectations are directly influenced by two fundamentally different antecedents- perceived service quality and customer organizational commitment 6 Hoffman, K. Douglas Kelley, Scott W. Rotalsky, Holly M. 1995 - Different failure categories - Different recovery strategies - Failure rating (magnitude) - Recovery rating - Customer retention In general it is possible to recover from almost any failure, regardless of its type or magnitude. Recoveries involving some form of compensation were rated most favorably. 16 Spreng, Richard a. Harrell, Gilbert D. Mackoy, Robert D. 1995 - Original service - Service recovery - Overall satisfaction - Repurchase intentions -word of mouth
Service recovery dominates overall satisfaction formation and positive intentions (repurchase and word of mouth) 10 & 14 Christo Boshoff 1997 - Level of atonement - Time delay - Organizational level of person handling the - Improvement in customer satisfaction
Level of atonement has the most significant impact on service recovery, followed by speed of recovery. The level of the person handling the complaint does not seem to
complaint matter Smith & Bolton 1998 - Type of failure - Severity - Service recovery - Satisfaction
- Intentional loyalty Cumulative customer satisfaction influences repatronage intentions. 10 Tor Wallin Andreassen 2000 - Customer disconfirmation - Equity - Initial negative satisfaction effect - Satisfaction with service recovery
Disconfirmation and equity both have a significant effect on satisfaction with the service recovery. The negative impact from the negative affect caused by the initial service failure on SSR was not confirmed.
6 Weun, Seungoog Beatty, Sharon E. Jones, Michael A. 2004 - Interactional justice - Distributive justice - Severity - Satisfaction with the service recovery
- Satisfaction with the service recovery - Post-recovery trust, commitment and negative WOM - Customer trust and commitment to the organization - WOM
The perceived severity of the service failure has a negative influence on SSR, trust, and commitment and a positive influence on negative WOM.
3 Wirtz, Jochen Mattila, Anna S. 2004 Distributive, procedural and interactive fairness - SSR - repatronage intent and negative WOM
Offering compensation might not add value in situations where the recovery process is well implemented (immediate recovery with an apology). Satisfaction as a mediator explained the relationship between recovery dimensions and post recovery behaviors (repurchase intent and negative WOM). 3, 4 & 12 Chebat, Jean Charles Slusarczyk, Witold 2005 Distributive, procedural and interactive justice - Actual loyalty - Loyalty vs. Exit (via emotions)
Justice affects customer loyalty through emotions.
1, 5 & 13 Bonifield, Carolyn Cole, Catherine 2007 Consumer appraisals about service failure Relatory and conciliatory behaviors (intentional)
Anger plays a powerful role in explaining retaliatory post-purchase behaviors. An intervention, which reduces this anger, decreases intentions to retaliate. 11 Grewal, Dhruv Roggeveen, Anne L. Tsiros, Michael 2008 Compensation Repurchase intentions moderated by the stability of the failure and locus of responsibility conditions
When a company is
responsible for the failure, the effectiveness of the
compensation varies depending on the stability of the failure. When the company is responsible for the complaint and the
complaint is stable, compensation enhances repurchase intentions. DeWitt, Nguyen, Marshall 2008 Justice perception - Positive and negative emotions - Attitudinal Loyalty and behavioral loyalty
High levels of justice lead to positive emotions and create a positive attitude towards the company. Low levels of justice lead to negative emotions and create a lower behavioral loyalty (exit).
1, 2 & 4 del Río-Lanza, Ana Belén Vázquez-Casielles, Rodolfo Díaz-Martín, Ana Ma 2009 - Distributive justice - Procedural justice - Interactional justice - Negative emotions with the SR - SSR
Procedural justice has the strongest direct effect on satisfaction. Procedural justice exerts both a direct effect on satisfaction and an indirect one (via emotions).
1,2 & 3 Komunda, Mabel Osarenkhoe, Aihie 2012 - Service recovery - Communication - Customer satisfaction with complaint handling - Customer loyalty (re-buy intentions) Communication has a significant relationship with service recovery. 10 & 14 Chuang, Shih-Chieh Cheng, Yin-Hui Chang, Chai-Jung Yang, Shun-Wen 2012 - Service failure (process- or outcome oriented) - Failure magnitude - Recovery efforts (tangible or psychological)
SSR When customers encounter an
outcome-oriented failure, customers are likely to be more satisfied with tangible recovery efforts. When encountering a process-oriented failure, they are more satisfied with psychological recovery efforts. Moreover, when encountering a serous process or outcome-oriented failure, there is no difference in customer satisfaction between psychological and tangible recovery efforts.
7 L. Jean Harrison-Walker 2012 - Different attributions - Emotions (Anger, frustration, irritation, disappointment and regret) - Repatronage intention - Share of Wallet - Reconciliation - Negative WOM
The emotional reaction of the customer depends on the customers’ perception of why the service failure occurred in the first place. Behavioral outcomes therefore depend directly on the negative emotion and indirectly on the customers’ perception of the cause of the failure.
11 & 15
2.2 Customer responses to service failures
After the evaluation of the service failure, people will have negative emotions (del Río-Lanza et al., 2009; DeWitt et al., 2008). Chebat and Slusarczyk (2005) argue that not only emotions are experienced, but also that the emotions have an influence on the intentions and behavior consistent with the feelings they experience with the recovery. Wetzer et al. (2007) identify six emotions customers could experience in response to a service failure. These emotions include anger, frustration, irritation, disappointment, regret and uncertainty. When a customer
experiences a service failure he or she will have the tendency to determine the reason why the occasion happened (Jean Harrison-Walker, 2012). It is interesting to know the emotional effects of customers after a service failure and service recovery, but since the costs of gaining a new customer are way higher than the costs of retaining a customer, companies want to know if customers stay loyal and under what conditions (Spreng et al., 1995). In other words, the managerial relevance of knowing emotions is lower than knowing the determinants of customer loyalty. An important determinant of customer loyalty is the satisfaction level of the customer. In current research a clear and consistent relationship is found between satisfaction and repurchase intentions. (see Yi, 1990, p.104 for a review). If the expectations are not met by the perceived quality, customer disconfirmation will be the result and this will subsequently have a negative effect on satisfaction (Andreassen & Andreassen, 2000). For marketers it would be very helpful to know what the effects of a service failure are on customer (intentional) behavior. Since it is clear that a service failure has a negative effect on satisfaction and intentional loyalty this study will first check for this relation with hypothesis 1a and 1b.
H1a. A service failure is negatively related to satisfaction. H1b. A service failure negatively related to intentional loyalty.
As already described in the introduction, current literature about the effects on actual loyalty is very scarce. In most studies subjects react to manipulated scenarios of service failures. The reaction of the subject to the manipulated service failure situation is only measurable in behavioral intent, which does not reflect their actual behavior when the situation would be in
real life. Valuable to know is the effect of service failures on actual behavior. Furthermore, it would be interesting to know whether actual behavior is in line with the intentional behavior of customers. There exists a time gap between the service failure and the repeat purchase of the service. Within this time gap a customer could change their attitudinal state, which could lead to different behavior. Also, due to switching costs, a dissatisfied customer could remain with the company, even though they could have expressed the intent to quit (Chebat & Slusarczyk, 2005). In fact, the direct relationship between service failure and actual loyalty has not been studied yet. Therefore, this study will also check for the relationship between service failure and actual loyalty with hypothesis 1d.
H1c. A service failure is negatively related to actual loyalty.
2.3 Customer responses to double deviations
It is possible that a customer could experience the same service failure for the second time. This means that the customer now experiences a failure in the recovery process. Johnston & Fern (1999) describe this as ‘double deviation’ from expectations. It has not been researched yet what the difference in effect is compared to single deviations. For a double deviation it is expected that the negative effect on satisfaction, intentional loyalty and actual loyalty will be stronger compared to first time complaints (del Río-Lanza et al., 2009). This is expected because the expectations of customers in case of a double deviation situation are higher, and thus harder to meet (Hart et al., 1990). Therefore hypothesis 2 has been developed.
H2. A double deviation is more negatively related to satisfaction, intentional loyalty and actual loyalty compared to a single deviation.
Figure 2
2.4 Company’s response to service failures
After a service failure companies will often try to offer a solution for the error to remain the customers. “The actions a service provider takes to respond to service failures and the process by which the firm attempts to rectify the failure is called service recovery (SR)” (Kelley & Davis, 1994). It is likely that the recovery process is the last experience in time with the company, which could result in a recency effect (Baddeley & Hitch, 1993). Spreng et al., (1995) describes this effect as: “The effectiveness of the service recovery effort may have a greater effect on intentions than the original service failure.” Therefore, it could be that the satisfaction with the recovery will have a greater impact on repurchase intentions and behavior than the satisfaction with the original service failure. On the other hand, an ineffectiveservice recovery could have a potential effect of increasing dissatisfaction (Spreng et al., 1995).
People accept that mistakes sometimes happen, especially in the service industry. The organization’s response to the failure (solution/compensation/explanation) is most likely the cause of dissatisfaction. The satisfaction with the recovery could be seen as a function of: (1) negative effect caused by initial service failure, (2) expectations of service recovery, (3) perceived quality of service recovery, (4) disconfirmation of expectations and (5) perceived fairness of outcome of service recovery (Andreassen & Andreassen, 2000). Disconfirmation occurs when the perceived quality of the recovery does not meet expectations of the customer.
Another antecedent to SSR is equity, which is a relative dimension. “In equity theory the outcome of the interaction is seen as a function of input to the interaction and relative to the output of the other party in the interaction” (Andreassen & Andreassen, 2000). In the service industry an example could be the amount of money a customer pays for a particular service (output for customer) relative to service provider’s outcome, the service itself (input for
customer). (Dis)satisfaction is the result of the summary of the equity/inequity by the customer.
Justice theory
Rio-Lanza et al., (2009) extended the literature by adding the effects of justice on satisfaction. This study focuses more on the way individuals evaluate the service and not the service itself, which is called the effect of perceived justice. According to Rio-Lanza et al., (2009) the level of (in)justice has an effect on emotional response and satisfaction. Up to now, justice is divided into distributive justice, procedural justice and interactional justice.Distributive justice refers to tangible compensations for the service failure offered by the company to the customer.
Examples of compensations are refunding money, discount (for further purchases) or delivering the service again in a proper way. Procedural justice refers to the procedure followed by the firm of the service delivery. Important aspects are accessibility, speed, process control, delay and flexibility to the customer. The third lever is interactional justice and includes customers’ perception about employees’ empathy, courtesy, sensitivity, treatment and the effort the
employee spend on the solution of the failure (Chebat & Slusarczyk, 2005; del Río-Lanza et al., 2009; DeWitt et al., 2008). The three dimensions of justice have positive effects on SSR. Striking is that del Río-Lanza et al., (2009) reported the strongest positive effect of interactional justice compared to distributional and procedural justice, even though there was a low level of personal interactivity. This study was set in the cell phone industry where the firm responded to cell-phone users who have experienced a failure. Chebat & Slusarczyk (2005) did also find a predominant role of interactional justice. This study pointed out that speed of the recovery was a basic requirement. A quick recovery does not automatically imply a more positive effect. In contrast, when the recovery happened too slow, the effect was negative. Furthermore,
distributive justice, the most tangible dimension of justice, has a positive effect on satisfaction and loyalty (Chebat & Slusarczyk, 2005). Wirtz & Mattila (2004) use compensation, response speed and apology to represent the three dimensions of justice. However, this is not how customers perceive the solution for the service failure, but is reaction to the recovery process. This study will focus more on the actual respond to the failure instead of how customers perceive the complaint handling. The results of the studies about perceived justice are very fruitful, but measure only how customers perceive the solution in terms of speed, compensation and interaction. It says little about how customers react to different treatments. For example, when a customer fills in a survey question with the answer that he/she got what he/she deserved or that the result of the complaint was fair, it says nothing about what customers received as tangible solution or if they even received a tangible solution. Moreover, when a customer tells that the company responded quickly to the problem a company does only know the opinion about recovery speed, but not what the actual recovery speed was in time. Also, if a respondent says that the employees of the company were courteous and pleasant to deal with, a company only knows if their employees are interacting in an orderly fashion way. But it is still not known if the employees made an excuse or explanation for example. This is a shortcoming of the literature that connected perceived justice with service failures and the recovery process. This is also recognized by del Río-Lanza et al., (2009) who mentioned that including the response of customers depending on the treatment would be very interesting to know. Furthermore, this study also mentioned that severity of the complaint should be included since this might also influence the reaction of customers.
2.5 Service recovery attributes
Keaveney (1995) shows in his research that the most important cause of switching to another service provider is a service failure. However, a good recovery could lead to switching
resistance, so the service recovery is crucial to keep customers satisfied and loyal (Chuang et al., 2012). Multiple approaches of recovery strategies are applied in the literature. In the first period of service failure research the method of critical incident technique (CIT) was developed to
categories the recovery strategies into eight types. (Scott W. Kelley et al., 1993). Wagner, Bolton, & Smith (1999) divide recovery strategies into three categories: compensation
(economic recovery e.g. discount or coupons), apology (psychological recovery e.g. excuse or explanation) and response speed. Miller, Craighead, & Karwan (2000) further differentiate two types of service recovery: psychological and tangible. In psychological recovery, an apology or explanation is offered. In tangible recovery compensation is provided. Based on this research, a psychological solution is treated as excuse and explanation and a tangible solution is treated as discount/compensation. Based on multiple studies, this study also includes recovery speed as third attribute of service recovery (Boshoff & Christo, 1997; Wagner et al., 1999; Wirtz & Mattila, 2004).
Linking service recovery attributes and nature of the complaint to satisfaction with the service recovery (SSR)
Service recovery: psychological and tangible
It is known that the three justice levels (interactional, procedural and distributional) have a positive impact on SSR (del Río-Lanza et al., 2009; Weun et al., 2004). What is not known is the direct impact of the solution on SSR in a real life situation. Distributional justice refers to the assignment of tangible resources to rectify and compensate for a service failure (Chebat & Slusarczyk, 2005; del Río-Lanza et al., 2009). Within the justice theory a high distributional fairness relates to a fair compensation for the customers’ loss. But the actual tangible recovery (e.g. compensation or discount) is unknown. Therefore it is interesting to know what the effects are of a tangible recovery on SSR. Since distributional justice has a positive impact on
satisfaction and since it refers to tangible resources, a positive impact of a tangible recovery on SSR is expected. Furthermore, interactional justice refers to the employees’ psychological reaction to solve the problem. Again, the effect of the actual reaction of the employee is still unknown. These studies only show that whenever the customer saw the psychological response as fair, this will positively affects SSR. Since interactional justice positively affects the
of psychological recovery (e.g. explanation and/or excuse) on SSR is expected. On the basis of the aforementioned information, hypothesis 3a en 3b are established:
H3a. A psychological recovery has a positive impact on SSR. H3b. A tangible recovery has a positive impact on SSR. Recovery speed
In various studies (e.g. Boshoff & Christo, 1997; Wirtz & Mattila, 2004) recovery speed is one of the attributes of service recovery. They found that a faster response positively impacts the SSR. Furthermore, Chebat & Slusarczyk (2005) concluded that if the service recovery is provided quickly, it does not have positive effects. He qualifies a fast recovery as “basic
requirement”. In other words, customers expect the recovery to be quick. But, when the recovery happened to slow, the reaction was negative. Furthermore, studies that used Justice Theory to analyze the recovery process did use recovery speed to measure procedural justice (Chebat & Slusarczyk, 2005; del Río-Lanza et al., 2009). In these studies they asked the customers if they perceived the recovery as fast. If they perceived it as fast, this was reported as high procedural justice. Consequently, higher levels of justice positively impact satisfaction. Interesting to know would be the actual recovery speed in a real life situation. Taken together, the following
hypothesis is proposed:
H4. A slow recovery is negatively related to SSR.
2.6 Nature of service failure Severity
Several studies about service recovery have demonstrated that the severity of the service failure is an important factor in determining the customer satisfaction following the recovery (e.g. Chuang et al., 2012; Weun et al., 2004) . A more severe service failure leads to a higher perceived loss, which leads to a lower satisfaction level (Harris, Grewal, Mohr, & Bernhardt, 2006). Also, it is more difficult to recover a service failure which customer see as severe (Mattila, 1999). Most studies used a survey method or experiment design. In the survey
approach respondents had to recall an incident that happened within the last 12 months.
Respondents had to report their perceived severity of the incident. A disadvantage of this method is that respondents are more likely to recall a failure, which is on top of mind, and could be more severe in basis. In an experiment design the limitation is clear, the reaction to a more severe failure is negative, but would customers react in the same way in a real live situation? To answer this question and to fill in this gap, this study does not use the perceived severity, but listed different service failures into five different severity categories. This will be further explained in the methodology section. A more serious service failure leads to a narrower customers’ tolerance zone, which potentially leads to customer dissatisfaction (Weun et al., 2004). Therefore, it is expected that the severity of a service failure will have a negative impact on SSR. Hence, current study proposes the following hypothesis:
H5. The severity of a service failure is negatively related to SSR. Service failure: outcome- and process related
The two types of service failures are already shortly discussed in the introduction section. Wagner et al., (1999) identify two types of service failures: outcome- and process-related failures. The first involves a failing in delivering the core service, whereas the second involves a flaw in the delivery of the service. For example, an outcome related failure could be a cold meal in a restaurant and a process related failure could be an unfriendly waiter who did not ask if everything was ok during the diner. “When confronted with different forms of service failures, customers will have different satisfaction feedbacks as outcome- and process-related service failures reflect different forms of loss” (Lin, 2006). As found by Smith & Bolton (1998), the impact on dissatisfaction is larger for outcome failures than for process failures. This study used an experimental study design in a restaurant and hotel setting. The example being used for an outcome-related failure in the hotel setting was a customer traveling on an important business trip who arrived at the hotel and found out that the hotel was overbooked. As a consequence she had to stay in another hotel several miles away. In this example the outcome of the service is different from the expected outcome. For the process-related failure in the hotel setting, the
example being used was a customer who arrived at the hotel and found out that the room as not been cleaned. As a result, the customer got assigned to another (clean) room. In these
experiments the result might be obvious that outcome-related service failures will have a more negative effect on satisfaction than process-oriented service failures. The question rises if these effects are the same in a real life situation. Also, would the same effects occur in other industries where process-oriented failure might have a greater impact? Waiting time in a telephone menu might be very frustrating for many people and is a process-oriented failure. Therefore it would be valuable to know if the same effects will happen in a real life setting and in a different industry. Taken together, the following hypothesis is proposed:
H6: An outcome-related service failure has a more negative impact on SSR compared to a process-related service failure.
2.7 Linking service recovery attributes and the nature of the service failure to actual loyalty
Since recovery efforts also have a positive influence on customer loyalty (Komunda & Osarenkhoe, 2012; Weun et al., 2004) and since SSR is a true antecedent for actual loyalty (Wirtz & Mattila, 2004) the same effects of the main constructs SSR are expected on actual loyalty. Current research did only report effects on intentional loyalty (except from the study of Chebat & Slusarczyk (2005) that investigated the effects of justice on actual loyalty). Since there is no study yet that tested the relationship of the service recovery attributes and nature of the service failure (e.g. severity and type of failure) on actual loyalty, there is a strong need for testing these effects. Current study will try to link this gap by investigating the effects of the main constructs on actual loyalty. Summarizing the aforementioned, the following hypotheses are proposed:
H7a. A psychological recovery has a positive impact on actual loyalty. H7b. A tangible recovery has a positive impact on actual loyalty. H8. A slower recovery is negatively related to actual loyalty.
H9. The severity of a service failure is negatively related to actual loyalty.
H10. An outcome-related service failure has a more negative impact on actual loyalty compared to a process-related service failure.
2.8 Linking SSR to actual loyalty
Customers’ intentional loyalty is higher when customers are more satisfied with the
organization’s service failure recovery (Smith & Bolton, 1998). But as already mentioned in sections before, a lot can happen in the time between the complaint and the repeat purchases of the service, which could lead to switching behavior. Examples are alternative suppliers and switching costs, which will vary per industry (Smith & Bolton, 1999). On the one hand, a prior study has shown that the accuracy of predictions about the relation between intentions and actual behavior is accurate (Morowitz & Schmittlein, 1992). On the other hand, many studies report that more research is needed to completely understand how customers’ SSR influences their actual behavior (Smith & Bolton, 1999). To test if there is indeed a positive relationship between satisfaction with the service recovery and actual loyalty, the following hypothesis is established:
H11. Satisfaction with the service recovery is positively related to actual loyalty.
To summarize, hypotheses 3 to 11 with its relationships among the different constructs investigated in this study are shown in figure 3.
Figure 3
3. Methodology !
In this section, a summary is presented of how the study has been executed. The dataset, measurement of constructs and the statistical analyses are clarified.
3.1 Participants/sample
Whereas most service complaint-handling research were laboratory studies, this study made use of a database analysis. Two datasets were used with a purpose to investigate the direct
relationship between service failure and intentional loyalty, satisfaction and actual loyalty on the one hand. And, the main purpose on the other hand, was to investigate the relationship between type of failure, type of recovery, recovery speed and severity on SSR and actual loyalty. This study is set in the automotive industry and the data is provided by a big car dealership in Rotterdam in the Netherlands. The validity of this research is very high because a well-known third party conducted the data of dataset 1. This data is based on surveys sent to customers who
evaluated the after sales service (car maintenance or repair) provided by the company. Dataset 1 exists from 6139 surveys consisting the years 2011 till 2015. Eighty per cent of the respondents were male. Of these 6139 surveys, 472 were listed as complaints and of those 472 complaints, 166 were double deviations. In order to test if there is a more negative effect on recommendation intention, intentional and actual loyalty in case of a double deviation compared to a single deviation, a separate database was used consisting only complaints.
A different database is extracted from the ‘complaint management system’ of the company to measure the influence of type of recovery, type of failure, recovery speed and severity on SSR and actual loyalty. This is a database where every complaint is tracked, traced and resolved. Even though this dataset has an extensive amount of data, it does not include consumers’ demographic data. In this database the solution is presented and recorded by the service employee who dealt with (owner of the complaint) the complaint. Furthermore the satisfaction of the customer after the solution is included per complaint. Dataset 2 exists from 1471 complaints handled in the complaint management system, also consisting the years 2011 till January 2015. These datasets are appropriate for present study because the needed constructs can be identified after recoding. Even though this study focuses on the automotive industry, the results are generalizable to other contexts. This sector is ranked second in terms of frequency of complaints (right after restaurants and ahead of banks and dental/medical services), as noted by Tax, Brown, & Chandrashekaran (1998).
3.2 Measures
Service failure and double deviation
To decide whether the evaluation of the service is no failure, a failure or a double deviation, the first question asked in the survey is about the overall satisfaction regarding the service. “If you think about your experience from your last workshop visit, how satisfied are you – taken
everything together - about the services of your dealer of this workshop visit. An example of the survey is provided in the appendix section. A 5-point Likert scale was used with five possible responses: extremely satisfied, very satisfied, satisfied, not so satisfied and totally not satisfied.
Whenever the answer is ‘not so satisfied’ or ‘totally not satisfied’ the evaluation is reported as a complaint. This study treats a complaint as a service failure. To decide whether a service failure is a double deviation, three questions were asked in the survey; (1) have you been to a dealership before for the exact same reason? If yes; (2) did you go to <Dealership>? If yes; (3) was that due to a previous repair that was not performed correctly or incomplete at <Dealership>? If
answered yes to the three questions, the complaint was recorded as double deviation.
Satisfaction and Intentional loyalty
As stated in the previous section, the overall satisfaction is determined by the question: “If you think about your experience from your last workshop visit, how satisfied are you – taken everything together about the services of your dealer of this workshop visit?” The intentional loyalty is determined by the question: “Would you visit this this dealer again for your next repair or service maintenance?” Again a 5-point Likert scale is used for the last two questions with five possible responses: sure, probably, maybe yes maybe no, probably not and certainly not.
Actual loyalty
To check whether a customer is actual loyal, a database was created by the company with all (276.688 in total) the invoices per license plate of the years 2011 YTD. Taking into account that every maintenance or repair is coupled with an invoice number and that the date of the invoice is the same date as the date of the visit, this database was sufficient to examine if a customer came back after a service failure for maintenance or repair. With the VLOOKUP function in Microsoft Excel the license plate number was looked up and the date of invoice was compared to the complaint date. When there existed an invoice and the difference between the two dates was more than 90 days, the customer came back for another service or maintenance action, thus stayed loyal to the company. All the cars in this database have a maximum service interval of a year. Therefore the complaints of the last year were excluded from the database in order to check for actual loyalty in a proper way. Otherwise, it was not possible to check for actual loyalty for the complaints after January 2014. In dataset 1, 472 of 6139 questionnaires were recorded as a
complained to the dealership, 291 stayed loyal to the company. Of the 1471 complaints in dataset 2, 934 stayed loyal to the company. After excluding the complaints after January 2014 in dataset 2 to ensure that some actual loyal customers were not ‘missed’, the dataset contained 1110 complaints. Of those complaints 732 customers stayed loyal.
Type of failure: process- and outcome-related
In dataset 2, the complaints were listed among 13 categories. To decide whether a complaint was a process- or outcome-related complaint, three managers of the company gave input to decide whether a complaint sort belonged to an outcome- or process-related failure. Every complaint category was then listed as process- or outcome oriented complaint. Table 2 provides an overview of the 13 categories with the associated type of failure. Table 2 provides an overview of the 13 categories as listed in the ‘complaint management system’.
Table 2
Category Outcome- versus
Process-related service failure
Double deviation Outcome
Customer treatment Process
Appointments not met Outcome
Implementation activities Outcome
Guarantee / leniency Outcome
Billing Process
Communication Process
Waiting time counter / telephone Process
Waiting time planning Process
Replacement vehicle Outcome
Pickup and delivery service Outcome
Quality product Outcome
Type of solution
In dataset 2 the solution is recorded per complaint as a text by an employee who handled the complaint. The types of recoveries (Boshoff & Christo, 1997; Wirtz & Mattila, 2004) are extracted from the added notes of the employee responsible for complaint using a the text filter “contains” in Excel. This is done with the help of a list of words, which indicates the type of solution. The researcher in association with three managers of the company prepared the list of words. The types of recoveries are tangible recovery and psychological recovery. 13 words were used to indicate if a solution was a tangible one and for the psychological solution also 13 words were used. To be sure about the outcome all the texts were read twice to secure the quality. This was needed due to the fact that for example both ‘discount’ and ‘no discount’ will be seen as a tangible solution. The list of words is provided in the appendix section.
Severity
The severity of the complaint (Weun et al., 2004) will be extracted from the 13 possible type of complaints as listed in dataset 2 from the complaint management system. To determine the severity three managers of the company listed the severity per type, the highest manager made the final list in the end. Severity is measured in a scale of 1 to 5. One is the lowest level of severity and five the highest level.
Table 3
Complaint type Severity manager 1 Severity manager 2 Severity manager 3 Double deviation 5 5 5 Customer treatment 2 5 3
Appointments not met 5 4 5
Implementation activities
5 5 5
Billing 2 2 1
Communication 2 4 2
Waiting time counter / telephone 1 3 2 Waiting time planning 1 2 1 Replacement vehicle 2 2 1
Pickup and delivery service
2 1 1
Car not ready on time 2 2 3
Quality product 3 3 4
Recovery speed
The time in net work days was calculated in Excel as the difference between the complaint date and the date the complaint was settled.
Satisfaction with the service recovery
Satisfaction with the service recovery is listed in four possible categories: extremely satisfied, satisfied, neutral and not satisfied. The employee who handled the complaint indicated the level of satisfaction of the customer after the failure was handled.
3.3 Data analysis
At first the data will be prepared for analysis in both Microsoft Excel and SPSS Statistics 21 (IBM SPSS Statistics for Mac 21, 2012). This is done by cleaning the data and excluding
missing and/or wrong values. The missing license plates were excluded as well as wrongly noted license plate numbers. This is done because the license plate number is the leading link to check for actual loyalty in the database with all the invoices. Also wrongly noted dates were excluded because both recovery speed and actual loyalty need correct dates to be reliable. In model 1 the independent variable will be service failure and the dependent variables will be satisfaction, intentional loyalty and actual loyalty. In model 2 the independent variables will be types of
recoveries, types of failures, severity and recovery speed. The dependent variables will be SSR and actual loyalty.
At first, model 1 will be analyzed by using a correlation matrix. To measure the relations between the independent variable complaint and the dependent variable intentional loyalty a linear regression is performed. In the dataset the level of satisfaction depends whether customers have a complaint, while in a real life setting a complaint should lead to a lower satisfaction (Andreassen & Andreassen, 2000). Since criteria for complaint is overall satisfaction level, satisfaction is excluded from the linear regression analysis as depended variable of complaint. To measure the relation between complaint and actual loyalty a binary logistic regression is performed. This is done since actual loyalty is a binary variable and a binary logistic regression fits better in case of a binary dependent variable (Osborne, 2008)
As discussed in the sample section a separate database only including the complaints is used to test if a double deviation has a more negative effect intentional and actual loyalty. Again a simple linear regression is performed to measure the relationship between the independent variable double deviation and the dependent variable intentional loyalty. Also, a binary logistic regression is performed to measure the relationship between the independent variable double deviation and the dependent variable actual loyalty.
Second, model 2 will be analyzed using a correlation matrix to investigate the relationships between the main constructs. To measure the relationship between the independent variables types of recoveries, recovery speed, types of failures and severity of complaint and the
dependent variable SSR an ordinal regression is performed. This analysis will better suit the data because SSR is measured on an ordinal scale (O’Connell, 2006). To measure the relation
between the dependent variables types of solutions, recovery speed, types of failures and
severity on the dependent variable actual loyalty, again a binary logistic regression is performed. Furthermore, the relationship between SSR as independent variable and actual loyalty as
discussed, for analyses with actual loyalty as dependent variable, dataset 2 is used without the cases of the last year to ensure actual loyalty will be measured in a proper way.
4. Results
4.1 Descriptive statistics model 1
The means, standard deviations and Pearson-correlations between variables are shown in table 4. As expected, it can be seen that the correlations between complaint and satisfaction (-0.83) and intentional loyalty (-0.72), show a moderate/high negative correlation. The correlation between double deviation and satisfaction and intentional loyalty is also negative, but less strong compared to complaint. Complaint and double deviation are also negatively correlating with actual loyalty, but the effect is less strong.
Table 4 Means (M), standard deviations (SD), and correlations between variables (N = 6138). Variables M SD 1 2 3 4 5 1. Complaint 0.08 0.27 - 2. Double Deviation 0.03 0.16 .58** - 3. Satisfaction 3.66 0.69 -.83** -.49** - 4. Intentional loyalty 91.40 33.84 -.72** -.43** .88** - 5. Actual loyalty 0.72 0.45 -.07** .03* .07** 0.10** - Note: *p < .05. **p < .01.
4.2 Hypothesis testing model 1
4.2.1 Service failure as effect on intentional and actual loyalty
As discussed in the method section, this study will not test hypothesis 1a since the level of overall satisfaction determines whether a filled in survey is a complaint or not. To test
hypothesis 1b, a simple linear regression was performed to test whether complaint could predict the level of intentional loyalty. See Table 5. To test hypothesis 1c, a binary logistic regression
was performed to test whether customers who complained are less likely to be actual loyal. See Table 6.
Table 5 Simple linear regression with intentional loyalty as dependent variable (N = 6138). Intentional loyalty Β SE β t Complaint -90.878 1.133 -.715** -80.21 R² 0.512 Note: **p < .001.
The model, in which complaint was entered as independent variable and intentional loyalty as dependent, is statistically significant F (1, 6138) = 6434.05, p < .001 and explains 51 % of the variance in intentional loyalty. The predictor complaint is statistically significant (β = -0.715, p < .001). Hypothesis 1b is confirmed.
Table 6 Binary logistic regression with actual loyalty as dependent variable (N=6139).
Actual loyalty 95% C.I. for Exp(B)
Β SE Exp(B) Lower Upper
Complaint -0.516 0.099 0.597** .491 .725
Note: **p < .0001. Omnibus tests of model coefficients (p = .000), .006 (Nagelkerke)
The model, in which complaint was entered as independent variable and actual loyalty as dependent variable, is statistically significant (p < .001). Customers who complain are 0.597 times more likely (40 % less likely) to be loyal than customers who did not complain. In other words, customers who complain are (1 / 0.597 = 1.675) times more likely not to be loyal. The contingency table for Hosmer and Lemeshow Test predicted 4133 customers to be actually loyal and 4133 were observed as actually loyal. In other words, the model predicted 100 % right. Without the model, thus only basing on the loyalty figures, only 72,1% will be predicted correctly. Hypothesis 1c is confirmed.
4.2.2 Double deviation as effect on intentional loyalty and actual loyalty
As already discussed in the sample description, a separate database is used to measure the effects of a double deviation compared to a single deviation. Again a simple regression is performed to measure the effect of double deviation on intentional loyalty. See table 7. Also, a binary logistic regression was performed to test whether customers who had a prior service experience will be less likely to be actual loyal. See Table 8.
Table 7 Simple linear regression with intentional loyalty as dependent variable (N=472).
Intentional loyalty
Β SE β t
Double deviation -4.954 2.848 -.080 -1.739
R² 0.006
The model, in which double deviation was entered as independent variable and intentional loyalty as dependent, is not statistically significant F (1, 470) = 3.025, p = .083. The predictor double deviation is not statistically significant (β = -.080, p = 0.083). Hypothesis 2b is rejected.
Table 8 Binary logistic regression with actual loyalty as dependent variable (N=472).
Actual loyalty 95% C.I. for Exp(B)
Β SE Exp(B) Lower Upper
Double deviation .411 0.117 1.202 .813 1.779
Note: .002 (Nagelkerke)
The Omnibus test of model coefficients is not significant (p = .355). The contingency table for Hosmer and Lemeshow Test predicted 107 customers to be actual loyal and 107 were observed as actual loyal. In other words, the model predicted 100 % right. Without the model, thus only basing on the loyalty figures, only 61,7% will be predicted correctly. Customers who
complained about a double deviation are 1.202 more likely (20 % more likely) to be loyal than customers who had a normal complaint. However, the model in which double deviation was
entered as independent variable and actual loyalty as dependent variable is not statistically significant (p = .356). Hypothesis 2 is rejected.
4.3 Descriptive statistics model 2 – Dataset SSR
The means, standard deviations and Pearson-correlations between variables are shown in table 9. As the correlation shows, severity is negatively related to SSR as expected, but the relationship is not significant. Outcome oriented failures are negatively related and statistically significant with SSR. Striking is the positive relationship between handling time and SSR since we expected a negative relationship. Furthermore, tangible solution is positively correlated with SSR and the relationship is significant. Contradictory, psychological recovery and SSR are negatively related and this relationship is also significant.
Table 9 Means (M), standard deviations (SD), and correlations between variables (N = 1471). Variables M SD 1 2 3 4 5 6 7 1. SSR 2.55 .97 - 2. Outcome related .37 .48 -.05* - 3. Process related .62 .48 .05* -1.00 - 4. Psychological recovery .20 .40 -.117** .02 -.02- 5. Tangible recovery .97 .17 .06* -.04 .04 -.03 - 6. Severity 3.50 1.73 -.01 .86** -.86** -.04 .05 - 7. Recovery speed 10.66 18.27 .06* .07* -.07* .05* -.03 -.10 - Note: *p < .05. **p < .01.
4.4 Descriptive statistics model 2 – Dataset Actual Loyalty
The dataset used in this descriptive statistics model is the same as the dataset in the last
descriptive model, but in this dataset one year from the end date of the dataset was deleted. The means, standard deviations and Pearson-correlations between variables are shown in table 10. The most important relationship seen in this correlation matrix is between SSR and actual loyalty with a positive and significant relationship.
Table 10 Means (M), standard deviations (SD), and correlations between variables (N = 1110). Variables M SD 1 2 3 4 5 6 7 8 1. Actual loyalty .66 .47 - 2. Outcome related .65 .48 .05 - 3. Process related .36 .48 -.05 -1.00** - 4. Psycho- logical recovery .15 .36 -.03 -.02 .02 - 5. Tangible recovery 0.96 .18 .01 -.04 .04 -.04 - 6. Severity 3.64 1.60 .06* .85** -.85** .01 -.04 - 7. Recovery speed 11.34 19.49 -.06* -.07* .07* -.09** .03 -.11** - 8. SSR 2.44 .95 .10** .04 -.04 .12** -.06* -.01 -.06* - Note: *p < .05. **p < .01
4.4 Hypothesis testing model 2
4.4.1 Testing the determinants of SSR: type of recovery, recovery speed, type of failure and severity.
To test hypotheses 3a, 3b, 4, 5 and 6, an ordinal regression was performed to test the effects of the different determinants of SSR. See table 11.
Table 11 Ordinal regression with SSR as dependent variable (N = 1471).
SSR 95% C.I. for Estimate
Estimate SE Odds Lower Upper
Ratio 1. Handling time .005 .381 1.005 .000 .010 2. Severity .186** .054 1.204 .081 .292 3. Process oriented -.770** .192 0.463 -.393 -1.147 4. Outcome oriented .770** .192 2.160 .393 1.147 5. Tangible recovery .585* .282 1.795 .032 1.138 6. Psychological recovery -.540** .119 0.583 -.774 -.307 Note: *p < .05. **p < .01, .032 (Nagelkerke)
Model fitting information shows a significant level (p = .000). Pearson Goodness-of-fit is not significant (p = .316). The estimate coefficients are logits, taking the exponent of the logits will
give the odds ratio. The odds ratio is a better way to interpret the impact of the independent variables on the dependent variable. In hypothesis 3a and 3b, it was expected that a
psychological and a tangible recovery both would have a positive effects on SSR. As seen in table 8 the coefficient of psychological recovery is negative (-.540) and is significant (p = .000). Taking the exponent of the logit the result shows that in case of giving a psychological solution compared to no psychological it is 0.583 times as likely a customer will have a higher level of SSR. As a result, hypothesis 3a is rejected. The coefficient of tangible solution is positive (.585) and significant (p = .038). Taking the exponent of the logit the result shows that providing a tangible solution compared to no tangible solution it is 1.796 times as likely to create a higher level of SSR. Therefore, hypothesis 3b is confirmed.
Again looking at table 11, the coefficient of handling time is positive (.005, p = .066). Even though the p value is above the cut off level, the significance levelis not far from the .05 significance level. Also, handling time is measured at a continuous scale. This means that one day extra handling time is 1.005 times the odds that 1 day extra handling day leads to one level higher satisfaction. But in case the handling time increases with 20 days, the odds ratio will be (EXP(.005*20)) = 1.105. Instead of a negative effect, the model shows a positive effect of a slower handling time, and is not significant. Therefore, hypothesis 4 is rejected.
In hypothesis 5 it was expected that a higher severity would lead to a lower SSR. Looking at table 11, it is seen there is a positive coefficient (.186) for severity and is significant (p = .001). Again taking the exponent of the logits will give the odds ratio. So customers with a more severe service failure, in terms of 1 level higher severity (e.g. from 2 to 3), are 1.204 more likely to have a higher level of SSR. Therefore, hypothesis 5 is rejected.
In hypothesis 6 it was expected that an outcome related service failure would have a more negative effect on SSR compared to a process related service failure. Looking at table 11, the results show a significant (p = .000) positive (.770) coefficient for outcome related failures indicating that outcome related service failures are 2.160 times more likely (compared to process oriented service failures) to lead to a higher level of SSR compared to process related service
failures. Contradictory, process oriented service failures show a negative (-.770) and significant (p = .000) coefficient. This means that process oriented service failures are 0.463 times more likely to lead to a higher level of SSR (compared to outcome related service failures). The results are exactly the opposite because outcome oriented failure was the reference group for process oriented failure in the ordinal regression is SPSS. Summarizing, hypothesis 6 is rejected.
4.4.2 Testing effects of the main constructs of SSR on actual loyalty
To test hypotheses 7a, 7b, 8, 9 and 10, a binary logistic regression was performed to test the effects of the determinants of SSR (type of recovery, type of failure, severity and handling time) on actual loyalty. See table 12.
Table 12 Binary logistic regression with actual loyalty as dependent variable (N=1110).
Actual loyalty 95% C.I. for Exp(B)
Β SE Exp(B) Lower Upper
1. Outcome related -.05 0.117 .951 .580 1.557 2. Psychological -.20 .176 .816 .577 1.152 related 3. Tangible recovery .12 .342 1.124 .575 2.197 4. Handling time -.01 .003 .994 .988 1.000 5. Severity .09 .076 1.093 .943 1.268 Note: .0.101 (Nagelkerke)
In the model the Omnibus tests of model coefficients is not significant (p = 0.101). Also, the Hosmer and Lemeshow Test is not significant (p = 0.279). In order to have a good model the significance level has to be higher than .05. So the model is good regarding this test. The contingency table for Hosmer and Lemeshow Test indicates that in case of 93 observed loyal customers, the model predicts 87.934 loyal customers. Without the model, thus only basing on the loyalty figure, only 66.0 % will be correctly predicted. However, looking at table 9, none of the direct effects of the determinants of SSR on actual loyalty are significant. Therefore
