Monetary and non-monetary overcompensation for severe service failures : impact on perceptions of fairness and negative-word of mouth intent

(1)

I

MONETARY AND NON-MONETARY OVERCOMPENSATION

FOR SEVERE SERVICE FAILURES:

Impact on perceptions of fairness and negative-word of mouth intent

Suzanne Flendrie (11110686) University of Amsterdam

Faculty Economics and Business Thesis MSc Business Administration Specialization: Marketing

Supervisor: Adriana Krawczyk

Date of submission (final version): January 27, 2017 Word count: 16123

(2)

II

Abstract

Research shows that severe service failures lead to very negative customer responses. Service firms should offer their customers overcompensation to minimize these negative outcomes. Within the context of denied service in the airline industry, this study examines the effect of overcompensation on customers’ negative word-of-mouth intent and to what extent this relationship is mediated by customers’ perceptions of distributive justice (fairness). A between-subjects design was used to test the hypotheses. Overcompensation level was manipulated at three levels (125, 150 and 200%) and tested with two overcompensation types (monetary and non-monetary). Full compensation (100%) was used as the control group. Data was collected from 254 consumers with a scenario-based online survey. This is the first study that compares similar levels of monetary and non-monetary overcompensation. The results show that service firms should offer consumers non-monetary overcompensation rather than offering monetary overcompensation or full compensation. Non-monetary overcompensation leads to significantly lower negative word-of-mouth intent than monetary overcompensation and full compensation. Furthermore, more compensation is not always better. Higher levels of overcompensation are not perceived as fairer than lower levels of overcompensation. Previous research indicates that perceptions of distributive justice fully mediate the relationship between overcompensation and negative word-of-mouth. This is the first study that shows that perceptions of distributive justice partially mediate the relationship between overcompensation and negative word-of-mouth intent. This research holds important managerial implications of tailoring service recovery strategies in the most effective and efficient way and improving business outcomes accordingly.

Key words: Overcompensation; Severe service failure; Perceived distributive justice;

(3)

III

List of Figures and Tables

Figures

Figure 1 Conceptual Model

Figure 2 S-shaped value curve

Figure 3. Visual representation of cell means for PoDJ

Tables

Table 1 Overview of the seven experimental conditions

Table 2 Descriptives, Correlation and Reliability Matrix

Table 3 Hierarchical Regression Model of PoDJ

Table 4a ANOVA Overcompensation Type and PoDJ

Table 4b ANOVA Multiple Comparisons Overcompensation Type and PoDJ

Table 5a ANOVA Experimental Conditions and PoDJ

Table 5b ANOVA Multiple Comparisons Experimental Conditions and PoDJ

Table 6 Results of ANCOVA test PoDJ

Table 7 Hierarchical Regression Model of NWOM intent (I)

Table 8a ANOVA Overcompensation Type and NWOM intent

Table 8b ANOVA Multiple Comparisons Overcompensation Type and NWOM intent

Table 9 Hierarchical Regression Model of NWOM intent (II)

(6)

VI

Statement of Originality

This document is written by Student Suzanne Flendrie 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 responsible solely for the supervision of completion of the work, not for the contents.

(7)

VII

Acknowledgements

I would like to express my gratitude for the support of several individuals in carrying out this research. First, I would like to thank all respondents for taking part in my online survey. This research would not have been possible without their participation. Second, I would like to thank my supervisor Adriana Krawczyk for her guidance and the knowledge that she shared with me. The feedback of Ms. Krawczyk really helped me to improve my research model and to improve the attainability of my study.

(8)

1

Introduction

In 2015, EasyJet, Britain’s biggest budget airline, oversold thousands of peak-season flights and, therefore, thousands of passengers were denied boarding. Despite the European flight regulations that require airlines to offer a new flight to denied passengers at the earliest opportunity, many passengers had to wait over 32 hours for a flight. In addition, European regulations require that anyone who is denied boarding should immediately be offered compensation ranging from €200 for short flights to €600 for long flights. Numerous passengers reported being very dissatisfied with the airline, as it took over 8 weeks to pay the required compensation (Calder, 2015). EasyJet might have been wise to sufficiently recognize the importance of a good service recovery strategy. Indeed, previous research shows that companies should build long-term relationships with customers, as it is less expensive to retain existing customers than to attract new ones (Spreng, Harrell and Mackoy, 1995; Maxham, 2001). Service recovery refers to a firm’s actions in response to a service failure (Grönroos, 1988). It is one of the most important aspects in building relationships with customers, as service recovery efforts could enhance customer satisfaction and behavioral intentions in the event of flawed service (Kotler and Keller, 2009).

In the event of a service failure, compensation has proven to be the main driver of customer satisfaction (e.g., Gelbrich and Roschk, 2011; De Ruyter and Wetzels, 2000; McCollough, Berry and Yadav, 2000; Weun, Beatty and Jones, 2004) and behavioral intentions (e.g., Conlon and Murray, 1996; Kwon and Jang, 2012; Maxham, 2001; Sparks and McColl-Kennedy, 2001) and the effect of compensation remains constant over time (Fang, Luo and Jiang, 2013). Besides, Maxham (2001) and Smith and Bolton (2002) show that service recovery is a key factor in building relationships with customers who were initially unhappy with the service encounter. Thus, due to its importance, the topic of compensation has received substantial attention in the marketing literature and has gained a fixed place in service recovery

(9)

2 strategies of various product and service providers (Kotler and Keller, 2009). Earlier research suggests that some compensation is better than no compensation, but there seems to be no agreement on the most effective type and amount of compensation in case of a service failure (Davidow, 2003).

Smith et al. (1999) argue that the effectiveness of compensation depends on the severity of the service failure. According to Weun et al. (2004, p. 135), “Service failure severity refers to a customer’s perceived intensity of a service problem. The more intense or severe the service failure, the greater the customer’s perceived loss”. Severe service failures consequently lead to more negative customer responses (Weun et al., 2004). To overcome the great loss that is perceived due to a severe service failure, firms should provide the customer with a relatively high amount of compensation. Consequently, severe service failures are very likely to make companies offer customers overcompensation (Noone and Lee, 2011). Overcompensation represents compensation with a higher cash value than the purchase price (>100%) (Estelami, 2000).

While much of the literature on compensation focuses on the effect of simple compensation (≤100%), there is little research about the role of overcompensation in service recovery (Noone, 2012). Firms aim to minimalize their service recovery costs. Therefore, unless overcompensation provides a substantial benefit compared to simple compensation, the extra expense may be unnecessary. Knowing whether overcompensation is more effective than simple compensation holds significant managerial value as it enables firms to manage service recovery costs in the most efficient way possible.

Most of the studies that incorporate overcompensation only focus on the effect of overcompensation on customer satisfaction and behavioral intentions (e.g., Mack, Mueller, Crotts and Broderick, 2000; Smith, Bolton and Wagner, 1999). The previous literature, thus, lacks understanding of the potential mechanisms influencing this relationship. Prior studies use

(10)

3 the theory of distributive justice to clarify the positive effect of compensation on customer satisfaction (e.g., Mattila and Patterson, 2004; Smith et al., 1999; Lind and Tyler, 1988). In line with these studies, Tax, Brown and Chandrashekaran (1998) show that compensation is the most important dimension influencing customers’ perceptions of distributive justice (PoDJ). Previous research mainly focuses on the influence of simple compensation on PoDJ; however, evidence regarding the effect of overcompensation on PoDJ is lacking (Noone, 2012). Hence, this research extends prior research by examining the impact of overcompensation on PoDJ.

Noone (2012) already partly fills the aforementioned gap by examining the mediating effect of PoDJ in the relationship between overcompensation and (NWOM) intent in the context of a severe service failure. Noone’s is the only study that proves the mediating effect of PoDJ on the relationship between overcompensation and negative word-of-mouth (NWOM) intent. She states that it would be valuable to replicate her study as it might increase its reliability. However, her study only determines the effect of monetary overcompensation (MO), not that of non-monetary overcompensation (NMO), on NWOM intent.

Although monetary compensation (e.g. a cash refund) is the most commonly used form of compensation (Davidow, 2003), research shows that monetary compensation does not suit every situation (Roschk and Gelbrich, 2014). Research suggests that non-monetary compensation (e.g. a complimentary product) should be used instead (Fu, Wu, Huang, Song and Gong, 2015). Noone (2012) states that the subject of NMO is absent in the literature and added this void to her implications for future research. This study fills this important gap that exists in the literature and will thus focus on NMO as well as on MO and their effect on PoDJ and NWOM intent.

NWOM plays a vital role in influencing firm reputation and it is thus essential for marketers to understand effective service recovery strategies to avoid NWOM behavior (Noone, 2012). If firms aim to effectively manage expenditures within their service recovery strategies,

(11)

4 it is of great importance to understand the effect of different forms and levels of overcompensation. Overcompensation might be perceived as fairer than simple compensation or a particular overcompensation type may lead to more NWOM behavior than another overcompensation type. Also, within these types, the level of overcompensation may have an impact on NWOM behavior.

Furthermore, the different levels and types of overcompensation involve certain costs. Gilly and Hansen argue that non-monetary compensation often involves lower costs (e.g. a complimentary product worth €40 in store only costs the firm €4 production costs) than monetary compensation (a cash refund of €40 actually costs the firm €40). Consequently, knowledge of the most effective overcompensation type enables managers to best manage their service recovery expenditures. This study thus has important managerial implications as it provides insights that will guide service firms to most effectively and efficiently manage their service recovery strategies and improve business outcomes accordingly.

Intrigued by the gap in the overcompensation literature and the managerial relevance, the research question of this study is: What is the impact of overcompensation type on negative

word-of-mouth (NWOM) intent and to what extent is this relationship mediated by perceptions of distributive justice (PoDJ)?

This study employs a between-subjects design. Overcompensation level is manipulated at three levels (125, 150 and 200%) and overcompensation type is manipulated at two levels (monetary and non-monetary). Full compensation (100%) is used as the control group. Data is collected from 254 consumers with a scenario-based online survey. The results are analyzed using the Statistical software Package for Social Sciences (SPSS). The method of Baron and Kenny (1986) is used to test for mediation.

This master thesis consists of four subsequent chapters. Chapter 1 provides an overview of the literature on the relevant topics. Based on this overview, hypotheses are stated. Chapter

(12)

5 2 outlines the research methodology. Consequently, the results of the survey data are presented in Chapter 3. Finally, Chapter 4 discusses these results, the limitations of the study and directions for future research and ends with an overall conclusion.

(13)

6

1. Literature Review

This chapter discusses the previous literature on the relevant topics and the hypotheses of this research. Sections 1.1 and 1.2 focus on the concept of overcompensation and the difference between MO and NMO. Sections 1.3 and 1.4 explain the theory of PoDJ and the general effect of overcompensation on PoDJ. Section 1.5 focuses on the impact of overcompensation type on PoDJ. Section 1.6 describes the current literature related to the effect of overcompensation levels on PoDJ. Section 1.7 focuses on the potential mediating effect of PoDJ on the relationship between overcompensation type and NWOM intent. The hypotheses that flow from the discussed theory in the framework are presented in a conceptual model in Section 1.8.

1.1. Overcompensation

The level of compensation is determined by the total cash equivalence of the monetary and non-monetary reimbursements. There are two main compensation levels: simple compensation and overcompensation. Reimbursement that is less than (<100%) or equal to (100%) the purchase price of the initial product or service represents simple compensation. Overcompensation, on the other hand, represents reimbursement with a higher cash value than the purchase price (>100%) (Gelbrich and Roschk, 2011).

Gilly and Hansen (1985) have a different way of grouping the levels of compensation; their study divides compensation into underbenefiting, equity and overbenefiting. The underbenefiting approach entails ignoring customer complaints or responding to complaints only with an apology; an approach that is mostly executed by firms with a cost leadership strategy. The equity approach intends to bring customers back to their pre-problem position; the firm does just enough to not lose anything on the transaction. The overbenefiting approach aims not only to bring the customer back to their pre-problem position, but also offer them something extra; a strategy aimed at enduring profit rather than controlling costs. The

(14)

7 overbenefiting approach is similar to overcompensation. Gilly and Hansen (1985) examine the effect of overcompensation in the hotel business. In their experiment, hotel guests who are denied service due to overbooking are offered a free dinner and a weekend stay. Their findings show that overcompensation has a significant positive effect on satisfaction, repurchase intent and positive WOM intent.

Megehee (1994) and Noone and Lee (2011) report mixed results regarding overcompensation and its effect on customer responses. Megehee (1994) studies the response of customers of dry cleaners to different levels of compensation, ranging from 50 to 300 percent of their bill. Her findings show that overcompensation increased satisfaction, but not repurchase intent. In their study on hotel guests who were denied service, Noone and Lee (2011) also indicate that cash-based overcompensation increased satisfaction, but not repurchase intent. Additionally, they show that credit-based (voucher-based) overcompensation does not increase customer satisfaction or repurchase intent. Furthermore, Garrett (1999) does not report any positive effects of overcompensation. In his study on product failures, customers are offered overcompensation ranging from 100 to 300 percent of the retail value of the product. The results of his study do not show that overcompensation significantly increases customer satisfaction, repurchase intent or positive WOM intent.

The results of the different studies on overcompensation show that there is no agreement about the effectiveness of overcompensation. Additionally, Noone and Lee (2011) show that the effect of overcompensation can vary depending on the type of overcompensation.

1.2. Monetary (MO) and Non-Monetary Overcompensation (NMO)

According to Estelami (2000), there are several types of compensation. First, firms may offer the complainant to keep the defective product or service (e.g., a mirror with a crack in it, a bad manicure treatment), to return it or to not pay for it. Then, firms may provide complainants with monetary and/or non-monetary reimbursement. Monetary reimbursement may entail a

(15)

8 refund or discount, whereas non-monetary reimbursement may entail a new product or service (e.g., a new mirror, a new manicure treatment) and/or a different product or service (e.g., complimentary transportation costs, a pedicure treatment).

The division into monetary and non-monetary compensation is also visible in the literature on overcompensation. Previous literature studied MO by means of cash- or credit-based compensation as a percentage of the original purchase price of the product or service (Megehee, 1994; Garrett, 1999; Noone, 2012; Noone and Lee, 2011). NMO is mostly studied as a complimentary service offered to customers in different hospitality businesses. The respondents in the experiment of Gilly and Hansen (1985) are offered a free dinner and weekend stay for being denied service at a hotel. In the study of Boshoff (1997), airline customers are offered a complimentary airline ticket for their missed flight connection because of a delay caused by the airline (some with a higher price than the original ticket price). In a study on the restaurant business by Hocutt, Bowers and Donovan (2006), customers are offered a free meal as compensation for being served the wrong entrée.

1.3. Perceptions of Distributive Justice (PoDJ)

Many studies show that complainants require a ‘fair’ problem resolution in the event of a service failure (e.g., Bies and Moag, 1986; Blodgett, Hill and Tax, 1997; Fu et al., 2015). The perceptions of fairness influences their satisfaction with the service provider as well as their future behavioral intentions (McColl-Kennedy and Sparks, 2003). Customers’ perceptions of the fairness of the firm’s service recovery efforts depend on distributive, procedural and interactional justice.

Distributive justice is defined as the extent to which the customer perceives the actual, tangible outcomes of the complaint as acceptable for offsetting the product or service failure (Blodgett et al., 1997; Smith et al., 1999; Zhao, Lu, Zhang and Chau, 2012). Procedural justice refers to the extent to which the customer perceives the visible policies and procedures used to

(16)

9 reach the actual outcomes as fair (Davidow, 2003; Thibaut and Walker, 1975; Hon and Lu, 2010). Interactional justice concerns the way the customer is treated throughout the service recovery process (Bies and Moag, 1986; Zhao et al., 2012); for example, whether there was a show of courtesy and respect (Davidow, 2003).

Prior studies use the theory of distributive justice to clarify the positive effect of compensation on customer satisfaction (e.g., Fu et al., 2015; Mattila and Patterson, 2004; Smith, Bolton and Wagner, 1999). In line with these studies, Tax et al. (1998) and Zhao et al. (2012) show that compensation is the primary dimension influencing customers’ PoDJ. Hence, the justice dimension that applies in this study is distributive justice. Previous research mainly focuses on the influence of partial or full compensation on PoDJ, but evidence regarding the effect of overcompensation on PoDJ is lacking (Noone, 2012).

1.4. Overcompensation and Perceptions of Distributive Justice (PoDJ)

According to equity and social exchange theories, exchange relationships should be in balance, meaning that resources should be exchanged equally (Cropanzano and Mitchell, 2005; Deutsch, 1973; Hatfield, Walster, Walster and Berscheid, 1978). In case of a service failure, the exchange relationship between the customer and the service provider has fallen out of balance. The unbalanced exchange relationship can be restored by providing the customer with an outcome in line with the magnitude of the experienced service failure (Smith et al., 1999). This is in line with the findings of Blodgett et al. (1997), who state that customers require fair solutions for service failures and their perceptions of the fairness of the solutions depends on how the actual, tangible outcomes (equity) compare to the inputs (costs). Oliver and Swan (1989) investigate the way consumers interpret equity and argue that equity is related to positive inequity. Using the egocentric hypothesis, Adams (1965) claims that person A feels less distress than person B when inequity is in person A’s favor, indicating positive inequity. Weick and Nesset (1968) were the first to support this proposition by stating that consumers prefer

(17)

10 overcompensation to under-compensation. Afterwards, Messick and Sentis (1983) describe this situation as advantageous inequity. Accordingly, consumers receive more equity when they gain more outcomes than the service or product provider. More recently, Gelbrich, Gäthke and Grégoire (2015) show that customers were more satisfied with overcompensation (in this case compensation levels between 100 and 168%) than with simple compensation (between 0 and 100%). Noone (2012) shows similar results: consumers who were offered overcompensation had higher PoDJ than consumers who were offered full compensation.

Because consumers are outcome-advantaged (compared to the service provider) when offered overcompensation, this study assumes that MO and NMO will lead to higher levels of PoDJ than full compensation.

1.5. Overcompensation Type and Perceptions of Distributive Justice (PoDJ)

Previous research is not clear about whether NMO can yield the same perceived fairness as MO (Noone, 2012). There are two main constructs when comparing MO and NMO: the immediacy effect and the certainty effect. According to Keren and Roelofsma (1995, p. 287), “The immediacy effect refers to the tendency of decision makers to amplify the significance of immediately (relative to delayed) experienced outcomes”. The immediacy effect is also called present bias (Benhabib, Bisin and Schotter, 2010). MO is an immediate outcome; it can be directly received by the complainant in cash or a bank transfer. In contrast, NMO may be discounted due to the delayed character of the compensation. Most NMO’s cannot be directly received, as the compensation is received later, for instance, in a complimentary service. Keh and Lee (2006) confirm this by stating that immediate gratification more effectively decreases customers’ anger in case of denied service than delayed gratification. Other studies also show that immediate outcomes may buffer dissatisfaction and even enhance customer loyalty (Smith et al., 1999; Maxham and Netemeyer, 2002).

(18)

11 The concept of immediacy is related to the certainty effect. The certainty effect includes the observation that “people overweigh outcomes that are considered certain, relative to outcomes which are merely probable” (Kahneman and Tversky, 1979, p. 265). In line with this, Mather, Mazar, Gorlick, Lighthall, Burgeno, Schoeke and Ariely (2012) argue that certain options are more favorable than uncertain options. Due to the nature and timing of receiving compensation, MO is perceived as a certain outcome. With NMO, on the other hand, the complainant is not guaranteed that the opportunity will occur to collect the promised compensation. Negative feelings towards the service provider due to a service failure may even decrease the probability that the complainant will repatronize the firm, thereby entirely devaluing the offered non-monetary compensation (Noone, 2012).

Additionally, there are other reasons why complainants are expected to be more satisfied with MO rather than NMO. Some complainants may feel that a complimentary service does not cost the firm as much as offering the price of the service in cash. To illustrate, a hotel night stay worth €100 might cost a hotel €50 (including costs such as clean bedsheets and cleaning staff) (Gilly and Hansen, 1985). Besides, Noone (2012) argues that some complainants may think that the firm also benefits from NMO. The service provider might receive an additional benefit if the customers use the complimentary service (e.g., a free hotel stay), but also spend an additional amount on another service (e.g., a breakfast). Complainants may also benefit more from MO than NMO, since MO is more flexible (in terms of its use) (Noone, 2012). In line with the theories and reasons that explain the advantages of MO, Mattila (2001) and Fu et al. (2015) show that MO creates higher PoDJ than NMO.

In contrast with the theories and reasons that explain the advantages of MO, Hoffman, Kelley and Rotalsky (1995) state that NMO yields a higher evaluation of customers’ service recovery than MO. Customers who receive a discount for food items they purchase at the moment of service (which corresponds to receiving money in return for the service failure) on

(19)

12 average rate the service recovery with a 7.75 on a scale of 1 to 10. Customers who receive complimentary meals, drinks or desserts on average rate the service recovery with an 8.05 on a scale of 1 to 10. However, their study does not directly compare the same level of monetary and non-monetary compensation. Therefore, it is not clear whether the discounts (monetary) on average express a similar, higher or lower monetary value than the complimentary products (non-monetary).

Noone (2012) compares two types of MO. She compared cash-based overcompensation with credit-based overcompensation and their effect on consumers’ PoDJ. Her results confirm the immediacy and certainty effects. Cash-based overcompensation generated higher PoDJ than credit-based overcompensation. According to Noone (2012), the most obvious reasons for this difference is the more immediate and certain character of cash-based overcompensation compared to the delayed and uncertain character of credit-based overcompensation. However, as in Hoffman et al. (1995), her study does not directly compare the same levels of MO and NMO.

The egocentric hypothesis states that overcompensation will yield higher PoDJ than full compensation (Adams, 1965). Additionally, the immediacy and certainty effects, as well as other characteristics of MO, suggest that MO will yield higher PoDJ than NMO.

H1: MO will yield significantly higher PoDJ than NMO.

1.6. Level of Overcompensation and Perceptions of Distributive Justice (PoDJ)

Prospect theory claims that people are more likely to focus on differences relative to a reference point rather than focus on absolute amounts (Kahneman and Tversky, 1979). Additionally, prospect theory asserts that people are more susceptible to losses than to gains. A service failure can be seen as a deviation from a reference point. In most cases, consumers do not expect a service failure and, therefore, the reference point is no service failure. The service failure is perceived as a loss, whereas a service recovery is perceived as a gain (Smith et al.,

(20)

13 1999). Based on mental accounting principles, consumers perceive service failure and recovery confrontations as mixed losses; a loss with a smaller gain (Thaler, 1985). Based on prospect theory, an S-shaped value curve (see figure 1) shows the value derived from the difference between a reference point and the gains or losses (Kahneman and Tversky, 1979). The S-shape value function can be explained by the decreasing impact as you move further away from the reference point. To illustrate, although the absolute difference (€10) is the same, the difference between €20 and €30 is perceived as larger than the difference between €200 and €210.

Figure 1. S-shaped value curve (Kahneman and Tversky, 1979)

The studies of Mack et al. (2000) and Noone (2012) confirm prospect theory and the mental accounting principles and their explanation for the decreasing marginal effect on customer evaluations of service recovery. In their study of restaurant businesses, Mack et al. (2000) conclude that overcompensation in service recovery attempts is not always needed after a service failure. They suggested that firms should look out for overkill strategies, as such strategies might jeopardize the credibility of the firm perceived by its customers. Noone (2012), who explored the differential effect of different kinds of monetary overcompensation (credit- and cash-based) on perceived fairness, reported similar results. She did not find a significant increase of perceptions of fairness with increases in the amount of both cash- and credit-based overcompensation. Similarly, Gelbrich et al. (2015) investigate the non-linear effect of

(21)

14 overcompensation amount on customer satisfaction and conclude that the increase in customer satisfaction marginally decreases when the compensation amount increases.

Estelami and De Maeyer (2002) also explain why higher levels of overcompensation will not lead to increased levels of PoDJ. Their study focuses on consumer responses to service provider overgenerosity, which they describe as “the act of giving customers value beyond their expectations” (p. 205). The results of their study show that low and moderate levels of generosity are mostly accepted by consumers, while overgenerosity triggers cognitive processes that negatively influence customer satisfaction. Overgenerosity might be perceived as going beyond the norms of profitable operations of firms, thus causing consumers to question why the firm is being overgenerous. In other words, the overgenerous actions might be perceived as unusual, unfair and suspicious by consumers. Participants in the qualitative part of the study of Estelami and De Maeyer (2002) stated things such as that they would feel ashamed by receiving the overgenerosity, that they would receive something that is unfair, or that they would question the motive of the seller.

Although MO and NMO are expected to yield higher levels of PoDJ than full compensation, based on prospect theory, mental accounting principles and the concept of perceived service provider overgenerosity, it is expected that the marginal impact of both MO and NMO decrease as the levels of MO and NMO increase.

H2a: PoDJ will not significantly increase with increases in the amounts of MO. H2b: PoDJ will not significantly increase with increases in the amounts of NMO

1.7. The Mediating Effect of Perceptions of Distributive Justice (PoDJ) on the Relationship between Overcompensation Type and Negative Word-Of-Mouth (NWOM) Intent

Word-Of-Mouth (WOM) is “oral, person-to-person communication between a perceived non-commercial communicator and a receiver concerning a brand, product, or a service offered for sale” (Lau and Ng, 2001, p. 163). It could entail product-related

(22)

15 conversations, informal communication, personal recommendations and interpersonal communication (Chevalier and Mayzlin, 2006). WOM is a consumer-dominated information channel and, therefore, the communicator is independent of the marketer. As a consequence, the information provided by the WOM communicator is perceived as trustworthy, reliable and credible (Chevalier and Mayzlin, 2006).

Several studies proved the effect of compensation on WOM behavior. Gilly and Hansen (1985) demonstrate that overbenefiting (overcompensation) has a positive effect on WOM. In their experiment, lower compensation leads to more NWOM than overbenefiting. Noone and Lee (2011) show that NWOM intent is triggered among affected customers by a service failure. Other studies scrutinize the buffering effect of service recovery strategies on NWOM intent after a service failure. They show that NWOM can be reduced with a recovery attempt, including compensation (e.g., Blodgett et al., 1997; Ro and Olson, 2014; Wirtz and Mattila, 2004).

Other studies examine the underlying mechanisms that influence customers’ post-service recovery satisfaction and behavior. Blodgett (1994) claims that perceived justice, including PoDJ, is the main factor that decides whether customers will perform NWOM behavior after a service failure. This claim is supported by Blodgett et al. (1997), who show that customers with lower PoDJ are more likely to engage in NWOM behavior and not return to the service provider. Goodwin and Ross (1989) prove that PoDJ has a positive effect on consumers’ willingness to return to an offending service provider. Likewise, De Ruyter and Wetzels (2000) claim that perceptions of procedural justice and PoDJ significantly enhance loyalty scores after service recovery. More recently, Noone (2012) shows that the relationship between NWOM intent and overcompensation is fully mediated by PoDJ.

Blodgett (1994) argues that distributive justice is not the only mechanism that influences the relationship between compensation and NWOM intent. He shows that when firms offer

(23)

16 their customers compensation, product importance is the second most important factor (after PoDJ) influencing NWOM intent. Product importance is a concept that identifies that consumers attach more value to some products or services than to others (Laurent and Kapferer, 1985; Prebensen, Woo, Chen and Uysal, 2013). The concept of product importance can be applied to the concept of overcompensation. MO and NMO might lead to different levels of product importance as some people might attach more value to money and discounts, whereas others might attach more value to a product or service. Consequently, consumers’ level of product importance influences their NWOM intent.

Similar research (Mano and Oliver, 1993; Russell and Pratt, 1980) shows that the degree of excitement that consumers experience while consuming a leisure service may be a main determinant of consumers’ satisfaction and behavioral intentions. Thus, since compensation is offered during consumption of a service (Sparks and McColl-Kennedy, 2001), the extent to which consumers are excited about the offered compensation influences their satisfaction with the service and their consequent behavioral intentions.

It is posited that people have higher PoDJ when they receive MO rather than NMO. PoDJ is a crucial factor that influences whether people engage in NWOM behavior. However, previous studies show the presence of other mechanisms that influence the relationship between overcompensation and NWOM intent. It is therefore expected that PoDJ mediates the relationship between overcompensation type and NWOM intent only partially instead of fully.

H3: PoDJ partially mediates the relationship of overcompensation type and NWOM

(24)

17

1.8.Conceptual Model

Figure 2. Conceptual Model H1 Overcompensation Type H 1 H 2 H2 H3 H3 H 3 H 3 NWOM intent PoDJ Overcompensation Level

(25)

18

2. Methodology

The hypotheses and research model were established and presented in the previous chapter. This chapter presents and explains the methodology of the research. Sections 2.1, 2.2 and 2.3 describe the research design, sample and procedure. Sections 2.4 to 2.8 focus on the development of the stimuli, pre-tests, manipulation, measures and statistical procedure.

2.1.Research Design

This study employed an experimental approach. The aim of an experiment is to study causal links; a change in one or more independent variable(s) leads to a change in one or more dependent variable(s) (Hakim, 2000). The independent variables for this study are overcompensation type and overcompensation level, which are both manipulated in the experimental conditions. The dependent variable is NWOM intent and the mediating variable (of the overcompensation type-NWOM intent relationship) is PoDJ. The data were collected with an online survey.

Furthermore, this research used a 2 (overcompensation type: MO versus NMO) x 3 (overcompensation level: 125, 150 and 200%) between-subjects design, because participants can only be part of the treatment group or the control group (Saunders and Lewis, 2012). Full compensation (100%) served as the control group. In total, the experiment had seven experimental conditions. Participants were randomly assigned to one of these seven conditions.

2.2.Sample

The relevant population for this study was airline customers. There was no sample frame available and, therefore, the probability of each subject being selected from the total population was unknown. Hence, this research used a non-probability sampling technique. More specifically, this research employed a convenience sampling technique; subjects were selected due to their convenient accessibility and closeness to the researcher. Besides that, the

(26)

19 convenience sampling technique was quick and inexpensive (Saunders and Lewis, 2012). The survey was written in Dutch and was only distributed among Dutch people. This national homogenous sample was used to avoid any cultural differences in relation to perceptions of service failures and NWOM behavior.

To guarantee that the respondents are familiar with airlines, they had to have used an airline before. According to information on the Dutch website Leeftijdgrens.nl (2016), the minimum age required to travel by plane is fifteen years. Consequently, people below the age of fifteen travel under supervision of airline staff, their parents or other adults. These people have probably not paid for their own flight and, therefore, will differ in terms of perceived loss (that was caused by denied boarding) compared to people who did pay for their own flight. To control for this, target respondents had to be at least fifteen.

According to Saunders and Lewis (2012), at least 30 subjects are required per condition to get a normal distribution. Since there are seven experimental conditions, the minimum sample size for this study was 210. This rule of thumb is similar to the average amount of subjects that are used for similar studies, which employ on average 29 subjects per experimental condition (Boshoff, 1997; Gilly and Hansen, 1985; Hocutt, Bowers and Donovan, 2006; Noone, 2012). This study ended up with 254 useful respondents, which is sufficient for the rule of thumb of Saunders and Lewis (2012).

2.3.Procedure

This research used a survey method in which respondents assessed a scenario containing a severe service failure with varying overcompensation types and levels. Scenarios were selected for several reasons. First, they eliminate the difficulties related to observation of service failures and the consequent service recoveries in the field, such as the time and expense involved, ethical issues, and managerial undesirability of intentionally forcing service failures. Second, scenarios decrease biases from memory gaps and the tendency of rationalization (Smith

(27)

20 et al., 1999). Third, scenarios decrease problems that involve individual variances in personal circumstances and responses to the research context. Consequently, using scenarios controls for extraneous and manipulated variables and decreases random noise in the dependent variable, thus improving the statistical and internal conclusion validity (Wirtz and Bateson, 1999).

Participants were obtained by sending an invitation via Facebook and e-mail. The invitations contained a link that forwarded the respondents to an online survey. The invitations for the online survey were sent to 363 people. Of these, 269 filled out the survey after the first invitation. The remaining 94 people received a reminder and of these, 26 people filled out the survey. Hence, the response rate was 81.27%. Participation in the study was voluntarily and the researcher did not provide the respondents with an incentive.

The survey contained four sections. First, the introduction of the survey informed the participants that the research’s purpose is to better understand customer satisfaction with services. The introduction was intentionally vague to minimize interviewer bias, which decreases the validity of studies (Saunders and Lewis, 2012). Besides that, the introduction explained the context of the study and the importance of honesty of responses. Lastly, it showed a confidentiality statement, which informed participants about the confidentiality of their answers. In the second part of the survey, participants had to answer two screening questions that checked whether the participants were eligible for the survey. The first question asked whether the respondent indeed used an airline before and the second question asked about their age. If the answer to the first question was no or the answer to the second question was below fifteen years, the survey was ended.

In the third part of the survey, participants were asked to imagine that they had booked a ticket at an airline and that they are denied boarding due to overbooking. Earlier research shows that two service recovery attributes influence the effect of compensation on customers’ post recovery responses: apology and explanation (Hocutt et al., 2006; Wirtz and Mattilla,

(28)

21 2004). It is thus important to ensure that respondents know that these attributes are present. Therefore, the scenario description mentioned the presence of an apology and an explanation. The scenario also controlled for the perception of waiting time at the airport. By mentioning the subsequent waiting time (seven hours), the respondents experienced a similar service failure. To illustrate, people who wait one hour at the airport for their new flight experience a less severe service failure than people who wait ten hours. Furthermore, participants received varying information regarding the type and level of (over)compensation corresponding to their experimental condition. In the fourth part of the survey, respondents were asked questions about their PoDJ, their NWOM intent and the control variables. The survey is found in Appendix 1.

2.4.Development of the Stimuli

Context of the Study

The context of this study is denied service at an airline due to overbooking. Smith et al. (1999) shows that consumers perceive denied service due to overbooking as a severe service failure. Service firms such as hotels, spas and airlines use overbooking strategies to maximize their revenues. British Airways admitted that they overbook one seat per 100 passengers (Massey, 2010), which proves that denied service due to overbooking in the airline industry happens.

Compensation Levels

The different (over)compensation levels were represented a percentage of the purchase price (€200) of the original flight ticket: 125, 150 and 200%. MO levels were expressed in monetary terms (€) and NMO levels were expressed as a complimentary service. The NMO levels were based on the same percentages as the MO levels. The motivation for this was to compare the two different types of overcompensation within a similar range of overcompensation amount. For instance, overcompensation at 200% equals a complimentary flight plus (1) €200 cash (MO), or (2) an upgrade to business class worth €200 (NMO). The

(29)

22 condition in which full compensation (100%) was offered to the respondents, served as the control group. An overview of the conditions can be found in Table 1.

Table 1. Overview of the seven experimental conditions

Level (%) MO NMO

100 Complimentary flight (€200)

125 Complimentary flight (€200) _{+ €50 cash} Complimentary flight (€200) _{+ upgrade worth €50}

150 Complimentary flight (€200) _{+ €100 cash} Complimentary flight (€200) _{+ upgrade worth €100}

200 Complimentary flight (€200)

+ €200 cash

Complimentary flight(€200) + upgrade worth €200

The overcompensation amounts that were initially set for this study (150, 200 and 300%) were similar to those tested in prior studies. In the study of Megehee (1994), compensation ranges from 50 to 300 percent. Garrett (1999) and Noone (2012) set levels of overcompensation ranging from 100 to 300 percent. The value of compensation in the study of Noone and Lee (2011) ranges from 50 to 200 percent of the original purchase price. In addition, these relative overcompensation amounts also reflect industry practices. For instance, Emirates’ policy for their passengers who are denied boarding is an alternative, next available flight to their destination, plus a voucher for a complimentary ticket on a similar Emirates route. Their compensation entails approximately 200% of the original ticket price (Emirates, 2016). However, the pretest results showed that respondents of the pre-test survey did not perceive the compensation as realistic. They rated the realism of the offered compensation with an average of 2.6 on a scale from 1 to 5. Therefore, the levels of overcompensation were adjusted to 125, 150 and 200% and another pre-test was executed in which the realism of the offered compensation increased to an acceptable average score of 3.9 on a scale from 1 to 5. A more extensive report of the pre-test results is explained in Section 2.5

(30)

23

Monetary Overcompensation (MO) and Non-Monetary Overcompensation (NMO)

MO represented an immediate cash payment. NMO entailed an upgrade to business class. Both forms of compensation are used by airlines (Emirates, 2016).

2.5.Pre-test

A pre-test was executed prior to the experiment to test whether the developed stimuli were interpreted correctly and whether the survey was clear to respondents. The online pre-test survey (see Appendix 2) was filled out by 32 participants, in line with the rule of thumb of 30 of Saunders and Lewis (2012). The pre-test evaluated several aspects of the actual research survey. First, to test whether participants perceived the service failure that was described in the scenario as severe, they were asked to rate the situation severity on a Likert scale ranging from 1 (not at all severe) to 5 (very severe). Second, to test whether the scenario and the compensation were perceived as realistic, participants were asked to rate the realism of the scenario and the compensation on a Likert scale ranging from 1 (not at all realistic) to 5 (very realistic). Third, to test whether the scenario and the compensation were perceived as easy to understand, participants were asked to rate the comprehensibility of the scenario and survey questions on a Likert scale ranging from 1 (not at all comprehensible) to 5 (very comprehensible). Fourth, to test whether the duration of the survey was perceived as acceptable, participants were asked to rate the duration on a scale from 1 (too short) to 5 (too long). Lastly, the pre-test contained space for additional comments.

Of the participants of the pre-test, 62% were female and 73% were under the age of 40 years (M = 36.52, SD = 9.28). Almost all variables in the pre-test had an acceptable score. On average, participants rated the service failure severity at 3.83 (close to 4: ‘moderately severe’), the realism of the scenario at 4.03 (close to 4: ‘moderately realistic’), the comprehensibility of the scenario at 4.00 (4: ‘moderately comprehensible’) and the comprehensibility of the survey questions at 4.17 (close to 4: ‘moderately comprehensible’). Moreover, the duration of the

(31)

24 survey was acceptable with a score of 2.83 (close to 3: ‘good’). However, participants rated the realism of the compensation with an average score of 2.6 (in between 2: ‘somewhat unrealistic’ and 3: ‘neither unrealistic nor realistic’). These scores were confirmed by the additional comment section. Twelve out of 32 participants mentioned that they found the compensation extraordinarily high. Consequently, the levels of overcompensation were adjusted to 125, 150 and 200%. A follow-up pre-test was executed with the same participants. The realism of the offered compensation increased to an acceptable average score of 3.9 (close to 4: ‘moderately realistic’). Lastly, the additional comment section provided the researcher with other feedback concerning the spelling and grammar of the survey text. The survey was adjusted accordingly.

2.6.Manipulation Check

To check whether the manipulations of the overcompensation conditions (MO vs. NMO) were effective, participants answered two questions regarding their expectations of the quality of the alternative flight. They were asked to rate their level of agreement (1: strongly disagree; 7: strongly agree) with the statements “I feel that I will receive the same level of service at the alternative flight as I would have received at this flight” and “I feel that the quality of my alternative flight will be superior to the quality of this flight”. People who receive an upgrade (NMO) should expect a higher quality of the alternative flight compared to their original flight.

2.7.Measures

The measures in the survey were obtained from English studies. Since the surveys had to be written in Dutch due to the Dutch speaking audience, the originally English measurements and scenarios were translated into Dutch. To guarantee that the content remained the same, an unrelated person translated the Dutch items back into English items. Differences between the original English items and the translated English items were used to revise the Dutch items.

(32)

25 This process is called back-translation (Wild, Grove, Martin, Eremenco, McElroy, Verjee-Lorenz and Erikson, 2005). An overview of the measures is found in Appendix 3.

Perceptions of Distributive Justice (PoDJ)

The mediating variable, PoDJ, was measured using the four-item (e.g., “The outcome I received was fair”), seven-point scale of Smith et al. (1999). This scale is anchored by strongly agree and strongly disagree and has a Cronbach’s alpha of 0.8.

Negative Word-Of-Mouth (NWOM) intent

The dependent variable, NWOM intent, was measured with Blodgett et al. (1997) three-item scale (e.g., “Given what happened, how likely are you to warn your friends and relatives not to book at this airline?”). This scale is anchored by 1 = not at all likely and 7 = very likely and has a Cronbach’s alpha of 0.79

Control Variables

Several control variables influence the way customers respond to complaint handling events (Hess, Ganesan and Klein, 2003; Frankel, 1981; Kolodinsky, 1993; Noone; 2012; Noone and Lee, 2011; Seo, 2005; Smith et al., 1999). The first control variable is gender. Prior research shows that men and women differ in terms of evaluating service failures (Kolodinsky, 1993; Smith et al., 1999; Hess et al., 2003). Iacobucci and Ostrom (1993) show that women are generally more interested in the relational aspects of a service encounter and men in core aspects. This influences their satisfaction with a service encounter (in this case a service recovery). The second control variable is age. Noone (2012) shows that age affects perceptions of service failure severity. Mather et al. (2012) suggest that the certainty effect (Kahneman and Tversky, 1979) is stronger for older people than for younger people. Since the two types of overcompensation differ in the extent to which they are certain, age is expected to have an influence on consumers’ satisfaction with an overcompensation type. Level of education is the third control variable. Hess et al. (2003) show that customers who completed a higher education

(33)

26 judge the service failure using a broader knowledge base, leading to higher perception of the service failure severity. These control variables were measured using single item scales.

The fourth control variable is perceived inconvenience of overbooking and this was measured using a two-item scale altered from Frankel (1981) and Seo (2005). The two items are “Having to change flight would cause me great inconvenience” and “It would not bother me to have to change flight”, r = 0.65. The scale ranges from 1 (strongly disagree) to 7 (strongly agree). Previous research also uses this control variable and shows that perceived inconvenience of overbooking in hotels significantly impacts satisfaction with the service recovery (Noone and Lee, 2011), PoDJ and NWOM intent (Noone, 2012). This control variables refers to whether people are really bothered by waiting for another flight. It is expected that people who don’t mind to wait for seven hours might have a different attitude towards a firm as opposed to people who are very impatient.

The last control variable is perceived service failure severity. As mentioned before, service failure severity influences customers’ perceived loss (Weun et al., 2004), attitudes (McCollough et al., 2000) and responses (Weun et al., 2004). Perceived service failure severity is, therefore, expected to influence the participants’ PoDJ as well as their NWOM intent. To test for the manipulation of a severe service failure, participants had to answer the question “To what extent do you perceive the service failure that is caused by the airline (overbooking) as severe?” This measure ranged from 1: not at all severe to 7: very severe. This control variable measures the extent to which people find the overbooking as a big mistake. To illustrate, some people might understand that an airline overbooks flights in order to overcome empty seats and these people will have lower perceptions of service failure severity than people who don’t understand why an airline would overbook a flight.

(34)

27

2.8.Statistical Procedure

Qualtrics was used to collect online survey data. The data were analyzed using the Statistical software Package for Social Sciences (SPSS). When testing, the desired level of confidence interval was 95%. Before testing the hypotheses, the sample characteristics were explored and five preliminary steps were executed. First, counter indicative items were recoded, frequencies were checked for each item, missing values were found and incomplete surveys were deleted accordingly. Second, the preliminary analysis checked for biases in the random allocation of participants into the different experimental conditions. No bias was discovered. Third, the preliminary analysis assessed the reliability of each of the scales used in the survey. All scales had an acceptable reliability. Fourth, a one-way ANOVA test was executed to assess whether the manipulations were successful. All manipulations were successful. Fifth, the data were tested for normality, skewness and kurtosis. The data showed some non-normality, skewness and kurtosis, but this was not considered as a problem due to the presence of a large sample and equal sample sizes per experimental condition. After the execution of the preliminary analyses, descriptive statistics and correlations were explored.

To test the hypotheses, hierarchical regression analyses were undertaken to test direct relationships and to test the hypothesized mediation effect. AVONA tests were performed to interpret the beta values of the categorical predictor variable. Furthermore, an ANCOVA test was executed to test for the effect of overcompensation type on PoDJ among different levels of overcompensation. A more thorough discussion of these steps is found in the results chapter.

(35)

28

3. Results

This chapter represents the results of the online survey. Section 3.1 presents the sample characteristics. Sections 3.2 to 3.5 describe the preliminary steps that were explained in the statistical procedure in the previous chapter. Sections 3.6 and 3.7 describe the descriptive statistics and correlations. Section 3.8 presents the results of the tested hypotheses presented in Chapter 2 and Section 3.9 presents an overview of the verdicts on the hypotheses.

3.1.Sample Characteristics

The sample consisted of 254 participants, of which 61% were female and 39% were male. The mean age of the participants was 34 years (Mage = 33.94, SD = 15.22). The mean age of females was 32 years (Mage = 31.88, SD = 14,57) and the mean age of males was 37 years (Mage = 37.15, SD = 15,74). The majority of the participants reported completing a university of applied science degree (44.8%) or a university degree (40.6%), followed by participants who completed an intermediate vocational education degree (9.1%). Finally, only 5.5% completed a secondary education program, including HAVO (1.9%) and VWO (3.6%).

3.2.Data Cleaning, Missing Values and Recoding

Frequencies were checked to examine any errors. Of the 295 cases, 41 were removed from the data set because of missing values or because they did not belong to the target group of the study. The latter criterion was checked by two screening questions that were present at the beginning of the survey. Next, the recoding of counter-indicative items applied to two items of PoDJ: PoDJ2 became rPoDJ2 and PoDJ4 became rPoDJ4. Additionally, overcompensation type was coded as 1 (full compensation), 2 (MO) and 3 (NMO). Overcompensation level (varying from 100 to 200%) was coded as 100, 125, 150 and 200.

(36)

29

3.3.Control Variables

Several preliminary analyses were executed to check for biases in the random allocation of participants into the different experimental conditions (see Appendix 4). First, a one-way ANOVA was performed to check whether significant differences in age exist among the seven experimental conditions (Field, 2009). Age did not significantly differ among the seven conditions, F (6, 247) = 1.305, p > .05. Second, four Chi-square tests of independence were conducted to determine whether there were significant differences among the conditions in (1) the number of males and females, (2) the level of education, (3) perceived service failure severity and (4) perceived inconvenience of overbooking (Field, 2009). Gender did not significantly differ among the seven conditions, χ² (6) = 5.158, p = .524. Education level did not significantly differ among the experimental conditions, χ² (18) = 25.281, p = .117. Service failure severity was not significantly different among the participants in the different experimental conditions, χ² (18) = 15.341, p = .638. Perceived inconvenience of overbooking did not significantly differ among the participants in the different experimental conditions, χ² (66) = 72,484, p = .273. These results show that no biases exist in the allocation of participants into the different experimental conditions. Hence, the random allocation was successful.

3.4.Reliability

Reliability is the degree to which analysis procedures or data collection techniques lead to consistent findings (Field, 2009). To examine the reliability of the scales, several scales were checked for their Cronbach’s alpha. Scales with a Cronbach’s alpha score above .70 were accepted. Next, the corrected item-total correlations should be above .30 to indicate that the individual items within the scale have a good correlation with the total score of the scale. Lastly, the Cronbach’s alpha if-item-deleted should not be .10 higher than the general Cronbach’s alpha. These alphas (see Appendix 5) were in line with the rules of thumb of Saunders and Lewis (2012). The Cronbach’s alpha scores are presented in Table 2.

(37)

30

Perceived Inconvenience of Overbooking

Perceived Inconvenience of Overbooking has a high Cronbach’s alpha of .844 (>.70). The corrected item-total correlations are both .733 (>.30). Both items did not have an alpha-if-item-deleted that was higher than the general Cronbach’s alpha (Δ <.10).

Perceptions of Distributive Justice (PoDJ)

The Cronbach’s alpha score of PoDJ is .792 (>.70). The corrected item-total correlations are .589, .605, .577 and .639 (all >.30). All four items did not have an alpha-if-item-deleted that was higher than the general Cronbach’s alpha (Δ<.10).

Negative Word-Of-Mouth (NWOM) Intent

NWOM intent has a high Cronbach’s alpha of .850 (>.70). The corrected item-total correlations are .767, .644 and .749 (all >.30). All four items did not have an alpha-if-item-deleted that was higher than the general Cronbach’s alpha (Δ<.10).

3.5.Manipulation Check

A manipulation test was executed to check whether participants perceived the two types of overcompensation as different. This manipulation test contained two questions: (1) I feel that I will receive the same level of service at the alternative flight as I would have received at this flight and (2) I feel that the quality of my alternative flight will be superior to the quality of this flight. Participants answered these questions on a scale from 1 (strongly disagree) to 7 (strongly agree). A one-way ANOVA was conducted (see Appendix 6) for each question to test whether significant differences exist among the experimental conditions (Field, 2009). The results of the first question revealed that participants in the MO condition (M = 5.19, SD = 1.645) expect that the level of service at the alternative flight will be similar to the original flight. This is significantly more than what participants in the NMO condition expected (M = 4.13, SD = 1.840), F (2, 251) = 9.918, p < .001. The results of the second question showed that participants in the NMO condition (M = 4.36, SD = 1.752) feel that the quality of the alternative flight is

(38)

31 superior to the quality of their original flight. Their scores for this question are significantly higher than for participants in the MO condition (M = 3.18, SD = 1.441), F (2, 251) = 14.375, p < .001. These results show that the manipulation of the overcompensation type was fruitful.

3.6.Normality, Kurtosis and Skewness

The data were checked for normality, skewness and kurtosis (see Appendix 7). All variables were significantly non-normal as all p values of the normality tests were below .001 (Field, 2009). Gender, Education, Perceived Service Severity, Perceived Inconvenience of Overbooking, PoDJ and Expected Quality 1 were negatively skewed. This means that the scores of these variables cluster at the high end. The other variables are positively skewed, which means that the scores cluster at the low end (Field, 2009). Gender, Age, Overcompensation Level, Expected Quality 1, Expected Quality 2 and NWOM intent show negative kurtosis. This indicates that the distributions of these variables are light-tailed. The other variables show positive kurtosis values and are, therefore, more heavy-tailed (Field, 2009).

Tabachnick and Fidell (2001) claim that non-normality and skewness will not substantially influence the analysis of data if the data is based on a reasonably large sample. Field (2009) states that non-normality can have a weak effect on the Type I error (incorrect rejection of null hypothesis), but only if the sample sizes are not equal. Since the sample sizes are equal, non-normality will not have an effect. In addition, Tabachnick and Fidell (2001) state that “Kurtosis can result in an underestimate of the variance, but this risk is also reduced with a large sample, +200” (p. 75). Considering the reasonably large sample in this study (n=254) and equal sample sizes, the risks of normality, skewness and kurtosis values were reduced.

3.7.Descriptive Statistics and Correlation Analysis

Table 2 presents an overview of the descriptive statistics and correlations. Correlation statistics were interpreted according to the theory of Field (2009). A first observation that can be derived from the table is that Perceived Inconvenience of Overbooking is strongly positively

(39)

32 correlated with Perceived Service Severity, r(254) = .507, p < .001. This indicates that the higher participants’ perceptions of the inconvenience of overbooking, the higher their perceptions of the service failure severity. Second, Perceived Service Severity is positively correlated with NWOM intent, r(254) = .414, p < .001. This means that if participants perceive the service failure as more severe, their intent to engage in NWOM behavior increases. Third, Perceived Service Severity is positively correlated with both Age, r (254) = 163, p < .01, and Education, r(254) = .165, p < .01. These correlations indicate that the older a person is, or the higher the level of education a person has, the higher that person’s perception is of the service severity. However, these correlations are weak.

Education was also weakly negatively correlated with Expected Quality of the Alternative Flight (1), r (254) = -.147, p < .05, meaning that the higher the level of education, the less people expected to receive the same level of service at the alternative flight. Fifth, Age and Education have a weak and negative relation, r(254) = -.169, p < .01. Sixth, Perceived Inconvenience of Overbooking is positively correlated with NWOM intent, r(253) = .433, p < .001. This shows that if participants perceive the overbooking as more inconvenient, their intent to engage in NWOM behavior increases.

Seventh, the two questions about the Expected Quality of Alternative Flight are negatively correlated with each other, r(253) = -.410, p < .001. This indicates that if people perceive the alternative flight to have the same quality, their expectation of the alternative flight being superior decreases.

Furthermore, Expected Quality of the Alternative Flight (1) is negatively correlated (r (254) = -.155, p < .05) and Expected Quality of Alternative Flight (2) is positively correlated (r (254) = .224, p < .01) with Overcompensation Type. These correlations are also weak. However, these correlations mean that the manipulation of the different types of overcompensation worked. This was already discussed in the previous analysis.

(40)

33 Eighth, Overcompensation Type is positively correlated with Overcompensation Level,

r(253) = .406, p < .001. However, since Overcompensation Type is a categorical variable, we

cannot make any further indications regarding the r value of these correlations. Ninth, PoDJ has weak negative correlations with Age (r (254) = -.161, p < .05), Perceived Service Severity (r = -.148, p < .05), and Perceived Inconvenience of Overbooking (r (254) = -.186, p < .01). These correlations mean that an increase in Age, Perceived Service Severity or Perceived Inconvenience of Overcooking leads to a decrease in PoDJ. On the other hand, PoDJ is weakly positively correlated with Overcompensation Type, but since Overcompensation Type is a categorical variable, we cannot make any further indications regarding the r value of this correlation.

Tenth, NWOM intent is weakly negatively correlated to both Overcompensation Level (r (254) = -.146, p < .05) and Overcompensation Type (r (254) = -.278, p < .01). The higher the overcompensation level, the less people are intended to engage in NWOM behavior. NWOM intent has a weak, positive correlation with Gender, r (254) = .135, p < .05, indicating that women’s NWOM intent, on average, are .135 higher than men’s NWOM intent. Lastly, PoDJ is negatively correlated with NWOM intent. This means that the higher participants’ PoDJ, the less likely they are to engage in NWOM behavior.

Monetary and non-monetary overcompensation for severe service failures : impact on perceptions of fairness and negative-word of mouth intent