Investigating Word of Mouth Behavior:
The Pursue for Recommending Consumers
- A research on constructs influencing Word of Mouth-
Investigating Word of Mouth Behavior:
The Pursue for Recommending Consumers
- A research on constructs influencing Word of Mouth-
Author: Thomas Stalman 1st-supervisor: Dr. M. Gijsenberg Department: Marketing 2nd-supervisor: Dr. M.C. Non Qualification: Master thesis External supervisor: M. Reitsma Completion Date: 06-02-2012 Organization: KPN
Adress: V Heemskerckstraat 2B Groningen Phone number: 0610073910
Management Summary
Preface
During the last day of my internship I had to choose between pretending to be working on my thesis or cleaning the toilets since the cleaners were striking. Obviously I chose the first option and decided to write this mind-blowing, stunning, and world changing preface. During the last five months I have been working on my master thesis on word of mouth behavior. While this is a very interesting subject I can say that I am sincerely glad that I have completed it. During my internship I spent most of the time writing my thesis. Furthermore, I participated in several interesting research projects of my colleagues. Since this thesis is the final part of my study I consider it as the end of a wonderful era. During the four and a half years that I studied I met tons of interesting people. Besides, I increased my knowledge, my analytical skills, and I have grown as a person. I would not have been able to write this thesis without the constructive feedback of my supervisor Maarten Gijsenberg, the help and suggestions of my company supervisor Meyke Reitsma, and the ongoing support of my girlfriend Suzanne. I hope you all enjoy reading my thesis and find the subject as interesting as I do.
Table of contents MANAGEMENT SUMMARY ... 3 PREFACE ... 4 TABLE OF CONTENTS ... 5 1 INTRODUCTION ... 6 1.1BACKGROUND PROBLEM ... 6 1.2PROBLEM STATEMENT ... 6 1.3INTRODUCTION COMPANY ... 8 1.4STRUCTURE ... 8 2 THEORETICAL FRAMEWORK ... 9 2.1WOM TYPOLOGY ... 9 2.2ANTECEDENTS OF WOM ... 12 2.3WOM AS A DRIVER ... 14
2.4SOCIAL NETWORK AND WOM ... 15
2.5ONLINE WOM ... 17
3 RESEARCH FRAMEWORK ... 18
3.1MAIN EFFECTS ... 18
3.2MODERATING EFFECT OF BRAND STRATEGY ... 21
3.3MODERATING EFFECT OF AGE OF THE RELATIONSHIP ... 22
3.4CONTROL VARIABLES ... 23 3.5CONCEPTUAL FRAMEWORK ... 24 4 RESEARCH DESIGN ... 25 4.1DATA COLLECTION ... 25 4.2CONSTRUCT MEASUREMENT... 25 4.3SCALE MEASUREMENT ... 26 4.4PLAN OF ANALYSIS ... 27 5 RESULTS ... 28 5.1OVERVIEW ... 28 5.2FACTOR ANALYSIS ... 28 5.3POOLING... 31 5.4ASSUMPTIONS ... 31 5.5RESULTS ... 34 5.6MODEL VALIDATION ... 39
6 CONCLUSION AND MANAGERIAL IMPLICATIONS ... 40
6.1CONCLUSION ... 40
6.2MANAGERIAL IMPLICATIONS ... 41
7 LIMITATIONS AND FUTURE RESEARCH ... 43
7.1LIMITATIONS ... 43
7.2RESEARCH DIRECTIONS ... 43
1 Introduction
In this chapter the background problem of the paper will be introduced, followed by the problem statement and research questions. Subsequently, a short introduction of the company where this report is written for will be given. This chapter ends with the organization of the rest of the paper.
1.1 Background problem
The average American engages in 120 word of mouth (WOM) conversations per week (Keller 2007). Furthermore, companies more and more use WOM in their marketing campaigns. For example, for the introduction of the ―No-Mess‖ pen, WD-40 hired Proctor & Gamble to promote its product through Proctor & Gamble‘s Vocalpoint, a panel of influential moms who were selected based on their ability to be ―connectors‖ (Neff 2006). Another example is Hasbro, who identified and used the ―coolest‖ kids in each school, to promote its new handheld video game called POX (Godes and Ofek 2004). Due to rapid changes in the social and technological environment, people communicate more and more with each other (Chen et al. 2011). Besides, firms are gaining increasing capacity to initiate and manage consumer social interactions directly (Godes et al. 2005). This has major consequences for companies and the influence that WOM has (Chen et al. 2011). WOM can be defined as ''informal, person-to-person communication between a perceived non-commercial communicator and a receiver regarding a brand, a product, an organization or a service" (Harrison-Walker 2001). WOM is called the world‘s most effective, yet least understood marketing strategy (Misner 1999). During the last years, WOM gained much attention from practitioners. Many books about this subject have been published (Trusov et al. 2009). Companies try to adapt WOM in their overall strategy and include measurements to investigate the intention to use WOM by their customers. However, there is still a lot to discover about this very interesting phenomenon.
1.2 Problem statement
Several scholars suggest that WOM may be among the most important aspects in relationship marketing (Christopher et al. 1991; Reichheld 2003; White and Schneider 2000). Despite the increasing interest and importance, there are relatively few studies directed at understanding factors that influence WOM (Brown et al. 2005). Therefore the following problem statement is formulated: “Which factors influence the degree of Word of Mouth used by customers from
This problem statement is divided into two research questions. First, the intention to use WOM by customers from three communications service providers is investigated. Those three providers all follow a different strategy. There is a price fighter, a ‗normal‘ provider, and a premium provider. There are clear differences between the providers. They focus for example on different segments and therefore it is likely that they all have different kinds of customers. This could result in different segments based on intention of WOM usage for each provider. No similar research has been performed yet. The following research question is formulated.
“Does the effect of the variables influencing the intention to use WOM differ between customers of a price fighter, a „normal‟ provider and a premium communications service provider?”
After this research question, the effect of brand and price related variables on the degree of WOM are measured. Most of the current research focuses on the direct effects of consumer‘s satisfaction and dissatisfaction with previous purchasing experiences on WOM (Brown et al. 2005). Some researchers find a positive effect of satisfaction on WOM (Blodgett et al 1993), while others find no direct relationship between satisfaction and WOM usage (Arnett et al. 2003). Other research for example, focuses on variables like opinion leadership, structure of networks, commitment and consumer identification (Brown and Reingen 1987; Brown et al 2005; Granovetter 1973; Wangenheim and Bayon 2003; de Matos and Rossi 2008). To get a good view on factors affecting WOM, there is a clear need to investigate other variables that might influence WOM behavior. This research investigates four antecedents that might influence the intention to use WOM. These factors are brand-self connection, self-image congruence, brand familiarity and perceived price perception. Based on literature, there is an indication that these variables might influence intention to use WOM (Brown et al. 2005; Verhoef et al. 2002). Therefore, they need to be investigated and the following research question is formulated.
“What for influence do brand-related and price-related variables have on the intention to use WOM”
1.3 Introduction company 1.3.1 KPN
Koninklijke PTT Nederland (KPN) is a Dutch communications service provider which mainly operates in the Netherlands and uses a multiple brands strategy in other countries in Europe. KPN offers consumers mobile phone, fixed phone, internet, and television connections. KPN has a large and diversified portfolio of brands. In the Netherlands KPN operates with four brands. At the first place KPN, which offers fixed and mobile connections for families. The next brand is Hi, which offers mobile connections for younger people. Furthermore, they operate under the brand Telfort, which is a price fighter in the mobile and internet market, and at last XS4all, which is a premium internet provider.
1.3.2 Market Environment
The market environment is changing rapidly. While a few years ago the use of internet was seen as new, nowadays many customers have smart phones with high speed internet access and see it as a commodity. KPN has a market share of approximate 50% in the fixed and mobile services market. The market share in the broadband internet market is approximate 41%. Rapid innovations and fierce competition characterize this market. In the second quarter of 2011, 42% of the consumers in the Netherlands used a smart phone (Telecompaper 2011). Compared to 2010 this is an increase of more than one third. This rapid increase forces communications service providers to enhance their business model.
1.4 Structure
2 Theoretical framework
2.1 WOM typology2.1.1 Definition of WOM
Early research already showed that WOM can be highly influential in building the market for new or improved products (Brooks 1957). However, relatively few studies are directed at identifying factors that influence WOM (Arndt 1967; Anderson 1998; Brown et al. 2005). The basic idea behind WOM is that information about products, services, stores, companies and other relevant topics can spread from one consumer to another (Brown et al. 2005). WOM refers to the dissemination of information through communication among people (Chen et al. 2011). Harrison-Walker (2001) defines WOM as ''informal, person-to-person communication between a perceived non-commercial communicator and a receiver regarding a brand, a product, an organization or a service". The view that WOM is organic, because it occurs between one consumer and another without direct prompting, influence, or measurement by marketers is the earliest and simplest understanding of consumer WOM (Kozinets et al. 2010). This theory is visualized in the organic interconsumer influence model below.
Figure 1: The organic interconsumer influence model
This model assumes that WOM occurs naturally among consumers when new market innovations are introduced, or effective advertising and promotion campaigns are executed (Bass 1969). The two most important aspects of WOM that have been examined are volume and valence (Mahajan et al. 1984; Mizerski 1982; Neelamegham and Chintagunta 1999). Volume measures the total amount of WOM interactions, whereas valence captures the nature of WOM messages, for example whether they are positive or negative.
2.1.2 Types of WOM
about a specific product or service. On the opposite there is negative WOM, characterized by discouraging a specific product or service. In the center there is neutral WOM. This kind of WOM only consists of details of a product or service. No judgment is given here. Positive WOM is found to be more persuasive than neutral WOM (Cheema and Kaikati 2010).WOM can be firm created, so called exogenous WOM, or endogenous which is created by consumers (Godes and Mayzlin, 2009). Exogenous and endogenous WOM are discussed in the following paragraphs.
2.1.3 Companies using WOM
Not only scholars are interested in WOM, also companies try to use WOM in their communication strategy. This so-called firm-created WOM is a hybrid between traditional advertising and consumer WOM. There are two interesting trends that can be witnessed. At the first place, companies try to engineer WOM. Firms take actions to increase the number of conversations that are taking place (Godes and Mayzlin 2009). The second trend is that firms attempt to identify the most important customers based on WOM, the ―key influencers‖ (Godes and Mayzlin 2009). WOM created as the result of firm‘s actions are referred to as exogenous WOM. Research on exogenous WOM shows that in some cases purely exogenous WOM is associated with higher sales (Godes and Mayzlin 2009). An interesting aspect for companies using exogenous WOM is that for products with low or moderate levels of awareness, loyal customers are not necessarily the cornerstones of a successful WOM campaign. This can be caused by the fact that the connections of loyal customers have probably been informed about the product for some time (Godes and Mayzlin 2009). This paper does not have the aim to investigate exogenous WOM. However, to give a coherent and integrated overview it is shortly discussed. Instead, this paper focuses on so-called endogenous WOM. These are conversations that naturally occur among consumers as a function of their experiences with the product.
2.1.4 Consumers using WOM
about products. According to this fact it is of marketer‘s interest to identify and attempt to influence these market mavens (Kozinets et al. 2010). This new insight of opinion leaders is captured in the linear marketer influence model (Kozinets et al. 2010). This model is visible below in figure 2.
Figure 2: The Linear Marketer Influence Model
According to the linear marketer influence model, marketers actively try to influence consumer WOM through the use of advertising and promotions (Kozinets et al. 2010). Because of the use of traditional advertising and promotion methods, this stage is referred to as a model of linear influence. Difference with the earlier organic interconsumer influence model is that in this model opinion leaders are specifically targeted as a result of new insights gained from research (Feick and Price 1987).
2.2 Antecedents of WOM
Most research on the antecedents of WOM focuses on the link between (dis)satisfaction and WOM usage. Based on the confirmation/disconfirmation paradigm, customer satisfaction can be described as the outcome of a comparison process between perceived product performance and previously held expectations (Oliver 1993; Wangenheim and Bayon 2004). When performance exceeds expectations, positive disconfirmation occurs and leads to satisfaction (Oliver 1993; Wangenheim and Bayon 2004).
Brown et al. (2005) found that satisfaction, commitment and identification exert significant influences on positive WOM intentions and behaviors. The effects of satisfaction and identification are at least partially mediated through commitment. Commitment can be defined as involving an exchange partner, who believes that an ongoing relationship with another is important enough, to warrant maximum efforts at maintaining it (Morgan and Hunt 1994). Commitment of consumers interacts with satisfaction in such a way that, the influence of satisfaction on positive WOM, becomes less positive at higher levels of commitment to the marketing organization (Brown et al. 2005). This suggests that consumer commitment mediated the effect of satisfaction on WOM. It is interesting to find that for higher levels of commitment, satisfaction may be inversely related to WOM behavior (Brown et al. 2005). Verhoef et al. (2002) also investigated the effect of commitment on WOM usage. They distinguished two components of commitment: affective and calculative commitment. Affective commitment is based on feelings of identification and loyalty (Gundlach et al. 1995), whereas calculative commitment is based on associated switching costs. (Geyskens et al. 1996). Research shows that affective commitment has a significant positive effect on WOM, while for calculative commitment no effect was found (Verhoef et al. 2002).
There exists a broad consensus that satisfaction has a direct positive influence on WOM. Several studies have shown positive WOM to be an outcome of high customer satisfaction ratings (Blodgett et al. 1993; Mittal et al. 1999; Richins 1983; Sundaram et al. 1998; Swan and Oliver 1989; Westbrook 1987). On the other hand, there are a few studies that find no direct relationship between the two constructs (Arnett et al. 2003; Bettencourt 1997; Reynolds and Beatty 1999). Overall, a U-shaped relationship whereby the extremes of satisfaction have the greatest influence on WOM is suggested to be true (Anderson 1998).
behavior (Dick and Basu 1994). This could be explained by the loyalty to WOM link. Bowman and Narayandas (2001) found that customers, who described themselves as loyal, were significantly more likely to engage in WOM. However, these customers were less likely to engage in WOM the higher their satisfaction. This suggests that loyal customers only engage in negative WOM and only when they are dissatisfied (Bowman and Narayandas 2001). This hypothesis is supported by De Matos and Rossi (2008).
Next to satisfaction and loyalty, other antecedents also have been examined. In a study of Verhoef et al. (2002) several variables and their effect on customer referral and number of services purchased have been examined. Based on the questions asked, customer referrals are assumed to be the same as WOM. Verhoef et al. (2002) also found proof for the relationship between satisfaction and WOM. Additionally, they tested the effects of trust and payment equity on WOM. Trust can be defined as the perception of confidence in an exchange partner‘s reliability and integrity (Morgan and Hunt 1994). Research shows that trust is positively related to WOM (Gwinner et al. 1999; Verhoef et al. 2002). Next to trust, the effect of payment equity on WOM has also been examined. Payment equity is the customers‘ perceived fairness of the price paid for their consumed services (Bolton and Lemon 1999). Payment equity is positively related to WOM (Verhoef et al. 2002; De Matos and Rossi 2008). Empirical research shows that quality has a positive influence on WOM (Bloemer et al. 1999; Boulding et al. 1993; Harrison-Walker 2001; Zeithaml et al. 1996). The customer‘s intentions in terms of recommendations are favorable when quality is high (Parasuraman et al. 1988; Zeithaml et al. 1996). Customers recommend the company to others when they perceive high quality and spread negative WOM when they perceive low service quality (de Matos and Rossi 2008).
Together with research on antecedents, research also has been performed on variables that could have a mediating or moderating effect. Verhoef et al. (2002) found no moderating effect of relationship age on the effect of trust, commitment, satisfaction, payment equity on the dependent variable WOM. A key premise underlying the moderating effect of relationship age is increasing customer confidence in one‘s impressions about the supplier as the relationship ages (Verhoef et al. 2002). In addition, research shows that commitment has a (partial) moderating effect on the effect of satisfaction and identification on WOM (Brown et al. 2005).
discussed in paragraph 2.1.4 also influence WOM. Furthermore, commitment was found to have a moderating effect while this was not proven for age of the relationship.
2.3 WOM as a Driver
It is likely to assume that consumers recommending your brand results in increasing sales. But is this true? How strong is this effect, and are there perhaps other related effects of WOM?
There is little debate as to whether WOM matters to the firm (Godes and Mayzlin 2004). One of the most widely accepted notions in consumer behavior is that word of mouth communication plays an important role in shaping consumers‘ attitudes and behaviors (Brown and Reingen 1987). A McKinsey & Company study found that 67% of the sales of consumer goods are based on WOM (Liu 2006). Multiple studies show that WOM is more influential and effective than traditional media in effecting purchase decisions (Katz and Lazarsfeld 1955; Brown and Reingen 1987). This is confirmed by Villanueva et al. (2008), who show that WOM customers add nearly twice as much long-term value to the firm compared to marketing-induced customers.
Research confirms the primacy of WOM as one of the key drivers of firm sales (Coleman et al. 1966; Arndt 1967; Engel et al. 1969). Coleman et al. (1966) found that the diffusion of the investigated product was the result of social contagion. While the research of Coleman et al. (1966) is often referred to as proof of the effect between diffusion and WOM, new analysis shows that this contagion effect disappears when marketing efforts are controlled for (Van den Bulte and Lilien 2001).
Nevertheless, other studies show the effect of WOM. Van den Bulte and Lilien (2003) for example, found in another study evidence of social contagion. Godes and Mayzlin (2004) showed that online conversations may offer an easy and cost-effective opportunity to measure WOM. Liu (2006) shows that WOM information offers significant explanatory power for aggregate and weekly box office revenue. Most of this explanatory power comes from the volume instead of its valence (Liu 2006).
damaging effects on shareholder value (Luo 2009). Furthermore NWOM could influence stock returns through decreasing brand equity (Luo 2009).
Concluding, WOM is one of the key drivers of sales. Research has proven this effect several times. On the other side, negative WOM has significant direct short and long term effects on firm cash flows and stock prices. In the following table an overview of antecedents and outcomes of WOM is given.
Table 1: Antecedents and outcomes of WOM
2.4 Social network and WOM
relational form. Tie strength is influenced by variables like the importance attached to the social relation, type of social relation and frequency of social contact (Granovetter 1973; Weimann 1983). In research, ties are either strong, weak, or absent. The specific strength of weak ties is found to be their significant role to be bridges. It allows information to travel, from one distinct subgroup composed of referral actors, to another in the broader social system (Granovetter 1973; Brown and Reingen 1987). On the opposite side, strong ties are more important at the micro level of referral behavior. When weak ties and strong ties were both available as source of information, strong ties were more likely to be activated than weak ties for the flow of information (Brown and Reingen 1987). Furthermore, strong ties were perceived by receivers as more influential than weak ties in decision making (Brown and Reingen 1987). As a result, the bridging function of weak ties is more conducive to the flow of information, whereas strong ties are more crucial to the flow of influence (Brown and Reingen 1987).
The idea that members of a consumer network exchange market messages and meanings, instead of a unidirectional flow, is the result of the evolutionary shift from the linear marketer influence model towards the network coproduction model (Kozinets et al. 2010). Another distinguishing characteristic of this model is the use of new tactics and metrics to deliberately and directly target and influence the consumer or opinion leader. Research shows that communal WOM does not just increase or amplify marketing messages; rather, marketing messages and meanings are systematically altered in the process of embedding them (Kozinets et al. 2010). The Network Coproduction Model is visible in figure 3 below.
Figure 3: The Network Coproduction Model
development and increasing importance of internet, but it is not limited to this domain (Kozinets et al. 2010).
2.5 Online WOM
3 Research Framework
3.1 Main EffectsThe degree of knowledge someone has of a brand differs for everyone. Brand familiarity reflects the brand-related experiences accumulated by the consumer (Alba and Hutchinson 1987). When the brand familiarity is higher, consumers have better knowledge of the organization (Keller 1993). Consumers with more knowledge are expected to have more confidence in their evaluations (Swann and Oliver 1989). Confidence serves as a psychological gatekeeper. It systematically determines whether people translate their beliefs into action (Berger and Mitchell 1989; Fazio and Zanna 1978; Pieters and Verplanken 1995). Studies on brand familiarity suggest that consumers react more favorably toward a familiar brand than they do toward an unfamiliar brand (Sundaram and Webster 1999). These customers are likely to endorse the company in order to keep cognitive consistency and justify their strong identification with the company (Brown et al. 2005). This should lead to more WOM usage. As a result the following hypothesis is formulated:
Hypothesis 1: Higher brand familiarity leads to higher intention to use WOM
identification and WOM (Brown et al. 2005). According to their future research direction, the following hypothesis is formulated:
Hypothesis 2: Brand-self connection has a positive effect on intention to use WOM
A related but different concept is self-image congruence. Whereas brand-self connection measures whether a brand is integrated into a consumer‘s concept, the self-image congruence investigates whether the personal self-image a consumer has of a brand, matches the view they have of their self. Both factors are related but (conceptually) very different. The above mentioned personal image is important. Products, suppliers and services are all assumed to have personal images, just as people do. It differentiates from functional attributes since it does not describe the product in terms of tangible costs and benefits, but more as the stereotype of the generalized users of that product (Sirgy et al. 2001). Self-image congruence positively affects brand loyalty directly and indirectly through functional congruity, product involvement and brand relationship quality (Kressmann et al. 2006). This self-image congruence and brand loyalty link has been confirmed by Sirgy et al. (2008). Brand loyalty influences intention to use WOM (Bowman and Narayandas 2001; De Matos and Rossi 2008). As a result, a direct effect between self-image congruence and intention to use WOM is expected (Brown et al. 2005). Self-image congruence affects customers‘ brand preferences and their purchase intentions (Mehta 1999). Furthermore, it facilitates positive behavior and attitudes toward brands (Sirgy 1982 1985; Sirgy et al. 1991 1997) and it is positively related to customers‘ product evaluations (Graeff 1996). Besides, self-image congruence is a strong predictor of brand satisfaction (Sirgy et al. 1997). Brand satisfaction is positive related towards intention to use WOM (Blodgett et al. 1993; Mittal et al. 1999; Richins 1983; Sundaram et al. 1998; Swan and Oliver 1989). Hence, a positive effect of self-image congruence on WOM is expected and the following is hypothesized:
Hypothesis 3: Self-image congruence has a positive effect on intention to use WOM
(Verhoef et al. 2002). This research tries to fulfill this future research direction by looking at the communication services sector. In line with earlier research (Verhoef et al. 2002) the following hypothesis is formulated:
Hypothesis 4: Payment equity is positively related to intention to use WOM
Empirical research has demonstrated that quality is a relevant predictor of WOM (Bloemer et al. 1999; Boulding et al. 1993; Harrison-Walker 2001; Zeithaml et al. 1996). When a customer‘s perception of service quality is high, the customer‘s behavioral intentions in terms of recommendations are favorable (Parasuraman et al. 1988; Zeithaml et al. 1996). Customers recommend the company to others when they perceive high quality and spread negative WOM when they perceive low service quality (de Matos and Rossi 2008). Therefore the following is hypothesized:
Hypothesis 5: Quality has a positive effect on intention to use WOM
Another important factor influencing WOM is satisfaction. The level of customer satisfaction has an influence on two purchase behaviors: repurchase intentions and WOM (Maxham and Netemeyer 2002; Oliver 1980; Ranaweera and Prabhu 2003). The likelihood of customers spreading WOM depends on the extent to which their expectations are exceeded (de Matos and Rossi 2008).This results in the following hypothesis:
Hypothesis 6: The level of satisfaction has a positive influence on intention to use WOM
Next to satisfaction, a related factor is loyalty. Loyalty is an antecedent of WOM because loyal customers are more likely to give positive recommendations of the company to the individuals in their reference group (de Matos and Rossi 2008). Furthermore, loyal customers have greater motivation for processing new information about the company and have stronger resistance to being persuaded by contrary information (Dick and Basu 1994). Consistent with earlier research the follow hypothesis is formulated:
Hypothesis 7: There is a positive effect of loyalty on intention to use WOM
commitment as a whole and does not make a distinction between calculative and affective commitment. Based on these findings the following is hypothesized:
Hypothesis 8: Commitment is positively related to intention to use WOM
The last independent variable that is taken into account is trust. Trust refers to a willingness to rely on an exchange partner in whom one has confidence (Moorman et al. 1993). Research has shown that higher levels of trust are associated with a greater tendency to offer favorable WOM (Garbarino and Johnson 1999; Gremler et al. 2001; Ranaweera and Prabhu 2003). A customer will be more likely to endorse a provider that he or she has previous experience with and confidence in (Gremler et al. 2001). In line with earlier research the following hypothesis is formulated:
Hypothesis 9: There is a positive effect of trust on intention to use WOM
3.2 Moderating Effect of Brand Strategy
After the main effects, the moderating effect of strategy of the brand is investigated. Three generic strategies are identified; low cost strategy, focus strategy and differentiation strategy (Porter 1980). No research on the (moderating) effect of strategy of the brand on intention to use WOM has been performed yet. However, relevant theory indicates there might be an effect (Wangenheim and Bayon 2004).
It is likely that a premium brand attracts consumers who are more demanding and also appreciate service more, than consumers choosing for a low cost brand. In turn, the brand following a differentiation strategy is also more likely to perform better than a low cost brand on aspects of quality and service (Porter 1980). According to the confirmation and disconfirmation paradigm, a service and quality oriented provider is more likely than a price fighter to exceed expectations (Wangenheim and Bayon 2004). This results in positive disconfirmation to occur and so leading to higher levels of satisfaction (Oliver 1993). Satisfaction subsequently positively influences WOM behavior (Blodgett et al. 1993; Mittal et al. 1999; Richins 1983; Sundaram et al. 1998; Swan and Oliver 1989).
and Oliver 1989). Following this line of reasoning the strategy of the brand is likely to have a moderating effect. Therefore the following is hypothesized:
Hypothesis 10: The effect of the investigated variables on the intention to use WOM is higher for consumers of the premium brand
3.3 Moderating Effect of Age of the Relationship
Earlier research found no support for the moderating effect of relationship age on the effect of commitment, trust, satisfaction and payment equity on the intention to use WOM (Verhoef et al. 2002). However, this paper takes other relational constructs into account and a different sector is analyzed. Therefore, the moderating effect of relationship age is also investigated in this study. The hypotheses are mostly based on one main stream of literature (Verhoef et al. 2002). Several studies show that length of relationship is positively related to confidence in one‘s evaluations of the partner (Swann and Gill 1997). As a result, the effect of a variable on the intention to use WOM is likely to differ throughout the relationship.
Customers are found to be more value conscious in lengthy relationships (Reinartz and Kumar 2000). In other words, customers with lengthy relationships will pay more attention to the price paid for their services. This logically results in a stronger effect of payment equity in lengthy relationships (Verhoef et al. 2002). This effect can be explained by the fact that in longer relationships, customers will perhaps have greater confidence in their evaluations of the price paid (Swann and Gill 1997). This results in relying more on payment equity in longer relationships. Therefore the following hypothesis is formulated:
Hypothesis 11a: Relationship age increases the positive effect of payment equity on intention to use WOM.
In the early phases of the relationship consumers don‘t have much experience with the company (Swann and Gill 1997). It is hard for consumers to form a good evaluation of the quality of the product. As the relationship develops, the richness of impressions increases and as a result the confidence in their evaluation increases (Swann and Gill 1997). Following this line of reasoning, as the relationship matures consumers tend to rely more on their evaluation of quality. This results in the following hypothesis:
Hypothesis 11b: Relationship age increases the positive effect of quality on intention to use WOM.
1997). Research on a related concept, customer satisfaction, shows that the effect is larger in lengthy relationships (Rust et al. 1999). Based on the above reasoning and related empirical evidence the following is hypothesized:
Hypothesis 11c: Relationship age increases the positive effect of loyalty on intention to use WOM.
Committed customers will have relatively intimate relationships with the company. Nevertheless, this intimacy develops over the long run (Dwyer et al. 1987). Consumers in a lengthy relationship have more information about the company which leads toward an increased richness of the customer‘s impression about the brand (Swann and Gill 1997). This increased richness leads to an enhanced confidence in the beliefs about the relationship (Gill et al. 1998). This confidence improves the impact of these beliefs (Berger and Mitchell 1989; Dick and Basu 1994; Pieters and Verplanken 1995). As a result, the effect of commitment increases as the relationship matures. Consequently, the following hypothesis is formulated:
Hypothesis 11d: Relationship age increases the positive effect of commitment on intention to use WOM
Trust is especially essential in the early phases of a relationship (Jap 1999). This is caused by the fact that customers do not have much experience with the company in the early phases of a relationship (Swann and Gill 1997). Therefore, it is harder for customers to base their behavior on the evaluation of their experience (Verhoef et al. 2002). As a result, customers will rely on the perceived trustworthiness (Garbarino and Johnson 1999). Therefore the following is hypothesized:
Hypothesis 11e: Relationship age decreases the positive effect of trust on intention to use WOM
Research shows that the effect of customer satisfaction on WOM is larger in lengthy relationships (Rust et al. 1999). This is in line with Bolton (1998), who reported that the positive effect of satisfaction on relationship duration is enhanced by relationship age. Consistent with earlier research the following hypothesis is formulated:
Hypothesis 11f: Relationship age increases the positive effect of satisfaction on intention to use WOM
3.4 Control Variables
use WOM, no explicit hypotheses are formulated. The control variables are solely included to account for some observed effects. Included in this research are the demographic variables gender, age and household size. Much research on differences between gender and several dependent variables have been performed. For example, women are more likely to demonstrate impulse purchasing behavior than men (Inman et al. 2009). It is likely that there is an effect between gender and intention to use WOM. Consumers in different age groups are likely to show diverse behavior. Children for example use fewer sources and less information when selecting products than grown-ups. (Capon and Kuhn 1980). On the other hand, elderly adults remember less product-related information than younger adults (John and Cole 1983; Stephens 1982).The last control variable included is household size. The size of a household can have major influences on a variety of variables. For example, household size influences in-store decision making (Inman et al. 2009). To account for the variance in the intention to use WOM caused by gender, age and household size these variables are included.
3.5 Conceptual framework
Based on the above stated hypotheses, the following conceptual framework is developed:
Figure 4: Conceptual Framework
4 Research Design
4.1 Data collectionThe data was collected through an online survey. A total of 19.000 customers of three communication services providers in the Netherlands received an email to participate in this survey. After one week, the customers who did not respond yet received a reminder. The selection procedure was randomized. A total of 1813 customers filled in the questionnaire. This results in a response rate of 9.5%. The other customers refused to participate, did not read the email, or did not receive the email due to incorrect contact information. After accounting for cases with missing values, a total of 1666 customers remained. From the participated customers 468 are from the normal brand, 501 from the price fighter and 697 from the premium brand.
4.2 Construct measurement
According to earlier research, a five-step procedure to develop construct measurements was used (Verhoef et al. 2002; Churchill 1979; Steenkamp and van Trijp 1991).
In step 1 a literature review was conducted to develop items. For brand familiarity three items were adapted from Kent and Allen (1994). In addition, four items that measure brand-self connection were adapted from Escalas and Bettman (2003). For self-image congruence, one item was adapted from Sirgey et al. (2001). This selected item outperforms other traditional measurements of self-image congruence (Sirgey et al. 2001). Also one item was adapted from Singh (1990) and Bolton and Lemon (1999) to measure payment equity. For service quality the dominant SERVPERF method is used, originally consisting of 22 items (Cronin and Taylor 1994). SERVPERF only uses the more concise performance-only scale (Cronin and Taylor 1994) from the well-known SERVQUAL method (Parasuraman et al. 1988). Six items that measure customer satisfaction were adapted from Oliver (1980, 1993). For the loyalty construct, two items from Zeithaml et al. (1996) were used. Besides, two items for commitment from Brown et al. (2005) were adapted. For the construct trust, three items from Crosby et al. (1990) were adapted. For measuring the dependent variable intention to use WOM, one item was adapted from Reichheld (2003).
items measuring service quality, and in specific the tangibles aspect, were replaced by three items measuring the product quality. The final questionnaire can be found in appendix 1.
In step 3 the questionnaire was pre-tested. It was distributed online among about 100 people. A total of 29 people responded, including 16 who have completed the questionnaire. There were three respondents who didn‘t know who their internet provider was. Furthermore, there were ten respondents who didn‘t complete the questionnaire. This high percentage of respondents not completing the questionnaire (34.5%) can be explained by the fact that a convenience sample is used. Due to the fact that social media and online platforms are used to gather respondents, there might be a lower commitment to complete the questionnaire. Most of the respondents were students and they usually share their internet connection. Because of this fact, it is possible that the subject of the questionnaire was not compelling enough for the respondents. The goal of this pre-test was to check the validity of the questionnaire and whether the questions are clear.
In step 4 the results of the pre-test were used to test the developed questionnaire. Based on earlier research (Verhoef et al. 2002), coefficient alpha and exploratory factor analysis are used to test the reliability and validity of the items. For brand familiarity, service quality, product quality, commitment and loyalty, all the measured items seem to load on their construct. However for brand-self connection, one item does not load significant on brand-self connection. In contrary, it looked like it was a better measurement of the construct brand familiarity. Due to the low number of respondents (16), the results have to be taken with great caution. Recommended is to have at least 50 respondents, or 5 per variable (Hair et al. 2009). Therefore it is decided not to delete any item based on the pre-test. However, it gives an indication to look extra careful at this item during the analyses.
In addition to the factor analysis, four respondents were asked what they thought about the questionnaire and if questions were unclear. Based on these qualitative interviews, two items measuring service quality were reformulated.
In step 5 the definitive questionnaire was distributed among 19.000 customers. To validate these measures, the renowned four-step procedure as described in Gerbing and Anderson (1988) was used. The next part of this paper starts with the results of this measurement validation.
4.3 Scale measurement
adopted from, all used this same seven point Likert-scale. Furthermore, it has proven to be a statistically good measurement system (Malhotra 2009). The question measuring the intention to use WOM is on an eleven point scale as suggested by Reichheld (2003).
4.4 Plan of analysis
5 Results
5.1 Overview
Out of the three providers, the customers of the premium provider show the highest intention to use positive WOM with an average score of 8.22, followed by the normal provider with an average of 6.48 and at last the price fighter with a score of 6.38. The distribution of the scores for the three providers, ranging from zero to ten, is visualized below.
Figure 5 WOM intention
Furthermore a boxplot of the distributions of respectively the normal, price fighter and premium provider are made and shown below.
Figure 6 Boxplot
It is clearly visible that the normal brand has many values of 6, 7 and 8. While the distribution of the price fighter is flatter and the premium provider is clearly negative skewed. This section will start with the explorative and confirmative factor analysis, followed by testing whether pooling is allowed, analyzing the assumptions of Ordinary Least Squares (OLS) regression and at last testing the hypotheses.
5.2 Factor Analysis
5.2.1 Exploratory factor analysis
confirmative factor analysis. Questions 2 and 6, regarding price equity and self-image congruence, are not included in the factor analysis. This because both questions are the only questions for respectively price equity and self-image congruence. Therefore, further reduction is not possible and they are excluded. Before explorative factor analysis is applied, the appropriateness is analyzed. With a high KMO value of 0.963 and a significant Bartlett’s test (p<0.05) factor analysis seems appropriate. With 1666 respondents for the factor analysis the N is by far large enough to perform factor analysis (Hair et al. 2009). To assess whether all the variables load sufficiently on at least one factor the communality scores are checked. Six variables with a score lower than 0.5 are evaluated and deleted. The deleted variables include two items intended to measure brand-self connection and four intended to measure service quality. One of these deleted items was also evaluated for deletion during the pre-test. The deleted items are reported in appendix 2. All the six deleted variables have in common that the mean lays around 4 (neutral score). This could indicate that consumers don’t have a clear opinion on the stated questions and therefore they don’t load on a specific factor.
5.2.2. Extraction of factors
Factor 1 Factor 2 Factor 3 Factor 4 Factor 5 Factor 6
Reliability Satisfaction Empathy Connection Product Quality Brand Familiarity Assurance Loyalty Responsive Commitment
V8.5,8.6 10.1-10.4, 10.9 11.1, 11.2, 11.7 V7,1-8,2, 8.6 V10.4,10.7 11.4-11.6 V4.2,4.4, 8.3-8.5 V9.1-9.3 V3.1-3.3 α: 0.945 α: 0.938 α: 0.827 α: 0.883 α: 0.880 α: 0.709
Table 2 Factor analysis
To assess the validity of this six factor solution, the exploratory factor analysis is re-executed but this time using a random split sample. The first sample suggests a five factor solution with service quality as one factor, while the second sample suggests a six factor solution with service quality divided into reliability and assurance on the one hand and empathy and responsiveness on the other hand. All the other factors stay the same. The results of this split sample factor analysis indicate that the validity of the model is good. The solution of the factor analysis is rotated using Varimax.
5.2.3 Confirmatory factor analysis
Fit index Value Treshold RMSEA 0.047 ≤ 0.07 Chi square 0.00 ≥ 0.05 Chi-square 1694(df473) SRMR 0.033 ≤ 0.08 AGFI 0.91 ≥ 0.90 CFI 0.99 ≥ 0.92
AIC 1989.88 Low as possible
Table 3 enhanced fit
Except for the Chi-square statistic, all of the indices have acceptable values. The Chi-square statistic is significant but this is caused by the large sample size. This model is the final model, however there are still modifications possible that are significant ( x2≥ 4). Due to the large
sample size significant modifications are almost always detected (Hair et al. 2009). The factor scores of this final model are used, instead of all the separate variables, to test the hypotheses.
5.3 Pooling
To find out whether the data of the three providers can be aggregated, Chow’s test for pooling is used. The degrees of freedom and residual sum of squares of the unpooled model are 1639 and 2735.53. For the fully pooled model the degrees of freedom are 1657 and the residual sum of squares is 2883.19. This results in a value of 4.91. Following the F-distribution F(18,1639) 4.91=0.00. Therefore H0 is rejected and pooling is not allowed. Since fully pooling is not allowed, several other options to model the data are investigated. A multilevel model with a random intercept using MLWIN 2.24 and a normal regression with dummy variables are estimated. However, following Chow’s test, pooling is also not allowed for the multilevel model and the dummy variables model. Therefore, three separate models are used to test the hypotheses.
Model Degrees of Freedom Residual Sum Squares F p-value
Unpooled 1639 2735,53 OLSDV 1655 2797,87 2.33 0.00 Multilevel 1654 2803,52 2.67 0.00
Table 4 Chow test
5.4 Assumptions
5.4.1 Normality assumption
To investigate whether the residuals are normal distributed, the residuals are plotted, the Kolmogorov-Smirnov test is used and the kurtosis and skewness are inspected. To investigate the distribution of the residuals the Q-Q plots are shown below for respectively the normal brand, the price fighter and the premium brand. Neither of them follows a normal distribution.
Figure 7 Normality Plot
This is confirmed by the significant Kolmogorov-Smirnov test. Investigating the distributions it is visible in table 5 that all three have a negative skewness and positive kurtosis. This indicates that all the three distributions are skewed to the left and have a sharp peak around the mean. This is also confirmed when looking at figure 5 on page 29.
Brand Kolmogorov-Smirnov Skewness Kurtosis
Normal brand 0.076 (p=0.00) -0.825 2.507
Price fighter 0.109(p=0.00) -1.274 5.468
Premium provider 0.062(p=0.00) -0.992 4.549
Table 5 Normality 1
To deal with this non-normality, the square root of the dependent variable is taken and the regression is estimated again. Since all the distributions show a negative skewness the dependent variable is reverse coded. After the transformation has been performed, the reverse of the dependent variable was taken again to enhance the interpretability of the results. Since the exact value of the dependent variable is on itself not meaningful like for example height or sales no information was lost with recoding the data. It is visible in the Q-Q plots below of respectively the normal brand, price fighter and premium brand that the residuals seem to be more normally distributed. Next to the square root, also the log(10), log(2) and ln are taken. However, the square root gave the best results.
To further investigate these distributions the Kolmogorov-Smirnov value, skewness and kurtosis are estimated again. The Kolmogorov-Smirnov test is not significant for the premium brand, indicating a normal distribution. For the other two brands it still indicates a non normal distribution. However, the normality assumption is less seriously violated as visible in the table below. With larger sample sizes it is easier to reject the normality hypothesis; therefore it is unappealing to follow strict rules with regard to this violation (Leeflang et al. 2000). As a result, no further transformations or changes are made.
Brand Kolmogorov-Smirnov Skewness Kurtosis
Normal brand 0.059 (p=0.00) 0.049 1.687 Price fighter 0.073 (p=0.00) 0.480 4.013 Premium provider 0.027(p=0.20) 0.200 1.055
Table 6 normality 2
5.4.2 Multicollinearity assumption
5.4.3 Homoskedasticity assumption
Next to no multicollinearity, homoskedasticity is another assumption of linear regression. To assess whether there is heteroskedasticity, the Quandt test is used. The Goldfeld-Quandt test is applied to test whether the error variances differ between the group who is less likely to recommend their provider (score of 0-6) and the group who are more likely to recommend their provider (score of 7-10). The distinction between 0-6 and 7-10 is made because this results in two groups of more or less the same size. As visible in the table below, the null-hypothesis is rejected for the normal brand and the price fighter. Therefore, the homoskedasticity hypothesis is rejected for both brands. There is no proof of heteroskedasticity for the premium provider.
Brand Goldfeld-Quandt
Normal brand 1.38 (p=0.00) Price fighter 1.53 (p=0.00) Premium provider 1.02(p=0.43)
Table 7 Goldfeld-Quandt test
To deal with this heteroskedasticity, generalized least squares is applied for the normal brand and the price fighter. All the scores on the variables for the respondents of the price fighter with a WOM score of at least 7, are divided by 0.245, the standard deviation of the residuals of this group. The scores of the group with a score lower than 7 are divided by 0.30, the standard deviation of the residuals of this group. For the normal brand the scores are divided by respectively 0.247 and 0.287.
5.5 Results
After the assumptions are met the following model can be specified:
ij j ij j ij j ij j ij j ij j ij j ij j j ij X X X X X X X X WOM 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 ij
WOM = intention to use WOM for customer i from brand j
X1 = Self-image congruence X2 = Price perception
X3-X8 = Factors 1-6 respectively
Payment Equity 14% Self-image congruence 4% Reliability/Assurance 20% Satisfaction/Loyalty 48% Empathy/Responsive 0% Connection/Commitment 5% Product Quality** 6% Brand Familiarity 3% 5.5.1 The normal brand
For the normal brand the whole model is significant. 69.6% of the variance is explained by this model. In the model the constant is significant. Furthermore, the payment equity, the reliability and assurance construct, the loyalty and satisfaction construct, and product quality are significant. Product quality is in the opposite direction of what was expected. For brand familiarity, the connection/commitment construct and self-image congruence no significant effect was found. Therefore, hypotheses 1, 2, 3 and 8 are rejected. For the payment equity, the reliability/assurance construct, and the satisfaction/loyalty construct a significant effect was found. As a result, hypotheses 4, 6, 7 and 9 are accepted. The quality hypothesis is partly confirmed since for the empathy/responsive construct no significant effect was found and for product quality a negative significant effect was found while the opposite was expected. No logical explanation can be given for this opposite effect. One would expect that the higher the product quality, the higher the willingness to recommend. The output of the model for the normal brand is visible in the table below.
Hypothesis Unstandardized Estimate Standard Error Significance
Intercept Brand Familiarity Connection/Commitment Self-image Congruence Payment Equity Reliability/Assurance Empathy/Responsive Product Quality Satisfaction/Loyalty - 1 2,8 3 4 5,9 5 5 6,7 -3.500 0.004 0.005 0.003 0.014 0.020 0.000 -0.006 0.042 0.076 0.003 0.004 0.002 0.004 0.005 0.004 0.003 0.005 0.000 0.242 0.250 0.268 0.000 0.000 0.911 0.039 0.000
Adj. R-square: 0.696 F-value:134.860(0.000) Table 8 Estimates normal brand
On the right side it is visible that satisfaction and loyalty have by far the largest influence. Product quality and brand familiarity are in the opposite direction. Self-image congruence has the
smallest effect
Figure 9 pie chart normal brand
Payment Equity 17% Self-image congruence 4% Reliability/Assurance 7% Satisfaction/Loyalty 53% Empathy/Responsive 4% Connection/Commitment 9% Product Quality 1% Brand Familiarity 5% 5.5.2 Price fighter
Just like for the normal brand, the whole model is clearly significant. For the price fighter 71.0% of the total variance is explained by this model. Next to the intercept, the payment equity, the satisfaction and loyalty construct, and the connection and commitment construct are significant. Since brand familiarity has no significant effect on the intention to use WOM, the first hypothesis is rejected. Hypotheses 2 and 8, regarding the brand-self connection and commitment, are accepted. After satisfaction and loyalty, the connection and commitment construct has the largest influence on the intention to use WOM. For self-image congruence no significant effect was found, therefore hypothesis 3 is rejected. The empathy/responsive construct, product quality and the reliability/assurance construct are not significant. Therefore, hypotheses 5 and 9 are fully rejected. The satisfaction and loyalty construct is significant and has the largest influence. Therefore hypotheses 6 and 7 are both accepted. An overview is provided in the table below.
Hypothesis Unstandardized Estimate Standard Error Significance
Intercept Brand Familiarity Connection/Commitment Self-image Congruence Payment Equity Reliability/Assurance Empathy/Responsive Product Quality Satisfaction/Loyalty - 1 2,8 3 4 5,9 5 5 6,7 -3.459 0.005 0.009 -0.003 0.014 0.007 0.005 0.000 0.044 0.059 0.003 0.004 0.002 0.003 0.005 0.004 0.003 0.004 0.000 0.070 0.010 0.216 0.000 0.153 0.282 0.886 0.000
Adj. R-square: 0.710 F-value:155.006 (0.000) Table 9 Estimates price fighter
Just like for the normal brand, satisfaction and loyalty have the largest influence on intention to use WOM. On the second place is payment equity. Connection and commitment also show a significant effect. However, the effect of payment equity is larger.
Figure 10 pie chart price fighter
5.5.3 Premium brand
With an F-value of 127.109 the whole model is significant. 59.2% of the total variance is explained by this model. Besides the intercept, payment equity, the reliability assurance construct, the satisfaction/loyalty construct, and the empathy/responsiveness construct are significant. The effect of empathy/responsiveness is in the opposite direction of what was expected. For brand familiarity, the connection/commitment construct, and self-image congruence no significant effect was found. Therefore, hypotheses 1, 2, 3 and 8 are rejected. Payment equity and the satisfaction/loyalty construct are clearly significant. As a result, hypotheses 4, 6 and 7 are accepted. Hypothesis 5 is partly accepted: For the reliability/assurance construct a positive significant effect was found, for product quality there is no significant effect, and for the empathy/responsive construct a negative significant effect. This negative effect can be caused by the fact that the more experienced internet users are member of the premium brand and that they don‘t appreciate empathy or responsiveness and want to be satisfied instead. Since the reliability and assurance construct has a positive significant effect, hypothesis 9 is accepted. The estimates are visualized in the table below.
Hypothesis Unstandardized Estimate Standard Error Significance
Intercept Brand Familiarity Connection/Commitment Self-image Congruence Payment Equity Reliability/Assurance Empathy/Responsive Product Quality Satisfaction/Loyalty - 1 2,8 3 4 5,9 5 5 6,7 -3.579 0.023 0.008 0.005 0.051 0.084 -0.034 -0.007 0.232 0.093 0.012 0.013 0.008 0.011 0.020 0.015 0.011 0.016 0.000 0.059 0.550 0.570 0.000 0.000 0.025 0.522 0.000
Adj. R-square: 0.592 F-value:127.109 (0.000)
Table 10 Estimates premium brand
The pie chart on the right shows the same pattern as that of the normal brand. Satisfaction and loyalty have the largest influence, followed by the reliability and assurance construct. The empathy and responsiveness construct is in the opposite direction.
Figure 11 pie chart premium brand
5.5.4 Main effects
For all the three brands no support was found for hypotheses 1 and 3, therefore the brand familiarity hypothesis and the self-image congruence hypothesis are fully rejected. For the brand-self connection partial proof has been found and as a result hypothesis 2 is partly accepted. Hypotheses 4, 6, and 7 are accepted. A positive significant effect has been found for payment equity and the satisfaction/loyalty construct for all the three brands. Hypothesis 5, 8, and 9 are partly accepted. For the quality construct, commitment, and trust partial proof has been found. An overview of the hypotheses is provided in the table below.
Hypothesis Construct Result
H1 Brand Familiarity Rejected
H2 Brand-self connection Partly accepted H3 Self-image congruence Rejected
H4 Payment equity Accepted
H5 Quality Partly accepted
H6 Satisfaction Accepted
H7 Loyalty Accepted
H8 Commitment Partly accepted
H9 Trust Partly accepted
Table 11 hypotheses
5.5.5 Interaction Effects
To investigate the interaction effect of the strategy of the brand, the slopes are tested for equality. The betas and standard errors are used in the following equation to obtain the T-statistic which is shown in the table below. Significant differences (p< 0.05) are marked with an asterisk. 2 2 2 1 2 1 . SE SE T (2)
Price fighter and normal Normal and premium Price fighter and premium
Payment equity -0,66461 -0,34454 -1,11969 Self-image congruence 2,097802* -0,67658 1,616516 Reliability/Assurance 2,322073* -0,17946 2,155249* Satisfaction/Loyalty -0,25653 1,806252 1,583571 Empathy/Responsiveness -0,74505 -1,32398 -2,01389* Connection/Commitment -1,47588 -0,68614 -2,34189* Product quality -1,94156 1,254287 -0,65978 Brand familiarity -0,26844 1,317137 1,158077
Table 12 Equality of coefficients
39 construct is significant higher for the normal brand compared to the price fighter. In order to test the relationship age hypothesis, the natural logarithm of the relationship age in years was added as a moderator variable. Due to privacy limitations, no individual level data was available for the premium brand. Therefore, the relationship age hypothesis could not be tested. The only significant (negative) interaction effect that was found was for the payment equity for the normal brand. This effect was in the opposite direction of what was expected. Therefore all hypotheses regarding the interaction effect of relationship age are rejected. This is in line with earlier findings (Verhoef et al. 2002). The estimates are in the table below.
Normal brand Price fighter
Variable Beta Std. Error Significance Beta Std. Error Significance Connection/Commitment -0.006 0.004 0.153 -0.001 0.002 0.492 Payment Equity -0.009 0.004 0.028* -0.001 0.002 0.607 Reliability/Assurance -0.002 0.015 0.905 0.002 0.004 0.602 Empathy/Responsive 0.019 0.012 0.104 -0.000 0.003 0.913 Product Quality -0.002 0.003 0.452 -0.000 0.002 0.951 Satisfaction/Loyalty 0.007 0.005 0.131 0.002 0.003 0.450
Table 13 Relationship age
5.6 Model Validation
To investigate the predictive validity of the model a holdout sample of 25% was taken. The remaining 75% is used to predict the intention to use WOM for the holdout sample. The mean absolute percentage error (MAPE) of this model is compared against the MAPE of the naïve model. The naïve model only consists of a constant, the mean of the intention to use WOM for the specific brand. The MAPE is visible in the table below.
Brand MAPE specified model MAPE naïve model
Normal brand 12.57% 14.16%
Price fighter 11.40% 16.22%
Premium provider 5.14% 6.74%
Table 14 MAPE
6 Conclusion and Managerial Implications
6.1 ConclusionThis paper investigated the effects of several antecedents on the intention to use WOM for three different brands. The strategy of the brands and relationship age were used as moderating variables. Several effects were found and will be discussed in this section.
6.1.1 Brand-related factors
For the investigated brand-related factors only one significant effect was found. For both brand-familiarity and self-image congruence no significant effect was found. Therefore hypotheses 1 and 3 are rejected. The only significant effect was for the connection and commitment construct for the price fighter. Since this construct consists of the brand-self connection and commitment, hypotheses 2 and 8 are partly accepted.
6.1.2 Product-related factors
For all the three brands, payment equity has a positive significant effect on intention to use WOM. Therefore, hypothesis 4 is fully accepted. The effects are comparable, however for the price fighter the relative largest effect was found. For the effect of quality on the intention to use WOM only partial proof has been found. As a result, hypothesis 5 is partly accepted. The quality construct consists of the reliability and assurance factor, the empathy and responsiveness factor, and product quality. The reliability and assurance factor has a significant effect on intention to use WOM for the normal brand and the premium brand. While for the empathy and responsiveness construct and product quality, no positive significant effect has been found. The two significant effects that were found, for both the premium brand and the normal brand, were negative.
6.1.3 Relationship-related factors
Satisfaction and loyalty are also combined into one construct. This satisfaction and loyalty construct is significant for all the three brands and also the most important antecedent for all the three brands. Therefore hypotheses 6 and 7 are jointly accepted. As mentioned before, commitment and connection together form a construct. This construct is only significant for the price fighter and therefore hypothesis 8 is partly accepted. Since hypothesis 9, trust, is part of the reliability and assurance construct, this hypothesis is also only partly accepted.
6.1.4 Moderating effects and control variables
of the variables and their standard errors are investigated. Only the effect of the reliability and assurance construct is significant higher for the premium brand compared with the price fighter, for the empathy and responsiveness construct and the connection and commitment construct the effect was in the opposite direction. Between the effects of the variables of the normal brand and the premium provider no significant differences were found. Concluding, hypothesis 10 is largely rejected. Due to privacy limitations no individual level data was available for the premium brand, because this is against their privacy policy. Therefore, the moderating effect of relationship age can only be examined for the normal brand and the price fighter. Only for the payment equity of the normal brand a negative significant effect was found, which is in the opposite direction of what was expected. Therefore, hypothesis 11 is fully rejected. The effect of the control variables age, gender and household size are also taken into account. Because of the same privacy limitations, the effect of age and gender are not taken into account for the premium brand. Only for the control variable age a positive significant effect was found for the price fighter. For gender and household size no significant effect was found.
6.1.5 Research Questions
Looking at the problem statement and research questions this research has proven that satisfaction, loyalty and payment equity influence the intention to use WOM for all the three brands. For the brand-self connection, commitment, quality and trust partial proof has been found. While for brand familiarity and self-image congruence no significant effect was found. For quality also negative effects have been found. The effects of the variables on the intention to use WOM differ across brands, indicating some sort of moderating effect of the strategy of the brand.
6.2 Managerial Implications
6.2.1 Relationship-related implications
usage. Satisfaction and loyalty have by far the largest influence on the intention to use WOM for all the three brands.
6.2.2 Product-related implications
Also the payment equity plays an important role for all the three brands. Payment equity comes on the third place of relative positive influence on intention to use WOM for the normal brand and the premium brand. It is on the second place for the price fighter. Furthermore, the normal and premium brand should focus on their reliability and assurance to increase the intention to use WOM by their customers. For the price fighter this effect is negligible. Investigating the reliability and assurance construct it is evident that there are two points were the normal brand can improve. Solving problems the first time right and putting the customer interest above their own interest are lagging behind. About 26% of the customers have the perception that the normal brand puts his interests above that of the customers. To improve this, the normal brand should give more attention to serving customers instead of selling products. They should focus on the specific benefits for a customer instead of properties and advantages of the product in order to create more value to the customer. A possible way to achieve this is by setting up customer segment sales teams instead of a more product focused sales department. Another possible solution is provided by internet. Internet offers the opportunity to have a conversation with the consumer. This could be used to give the consumers more a perception that the customers‘ interest is put above the interest of the brand. Furthermore, the first time right percentage of solving problems should be improved. Possibilities here are for example trainings for service employees or empowering the service employees to let them solve more problems their selves.
6.2.3 Brand-related implications
