Word-of-Mouth Preferences

(1)

Word-of-Mouth Preferences

Which factors determine Word-of-Mouth impact?

(2)

Word-of-Mouth Preferences

Which factors determine Word-of-Mouth impact?

Student: Jan Jacob Helder

Student Number: 1533193

Supervisor: Dr. Sonja Gensler

Second supervisor: Dr. Jenny van Doorn

Date: July 12, 2011

Faculty: Economics and Business

Course: MSc. Business Administration

Profiles: Marketing Management & Marketing Research

Prinsesseweg 27a

9717 BB Groningen

0642841198

(3)

Management summary

In this research, four constructs that influence Word-of-Mouth impact are investigated. Goal of the research is to find out which construct has the biggest influence on WOM impact. The report starts with a literature review based on academic sources discussing the constructs and their relevant attributes. Not all of these attributes are measured in this research since some of them are difficult to moderate or influence from a research point of view. Accessibility for example is not an issue since the information will be provided to the respondents and information is easy to find through modern communications. Also homophily (difficult to select similar network environments for every respondent) and argument strength (people will search for the best information available, so argument strength is probably always high when the message is accepted) are left out of the research.

The four constructs, network, sender, message content and message source are given attributes and levels which form scenarios presented to the respondents. The respondents made 12 choice task decisions, choosing between two scenarios at a time. By doing a choice based conjoint analysis, the importance scores for each attribute and the utility scores for each level are determined.

For the aggregated level, sender credibility turned out to be the most important construct influencing WOM impact. Second was sender expertise and the third important attribute was tie strength. Also, segmentation was performed to distinguish differences in preferences between different demographic groups. For young people, tie strength is the absolute most important construct influencing WOM impact. This group basically ignores someone’s credibility or expertise. Older people tend to value credibility and expertise the most. Finally, there was a large segment called deliberates, who tend to value all five constructs fairly even but value sender expertise the most.

(4)

Preface

In front of you lies my final work as a student. After six years of hard work, lots of fun and memorable events at the University and city of Groningen, an era will end with the handing in of this report. Two years after receiving my Bachelor degree in February 2009, I finish my study Msc Business Administration Marketing Management and Marketing Research. I would like to thank some people who made it possible for me to write this thesis and to graduate from university.

First of all, I want to thank my thesis supervisor, Dr. Sonja Gensler for providing me with suggestions, feedback and critical comments throughout the writing of this thesis. And for her patience during the process, which lasted longer than the both of us had initially expected. Without her help, this thesis would never have taken the path it took now. Furthermore, I would like to thank Dr. Jenny van Doorn, my second supervisor. Her feedback in the final stages of the process helped me to finish this report.

In addition, I want to thank my parents for giving me the opportunity to study at the university, their patience, and their faith in me during these six years. I know there were times when they completely had it with me due to bad results and slow progress, but they have always supported me and tried to motivate me. By finally finishing my thesis and graduating with a Master’s Degree, I hope to repay them for their continuous support, patience and trust.

Finally, I want to thank all my friends for the good times at Groningen, but also for their cooperation, advice and help during assignments, exam periods and the writing of our theses. Without the coffee and lunch breaks, social drinks and other activities, these six years would have been a lot harder to get through.

(5)

Chapter 1 - Introduction

1.1 - Background

Despite firms spending millions of dollars on advertising campaigns, the thing that often really makes up a consumer’s mind is free: word-of-mouth (WOM) recommendations from a trusted source (Bughin, Doogan and Vetvik 2010). Several studies have empirically demonstrated that WOM is more persuasive and more effective than traditional media channels (Godes and Mayzlin 2004; Herr et al. 1991; Nail 2005; Trusov, Bucklin and Pauwels 2009). According to recent research, marketers spent more than $1.54 billion on WOM marketing initiatives in 2008, and that amount is expected to rise to $3 billion in 2013 (Kozinets et al. 2010). Due to the great availability of product choices, consumers are overwhelmed by traditional marketing techniques and WOM cuts through the noise quickly and effectively. This is mainly because WOM comes from a sender which is often a strong tie like a friend, a co-worker, a relative or an objective and trustworthy expert on the subject who probably has some knowledge and expertise on the subject to give good and trustworthy arguments.

(7)

WOM might make consumers aware of a product they had not been aware of before, or it might persuade them by changing the expected utility they had of that product (Godes and Mayzlin 2009). According to Bughin, Doogan and Vetvik (2010), WOM accounts for 20 to 50 percent of all purchasing decisions and it has the biggest influence when consumers are buying a product for the first time or when the product is relatively expensive. Lilien and De Bruyn (2008) state that researchers have demonstrated that consumers choices and purchase decisions are influenced by personal conversations and informal exchange of information between peers (people in their network, such as friends, family, acquaintances and co-workers). If the mutual relationship is stronger (i.e. the tie strength is higher), WOM impact on choices and purchase decisions should be higher than when tie strength is weak.

Due to the Internet’s accessibility, reach and transparency, WOM has grown rapidly (Kozinets et al. 2010). One single opinion or review can reach literally thousands of possible consumers with the click of a mouse, and eWOM often occurs between people who do not have a relationship with one another (Dellarocas 2003; Goldsmith and Horowitz 2006; Kozinets et al. 2010). This anonymity allows consumers to share their opinions more comfortably without revealing their identities (Goldsmith & Horowitz, 2006). However, this also leads to credibility issues, with senders, sources or messages potentially being deceptive (Cheung et al. 2009). Also, source A could be credible while source B is transmitting the same message and is perceived unreliable, so message source (not to be mistaken with the sender, these are two different concepts) influences WOM impact as well.

(8)

The goal is to investigate which of these four constructs is the most important influencer of WOM impact.

The network construct is relevant because it is the place where the WOM communications are taking place. Different people interacting form a network with other people, in which the message can be transmitted. The message content is relevant because of the characteristics of the WOM message that are in it. WOM impact is likely to be higher when people communicate clear messages with solid arguments and critical remarks about products or services. Bughin, Doougan and Vetvik (2010) state that message content is the most important construct influencing WOM impact. The sender is the third construct that is relevant for WOM impact, the receiver of the WOM message must trust the sender and the sender should have expertise and credibility. Finally, message source is relevant for WOM impact because it is important to know if the information is coming from someone who knows the product from their own experience or from third parties who just repeat a message they have heard elsewhere.

1.2 - Problem Statement

The main research question that will be investigated in this research is:

“The network, sender, message content and message source are supposed to affect WOM impact. Which of these factors influences WOM impact the most?”

This main research question can be divided into the following questions: 1) What is the influence of the network on WOM impact?

2) What is the influence of the sender on WOM impact?

(9)

1.3 - Academic and practical relevance

Bughin, Doogan and Vetvik (2010) introduced the four constructs influencing WOM impact in their model to calculate WOM equity. This research will use that model as a starting point and will investigate which of these four constructs influences WOM impact the most. In the literature, there are studies which focus on the credibility of messages or the influence of networks regarding WOM. Consumer discussions have great potential to significantly influence readers/listeners/spectators who intend to use these recommendations for purchase decisions. These recommendations could shape readers’ attitudes toward a product, indirectly influencing the sales of a product. Most previous studies focus on the benefits of WOM (Cheema and Kaikati 2010). This research will hopefully help to gain insights in the effects of the different constructs (sender, message source, message content and network) on WOM impact. In this way, this research hopes to add some new theory and knowledge to the existing theory.

For managers, it will be interesting to know which constructs influence WOM impact the most. It also will be interesting to see if different segments show heterogeneous results between them regarding the importance of the different drivers. Managers could then use this information to target their target group with the right approach instead of using a general strategy for all segments.

1.4 - Structure of the thesis

(10)

Chapter 2 - Literature review

In this chapter, an overview of the literature on WOM and the four drivers of WOM impact will be given. In section 2.1, a short overview of general WOM literature is discussed followed by a review of the network construct in section 2.2. In sections 2.3, 2.4 and 2.5, the sender construct, the message content construct and the message source construct are discussed. Hypotheses will be formed based on the literature found and restrictions and boundaries relevant for this research. In section 2.6, the conceptual framework is displayed.

2.1 - General WOM literature

As already mentioned in the introduction, WOM is increasingly becoming important and is a very relevant subject nowadays. WOM has a strong influence on product and service perceptions, leading to changes in judgments, value ratings and the likelihood of purchase (Arndt, 1967). Although WOM is typically associated with face-to-face communication, it can be provided through all sorts of media such as telephone and Internet (Dellarocas 2003; Herr et al. 1991) Furthermore, since WOM is taking place independently from the organization, all social actors can provide WOM information, including family, friends, co-workers or even complete strangers (Brown and Reingen 1987). WOM can be initiated by two sides. For example, if someone is happy or unhappy about a product or service, they may want to inform others about their happiness or unhappiness regarding the advantages or disadvantages of a product or service. On the other hand, if someone has to make a purchase decision for a product or service and this person wants to be informed as good as possible then they might ask people they perceive to possess this information to share this information. Also, since WOM is company independent and is not only about promoting the product, it can contain both positive and negative comments (Bone 1995).

(11)

Misner, 1999; Nail 2005; Kozinets et al. 2010). In his research, Nail (2005) reports that consumer attitudes towards advertising dropped. He states that fewer people agree that advertisements are a good way to learn about new products and also fewer people buy their products because of advertisements. Finally, the number of people who find advertisements funny has also decreased heavily according to Nail (2005).

The nature of WOM was long described as fleeting because, due to its spontaneous and face-to-face manner, it used to vanish as soon as it was uttered. With the help of the Internet, WOM no longer vanishes instantly and does not have to be spontaneous (Breazeale 2008). It does not vanish instantly because everything can be searched for on the Internet and old opinions or reviews are saved indefinitely on the World Wide Web. Also, because of specific sites like consumer reviews forums, the actual WOM does not have to be spontaneous. Often, these sites plan ahead which products to review in a specific timeframe, like a week or a month. Breazeale (2008), states that the Internet plays an important role in WOM nowadays. Research has suggested that WOM such as consumer reviews (both online and offline) generate greater empathy, credibility and relevance than information generated by the organization itself (Bickart and Schindler 2001).

Due to the Internet, WOM is not limited in its domain. One single opinion or review can reach literally thousands of potential customers with the click of a mouse. No surprise that marketers have become interested in directly managing WOM activity (Kozinets et al. 2010). In this research however, WOM is defined as communications outside the influence of the organization and thus takes place between consumers only. Thanks to the Internet, personal experiences and stories are communicated among masses of people. According to Zhu and Zhang (2010), a major reason for consumers to use WOM is to obtain quality information in order to reduce risk. Studies have shown strong relationships between a product’s popularity and its perceived quality.

(12)

product quality is also high (Chen, Wu and Yoon 2004; DeSarbo et al. 2002; Zhu and Zhang 2010). However, Chatterjee (2001) does not agree and states that consumers interested in a less popular product are likely to search and access more WOM information to shield themselves from possible regret. Regret is higher if consumers buy a less popular product (for example from a lesser-known brand) that turns out to be inferior than if they choose a well-known brand that turns out not to be better than the lesser-known option (Simonson 1992).

In the following paragraphs, the four constructs that are investigated in this research will be discussed. As already stated in the introduction, the four constructs network, sender, message source and message sender are chosen based on the model used by Bughin, Doogan and Vetvik (2010) to calculate WOM equity. This model is displayed in figure 1 below. In this model, WOM impact is a variable that helps calculating WOM equity.

(13)

2.2 - Influence of the network on WOM impact

The network is the first construct discussed in this literature review taken from the model by Bughin, Doogan and Vetvik (2010). A network is a series of points or nodes which are interconnected by communication paths.1 Networks can interconnect with other networks and can contain sub networks. In WOM context, a network would be a group of people who are connected with each other through mutual interests and bonds. The network is the environment in which the WOM is taking place, using the communication paths between the nodes to be transmitted. For eWOM the network is much larger, compared to normal WOM. More contributors and audiences are involved and the reach extends to the Internet world instead of small-scale interpersonal connections (Dellarocas 2003; Goldsmith and Horowitz 2006; Kozinets et al. 2010; Cheung et al. 2009). Some examples of networks are social networks, professional networks and distribution networks. For WOM, social networks are the most relevant. People prefer to turn to other people rather than documents for information (Pelz and Andrews 1966; Mintzberg 1973). The position of the nodes is influencing an individual’s degree of receiving and transmitting information. Someone who has a lot of communication paths with other nodes would claim a more central role in the network than someone who does not have many connections allowing the former to access more information and have a bigger reach when transmitting.

For WOM to operate within a consumer network it appears to be influenced by tie strength (Granovetter 1973; Bansal and Voyer, 2000) Granovetter did not come up with a precise definition for tie strength. But in this research, tie strength is defined as the intensity of an interpersonal relationship between two human beings. Friends, family, and co-workers who a person sees and talks with often are the so called strong ties. Vague acquaintances, e-mail contacts and casual contacts often have a lower intensity of interpersonal contact with a person, and thus are called weak ties. Strong ties play an important role on the micro-level (Brown and Reingen 1987; Granovetter 1983).

Brown and Reingen (1987) found out that if someone is in a situation with both strong and weak ties available for information, the probability that the strong tie is preferred is

1

(14)

higher. They also found support that information coming from strong ties is more likely to influence receiver decision making than information coming from weak ties.

However, despite being weak ties, these people play an important role in sending information, especially on the macro-level (between networks) according to Brown and Reingen (1987) and Granovetter (1983). Weak ties bridge the gaps between networks (Granovetter 1983). Rogers (1995) states that weak ties play an important role in the transmitting of WOM regarding innovations and new products. Through these weak ties, new networks learn about the innovations and the product or service becomes more known to consumers.

Strong ties form networks and weak ties form a bridge to another network (Granovetter 1973). This is illustrated in figure 2; the person in the center has a group of people surrounding him, which are defined as strong ties. These strong ties have connections with other networks. For the original person, transmitting the WOM, these

interconnections with other networks are (probably) weak ties. However, through these connections, the WOM message is transmitted into other networks and results in a greater increase in WOM conversations. This effect, the passing on of WOM by one customer to another, is known as a ripple effect (Gremler and Brown 1999). Strong ties are likely to exchange information of higher economic value with each other than weak ties (Frenzen and Nakamoto 1993). Goldenberg et al. (2001) investigated the relative macro level impact of strong and weak ties and found that the latter may have a bigger impact even though the former are activated more frequently.

The second network attribute influencing WOM impact is homophily. Homophily is the tendency of individuals to associate and bond with similar others. The presence of

(15)

homophily has been discovered in a vast array of network studies.2 Gilly et al. (1998) mention that the similarity or dissimilarity of the consumers in terms of backgrounds, opinions, likes and dislikes will influence the WOM impact. Most human interaction will occur between demographically similar consumers (Lazarsfeld and Merton 1954; Brown and Reingen 1987; Lilien and de Bruyn 2008). Brown and Reingen (1987) stated that if the homophily between sender and receiver is higher, the likelihood of activation for information is higher. Empirical evidence suggests that consumers are likely to talk to similar sources and the influences of these homophilous (similar) sources are sometimes greater than that of expert sources. Also, similar consumers might have similar product needs compared to heterophilous (dissimilar) individuals, resulting in the most personally relevant product information (Brown and Reingen 1987). Brown and Reingen (1987) also suggest that homophilous sources of information will be perceived as more credible than heterophilous sources, which results in greater influence on WOM impact.

Accessibility is the probability that someone receives WOM messages and it is a key determinant of WOM impact on judgment and decision making (Feldman and Lynch, 1988). In this research accessibility is defined as the ease of contact with another person to receive WOM information.3 For example, if someone would like to buy a new car and a close friend works in a garage, the accessibility of WOM information for that person about cars would be high because he can easily contact his friend. The accessibility is related with the tie strength construct. If someone has a lot of strong ties, it would probably be easier for this person to gather the necessary information. If someone has only a couple of strong ties, there is a possibility that the necessary information is not available in this person's network. With help of the Internet, consumers can now easily publish their opinions, thoughts, feelings and viewpoints on products and services to the public at large (Andrew and Toubia 2010). Accessibility is important in an offline, as well as an online environment. Offline, the seeker of information will get that information from a source/peer as long as this peer is accessible. Being accessible also depends on the centrality of the source. If someone has a high degree of centrality, the

2

http://www.encyclo.co.uk/define/Homophily

3

(16)

number of linkages is higher which could mean that information is more accessible and WOM is more likely to reach the individual.

For this research, the homophily and accessibility constructs are difficult to manipulate, change or moderate from a researcher point of view. Since the information will be supplied to the respondent, accessibility is not an issue in this research. Furthermore, on the Internet, the necessary information is always only a few mouse clicks away and thus accessibility is not really relevant for WOM research nowadays. People in general know how to find the necessary information when they make decisions. Also, homophily is difficult to manipulate since it would be a very hard job to select two groups of respondents in which the members of one group are identical to each other based on backgrounds, opinions, likes and dislikes and those of the other group are dissimilar in order to test the influence of this construct. Also, the respondents do not have the opportunity to select the sender or source of the information they receive themselves in this research, which is mainly the idea regarding homophily, since it states that similar people will exchange higher volumes of information with each other. Therefore, the main focus will be on the tie strength construct, because this construct can be manipulated in order to see how respondents will react to information coming from different tie strength levels. Literature seems to agree that people with strong ties will interact more and share more valuable information which has a higher influence on the receiver decisions. Weak ties will be used to transmit this information between networks. Strong ties will have more influence on the micro level whereas weak ties play an important role on the macro level. In this research, the focus is on the micro level, so based on this, hypotheses 1 will be:

• H1: If tie strength is higher, its relative importance on WOM impact will be

higher

(17)

2.3 - The influence of the sender construct on WOM impact

The sender is the second construct taken from the model by Bughin, Doogan and Vetvik (2010) influencing WOM impact. In WOM context, message sender is defined as an interpersonal source which, unlike an impersonal source (like advertisements), shares information independently from the company. The sender is the person who transmits the WOM information to the recipient. Bansal and Voyer (2000) and Lau and Ng (2001), state that the characteristics of the sender are one of the influences of WOM impact. Lilien and de Bruyn (2008) give an example of how sender influence works. They state that when someone receives an e-mail and considers opening it before knowing the content of the message, the decision to open and read it cannot relate to message content (which is at that point unknown) but only the known characteristics of the sender like how familiar the sender is to the receiver, if he/she is perceived credible towards the receiver and if the receiver even likes the sender.

Furthermore, the receiver of the WOM must trust the sender and believe that he or she really knows the product or service in question (Bughin, Doogan and Vetvik 2010). The WOM sender is likely to be more relied on than other sources in consumer decision making because of its assumed objectivity (Gilly et al. 1998). There are a couple of characteristics that contribute to the sender's influence on WOM impact, namely sender expertise and sender credibility (Bansal & Voyer 2000; Gilly et al. 1998; Dean, Austin and Watts 1971; Harmon and Corney 1982, Cheung et al. 2009).

The first attribute is sender expertise. Sender expertise is defined as the degree of knowledge and experience the sender possesses with respect to the product, service or organization (Bone. 1995). Previous research has identified the perceived expertise of the sender as a key determinant for WOM impact (Bansal & Voyer 2000; Gilly et al. 1998).

(18)

organization (Gilly et al. 1998). Higher source expertise will lead to higher amounts of WOM generated (Bansal and Voyer 2000; Gilly et al. 1998).

Whether or not a sender is perceived as having high expertise is determined from an evaluation of the knowledge that person holds but also by his or her occupation, social training or experience (Gotlieb and Sarel 1991). However, in the online environment, such evaluations must be made from the relatively impersonal text-based resource exchange provided by actors in the site network. Knowledge of the individual's background and attributes is limited and evaluation will take place on a reduced number of characteristics (Brown, Broderick and Lee, 2007). In online environments, people tend to evaluate sender expertise based on online context and especially more on characteristics of the website (like the website's authority and if the info on the website was useful the last time the receiver used it) than the actual sender characteristics. Furthermore the website’s specific language (like: you know your stuff [sender], thanks for the info!) is an important cue to determine sender expertise (Brown, Broderick and Lee 2007).

(19)

In an online environment, credibility is always a major concern for WOM receivers. Offline, friends and family are generally trustworthy and reliable but online, receivers are not always able to evaluate WOM information in the way they would like (Cheung et al. 2009). Since electronic WOM (eWOM) arises from a possibly unlimited number of unknown participants, the high amount of unevaluated information makes the credibility of the information source uncertain. Frambach, Roest and Krishnan (2007), investigated online credibility for the banking sector and concluded that due to the unclear credibility of online environments, tie strengths of these sources are typically weak and when both online and offline channels are present, the offline channel is preferred over the online channel for all three stages of the buying process for mortgages. So, credibility is a very important construct in the banking sector. This could apply to other sectors as well.

Both sender expertise and sender credibility can be manipulated in this research. Sender expertise is a key determinant according to the literature and people with high levels of expertise are likely to create and transmit more WOM than people who do not have this expertise. Also, sender credibility is important, since a high credible source is more persuasive and more likely to be believed by the receiver according to the literature. Based on these findings, the following hypotheses are created to investigate if the academic findings hold for this research as well:

• H2: If sender expertise is higher, its relative importance on WOM impact will be

higher

• H3: If sender credibility is higher, its relative importance on WOM impact will be

higher

(20)

2.4 - Influence of the message content on WOM impact

Message content is the third construct taken from the model by Bughin, Doogan and Vetvik (2010) and discussed in this literature review. They stated that message content, or what is being said, is the most important factor influencing WOM impact. The message content must address important product or service features if it is to influence consumer decisions (Bughin, Doogan and Vetvik 2010).

The first attribute for message content is the strength and the persuasiveness of the arguments in the message. Argument strength is concerned with the quality of the perceived information (Cheung et al. 2009). Numerous studies have shown that argument strength will affect the attitude of the receiver (Cacioppo, Petty and Morris 1983; Wathen and Burkell 2002). It is the extent to which the receiver perceives the message to be valid and convincing in supporting its position. The WOM impact will be higher and a positive attitude will be developed when the message content is perceived to have good arguments (Cheung et al. 2009). Argument strength is especially important in an online environment. Since the evaluation of the credibility of the source is often difficult in an online environment, consumers judge the credibility of online recommendations based on the argument strength (Cheung et al. 2009).

(21)

Chevalier and Mayzlin (2006) found out that, on average, reviews tend to be positive. This suggests that people are more willing to share good experiences than bad ones. Chevalier and Mayzlin (2006) also add that the addition of new, favorable reviews results in an increase in sales. They also state that an incremental negative review is more powerful in decreasing sales than an incremental positive review is in increasing sales.

Since, especially in online environments, credibility is an issue as the source and the network are not easy to evaluate or are unknown, efforts on credibility through the message content is an important aspect which can easily deceive consumers. Dellarocas (2003) mentioned that many firms are taking advantage of online consumer reviews as a new marketing tool. Firms do not only regularly post their product information and sponsor promotional chats on online forums, such as Usenet (Mayzlin 2006), but also proactively induce their consumers to spread the word about their products online (Godes and Mayzlin 2004). Some firms even strategically manipulate online reviews in an effort to influence consumers’ purchase decisions (Dellarocas 2006). However, this is firm initiated WOM and is not a part of this research. It is shortly discussed to show how vulnerable online recommendations can be regarding credibility and valence.

A few negative messages can be helpful in creating a positive attitude towards the product or service. A single negative message itself can be harmful for product evaluations and it is expected that if all messages are negative, the influence may be larger than if all the messages are positive (Doh and Hwang 2009). However, one negative message in a 10-message set is not very harmful and can even be beneficial in WOM context because consumers may perceive the credibility to be higher if there are also negative comments to be found (Doh and Hwang 2009).

(22)

Two-sided messages have better communication effects and are more persuasive than one-sided messages (Wilkie and Farris 1975; Wilson 1976; Smith and Hunt 1978 and Syinyard 1981).

Summarizing, there is academic literature that states that message content is an important influence on WOM impact. Argument strength can influence consumers’ decisions, and valence and sidedness increase the credibility of the messages, which leads to higher WOM impact. A few negative comments or messages can make a consumer more informed. With knowledge about downsides of the product or service, the consumer might feel more prepared when making the purchase decisions. However, argument strength and valence are difficult to manipulate in this research. Therefore, the messages will be the same for each respondent regarding the arguments and the nature of the WOM. The sidedness attribute can be manipulated and thus is included. Respondents will evaluate situations with only positive remarks about the product or situations with both positive and negative remarks. Based on this, the following hypothesis can be created:

• H4: Two-sided messages have a higher relative importance for WOM impact then

one-sided messages

Like the other constructs, again the relationship between sidedness and WOM impact is expected to be positive, with an increase in the levels of the attribute resulting in an increase in WOM impact.

2.5 - Influence of the message source on WOM impact

(23)

reviews in order to create positive WOM (Chatterjee 2001; Dellarocas 2003). Because it is difficult to determine the quality and credibility of the eWOM based on characteristics of the communicators, consumers tend to use other cues to evaluate the eWOM (Greer 2003) like the platform to which WOM is posted (Senecal and Nantel 2004; Schindler and Bickart 2001). Examples of platforms are brand websites, comparison sites, personal blogs and review websites but also sales persons in shops.

When WOM is posted on a brand’s website, the consumer may perceive the reviewer being influenced by the brand (Fei and Phelps 2004). Fei and Phelps (2004) also found out that independent product review websites are generally known to be credible and free of marketer influence since the goals of these websites are to help consumers make informed buying decisions by providing a platform to share their product experiences (Lee and Youn 2009). Finally, Fei and Phelps (2004) stated that eWOM that is posted on an independent product review website may be more likely to reflect the reviewer’s feelings about the product or service instead of its actual performance. However, this will make these reviews more persuasive than recommendations posted on a website of the brand itself.

Lee and Youn (2009) carried out research investigating the impact of different environments. Their findings indicated that, other things being equal, there were differences between the reactions of consumers to messages on personal blogs compared to brand’s websites. Consumers exposed to the review posted on the personal blog were more likely to attribute the review to circumstances and less likely to recommend the product to friends than those who were exposed to the review either on the independent review website or the brand’s website (Lee and Youn 2009). Bickart and Schindler (2001) also did research and found out that online forums were distinctly more effective in generating product interest than corporate websites. Hypothesis 5 will be:

• H5: If source independence is higher, its relative importance on WOM impact

(24)

Also source independence is expected to have a positive relationship with WOM impact. An increase in independence will lead to higher WOM impact.

2.5 - Theoretical framework and general hypotheses

Based on the hypotheses mentioned in the paragraphs above, the following conceptual framework is created which forms the starting point of the empirical part of this research.

Figure 3 - Conceptual framework

All five attributes which can be manipulated are expected to have a positive relationship with WOM impact. There are also some general hypotheses which can help answering the research question.

• H6:Sender credibility is expected to have the highest importance regarding WOM

impact

(25)

• H7: Sidedness is expected to have the lowest importance regarding WOM impact

Sidedness is an attribute of message content which has hardly been discussed in the literature yet. It seems that sidedness has a lot of influence on factors like message credibility and awareness of the receiver. Obviously, these factors influence WOM impact, but from a WOM receiver point of view, the importance of sidedness is probably the lowest out of the five since consumers are probably mostly interested in products which have positive characteristics in order to avoid risk and make the right purchase decision.

(26)

Chapter 3 - Research Design

In this chapter, the research design of the study will be discussed. A research design is a framework or blueprint for conducting the research (Malhotra 2007). In this chapter, the necessary procedures for obtaining the information needed for the research are discussed. This chapter deals with two items, namely how the data is collected and how the data is analyzed. In section 3.1, the data collection methods are discussed followed by the study design in section 3.2. Finally, in section 3.3, the plan of analysis is explained.

3.1 - Data collection

In order to collect the data necessary for doing the analysis, a survey is used. The survey method of obtaining information is based on the questioning of respondents. Respondents are asked a variety of questions regarding their behaviour, intentions, attitudes, awareness, motivations and demographic and lifestyle characteristics (Malhotra 2007). The respondents are not selected by the researcher; instead the survey is communicated through social media like Facebook and Hyves in order to reach many respondents with little effort. This method is known as the snowballing4 method, which relies on referrals from initial respondents to generate additional respondents.

The survey for this research will be structured, which means that the questionnaire is prepared and the questions are asked in prearranged order (Malhotra 2007). A survey data collection method is the best option for this research because it is an efficient way to collect information from a large enough number of respondents. A lot of people can be reached in little time and with little effort. Also, a survey is useful for collecting the necessary wide range of information and this research covers a number of different constructs and dimensions. Furthermore, standardized questions make measurement more precise by enforcing uniform definitions upon the participants and the reliability is usually high because all respondents answer the same questions and subjectivity is eliminated. The constructs and dimensions from the theoretical framework will form the basis of the survey questions.

(27)

There are several ways of performing a survey, for example by personal interview, over the phone, written, via mail and via the Internet. An online Internet survey will be used in this research in order to collect the necessary data. The online survey is favored as this method makes it possible to collect large amounts of data in a relatively short period of time and because of the fact that participants feel relatively anonymous when filling in an online survey (Cooper & Schindler 2003).

The target group for this survey can be characterized as people older than 12 years. Dependence techniques will be used to analyze the data, this may be defined as one in which a variable or set of variables is identified as the dependent variable to be predicted or explained by other variables known as independent variables (Hair et al. 2006). In this research, WOM impact is the dependent variable and the constructs and dimensions are the independent variables influencing it.

3.2 - Study design

In order to determine which construct and dimensions are the most important influences on WOM impact, a Choice Based Conjoint analysis will be performed. A conjoint analysis attempts to determine the relative importance consumers attach to salient attributes and the utilities they attach to the levels of attributes (Malhotra 2007).

This data is gathered by presenting the respondents two scenarios at a time. These scenarios consist of attributes and levels related to the constructs. The respondents are asked to evaluate these scenarios and select the one they prefer. By doing the research this way, the scenarios are evaluated as a bundle of attributes. The different attributes are given levels in order to create the scenarios. These levels are displayed in table 1.

Table 1 - Overview of attributes and levels

(28)

because it is a relatively new product and people are not yet completely familiar with it to make it a low involvement purchase decision. This makes it a high influence purchase decision which is not made in a split second. The story that will be used is shown below:

By changing the text that is displayed in bold, the different levels will be included into the different profiles. The situation stays the same throughout the research, only the characteristics will change. In order to create the different profiles, Sawtooth is used. This software transforms the five different attributes and their levels into 24 different profiles. In this research, the respondent will repeatedly have to choose one option from a choice set of two options presented to him/her, which is called a pairwise comparison (Hair 2005). Each stimulus will be a scenario including the attributes and their levels mentioned above. The respondent is asked to indicate which profile has his/her preference. The five factors above and their respective levels will be transformed into 24 different profiles. These 24 profiles are divided into twelve groups of two and the respondent is asked to choose the favorite of the two options for each group, leading to twelve preference decisions. The 24 profiles can be found in the questionnaire, which is in Appendix 1.

A [Tie strength] tells you that he read [Source independence] and states that the

new smartphone from Samsung would be a good choice. The sender has [Sender

Expertise] himself. The sender repeats what he [source] and mentions [Sidedness]

(29)

3.3 - Plan of analysis

The first step in a survey research is to check the data and prepare it for analysis. Important is to check all surveys for completeness and quality. The main purpose is to identify surveys filled in incompletely and delete them to make sure the data does not have missing values leading to biased results.

(30)

Chapter 4 - Results

In this chapter, the results of the explained research in chapter three will be discussed. The dataset is explored with the help of descriptive statistics followed by a conjoint analysis of the data. In this conjoint analysis, also segmentation is performed to distinguish clusters and differences in preferences between these clusters. The data will be used to determine if the hypotheses created in the literature review are supported or not.

4.1 - Descriptive statistics

In this section, descriptive statistics are used to get familiar with the database and the characteristics of the respondents. The dataset exists of 205 respondents and these respondents will be described by using the demographics gender, age and education.

Of the 205 valid respondents in this dataset, the gender is surprisingly well distributed. In total, 105 male and 100 female respondents completely filled in the survey. This corresponds with 51.2% male respondents and 48.8% female respondents.

The average age of the respondents is 31.10 years old. The oldest respondent was 75 years old, and the youngest respondent was 13 years old. The median age is 25 years. As can be seen in figure 4the age classes 10-20 (22.4%) and 20-30 years (40.5%) are accounting for a large part of the respondents in this research. The smallest age class based on the respondents in this research is 71-80 years old (0.5%) followed by 61-70 years old (3.9%) and 51-60 years old (8.8%).

In the survey, a question was included to gather data about the highest level of education that was attended by the respondents. The highest percentage was for University with

(31)

32.2% followed by HBO (22.4%) and VWO (18.0%). In figure 5, the scores per education type are shown.

4.2 - Choice-based conjoint analysis

In this section, the results of the conjoint analysis performed on the data will be dis-cussed. Conjoint analysis is a statistical technique used to determine how people value different features that make up a specific product or service (Hair et al. 2001). Conjoint analysis in this research is used to determine the importance of the attributes in the scenarios discussed in the literature and the research design. First, the goodness-of-fit statistics are discussed followed by the importance of the attributes. In this section, the hypotheses will also be answered. Thirdly, segmentation analysis is performed and discussed for the respondents in the dataset. These segments and their different preferences will be discussed next. Finally, the segments will be named using the outcomes of the analysis and their preferences.

Respondents education 0 5 10 15 20 25 30 35

University HBO VWO HAVO MBO other

Education type P e rc e n ta g e

(32)

4.2.1 - Goodness-of-fit

In the table below, some goodness-of-fit variables are displayed regarding the dataset. To determine the goodness of fit, the R2 variable and the BIC and CAIC statistics are being taken into account. In general, the higher the R2 variable, the better it is for the model. For the BIC and CAIC statistics, a lower value of these statistics shows the better comparison of the model.5 As can be seen in table 2, a three segment solution would be right for this research, based on the BIC and CAIC scores. Also, the R2 variable for a three segment solution is three times larger than for the aggregated level (all respondents in one segment). This means that more variance is explained in a three cluster solution compared to the aggregated level.

Table 2 - Goodness of fit statistics

4.2.2 - Parameter estimates

In table 3, an overview of the utility scores per segment and the aggregated level are given. The first notable outcome is that all segments value word-of-mouth from a good friend above word-of-mouth from a vague acquaintance. For segment 3, this attribute is the most important influence on WOM impact.

Table 3 - Utility scores

(33)

The preference for the attribute sidedness is also equal across all three segments with each preferring both positive and negative characteristics over only positive characteristics.

Regarding the hypotheses created in chapter two, these statistics can help to determine whether the hypotheses are supported or not. For each segment, the utility score for the highest tie strength level ‘good friend’ is higher than the utility score for ‘vague acquaintance’. This means that Hypothesis 1: If tie strength is higher, its relative

importance on WOM impact will be higher is supported. The respondents have chosen

more scenarios and thus accepted the WOM information in which tie strength was higher, than in which tie strength was low the and thus it can be concluded that higher tie strength results in higher WOM impact.

Hypothesis 2: If sender expertise is higher, its relative importance on WOM impact will

be higher is partially supported by the findings in this research. Only segment 3 seems to

value a lower level of sender expertise when accepting WOM information. The other two segments give the highest utility scores to the highest expertise level and thus will WOM impact for these respondents be larger when the sender expertise is higher.

The same holds for hypothesis 3: If sender credibility is higher, its relative importance on

WOM impact will be higher which is also partially supported. For segment 3, sender

credibility does not seem to matter much since the highest utility score is for the lowest level. For segments 1 and 2 however, the hypothesis is supported since the highest level of sender credibility receives the highest utility scores and thus will WOM impact for these respondents be bigger when the sender credibility is higher.

Hypothesis 4: Two-sided messages have a higher relative importance for WOM impact

then one-sided messages, is supported. For each segment and the aggregated level, the

(34)

Finally, hypothesis 5: If source independence is higher, its relative importance on WOM

impact will be higher is partially supported. Segment 2 and the aggregated level value the

highest level with the highest utility score and these respondents are thus accepting information from a higher independence source sooner, leading to bigger WOM impact. However, segments 1 and 3 do not value the highest level with the highest utility score, leading to the hypothesis only being partially accepted.

4.2.3 - Importance of attributes

To determine which attribute in the model is the most important in the eyes of the respondents, LatentGold offers importance statistics. The statistics in table 4 stand for percentages of importance. The most important factor is the one with the highest percentage and the least important factor is the one with the lowest percentage. As can be seen in table 4, for the aggregated level, sender credibility has the highest score followed by sender expertise and tie strength. Sidedness is not considered very important by the respondents and

source

independence is the construct

which is

considered the least important by the respondents in the dataset. For cluster 1, the most important attribute is sender expertise with 31.1%. Similar to the aggregated model (without any segmentation), the most important attribute for cluster 2 is sender credibility with 41.9%. Finally, for cluster 3, tie strength is the most important attribute with 57.4%.

Regarding the last two hypotheses, these statistics can help to determine whether these hypotheses are supported or not.

Hypothesis 6: Sender credibility is expected to have the highest importance regarding

WOM impact is partially supported. For the aggregated level and segment 2, sender

credibility is indeed the attribute with the highest relative importance. Respondents in segments 1 and 3 however, value other attributes with higher relative importance scores.

(35)

Hypothesis 7: Sidedness is expected to have the lowest importance regarding WOM

impact is only supported for segment 3. For segments 1, 2 and the aggregated level,

sidedness is more important than expected.

4.2.4 - Segment size

In a three-cluster solution, the class sizes are 63.9% for cluster 1, 26.34% for cluster 2 and 9.75% for cluster 3. So cluster 1 is by far the largest cluster with 131 people in it. Cluster 2 has 54 people in it, and cluster 3 consists of the remaining 20 people. Such a distribution with one very large segment is far from ideal, which forms a limitation in this research.

4.2.5 - Demographic differences per segment

Segment 1 (131 people)

The largest segment that the cluster analysis forms is segment 1. In this segment, 131 out of 205 respondents are placed. The average age of the 131 respondents in segment 1 is 31.02 years with the oldest respondent being 75 and the youngest 13. The average ageclass is 2.55, which means that 58% of the respondents in this segment is younger than 30, and 11.5% is older than 40. The segment is quite spread out in relation to the different ages of the respondents. In the segment, 51.9% of the respondents are male, 48.1% female. The average education attended is 2.69, which means that the majority of the respondents attended University or HBO (51.1%). 31.3 % attended VWO or HAVO, only 13.7% attended MBO and 3.8% had a different education than the five answer options.

(36)

attended is 2.52, which means that 63% attended university or HBO, 20.4% attended VWO or HAVO, 7.4% attended MBO and the remaining 9.3% attended a different education than the five answer options.

The final segment, number three is a relatively small segment which consists of 20 out of 205 respondents. The average age is 22.8 years with the oldest respondent being 50 years old and the youngest 16 years old. The average age class is 1.65, which means that 95% of the respondents in this segment are younger than 30, with only 5% being older than 30. In segment 3, 50% of the respondents are female and 50% are male. The average education attended is 2.35. University or HBO is attended by 55% of the respondents, VWO or HAVO by 40% and MBO is attended by 5%. The other option was not selected by people in this segment.

4.2.6 - Hit rates

By calculating hit rates, an analysis can be done to find out if the aggregated model predicts outcomes better than the segmented model. The hit rate is a measurement for the quality of the model that compares the predicted values with the actual choices made by the respondents (Hair, 2005). The hit rate is calculated by matching the observed choices with the predicted choices. There is a hit if the observed choice matches the predicted choice; otherwise, there is a miss. LatentGold gives statistics for the hit rates.

For the aggregated model, it predicts 1581 (765+816) out of 2460 outcomes correctly, leading to a hit rate of 64.27%. In the segmented level, 1770 outcomes (805+965) out of 2460 are predicted correctly leading to a hit

rate of 71.95% which is a 7.68% increase.

So, the segmented solution is better than the aggregated solution.

(37)

4.3 - Segments discussion

In this section, the three segments will be discussed based upon the findings in the research. Both their demographic values as their preferences and valuations for the four constructs will be used to come up with a suitable description for each segment

Segment 1

The respondents in segment 1 do not really have an outspoken preference for any of the attributes compared to the other segments. According to the performed analysis they value sender expertise the most (31.1%) followed by sender credibility (20.47%) and sidedness (18.96%). As mentioned before, this segment exists out of 63.9% of the total respondents. 51.1% of the respondents in this segment attended University or HBO which is lower than the two other segments. These respondents value sender expertise the most, followed by sender credibility. The respondents in this segment will be called the 'deliberates'. Their preference for expertise and credibility shows that they value information coming from a reliable and knowledgeable sender. Since they also value the other three attributes relatively highly, they tend to make careful decisions based on as much information as possible, while taking all attributes into account.

Segment 2

(38)

Segment 3

(39)

Chapter 5 - Conclusion and discussion

The main goal of this research is to investigate which construct has the biggest influence on Word-of-Mouth impact. In order to do this, this research consists of four parts; a literature review, the identification of four types of constructs that influence WOM impact, the analysis of these constructs and their importance and the segmentation of the respondents and their preferences. WOM is a relatively new subject in marketing, which has been getting increasingly more attention from researchers over the past years. However, the concept of WOM itself was already investigated in the 1950s. Although studies have investigated WOM, most tend to focus on the benefits of WOM but not on the constructs that influence WOM impact. In the literature review, an overview of the relevant constructs and their attributes influencing WOM impact are discussed. However, not all of these attributes were included in the research, since some were hard to influence, manipulate or change from a research point of view to get the necessary data. Data for this research was collected through an online questionnaire filled in by 205 respondents. The survey asked the respondent to make a choice based on two different scenarios and repeat this for eleven other choice sets. This data was analyzed by doing a Choice Based Conjoint Analysis. This chapter of the report summarizes and discusses the findings in relation to literature. Furthermore, the managerial implications of this thesis’ results will be discussed, together with the limitations and directions for further research.

5.1 - What is the influence of the network construct on WOM impact?

(40)

and vague acquaintance is negative for the aggregated level as well as the three different segments. Hypothesis 1 was supported since higher tie strength led to higher WOM impact. The higher utility score for good friend compared to vague acquaintance supports this finding. It can be concluded that respondents in this research value tie strength as an important factor, but it is not the most important construct on WOM impact.

5.2 - What is the influence of the sender construct on WOM impact?

For measuring the influence of the sender construct on WOM impact, two sender attributes were included in the scenarios: sender expertise and sender credibility. For the aggregated level, sender credibility is the most preferred attribute and sender expertise is the second most preferred attribute. Between the segments, differences are present. For segments 1 and 2, sender expertise is the most preferred attribute and sender credibility is second most preferred. In segment 3, sender expertise is the second most important attribute behind tie strength. Sender credibility is the least preferred attribute for this segment. The sender expertise attribute consists of three levels. For the aggregated level and segment 1 and 2, the highest level (knows a lot about smartphones) of sender expertise also gets the highest utility score. Only for segment 3, the utility score is highest for the second level (knows a little about smartphones). The respondents in this segment value their friends’ opinions the most and do not seem to care much if the sender has the right knowledge to share information with them in order to accept WOM messages from these people.

(41)

For sender credibility, two levels are created in the scenarios. Like sender expertise, a similar trend can be discovered. For the aggregated level and segments 1 and 2, the highest level of credibility gets the highest utility score. For segment 3, the lowest level gets the highest utility score. This is similar to the sender expertise attribute and likely to be because the respondents in this segment give the highest value to messages coming from their strong ties, and do not seem to care if these people have been credible in the past when telling them about products. It can be concluded that higher sender credibility will lead to higher utility scores and thus increase the probability that the WOM impact will be bigger and the information received is accepted by the respondent.

So, hypothesis three is supported by the findings in this research. This is in line with the literature found since people tend to take advice from senders they perceive being credible and trustworthy (Dean, Austin and Watts 1971; Harmon and Corney 1982; Cheung et al. 2009). Finally, hypothesis 6 predicted that sender credibility would be the most important influence on WOM impact. However, this hypothesis is only partially supported. Only for the aggregated level and segment 2, sender credibility had the highest importance score.

5.3 - What is the influence of the message content construct on WOM impact?

The message content construct is measured by the sidedness attribute in the scenarios. Sidedness means that either only positive characteristics are mentioned in the message or negative characteristics are mentioned as well. For the aggregated level, sidedness is the fourth preferred attribute. For segment 1, sidedness is third important and the attribute fourth important for respondents in segment 2. Finally, for segment 3, sidedness is the least important attribute. The sidedness attribute consists of two levels (both positive and negative characteristics, and only positive characteristics).

(42)

section 2.4 that higher sidedness will lead to more persuasive WOM messages (Wilkie and Farris 1975; Wilson 1976; Smith and Hunt 1978 and Syinyard 1981). Hypothesis 7 predicted that sidedness would be the least important construct in this research. However, this turned out not to be true since the hypothesis was not supported and sidedness was overall and for the different segments not the attribute with the lowest relative importance. Only respondents in segment 3 found sidedness the least important.

5.4 - What is the influence of the message source construct on WOM impact?

Finally, the message source construct is measured through the attribute source independence which consists of three levels. For the aggregated level, this attribute is the least preferred attribute. For segment 1 and 3, source independence is fourth important and for segment 2, source independence is least important. Source independence is measured on three different levels (Consumer reviews, salesperson in a shop and TV advertisement). For segment 1 and 3, the utility scores for consumer reviews are the highest, for segment 2 and the aggregated level, the utility scores for salesperson in a shop are the highest.

Half of the respondents have a higher probability of accepting the WOM information when the source independence is higher. However, for the other half of the respondents, this does not hold. However, the hypothesis cannot be rejected, because the third level (TV advertisements) does not seem to be a significant factor. These findings are in line with the found literature. People tend to believe independent review sites and are more probable to pass this information on to others (Lee and Youn 2009). In this research, the consumer reviews are the most preferred source for WOM information and according to Fei and Phelps (2004); this is because independent product review websites are generally known to be credible and free of marketer influence.

5.5 - Managerial implications

(43)

analyze information critically and prefer to know negative characteristics about the product. For managers, this could create an opportunity to reach this group of people through experts and their representatives, making sure that people selling their products in shops have the necessary knowledge, credibility and are sure to also mention negative characteristics in order to gain the consumers’ trust.

Segment three exists of young people who almost completely trust the judgment of their friends. They do not care about expertise, credibility or sidedness, they just want to know what their friends think and will accept information coming from them quickly. By influencing the so called opinion leaders in their social groups, these groups can be targeted rather easily. The influential youngsters (opinion leaders) in these groups need to be pushed to share the product characteristics and information with their friends in order to increase sales and thus market shares.

(44)

5.6 - Limitations

This research deals with a number of limitations that could influence the results from this research differing from the real situation. The first limitation in this research is the exclusion of a none option in the survey. A none option is normally common in a CBC analysis but it was decided to leave it out because target of the research was to see which attributes and levels were preferred over others. However, respondents did not have the possibility to choose either of the two scenarios and were forced to make decisions even if they did not like any of the two scenarios. The second limitation is the sample size. With only 205 respondents, the results and opinions may be biased and not as accurate as could have been with a far larger sample size. The third limitation is the composition of the sample. Many respondents had the same demographic characteristics and are mostly students, leading to possible biased results for the complete sample. The fourth limitation is that this research only discusses five constructs influencing WOM impact into account. Consumers might be influenced by completely different characteristics in WOM messages that are not included in this research.

5.7 - Directions for further research

This research is far from conclusive and other areas can be investigated in order to understand WOM impact better. An interesting direction for further research based upon this research could be to include more constructs to investigate how they relate to the findings in this report and if there are other constructs that have larger preferences and influences on WOM impact than the constructs investigated in this research.

Moreover, it would be interesting to do the same research with a lot more respondents. With such a small sample, segmentation is not always useful and there might be more segments to be discovered who have different preferences and characteristics than the ones found in this research.

(45)

For example: there might be other aspects in the construct sender credibility besides if someone gave good or bad advice in the past. It could be interesting to investigate this to increase understanding of the different constructs and to find out the sub-constructs of which they exist.

Word-of-Mouth Preferences

Word-of-Mouth Preferences

Which factors determine Word-of-Mouth impact?

Word-of-Mouth Preferences

Which factors determine Word-of-Mouth impact?

Management summary

Preface

Table of contents

Chapter 1 - Introduction

Chapter 2 - Literature review

Chapter 3 - Research Design

Chapter 4 - Results

Chapter 5 - Conclusion and discussion