The impact of source credibility on review usefulness: the moderating role of customer and firm response

The impact of source credibility on review

usefulness: the moderating role of

customer and firm response

Author: Koen Wegstapel

Date:

January 11, 2016

Faculty of Economics and Business

2

The impact of source credibility on review

usefulness: the moderating role of

customer and firm response

University of Groningen

Faculty of Economics and Business

MSc Marketing Intelligence & MSc Marketing Management

Author:

Koen Wegstapel

Date:

January 11, 2016

Address:

Nieuwe Boteringestraat 45

9712 PH, Groningen

Phone number:

+31(0) 653797556

Email address:

k.wegstapel@student.rug.nl

Student number:

2586312

3

MANAGEMENT SUMMARY

This study examines the relationship between source credibility and the perceived review usefulness, with a focus on negative online consumer reviews. A new developed tool on review websites is a reputation mechanism for reviewers, based on earlier written reviews. This reputation mechanism presents the acquired source credibility of the reviewer. This research gives valuable insights into which influence the presence of this tool has on the perceived review usefulness. In addition, more and more often there is the opportunity to respond on an online consumer review. Both consumers and firms can provide these responses. However, interaction effects of different responses between the relationship of source credibility and review usefulness have not been considered before. This research observes the provided content on a review site as a form of dialogue between consumers and firms. This contributes to a possible increase of the perceived review usefulness. Hence, this leads to a greater perceived confidence in online reviews and eventually in better guidance during online shopping.

235 respondents participated in this study by the means of a choice-based conjoint analysis. Via this study the preferences and relative importance of the attributes are measured among respondents. Research results show that source credibility, based on someone’s prior provided reviews, has a positive influence on the perceived review usefulness. People attach great value to a response to a negative online consumer review, whether this comes from a consumer or a firm. The perceived review usefulness increases when people find a response on a negative online consumer review. Online retailers should encourage consumers to provide a response. Given that there is a response, people prefer a consumer response instead of a firm response. A presented disconfirming (positive) consumer response leads to a decreased influence of source credibility on review usefulness. However, this finding only holds in the situation of high source credibility compared to low source credibility. Results shows that people have more confidence in content provided by consumers than provided by firms.

4

ACKNOWLEDGMENTS

This thesis is the last achievement of my master Marketing Management and Marketing Intelligence and marks the end of my time as a student. After a turbulent period and after many hours of work, the time has come to finish my study and to start my professional career. First of all, I am very grateful to my supervisor dr. Hans Risselada for all provided feedback during the scheduled group meetings. Subsequently, I want to thank my thesis group for the support during the feedback moments. Furthermore, I would like to thank my dear friend Marten van der Zee for all valuable discussions and huge support during the entire process. Finally, a special word of thank to my parents Gerard Wegstapel and Gerda Wegstapel, without them I would never have been able to achieve this.

I hope you enjoy reading my thesis.

5

TABLE OF CONTENTS

2.1 Word of Mouth and Electronic Word of Mouth... 10

2.2 Online consumer reviews ... 11

2.2.1 Negative online consumer reviews ... 12

2.3 Source credibility ... 14 2.4 Response type ... 15 2.5 Conceptual model ... 20 3. RESEARCH DESIGN ... 21 3.1 Research method ... 21 3.2 Data collection ... 21 3.3 Study design ... 23 3.4 Plan of analysis ... 24 4. RESULTS ... 26 4.1 Characteristics ... 26 4.2 Main effects CBC ... 27 4.3 Interaction ... 29 5. DISCUSSION ... 34 5.1 Theoretical implications ... 34 5.2 Managerial implications ... 37

6. LIMITATIONS AND FURTHER RESEARCH... 38

7. REFERENCES ... 39

8. APPENDICES ... 46

Appendix A: Main effect model ... 46

6

1. INTRODUCTION

Our current society consists of much empowered consumers (Broniarczyk & Griffin, 2014). The number of people connected to the Internet still increases day by day and at this time there are seven billion different devices connected with more than three billion consumers spread over the world (International Telecommunication Union, 2014). Everybody is connected with one another and this development allows for a growing availability of electronic word of mouth (eWOM). EWOM contains ‘’information spread by customers via Internet about goods, brands and companies’’ (Babic, Sotgiu, Valck & Bijmolt, 2015). An important element of eWOM is that it is in most cases a consumer-to-consumer interaction. Information is now available as a combined resource from different sources, created by connected consumers through the Internet. The content they write can be described as social capital (Paxton, 1999). A major aspect for consumers during online shopping is gathering information about different product attributes and they also search for recommendations provided by other users (Lee, Park & Han, 2008). One of the most consulted sources of eWOM during online shopping and therefore one of the most important categories of eWOM are online customer reviews (OCR) (Babic et al., 2015). Nielsen (2012) shows that 70% of consumers trust online reviews. So people see reviews as an important source to consult information. The majority of firms have recognized this fact and they allocate a higher marketing budget than ever before, trying to manage the eWOM process on their websites to present the information clearly as possible (Moorman, 2014). Despite this development, firms have not yet figured out a formula for successful eWOM management (Babic et al., 2015).

The relevance and the influence of OCR are broadly discussed in literature (Zhu & Zhang, 2010). The abundance of studies regarding OCR focuses on positively rated reviews and their different aspects. Chevalier and Mayzlin (2006) show that positive (valence) OCR has positive significant influences on sales and this is confirmed by a study of Zhang, Ma and Cartwright (2013). The volume and dispersion of positive OCR has proven to be a positive significant influence on sales and review usefulness (Liu, 2006). Alongside the positive OCR there are also negative OCR on review pages available. People find both positive and negative reviews useful (Sen & Lerman, 2007).

Even though less research has been done regarding negative eWOM, contradicting findings have been found. It can be stated that prior research disagrees on the question of what influence negative OCR have on brand evaluations, purchase intentions and even on review usefulness.

7 counter normative (Kanouse & Hanson, 1972). The negative cues that appear tend to attract more attention, more than positive cues (Kanouse & Hanson, 1972; Sen & Lerman, 2007). This effect is called ‘’ negativity effect’’ and can be explained by the fact that negative information contains more diagnostic value. According to Sen and Lerman (2007) consumers tend to pay more attention to negative eWOM than to positive eWOM and they conclude that consumers find negative information more useful regarding product evaluations. East, Hammond and Lomax (2008) show with their research that positive eWOM and negative eWOM are similar regarding advice giving behaviour, with a different effect with respect to their choice. If consumers consult OCR to find support for their positive attitude, negative information has been shown to have a greater impact.

A very interesting development with respect to OCR is that e-commerce sites have developed a new tool, a form of reputation mechanism. With this tool people are able to quantify the quality of the reviewer. Because this is such a new development, knowledge about the influence of this reputation mechanism is currently lacking. Prior research from Floyd et al. (2014) shows that people, which show great expertise, have a strong influence on the product sales. It is likely that source credibility for negative OCR also seems to be a great influencer and that higher source credibility will be experienced as more useful. Usefulness can be described as ‘’peer-generated evaluation of a product facilitating the purchase decision process of consumers’’ (Mudambi & Schuff, 2010). Ghose and Ipeirotis (2007) show that people who consult online reviews want to see confirmation of the validation of the product. If there are no comments on the product, the review is perceived as less useful. In addition, Mudambi and Schuff (2010), show that review extremity, so a very positive or negative review, positively affect the perceived review helpfulness and usefulness.

8 Therefore, it is expected that the influence of source credibility of a negative OCR on review usefulness, is moderated by response type. The focus of this research is on the influence of source credibility on review helpfulness and on the interaction effects between response types and this relationship. Build on this information; the aim of this paper is to provide an answer to the following research question:

‘’To what extent does a customer/ firm response affects the relationship

between negative OCR provided by highly credible/ low credible sources and

review usefulness?’’

This research builds on existing literature by identifying the influence of source credibility of the reviewer followed up by a firm or a customer response, which will lead to a different degree of usefulness for consumers. To understand how differences regarding source credibility and different response types influence review usefulness, participants in this study were confronted with different combinations of reviewer source credibility combined with different type of responses. Through focusing on the probable mitigating effects of different response types, the most beneficial way to respond to a negative OCR from a consumer’s perspective is investigated.

Hence, this study investigates the influence of the different responses on the relationship between source credibility and review usefulness. This is based on the Heuristic-Systematic Model (HSM). This theory shows that people processes information via two modes; heuristic and systematic. Via this way people base their opinion about the perceived validity and usefulness of the provided information (Watts & Zhang, 2008).

Prior research on negative OCR has been executed for different product categories and provides different results. Negative OCR for utilitarian products are compared to hedonic products more focused on product-related content instead of personal taste, that is why OCR for hedonic products are perceived as less valuable (Kikumori and Ono, 2013; Sen and Lerman (2007). Sen and Lerman (2007) show that people perceive negative reviews as more useful than positive reviews for utilitarian products. Based on these arguments this paper will focus on utilitarian products.

9 The results of this study show that when consumers write a negative OCR and they have built up high credibility with their previous review behaviour, they have a stronger positive influence on review usefulness compared to consumer with low credibility. This research adds different dimensions to the literature by further exploring the influence of negative OCR written by different levels of credibility on review usefulness and this is the first research that considers the phenomenon of OCR as a form of a dialogue. This study provides evidence that a response, whether these come from a firm or consumer, is positively related to review usefulness compared to a situation in which people do not find a response. However, people attach most value to a consumer response with respect to review usefulness. This paper shows that there is an interaction effect between consumer response and the relationship between source credibility and review usefulness.

A positive consumer response (disconfirming) on a negative OCR decreases the influences of source credibility on review usefulness. The presence of disconfirming information fosters systematic processing because people perceive this information as intriguing and hence affects the balance between systematic and heuristic processing. This only applies in the situation of high source credibility compared to low source credibility. A negative consumer response (confirming) increases the influence of source credibility. Both a positive and negative firm response increases the influence of source credibility on review usefulness. This only implies in the situation of moderate source credibility compared to low source credibility. However, the firm results must be interpreted carefully, because these findings are just marginally significant. These findings extend literature and provide new research directions.

The managerial relevance of this paper is to show practitioners how they can optimize the appearance of negative online reviews. If people have built up high source credibility, which is presented via a reputation mechanism on an e-commerce site, than this increases review usefulness. Practitioners must be aware that people with high credibility have a strong influence with respect to the content they spread. To decrease the influence of high source credibility on the perceived usefulness, managers and online retailers could encourage people to provide disconfirming responses on negative OCR.

10

2. THEORETICAL FRAMEWORK

This section presents an overview of existing literature about the phenomenon ‘online consumer reviews’. First the topics of word of mouth (WOM) and electronic word of mouth (eWOM) are discussed. Secondly, negative OCR and the possible influence of different kind of responses will be given in order to provide a comprehensive understanding of OCR. Finally a conceptual model will be presented.

2.1 Word of Mouth and Electronic Word of Mouth

WOM within marketing refers to consumers, which provide each other with information about a certain brand, service, good or company (Babic et al., 2015). This description of WOM is in line with the description by Liu (2006) as ‘’informal communication between different consumers about services and products’’. An important factor is that companies do not play any role in this form of communication. Arndt (1967) showed, as one of the first researchers, that positive WOM has a positive effect on the purchase probability whereas negative WOM decreases the purchase probability. It is determined that traditional WOM influences consumers purchase decisions, both pre-purchase and the post-purchase product perception (Matos and Rossi, 2008). In the last decade the traditional WOM has undergone a transformation towards different types of eWOM. This phenomenon can be described as ‘’Internet-mediated written communications among potential and current consumers’’ (You, Vadakkepatt & Joshi, 2015).

Currently, there is an enormous increase in people sharing and exchange information about products on the Internet and this explosive grow of available information sources allows consumers these days to rely on information during their online purchase process even more (Werbler, 2009). Compared to the traditional WOM there are some major differences for both companies as well as consumers. An important difference from a company’s perspective is that eWOM conversations are easier to track because the conversations take place on the Internet (Godes & Mayzlin, 2004). Another important difference compared to WOM is that the volume of eWOM is much greater and the reach potential is radically enlarged (Dellarocas, 2003; Zhang et al., 2013). The dissemination of positive and negative eWOM communication is exceptionally fast and this form of communication takes place within a network of people (King, Racherla & Bush, 2014).

11 Despite all the differences between WOM and eWOM it is clear that eWOM is incredibly useful in the decision-making and purchase process of customers (Zhang et al., 2013).

2.2 Online consumer reviews

The phenomenon eWOM not only encompasses OCR, nonetheless it can be seen as one of the most important categories of eWOM (Babic et al., 2015). During the process of online shopping there is always a certain amount of risk for the specific buyer. One of the possibilities consumers have to reduce this perceived risk is to consult OCR. This is confirmed by the study of Zhu and Zhang (2010), wherein they show that OCR has a major role within the online product decision-making process of the customer and hence are experienced as useful. An essential part for consumers during the purchase process is the gathering of product attribute value information and to read different consumer opinions and recommendations provided by different people (Lee et al., 2008). OCR are found to have two different functions; they act as a recommender and as an informant. From an informative perspective, the OCR contains information, which corresponds with the situations in which sellers provide information. From a recommendation point of view, OCR are more focussed on consumer oriented information, regards product performance from a customer’s point of view and provides a description of different usage situations (Lee et al., 2008).

Most research regarding positive OCR concludes that these reviews have a stronger positive effect on sales than all other marketing-mix instruments (Chevalier and Mayzlin, 2006; Chen, Wang and Xie, 2011). Some scholars found evidence for the positive influence of number of positive OCR on sales (Ho-Dac, Carson and Moore, 2014; Liu, 2006), while others showed that the main predictor of OCR is the valence on sales (Dellacrocas, Zhang and Awad, 2007). Review helpfulness and usefulness are a measure of perceived value within the purchase decision-making process (Mudambi & Schuff, 2010). OCR provides people with diagnostic value across different stages of this process. Chen, Dhanasobhon and Smit (2008) found that when people experience a review as more helpful or useful, that this review has a stronger influence on consumers’ decisions making process.

12 This paper will differentiate from existing work because within this research the provided negative OCR are guaranteed written by consumers. In addition, this work focuses solely on negative OCR whereas prior research mostly provided both negative as possible OCR. Hence, the sequel of this paper will focus on negative online reviews written by consumers and the support will be set out in the next sub chapter.

2.2.1 Negative online consumer reviews

Floyd et al. (2014) show that positive OCR mostly enhance the expected attitude and quality of products for customers. Negative OCR includes complaints, product defamation and has an unfavourable impact towards consumers’ product attitudes. The attribution theory (Kelley, 1973) is used in prior research of eWOM and OCR and tries to explain what causes the effect of eWOM (Sen & Lerman, 2007; Lee & Song, 2010). This theory delineates the process of how people draw causal inferences based on the communication of other people. Based on this theory, a reader can either judge the motivation of the sender as stimulus attributed or as non-stimulus-attributed (Lee & Song, 2010). The motive for writing a negative OCR could either be to write about the product’s low quality (stimulus) or, for example to try to improve the negative provided content (non-stimulus).

13 Sen and Lerman (2007) show with their results that negative OCR are more helpful than positive OCR for utilitarian products. It is clear that for different researches a distinction is made between product characteristics, especially between utilitarian and hedonic products. Hedonic products can be characterized as products wherein the consumption is related to a sensory experience and sensual pleasure, whereas the consumption of utilitarian products focuses on a functional and practical task (Cheema and Papatla, 2010). Sen and Lerman (2007) note that in case of negative OCR as applied in this paper, the negative OCR for utilitarian products are related to product-related content and negative OCR for hedonic products are more based on personal taste. Therefore, people experience them as less value-added content. In all studies above, there were both negative as well as positive OCR displayed to consumers. The focus of this research is solely on negative OCR and based on above-mentioned arguments the focus is on utilitarian products. The main reason for this decision is to exclude personal tastes of people within the negative OCR and ensure that the content is only about product-related issues.

Review usefulness can be explained by the extent to which people accept the provided content and consider it as meaningful. Hence, usefulness shows a great similarity with information adoption. This description is on an individual level of information processing perspective that contemplates how meaning is ascribed to content (Goodman & Darr, 1998). People ascribe meaning to content in their memory. The organizational memory of humans is a highly valuable organizational asset (Walsh & Ungson, 1991). In the context of this research, the content of an online review website is invaluable for people who consult OCR. Nevertheless, these reviews are only valuable in so far as the provided content of the reviews are used. OCR websites are content-based. Hence, review usages implies that content will be adopted by readers (Watts & Zhang, 2008). Therefore, a theory is sought for the adoption of negative OCR, or how such a review is seen as useful.

14

2.3 Source credibility

It is very likely that the source, the person writing the OCR, has an impact on the review usefulness. The influence of source characteristics originates from persuasion research in social psychology, which shows that a source characteristic can affect and enhance the influence effect (Hovland, Janis & Kelly, 1953). They show with their source credibility model that when sources demonstrate a higher level of trustworthiness and expertise they are considered to be more credible and have a greater persuasiveness, this results in a more useful review.

Consumers commonly form an opinion regarding the added value of an online platform with online customer reviews from more than just the written content. They attach importance to additional provided information about the sender of the review (Babic et al., 2015). Information provided by high credible sources are more likely to be and more easily accepted because they include a high level of expertise and trustworthiness (Bearden & Etzel, 1982; Hovland et al., 1953; Lee, Park & Han, 2011). Lee et al. (2011) found that trustworthiness is important; high credible sources positively influences consumers’ attitudes towards brands. So it is likely that highly credible sources positively affect review usefulness.

A recent study of Floyd et al. (2014) shows that positive OCR provided by sources with greater expertise provide higher product sales. With source credibility in their research they refer to the source platform on which different OCR are provided and to the content of the review. This research will shift the attention to the reviewer and this is in line with a recently developed reputation mechanism of firms, which is able to quantify the quality of the review providers (King et al., 2014). This reputation quality is based on previous written content and other readers determine this quality ranking. This new development could have important consequences for the influential power of OCR (King et al., 2014).

15 When readers of negative OCR carefully read the message, they are engaged in systematic processing. If people experience the provided OCR as more relevant and truthful, they will perceive the review as more relevant and hence of a higher quality (Xu & Chen, 2006). So it is likely that a negative OCR written by someone with higher source credibility positively influence the adoption of the information and hence increases review usefulness.

If the reader of an OCR thinks that the review is credible, this person will have more confidence in the comments stated in the messages and will consider them more during the purchase decision process (Nabi & Hendriks, 2003). Connors, Mudambi & Schuff (2011) found that if reviewers show more expertise, this positively increases review helpfulness. In addition, McKnight and Kacmar (2006) show a positive effect of source credibility and the willingness to accept the information on a website. With respect to this research, it is very likely that consumers who perceive the negative OCR as more credible will trust and ‘use’ the content of this particularly review. If the source is perceived as less credible, this effect will be the other way around; they will be very unlikely to follow the written content. Consumers always search for information provided by people who are perceived as more knowledgeable (Park and Lessig, 1981). Taking the previous arguments into consideration, it is expected that source credibility has a positive effect on review usefulness. Negative OCR written by credible reviewers are expected to have a stronger positive effect on review usefulness. Hence,

H1: Source credibility positively affects review usefulness.

2.4 Response type

While the influence of negative OCR and source credibility has already received some attention within the scientific literature, the influence of response to negative OCR is less settled. The research on the operationalization and the phenomenon of the response itself in the context of OCR is underexplored. Negative information about different products, brands and companies are well presented on websites, which contains OCR. In prior research about negative OCR, this phenomenon was never seen as a dialogue. First different possible responses will be discussed and then the potential impact of these responses will be set out.

Consumer response

16 can be seen as a positive response and a confirming review can be seen as a negative response. This paper will focus on two types of consumer responses, a positive and a negative consumer response. The positive response contains legitimate product information trying to convince other consumers to disregard the negative information and the negative response confirms the negative written content. Mathwick, Wiertz and de Rutyer (2007) did research on social capital production in a virtual peer-to-peer problem solving (P3) community, related to consumer experiences. Social capital can be described as an aggregation of resources where groups and individuals gain from connections to each other (Paxton, 1999). In virtual communities people try to find solutions, exchange experiences and try to build expertise by the use of textual conversations (Matchwick et al., 2007). This context can be compared with pages that cover OCR. The results of Matchwick et al. (207) shows that when there is no response given to a negative comment by a sponsored comment (firm response), individual problems lead to collective action. Within the community people will cooperate with each other, reacting to one another’s message. They spend their content to suggest competitive alternatives in order to put pressure on the specific firm trying to force a response.

If someone responds with a positive response, then one can speak of divergent opinions. Babic et al. (2015) show that divergent opinions of consumers on a review page increase the perceived uncertainty about the performance of a product and so negatively affect a firm’s bottom line. This can be linked to recommendation consistency. This indicates the extent to which a review is consistent with the contributor of the first message, concerning the evaluation of the same product (Zhang & Watts, 2003). In a context of an online consumer decision forum they found, that if recommendations were consistent, readers rated the reviewer credibility as higher. If the eWOM recommendation was inconsistent for the same product, the reader perceived the first opinion as less credible and they got confused (Zhang & Watss, 2003).

From a HSM perspective, disconfirming information stimulates systematic processing because it is both upsetting and intriguing (Watss & Zhang, 2009). The disconfirming information challenges the reader of a review to investigate the response more precisely to understand differences between the given information (Vandenbosch & Higgins, 1996). It forces the reader to think about the contradictions and therefore evaluate the message’s validity (Maheswaran & Chaiken, 1991).

17 Their research findings demonstrate that when readers in the online community are confronted with the presence of disconfirming information, they rely to a lesser extent on heuristic processing, as implied by the decreasing influence of source credibility (Watts & Zhang, 2009).

Therefore, it is expected that the heuristic cue of source credibility of negative OCR will be less influential during the readers ‘validity assessment’ than when they are confronted with disconfirming information (a positive consumer response). This leads to the following hypothesis:

H2: Source credibility of negative OCR has a less positive effect on review usefulness after a positive

consumer response compared to a negative consumer response

Firm response

In the light of firm response to negative eWOM some research has been done. Not specifically about the effectiveness of a firm’s response to negative OCR but to negative eWOM in general. The interventions a firm can take to tackle negative eWOM are referred to as webcare activities. The concept of webcare has not been defined within the scientific literature. Noort and Willemsen (2012) are one of the first trying to define this concept. They describe webcare as engaging in an online interaction with complaining consumers by actively providing feedback regarding the question or complaints.

According to Kerkhof, Beukeboom and Utz, (2010), a company can react to negative eWOM because a customer specifically asks for a firm response (i.e. reactive webcare) or they can react proactively, without a special demand from the consumer. Negative OCR can be seen as a trigger event, which has such a powerful force that it can negatively affect innumerable customers or potential consumers (Noort & Willemsen, 2011).

18 Despite the fact that the current literature regarding the effect of response on negative eWOM is still in its infancy, there is some empirical evidence that shows that a specific form of webcare can provoke positive consumer responses after a firm confronts negative eWOM. Lee and Song (2010) showed participants different sorts of negative eWOM messages and these messages were either followed up by a firm response or a non-response. Within the response the firm tried to debunk the complaint mentioned in the eWOM. Their results show that a firm’s response compared to a non-response creates a more favourable effect on how consumers evaluate the specific firm. In addition, Kerkhof et al. (2010) show that any kind of firm reaction (financial compensation or an apology) to negative eWOM raise positive responses.

Van Noort and Willemsen (2011) did a study with respect to online damage control when companies try to counter complaints as expressed in negative eWOM. They also made a distinction as initiated by Kerkhof et al. (2010) between reactive and proactive responses. An important aspect during their research was the degree of conversational human voice in the responses. According to Kelleher and Miller (2006) conversational human voice is an important factor when firms want to create favourable brand responses during computer communications. Kelleher and Miller (2006) argue that firms show a lot of conversational human voice when they communicate in a way that they provide feedback without being critical with respect to the content provider. The research results of van Noort and Willemsen (2011) show that consumers evaluate a brand more positively after a response compared to a non-response and this result is even more positive when the response contains a conversational human voice.

Based on those different studies, it is evident that a firm even as a consumer has several response options to negative OCR. Van Noort and Willemsen (2011) suggest a response with conversational human voice, which could be compared to a diagnostic response as suggested by Ahluwalia et al. (2000) in which a firm shows understanding for the situation. A response with conversational human voice is comparable with a negative response (confirming response) as mentioned in the customer response subchapter in which the firm can’t deny the reason for writing a negative OCR. A counterargument response in which a firm counter argues the negative OCR is comparable with a positive response where they try to improve the negative content (disconfirming response).

19 decreases the effect of source credibility on the adoption of information since people rely to a lesser extent on heuristic processing. Therefore, it is expected that the heuristic cue of source credibility of negative OCR will be less influential during the readers ‘validity assessment’ when they are confronted with disconfirming information (a positive firm response). Hence, the following hypothesis will be tested:

H3: Source credibility of negative OCR has a less positive effect on review usefulness after a positive

firm response compared to a negative firm response

Furthermore, when readers do not find any response to the negative OCR (neither a consumer or a firm response), they might infer why nobody provides an answer. So they must be satisfied with one given negative OCR from a certain source. They cannot find any confirmation, divergent opinion or additional information. People search either for a one sided message, or a two-sided messages (Cheung, Sia & Kuan, 2012). A positive response will provide people with two-sided information, whereas a negative response strengthens the one-sided information. In addition, previous research shows that review quantity is positively related to information credibility and so positively influences review usefulness (Chunhua & Yezheng, 2009; Xiaoring et al., 2011). Following this line of reasoning, a response enlarges the quantity of eWOM and enlarges the information of a single review and so positively influences review usefulness compared to a non- response. This leads to the following hypothesis:

H4: A response is positively related to review usefulness.

20

H5A: Source credibility of negative OCR has a stronger negative effect on review usefulness after a positive consumer response compared to a positive firm response.

H5B: A consumer response has a stronger effect on review usefulness than a firm response

2.5 Conceptual model

Figure 1: Conceptual framework

21

3. RESEARCH DESIGN

This chapter provides the research design of this study. First, the choice for the method of this research will be justified. Subsequently, the data collection procedure will be discussed. Lastly, the study design and plan of analysis will be explained.

3.1 Research method

This study is conducted via a Choice-Based Conjoint analysis (CBC). A Conjoint analysis provides respondents with different ‘choices’ and commonly registers the evaluation of the corresponding attributes and levels. Statistical methods are used to dissect the preferences (Eggers and Sattler 2011). Those preferences can be developed for any type of product, services, website etc. (Hair, Black, Babin, Anderson, & Tatham, 2010). In the case of this study consumers will be provided with different negative OCR. During a Choice-Based Conjoint analysis, respondents get continuously shown sets of alternatives, from which they have to select their most coveted alternative (Mccullougii, 2002). Within this study, respondents are forced to choose between three alternatives. Source credibility, response type and response source are attributes which influence the review usefulness and so the preference for a type of dialogue with respect to negative OCR. The most useful review (so the most preferred choice) is based on the utilities of the attributes with the corresponding levels with the specific option.

3.2 Data collection

The survey was created in Preference lab, which contains all different choice-sets as well additional questions to measure control variables. The survey was spread via mail, social media networks and personal contact. People were also recruited in the university library.

Conjoint attributes

22

Attribute Attribute levels

Consumer source credibility Low (1 out of 5 stars) Moderate (3 out of 5 stars) High (5 out of 5 stars)

Response Consumer positive Consumer negative Firm positive Firm negative Non-response

Table 1: Conjoint input

The attribute ‘consumer source credibility’ was manipulated by creating three different attribute-levels concerning different credibility. A pre-test (n=15) confirmed the differences between the credibility. Respondents were asked, after being confronted with different OCR written by different people, whether they would indicate the credibility of the source as ‘ high’, ‘moderate’ or ‘ low’. This was manipulated in the program Photoshop, as different sources were labelled with a number of stars (High was 5 out of 5 stars as stated in table 1 above). They confirmed the pre-expected differences among the sources. The attribute levels of response consist of different kind of responses provided by different sources. In order to be sure that a negative consumer response really was perceived as negative, a pre-test (n=15) was conducted to be sure that all different responses were perceived as they were intended. After the pre-test small changes were needed with respect to the word choice.

Control variables

23

3.3 Study design

This study used a conjoint analysis, which included 14 different choice sets with 3 alternatives per choice set. In total there were 235 respondents recruited for this research and together they stand for 3290 choices. In order to find interaction effects between attributes a high number of choice sets and respondents are needed (Sawtooth Software, 2013). To make sure that respondents keep their attention and focus, respondents could leave their email address and win a voucher for bol.com. An optimal choice design for a conjoint analysis is balanced, minimizes overlap and is orthogonal (Sawtooth Software, 2003; Charzan & Orme, 2000). The article of Charzan and Orme (2000, p.171) also state that a random design method is the most accurate way to assess and estimate interaction effects. The aim of this paper is to discover interaction effects of different responses between source credibility and review usefulness, so the random design method is used during this study. The software Preference Lab provided the choice sets used for the survey. This programme automatically accounts for a random and balanced design, so all attribute levels are shown approximately an equal number of times in an orthogonal way (Eggers & Sattler, 2011). Thus, all attribute-levels are measured independently from all the other effects (Sawtooth Software, 2013). The different attributes and levels are created in Photoshop, so the choices consist of different layers. An example of a choice set is shown below in figure 2.

24

Procedure

Respondents received a link, which redirected them to the online survey. Before starting the survey, people were asked to imagine a situation in which they are orienting for the purchase of a digital camera. Back at home they went surfing on the Internet and searched for OCR. Before they had to make any decision it was highlighted that people needed to look carefully at the different choices and provided information. The combination of attribute levels varied, this resulted in different options per choice set. Subsequently, people were asked to select the most useful combination of negative OCR written by varying credible writers and response. A no-choice option was not included because this would give the respondents the opportunity to skip difficult choices.

3.4 Plan of analysis

To be able to estimate the value of each respondent on each attribute, the utility, the following equation will be used:

𝑈

∑ 𝛽

𝑘=1

𝑋

Utility (U) of all respondents of review usefulness (j) is the sum of the utilities (β) of the different explanatory variables (X; k=1,..K).

In order to be able to assess the influence of different responses (firm/consumer/ non), the attribute response is taken apart. New attributes have been created in SPSS, so the effect for a response compared a non-response generally, a firm and a consumer response on review usefulness could be determined.

Table 2: New response attributes

Attribute Attribute levels

Response Yes No

Consumer response Positive Negative

25 The conjoint data was analysed with the software LatentGold. When someone was confronted during the survey with a consumer response, then the other options; firm and non-response, received a missing value in SPSS in the consumer response variable. The same procedure was executed the other way around for the firm response variable. These changes were needed in order to ensure that the effects were not measured twice. This procedure was necessary in order to separate the original ‘response’ attribute. Via this way the main effects of source credibility, a response in total compared to a non-response, a consumer response and a firm response on review usefulness could be determined. These variables were included for the main effect model.

Model fit

The model fit can be compared between two different models to see if there is a significant improvement. The main effect model and the model including the moderation effects can be assessed by comparing the goodness of fit, where the information criteria AIC is one of the most important. Vermunt and Magidson (2005) state that the AIC takes the parsimoniousness of the model into account. In addition, the AIC corrects for the number of parameters, because normally a model with more parameters will fit better. A lower value of the AIC represents a better fit (Vermunt & Magidson, 2005).

Interaction effects

26

4. RESULTS

Within this chapter the results of the conjoint analysis are presented. First, the sample characteristics are discussed. Then, the main effects of the CBC analysis are shown. Finally the interaction effects between the attributes are studied.

4.1 Characteristics

The sample for this research consists of 235 respondents. This exceeds the minimal required number of 200 respondents, in order to be able to make a sufficient population representation (Hair & Black, 2009). Table 3 shows different respondent characteristics.

Table 3: sample characteristics

The sample is not evenly distributed regarding gender. The largest part of the sample is male (n= 154, 65,5%), whereas the minority is female (n=81, 34,5%). The vast majority belongs to the youngest age category (n=108, 46%). The youngest person within this sample is 18 years old, whereas the oldest person is 64. The mean age is 28.

27 Furthermore, this sample is well educated. 25,5% (n=60) has a HBO education while 57% (n=134) has a WO education. Therefore it is not surprising that the largest part within this sample is a student (n=127, 54%). In addition, the second largest group consists of people who are in-paid employees (n=68, 28,9%).

4.2 Main effects CBC

This section discusses the main effects model of the Choice-based Conjoint analysis. Subsequently different insights will be provided about the relative importance of the different attributes. Hence, all main effects are elaborated and compared to the relating hypotheses.

Main effects model

The estimation of the model showed that all the attributes are highly significant (Appendix A). This indicates that it is not necessary to drop an attribute from the model. The result of the main effects model compared to the naïve model is summarized in table 4. The difference between the Log likelihood of the main effects and the naïve model shows a χ2 of 1127,37. This indicates that the main effects model is able to make better predictions compared to the naïve model (p <0.01). Since all of the attributes remain part of the model, the preferences are estimated for the four attributes remaining.

Table 4: Full model compared to naïve model Selection

Each effect per attribute can be included differently in the model. They can either be linear, nominal or quadratic. Since the attribute level of source credibility is composed of levels representing; low, moderate and high credibility (imaged by a star rating) it is not realistic to consider this attribute as linear, so it is included as part-worth. Furthermore, the response attribute (yes/no), firm response and consumer response are nominal variables, so these attributes will also be included as part-worth models. Figure 3 shows the part-worth estimation for all included attributes.

Model LL Npar R2 R2ad χ2 df P value AIC Hit rate

Naïve -3614,43 33.33%

Main

28

Figure 3: Part-worth estimation attributes Main effects model

Table 5 below provides an overview of all part-worth estimated values for the different attributes levels. The review usefulness attributes were all significant at a 1% level. It can be concluded that people attach the greatest value to if there is a response to a negative OCR, with a relative importance of 45.56%. A consumer response is the second most important attribute with a relative importance of 23,42%. Thereafter source credibility is the third most important with a relative importance of 16,07%. However, note that the firm response attribute and the consumer response attribute are conditional to whether there is a response. Interpreting the attribute importance of both attributes should also be done conditionally. Interpreting the values as a percent is not possible anymore. Because the attributes firm and consumer response are conditional to if there is a response, the magnitude of the importance (the range) is now what is need to be compared.

Table 5: Main effects model (*** significant at the 1% level) -1 -0,5 0 0,5 1 Sou rce … Lo w Mo d er ate _High Re sp o n se _Yes _No Firm r es p o n se N egat iv e Po sitiv e Con su m e r… N egat iv e Po sitiv e

Attribute Utility P-value Range Relative importance

29

Hypotheses testing

To be able to determine the main effects, all different attribute-levels of the full model were examined. The part-worth’s estimates are represented in figure 3 and table 5. Regarding the attribute source credibility it was found that low source credibility has a negative effect (β -0.2968) on review usefulness, relative to the mean. Contrary to low source credibility, high source credibility (β 0.2978) has a positive effect. Therefore it can be stated that an increasing source credibility, has a positive effect on review usefulness. Hence, hypothesis 1 is supported. Subsequently, within the attribute response it is clear that a response, irrespectively a consumer or a firm response (β 0.8431) has as positive effect, whereas a non-response has a negative effect (β -0.8431) on review usefulness. This finding supports hypothesis 4. To be able to compare the different influences of firm and consumer response, we first see the differences between the relative importances. It turns out that people within this research attach more value to a consumer response. The range of the attributes confirms this. Negative consumer response to a negative OCR has a stronger positive effect (β 0.4333) on review usefulness than a negative firm response (β 0.2767). In addition, a positive consumer response has a stronger negative effect on review usefulness (β -0.4333) than a positive firm response (β -0.2767). Hence, hypothesis 5B is supported. Either a positive or a negative response of a consumer has a stronger influence (positively and negatively) on review usefulness compared to a positive and a negative firm response.

Validity

The main effects model shows a hit-rate of 55,6% (table 4) So in approximately 56% of the cases, the main effects model predicts accurately. Compared to the naïve model in which a selection of the three alternatives provides us with a 33.33% hit-rate, the main effect model has a relatively good prediction power.

4.3 Interaction

30

Full model and selection

To be able to estimate the preferences of the respondents, while capturing for the effect of interaction of response between source credibility and review usefulness, the new variables were included in the full model. To see if the model with interactions improves the model fit, a comparison is made with the main effects model. Table 6 shows a comparison between both models.

Table 6: Main effects model compared to full model

The full model, which accounts for the interaction effects between the attributes source credibility and response, shows an increase in the log-likelihood value and is thus closer to zero. This indicates a better model fit. In addition, the R2 and the R2ad are also higher for the full model with interactions

included compared to the main effects model. Table 6 shows that the full model presents a better AIC value, so it is evident that the model fit is improved.

As stated before, the attributes can differ with respect to the type of model. Since the non-response effect is already in the response attribute, the attributes consumer response and firm response are treated as numeric. The variables that contain the moderator effect should also be treated as numeric. Table 7 on the next page provides an overview of the estimated full model containing the interaction variables.

Model LL R2 R2ad AIC Hit rate

Full model -3032,90 0,1611 0,1590 5613,811 56,8%

31

Attribute Utility Wald P-value

Source credibility 146,8118 0.00*** Low -0,2874 Moderate -0,0033 High 0,2907 Response 432,3328 0.00*** Yes 0,848 No -0,848

Consumer response (positive) -0,4233 191,7942 0.00***

Moderatingconspos*lowcr 0,2337 18,7654 0.00*** Moderatingconspos*modcr 0,0443 0,6994 0,4 Moderatingconspos*highcr -0.278 Moderatingconsneg*lowcr -0.2337 Moderatingconsneg*modcr -0,0443 Moderatingconsneg*highcr 0.278

Firm response (positive) -0.2567 44,0905 0.00***

Moderatingfirmpos*lowcr -0,0698 0,7526 0,39 Moderatingfirmpos*modcr 0,0752 2,7562 0,097* Moderatingfirmpos*highcr 0.0054 Moderatingfirmneg*lowcr 0.0698 Moderatingfirmneg*modcr -0,0752 Moderatingfirmneg*highcr 0.0054

Table 7: Full model with interactions (*significant at the 10% level *** significant at the 1% level)

The parameters of the main attributes only differ marginally from the main effects model and are therefore not discussed again. As stated earlier, the effect of source credibility, consumer response and firm response are significant. Higher source credibility increases review usefulness just as a negative firm/consumer (table 7 provides the utility for a positive response) response. Consumer response and firm response are now coded as numeric, because the effect of non-response is already in the response attribute. Since high credibility is the reference level, the effects are related to high source credibility. Given that effect coding is used, the effects for the moderating effect between a positive consumer/firm response and high source credibility, even as the moderating effect between source credibility and a negative consumer/firm response are calculated manually. The effect for high source credibility*positive consumer response and for firm response, is the sum of the two levels given in table 7 and multiplied with -1. The moderating effects for the negative responses (consumer/firm) and the different levels are source credibility, are the corresponding utilities for positive responses multiplied with -1.

32 consumer response, this works the other way around. When a consumer response is negative on a negative OCR, so a confirming response, the stronger the effect of source credibility on review usefulness.

As an example, assume that a consumer give a positive response to a negative OCR, then the overall effect of source credibility on review usefulness becomes (Appendix B for the other calculations): Source credibility Utility Utility moderator Total utility Total range moderator+source:

Low -0,2874 0.2337 -0.0537

Moderate -0,0033 0,0443 0.0410

High 0,2907 -0.278 0.0127

0.0947

Total range source:

0.5781

Table 8: moderation effect positive consumer response*source credibility on review usefulness

Hypotheses testing

The range of utilities for source credibility through a positive consumer response becomes smaller compared to the range of the main effect of source credibility without interaction. This means in total that a positive consumer response compared to a negative consumer response decreases the importance of source credibility (the range becomes smaller for positive, for positive wider). The ranges of the other part-worth’s stay approximately the same compared to the main effects model.

After including the interaction variables in the model, the new relative importance is calculated for each attribute as shown in table 9. It shows that people attach most value to a negative consumer response. For a firm response it works the other way around, were people attach slightly more value to a positive firm response.

Table 9: Relative importance

Attribute Range Relative importance

33 The effect of low source credibility relative to high source credibility is highly significant moderated by a positive and negative consumer response. Moderate source credibility relative to high source credibility is not significantly moderated by consumer response. A positive consumer response (disconfirming response) compared to a positive consumer response (confirming response) decreases the impact of source credibility on review usefulness. This means that hypothesis 2 is partially supported. Source credibility of negative OCR have less effect on review usefulness after a positive consumer response compared to a negative consumer response, but only in in the case of high source credibility related to low source credibility.

The effect of moderate source credibility relative to high source credibility is only marginally moderated by a positive and negative firm response (significant at the 10% level). Low source credibility relative to high source credibility is not significantly moderated. Contradicting results are found related to hypothesis 3, instead of the expected decrease of the impact of source credibility on review usefulness after a positive firm response, a positive firm response increases the importance of source credibility on review usefulness. Whereas the range becomes smaller for source credibility after a negative firm response, the ranges become larger after a positive firm response. Hence, hypothesis 3 is not supported.

Hypothesis 5A states that effect of source credibility of negative OCR decreases stronger after a positive consumer response compared to a positive firm response. This hypothesis is confirmed given that the range for review usefulness, after including the moderator positive consumer response in the full-model, becomes much smaller as the original range for review usefulness compared with a positive firm response (Appendix B). That means in total, a positive consumer response decreases the effect of source credibility on review usefulness.

Validity

34

5. DISCUSSION

This research was executed in order to examine the influence of consumer and firm response on the relationship between source credibility and review usefulness. First the different main effects of source credibility and the different responses on review usefulness were studied. Thereafter the moderating effects of consumer and firm responses on the influence of source credibility on review usefulness were analysed. These findings will be discussed in the theoretical and managerial implications.

5.1 Theoretical implications

Main effects

The main effects are in line with the expectations from literature; source credibility has a positive influence on the usefulness of negative OCR. A higher utility was found for high source credibility compared to low source credibility. The results confirm prior research suggestions of King et al. (2014) in which they state that the presence of the recently developed reputation mechanism on review websites in which the quality of the reviewer is determined, could have an important influential power on the perceived usefulness of OCR. This research confirms the fact that people are more likely to accept information provided by credible sources, because they show a higher level of expertise and are therefore more trustworthy (Bearden & Etzel, 1982; Hovland et alk., 1953; Lee, Park & Han, 2011; Reichelt, Sievert & Jacob, 2014).

In this research, people were confronted with reviews provided by people with different source credibility levels. The author’s name was displayed as well as a star rating based on the quality of his previous reviews (the source credibility). By displaying these features people get a sense of the credibility of the author (Watts & Zhang, 2009). This design seems to facilitate heuristic processing, therefore people uses the source credibility cue as a simple decision rule. This is a legitimate explanation as to why a strong significant positive relation between source credibility and review usefulness is found.

35 credibility and usefulness, it enlarges the information of a single review (Chunhua & Yezheng, 2009; Xiaoring et al., 2011).

Prior research shows that consumer generated content is perceived as more credible than firm generated content (Bickart and Schindler, 2001). This research shows that this finding also holds for the different consumer responses compared to firm responses, either for a positive or a negative response. A possible explanation for this finding is that firms could communicate false promises or that their responses motives are based on self-motivated motives (Akdeniz, Calantone and Vorhees, 2013). For a disconfirming response this would be logical, but this is even the case for a confirming, negative firm response. A negative confirming firm response has less positive influence on review usefulness compared to negative confirming consumer response. A possible explanation could be that when a service failure is attributed to a firm, the anger caused through this service failure is higher than the positive feelings that arise in a customer when firms admit the fault (Gelbrich, 2010).

Interaction of responses

An important finding and contribution of this study is the interaction effect of response on the relationships between source credibility and review usefulness. This study is the first one that tries to examine the interaction of reviews as a form of dialogue. The findings show that when there is a positive consumer response (disconfirming) on a negative OCR that the influence of source credibility decreases on review usefulness. However, this is only the case in the situation of high source credibility related to low source credibility, not in the case of moderate source credibility related to low source credibility.

36 different cognitive processes such as differentiation, bolstering and denial (Abelson, 1983). This research’s results shows that if people want to resolve inconsistency between the first observed OCR and the disconfirming OCR that there is a high level of systematic processing which is a difficult cognitive task. This is in line with previous dual-process research (Jain & Maheswaran, 2000). As mentioned earlier, this only applies in the situation of high source credibility compared to low source credibility. When people were confronted with a confirming response, so a negative response, the influence of source credibility on review usefulness increased. After a confirming consumer response people do not have to resolve inconsistency between the first provided OCR and the response, so this results in a less difficult cognitive task and hence in a lower level of systematic processing. The balance between systematic information and heuristic processing is affected in a different way.

37

5.2 Managerial implications

Next to the theoretical implications, the results of this study have some managerial implications for online retailers which offer review opportunities on their website. One important outcome is that when review websites operationalize a reputation mechanism for reviewers in which credibility is displayed, this will have a positive impact on the perceived review usefulness. Managers should be aware of the influential power of source credibility on review usefulness for customers. They can optimize their e-commerce website in such a way, that it facilitates customer choices in for them the most appropriate way. They can achieve this by adding a filter, so the highest source credibility authors are presented on top of the review page.

This research explained the credibility cue by the description underneath the name of the reviewer taken together with the provided star rating. These technical features show managers that the cue used in this study adequately reflects the impact of source credibility. Managers can incorporate the used cue for source credibility in this research on their website so that they provide an accurate representation of the writers source credibility.

This research shows that when people are confronted with a negative OCR and subsequently with a disconfirming (positive) response, the influence of source credibility on review usefulness decreases from a consumer’s perspective. Therefore, in order to decrease the influence of a high credibility source of a negative review on the perceived usefulness, online retailers could stimulate people to react contradictorily to a negative OCR. They can stimulate people who have experienced the product differently to give a response.

38

6. LIMITATIONS AND FURTHER RESEARCH

This research contains a number of limitations. It is possible that some of these limitations can be tackled by future research projects. The first limitation of this study is the research design. The chosen graphical rendering of the review page during the conjoint analysis could have biased the results. The shown product (camera) and the representation of the attributes (colour, form, size etc.) could have an impact on the choices of the different respondents in this study. So further research should affirm the results of this study and the shown attributes should be presented in a distinctive setting.

A second limitation of this study is the limited number of attributes in combination with the high number of choice sets. Hence, people could get the feeling that the different pages were very much alike. it is quite possible that people were not able to keep their focus and attention, which could have resulted in them getting annoyed. This may have resulted in less concentrated participants and therefore a biased result. Future research could investigate the influence of more attributes on review usefulness, thus creating more diversity in the study. As an extra check, scholars should measure how focused people were during the participation.

A third limitation refers to the significant influences of the moderator response on the relationship between source credibility and review usefulness. It can be stated that different influence effects of response and response sources clearly exists. No expectations were formulated for the influence of a confirmative response, which shows to strengthen the relationship between source credibility and review usefulness. However, this study fails to address the origin of these differences. Further research could focus the underlying causes by which these differences are causes.

A fourth limitation of this study is that a holdout set was not used. In order to be able to measure the external validity, values should be compared against a holdout sample. Because there was not a holdout set included in the conjoint, no comparison could be made of the hit rate.

39

7. REFERENCES

Abelson, R. P. (1983). Whatever became of consistency theory?. Personality and Social Psychology

Bulletin.

Ahluwalia, R., Burnkrant, R. E., & Unnava, H. R. (2000). Consumer response to negative publicity: The moderating role of commitment. Journal of marketing research, 37(2), 203-214.

Akdeniz, B., Calantone, R. J., & Voorhees, C. M. (2013). Effectiveness of marketing cues on consumer perceptions of quality: The moderating roles of brand reputation and third‐party information.

Psychology & Marketing, 30(1), 76-89

Arndt, J. (1967). Role of product-related conversations in the diffusion of a new product. Journal of

Marketing Research, 3(4), 291-295

Babic, A., Sotgiu, F., Valck, K., & Bijmolt, T. (2015) (forthcoming, Journal of Marketing Research), “The Effect of Electronic Word of Mouth on Sales: A Meta-Analytic Review of Platform, Product, and Metric Factors” [Dissertation Essay 2]

Bearden, W.O. & Etzel, M.J. (1982), “Reference group influence on product and brand purchasedecisions”, Journal of Consumer Research, 9(2), pp. 183-94.

Blazevic, V., Hammedi, W., Garnefeld, I., Rust, R. T., Keiningham, T., Andreassen, T. W. & Carl, W. (2013). Beyond traditional word-of-mouth: an expanded model of customer-driven influence. Journal

of Service Management, 24(3), 294-313.

Broniarczyk, S. M., & Griffin, J. (2014). Decision difficulty in the age of consumer empowerment. Journal of Consumer Psychology, 24(4), 608-625

Chaiken, S. (1980). Heuristic versus systematic information processing and the use of source versus message cues in persuasion. Journal of personality and social psychology, 39(5), 752.

Chaiken, S., & Eagly, A. H. (1989). Heuristic and systematic information processing within and.

Unintended thought, 212.

Charzan, K., & Orme, K. (2000), “An Overview and Comparison of Design Strategies Choice-Based Conjoint Analysis,” Sawtooth Software Research Paper Series, 98382 (360).