2
Master Thesis
Msc Marketing, Semester I, 2013-2014
eWOM and Online Consumer Reviews
“The Effects of Perceived Reviewer Expertise and Message Content on
OCR Credibility”
Frank Posthumus
s1690183
Parkweg 60a, Groningen
f.l.posthumus@student.rug.nl
+31 (0)6-28709112
Supervisor: Dr. J.A. Voerman
3
I. Executive Summary
This research is directed towards the field of electronic Word of Mouth (eWOM), and focuses on Online Consumer Reviews (OCR’s). Before the Internet revolution, consumers’ information sources on products were limited to information provided from the side of the seller or producer, and direct interpersonal communication. The latter form, also known as Word of Mouth (WOM), has experienced a revolution in form and size since the introduction of the Internet. The WOM-form that is dominating nowadays (at least, in volume) is eWOM. Besides, eWOM is becoming more and more influential in day-to-day purchasing decisions, as consumers are becoming more connected through social platforms as Twitter, Facebook, Pinterest and more, where they can share their opinions. Online Consumer Reviews (OCR’s) are an often used form of eWOM, and can best be described as online evaluations or ratings, provided by users of a certain product or service
.
Credibility of such eWOM communications is vital for the actual adoption of the review or advice by the recipient, thus
for eventual purchase levels. This research identified two main factors that can be of
influence on OCR credibility: perceived reviewer expertise, and message content. Possible
moderating influence from involvement with the product and the recipient’s style of
information processing are identified. This leads to the following problem statement:
“How do the perceived expertise of a reviewer and the content characteristics of a review affect the perceived credibility of online consumer reviews, and how is this effect moderated by a consumers’ product involvement and style of information processing?”
4 When the effect of OCR content on dependent variable Content Credibility is concerned, it is striking that the review with purely product-specific elements is perceived as less credible. This contrasts the findings of the above mentioned studies. Furthermore, and perhaps the most striking, is the fact that OCR with a combination of generic and specific elements is perceived as most credible. This combination of elements was, to the best knowledge of the author, not yet described in the existing? literature, and hints that consumers value a combination of both factual information and information that stems from more personal experiences.
An analysis of the effect of perceived reviewer expertise on OCR credibility shows that generic OCR’s are less credible coming from expert sources, and that product-specific OCR’s show a steep fall in credibility when they come from a non-expert source. Strikingly, the combination of generic and specific content from a non-expert source is clearly perceived as more credible when compared to the same OCR from an expert source.
5
II. Preface
This master thesis that lies before you is the end product of a 5-month process, in which I learned many things about marketing, academic research, and last but not least, about myself as a person. Moreover, it is the end of study career that began in 2007, as a Bachelor student International Business and Management. Over these first years, marketing soon became the most interesting and challenging study direction for me. I am very grateful that I had the opportunity to write my thesis on the topic that I selected as my first choice. The variety the field of research on word of mouth offers is perhaps best illustrated by the many different research directions that I discussed at the table of dr. Voerman. Together, we found a research direction that was both interesting and challenging. I would like to thank a few people for their help and support for the last months, as I would have never accomplished completing the thesis in my own. First of all, I would like to thank Liane Voerman for her guidance in selecting interesting research directions, and her very devoted and enthusiastic group sessions. Furthermore, I would like to thank my girlfriend Judith for her motivating support , and her refreshing and professional comments on my work. Also, I would like to thank my friends, who helped me to get my mind off the thesis when stress levels went up a little. And last but definitely not least, I thank my parents for their loving (financial) support during my whole study career.
Before starting the thesis, I was unsure whether I was going to enjoy it. In hindsight, I can say that I am surprised in how much I enjoyed it. Admitted, it is not my number one hobby, but the amount of satisfaction I got from creating my own piece of academic research is something I did not expect on forehand.
I hope you will enjoy reading my thesis. I surely enjoyed writing it for you.
Frank Posthumus
6
Contents
1. Introduction ... 9
1.1 An Evolution of WOM ... 9
1.2 Electronic Word of Mouth ... 10
1.3 Online Consumer Reviews ... 11
1.3.1 OCR Credibility ... 12
1.3.2 Factors Affecting OCR Credibility: Source Effects ... 13
1.3.3 Factors Affecting OCR Credibility: Message Content ... 14
1.3.4 Moderating Consumer Characteristics ... 14
1.4 Problem Statement ... 15
1.4.1 Research Questions ... 15
1.4.2 Narrowing the Scope: OCR’s on Objective Platforms ... 16
1.5 Academic Relevance ... 16
1.6 Research Structure ... 17
2: Literature Review ... 18
2.1 Perceived Reviewer Expertise ... 18
2.1.1 Display of Reviewer Identity Cues as Proxy for Source Expertise ... 19
2.2 Message Content ... 19
2.3 Covariates ... 21
2.3.1 Style of Processing ... 21
2.3.2 Product Involvement ... 22
2.4 Hypotheses and Conceptual Model ... 24
3. Methodology ... 25
3.1 Research Method ... 25
3.1.1 Content Manipulation ... 26
3.1.2 Source Manipulation ... 26
3.2 Survey and Scales ... 27
7
3.2.2 Independent Variables ... 28
3.2.3 Covariate 1: Style of Processing ... 28
3.2.4 Covariate 2: Product Involvement ... 28
3.3 Sample Characteristics ... 29
3.4 Manipulation Checks ... 30
3.5 Plan of Analysis ... 32
3.6 Validity of Measurement Scales ... 33
3.6.1 Reliability Problems ... 33
3.7 Normality Test ... 34
3.8 Homogeneity of Regression Slopes ... 34
3.8.1 Alternative Model Including Source Expertise ... 35
4. Results and Analysis ... 37
4.1 Final Analysis: ANCOVA ... 37
4.2 Alternative Model Including Source Expertise ... 39
4.3 Testing Hypotheses. ... 41
4.3.1 Hypotheses for Main Effects... 41
4.3.2 Hypotheses Including Covariates ... 42
5. Discussion ... 43
5.1 Discussion of Main Effects ... 43
5.2 Discussion of Moderating Influence of Covariates ... 44
5.3 Managerial Implications ... 45
5.4 Limitations and Future Research Directions ... 46
6. References... 47
Appendix 1: Scales and Validity Check ... 52
Appendix 2: Testing Assumptions ... 61
Appendix 3: ANCOVA Analyses ... 65
9
1. Introduction
Imagine the following scenario: You are planning to visit a restaurant to take someone out on a romantic date. Before making the reservation, you visit a well-known restaurant reviewing website, say iens.nl. There are several reviews written about that specific restaurant. However, the reviews vary greatly in length and specificity, and in many cases it is hard to judge whether the reviewer can be trusted to have an objective opinion. The website awards stars to reviewers that frequently post messages, but is that a proxy for a good and credible review? Also, some reviews are quite short and give only generic information saying how nice the atmosphere was and how nicely the host treated them. Others are lengthier and give an in-depth analysis of the food provided, without actually saying whether it was a nice romantic restaurant for a date. In other words, which reviews do you deem more credible, and therefore help you to make your decision on the reservation? Do the awarded stars and the specificity of the review play an important role?
The scenario described above poses one of the many issues with online consumer reviews that marketers and researchers are facing today. Although the scenario sketches a specific situation (i.e. restaurant choice), it can be replaced by many other decisions consumers make every day, for instance making a decision on purchasing an iPad, for which they then search the Internet for information.
1.1 An Evolution of WOM
Consumers are always seeking for information on products and services, for as long as we can remember. Before the introduction of the Internet, consumers’ information sources on products were limited to information provided from the side of the seller or producer, and direct interpersonal communication. The latter form, also known as Word of Mouth (WOM), has experienced a revolution in form and size since the introduction of the Internet. Before the age of the Internet began, WOM was only distributed through face to face personal interaction. The definition of WOM communication in the offline era, also referred to as ‘traditional WOM’, was very straightforward, and could best be described as spoken words with one friend or relative in a face-to-face situation, sharing product information with each other (Bickart and Schindler, 2001).
10 consumers are becoming more connected through social platforms as Twitter, Facebook, Pinterest and more, where they can share their opinions (Liu, 2006, Dellarocas, 2003). Thus, it seems that interpersonal communication is shifting from the offline to the online environment.
Interpersonal communication, and WOM as a part of it, is one of the most ancient and credible sources of information, it is even identified as one of the key drivers for successful business (Godes and Mayzlin, 2004) and to have a strong impact on consumers’ product judgements (Herr, Kardes and Kim, 1991). However, not all sources of WOM have an equally strong influence on the opinion of consumers. A recent study on eWOM by online marketing researcher eMarketer.com (2010) shows that consumers put more absolute trust in a blog (26%) or Facebook post (23%) from people they know, than in blogs or Facebook posts from brands or companies (11% and 9% respectively). This signals that there might be a significant distinction in trust between various sources of eWOM, and asks for a closer look.
1.2 Electronic Word of Mouth
Electronic word of mouth, or eWOM, is defined by Hennig-Thurau et.al. (2004, p.39) as “any positive
or negative statement made by potential, actual, or former customers about a product or a company, which is made available to a multitude of people and institutions via the internet”. Although the
principal is similar to traditional WOM, there are several differences that are important for a complete understanding of eWOM.
eWOM has many differences in relation to traditional WOM. Among these differences is the fact that eWOM is larger in number, and can be found through several different sources. Positive as well as negative messages are presented together, as opposed to traditional WOM where it is most often one single source at the time which is either positive or negative in valence (Chatterjee, 2001). Perhaps the greatest difference is the reach of each eWOM communication, which is potentially unlimited.
11 14 other consumers (Burton-Marsteller 2001), and this group rapidly expanded considered the growth of Internet penetration and computer literacy since then (Sun et.al., 2006).
Recent studies have proven that the majority of our purchase decisions are influenced by word of mouth communications, where especially eWOM is growing in volume and importance (e.g. Reitsma, 2010; Ratchford, Talukdar, and Lee, 2007). It is even argued that WOM communications are more credible, easier accessible through social networks, and of stronger influence on consumer behaviour than company-issued signals (e.g. TV or newspaper ads) (Mitchell and Khazanchi, 2010). It is clear that the growth of online review platforms has strategic implications for managers (Chen and Xie, 2008). Managers can use information from online reviews by e.g. analysing which product characteristics affect their online ratings, and how they can use these ratings as a marketing communications instrument. It is even found that eWOM can have a positive role in the overturning of product failure (Sridhar and Srinivasan, 2012).
1.3 Online Consumer Reviews
Online Consumer Reviews (OCR’s) are an often used form of eWOM, and can best be described as online evaluations or ratings, provided by users of a certain product or service (Cheung, Sia, and Kuan, 2012). This can be through common review sites (e.g. the Consumentenbond in The Netherlands), product-specific online platforms as IMDb (movies) or tweakers.net (electronics), or even through review boards on the seller’s company website. OCR’s provide a consumer-generated addition to the product-specific information from the side of the seller, and are more consumer-oriented. OCR’s describe product attributes in usage situations, and sometimes provide consumers with information which the seller is not always willing to mention or explain (Bickart and Schindler, 2001, Lee, Park, and Han, 2008). This new form of information poses interesting questions for firms as well as for researchers.
12 relative anonymity of eWOM it is hard to identify experts or opinion leaders, which means that consumers often have to rely on the opinions of complete strangers (Brown et.al., 2007).
Although the growing supply of online, seemingly independent, reviews appears to be a great asset for consumers, there may be some negative aspects to it. An overload of, often anonymous, online reviews makes it difficult for consumers to discern the source’s identity and origin, and these source effects have thus far not been extensively researched (Dou et.al., 2012). Early research identified that the degree of expertise of the reviewer/source is of influence on the perceived credibility of a WOM message (e.g. Sternthal, Dholakia, and Leavitt, 1977, Dholakia and Sternthal, 1977). The apparent doubts that are made on the credibility of OCR’s and eWOM in general, and the many factors that have proven to play a role in credibility of traditional WOM, give grounds to investigate on the credibility of WOM in the online environment, and in specific OCR’s.
1.3.1 OCR Credibility
The topic of credibility and trust in online environments has been subject to extensive research as the volume of available online information increased over the past decades, and issues of credibility of that information were put forward. The information consumers find nowadays on various topics is not always as reliable as they might think, and might even be perceived to be deceptive and manipulated by commercial parties (Cheng and Zhou, 2010). Research by Chau et.al. (2007) identifies that trust is a key factor in the online decision-making process, and that increased reliability of a website leads to increased purchasing intentions. This is an important finding, as the changes in information searching of consumers made that findings on persuasiveness of traditional WOM are less applicable to the current situation of online information searching.
Credibility of eWOM communications is vital for the actual adoption of the review or advice by the recipient. Recent empirical research by Cheung et.al. (2009) showed that the credibility of an eWOM message is a significant determinant of the eventual adoption of eWOM. Depending on the review topic, this finding is highly important in predicting actual levels of sales or product acceptance. This is supported by early communication research where highly credible message sources are significantly more influential in short term message adoption than sources with low credibility (Hovland and Weiss, 1951).
13 provides researchers with a wide range of possibly interesting research topics. Since there are many factors that can influence perceived OCR credibility, a choice has to be made. Therefore, this research will focus on factors that can influence OCR credibility, one of the factors being the expertise of the reviewer (i.e. the source). The following sections describe the factors chosen to be in the scope of this research.
1.3.2 Factors Affecting OCR Credibility: Source Effects
As consumers gather more information online than offline nowadays, there is a need to study the effectiveness of online communications in more detail. An important concept that emerges in literature on WOM is the concept of tie-strength. Tie-strength is described by Duhan et.al. (1997) as follows: “Tie-strength of a relationship is defined as strong if the source is someone who knows the
decision maker personally”. From this definition, it can be derived that the better the perceived
knowledge of the source, the higher the perceived credibility of the eWOM message will be.
Before the introduction of eWOM, Brown and Reingen (1987) researched the importance of tie strength in the flow of WOM. They found that strong ties are more likely to be to be activated for WOM than weak ties, and are therefore deemed more important for decision-making. Empirical evidence for this was again found by Bansal and Voyer (2000), who identified tie-strength as a significant influencer on purchase decision. When we import this view, we might assume that face-to-face communication (strong tie) is more influential in decision making than eWOM communications (weak tie), since online communication is often less personal (Brown, Broderick and Lee, 2007). The weaker ties between sender and receiver in an online environment, and the increased anonymity of eWOM communications, raise issues on the perceived credibility of OCR’s (Cheng and Zhou, 2010). Since the tie strength appears to be lower in an online context, information about the source of an OCR, e.g. stars awarded to the reviewer, might be a factor that can strengthen the feeling of familiarity with the source, and consequently increase OCR credibility.
14
1.3.3 Factors Affecting OCR Credibility: Message Content
OCR’s come in many different forms and sizes. The content of each OCR is different, as there is no standard format for an OCR, and is dependent on the style and preferences of the reviewer. Some OCR’s can be regarded as generic, and give a relatively superficial evaluation of a product or service. In contrary, other reviewers are keener to provide very factual and product-specific information. It can be assumed that generic information can be perceived as more subjective, which provides reasons to doubt the credibility of such an OCR. A review that is more logical and persuasive, and based on factual arguments is in general of higher quality than more personal, subjective OCR’s (Park, Lee and Han, 2007).
1.3.4 Moderating Consumer Characteristics
Besides the proposition that different types of OCR content can lead to differences in perceived credibility, there are different ways in which consumers interpret or evaluate an OCR. Two important determinants for this are involvement with the product and information processing style of the recipient. Product involvement determines the importance of specific aspects of an OCR. Petty and Cacioppo (1983, 1984) developed the Elaboration Likelihood Model, which describes how consumers with low involvement use the so-called peripheral route to process information, whereas involved consumers use the central route. Both routes focus on different characteristics of the message, e.g. involved consumer pay more attention to the quality of the arguments, and low-involved consumers are more focused on heuristics, for example a celebrity who endorses the product.
An equally interesting construct that determines how a message is perceived is the Style of Processing, first described by Chaiken (1980). She divides the human style of processing into a heuristic and a systematic style of processing. The former accounts for consumers that are focusing on non-content cues, e.g. a source’s identity, and the systematic style of processing pays attention to the quality of the arguments and messages.
15
1.4 Problem Statement
The topic of eWOM and in specific in offers many opportunities for researchers. However, a choice has to be made to limit the scope of this research due to limits in time and space available. This very research pertains to the issue stated in the following problem statement:
How do the perceived level of expertise of a reviewer and the content characteristics of a review affect the perceived credibility of online consumer reviews, and how is this effect moderated by a consumers’ product involvement and style of information processing?
The purpose of this research is to investigate whether the perceived expertise of the source (i.e. the reviewer), and the level of specificity of the OCR message are of influence on the perceived credibility of the OCR. The chosen moderating variables will help to find out whether product involvement and style of processing of a recipient is of influence in how the OCR is perceived. The chosen dependent variable is perceived OCR credibility, since this is proven that credibility leads to increased sales.
1.4.1 Research Questions
To find answers for this problem statement, the following research questions must be answered: RQ1: How does the perceived expertise of the reviewer influence perceived OCR credibility?
RQ2: How does the specificity of the content of an OCR have an effect on perceived OCR credibility? RQ3a: How does the degree of product involvement moderate the relationship between reviewer expertise and perceived OCR credibility?
RQ3b: How does the style of processing of an OCR recipient moderate the relationship between reviewer expertise and perceived OCR credibility?
RQ4a: How does the degree of product involvement moderate the relationship between OCR content and perceived OCR credibility?
RQ4b: How does the style of processing of an OCR recipient moderate the relationship between OCR content and perceived OCR credibility?
16 To find answers on these questions, respondents will be asked to answer questions, after they have been shown an OCR. They are asked to imagine themselves being in a specific situation where they have to make a purchase decision. The sketched situation will pertain to a certain product category that should be familiar to all respondents.
1.4.2 Narrowing the Scope: OCR’s on Objective Platforms
One other aspect of OCR’s that can be regarded to have source effects is the type of website or platform the OCR is displayed on. There is research that supports that not all eWOM platforms can be regarded as equally influential or credible. For example, there are many high-regarded independent review sites, and also company websites that facilitate consumers to post reviews and have discussions. However, there are many websites that consumers are not greatly familiar with, but that do provide OCR’s. The question now is whether they are regarded as equally influential, thus, does the eWOM platform influence the perceived credibility of the message? Dou et.al. (2012) state that the identity of reviewers can be overlooked or lost in the overload of information present on the website, causing the characteristics of the website to become a more important cue for credibility. Research by Brown et.al. (2007) adds to this, as they prove that consumers see websites as being a primary actor in online networks, suggesting that credibility partly stems from the website characteristics.
Although these questions on type of platforms are interesting, the majority of the research done so far point in the same direction, namely that objective platforms are seen as more credible (Senecal and Nantel, 2002 ; 2004), and that reviewers that were internally motivated (e.g. by monetary rewards) are found less credible (Rifon et.al., 2004). Therefore, OCR platform will not be further elaborated on in this research.
1.5 Academic Relevance
17 evaluation of eWOM communications as they are in WOM (Cheung et.al, 2009). Furthermore, so far no study that researched message content seems to have studied the effects on credibility of both weak (non-factual and intangible) and strong (factual and tangible) arguments combined. Instead, most studies used message sidedness and/or consistency as varying message characteristics (e.g. Cheung, Sia, and Kuan, 2012), and this research tries to fill in this void.
1.6 Research Structure
18
2: Literature Review
The following chapter will review the extant literature on the various topics that are discussed in this research. First, the literature on electronic Word of Mouth will be discussed, followed by the more in depth review of literature on reviewer characteristics, message content, and OCR Credibility. The possible moderating effect of product involvement and style of processing will also be discussed. The theoretical background will be used to formulate hypotheses, which are the basis for the later analysis of the results. The concept of credibility is highly complex, and cannot be determined by one factor alone. Source, receiver, and message characteristics interact in the eventual determination of the credibility of a message (Wathen and Burkell, 2002), and the following chapter provides a theoretical perspective on a selection of potentially influential factors.
2.1 Perceived Reviewer Expertise
One of the determining factors in overall perceived credibility of a message is source credibility (Buda and Zhang, 2000). A definition of credibility proposed by Freedman et.al. (1981) makes this more clear, as he describes credibility of communication as how expert the communicator is perceived in the area of concern, and also as how trusted by the individual receiving the communication. More research indicate that expertise of a communicator can indeed influence how a message is perceived (e.g. Sternthal et.al., 1978, Dholakia and Sternthal, 1977). The effect of source credibility in online reviews is an important aspect in the prediction of the overall credibility of OCR’s.
19
2.1.1 Display of Reviewer Identity Cues as Proxy for Source Expertise
The anonymity offered by the Internet makes it doubtful whether every review is perceived as equal in quality, as all Internet users are free to post whatever they desire. Hence, consumers approach OCR’s with more care and are cautious to put trust in a source (Kusumasondjaja et.al, 2012). The credibility of a message, and whether a reviewer is regarded as an expert on a specific subject, is determined from an evaluation of the knowledge the reviewer holds (Gotlieb and Sarel, 1991). However, OCR’s are often posted in environments that hold little cues to determine the knowledge and background of the reviewer, and subsequently make it more difficult to identify experts and opinion leaders (Brown et.al., 2007). A more salient cue to determine the source expertise may be a rating the reviewer is given by other users. This reviewer reputation system is used to convey credibility information about the reviewer. A common used cue is the posting history of the reviewer, or a rating awarded by the administrator of the platform (Cheung et.al, 2009). An example of this system is the restaurant reviewing website iens.nl, where reviewers are awarded stars that resemble their number of posts on the website. The accompanying text (e.g. apprentice taster for zero stars, top taster for five stars) indicates their level of expertise in fine dining. Although the number of posts does not necessarily imply that the reviewer is an expert on the subject, these cues can function as simple heuristics that help eWOM users establish trust in a review.
The literature provides several factors that culminate in source credibility. To limit the scope of this research, we focus on the perceived level of expertise of a reviewer. From the extant literature on the importance of source credibility, it can be concluded that the opinions of more experienced reviewers can be expected to have greater influence than those of less experienced or novice users of a product. Therefore, the following is hypothesized:
H1: A higher degree of perceived source expertise leads to a higher perceived overall credibility of the
OCR.
2.2 Message Content
20 Park, Lee and Han (2007) state that the credibility of online reviews is partly determined by the quality of the review based on the review’s content, where a high-quality review features content that is more logical and persuasive, and is supported by factual information. This last observation is in accordance with the arguments of Hunt, Smith and Kernan (1985), who argue that the content of a message is perceived as high in quality when the message is stated in terms of tangible product features. Opposed to these high-quality messages are the low-quality messages, which are stated in intangible product features. When these views are translated to OCR’s, it is arguable that reviews with a more factual message content should be perceived as more credible than reviews that feature generic arguments that are more superficial. A study that confirms the proposition that strong messages positively affect OCR credibility is done by Cheung, Sia, and Kuan (2012). In their study, the strength of a message is the review cue with the strongest influence on OCR credibility. This finding is translated to this very research as the basis for the choice of message content as a possible determinant of OCR credibility. A manipulation of the OCR content through using generic (weak) and product-specific (strong) arguments, both separate and combined, should create a difference in perceived OCR credibility.
Furthermore, Chu and Kamal (2008) argue that there is the possibility of interaction between perceived source credibility and the content of a message. The authors suggest that message recipients put less effort in scrutinizing a message when a source is perceived as credible. In contrast, recipients are unsure of the credibility of a message when the source’s expertise is unclear, and consequently sense a higher need to scrutinize the message more carefully, in order to evaluate the validity and quality of the message. The same authors propose that manipulation of the argument quality is a sound way to evaluate the believability and acceptance of a message. When looking at OCR’s in specific, this means that there should be differences in perceived credibility of an OCR depending on the strength of the arguments, and possible interaction exists with the perceived reviewer expertise. OCR’s differ greatly in type and length of message, and a possible way for manipulating the message quality is to make a distinction between generic information about a product or product experience, versus a more elaborate evaluation of product-specific information. The former can be categorized as being an OCR with weak arguments, the latter as an OCR with strong arguments. Therefore, the following is hypothesized:
H2: The higher the degree of product specific information, the more the OCR will be perceived as
21
2.3 Covariates
This research investigates not only the effects of source expertise and message content on OCR credibility, but also how this effect is moderated by specific characteristics of the review recipients. The chosen moderating variables for this research are a recipient’s (1) style of processing, and (2) product involvement. These covariates are based on findings presented in literature on consumer psychology, and can be related to the perceived credibility of reviews in an online setting.
2.3.1 Style of Processing
The first covariate in this research pertains to the style of information processing of the OCR reader. Research by Chaiken (1980) puts forward the two dominating styles that consumers have for information processing: the heuristic and the systematic style of processing, or the HSM model. The characteristics of both styles are described by Chaiken (1980, p.752) as the following:
Heuristic Style of Processing: “According to the heuristic view of persuasion, recipients exert
comparatively little effort in judging message validity: Rather than processing argumentation, recipients may rely on (typically) more accessible information such as the source’s identity or other non-content cues in deciding to accept a message’s conclusion.”
Systematic Style of Processing: “According to a systematic view, recipients exert considerable
cognitive effort in performing [message evaluation]: They actively attempt to comprehend and evaluate the message’s arguments as well as to assess their validity in relation to the message’s conclusion.”
22 can occur, indicating that recipients may reject useful reviews they would otherwise have accepted, vice versa.
When speed, and not accuracy, of information seeking is paramount, one of the heuristic cues that can be relied on by consumers are source factors. The impact of source factors, of which source expertise is a specific part and subject to this research, has a direct influence on message acceptance when a recipient uses a heuristic style of processing. In contrary, source factors have only an indirect effect on message acceptance when a recipient uses a systematic style of processing (Chaiken, 1980). This view is shared by Cheung, Sia, and Kuan (2012), who argue that source factors are important when involvement and willingness to put in effort is low, and that message content is important when involvement and willingness to put in effort are high.
The mere opposite from the heuristic view is the systematic view. With the systematic style of processing, it is argued that not source characteristics, but argument and message quality are determinants of the acceptance of a message (Chaiken, 1980; Priester and Petty, 2003; Chu and Kamal, 2008). Recipients of a message that use a systematic style of processing actively attempt to link the arguments in the message to the knowledge they already possess (Bohner, Moskowitz, and Chaiken, 1995) When translated to this specific research, it is proposed that the content of a message is of stronger influence on OCR credibility for recipients that use a systematic style of processing. Due to the conflicting nature of both styles of processing, and the estimated difference in importance of source expertise as well as message content on the perceived credibility of an OCR, the following is hypothesized:
H3a: A high degree of product-specific information will lead to a higher perceived OCR credibility for
recipients with a systematic style of processing than for recipients with a heuristic style of processing.
H3b: High perceived reviewer expertise will lead to higher perceived OCR credibility for recipients with
a heuristic style of processing than for recipients with a systematic style of processing.
2.3.2 Product Involvement
23 product. The reasoning behind the model is that consumers are seen as ‘cognitive misers’ who use their ability to reason only for messages that seem relevant to them, and that they use mental shortcuts for less relevant messages, which saves energy and time. The model can be used to estimate how likely a consumer is to intensively elaborate on the information he receives.
It is argued that message recipients, thus consumers that read an OCR, who are less involved in the product category and who are relatively novice users of a product place more value on information that the ELM model considers as peripheral information (Petty and Cacioppo, 1986). This information can exist of non-content cues such as length of the argument, and information disclosed about the identity of the source, e.g. its level of expertise. The same authors argue that consumers engage in more detailed elaboration when they have higher motivation and ability to elaborate on the provided information.
The ELM model of Petty and Cacioppo (1981, 1986) is used by Cheung, Sia, and Kuan (2012) to study the effects of involvement on OCR credibility. This study takes the principals from the ELM model, which dates from the time were only traditional WOM existed), to an online context. Among other variables, they propose that the argument quality is cue processed through the central route, and that source credibility is a peripheral cue. This assumption matches the variables in the scope of this research, since message content is a co-determinant of argument quality, and perceived reviewer expertise is that for source credibility.
The proof that involvement has a role in which cues consumers deem important, is highly applicable to this very research. The earlier hypothesized effects of OCR content and reviewer expertise on OCR credibility might be moderated by the degree of involvement a consumers has with a product. The studied literature provides grounds to believe that involvement in the product that is discussed can have a moderating role in the effect of reviewer expertise on OCR credibility. Therefore, the following is hypothesized:
H4a: An OCR with a high level of product-specific information will be perceived as more credible than
an OCR with generic information when a recipient has a high level of product involvement.
H4b: An OCR from a source with a high level of perceived reviewer expertise will be perceived as
more credible than an OCR from a source with low perceived reviewer expertise when the recipient has a low level of product involvement
24 when the likelihood to elaborate goes up or down. In contrast, HSM suggests that the impact of both heuristic and systematic processing can increase when the elaboration likelihood increases. As long as there is no conflict between the two styles of processing (e.g. when an expert source posts a weak message), HSM holds that a heuristic style of processing adds to the impact of any systematic processing that occurred before (Petty, 1994). In summary, IN ELM there is either one or the other processing route with a dominant impact, whereas HSM offers the opportunity for both styles of processing to co-occur and have a joint impact on message evaluation (Petty, 2004).
2.4 Hypotheses and Conceptual Model
Table 1: Overview of Hypotheses
Figure 1: Conceptual Model
Hypothesis
H1 A higher degree of perceived reviewer expertise leads to a higher perceived overall credibility of the OCR.
H2 The higher the degree of product specific information, the more the OCR will be perceived as credible.
H3a A high degree of product-specific information will lead to a higher perceived OCR credibility for recipients with a systematic style of processing than for recipients with a heuristic style of processing
H3b High perceived reviewer expertise will lead to higher perceived OCR credibility for recipients with a heuristic style of processing than for recipients with a systematic style of processing
H4a An OCR with a high level of product-specific information will be perceived as more credible than an OCR with generic information when a recipient has a high level of product involvement.
25
3. Methodology
This chapter outlines and describes how the research will be conducted. First, the type of research will be described and introduced, followed by the data collection method and types of analyses that are used. An overview of the measures for the various concepts in the conceptual model will be provided, along with the questions and scales that are used to compose the survey for the data collection.
3.1 Research Method
This research will be using quantitative data collected through the use of a between-subject experimental design. A scenario study with a 3 (message content) x 2 (source expertise) between-subjects factorial design is used to study the research questions. The design will also include two covariates, i.e. a consumers’ 1) style of processing and 2) product involvement. A survey will be digitally provided to a random sample of Dutch people.
The survey will be composed and distributed online, with the clear advantages it has over manual distribution in terms of flexibility, speed, reach, and ease of data processing through statistical computer software. The survey will be distributed mainly via social media like Facebook and Twitter, and furthermore via email. By sharing a link to the survey, the data collection should enable respondents to share it within their networks and that will help to reach the necessary number of respondents within reasonable time.
The survey contains the elements earlier described in the conceptual model (see chapter 2, figure 1). Each respondent will see one review
written by a consumer. Within this review, message content and source credibility are manipulated, resulting in six different reviews. The review will be on the iPad 2, since products in this category are likely to be known in the sample group. Following the conceptual model and hypotheses, the manipulations will concern the message content and the expertise of the source. The six groups that can be distinguished due to the manipulations are listed in Table 2. The exact ways in which the manipulations are done are explained below.
Table 2 Manipulation Groups
Survey Group # OCR Content
Generic Specific Both
Source Expertise Expert 1 2 3
26
3.1.1 Content Manipulations
The OCR that is used for this research is based on a review of one of Wehkamp.nl, one of the largest online stores in The Netherlands. The content manipulation has to ensure that there will be one OCR with only generic information about the product, without giving much detailed information about the iPad, one OCR with exclusively product-specific information, and one with both generic and specific elements. The reviewer in this generic OCR states how happy he is with the product, and that the iPad really is an enrichment to his life. This OCR is short in length, and is intended to be perceived as an OCR with low message and argument quality (Hunt, Smith, and Kernan, 1985). The second OCR will feature much more detailed information about the iPad by giving information about new features and enhanced performance through new microchips. This OCR has more length, and is supposed to be the OCR with high message and argument quality. The third OCR will be the OCR with elements of both the generic review and the product-specific review.
This set-up for the content manipulation is similar to the method used by Hunt, Smith and Kernan (1985). The authors suggest that strong arguments contain tangible and verifiable claims about the product, i.e. the product-specific OCR, and that weak arguments contain intangible product claims, i.e. the generic OCR. The textual elements are selected using this claim, and examples of strong and weak messages provided by Park, Lee and Han (2007).The third OCR containing both generic and specific elements is an addition to the research of both previously mentioned studies, to see whether a combination of the two argument types yields surprising results.
3.1.2 Source Manipulation
The other factor subject to manipulation is the source of the OCR message, i.e. the reviewer or writer. Several items were added to the basic review, in order to ensure the correct manipulation effect. The basic review from Wehkamp.nl did not include a star rating or any other clues about the identity of the reviewer, as it only featured an avatar, a reviewer nickname and an age category (e.g. 40-49). In addition to the already present clues on source identity, three items were added:
1. A star rating of the reviewer, accompanied by a short descriptive text. The OCR with the manipulation for high source credibility was awarded 5 stars and the description ‘Top Reviewer’. The second OCR, aimed to be perceived as a source of low expertise, was awarded zero stars and was described as ‘New Reviewer’.
27 3. The field of expertise of the reviewer. The expert source was given the tag ‘field of expertise:
electronics’, whereas the non-expert source received the tag ‘field of expertise: unknown’. As an example, one of the manipulated OCR’s is displayed in Figure 2. This is the OCR for group 3, this with an expert source and both generic and specific elements. In Figure 2, the generic elements are circled with a blue line, and the specific elements are circled with a red line. The manipulation for source expertise has a green circle. A complete overview of the different manipulated OCR’s belonging to the 6 groups can be found in Appendix 4.
3.2 Survey and Scales
The different concepts and variables are all measured trough different scales. Each of these scales is derived from previously published literature on similar subjects, and are proven to be effective measured of the various concepts. After the data has been collected, the scales will be tested for internal consistency using the Cronbach’s Alpha statistic. A summary of the variables, items and reliability scores are displayed in Table 3. A complete overview of all used scales and questions can be found in Appendix 1.
3.2.1 Dependent Variable
The dependent variable in this research is perceived OCR credibility. The measure for this variable is derived from Tate et.al. (2006), who used it to measure the credibility of blogs. The overall concept of credibility is measured on 9 items, and is split up into two parts: Source credibility and content
28 credibility. Source credibility has 5 items, and content credibility has 4 items. All items are phrased into questions that can be answered on a 7-point Likert scale.
3.2.2 Independent Variables
The two independent variables in this research are 1) message content and 2) perceived reviewer expertise. Both are predetermined, or fixed, variables, and are correspondingly processed in the survey. Following the 3 x 2 factorial design, each of the six groups is shown a different condition (see Table 2). Message content is a fixed number of either 1, 2 or 3, corresponding to with 1) generic, 2) product specific, or 3) a combination of both types of information. Perceived reviewer expertise is a predetermined binary variable with 1) an expert reviewer and 2) a non-expert reviewer.
3.2.3 Covariate 1: Style of Processing
The first covariate is the style of processing as described by Chaiken (1980). The two styles, heuristic and systematic, are measured through a set of items as used by Griffin et.al. (2002). These items were selected from various previous researches, and ask respondents to rate their agreement or disagreement on a 7-point Likert scale. The questions pertain to the way the respondents deal with information that is presented to them, and are proposed as statements to which they can rate their level of agreement.
3.2.4 Covariate 2: Product Involvement
29
Table 3: Overview of variables, constructs, measurement scales, and reliability analyses
Variable/Construct Measurement Cronbach’s Alpha Dependent Variable Source Credibility 1: Trustworthy 2: Credible 3: Honest* 4: unquestionable 5: Conclusive
*This item was deleted from the scale to increase the reliability
7-point Likert Scale .824
After deletion item 3: .841 Dependent Variable Content Credibility 1: Believable 2: Authentic 3: Reasonable 4: Convincing
7-point Likert Scale
.812
Manipulation Checks
Reviewer Expertise:
- I think the reviewer is an expert on this subject
Message Content
- The content of this review gives me much detailed information on the product’s specifications
7-point Likert Scale
Systematic Style of Processing 5 items, e.g.:
1: After I encounter information about this topic, I am
likely to stop and think about it
2: If I need to act on this matter, the more viewpoints
I get the better
7-point Likert Scale .706 Heuristic Style of Processing
4 items, e.g.:
1: When I encounter information about this topic, I
focus on only a few key points
2: If I have to act on this matter, the advice of one
expert is good enough for me
7-point Likert Scale .466 Product Involvement
1: I have a strong interest in this product category 2: This product category is very important to me 3: This product category matters a lot to me
7-point Likert Scale .923
3.3 Sample Characteristics
30
Valid Missing Age 189 8
Gender 188 9
Table 4: Age and Gender Count
The average age of all respondents is 32.91 years, and the female part of the sample is slightly older. 9 respondents did not fill in their age. An overview of age statistics sorted by gender can be found in Table 5. Furthermore, Table 6
provides an overview of the sample characteristics sorted for the six groups of the 3 x 2 factorial design.
N Mean Age Minimum Age Maximum Age
Male 122 31,65 19 73
Female 66 35,24 20 67
Total 188 32,91
Table 5: Age Statistics, sorted by gender
Group Source/Content % Male % Female Average Age 1 Expert/Generic 64.7 35.3 31,74 2 Expert/Specific 60.0 40.0 33,44 3 Expert/Both 67.7 32.3 30,90 4 Non-Expert/Generic 78.1 21.9 33,68 5 Non-Expert/Specific 67.7 32.3 35,35 6 Non-Expert/Both 48.3 51.7 31,93 Total: 64.6 35.4 32,86
Table 6: Age and Gender characteristics per group
3.4 Manipulation Checks
Inherent to the 3 x 2 factorial design are manipulations to the survey that is presented to the respondents. For this research, this means that manipulations on as well the perceived level of expertise of the source as the specificity of the OCR content (see section 3.1.1 and 3.1.2). To test whether the manipulations were successful it is necessary to perform a series of statistical tests. The full output for these tests can be found in Appendix 1 (tables 10-13).
The first manipulation, perceived source expertise, was done using a fixed variable with either a 1 or a 2 as value. A value of 1 indicated an expert source, and a value of 2 represented a non-expert source. The successfulness of the manipulation was tested by performing a 2 x 3 ANOVA test, to check for possible effects that the manipulation of the OCR content might have on the perceived level of expertise of the source. The question used for this manipulation test is formulated as “I think
the reviewer is an expert on this topic”. The results of this test, displayed in table 7, show a negative
31 Furthermore, the 3 x 2 ANOVA test showed a significant effect of the message content manipulation on the manipulation check question for source expertise, F(2, 191)=57.622, p=.000. This test result can explain how the manipulation for perceived reviewer expertise has failed, as the respondents apparently included the OCR content in their evaluation of the level of expertise of the reviewer. This is merely a suggestion, and cannot be taken into account in the proceeding of the research, since the data collection already took place. Since this can provide useful insights for future research, it is briefly discussed in chapter 5.
The failed manipulation has consequences for the later analysis of results. In the case of a successful manipulation of message content and an unsuccessful manipulation of source expertise, there will be two models that have to be tested. Firstly, there will be tests on a model that only includes message content as an independent variable. Secondly, there will be an alternative model with both message content and source expertise included. Although the latter model is not fit for solid conclusions to be drawn, it can still provide leads and evidence for future research purposes.
Furthermore, the manipulation of the OCR content must be tested. Similar to the manipulation test for source expertise, a 2 x 3 ANOVA test is used, where only a significant main effect of message content is expected. The question for this manipulation check is “This review provides me with much
product-specific information”. The results of this test are displayed in Table 8. The manipulation for
message content is successful, as F(2, 191)=48.695, p=.000. The LSD Post-hoc test shows similar significant results. These results indicate that the manipulation of the OCR content was successful, and that the respondents rated the specificity of the OCR as was intended when the survey was designed. The means are M=1.630 (generic content), M=4.065 (specific content), and M=3.922 (generic and specific content).
Table 7: Manipulation Check for Source Expertise
Variable Df F Sig.
Expert 1 2.772 .358
Content 2 57.622 .000
Expert*Content 2 .101 .970
Error 191
LSD Posthoc Test Mean difference Sig.
32 The effect of source expertise on the manipulation check for message content is in line with the findings on the failed manipulation of this variable, as is described above. The score from the 2 x 3 ANOVA test, F(1, 191)=2.772, p=.001, is significant, which can be seen as confirmation for the assumed wrong formulation of the manipulation check for source expertise. Implications and possible causes of this will be discussed in chapter 5.
The overall evaluation of the manipulation checks show varying results. It can be concluded that the manipulation on expertise is insufficient, and that no valid conclusions can be drawn on any effects of perceived reviewer expertise on OCR credibility. In contrary, the manipulation for message content was successful, providing grounds to make valid assumptions about the effect of OCR content on OCR credibility.
3.5 Plan of Analysis
In the proceeding of this chapter the validity of the measurements scales will be tested on internal reliability, and possible validity problems will be solved. A normality check will be performed, and there will be a test on the homogeneity of regression slopes. This latter test is necessary to check whether there are interaction effects between independent variables and covariates. Following these tests, there will be a plan of analysis for the final analysis in chapter 4. The most essential figures will be presented alongside the text, and a more detailed overview of the analysis can be found in Appendix 1 and 2. Chapter 4 will feature the final analysis, which will be done using the ANCOVA method. The output for these analyses is displayed in Appendix 3.
It is important to note that the manipulation check has consequences for the structure of the analysis. Since the manipulation for perceived reviewer expertise has not been successful, there will
Table 8: Manipulation Check for Message Content
Variable Df F Sig.
Expert 1 11.659 .001
Content 2 48.695 .000
Expert*Content 2 .732 .482
Error 191
LSD Posthoc Test Mean difference Sig.
33 be separate sections in both the analysis of homogeneity of regression slopes and in the final analysis in chapter 4. Practically, this implicates that there will be a main model with only the independent variable message content, and an alternative model where source expertise is added.
3.6 Validity of Measurement Scales
The different concepts in this research are measured on scales derived from earlier researchers on the same topics (see chapter 3). The scales have to be tested whether these scales are accurate enough to use for the analysis of the data. This is done by using Cronbach’s Alpha reliability test for internal validity.
As can be seen in Table 3, nearly all scales meet the required threshold of .600 for the Cronbach’s Alpha test. The measurement scale for ‘source credibility’ scored higher after the deletion of the third item, which was ‘I think this review is unquestionable’. Therefore, the item was deleted from the scale, and a new variable was created, called ‘SourceCredibilityNieuw’. This variable consists of items 1, 2, 4 and 5, and will replace the initial variable for source credibility in the proceedings of the data analysis.
3.6.1 Reliability Problems
34 of processing. From here on, this item will be referred to as ‘heuristic’ or ‘heuristic style of processing’, as it is the closest representation for the covariate heuristic style of processing.
3.7 Normality Test
One of the assumptions for ANCOVA is a normal distribution of the data. To test for this normality, skewness and kurtosis are the important statistics, and these should be in order for as well the independent variables as the covariates. Skewness indicates whether the data is asymmetrical and thus deviates from a normal distribution. The kurtosis statistic shows whether the distribution of the data is flattened or peaked. it is widely accepted that when the values for skewness and kurtosis fall between -1 and 1, normality is assumed. The values for skewness and kurtosis are listed in Table 9. A more elaborate display of these statistics can be found in Appendix 2.
Table 9: Normality Test
Variable Skewness Std. error Skewness Kurtosis Std. error kurtosis Source Credibility .181 .175 -.539 -.347 Content Credibility .048 .173 -.389 .345 Covariates Product Involvement -.258 .173 -.658 .345 Systematic Processing -.470 .175 -.097 .348 Heuristic Processing -.413 .173 -1.050 .345
The results from Table 9 suggest that the majority of the data is normally distributed, except for the proxy for heuristic processing (kurtosis = 1.050). Since there have been issues with the reliability of this scale before (see 3.4.1, it is decided to allow this score, and normality for all variables is assumed.
3.8 Homogeneity of Regression Slopes
35 The test for homogeneity of regression slopes is performed by making a custom model featuring all possible interaction effects of the independent variables and the covariates. Note that interaction effects between the covariates are not included in the model, and also note that the independent variable ‘Source Expertise’ was excluded from the main research due to a failed manipulation. This leaves ‘Content’ as the only independent variable. Table 10 provides an overview of the interaction effects between covariates and the independent variable, sorted for dependent variable ‘Source Credibility’. Table 11 displays the statistics for dependent variable ‘Content Credibility’.
Table 10 and 11 show that there are no significant interaction effects between the covariates and the independent variables. This means that the relationship between the covariates and the dependent variable is not significantly different across all groups, which is important to carry on with this research. Now that the assumption for normality is met, and there is homogeneity of regression slopes, it is possible to perform the final ANCOVA analysis (chapter 4)
3.8.1 Alternative Model Including Source Expertise
The test for homogeneity of regression slopes for the alternative model including the source expertise is similar to the model with only message content as independent variable. Table 12 shows that the alternative model does not show signs of interaction between the covariates and the independent variables, as all p-values are not significant at the .05 level. Therefore, this model is also fit for the ANCOVA analysis, which will be performed and described in chapter 4.
Content Credibility Df F Sig.
Content*Involvement 2 .061 .941
Content* Heuristic 2 .067 .935
Content*Systematic 2 .486 .616
Error 181
Table 11: Interaction Effects Between IV and Covariates for Source Credibility
Source Credibility Df F Sig.
Content*Involvement 2 .191 .826
Content* Heuristic 2 1.213 .300
Content*Systematic 2 1.798 .169
Error 178
36
Table 12 Interaction Effects Between IV and Covariates for Source and Content Credibility
Source Credibility Df F Sig.
Content*Involvement 2 .033 .967 Content* Heuristic 2 .814 .445 Content*Systematic 2 1,640 .197 Expert*Involvement 1 .019 .889 Expert*Heuristic 1 .016 .899 Expert*Systematic 1 .728 .295 Error 166
Content Credibility Df F Sig.
37
4. Results and Analysis
This chapter will describe the final data analysis using the ANCOVA method, and will provide the analysis of the results and the necessary figures and tables. As indicated earlier, there are two models that are tested: 1) a model with only message content as independent variable, and 2) a model where the independent variable source expertise is added.
4.1 Final Analysis: ANCOVA
The analysis of covariance, or ANCOVA, is used to test interaction effects between independent variables and a dependent variable, with the possibility for other factors, i.e. the covariates, to influence this interaction as well. Initially, this research included two independent variables and two covariates. However, the manipulation for source expertise was not significant, and therefore this independent variable cannot be a part of the main analysis. Despite this fact, it is still useful to look at the effects of perceived source expertise on OCR credibility, since the literature suggests that there is indeed a relationship between the two variables. For the sake of validity of the research, there will be two analyses: one analysis with only message content as independent variable included, and one alternative analysis where source expertise is included together with message content. Although no conclusions can be based on a model including this variable due to the failed manipulation, it can provide clues on a possible interaction effects that can be researched more thoroughly in the future.
Since the dependent variable ‘Perceived OCR Credibility’ is divided into Source Credibility and Content Credibility, there will be two ANCOVA tests. The first test will test for effects on Source Credibility, and the second test will test for results on Content Credibility.
38 The right part of Figure 31 illustrates the effect of Message Content on dependent variable Content Credibility. Interestingly, the mean for generic message content is higher than for product-specific content, and the highest mean is for an OCR where both text elements are present. It is interesting to note that this latter observation is the case for Content Credibility as well as for Source Credibility.
Figure 3: Visual representation of ANCOVA output
1
Covariates appearing in the Source Credibility model are evaluated at the following values: Involvement = 4,1614, Proc.Systematic = 4,8442, Proc. Heuristic = 4,44 Covariates appearing in the Content Credibility model are evaluated at the following values: Involvement = 4,1779, Proc.Systematic = 4,8518 Proc. Heuristic = 4,46
Table 13: Tests of Between-Subjects Effects for Source Credibility and Content Credibility
Source Credibility Df F Sig.
Corrected Model 5 1.668 .144 Intercept 1 23.544 .000 Involvement 1 .752 .387 Proc. Systematic 1 3.892 .050 Proc. Heuristic 1 .580 .447 Content 2 .899 .413 Error 184 Total 190 Corrected Total 189
Content Credibility Df F Sig.
39
4.2 Alternative Model Including Source Expertise
As described earlier, the manipulation for expertise was insufficient to keep this independent variable in the scope of this research. There are, however, indications that perceived reviewer expertise is of possible influence on OCR credibility. Therefore, it is decided to show an alternative analysis, where this independent variable is still included.
Firstly, the test for interaction between covariates and independent variables is repeated. This test for homogeneity of regression slopes, displayed in paragraph 3.8.1 and Appendix 2 (Tables 4 and 5), showed no significant results. This confirms that the covariates and independent variables can be used for the ANCOVA analysis. The ANCOVA test is performed, and the results are displayed in Table 14. The full output of this test are displayed in Appendix 3 (Tables 3 and 4).
The effects of source expertise on the dependent variables Source Credibility and Content Credibility are F(1, 181)=.041, p=.840, and F(1, 184)=.527, p=.469 respectively, and are therefore not significant at the .05 level. This indicates that there are no significant effects of perceived reviewer expertise on the credibility of an OCR. Furthermore, the interaction between the independent variables
Table 14: Tests of Between-Subjects Effects for Source Credibility and Content Credibility with Source Expertise included
Source Credibility Df F Sig.
Corrected Model 8 1.404 .197 Intercept 1 21.477 .000 Involvement 1 1.020 .314 Proc. Systematic 1 3.761 .054 Proc. Heuristic 1 .251 .617 Content 2 .922 .400 Expert 1 .041 .840 Content*Expert 2 1.434 .241 Error 181 Total 190 Corrected Total 189
Content Credibility Df F Sig.
40 (Content*Expert) has no significant effect on Source Credibility (F(2, 181)=1.434, p=.241) or Content Credibility (F(2 ,184)=.822, p=.441). Similar to the ANCOVA analysis with message content as only independent variable, a visual representation of the results is provided in Figure 4.
Figure 4: Visual representation of Source Expertise effects combined with Message Content
From Figure 4 it can be derived that an expert source is perceived to be the most credible (on both Source and Content Credibility2) when the OCR contains many product-specific elements. However, when the same message is provided by a source that cannot directly be classified as an expert, the credibility of the OCR is far less, as is indicated by the downward slope of the green line in Figure 4. The exact opposite holds for OCR’s with both generic and product-specific elements. The yellow line corresponding to this type of OCR has a steep upward slope, indicating that non-expert sources are perceived as more credible when they combine generic elements with specific information about the product.
2
