What affects the source credibility of an online consumer review?
How does homophily based on different types of source characteristics, for either search products or experiential products, affect the perceived source credibility and of an onlineconsumer review?
By
LAURA KRIJNSE LOCKER Meeuwerderweg 57 9724 EM Groningen (06) 13985659 l.m.krijnse.locker@student.rug.nl student number S2516527 January 2019 University of Groningen Faculty of Economics and Business
MSc. Marketing Management
3 ABSTRACT
Due to the increasing amount of falsified information on the internet the trustworthiness of online information is affected. Therefore, credibility of online consumer reviews is also affected. Since online consumer reviews have been found to affect the performance of a business, it has become topic of interest with increasing importance for companies.
This study tested the effect of the presence of a similar demographic characteristic (age) and the effect of the presence of similar psychographic characteristics (based on VALS-typology) on perceived source credibility. This was tested for both a search product (sport shoes) and an experience product (restaurant visit).
A digital survey was conducted among 282 participants. The collected data is analyzed with various statistical tests such as ANOVA and Regression analysis. The results show that the presence of psychographic characteristics similar to those of the respondents attached to an online consumer review (OCR) leads to higher perceived source credibility. Additionally the results show that this relationship is fully mediated by perceived homophily (similarity)
The findings of this study contribute to the research field of electronic word of mouth (eWOM), more specifically, online consumer reviews. This study provided insights that can be used by businesses to enhance the perceived source credibility of the online consumer reviews of their products.
Keywords: Electronic word of mouth (eWOM), online consumer reviews (OCRs), Source
identity, Source credibility, Homophily, Similarity, Review trustworthiness
Research topic: Improving the perceived source credibility of online consumer reviews
(OCRs).
4 TABLE OF CONTENTS
CHAPTER 1: INTRODUCTION ... 6
1.1 Electronic word of mouth (eWOM) and online consumer reviews (OCRs) ... 6
1.2 Credibility of eWOM and OCRs ... 7
1.3 Source credibility ... 8
1.4 Homophily as an antecedent of source credibility ... 10
1.5 Type of product ... 10
1.6 Problem statement ... 11
1.7 Research questions ... 11
CHAPTER 2: THEORETICAL FRAMEWORK ... 13
2.1 The effect of perceived homophily on perceived source credibility ... 13
2.2 The effect of different types of source characteristics on perceived homophily ... 14
2.6 The effect of type of product ... 16
2.4 Other variables ... 18
2.4.1 The effect of conformity on the relationship of homophily and source credibility ... 18
2.4.2 The effect of eWOM skepticism on perceived source credibility ... 19
2.6 Conceptual model & Hypotheses ... 20
CHAPTER 3: METHODS ... 22
3.1 Participants and design ... 22
5
3.5.3 Credibility ... 29
3.5.4 Extra variables ... 30
3.5.4.1 Need for conformity ... 30
3.5.4.2 eWOM skepticism ... 30
3.5.5 Manipulation check ... 31
3.5.5.1 Type of product ... 31
3.5.6 Additional checks ... 31
3.6 Plan of analysis ... 32
3.6.1 Preparing for data analysis ... 32
3.6.2 Factor analysis ... 33
3.6.3 Reliability analysis ... 35
3.6.4 Correlations ... 36
3.6.5 Analysis of variance (ANOVA) ... 36
3.6.6 Mediation analysis ... 36
3.6.7 Bootstrapping ... 41
3.6.8 Multicollinearity ... 41
CHAPTER 4: RESULTS ... 42
4.1 ANOVA: DV = Source credibility ... 42
4.2 ANOVA: DV = Homophily ... 43 4.2 Mediation analysis ... 45 4.3.1 Step 1: path c ... 47 4.3.2 Step 2: path a ... 47 4.3.3 Step 3: path b ... 49 4.3.3 Step 4: path c’ ... 49 4.2 Hypothesis testing ... 50 CHAPTER 5: DISCUSSION ... 53 5.1 General conclusion ... 50
5.2 Managerial and academic implications ... 51
5.3 Limitations and recommendations for further research ... 56
REFERENCES ... 57
APPENDICES ... 65
APPENDIX A ... 65
APPENDIX B ... 73
6
CHAPTER 1: INTRODUCTION
1.1 Electronic word of mouth (eWOM) and online consumer reviews (OCRs)
When making a purchase decision, consumers often experience some level of uncertainty or risk (Pihlaja et al, 2017). Consequently, consumers develop strategies to decrease such perceived risk so that they can act with relative confidence in those situations where they feel like they lack adequate information (Woodside & Delozier, 1976). One way consumers do so is through acquiring and evaluating information that can help them in making their purchase decision (Taylor, 1974; Zhang et al, 2010).
A primary source of information used by consumers is that given by other consumers in the form of word of mouth (WOM) (Kimmel & Kitchen, 2014). WOM is ‘the act of consumers providing information about goods, services, brands or companies to other consumers’ (Babic et al, 2016). The concept is typically based on a friend or relative providing such information in a face-to-face situation (Bickart & Schindler, 2001). However, the rise of the web has facilitated that consumers can more easily disseminate WOM information to individuals other than just their friends or relatives (Xiang & Gretzel, 2010. Since October 2019, more than 4.48 billion individuals are active web users (Statista.com, 2019). Therefore, when WOM information is provided on the internet instead of face-to-face it can be consulted by many more different individuals, which are not necessarily friends or relatives. This spreading of information about goods, services, brands or companies to other consumers through the internet is defined as electronic word of mouth (eWOM), and is said to represent one of the most significant developments in contemporary consumer behaviour (Babic et al, 2016). The concept has proven its importance in relation to business performance, as eWOM has been found to affect product sales and thus market share (Chevalier & Mayzlin, 2006), customer value and loyalty (Gruen et al., 2005) and also the success of new product introductions (Clemons et al., 2006).
7 for local businesses by 2018 (Brightlocal.com). As retailers have started to recognize the increasing importance of this way of influencing consumers, many of them have been prioritizing on inducing, collecting, and displaying eWOM and OCR’s as part of their efforts to stimulate sales (Babic et al, 2016).
However, an important concept related to information on the internet is trust. Throughout the past few years, more people have been confronted with the occurrence of information found on the internet not always being true. The phenomenon of ‘fake news’, which has received growing attention over the past years, is an example of the spread of dishonest information through the internet. It can be defined as the spread of fabricated news media through channels which lack accuracy and credibility (Lazer et al, 2018). Fake news has received more attention especially because of its appearance in the political environment. It was recorded that right before the 2016 election in the United Stated of America, there was a significant spike in the amount of false information that the average American accounted (Vosoughi et al., 2018). As more people have become aware of the occurrence of fake news, trustworthiness is likely to be affected. Indeed, the results of the Edelman trust barometer, which is used to research the general perception of trust every year, presented that 73% of individuals worry about false information or fake news being used as a weapon (Edelman trust barometer, 2019). Additionally, the Edelman trust barometer showed that overall, there is a progression in people who often do not know what information is true and what is not (Edelman trust barometer, 2018).
1.2 Credibility of eWOM and OCRs
Similar to news media being mistrusted because of fake news, eWOM has also been subject of the prevailing feeling of mistrust towards information found on the internet. Since eWOM has been found to be very important for a business, there might be incentives for businesses to influence OCRs in their favour. As a result, there is an increasing amount of examples that show that also OCRs have been manipulated containing untrustworthy information.
8 additional report that shared their findings of an analysis of reviews on Amazon of hundreds of popular tech products in 14 online categories (The Guardian, 2019). They found that for many reviews there was no evidence that the reviewer had even bought or used the item. Additionally, many items on the website received a high number of five-star ratings, which is the highest rating that can be given to a product, posted in a very short space of time. Both observations were seen as indicators of inauthentic reviews signalling that the reviews are not to be perceived as credible. (The Guardian, 2019).
Another example is of the US based skin care brand Sunday Riley, which was recently investigated after a whistle blower stated that the company’s employees were forced to write fake reviews on beauty store Sephora’s website to boost sales between November 2015 and August 2017. The allegation done by the whistle blower was confirmed by the investigation lead by the Federal Trade Commision, as they found that both the CEO and the account manager of Sunday Riley had written explicit instructions to employees how to write fake reviews (Huffpost.com, 2019; Washingtonpost.com, 2019).
The increase of attention for practices illustrated by these examples, has led to increasing scepticism of the credibility of eWOM information (Kim et al., 2018). This is because of trustworthiness being an important component of credibility (Ohanian, 1990). Therefore credibility of information has also become a key concern in today’s online environment (Sundar, 2008).
Credibility is shown to be strongly related to important constructs such as informational usefulness, helpfulness of the review and information adoption (Cheung & Thadani, 2012). Since consumers nowadays tend to rely heavily on online consumer reviews for their purchase decision (Zhang et al, 2010), review credibility is an important facet for organizations to focus on. eWOM review credibility is conceptualized as the perceived degree to which an eWOM review provides accurate and truthful information (Cheung & Thadani, 2012).
1.3 Source credibility
9 review channel (Xiang & Gretzel., 2010), credibility derived from the review source will be the main focus for the purposes of the current study.
Source credibility is defined as a message receiver’s perception of credibility of a message source, reflecting nothing about the message itself (Chaiken, 1980; Cheung & Thadani, 2012). Similar to message credibility, source credibility has been found to have a positive effect on purchase intention (Wu & Wang, 2011). In eWOM, different studies have allocated source credibility to different types of sources. For example, source credibility has been derived from the platform of the electronic word of mouth, like social media websites (Westerman et al, 2014). However, this study focuses on source credibility attributed to the individual reviewer, or in other words, the writers of the reviews.
Generally, WOM is perceived more trustworthy and credible opposed to other marketing communication tools because of the familiarity of the source and because it does not have any commercial intent (Wu & Wang, 2011). However, a fundamental difference between traditional WOM and eWOM is the relationship between the source and the receiver of the word of mouth (Tham et al., 2013). With traditional WOM, the source and the receiver are often friends or relatives. Additionally, subtle credibility cues can be derived from the face-to-face contact that transpires when exchanging WOM (Buttle, 1998). However, with eWOM, recommendations are given by anonymous individuals in text (Kim et al., 2018). Consequentially, there are limited personal credibility cues to be noticed, which might result in more perceived difficulty in determining the reliability of eWOM information (Chatterjee, 2001; Kim et al., 2018). Although the identity of the source in eWOM is more ambiguous, a source is still a very important variable to consider when trying to understand people’s online review evaluation (Dou et al, 2012). It has been found that knowing the identity of information providers lead to higher perceived credibility and better recall of the information (Wathen et al, 2002). When the personal identity of the reviewer is unclear, the credibility of the review is found to be negatively affected (Dou et al, 2012).
10 characteristics can have direct effects on the reader of the message, independent of the content of the message (Forman et al. 2008).
The concept of source credibility has been elaborated throughout different research. Fundamental research by Hovland et al. (1953), Wu & Shaffer (1987) and Ohanian (1990) defined credibility as a two-part construct. They state that credibility is mainly based on trustworthiness and expertise. This definition is also supported by more recent research (Tseng & Fogg, 1999; Weathers et al., 2015). Trustworthiness is defined as truthfull, well-intentioned and unbiased. Expertise is defined as knowledgable, experienced and competent (Tseng & Fogg, 1999). When individuals evaluate the credibility, they evaluate both dimensions of trustworthiness and expertise.
1.4 Homophily as an antecedent of source credibility
Although the level of trustworthiness and expertise have been found to be the main dimensions of source credibility, homophily has also been proven to positively affect source credibility (Babic et al, 2016) and it is proven to be based on evaluating identity cues of the reviewer. The concept of homophily is based on similarity. When we feel similar to someone else, we like them more compared to when we do not feel similar (Byrne & Griffith, 1969; Cialdini, 1984).
Thus, reviews that can be identified to be written by similar others have a high level of homophily for the reader of the review. This means that there is a high level of ‘attraction of individuals with similar traits to one another’ (McPherson et al, 2001). Overall, when source characteristics which imply homophily details are provided with an OCR, eWOM’s effectiveness is boosted (Pihlaja et al, 2017). People are indeed more willing to accept and trust information, when the reviewer‘s characteristics are similar to their own (Cheung et al., 2009).
11 1.5 Type of product
However, online consumer reviews can be written for different types of products. Different types of products might result in a different level of the effect of the presence of similar source characteristics attached to an OCR.
For example, when consumers decide to book a hotel for a holiday with their children, they will more likely believe in hotel reviews that are written by similar consumers who also have children, because they have similar needs (Kusumasondjaja et al. 2011). However, when consumers are buying a new shirt, the identity of the source of the review written on the t-shirt might be much less relevant to them.
For the categorization of different types of products, fundamental research was performed by Nelson (1970). He identified two different types of product categories, labelled search and experience goods. He developed a continuum ranging from ‘easy to evaluate’ to ‘difficult to evaluate’. Based on the location on the continuum, the products are classified as either search goods or experience goods. Search goods are defined as ‘those dominated by product attributes for which full information can be acquired prior to purchase’ (Klein, 1998). Experience goods are ‘dominated by attributes that cannot be known until purchase and use of the product or for which information search is more costly and/or than direct product experience’ (Klein, 1998). Additional to these two types of products, credence attributes were added to this classification of goods (Darby & Karni, 1973), which were defined as attributes which cannot be acquired, even post-purchase (Klein, 1998). Consumers are more involved and thus perceive more risk when buying a credence product and an experience product compared to a search product (Wu & Wang, 2011), therefore evaluations regarding the source might be different for the types of products. The current study primarily focuses on the distinction between search and experience products, meaning credence goods will not be taken into account.
1.6 Problem statement
12 credibility, therefore attaching source characteristics to OCRs that could imply similarity could be an interesting tool for managers to fight the problem credibility of online consumer reviews. Since not a lot of research has shown what type of source characteristics, businesses have not successfully implemented this strategy yet. This study aims to give more insights in the field of effectively managing online consumer reviews by adding the appropriate source characteristics.
1.7 Research questions
The objective of the current study is the answer the following main question. Additionally, a set of research questions are proposed which will be answered throughout the report.
How does the presence of similar source characteristics attached to an online consumer review for different types of products affect perceived source credibility.
Research questions:
1. What is the effect of perceived homophily on perceived source credibility?
2. Is the effect of the effect of perceived homophily on perceived source credibility moderated?
3. What is the effect of different types of similar source characteristics on perceived homophily?
4. Is the effect of the presence of similar source characteristics attached to an OCR on perceived source credibility affected by type of product?
5. What are the effects of possible control variables with regard to perceived source credibility?
13
CHAPTER 2: LITERATURE REVIEW
This chapter is structured as follows. First of all, existing literature in relation to homophily and source credibility will be reviewed. This will be followed by the literature on the different effects of type of product. Further more, additional variables which could affect the
hypothesized effects are treated.
2.2. The effect of homophily in perceptions on source credibility
As mentioned before, the lack of identity cues in the online environment makes the presence of source characteristics attached to an online consumer review very relevant and can have significant effects how consumers respond to messages (Chaiken 1980, Mackie & Worth, 1990). However, it is suggested that the mere presence of source characteristics alone might backfire.
Wathen et al. (2002) found that when message sources were chosen to be similar in attitudes and context as their message audience, it leads to a higher credibility as well as a higher recall of information. Additional research suggests that the effect of homophily on information credibility not only applies in the offline environment, but also applies in the online environment (Wang et al. 2008). For example, homophily is shown to drive credibility on online health discussion groups as well as on websites (Wang et al, 2008).
Sundar (2008) states that similarity is likely to influence the perception of relevance of the information and therefore contribute to the credibility of the information. Individuals check similarity cues to distinguish between information which is interesting and relevant to them. For example, the opinion of a 16 year old about his/her stay at a hotel might not be very relevant for a an elderly couple looking to book a hotel, they will more likely value reviews written by people of similar age as more relevant. This perceived relevancy of reviews written by similar individuals will then contribute to the perceived credibility of the information (Sundar, 2008).
14 Based on this prior notion, it is suggested that there is a positive relationship between homophily and source credibility. Additionally, it is suggested that the positive effect of source characteristics on the level of source credibility is fully mediated by homophily.
Therefore the following hypotheses are presented.
Hypothesis 1: There is a positive effect of perceived homophily on perceived source credibility of an online consumer review (OCR).
Hypothesis 2: The effect of presence of characteristics of a source attached to an online consumer review (OCR) on the level of perceived source credibility is fully mediated by
homophily.
2.2. The effect of different types of source characteristics on homophily
As mentioned before, homophily is based on similarity. For similarity to become apparent, source characteristics need to be attached to an online consumer review. However there are different types of source characteristics defined which have an effect on homophily.
Several studies showed that the feeling of perceived homophily can be based on different traits. A seminal study by Lazarsfeld & Merton (1954) differentiates between status homophily and value homophily. McPherson et al. (2001), revisites the dimensions and also defines the dimensions of status and value homophily. This research additionally identifies the characteristics linked to the dimensions of status and value homophily and their effect on the level of homophily.
Additionally, McCroskey et al. (1975) measures the construct of perceived homophily through four different dimensions. They define homophily based on similarity in attitude, background, value and appearance. An improvement of this scale was developed in 2006, which concludes homophily to be mainly based on attitude, appearance and background characteristics (McCroskey et al., 2006). Since evaluations of similarity based on appearance are not possible in the current study, appearance will not be taken into account.
15 education, occupation, or behaviour patterns. McPherson et al. (2001) found that in the offline world race and ethnicity create the strongest level of homophily, followed by age, religion, education, occupation and gender.
Value homophily, however, is a more ambiguous concept. It is defined as ‘including a wide variety of internal states presumed to shape our orientation toward future behaviour’, and is based on values, attitudes, and beliefs. (Lazarsfeld & Merton, 1954; McPherson et al, 2001). These characteristics on which value homophily is based are typically identified as psychographic characteristics.
While comparing the dimensions distinguished by McPherson et al. (2001) and the dimensions distinguished by McCroskey et al. (2006), we can conclude they are very similar dimensions.
Both status and background homophily can be said to be based on demographic characteristics describing sex, age, ethnicity etc. Additionally, both value and attitude homophily can be said to be based on more ambiguous psychographic characteristics describing atittudes, interests and values. Therefore, the current study will distinguish between demographic characteristics and psychographic characteristics as bases of homophily.
The following two hypotheses show the effect of adding similar types of source characteristics, more specifically demographic versus psychographic characteristics, on homophily. The main effects are rather straightforward, but the relative effect of the different types of characteristics on perceived homophily is more interesting.
Hypothesis 3: The presence of similar demographic characteristics of the source attached to an online consumer review (OCR) has a positive effect on perceived
homophily.
Hypothesis 4: The presence of similar psychographic characteristics of the source attached to an online consumer review (OCR) has a positive effect on perceived
homophily.
16 businesses, as it might indicate which type of source characteristic they should ask consumers to share.
Bisgan et al (2012) identify and summarize the differences of the occurrence of homophily in an offline environment and an online environment. One important difference that they identified is that in the offline physical world mostly demographic characteristics such as age, gender, education and social status are used to study homophily, while in the online environment these sociodemographic dimensions often are not available or cannot be trusted. This is supported by Brown et al. (2007), who suggests traditional characteristics causing offline homophily have typically been based information that is often missing, reduced, camouflaged, or even intentionally falsified.
In other words, demographic information is information that is not often shared on the internet. Additionally, this type of information on the internet is more susceptible to deceit. Therefore interests of individuals are one of the strongest factors for evaluating homophily in the virtual world (Bisgin et al., 2012).
Hypothesis 5: The positive effect of the presence of similar psychographic characteristics on homophily is stronger than the positive effect of the presence of similar demographic
characteristics.
2.3. The effect of type of product
Mudambi & Schuff (2010) found that type of product, defined as either search or experience, affects information search and evaluation by consumers. Additionally, they found a difference in perceived credibility between search and experience goods when researching for the effect of review extremity. The difference found between the level of credibility perceived between the two types of products might imply this might also be possible in relation to source characteristic of a review.
17 Indeed, individual involvement is a significant influential factor for differences in individual attitudes (Wu & Wang, 2011). Involvement is defined as ‘product relevance to consumer needs and values (Zaichkowsky, 1985, p.342). The current study focuses on purchase decision involvement (PDI), which is defined as ‘the amount of interest and concern that a consumer brings to bear on a purchase-decision task’ (Mittal, 1889, p.150). According to the Elaboration Likelihood model (ELM) the information-processing route of consumers is related to the degree of involvement. The ELM distinguishes between two types of processing: central route-processing and peripheral-route route-processing. Central route route-processing is related to systematic processing, which is actively trying to understand and evaluate arguments provided (Chaiken, 1980). Peripheral route processing is related to heuristic processing, which can be described as evaluation based on little effort and more accessible information (Zhang & Watts, 2003; Chaiken, 1980). Consumers who are highly involved often participate in central-route processing, while consumers who have a low degree of involvement often participate in peripheral-route processing (Petty & Cacioppo, 1986).
When central route processing occurs, evaluations are based on aspects like arguments or message quality, while evaluations through peripheral route processing are based on simple cues in the context such as the source’s identity or other non-content cues. (Petty & Cacacioppo, 1986; Chaiken, 1980). Source characteristics attached to an OCR can be defined as such peripheral cues which are thus more often used for evaluation by individuals who have low involvement.
Involvement is also strongly related to risk: highly involved individuals experience more perceived risk, while individuals with low involvement experience less perceived risk (Wu & Wang, 2011). When an individual scores high on uncertainty avoidance, they seek stability, predictability and low risk (Hofstede, 1984). Uncertainty avoidance is defined as “the extent to which people feel threatened by ambiguous situations and create beliefs and institutions that try to avoid these” (Hofstede & Bond 1984, p. 418).
Similar source characteristics attached to an OCR might positively effect source credibility and could be seen as a quality insurance.
18 who are highly involved additionally experience more perceived risk, they might be looking for cues like similar source characteristics in support of the credibility and thus quality of the message. However, as highly involved individuals will also be likely to evaluate the message content and argument quality, the mere presence of similar source characteristics attached to an OCR alone might not be enough as an indication of source credibility.
When individuals are buying a search product, they are likely to participate in low purchase decision involvement which results in peripheral route processing. Subsequently, the individuals will focus on peripheral cues like source characteristics attached to an OCR for evaluating the OCR. Therefore, the similarity in source characteristics might be more salient to these individuals. As they additionally perceive lower risk and have a lower level of uncertainty avoidance, they are not likely to look for additional cues for credibility and are likely to derive more similarity and thus source credibility from the mere presence of similar source characteristics attached compared to individuals buying an experience good.
Hypothesis 6: The effect of the presence of similar source characteristics, more specifically demographic versus psychographic characteristics, on perceived source credibility is stronger when a review is written for search products in comparison to
experience products.
Based on prior notions, it is also expected that these differences in effects for type of product are stronger for the presence of similar demographic characteristics attached to an OCR compared to the presence of similar psychographic source characteristics. As it is suggested before that similar demographic source characteristics will be a less strong base for homophily and source credibility.
Hypothesis 7: The moderating effect of type of product on the effect of the presence of similar source characteristics on perceived source credibility is stronger for
19 2.4 Other variables
2.4.2 The effect of conformity on the relationship of homophily and source credibility
The basic premise of conformity is behaving in a specific way to meet group expectations as well as gaining group approval (Bearden et al, 1989). There are two types of conformity: normative conformity and informative conformity (Deutsch and Gerard, 1955). Normative conformity is based on conforming to the expectations of others while informative conformity, is based on conforming to information provided by others because you want to be right and you think this is the truth (Tsao et al, 2015).
An important aspect related to conformity is the perceived characteristics of the source (Mackie & Skelly, 1994; Castelli et al, 2001). It is suspected that perceived similarity with the source and conformity are positively related (Castelli et al, 2001). When individuals conform to a specific objective provided by another individual, they also implicitly view this objective as more credible (because more accurate) (Castelli et al., 2001).
It is expected that the positive relationship between homophily and credibility will be (positively) moderated by the need of conformity of an individual.
Therefore the following hypothesis is suggested:
Hypothesis 8: The higher the need for conformity, the higher the positive effect of perceived homophily on perceived source credibility.
2.4.3. The effect of eWOM skepticism on source credibility
Skepticism is strongly related to credibility, as both concepts rely heavily on the level of trustworthiness. Literature on consumer skepticism is mostly focused on skepticism towards advertisements. Skeptical consumers have been found to be more disbelieving of advertising compared to less skeptical consumers (Obermiller and Spangenberg, 1998). In line with this research with regard to advertisements, skepticism has been found to be negatively correlated to source credibility (Pan, Meng, Lee , 2017).
20 (Flanagin & Metzger, 2007). As previously mentioned, individuals have difficulty recognizing falsified information on the internet, and therefore are more likely to be skeptic towards it.
Previous research suggests that skepticism towards eWOM communications may be an influential factor in online interactions (Dou et al., 2012). Therefore it is an important concept to consider in the current study. Based on the prior research described, it is expected that when individuals record a high level of skepticism towards eWOM, their credibility judgements of an online consumer review will consequentially be lower compared to individuals who are not skeptic towards eWOM. Thus it is expected that there is a negative relationship between eWOM skepticism and perceived source credibility. Therefore, the following hypothesis is suggested
Hypothesis 9 : The higher the level of skepticism towards eWOM, the lower the level of perceived source credibility perceived by the individual.
Based on the prior notions, the following hypothesis have been suggested (Table, 1). The study aims to find support for these. All suggested hypotheses of the effects of the different concepts are portrayed in the conceptual model (Figure, 1), except for the relative effects suggested in hypothesis 5 and hypothesis 7.
Table 1 Hypotheses
Hypotheses Hypothesis
1
There is a positive effect of perceived homophily on perceived source credibility of an online consumer review (OCR).
Hypothesis 2
The effect of presence of similar characteristics attached to an online consumer review (OCR) on the level of perceived source credibility is fully mediated by homophily.
Hypothesis 3
The presence of similar demographic characteristics of the source attached to an online consumer review (OCR) has a positive effect on perceived homophily.
Hypothesis 4
The presence of similar psychographic characteristics of the source attached to an online consumer review (OCR) has a positive effect on perceived homophily.
Hypothesis 5
The positive effect of the presence of similar psychographic characteristics on homophily is stronger than the positive effect of the presence of similar demographic characteristics.
Hypothesis 6
21 credibility is stronger when a review is written for search products in comparison to experience products.
Hypothesis 7
The moderating effect of type of product on the effect of the presence of similar source characteristics on perceived source credibility is stronger for demographic characteristics compared to psychographic characteristics. Hypothesis
8
The higher the need for conformity, the higher the positive effect of perceived homophily on perceived source credibility.
Hypothesis 9
The higher the level of skepticism towards eWOM, the lower the level of perceived source credibility perceived by the individual.
22
CHAPTER 3: METHODS
This chapter will give an outline of the design of the study that is developed to test the hypotheses proposed in chapter 2. Additionally, both the procedure and the method used for data collection is described. Furthermore, an operationalization of the relevant variables that will be measured and an overview of the steps taken for data analysis will be provided.
3.1 Participants and design 3.1.1 Participants
The current study is based on convenience sampling, which means that the selection of the participants is based on participants who are accessible, easy to measure and cooperative (Malhotra, 2010). As a result of convenience sampling, the study will be very likely to gain respondents belonging to the age group of 18 to 25 years.
Since the manipulation of the demographic characteristic could become very complicated and as a result more susceptible for possible inaccuracies in the measurement, it was decided to limit the participant population for this study to a specific age group. The target population are individuals who are 18 to 25 years old. However, limiting the group of participants exclusively to this age group makes the results of this study less generalizable, which will affect the external validity of the study (Malhotra, 2010). This will be taken into account when interpreting the results of the study.
3.1.2 Experimental design
This study is based on a survey, which tests the causal relationship between the independent variables (Demographic characteristic, Psychographic characteristic and Type of product) and the dependent variable(s) (overall feeling of homophily and perceived source credibility).
23 The study is 2 (Demographic characteristic attached to OCR/ Demographic characteristic not attached to OCR) x 2 (Psychographic characteristic attached to OCR/ Psychographic characteristic not attached to OCR) x 2 (Search good/ Experience good). Table 2 shows a clear representation of the design of the study.
Table 2 Experimental design
Psychographic characteristic attached to OCR
Yes No
Search Experience Search Experience
Demographic characteristic attached to OCR
Yes Condition 1 Condition 5 Condition 2 Condition 6
No Condition 3 Condition 7 Condition 4 Condition 8
3.2 Procedure 3.2.1 Sampling
The effect of random assignment on the internal validity of the study only works as long as the sample size is large enough (Aronson et al., 1998). Therefore between-subjects design should have at least 15 participants per condition (Aronson et al., 1998). Additionally it is important to have a large enough sample so that the results of the study will be more generalizable. The current study aimed to collect at least 240 participants so that even after data cleaning there would be enough participants’ responses left to draw valid conclusions from. Eventually there were 558 participants recorded in total, of which 321 finished the whole survey and another 46 finished the survey up to the manipulation check. Which resulted an amount of 367 responses in total before data cleaning, that could be used for further analysis.
24 3.2.2 Survey design
When participants clicked on the link of the study, they were first introduced to the study. However, the information about the actual study given beforehand was kept very limited, as more extensive information might make participants aware of the study’s purpose. When participants already have an idea of the hypotheses that are researched, their responses will be likely to be affected (Nichols & Maner, 2008). Consequently, participants were only informed about first having to fill in some basic questions, which would be followed by an online consumer review on which some additional questions would be asked.
Followed by this, participants were informed about the duration of the survey and the incentive which can be won. Since concern about confidentiality of answers is found to contribute significantly to survey nonresponse and response quality (Singer, 1993), participants were also informed about the anonymity and confidentiality of their answers, and the option to terminate their participation at any point of time. By clicking on the link forwarding the participants to the first questions of the survey, participants agreed to the conditions of filling in the survey and to participate in the study.
3.2.3 Condition referral
As this is a 2x2x2 between-subjects design, there were 8 different conditions a participant could be assigned to. Based on the information provided through the answers given on the first set of questions of the study (Appendix A), from which their demographic characteristic (age group) and their psychographic characteristic (VALS-label) were derived, the participants were randomly assigned to just 1 of 8 different conditions.
3.3 Stimulus 3.3.1 Review
25 The length and the look of the review were similar to reviews produced for both product types to enhance external validity. Additionally, the textual content of the review was controlled for by it similar across the different conditions. In this way, the chance of a variance of outcomes that could be accounted to the textual content of the review is minimized. However, because of the manipulation of the different types of products, the reviews written do vary between those conditions, this will be taken into account when interpreting the results.
Figure 2 Stimulus condition 1
Figure 3 Stimulus condition 5
3.4 Manipulations
Both source characteristics and type of review were manipulated. Source characteristics attached to the review were manipulated as either a demographic characteristic or a psychographic characteristic, based on the first set of questions answered in the survey.
3.4.1 Demographic characteristic
26 straightforward characteristic which is easily measurable and which is also appropriate and common to be attached to a review.
3.4.2 Psychographic characteristic
For the manipulation of the psychographic characteristic, the VALS-typology developed at SRI international by Mitchell (1981) was found most appropriate in relation to the current research. VALS is a typology which differentiates between 8 different so-called ‘lifestyles’, which consist of different combinations of psychographics like values, interests and attitudes. The differentiated lifestyles have very clearly showed a distinction between possible psychographic characteristics, like different sets of values. Therefore it was found as an appropriate typology for the manipulation of a psychographic characteristic.
Figure 4 Survey question VALS
The participants were shown a list of different types of labels of the lifestyles with the related values (Figure 4). They were asked to choose one out of the 8 lifestyles which was most applicable to them. Subsequently, the manipulated review would show the label chosen by the participant when assigned to the relevant condition.
27 3.4.3 Type of product - Online search/experience
Although the definitions provided for search and experience products are still widely accepted, Klein (1998) suggested that the internet might influence the classification of the products so that experience goods would transform in search goods.
However, it was also found that access to the internet does not result in significant shifts between search, experience and credence classifications because there are also counterforces against such shifts prevalent on the internet (Nakayama, Sutcliffe, Wan, 2010).
Nonetheless, to ensure the manipulation of type of product in the current study is done accordingly, the type of search product and the type of experience product were chosen based on similar research in relation to online consumer reviews and the internet. The current study selected search and experience products based on the research of Bei et al., (2004), who tested type of goods and identified a set of search goods and a set of experience goods for in the online environment. A visit to a new restaurant was identified as an experience product and athletic shoes (sport shoes) were identified as a search product. Consequently, these types of goods will be used for the manipulation in the current research.
3.5 Measurement
28
Table 3 Operationalization of variables
Construct Source Items Scale Eigenvalue
and variance Chronbach’s alpha Perceived homophily McCroskey, Richmond & Daly (1975) McCroskey et al. (2006) The reviewer: 1. Thinks like me. 2. Behaves like me. 3. Is similar to me.
4. Has thoughts and ideas similar to mine.
5. Treats people like I do. 6. Is like me.
7. Shares my values.
8. Has a lot in common with me.
7 point Likert scale: ranging from Strongly agree to strongly disagree 5,08 63,48% explained 0,92 Perceived credibility of a review
Xu (2014) 1. The review is credible. 2. The review is believable. 3. The review is trustworthy.
7 point Likert scale Strongly agree to strongly disagree 2,41 80,26% explained 0,88 Source credibility Cheung et al (2008). (Wu & Shaffer)
1. The review is provided by a knowledgeable source. 2. The review is provided by an expert source.
3. The review is provided by a trustworthy source.
4. The review is provided by a reliable source. 7 point semantic differential scale 2,80 70,03% explained 0,85 Need for conformity Bearden et al. (1989) Tsao & Hsieh 2015
1. I rarely purchase the latest trend products until I am sure my friends approve of them.
2. It is important that others like the products/services I choose to buy. 3. When purchasing
products/services, I generally select those that I think others will approve of.
4. When purchasing a product/service, I generally purchase those
recommended by my friends. 5. I like to know what products/services make good impressions on others.
6. I achieve a sense of belong ing by purchasing the same products/services that others purchase.
7. If I want to be like someone, I often try to buy same products/services that they buy.
8. I often identify with other people by purchasing the same
product/services that they purchase. 9. To make sure I purchase the right product/service, I often observe the products/services others have purchased.
10. If I have little experience with a product/service, I often ask my friends about it.
29 12. I frequently gather information
from friends or family about a product/service before I buy. eWOM Skepticism Zhang, X., Ko, M., & Carpenter, D. (2016)
1. We can hardly depend on getting the truth from most online reviews. 2. Online reviews are not generally truthful.
3. In general, online reviews don’t reflect the true picture of a subject. 4. Online reviewers care more about getting you to buy things.
5. Most online reviews are intended to mislead’.
6. People writing online reviews are always up to something.
7. People writing online product reviews are not necessarily the real customers.
8. People write online reviews pretending they are something else. 9. Different reviews are often posted by the same person under different names. 7 point Likert scale Second order factor 2,0 66,63 % 0,75 Manipulation Check: type of product
N/A 1. Did you notice the reviewer characteristics attached to the review?
Yes No I’m not sure
N/A N/A
Information check: Source characteristic
N/A 1. The product for which you just saw a review has attributes that you mostly…
7-point Sementic differential scale ranging from Can find out beforehand to Have to experience during consumption. N/A N/A Demographic characteristic: Gender
N/A 1. What is your gender? Male Female
N/A N/A
3.5.1 Perceived homophily
30 The attitude-homophily scale used more ambiguous items which could be answered based on both the demographic characteristic and the psychographic characteristic attached to the OCR. Additionally, background homophily and attitude homophily (as well as status homophily and value homophily) have been expected to covary in many communication settings (McCroskey et al, 2006).
The scale was found to be effective in measuring attitude homophily with an alpha of 0,916.
3.5.2 Source Credibility
For the measurement of the dependent variable source credibility, a scale defined by Cheung et al. (2008) was used which was based on a scale developed by Wu and Schaffer (1987). The scale was found to be effective in measuring source credibility for the current study and sample with a Chronbach’s alpha of 0,853. All items of the scale used were computed into one new variable used for further analysis.
3.5.3 Credibility
To be able to support the relationship defined between source credibility and credibility, the construct of credibility was also measured in the current study. The correlation matrix (Appendix C11) showed that indeed the correlation between credibility and source credibility is significant.
For the measurement of perceived review credibility a scale was used developed by Xu (2014). The scale of Xu (2014) was chosen as a measurement for perceived review credibility. Xu (2014) used a 3-item scale which was measured on a 7-point Likert scale ranging from ‘strongly disagree’ to ‘strongly agree’. The items of the scale measured the perceived level of credibility, believability and trustworthiness of the review. The scale was found to be effective in measuring perceived review credibility for this study and sample with a Chronbach’s alpha of 0,876. All items of the scale used were computed into one new variable used for further analysis.
3.5.4 Extra variables
31 To measure the need for conformity, a scale with relatively many items was used as there were no scales perceived more adequate for measuring the full construct of conformity in relation to the current study.
The scale used in the current study was initially developed by Bearden et al. (1989) but was custumized by Tsao and Hsieh (2015) to measure conformity towards eWOM. The scale consists of 12 items measured on a 7-point Likert scale ranging from ‘Strongly disagree’ to ‘Strongly agree’. Eight out of twelve items measured normative conformity and the other four items measured informative conformity.
Although Tsao and Hsieh (2015) showed high internal consistency of their scale measuring conformity towards eWOM, the factor analysis in the current study showed that the different items of the scale loaded on three different factors (Appendix C4). Based on the conceptualization of conformity in the current study, one of the three factors was chosen as a measurement of conformity towards eWOM. Subsequently, the items which the rotated components matrix showed that loaded on that particular factor were computed in a new variable for conformity used in further analysis.
3.5.4.2 eWOM scepticism
Skepticism was measured as a control variable. Although many different scales have been defined for measuring scepticism, the scale defined by Zhang, Ko, & Carpenter (2016) was perceived most suitable for this research. This scale measures the scepticism of consumers towards electronic word of mouth, thus very appropriate for the current study.
eWOM scepticism was measured by nine items on a 7-point Likert scale ranging from ‘strongly disagree’ to ‘strongly agree’. Factor analysis showed that the scale loaded on three different factors, this is in line with the literature which composed the scale out of three factors: Truthfulness, Motivation, Identity. Since the literature showed that eWOM scepticism should measured by these three factors, it was decided to perform a second order factor analysis to
check whether the three different factors together loaded on one main factor: skepticism.
32 different levels showed a Chronbach’s alpha of 0,749, indicating the effectiveness of the scale used for measuring scepticism in the current study.
3.5.5 Manipulation check
Type of product
To analyse whether the manipulation in the current study was performed correctly, a manipulation check can be used. One of the manipulations was checked by asking the question based on the conceptualization of both search and experience goods (Nelson, 1970).
Using an independent samples t-test, the respondents were checked for the different conditions whether indeed they thought of search and experience goods as different type of goods in relation to the judgements of the attributes of the goods.
Levene’s test for equality of variances shows very small F values, and not statistically significant for both variables. Therefore the assumption of homogeneity of variance still holds. Consequently, the results of the t-test can be successfully interpreted. The results of the independent samples t-test show that indeed there is a significant differences in responses for the types of products. The respondents scored the type of products according to their definition, therefore the manipulation of type of product was performed correctly.
3.5.6 Additional checks
3.5.5.1 Source characteristics check
Additionally, a check was done whether the respondents were conscious about the source characteristics attached to the review.
3.5.5.2 English fluency check
33 the survey, a question asking participants to score their English fluency was asked (Appendix, A).
3.6 Plan of analysis
In this plan of analysis, the steps taken for the analysis of the data are described and explained.
After data collection was completed, data was converted to SPSS.
3.6.1 Preparing data for analysis
Predominantly of the analysis based on the data, data cleaning is in order. Data cleaning is considered essential for adequate quality of the results. If data cleaning is not performed well, results could be biased as well as interpreted incorrectly (Malhotra, 2009).
In total 558 respondents were recorded in total for this study. However, 44% of these respondents failed to finish the survey. The data of the respondents who finished the survey up till the at least the manipulation check, was treated with pairwise deletion, meaning that data of variables they did provide, were used in the analysis (Malhotra, 2009).
In the first question of the survey (appendix, A) respondents were asked to indicate whether they belonged to the age group required for the study or not. Respondents who indicated that they were not 18-25 years old, were immediately forwarded to the end of the survey. However, these responses were still labelled as recorded responses by Qualtrics therefore they needed to be cleaned from the dataset.
Furthermore the dataset was checked for any missing data, weird data and any outliers. Based on the English fluency check (Appendix, A), 4 respondents were found recording their level of English significantly low. Therefore their responses to the survey were perceived not reliable and thus not used for further analysis. Additionally, the data from respondents who failed the attention checks in the scales of conformity and scepticism in the survey was not used. Consequently, further analysis was based on the remaining data from 282 respondents.
34 variables. These are values which only take on two values, such as 0 and 1 (Malhotra, 2010). The dummy coding helps for further analysis.
Table 4 shows the descriptive statistics based on gender for each condition. The data recorded for this study was mainly based on female respondents (76,6% of the total respondents). This should be taken into account when interpreting the results. Additionally, not all conditions contain the exact same number of respondents, this should also be taken into account when interpreting the results.
Descriptive statistics per condition
Table 4 Desciptives per condition
Descriptives per condition
Male Female Total
Condition 1 8 (29,6%) 19 (70,4%) 27 (100%) Condition 2 7 (17,5%) 33 (82,5%) 40 (100%) Condition 3 5 (16,1%) 26 (83,9%) 31 (100%) Condition 4 8 (22,9%) 27 (77,1%) 35 (100%) Condition 5 14 (36,8%) 24 (63,2%) 38 (100%) Condition 6 11 (26,2%) 31 (73,8%) 42 (100%) Condition 7 4 (11,8%) 30 (88,2%) 34 (100%) Condition 8 9 (25,7%) 26 (74,3%) 35 (100%) Total 66 (23,4%) 216 (76,6%) 282 (100%) 3.6.2 Factor analysis
Subsequently, factor analysis will be performed. Factor analysis is adequate for testing the hypothesized dimensions of the scales on whether they also exist in the current dataset (Malhotra, 2010). Factor analysis will be performed for each variable and its items.
35 Additionally, Bartlett’s test of sphericity is performed, which tests the correlation of the variables (Hair, 2006). If the outcome is significant, H0 can be rejected which means correlations can be established.
Subsequently, the percentage of variance of a given variable shared with all other variables included in the analysis, defined as communalities, will be inspected (Hair, 2006). Communalities are values <1. The rule of thumb for communalities is that they should at least be 0,4. When all prerequisites are met, factor analysis is appropriate.
For determining the number of factors per variable, there are several criteria which should be considered. First of all, the eigenvalues, which represent the total variance explained by each factor, should meet a minimum of 1 (Hair, 2006). Additionally, the total variance explained should be at least 60% and each factor should explain a minimum of 5% of the total variance each (Hair, 2006). Ultimately, a scree plot could be inspected for determining the appropriate amount of factors.
Preferably, all items correlate on one factor, as this represents the adequacy of the scale for this dataset. However, factor analysis could show more underlying dimensions of the scale. The rotated component matrix will show more clearly which items load to which dimensions (Hair, 2006).
Subsequently, the factor and its items which are perceived as measuring the aspect most in line with the conceptualization of the construct in relation to the current study will be chosen for measuring the construct. Additionally, the Chronbach’s alpha (outcome of reliability analysis) for the different factors is taken into account when deciding on which factor to continue analysis with.
36 3.6.3 Reliability analysis
Proceeding, reliability analysis will be conducted to measure the internal consistency of the factors. Chronbach’s alpha is inspected for a value which meets the prerequisite of 0.6 (Hair, 2006). When the prerequisite is not met, it indicates unsatisfactory internal consistency of the dimensions (Malhotra, 2010). When ‘Chronbach’s alpha if item deleted’ indicates that the internal consistency of the dimensions will increase when a specific item is deleted, the item will be deleted. However, when there is only a very small increase, the item will not be deleted to stay in line with the literature. After the factor and reliability analysis, a new variable will be computed based on the appropriate items.
3.6.4 Correlations
Before further analysis, correlations between the variables were inspected. When a correlation is shown to be significant between variables, it can be concluded that they relate. The correlations identified can be used for the support of relationships defined in chapter 2, like the relationship between credibility and source credibility (Appendix 1C1).
3.6.5 Analysis of variance (ANOVA)
Following the reliability analysis, the analysis of variance (ANOVA) will be conducted. ANOVA is adequate for testing the influence of certain treatments. More specifically, it offers a way to examine the differences in the mean values the dependent variable as a result of the influence of the independent variable(s) (Malhotra, 2010).
If the test is significant, the H0 (all means are equal) can be rejected and it can be concluded that the mean value of the dependent variable will be different for different categories of the independent variable (Malhotra, 2010). However, ANOVA does not show for which category the mean is different, therefore post-hoc multiple comparisons should be performed. Since the experimental variables tested in the current study just have two levels, an additional ANOVA will be performed which tests the effect of the different conditions on the variable tested. The post-hoc tests of this ANOVA will show the significant differences of influence on the dependent variable between all the possible pairs of conditions (Hair, 2006 p.473).
37 effects will not be interpreted for analysis as they cannot be tested meaningfully as the different main effects will depend on each other. However, when there are no significant interaction effects, the main effects are appropriate for further analysis.
3.6.6 Mediation analysis
The current study suggests a mediation effect. Mediation analysis investigates how independent variable X causes dependent variable Y, or in other words, what is the mechanism M by which independent variable X influences dependent variable Y (Hayes, 2017). In the current study it is hypothesized that the effect of the independent variables (the presence of demographic/ psychographic characteristics attached to an OCR) on the dependent variable (source credibility) is only there when the mediating variable (homophily) is present. The process model designed by Hayes (2017) could be used for testing the full model of the current study.
Table 5 Full model, Regression formula
Table 6 Definitions of variable labels
Hayes (2017) designed many different models which can be used to test mediation in SPSS. A model which reflects the full model of this study (Table, 5) was selected: Model 51. Although
Full model M = β0 + (β1*X1) + (β2*X2) + (β3*X3)+ β4(X1*X3) + β5(X2*X3) + Y= β0 + (β1*M) + (β2*Z) + β3(Z*M) + β4* V1 + β5* V2 Variable Description Dependent variable
Y Perceived source credibility Independent
variables
X1 Type of product (Search/Experience)
X2 Presence of demographic characteristic (Yes/No)
Moderating EV
X3 Presence of psychographic characteristic (Yes/No)
Mediator M Perceived homophily Moderator Z Need for conformity Control
variables
V1 eWOM scepticism
38 this model does include a three-way interaction and a two-way interaction which is originally not tested in the current study, this model was perceived most appropriate for testing the hypothesized mediation compared to other models designed by Hayes (2017).
Unfortunately, model 51 was not available in SPSS, which resulted in not being able to test the full model of the current study with the help of Hayes.
To check whether ‘homophily’ indeed mediates between the experimental variables and source credibility, the four-step approach as defined by Baron & Kenny (1986) has been followed. In this approach, four different requirements should be met, based on the paths as depicted in figure 5 and 6.
Figure 5 Baron & Kenny (1986), path c
Figure 6 Baron & Kenny (1986), path a,b, c'
1. The effect of independent variable X on dependent variable Y (c) should be significant. 2. The effect of independent variable X on mediator M (a) should be significant.
3. The effect of mediator M on independent variable Y should be significant.
4. The effect of mediator M on the effect of independent variable X on dependent variable Y (c’) should be not significant, which means there is full mediation, or lower, which means there is partial mediation.
The models of Baron & Kenny (1986) shown in figure 5 and 6 are not representative for the current study, there are several differences. The current study does not just have one
independent variable, but two: presence of similar demographic source characteristics (X1) and
presence of psychographic source characteristics (X2). Additionally, the effects of these
39
(X3). Furthermore, the relationship between the proposed mediator homophily (M) and the
dependent variable source credibility (Y) has been suggested to be moderated by need for
conformity (Z). Finally, two control variables were added to the model, eWOM scepticism (V1)
and gender (V2). Therefore the base models as shown in figure 5 and 6 have been modified to
models with these additional variables added, so that the models are more representative of the current study. The modified models are shown in figure 7 and 8.
Figure 7 Modified model based on Baron & Kenny (1986), path c
Figure 8 Modified model based on Baron & Kenny (1986), path a,b, c'
40 regression will also show the value of the R-squared and adjusted R-squared. The R-squared is used to check how good the model is in explaining the variance. For each path defined by Baron and Kenny (1986), the analysis will start with a small regression model and add variables to check whether the fit of the model improves by adding the variables. This can be checked by comparing the adjusted R-squared between the models (Malhotra, 2010). The F will show the overall fit of the model tested (Hair, 2006).
The different models for each path tested are described in table 7-10.
Table 7 Step 1 Baron & Kenny (1986), Model formula's
Step 1: Path c
Model Description Formula
Model 1 The main effect of the presence of similar demographic characteristics attached to an OCR (X1) on source
credibility (Y), and the main effect of the presence of similar psychographic characteristics attached to an OCR (X2) on source credibility (Y).
Y= β0 + (β1*X1) + (β2*X2)
Model 2 Model 1 with the main effect and interaction effects of the moderating experimental variable type of product (X3) included.
Y = β0 + (β1*X1) + (β2*X2) +
(β3*X3) + β4(X1*X3) +
β5(X2*X3)
Table 8 Step 2 Baron & Kenny (1986), Model formula's
Step 2: Path a
Model Description Formula
Model 3 The main effect of similar demographic characteristics attached to an OCR (X1) on homophily (M), and the
main effect of similar psychographic characteristics attached to an OCR (X2) on homophily (M).
M= β0 + (β1*X1) + (β2*X2)
Model 4 Model 3 with the main effect and interaction effects of the moderating experimental variable Type of product (X3) included.
M = β0 + (β1*X1) + (β2*X2) +
(β3*X3) + β4(X1*X3) +
β5(X2*X3)
Table 9 Step 3 Baron & Kenny (1986), Model formula's
Step 3: Path b
Model Description Formula
Model 5 The main effect of perceived homophily (M) on perceived source credibility (Y1).
Y= β0 + (β1 *M) Model 6 Model 5 with moderator Conformity (Z) included and
with the interaction effect of Homophily (M) and Conformity (Z) included.
Y= β0 + (β1*M) + (β3*Z) + β3(Z*M)
Model 7 Model 6 with the control variables eWOM scepticism (V1) and Gender (V2) included.
41
Table 10 Step 4 Baron & Kenny (1986), Model formula's
Step 4: Path c’
Model Description Formula
Model 8 Model 1 with the mediating variable added Y= β0 + (β1*X1) + (β2*X2) +
(β3*M)
Model 9 Model 2 with the mediating variable added Y = β0 + (β1*X1) + (β2*X2) +
(β3*M) + (β4*X3)+ β5(X1*X3)
+ β6(X2*X3)
3.6.7 Bootstrapping
The outcomes of the regressions can be validated through bootstrapping. Bootstrapping is a statistical method used to look at errors of the analysis (Hair, 2006). When bootstrapping is performed, a large number of subsamples (N=1000) is drawn, and for each sample, a model is estimated. Subsequently, all the different estimates are combined which provides the “best” mean for each estimated coefficient derived from all the different models. Additionally, the expected variability is provided (Hair, 2006).
The outcomes of bootstrapping showed that the significance levels for each almost do not change when bootstrapping is performed (Appendix C12.1-C12.9). Therefore the outcomes can be perceived valid.
3.6.8 Multicollinearity
