INFORMATION PROCESSING IN ONLINE REVIEWS
Minke Mennen
MSc. Marketing Intelligence
A CHOICE-BASED CONJOINT ANALYSIS FOR THE EFFECT OF PERIPHERAL CUE AND CENTRAL CONTENT FACTORS ON A
CONSUMER’S TRUST OF ONLINE REVIEWS
Yes! I trus t
online re views
I NFORMATION P ROCESSING IN O NLINE R EVIEWS :
A CHOICE-‐BASED CONJOINT ANALYSIS FOR THE EFFECT OF PERIPHERAL CUE AND CENTRAL CONTENT FACTORS ON A CONSUMER’S TRUST OF ONLINE REVIEWS
Minke M.H.E. Mennen University of Groningen
Faculty of Economics and Business MSc. Marketing Intelligence
June 22, 2015
Kapelstraat 10 6039 RT Stramproy Tel: (06) 30776578
m.m.h.e.mennen@student.rug.nl Student number: 2528304
Supervisors
First supervisor: Dr. H. (Hans) Risselada
Second supervisor: J.E.M. (Erjen) Van Nierop
M ANAGEMENT S UMMARY
Currently, most of the consumers who shop online read and use opinions of other consumers in their decision-making process. Therefore, it is very important to know what factors are most important in trusting an online review. This study helps the management of online retailers to optimize their online review environment. Prior studies indicate that consumers have different abilities in information processing. This study explores to what extent peripheral cues and central content factors affect a consumer’s trust of an online review and whether this differs for consumers in different mood states or with different personalities. The main goal of this study is to find the most important attributes when it comes to trusting an online review.
Since consumers differ in their preferences and behavior, three segments with different preferences were found in this study, namely ‘observational learners’, ‘text critics’ and
‘independent consumers’. Furthermore, six different attributes are expected to affect a consumer’s trust and are therefore compared. Results show that central content factors (e.g.
the complexity of the content) are more important than peripheral cue factors (e.g. the number of people who reviewed a product) for consumers when trusting an online review. Lexical complexity is found as the most important attribute, followed by aggregated rating score volume, two-sidedness, reviewer’s expertise, real name exposure and review helpfulness volume. They all directly affect a consumer’s trust. The most striking result is that respondents prefer online review with a high level of lexical complexity which is inverted to what was expected. Two-sided reviews boost trust in online reviews. Aggregated rating score and review helpfulness volume should be as high as possible, i.e. the higher the volume, the more the online review is trusted. Reviews with a product expert label and with a real name are preferred. The management of online retailers can use this information to create guidelines for people who are willing to write a review where after the written online reviews will lead to higher trust by consumers who are reading the online reviews. Moreover, the effect of the central content and peripheral cue factors on trust differs per segment. ‘Observational learners’ are focused on volumes in trusting online reviews and prefer content which does contain mainly subjective information and a little product attribute information, ‘text critics’
highly prefer online reviews that contain a lot of product attribute information so that these
reviews have a high level of lexical complexity, and the ‘independent consumers’ prefer both
positive and negative arguments in their reviews.
A BSTRACT
The aim of this study is to investigate to what extent central content and peripheral cue factors affect a consumer’s trust of online reviews among different types of personality and mood. A choice-based conjoint analysis on a sample of Dutch respondents (n=223) was performed, both on the aggregated and segment level. Results show no differences with regard to personality traits and mood. The aggregated results show that lexical complexity is the most important attribute followed by aggregated rating score volume and two-sidedness. It can be concluded that central content factors are more important than peripheral cue factors in a consumer’s trust of an online review. Respondents were grouped in three different segments, namely ‘observational learners’, ‘text critics’ and ‘independent consumers’. This study contributes to the field of online reviews and information processing. Each segment has different preferences and therefore this study has important information for strategic management. Moreover, this study gives recommendations for online retailers to optimize their online review environment in web shops.
Key words: trust, central content, peripheral cues, online reviews, mood, personality traits,
information processing, choice-based conjoint analysis
P REFACE
Hello reader,
My final chapter! This thesis is my final phase of the MSc. Marketing Intelligence where after I will be graduated and I am planning to start a career.
Six years ago I started my HBO bachelor in Tilburg. Next, I decided that I would like to have master degree. I decided to move to the other side of the country and started my pre-master in Groningen which resulted in an admission to the MSc. Marketing Intelligence track. It was a year of hard work where I learned a lot. But one to remember!
First of all I would like to thank Hans Risselada, my supervisor, for his support and great feedback during the last five months. I also would like to thank my fellow students of the thesis group ‘information overload’. We supported and helped each other when necessary. In particular, I would like to thank my family and friends. They supported me out-and-out during my six years as a student.
Enjoy reading.
Stramproy, June 2015
Minke Mennen
T ABLE OF C ONTENT
1. Introduction ... 1
2. Research Framework ... 4
2.1. Defining Trust In Online Shopping ... 5
2.2. Information Processing Theory ... 5
2.2.1. Central Content Of Online Reviews ... 6
2.2.2. Peripheral Cues Of Online Reviews ... 8
2.3. Moderator ... 11
2.3.1. Personality Traits ... 11
2.3.2. Mood ... 12
2.4. Conceptual Model ... 13
3. Research Design ... 14
3.1. Research Method ... 14
3.1.1. Attributes And Levels ... 15
3.1.2. Personality Traits Moderator Measurement Scale ... 17
3.1.3. Mood Moderator Measurement Scale ... 18
3.2. Choice Design ... 18
3.3. Data Collection ... 19
3.4. Plan Of Analysis ... 20
4. Results ... 22
4.1. Sample Characteristics ... 23
4.2. Construct Validity ... 24
4.3. Choice-Based Conjoint Analysis ... 24
4.4. Model Selection ... 27
4.5. Validity ... 28
4.6. Segment Interpretation ... 29
4.6.1. Segment 1- ‘Observational Learners’ ... 30
4.6.2. Segment 2 – ‘Text Critics’ ... 31
4.6.3. Segment 3 – ‘Independent Consumers’ ... 32
4.7. Hypothesis Testing ... 33
5. Discussion ... 34
5.1. Theoretical Implications ... 35
5.1.1. Central Content ... 35
5.1.2. Peripheral Cues ... 36
5.1.3. Moderators ... 37
5.2. Managerial Implications ... 37
5.3. Limitations And Future Research ... 39
6. References ... 41
1. I NTRODUCTION
Online shopping is becoming more popular. Over the last 10 years the amount of online shopping has increased from 2.8 billion euro in 2005 to 10.6 billion euro in 2013 in the Netherlands (Thuiswinkel.org, n.d.). Thuiswinkel.org (n.d.) is expecting that in 2017, 27% of the purchases will be made online which will increase to 36% in 2020.
When engaging in e-commerce (i.e. shopping online), consumers are exposed to a large amount of product-related information and many stimuli to buy a specific product. This exposure of product-related information on the Internet can also be called the “participative web” (OECD, 2007). Many large online retailers, including, Bol.com, Wehkamp.nl, and Coolblue.nl are using this participative web by publishing product-related information on the products they offer (Gupta & Harris, 2010). Some examples of the participative web are communicating and opinion giving through social networks, blogs, forums and through online product reviews. There are two types of product-related information: firm-generated and user- generated content. Firm-generated content (FGC) consist of detailed information about the product’s features and price whereas user-generated content (UGC) is additional information created by users of a certain product, service or experience to share with others, i.e. online reviews (Tang, Fang & Wang, 2010). This study will dive deeper into online reviews.
Previous studies investigate already the importance of reviews in online shopping and the positive impact online reviews have on retail performance (Chevalier & Mayzlin, 2006). The presence of reviews in online shopping environments increases the amount of visits that will be converted into sales (Ludwig, De Ruyter, Friedman, Brüggen, Wetzels & Pfann, 2013; Zhu
& Zhang, 2010). Currently, researchers have become more interested into what elements of online reviews are perceived as more or less influential by consumers (Purnawirawan, De Pelsmacker & Dens, 2012).
When consumers shop online they feel some uncertainty in decision-making (Hu, Liu &
Zhang, 2008). This uncertainty can be reduced by getting more information and details about a product from informal, personal and buyer-oriented sources (e.g. online reviews) and thereby creating more trust (Hu et al., 2008; Roselius, 1971; Burton & Khammash, 2010).
Besides, if an online shop is not perceived as trustful, no purchase will result (Ang & Lee,
2000). Trust is an important concept in online reviews since it influences the extent to which a
consumer adopts information (Cheung, Sia & Kuan, 2012). Besides, consumers feel more
confident when finding reviews for their specific product online (Elberse & Eliashberg, 2003;
Zhu & Zhang, 2010) which makes the knowledge base around a product more trustful (Purnawirawan et al., 2012).
Therefore, it is important to understand what attributes of online reviews consumers trust on most in online shopping environments. Prior research already discussed some very important elements within online reviews, e.g. the valence (Kusumadrojaja, Shanka & Marchegiani, 2012) and volume (Dellarocas, Zhang & Awad, 2007). However, there may be more elements within online reviews which affect a consumer’s trust. Such as real name exposure may influence a consumer’s interpretation of a given review.
A lot of research has already been done with regards to advertising and how consumers process information. However, little research has been done on information processing in online reviews. According to a market study of Channel Advisor (2010), 92% of consumers who shop online read and use opinions of other consumers in their purchase decision-making.
Chen, Shang and Kao (2009) show that reviews affect purchasing behavior since consumers have different abilities in processing information (Chen & Lee, 2008). According to the dual process theory, there are two routes of information processing, namely one which takes low effort in information processing whereas the other takes high effort in information processing (Baek, Ahn & Choi, 2013). In this study, the elaboration likelihood model (ELM) of Petty, Cacioppo and Schumann (1983) is used to classify the different factors that influence a consumer’s trust with regard to online reviews, namely peripheral cue factors for a peripheral route in information processing and central content factors for a central route in information processing. Baek et al. (2013) investigated the helpfulness of online reviews based on review cues and readers’ objectives. However, other factors are not elaborated yet. This study focuses on trust and how this is influenced by the central content and peripheral cue factors of online reviews. Prior studies show already that after reading a message, a consumer will rely to a certain level on that message (Purnawirawan et al., 2012). Therefore it is important to know how consumers process information from online reviews and which factors, peripheral cues or central content, within online reviews they trust most on.
First, central content of online reviews, e.g. arguments and argument quality in a central route of information processing, are evaluations of previous shoppers which serves as a source of information for consumers (Zhu & Zhang, 2010; Chevalier & Mayzlin, 2006; Hu et al., 2008;
Baek et al., 2013). It gives consumers an opportunity to read information about how previous
about the quality of the product (Hu et al., 2008; Robson, Farshid, Bredican & Humphrey, 2013; Ziegele & Weber, 2015). Consumers perceive information provided by their peers as more trustful compared to information provided by marketers (Herr, Kardes & Kim, 1991).
Therefore, it is interesting to investigate different central content factors and to what extent those factors affect a consumer’s trust. For central content of online reviews, lexical complexity and two-sidedness will be explored. Second, peripheral cues of online reviews, e.g. non-content cues or source identity in peripheral route of information processing, enable consumers to assess a product quickly on a few cues without spending too much time on it in.
(Robson et al., 2013; Ziegele & Weber, 2015; Baek et al., 2013). It gives a short impression about the product’s overall performance. In this study, for peripheral cues the aggregated rating score volume, review helpfulness volume, real name exposure and reviewer’s expertise will be explored. Peripheral cue factors receive little attention in the literature. Hence, the amount of influence of the peripheral cue factors on a consumer’s trust compared to central content factors is interesting and therefore, worth studying.
Consumers evaluate and process information of online reviews differently depending on their personality and mood. Chen and Lee (2008) found in their study that when consumers are more agreeable and conscientious, a central route for website contents would be used whereas a peripheral route to website contents is used when consumers have greater emotional stability, openness to experience and extraversion. Second, consumers tend to focus on peripheral cues or central content factors depending on their mood (Bless, Bohner, Schwarz &
Strack, 1990). Consumers who are in a positive mood are more likely to use peripheral route whereas consumers who are in a more negative mood are more likely to use central route in information processing (Bless et al., 1990). Thus, personality traits and mood are included as potential moderators in this study. It is expected that central content factors of online reviews are more trusted by consumers who are in negative mood state and for consumers who are more agreeable and conscientiousness. Besides, it is expected that peripheral cue factors of online reviews are more trusted by consumers who are in a positive mood state and for consumers who have greater emotional stability, openness to experience and extraversion.
Concluded, the aim of this research paper is to answer the following research question:
To what extent do central content and peripheral cue factors affect a consumer’s trust of online reviews among different types of personality and moods?
In this study, data is gathered by means of a survey among 223 Dutch participants. First, a
choice-based conjoint (CBC) experiment with 13 choice sets has been conducted in order to
reveal consumer’s preferences and to investigate the extent to which the different attributes, central content and peripheral cue factors, affect a consumer’s trust with regard to online reviews on both, the aggregated and the segment level. Second, respondents were asked to rate 10 statements about their personality (BFI-10 scale) on a five-point likert scale and 4 items about their mood on a seven-point semantic differential scale.
Results show that central content factors are more important than peripheral cue factors in trusting online reviews. Lexical complexity is in particular very important where a high level of lexical complexity is preferred. Moreover, all peripheral cues and central content factors do affect a consumer’s trust of online reviews. It was very surprising that no significant differences were found between consumers in different mood states and with different personality traits. However, three different segments were defined with different consumer’s preferences, namely ‘observational learners’, ‘text critics’ and ‘independent consumers’
This study has both theoretical and practical relevance. Prior research has already focused on some important elements within online reviews. But, literature is lacking in making a clear distinction between peripheral cue factors and central content factors and to what extent consumers trust those two. Moreover, factors that affect a consumer’s trust within peripheral cues and central content information are underexplored in existing literature. This study has also strong practical relevance. This study gives managers insights in how they can optimize their online review environments. It has clear implications on how the management of an online retailer should design and manage their online review environment in such a way that it is as trustful as possible for their consumers, on the aggregated but also on the segment level.
2. R ESEARCH F RAMEWORK
This study investigates the extent to which central content and peripheral cue factors of online
reviews affect a consumer’s trust. First, trust in online shopping is explained. Second,
information processing theory is explained. Third, central content factors (lexical complexity
and two-sidedness of a review) are elaborated. Fourth, peripheral cues factors (aggregated
rating score volume, review helpfulness volume, real name exposure and reviewer’s
expertise) are investigated. Fifth, the expected moderators, the five-factor model of
personality and mood, are introduced. Finally, this section will end with the conceptual
model.
2.1. DEFINING TRUST IN ONLINE SHOPPING
Lack in trust is one of the largest barriers for consumers to shop online (Wang & Emeurian, 2005). Ang & Lee (2000) also argue that trust is a crucial concept in online shopping. It is important to understand how consumers are able to trust online information sources (Racherla, Mandviwalla & Connolly, 2012). Consumers perceive, process and respond to online reviews differently. This influences the extent to which the central content or peripheral cue factors of online reviews affect a consumer’s trust of online reviews (Cheung, Luo, Sia & Chen, 2009; Cheung et al., 2012; Racherla et al., 2012).
Keen, Balance, Chan and Schrump (1999) argue that trust “is so easy to talk about but so hard to pin down” (in McKnight, Choudhury & Kacmar, 2002; p. 334). Therefore, a clear definition of the concept trust is necessary. Trust can be defined in many different ways. Trust can be seen as the heart of relationships (Gefen, Karahanna & Straub, 2003), e.g. the relationship between the consumer and the company. According to the Oxford English Dictionary (1971), trust is defined as ‘‘confidence in or reliance on some quality or attribute of a person or thing or the truth of a statement’’ (p. 3423).
In online reviews, information about a reviewer’s actions, thoughts or motives is missing which creates a problem of information asymmetry (Racherla et al., 2012). Consumers have to resolve this asymmetry to make optimal decisions. The following definition of trust is best applicable to this study: trust is the extent to which consumers are willing to rely on information of online reviews and feel confident even when it could have negative consequences (i.e. a potentially misleading or false review) (McKnight et al., 2002; Racherla et al., 2012).
2.2. INFORMATION PROCESSING THEORY
Consumers have different abilities in processing information, i.e. how people deal with online reviews (Henry, 1980; Chen et al., 2009; Chen & Lee, 2008). Dual process theories make a distinction between two types of information processing, one which takes low effort in information processing whereas the other takes high effort in information processing (Baek, et al., 2013). ELM argues that there are two routes of information processing, namely the central and the peripheral route. In the central route, persuasion of a message is determined by the quality of arguments (Petty et al., 1983). Information processing occurs systematic, e.g.
understanding and evaluating a message’s arguments, and requires high effort and many
thoughts (Chen & Chaiken, 1999). In the peripheral route, persuasion of a message is
determined by the positivity of peripheral cues (Petty et al., 1983). Information processing is based on heuristics, e.g. source identity or non-content cues, and requires low effort and not so much thought (Chen & Chaiken, 1999). This study applies ELM to divide information in online consumer reviews into central content for central route in information processing and into peripheral cues for peripheral route in information processing. Thus, it can be argued that there may be differences in consumers who are more affected by central content factors in online reviews and in consumers who are more affected by peripheral cue factors in trusting online reviews. Next, central content factors and peripheral cue factors are elaborated.
2.2.1. CENTRAL CONTENT OF ONLINE REVIEWS
Currently, consumers have the freedom to share their opinion about products, services and/or experiences online. Central content factors are the arguments and their quality within a review which are mainly used in a central route of information processing (Baek et al., 2013). Within online reviews central content factors are peer-generated product evaluations published on a company or third party web sites (Mudambi & Schuff, 2010). Thus, consumers of a certain product publish their experiences online which then serve as a source of information for other consumers who have not yet experienced the product (Kusumasondjaja et al., 2012).
Central content within online reviews has become an important source of information for customers to discover the real quality of a product (Zhu & Zhang, 2010; Chevalier &
Mayzlin, 2006; Hu et al., 2008). It offers consumers an additional level of detail in information providing indirect experiences to consumers to reduce uncertainties of product quality (Hu et al., 2008; Robson et al., 2013). Herr et al. (1991) shows that information derived from previous consumers is perceived as more trustful than information derived from marketers. In this study, 2 central content factors are explored, namely lexical complexity and two-sidedness. Prior research already investigated to what extent these two factors contribute to the helpfulness of an online review (Korfiatis, Garcia-Bariocanal & Sánchez-Alonso, 2012;
Schlosser, 2011). Therefore, it is plausible that these two factors also have an effect on a consumer’s trust of online reviews.
LEXICAL COMPLEXITY
Online reviews often provide a consumer’s personal experience about a product (Hu, Bose,
Koh & Liu, 2012). They differ in terms of linguistic structure and semantic content i.e. the
level of lexical complexity. Online reviews with a high level of lexical complexity consist of
more technical terms, longer words and complex sentences (Jensen, Averbeck, Zhang &
Wright, 2013; Putrevu, Tan & Lord, 2004).
Lexical complexity is part of the readability of an online review, i.e. how easy an online review is to read (Consti-Ramsden, Durkin & Walker, 2012). When a review is easy to understand (i.e. low level of lexical complexity), it contains more subjective information instead of product attribute information (Korfiatis et al., 2012). Reviews that are easy to understand are more helpful than reviews that are difficult to understand (Korfiatis et al., 2012). Prior research argues that the level of cognitive effort in information processing, and in particular the level of cognitive fit, explains why a review should have a low level of lexical complexity (Park & Kim, 2009). Thus, when a message is difficult to understand, the consumer (i.e. the reader of the message) will draw a negative conclusion about the communicator (i.e. the writer of the review) (Schindler & Bickart, 2012). The information processing capacity is inefficient and ineffective since consumers are not able to use appropriate cognitive processes when the message has a high level of lexical complexity (Park & Kim, 2009). Jensen et al. (2013) also found in their study that a high level of lexical complexity decreases a reviewer’s credibility. Therefore, it is expected that an online review with a high level of lexical complexity has a negative influence on a consumer’s perceived trust of online reviews, as it is likely that they are considered as less credible and less helpful.
The following hypothesis is investigated:
H
1: Lexical complexity has a negative influence on a consumer’s trust.
TWO-‐SIDEDNESS
Although, most of the reviews online are one-sided: they are positive or negative about a certain product (Jensen et al., 2013). Two-sided reviews provide more diverse information than one-sided reviews (Schlosser, 2011). According to Jensen et al. (2013), two-sidedness is providing both, positive and negative arguments in an online review. They illustrate the performance of a certain product from two sides.
The reliability of two-sided reviews has been studied mostly in the context of advertising.
Many studies found that putting some negative information in an advertisement enhances the reliability of the advertisement (Boher, Einwiller, Erb & Siebler, 2003; Kamins & Assael;
1987). Schlosser (2011) explains in her study that negative aspects in a review of a product
show that the reviewer does not only have self-interest in selling its product but wants to show
all aspects. Consequently, consumers find the online review as more trustful than when it
shows only one side (Schlosser, 2011). Besides, it increases the likelihood that expectations which were made upfront take place since both, positive and negative aspects are mentioned.
Therefore, it can be hypothesized that:
H
2: Two-sidedness has a positive influence on a consumer’s trust.
2.2.2. PERIPHERAL CUES OF ONLINE REVIEWS
Peripheral cues are heuristics, e.g. non-content cues or source identity, in a peripheral route of information processing (Petty et al., 1983). They require low cognitive effort in persuasion and are focused on simple acceptance or rejection cues. Peripheral cues are an “easy draw on the eye of the brain” (Robson et al., 2013; p.6). It enables consumers to process information quickly since peripheral cues are easy and fast to judge (Robson et al., 2013). Peripheral cues give a short impression about the product’s overall performance. In this study, 4 peripheral cue factors are explored, namely aggregated rating score volume, review helpfulness volume, real name exposure and reviewer’s expertise. Prior research focuses mainly on the extent to which these peripheral cues are considered as helpful or credible (Jensen et al., 2013; Chen, Wang & Xie, 2011). So, it is plausible that these four factors also have an effect on a consumer’s trust with regard to online reviews.
AGGREGATED RATING SCORE VOLUME
The concept of observational learning might enter when consumers process the aggregated rating score volume. Observational learning is an action- or behavior based social interaction in which consumers can learn from and are affected by other consumer’s opinions (Chen et al., 2011). Observational learning gives consumers limited information, i.e. only the actions.
For example, when choosing a restaurant, a person is influenced by how many people are already dining in the restaurant, even without knowing those people and why they choose that restaurant (Becker, 1991). Observational learning fits with heuristic information processing because consumers observe and process only the actions of other consumers, which reflect the overall popularity of a certain product, in decision-making (Qui & Li, 2010; Purnawirawan et al., 2012).
Previous studies confirm that people trust more on a product if it is already recommended by
other consumers (Kriby, 2000; Purnawirwan et al., 2012). For example, if 100 people
recommend a product, the average likability will be more trustful than if only 10 people
recommend it. Besides, products or services that are recommended by other consumers are
selected twice as often as products and services that didn’t receive recommendations (Senecal
& Nantel, 2004). Although the aggregated rating score volume does not provide direct information about the trustworthiness of the rating score, prior research argues that a higher rating volume (i.e. the more people rate a product), matches better to the true quality of a product which creates a higher level of consumer trust (Shubert & Ginsberg, 2000; Chen, Wun & Yoon, 2004). Thus, the following hypothesis will be tested:
H
3: Aggregated rating score volume has a positive influence on a consumer’s trust.
REVIEW HELPFULNESS VOLUME
Observational learning might also enter when consumers process the review helpfulness volume. Based on the total number of review helpfulness votes (i.e. … people found this review helpful), someone finds a review very helpful by observing how many other people found a certain review helpful. A helpful review is “a peer-generated product evaluation that facilitates the consumer’s purchase decision process” (Mudambi & Schuff, 2010). It is a rating which measures if the review was helpful to other consumers. Consumers can learn from and be affected by reviews which are written by other consumers (Chen et al., 2011). To evaluate a review, review helpfulness volume is another factor within peripheral cues that gives consumers information about other consumer’s actions (Mudambi & Shuff, 2010). Chen et al. (2011) argue that a person is more likely to act according to the information of previous consumers (e.g. helpfulness of a review) if more consumers already found a certain review helpful. Concluding, although observational learning only gives ratings (actions) and not the reasons behind, it can have a large impact on consumer decisions. “Actions speak louder than words”, and therefore a review with more helpfulness votes might be perceived as more trustful relative to a review with no helpfulness votes (Chen et al., 2011; p. 240). So, it is hypothesized that:
H
4: Review helpfulness volume has a positive influence on a consumer’s trust.
REAL NAME EXPOSURE
When reading online reviews, consumers do not have a lot of information about a reviewers’
action, thoughts and motives (Racherla et al., 2012). Although, many online reviews provide
some kind of information about the reviewer as well as they provide information about the
product (Forman, Ghose & Wiesenfeld, 2008). Mostly, online reviews are provided by other
consumers with whom the reader has no relationship (Chevalier & Mayzlin, 2006).
Source credibility has an important role in online information adoption (Briggs, Burford, De Angeli & Lynch, 2002). It is an important cue in processing information (Cheung et al., 2012). Prior research argues that social information about the reviewer is often used as heuristic in the decision-making process (Chaiken, 1980). Gershoff, Broniarczyk and West (2001) argue that source information is important for determining a review’s usefulness.
Thus, source information, e.g. a reviewer’s identity, plays an essential role in shaping how consumers respond to an online review when processing information (Wood, 2000). One element of source information is a real name exposure. Exposing a reviewer’s real name enhances source credibility, which leads to information credibility and usefulness (Sussman &
Siegal, 2003; Forman et al., 2008). Exposing a reviewer’s real name is viewed as more favorable and useful than showing a nickname or being anonymous (Forman et al., 2008).
Racherla et al. (2012) argued that reviews which are anonymous in nature, i.e. not showing a reviewer’s real name, are harder to assess on quality and are therefore perceived as less influential.
This study proposes that reviews with a real name exposure will be perceived as more trustful than review with a nickname or anonymous review, as it is likely that anonymous reviews are harder to assess on quality and nicknames are less useful. Thus, it is hypothesized that:
H
5: Real name exposure has a positive effect on a consumer’s trust.
REVIEWER’S EXPERTISE
As stated earlier, everyone can place an online review: from novice to expert. Unfortunately, there are more novices than experts and therefore, experts are in the minority of writing reviews. Consumers do not limit themselves to the content and cues in assessing an online review, but they also take the credibility of the reviewer into account (Hu et al., 2008).
Expertise is defined as the extent to which a source is capable of providing correct information and expertise to induce persuasion (Bristor, 1990, in Bansal & Voyer, 2000).
Dholakia and Sternthall (1977) found that expertise is a major dimension in how consumers trust a certain source. Jensen et al. (2013) argued that a review which is written by an expert is perceived as more credible than when there is no reviewer’s expertise. Hu et al. (2008) also state that reviews written by reviewers with more knowledge and higher quality content are perceived as more credible and trustworthy.
In this study, it is expected that a review written by an expert is perceived as more trustful, as
it is likely that they are considered as more credible and trustworthy. So, it is hypothesized that:
H
6: Reviewer’s expertise has a positive influence on a consumer’s trust.
2.3. MODERATOR
2.3.1. PERSONALITY TRAITS
Personality traits argue that people differ in their motivation to process information (Chen &
Lee, 2008). Personality traits describe people in terms of thoughts and emotions (Parks- Leduc, Feldman & Bardi, 2015). According to McCrae and John (1992), there are five basic dimensions in which individuals differ in their enduring, interpersonal, attitudinal and motivational styles, namely extraversion, agreeableness, conscientiousness, neuroticism and openness to experience. In literature, those five basic dimensions are often mentioned as the five-factor model (FFM) of personality (table 1) (McCrae & John, 1992; Sadowski &
Cogburn, 1997; Thoms, Moore & Scott, 1996; Parks-Leduc et al., 2015).
Table 1. Five-‐Factor Model of Personality
Dimension A person’s degree of …
Extraversion Assertiveness, sociability and energy.
Agreeableness Friendliness, cooperativeness, being courteous, flexible and trusting.
Conscientiousness Dependability, achievement orientation and perseverance.
Neuroticism Emotional stability, anxiety, self-confidence, self-consciousness.
Openness to experience Imaginativeness, curiosity, open-mindedness and being artistic.
According to the ELM, it differs on a person’s level of elaboration how he/she deals with a
persuasive message (Chen & Lee, 2008). Consumers with greater emotional stability,
openness to experience and extraversion are having a more positive attitude toward hedonic
value-seeking behaviors (e.g. fun, playfulness) whereas consumers who are more agreeable
and conscientious are having higher utilitarian value-seeking behaviors (e.g. cognition,
evaluating information) (Karl, Peluchette & Harland, 2007). Thus, when consumers have
greater emotional stability, openness to experience and extraversion, a peripheral route for
website contents would be used whereas a central route to website contents is used when
consumers are more agreeable and conscientious (Chen & Lee, 2008). Therefore, it is
expected that the FFM of personality has a moderating role in this study. It is expected that
consumers who are more agreeable and conscientious trust more on the central content factors
within online reviews whereas consumers who have greater emotional stability, openness to
experience and extraversion trust more on peripheral cue factors, as it is likely that consumers who are more agreeable and conscientious follow a central route in information processing whereas consumers who have greater emotional stability, openness to experience and extraversion follow a peripheral route in information processing.
H
7a: For consumers who are more agreeable and conscientious, central content factors are more influential in trusting an online review than peripheral cue factors.
H
7b: For consumers with greater neuroticism, openness to experience and
extraversion, peripheral cue factors are more influential in trusting an online review trust than central content factors.
2.3.2. MOOD
Mood is a consumer’s affective state representing positive or negative feelings in a specific situation, e.g. advertisement exposure, brand selection etc. (Eagly & Chaiken, 1993; Gardner, 1985). Mood is subjectively perceived, low in intensity and lacking of source identification (Cohen, Pham & Andrade, 2008; Gardner, 1985). For example, someone may be in a bad or cheerful mood without knowing why he/she feels that way (Gardner, 1985). Small changes in the environment may influence a consumer’s mood (e.g. weather, exposure to happy versus sad information, hormonal activity) (Cohen et al., 2008).
Mood does not only influence a consumer’s judgment or decision-making but it also influence the way information is processed (Bless et al., 1990; Hüttermann & Memmert, 2015).
Consumers who are more in a negative mood use a more systematic and effortful information
processing in an attempt to improve their negative mood state (Clark & Isen, 1982; Sedikides,
1994). Negative moods ensure that consumers perform a more data-driven and analytical
form of reasoning while relying less on general knowledge structures such as scripts and
stereotypes (Bless et al., 1990). Consumers who are in a positive mood may spend less effort
in information processing to protect their current mood and use heuristic information
processing (Clark & Isen, 1982; Isen, 1984). Cohen et al. (2008) showed that consumers who
are in a positive mood seem to perform a less data-driven and thorough mode of processing
while promoting greater flexibility and creativity in solving problems (Cohen et al., 2008)
Therefore, it is expected that mood has a moderating role in this study. It is expected that
consumers who are in a positive mood trust more on the peripheral cues of online reviews
whereas consumers who are in a negative mood trust more on the central content of online reviews.
H
8a:For consumers with a positive mood, peripheral cue factors are more influential in trusting an online review than central content factors.
H
8b:For consumers with a negative mood, central content factors are more influential in trusting an online review than peripheral cue factors.
2.4. CONCEPTUAL MODEL
Based on the these theoretical expectations, a conceptual model (figure 1) is created with the different factors, central content and peripheral cues, and the two moderators, that are
expected to affect a consumer’s trust of online reviews.
Figure 1. Conceptual Model
Lexical complexity
Two-sidedness Central
content
Aggregated rating score volume
Helpfulness of a review volume
Reviewer’s expertise Peripheral cues
Trust
Personality traits Agreeableness Conscientiousness
Neuroticism Extraversion Openness to experience H1-H2
H3-H6
H7b H7a
Mood
H8b
H8a
Real name exposure
3. R ESEARCH D ESIGN
In this section the research design is discussed. First, the chosen research method, choice- based conjoint (CBC), is explained. Second, data collection is elaborated in which the sampling method, attribute and levels measurement scale and the moderator measurement scale are explained. Third, a plan of analysis is presented.
3.1. RESEARCH METHOD
Conjoint analysis is a multivariate preference measurement technique to gain insights in underlying motives and preferences for consumers’ actions (Vriens, 1994). It measures trade- offs among multi-attributed products and services (Green & Rao, 1971; Green & Srinivasan, 1990). Conjoint analysis has a decompositional structure. Respondents evaluate products or services on different product attributes which are specified in terms of levels (Eggers &
Sattler, 2011; Loviere & Woodworth, 1983; Cattin & Wittink, 1982). Conjoint analysis is for this study the most appropriate because the main purpose is to investigate the extent to which consumers trust different attributes (i.e. factors) of central content and peripheral cues.
Conjoint analysis makes it possible to analyze data on the aggregated and on the segment level. Analyzing data on segment level gives opportunities in defining groups of respondents who differ in making choices on how they trust attributes. Furthermore, different types of consumer behavior can be explained after comparing the different segments (Papies, Eggers
& Wlömert, 2011). Therefore, this study analyzes data on both aggregated and segment level.
This study uses a CBC method in which respondents choose the most preferred option in a presented selection of stimuli (Loviere & Woodworth, 1983). Choosing one option is a natural task for respondents compared to traditional conjoint methods where respondents have to rate or rank products (Toubia, Houser & Simester, 2004; Eggers & Sattler, 2011; Johnson
& Orme, 1996). Thus, CBC is a very effective and accurate method (Toubia et al., 2004).
CBC makes it able to create different choice sets with the central content and peripheral cue
factors in which consumers have to choose the alternative they trust most. With CBC it is
possible to measure the relative importance of the different levels of attributes and by
multiplying the utilities of the attribute levels the relative degree of trust is obtained (Johnson,
1974).
3.1.1. ATTRIBUTES AND LEVELS
In this study, online reviews are decomposed into their main factors, the six different attributes. The selection of these attributes is based on insights from literature research. The product used in this study was a tablet. A tablet was chosen in this study since it is a search good. Thus, consumers can compare and evaluate information about the product’s quality before buying it (Mudambi & Schuff, 2010).
For central content, the main attributes in this study are lexical complexity and two-sidedness.
For peripheral cues, the main attributes in this study are aggregated rating score volume, review helpfulness volume, real name exposure and reviewer’s expertise. Table 2 presents an overview of the different attributes and their levels which are used in the CBC.
Table 2. Attribute and Attribute Levels
Attributes Level 1 Level 2 Level 3
Central content factors
Lexical complexity Low Medium High
Two-sidedness One-sided Two-sided
Peripheral cue factors
Aggregated rating score volume 2 19 43
Review helpfulness volume 0 2 5
Real name exposure Anonymous Nickname Real name
Reviewer’s expertise No label Top reviewer Product expert
The data collection design had accounted for minimal overlap and orthogonality (Huber &
Zwerina, 1996). However, a problem in level balance could arise due to the so called ‘number of levels’ effect, i.e. number of levels are not distributed equally across the different attributes (Eggers & Sattler, 2011). This can lead to a higher relative importance of the attributes with more levels. In this study only one attribute, two-sidedness, had two levels compared to the five other attributes with three levels. There is a small difference of one level and therefore it is expected that the number of level effect will not cause disruption in the data.
Documented qualitative research on the top 3 online web shops in electronics in the Netherlands (i.e. Bol.com, Wehkamp.nl and Coolblue.nl) has been done in order to operationalize the different levels within this CBC. It is important to mention that the valence of the reviews used in this study is kept constant since the effect of valence is not the aim of this study. Thus, all reviews contain of positive content in which the levels per attribute vary.
First, central content attributes will be explained. Lexical complexity consists of three levels,
low, medium and high lexical complexity. Online reviews of different tablets from these 3
online web shops were compared and inspected to create the three levels of lexical complexity in this study. A low level of lexical complexity only contains of subjective information whereas a high level of lexical complexity contains product attribute information (Korfiatis et al., 2012; Park & Kim, 2009). A medium level of lexical complexity is in between a low and high level of lexical complexity and therefore contains both, subjective and product attribute information. Two-sidedness has two levels (Jensen et al., 2013). On the one hand a review can be two-sided, i.e. contains both positive and negative elements in the message. On the other hand a review can be one-sided, i.e. a positive or negative message. In this study, the one-sided review is positive. Second, peripheral cue attributes will be explained. Aggregated rating score volume and review helpfulness volume was inspected from different tablets on the websites of those three online retailers which gave input in deciding the different levels for these two attributes. The utility that respondents associate with aggregated rating score volume is likely to depend on the number of people that already give the product a rating score (Kriby, 2000; Purnawirwan et al., 2012). All the levels consist of a 4-star rating, so the valence is kept constant, whereas the rating score volume changes between the three levels since volume is of only interest in this study. The three levels of aggregated rating score volume are: ‘(2)’, ‘(19)’, ‘(43)’. The utility that respondents associate with review helpfulness volume is likely to depend on the number of people that found a certain review helpful (Chen et al., 2010). Helpfulness of a review consists of three levels: ‘this review was helpful to me?
Yes (0), ‘this review was helpful to me? Yes (2)’ and ‘this review was helpful to me? Yes (5)’. To capture consumer preferences on trust with regard to real name exposure, three attribute levels are used: anonymous, nickname and real name (Forman et al., 2008; Racherla et al., 2012). The anonymous level will be presented in the experiment with ‘Anoniem’, nickname will be presented with ‘Ray 058’, and real name will be presented with ‘Floris van Hamel’. To capture reviewer’s expertise with regard to trust, a reviewer can receive three labels, namely no label, ‘product expert’ label and ‘top reviewer’ (Hu et al., 2008). A reviewer can receive such a label from the online retailer. A reviewer receives a ‘product expert’ label when he/she has a lot of knowledge about the product. Reviewers who receive a
‘product expert’ label mainly write high quality reviews. A reviewer receives a ‘top reviewer’
label when he/she has a lot of media exposure on this online shop, i.e. the reviewer is very
experienced in writing online reviews for the web shop (Hu et al., 2008).
3.1.2. PERSONALITY TRAITS MODERATOR MEASUREMENT SCALE
The FFM of personality traits, extraversion, agreeableness, conscientiousness, neuroticism, and openness to experience could have a moderating role in this study. To explore this moderating role, consumers are exposed to the Big Five Inventory-10 (BFI-10). The BFI-10 is a measurement scale to measure a respondent’s personality in one minute (Rammstedt &
John, 2007). This short measurement scale is chosen since other personality scales consist of more statements so it takes longer to complete, namely the BFI-44 contains of 44 statements (John, Donahue & Kentle, 1991) and the NEO-PR-I contains of 60 statements (Costa &
McCrae, 1992). Ramstedt and John (2007) found in their study that the BFI-10 correlates part-whole with the BFI-44, has retest reliability and structural and convergent validity with NEO-PR-I and its facets. Rammstedt and John (2007) selected two items per personality trait from the BFI-44 following five criteria: (1) both, the high and the low pole of each factor were selected, so that each BFI-10 scale consists a true- and false-scored item. (2) Two items that measure core aspects of a dimension were chosen but were not highly redundant in content. (3) English and German-language versions were constructed for cross-cultural research. Items were selected based on two criteria (4)(5): “their corrected item-total correlations with the full BFI-scales and the simple structure pattern of their loadings in factor analysis of all 44 items” (p. 205). Table 3. represents the BFI-10 with a Dutch translation (Denissen, Geenen, Van Aken, Gosling, & Potter, 2008). The ten BFI-10 statements had to be answered on a five-point Likert scale from totally disagree to totally agree i.e. “helemaal mee oneens tot helemaal mee eens”.
Table 3. Dutch Big Five Inventory-‐10 (BFI-‐10)
Personality Trait I see myself as someone who… Ik zie mezelf als iemand die…
Extraversion … is reserved … terughoudend is
Agreeableness … is generally trusting … mensen over het algemeen vertrouwt Conscientiousness … tends to be lazy … geneigd is lui te zijn
Neuroticism … is relaxed, handless stress well … ontspannen is, goed met stress kan omgaan
Openness to experience … has few artistic interests … weinig interesse voor kunst heeft Extraversion … is outgoing, sociable … hartelijk, een gezelschapsmens is Agreeableness … tends to find fault with others … geneigd is kritiek te hebben op
anderen
Conscientiousness … does a thorough job … grondig te werk gaat
Neuroticism … gets nervous easily … gemakkelijk zenuwachtig wordt Openness to experience … has an active imagination … een levendige fantasie heeft
3.1.3. MOOD MODERATOR MEASUREMENT SCALE
To explore the moderating role of mood, respondents are exposed to four items on a seven- point semantic differential scale (table 4) (Brunner, 2009). The scale measures someone’s general mood at a particular point in time. Thus, the scale measures if someone’s in a good/positive or bad/negative mood at the moment they fill in the scale.
The scale was originally developed by Allen and Jansizewski (1989). They found an alpha of .72 for their scale. Roehm and Roehm (2005) used the word unhappy, as is used in this study, instead of sad and found alphas ranging from .80 - . 91.
Table 4. Mood scale.
At this moment I am feeling … Op dit moment voel ik me …
Good / Bad Goed / Slecht
Unpleasant / Pleasant Onprettig / Prettig
Happy / Unhappy Gelukkig / Ongelukkig
Negative / Positive Negatief / Positief
3.2. CHOICE DESIGN
An important topic in CBC is the amount of choice tasks to give to each respondents. Too many choice tasks may result in bias due to fatigue-effects whereas too few will reduce precision (Johnson & Orme, 1996). Therefore, it is important to find balance in the amount of choice tasks. According to Hair, Black, Babin and Anderson (2010), the upper limit on number of attributes of a CBC is six. Second, the amount of levels per attribute and the number of alternatives per choice set should be between two and five. This study satisfies these criteria.
The minimum number of choice sets that must be evaluated by each respondent can be calculated by the following formula (Hair et al., 2010):
𝑀𝑖𝑛𝑖𝑚𝑢𝑚 𝑛𝑟. 𝑜𝑓 𝑐ℎ𝑜𝑖𝑐𝑒 𝑠𝑒𝑡𝑠 = 𝑇𝑜𝑡𝑎𝑙 𝑛𝑟. 𝑜𝑓 𝑙𝑒𝑣𝑒𝑙𝑠 𝑎𝑐𝑟𝑜𝑠𝑠 𝑎𝑙𝑙 𝑎𝑡𝑡𝑟𝑖𝑏𝑢𝑡𝑒𝑠 − 𝑁𝑟. 𝑜𝑓 𝑎𝑡𝑡𝑟𝑖𝑏𝑢𝑡𝑒𝑠 + 1. This study contains of six attributes with three alternatives per choice set. Thus the minimum number of choice sets calculated by the formula of Hair et al. (2010) is thirteen (18-6+1). Eggers (2015) found in his study, based on empirical research, that respondents are also willing to respond to thirteen different choice-sets when there are six different attributes with three alternatives per choice set. Therefore, thirteen choice sets is the optimal number of choice sets in this study.
This study does not contain a no-choice option for several reasons. First, the no-choice option
should only be included whenever price-senstivity, willingness-to-pay or purchase intentions
are measured which is not the case in this study (Eggers & Sattler, 2011). Second, consumers might choose the no-choice option to avoid difficult decisions. Third, consumers might feel that some information (attributes) are missing and therefore prefer the no-choice option (Gunasti & Ross, 2009). In this study, a no-choice option doesn’t add realism to the task and therefore it will not be included (Johnson & Orme, 1996).
The dependent variable in is trust. In this study, trust is defined as the extent to which consumers are willing to rely on central content or peripheral cues of online reviews and feel confident even when it could have some negative consequences. Thus, trust is measured by asking respondents to answer the following question at each choice task: “on which review would you trust most?” Figure 2 represents an example of a choice set.
Figure 2. Example of a choice set
3.3. DATA COLLECTION
Data in this study is collected a self-administrated questionnaire (appendix 1). Data is
collected via the Internet. This channel was chosen since it is an easy and fast way to
distribute the questionnaire. Besides, online data collection makes it possible to expose
respondents randomly to choice sets. Randomization has the advantage that it is possible to
aggregate data question-by-question. The data presented in this study is collected via preferencelab.com. Preferencelab.com is chosen because it is an online tool for conducting and analyzing surveys which is specialized in consumer preferences via CBC analysis.
Preferencelab.com provides optimal choice sets by randomization. In this study 729 profiles could be designed since there are 6 attributes with each three levels (=6
3). It is not manageable to ask each respondent 729 profiles so the number of profiles is reduced to 13 choice sets via a fractional factorial design. By applying a fractional factorial design, preference lab accounts for minimal overlap and an orthogonal design.
The online questionnaire starts with an introduction of this study. Second, respondents had to ask several demographic questions which were followed by the questions about a respondent’s current mood. After those general questions a short explanation of CBC is given.
Next, respondents were exposed to the thirteen different choice sets in which they had to choose every time one of the three options given. Lastly, the questionnaire contains of 10 statements about personality.
A total sample of 223 Dutch respondents was collected in April 2015. Respondents were approached via e-mail, Facebook and LinkedIn. According to Hair et al. (2010) a sample size of 200 respondent is adequate for conjoint analysis. Results are estimated in Latent Gold Choice.
3.4. PLAN OF ANALYSIS
This study tests the conceptual model (figure 1) whereby it is investigated if there is a relation between different attributes and trust and the impact of moderators on this relationship.
The first step after collecting the data is to test the construct validity of items that consist of
more than one item, namely mood and the items within personality traits. Construct validity
measures whether the measurement scales for one item actually represent the investigated
construct (Hair et al., 2010). Constructs validity of constructs with more than two items are
measured via convergent validity. Reliability measures show whether or not the separate
measurements can be combined into one construct. Cronbach’s alpha should be larger than
α=0.6. Construct validity of constructs with only two items are measured via bivariate
correlation. If the two items correlate with each other it is allowed to combine them into one
construct. If the items can be combined into one construct they will be used as covariate in the
CBC analysis to measure whether they have a moderated relationship between the attributes
The second step in conjoint analysis is the specification of the model. Respondents base their choices on overall utilities of alternatives within a choice set. These utilities are summed up part-worth estimates for a specified set of levels (Hair et al., 2010). Respondents will choose the alternative with the highest utility.
The dependent variable is the by respondents chosen alternative (i.e. review they trust on most) selected from the choice set and the explanatory variables are the peripheral cue and central content attributes. The following formula for the utility of review (U) per segment (i) is used for the initial model:
𝑈!= 𝛽!!𝐴𝑁𝑂 + 𝛽!!𝑁𝑁 + 𝛽!!𝑅𝑁 + 𝛽!!𝐿𝐶 + 𝛽!!𝑀𝐶 + 𝛽!!𝐻𝐶 + 𝛽!!𝑂𝑆 + 𝛽!!𝑇𝑆 + 𝛽!!𝐿𝑅 + 𝛽!"!𝑀𝑅 + 𝛽!!!𝐻𝑅 + 𝛽!"!𝐿𝐻 + 𝛽!"!𝑀𝐻 + +𝛽!"!𝐻𝐻 + 𝛽!"!𝑁𝐿 + 𝛽!"!𝑃𝐸 + 𝛽!"!𝑇𝑅
Where, i = 1,2,3,... I
U = Utility i = Segment
ANO = Anonymous NN = Nickname RN = Real name
LC = Low lexical complexity MC = Medium lexical complexity HC = High lexical complexity OS = One-sided
TS = Two-sided
LR = Low aggregated rating score volume MR = Medium aggregated rating score volume HR = High aggregated rating score volume LH = Low review helpfulness volume MH = Medium review helpfulness volume HH = High review helpfulness volume NL = No reviewer’s expertise label PE = Product expert label
TR = Top reviewer label
