How can online reviews facilitate decision making:
The impact of semantic and linguistic characteristics
on the helpfulness of online consumer reviews
Dániel Hegedüs
June 20, 2016
University of Groningen
Faculty of Economics and Business
2
How can online reviews facilitate decision making:
The
impact of semantic and linguistic characteristics on the
helpfulness of online consumer reviews
University of Groningen
Faculty of Economics and Business
MSc Marketing Management
Author: Dániel Hegedüs
Date: June 20, 2016
Address: Illegaliteitslaan 166
9727 EG, Groningen
Phone number: 0031 641784136
E-mail address: d.hegedus@student.rug.nl
Student number: 2991861
First supervisor: Dr. J. A. Voerman
3
Management Summary
This study examined the effects of semantic and linguistic online consumer review characteristics on the helpfulness of these reviews. More specifically, it investigated how online reviews can assist in the decision making of consumers. In terms of the semantic characteristics, an online review can be objective or subjective depending on the extent to which it is conveying factual product features or on the contrary, is mainly based on biased personal opinions. A review can also be differentiated on its concreteness. In this regard, a concrete online review provides readers with specific product experiences, while in contrast, an abstract review includes unclear and ambiguous statements. We can also distinguish between high quality linguistic style which supports the statements conveyed in the review by a proper use of language, and low quality linguistic style with grammar and spelling mistakes, and also using inexpressive slang words. These three main characteristics influence the argument quality and diagnosticity of the online reviews, that is, how the arguments in the reviews are supported by facts, and how the reviews help in reducing the ambiguity and uncertainty of the consumers, respectively. These two concepts in turn lead to the helpfulness of online consumer reviews.
To examine these effects, an online survey was conducted based on a mixed within-participants experimental design, in which a total of 226 respondents participated. This measured the helpfulness of online reviews perceived by consumers, depending on the different quality conditions of the review characteristics.
The results show that objective reviews were perceived more helpful than subjective reviews, concrete reviews were more helpful than abstract ones and reviews with high quality linguistic style were seen as more helpful than the ones with low quality linguistic style. Therefore, all the inspected main effects of the independent variables on argument quality, diagnosticity and in turn on helpfulness were found to be positively significant. Moreover, argument quality mediates the effect of the independent variables on diagnosticity, and the effect of argument quality on helpfulness is partially mediated through diagnosticity.
4
Acknowledgements
With this thesis, I am reaching the end of my studies at the University of Groningen. This year, which I have spent with my Master’s Marketing Management programme, has been one of the most challenging in my life so far, but at the same time I am convinced that I have learned the most during this period, and that this knowledge will greatly help me in the coming years.
First of all, I would like to thank my supervisor, Dr. Liane Voerman for her guidance and valuable feedback over the past months. Also, I would like to thank my fellow thesis group members for their useful advice and suggestions during the meetings.
Furthermore, I would like to thank my friends for distributing the survey of this thesis, and also my respondents for taking the effort to answer my questionnaire.
Finally, I would like to thank my parents, to whom I am really grateful for their help and their enormous support during this time.
June, 2016
5
Table of Contents
1. Introduction
7
1.1 Helpfulness of online reviews 8
1.2 The valence and length of online consumer reviews 9
1.3 The semantic and linguistic characteristics of online reviews 9
1.3.1 Semantic characteristic: Objectiveness 10
1.3.2 Semantic characteristic: Concreteness 10
1.3.3 Linguistic characteristic: Linguistic style 10
1.4 Elaboration likelihood Model 10
1.5 Search goods and experience goods 11
1.6 Problem statement 11
1.7 Research questions 12
1.8 Academic and managerial relevance 12
1.9 The structure of the thesis 13
2. Theoretical Framework
14
2.1 Diagnosticity 14 2.2 Argument quality 15 2.3 Objectiveness 16 2.4 Concreteness 18 2.5 Linguistic style 192.6 Interaction effects of the independent variables 20
2.7 Frequency of online shopping and frequency of reading online reviews 21
2.8 Conceptual model 22
3. Research methodology
24
3.1 Research design 24
3.2 Population 25
3.3 Design of the survey 26
3.4. Operationalization of concepts 28
3.5. Factor Analysis and Reliability Analysis 31
3.6. Plan of data analysis 32
6
4. Results
36
4.1 Condition means 36 4.2 ANOVA 37 4.3 Regression analysis 37 4.4 Mediation analysis 40 4.5 Hypothesis testing 425. Discussion
47
5.1 Theoretical implications 47 5.1.1 Main effects 47 5.1.2 Interaction effects 49 5.2 Managerial implications 505.3 Limitations and directions for further research 50
6. References
52
7
1. Introduction
With the ubiquitous use of the web 2.0, consumers can easily create and share large amounts of information in the form of electronic word-of-mouth (eWOM, which is one manifestation of the wide range of user-generated-content), telling other consumers about different kinds of products and services (Cheung & Thadani, 2012). Hennig-Thurau et al. (2004) define electronic word-of-mouth communication or eWOM as “any positive or negative statement made by potential, actual, or former customers about a product or company, which is made available to a multitude of people and institutions via the Internet.” Online consumer reviews (OCRs, online product reviews or simply reviews) are one specific form of eWOM, which can be described as information displayed by consumers who have acquired and have been using a particular product, and which encompass the “experiences, evaluations, and opinions” of consumers (Park et al., 2007). According to them, these pieces of product- and service-related information on the internet allow consumers (in addition to the information provided by companies) to read and assess information related to the actual usage experiences of reviewers, based on which they can evaluate and compare product attributes and benefits that can help them in their decision making processes.
8
So, on many platforms, similar to Amazon.com, there are often hundreds of reviews for each product, out of which only a small proportion are concerned highly helpful, as is also reflected in the high number of helpfulness votes these reviews receive. Therefore, it is an interesting question to examine what makes a consumer review really helpful for readers, and what are the specific key features that really facilitate the decision making of consumers.
1.1 Helpfulness of online reviews
Helpfulness of online product reviews can be defined as “the extent to which consumers perceive the product review as being capable of facilitating judgment or purchase decisions.” (Li et al., 2013). Helpfulness can be considered a highly central concept in the research of online reviews posted on various consumer platforms, since it can have an influence on many other related entities, such as the number or ratio of helpfulness votes of particular reviews which can help readers to make a decision more quickly, since this way they only need to read a small number of reviews (Cao et al., 2011). Helpfulness can also determine the readership of OCRs, more specifically, reviews that are judged to be highly helpful will be read by more people, since many websites, such as Amazon.com position these reviews in a prominent field (Lee, 2013). Review helpfulness may also lead to higher purchase intentions, as helpful reviews with high argument quality that approve of a product or service, would likely help readers to make an informed decision and prompt them to consider buying that particular offering (Zhang et al., 2014). Consequently, helpfulness can have a significant influence on the sales of these offerings (Hu et al., 2014).
One essential determinant of helpfulness is diagnosticity, which encompasses the extent to which a review can lessen readers’ uncertainty by providing a clear explanation about the focal product (Weathers et al., 2015). More specifically, how can these reviews help readers distinguish between different assumptions concerning the product (Herr et al., 1991).
9
diagnosticity. Since the different characteristics examined in this study influence both argument quality and diagnosticity, argument quality mediates the effect of these characteristics on diagnosticity.
Based on the aforementioned findings, it is an intriguing question to investigate, what drives the argument quality and the diagnosticity of online consumer reviews and also what is the relationship between these concepts. More specifically, how does argument quality influence diagnosticity and the helpfulness of online reviews, and also what effect does diagnosticity have on helpfulness?
1.2 The valence and length of online consumer reviews
In the widespread research on the different antecedents of review helpfulness, there is a large pool of studies which has examined the effect of valence and length on the helpfulness of online reviews (Pan & Zhang, 2011; Cao et al., 2011; Baek et al., 2012). However, in this study I am particularly interested in the semantic and linguistic characteristics of the text conveyed in online consumer reviews, since it is likely that neither valence, nor length can fully comprehend the different aspects that determine the actual helpfulness of each online review in the decision making of readers, but only guide them to make a quick and superficial judgement from a number of reviews. In relation to this issue, Chevalier & Mayzlin (2006) have found that readers of online reviews prefer textual information included in the reviews compared to the numerical, average star ratings presented on websites, which expresses the mean or the distribution of review valence. In addition, Pavlou & Dimoka (2006) argue that textual comments can provide readers with more helpful information than simple numerical ratings in online marketplaces. Therefore, it would be interesting to assess the two groups of textual features, namely the semantic and the linguistic characteristics of online reviews (Ludwig et al., 2013).
1.3 The semantic and linguistic characteristics of online reviews
10 1.3.1 Semantic characteristic: Objectiveness
One of the two major semantic characteristics of online reviews that determine their helpfulness is objectiveness, which can be defined as a review containing statements using factual information on product characteristics, allowing for a product description that augments and reflects on the product description provided by the company (Ghose & Ipeirotis, 2011). In addition, objectiveness refers to reviews using logical, understandable and reason-based information on the focal products (Park et al., 2007). Thus, objectiveness mainly focuses on what are the extant features of the product, how does it actually perform and what are its capabilities, as opposed to making evaluations based on feelings and highly personal opinions.
1.3.2 Semantic characteristic: Concreteness
The other major semantic characteristic of online reviews is concreteness, which stands at one end of the scale of abstractness, and can be described as the extent to which reviews provide the readers with specific experiences, from which they can analyse information more easily and can determine the actual product performance more accurately (Li et al., 2013). Therefore, concreteness involves the clear depiction of the usage experiences, in contrast to vague and inexpressive statements.
1.3.3 Linguistic characteristic: Linguistic style
The third main antecedent of helpfulness is linguistic style, which can be defined as “the specific wording choices made within each of the review’s statements” (Schindler & Bickart, 2012). Review style can be distinguished from the factual and affective content of the reviews, and mainly conveys the use of function words (Ludwig et al., 2013). Hence, the style of the reviews helps readers interpret and understand the statements that depict the usage situation.
1.4 Elaboration likelihood Model
11
right product, they are motivated to evaluate arguments that are relevant to their search, while in cases of low-involvement, they mainly base their decision on simple cues.
Therefore, argument quality has a larger influence on consumers’ attitudes in high-involvement decision-making situations (Petty et al., 1983). In other words, consumers employ a “systematic information processing strategy” in high involvement situations and a “heuristic processing strategy” in low involvement situations (Chaiken, 1980). Since the process of careful reading and thoughtful analysis of online reviews with the aim of making a well-informed and accurate decision is based on the high argument quality and the high level of diagnosticity of reviews, it could be argued that the assessment of review helpfulness is relevant in decision making situations where systematic processing is applied. Consequently, in this study the different online review characteristics and their influence on review helpfulness are going to be examined with regards to products and decision making situations that require systematic processing from the readers.
1.5 Search goods and experience goods
The numerous review characteristics exert their influence in a different way on the helpfulness of the reviews in cases of two major product types, namely if the particular products are search goods or experience goods. Search goods refer to those products, from which information can be acquired and the quality of which can be determined before purchase, such as electronics, while in the case of experience goods, one has to purchase the product to assess its quality, such as with books or music (Nelson, 1974).
It can be argued that the objectiveness and the concreteness of reviews lead to helpfulness particularly in the case of search goods. Mudambi & Schuff (2010) assert that readers of reviews on search goods prefer more factual, detailed and objective information. The findings of Jiménez & Mendoza (2013) confirm this, showing that more detailed information engenders a high level of diagnosticity only in the case of search goods. Accordingly, the present study is analysing the effects of various textual online review characteristics on their helpfulness only in the category of search goods.
1.6 Problem statement
12
helpfulness of an online consumer review that in itself can significantly facilitate the decision making process of an involved consumer. Therefore, the following research question is formed:
How do the objectiveness, concreteness and linguistic style of online consumer reviews influence the helpfulness of reviews written on search products?
1.7 Research questions
Based on the aforementioned problem statement, these additional research questions are developed:
How does the increase in perceived content diagnosticity influence the helpfulness of
online reviews?
How does the change in perceived argument quality effects review helpfulness? How does the objectiveness of online reviews influence their helpfulness? How does concreteness of online reviews affect their helpfulness?
In what ways does linguistic style influence the helpfulness of online reviews?
1.8 Academic and managerial relevance
This research contributes to existing literature by inspecting the individual and shared effect of the semantic and linguistic features of online consumer reviews on their helpfulness. Although a large pool of studies exists which investigates the helpfulness of online reviews, most of these focus on the valence or length of the reviews, and only a small number of articles explore the semantic or linguistic characteristics of online reviews. Even though the latter examine the effects of these characteristics, they do not study their combined effect in a particular online consumer review. Therefore, the present research adds to the existing literature by showing that all objectiveness, concreteness and linguistic style have a positive effect on online review helpfulness. However, only the joint effect of objectiveness and concreteness is stronger than the sum of their individual influence.
13
other readers would be able to filter the reviews based on which aspects they find the most helpful and contribute to their decision making the most.
1.9 The structure of the thesis
14
2. Theoretical Framework
The core theories, which have been introduced in the previous chapter, and which comprise the conceptual model of this research, are presented in greater detail in the following theoretical framework. Firstly, the two main determinants of helpfulness, that is, the dependent variable diagnosticity and the mediator argument quality are described more thoroughly. Subsequently, the three independent variables, the antecedents of helpfulness are going to be explained, namely objectiveness, concreteness and linguistic style. Finally, based on the relationship between the independent variables, the mediating and the dependent variable, and also the final outcome variable, the hypotheses of this framework are formed that lead to the conceptual model, and will also allow for the formulation of a particular research method that is meant to address the questions that are developed here.
2.1 Diagnosticity
How does the increase in perceived content diagnosticity influence the helpfulness of online reviews?
One of the two elements that express review helpfulness is diagnosticity. It can be defined as how the information conveyed in the reviews can aid readers in their decision making process to come to a well-established conclusion and choose the right alternative (Lynch et al., 1988). According to Weathers et al. (2015), the diagnosticity of reviews greatly contribute to their helpfulness, through reducing the readers’ uncertainty by providing them with sufficient information about the product and also through reducing equivocality by giving clear explanations about the information and claims employed. In other words, diagnosticity concerns the degree to which information can “discriminate between alternative hypotheses, interpretations, or categorizations” (Herr et al., 1991). Moreover, Weathers et al. (2015) find that in addition to the provision of simple product evaluations, reviews can present information pertaining to the actual and concrete usage experiences of the reviewers, therefore adding to the helpfulness of the reviews. Furthermore, Mudambi & Schuff (2010) posit that review depth (measured in the number of words used) has a significant effect on review helpfulness due to the fact that it evokes an increased level of information diagnosticity, which in turn leads to more helpful reviews, especially for search goods.
15
quality, diagnosticity and helpfulness, the latter being the outcome of the former two variables. Consequently, it is hypothesised that the influence of argument quality on helpfulness is mediated through diagnosticity to a certain extent.
H1: The effect of diagnosticity on helpfulness is mediating the effect of argument quality on helpfulness.
2.2 Argument quality
How does the change in perceived argument quality effects review helpfulness?
The other determinant of helpfulness is argument quality, which also has an effect on diagnosticity. According to Cheung et al. (2008), argument quality consists of four main elements: relevance, timeliness, accuracy and comprehensiveness, out of which relevance and comprehensiveness are the strongest determinants of the helpfulness of online reviews. Cheung & Thadani (2012) define relevance as the degree to which information in the reviews are “applicable and useful for decision making” and comprehensiveness as the completeness of the presented messages.
The study of Racherla (2012) argues that reviews, which contain strong claims that are objective, relevant and well-justified, tend to be perceived as more credible and are more convincing, which can become highly important in a decision making situation, where the readers are searching with the aim of gathering detailed information on a particular product.
Park et al. (2007) posit that the quality of online reviews has a positive effect on the purchase intention of consumers due to their persuasiveness, which in a high-involvement situation can relate back to their helpfulness. They refer to review quality as content and information characteristics, such as how relevant, understandable, sufficient and objective these reviews are. Therefore, high-quality reviews are ones, which aid the evaluation of consumers with arguments based on the factual characteristics of products, thus being more logical and persuasive in their content.
16
It can be asserted that reviews the opinions and claims of which are well-grounded in arguments, are higher in their level of helpfulness. Willemsen et al. (2011), who examined the effects of content characteristics on helpfulness, find that argument diversity, which can be considered as a part of argument quality, is a significant determinant of the helpfulness of online reviews. Building on the work of Eisend (2007), they find that reviews that feature both sides of an argument, thus are higher in argument diversity, are perceived more helpful, since the provision of both the advantages and the downsides of a specific product indicates a concrete and real experience, which can make reviews more authentic and convincing. Weathers et al. (2015) show that reviews which accommodate both positive and negative arguments are perceived as more helpful.
Argument quality has a significant role in how much consumers rely on online product reviews in cases of high-involvement due to their reduced level of uncertainty, which in turn can make the reviews highly diagnostic for the readers (Racherla, 2012).
H2: Argument quality is mediating the effects of objectiveness, concreteness and linguistic style on diagnosticity.
H3: An increase in the level of argument quality results in a higher level of helpfulness.
2.3 Objectiveness
How does the objectiveness of online reviews influence their helpfulness?
17
information, readers can make their own assessments, free from the subjective opinion of the reviewer.
Park et al. (2007) argue that objective reviews are of higher quality, since these contain logical, understandable and reason-based pieces of information on the products compared to the more subjective and emotional reviews that lack factual information and are only recommending the particular product. The research of Salehan & Kim (2016) supports this notion by stating that sentiment in reviews decrease their helpfulness, since expressed emotions can be considered less rational, and consequently less helpful. Lee & Koo (2012) further corroborate these findings by showing that consumers assessing online information sources believe objective information, such as facts referring to product characteristics more.
Liang et al. (2014) support the aforementioned findings, since in their research they have shown that reviews which are more descriptive and objective, including words referring to measurements such as quantifiers and space words (referring to location and size) are perceived more helpful in contrast to reviews that convey more evaluative and subjective expressions and often refer to the self of the reviewer. Similar to the reasoning of Schindler & Bickart (2012), they assert that these differences in helpfulness originate from the different levels of information processing and that objective information, which allows the readers to form their own evaluations, facilitates elaborated processing, while subjective claims inhibit this process by driving readers in the direction of a specific point of view.
Online reviews that provide descriptions on the focal products, including different attributes, can use objective information as the foundation of claims and arguments employed in the review, therefore adding to its argument quality (Park & Kim, 2008).
H4: Higher online consumer review objectiveness leads to higher argument quality.
Objective information in online reviews that refers to easily measured characteristics can help readers reduce uncertainty and ambiguity, therefore can facilitate their decision making through greater diagnosticity (Lee & Koo, 2012).
18 2.4 Concreteness
How does concreteness of online reviews affect their helpfulness?
Concreteness is the other semantic characteristic that strongly influences the elements of helpfulness of online reviews. We can distinguish between the level of abstractness of the reviews, that is, whether they are describing a product in a more concrete or in a more abstract way. Concreteness can be described as the extent to which reviews provide the readers with specific experiences that help them analyse product-related information in their decision-making process (Li et al., 2013). The reason behind these differences in the reviews is the level of congruence between the actual product experience of the reviewers, and their attitudes and expectations towards the product. Particularly, when a product experience is congruous with the brand attitudes of the reviewers, they would write a more abstract review, while when the experience is inconsistent with their expectations, they tend to write a more concrete product review. This level of abstraction can influence the assumptions readers make on the reviewers’ attitudes towards the product. More specifically, positive product experiences described in an abstract way suggest a more positive initial attitude of the reviewer, while negative experiences described in a more abstract way suggest less positive attitudes for the product (Schellekens et al., 2010). Consequently, based on these findings, it might be argued that readers can comprehend the circumstances of the reviews and make a more precise evaluation of the actual performance of the product.
Moreover, Li et al. (2013) have found that concrete product reviews are more helpful for readers than abstract ones. They describe concrete reviews as ones that provide the reader with specific characteristics, from which they can extract information more easily and can judge the actual product performance more accurately. On the other hand, more abstract reviews include expressions such as “I am really satisfied with this product, it works well for me”, which can lead to more ambiguous evaluations, therefore these would be less helpful for the readers. Park et al. (2007) confirm this notion by positing that high-quality reviews provide consumers with specific, fact-based product information.
Concrete reviews provide the readers with highly detailed information about the product experiences of the reviewer, which can effectively support the claims the reviewer makes, which contributes to the decision making process of the reader (Li et al., 2013).
19
Concrete reviews show specific characteristics and usage experiences, out of which readers can elicit information more easily and can assess the actual product performance more accurately, which leads to uncertainty reduction via a high level of diagnosticity (Li et al., 2013).
H7: An increase in online consumer review concreteness leads to an increased perceived diagnosticity.
2.5 Linguistic style
In what ways does linguistic style influence the helpfulness of online reviews?
The third independent variable, linguistic style refers to the “the specific wording choices made within each of the review’s statements” according to Schindler & Bickart (2012). It is important to note that review style can be distinguished from the factual and affective content of the reviews, and mainly conveys the use of function words (Ludwig et al., 2013). These function words among others include pronouns, conjunctions, auxiliary verbs, articles and prepositions in contrast to the words mainly referring to the content of the reviews such as regular verbs, nouns, adjectives and adverbs (Tausczik & Pennebaker, 2010).
The significance of stylistic characteristics comes from the fact that using a particular linguistic style that is similar to and relevant for the target group of a specific product category can have a greater impact on the decision making of the readers according to Ludwig et al. (2013). They assert that linguistic style (in addition to and in relation to content) has diagnostic effects, thus it can help readers understand why the reviewer used certain claims and put these into the context of the product usage. This is highly related to the term of vicarious expression which also indicates that a specific linguistic style, which also conveys a rich and detailed depiction of the usage experiences, can aid the readers to more easily interpret the reviewer’s point of view (Li et al., 2013).
20
use of expressive slang words can be helpful, which in some cases can lead to more accurate descriptions compared to common words.
On the other hand, Schindler & Bickart (2012) have shown that some types of style characteristics can weaken the helpfulness of online reviews such as spelling and grammar mistakes, repetition or inexpressive, generalizing slang words.
These findings are corroborated by other studies, which show that the quality of linguistic style can influence the helpfulness of reviews. Korfiatis et al. (2012) capture this quality (using the term readability) as how much cognitive effort does it take to read and comprehend a review, and they also find that it has a significant positive effect on perceived review helpfulness. The study of Ghose & Ipeirotis (2011) confirm these findings, asserting that easy to read reviews are considered more helpful and are more influential. They further show that the number of spelling errors are inversely proportional with review helpfulness. Therefore, it can be hypothesised that reviews possessing a high quality linguistic style which supports the claims put forward in the review by conveying sound grammar and spelling can lead to higher argument quality.
H8: Reviews possessing a high quality linguistic style tend to have higher argument quality.
A high quality linguistic style conveyed in an online review that helps readers understand the underlying reasons for particular claims conveyed in the review, can reduce equivocality and therefore make the review more diagnostic and more helpful in terms of facilitating the decision making process of the readers (Schindler & Bickart, 2012). Thus, it can be hypothesised that reviews possessing a high quality linguistic style, which indicate the reason behind why certain claims are made in the review, and as a result, are expressive of the context and also convey sound grammar and spelling, can reduce ambiguity and lead to higher diagnosticity.
H9: Reviews possessing a high quality linguistic style lead to higher diagnosticity.
2.6 Interaction effects of the independent variables
21
effects, all three independent variables have an interaction effect on the other two independent variables.
H10: There is a positive interaction effect between the independent variables in the effect of objectiveness on argument quality.
H11: There is a positive interaction effect between the independent variables in the effect of
objectiveness on diagnosticity.
H12: There is a positive interaction effect between the independent variables in the effect of concreteness on argument quality.
H13: There is a positive interaction effect between the independent variables in the effect of concreteness on diagnosticity.
H14: There is a positive interaction effect between the independent variables in the influence of linguistic style on argument quality.
H15: There is a positive interaction effect between the independent variables in the influence of linguistic style on diagnosticity.
2.7 Frequency of online shopping and frequency of reading online reviews
22
H16: Consumers who shop online more often, will find online consumer reviews to have higher argument quality than those who shop online less frequently.
H17: Consumers who read online reviews more often, will find online reviews to have higher argument quality than those who read these review less frequently.
H18: Consumers who shop online more often, will find online consumer reviews more diagnostic than those who shop online less frequently.
H19: Consumers who read online reviews more often, will find online reviews more diagnostic than those who read these review less frequently.
H20: Consumers who shop online more often, will find online consumer reviews more helpful than those who shop online less frequently.
H21: Consumers who read online reviews more often, will find online reviews more helpful than those who read these review less frequently.
2.8 Conceptual model
Based on the hypotheses that describe the effects between the independent variables (objectiveness, concreteness and linguistic style), the mediator (argument quality) and the dependent variable (diagnosticity), and also the final outcome variable (helpfulness), a conceptual model of this theoretical framework is created.
In this model, which can be seen in Figure 1, one of the two semantic characteristics that positively influence review helpfulness is objectiveness, which refers to the factual and descriptive information (Ghose & Ipeirotis, 2011) and the reason-based logical claims conveyed in online reviews (Park et al., 2007). The other semantic characteristic that bolsters online review helpfulness is concreteness, which shows how much a review equips readers with detailed and specific experiences about the focal products (Li et al., 2013). The third review characteristic that leads to higher levels of review helpfulness is linguistic style, which involves the particular set of words chosen to be included in the review (Schindler & Bickart, 2012).
23
uncertainty of the readers about a product (Weathers et al., 2015). Diagnosticity is influenced by all three textual characteristics (objectiveness, concreteness and linguistic style) and also by the mediator argument quality. In addition, helpfulness, which is both influenced by argument quality and diagnosticity, is the final outcome variable that encapsulates to what extent the characteristics of online reviews help consumers in their decision making.
Moreover, two control variables are introduced into the model to account for the variation caused by the individual differences of the respondents in this study. These variables are frequency of online shopping and frequency of reading online reviews.
24
3. Research methodology
In this chapter the research method is going to be presented, which builds on the theoretical framework in the previous section. First, the research design is introduced, followed by the specific population that has been used for the research. After that, the operationalization including the measurement of concepts applied in the survey are described and finally the planned process of data analysis is presented.
3.1 Research design
In the following study, the two levels of each of the three independent variables or attributes add up to a 2x2x2 experimental design with a total of eight profiles, as can be seen in Table 1. The online review attribute of objectiveness has two levels, that is, reviews can be either objective or subjective. The attribute of concreteness has also two levels, therefore reviews can be concrete or abstract. In addition, all of these aforementioned semantic review attributes can be combined with a high or low quality linguistic style.
Linguistic style
High quality Low quality
Objectiveness
Objective Subjective Objective Subjective
Concreteness
Concrete Profile 1 Profile 3 Profile 5 Profile 7
Abstract Profile 2 Profile 4 Profile 6 Profile 8
Table 1 – Stimulus profiles
In this research, examining the hypotheses introduced in the theoretical framework, a mixed experimental design is conducted, which combines a between-participants and a within-participants design. In a between-within-participants design, the conditions are randomly assigned between respondents so that each of them receive only one condition, while in a within-participants design, all respondents receive stimuli from all the different conditions (Aronson et al., 1998). In this particular mixed design, respondents are randomly assigned three of a total of eight conditions.
25
still might be considered too long for being rated in numerous evaluation tasks, and would require too much effort from the respondents. In this study, a full profile design is applied, which means that each stimuli are presented by using all three attributes determined by the design (Malhotra & Birks, 2007).
3.2 Population
The population of this study, from which the sample of participants is selected, consists of consumers between the age of 18 and 65, who are active users of online opinion platforms featuring large numbers of online consumer reviews. This range was chosen because a large proportion of companies, whose products are present on these opinion platforms, define their target groups between these two ages, therefore the respondents of this study would be in the age category that is also relevant for the operators of these platforms. These respondents are reached through a snowball sampling method, therefore they are initially approached via social media and personal messages, and additional respondents are collected by previous participants (Malhotra & Birks, 2007). In order to be able to make a reliable examination of the effects of the independent variables on the dependent variable, a number of 30 responses are needed for each condition, which would add up to a total of 240 participants. Although a minimum sample size of 30 subjects in each of the different conditions is suggested as a general rule, this might not accurately explain the statistical power of the experiment, which also depends on many other aspects (Sawyer & Ball, 1981). In this particular study, which is based on a mixed participant design, each participant receives 3 of the total of 8 stimulus profiles in a randomized design. This means that each of the 8 stimulus profiles appears in an equal distribution for a large number of respondents, even with only around 200 respondents, allowing for a sufficient amount of data to be collected and later analysed.
26
to the frequency of online shopping, the majority, two-thirds of those taking part in the survey, have ordered a product online 1-4 times in the past month. In addition, two-thirds of all respondents stated that they tend to read online consumer reviews “most of the time” or “always” during online shopping. Furthermore, there are no outstanding differences among these demographic groups in terms of the mean differences in general or by each condition.
Age Frequency Percentage Gender Frequency Percentage
18-24 60 26.5 Male 85 37.6 25-34 55 24.3 Female 141 62.4 35-44 54 23.9 Total 226 100.0 45-54 42 18.6 55-64 15 6.6 Total 226 100.0
Education Frequency Percentage Orders in past month
Frequency Percentage Less than high school 2 .9 Not even once 57 25.2
High school 40 17.7 1-4 times 145 64.2
Vocational school 25 11.1 5-10 times 20 8.8 Bachelor’s degree 76 33.6 More than 10
times
4 1.8
Master’s degree 55 24.3 Total 226 100.0 Professional degree 13 5.8
Doctoral degree 9 4.0
Other 6 2.7
Total 226 100.0
Reads reviews Frequency Percentage
Never 6 2.7
Sometimes 48 21.2
About half the time 24 10.6 Most of the time 87 38.5
Always 61 27.0
Total 226 100.0
Table 2 – Descriptive statistics
3.3 Design of the survey
27
before purchase. The respondents of this survey are asked to read the reviews carefully, as the presented questions are referring to the various characteristics of these reviews. At this stage, 3 randomly chosen stimuli of the stimulus material (see Appendix A) with a fixed set of questions on the same page with each of these reviews are shown that examine the perceived argument quality and diagnosticity of the presented online consumer reviews (see Appendix B). The review stimuli are written on the same product and are referring to the same product aspects, but are different on the three attributes employed in the research design, such as objectiveness, concreteness or linguistic quality (see Figure 2).
Figure 2 – Stimuli examples
In the second part of the survey, a manipulation check is conducted in order to examine that the respondents perceived the differences in the experimental variables in the same way as these were intended. More specifically, they were presented randomly with 3 of the 8 textual reviews, each review with three questions asking about all three independent variables (see Appendix B). Unfortunately, due to the limitations of the Qualtrics program, it was not feasible to show those exact same three stimuli to a particular respondent that he or she has seen in the first part, since randomization can only be applied to question blocks and when this line of randomization is interrupted (for instance by the second and the third randomly chosen stimuli in the first part), it cannot be continued in the second part of the survey.
Finally, participants were asked questions related to the background information about them, such as demographic data, information on internet usage and the number of online purchases made in a certain time period.
High quality linguistic style, objective and abstract review:
“The price of this camera is quite modest, and includes the full set. Its battery is enough for more than a day and takes really high quality pictures. It is also suitable for low-light settings.”
High quality linguistic style, subjective and concrete review:
28 3.4. Operationalization of concepts
In the survey, after the stimuli are presented, the concepts of the mediator, argument quality and the dependent variable, diagnosticity, both of which can express the helpfulness of online reviews, are measured, as is shown in Table 3. All the questions are assessed on a 7-point Likert scale, through which participants need to evaluate to what extent do they agree with the listed statements, where 1 means “Strongly disagree” and 7 means “Strongly agree”. The use of 7-point Likert scales gives enough room for respondents to express their actual perceptions and attitudes, but at the same time it does not require great effort from them to make an accurate choice. In addition, the consistent use of one type of scale prevents them from becoming confused about various forms of measurement and getting distracted from the content of the research.
There are several studies that define the constructs of, and use scales for the mediator, argument quality and the dependent variable, diagnosticity. However, many of these are irrelevant or inapplicable for this particular study, mainly because these are expressing qualities that are similar to the independent variables or the other measured variable, thus their use would lead to confounding results. For instance, as can be seen from Table 3, the research of Cheung & Thadani (2012) uses relevance as a construct of argument quality, and which expresses that the messages employed in the reviews are useful for decision making, which in fact is closer to the definition used in this study for diagnosticity. In addition, the constructs applied by Racherla (2012) follow a structure (that is, the claims applied in a review are supported by sufficient evidence and it is explained why these claims should be accepted) that cannot be applied to the stimuli in the present study due to the shortness of these texts. Therefore, argument quality is measured on 7-point Likert scales using the definition and scales of Zhang et al. (2014), and based on the characterization of high-quality arguments by Park et al. (2007).
29
Constructs and source Definition Item Cronbach’s Alpha Argument quality (Zhang et al., 2014) Perceived informativeness and persuasiveness
1. The online review
above is highly
informative.
2. I find this review persuasive. .927 .929 (if Q2 deleted) • Logical • Reasons based on facts (Park et al., 2007)
3. I think this review is logical.
4. This review supports its evaluations with reasons based on the facts about the product.
• Relevance • Timeliness • Accuracy • Comprehensiveness (Cheung & Thadani, 2012) -Messages are applicable and useful for decision making
-Messages are current, timely and up-to-date -Reliability of the messages/arguments -Completeness of messages 5. The messages
employed in this review are applicable and useful for my decision making 6. The content of this review is current, timely and up-to-date
7. The arguments used in this review are reliable 8. The messages used in this review are complete
• Claim - assertion • Data - evidence • Backing
(Toulmin, 1958 in Racherla, 2012)
-Put forward for acceptance -Supporting the claim
-Explaining why the claim and data should be accepted
30 Diagnosticity
(Herr et al., 1991)
The degree to which information can “discriminate between alternative hypotheses, interpretations, or categorizations”
0. This review helps me in deciding between alternative options .900 .905 (if Q5 deleted) • Facilitating decision making (Li et al., 2013)
5. The review above helps me in
distinguishing between the advantages and disadvantages of this product.
• Evaluation • Familiarization • Understanding
(Jiang & Benbasat, 2007)
6. This review doesn’t help me to evaluate the product. (R)
7. This review helps in familiarizing me with this product.
8. This review allows me to understand the performance of the product. • Uncertainty reduction • Equivocality reduction • (Weathers et al., 2015)
5. This review presents many product features without subjectively evaluating them
6. This review provides information about how the review writer has used the product
31 3.5. Factor Analysis and Reliability Analysis
For the four items measuring argument quality, a factor analysis has been conducted (see Appendix C). The Kaiser-Meyer-Olkin (KMO) measure resulting in a value of .849 that is higher than the suggested .5 level shows that the analysis will provide valid outcomes due to its high sampling adequacy (Malhotra & Birks, 2007). In addition, all of the items are correlated with each other at a level lower than .9, thus there is no risk of multicollinearity in this case. The communalities are higher than .7 on all of the items, which is higher than the necessary level of .4, therefore they have sufficient explanatory power after extraction. The factor analysis yielded one component for these four items explaining 82% of the variance. The subsequent reliability analysis has a Cronbach's Alpha of .927, which is far higher than the accepted level of .6 (see Table 3 and Appendix C), therefore these items have a high internal consistency (Malhotra & Birks, 2007). Even with the deletion of question 2, the Cronbach's Alpha would increase by a negligible rate to .929.
In the case of diagnosticity, the KMO measure provides a value of .825, showing a high sampling adequacy. The correlation between any two of these four items used for this scale remain below .9, therefore no multicollinearity is present here. The communalities after extraction are all above .6, thus providing ample explanatory power. The factor analysis extracted one factor with 77% of the variance explained. The conducted reliability analysis yielded a Cronbach’s Alpha of .900 that indicates high internal consistency among these items. Only with the deletion of question 5 (item 1), would the value slightly increase to .905 (see Table 3).
In addition, a third variable, helpfulness was measured in the survey with question 9 (“I find this review helpful in my decision making.”) that is strongly correlated to argument quality and diagnosticity, as is shown in Table 4. This can be considered as the final outcome variable in the model, which is influenced by both of the former two concepts, argument quality and diagnosticity. The following analyses will also examine whether there are substantial differences among all these measured variables in terms of the explanatory power of the experimental variables and also the interaction effects of the latter.
32
deletion of which would slightly increase the Cronbach’s Alpha for diagnosticity. This is likely due to the fact that this question is referring to how the review helps in the distinction between the advantages and disadvantages of the product, and a number of respondents assigned very low values even for the high quality reviews on this question, since the stimuli is only focusing on the advantages of the focal product.
Q1 Q2 Q3 Q4 Q5 Q6R Q7 Q8 Q9
Q9 .809 .731 .780 .747 .642 .760 .802 .842 1.000
Table 4 – Correlation matrix
3.6. Plan of data analysis
This online survey has been completed by 246 respondents, which provide sufficient data and makes it possible to receive applicable results in the consecutive analyses to be conducted. However, because several participants missed the validation in the form of the reversely worded question 6 (“This review doesn’t help me to evaluate the product.”), which presumably arises from the fact that they did not read the item questions carefully, and which makes their answers inadequate, therefore these responses have been deleted from the data analysis. In addition, a small number of participants did not fit the filtering condition between the ages 18 and 65. Accordingly, a number of 226 responses were used for the analyses. Moreover, some participants missed the validation only in one of the three sets of questions shown to them, and in the rest of the cases, their responses showed that they filled in the survey paying enough attention. Thus, in their situation, only those sets of questions (on one particular stimuli) were deleted that they missed.
In terms of coding, the three independent variables have been coded into dummy variables so that the subsequent analyses can be carried out. Regarding objectiveness, objective profiles received a value of ‘1’ and subjective profiles received a value of ‘0’. In respect of concreteness, concrete profiles were assigned a value of ‘1’ and the abstract profiles a value of ‘0’. As far as linguistic style is concerned, high quality profiles were given a value of ‘1’ and low quality profiles were given a value of ’0’.
33
independent variables, namely objectiveness (OB), concreteness (CO) and linguistic style (LI), then in the following models the interaction effects of objectiveness and concreteness, objectiveness and linguistic style, and after that the interaction between concreteness and linguistic style are added to see how much each of these variables contribute to the whole model and add to the explained variance. Finally, the two control variables are added to measure how often respondents shop online (BUYING) and how often they read online consumer reviews (READING). The full models are constructed as follows:
Argument quality =β0 + β1OB + β2CO + β3LI + β4OB*CO + β5OB*LI + β6CO*LI +
BUYING + READING + ε
Diagnosticity = β0 + β1OB + β2CO + β3LI + β4OB*CO + β5OB*LI + β6CO*LI +
BUYING + READING + ε
Helpfulness =β0 + β1OB + β2CO + β3LI + β4OB*CO + β5OB*LI + β6CO*LI + BUYING
+ READING + ε Where OB = Objectiveness CO = Concreteness LI = Linguistic style OB*CO = Objectiveness*Concreteness OB*LI = Objectiveness*Linguistic style CO*LI = Concreteness*Linguistic style
BUYING (control) = Frequency of shopping online
READING (control) = Frequency of reading online consumer reviews
34
Figure 3 – PROCESS Model 10 – Conceptual and statistical diagram Source: own edit, based on Hayes (2013)
3.7. Manipulation check
Before the analyses could be conducted, a manipulation check was executed in order to see whether the independent variables were perceived by the respondents as initially intended, and whether these variables can be further used separately or need to be merged. The results of the independent samples t-tests show (see Table 5 and Appendix D) that in the cases of all three independent variables, the profiles with the intended higher quality (objective, concrete and high quality linguistic style) were indeed rated significantly higher than the lower quality counterparts (subjective, abstract and low quality linguistic style), therefore all three independent variables can be applied for the subsequent analyses. These differences in means were examined on argument quality, diagnosticity and helpfulness combined, since these tests asked respondents about their perception of the reviews only on the basis of the independent variables.
Objectiveness N Mean St. Deviation Difference Significance
Objective 333 4.7357 2.01107
1.72704 .000
Subjective 345 3.0087 2.03385
Concreteness N Mean St. Deviation Difference Significance
Concrete 341 5.2170 1.75896
2.95588 .000
35
Linguistic style N Mean St. Deviation Difference Significance
High quality 340 4.6265 1.95794
2.21227 .000
Low quality 338 2.4142 1.92594
Table 5 – Manipulation check on all measured variables combined
36
4. Results
In the following chapter, the results of the analyses which are based on the data from the surveys are going to be presented. First, the differences in the means by the different conditions are presented, after that the results of the analysis of variance are shown, then the outcomes of the regression analysis are provided. In addition, the results of the mediation analysis are given. Finally, these findings of these analyses are compared with the hypotheses in the hypothesis testing.
4.1 Condition means
Table 6 shows the means of all 9 questions in the different conditions on all three measured variables, that is, argument quality, diagnosticity and helpfulness separately. In most of the cases, a substantial difference can be seen between two conditions, which only differ on one of the independent variables. However, when comparing objective and abstract reviews with subjective and concrete reviews at the same quality level of linguistic style (namely Profile 2 vs. 3 and Profile 6 vs. 7) in order to see if objectiveness or concreteness has a greater importance for consumers, there is no significant difference between each pair of conditions.
Linguistic style
High quality Low quality
Objectiveness
Objective Subjective Objective Subjective
Concr
et
en
ess
Concrete Profile 1 Profile 3 Profile 5 Profile 7
Mean SD Mean SD Mean SD Mean SD
Argument Q. 5.7107 1.09161 4.3544 1.51044 4.7333 1.43103 2.9345 1.25937
Diagnosticity 5.2809 1.14056 3.8671 1.37796 4.5833 1.32245 2.9613 1.30655
Helpfulness 5.5506 1.31435 3.9620 1.86357 4.7733 1.74428 2.7024 1.61901
N 89 79 75 84
Abstract Profile 2 Profile 4 Profile 6 Profile 8
Mean SD Mean SD Mean SD Mean SD
Argument Q. 4.1860 1.51271 2.5546 1.33677 2.8293 1.47922 1.7685 1.11928
Diagnosticity 3.9848 1.53626 2.4483 1.29351 2.8780 1.52916 1.7870 1.10899
Helpfulness 4.0854 1.93219 2.8046 1.68329 2.6829 1.70604 1.9630 1.47855
N 82 87 82 81
37 4.2 ANOVA
The results of the ANOVA conducted on all three measured variables, namely argument quality, diagnosticity and helpfulness, show that the direct effect of all the three independent variables, that is, objectiveness, concreteness and linguistic style are significant with objectiveness and concreteness having a similarly strong effect and linguistic style a somewhat weaker but still considerable effect (see Appendix E). However, regarding the interaction effects between the independent variables, there is a small but significant interaction effect between objectiveness and concreteness, but only on helpfulness. This implies that with a high level of objectiveness, the incremental effect between a low and high level of concreteness on the perceived helpfulness of the review is greater than when there is a low level of objectiveness (see Figure 4).
Figure 4 – Interaction effect between Objectiveness and Concreteness
4.3 Regression analysis
38
have a moderately high R Square value, adding the interaction effect variables to the models which include only the direct effects, actually decreases the R Square value for argument quality and diagnosticity and it only slightly increases R Square in the case of helpfulness (see Appendix F).
Argument Quality Model 1 Model 2 Model 3 Model 4 Model 5
Constant 1.532 1.589 1.607 1.640 1.650 Objectiveness .400*** .370*** .360*** .360*** .360*** Concreteness .437*** .406*** .406*** .388*** .387*** Linguistic style .311*** .310*** .300*** .283*** .279*** Objectiveness*Concreteness .053 .053 .052 .052 Objectiveness*Linguistic style .017 .016 .016 Concreteness*Linguistic style .031 .031
Frequency of online purchases .048*
Frequency of reading online reviews -.023
R2
.456 .457 .457 .458 .460
R2 Adjusted .454 .454 .453 .453 .454
F 183.363 137.828 110.134 91.754 69.298
Significance .000 .000 .000 .000 .000
Table 7 – Regression analysis: Argument Quality
39
Diagnosticity Model 1 Model 2 Model 3 Model 4 Model 5
Constant 1.644 1.693 1.725 1.704 1.716 Objectiveness .411*** .383*** .365*** .365*** .365*** Concreteness .406*** .378*** .377*** .389*** .388*** Linguistic style .246*** .245*** .227*** .239*** .233*** Objectiveness*Concreteness .049 .049 .050 .049 Objectiveness*Linguistic style .031 .031 .031 Concreteness*Linguistic style -.020 -.020
Frequency of online purchases .062**
Frequency of reading online reviews -.033
R2 .402 .403 .403 .404 .408
R2 Adjusted .400 .399 .399 .398 .400
F 146.926 110.397 88.296 73.509 55.935
Significance .000 .000 .000 .000 .000
Table 8 – Regression analysis: Diagnosticity
The results show that the direct effects of the independent variables are all significant in all three cases at a significance level of .000. Similar to the results of the ANOVA, it can be seen that objectiveness and concreteness have a comparable positive effect and linguistic style has a slightly weaker but considerable positive effect in the model. However, there is a moderate and highly significant positive interaction effect between objectiveness and concreteness, but only in the case of helpfulness. In respect of the two control variables, only the one regarding the frequency of online purchases shows a slight but significant positive effect in the cases of argument quality and diagnosticity. This means that the respondents who order products online more often, might perceive online ratings more helpful in their decision making than the ones who only rarely shop online.
Helpfulness Model 1 Model 2 Model 3 Model 4 Model 5
Constant 1.637 1.845 1.856 1.831 1.842 Objectiveness .348*** .248*** .243*** .243*** .243*** Concreteness .335*** .234*** .234*** .247*** .246*** Linguistic style .268*** .264*** .259*** .271*** .266*** Objectiveness*Concreteness .174*** .174*** .175*** .174*** Objectiveness*Linguistic style .009 .010 .009 Concreteness*Linguistic style -.022 -.021
40
Frequency of reading online reviews -.029
R2 .311 .321 .321 .321 .324
R2 Adjusted .308 .317 .316 .315 .316
F 98.588 77.369 61.809 51.465 38.933
Significance .000 .000 .000 .000 .000
Table 9 – Regression analysis: Helpfulness
4.4 Mediation analysis
In the following, the results of the mediation analyses will show how the independent variables influence the dependent variable, diagnosticity through the mediator, argument quality. Thus, now the relationship between argument quality and diagnosticity is also discussed. In the mediation analyses, which were conducted in SPSS with the PROCESS macro by Hayes (2013), the effects of the three independent variables were measured separately with the other two independent variables treated as moderators (see Table 10 and Appendix G).
Corresponding to the previous findings, these analyses show that all independent variables have a significant positive effect on the mediator (a path) and also the mediator has a significant positive effect on the dependent variable (b path) in all three cases. Therefore, the indirect effects (ab path) are also significant and show a positive direction in all three analyses.
In this specific mediation analysis (Model 10), the direct effect of the independent variable on the dependent variable (c’ path) is conditional on the levels of the two other included independent variables, considered as moderators. In the case of objectiveness, the direct effect is positive and significant at all levels, with a significance level of at least 90% (see Figure 5). It can be noticed that the direct effect of objectiveness is slightly higher at high levels of linguistic style than at low levels of linguistic style, while this stronger direct effect of objectiveness with concrete reviews compared to abstract reviews is only marginal.
41
Regarding concreteness, its positive direct effect on diagnosticity is only significant at 90% with low levels of linguistic style, and it is slightly higher with objective reviews than with subjective ones. In respect of linguistic style, it has a small significantly (90%) negative effect on diagnosticity only in the case of subjective and concrete reviews. Consequently, it can be claimed that there is a strong partial mediation of argument quality on the effect of objectiveness on diagnosticity, and the effect of concreteness and linguistic style on diagnosticity are partially or fully mediated through argument quality, being contingent on the different conditions of the moderator variables (see Figure 6 and 7).
Figure 6 – Mediation analysis on Concreteness
OB CO LI a b ab c (calculated) c’ Objectiveness - 0 0 1.3169*** .8092*** 1.0657 1.2553 .1896* - 0 1 1.1219 1.3765 .2546** - 1 0 1.2459 1.4505 .2046* - 1 1 1.3021 1.5717 .2696** Concreteness 0 - 0 1.4193*** .8102*** 1.1499 1.3387 .1888* 0 - 1 1.2557 1.2590 .0033 1 - 0 1.3285 1.5361 .2076* 1 - 1 1.4342 1.4564 .0222 Linguistic style 0 0 - 1.0356*** .8102*** .8390 .8225 -.0165 0 1 - .9481 .7461 -.2020* 1 0 - .8956 .9452 .0496 1 1 - 1.0047 .8688 -.1359
42
Figure 7 – Mediation analysis on Linguistic style
Moreover, when a mediation analysis is conducted on all three measured variables, where argument quality is considered the independent variable, diagnosticity the mediator and helpfulness the dependent variable, it can be seen from Table 11 and Figure 8 (and also from Appendix G) that both the direct and indirect effects are highly significant. Therefore, the effect of argument quality on helpfulness is partially mediated through diagnosticity with a substantial direct effect. a b ab c (calculated) c’ X = Argument quality M = Diagnosticity Y = Helpfulness .8413*** .6658*** .5602 .9384 .3782***
Table 11 – Mediation analysis: Measured variables
Figure 8 – Mediation analysis on Argument Quality
4.5 Hypothesis testing
As can be seen from Table 12, hypothesis H1 refers to the effect of diagnosticity on helpfulness
and consequently also its mediating role on the effect of argument quality on helpfulness. Since the mediation analysis shows a reasonably strong, positive and highly significant effect of diagnosticity on helpfulness, Hypothesis H1 is confirmed.
Hypothesis H2 states that argument quality mediates the effects of objectiveness, concreteness
43
that the effects of all three of these independent variables on diagnosticity are strongly mediated through argument quality. More specifically, the effect of objectiveness is partially mediated at every level of the moderator variables, and both the effects of concreteness and linguistic style are partially or fully mediated through argument quality depending on the different levels of the moderator variables. Since all these mediation effects by argument quality are significant at a level of .000, hypothesis H2 is confirmed.
Hypothesis H3 concerns the effect of argument quality on helpfulness. Based on the result of
mediation analysis, argument quality has a moderate and positive, but highly significant direct effect on helpfulness, therefore hypothesis H3 is accepted.
Regarding one of the main effects of the independent variables on the measured variables, hypothesis H4 claims that a higher level of online consumer review objectiveness leads to higher
argument quality. The significant result of the regression analysis shows that objectiveness has a positive effect on argument quality with a standardized Beta of .400 at a significance level of .000, therefore hypothesis H4 is confirmed. In addition, hypothesis H5 asserts that higher level
of review objectiveness leads to higher diagnosticity. According to the regression analysis results, objectiveness has a significant (.000) positive effect with a standardized Beta of .411, thus hypothesis H5 is confirmed.
Hypothesis H6 states that an increase in review concreteness increases its argument quality. The
result of the regression analysis shows that concreteness has a positive effect on argument quality with a standardized Beta of .437 at a significance level of .000. As a consequence, hypothesis H6 is accepted. Also, hypothesis H7 asserts that an increase in online consumer
review concreteness results in an increased level of diagnosticity. Based on the results of the regression analysis, concreteness has a significant (.000) positive effect on diagnosticity with a standardized Beta of .406. Thus hypothesis H7 is accepted.
Hypothesis H8 claims that consumer reviews having a high quality linguistic style, also have a
higher level of argument quality. Based on the regression analysis, linguistic style has a positive and significant (.000) effect on argument quality with a standardized Beta of 311, consequently hypothesis H8 is accepted. In addition, hypothesis H9 states that consumer reviews that bear a
