Was this review helpful to you? An enquiry into user reviews helpfulness. How perceived user helpfulness is influenced by different expressions of affect and product attributes.
-Master’s Thesis-
-Graduate School of Communication-
Author: Blagoy Velkov Student ID -10842101
Supervisor: Dr. Aart S. Velthuijsen
Master’s Programme: Communication Science Track Specialization: Persuasive Communication Date of Completion: 29th January, 2016
Abstract
The present research advanced a model to test whether perceived user review helpfulness can be influenced by the expression of positive or negative affect and if this relationship can be
moderated by the type of product attribute described in the review. One hundred ninety five people took part in an online experiment with a 2 (positive vs. negative affective expression) by 2 (search or experience attribute type) between subjects design. The results demonstrated no statistically significant difference in the mean scores on overall perceived helpfulness between reviews containing positive and negative affective expression. Furthermore, there was no statistically significant relationship between the scores of reviews about search attributes containing positive affective expression and reviews about experience attributes containing positive affective expression, neither there was a statistically significant difference on the scores of the overall perceived helpfulness between reviews about search attributes containing negative affective expression and reviews about experience attributes containing negative affective expression. The results also demonstrated that the difference in overall perceived helpfulness between negative and positive affective expression is of similar magnitude for reviews about experience and search attributes. The current study adds to the to the scientific knowledge as being one of the first of its kind to research the combined effect of type of affective expression and type of reviewed product type with the use of an online experiment. Furthermore, this is one of the first studies which research the difference between search and experience attributes in one and the same product.
Introduction
Recent advancements in technology have made it is easier than ever to share an opinion about a product with people all around the world (Chen & Xie, 2008; Duan, Gu, & Whinston, 2008). In turn these changes have led to a huge rise in the accumulation of user generated
reviews, especially in the last decade. More and more businesses (Chen & Xie, 2008; Duan et al., 2008) have started providing user generated reviews in order to facilitate consumers in their online or offline shopping decisions. Being perceived as more credible than the official
information delivered by corporations or companies is one of the main reasons why user reviews play such an important role in modern market activities (Lee & Koo, 2012; Sen & Leerman, 2007). Even though consumers can be suspicious or skeptical towards the specifications and the advertising efforts made by the producers of certain products (Coker & Nagpal, 2013), they have fairly strong trust in other consumers, expecting them to be honest and straightforward about the qualities of the commodities they are describing (Casaló, Flavián, Guinalíu, & Ekinci, 2015; Lee & Youn, 2009).Therefore, consumer generated feedback on products is in high demand and is greatly appreciated by customer in numerous areas of online shopping (Casaló et al., 2015).
When consumers face a situation wherein they are in doubt which product or service they should choose, they need more information in order to resolve these product related uncertainties (Lee & Koo, 2012). One very efficient way of acquiring this information is by reading user generated reviews on the Web. These reviews can help consumers familiarize with different products or services and decrease the risk of eventual bad market decision (Lee & Koo, 2012). Furthermore, while company-provided information usually concerns very technical aspects of products and compares them according to their technical aspects and capabilities, user reviews reflect the personal experience of a consumer with a specific product or service and the extent to
which this experience managed to match his initial expectations (Chen & Xie, 2008). User generated reviews put the importance on the consumer and his or her needs , they help to overthrow the seller-centric marketing communication model, replacing it with a consumer-centric one (Lee & Koo, 2012), while facilitating the growth of trust between fellow consumers and the community spirit among users of specific products and services. Furthermore reviews are usually an easily reachable resource of information, they are undemanding to understand and can be accessible at will, making them a simple and convenient way for consumers to inform
themselves about products (Lee & Koo, 2012).
According to previous literature user reviews can affect sales through either increasing awareness for certain products or through their persuasive effects (Duan et al., 2008). The accumulation of user reviews can increase the awareness of consumers about certain products and services, while the content of the reviews can shift/alter users' product evaluation in a positive or negative direction (Duan et al., 2008).When users are in search of information, they want to decrease the risk associated with a certain decision about a product and/or service (Lee & Koo, 2012). Therefore they need data with high informative and diagnostic value which they can trust and base their decision on. However this can prove to be a challenging task, especially because of the anonymity that the Internet provides to the authors of user reviews (Lee & Koo, 2012). The companies that provide user reviews assist consumers in their quest for reliable information by the provision of helpfulness rating attached to every review, a measure that helps people to discriminate between high and low diagnostic reviews (Connors, Mudambi, & Schuff, 2011; Mudambi & Schuff, 2010).
One under researched area about user review helpfulness is to what extent the expression of affect can influence people’s evaluation of the review and increase or decrease its perceived
informative value (Kim & Gupta, 2012). Understanding how affect steers people’s evaluation of online user reviews helpfulness can be beneficial for both practitioners and researchers because affective expressions are abundant in eWOM (Kim & Gupta, 2012). Furthermore there is strong evidence that people can easily interpret and discriminate between different affective cues in the context of online communication (Yin, Bond, & Zhang, 2014) and their evaluation of the
perceived review helpfulness can be influenced by the expression of affect in the review (Kim & Gupta, 2012; Mudambi & Schuff, 2010). Online shopping sites are in constant competition with each other to secure more customers and providing your clients with easy and transparent way to discriminate between highly and lowly helpful user reviews is one way to earn an edge over your competitors (Connors et al., 2011; Mudambi & Schuff, 2010; Pan & Zhang, 2011).
Even though previous research suggest that people can be influenced by affective expressions in user reviews (Kim & Gupta, 2012; Mudambi & Schuff, 2010; Pan & Zhang, 2011), there is still plenty to be researched on the topic. A particularly interesting area of
research is how emotional expressions shape reviews and what is the moderating role of the type of product on the effect emotions have on the perceived user review helpfulness? The division between product search and experience products is a classification which is often used in marketing research (Mudambi & Schuff, 2010). The first type of products can typically be assessed before purchase, while the latter type can be evaluated mostly based on personal
experience (Chua & Banerjee, 2014; Nelson, 1970). Even though there is some existing research that explores the combined effect of affective expression and type of product attributes described in the user review, still to the best of our knowledge, there are no such studies which utilize online experiments as their data collection method. What is more, this research in one of the first in the field which examine the influence of search and experience attributes using a single
product that combines both of these different types of attributes. This leads to the proposition of the following research questions.
Research questions
How does the expression of positive and negative affect influence the perceived helpfulness of online user reviews?
How is this effect different for reviews that describe predominantly search product attributes and reviews which describe predominantly experience attributes?
For an illustration of the proposed research questions, please check the conceptual model of the study presented below. The independent variable type of affective content (positive vs. negative) will exert influence on the dependent variable – overall perceived helpfulness. This relationship is expected to be moderated by the other independent variable of the study – type of reviewed product attribute (search vs. experience).
Theoretical background:
Conceptual model of the study
Type of affective expression Type of reviewed product attribute Overall perceived helpfulness
Perceived Review Helpfulness
As it was already mentioned, one of the most important aspects of user reviews is that they help consumers improve their knowledge about different products and services and
therefore decrease the risk of possible bad decisions when it comes to eventual purchases (Lee & Koo, 2012). Review helpfulness can be defined as the perceived value of the information that helps online users to discriminate between different alternatives for products or services
(Mudambi & Schuff, 2010). Another, similar definition is - the value that is added by the specific piece of information to the decision making process (Connors et al., 2011). Customer reviews can provide assistance on multiple stages of the purchase process – from the shaping of the consumer need, through the evaluation of possible alternatives, to the post-purchase evaluation of the product (Mudambi & Schuff, 2010). Once a purchase necessity is acknowledged,
consumers often need some additional information in order to better differentiate between
competing products and services and choose the best possible option (Mudambi & Schuff, 2010). To this end consumers will need highly diagnostic information.
According to Li, Huang, Tan and Wei (2013), there is still no clear-cut conceptualization of helpfulness in the existing literature. Mudambi and Schuff (2010) use the consumer-generated measure of helpfulness provided directly by the websites from which they obtain their data. The helpfulness rating in these websites is generated when consumers are asked the question “Was this review helpful for you” (Yes/No). Then, the number of helpful votes (Yes) is divided by the number of total votes (Mudambi & Schuff, 2010). Some researchers (Li et al., 2013) have argued strongly against this way of measuring helpfulness, because of the so called “winner circle bias” – reviews which are evaluated as more helpful usually get promoted higher up and are made more visible to consumers, hence they get more attention and respectively more “it was helpful”
votes. This starts a self-perpetuating circle of ever increasing helpful votes number (Li et al., 2013). Moreover, the reviews published on earlier dates have more exposure than later reviews and therefore higher chance (“early bird bias”) to receive positive evaluation (Li et al., 2013). Li and his colleagues (2013) advise that in order to really understand what makes an online review helpful, it would be better to study what types of form and content users themselves evaluate as helpful when they have to evaluate a specific user review. The present research will try to follow this advice by manipulating certain characteristics in user reviews and consequently examine their effects with an online experiment.
Providing useful and diagnostic information for their customers can be viewed as a key competitive advantage that all websites want to have in order to secure loyal and satisfied consumers (Connors et al., 2011). There is research evidence that reviews that are voted as more helpful exert bigger influence on consumers purchase decisions than less helpful reviews (Chen, Dhanasobhon, & Smith, 2008). What is more, consumers often base their subsequent actions on a small subset of helpful reviews (Mudambi & Schuff, 2010; Yin et al., 2014). It is a common policy of successful on-line shopping sites to try and promote most helpful user reviews and show them first to information-seeking customers (Yin et al., 2014). For example, it is estimated that Amazon.com increased their income with 2.7 billion dollars, just by adding the question “Was that review helpful for you?” and progressing the ones that are most helpful (Yin et al., 2014). Therefore it’s crucial to understand what characteristics of the reviewer and the content of the review make it helpful for consumers.
Previous literature on helpfulness has identified several dimensions of helpfulness (Li et al., 2013; Yin et al., 2014). Information diagnosticity is one of the most commonly occurring ones, defined as the “the extent to which a given piece of information discriminates between
alternative hypotheses, interpretations, or categorizations” (Herr, Kardes, & Kim, 1991, p. 457). Another one is perceived source credibility, or to what extent the reviewer is perceived as
credible and trustworthy (Li et al., 2013). The third one, perceived vicarious expression, refers to the extent to which the reviewer is able to realistically convey the experience he or she had with the product so that experience can be fully felt and understood by consumers (Li et al., 2013).
In the present research the focus will be only on the first and the third dimension, because all participants are going to be presented with one and the same reviewer’s profile/persona and therefore the scores on this dimension are expected to be controlled for. These dimensions are going to be operationalized in the same way Li and his colleagues (2013) did in their research on the relationship between user reviews helpfulness and source and content features.
Emotions
Before the influence of emotions on eWOM is discussed in greater detail, it is important to first make the distinction between two similar and related constructs – emotions and moods. Both terms are usually categorized as affective processes (Yin et al., 2014). Even though in everyday language these two words are used somewhat interchangeably, there is a difference in their scientific usage. Moods commonly refer to a somewhat persistent feeling state, the causes of which are often blurry, unspecified and characterized by low arousal (Yin et al., 2014). On the other hand, emotions are generally shorter, more intense feeling states and their existence can be attributed to a specific thought or event (Yin et al., 2014). The term that is going to be used most in the present research is going to be affect, because it represents a more general description of a feeling state. Taking into account that affective expressions in user reviews are rarely high in arousal, but their existence can be typically attributed to a specific cause (the reviewed product),
the use of more general term such as affect is advisable. Even though the word emotion is used occasionally in the current research paper, it is done so as a synonym of the more general term affect, due to the fact that the feelings demonstrated in reviews are hardly ever intense enough to qualify as typical emotions such as happiness, surprise, anger, fear, sadness, disgust.
Affect has been identified as an important field of study in eWOM, because it has been proven that it can drive consumer behavior in online context (Jones, Ravid, & Rafaeli, 2004; Ludwig et al., 2012). Past research has demonstrated that affective information is more readily accessible than facts and other descriptive information (Ludwig et al., 2012; Zajonc, 1980) and affective content is usually spotted and processed faster than cognitive assessments (Ludwig et al., 2012; Zajonc, 1980). What is more, previous studies suggest that when people are presented with affective cues, they can trigger automatic affective responses and reactions which typically require small mental effort and subsequently shape consumer’s attitudes and decisions (Cohen, Pham, & Andrade, 2008; Ludwig et al., 2012). Data from past research support the statement that emotions can influence people’s attitudes towards products and brands (Lau‐Gesk & Meyers‐ Levy, 2009; Ludwig et al., 2012).
The presence of affect in user reviews has been proven to influence conversion rates (Ludwig et al., 2012), information diagnosticity and helpfulness (Kim & Gupta, 2012a; Ludwig et al., 2012; Yin et al., 2014), and product evaluations (Kim & Gupta, 2012b). According to Kim and Gupta (2012), affective expressions is a common occurrence in eWOM, since there are no negative sanctions on expressing it online, due to the anonymity associated with the medium (Kim & Gupta, 2012b). Even though there are no personal ties between creators and consumers of online reviews, persons are still able to fully comprehend and integrate the affect they come across in these reviews into a coherent and logical story, which they can interpret clearly and
correctly (Kim & Gupta, 2012a). Affective content can be particularly effective in driving people’s review evaluations when they lack the expertise or the motivation to process the
information in the review, when the information presented is ambiguous and ambivalent or when they lack the cognitive resources to process the data due limited cognitive resources like
distraction or similar reasons (Ludwig et al., 2012).
According to some researchers, when it comes to consumer generated reviews, people tend to attribute the reviewer’s negative expression of affect to the reviewer’s predispositions and personal characteristics and positive expressions of emotions to product related
characteristics (Kim & Gupta, 2012). Some researchers (Ludwig et al., 2012) claim that highly expressive positive or negative emotions can decrease the perceived diagnosticity of consumer generated reviews. Kim and Gupta (2012), on the other hand, established that a single negative review can decrease the perceived information diagnosticity and product evaluation, while a string of negative reviews can increase the information diagnosticity and will again decrease product evaluation. When a single review contains an expression of positive emotion, this won’t significantly affect the perceived helpfulness of the review (Kim & Gupta, 2012). Helpfulness will be boosted when a string of reviews contacting expression of positive affect are presented to the consumer (Kim & Gupta, 2012). A lot of studies have demonstrated that negative affect exerts stronger influence on perceived review helpfulness than positive affect (Covey, 2012; Kim & Gupta, 2012; Ludwig et al., 2012; Pan & Zhang, 2011), with negative information often perceived as more diagnostic, because negative information is usually observed less often than positive information (Lee & Koo, 2012). It’s important to note that the subject of this research are moderate expressions of affect, mainly because reviews with too extreme expressions of
emotions are not going to be perceived as helpful. Therefore, we can posit the following hypothesis:
H1: Reviews containing negative affect expression will be perceived as more helpful than
reviews containing positive affective expression.
Search and Experience Attributes
One of the most prominent divisions of products in marketing literature is the one made by Nelson in 1970. He developed the SEC framework, which is still in use today. According to this framework – there are three types of products – search, experience and credence products (Girard & Dion, 2010; Nelson, 1970). For the purpose of the current research only search and experience products will be examined, mainly because there is no existing research on user reviews about credence attributes, since it is very hard to verify or falsify claims in user reviews concerning credence attributes.
Search products or also known as utilitarian products, represent the practical side of consumption and they are usually purchased because of their certain pragmatic and functional attributes (Chua & Banerjee, 2016; Nelson, 1970). Their product value can be typically assessed before purchase, because search attributes are quantifiable, measurable and objective (Nelson, 1970). These assessments are rational and cognitively driven (Pan & Zhang, 2011). Search product reviews tend to elaborate more on the objective and functional characteristics of the products and how these products meet the expected functional needs of consumers (Chua & Banerjee, 2016; Pan & Zhang, 2011). Examples of search products include photo cameras, photo printers, cell phones (Chua & Banerjee, 2016).
Experience products or also known as hedonistic or experiential products on the other hand tend to satisfy primarily the self-indulgent needs of consumption (Nelson, 1970). The quality of these products is hard to be assessed before consumption and depends highly on the individual tastes of consumers. The user appraisal of these products is generally subjective and emotional, varying significantly between different consumer tastes (Pan & Zhang, 2011). Therefore the reviews of experience products tend to focus primarily on the different aspects of the sensory experience people get out of these products and are they are a reflection of the people’s personal preferences about these products (Pan & Zhang, 2011). Typical examples of experience products are books, music, visual arts (Chua & Banerjee, 2016).
Largely due to the advancement of the internet and the increased number of user
generated reviews, nowadays there is more and more information available, for a broader set of products characteristics, than ever before (Zhang, Ma, & Cartwright, 2013). Therefore, the difference between search and experience products has become blurry and not so clear cut as it used to be in the past (Zhang et al., 2013), because today even information for experience attributes can be sought and found (Zhang et al., 2013). However several researchers have proven that this is still an important and valid distinction (Girard & Dion, 2010; Mudambi & Schuff, 2010), mainly because people tend to process and evaluate information about search and experience products differently (Chua & Banerjee, 2014, 2016; Pan & Zhang, 2011). In the next subsection the different ways of processing information about search and experience products will be examined in greater detail.
User generated reviews for search and experience products and their perceived helpfulness
Researchers have demonstrated that in general consumers find user reviews about search product to be perceived as more helpful than user reviews about experience products (Mudambi & Schuff, 2010; Pan & Zhang, 2011). Researchers base this finding on the observation that people are generally inclined to attribute the content of reviews about experiential products more to the reviewer’s personality than to the characteristics of the product itself (Pan & Zhang, 2011). Due to their subjective nature, experience product reviews have a bigger share of extreme ratings than of moderate ones (Mudambi & Schuff, 2010). However, precisely the reviews containing moderate evaluations are perceived as more helpful when it comes to experiential products. The exact opposite is true when it comes to search products – in that situation extreme positive or extreme negative reviews are credited as more helpful than moderate ones (Chua & Banerjee, 2014; Pan & Zhang, 2011). If users encounter highly positive or highly negative evaluation of an experiential product they will most likely attribute these feelings to the reviewer’s personal dispositions (Mudambi & Schuff, 2010). Because search product reviews contain more objective and measurable information people are less likely to discredit the information they contain even if this information is presented in highly polarized manner (Pan & Zhang, 2011). Therefore we can posit our first two hypotheses:
H2: Reviews for search products containing expression of negative affect will be voted as
more helpful than reviews for experiential products containing negative effect.
H3: Reviews for search products contacting expression of positive affect will be voted as
According to previous research the demonstration of affect in user reviews usually has stronger influence on the perceived helpfulness of consumer generated reviews about experience attributes, than user reviews that are for search attributes (Pan & Zhang, 2011). As it was already mentioned above, people have the tendency to perceive extreme arguments for search attributes as more trustworthy than for experience attributes, mainly because of the fact that claims for search attributes can be easily verified or falsified (Chua & Banerjee, 2014), due to the quantifiable, measurable and objective nature of all search characteristics (Chua & Banerjee, 2016). Taking into account that people will still perceive negative information as more diagnostic than positive information, most probably the difference in perceived helpfulness between positive and negative affective conditions will be greater for experience than for search attributes. To add to that, negativity bias is usually stronger for experience than for search products (Lee & Youn, 2009). Hence the following hypothesis is proposed.
H4: There is going to be a higher difference in perceived helpfulness between positive
and negative affect for reviews of experience than for reviews of search products.
Methods
Stimuli
The stimuli were developed so that they can look as close as possible to real customer reviews published in Amazon.com. The choice of a smartphone as a reviewed product was motivated by the reason that smartphones combine both search and experience attributes. For example the camera of a smartphone has both objective, measurable and quantifiable qualities (typical for search attributes) and characteristics that can only be experienced like the quality of the photos made, their clarity or crispness (experience attributes). Other experiential attributes
can relate to the smartphone design or user experience (experience attributes). The information across the four different conditions was kept as comparable as possible in order to avoid the influence of external factors to contaminate the achieved results.
On Figure 1 it can be observed how the stimulus material presented to the participants in the positive affective condition about search attributes looked like. The other 3 versions of the stimuli can be seen at the Appendix at the end of the thesis.
Figure 1. Stimulus material for people in the positive affective condition about search attributes. Participants
Because of money and time restraints the present research utilized a convenience sample. The inclusion criteria of the study were: all participants should be at least 18 years of age and all participants should be fluent in English, because otherwise they would not understand the survey questions. One of the respondents was excluded because of answering with the highest possible values to all questions. The final sample consisted of 194 respondents. Of them 46,9 % were male and 53,1 % were female. The participants were between 18 and 62 years old, with an average age of 26,60. Participants didn’t receive any financial reward for their partaking in the experiment.
Design
The experiment was conducted online, utilizing the web platform Qualtrics and it consisted of two between subject factors (type of affect: negative vs. positive affect, type of attribute: search vs. experience attribute). The four groups were compared on the helpfulness ratings they gave to the different user reviews they were exposed to.
Procedure
The subjects of the study were advanced via various channels – email, social media postings (Facebook and Twitter) or in person. They were briefed with short summary about the purpose of the experiment and then asked to fill in the survey in Qualtrics.
Just before the commencement of the survey the participants were asked to sign the informed consent that provided the researchers with the legitimate right to use their data for research purposes on the behalf of the University of Amsterdam. The respondents who agreed to participate in the study then were randomly assigned to one of the four conditions of the
experiment. These conditions included as follows:
Reviews about search attributes with expression of positive affect - 46 respondents, 23,6% of the total respondents
Reviews about search attributes with expression of negative affect - 51 respondents, 26,2% of the total respondents
Reviews about experience attributes with expression of positive affect - 51 respondents, 26,2% of the total respondents
Reviews about experience attributes with expression of negative affect - 47 respondents, 24,1% of the total respondents
The study subjects were requested to carefully read the reviews presented to them and subsequently all of them had to answer questions measuring the dependent variable of the study, which was – to what extent every type of review was perceived as helpful by the respondents. After these questions were answered there was a section containing control and demographic questions. After that respondents were thanked for their participation and efforts, debriefed and left with the option to receive the results of the study via e-mail.
Measures
Dependent variable
Perceived review helpfulness – in order to measure perceived review helpfulness the scale of Mengxiang Li, Liqiang Huang, Chuan-Hoo Tan, and Kwok-Kee Wei (2013) was used. The scale measures the three dimensions of review helpfulness – perceived credibility, perceived vicarious expression and perceived diagnosticity. Because the goal of the research is to compare the consumer generated reviews only according to their textual contents, the perceived credibility dimension is not going to be measured, because it refers mainly to the perceived credibility of the reviewer’s persona. The perceived diagnosticity dimension contains three bipolar 5-point agree-disagree (Likert) items that measure – to what extent the respondents felt that the review helped them to familiarize with the product (1), to evaluate the product (2) and to understand how the product performs (3). The perceived vicarious expression dimension again consists of three bipolar 5-point agree-disagree (Likert) items which measure the extent to which consumers perceived the review as helping them to feel what the reviewer wants to communicate (1), to envision what the reviewers tries to communicate (2) and to imagine what the reviewer want to communicate (3). Motivated by the use of these scales in previous research (Li et al., 2013), the
perceived helpfulness score was calculated as the average scores of all items in the two dimesons. A PCA with Oblimin1 rotation demonstrated that the items on the scale formed two dimensions which corresponded exactly to the dimensions already described by (quote). The perceived diagnosticity dimension had the following characteristics - EV = 2.51; R2 = .42; α = .74 with a mean score of 3.23. The perceived vicarious expression dimension had the following characteristics EV = 1.26; R2 = .21; α = .65 with a mean score of 3.77. The scale as whole had Cronbach’s α=.72 and mean score 3.50.
Control variables
Smartphone ownership – this question is set to measure if people own smartphones or not, because smartphone ownership can influence subsequent evaluations of review helpfulness, because they will have more explicit evaluations about smartphones. This question has two possible answer options – yes/no. 94.8% of the people (184) owned a smartphone and only 5.2% (10) of the respondents– did not.
Experience with the reviewed phone (Nexus 5X) – if people have experience with the phone at hand, they will compare the information provided in the review with their own experience and maybe that can influence their subsequent evaluations. This measure was
operationalized with two questions. The first question was - “Have you heard of Nexus 5X?” and there were 3 possible answer options – Yes/Not Sure/No. 49% (95) of the people in the sample were unaware of the existence of Nexus 5X, 27,8% (54) were not sure and 23,2% (45) answered positively to the question. The second question was “If the answer to the previous question was
1
The choice of type of rotation was motivated by the fact that the two dimension were expected to and proved to be related.
"yes", what information did you get about Nexus 5x?”. This question had a bipolar, semantic-differential scale ranging from Extremely Negative (1) to Extremely Positive (7). 66 people in total provided answers on that question and the mean score on the scale was 4,45.
Trust in user reviews – Trust in user reviews has been measured by using two items. The first item was the question “Do you use reviews written by other customers in order to inform yourself for technology products?”, with possible answers
1. Never 2.6 % (5) 2. Rarely 14.4% (28) 3. Sometimes 28.4% (55) 4. Often 38.7% (75) 5. All of the times 16% (31)
The other item was “How helpful do you think user reviews are in general?”, which had a bipolar, semantic-differential scale ranging from Not at all helpful (1) to Very Helpful(7). The mean score for that question was 5.20.
Analysis
Covariates
Before the execution of the main analysis a regular check if some of the control variables exert influence on the final results was executed. According to the standard recipe for data analysis provided by GSC, this check was performed for every control variable in the following manner: first, there was a check if the control variable had a statistically significant relationship with the one of the two independent variables. If there was a statistically significant relationship between the independent variable and the control variable there was a second step which
consisted of checking for statistically significant relationship between the control variable and the dependent variable of the study. If the control variable had statistically significant
relationship with both the independent and the dependent variables, the particular control variable was included in the overall model as a covariate.
Age and Gender – There was no significant statistical difference between the gender and age composition of the different conditions and therefore these variables were not included in the model as covariates.
Smartphone ownership - smartphone ownership proved to be significantly different across the negative and positive affective conditions X2 (1, N = 194) = 3.93, p = .047.
Furthermore, a independent sample T-test, with smartphone ownership as independent variable and overall perceived helpfulness as dependent variable, demonstrated that smartphone owners perceived the reviews as significantly more helpful Mowners=3.54, SEowners = .04, than the people
who didn’t own a smartphone Mnon-owners=2.92, SEnon-owners = .26. This proved to be a statistically
significant difference with t(9.45) = -2.39, p = .040. Therefore the variable smartphone ownership has been included as a covariate in the overall model.
Experience with the reviewed phone (Nexus 5X) - There was no statistically significant relationship between the people’s previously heard about Nexus 5X, not heard or unsure if they have heard about the Nexus 5X and the valence or product attribute condition people were subscribed to at the beginning of the experiment, therefore the variable – Have you heard about Nexus 5X before wasn’t included in the overall model as a covariate. However the evaluation of previously heard information was different across the two affective conditions. People in the positive affective condition evaluated the information they have received for Nexus 5x as more
positive Mpositive= 5.17, SEpositive= .18, than the people in the negative affective condition
Mnegative=3.89, SEnegative=.273. This proved to be a statistically significant difference - t(64) =
3.684, p<.001. Bivariate correlation didn’t demonstrate any statistically significant relationships between previous information received about Nexus 5X and overall helpfulness, perceived vicarious expression and perceived diagnosticity. Therefore the previous information which people received about Nexus 5X, was not included as a covariate in the model.
Trust in user reviews - There was no significant statistical relationship between the frequencies with which people use user reviews and the affective condition or the attribute feature they were placed in. Therefore this variable was not included as a covariate in the subsequent analysis. Even though there was no significant statistical relationship between the extent to which people perceive user reviews as helpful in general and the affective or attribute type condition they were subscribed to, there were strong theoretical grounds to include this variable in the overall model, because it is most likely that people who perceive the reviews more helpful in general are going to give higher scores on the overall perceived helpfulness on the user reviews they were presented with. Furthermore there was a moderate bivariate
correlation between overall perceived helpfulness and the user reviews perceived helpfulness in general r= .29, p> .001. Therefore this variable was included as a covariate in the model.
Hypothesis testing
In order to test the four proposed hypotheses of the study ANCOVA was used with type of affective expression (positive and negative) and type of product attribute (search and
experience) as independent variables and overall perceived helpfulness as a dependent variable. Smartphone ownership and the extent to which people perceive user reviews to be helpful (“How
helpful do you think user reviews are in general?”) in general were included as covariates. The test demonstrated no statistically significant main effect for either type of affective content F (1, 188) = 1.826, p=.178, η2 = .01, nor attribute type F (1, 188) = .435, p= .510, η2 = .002. The two-way interaction effect of type of affective content and type of attribute was also not significant - F (1, 188) = .012, p= .912, η2 =.00. The covariate variable smartphone ownership proved to statistically significantly influence the scores on the overall perceived helpfulness scale. People with smartphones tended to perceive the user reviews presented to them as 0,68 points more helpful than people who didn’t own a smartphone. This proved to be a statistically significant difference - F (1, 188) = 9.88, p> .001, η2 =.05. The other covariate variable – the extent to which people perceive user reviews as helpful in general proved to exert statistically significant difference on the perceived helpfulness scores that people were giving as well. With every point increase on the general helpfulness of reviews score people tended to score with 0.11 points more on the overall perceived helpfulness of the reviews they were exposed to in the experiment. This proved to be a statistically significant difference - F (1, 188) = 13.380, p> .001, η2 =.07.
The first hypothesis of the study claimed that people will perceive reviews containing negative expression of affect as more helpful on average than reviews with positive expression of affect. Considering that the main effect for type of affective expression proved not to be
statistically significant this hypothesis had to be refuted. In fact the observed results, even though not significant, point in the opposite direction with the average score for positive reviews
Mpositive=3.55 which is a score which is slightly bigger than the overall perceived helpfulness
score people gave on average for negative reviews – Mnegative=3.46.
The second hypothesis of the study stated that people will perceive reviews about search attributes containing negative affective expression as more helpful than reviews about experience
attributes containing negative affective expression. Considering that interaction affect for type of affective content and type of product characteristic was not significant, this hypothesis had to be refuted. Moreover when the pairwise comparisons were examined the difference between search and experience reviews containing negative affective expression was -0.06, which was again not statistically significant with p=.584.
The third hypothesis of the study stated that people will perceive reviews about search attributes containing positive affective expression as more helpful than reviews about experience attributes containing positive affective expression. Like with the previous hypothesis, because of the statistically insignificant interaction effect between or type of affective content and type of product characteristic, this hypothesis had to be refuted. When the pairwise comparisons were examined, the difference on their overall perceived helpfulness between search and experience products containing positive emotion was 0.04, which was not a statistically significant
difference with p= .701.
According to the fourth hypothesis of the study, the difference in the overall perceived helpfulness between negative and positive affective condition was going to be higher for experience products than for search products. The interaction effect between type of product attribute and type of affective content was statistically insignificant, therefore this fourth hypothesis of the study had to be refuted. Examining the pairwise comparisons, it can be observed that the difference between the positive and negative reviews for search attributes was 0.12, and the between the positive and negative reviews for experience attributes was 0.10, so therefore it can be concluded that the observed results were counter the proposition advanced in the fourth hypothesis.
Conclusion and discussion
In the present study an online experiment was conducted in order to prove if people respond differently when they encounter positive or negative affective expressions in user reviews and if this difference is changed when the reviews cover predominantly search or experience attributes. In total 195 people took participation in the experiment.
The contribution of the current study to the existing research on user review helpfulness is mainly in the utilization of a new methodological approach. It’s one of the first studies to examine the combined effect of affective expressions and type of reviewed product attribute, using an online experiment as method of data collection. Generally researchers on user reviews collect their data utilizing one or more of three approaches. The first approach is the conduction of an online experiment (Kim & Gupta, 2012a; Lee & Koo, 2012; Yin et al., 2014). The second one, used mostly for reviews concerning a specific type of emotions, is the use of automated programs for data collection that go through a data set of reviews and on the basis of predefined rules distinguish and categorize the reviews based on the rules specified beforehand (Ludwig et al., 2012). The downside of the second data collection method is that sometimes the algorithm has troubles with the correct recognition of the demonstrated affect that is present in the user review. Especially hard for algorithmic categorization are the cases when the user expresses irony and therefore even though a person can correctly distinguish between real emotional expression and ironical one, this is not so easy for the computer. The third approach is to collect some or all the reviews for a specific product, chosen according to a specific criteria – type of product type that is reviewed, most popular products on a specific site (Chua & Banerjee, 2014), etc. However, when review helpfulness is examined in such a way,there is a high risk that the results can be influenced by the “winner-circle bias” or the “early bird bias”. Therefore, an
online experiment which examines the effect of type of affective expression and product type can produce far more reliable research results.
Generally in the existing literature on user reviews, when the relationship between user reviews helpfulness and the type of product attribute is researched, scientists compare different products, which can be described as predominantly search or experiential products (Chua & Banerjee, 2016; Girard & Dion, 2010; Nelson, 1970). However, according to some scientists the difference between the two types of products is becoming more and more blurry and unclear (Zhang et al., 2013). To some extent the lack of statistically significant findings in the current study is an indication of that trend as well. It makes sense, therefore, to talk not about search and experience products, but about search and experience attributes or characteristics, especially since most items combine both of them at the same time. Wine for example is usually described as an experiential product, but it also has some objective and searchable characteristics such as price, year of production, country of origin and so on. What is more, when search and experience attributes are examined in one and the same product, this makes the job of the scientist easier, because the different stimulus materials are as close to one another as possible, which helps to control for additional factors that can influence the study results. Therefore, introducing this new approach to the research of search and experience attributes is paving the road for even more interesting scientific finding in the future.
The current study had four different hypotheses. The first hypothesis of the study stated that people perceive user reviews with positive affective expression as less helpful than user reviews with negative affective expression. Based on the theoretical foundation that people often perceive negative information as more diagnostic because of its comparative rarity to positive information this hypothesis failed to reach statistical significance. However, when it comes to
perceived diagnosticity, completely opposite effect was observed – people tend to find positive information significantly more diagnostic than negative information. A possible explanation to this rather unexpected finding can be found in the research made by Kim and Gupta (2012). The researchers use the attribution theory which stipulates that people tend to attribute their positive behaviours and others’ negative behaviours to internal factors such as personality traits and to attribute their own negative behaviours and others’ positive behaviours to external factors, such as specific situational constraints or characteristics. According to their line of reasoning when people read user reviews, they tend to attribute negative ones to people’s own personal dispositions and idiosyncrasies. On the other hand when consumers come across reviews containing positive information, they attribute that information to the qualities of the product (Kim & Gupta, 2012a). It is possible that this mechanism made the respondents in the online experiment evaluate the negative information presented to them as less diagnostic and
respectively less helpful. Furthermore in the same research the scientists claim that when people encounter one negative review it usually lowers the perceived helpfulness of the specific review. It is completely different story, however, when people encounter several reviews and all of them express negative evaluation of the product. In that case people interpret these negative
expressions as being influenced by the specific characteristics of the product, not by the personal characteristics of the reviewers because they are verified by other users.
According to previous literature, extreme expressions of affect can be detrimental for the perceived review helpfulness when it comes to experience attributes (Chua & Banerjee, 2014; Pan & Zhang, 2011). Generally people tend to prefer user reviews about experience attributes to be more moderate, as moderation is perceived to be giving better and more objective information about the different qualities of the product. Because of the strictly functional and quantifiable
nature of search products, the helpfulness ratings of their user reviews are usually not as affected as the user reviews about experience attributes by the demonstration of extreme affect, no matter if this affect is positive or negative. Therefore, it was expected that user reviews containing positive affective expression about search products will have higher perceived helpfulness ratings than user reviews containing positive expression of affect, but describing experience products. However, this hypothesis failed to reach statistical significance. Similar to the second hypothesis of the study, the third hypothesis stated that user reviews about search attributes containing positive affective expression will have higher perceived helpfulness than user reviews about experience attributes containing positive expression of affect. This hypothesis also failed to reach statistical significance. In both hypotheses the average perceived helpfulness scores were really close to each other (not bigger than 0,06 points on a scale varying from 1 to 5), indicating that when it comes to affective expression people are more influenced by the valence of the emotion than the type of product characteristic that was described in the review, which was made even more evident by researching both types of attributes in one and the same product.
The fourth and final hypothesis of the study advanced the statement that there is going to be a bigger difference in perceived helpfulness between the positive and negative affective conditions for experience, rather than search products. Previous theory has suggested that affective content is more likely to affect experiential products more than search products and therefore if negative information was going to be perceived as more helpful than positive in bigger proportion for experiential products than for search products. However, this hypothesis failed to reach statistical significance. As it was said before a possible explanation for this lack of findings can be the fact, that people are far more influenced by the type of user emotion
Managerial implications
Even though the study didn’t reach any statistically significant results, there is still an important take-away message for practitioners. Affective expressions in user reviews matter and they can change the evaluation of a user review. This is particularly true when it comes to perceived diagnosticity dimension. An extra investigation into the data demonstrated that people perceive reviews with positive affective expression as more diagnostic than reviews with
negative affective expression and this difference was statistically significant. Therefore,
managers should try to promote reviews with positive affective for products, because consumers will perceive them as more diagnostic, compared to the ones containing negative expression of affect.
Limitations and directions for future research
Because of the time and resource constraints of the study there was no pretest of the stimulus material. Therefore, there is the possibility that the different stimulus materials are not fully comparable with each other, in terms of the intensity of the affect that was shown. This could have influenced participants in a way that they perceive the negative emotion of the study as more vivid than the positive, hence the negative reviews as more driven by emotion and therefore less helpful. A future research with extensive pretest of the stimulus material can ensure that the different affective conditions are similar to each other in terms of the intensity of the expressed affect. Furthermore if the texts of the reviews were tested without the use of affective words and expressions, that was going to give a good baseline to compare to what extent the texts for experience and search attributes differ in their perceived helpfulness for the users, when there is no emotion expressed.
A much needed extension of this research will be a future study in which the respondents are exposed not to only one isolated review, but to series of reviews. This is going to be much closer to what a consumer who is searching for a product generally experiences in reality, where he or she is usually encountered by series of user reviews, which shape his perception of a product. With such an experiment, researchers can be more certain if negative information is perceived as more diagnostic and helpful than positive information, because right now the results are mixed and inconclusive. Furthermore, the proposed study will establish what critical mass of user reviews in one direction is needed in order to push a consumer in a certain line of thought about a product.
References
Casaló, L. V., Flavián, C., Guinalíu, M., & Ekinci, Y. (2015). Avoiding the dark side of positive online consumer reviews: Enhancing reviews’ usefulness for high risk-averse travelers. Journal of Business Research, 68(9), 1829–1835.
http://doi.org/10.1016/j.jbusres.2015.01.010
Chen, P., Dhanasobhon, S., & Smith, M. D. (2008). All Reviews are Not Created Equal: The Disaggregate Impact of Reviews and Reviewers at Amazon.com. Heinz College Research, 1–32. http://doi.org/10.2139/ssrn.918083
Chen, Y., & Xie, J. (2008). Online Consumer Review: Word-of-Mouth as a New Element of Marketing Communication Mix. Management Science, 54(3), 477–491.
http://doi.org/10.1287/mnsc.1070.0810
Chen, Y., & Xie, J. (2008). Online Consumer Review: Word-of-Mouth as a New Element of Marketing Communication Mix. Management Science, 54(3), 477–491.
http://doi.org/10.1287/mnsc.1070.0810
Chua, A., & Banerjee, S. (2014). Developing a Theory of Diagnosticity for Online Reviews. Proceedings of the International MultiConference of Engineers and Computer Scientists IMECS, I.
Chua, A., & Banerjee, S. (2016). Helpfulness of user-generated reviews as a function of review sentiment, product type and information quality. Computers in Human Behavior, 54, 547– 554. http://doi.org/10.1016/j.chb.2015.08.057
Behavior. In Handbook of Consumer Psychology (pp. 297–348). http://doi.org/10.4324/9780203809570
Coker, B., & Nagpal, A. (2013). Building-Up versus Paring-Down: Consumer Responses to Recommendations When Customizing. Journal of Retailing, 89(2), 190–206.
http://doi.org/10.1016/j.jretai.2012.11.002
Connors, L., Mudambi, S. M., & Schuff, D. (2011). Is it the review or the reviewer? A multi-method approach to determine the antecedents of online review helpfulness. Proceedings of the Annual Hawaii International Conference on System Sciences, 1–10.
http://doi.org/10.1109/HICSS.2011.260
Covey, J. (2012). The role of dispositional factors in moderating message framing effects., 33(1), 52–65. http://doi.org/10.1037/a0029305
Duan, W., Gu, B., & Whinston, A. B. (2008). Do online reviews matter? — An empirical investigation of panel data. Decision Support Systems, 45(4), 1007–1016.
http://doi.org/10.1016/j.dss.2008.04.001
Girard, T., & Dion, P. (2010). Validating the search, experience, and credence product classification framework. Journal of Business Research, 63(9-10), 1079–1087. http://doi.org/10.1016/j.jbusres.2008.12.011
Herr, P. M., Kardes, F. R., & Kim, J. (1991). Effects of Word-of-Mouth and Product-Attribute Information on Persuasion: An Accessibility-Diagnosticity Perspective. Journal of Consumer Research, 17(4), 454–462. http://doi.org/10.2307/2626839
online interaction spaces: A theoretical model and empirical exploration. Information Systems Research, 15(2), 194–211. http://doi.org/10.1287/isre.1040.0023
Kim, J., & Gupta, P. (2012). Emotional expressions in online user reviews: How they influence consumers’ product evaluations. Journal of Business Research, 65(7), 985–992.
http://doi.org/10.1016/j.jbusres.2011.04.013
Lau‐Gesk, L., & Meyers‐Levy, J. (2009). Emotional Persuasion: When the Valence versus the Resource Demands of Emotions Influence Consumers’ Attitudes. Journal of Consumer Research, 36(4), 585–599. http://doi.org/10.1086/605297
Lee, K. T., & Koo, D.-M. (2012). Effects of attribute and valence of e-WOM on message adoption: Moderating roles of subjective knowledge and regulatory focus. Computers in Human Behavior, 28(5), 1974–1984. http://doi.org/10.1016/j.chb.2012.05.018
Lee, M., & Youn, S. (2009). Electronic word of mouth (eWOM): How eWOM platforms influence consumer product judgement. International Journal of Advertising, 28(3), 473– 499. http://doi.org/10.2501/S0265048709200709
Li, M. X., Huang, L. Q., Tan, C. H., & Wei, K. K. (2013). Helpfulness of Online Product Reviews as Seen by Consumers: Source and Content Features. International Journal of Electronic Commerce, 17(4), 101–136. http://doi.org/10.2753/JEC1086-4415170404
Ludwig, S., Ruyter, K. De, Friedman, M., Brüggen, E. C., Wetzels, M., & Pfann, G. (2012). More than Words: The Influence of Affective Content and Linguistic Style Matches in Online Reviews on Conversion Rates. Journal of Marketing, 77(January), 1–52. http://doi.org/10.1509/jm.11.0560
Mudambi, S. M., & Schuff, D. (2010). What Makes a Helpful Review? A Study of Customer Reviews on Amazon.com. MIS Quarterly, 34(1), 185–200. Retrieved from
http://ssrn.com/abstract=2175066
Nelson, P. (1970). Information and consumer behavior, J. Poiit. Con., 78(2), 311–329.
Pan, Y., & Zhang, J. Q. (2011). Born Unequal: A Study of the Helpfulness of User-Generated Product Reviews. Journal of Retailing, 87(4), 598–612.
http://doi.org/10.1016/j.jretai.2011.05.002
Sen, S., & Leerman, D. (2007). Why are you telling me this? An examination into negative consumer reviews on the web. Journal of Interactive Marketing, 21(3), 2–20.
http://doi.org/10.1002/dir
Yin, D., Bond, S., & Zhang, H. (2014). Anxious or Angry? Effects of Discrete Emotions on the Perceived Helpfulness of Online Reviews. MIS Quarterly, 38(2), 539–560.
Zajonc, R. B. (1980). Feeling and thinking: Preferences need no inferences. American Psychologist, 35(2), 151–175. http://doi.org/10.1037/0003-066X.35.2.151
Zhang, L., Ma, B., & Cartwright, D. K. (2013). The impact of online user reviews on cameras sales. European Journal of Marketing, 47(7), 1115–1128.
Appendix: s ears das das d
1. Review about search attribute containing positive affective expression
2. Review about search attribute containing negative affective expression
