THE EFFECTS OF E-WOM VARIANCE ON SALES
PERFORMANCE IN FASHION INDUSTRY: THE
MODERATING ROLE OF BRAND STATUS
Bunga Ghassani
11385871
MSc Business Administration
Entrepreneurship and Management in the Creative Industries
Supervised by R.G.H.J. (Rens) Dimmendaal MSc
January 24
th2017
University of Amsterdam
STATEMENT OF ORIGINALITY
This document is written by Bunga Ghassani-11385871, who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in
creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
ABSTRACT
Electronic word-of-mouth (eWOM) communication through online consumer reviews in e-commerce platforms have become a trusted source of information for consumers before making an online purchase decision in order to reduce their uncertainty of online purchases, as the perceived risk is relatively high. Online consumer reviews give every consumer the opportunity to share their opinions and evaluate what other consumers say about a particular product. Most of the online retailers allow consumers to give their evaluation through product ratings. Variance of ratings measures and shows the heterogeneity of consumer reviews. Previous findings have shown some inconsistency regarding the effects of this rating valence variance on sales. However, the most common findings show that it has a negative effect on sales. This study aims to extend the research on this topic by analyzing the effects of variance in rating valence on sales when it is moderated by brand status and to mitigate information asymmetry between brands and consumers due to rating variance. In this study, rating valence variance is measured by average ratings and standard deviation ratings during the ‘training period’ or the time when people give their reviews genuinely with less influence from other reviewers. Whereas, the sales perfomance is measured by volume of rating reviews during the ‘sales period’ or the time when consumer purchasing decision may be influenced by the prior reviews. The empirical setting of this study is fashion industry. The sample for this research is taken from dataset of shoes category in Amazon Fashion, by excluding data of products with less than ten reviews. The total number of data collected from Amazon database is 479. OLS regression is used to process the data. The results of this research show that rating valence variance has a significant and negative effect on the sales performance of a product. Standard Deviation Ratings is found to have stronger negative value than Average Ratings. In addition, when the effect is moderated by brand status, it becomes negatively stronger on the sales of Non-Top Brands than the Top Brands. However, this result is not found to have a stronger negative influence on the sales of women’s products compared to men’s products.
TABLE OF CONTENTS
1. Introduction... 5
1.1 Conceptual Framework... 9
2. Literature Review ... 10
2.1 Electronic Word-of-Mouth (eWOM)... 10
2.2 Brand Status... 15
2.3 Gender ... 17
2.4 Fashion... 18
3. Research Design and Methodology ... 19
3.1 Research setting ... 19 3.2 Data Collection... 20 3.3 Methodology... 21 4. Results... 21 4.1 Descriptive statistics... 22 4.2 Correlations... 24 4.3 Regression analysis...\... 25 5. Discussion... 30
5.1 Theoretical foundations and hypotheses ... 31
5.2 Limitations, future research, and contributions to the literature ... 34
5.3 Practical Implications... 35
6. Conclusion ... 36
1. INTRODUCTION
The development of Web 2.0, has changed the way consumers share and acquire information from online stores. The growth of e-commerce continues to rise since 2014 (BusinessInsider, 2016). Along with the growth of e-commerce, electronic word-of-mouth (eWOM) communication has become a major source of information for consumers before purchasing a product. It begun in 2005 until 2010 era, when consumer started to shape the digital, social media, and mobile marketing (DSSM) where eWOM serves as an individual expression that matters to marketing (Lamberton and Stephen, 2016). EWOM consists of several types, one of them is online consumer review, in which different consumers can post any comment or opinion based on their experiences after purchasing a product and become the reference for other consumers. Nowadays, online reviews are perceived as a trusted source where consumers regularly rely on, before making purchase decisions in order to reduce the feeling of uncertainty prior to purchase as the perceived risk is relatively high (Langan et al., 2017; Almana and Mirzha 2013; Zhu and Zhang, 2010).
Most of the e-commerce retailers such as Amazon.com, Bol.com, and Walmart.com are allowing consumers to give their evaluation of a product through product ratings and descriptive reviews. As a result, these e-commerce retailers are able to show the average of all ratings to the future consumers (Wang et al., 2015 and Sun, 2012). For instance, Amazon presents the ratings review in bar charts. The average rating of a product is considered helpful by consumers prior to purchase decision, as the higher average ratings, the better quality of products (Sun, 2012). Thus, these online reviews created by consumers can work as “sales assistant” which potentially lead to an increase in product sales (Moe and Trusov, 2011; Chevalier and Mayzlin, 2006).
In most cases, each consumer does not give the same rating and the potential consumers can easily observe the dynamic of this rating distribution because some e-commerce retailers like Amazon make it salient to them and this may not always be helpful for them in order to reduce the risk and their uncertainty (Langan et al., 2017; Wang et al., 2015). Basically, the variance of ratings measures and shows the heterogeneity of consumer reviews. In addition, it can be easily obtained by summarizing the online ratings from each product. Given the fact that variance of
ratings is easy and cheap to obtain, it is important for practitioners to understand how the variance of ratings influences purchase decisions and how to create better demand forecasts to increase the sales (Langan et al., 2017).
The effects of online reviews on sales have gained researches interest over the last decade. The findings of Zhu and Zhang (2010) make an important contribution by investigating the effects of online consumer reviews variance on new product sales. They investigate the effect by looking at the popularity of video games. The results show that the less popular games get more benefit from online consumer reviews compared to the popular games. But, the online consumer reviews variance affects the less popular games more negatively than the popular ones. Moreover, the most common online consumer review metrics to be researched are volume, valence and variance. Previous scholars have studied the relationship each of these metrics towards sales and also investigated the combination of these metrics. For instance, Roy et al., (2017) find that valence increases the sales on software products. Furthermore, similar results are also found in the movie and book industry (Chintagunta et al., 2010; Chevalier & Mayzlin, 2006; Zhang and Dellarocas, 2006). Additionally, higher volume is found to generate more sales on movies and software (Zhou & Duan, 2012; Duan, Gu, and Whinston, 2008; Liu, 2006). Meanwhile, other researchers like Sun (2012) and Khare, Labrecque, and Asare (2011) examine the combination of valence and variance on book retailers and movies. They find that high variance with low valence reviews are increasing consumer’s intention to watch or to buy a product. On the contrary, high variances with high valence reviews are decreasing consumer’s desire to purchase a product. Thus, these findings are interesting because most of studies show that valence and volume affect the product sales as well as the combination of valence and variance.
According to recent studies, the effect of user reviews variance towards sales is found to be inconsistent. Some studies find a positive relationship between the online consumer reviews variance and the sales. For instance, Sun (2012) on her research examines the informational role of the variance and its interaction with product ratings. According to her results, a higher variance positively influences the consumer’s demand, only when the average rating is low. Additionally, she also finds a higher standard deviation of ratings improves the sales rank of a book in Amazon, when the average rating is lower than 4.1 stars. Similarly, Moe and Trusov (2011)
testify the impact of rating dynamics on both subsequent rating behavior and beauty product sales. The results show the ratings dynamics have the potential to significantly influence the sales of beauty products because product ratings do not always accurately reflect product performance.
On the contrary, other studies discover the negative effect of variance on the sales. One of them is a study by Langan et al. (2017) confirm that rating valence variance is negatively influence the sales by examining the interaction between review valence (average rating) and variance in the context of intrinsic and extrinsic cues. It shows that higher levels of review variance lower the purchase intention of utilitarian than hedonic products, when a product has low average rating. The reviews variance leads to consumer uncertainty, which negatively influence their decisions process (Langan et al., 2017). Meanwhile, Wang et al. (2015) discover the effect of reviews variance to be a double-edged sword, which can either hurt or help products sales, depending on the critic reviews variance and other quality signals. All the results of these studies encourage this research to examine further about the effect of rating valence variance on sales.
Nonetheless, when a product receives high dispersion of reviews, consumers tend to look at other signal to reduce their uncertainty and help them in making a purchase decision. It indicates the reviews variance has stronger negative influence on the sales. The brand is one of the signals that consumers usually use as a signal of quality and performance. In fact, Ho-Dac, Carson, and Moore (2013) on their research, examine the moderation effects of brand strength and category maturity on the positive and negative online customer reviews and they find that the strong brands are more protected from negative information and its performance is not affected by eWOM. Further, the strength of the brand is often associated with the brand status. The classification of brand status whether it is a low or high-perceived status is also determined by the image of the brands, popularity and consumer’s assessment (Miller and Mills, 2012). Higher status brands are considered to have more benefits than the lower status brands. Moreover, little is known about the effect of high dispersion in ratings on the sales of the products from low status and high status brand. Hence, it needs further examination whether the brand status moderates the effect of rating valence variance on the sales.
The current study attempts to clarify the effect of variance in rating valence, as it tends to have stronger negative effect on the sales performance of a product by also considering the effect of the brand status as the moderating variable. The standard deviation has long been used to measure the variance, following the purpose of this research as it is focusing on the rating valence variance, the average of ratings is added to measure the effect of the rating valence variance on the sales. As most of the studies mentioned, only investigate the effects on experience goods such as movies, books, television shows, software, and video games from the online retailers, this study also attempts to explore the effect within a new research context in another experience good by studying the fashion industry. The empirical setting of this study is Amazon Fashion. The top brands and non-top brands from Amazon Fashion are used as the indicators to measure the brand status. In this case, fast fashion brands usually collaborate with mass online retailers like Amazon in order to grab more consumers and sell their products quickly. Consequently, consumers may find it harder to know the real quality of the product or get the perfect size as they cannot check or try the physical product like in offline stores. As a result, they have to rely on the online reviews to get more information about the product. However, when there is variance on the reviews, it is expected to have negative effect on the consumer decision process because it gives them higher uncertainty, which affects the sales performance of that product. Thus, they have to rely on other quality signals such as brand status and to make inferences about the product quality. The above leads to the following research question: “How does brand status mitigate the negative effect of eWOM variance towards sales performance?”
Further, each consumer has a different response towards online consumer reviews as their trusted source of information to evaluate a product, in which can influence their purchasing decision. In addition, previous studies find that women and men respond the online consumer reviews differently and both of them have different online shopping behavior. Women tend to have higher perceived risk and they are more hesitant than men to make an online purchase, so women really pay attention on the consumer reviews (Abubakar et al., 2017; Bae and Lee, 2010). Supporting to these findings, Bae and Lee (2010) as well as Awad and Ragowsky (2008) investigate the effect of online consumer reviews on consumer’s purchase intention based on their gender. They show that the effect of online consumer reviews is found to have
stronger influence on women and the negative effect is also found to be stronger for women than men. Nevertheless, a higher variance on the reviews means a higher uncertainty for consumers, thus it is expected to have stronger negative effect on women than men. This research tries to look at the influence of rating valence variance closely from gender perspective. Therefore, in order to find out whether the negative effects of rating valence variance on the sales which moderated by brand status is stronger for women than men as well as enriching the results, the study includes gender as the controlling variable.
This study makes several contributions to the theories and practices. This research helps to shed a light on the inconsistent findings about eWOM variance on the previous literature and extend our understanding of these concepts by providing a unifying framework that helps practitioners understand of the interaction between variance with rating valence, as well as to bring new insights for professionals and academia from a broader perspective by investigating the moderating role of brand status and the role of gender as the controlling variable, which plays an important role to increase the sales of a product. Basically, the aim of this study is to clarify further and mitigate the information asymmetry between brands and consumers due to rating valence variance. Moreover, the findings of this study give beneficial insights on how to generate more sales and sustain a competitive advantage in the fashion industry.
This thesis is structured as follows: first, the literature review offers a thorough review of the relevant scientific theories, which leads to the development of hypotheses in order to answer the research question. Second, the research design and the methodology describe how the data is collected and indicate important characteristics. Third, the results from the data collection and analysis are presented. Fourth, the results are discussed and compared with the current literature. Fifth, the limitations and recommendations for future research are given. Sixth, the conclusion of this thesis provides a summary of the overall results and answers.
1.1 Conceptual Framework
Figure 1 Conceptual Framework
This figure illustrates the basic conceptual framework of the research. EWOM variance through rating serves as independent variable and measured by the average ratings and the standard deviation ratings. Sales performance represents the dependent variable is measured by number of ratings. Meanwhile, brand status works as a moderating variable between the rating variance and the sales performance. The effect is expected to be stronger for the Non-Top Brands than Top Brands. This study also includes gender as control variable, which effect is expected to be stronger on women.
2. LITERATURE REVIEW
This section provides literature overview from previous studies. It starts by introducing the theories of eWOM. Following the subsection, the literature on eWOM variance is explained. Third, the concepts of sales are provided and followed by the literature review of variance on sales. Fourth, comprehensive literature on brand status as a moderation variable is presented. In the next following subsections, several concepts of fashion industry and its relation to e-WOM are evaluated. The theories of gender as control variable are explained.
2.1 Electronic Word-of-Mouth (eWOM)
Electronic word-of-mouth (eWOM) “electronic consumer-to-consumer interaction regarding a brand or product”, performs a vital role in the way customers communicate with one to another on the Internet (Petrescu and Korgaonkar, 2011;
EWOM Variance
-‐Average Ratings
-‐Standard Deviation Ratings
Sales Performance
-‐ Number of Ratings
Gender
-‐ Women > Men
Brand Status
-‐ Non-‐Top Brands > Top Brands H1 (-)
H2a (-)
Brown et al., 2007). There are some definitions of eWOM given by previous scholars. EWOM refers to the both positive and negative online content about a product or entities, which is generated by people in online environment and made it accessible to other people (Hennig-Thurau et al., 2004). In addition, Litvin et al. (2008) describe it as informal communication between the consumers through the Internet about the use or characteristic of products or services or the companies. Thus, this research defines eWOM as an interaction among consumers in online environment about products, services, or entities, which content can be either positive or negative.
Consumers who engage in eWOM are triggered by the benefits of online interaction, economic incentives, and self-worth enhancement (Mishra, 2017). It can be diffused in many forms of social interaction such as online forums, blogs, review sites, and social media sites like Instagram, Twitter, YouTube, and Facebook, but online reviews and ratings are the most adopted (Floyd et al., 2014; Petrescu & Korgaonkar, 2011). Moreover, most of e-commerce platforms have actively tried to harness electronic word-of-mouth (eWOM) as a new marketing tool by allowing their consumers to post personal product evaluations by giving ratings and reviews on their websites which can be evaluated by other customers as well (Floyd et al., 2014).
Electronic WOM has certainly been a highly effective marketing strategy and it is considered by marketers as an important source of product information that influences consumer behavior, purchase decision and reduce the perceived risk (Lamberton and Stephen, 2016; Almana, 2013). As an example, Amazon.com has facilitated the eWOM interaction by encouraging its consumers to post their product reviews, and now the number of the product reviews have skyrocketed over 10 million consumer reviews across product categories on its website. Amazon’s online product reviews are widely used and considered as one of the most useful features for consumers (Floyd et al., 2014).
Indeed, eWOM has several advantages for both consumers and retailers such as the information bias can be diminished because the source of information is from different consumers and the content created is easily accessible through the Internet and can be controlled by online retailers (Floyd et al., 2014). All these benefits given by eWOM create the assumptions to the online retailers that by integrating online product reviews into their marketing strategies will significantly influence consumer’s purchasing decisions and increase their product sales which resulting to profits improvement (Utz, 2010). However, eWOM can be double edged-sword because
people can give any positive and negative statements about a product or a company. Previous studies found online product reviews strongly affect retailers performance (Roy et al., 2017; Cui et al., 2012; Amblee and Bui, 2011; Zhu and Zhang, 2010; Chevalier and Mayzlin, 2006) but the results are mixed, it may hurt retailer sales performance (Wang et al., 2015; Zhu and Zhang, 2010) or increase their sales performance (Wang et al., 2015; Sun, 2012; Moe and Trusov, 2011; Chevalier and Mayzlin, 2006) depend on the eWOM metrics used and the research context (Wang et al., 2015).
Volume, valence, and variance are the most common eWOM metrics to be studied which allow firms to harness these new sources of information for decision support and influence the consumer behavior towards products. As an example, Amazon.com typically summarize the numerical ratings by presenting the average ratings (the valence) in stars and the distribution of those ratings in a bar chart (the variance) on the top of the product page with the amount of volume reviews next to each bar. So potential buyers can easily find the valence, volume and variance of reviews and thus acquire valuable information about the product they want to buy. If they want to seek further information, they can click individual reviews and read the corresponding overall ratings and text messages. In practice, previously posted ratings affect individual’s posting behavior whether they want to follow the crowd or stand out from the crowd by giving different opinion (Wendy and David, 2012). In other words, the reviews given by consumer who made their purchasing decision after reading previously posted ratings are influenced by those prior reviews.
The quantity of online reviews or volume measures the total amount of eWOM interactions (Liu, 2006) and previous studies relatively show that it positively affects the product sales (Duan, Gu, and Whinston, 2008; Liu, 2006; Godes and Mayzlin, 2004). Moreover, the result is also applicable on the quality of online reviews or the valence, which is the average rating of the product that contains positive and negative opinions (Chintagunta, Gopinath, & Venkataraman, 2010; Moon, Bergey, and Iacobucci, 2010; Chevalier and Mayzlin, 2006). However, the effect of positive and negative review may vary depending on the product type such as high and low involvement or search and experiential goods (Langan et al., 2017). The brand strength is also considered to moderate the effect,in which positive online reviews have greater influence on products with weak brands (Ho-Dac et al., 2013). Meanwhile, according to prior studies, the effect of variance towards sales
performance is relatively sparse. Some studies show that the variation in ratings hurts the sales of less popular games (Zhu and Zhang, 2010). Conversely, Sun (2012) gives another perspective from books context, in which the rating variance is found to have positive effect towards the sales. In addition, other studies also find similar result that the dispersion of ratings increases the sales of beauty products (Moe and Trusov, 2011) and beer brands (Clemons, Gao, and Hitt, 2006). Meanwhile, Wang et al., (2015) study the variance on music, books, and digital cameras find that the overall effects of variance can negative, insignificant or positive. All in all, despite the significant insights provided by prior studies, eWOM has been an interesting topic to be researched further because the emerging studies of this topic always generate new findings.
2.1.1 Variance
Every piece of information is useful for potential buyers in order to reduce their risk and uncertainty, especially when they are not attached to any particular products yet. In some cases, online reviews given by consumers can be varied. Moreover, consumers tend to pay more attention on the overall ratings instead of the narrative evaluation because rating is considered as easy-to-process information and consumers do not want to spend their time to read the consumer reviews carefully (Wang et al., 2015). Variance in rating valence is a condition when there is inconsistency in the rating reviews. A general consensus holds that positive rating valence benefits product sales and the effect of variance has a tendency to hurt the sales, yet the current results are mixed.
Consequently, if the consumer’s rating feedback is highly inconsistent, it may harm the sales due to higher uncertainty that consumer faced (Rosario et al., 2016). Supported by Zhu and Zhang (2010) on their study about the effect of rating variance in online games, in which they find that variation in ratings has a negative impact on sales, especially for unpopular online games. Additionally, in Langan et al., (2017) research, they illustrate the effects of interaction between review valence (average rating) and variance in books context and discover that the non-simultaneously
presented valence and variance information can be mistakenly assume by the consumer and it affects their decision process. For instance, if there are two books received the same average ratings and the same volume of reviews. The first book got
1000 consistent four-star ratings, while the second book got 600 five-star ratings, 200 four-star ratings, and 200 one-star ratings. It is shown that the high levels of variance lead to misperceptions about the average rating of a product and leads to consumer uncertainty, which negatively influence their decisions process. It is found that higher levels of review variance lower the purchase intention of utilitarian than hedonic products, when a product has low average rating. Their interesting finding is the brand equity found to eliminate the negative effects of high review variance towards purchase intention. Alternatively, both high and low brand equity products have no influence on purchase intention when the review variance is low. In other words, consumers tend to rely on brand equity when the review variance is high, but they trust the consumer review when the variance is low. Likewise, Khare et al. (2011) find that high variance reduces consumer’s intention to watch the movie, when the average rating is negative. In sum, the variance in rating valence is found to have negative influence on the sales, especially when the average rating is low.
On the contrary, studies done by Moe and Trusov (2011) and Clemons, Gao, and Hitt (2006) show the opposite result. High variance in ratings can also be positively correlated with sales growth because the ratings do not always accurately reflect product performance and it also indicates a successful hyper-differentiation of product. Similarly, Sun (2012) presents that a higher variance represented by a higher standard deviation of rating positively influences the sales, only when the average rating is lower than 4.1 out of 5 stars because it is associated with a niche product. However, when the average rating is higher than 4.1, the variance tend to have negative influence on the sales. Further, Wang et al., (2015) extend the result of previous studies by discovering user reviews variance effects in movies, digital cameras, and books industries can be negative, insignificant, or positive, depending upon the variance of critic reviews and other quality signals such as product cost and product extension. Despite of the mix findings, the effect of rating variance is expected to have stronger negative influence than the rating valence, although it can be positive under certain circumstances. In short, no matter how low or high the average rating is, the rating variance has stronger negative effect on the sales performance.
There are some other conditions and product characteristics in which different effects of variance occur and influence consumer evaluations. According to Langan et
al., (2017) and Park and Park (2013) studies, high variance reviews may increase or decrease product evaluations depending on consumer’s prior expectations and the characteristic of the product whether it is searched, experienced, utilitarian or hedonic. Moreover, consumer who has a high pre-commitment to a product will not be influenced by variance (Khare et al., 2011). Based on these studies, it can be concluded that the effects of online consumer reviews variance may differ under certain conditions and depending on the characteristic of the product itself. As the variance gives higher uncertainty to consumers, thus the effect of variance in rating valence is expected to be negative towards the sales performance. Therefore, the following hypothesis is proposed.
H1 : Variance in rating valence has negative effect on product sales 2.2 Brand Status
Status is obtained through judgments and known as a signal of recognition. In the process of status judgment, individuals judge an entity based on particular assessment whether it is classified into certain categories or ranked in social classes which in result generates privilege or discrimination(Bitektine, 2011; Washington & Zajac, 2005). Washington and Zajac (2005) define status as ‘special benefits or known as privileges, which are granted and enjoyed by socially constructed high-status individuals, groups, organizations, or activities and not necessarily based on their performance or quality in a social system. Furthermore, brand status is defined as a perception given by consumers based on their assessment to a brand, whether it displays high levels of quality, prestige, luxury, and symbolizes success, which affects their feelings about the brand and their purchasing decision (O’Cass and Frost, 2002). As the needs of social recognition matter in nowadays society, status which has been known to have beneficial effects becomes more important not only for individual but also for the brands to prove their existence. In order to signal their social status, consumers often take brand names into consideration when purchasing goods, because brand names have always been attached to status and it determines the status level of an individual who owns it (Mazali and Rodrigues-Neto, 2013). This condition is called as status consumption that refers to consumers who have a positive feeling after purchasing a product from a particular brand for the sake of their status and self‐image (Aron and Hmily, 2002).
Fashion industry is a good example for this case because fashion brands are linked to social stratification, in which affects consumer’s risk taking and purchasing behavior (Mazali and Rodrigues-Neto, 2013; Becker et al., 2005). As consumer tend to pay more attention what they are familiar to, branding is really important in fashion industry, not only to create impression of what the brand want to be perceived and make consumers aware of the brand existence, but also to present the status and deliver the experience to the consumer. The status of the brand whether it is classified as low or high-perceived status is also determined by the image of the brands, popularity and consumer’s assessment (Miller and Mills, 2012). In result, some brands are willing to give more efforts to get a better status and to be perceived as the top brands because product or brand with high status could charge higher price and get some other privileges such as awareness, trust, perceived reputation, and quality
(Washington and Zajac, 2005; Benjamin and Podolny, 1999).
For instance, giant online retailer like Amazon has its own strategy that takes reputation to a new level. Amazon divides the status of their brands into two categories top brands and non-top brands. By choosing the option menu or the featured brands page, consumers can see the list of the Amazon top brands from each product category. However, there is a fee of partnership need to be paid to Amazon by the brands to be listed as top brands and get the privileges such as extra protection against counterfeit from the third-party sellers and better merchandise placement (Racked, 2014). These privileges might be beneficial for both the ‘well-known brands’ and the ‘unpopular brands’. Most of the ‘well-known brands’ pay the fee to protect their brands and maintain their reputation. Meanwhile, for the ‘unpopular brands’ it is good for them to raise their status, perceive reputation and consumer’s awareness, without being judged of their items quality based on information available on the Internet (DigidayUK, 2017; De Figueiredo, 2000).
As most of online retailers have enabled the consumers to give their opinion on their platforms, the reviews given can be positive or negative. The positive reviews will benefit the brands as they get the cheapest form of marketing. On the other hand, negative reviews will be more powerful at influencing the consumers and harmful for the brands (Cukul, 2015). Further, when the reviews are varied, consumers tend to use other quality signals such as brand to reduce their uncertainty that affect their decision process. This indicates the reviews variance has stronger negative influence on the sales. Ho-Dac, Carson, and Moore (2013) on their study find that strong brands are
more protected from negative information. In this case, the negative effect is expected to be lower for higher status brands as it has more benefits than lower status brand. Thus, the following hypothesis is proposed.
H2a : Variance in rating valence has a stronger negative effect on product sales of products with low brand status than for products with high brand status 2.3 Gender
Each gender has been identified to have different effects on the perception of eWOM and trust in online shopping (Abubakar et al., 2017). Men and women are found to have a gender gap as they respond the consumer reviews differently and have different online shopping behavior, which affect their purchasing decision. In online shopping environment, females tend to have higher perceived risk and more risk-averse than males, so they really rely on what other consumers are saying regarding the products, while men consider giving a voice or share their experiences is more important than trusting other people reviews. Moreover, males have more positive attitudes towards online shopping and they have a tendency to trust and be satisfied with all the consequences of online shopping (Abubakar et al., 2017; Pascual-Miguel et al., 2015; Bae and Lee, 2010; Olsen and Cox, 2001). In detail, Bae and Lee (2010) as well as Awad and Ragowsky (2008) on their investigation about the influence of online consumer reviews on consumer’s purchase intention based on their gender, discover that the effect of online consumer reviews is found be stronger influence on women and the negative effect is also found to be stronger for women than men. Nevertheless, the inconsistency of reviews gives a higher uncertainty for consumers, thus the effect is expected to be negative on women than men due to their different responses toward consumer reviews. Thus, this research includes gender as the controlling variable to enrich the results by considering about the influence of rating valence variance closely from gender perspective and testify whether the negative effects of rating valence variance on the sales, which moderated by brand status is stronger for women than men.
2.4 Fashion
Fashion items are often attributed with current trends and how people want to be perceived in society, in which evaluation requires personal examination (Jung et al. 2014). It is also known as positional good and identity good, which differentiate the owners with other people and acts as a status symbols at the society.
According to Kawamura (2005), fashion is a conceptual tool to understand the nature of the relationship between people and cultural objects. In this sense, different culture has different type of fashion style, which represents the origin of people. Additionally, Odom et al. (2009) and Verbeek (2005) tend to emphasize fashion as more closely related to product symbolic qualities, in particular the ways that products are used to express identity and lifestyle. Thus, fashion is determined as one that has a symbolic (psychological) value, aesthetic, and cultural meanings, as well as emotional needs as individual and social beings for the consumers.
Brand is really important in fashion industry because it basically creates customer feelings and the dynamics of the fashion process are important to understanding a consumer’s brand experience across times and situations (Kim, 2012). By coordinating all of the branding signals, consumers can feel the experience of fashion brands as a whole. Nevertheless, the brand experiences may vary on every consumer from different cultures and particular brands are considered to be more valuable in one country than in others, due to image and symbolic meanings attached to the brands, which formed by local culture the economic and social values differ from one to another (Jung and Sung, 2008).
Beside known as experience goods, fashion items are also classified as specialty goods because initially, consumers need to invest physical effort to be able to purchase these items (Murphy and Enis, 1986). Further, consumers for this type of good are willing to make an extra effort and ignore other alternatives to obtain their preferred products and brands to satisfy their distinct need (Heuer et al., 2015) Despite that many fashion brands have entered the online market, consumers are no longer need to give their physical effort by going to the store, but they are faced to higher degrees of perceived risk in online shopping such as risk of undelivered and defective products, as a result of their inability to try and check the physical products (Biswas and Biswas 2004, Kirmani and Rao, 2000). For this reason, consumers strongly rely on product information available on the web and other consumer
reviews to reduce their anxiety before making an online purchase (Langan et al., 2017; Heuer et al., 2015).
In practice, product ratings do not always reflect the sales performance of fashion items because it may be influenced by other factors as well, such as gender as women and men have different shopping behavior and different response to the reviews and consumer judgment or personal examination because fashion is closely associated with consumer behavior, culture, style and lifestyle practices. Lastly, branding is identified as one of the strongest signals that consumers use in e-commerce (Heuer et al., 2015; Baye and Morgan, 2009). All in all, the effects of rating variance on sales performance of fashion items may be moderated by the brand status and controlled by gender of consumers.
3. RESEARCH DESIGN AND METHODOLOGY
This part is started with research setting and sample. Second, data collection is explained. Third, the explanation and operationalization from each variable are provided. Lastly, the method for this research is presented.
3.1 Research Setting
The empirical setting of this research is fashion brands in Amazon.com, where all the brands are classified to be part of Amazon Fashion. Amazon.com is chosen because it is one of the largest and most popular online retailers and it has extensive consumer reviews system. In Amazon Fashion, they classified the brands as top brands and non-top brands, where consumers can sort the type of products based on the brands or through the featured products page, which displays the products from the top brands. Afterwards, consumers can see all the list of top brands for the type of product they choose. The ones are not in the list of top brands are considered as non-top brands. Thus, this classification of the brands whether it is non-top brands or non-non-top brands represents the brand status and it is used as the sample population for this study.
Amazon Fashion does not only divide the category of products, but also the products into some departments based on age and gender, such as women, men, boys,
girls, and baby. Following the goal of this research, the study only focuses on men and women categories. Further, in order to get more focused result and testify the effects on both gender equally, this research only limits to one product category, which available on both men and women departments as the research sample. Hence, shoes category is chosen as the sample for this research.
Following the past research on consumer reviews (Zhu & Zhang, 2010), this research will use three review variables: the average rating, the standard deviation of rating variance, and the total number of reviews posted. Although, the sales data is not accessible, Amazon displays the volume of reviews for each product, which is considered as the proxy of sales or represents the number of product sold. Accordingly, the rating volume is used as an indicator of sales response.
3.2 Data Collection
This quantitative research uses Amazon product database as the main source of data. However, for this study the data is taken from dataset of ‘Clothing, Shoes, and Jewelry' category. The access to obtain this database is granted by Julian McAuley (2016). This dataset consists of product reviews (ratings, text, and helpfulness votes), metadata (descriptions, category information, price, brand, and image features), and links (related products) from Amazon, with the time span from May 1996 to July 2014. As there are some un-reviewed products in that period, this study excludes all the products without reviews information.
Further, there are four types of variables to test in this study. Started with rating variance as independent variable. It is measured by the average ratings and the standard deviation of ratings. In order to be able to calculate the standard deviation, this study excludes the data of products with less then ten reviews. Second, the sales performance as the dependent variable is operationalized by volume of rating reviews as the proxy of sales because in Amazon only consumers who have bought the products can give their personal reviews. In this case, the period of data taken from the rating is divided as the ‘training period’ and ‘sales period’. The data for average ratings and the standard deviation ratings are taken during this ‘training period’ or the time when people give their reviews genuinely with less influence from other reviewers. It is using the assumption of the first twenty percent of the rating volume. The number of ratings to measure the sales is taken on the ‘sales period’ because after the ‘training period’, consumer purchasing decision might be influenced by the prior
reviews (Wendy and David, 2012), so then the number of reviews after the training period can be used for the dependent variable of sales. After sorting the necessary data, the total number of data collected is 479. Third, this study uses brand status as the moderating variable, because it is expected to moderate the relationship between rating variance and sales performance and it is operationalized by two indicators, those are top brands and non-top brands from the shoes category in both men and women’s departments. Fourth, Gender is considered to have influence on eWOM (Abubakar et al., 2017). In this study, gender is used as the controlling variable and operationalized by using dummy variable women (1) and men (0) determined by each product from both men and women’s section.
3.3 Methodology
In order to achieve the purpose of this study and to test hypotheses, database research design is conducted to measure the effect of rating variance towards sales performance. Single cross sectional sampling method is chosen, where the sample data will be obtained once (Malhotra, 2010). The first step is by sorting the unnecessary data and removing all the duplicate products from the big data. Second, the descriptive statistics analysis is used to find out the demographic characteristics of the sample. Before running the multiple regression analyses, there are some assessments for the variables such as, multicollinearity, linearity, normality, homoscedasticity test to make sure the data is valid for regression analyses and avoid future analysis problems. As the measurement scale for both the independent and dependent variable is ratio, the use of ordinary least square (OLS) regression for data analysis is applicable to see the causal relationship of independent and dependent. SPSS 23 is used to run all the statistical analysis from descriptive analysis to multiple regression.
4. RESULT
This section shows the comprehensive result of data analysis by using multiple regression. It is started with descriptive statistic analysis of the sample. Second, the
correlations between variables are presented. Third, the results of classic assumptions test before running the regression and the results of the multiple regression are discussed.
4.1 Descriptive Statistics
On the first table, the average ratings are classified into 5 categories by dividing the score gap between the minimum (1) and maximum (5) score with the rating number (5 stars) to get the interval score. This table helps to determine whether a product is classified as poor, fair, good, very good, or excellent based on its average rating.
Table 1 Average Ratings Classification
Rating Interval Description
1 – 1.8 Poor
1.81 – 2.6 Fair
2.6 – 3.4 Good
3.41 – 4.2 Very Good
4.21 – 5 Excellent
Table 2 Descriptive Statistics Summary
Variable N Min Max Mean Median
Standard Deviation
Ratings 479 0.24 3.87 1.21 1.213 Average Ratings 2 4.94 4.04 4.143 Number of Ratings 7 13262 433.48 66
Table 2 displays the descriptive statistics summary of the independent variables and dependent variable. It contains the N as the total number of sample, the minimum value, the maximum value, the mean, and the median from each variable. The total number of sample for this research is 479. As one of the variable to measure the variance, standard deviation ratings assess how spread out the data from the average (mean). A high standard deviation means the data is less reliable as it is widely spread with the value (Standard Deviation ≥ 1), while a low standard deviation means the data is more reliable as it tends to be very close to the average with the value below 1 (Malhotra, 2010). According to table 2, the lowest standard deviation value is 0.24 and the highest standard deviation value is 3.87. Both the average (1.21)
and the median (1.213) of standard deviation have similar value that means most of products are considered to have high variations on their ratings.
The average ratings classification table helps to interpret the descriptive statistic result of average ratings. The lowest average rating (2) of a product in the data is classified as fair and the highest average rating (4.94) is categorized as excellent. Based on the mean and median values (4.04 and 4.143), most of the products fall into the very good category based on the consumer’s opinion.
In this study, number of ratings represents the number of product sold. The statistical data shows there is a huge gap between the minimum value (7) and the highest amount (13262) of a product sold. The product with this highest value is considered as the best selling product from the product category. Besides, the mean is often get affected by the outliers, as a result the value becomes bigger than the reality (433.48). In this case, median that shows the central value of the data is more reliable than the mean to reflect the average number of product sold with the value of 66.
Table 3 Descriptive Statistics of Dummy Variables Variable Indicators Frequency Percentage % Brand Status Top Brands 311 64.7
Non-Top Brands 169 35.3
Gender Men 174 37.3
Women 305 63.7
N 479
Table 3 shows the frequency and the percentage from the descriptive statistics analysis of the variables that are using dummy. It gives a picture about the sample data used for this research. According to the table above, most of the shoes are purchased from the Top Brands category with 311 products sold or 64.7% of the total sample. The table also presents the buyer distribution based on gender. Based on the statistic, women’s shoes are bought more than men’s shoes with the frequency comparison of 305 to 174. It can be concluded that women are the majority of the buyer of shoes products with 63.7% of the total percentage compare to men.
4.2 Correlations
Table 4 Means, Standard Deviation, and Correlations
Variables Mean SD Min Max 1 2 3 4
1 Sales: Number of Ratings 433.48 1465.9 7 13262 2 Average Ratings 4.044 0.54 2 4.94 .073 3 Standard Deviation Ratings 1.21 0.44 0.24 3.87 -.012 -.746** 4
Brand status: Non-Top Brands (1) and
Top Brands (0)
-.127** -.025 -.020
5 Gender: Women (1)
and Men (0) -.098* -.048 .017 .122**
** Correlation is significant at the 0.01 level (2-tailed) * Correlation is significant at the 0.05 level (2-tailed)
Before running the regression, it is beneficial to test the correlations between variables. It is used to determine whether the relationship exists between X and Y and summarize the strength of association between two metric. Table 4 consists of some information such as means, standard deviations, minimum, maximum and correlations.
The mean of the sales means the total average of product sold, which gathered from the total number of rating reviews after the ‘training period’ is 433.48 and the standard deviation is 1465.9, showing that the amount gap of product sold between each product is quite high, ranging from 7 as the minimum value and 13262 as the maximum value. Next, the total mean of the average ratings given by consumers is 4.044 and the standard deviation of 0.54 is comparatively low with the minimum average of 2 and maximum average of 4.94. The mean of standard deviation ratings is 1.21 with a comparatively low of standard deviation as well between each product (0.44), ranging from the lowest variance of 0.24 and the highest variance of 3.87.
Based on table 2, the brand status and the sales have significant negative correlations with the significance value (0.003) below 0.05 at the 0.01 level or 99% confidence level for two-tailed test. The strength for this correlation is -0.127 or 12.7%. This could suggest that brand status is one of the things to consider by
consumers when it comes to purchasing a product. However, there is stronger negative correlation for the non-top brands and the sales, which means there is an inverse relationship between these two variables, in which when one variable increases and the other decreases or when the non-top brand is high, the correlation with sales is negative.
Variable gender and sales also have Sig. < 0.05 and the strength of its correlation is -0.098 or 9.8% at the 0.05 level or 95% confidence level for two-tailed test, this means there is a significant negative correlation between gender and sales. In short, an increase number in women decreases the sales of the product. Likewise, the significance value for the standard deviation ratings and the average ratings is 0.000 at the 0.01 level or 99% confidence level for two-tailed test. The value is below the critical value 0.05 but the strength of correlation is -0.746 or 74.6%. This value means both standard deviation ratings and average ratings have significant negative correlations. In other words, when the average ratings is high, the correlation with sales is positive but when the standard deviation ratings is high, the correlation becomes negative. However, the control variable gender is significantly and positively correlated with the moderation variable brand status at the 0.01 level or 99% confidence level for two-tailed test, with 0.122 or 12.2% the strength of its correlations. These correlations could suggest that women as the majority of the consumers generally consider about the status of the brand, which in turn can result in higher sales of shoes from the top brands.
4.3 Regression Analysis
Before running the OLS regression analysis, the variables are checked whether it is valid to be examined further with regression. First, by assessing the multicollinearity problems between independent variables. The collinearity tolerance value has to be above 0.2 for both variables (Malhotra, 2010). It is found that the collinearity tolerance value for average ratings and standard deviation ratings is 0.433, which means both independent variables have no multicollinearity problems and valid for the regression analysis. Since both independent variables have no correlation to each other, the regression analysis can be continued.
Table 5 Results of Regression Analyses of the Independent Variables and the Dependent Variable Variables Adjusted R2 F-Test Coefficients Independent Variable Dependent Variable F sig. Standardized Coefficients Beta t-Value sig. Average Ratings Number Ratings 0.672 2.285 0.014 0.145 2.122 0.034 Standard Deviation Ratings -0.152 -2.471 0.014
Model summary, ANOVA and regression coefficients are reported Two-tailed test p < 0.05
The first part shows the multiple regression result of the independent variables toward the dependent variable. The Adjusted R Squared, which represents the percentage of contribution of the independent variables on dependent variable for this model is 0.672 that means 67.2% of the variation value of the dependent variable (Number Ratings) can be explained by the variation value of independent variables used in the model (Average Ratings and Standard Deviation Ratings) and the rest of 32.8% is explained by other variables that are not included in this research model. The value is often used to represent the goodness of fit of the model and it becomes more accurate if the value is closer to 1. It indicates, the model has a good fit with the data and 67.2% accurate to predict the result.
The F-test shows the F value of this model is 2.285 with the significance value (0.014) below 0.05. Hence, this model is significant to test the first hypothesis and predict the relationship of Average Ratings and Standard Deviation Ratings on Number Ratings. The coefficients results are used to analyze the relationship of these variables. It is shown that the t-value for Average Ratings is 2.122 with the significance value of 0.034. This t-value is above the critical value ±1.96 at the 0.05 level or 95% confidence level for two-tailed test, which means Average Ratings has positive and significant effect on the Number Ratings. On the contrary, the t-value of Standards Deviation Ratings shows negative value -2.471 (-t-value < -1.96) at the 0.05 level or 95% confidence level for two-tailed test. This value means that the Standard Deviation Ratings has a significant negative effect on the Number Ratings. In other words, the relationship effect is significant for both independent variables but each variable has different effect towards dependent variable. When the Standard Deviation Ratings is high, the effect on sales will be negative, meanwhile when the
Average Ratings is high, the effect will be positive on sales. However, based on standardized coefficients beta and t-value, the Standard Deviation Ratings has bigger value than the Average Ratings, which means it has stronger negative value than Average Ratings. Hypothesis 1 suggests that variance in rating valence has negative effect on sales performance. Therefore, hypothesis 1 is supported.
Table 6 Results of Regression Analyses of the Interactions between Moderation Variables and the Independent Variables on the Dependent Variable
Variables Adjusted R2 F-Test Coefficients Interactions Moderation Variables Dependent Variable F sig. Standardized Coefficients Beta t-Value sig. Average Ratings x Non NTB (1) TB (0) Number Ratings 0.226 3.086 0.016 0.087 1.906 0.005 Standard Deviation Ratings x Non NTB (1) TB (0) -0.122 -2.557 0.011
Model summary, ANOVA and regression coefficients are reported Two-tailed test p < 0.05
In the second model, the interactions with Brands Status as moderation variable are included. The first interaction of Brand Status by its dummy variable Non-Top Brands-Top Brands with Average Ratings is called moderator 1 and its interaction with Standard Deviation Ratings is called moderator 2. This model is used to see the moderation effects of Brand Status and test hypothesis 2a, which suggests that variance in rating valence has a stronger negative effect on the sales of products with low brands status than for products with high brand status. The analysis starts from the Adjusted R Squared that shows 0.226 or 22.6% of the variation value of the dependent variable (Number Ratings) can be explained by the variation value of moderation variable interactions and the rest of 77.4% is explained by other variables that are not included in this research model. This indicates, the accuracy of the model is only 22.6% or considered has a low good fit with the data. However, the main goal of this study is to understand the relationships between the variables. Thus, the low value of the Adjusted R Squared is not a problem for this study because the coefficient values are more relevant to predict the relationships.
The interactions model with moderation variable is significant to test the second hypothesis with the F-value 3.086 and significance level below the critical value (0.016 < 0.05). The coefficients of regression present different result on each interaction. First, the t-value for moderator 1 is 1.906 (t-value > 1.96). It indicates that the interaction of Average Ratings with Non-Top Brands - Top Brands on Number Ratings has a positive effect but not significant. Second, the coefficients for moderator 2 have negative and significant effects on the Number Ratings, indicated from the t-value (-2.557) below the critical value of -1.96. It suggests that for the Non-Top Brands an increase in Standard Deviation Ratings has a stronger negative effect on the Number Ratings or the sales performance of the products than the Top Brands. In essence, the brand status significantly moderates the effects of rating valence variance toward the sales of products. However, the effect of rating valence variance is different for Non-Top Brands and Top Brands. The same result as the first model, standardized coefficients beta and t-value of the Standard Deviation Ratings is bigger than the Average Ratings. However, since the effect of moderator 1 is not significant, thus the moderator 2 is used to represents the effect on sales. Hence, hypothesis 2a is supported as this finding is matched with the hypothesis that suggests variance in rating valence has a stronger negative effect on the product sales of the low brands than the high brands and the illustration of the interactions effects is on Figure 2 below.
Figure 2 Interactions between Rating Valence Variance and Brand Status
Nu m be r R at in gs
Rating Valence Variance
Non-Top Brands Top Brands
Table 7 Results of Regression Analyses of the Interactions between Control Variables and the Moderation Variables on the Dependent Variable
Variables Adjusted R2 F-Test Coefficients Interactions Control Variables Dependent Variable F sig. Standardized Coefficients Beta t-Value sig. Moderator1 x Women (1) Men (0) Number Ratings 0.193 5.372 0.005 0.132 2.187 0.029 Moderator2 x Women (1) Men (0) -0.130 -1.333 0.183
Model summary, ANOVA and regression coefficients are reported Two-tailed test p < 0.05
The third part includes the control variable gender with dummy variable Women-Men in the interactions model. Further, hypothesis 2b can be tested by this model, which suggests that the variance in rating valence has a stronger negative effect on the sales of women’s products than men’s products with low brands status. Following the Adjusted R Square, the variation value of control variable interactions are able to explain the variation value of the dependent variable (Number Ratings) as much as 19.3%, and the rest 80.7% is explained by other variables that are not included in this model. This means, the model has a low good fit with the data and the accuracy is 19.3%. As mentioned before, this value does not have much influence on the expected relationships of the variables, but it helps to understand the strength of the model.
The F-value for the interactions model of both moderator 1 and moderator 2 with Women-Men is 5.372 (t-value>1.96) and its significance value is 0.005 (sig. < 0.05). It indicates that this model is significant to test the third hypothesis. The first interaction is between the moderator 1 and dummy variable Women-Men. According to coefficients of regression, the t-value for this interaction is 2.187 and the significance value is 0.029. It suggests that the interaction of Moderator 1 with Women-Men has positive and significant effect on Number Ratings. Specifically, an increase in Moderator 1 has a stronger positive effect on the Number Ratings or sales performance of women’s products than men’s products. However, the interaction between moderator 2 and dummy variable Women-Men presents the opposite result. It exhibits negative and not significant effects on Number Ratings, indicated from the
t-value (-1.333) above the critical value -1.96. Thus, the H0 is accepted. In brief, the Moderator 1 has stronger positive value on sales of women’s product as the standardized coefficients beta and t-value show bigger value than Moderator 2. An increase in Average Ratings of the Non-Top Brands has a stronger positive effect on the number ratings or the sales performance of women’s products than men’s products. It actually suggests the opposite effect of the hypothesis 2b. Consequently, hypothesis 2b is not supported.
Table 8 Results of Regression Analyses of the Interactions between Independent Variables on the Dependent Variable
Variables Adjusted R2 F-Test Coefficients Interactions Independent Variables Dependent Variable F sig. Standardized Coefficients Beta t-Value sig. Average Ratings x Standard Deviation Ratings Number Ratings 0.321 2.631 0.011 -0.517 -2.351 0.019
Model summary, ANOVA and regression coefficients are reported Two-tailed test p < 0.05
This last model includes the interaction between the two independent variables and all of the variables to confirm the negative effect of the variance in rating valence. The Adjusted R Squared explains the accuracy of this model and the variation value of number ratings can be explained by the variation value of the independent variables interactions as much as 32.1%. The result presents that when all variables are added, the t-value of the interaction between Average Ratings and Standard Deviation Ratings is -2.351 with significance value of 0.019. This implies that the effect of rating valence variance remains negative and significant, when both variables simultaneously affect the number ratings.
5. DISCUSSION
In this section, the results from the previous analysis section are discussed with some theories. It starts with a brief summary of the theoretical foundations of this research. Second, the results of the hypotheses are explained. Third, limitations of
this study are described along with the recommendations for future research and the contributions to the literature. Last part of this section presents the practical implications for practitioners.
5.1 Theoretical Foundations and Hypotheses
This research discovered the effects of variance in rating valence on sales and also looked at the interactions of the brand status and gender in the empirical setting of online fashion retailer. Previous studies have studied the concepts of online consumer reviews variance (Langan et al., 2017; Wang et al., 2015; Sun, 2012; Moe and Trusov, 2011; Clemons, Gao, and Hitt, 2006), but their findings are relatively sparse and mostly only investigate the effects on experience goods such as movies, books, television shows, software, and video games from the online retailers. This research extends the concepts of variance in rating valence by investigating the moderating effect of brand status, which controlled by gender in a setting where consumers are care about the brand status. According to those literature, the effects of variance in rating valence was expected to be negative toward sales, as it is found to increase consumer’s uncertainty. Additionally, the hypothesis expected the negative effects when it is moderated by brand status and these negative effects expected to be stronger for women. The results of the regression analyses showed the two out of three hypotheses are supported. The negative effect is not found to be stronger for women and the result was actually the opposite from what this research had expected.
The first hypothesis suggested that variance in rating valence has negative effect on the shoes sales and the result was matched with the hypothesis. There were two variables used to measure the effect, those are the average ratings and standard deviation ratings. The negative effect was expected to be stronger to influence the sales. However, both of these variables were found to be significant but the negative effects of standard deviation ratings are stronger than the positive effect of average ratings at influencing the sales. In short, the results suggest that an increase in variance reflected by standard deviation ratings would lead to a decrease of the total number of shoes sold.
There are several explanations for these results. First, consumers may feel insecure because the information that they rely on is not consistent. As stated on the previous literature, higher variance hurts the sales due to higher consumer’s uncertainty (Rosario et al., 2016). Basically, the effect of average ratings is in line
