The effectiveness of eWOM on sales for hedonic and utilitarian products : an empirical study in mobile application marketplace

(1)

The Effectiveness of EWOM on Sales for Hedonic and

Utilitarian Products: An Empirical Study in Mobile

Application Marketplace

University of Amsterdam

Faculty Economics and business

MSc Business Administration -- Entrepreneurship and Management in the Creative Industries

First supervisor: Rens Dimmendaal Name: Yiner Xu

Student number: 11387432 Date: 23-06-2017

(2)

Statement of originality

This document is written by Student Yiner Xu 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.

(3)

Abstract

This study examined the effect of electronic word-of-mouth on product sales for utilitarian and hedonic mobile applications and the moderators which affect the relationship between eWOM, perceived risk and sales. Analyses of panel data of 382 paid apps from Google Play reveal that eWOM has a stronger effect on sales of utilitarian products than hedonic products. This finding attributes to the fact that utilitarian products are evaluated by instrumental attributes (Dhar and Wertenbroch, 2000) so that consumers can get information about product quality more easily from eWOM. On the contrary, evaluated by subjective and personalized metrics (Dhar and Wertenbroch, 2000), eWOM of hedonic products lack indication of product’s quality, thus the sales will not increase if consumers still perceive higher functional risk (Babic et al., 2016). Thus managers should pay more attention to eWOM management for utilitarian products while improve the attractiveness of trials of hedonic products to reduce consumer perceived risk.

Key words: electronic word of mouth, product sales, perceived risk, hedonic product, utilitarian product, trialability, price.

(4)

1. Introduction

With the help of the advent and ubiquity of the Internet, consumers have explored new methods to learn knowledge about products before purchase (Lynch and Ariely 2000). In this era of online shopping, electronic word-of-mouth (eWOM) has become a essential information source that consumers use to plan their purchase choices. Because it is difficult for consumers to evaluate the products or the services with unknown quality, they tend to seek and rely on some information sources which seem trustworthy so that the uncertainty they feel will be reduced by consulting with others (Roselius 1971). These sources of consumer-generated information (e.g online product reviews) which includes personal experiences and suggestions from users of the products or services could facilitate to reduce the risk and uncertainty perceived by consumers as well as help them decide whether to purchase the products or services.

Until now, the eWOM’s impact on firm performance has been studied in an increasing number of academic works. According to the meta-analytic review by Babic et al. (2016), around 100 studies have investigated whether and to what extent eWOM is linked to the product sales in various industries during the past 15 years. For example, Chevalier and Mayzlin (2006) examine the positive effect of improving consumer reviews on sales of book and Dellarocas et al. (2007) examine this effect in movie industry. In addition, it is previously suggested that consumers perceive different levels of functional and financial risks for diverse products and services (Wangenheim and Bayón, 2004). In order to offer a further insight, Babic et al. (2016)

(7)

propose and examine that diverse moderators do influence eWOM effectiveness differently across platforms and product categories. Among the four product characteristics they studied, hedonic products are claimed to have higher functional risk than utilitarian products. Since eWOM is one of the most important tools to reduce perceived risk and further influence product sales, in this paper I expect that eWOM has a greater impact on sales of hedonic products than on sales of utilitarian products.

In this study, I also pay attention to two moderators that are able to affect the effectiveness of eWOM on perceived risk and product sales. First moderator is trialability. In fact, perceived functional risk can be moderated by providing trialability to consumers. The freemium strategy adopted by companies reduces functional risk by offering consumers opportunities to try their simplified or incomplete products for free so as to stimulate their willingness to buy a full-functional or upgraded products (Liu et al., 2014). The effectiveness of eWOM will be reduced when consumers can examine the product quality by experiencing the product themselves. Thus I predict that when product has a free trial, the effectiveness of eWOM on sales will also decrease for both utilitarian and hedonic products. The second moderator is price. It is suggested by Lin and Fang (2006) that in the case of high financial risk, consumers will rely more heavily on eWOM. However, there are insufficient research have studied the moderating role of price in the effectiveness of eWOM, therefore this study is designed to investigate whether the effectiveness of eWOM on sales for utilitarian and hedonic products will increase when the prices of

(8)

products increase.

Although there are already conclusive researches investigated the effect of eWOM on reducing risk and affecting sales, there is lack of study considering the difference in the effectiveness of eWOM on sales for products with different nature. To extend the extant knowledge about the relationship between eWOM, perceived risk and sales with insight of different product natures, the main research question of this paper is “Whether the effectiveness of eWOM on sales varies for products with utilitarian and hedonic nature, and if so, what moderator affects the relationship between eWOM, perceived risk and sales. ”

Empirically I will do this research by studying a panel data in mobile application market. The first reason for choosing this industry is that the mobile application market is now seen as a booming and innovative industry, and it is now earning increasing willingness of consumers to purchase for its products. According to the statistics provided by new zoo (global mobile market report 2016), there are 2.3 billion smartphone users worldwide and the global app revenues will reach $44.8 billion in 2016 and grow to $80.6 billion by 2020. Since mobile apps have inevitably become popular among consumers from all generations, the coverage of its audiences and the prospect of this industry attracts researchers become increasingly interested in investigating this industry and so do I. Earlier works have tested the relationship between eWOM and the sales in the mobile application ecosystem, more practical insights about the effect of product nature are needed for better understanding the market dynamics.

(9)

Second reason is that there are plenty categories of mobile apps nowadays, including games, education, business, lifestyle and countless others. Since consumers’ choices based on needs that can be holistically classified into two types by nature: hedonic and utilitarian (Dhar and Wertenbroch, 2000). Consumers have hedonic products and services which focus on the consumption experience (Hirschman and Holbrook, 1982), which reflecting the need for pleasure, fun and excitement. Conversely, utilitarian products and services are mainly instrumental and functional in nature. In mobile application market, mobile apps are ideally designed to be basically categorized into hedonic and utilitarian subgroups based on consumers’ needs differences. Consumers have hedonic apps which are mainly games and entertainment apps for excitement and fun, they can also download utilitarian apps that are mostly used for practical purposes. Having these apps with such specific and obvious nature that this paper is going to study, I believe the data set based on these samples will enjoy a great extent of validity for examining my hypotheses.

The remainder of this paper is organized as follows. Section 2 reviews the literature related to this paper. Then research framework and hypotheses are in section 3. Depending on literature review, I state four hypotheses which are related to the traditional theories of the relationship between eWOM, perceived risk and product sales as well as moderators of trailability and price. Section 4 is methodology, the data set used for the empirical analysis consists of a panel data of 382 paid apps from April 2016 to February 2017. The results of the study are presented in section 5 and the discussion and conclusion of these results are in section 6. The last section is the

(10)

limitation of this paper and suggestions for future research.

2. Literature Review

2.1 Hedonic and Utilitarian Product Nature

2.1.1 Definition

To classify products with their nature, consumers’ attitude towards different product types is among the most essential matters to be considered. It is suggested that consumers’ attitudes and consumption behaviors are inherently bi-dimensional. As Babin et al. (1994) have described in their study that consumers' evaluations of a shopping experience are assessed along two important dimensions: hedonic and utilitarian value. Therefore, when value is considered from experiential perspective, it is recognized as related intimately to hedonic senses. On the other side, when value is considered from instrumental perspective, it is recognized as related intimately to utilitarian outcomes.

Hedonic consumption is defined as those facets of consumer behaviors related to the multisensory, fantasy and emotive aspects of product use (Hirschman and Holbrook, 1982). In marketplaces, consumers have hedonic products and services which focus on the consumption experience and provide more sensory feeling such as fun, pleasure and excitement. Driven by attitudes and evaluation toward different nature of products, consumer choices based on needs include another type (Dhar and Wertenbroch, 2000) -- utilitarian consumption is always concerned with “expectations

(11)

of consequences”. Consumers have utilitarian products and services which are mainly equipped with instrumental or functional nature.

This paper extends the understanding of hedonic and utilitarian product nature with another popular theory about product nature. Nelson (1970) classifies products into search and experience products according to consumers' ability to obtain product quality information before purchase. In fact, consumers of e-commerce are usually uncertain about whether the products they purchase online will fit their needs or perform as they expected. The classification of search and experience products helps to explain consumers’ relative attitudes and consumption behaviors.

The hedonic and utilitarian products can be compared as well as related to experience and search products to a certain extent. Search products can be evaluated by specific attributes, functions and performance before purchase, and consumers tend to be objective and prefer to use a systematic decision-making process when assessing a search product. The utilitarian products have most characteristics of search products because consumers would evaluate the products objectively and emphasize on functional metrics so that they get useful information to achieve the practical goal of purchase. However, experience products offer feeling or experiencing, which are more difficult to describe by specific attributes, and may render varied experiences across consumers (Cui et al., 2012). The evaluation of experience products by consumers tends to have subjective variables and more personal-differentiated information about product quality. Most of hedonic products are mainly considered as experience products because the evaluation standards are personalized and emphasize

(12)

sensory feelings of consumption experience, which means it is relatively more difficult to measure product quality before purchase.

2.1.2 Effect of Product Nature

There are increasing amount of researches studying the effect of product nature on consumer choices and consumption behaviors.

Consumers’ evaluation of their purchases for hedonic and utilitarian products differentiate in criterion. Hedonic value provided by hedonic products is more subjective and personal compared to utilitarian one for representing a need for fun and playfulness rather than a need for completing the task (Hirschman and Holbrook, 1982). Based on different criterion, consumer satisfaction is significantly influenced by hedonic and utilitarian values (Ryu et al., 2010), which would eventually impact consumers behavioral intentions. To be specific, consumers would highly value the utilitarian aspects during necessary consumption behavior. When the consumption is pleasure-oriented, consumers will value the hedonic aspects highly.

Hedonic nature of products also can be a more effective stimuli for consumers to spend money compared to utilitarian nature. Dhar and Wertenbroch (2000) suggest that products and brands that are highly valued on the hedonic dimension instead of the utilitarian dimension, in addition, hedonic value is more able to charge consumers for a price premium, and consumers will value the hedonic products higher than utilitarian ones relative to market price. There is former argument propose that higher price leads to increase in price sensitivity of consumers in both traditional stores and

(13)

online stores, only different in amount (Degeratu et al., 2000).

With my research question designed to explore the effectiveness of eWOM on sales for products with different nature, I choose hedonic and utilitarian nature as the main classification metric of the product groups which can be valued from experiential and instrumental perspectives and further explained by related theories including the theory of experience and search products. These reviews related to product nature above are aimed to facilitate the empirical analysis design and implication of related theories on results discussion.

2.2 Perceived Risk

A consumers' perceived risk is a major barrier for consumers who are considering whether to make a purchase. It is difficult for customers to evaluate the products with unknown quality, online shopping is even more risky when compared to traditional transaction with which consumers could assess the products directly in a store. Kim et al. (2008) propose that perceived risk, together with trust and perceived benefit, are strong determinants of a consumer's e-commerce purchase decision, therefore exploring and managing perceived risk play essential roles in e-commerce marketplace as well as in this research.

Wangenheim and Bayón (2004) distinguish two dimensions of perceived risks: functional/financial and psychological/social due to the fingding that there are strong correlations within these two dimensions of perceived risk (Kaplan et al., 1974). Functional risk refers to the uncertainty about a product’s functional performance.

(14)

Financial risk relates to the financial loss consumers pay to fix or replace the product with poor performance. Psychological and social risk are usually fused and treated as one, the former is how the individual perceives himself after made a bad purchase while the latter is used to refer to the consumer's perception of how others will react to his choice. The research question of this paper is more related to the objective dimension, thus I will discuss functional and financial risk rather than psychological and social risk.

Diverse products have different degrees of perceived risk, for example, high-involvement products are typically characterized by a higher level of perceived risk relative to low-involvement products (Deshpande and Hoyer, 1983). Consumers will perceive different levels of functional and financial risks when they purchase for products with different nature (Wangenheim and Bayón, 2004). From the perspective of experience and search products, because more uncertainty about functionality is inherent to experience products, they are associated with higher risk than search products. This is followed by the perspective of hedonic and utilitarian products that functional risk is demonstrated higher for hedonic products than utilitarian products also because it is usually more difficult to assess the performance quality before purchase (Dhar and Wertenbroch, 2000). During consumer purchasing process, perceived risk can be moderated by several externalities.

(15)

2.3 EWOM and its Effect On Sales

2.3.1 Definition

Focusing on studying the effect of word of mouth in e-commerce marketplace, this paper will follow the definition provided by Henning-Thurau et al. (2004), which is the most commonly used one, it refers that eWOM is “any positive or negative statement made by potential, actual, or former customers about a product or company, which is made available to a multitude of people and institutions via the Internet.” The most common eWOM we see on diverse websites are customer reviews and ratings about the products. Consumer review refers to the verbal information provided by consumers to describe the products they have purchased. And consumer rating could be the score or the star number which represent consumers’ satisfaction or dissatisfaction to the product, which is the main study object of this paper.

2.3.2 Metrics of eWOM

EWOM consists of several divergent metrics to measure the impact. Among all the metrics of eWOM, volume and valence are classical dimensions that have been the most frequently studied while the variance dimension is less frequently addressed (De Maeyer, 2012). The metric variance refers to the dispersion of the customer reviews and ratings.

Volume

(16)

2006) about a particular object. The volume of eWOM could increase consumers’ awareness of products by delivering information about how many other people also used the products. The popularity showed by volume of eWOM would thus reduce consumers’ uncertainty and eventually lead to an increase in sales (Chintagunta et al., 2010). This effect on sales can be explained as Godes and Mayzlin (2004) suggested that if consumers find more conversation about a product, they are more likely to be informed and tend to put the product into purchase consideration. This paper treats volume of ratings as the specific metric established to facilitate consumers to investigate the popularity of a product based on the argument that the popularity of a product could be told from the amount of feedback (De Maeyer, 2012).

Valence

The valence of eWOM refers to the numerical value of a customer feedback (De Maeyer, 2012), or in another word, captures the nature of WOM messages (Liu, 2006). It can be the average of numerical ratings given by customers, sometimes it can be positive (e.g. The percentage of five-star ratings) and sometimes being negative (e.g. the percentage of one-star ratings). It can also be captured with volume at the same time such as the number of one-star ratings (Chevalier and Mayzlin, 2006). Valence is a classical dimension that is linked to the sales of a product because positive eWOM enhances consumers’ expected quality, whereas negative eWOM reduces it. The effect of valence can be described as a “persuasive” effect on consumer attitudes (Liu, 2006).

(17)

study the metric of variance because variance of eWOM has been studied sparsely in the past and the results of such studies are inconclusive (De Maeyer, 2012).

2.3.3 Effect of EWOM on Perceived Risk and Product Sales

Research shows that instead of being surrounded by all aspects of perceived risks, consumers look for information about the product initiatively to reduce these risks associated with the buying decision (Guo, 2001). When a consumer perceives more financial risk, he or she will seek information whether the product is price-worthy, and when a consumer perceives more functional risk, he or she will look for information relates to product attributes and performance. Among various sources, customized information and WOM communication influence consumers more than other types of information because consumers tend to believe in and rely on other consumers’ experience (Ha, 2002). Moreover, people who perceive higher risk tend to seek these WOM information more actively compared to those who perceive lower risk (Arndt, 1967). Unlike traditional commerce, e-commerce is now providing effective information for consumers to reduce search cost. Although consumers perceive various risks, and these risks influence consumers’ purchase decisions (Antony et al., 2006), the wealth of information online facilitates consumers to engage in risk reduction behaviors and make optimal purchasing decisions.

Nowadays the effect of eWOM on business performance is studied and applied broadly, for example, the extant studies have examined the effect of eWOM on the success of new product introductions (Clemons et al., 2006). More specifically, in

(18)

different industries such as books (Chen et al., 2004), films (Dellarocas, 2007), online video games (Yang and Mai, 2010), cameras (Zhang et al., 2013) and plenty others across product categories. Later in the meta-analytic review of Babic et al. (2016) which integrates more than 100 studies and find that consumers use eWOM to reduce their perceived risk and make optimal purchase decisions, according to the study results, eWOM is positively correlated with product sales.

Even though both the volume and valence of reviews are important factors for affecting product performance, the extent of their influence tends to be different. For example, in the study of Huang et al. (2009), their research results show that experience products involve greater depth (time per page) and lower breadth (total number of pages) of search than search products in particular. They indicate that the total eWOM from other consumers and multimedia has a greater effect on consumer purchase behavior for experience products than for search products. This should because consumers of search products find it is easy to get information about search attributes even with less eWOM and they are less willing to spend time digging deeper. This conclusion has been extended later by the work of Cui et al. (2012). They find that the valence of eWOM has a stronger effect on search products, while the volume of eWOM is more effective for experience products. Their finding again explains the fact that consumers for search products seek for a explicit score about utilitarian attributes while consumers for experience products prefer to rely on more eWOM for other users’ experimental feelings as reference.

(19)

the effectiveness of eWOM on sales varies for products with utilitarian and hedonic nature.

2.4 Moderators in Relationship between EWOM, Perceived Risk and Sales

2.4.1Trialalibily

The freemium revenue model (Liu et al., 2014) is now one of the most popular pricing strategies in mobile application marketplace, in which users can try the application for free and consumers can purchase for extra features or using the application after the tryout period has been expired. Consumers actually perceive less risk after they have used the product trials. This can be explained by product trialability, the consumers' ability to try the product on a limited basis before making the adoptions decision (Rogers, 1995), can provide a minimize-risk option for consumer to evaluate product attributes and fitness directly, which would have significant effect on consumers beliefs and purchase intentions (Agarwal and Prasad, 1997).

It is stated that product trials affect sales positively by minimizing product uncertainty and perceived risk (Bawa and Shoemaker, 2004). However, according to the finding suggested by You et al. (2015) that the effect of eWOM is greater for products with low trialability than for those with high trialability in the consumer’s decision-making process for purchase. This finding implies that effect of eWOM can be moderated by trialability to a certain extent. Consumers perceive different levels of risks across diverse products and they seek ways to overcome these barriers. This

(20)

paper thus tries to test whether the finding of You et al. (2015) is valid for both hedonic and utilitarian products.

2.4.2 Price

There is another pricing strategy that mobile apps usually apply -- micro-pricing. Nowadays most of the applications available in the market are either free or cost less than two dollar, consumers perceive less financial risk towards these relatively cheap apps.

Not only the perceived risks during consumer purchase decision but also the effectiveness of eWOM will be moderated by lower price of products. Hyrynsalmi et al. (2015) find that when the product is reasonably cheap, customers are more willing to save her or his time and take a risk with the product because consumers may spend their money without expectation of return and thus prefer to save their time on exploring the ratings and reviews. In contrast, consumers heavily rely on eWOM when they perceive high financial risk (Lin and Fang, 2006). To investigate if customers spend more time on studying the product and its word-of-mouth when the price of an application increases, this paper is going to empirically test whether the effectiveness of eWOM increases when prices of products are higher.

Furthermore, consumers react to prices of products differently as suggested by Tellis (1988) that price elasticities varied substantially across product categories, and he suggests that these differences are the result of levels of product risk in different product categories. Therefore this paper is going to test whether the effectiveness of

(21)

eWOM on sales varies for products with hedonic and utilitarian nature, in addition, to control the financial risk consumers perceive, the results will be analyzed between higher and lower price product groups.

The literature reviewed above show that the with/without trialability, higher/lower price are notable moderators in the relationship between eWOM, perceived risk and sales. To test whether their moderating effect varies between hedonic and utilitarian products, this paper is going to control these variables to analyze the results respectively.

3. Hypotheses and Research Framework

3.1 Effectiveness of EWOM on Sales of Products with Different Nature

In the marketplace, word-of-mouth is considered credible and valuable thus influences consumers’ product evaluation and purchase decision processes through clarification and feedback opportunities (Arndt, 1967). My study will focus on investigating electronic word-of-mouth, which saves the face-to-face pressure of traditional word-of-mouth and employs more convenience than traditional information sources by breaking the constrains of spreading speed. It is even more influential than other information sources such as advertising and social medias because it is able to reach the target audiences more efficiently and accurately (Phelps et al., 2004). In the meta-analytic review of Babic et al. (2016), they have systematically examined the positive correlation between eWOM and sales, and they

(22)

demonstrate that consumers rely on eWOM because it reduces their perceived risk and helps them choose the product matched the most, which further affects the product performance. Volume and valence of eWOM are two classical metrics have been studied frequently, they are the major variables volume of ratings and average ratings in my study to explore the effect of effectiveness of eWOM on sales for products with different nature.

My first hypothesis is based on the finding of Babic et al. (2016) that hedonic products, whose value is considered from experiential perspective, have higher functional risk than utilitarian products whose value is considered from instrumental perspective. Since the nature of hedonic products hinders consumers from getting sufficient information about the product quality simply by the description about their attributes, consumers who perceive functional risk thus tend to rely more on eWOM to reduce their uncertainty about the products’ functional performance. Therefore, I predict that eWOM has a greater impact on reducing consumer perceived risk thus affect the sales of hedonic products more than utilitarian products.

Hypothesis 1a: The volume of ratings has a greater impact on sales of hedonic products than on sales of utilitarian products.

Hypothesis 1b: The average ratings has a greater impact on sales of hedonic products than on sales of utilitarian products.

3.2 Moderating Role Played by Trialability

(23)

(Rogers, 1995), which offers consumers opportunities of trying the limited product before purchase. The freemium strategy adopted by companies reduces functional risk by offering consumers opportunities to try their simplified or incomplete products for free so as to stimulate their willingness to buy a full-functional and upgraded products. When the product has higher trialability, consumers can evaluate the quality by experiencing the product themselves instead of heavily rely on other consumer-generated information, thus the functional risk as well as the effectiveness of eWOM are reduced by trailability (You et al., 2015). To re-examine this theory with insight of product nature, I predict that when there is a free trial option, the moderating role of trailability in relationship between eWOM and sales works for both utilitarian product and hedonic product.

Hypothesis 2a: The effectiveness of eWOM on sales for utilitarian products decreases when there is a free trial option.

Hypothesis 2b: The effectiveness of eWOM on sales for hedonic products decreases when there is a free trial option.

3.3 Moderating Role Played by Price

There are insufficient studies address the influencing role of price in the effectiveness of eWOM and Hyrynsalmi et al. (2015) are among the first to test the moderating impact of micro-prices on the relationship between eWOM and sales. They find that consumers are more willing to take the risk and try products without referring the peer-generated information when the price is low enough, which means

(24)

the effectiveness of eWOM is moderated by micro-prices. On the other side, Lin and Fang (2006) suggest that when consumers perceive high financial risk, they will rely more heavily on eWOM. Therefore, to investigate the moderating role of higher prices on the relationship between eWOM and sales, this study predicts that when the product has higher price, the effect of eWOM on sales increases.

Hypothesis 3a: Higher price of product increases the effect of the volume of ratings on sales.

Hypothesis 3b: Higher price of product increases the effect of the average ratings on sales.

When prices increase, will the effectiveness of eWOM on sales for products with different nature both increase or alter towards different directions? To specifically investigate the moderating role of higher price with insight of product nature, I predicts:

Hypothesis 4a: When the prices of products are higher, the effectiveness of eWOM on sales for utilitarian products increases.

Hypothesis 4b: When the prices of products are higher, the effectiveness of eWOM on sales for hedonic products increases.

To better illustrate the conceptual framework of this study, I provide an integrated view in Figure 1.

(25)

Figure 1. An integrative framework on the effect of eWOM on sales for products with different nature

4. Methodology

4.1 Sample and Data Collection

To test my Hypotheses, I collected the data of this study from androidrank.org. (http://www.androidrank.org) which is a service website provides statistics of android applications, app rankings and app rank reviews based on public data of Google Play. This website publishes various ranklists for both applications and application developers. These app ranklists include popular apps which achieved 5,000,000 installs or 10,000 ratings in “All applications”, “Paid apps” and “Free apps” lists. There are 26751 apps showed in the ranklists, 382 of them are paid apps and others are free apps. Although the majority of apps are free apps, consumers seldom feel functional or financial risk towards these apps because their very little cost. Therefore, this study is going to use the “Paid apps” ranklist alone as the resource to construct my data set to better support and examine my hypotheses about the effect of hedonic nature on the relationship between eWOM, perceived risk and sales.

I use a web crawler to scrape and collect the frequently updated data from April 2016 to February 2017, the data set includes following variables of apps: 1) category,

(26)

6) Price and 7) Price history statistics. Because androidrank collects the statistics base on a basic frequency of two months but on random date, thus every app get its data uploaded at different time. My analysis results could be flawed by this flexible collection-intervals. For example, both volume of rating of app A and app B have their data collected in April but on different date. If A’s data is collected on 1 April while B’s data is collected on 10 April, these two numerical value cannot be compared directly because during this collection interval the data of app A has already changed. To make sure that every sample included in the data set has its actual value at specific time, it is necessary to minimize the deviation of data caused by flexible collection-intervals. Therefore, for samples without available data on the selected day (the first day of every two months), I collected the data on (1) the closest date with data before the selected day and (2) the closest date with data after the selected day. Based on the collected data, I calculate the growth or decrease rate of this variable and deduce the value on the selected day.

Since I need to compare two different app groups with hedonic and utilitarian nature, I create a dummy variable to indicate whether the app is hedonic or not. In Google Play, there are 33 application categories in total. In this study, the “Games” and “Entertainment” categories are defined as hedonic categories because all apps included in these categories are enjoyable and appeal to the senses as well as experiences (Dhar and Wertenbroch, 2000), consumers assess them more subjectively.

The left categories “Books and Reference, Business, Comics, Communication, Education, Finance, Health and fitness, Libraries and Demo, Lifestyle, Medical,

(27)

Music and Audio, News and Magazines, Personalization, Photography, Productivity, Shopping, Social, Sports, Tools, Travel and Local, Weather, Art & Design, Auto & Vehicles, Beauty, Dating, Events, Food & Drink, House & Home, Maps & Navigation, Parenting, Video Players” are defined as utilitarian categories. Because most apps included in these categories are assessed more objectively with the criterion whether they are useful and practical. Even for some categories seem “hedonic”, for example, the category of comics, apps of this category serve platforms that collect variety of comics which are hedonic, but these apps themselves are functional and consumers will evaluate these apps with mainly objective criterion like whether they provide more choices and update frequently. Furthermore, the social category of app may also cause disagreement that it should be included in hedonic, but until now most of apps within this category are for communicating and interacting with friends, they serve more instrumental functions rather than pure pleasant, thus this study defines it as utilitarian.

To test the moderating effect of trialability, I check all the 382 paid apps by myself to create a dummy variable to indicate whether the product has a free trial option or not. To investigate the influencing role of price, I classify all paid apps into three groups with their prices, every group has approximate one-third number of paid apps. They are group 1 ($0-$2), group 2 ($2-$4) and group 3 ($4-), which represented lower price product group, medium price product group and higher price product group respectively. This classification aims to exclude the samples with medium prices that might affect the significance of study results during analysis and allow the

(28)

discrepancy between the study results more distinguished by comparing the lower price group and the higher price group.

4.2 Variables Dependent Variable

The dependent variable of this study is the sales of product. From the available data displayed by Androidrank, the install amount of every app can be seen as the sales of this app, because in the meta-analysis of Babic et al. (2016), 11% of the total 96 studies operationalize variable of sales as “number of customers or product units sold”. And in my study, the install amount of apps fits the concept of product sales because it represents the actual demand of these apps, in another word, the number of product units sold. However, due to the collecting constrains related to its own capability, Androidrank actually releases the estimated install amount of every app they tracing based on the statistics from Google Play in a frequency of every two months but with flexible time intervals. And because the mobile application marketplace is placed in highly dynamic changes at any time, the install amount of app remained “estimated” but it is still a valid variable to study. Empirically, this study uses log (Sales) as dependent variable because log transformation converts the relationship between variables into a linear form for empirical estimation and helps to improve the distribution of variables in linear regression analysis. Moreover, with log transformation, the elasticity of independent variable can be explained as the change of percentage amount, which is more reasonable for larger data like sales.

(29)

The independent variables I am going to use in this study include log(Volume of ratings) and average ratings.

According to the previous studies, since the volume of eWOM measures the total amount of eWOM interaction (Liu, 2006), and the popularity showed by volume of eWOM reduces consumers’ perceived risk and eventually lead to an increase in sales (Chintagunta et al., 2010), the volume of eWOM is one of the most essential metrics that has been examined to affect sales. To investigate the effect of product nature on relationship between eWOM, perceived risk and sales, this study uses the cumulative number of ratings which indicates the total number of ratings at a particular time, including past periods and, in some cases, the current period to measure the volume of eWOM. Moreover, to improve the distribution of variables in linear regression analysis and explain the elasticity of the dependent variable with the amount of percentage change of estimated coefficient, this study empirically uses log(Volume of ratings) as independent variable.

Besides the volume of eWOM, another metric of eWOM is also a classical dimension that has been the most frequently studied (De Maeyer, 2012) -- the valence of eWOM, which refers to the numerical value of a customer feedback. To investigate whether the average score rated by consumers for products with different nature will affect consumers’ purchasing behavior, this study uses the data of average rating as an average aggregate measure to indicate the valence of eWOM.

Control Variables

(30)

on sales across all product categories (Babic et al., 2016), I use product nature to classify different groups of products. According to their nature, products can be broadly classified as either hedonic or utilitarian. In addition, I use data of trialability and price group of product to classify subgroups based on all my samples. These control variables help to avoid interference and make the comparison between subgroups more representative, thus the analysis results could be more significant. Hedonic or utilitarian nature of product and trail option of product are dummy variables while price of product has three subgroups which the two subgroups with most difference will be used. Although this study doesn’t focus on investigating product life cycle as Cui et al. (2012), the problem that the metrics of eWOM may be influenced by the previous performance such as cumulative ratings and sales exists. Thus I use data of age of app, the releasing year of the app, as control variable to avoid the influence exerted by this endogeneity problem.

5. Results

5.1 Preliminary Analysis

Table 1: Means, Standard Deviations and Correlations

Variables M SD 1 2 3

1. lon (Sales) _12.7037 _.96038

-2. log (Volume of ratings) _10.0793 _.80215 _.760**

-3. Average ratings _4.409627 _.2863527 _-.156** _.028

-**. Correlation is significant at the 0.01 level (2-tailed). *. Correlation is significant at the 0.05 level (2-tailed).

(31)

paid apps, including 175 hedonic apps and 207 utilitarian apps, and made a total of 2292 observations. Table 1 provides the descriptive statistics, including the mean and standard deviation of the dependent and independent variables as well as results of correlation analysis. The ln (volume of ratings) is positively correlated to ln (sales) with coefficient of 0.76 (p ≤ 0.01), which is correspond to my expectation that volume of ratings and sales vary in the same direction. On the other hand, average ratings is negatively correlated to ln (sales), but in relatively low coefficient of 0.156 (p ≤ 0.01), which means when one of the variables increases, the other will decrease a little. 5.2 Panel Data Analysis

To test Hypothesis 1, I use 1050 observations of 175 hedonic apps and 1242 observations of 207 utilitarian apps collected from April 2016 to February 2017 to linear regression in mixed effects model respectively.

Table 2: Regression Result of Products with Different Nature

Model/fitness _{(Utilitarian)}Model 1 _(Hedonic)Model 2 Chow-test/_t-test

Adjusted R2 _0.644 _0.633 F-value 562.310 451.090 27.94 Sig. F 0.000 0.000 0.000 Control variables β Std. Err. β Std. Err. Price 0.005* _0.003 _0.043*** _0.007 _4.981*** Age 0.095*** _0.013 _0.073*** _0.012 _1.209 eWOM metrics log (Volume of ratings) 1.068*** 0.023 0.795*** 0.020 8.937*** Average ratings -0.749*** _0.068 _-0.526*** _0.056 _2.531*** (Constant) 4.817*** _0.360 _6.545*** _0.327 Note:***_{Significant at ≤ 0.01;}**_{significant at ≤ 0.05;}*_{significant at ≤ 0.1.}

(32)

To test that whether the assumption satisfied, the scatterplot of residual analysis is attached in Appendix Figure 2 and Figure 3, the residual plot did not show significant violations. And there is no multicollinerity problem existed to affect the stability of my models and results.

Table 2 shows that both Model 1 and Model 2 are significant and the adjusted R2

of Model 1 and Model 2 are 0.644 and 0.633, thus both models for utilitarian and hedonic products have predictive validity. In Model 1, the β of log(Volume of ratings) is 1.068 (p ≤ 0.01), which indicates that when the volume of ratings increased by one percent, the sales of utilitarian products is expected to increase by 1.068%. The β of average ratings is -0.749 (p ≤ 0.01), which indicates that when average ratings increased by 1 unit, the sales of utilitarian products is expected to decrease by 74.9%. In Model 2, the β of log(Volume of ratings) is 0.795 (p ≤ 0.01), which indicates that when the volume of ratings increased by one percent, the sales of utilitarian products is expected to increase by 0.795%. The β of average ratings is -0.526 (p ≤ 0.01), which indicates that when average ratings increased by 1 unit, the sales of utilitarian products is expected to decrease by 52.6%.

The Chow test of equality indicates that the two regression models are significantly different in terms of their model structures and parameter coefficients (F = 27.94, p ≤ 0.01). Based on the parameter estimated and the t-test results, log (the volume of ratings) has a stronger positive effect on utilitarian products than on hedonic products (1.068 vs. 0.795, p ≤ 0.01), which indicates that the volume of ratings has a greater impact on sales of utilitarian products, thus rejecting H1a.

(33)

Meanwhile, the average ratings has a stronger negative impact for utilitarian products than for hedonic products (-0.749 vs. -0.526, p ≤ 0.01), which indicates that the average ratings has a greater impact on sales of utilitarian, thus H1b is rejected.

Concluding, hypothesis 1 is completely rejected. My results indicates that eWOM has a greater impact on sales of utilitarian products more than hedonic products.

Before estimating, the scatterplot of residual analysis is attached in Appendix Figure 4-7 to check whether the assumption hold, the residual plot did not show significant violations. And there is no multicollinerity problem existed to affect the stability of my models and results.

To test Hypothesis 2, I classify all my observations into four seperate groups: utilitarian products without trial, utilitarian products with trial, hedonic products without trial and hedonic products with trial. The regression results of utilitarian products are displayed in Table 3 and that of hedonic products are in Table 4.

Table 3: Regression Result of Utilitarian Products with Different Trailability

Model/fitness Model 3 (Without Trial) Model 4 (With Trial) Chow-test/ t-test Adjusted R2 _0.650 _0.651 F-value 159.53 419.65 5.44 Sig. F 0.000 0.000 0.001 Control variables β Std. Err. β Std. Err. Price -0.002 0.003 0.021* _0.008 _2.693** Age 0.156*** _0.022 _0.076*** _0.017 _2.856*** eWOM metrics log (Volume of ratings) 1.061*** 0.042 1.081*** 0.027 0.405 Average ratings -0.935*** _0.130 _-0.666*** _0.080 _1.765** (Constant) 5.573*** _0.611 _4.325*** _0.443 Note:***_{Significant at ≤ 0.01;}**_{significant at ≤ 0.05;}*_{significant at ≤ 0.1.}

(34)

Table 3 shows that Model 3 is significant and can explain 65% of the variance in sales, Model 4 is significant and can explain 65.1% of the variance in sales, both models for utilitarian products without and with trial have predictive validity. In Model 3, the β of log(Volume of ratings) is 1.061 (p ≤ 0.01), which suggests that when the volume of ratings increased by one percent, the sales of utilitarian products without trial is expected to increase by 1.061%. The β of average ratings is -0.935 (p ≤ 0.01), which suggests that when average ratings increased by 1 unit, the sales of utilitarian products without trial is expected to decrease by 93.5%. In Model 4, the β of log(Volume of ratings) is 1.081 (p ≤ 0.01), this shows that when the volume of ratings increased by one percent, the sales of utilitarian products with trial is expected to increase by 1.081%. The β of average ratings is -0.666 (p ≤ 0.01), this shows that when average ratings increased by 1 unit, the sales of utilitarian products with trial is expected to decrease by 66.6%.

The Chow test of equality indicates that the two regression models are significantly different in terms of their model structures and parameter coefficients (F = 5.44, p ≤ 0.01). Based on the parameter estimated and the t-test results, the regression coefficient of log (volume of ratings) does not have significant difference between utilitarian products without and with trials. However, the average ratings has a stronger negative impact for utilitarian products without trial than for utilitarian products with trial (-0.935 vs. -0.666, p ≤ 0.01), thus H2a is partially supported.

(35)

without and with trial have predictive validity. In Model 5, the β of log(Volume of ratings) is 0.810 (p ≤ 0.01), which indicates that when the volume of ratings increased by one percent, the sales of hedonic products without trial is expected to increase by 0.81%. The β of average ratings is -0.382 (p ≤ 0.01), which indicates that when average ratings increased by 1 unit, the sales of utilitarian products without trial is expected to decrease by 38.2%. In Model 6, the β of log(Volume of ratings) is 0.760 (p ≤ 0.01), which indicates that when the volume of ratings increased by one percent, the sales of hedonic products with trial is expected to increase by 0.760%. The β of average ratings is -0.981 (p ≤ 0.01), which indicates that when average ratings increased by 1 unit, the sales of hedonic products with trial is expected to decrease by 98.1%.

The Chow test of equality indicates that the two regression models are significantly different in terms of their model structures and parameter coefficients (F

Table 4: Regression Result of Hedonic Products with Different Trailability

Model/fitness Model 5 (Without Trial) Model 6 (With Trial) Chow-test/ t-test Adjusted R2 _0.666 _0.625 F-value 357.10 140.80 13.01 Sig. F 0.000 0.000 0.000

control variables β Std._Err. β Std._Err.

Price 0.072*** _0.009 _0.010 _0.012 _4.244*** Age 0.125*** _0.016 _0.006 _0.019 _4.807*** eWOM metrics ln (Volume of ratings) 0.810*** 0.025 0.760*** 0.034 1.207 Average ratings -0.382*** _0.064 _-0.981*** _0.111 _4.675*** (Constant) 5.414*** _0.387 _9.408*** _0.585 Note:***_{Significant at ≤ 0.01;}**_{significant at ≤ 0.05;}*_{significant at ≤ 0.1.}

(36)

= 13.01, p ≤ 0.01). Based on the parameter estimated and the t-test results, the regression coefficient of log (volume of ratings) does not have significant difference between hedonic products without and with trials. However, the average ratings has a stronger negative impact for hedonic products with trial than for utilitarian products without trial (-0.981 vs. -0.382, p ≤ 0.01), thus H2b is rejected.

Concluding, hypothesis 2 is mostly rejected, with H2a partially supported that the effect of average ratings on sales for utilitarian products decreases when there is a free trial option.

Before estimating results of H3, the outcomes of residual analysis is attached in Appendix Figure 8 and Figure 9 to test whether the assumption satisfied, the residual plot did not show significant violations. And there is no multicollinerity problem existed to affect the stability of my models and results.

To test Hypothesis 3, I use the 775 observations from lower price group and 658 observations from higher price group to do linear regression with mixed-effect model respectively. The results is displayed in Table 5.

of Model 7 and Model 8 are 0.613 and 0.667, thus both models for utilitarian and hedonic products have predictive validity. In Model 7, the β of log(Volume of ratings) is 0.943 (p ≤ 0.01), which indicates that when the volume of ratings increased by one percent, the sales of products with lower price is expected to increase by 0.943%. The β of average ratings is -0.501 (p ≤ 0.01), which indicates that when average ratings increased by 1 unit, the sales of products with lower price is expected to decrease by

(37)

50.1%. In Model 8, the β of log(Volume of ratings) is 0.900 (p ≤ 0.01), which indicates that when the volume of ratings increased by one percent, the sales of products with higher price is expected to increase by 0.9%. The β of average ratings is -0.720 (p ≤ 0.01), which indicates that when average ratings increased by 1 unit, the sales of products with higher price is expected to decrease by 72%.

Table 5: Regression Result of Products with Different Price

Model/fitness Model 7 (Lower Price) Model 8 (Higher Price) Chow-test/ t-test Adjusted R2 _0.613 _0.667 F-value 307.79 330.16 4.46 Sig. F 0.000 0.000 0.004

Trialability 0.054 0.046 -0.121*** _0.047 _2.669*** Age 0.0673*** _0.015 _0.057*** _0.017 _0.464 eWOM metrics ln (Volume of ratings) 0.943*** 0.028 0.900*** 0.026 1.127 Average ratings -0.501*** _0.068 _-0.720*** _0.092 _1.908** (Constant) 5.063*** _0.415 _6.687*** _0.493 Note:***_{Significant at ≤ 0.01;}**_{significant at ≤ 0.05;}*_{significant at ≤ 0.1.}

The Chow test of equality indicates that the two regression models are significantly different in terms of their model structures and parameter coefficients (F = 4.46, p ≤ 0.01). Based on the parameter estimated and the t-test results, the regression coefficient of log (volume of ratings) does not have significant difference between products with lower price and higher price. Meanwhile, the average ratings has a stronger negative impact for products with higher price than for products with lower price (-0.720 vs. -0.501, p ≤ 0.01), thus H3b is supported.

(38)

increases the effect of the average ratings on sales.

To test Hypothesis 4, for the test that whether the assumption hold, the scatterplot of residual analysis is attached in Appendix Figure 10-13, the residual plot did not show significant violations. And there is no multicollinerity problem existed to affect the stability of my models and results.

I classify all my observations into four seperate groups: utilitarian products without lower price, utilitarian products with higher price, hedonic products without lower price and hedonic products with higher price. The regression results of utilitarian products are displayed in Table 6 and that of hedonic products are in Table 7.

of Model 9 and Model 10 are 0.625 and 0.652, thus both models for utilitarian

Table 6: Regression Result of Utilitarian Products with Different Price

Trialability -0.178*** _0.066 _-0.149** _0.047 _0.310 Age 0.078*** _0.021 _0.073*** _0.017 _0.146 eWOM metrics ln (Volume of ratings) 1.086*** 0.040 1.050*** 0.026 0.600 Average ratings -0.510*** _0.112 _-1.027*** _0.092 _2.864*** (Constant) 3.813*** _0.596 _6.468*** _0.493 Note:***_{Significant at ≤ 0.01;}**_{significant at ≤ 0.05;}*_{significant at ≤ 0.1.}

(39)

β of log(Volume of ratings) is 1.086 (p ≤ 0.01), which indicates that when the volume of ratings increased by one percent, the sales of utilitarian products without lower price is expected to increase by 1.086%. The β of average ratings is -0.510 (p ≤ 0.01), which indicates that when average ratings increased by 1 unit, the sales of utilitarian products with lower price is expected to decrease by 51%. In Model 10, the β of log(Volume of ratings) is 1.050 (p ≤ 0.01), which indicates that when the volume of ratings increased by one percent, the sales of utilitarian products with higher price is expected to increase by 1.05%. The β of average ratings is -1.027 (p ≤ 0.01), which indicates that when average ratings increased by 1 unit, the sales of utilitarian products with higher price is expected to decrease by 102.7%.

The Chow test of equality indicates that the two regression models are significantly different in terms of their model structures and parameter coefficients (F = 2.95, p ≤ 0.01). Based on the parameter estimated and the t-test results, the regression coefficient of log (volume of ratings) does not have significant difference between utilitarian products with lower and higher price. However, the average ratings has a stronger negative impact for utilitarian products with higher price than for utilitarian products with lower price (-1.027 vs. -0.510, p ≤ 0.01), thus H4a is partially supported that when the prices of products are higher, the effectiveness of average ratings on sales for utilitarian products increases.

Table 7 shows that both Model 11 and Model 12 are significant and the adjusted R2 _{of Model 11 and Model 12 are 0.655 and 0.701, thus both models for hedonic}

(40)

the β of log(Volume of ratings) is 0.778 (p ≤ 0.01), which indicates that when the volume of ratings increased by one percent, the sales of hedonic products with lower price is expected to increase by 0.778%. The β of average ratings is -0.528 (p ≤ 0.01), which indicates that when average ratings increased by 1 unit, the sales of utilitarian products with higher price is expected to decrease by 52.8%. In Model 12, the β of log(Volume of ratings) is 0.808 (p ≤ 0.01), which indicates that when the volume of ratings increased by one percent, the sales of hedonic products with higher price is expected to increase by 0.760%. The β of average ratings is -0.498 (p ≤ 0.01), which indicates that when average ratings increased by 1 unit, the sales of hedonic products with higher price is expected to decrease by 98.1%.

Table 7: Regression Result of Hedonic Products with Different Price

Trialability 0.321*** _0.063 _-0.099 _0.070 _4.446*** Age 0.094*** _0.023 _0.073*** _0.023 _0.649 eWOM metrics ln (Volume of ratings) 0.778*** 0.037 0.808*** 0.032 0.601 Average ratings -0.528*** _0.087 _-0.498*** _0.119 _0.204 (Constant) 6.643*** _0.537 _6.595*** _0.630 Note:***_{Significant at ≤ 0.01;}**_{significant at ≤ 0.05;}*_{significant at ≤ 0.1.}

The Chow test of equality indicates that the two regression models are significantly different in terms of their model structures and parameter coefficients (F = 5.30, p ≤ 0.05). However, based on the parameter estimated and the t-test results,

(41)

neither the regression coefficient of log (volume of ratings) nor average ratings has significant difference between hedonic products with lower price and higher price, thus H4b is rejected.

Concluding, hypothesis 4 is mostly rejected, it is supported that when the prices of products are higher, the effectiveness of average ratings on sales for utilitarian products increases.

6. Discussion and Conclusion

The main research question in this study is whether the effect of eWOM on sales varies for products with utilitarian and hedonic nature, and if so, what moderator affects the relationship between eWOM, risk and sales of different product groups. My findings provide empirical evidence that helps to answer these questions.

First take a look at the holistic results of this study, I find that for log (volume of ratings), not only it’s correlation with log (sale) is positive, all the estimated coefficients of log (volume of ratings) in every model are positive. In contrast, for average ratings, it is found negatively correlates with log (sales), and all the estimated coefficients of average ratings in every model are negative. The finding that volume of eWOM is positively related to sales is fully expected because it is consistent with most results from extant studies on effect of the volume of eWOM on sales (Babic et al., 2016). This can be explained by the fact that consumers perceive less risk towards products with larger amount of peer-generated information, in another word, consumers inherently tend to follow others and buy popular products. Interestingly,

(42)

the effect exerted by valence of eWOM on sales is negative in my finding. There are previous researches have found that valence of eWOM has no impact on sales ( Chen et al., 2004; Amblee and Bui, 2011) and Hyrynsalmi et al. (2012) also find a weak , negative correlation between ratings and downloads in mobile market. I here propose two possible explanations for this result. First, this result may attribute to the fact that the average ratings generated by consumers is relatively high, or in another word, consumers may not find these ratings informative because they lack of variability. The mean of average ratings showed in Table 1 is 4.41 and the standard deviation is only 0.29, which means most ratings of apps are higher than 4 and seen as positive ratings. This lack of variability in consumer ratings makes valence of eWOM a less persuasive metric for consumers which is not able to contribute to increase in sales. Second, the valence of eWOM expresses only consumers’ relative sentiment about the product (Babic et al., 2016), therefore, consumers may neglect the ratings scored by other users and loyal to their own tastes to only purchase for products they favor. It is also possible that when the average rating of a product is relatively high, consumers may doubt whether this metric of eWOM has been manipulated by some companies.

To specifically discuss my hypotheses, for H1, previous studies suggested that consumers perceive different levels of functional risks for diverse products with different natures (Wangenheim and Bayón, 2004), they thus rely on eWOM generated by other users differently. Extending these theories with specific product nature, my first hypothesis is fully rejected and indicates that both volume of eWOM and valence of eWOM have greater impact on sales of utilitarian products than on sales of hedonic

(43)

products. This result can be explained by the fact that although consumers of hedonic products perceive higher functional risk (Babic et al., 2016), they are less able to reduce perceived risk with the help of eWOM compared to consumers of utilitarian product. To be specific, it is suggested by Dhar and Wertenbroch (2000) that consumers’ evaluation of their purchases differentiate in criterion between hedonic and utilitarian products. Utilitarian products are evaluated by instrumental attributes, and users tend to be objective and prefer to use a systematic decision-making process. On the other hand, the evaluation of hedonic products is more difficult and less indicative of a product’s quality because users tend to have subjective metrics and express their personalized senses. Therefore, consumers of utilitarian products find it more easy to get practical information about product quality from eWOM and are more likely to decide their purchases base on the peer-generated information.

For the results of my second hypothesis, I have H2a partially supported and H2b rejected. For both utilitarian and hedonic products, the effect of volume of ratings on sales has not been moderated by trialability that a larger amount of eWOM keeps leading to a higher sales. Even they have free trials, consumers are still influenced by the popularity of products to a similar extent when they making purchase decisions. However, for utilitarian products, the effect of average ratings has been moderated by trialability that a higher average rating causes less decrease in sales. This finding is consistent with You et al. (2015) and indicates that when consumers have opportunities to try products before purchase, they care less about the ratings scored by others, and may tend to decide whether to buy them after they try the products. On

(44)

the contrary, for hedonic products, average ratings has a stronger effect on sales when they have free trials. This may because consumers expect the feedback of hedonic products to be more diverse. When a product has free trial, more consumers will be attracted by both trial and formal version of the product, thus it will have more feedback from users. Since the evaluation of hedonic products are subjective and personalized, if consumers find the product has a trial which means has been tried by more people but has a really high average ratings which means most of the ratings are strongly positive, they will doubt the authenticity of this metric and become less willing to buy it.

For H3, I expected that all eWOM metrics will be more effective when the price of products is higher because when consumers perceive more financial risk, they are supposed to rely more heavily on eWOM (Lin and Fang, 2006). H3b is supported that the negative effect of average ratings exerted on sales will increase when the price of product is higher. When consumers are checking information for products with higher prices, if the average ratings of the products are higher, less consumers will buy the products. Meanwhile, the result of H3a indicates that no matter the price of product is low or high, the effectiveness of volume of eWOM does not change significantly. That is to say, when consumers find the product has more eWOM, they are more willing to buy it regardless of the price differences.

For my last hypothesis, I have H4a partially supported and H4b rejected. For utilitarian products, the effect of volume of eWOM on sales has not been moderated by higher prices. However, consumers of utilitarian products are more affected by

(45)

average ratings when the prices of products are higher. When consumers find the average rating of a utilitarian product is higher, they are less likely to buy it. This is also consistent with the expectation that when consumers perceive more financial risk, they will rely more on eWOM (Lin and Fang, 2006). For hedonic products, the effect of both volume and valence of eWOM on sales have not changed significantly when the prices of products are different. This can be explained by the finding of Wakefield and Inman (2013) that consumers are relatively less sensitive to price in products that are perceived as primarily hedonic in nature because they value the experimental gain more over the financial cost. Therefore, the effect of eWOM on sales for hedonic product will not be moderated by price.

To sum up, there are several main findings of this study:

First, from the perspective of metrics of eWOM, volume metric is more effective than valence metric because it’s effectiveness is not moderated by trailability or price but only affected by hedonic nature in this study. Compared to the stability owned by the effectiveness of volume metric, valence metric is more frequently affected by product nature and moderators. This finding re-confirms the theory from Chintagunta et al. (2010) and Babic et al. (2016) that consumers tend to take the popularity of products showed by volume metric of eWOM as main reference to reduce their perceived risk while the nature (positive or negative) of eWOM does not matter so much, thus the sales of products which have more volume of eWOM will be higher.

Second, from the perspective of product nature, this study contributes to extend the existing knowledge and suggests that sales of hedonic products is less affected by

(46)

eWOM compared to that of utilitarian products. This may attribute to the fact that consumers are less able to get indication of product’s quality from hedonic products’ eWOM, thus the sales will not increase if consumers still perceive higher functional risk (Babic et al., 2016). In addition, for utilitarian products, consumer will be affected less by eWOM’s valence metric when they have higher trialability while consumers will be affected more by eWOM’s valence metric when they find the prices of products are higher. For hedonic products, consumers purchase behavior will not be affected by price difference.

These findings also offer interesting managerial implications. From the perspective of eWOM’s metrics, my finding implies that managers should monitor eWOM wisely since the allocation of resources to eWOM management is justified, especially for volume metric. For product managers, they should pay more attention to eWOM management for utilitarian products and even greater for these with higher prices. For hedonic products, managers should lay emphasis on improving the attractiveness of trials by improving their quality so that consumers perceive less functional risk after they tried the trials and have more willing to buy the products.

7. Limitations and Future Research

This study has several limitations that need to be aware by the readers.

First, as is generally for most empirical research, the results of my study are subject to interpretation and are limited to the data available. With androidrank providing statistics of the most popular apps, the number of valid sample is limited.

(47)

This insufficiency in samples makes the classification of subgroups more difficult and is less able to provide a holistic view for my studying. Second, the web crawling can only control the process of data gathering but not the changing information. There will be certainly some missing targets, for example, some mobile apps quit the market or be accidentally filtered out because some regional restrictions. Third, to deal with the dissymmetry of time for data collection, I calculate the growth or decrease rate of specific variable and deduce the value on the first day of every two months. This calculation process inevitably causes some errors that probably affects my research analyses. Last, I defined specific app categories as “hedonic” and “utilitarian” in this study, a more accurate classification standard between these two product nature will ideally make my study more precise.

In future research, the extent of moderators’ effect on relationship between eWOM, perceived risk and product sales could be an interesting topic since this study just explore the difference of effect for different product groups, an extension based on the results of this study is recommended. In addition, to study the influence of price on affecting the effectiveness of eWOM on sales for products with different nature, it is better to have the independent variable of price changes. Since the prices of products change frequently in dynamic marketplace, the observation on consumer reaction towards price changes is more interesting and could be used to extend this study. Moreover, since the study target of this paper is product, future research is recommended to explore the effectiveness of eWOM on sales from the perspective of services. Because Murray and Schlacter (1990) indicate that in traditional commerce,

(48)

there is a higher level of risk associated with the purchase of services than products, primarily because services are intangible, less standardized and seldom equipped with guarantees. It is an interesting topic to investigate whether this theory can be explained with the effect of eWOM.

The effectiveness of eWOM on sales for hedonic and utilitarian products : an empirical study in mobile application marketplace