Consumer reviews and product sales : the impact of product price, size and information vividness

(1)

Consumer reviews and product sales

The impact of product price, size and information vividness

Faculty of Economics and Business

Amsterdam, June 24, 2016

Student: Jens Jansma Student number: 10248269

Master Thesis Master in Business Administration

Specialization: Entrepreneurship and Management in the Creative Industries Supervisor: Dr. B. Kuijken

Academic year: 2015-2016 Semester 2, Block 3

(2)

2

Statement of Originality

This document is written by Jens Jansma 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)

3

Abstract

This study contributes to the theory on (e)WOM and its effect on product sales. More specifically, the objective is to provide insight into the effects of product characteristics that alter the level of perceived risk of the consumer on this relationship. The empirical setting is the US mobile games industry. Data on mobile games was retrieved from the US Google Play Store for statistical analysis. The conditional effect of three product characteristics was

measured: product price, product size, and the vividness of product information. The findings suggest a strong influential role of the volume (total amount) of reviews for the number of downloads. The different levels of product price, size and vividness of product information each had a marginal or insignificant impact on this effect. The valence (average rating) of reviews had a limited effect on the amount of downloads, however, this effect increases as the mobile games become more expensive. Interaction effects of product size and the vividness of information for the valence of reviews were either insignificant or marginal. Game developers should therefore mainly focus on the volume of reviews, and for higher priced games on the valence as well. Further practical and theoretical implications are discussed.

(6)

6

1. Introduction

The number of people owning mobile devices such as smartphones and tablets is growing rapidly. Forecasts predict that one third of the world’s population, or 2.5 billion people, will own a smartphone in 2018, representing a growth of nearly 65 percent in comparison to the number of users in 2014 (eMarketer, 2014 ; Statista, 2016). As a result, the market for mobile applications (apps) has become very attractive. In 2015 the global gross revenue from mobile apps was US$41.1 billion and it is expected to exceed US$100 billion in 2020 (AppAnnie, 2016).

The app stores from which these mobile applications can be downloaded, provide the consumer with different kinds of information about the applications, including reviews. These reviews are written by consumers that have already used a particular application. The sharing of their personal experiences with the product can be considered an electronic form of word of mouth, called eWOM (Chevalier and Mayzlin, 2006). Traditional WOM was described by Arndt, one of the first to research the impact of WOM on consumer behavior, as face-to-face communication about products, brands, or services without commercial intentions from the communicator, as perceived by the recipient of the communication (Arndt, 1967). Stern adds to this that WOM is an unpaid or unsponsored form of conversation about consumption matters, with the goal of ‘conducting the business of life’ (1994, p. 7). When consumers are presented with the experiences of other consumers while browsing through, for example, an app store, they are likely to take the opinions of others in consideration during their own decision making process (Liang, Li, Yang and Wang, 2015). Therefore, the expected

relationship between (e)WOM and sales or financial performance in general is an extensively studied one in marketing literature.

(7)

7 It has been shown to be an influencing factor for product success and adaption by the

consumer (Dellacoras, 2003 ; Godes, et al., 2005 ; Harrison-Walker, 2001 ; Leskovec,

Adamic and Huberman, 2007), as well as a crucial factor in the acquisitions of new customers (Trusov, Bucklin and Pauwels, 2009). In addition, ratings are shown to drive sales, although this effect is only significant for a limited time (Moe and Trusov, 2011 ; Zhu and Zhang, 2006). Zhu and Zhang (2010) provide an overview of empirical studies from 1983 to 2008 on the relationship between online consumer reviews and product sales. The majority of these studies used data from the movies industry and book publishing industry, but still obtained varying results. Some studies found eWOM to have an influencing effect on product sales, whereas others only observed it to be a predictor.

Through their research in the video game industry, Zhu and Zhang (2010) find product and consumer characteristics to be a potential explanation for these different findings.

According to the authors, online WOM is more influential for less popular products, which implies that the reviews are more relevant when there is an absence of other sources of information. In addition, eWOM is of greater influence when users are more experienced on the Internet (Zhu and Zhang, 2010). These findings suggest that the role of eWOM should be especially relevant for the mobile applications industry. Since the online stores immediately offer an overview of consumer reviews, these become a primary source of information. In addition, one could expect that an online store for digital products such as mobile applications is visited by a large amount of consumers with relatively high internet experience. Insights offered by a study of Liang et al. (2015) are in support of this statement. The researchers recently studied online, written reviews of mobile applications in terms of the effect of their content on sales. Liang et al. (2015) discovered that both product quality reviews and service quality reviews significantly affect sales. This research, however, only addressed the top 500 mobile applications for the iOS app store. Therefore, the less popular applications were not

(8)

8 taken into account.

It is not surprising that product and consumer characteristics can make a substantial difference in an online purchase environment. Especially in stores such as those for mobile applications, selling products for digital and online use through an online channel, the presence of consumer reviews and ratings is very strong. As the online world continues to develop, research in this area is very relevant and worthwhile. Now that it is known that consumer behavior is not only affected by (e)WOM, but that product popularity and consumer internet experience are able to set conditions under which this effect is amplified, the question rises whether other factors can further optimize those conditions. The effect of consumer reviews on the online purchasing behavior of consumers is not likely to solely depend on product popularity and internet experience. There are many other factors that could potentially affect this relationship. The literature on consumer behavior considers influences from

psychology, sociology, social anthropology, marketing, and economics (Cheung, Chan and Limayem, 2005).

An important facet in consumer decision making is risk, and the extent to which the consumer is averse to this risk. During purchase decisions the consumer is subject to financial, product performance, physical, social, and time risk (Jiuan Tan, 1999). Although consumers differ in the extent to which they are risk averse, in general all consumers are risk averse (Yan, 2010). This is true for both online and offline consumers. In addition, consumers will always make an effort to minimize the risk before purchase, for example by collecting additional information (Yan, 2010 ; Bao, Zhou and Su, 2003). This suggests that product characteristics that can alter the level of perceived risk, may affect the relationship between online reviews and product sales. Therefore, the research question addressed in this study is:

(9)

9

‘How do product characteristics that alter the level of risk for the consumer affect the relationship between word of mouth and product sales?’

The research area for this question is the mobile games industry. Common characteristics that differ among mobile games are the price and the size of the product. In addition, some games include a game trailer to the product information, whereas others do not. Each of these three characteristics can alter the perceived risk for the consumer. For price this is due to an

increase of financial risk, for size to an increase of the risk of losing smartphone functionality, and for the vividness of information to a reduction in the risk of having incomplete product information. The influence of each of these characteristics on the relationship between eWOM and product sales will be tested. In order to uncover these relationships, data from the Google Play Store of the US will be retrieved and analyzed.

The extant literature on (e)WOM has mainly focused on its effects, and to a lesser extent on its antecedents and motivations for engaging in it. The conditions under which (e)WOM is more or less effective are somewhat overlooked. Overall, this research aims to provide a deeper understanding of the effect of (e)WOM on product sales and the different elements that potentially affect this relationship. The outcomes of this study can contribute to the theory on the relationships between (e)WOM and sales, in particular for the creative industries. Specifically, it can provide future research directions and insights in the potential drivers of eWOM influence in terms of product characteristics and the information provided to the consumer. Since the relationship between eWOM and sales has been widely recognized already, especially this latter part is very relevant for practice. Internet marketers could be provided with insights that specify under what circumstances eWOM is especially relevant. In addition, they could improve their understanding of how to manage the information they provide about a product. In this way marketers can optimize their online strategy, which could

(10)

10 help them to increase product sales.

This research paper will be structured in different sections. First of all, the literature review will establish a clear understanding of (e)WOM and its relation to sales and product success. In addition, it will address different theories on which hypotheses will be based that predict the influence of different product characteristics and information on sales and eWOM. Hereafter, the data and method section will provide information about the dataset and the method of the analysis. Next, in the results section the analysis will be discussed and the outcomes will be explained. In the discussion section, the findings will be critically evaluated and the research question will be answered. Finally, the findings and contributions of the research will be summarized in the conclusion.

(11)

11

2. Literature Review

This section will provide a detailed background of the relationship between WOM and sales. In addition, an elaboration of the product characteristics price, size and the vividness of product information will follow. This will be the basis for the formation of the hypotheses.

2.1 Word of Mouth

2.1.1 Traditional and Electronic Word of Mouth

As discussed in the introduction, traditional word of mouth is defined as unpaid or sponsored person to person communication about a product, brand or service, without commercial intentions (Arndt, 1967 ; Stern 1994 ; Buttle, 2011). The information that is exchanged in these conversations is not biased due to the involvement of marketing. With the introduction of the Internet and its continuous development, new ways for consumers to acquire such unbiased product information have emerged. The gathering and sharing of information, opinions, and experiences related to consumption online is considered a form of electronic word of mouth. There are some important aspects that distinguish eWOM from traditional WOM. The definition of eWOM by Hennig-Thurau, Gwinner, Walsh and Gremler (2004, p.32) clearly exposes these: ‘’eWOM communication 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.’’ As opposed to traditional WOM, eWOM is available to a large, unspecified audience for an indefinite period of time. Therefore, it could be expected that eWOM can be of as much or even greater influence than traditional WOM (Hennig-Thurau et al., 2004). Traditional WOM is known to be more effective than advertising and personal selling and has an important influence on the consumer’s product choice (Gruen, Osmonbekov and Czaplewski, 2006 ; Richins, 1983).

(12)

12 Bickart and Schindler (2001) confirm that eWOM is likely to be more credible and relevant for customers than online content produced by marketers. Together with the finding that the motivations for eWOM are similar to WOM, this suggests that both the online and offline variant behave very similarly (Gruen et al., 206 ; Hennig-Thurau et al., 2004).

2.1.2 Reviews and Performance

As mentioned, the relationship between (e)WOM and different performance indicators is an extensively studied one. Zhu and Zhang (2010) examine several of these in their overview of recent studies. In this research, the focus will be on the relationship between eWOM and performance in terms of reviews and sales. Current research on this relationship distinguishes between two important characteristics of WOM: volume and valence. Volume is the total amount of WOM about one specific product, service, or company. Valence represents the evaluative side of WOM and shows whether the content is positive, negative, mixed, or

neutral (Mahajan, Muller and Erin, 1984 ; Neelamegham and Jain, 1999). In the case of online reviews, volume would be represented by the total number of reviews. Valence could be represented by both the content of the reviews and the rating provided by the reviewer.

The effects of the volume and valence of online reviews are often studied in relation to different product types. This is usually done on the basis of the search/experience

classification paradigm, which offers an effective way for assessing the potential of the

Internet as a marketing channel (Weathers, Sharma and Wood, 2007). The distinction between both product types is made on the basis of the ability to assess the product quality before purchase. Search goods are relatively low in physiological and emotional involvement, which enables consumers to make their purchase decision on the basis of simple, verbal descriptions, containing specific product attributes. Experience goods, on the other hand, are more

challenging to evaluate, as the consumer needs to use his or her senses, or experience the product, to assess the quality of the product (Weathers et al., 2007 ; Rosa and Malter, 2003 ;

(13)

13 Cui, Lui, and Guo, 2012).

The volume of WOM is generally found to be a factor that significantly influences product sales (Chen et al., 2004 ; Liu, 2006 ; Yang, Kim, Amblee, and Jeong, 2012).

Especially experience goods tend to be influenced by the volume of reviews. A large amount of reviews indicates that the product is popular and increases the awareness. Especially in the early stage of the product’s life cycle the effect on sales is large (Cui et al., 2012). Valence, on the other hand, is more influential for the sales of search products, although studies on the valence of reviews are quite inconsistent in their findings. The difference in the effects for both product types is explained to be due to the inability to ‘feel’ experience products through an online channel. Although other users share their evaluations, these are usually perceived as telling more about the personality of the reviewer than the quality of the product. This is due to the nature of the consumption of experience goods, which is an interaction between the customer’s personal characteristics and the experience good (Yang and Mai, 2010). Therefore, the volume of the reviews is more influential, as it is used as a proxy for popularity, which in turn is an indicator of product quality (Cui et al., 2012).

A different theory, however, states that others’ opinions are important particularly for experience goods, because they offer an indirect experience that can help to overcome uncertainty about the product quality (Park and Lee, 2007). Again, this is argued to be especially relevant for the online environment, which is unable to express sensory attributes. The experiences shared by other consumers are suggested to be able to make experiential attributes transform to search attributes (Klein, 1998). Yang and Mai (2010), however, find that online reviews are not capable of completely succeeding in this transformation due to a lack of trust in the experience attributes of the reviews. Consumers are confident about the search attributes in the reviews, but have less trust in low-level experience attributes, and little or no trust in high-level experience attributes. It is probably due to this, that Yang and Mai

(14)

14 (2010) also find that a large installed base makes consumers trust a product’s quality. As is true for a large volume of reviews, the installed base serves as an indicator for popularity. Furthermore, Yang and Mai (2010) argue that this finding might offer an additional explanation for the general observation of many studies that the volume of reviews has an important impact on product sales, whereas the valence of the reviews has not. A large installed base, which usually also results in a high volume of reviews, can make the valence of reviews insignificant.

2.1.3 Positive vs. Negative WOM

In addition to the different effects of the volume and valence of reviews, research has also examined the relative strength of positive and negative reviews. Positive reviews indicate that a product is of high quality, whereas negative reviews present reasons not to buy a product. Therefore, it is important to make a distinction between positive and negative reviews as well, rather than to only focus on valence in general .

Research on this aspect is very univocal, with the overall conclusion that negative WOM is more powerful than positive WOM (e.g. Arndt, 1967 ; Yang and Mai, 2010 ; Cui et al., 2012 ; Park and Lee, 2012). Negative WOM is found to have a relatively strong effect on consumers’ attitude towards brands, as well as on their purchase decisions (Arndt, 1967 ; Richins, 1983 ; Park and Lee, 2012). Negative, traditional WOM is reported to be spread to twice as fast compared to positive WOM (Buttle, 2011). Research by Desatnick (1987) indicated that customers dissatisfied with the service provided to them, would not repurchase. Furthermore, these customers spread their negative experiences with 9 people or more. About 13% even shared it with more than 20 people.

Negative information is found to attract more attention in general and it is examined more closely than positive information (Homer and Yoon, 1992). It is generally accepted that this is due to the ‘negativity bias’, which is a ‘psychological tendency for people to give

(15)

15 greater diagnostic weight to negative information in making evaluations’ (Cui et al., 2012, p.46). The negativity bias is said to occur due to the relatively large amount of positive stimuli in an individual’s social environment, which causes negative cues to attract more attention, as they are against standards.

2.1.4 Product and Consumer Characteristics in relation to WOM

The most striking part of the inconsistent findings in the research of the impact of valence of WOM on product sales is that studies even report different outcomes within a particular industry. For example, this is true for research in the motion picture industry (Liu, 2006 ; Duan, Gu and Whinston, 2008), and for research on online book retailers (Chen et al., 2004 ; Chevalier and Mayzlin, 2006). Zhu and Zhang (2010), also recognized the different outcomes of research on the effects of WOM and were the first to empirically test the differential effects of WOM on products within the same product category. More specifically, the authors

examine how product characteristics and consumer characteristics affect the relationship between online consumer reviews and product sales. Product popularity is the product characteristic of which Zhu and Zhang (2010) researched the impact on the relationship between online consumer reviews and product sales. The authors measured product popularity on the basis of sales in the most relevant stage of the product lifecycle. The consumer

characteristic that the authors researched was the Internet experience of the consumer. Through their empirical examinations, Zhu and Zhang (2010) found that reviews are more influential for less popular products. They argue that this is due to a scarcity of other sources of information, which makes the role of reviews more important. Consumers want to have quality information in order to reduce uncertainty about the product before purchase and this information is more difficult to obtain for less popular products (Bolton, Katok and Ockenfels, 2004). Furthermore, popularity in itself already signals quality and represents social cues, which both reduce risk (Caminal and Vives, 1996 ; Zhu and Zhang, 2010 ;

(16)

16 Bearden, Netemeyer and Teel 1989).

In addition, online reviews are more effective when consumers are more experienced on the Internet (Zhu and Zhang, 2010). These consumers are more likely to consult the internet to acquire product information and have less trouble with accessing the information than less experienced consumers. Consumers with higher Internet experience therefore use the Internet as their primary source of information and have greater confidence in the Internet (Zhu and Zhang, 2010).

These two findings are interesting for the research in the mobile games industry. Zhu and Zhang studied the effects of eWOM on the sales of offline, tangible products, sold by retail chains. Their findings suggest that reviews would be of greater influence on products that are sold and used online, and this influence will increase over time. This is not only due to a growing number of Internet users with growing experience, but also due to the improved accessibility of online information. The Google Play Store is a good example of an online operating store that only sells digital products that must be downloaded. The Play Store presents the consumer with an overview of the total number of reviews and the distribution of ratings for each individual application. The consumer can choose to sort reviews on the basis of relevance, rating and date. Such developments make reviews more likely to be among the primary sources of information. In addition, it can enable even inexperienced Internet users to be influenced by online reviews, as better decisions can be made with substantially less effort (Zhu and Zhang, 2010 ; Häubl and Trifts, 2000).

In sum, consumers are presented with ratings and reviews about a mobile application in the Google Play Store, whether they feel that there is a lack of other information or not. In addition, the information becomes more accessible, even for inexperienced users, and Internet experience is also growing over time. Therefore, it is assumed that the relationship between online reviews and the sales of mobile games is highly relevant. Furthermore, is has been

(17)

17 shown that product characteristics can alter the effect of reviews. When targeted at popular products and consumers with higher Internet experience, both the volume and valence of reviews have an influencing role. This leads to the study of the potential influencing effect other product characteristics, that will be addressed in the next paragraph.

2.2 The Effects of Product Price, Size, and Information Vividness

This paragraph will address theory around the price and size of products and the vividness of product information. Each of these subjects are expected to have an impact on the influence of online consumer reviews on product sales. These expectations will result in the formulation of three hypotheses.

2.2.1 Product Price

The price of a product has an important signaling function for the product quality as perceived by the consumer. It is one of multiple potentially useful indicators, that is always available to the consumer (Zeithaml, 1988 ; Bagwell and Riordan, 1991). Usually, high quality goods are brought to the market with a high initial price, that will be lowered over time. This is done in order to first signal high quality through the initial price. Once the product quality is better known due to the diffusion of the product, it becomes easier for a firm to communicate its high quality attributes and the product price can be lowered (Bagwell and Riordan, 1991). The diffusion of the information about the product quality results from the adaption of the product by the customer. Other consumers can acquire such information about a product by consulting reviews and ratings of previous customers. It can be expected that as the price increases, consumer will more critically evaluate the product before purchase. This is due to the general tendency of consumer to be risk averse, and as the price level increases, the financial risk of making an inaccurate judgement increases (Rao and Monroe, 1989 ; Yan, 2010). Risk aversion motivates consumers to undertake actions in order to reduce their risk during the

(18)

18 purchase decision process. One of the actions that consumers can take is to collect additional information (Yan, 2010 ; Bao et al., 2003). Online reviews and ratings could be important sources of this information. In fact, risk reduction is found to be the strongest driver for consumer to engage in WOM communication (Hennig-Thurau and Walsh, 2003).

In addition, many mobile games are offered for free in the mobile application stores. These games generate income through, for example, advertising or the possibility to purchase additional functionalities within the application. Experience goods, are found to best

promoted through offering a direct experience (Yang and Mai, 2010). For a mobile game, this could be a free trial or free game for which additional functionalities can be bought.

Whenever, such a free application does no fully satisfy the customer’s needs, the customer has lost time, but has, at least not directly, lost any money. Obviously, this is different for mobile games that must be purchased. In order to reduce the risk of losing money on such an application, the consumer can try to reduce the uncertainty about the quality of the application through consulting online reviews and ratings. As mentioned before, these reviews could serve as an indirect experience, transforming experience attributes towards search attributes (Park and Lee, 2007 ; Klein, 1998).

Taking the above arguments into account, consumers are likely to more critically evaluate their purchase decision by assessing the product quality more carefully. Therefore, it is expected that consumers rely more on reviews and ratings, strengthening the relationship between reviews and sales. This is formulated in the following hypotheses:

Hypothesis 1a: As a product’s price increases, the strength of the relationship between the volume of online reviews and product sales increases.

Hypothesis 1b: As a product’s price increases, the strength of the relationship between the valence of online reviews and product sales increases.

(19)

19

2.2.2 Product Size

Another product characteristic that is expected to alter the relation between online reviews and sales, is product size. Mobile phones only have a limited amount of storage space and downloading an applications takes up a part of it. The larger a particular application is, the less storage space is left for other applications and personal files and media. In addition, the performance of the smartphone might decrease in terms of a decrease in speed and battery life (Zhong et al., 2014). Consumers of mobile games applications face the trade-off between maintaining storage space and acquiring a particular application, or have to choose between different applications. Facing such a trade-off means that the consumer might not be able to have everything he or she wants (Campbell and Kelly, 1994). The theory of loss-aversion goes beyond this trade-off theory. The loss-aversion theory states that consumers would rather not lose something they currently own, than they would like to gain something they currently don’t own (Camerer and Loewenstein, 2011). In other words, the consumer dislikes losing something more than he or she likes to gain something. In effect, it should be less likely for a consumer to make an uninformed decision to acquire an application that is relatively large in size.

Following the same line of reasoning as for hypothesis 1, the consumer is expected to more critically evaluate the decision to download an application, as the size of the file

increases. Therefore, the consumer is expected to rely more heavily on reviews and ratings in order to reduce the uncertainty of the product quality. These arguments lead to the following hypotheses:

Hypothesis 2a: As the size of a mobile application increases, the strength of the relationship between the volume of online reviews and product sales increases.

(20)

20

Hypothesis 2b: As the size of a mobile application increases, the strength of the relationship between the valence of online reviews and product sales increases.

2.2.3 The Vividness of Product Information

Assessing the quality of an experience good is most accurately done by the consumer actually experiencing it (Yang and Mai, 2010). This enables the consumer to evaluate the quality of the product on the basis of his or her own perceptions, rather than on those of others. Such direct experiences can be provided through, for example, free trials, trailers, and pictures (Yang and Mai, 2010). In order to evaluate experience goods, consumers need to use their senses. The vividness of the product information, relates to the level of sensory information that is provided (Weathers et al., 2007). Vivid information can assist consumers in recalling the performance of previously used, similar products. In addition, visual images can help to overcome a lack of ‘touch’ information and help consumer to trust their judgements about the product’s quality (Shelder and Manis, 1986 ; Peck and Childers, 2003). Furthermore, it increases involvement and creates stronger, and more positive attitudes (Griffith and Krampf, 1999 ; Coyle and Thorson, 2001). Empirical research on the use of product pictures confirms that providing such pictures is an effective means for lowering the consumer’s uncertainty about product quality. Therefore, retailers that sell experience goods online are advised to increase the vividness of product information (Weathers et al., 2007).

For most mobile applications in the online store, images are provided. However, for only a part of the applications a video, or trailer, is provided as well. For movies, trailers are the most important tool for generating audience interest prior to their release. In addition, trailers are the most influential promotional medium for setting consumer expectations (Finsterwalder, Kuppelwierser and de Villiers, 2012 ; Hixson, 2006). Although a movie is a product quite different from a mobile game, a trailer can offer an indication of what the actual product is like when it is watched or played. In addition, a movie clip can be more interactive

(21)

21 and representative of the actual product than product pictures alone. As it offers moving images and sound, the vividness of the information is higher than for product pictures.

In sum, product information that includes a trailer, or video, is likely to have a higher level of vividness. The consumer is presented with more, and richer information that

resembles the actual product experience. Therefore, the uncertainty about the product quality and the associated risk for the consumer is reduced. It is expected that this also reduces the need for the consumer to consult product reviews and ratings. These expectations lead to the following hypotheses:

H3a: Whenever a trailer is included in the product information, the strength of the relationship between the volume of online reviews and product sales diminishes. H3b: Whenever a trailer is included in the product information, the strength of the relationship between the valence of online reviews and product sales diminishes.

All above described hypotheses are visually presented in figure 1. The next section will discuss the data and method that are used to empirically test the proposed hypotheses.

(22)

22

3. Data and Method

This section will provide information about the design of the empirical research and the data collection. The data source and the way in which the data was collected will be discussed, as well as the different variables and their respective measures. Finally, the statistical method for testing the data will be presented.

3.1 Research design

The empirical setting of this research is the online environment of mobile applications stores for mobile games. The main objective is to test whether the positive relationship between product sales and the volume and valence of reviews that are displayed in such online stores is moderated by the product’s price, size and the presence or absence of a game trailer. In order to accomplish this, and to eventually provide an answer to the research question, an

explanatory research design is used. By collecting measurable, quantifiable data from a real life, online setting, the nature of actual relationships between the variables can be statistically tested (Saunders and Lewis, 2012).

3.2 Sample

The relationships that are intended to be measured are all based on the behavior of consumers of mobile games. In a sense, all these consumers could be regarded to as the population under investigation. However, the data that will be used to test this behavior consists of statistics about individual, mobile games. Therefore, the population of the data that will be tested, is the total of all mobile games that are available in online stores. The mobile devices on which these games can be downloaded, run on Android, iOS, Windows, or Blackberry operating systems. Of these systems, Android and iOS together have a market share of over 95% worldwide, of which around 83% was accounted for by Android in 2015 (Topmobiletrends,

(23)

23 2015 ; IDC, 2015). The largest stores for mobile applications are the Google Play Store (for Android systems) and Apple’s App Store (for iOS systems), currently offering 2.2 million and 2.0 million applications respectively (Statista, 2016). A number of these applications are offered both on Android and iOS systems. In addition, the applications offered in the application stores differ per country or region.

The sample that is used in this research consists of all the games that were offered in the Google Play Store in the US at the time of collection. This sample should be

representative for the entire population of mobile games offered in the US, as the Google Play Store is the largest provider for the largest amount of smartphone users. The data connected to this sample was acquired by a web scraping tool that was developed by iQU, which is a Dutch marketing company for mobile and online games. A web scraping tool can collect information from websites automatically. The collected data can all be found on the webpage of an

individual mobile game in the Play Store and is freely available.

The main reasons for choosing this sample is due to the availability of the dataset for this research. In addition, the dataset is up-to-date and contains all games that are offered by the largest mobile application store. Therefore, the dataset is likely to be of a sufficient level of credibility, as well as representative for the population.

3.3 Variables

The data that were collected by iQU include all specifics that are provided on the webpage of an individual mobile game. Not all these data are relevant for the variables in this research. Therefore, an overview of the different variables and their corresponding measures that are used in this research is provided here:

(24)

24 Dependent variable:

Product sales - This variable is represented by the number of installations. The volume of

these installations is measured on the basis of the minimal and maximal install range borders. These provide the minimal number of installs and the maximum number of installs for a particular mobile game. The values that these range borders can take are: 0, 1, 5, 10, 50, 100, 500, 1000, 5000, 10000, 50000, 100000, …, 500000000. The lower install range border is always one category under the higher install range border. The higher install range border does not take any value when the lower install range border is zero. Since the actual number of installations are somewhere in between the lower and upper range border, the mean of these two borders is computed as the variable that represents the amount product sales.

Independent variables:

Volume of reviews - The first independent variables, the volume of reviews, is measured on

the basis of the total amount of reviews. The total amount of reviews can be any discrete value.

Valence of reviews - The second independent variable, the valence of reviews, is measured by

the average rating given to an individual game. Each user can give a rating of one to five stars, the average of all these rating is provided for each game as a number rounded to one decimal place.

Moderator variables:

Product price - The first moderating variable, the price of the product, is measured in amounts

of US dollars. The price of free games is labeled as zero. In addition, each price is specified in cents.

(25)

25

Product size - The second moderating variable, the size of the product, is measured in terms

of megabytes. The size in megabytes is rounded off to a discrete value without any decimal places. Games that have a size of less than one megabyte are rounded off to one. Games that are specified in gigabytes are multiplied by thousand in order to express those in megabytes.

Vividness - The final moderating variable, the vividness of the product information, is

measured on the basis of whether the product information includes a game trailer. In order to establish a distinction between a category of games with a trailer and one without a trailer, a coding variable is created. A ‘0’ is assigned to a game without a trailer, whereas a ‘1’ is assigned to a game with a trailer.

Control variables:

Developer size - There are many different developers that offer one or more mobile games in

the Play Store. Large developers have produced multiple games, whereas others offer only one. Large developers might be better known than the smaller developers among consumers, for example for quality. Therefore, the size of a developer might alter the likeliness of

consumer to buy a mobile game. In order to control for this, a dummy coding variable is made that indicates whether a developer has produced one (indicated by a ‘0’) or multiple (indicated by a ‘1’) mobile games.

Genre popularity - Another potential explanation for a higher sales volume is the popularity

of certain genres in which mobile games are classified. Well over one third of the mobile games is represented by only four genres: action, arcade, casual, and puzzle. Each of these genres accounts for at least 7% of the total amount of games. In order to control for the popularity of the genres, two categories are created with dummy coding. The category presented by ‘0’ represents low-popularity games, whereas a value of ‘1’ represents high-popularity games.

(26)

26

Category Spanning - The final variable that is controlled for is category spanning. In this case,

category spanning refers to the presence of one game in multiple genres. Some games can only be found under one genre in the Google Play Store, whereas others are assigned to two genres. Games that are present in two categories are likely to have greater visibility. This might impact the amount of product sales. Therefore, a control variable for category spanning is created through dummy coding. A value of ‘0’ is assigned to games that only appear in one genre, whereas a ‘1’ is assigned to games that appear in two genres.

3.4 Data Analysis

The actual analysis of the data will be carried out with IBM’s statistical computing software ‘SPSS’, version 22. The results section will explain in detail which statistical tests were used.

3.5 Strengths and Limitations

The method used in this research has some important strengths and weaknesses. First of all, the reliability of the data is very high. Due to the fact that the measures are real life, objective numbers from the complete range of mobile games offered in the largest mobile applications store, human errors in interpreting this data are excluded and the sample size is very large. Therefore, the data collection method and the analysis procedures are likely to produce consistent findings (Saunders and Lewis, 2012).

The validity of the data is somewhat different. For the internal validity applies that the large sample size is likely to give a good representation of the population and the data

collection is not subject to any disturbing factors. In addition, the criterion validity is close to ideal, as the measures are real world observations, given that Google provides actual numbers (Field, 2013). However, cross-sectional research methods are limited in their ability to

uncover causal relationships. This is due to the ambiguity of the causal direction, which might be the opposite of the hypothesized relationship (Saunders and Lewis, 2012). For example, a

(27)

27 higher price could signal better quality to the consumer, which increases sales and the number of reviews.

The external validity refers to the generalizability of the findings to other research settings. For the US, the generalizability is likely to be quite high. This, again, is due to the large sample size. However, it could be questioned whether the behavior of the users of Apple smartphones is different from that of users of Android phones. In addition, the implications might not be valid for other regions or countries due to cultural differences. People in other regions might respond differently and the mobile games that are offered might be significantly different. The cultural similarities and difference of countries can be measured with the

cultural dimensions that were introduced by Geert Hofstede (Hofstede, 1983). However, in order to be sure of the results in other countries, the study should be replicated there.

Finally, in order to more clearly understand the effects, the actual number of installs would have been more accurate than the intervals in which these are measured now. The same applies to the variables that are controlled for, which are each divided into one of two possible groups. A more accurate representation would have provided with continuous variables.

(28)

28

4. Results

This section will present descriptive statistics of the data, comment on the distribution of the data, explain the statistical methods used for testing the hypothesis, and provide the results of those tests.

4.1 Descriptive statistics and data distribution

Product sales – The lower install range border represents the number of product sales. The

total amount of observations is N = 21,339, there were no missing values and the mean is M = 862,807. A histogram of the frequency distribution shows that the distribution is multimodal. In order to test for normality, a Kolmogorov-Smirnov test was used. The lower install range border score is D(21,339) = 0.450, p < .001, meaning that the scores are significantly non-normal (Field, 2013).

Volume of reviews – The total amount of observations for the volume of reviews also is N =

21,339. There were no missing values and the mean is M = 34,969. The graph of the scores shows a positively skewed distribution. The Kolmogorov-Smirnov test indicates that the scores for the volume of reviews is D(21,339) = 0.456, p < .001, meaning that the scores are significantly non-normal (Field, 2013).

Valence of reviews – The average rating per game has a total amount of observations of N =

21,339. Of these, 1434 (6.7%) values were missing due to the fact that not a single rating was given to these games. The mean of the average rating is M = 4,11, showing that the average rating of all games is fairly high. The Kolmogorov-Smirnov test reveals that , D(19,905) = 0.084, p < .001. Therefore, it can be concluded that the scores are significantly non-normal (Field, 2013).

(29)

29

Product price – For product price, there are two categories: paid and free games. The total

amount of observations is N = 21,339, and no missing values were detected. The mean score, M = 0.27, indicates that 27% of all games has to be paid for. The Kolmogorov-Smirnov test indicates that D(21,339) = 0.459, p < .001. Therefore, the scores are significantly non-normally distributed (Field, 2013)

Product size – The product size, indicated in megabytes, has a total number of observations of

N = 21,399, with 1233 (5.8%) missing values. These values are missing, because some games do not specify the size of the file due to differing sizes for different mobile devices. The mean score is M=45,41. Plotting a graph shows a positive skew. The normality test of Kolmogorov-Smirnov tells that the scores are significantly non-normal, with D(20,106) = 0.352, p < .001.

Vividness – There are two values for the vividness of product information: with video and

without video. The total number of observations is N = 21,399, with no missing values. The mean score is M = 0.33, meaning that 33% of all games have product information that includes a video. The Kolmogorov-Smirnov test reveals that the scores are significantly non-normal, with D(21,399) = 0.427, p < .001.

In conclusion, the scores of all variables are found to be non-normally distributed. It is, however, suggested that large samples reduce the risk of skewness and kurtosis making a substantive difference in the analysis. Tabachnick, Fidell and Osterlind (2001) report that this is true for samples with over 200 cases. In addition, Field (2013) explains that normality tests are more likely to result in significant findings for larger samples. However, this should not be the case due to what is known as the central limit theorem. The central limit theorem states that as the size of the sample increases, the assumption of normality becomes less relevant, because the sample will be normal despite of what the population looks like. Despite these theories and the large sample size, the severity of the non-normality calls for a transformation

(30)

30 of the data. Therefore, the natural logarithm is used to standardize the distribution for each of the (non-dummy coded) variables.

4.2 Correlations

Table 4.1 shows the correlation matrix, displaying all correlations coefficients for the all combinations of variables. Many of the variables are significantly correlated, but the most outstanding is the correlation coefficient of product sales and the volume of reviews, which approaches a perfect, positive relation (r = 0.95, p < 0.01). The valence of reviews on the other hand is, although weakly, negatively correlated with product sales (r = -0.18, p < 0.01), which is against expectations. This suggests that the volume of reviews is much more relevant for product sales than the valence of reviews. In addition, product sales has a weak negative correlation with price (r = -0.18, p < 0.01), which means that as the product price increases, the amount of sales tends to decrease. Product size and the vividness of information are weakly, but positively related to product sales. Finally, the control variables are all significantly related with product sales. Product popularity is negatively and weakly correlated with sales, whereas category spanning and developer size are positively related. Only the latter has a meaningful correlation strength (r = 0.24, p < 0.01).

The volume of reviews has noteworthy positive levels of correlation with developer size, product size, and information vividness. This suggest that higher volumes of reviews are associated with ‘larger’ developers, larger product size, and product information that includes a video. The valence of reviews, on the other hand, does not correlate strongly with any of the variables.

The independent variables product price, size and vividness of product information are all positively correlated. Higher values of any of these variables go along with higher values of each of the other variables. Finally, the independent variables share no noteworthy

(31)

Table 4.1: Means, Standard Deviations, Correlations

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

Variables M SD 1 2 3 4 5 6 7 8 9 1. Genre popularity 0.36 0.48 - 2. No. of genres .07 0.25 -0.48** - 3. Developer size 0.67 0.47 -0.49** 0.08** - 4. Product sales 9.18 4.43 -0.13** 0.03** 0.24** - 5. Reviews 5.57 3.65 -0.10** -0.00 0.21** 0.95** - 6. Ratings 1.6 0.13 0.09** -0.00 -0.06** -0.18** -0.07** - 7. Price 0.29 0.53 0.07** 0.15** 0.05** -0.18** -0.12** 0.05** - 8. Size 3.15 1.07 -0.02* 0.09** 0.16** 0.15** 0.21** 0.04** 0.21** - 9. Vividness 0.33 0.47 0.03* 0.07** 0.07** 0.24** 0.29** 0.06** 0.24** 0.28** -

(32)

4.3 Regression Analysis

Before testing the hypotheses, first the linear relation between the independent variables, volume and valence of reviews, and the dependent variable product sales will be explored. In order to find out the ability of reviews to predict the level of product sales, two hierarchical regression analyses will be carried out. The first regression will be carried out to test the predictive ability of the volume of reviews. In the first step of the regression, the model will only contain control variables. The three main control variables are genre popularity, category spanning, and developer size. As a fourth control variable, the valence of reviews is added to the model. The results of this first hierarchical regression are presented in table 4.2. The first model of this regression explains 10% of the variance of product sales, and is statistically significant with F (4; 17,812) = 511.42, p < .001. In the second model the volume of reviews is included. This model explains 90% of variance in product sales, and is statistically

significant with F (5; 17,811) = 137,954.46. The second model, thus, increased the explanatory power substantially (R² Change = .79, p < .001). The independent variable volume of reviews was statistically significant with a Beta value of β = 0.92 (p < .001).

In the second hierarchical regression, the valence of reviews is treated as the

independent variable, whereas the volume of reviews is added to the control variables of step 1 of the regression. Table 4.3 displays the results. The first model accounts for 89% of the variance in product sales and is statistically significant with F (4; 17,812) = 34,714.34; p < .001. The second model only marginally increases the explanatory power of the model to a total of 90% (R² Change = .01, F (5; 17,811) = 32,168.75; p < .001). In addition, the valence of reviews is statistically significant with β = -0.11 (p < .001). In sum, it can be concluded that the volume of reviews has a very strong explanatory power for the variance in the product sales, whereas the valence of reviews makes only a marginal difference.

(33)

33

Table 4.2: Hierarchical Regression Model of Product Sales for Review Volume

R R² R² B SE β t Step 1 .32 .10* Genre Popularity -.99 .06 -.12* -16.08 No. of genres .04 .12 .00 .31 Developer size 2.06 .06 .23* 32.68 Review valence -5.02 .23 -.16* -22.30 Step 2 0.947 0.90* 0.79* Genre Popularity -.29 .02 -.03* -13.69 No. of genres .50 .04 .03* 12.36 Developer size .28 .02 .03* 12.82 Review valence -3.35 .08 -.11* -43.96 Review volume 1.12 .00 .92* 371.42

(34)

34

Table 4.3: Hierarchical Regression Model of Product Sales for Review Valence

R R² R² B SE β T Step 1 .94 .89* Genre Popularity -.36 .02 -.03* -16.24 No. of genres .49 .04 .03* 11.63 Developer size .33 .02 .04* 13.97 Review volume 1.12 .00 .93* 355.87 Step 2 .95 .90* .01* Genre Popularity -.29 .02 -.03* -13.69 No. of genres .50 .04 .03* 12.36 Developer size .28 .02 .03 12.82 Review volume 1.12 .00 .92* 371.42 Review valence -3.40 .08 -.11* -43.96

(35)

35 4.4 Hypotheses testing

4.4.1 Hypothesis 1

The first hypothesis regards the effect of a product’s price on the relationship between online reviews and product sales in terms of their volume (H1a) and their valence (H1b). In order to test whether these two independent variables interact with the moderating variable product price, the PROCESS procedure by Hayes (2012) was used in SPSS. Table 4.4 displays the results of the analysis for H1a.

Table 4.4a: Moderator analysis H1a

Variable Coefficient SE t P Constant 8.97 .02 438.05 .000 Price -.61 .01 -40.81 .000 Volume 1.11 .00 383.92 .000 Interaction -.12 .01 -22.14 .000 Popularity genre -.27 .02 -12.68 .000 Category spanning .58 .03 17.88 .000 Developer size .36 .02 15.41 .000 F(6;19,090) = 32,796 p < .001 R² = .90

The results indicate that the complete model of hypothesis 1a is significant with F (6; 19,090) = 32,796, p < .001. An examination of the interaction effect reveals that the regression

coefficient for the interaction effect is, b = -0.12. This value is statistically different from zero, with t (19,905) = -22.14, p < .001. Table 4.4b specifies the conditional effects at different values of price. These figures show that the interaction effect is larger for lower priced games.

(36)

36

Table 4.4b: Conditional effects at levels of price for H1a

Price Unstandardized

Boot Effects

Boot SE Boot LLCI Boot ULCI

-.29 1.14 .00 1.14 1.15

.00 1.11 .00 1.10 1.11

.53 1.04 .00 1.03 1.05

Surprisingly, the relationship between the volume of reviews and product sales is larger when games are lower priced. However, the additional variance of product sales explained by the interaction effect is only 0.02% (R² Change = .002, p < .001) In conclusion, although an interaction effect was observed, no support was found for hypothesis 1a, due to the conflicting direction of the effect compared with the hypothesized direction.

The results for the regression test for hypothesis 1b are presented in table 4.5a.

Table 4.5a: Moderator analysis H1b

Variable Coefficient SE t P Constant 8.43 .06 152.98 .000 Price -1.91 .05 -41.43 .000 Valence -4.70 .40 -11.84 .000 Interaction 8.74 .58 15.04 .000 Popularity genre -.77 .06 -12.16 .000 Category spanning .53 .10 5.55 .000 Developer size 2.13 .06 34.15 .000 F(6;17,810) = 579.49 p < .001 R² = .17

(37)

37 The complete model is significant with F (6; 17,810) = 579.49, p< .001. The model’s ability to explain the variance in product sales is only 17% and the inclusion of the interaction effect accounts for about 2% (R² Change = 0.02 ,p < .001). In addition, the interaction coefficient, b = 8.74, is significantly different from zero, with t (17,810) = 15.04, p < .001.

Table 4.5b specifies the values of the conditional effect different price levels. On the basis of these numbers it can be concluded that with higher price levels the relationship between valence and sales is, indeed, intensified. Therefore this analysis provides support for hypothesis 1b.

Table 4.5b: Conditional effects at levels of price for H1b

Price Unstandardized

Boot Effects

-.30 -7.31 .52 -8.33 -6.28

.00 -4.7 .40 -5.48 -3.92

.81 2.39 .39 1.64 3.15

2.11 13.71 1.03 11.70 15.73

The second hypothesis regards the effect of the size of a product on the relationship between product sales and reviews in terms of their volume (H2a) and their valence (H2b). The same statistical procedure was uses as for hypothesis 1. The results of the moderation tests from the PROCESS procedure by Hayes (2012) for hypothesis 2a are presented in table 4.6a. The interaction coefficient is b = -0.05. This score is significantly different from zero with t (18,009) = -19.78, p < .001. The interaction effect adds only 0.2 % to the total of 90% of variance in product sales that can be explained by the model (R² Change = .002, R² = 0.9, F(6;

(38)

38 18,009) = 28,912, p < .001). Table 4.6b provides an overview of the interaction effect at different values of size. These results indicate that as the size increases, the strength of the effect diminishes. The interaction effect, thus, works in the direction opposite from the hypothesize one. Therefore, hypothesis 2a is unsupported.

Variable Coefficient SE t P Constant 8.88 .02 419.92 .000 Size -.24 .01 -25.93 .000 Volume 1.16 .00 389.84 .000 Interaction -.05 .00 -19.78 .000 Popularity genre -.30 .02 -13.62 .000 Category spanning .56 .04 16.12 .000 Developer size .37 .02 15.83 .000 F(6;18,009) = 28,912 p < .001 R² = .90

Table 4.6b: Conditional effects at levels of size for H2a

Size Unstandardized

Boot Effects

-1.07 1.22 .00 1.21 1.22

.00 1.16 .00 1.15 1.16

(39)

39 Table 4.7a displays the results of the moderation analysis for hypothesis 2b. This model is statistically significant and is capable of explaining a total of 12%. In addition, the interaction coefficient is statistically significant with b = 1.44, t (16,742) = 4.02, p < .001. However, the inclusion of the interaction effect increases the R² with only 0.25% (R² Change = 0.0025, with p < .001).

Table 4.7a: Moderator analysis H2b

Variable Coefficient SE t P Constant 8.58 .06 147.46 .000 Size .37 .03 13.85 .000 Valence -5.04 .40 -12.65 .000 Interaction 1.44 .36 4.02 .000 Popularity genre -1.01 .07 -15.47 .000 Category spanning -.08 .12 -.72 .47 Developer size 1.92 .07 29.32 .000 F(6;16,742) = 337.07 p < .001 R² = .12

Table 4.7b displays the conditional effects at different levels of size. The Johnson-Neyman significance regions test, reveals that the interaction effect is significant up until a value of 2.22 for game size. The table shows that the correlation effect is decreasingly negative as the size of the games increases. At values from 2.83 the effect becomes positive, but is non-significant. In conclusion, since the relationship between the valence of reviews and product sales intensifies as the size of a game increases, there is support for hypothesis 2b. Although the interaction coefficient is only significant with negative values, the direction of the effect is

(40)

40 in line with the hypothesized one. Still, the interaction effect is very limited in its ability to explain the variance in product sales.

Table 4.7b: Conditional effects at levels of size for H2b

Size Unstandardized

Boot Effects

-1.08 -6.58 .50 -7.57 -5.59

.00 -5.04 .40 -5.82 -4.26

1.07 -3.50 .60 -4.67 -2.32

2.07 -2.07 0.90 --3.82 -.31

The third hypothesis regards the effect of the vividness of product information on the

relationship between products sales and reviews in terms of volume (H3a) and valence (H3b). Again, the moderation tests is conducted with the PROCESS macro of Hayes (2012). The results for hypothesis 3a are presented in table 4.8a. The hypothesized interaction is found to be statistically significant, with b = -0.09 (t (19,905) = -16.50, p < 0.001). The interaction accounts for only 0.01% of the total 90% of variance that can be explained by the model, but is significant with F(6; 19,090) = 32,254, p > .001). The conditional effects at the two

different levels of vividness are presented in table 4.8b. These outcomes suggest that the effect of the volume of reviews is stronger for games without a video in the product information. This is in line with the hypothesized direction and therefore hypothesis 3a is supported.

(41)

41

Variable Coefficient SE t P Constant 9.08 .02 438.67 .000 Vividness -.36 .02 -16.70 .000 Volume 1.15 .00 396.73 .000 Interaction -.09 .00 -16.50 .000 Popularity genre -.29 .02 -13.07 .000 Category spanning .50 .03 14.62 .000 Developer size .32 .02 13.68 .000 F(6;19,090) = 31,254 p < .001 R² = .12

Table 4.8b: Conditional effects at levels of vividness for H3a

Vividness Unstandardized

Boot Effects

No video 1.18 .00 1.77 1.19

Video 1.09 .00 1.08 1.1

The results for hypothesis 3b are presented in table 9a. For this hypothesis, the interaction effect is significant, with b = 2.91 (t (17,810) = 482.76, p < .001). The complete model with the interaction is capable of explaining 15% of the variance in product sales. The interaction effect, however, only accounts for 0.01% (p = .001) of this. Table 12 presents conditional effects at levels of vividness for H3b. These findings shows that the direction of

(42)

42 the interaction effect is contrary to the hypothesized direction, however its values are negative both for both options. The relation between the valence of reviews and product sales is also stronger for games with information that includes a video. Therefore hypothesis 3b is not supported.

Table 9a: Moderator analysis H3b

Variable Coefficient SE t P Constant 8.72 .06 154.98 .000 Vividness 1.76 .06 28.77 .000 Valence -4.96 .40 -12.34 .000 Interaction 2.91 .91 3.19 .001 Popularity genre -1.07 .06 -16.93 .000 Category spanning -.14 .11 -1.25 .211 Developer size 1.96 .06 31.15 .000 F(6;17,810) = 482.76 p < .001 R² = .15

Table 9b: Conditional effects at levels of vividness for H3b

Vividness Unstandardized

Boot Effects

No video -5.98 .44 -6.84 -5.12

(43)

43

5. Discussion

In this section the results will be interpreted in the light of the discussed literature. Additional theory will be introduced in order to clarify non-hypothesized results. Furthermore, the contributions to theory and practice will be discussed. Finally, the limitations are discussed and suggestions for future research are provided.

5.1 Main Findings

The hypotheses of this research were centered around the theory that the volume and valence of eWOM, or reviews, are capable of driving product sales (e.g. Chavelier and Mayzlin, 2006 ; Chen et al., 2004). Previous empirical research is rather conclusive about the existence of the effect of volume, whereas those for valence are somewhat mixed. The empirical findings of this research are in line with previous research, as they suggest that the volume of reviews has a strong impact on product sales. The effect of the valence of reviews is, however, very limited. One reason for the limited effect of the valence of reviews could be the unknown credibility of the reviewer. Research on traditional WOM showed that WOM is more influential when there is a strong relation between consumers and the source (Granovetter, 1973). In addition, strong ties are more likely to be consulted (Bansal and Voyer, 2000). Empirical findings show that, when presented with the choice, consumer rather source information from an offline channel than an online one (Frambach, Roest, and Krishnan, 2007). It is argued that this is due to the inability to assess the credibility of an eWOM source. For the mobile games offered in the Google Play Store, WOM is unlikely to appear in offline settings such as stores, newspapers, or magazines. Although, the online reviews displayed in the Play Store are the major and most apparent source of WOM, the credibility of the

reviewers can still not be determined. This might reduce its influence, even though consumers cannot choose to use an offline WOM channel.

(44)

44 Another potential explanation for the lack of explanatory power of review valence is the lack of sensory information that reviews can provide, which is especially problematic for experience products (Yang and Mai, 2010). As a result, the consumer focuses on the volume of reviews which can be used as a proxy for popularity. The same is true for high product sales and a large installed base. Zhu and Zhang (2010) showed that under low levels of popularity, the valence has a significant impact on sales. It is argued that this is due to a lack of other information. Theory on risk avoidance is in line with this argument, as it states that consumers attempt to minimize their risk during a purchase decision and that this is a strong driver for engaging in (e)WOM communication (Yan, 2010 ; Hennig-Thurau and Walsh, 2003). So in the absence of popularity of the product, the consumer might source additional information in the valence of reviews in order to reduce their risk.

The goal of this research is to find out whether other product characteristics that are likely to affect the risk of the consumer might set conditions under which the relationship between the volume and valence of reviews on the one hand, and product sales on the other, increases. The researched characteristics are product price, product size, and vividness of product information. The hypothesized effects and the actual findings are discussed below.

The first hypothesis consists of two sub-hypotheses: As a product’s price increases, the

strength of the relationship between the volume (1a) and the valence (1b) of online reviews and product sales increases. The main line of reasoning behind this hypothesis is that as the

price of the product increases, the financial risk of making an inaccurate judgement increases (Rao and Monroe, 1989 ; Yan, 2010). In order to reduce the risk, the consumer is expected to rely more heavily on reviews and ratings.

The analysis found no support for H1a, since the actual interaction effect is opposite from the hypothesized effect. The strength of the relationship between review volume and

(45)

45 product sales is actually higher for lower priced games. The effect is, however, found to be very limited and therefore it is questionable whether it makes a substantial difference in practice. Still, the lack of the hypothesized, positive interaction is noteworthy.

Hypothesis 1b, on the other hand, is supported by the results. This effect accounted for a two percent increase in explanatory power of the model. The effect of review valence on product sales is, thus, stronger for higher priced products.

It is notable that the effect of price is different for the volume and the valence of reviews. A potential explanation for the weak interaction effect for the former might be that the relation between volume and sales is already remarkably strong, which leaves little room for improvement. It remains, however, remarkable that the interaction effect is slightly weaker with higher prices for the volume of reviews, whereas it is stronger for the valence of reviews. A potential explanation might be that the risk perceived by the consumer is, indeed, increased. Therefore, the consumer might seek additional information (Bao et al., 2003). This might result in additional reliance on ratings, which could offset a small part of the reliance on the number of reviews. In addition, it could be that due to relatively low prices (80% under $1.00) the financial risk is so unsubstantial that the consumer just buys and experiences the product in order to acquire information (Murray 1991). This would, however, not explain the increase in the effect of review valence. Another possibility is that the price already signals quality to the consumer (Zeithaml, 1988). This could reduce the impact of review volume, and the valence might be addressed to confirm the expected level of quality.

The second hypothesis also consists of two parts: As the size of a mobile application

increases, the strength of the relationship between the volume (2a) and valence (2b) of online reviews and product sales increases. The underlying theory for these hypotheses is similar to

Consumer reviews and product sales : the impact of product price, size and information vividness