The effect of new product preannouncement timing on pre-release eWOM and early market performance : an empirical investigation in Indie games industry

Master of Science Thesis Business Administration

The Effect of New Product Preannouncement Timing on

Pre-release eWOM and Early Market Performance:

An Empirical Investigation in Indie Games Industry

Yunrui Ji (11374616) June 23, 2017 (Final Draft)

Entrepreneurship and Management in the Creative Industries Track University of Amsterdam

First Supervisor: Rens Dimmendaal Second Supervisor: Frederik Situmeang

Statement of originality

This document is written by Student Yunrui Ji who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

Abstract

Why are some new products more successful than others? This paper takes the perspective of innovating firms to understand what actions may affect new product success. In particular, the research investigates the potential impact of preannouncement timing decision by focusing on the time period before and shortly after product release. The main objective is to understand whether and how preannouncement timing decision influences consumer word-of-mouth and market performance of new products. Based on the empirical setting of indie game industry, the paper is able to examine the relationship among preannouncement timing, volume of online word-of-mouth before release (i.e. pre-release eWOM) and new product’s early market performance. It is found that new products with a preannouncement behavior closer to release date have greater amount of pre-release eWOM. The increased eWOM in turn attracts more players shortly after release and delivers better early market outcomes. The failure for an early preannouncement decision to gather more online discussions may relate with the fact that customers can quickly lose interest over time. The findings contribute to the literature of preannouncement, word-of-mouth and new product success by linking the three streams together, and shed some light on how innovating firms and individuals can strategically choose the optimal preannouncing time to boost product buzz and improve early market performance.

Table of Contents

3.1 NPP TIMING AND PRE-RELEASE EWOM ... 19

3.2 PRE-RELEASE EWOM AND EARLY MARKET PERFORMANCE ... 21

3.3 NPP TIMING AND EARLY PRODUCT PERFORMANCE ... 22 4. METHODOLOGY ... 22 4.1 RESEARCH SETTING ... 22 4.2 DATA COLLECTION ... 24 4.3 VARIABLES ... 27 4.3.1 Dependent variable ... 27 4.3.2 Independent variables ... 27 4.3.3 Control variables ... 28 4.4 METHOD ... 32 5. RESULTS ... 34 5.1 DESCRIPTIVE STATISTICS ... 34 5.2 LOG TRANSFORMATION ... 36 5.3 CORRELATION ANALYSIS ... 37 5.4 HYPOTHESES TESTING ... 40 5.4.1 H1: NPP timing and pre-release eWOM ... 40 5.4.2 H2: pre-release eWOM and early market performance ... 42 5.4.3 H3: NPP timing and early market performance ... 44 5.4.4 Robustness check with full sample ... 46 6. DISCUSSION ... 46 6.1 H ... 46

6.3 MANAGERIAL IMPLICATIONS ... 53 6.4 LIMITATION AND FURTHER RESEARCH ... 55 7. CONCLUSION ... 58 REFERENCE ... 59 APPENDIX 1 – DETAILED DESCRIPTIVE STATISTICS ... 65 APPENDIX 2 – OUTLIER ANALYSIS ... 67 APPENDIX 3 – RESIDUAL ANALYSIS ... 68 APPENDIX 4 – ROBUSTNESS CHECK ... 69

1. Introduction

New products have experienced high failure rate to be accepted by people or make profits (Dijksterhuis, 2016). Around 40% of new offerings fail across different industries (Hienerth and Lettl, 2011). Many of them already failed before entering the market and reaching their first customers. The Internet allows entrepreneurs to promote without large marketing campaigns or media coverage. But the digital era has also made it more difficult with overwhelming information (Godes et al., 2005). It is hard to stand out when everyone has a voice.

But different voices from consumers online can serve as an important marketing tool. Internet-mediated communications among consumers, the electronic word of mouth (eWOM), is frequent and important in influencing consumer decision-making process (Berger, 2014; You et al., 2015). Consumers increasingly search online for information to reduce uncertainty and find social assurance (King et al., 2014). Online discussions prove to be more effective in increasing product awareness and persuasiveness (Bickart and Schindler, 2001), as consumers tend to trust customer-generated contents over traditional media information (You et al., 2015). Under proper management, innovating firms and individuals can leverage eWOM into their competitive advantages for future success. As a result, (e)WOM has attracted considerable interest in the past decade (King et al., 2014). Academics examined its

characteristics, antecedents and effects to better control “the world’s most effective, yet least understood marketing strategy” (Misner, 1999).

The prevalence and significance of eWOM continues to grow over time (Hewett et al., 2016). Information about products can appear even before the products are available on the market. People start talking about concept cars and next-generation software that will be released months or years later in the future. Firms also deliberately reveal their new ideas and early prototypes to prevent competition and delay consumer purchase (Lilly and Walters, 1997). The important role of eWOM in communication environments (Hewett et al., 2016) calls for an extensive understanding of its dynamics and influences over different periods of a product’s life, from idea development to commercialization. While conceptual and empirical researches looked into the period from early to late adoption, discussions before product release is often neglected. Studies provided evidence for firm’s ability to manage online social interactions (Godes et al., 2005), but failed to further explain how to manage the eWOM. What strategies firms adopt will influence eWOM and eventually the product market performance? What actions may affect consumer attitudes and stimulate the motivation to spread good words around? To be more specific, what information to stress in promotion, what platforms to choose and above all, when to make the first public appearance to start the whole process of eWOM?

The paper centers around the last question and focuses on the preannouncement decision of individual innovators in the indie game market. Indie game developers can either introduce their games when almost completed, or reveal them as early as the stage of basic concepts. The gap between first announcement and official release varies along this strategic decision. The research therefore tries to analyze (1) whether the preannouncement timing will influence the accumulation of eWOM before release, and (2) whether this pre-release eWOM will in turn affect the early market performance of the new products.

This paper is among the first attempts to understand the influence of firm actions on pre-release eWOM and the consequent market performance. The research combines theories from eWOM and preannouncement, and will complement the existing literature from the viewpoint of firms and the time frame before release. The findings will serve as the theoretical foundation for further researches of other relevant strategies. Individual developers and firms alike can find beneficial references to pick the ideal time for public announcement. The managerial implication on announcement decisions can be further generalized for other contexts like independent movies and music.

The rest of the paper is organized as follows. Section 2 presents the literature review on eWOM with user communities in particular and on new product preannouncements. Section 3 bases on the review and constructs a conceptual framework with hypotheses for subsequent research. Section 4 explains the methodology and research design while

Section 5 shows the analysis results. A further discussion together with implications, limitations and possible directions for future studies are included in Section 6. The paper ends with a conclusion in Section 7.

2. Literature Review

2.1 e-WOM

eWOM, also referred to as online buzz, is defined as “any positive or negative statement made by potential, actual, or former customers about a product or company, which is made available to a multitude of people and institutions via the Internet” (Hennig-Thurau et al. 2004, p 39). These information exchanges via eWOM influence consumer behaviors and the consequent product performance (Berger, 2014). Researches acknowledged the impact of both volume and valence of eWOM, as more eWOM with higher level of positivity can gain more sales. The increased sales further generate more eWOM and establish a positive feedback mechanism (Duan et al., 2008). In particular, consumers are more likely to adopt and be satisfied with a product of positive eWOM, whereas negative words can undermine product image and lead to product failure (East et al., 2008; Moldovan et al., 2011).

eWOM becomes more frequent and important as consumers rely more on each other. King et al (2014) attributed the development of online retailing and the extensive use of social media to the popularity of eWOM. The rationale of consumers’ increasing

reliance to eWOM lies in its ability to address different needs.

2.1.1 Consumer needs in decision-making process

The existing literature identifies three major needs when consumers make adoption decisions: a) to obtain more information about the product; b) to reduce efforts in search and evaluation process and c) to find assurance for the (potential) adoption of the product (King et al., 2014).

As the variety and complexity of offerings increase, consumers require more product-related information in usability, technical and financial aspects (Talke and O’Connor, 2011). Relevant messages online provide a deeper understanding over product attributes. Consumers are thus able to form a clearer perception regarding the relative advantages, compatibility, usage difficulty, profitability and costliness. They become more likely to adopt innovative products with the lowered uncertainty and barriers. (Talke and O’Connor, 2011).

Though in need of more information, consumers become more demanding as they are reluctant to spend more efforts. On the one hand, the development of Internet results in an explosion of available information. On the other, the excessive information varies greatly in its usefulness and reliability (Godes et al., 2005). This information overload requires not only more physical and psychological efforts to search, but also more knowledge reserves to evaluate and select truly helpful information. Consumers are desperate to simplify the process (Godes et al., 2005).

The social response towards a product also exerts a significant impact in consumer decision-making process. People actively seek for social assurance through observational learning, as they expect to benefit from complying choices with others. Popularity, even when implicitly indicated, can act as a signal of quality (Salganik and Watts, 2008).

In general, consumers wish to acquire more information with less needed efforts to reduce potential risks in functional, economic, social and psychological concerns (Ram,1989). Meeting the major consumer needs leads to a greater willingness for product adoption (King et al., 2014).

2.1.2 Characteristics of eWOM in addressing consumer needs

The distinct characteristics of eWOM make it particularly effective in satisfying these needs.

With the development of Internet, the volume and reach of eWOM are unprecedented (King et al., 2014). A vast amount of information is able to travel across communities and borders to reach millions of consumers in a short period of time. The persistence and observability of eWOM further enhance the information availability, as online messages remain in public and are assessable whenever a consumer seeks for opinions (King et al., 2014). When information from the vendor turns out to be limited or unavailable, online word of mouth serves as the source of additional information (Berger, 2014).

eWOM presents in the form of straightforward expressions like written reviews and average ratings. The textual nature provides more time to reflect and requires clearer and more logical articulation than oral communication (Berger, 2014). Messages written in words are better constructed and carry more implicit information on the senders in the syntactic and semantic properties (King et al., 2014). It is also less likely to misinterpret the dominating attitudes with the visualization of numeric rating scales (Chevalier and Mayzlin 2006). The salience of information reduces the effort in evaluating eWOM’s credibility (King et al., 2014) and allows consumers to grasp important feedback quickly and easily (Godes et al., 2005).

Despite the anonymous nature of online media, Bickart and Schindler (2001) discussed in their study on Internet forums that information in virtual communities are more credible, relevant and emphatic than marketer-generated information. People exchange opinions online, and reinforce ideas with each other as they conduct further discussions. Fellow consumers are expected to have no vested interests in the product and no manipulation incentives in their shared opinions (Bickart and Schindler, 2001). The contents generated by them are thus considered as an authentic reflection of typical product performance and social trends (Bickart and Schindler, 2001). People observe or infer adoption behaviors accordingly to seek legitimacy for their own purchase intention or action. Reading and sharing opinions also generate empathy between senders and receivers and help signal identities within particular domains (Berger,

2014). The presentation of a direct view on consumer opinions therefore makes eWOM a trustworthy source for social assurance and reassurance of any (potential) purchase behaviors.

2.1.3 Influencing factors of eWOM

While academics have examined the impacts of eWOM in product success and its effectiveness in meeting consumer needs, studies about what affects eWOM are much recent (Chen and Berger, 2013). Researches to analyze the initiation and evolution of eWOM identified various influencing factors, including motivations, social network structures, product and contexts (Berger, 2014; Berger and Schwartz, 2011; You et al., 2015). Berger (2014) argues that word-of-mouth in general is motivated by five self- (rather than other) serving goals. The written nature and large audience size of eWOM further encourages the psychological goals of impression management and persuasion. Yet the influence of different individuals will not remain the same. Well-connected individuals (“hubs”) prove to be most effective for seeding strategies, as they are more active and possess a higher number of connections (Hinz et al., 2011). Product itself may compromise the effect of eWOM (Moldovan et al., 2011). Publicly consumed products, for instance, provide additional information on previous purchases and reduce the value of information gathered through eWOM (You et al., 2015). Experience products with low trialability, on the other hand, may encourage more pre-purchase information searches. Both the volume and valence of e-WOM become more effective

under this situation (You et al., 2015). Contextual factors also play a role in affecting eWOM, as different industries, communication channels and time periods influence the elasticity of eWOM on product sales (You et al., 2015; Berger, 2014).

While the current literature generates a rich understanding on how eWOM can affect product performance and how itself is influenced, the focus is mainly on consumer statements about products already in the market (Hennig-Thurau et al., 2010). Substantial discussions can appear far in advance of a product’s release, and will further influence the subsequent posting behaviors (Moe and Schweidel, 2012). Yet relevant research on pre-release eWOM is rare and centered on the movie industry alone (Liu, 2006; Houston et al., 2008). Both Liu (2006) and Houston et al (2008) provided evidence that early buzz for to-be-released movies can predict the opening-weekend and the overall box office success. But it remains unknown about whether these findings can be further generalized for other industries and for unfinished products. What stimulates a product’s pre-release eWOM when it is still in development and how do these early buzz in turn influence adoption decisions? Closer examination is needed to explore the determinants and impacts of pre-release eWOM under different contexts.

2.2 Virtual communities

Though people can search and receive online information from various sources, more than half of online users prefer to visit the same sites repeatedly (Rothaermel and

Sugiyama, 2001). Virtual communities have the potential to attract more regular visitors and serve as the major locus for eWOM.

Following the original attempt of Tönnies to discuss the concept of community (1912; 1967), numerous researchers went further to define virtual communities based on the key elements and specific characteristics (Hagel and Armstrong, 1997; Rothaermel and Sugiyama, 2001). Ridings et al. (2002, p273) embraced the common attributes of previous definitions and offered a comprehensive way to define virtual community as “groups of people with common interests and practices that communicate regularly and for some duration in an organized way over the Internet through a common location or mechanism”. Although virtual communities are essentially social aggregations similar with real-life social networks (Rheingold, 1993), Hienerth and Lettl (2011) suggested two major ways that distinguish virtual communities: (1) unlike egocentric social networks, virtual communities center around specific topics; and (2) community members drawn by same interests generally possess more related knowledge and have more needs within the particular domain than average people.

2.2.1 Characteristics and benefits of virtual communities

The distinguishing characteristics of virtual communities make them beneficial for innovation development, adoption and diffusion (Hienerth and Lettl, 2011).

consumer needs, assess market potential and get inspiration and support for new ideas (Hienerth and Lettl, 2011; Franke and Shah, 2003; von Hippel, 2001). As von Hippel (2001) discovered in his study, customer needs evolve over time. They learn along the adoption process and constantly change their preferences. New product developers observe the online discussions to obtain messages on current problems and future trends. Yet the information on user needs is sticky: it is costly to transfer the community messages to outsiders, and even more difficult to transfer necessary information all at once (von Hippel, 2001). Developers need to participate in the communities in order to understand and collect the needed information. Virtual communities provide not only information and related context for user needs, but also an opportunity for innovators to experiment with incrementally adapted prototypes and evaluate the market potential (von Hippel, 2001; Hienerth and Lettl, 2011).

The benefits of belonging to a community go further as members provide support for innovators. Networks are critical for firm survival and performance in entrepreneurship process (Hienerth and Lettl, 2011). As a specific type of social network, virtual communities give access to more information and resources (Hienerth and Lettl, 2011). Community members provide quality assistance either by direct help or by giving indirect reference for outside resources (Franke and Shah, 2003). Crucial resources including funds, equipment and multidisciplinary knowledge may not be reached among individual social circles (Hienerth and Lettl, 2011; Franke and Shah,

2003). Innovators in general find the received assistance satisfactory and helpful (Franke and Shah, 2003). The studies of sources of innovations also identified the member’s role in inspiration, flaw identification and even psychological support (Hienerth and Lettl, 2011). To some extent, innovators and community members together deliver a marketable new product.

On the other hand, community members feel more engaged as they exchange experiences with each other (Hagel and Armstrong, 1997). The shared interests and unmet desires of members strengthen the ties with one another and increase their willingness to communicate. Their extensive knowledge allows them to conduct in-depth discussions within the community, as studies found it easier for community members and especially lead users to adopt complex innovations (Schreier et al., 2007). Community members also play an active role in dispersing information across communities to reach broader readers. Hienerth and Lettl (2011) advocated the importance of involving key community members for “crossing the chasm”, because consumers tend to find the information more convincing with the expertise of these key members. Community members actively engage in discussions and disseminate the word to increase the activation in the outside world. Virtual communities are effective in stimulating WOM and generating greater awareness and persuasiveness for new products (Hagel and Armstrong, 1997).

While virtual communities can be helpful for new product development and adoption, innovators cannot benefit unless users are willing to get involved. Academics conducted researches to understand what stimulate community communications and contributions. Prior work has shown the underlying motives as a mix of intrinsic and extrinsic rewards (Hienerth and Lettl, 2011). The extrinsic rewards refer mainly to the material rewards. Its effect is however limited, especially when most community members reveal information and provide assistance free of charge (Franke and Shah, 2003). The intrinsic rewards comprise of community factors and personal psychological benefits (Franke and Shah, 2003). Community members help each other out of expectation of reciprocity or pure fun and enjoyment (Franke and Shah, 2003). Studies on open source software also identified reputation as a key driver of voluntary efforts within communities (Jeppesen and Frederiksen, 2006).

In general, the scientific literature demonstrates the benefits of virtual communities and what motivates member participations and contributions. Practitioners have also recognized the critical role of virtual communities and started to seek ways to get engaged. Firm’s participation and management of these communities can be as simple as on-going additions of new contents or the filtering work of site contents. Godes et al (2005) shed some light on firm’s management on social interactions (SI), stating “at least some of the SI effects are partially within the firm’s control”. While confirming the potential of firm management, they identified a variety of research opportunities in

the area, including the appropriate extent of firm intervention and the optimized time to step in. The current paper follows this stream and explores the impact of firm actions on the participation motivations and the consequent contribution behaviors. In other words, this paper takes a further step and examines how innovating firms and individuals can stimulate the motivational rewards and maximize community contributions; and how they can leverage the acquired resources and information to progress through the process of idea development and diffusion.

2.3 Launch decision / Preannouncement

For new products, it is important to make a good entrance that can build a beneficial first impression and stimulate further attention and discussion. Launch strategies concern “what”, “when”, “where”, “why” and “how” to launch and occur as early as in the fuzzy front end of the new product development process (Hultink et al., 1997; Frattini et al., 2014). In particular, the preannouncement practice is getting more widespread across different industries, with Eliashberg and Robertson (1988) observed more than half of preannouncements for new products in their study. While preannouncements can occur when entering new markets, changing prices, expanding production or introducing new products (Kohli, 1999), this paper focuses on the last issue. The action proves to be closely related with new product success (Lilly and Walters, 1997).

New product preannouncement (NPP) is defined as “a firm's communications about one or more new products or services to be introduced to the marketplace by the firm at some future date” (Lilly and Walters, 1997). The concept should be explicitly distinguished from new product announcement (NPA), as the former appears far in advance and exhibits a higher degree of uncertainty (Su and Rao, 2010). The target audience can be consumers, competitors, suppliers or other shareholders like potential investors, but the statements are available in public and the information will reach more target groups nonetheless (Lilly and Walters, 1997).

2.3.1 Antecedents of NPP

Firms preannounce to appeal to potential customers and to preempt competitive behaviors (Eliashberg and Robertson, 1988). The knowledge of interesting products available in future markets encourages consumers to postpone their purchase of current options (Lilly and Walters, 1997). Competitors find it less tempting to enter when the market recognizes the preannouncing firm as the industry innovator (Lilly and Walters, 1997). Preannouncing is also a good way for firms to test market potential and allow channel partners to prepare complementary products and systems (Lilly and Walters, 1997). The generated market knowledge helps to adjust following strategic actions such as the optimal launching time (Lilly and Walters, 1997; Eliashberg and Robertson, 1988).

competition (Kohli, 1999). The potential risks of information leakage and more sufficient time for competitors to react will prevent firms from promoting their new products too early in the development process. The nature of high uncertainty makes it common for preannounced products to be delayed (Bayus et al., 2001). Firms bring damage to their reputation when they fail to deliver as promised (Eliashberg and Robertson, 1988; Lilly and Walters, 1997). The advantages of preannouncement are further negated when firms expect to suffer from the cannibalization of their own existing offerings (Eliashberg and Robertson, 1988). It is a risk-benefit tradeoff on whether to preannounce, when to do it and what to include (Eliashberg and Robertson, 1988; Su and Rao, 2010).

2.3.2 NPP timing and the influencing factors

Among various NPP decisions, Kohli (1999) suggested the timing of preannouncement(s) as the most critical concept. He defined preannouncement timing as “the interval between first such (planned) communication about the product and the date the product is ready for delivery to customers” (Kohli, 1999). Delivering the product to the market is also referred to as new product launch (NPL) (Su and Rao, 2010). All advantages and disadvantages of NPP link closely to the timing decision.

Lilly and Walters (1997) examined the influencing factors of the preannouncement timing and identified four main factors, related with competitors, consumers, and the product and firm itself. The competitive environment serves as the exogenous

influencer. Firms find it less risky to flaunt their new ideas far in advance when competitors are expected to retaliate more slowly and less intensively (Kohli, 1999). Buyer loyalty to existing competing products and the innovativeness and complexity of the new product all tend to increase the preannouncement timing (Lilly and Walters, 1997). Consumers incur higher switching costs and require more time to get familiar with these new offerings (Kohli, 1999). Eliashberg and Robertson (1988) suggested that firm size and market share are negatively related with the preannouncement timing. Large firms with early preannouncements put their current offerings and market positions at risk. The potential accusation of “market overhanging” also dissuades them to preannounce too early, or to preannounce at all (Eliashberg and Robertson, 1988). The four factors interact with each other. More complicated products may increase the learning requirements for potential consumers. Firms would also turn to late preannouncements when they can not determine the final attributes of the new product or predict the delivery date with accuracy (Kohli, 1999). Lilly and Walters (1997) found in their interviews that mangers will choose late preannouncements when the gap between “feature freezing” and the actual introduction is small.

While both practitioners and academics recognized the importance of preannouncement and its related timing decision, most literature explored the goals and influencing factors (Su and Rao, 2010). Whether these goals are achieved and how effective is the preannouncing strategy remains unclear. In other words, the study on

the impact of NPP is lagged behind. Schatzel and Calantone (2006) identified this research gap and investigated the outcome of a firm’s NPP behaviors. Their findings indicated a positive influence of preannouncement behavior on new product success, where a greater level of preannouncement behavior generates favorable market expectations that in turn engender product success. Joenssen et al (2014) linked NPP timing with crowdfunding, suggesting that a crowdfunding project executed closer to its estimated delivery time are more likely to succeed and meet its funding goal. This paper bases on these prior works and further complements the NPP literature by exploring whether and how NPP timing affects new product success.

2.4 Concluding

The review of eWOM and virtual communities literatures acknowledges their significance for new product success and identifies a general research gap: a lack of understanding on the influencing mechanism of new product developers. How can these innovating firms and individuals intervene the influencing factors of online discussions? How can their strategic actions shape eWOM and virtual communities to establish a consensus in favor of the new product? And how can they leverage the beneficial buzz to deliver satisfying product performance? The role of innovators in affecting the community behaviors and eWOM remains a mystery in the scientific field. The paper particularly chooses the NPP timing as the focal strategic action and centers around the pre-release period. As the academics overlooked this dynamic period, the impact of the

meta-layer members that link innovators with end users is underestimated. King et al (2014) stressed the importance of the meta-layer communities on product diffusion and early adoption, but failed to dig deeper into the correlation between firm strategies and the corresponding community eWOM. The paper therefore combines the three research streams and tries to fill the gap by answering how preannouncement timing (i.e. early or late preannouncement) may influence the new product’s early market performance, indirectly through shaping the new product’s eWOM in virtual communities before official launch.

3. Hypotheses

3.1 NPP timing and pre-release eWOM

It is plausible to assume that an early NPP with a longer time period should generate more discussions. It takes time to spread words around and reach more people in the first place (King et al., 2014). The existing online messages further attracts others to provide their opinions. A positive loop is likely to form over time, as a greater number of eWOM generates even more discussions (King et al., 2014). The extended time period between NPP and NPL also allows the innovator to provide more related information and refine their presentation of the innovative product with following updates. Yet the choice of whether, when and what to update depends on the innovator’s strategies rather than the length of pre-release time.

The early NPP not only relates with a longer pre-release period, but also exposes the new product to public at its early development stage. This early exposure can be a double-edged sword. On the one hand, potential customers are more likely to post for impression management (Berger, 2014). Talking about original new ideas signals their expertise, as they stay “in the know” and are able to identify interesting and quality products even before they are completed (Marchand et al., 2016). The complacency of being among the first few to share further increases when potential customers have an opportunity to offer advices and become co-creators. On the other hand, an early communication increases the risks as the products are not fully developed. The presentation of unfinished products may not be well-polished or informative enough to impress people and establish advantageous product expectation. Potential customers dissatisfied with the under-developed products can either refuse posting, or provide negative feedbacks. The latter further worsens the situation by creating a more negative environment that discourages future discussions (Moe and Schweidel, 2012). Even in the case of positive discussions, the immediate WOM created by interest alone may not generate on-going WOM (Berger and Schwartz, 2011). The hype of buzz can easily wear out when potential customers lose interest and patience over a long development process.

Therefore, an early NPP allows the innovators to grab more attention and to better organize the information to present the products (Berger, 2014). Customers are more

willing to post comments when they can easily get more information that are well structured. It is also possible that the risk of preannouncing too early with an unfinished product may create a negativity spiral and stop potential customers from posting. And the interest and curiosity of customers can run out throughout the prolonged development process. The potential effect of NPP timing on pre-release eWOM is not clear at this point. As the impact can be detected in either direction, the paper makes the first hypothesis as follows:

H1: An early new product preannouncement positively influences the volume of

pre-release eWOM.

3.2 Pre-release eWOM and early market performance

Extensive studies have proved that a greater amount of eWOM positively affects purchase behaviors and the consequent product sales (Duan et al., 2008; Marchand et al., 2016). The finding remains valid across different contexts, as Bughin et al (2010) reported in their paper that “consumer-to-consumer word of mouth generates more than twice the sales of paid advertising in categories as diverse as skincare and mobile phones.”

In particular, the eWOM should be mostly related with early market performance, when other information is yet limited. Hennig-Thurau et al (2015) proved the “Twitter effect” on early adoption, as they found early tweets on a movie to be positively related with its open-weekend box office performance. They stated that consumers consider

sources like Twitter as the “first resort” to reduce the information asymmetry right after release. The volume of pre-release eWOM proves to be especially important for early adopters. A large quantity of online discussion communicates about how many people are aware of and interested in a focal product. When the main motivation for early adopters are gaining status and self-enhancement, they may base their purchase decisions more on the volume of buzz than the product quality (Marchand et al., 2016). The following hypothesis is therefore derived:

H2: The volume of pre-release eWOM positively influences the product’s early

market performance.

3.3 NPP timing and early product performance

Based on the first two hypotheses, as well as the above discussions in the literature review section that link NPP timing closely with product success, I arrive at the third hypothesis, suggesting that an early NPP decision will lead to better early market performance, and that this effect is mediated by the volume of pre-release eWOM:

H3: An early new product preannouncement positively influences the product’s

early market performance through increasing the volume of pre-release eWOM.

4. Methodology

4.1 Research setting

distributed on a digital platform for PC gaming.

The main purpose when selecting a setting is to find an industry where NPP is a common strategy in promotion and will play a significant role when utilized.

Independent production, where small, entrepreneurial firms are established to carry out individual projects (Robins, 1993), meets the criteria. Unlike mainstream corporations, independent firms (the “indies”) have tighter budgets and limited resources to organize large marketing campaigns (Johns, 2006). Indie producers and developers rely more on other promotion strategies to create eWOM for their products. Another feature that distinguishes independent productions is the level of innovation (Robins, 1993). Smaller, independent firms tend to worry less about their market positions and are better able to innovate. These creative products are also more niche-oriented (Newman, 2009) and require longer time for consumers to understand and accept. As a result, an extra promoting period before release (and thus the NPP decision) is beneficial for the indies to prepare their consumers and generate buzz.

While independent production is common in fields of music, movies and games, I use video games as the product examples. Like other experiential products, which consumers buy and use to get intangible experience and enjoyment (Cooper-Martin, 1992), the quality of video games is difficult to judge before an actual purchase. Other’s opinions are especially important, as they transform the experience attributes into search attributes that provide useful information for potential customers (Yang and Mai,

2009). The development of Internet allows video games to be digitally distributed. When products are distributed online, it is easier for consumers to find all relevant information and to post their own opinions on the Internet. The amount of eWOM on video games is therefore large and should be representative to measure the overall word-of-mouth.

The video game industry also features an exponentially decaying patterns in product diffusion (Marchand et al., 2016). As the demand for a new game is the highest at the moment of release and immediately after, the instant success of indie games is significant for the overall market performance.

Taken together, I selected the industry of indie games as the empirical setting to test the hypotheses. Indie games are defined as video games developed by individuals or small-scale studios without financial support of a publisher (Martin and Deuze, 2009). The independent, experiential nature and the digital distribution method amplify the role eWOM plays, and the decaying distribution pattern further increases the strategic importance of early market performance. It is thus common for indie game developers to preannounce and gain as much support and buzz before release as possible. The benefits of an early NPP, if any, are very likely to be observed in this particular industry.

4.2 Data collection

The study is a quantitative research that chooses a digital distribution platform named Steam as the data source and focuses particularly on games released through its

Greenlight project.

Steam Greenlight

Steam was developed by Valve Corporation in 2003 and has operated as the largest digital distribution platform for PC gaming. Besides the basic services of game installation and automatic update, the platform also provides social networking services with community features like friend lists, chat function and cloud saving.

Steam first introduced Greenlight in August, 2012 to replace the previous submission system (Valve, 2012). When the old way to select games for distribution by a small expert team failed to bring the games players were truly interested in (Steam, 2017), Steam introduced the Greenlight project to let its users help pick the games. The project allows developers without established publishers to post their games in the Greenlight community, which indicates that games submitted through Greenlight must be indie games. There is no strict regulation upon how completed the game should be; developers post information, screenshots and videos for their game and start creating an active community along the development process. Once a game is submitted and appears on the Greenlight community, users with a Steam account can post comments and/or vote among the three options (yes, no thanks or ask me later) to indicate their interest. As a result, every game has its own ratio of yes-to-no votes, and Steam prioritizes games based on this ratio. While the specific criteria remain unknown, games that get into the top 100 are most likely to be “greenlit” and can be released on Steam

once completed. Developers can continue to post updates and communicate with other community members from submission to final release.

Time span

The paper picks all games released through Steam Greenlight in the year of 2014. The chosen one-year time frame is to avoid any additional variance in longer periods that can distort the estimation. Furthermore, as it takes months to years for a game to develop, games submitted from the beginning of Greenlight in late 2012 might be ready for release in 2013 and 2014. The project itself needed time to develop mature mechanisms and become popular among game developers. In short, the choice of 2014 can guarantee a sizable and representative sample.

Access to data

So far, there is no existing database for Steam Greenlight with integrated data. Related information is publicly available online with official APIs and third party websites. To be more specific, information on games, developers and comments can be found on the game homepage in Greenlight community. For important dates, prices, players and review scores, Steam’s web APIs and other third-party websites have generalized the data and presented with tables and graphs. I scraped the data from the websites above to establish the dataset for this study.

The collection process started with a third-party website (greenlightupdates.com) to obtain all released games and their unique ids in 2014. The ids were then used to

match number of comments, announcements and other descriptive information on the game’s Greenlight homepage, prices on Steam official API, review scores on Steam Spy API and player numbers on steamcharts.com.

4.3 Variables

4.3.1 Dependent variable

The dependent variable is the average number of players for a game in the first month after release. When searching for a proxy variable to represent the product sales at the early stage, I chose the number of first month players instead of owners. The number of owners stands for how many people owns a game in their Steam accounts, either bought on Steam, in retail or in bundles with other products. It is possible that users get a game in their Steam accounts as a promotion gift, but never install to play it. Though the owner number seems to better represent product sales, it can be inaccurate. The number of players instead shows how many users actually obtain, install and experience a game. It performs better in representing true market performance.

4.3.2 Independent variables

The two independent variables are the NPP timing and the volume of pre-release eWOM, while the latter also serves as the mediator variable for the effect of NPP timing on first month player numbers. The paper treats the submission date as a game’s first preannouncement time, and uses the number of days between a game’s submission and

final release to represent the NPP timing. The volume of pre-release eWOM is measured by the sum of all comments posted in a game’s Greenlight homepage. Once a game is released on Steam, a new community will be established in Steam Community, and consumers entering the original Greenlight page will be redirected to the new community with a pop-up notice. Hence the number of comments will not increase much after release and can be qualified to stand for pre-release eWOM.

4.3.3 Control variables

The study controls for four variables: pre-release updates, game price, user review score and game-specific factors.

Pre-release updates

Developers do not always provide updates after early or late preannouncement, but the update behavior can influence the accumulation of pre-release eWOM. Berger (2014) stated that the on-going addition of valuable new contents will help to keep the current members interested and to attract more attention from new ones. The study therefore includes the pre-release updates as a control variable and measures it in terms of the total number of announcements by game developers between NPP and NPL.

Game price

Price of a game can be an important predictor of market performance (number of players in this case). Following the economic theories, demand for an ordinary product is elastic with respect to its price, where consumers are less likely to purchase with a

higher price, all else being equal. The paper includes the full price of a game (in US dollars) to control the expected impact of price on player numbers.

Review score

Another factor that has a potential influence is the quality of a game. A game’s quality may affect both how much users want to comment before release and how much they are willing to purchase and experience the game after release. The paper uses the percentage of positive user reviews to deduce game quality. A larger percentage results in a higher review score, and indicates a better quality of a game in the view of users. For example, a game with a user review score of 90% means that 90% of its user reviews are positive. This game is considered to have a higher quality than another game with a user review score of 60%.

The quality can vary during the development process, and should be more stable when the game is eventually released. Given that the historic data on user response that can reflect game quality is not available either in the pre-release period or shortly after release, the paper uses the latest review score as a proxy to deduce game quality in the view of potential customers during the period that is analyzed. As a result, the data in this variable is observed at the time of data collection. A potential ex-post conditioning bias (Horst et al., 2001) can occur and needs to be further addressed in future researches.

Game-specific factors

affecting the accumulation of pre-release eWOM and the post-release market performance. In this study, I include and control for a game’s creativity, art style, genre and player mode based on user tags.

Steam allows for search by most popular user tags. As a result, games with such tags have a greater chance to be exposed to customers and thus more likely to be purchased and experienced. The user tags in general can have an impact on game sales and player numbers. A potential bias can occur, as more popular games are more likely to get more tags with a larger variety. It is possible that less popular games may not have specific tags even when it has the specific features. However, the paper considers the effect of this bias rather limited. To elaborate with an example, a lack of tags related with creativity for a less popular game indicates that the game is not considered creative by its small group of players. While this may not represent the opinion for a larger group of players, the lack of certain tags still has an impact as it does represent some customer opinions.

Creativity refers to whether a game would be recognized by customers as creative. A game is measured as creative if it is tagged by users as “visual novel”. Previous studies have proven the role of product originality in increasing WOM amount (Moldovan et al., 2011). Hence, an innovative game is expected to generate more discussions and influence the consequent purchase decisions.

of a game as retro and (2) whether a game evokes emotion by presenting an atmosphere that triggers a feeling of horror. Games tagged as “pixel”, “retro”, “classic”, “arcade”, “1990s” and “1980s” are recognized as retro. As Esposito (2005) stated in his study, consumers tend to be nostalgic about video games from past times. A game with visual features that resembles old games are more likely to have an impact on whether customers are interested to talk about or to play it. Games tagged as “psychological horror”, “horror”, “survival horror” and “zombies” are recognized with high emotionality in the sense of horror. Berger and Milkman (2012) found in their study that contents are more likely to go viral if they can evoke more emotions such as amusement or awe. Accordingly, it can be assumed that customers are more likely to talk about or purchase games that looks terrifying. While the user tags on Steam also includes other topics related with artistic features, such as story telling (“rich story”) and specific styles (“cyberpunk”), these tags are less popular or only targets niche players. The paper therefore focuses on controlling the art style in terms of retro and horror.

The genre measures whether a game is in the popular genres of action, adventure and casual, which are the top three most used genre tags on Steam. Video games can be classified in a variety of genres such as strategy and card games (Elliott et al., 2012). Previous studies on video games included different genres as control variables based on different criteria, and have proved the impact of specific genres on game sales and

volume consumer reviews (BI Situmeang et al., 2014; Marchand et al., 2016). For instance, Marchand et al (2016) included the genre of first-person shooter as they expected the shooter games to target young customers and may largely increase the volume of consumer reviews. BI Situmeang et al (2014) based on the global unit sales and included the most popular genres of shooters, racing and sports. Following this, I included the genre of action, adventure and casual as the control variable based on the popularity of user tags on Steam. Games fall in the three genres are assumed to attract more attention and have a bigger chance to be discussed and purchased. A game is counted as in the popular genres if it is tagged as any one of the three genres.

The player mode indicates whether a game provides a multiplayer function. The paper includes it as a control variable when previous researches proved that whether a game allows a multiplayer feature had an impact on game sales and consumer reviews (BI Situmeang et al., 2014; Marchand et al., 2016). The nature of multiplayer games encourages social interactions among players and requires a larger amount of players than single-player games (Ducheneaut and Moore, 2004). The player mode therefore can influence the number of players in the first month after release. As customers may have different propensity towards a multi-/single-player game, the player mode itself can also affect how interested potential customers are and how many pre-release discussions will be generated.

The game-specific variables were included as dummy variables. The rest of the variables are scale variables and have values that are either positive or equal to zero. OLS regression was used to analyze data and test hypotheses.

A key consideration for the empirical model is to apply a logarithmic transformation. Following extant research, I conducted a natural log transformation after descriptive analysis for the dependent variable of first month players, the independent variables of NPP timing and pre-release eWOM as well as the control variables of pre-release updates and prices (Marchand et al., 2016; Ye et al., 2009). The reason to perform the log transformation is that the estimated coefficients can directly reflect the elasticity of the explained variables with respect to changes in the explanatory variables (Elberse and Eliashberg, 2003). In other words, the analysis result after log transformation directly shows that a 1% increase in NPP timing can have a n% effect in first month player numbers (n depends on the specific coefficients); while the analysis result before transformation indicates the effect of one extra day in the NPP timing on the absolute number of first moth players. In practice, an extra day can have larger impact when the original NPP timing is three days than it is one year. The log-transformed analysis is therefore more desired as it provides a better understanding for real world situations. Another advantage of log-transformation is that the original relationship can be converted into a linear form for further empirical estimation (Duan et al., 2008).

data. A correlation analysis and regression analyses were conducted with the log-transformed variables. In particular, hierarchical multiple regressions were performed to test H1 and H2 and a simple mediation analysis to test H3. SPSS 24 was used for all statistical analyses.

5. Results

5.1 Descriptive statistics

In total 507 games released in 2014 through Steam Greenlight were collected. A frequencies test followed and ensured that there were no missing data or errors. The most important findings are discussed below, for a more detailed description and statistics check Appendix 1, Table 6.

First month players

The mean value of the average first month player numbers is 44.05, but the number largely varied from almost zero to more than 1,500 with a standard deviation of 141.49. The skewness and kurtosis tests indicated that the data was highly skewed right (median=5.07) with a sharp peak. In other words, while most games had a relatively small amount of average players in the first month, a few games stood out and attracted huge crowds.

NPP timing

official release (mean=359.93). The distribution of NPP timing had a broad peak, which suggested that, depending on the strategic choice of developers and the practical development process, the time between NPP and NPL can last from less than a month to more than two years and any durations in between.

Pre-release comments

The mean of pre-release comments was 708.62 (minimum=5; maximum=9733; standard deviation=1090.39). The data was positively skewed (median=370) and have a high kurtosis value. This indicated that games differ greatly in their ability to motivate online discussions. Only a small percentage of games generated heated discussions in the Greenlight community.

Pre-release updates

On average, game developers posted around seven announcements during the pre-release period. The data was positively skewed with a long tail (median=5), where most developers preferred not to post (N=63, %=12.4) or to post no more than ten announcements (N=404, %=79.7). While games differed much in their NPP timing decisions, the differences in pre-release updates (i.e. the number of announcements) were relatively small.

Prices

9.8% of the indie games (N=50) could be downloaded for free. The average price was US$8.33, and the price could be as high as US$69.99. The distribution was skewed

right (median=6.99) and had a sharp peak in the lower prices, where most paid games were charged at US$9.99 (N=97, %=19.1) and US$4.99 (N=83, %=16.4).

Review scores

The mean value of user review score was 69.98% and the median value was 74%, suggesting that user reviews for indie games were mostly positive. Yet these user responses varied from very negative (minimum=11.43%) to completely positive (minimum=100%). There was huge difference in quality across games.

Game-specific factors

Among the 507 games in the dataset, 10 (2%) were tagged related with creativity, 104 (20.5%) were tagged related with retro, 71 (14%) were tagged related with emotionality of horror and 406 (80.1%) were tagged related with popular genre. 109 (21.5%) games provided an online multiplayer feature.

5.2 Log transformation

Before proceeding to further correlation and regression analysis for testing the hypotheses, a natural log transformation was conducted for the dependent variable of

first month players, the independent variables of NPP timing and pre-release eWOM as

well as the control variables of pre-release updates and prices. The operationalization of all variables after transformation is presented in Table 1.

An outlier analysis was performed to detect whether there are any extreme values. The method of z-score, where the distance between an original observation value and a

mean is measured in standard deviation units, was implemented to defined outliers (Thompson, 2006). Following extant researches, the paper considered an observation as extreme when the corresponding z-score was greater than 3.0 (Thompson, 2006). Seven observations were found as outliers, with extreme values in either pre-release comment numbers or review score (Appendix 2, Table 8). Due to the limited data available, these outliers were removed, as they were considered not representative for the whole sample and might distort the estimation. The final research database consisted of 500 released games. The following correlation and regression analyses were conducted with the final dataset. The original sample was used to do the same analyses later as a robustness check.

The descriptive statistics of the final dataset were presented in Appendix 1, Table 7. 5.3 Correlation analysis

As Table 2 shows, the pre-release eWOM was found significantly correlated with NPP timing, pre-release updates and some game-specific factors. To elaborate, there was a negative correlation of -.12 (p<0.01) between ln_comments and ln_time, suggesting that an earlier preannouncement significantly related with fewer pre-release eWOM. This may indicate that the negative impacts of an early NPP discussed in hypothesis section above outweigh the positive impacts. The ln_announcements positively correlated with ln_comments (r=.19, p<0.01), which meets the expectation that games with more pre-release updates have more online discussions before release.

The art style, measured in retro (r=.11, p<0.05) and horror (r=.16, p<0.01), as well as

genre (r=.20, p<0.01) of a game also presented a positive correlation with ln_comments.

It is more likely to find a game with more pre-release eWOM to be retro, creating a terrifying atmosphere and/or in one of the three popular genres.

Table 1. Description and distribution for model variables.

The price (r=-.24, p<0.01) were found significantly and negatively correlated with

ln_player_m1, while review_score (r=.29, p<0.01), ln_comments (r=.33, p<0.01) and

genre (r=.13, p<0.01) were found significantly and positively correlated with

ln_player_m1. The findings are logical, as it is expected to observe a game with a larger

amount of players to have lower price, more positive user reviews, more pre-release eWOM and be among one of the most popular genres.

Variables Descriptive Titles Operationalization

1. ln_player_m1 First month players Log-transformed average number of players of a game in the first month after NPL.

2. ln_comments Pre-release eWOM Log-transformed sum of user comments of a game between NPP and NPL

3. ln_time NPP timing Log-transformed number of days between the NPP and NPL

of a game.

4. ln_announcements Pre-release updates Log-transformed sum of announcements by a game developer

between NPP and NPL.

5. ln_price Price Log-transformed non-discounted US price of a game.

6. review_score Review score Percentage of positive reviews based on all user reviews until the time of data collection.

7. creativity Creativity A game is recognized by users as creative. Dummy variable.

8. retro Retro A game is recognized by users as retro. Dummy variable.

9. horror Emotionality of horror The atmosphere of a game is recognized by users as able to create a feeling of horror. Dummy variable.

10. genre Popular genre A game is among the most popular genres. Dummy variable.

Table 2. Means, Standard Deviations, Correlations.

Variables M SD 1 2 3 4 5 6 7 8 9 10 11 1.ln_player_m1 1.84 1.92 - 2.ln_comments 6.07 .81 .33** - 3.ln_time 5.70 .73 -.04 -.12** - 4. ln_announcements 1.71 .91 .05 .19** .00 - 5. ln_price .64 4.11 -.24** -.05 .02 -.01 - 6. review_score .70 .19 .29** .06 -.04 -.02 .04 - 7. creativity - - .02 .04 -.04 -.04 .01 .05 - 8. retro - - .08 .11* -.09* .03 .05 .23** -.04 - 9. horror - - .01 .16** .05 .04 -.06 .02 -.02 .05 - 10. genre - - .13** .20** -.03 .08 -.02 .20** .04 .15** .16** - 11. player_mode - - -.04 .02 .06 .05 -.09* -.13** -.08 -.14** -.03 -.09* _-

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

5.4 Hypotheses testing

To test the three hypotheses, two hierarchical multiple regressions (H1 and H2) and a simple mediation analysis (H3) were conducted.

5.4.1 H1: NPP timing and pre-release eWOM

A hierarchical multiple regression was performed to test H1 (An early new product

preannouncement positively influences the volume of pre-release eWOM) controlling

the pre-release update behavior of its developer, review score and game-specific factors. A residual analysis was conducted to test whether the assumptions for OLS regression can be met. An investigation of the residual plot (Appendix 3, Figure 1) suggested that there were no significant violations to severely undermine the credibility of the model and results.

The control variables were entered in the first step, significantly explaining 9.9% of variance in the pre-release eWOM (F(7,492)=7.697; p<0.001). The variable of NPP

timing was entered at step two. With all the variables, the model as a whole can

significantly accounted for 11.1% of variance in the number of eWOM before release (F(8,491)=7.669; p<0.001). The introduction of NPP timing accounted for the additional 1.2% increased in the overall explanatory capability of the model (R2 Change=.012, p<0.01). When holding other variables constant, the number of pre-release eWOM will decrease by 1.2% (β=-.12, p<0.01) if the time gap between NPP and NPL is 10% longer. The results further showed that with 10% increase in the

number of pre-release updates, the number of pre-release eWOM will increase by 1.5% (β=.15, p<0.001). Games that created an atmosphere of horror obtain 31.4% (β=.31, p<0.01) more pre-release eWOM than non-horror ones. And games belonging to one of the three popular genres will accumulate 30.9% (β =.31, p<0.001) more pre-release eWOM than games in less popular genres.

Table 3. Regression Results for Testing H1.

Variable R R2 _R2 _{Change β SE t} Step 1 .31 .10*** ln_announcements .15*** .04 4.01 review_score .09 .19 .45 creativity .30 .25 1.21 retro .16 .09 1.76 horror .30** _{.10 2.96} genre .31*** .09 3.47 player_mode .10 .09 1.12 Step 2 .33 .11*** _.01** ln_announcements .15*** _{.04 4.04} review_score .08 .19 .42 creativity .27 .25 1.11 retro horror genre player_mode .14 .31** .31*** .11 .09 .10 .09 .09 1.54 3.12 3.44 1.23 ln_time -.12** _{.05 -2.62}

Note. Statistical significance: **p<.01, ***p<0.001

In general, H1 is not supported, as a greater time gap between NPP and NPL (hence an earlier NPP) will not increase the number of comments, but instead decrease it. Following the discussion when proposing the hypothesis above, this result can be

interests exceeds the positive impacts, causing an overall negative influence of timing on the generation of pre-release eWOM. Other potential explanations and related theories are further discussed in the following discussion section.

5.4.2 H2: pre-release eWOM and early market performance

To test H2 (The volume of pre-release eWOM positively influences the product’s

early market performance), another hierarchical multiple regression was performed.

The regression analysis controlled for prices, review scores and other game-specific factors and was intended to explain the variance in first month player numbers with the number of pre-released eWOM.

Similarly, a residual analysis was conducted and the residual plot did not show severe violation of assumptions (Appendix 3, Figure 2).

In the first step of the hierarchical multiple regression, three set of predictors were entered: prices, review scores and game-specific factors. A statistic significant model was found (F(7,492)=12.876, p<0.001), which explained 15.5% of the variance in first

month player numbers. After entry of pre-release eWOM at Step 2, the total variance

explained by the model as a whole was 24.2% (F(8,491)=19.641; p<0.001). The introduction of the pre-release eWOM contributed to an additional 8.8% variance in

first month players, after controlling for the other variables (R2 change =.088;

F(1,491)=56.778; p<0.001).

suggested by the results, a 10% increase in the volume of pre-release eWOM can increase the number of first month players by 7.3% (β=.73, p<0.001), all else being equal. Improvements in the review score can also attract more first month players (β=2.82, p<0.001), while higher game prices will decrease the player number (β=-.11, p<0.001).

Table 4. Regression Results for Testing H2

Variable R R2 _R2 _{Change β SE t} Step 1 .39 .16*** ln_price -.12*** .02 -6.12 review_score 2.86*** _{.44 6.53} creativity .06 .57 .11 retro .08 .20 .40 horror -.10 .23 -.44 genre .33 .21 1.59 player_mode -.06 .20 -.33 Step 2 .49 .24*** _.09*** ln_price -.11*** _{.02 -6.13} review_score 2.82*** _{.42 6.77} creativity -.13 .54 -.25 retro horror genre player_mode -.04 .33 .08 -.14 .19 .22 .20 .19 -.21 -1.46 .41 -.76 ln_comments .73*** _{.10 7.54}

Note. Statistical significance: ***p<0.001.

The game-specific factors related with horror and genre, which were found influential for pre-release eWOM in the test for H1, did not have a significant impact on player numbers in this model. In other words, the art style in atmosphere and the