User innovation and network effects: the case of video games

Abstract

1. Introduction

Information technology has deeply changed the way producers and consumers interact, enabling new forms of collaboration in the product development process (Füller et al., 2010), and redefining the whole concept of consumer (Cova and Dalli, 2009). Producer–consumer collaboration has been known for a while now and concerns information products as well as physical products (von Hippel, 2005). This kind of outsourcing has dramatically grown in scale and scope especially within the digital context (Arakji and Lang, 2007). The interaction between user and producer gives rise to a new “user and producer innovation paradigm,” in which the traditional microeconomic demand side of the market becomes a source of innovative designs and products (Gambardella et al., 2017) that might complement the producer’s ones. The mutually beneficial partnership between user and producer may be justified by the positive consumption externalities that typically, but not exclusively, characterize network industries (von Hippel, 2005). The situation in which user–producers supply a product that is complementary to a primary product making it available to peers, is described by Katz and Shapiro (1994) as a “system”, that is a collection of components that may work together (possibly by means of an interface). Systems pose challenges for coordination among consumers because of the network effects that may generate inside them. For the user–producers group further concerns arise, regarding consumers’ incentive to participate to the complements’ production and contribute to the system’s innovation process.

The increasing role that networks and platforms play in a multitude of settings suggests the need for further research (McIntyre and Srinivasan, 2017). Llanes (2019) studies the openness decisions of for-profit firms in presence of user innovation and calls for the development of dynamic models that take into account the strategic decision of the demand side of the market (user–producers). In this article, we tackle such demand-side perspective to provide a comprehensive assessment of the network externalities generated when users of a good are also “complementors” that is, producers of the product’s complements. On a conceptual ground, the contribution lies in arguing that user–producers, because of their two roles, may be subject to conflicting direct network effects. Consistent with this theoretical argument, the article’s main empirical contribution is to show that entry by user–producers follows a non-monotonic pattern with respect to their number, first rising and then falling. This conclusion stems from considering the motivations of producers who are also users of a task divisible, non-rival good and who have reasons to value the presence of other user–producers (Baldwin and Clark, 2006) and incentives to contribute goods to the common pool for free (Johnson, 2002). This theoretical framework allows going beyond Boudreau and Jeppesen (2015)’s empirical result that network effects on the user–producer side are merely negative.

In addition, this article teases out the concept of “interactive network effects” proposed by Shankar and Bayus (2003) and investigates whether network effects generated by user innovation affect consumer reaction to a given pricing decision. Results concerning indirect network effects confirm previous studies’ findings that indirect network effects between users and complementors are positive in both directions and that direct (same-side) network effects are positive for users.

The methodological contribution stems from a consistent estimation strategy, which uses a three-stage least squares procedure to estimate a system of equations in which both the primary product’s demand and the newly created complements are endogenously determined.

The empirical evidence concerns the video game industry and, in particular, the complementary game modifications (“mods”) for game engine platforms. The video game sector is characterized by a high degree of user innovation and thus provides an excellent field of study (Marchand and Hennig-Thurau, 2013). The analysis of a unique panel database of weekly video games’ sales allows an accurate estimation of the demand dynamics. In addition, the analysis focuses on the retail market, thus contributing to the scant literature on sales drivers, about which “we know virtually nothing” (Marchand, 2016: 143).

2. Literature review and hypotheses development

In this section, we formulate hypotheses on the consumer demand in the user–producer system in light of the literature on network effects and according to the classification in direct and indirect network effects. User–producers formulate expectations on the basis of the existing installed base and the number of third parties (the “crowd”) to plan their future consumption decisions as consumers (Farrell and Saloner, 1986) and their creative activity, as producers (Johnson, 2002).

Useful insights for inferring the possible implications of the coincidence of consumers with unpaid producers come from studies related to open innovation (Lerner and Tirole, 2002) and, as suggested by Johnson (2002), from the literature on public or club goods (Samuelson, 1954; Buchanan, 1965).

2.1 Indirect network effects in the user–producer system

In a system where users produce a complement to the firm’s product, user innovation and firm production are strategically connected. Users provide their marginal contribution to the product development by making the complements they create freely available to firm’s customers. In this situation, consumption externalities take the form of “indirect network effects” because they come to pass indirectly, through the impact of one consumer’s adoption decision on the future supply of other components. For example, the decision to purchase the primary product will be based on the prospective availability of compatible complements, as described in the “hardware/software paradigm” (Katz and Shapiro, 1985). User innovation benefits the firm as the primary product’s value stems from the volume and diversity of available complements (Gawer and Cusumano, 2014).

The traditional finding of the literature on network effects, that a larger base of “hardware” owners will lead to greater “software” sales, or to a greater variety of software or to software of high quality, still holds in case complements are supplied at a zero price by user–producers, as long as they are produced subject to declining marginal cost, due to traditional economies of scale or learning-by-doing. Thus, in the user–producer system, indirect network effects are supposed to work in the standard way, with the demand for complement increasing with the supply of primary products that, in turns, increases with the availability of complements.

We expect that when user–producers offer free complements to a primary product, indirect network effects take place according to the following hypothesis:

H1: In markets with free, user-generated complements, indirect network effects arise:

H1a: An increase in the number of users (hence units sold) has a positive impact on the generation of new complements;

H1b: New complements have a positive impact on the number of units sold.

2.2 Direct network effects in the user–producer system

In the user–producer system, direct network effects affect both segments of users-only and user–producers (complementors).

Let us focus on the users-only segment first. When users and producers are strictly segregated, but users are part of a network, an additional user may increase the value of the network to other users. Such systems are said to display “positive network externalities” (Arthur, 1989). Positive “direct” network effects may descend from the physical or technological characteristics of the network, like in the public telephone system (Rohlfs, 1974), or reflect a bandwagon effect, generated from the desire of individuals to possess goods in order to conform to people they wish to associate with (Leibenstein, 1950). Consumption network effects may also be negative, as a consequence of congestion or interference or because snob and vain consumers may lose interest in a product that is adopted more widely (Shy, 2011).

In the user–producer system, where some users may provide innovation that benefits other consumers, it is reasonable to assume a positive relationship between consumer demand and the volume of already sold products. The more users expect the product of one firm to be adopted, the more they expect it to incorporate user innovation and are more willing to purchase it (Llanes, 2019). Thus, the willingness to adopt a firm’s products depends on its expected number of users.

We then hypothesize the following direct network effect on the users side:

H2a: The number of users has a positive impact on the number of new users (hence units sold).

As for complementors, direct network effects may be initially analyzed in the light of the standard microeconomic result concerning pure-producers. When users and producers are strictly segregated and producers sell their goods to users at a positive price, producers of a homogeneous good will be harmed by competition from other producers because, as more producers enter the market, the equilibrium price declines and so the producers’ surplus. Thus, an increase in the number of producers would reduce the value of entering the market to other producers, which is implicitly a negative network effect.

Economic theory holds that producers may supply goods even for free, both to promote their careers and for ego gratification (Lerner and Tirole, 2002). The career concern incentive refers to the possibility to improve their job prospects or gain better access to the venture capital market. The ego gratification incentive reflects a desire for peer recognition. Provision of a free good creates a signal of the “quality” of the producer. However, if many producers are signaling the same users, then as the number of producers increases, the quality of the signal from each one declines. The producers’ incentives to invest in the development of new products or to enter the market decline as well (Aghion et al., 2005), resulting again in a negative direct network effect for producers. Signaling mechanisms are the main motivation on which Boudreau and Jeppesen (2015) build their hypothesis that development rates decrease as the crowd of unpaid producers increases in size.

However, the picture is more complex in the user–producer system, given that producers are also users. A user–producer may welcome, not just fear, the presence of other producers, because he/she may be able to build on their innovation, include their contributions in his/her own system and possibly enrich it with different components. Johnson (2002) and Baldwin and Clark (2006) show that this is the case in open source communities in which users–innovators have the ability to divide up tasks while sharing output. These “communities” of innovators are larger and more active the more modular and option-rich the underlying designs (Baldwin and Clark, 2006). Likewise, user–producers design heterogeneous complements with option value to other potential innovators, and make them freely available through the Internet. By devoting effort to a project in which their contributions become a public good once developed, each user–producer participates only marginally to the product development and his/her incentives to contribute depend largely on the developer base’s size (Johnson, 2002). Since user–producers can innovate more when they access more knowledge, user incentives to provide new complements positively depend on the cumulative number of the already available user-generated complements (Llanes, 2019).

In sum, conflicting economic forces affect the behavior of user–producers. Depending on the size of the network, one or the other effect may dominate and give raise to non-monotonic, same-side network effects. Larger networks allow a wider variety of possibilities for strategic interactions but also of rivalry, which makes it plausible to assume that the relationship between innovation and the number of developers follows an inverted U-shape pattern (Boudreau, 2010). Giordani et al. (2018) describe a similar behavior in open collaborative innovation communities, in which members’ degree of involvement increase with communities’ size when they are small and decrease when they grow large.

We then formulate the following hypothesis on direct network effects on the user–producers side:

H2b: The number of complements has a positive impact on the generation of new complements up to a certain threshold, but is negative beyond that threshold.

2.3 Interactive network effects in the user–producer system

The consumers demand for a “composite” product, which is made of the primary good and its complements, is a function of variables related to the former (namely, its price and quality) and the latter (in particular, its variety and quality). Actions taken by the firm producing the primary product affect the supply of complements through the effect of these actions on consumer demand for the whole product (Gupta et al., 1999). In particular, a smaller network of ancillary services or complementary product available for a primary good will reduce the consumer’s willingness to pay for that good (Katz and Shapiro, 1985). Vice versa, a larger installed base of compatible products would make the product more valuable to consumers and produce an increasing pressure on price, just like the availability of open source software allows firms to charge a larger price for hardware (Di Gaetano, 2015) and, possibly, for proprietary versions of the software itself, whose quality is improved by the development effort of the OS community (Comino and Manenti, 2011).

Thus, the theoretical literature suggests that network effects alter consumer reaction to a given marketing mix decision, taking the form of “interactive network effect”, as defined by Shankar and Bayus (2003). In their empirical analysis, the demand for a (videogame) platform captures the effect of the customer network on the effectiveness of a firm’s price decision through the platform’s price elasticity. Price sensitivity is proven to decrease as network size increases because “customers will be willing to pay more for a product supported by a large network of users due to an expected increase in complementary products” (p. 379). In this article, we leverage Shankar and Bayus (2003)’s intuition and explicitly test whether the primary good’s price sensitivity decreases as the network of free complementary products expands. As user innovation increases, the demand for the primary product becomes less sensitive to changes in price since consumers are better off when they can benefit from free innovation at a larger extent.

Given these premises, we formulate the following hypothesis about interactive network effects in the user–producer system:

H3: New complements reduce the price elasticity of units sold.

In the user–producer system, the decisions individuals make about innovation (as producers) and consumption (as users) are inextricably linked and so are the network effects involving them. We depict this framework in Figure 1. The white arrows represent the indirect network effect that the user base exerts on complements (H1a) and that complementors exert on the user base (H1b), possibly through price (H3); the black arrows represent the direct network effects that take place on the user base side (H2a) and on the complementors side (H2b).

Figure 1.

Hypotheses on network effects in the user–producer system. In the user–producer system, the two sides, users and complementors, are interested by network effects. The white arrows represent the indirect network effects that users exert on complementors (H1a) and that complementors exert on users (H1b), possibly through price (H3); the black arrows represent the direct network effects that users (H2a) and complementors (H2b), respectively, exert on each other.

Open in new tabDownload slide

3. Empirical context: network effects in the videogames industry

The video game industry provides a suitable field of study to analyze consumer demand in user–producer system and in the light of network effects. Katz and Shapiro (1994) report that producers of video games’ hardware and complementary software, aware of the under-utilization risk deriving from the positive adoption externalities affecting the system, sponsor the network by making investments in its growth. Firms like Nintendo, Sega and Atari (who sell proprietary hardware and complementary software), take lower margins on hardware than software with the aim to sustain a large proprietary network through hardware sales and thus stimulate future software revenues. In some cases, hardware producers have created “open system” by allowing third-party software developers to supply components for the sponsor’s system, with the aim to expand the network by convincing consumers that software will be inexpensive in the future.

The video games sector presents a high degree of innovation (Marchand and Hennig-Thurau, 2013) to which users contribute by offering free labor, “voluntarily given and unwaged, enjoyed, and exploited” (Terranova, 2000: 33) through vibrant online communities. One expression of users creativity in the video games environment is the so-called practice of modding, the act of modifying an existing hardware or software, which intensely permeates the modern-day game culture (Jeppesen, 2004; Nieborg and van der Graaf, 2008). Building on existing computer game titles, some users—called modders, or complementors—modify the original source-code in order to perform functions not necessarily authorized or imagined by the original manufacturer, to respond to their personal preferences or to create entirely different gaming experiences (Münch, 2013). These modifications, usually abbreviated as mods, can range from minor alterations (e.g. add-ons, patch) to more extensive variations such as partial or total conversion and, once generated, are made available to the online community for free as complementary contents to the original PC version of the game. Mods give players the chance to enjoy new visual effects and radical changes and improvements that usually are made possible only by a new system generation.

In the user–producer system, mods represent a “user-complemented market,” as they are user-created complements, involving modifications built onto or into producer products, and are diffused peer to peer. In the following, the theoretical hypotheses on network effects are discussed in light of the relevant empirical literature on the subject.

3.1 Indirect network effects

Prügl and Schreier (2006) prove the existence of a demand for user generated complementary contents for a popular computer game (H1a). This is similar to the indirect effect described by Boudreau and Jeppesen (2015) for unpaid complementors, which is built on those motivations that relate to usage growth, namely signaling and reputation: new mods will increase in response to growing user base. The same result is tested in this article, as we make the hypothesis that indirect network effects in the video games and mod system work in the standard way.

The research on the effects that complements exert on the video games (H1b) is quite extensive. The value of a console increases with the number and variety of complementary video games’ titles (Schilling, 2003; Venkatraman and Lee, 2004). Likewise, Boudreau and Jeppesen (2015) find that video game platform usage responds positively to an increase in free complements. Therefore, the empirical analysis presented here includes the estimation of the effect that new mods exert on the user base.

3.2 Direct network effects

On the video game side, network externalities arise when the gaming experience improves, or when the probability of starting to play increases, due to an increase in the number of users (H2a). This size effect may reflect a “consumption effect” between interconnected consumers for multi-player and online games and an imitative effect for single-player games (Majumdar and Venkataraman, 1998; Zhu and Zhang, 2010).

On the complementors side, direct network effects were shown to be negative because of the increased competition among complementors by Boudreau and Jeppesen (2015). Wu et al. (2013) find a similar negative direct network effects for complements sold at a non-zero price to players of free online videogames. Such a negative direct network effects may reflect a decrease in consumer perceived utility as the number of complements increases.

To our knowledge, in the literature on video games and their free modifications, the hypothesis of positive, direct network effects on the complementors side is absent. In contrast, consistent with our theoretical argument, we claim that a mod has a positive network effect to the complement itself up to a certain size of the modder crowd; beyond that size, the same side network effect of a mod becomes negative (H2b). In other words, a plot of mod development rate versus the total amount of available mods has an inverted U shape.

3.3 Interactive network effects

Complements have an interactive network effect on video games sales through reduced price elasticity. Shankar and Bayus (2003) confirm the prediction of the literature on network effects that price sensitivity decreases with network size and show that an increase in the consoles’ customer base reduces the price sensitivity of the demand for hardware. This article expands the analysis of Shankar and Bayus (2003) by showing that the price sensitivity of video games consumers decreases as new mods are made available. This is consistent with Binken and Stremersch (2009)’s findings that high-quality games may increase console sales and postpone the satiation effect that causes a rapid decline in the hardware’s lifecycle.

4. Dataset

The dataset includes five single player, best-selling video games sold in Italy through retailers from the 12th week of 2006 to the 48th week of 2014 and matches two sources of information. GfK Italy provided information on weekly sales, prices and units sold, while the intensity of user innovation through modding activity was retrieved from publicly available online sources.

For each game, distinct data on sales and prices are available by platform (PC or PS3 or X360 consoles). Thus, the sample is an unbalanced panel composed by 15 cross-sectional units (5 games × 3 platforms) and 4830 total weekly observations. As to the temporal dimension, the first week (t = 1) corresponds to the week of release of the game, the latest week to the last one with positive sales, in the time-window provided by GfK (with a maximum length of t = 453).¹

The dataset does not include online sales, which nonetheless constitute a great part of the industry revenues, although in Italy the traditional retail outlets still dominate the market. In 2015, Italy’s video games sales (referred to the software only) were approximately equal to 570 million euros and the retail segment accounted for 61.55% of it, or 350 million euros (AESVI, 2015). Information on retail sales are usually difficult to access and limit the research on distribution issues, which “remain unexplored for the video game industry” (Marchand, 2016: 150). Retail sales of the video game titles in our sample account for 47 million euros in the analyzed period.

To investigate the impact of users’ innovation on the video game demand, information about the number of available mods for each game in each time period was manually added. Until some recent exceptions (such as Fallout 4, launched at the end of 2015), the mods have been an exclusive opportunity for the PC version of the video game.² Thus, this information refers to the PC platform only (5 cross section units and 1610 weekly observations).

The sample games are similar for genre (action or adventure) and popularity both in terms of sales volume and in terms of number of mods, which are diffused mostly through the Nexusmods.com and the Moddb.com fan sites. The concentration of mods in few databases facilitated the collection and the validity of the information. At the end of 2014, for each game, all available mods were sorted by the number of users’ endorsements for the Nexusmods.com’s mods and, in absence of information about endorsements, by the total number of downloads for the Moddb.com’s ones. Then, the first 300 mods were allocated to the corresponding week of release, leading to the creation of two variables. The variable modnew indicates the number of new mods created during each week, whereas the variable modcum is the cumulative number of mods.

For the five PC platform games, along with the number of weekly sold units (variable units) and the average weekly price (variable pricePC), the database includes a variable priceCON computed as the average weekly price charged for the corresponding non-PC platforms’ games. Additionally, to measure the network effect generating from the platform’s usage, two additional variables are created, the first one focusing on the PC platform only (KPC) and the second one on all but the PC platforms (KCON), respectively defined as the cumulative number of PC or console units sold in each time period.

Table 1 summarizes the variables’ descriptive statistics and the Pearson’s pairwise correlation among them. Over the entire sample period of 453 weeks, sales for PC are on average equal to 141.61 U/week, with a maximum of 6063 U reached by a game in its first week in November 2011, and present a high variability (the standard deviation is equal to 313.07 U). The price of video games for the PC platform (variable pricePC), on average equal to 27.77 euro, is slightly lower than the average price registered on alternative platforms (variable priceCON), equal to 34.05 euro. As for the market share, the PC platform represents about one-third of the cumulative sales through other platforms (PS3 and X360), as captured by variables KPC and KCON, on average equal to 35,176 and 110,996 units, respectively.

Figure 2 helps better understanding the dynamics of the retail market for the five analyzed video games, which share some common trends. The distribution of sales across time is skewed to the right and concentrates in the first three months after release, despite some games (particularly Games 2 and 4) show pronounced sales peaks even after 100 weeks since their release. The rapid decline of sales is a tendency affecting the video games as well as the entertainment industry in general: most of the sales intensify in the very few weeks after a new product’s release and show regular peaks during the Christmas season in the subsequent years, as the spiky time series of video games sales reveals. The performance of software titles also depends on the technical characteristics of the console for which they are designed. During the past 30 years, innovation has greatly improved the game platform performance resulting in a huge increase in the average and maximum sales of every new platform launched. Because games are designed for a specific platform, the peculiar lifecycle of each platform transfers to the games, making video games a “cyclical business” (Marchand and Hennig-Thurau, 2013) and contributing to the time dependent characteristic of game revenues.

Figure 2.

Video games units and price. For each sample game, the black line represents the weekly sales per PC (in units, left-hand side axis in all graphs), the gray line the weekly price per unit (in euro, right-hand side axis in all graphs)

Open in new tabDownload slide

Figure 2 reveals for video games prices a decreasing trend similar, although slower, to that of sales. Interestingly, sharp demand peaks sometimes correspond to notable price reductions: although possible price promotions might give a one-shot boost, they cannot invert the downward trend in the demand. The strong influence exerted by the time factor in both sales and prices series justifies the significant, positive correlation between them. According to Marchand (2016), the positive relationship between higher prices and higher sales of video games depend on the high technological standard and production budgets that transfer into higher retail prices for best-selling video games (e.g. Call of Duty, Grand Theft Auto).

In Figure 3, video games units sold are compared with, instead of price, the mods created for each game in each time unit and a similar trend emerges: users’ innovation is more intense in the period immediately after the release of the software. However, for all the five games, new mods are made available during the whole analyzed period. In the context of the overall decline of the retail market, it is not immediate to detect a relationship between units sold and mods, and the causality direction itself is not clear. On the one hand, the availability of new, free of charge, mods can increase the appeal of the original product and attract new users. On the other hand, a greater number of units sold implies an expansion of the users’ community network and spurs the developers’ incentives to generate new mods. The aim of the empirical model is to deal with such complexity and disentangle all possible network interactions.

Figure 3.

Video games sales and mods. For each sample game, the black line represents the weekly sales per PC (in units, left-hand side axis in all graphs), the gray line the number of new weekly mods (right-hand side in all graphs).

Open in new tabDownload slide

5. Methodology

The empirical model consists of a system of two main equations. The first equation models the demand function in the retail video game market in order to obtain the evaluation of, on the one hand, the game network effect on the user side and, on the other hand, the impact of the intensity of users’ innovation on the video game demand. The second equation models the determinants of new mods and thus enquires the achievement of network effects on the complementors side. To facilitate the comparison of results with previous literature, the system of equations takes three different specifications (Models I–III). The baseline model (Model I) includes the standard hypotheses that direct and indirect network effects take a linear form (as in Boudreau and Jeppesen, 2015). The succeeding models highlight the main contribution of the article: Model II includes the hypotheses of non-linearity of network effects on the complementors side, while Model III integrates within the user–producer system the idea of interactive network effects (Shankar and Bayus, 2003).

5.1 Model I

In equation (1-I), units depend on the user network base, KPC (H2a), and on modnew (lagged by one week in order to reduce the risk of endogeneity) (H1b). The own-price and cross-price elasticities can be estimated including both the average price of each title’s PC versions (pricePC) and that of the console equivalent versions (priceCON).

In equation (2-I), the users’ network base (KPC), represented by all customers that have already bought a copy of the video game’s PC version, is supposed to be positively related with the potential appeal of the video games market, thus providing incentive to the creation of new mods (H1a). As to the direct network effects, the effect of the existent stock of mods (modcum) on the generation of new ones is modeled as linear.

In both equations, the panel structure of the data is exploited by using game and time period fixed effects to unambiguously control for cross-sectional variations and general macro industry trends (in the model, terms $πi$ and $σt$⁠, respectively).

5.2 Model II

In Model II, equation (1-II) coincides with previous equation (1-I), whereas equation (2-II) includes the quadratic effect of modcum. The cumulative number of the already generated mods (modcum) may first motivate modders to create new mods and then, when the crowd has reached a considerable size, may discourage them from doing it, acting as a competitive threat to the developers. In other words, the effect of the existent stock of mods on the generation of new ones is expected to be non-monotonic with an inverted U shape (H1b).

5.3 Model III

Finally, in order to capture the possibility that a novel supply of complementary goods positively affects the demand of video games by also reducing the video game’s own-price elasticity (H3), Model III incorporates the interaction term pricePC*modnew in equation (1-III). The second equation remains unchanged so that equation (2-III) coincides with equation (2-II).

6. Results

Table 2 reports the results of three-stage least squares estimates for Models I, II and III.³ The fit improves noticeably from Models I to II (R² increases from 0.481 to 0.543 in equation 2) and marginally from Model II to Model III (R² goes up from 0.823 to 0.827 in equation 1).

After controlling for the temporal variation of sales (with the inclusion of time dummies) and for each game idiosyncratic characteristics (by including game fixed effects), the video game demand function shows a regular negative relationship between quantity and price, as attested by the coefficient of pricePC, which is significant at the 1% level of confidence in all models. On the contrary, the cross-price elasticity is not significant.

In Model I, indirect network effects are confirmed to work as expected, though the impact of the number of users on the generation of new complements (H1a) is only slightly significant (P < 0.1). When looking at direct network effects, they are positive on the users side (H2a) but negative on the complementors side. Thus, as the number of existing mods increases, negative incentives to create new mods seem to prevail, in line with the results of Boudreau and Jeppesen (2015). However, in Model II, the inclusion of the quadratic term of the number of complements allows testing and confirming the presence of a non-linear effect (H2b). Thus, the negative incentives to create new mods prevail only after reaching a critical point. The new specification also improves the significance of the coefficient associated to KPC (P < 0.01). Model III pushes the analysis further by estimating the interactive network effect and shows that new mods reduce the price elasticity of units sold (H3). This finding is consistent with Shankar and Bayus (2003)’s result that price sensitivity decreases as network size increases and actually provide an empirical confirmation that this happens because a larger user base is expected to spur the supply of complementary products.

Model III provides a comprehensive view of the hypothesis testing and allow a widespread discussion of the coefficients’ magnitude.

The video game’s user base (KPC) has a positive effect on the demand of video games (Hypothesis H2a) and on the generation of new complements (Hypothesis H1a). In equation (1-III), KPC’s coefficient is equal to 0.002, thus confirming that the popularity of the game fuels sales also through the retail distribution channel. In equation (2-III), KPC’s coefficient is equal to 0.0000315, evidencing that modders are sensitive to the size of the video gamers set, who represent the audience to which they direct their creative efforts and offer their production. While the size of both coefficients associated to KPC is relatively small, this is due to the relatively high value of KPC itself (cumulative number of units sold) with respect to the dependent variables. If one computes elasticities at the mean values (taken from Table 1), the magnitude of these network effects is clearer. In particular, a 10% increase in the user base is likely to favor a 5% increase in new units sold and a 12% increase in the creation of new mods.

Differently, modders’ response to an increase in the size of the modders’ crowd (explicated in equation 2-III) is non-monotonic (H2b). The coefficients of modcum and its square (positive and negative, respectively) describe an inverted U shape, with the turning point reached when the cumulative number of mods is equal to 110, which falls in the second decile of the distribution of the variable modcum. When the crowd is small, modders have an incentive to contribute to the user experience with new mods because their visibility is high and the production of their work rare. When the crowd size increases, the modders’ signal blurs and video gamers search costs for new mods increase. As a consequence, the complementors’ incentive to create new mods decreases.

The modders’ creative flow turns out to be an important source of video games sales: every new mod has a positive impact on the video game demand and it contributes to smoothing consumer price sensitivity, as the positive coefficients of modnew (=7.656) and pricePC*modnew (=0.336), in equation (1-III), show. Computing elasticities at mean value, a 10% increase in new mods creation raises the units sold, on average, by 0.5%. Thus, Hypotheses H1b and H3 are also confirmed: in principle, new complements stimulate the market demand for the original game title and provide producers with a valid tool in terms of pricing. In the context of the typical dynamics observed in the video game industry, where new titles tend to exhibit a short market-life, this represents a key strategic opportunity. In the presence of intense users’ innovation, the life cycle of the product is enhanced and strategies based on price promotions can be postponed.

7. Discussion and conclusions

This article analyses the network effects arising in a “free complemented market,” in which consumers produce innovative complements for a primary good and make them available to the primary good’s buyers.

A fair assessment of users’ incentives to create new complements, and of the impact on the primary good sales, must rely on a comprehensive analysis of the interaction between all possible network effects arising in the user–producer system. As illustrated in Figure 1, the flows of demanded complements and of primary goods represent the marginal increment to the crowd and installed base, respectively. Network effects arising from an increase in the crowd and the installed base, and measured in terms of new complements and sales flows, configure a circular (and potentially self-reinforcing) process. Two different size effects concern the creation of new complements: a direct one, whereby variations in the crowd of complementors influence the complement development rate (in a non-monotonic fashion, as seen) and an indirect one, according to which an increase in the installed base encourage complementors to create new complements. At the same time, a flow of new complements stimulates the primary product demand, also through reduced price elasticity.

In this article, network effects are empirically measured in terms of product demand’s variation and on the basis of a unique panel dataset of weekly observations. The frequency of data collection represents a major contribution to the literature on network goods, in which network effects are often analyzed with lower frequency data (Varian, 2003). Recent empirical works on network effects make use of monthly observations of mods at best (for example, Boudreau and Jeppesen, 2015; Marchand, 2016). Weekly time periods provide a period short enough to capture the short development cycles of complements and the precise dynamics of consumer demand.

Two main findings emerge. First, user-generated complements spur the demand for the original product and smooth consumer’s price sensitivity, thus contributing to an increase in the product popularity and diffusion. Second, direct network effects arise on the user–producers side and design an inverted U shape of the innovation rate as a function of the complementors’ crowd: complementors positively respond to an initial increase in the number of their peers and then decrease the pace of their innovation activity when the crowd becomes bigger. Nevertheless, the innovation rate’s non-monotonic pattern finds a “counter strength” in the boost that user innovation gives to product sales.

7.1 Theoretical implications

The article offers a conceptual framework for modeling the demand for a primary good whose complements are supplied by user–producers, as a function of the network effects arising in such a “system.” In the user–producers system, both direct and indirect, positive and negative network externalities may take place. The article suggests that direct network externalities on the complementors side follow a non-monotonic pattern, being positive when the crowd of complementors is small but negative when it reaches a given threshold. The complementors’ need to signal their ability to an increasing crowd provides them with an incentive to engage in development activity but as soon as the audience of peers reaches a certain level, the effectiveness of signal diminishes together with their innovation rate.

In addition, user innovation is shown to reduce the primary product’s elasticity. The interaction of network effects with a product pricing strategy gives rise to the so-called “interactive network effects” (Shankar and Bayus, 2003). The article investigates the mechanism behind this result and finds that consumer price elasticity decreases for product exhibiting a large user base because consumers expect a greater availability of user-generated complements.

The empirical analysis confirms that in a user–producer system, the interplay of direct, indirect and interactive (through price) network externalities may lead to an increase in the demand for the primary good.

7.2 Managerial implications

The article’s findings imply that producers may experience an increase in their revenues when complements to the products they sell are made available in a user-complemented market. The sales growth derives from the product’s popularity among customers and complementors, as well as from the effect that a continuous flow of innovative complements exerts on the product appeal, also in terms of price. Especially for industries characterized by the frequent introduction of new products, the achievement of higher level of revenues represents a key strategic opportunity since “a firm’s rate of growth is a crucial predictor of both current and future profits […], as well as a key determinant of firm survival in dynamic industries” (Chacko and Mitchell, 1998: 736).

In the presence of intense users’ innovation, the product’s life cycle is enhanced and strategies based on price promotions can be postponed. By taking a holistic perspective over the customers to capture their potential as innovators and their preferences for user-created complementary contents, producers are offered a chance to contrast the decline in sales typical of short cycle products, because user innovation is sustained by, and contributes to, the product’s sales over time.

Network effects arising within the user–producer system and, finally, the additional value that user-generate products may bring into traditional businesses, motivate producers to coordinate their innovation activity with complementors and, possibly, to make decision over the division of labor with them (Gawer and Cusumano, 2014). For example, for video games, von Hippel and Katz (2002) and von Hippel (2016) suggest that producers should focus their development efforts on the game engine software and leave the development of mods to the gamer customers, in case supporting them with the provision of design tools and innovation environments. This is consistent with the prediction that, if the network externalities generated by the sales of complementary goods (or by original product’s quality enhancement) are sufficiently strong, there is incentive for an exclusive holder of a technology, or code, to freely license it (Economides, 1996). Comino and Manenti (2011) say that an appropriate definition of the open source software licensing terms of distribution helps firms balance the opposing effect of going open source. Along these lines, Di Gaetano (2015) analyzes under what circumstances hardware firms have incentives in investing in open source software projects. Giordani et al. (2018) show that the innovative performance of open collaborative communities rests on its own characteristics, in terms of strength of social motivations and/or of protection of the community-produced knowledge, and also on factors outside its scope, namely the characteristics of the (alternative) institutional setting based on intellectual property rights.

Although the evidence presented in this article refers to a situation in which complements are diffused through the Internet, results show that the consequences of such phenomenon oversteps the digital bounds and spread across traditional distribution chains such as retail outlets, where the product is sold. Similar network effects may then arise in sectors where technology plays a minor role but user communities contribute to fuel new product development in the form of complementary items. User designs of Lego novel constructions and of Harley Davidson parts and decorations represent well know illustrations of such situations.

7.3 Limitations and future research

The empirical evidence presented in this article is based on a panel dataset including fifteen cross-sectional units referred to five video games’ titles and three different platforms. Incidentally, also Wu et al. (2013) verify their analytical findings with reference to the same number of video games. The fact that in Wu et al. (2013), as declared by the authors themselves, “market evidence is merely supportive” (p. 166), does not depend on the number of observations but on their empirical approach, which is meant to be only descriptive and exemplificative of the theoretical model. In this article, the focus on a limited number of video games allows a precise monitoring of the modding activity. Moreover, the empirical strategy leverages the relatively long period of time for which data are available (453 weeks) to overcome the limits imposed by the relatively small cross-sectional dimension. Future research could be based on a more extensive dataset including a greater number of video games’ titles. In addition, user innovation could be measured by also including mods for platforms different from PC (that were not available until recent times).

Recent contributions have highlighted the role played by complement quality and variety (and not only quantity) in generating value through network effects (see e.g. Cennamo, 2018). Future research could usefully introduce a measure of the complement quality and variety in the assessment of the overall network effects arising in the user–producer system.

As expression of the users’ creativity and dedication, complements reflect the profile of their creators. Boudreau (2010) provides evidence that, by granting the access to the platform’s core technology to independent developers of complementary components, platform owners drew on a diverse set of capabilities. McIntyre and Srinivasan (2017) remind the critical role of complementors in consumer adoption decisions and pray for a complete assessment of the complementors’ attributes that might proxy their incentives and ability to extend support to platforms. Information about the complementors’ profile would help assessing the relative importance of complementors’ and complements’ characteristics in the generation of the network effects.

[email protected]

Footnotes

Acknowledgements

We are very grateful to an anonymous referee who provided a constructive review on the first version of the paper. We are especially grateful to the Editor, Carliss Baldwin, for her extremely generous and insightful comments that significantly improved the article. We received invaluable suggestions from Fabio M. Manenti, Gustavo Llanes and Leonardo Madio. Samuele Giacomo Pontiroli provided excellent research assistance. We received financial support by the Department of Economics and Business of University of Piemonte Orientale. All errors are our own.

References