Building on the resource-based view of the ? rm, we advance the idea that a ? rm’s customer network can be a strategic asset. We suggest that network effects are a function of network size (i. e. , installed customer base) and network strength (i. e. , the marginal impact of a unit increase in network size on demand). We empirically study these network effects in the 16- bit home video game industry in which the dominant competitors were Nintendo and Sega.
In the spirit of the new empirical IO framework, we estimate a structural econometric model assuming the data are equilibrium outcomes of the best ? tting noncooperative game in price and advertising. After controlling for other effects, we ? nd strong evidence that network effects are asymmetric between the competitors in the home video game industry. Speci? cally, we ? nd that the ? rm with a smaller customer network (Nintendo) has higher network strength than the ? rm with the larger customer base (Sega).
Thus, our results provide a possible explanation for this situation in which the ? rm with a smaller customer network (Nintendo) was able to overtake the sales of a ? rm with a larger network size (Sega). Copyright ? 2002 John Wiley & Sons, Ltd. INTRODUCTION In many industries, the network of consumers using compatible products or services in? uences the bene? ts of consumption.
Positive network effects arise when the consumer utility of using a product or service increases with the number of users of that product or service. The tele- phone system is a widely used example since Key words: network externalities; technology lock-in; new empirical industrial organization *Correspondence to: Barry L. Bayus, Kenan-Flagler Business School, University of North Carolina, CB3490, Chapel Hill, NC 27599, U. S. A. it seems clear that the value of being part of the network rises as the network size increases. Consumption bene?
ts can also arise in markets where a large customer network leads to increases in complementary products and services, which in turn leads to increased consumer utility (e. g. , see Farrell and Saloner, 1985; Katz and Shapiro, 1985). Prominent examples of industries thought to exhibit network effects include automated bank teller machines, computer hardware and software, videocassette recorders, video games, and Internet web browsers. Not surprisingly, network exter- nalities and the implications of having a large installed customer base are receiving increased Copyright ? 2002 John Wiley & Sons, Ltd. Received 3 November 1999.
Final revision received 14 August 2002 376 V. Shankar and B. L. Bayus attention by strategy researchers (e. g. , Hill, 1997; Schilling, 2002).
As noted by Majumdar and Venkataraman (1998), the literature related to network effects broadly tackles three categories of research ques- tions: (1) technology adoption decisions (e. g. , what factors are related to whether and when a new technology is adopted); (2) technology compatibil- ity decisions (e. g. , what factors in? uence a ? rm’s decision to seek compatibility); and (3) decisions among competing incompatible technologies (e. g. , what factors are related to consumers’ choices among rival incompatible products within a single product category).
While theoretical research has addressed all three of these categories, empirical research has been limited to the ? rst and second categories of questions (e. g. , see the review by Economides, 2001).
With the exception of a few industry case stud- ies (e. g. , Gabel, 1991; Grindley, 1995), we are unaware of any published studies that empiri- cally investigate the nature of network effects in an industry with multiple competing product technologies that are incompatible.
Consequently, the purpose of this paper is to explore the third category of research questions that has received scant empirical attention; i. e. , we investigate the possible network effects that might exist for a set of competing ? rms with incompatible product technologies. This general situation is important since many markets have more than one prod- uct standard in equilibrium.
For example, currently in the PC market there are three major operat- ing systems (Windows, Mac, and Linux) and in the cellular phone market there are three stan- dards (CDMA, TDMA and GSM). Even the tele- phone system initially had multiple, competing networks that were incompatible (e. g. , Mueller,1997). Important questions in this context include the following.
Do network effects exist within each competing product technology? What is the nature of these network effects? Are these net- work effects symmetric across ? rms? What are the implications of network effects on the out- come of competition among ? rms with incompat- ible technologies? A more extensive version of this paper that includes a detailed discussion of the literature, video game industry, modeling approach, sta- tistical estimation issues, and implications is in Shankar and Bayus (2002).
A THEORETICAL FRAMEWORK FOR NETWORK EFFECTS AND COMPETITION The resource-based view suggests that ? rm capa- bilities and resources are related to long-term com- petitive advantage (e. g. , Wernerfelt, 1984; Bar- ney, 1991). Firms can achieve competitive advan- tage through heterogeneous, rare, and dif? cult- to-imitate assets or resources. The resource-based view goes on to suggest that how the ? rm uses its assets is a key determinant of a sustainable com- petitive advantage. In this paper, we propose that a ? rm’s customer network is an important strate- gic asset that can be used to gain a competitive advantage.
A customer network helps a ? rm gain an advantage by creating an isolating mechanism. An isolating mechanism is a phenomenon that pro- tects a ? rm from imitation and preserves its rent streams (Rumelt, 1984). Particularly for incompat- ible product technologies, installed customer bases are heterogeneous across competitors, and are rare and dif? cult to imitate. The effects associated with a customer network are not only a function of network size, but also network strength. A ? rm’s network size is equiv- alent to its installed user base, whereas network strength can be viewed as the marginal impact of a unit increase in network size on demand.
Drawing on the community focus theory of Feld (1981) and the social ties of belonging and sharing embedded within groups as described by Homans (1974), the source of a ? rm’s network strength stems from the customers in its installed base. Importantly, network strength may be based on virtual or phys- ical customer ‘communities’ and can vary across ?rms (e. g. , Balasubramanian and Mahajan, 2000). Particularly strong customer networks share a com- mon, underlying (actual or perceived) bond along some important dimension (e. g. , personal interests, demographic characteristics, fanatical product loy- alty).
While some ? rms are pleasantly surprised with the existence of high network strength for their products (e. g. , Apple, Harley-Davidson), oth- ers attempt to actively create, manage, and lever- age their network strength (e. g. , Saturn, Ama- zon. com). For a more complete discussion, see Rosen (2000). A ? rm’s network strength is a strate- gic asset because the social ties among members in such a customer network constitute an imper- fectly imitable socially complex resource (Barney, 1991). In addition, loyalty among members of a Copyright ? 2002 John Wiley & Sons, Ltd. Strat. Mgmt. J. ,24: 375–384 (2003).
Research Notes and Commentaries 377 ?rm’s customer base can make network strength a strategic asset (Wernerfelt, 1984). As noted earlier, an installed customer base can positively affect demand when the utility of using a product increases with the number of users of that product, or when a large customer network leads to increases in complementary products and services. In addition, an existing customer net- work might in? uence the effectiveness of a ? rm’s marketing mix decisions such as price and adver- tising.
For example, customers are willing to pay a price premium for Microsoft’s Excel, a spreadsheet product that boasts a large network of users (Bryn- jolfsson and Kemerer, 1996). Similarly, through its ‘Friends and Family Long-Distance Calling Plan’ (which increased the bene? ts to users when more users joined), MCI dramatically increased the effectiveness of its limited advertising budget (Wall Street Journal, 1995). In each of these cases, cus- tomer response to a given marketing mix decision (e. g. , price and advertising elasticities) is a func- tion of the ? rm’s customer network.
Thus, network effects can be direct, that is, the direct effect of an installed customer base on demand, or interactive, that is, the effects operate through the interaction of an installed customer base with one or more marketing mix variables such as price and advertising. These interactive network effects are important to consider since they impact the ? rm’s marketing mix decisions. The total network strength of a ? rm is re? ected in its direct and interactive network effects.
THE HOME VIDEO GAME INDUSTRY We empirically explore the nature of network effects in the 16-bit home video game industry. The network effects associated with a large cus- tomer base of hardware users are very important in this industry since they are typically associated with increased complementary products (e. g. , soft- ware titles, licensed products, television cartoon shows, videos and movies), which in turn leads to greater utility and thus greater hardware demand.
There are also bene? ts to a large user base from the word-of-mouth discussions of game strategies and experiences that take place between users of the same hardware system, as well as from the bor- rowing and swapping of games (e. g. , Monopolies and Mergers Commission, 1995). The two primary competitors in 16-bit hardware systems, Sega and Nintendo, offered incompati- ble product technologies. These product technolo-gies were not backward (or forward) compatible with other systems offered in either ? rm’s prod- uct line.
Firms in this market did not compete by changing their 16-bit product, but instead competi- tion primarily involved varying hardware price and advertising. The business strategies of Nintendo and Sega centered on their hardware systems, and these ? rms did not exhibit long-term strategic pric- ing or advertising behavior. By 1993, Nintendo had shifted its emphasis from the 8-bit NES to the 16-bit SNES, and had survived a government antitrust investigation due to its large installed base of 8-bit systems.
These two ? rms had asymmet- ric installed customer bases (at the end of 1992, Sega had an installed base of 6. 9 million units and Nintendo only had an installed base of 4. 2 million units of 16-bit systems). Each ? rm made different hardware pricing and advertising decisions during this period, and obtained different outcomes; i. e. , the ? rm with the smaller installed customer base of 16-bit systems (Nintendo) was able to eventu- ally overtake the ? rm with the larger installed base (Sega) in monthly demand (see Table 1).
A MODEL OF NETWORK EFFECTS AND COMPETITION Given the advances in game theory indicating that market outcomes (e. g. , demand) and pro? tability are not only a function of broad structural variables but also signi? cantly related to market- and ? rm- speci? c characteristics (e. g. , the different demand and cost structures of competitors, the order of decisions by rivals) as well as rival ? rms’ strate- gic decisions (e. g. , Moorthy, 1993), to more fully understand the impact of a ? rm’s strategic deci- sions on its performance we have to simultane- ously understand its effects on demand, costs, and competitor reactions.
To do this, researchers within the ‘new empirical industrial organization’ (NEIO) tradition develop and estimate structural econo- metric models where ? rm decisions are based on pro? t maximization and the decisions of competing ?rms are interdependent (i. e. , the strategic deci- sions of one ? rm cause a reaction from its com- petitor).
There are several advantages to the NEIO approach (e. g. , Kadiyali, Sudhir, and Rao, 2001). Copyright ? 2002 John Wiley & Sons, Ltd. Strat. Mgmt. J. ,24: 375–384 (2003) 378 V. Shankar and B. L. Bayus Table 1. Annual summary of data for the U. S. home video game industry 1993 1994 Through August 1995 Nintendo 16-bit unit sales (in millions) 1. 91 1. 66 0. 52 Sega 16-bit unit sales (in millions) 2. 59 2. 03 0. 45 Nintendo 8-bit installed base (in millions) 25. 7 26. 0 26. 2 Nintendo 16-bit installed base (in millions) 4. 8 6. 6 8. 0 Sega 16-bit installed base (in millions).
7. 6 10. 1 11. 7 Nintendo 16-bit top 10 software sales (in U. S. dollars, millions) 19. 2 66. 0 36. 0 Sega 16-bit top 10 software sales (in U. S. dollars, millions) 56. 4 50. 4 34. 8 Nintendo advertising expenditures (in U. S. dollars, millions) 46. 4 47. 2 22. 5 Sega advertising expenditures (in U. S. dollars, millions) 46. 9 40. 7 11. 6 Nintendo 16-bit average price (in U. S. dollars) 120 115 122.
Sega 16-bit average price (in U. S. dollars) 112 118 114 Since structural models are based on a behav- ioral theory of ? rms (e. g. , pro? t maximization), the estimated parameters have economic mean- ings that can be directly interpreted. The estimated parameters of structural models are invariant to policy changes (due to the simultaneous consid- eration of demand, costs, and competitive reac- tions), allowing for ‘what if’ analyses associated with changes in a ? rm’s decision variables.
The structural approach also provides an opportunity to empirically test alternative theories of strategic interaction since the best-?tting model can be con- sidered to represent the particular market situation being studied. These advantages, however, come at a cost. Since NEIO studies consider greater details associated with the competition between ? rms in a particular situation, they are really only case stud- ies that do not offer clear generalizations.
Instead, generalizations come from the replication of NEIO results across similar competitive situations (e. g. , Kadiyali et al. , 2001). Given our interest in studying the possible net- work effects for competing ? rms with incompati- ble product technologies, it is important to consider the different demand structures of the competi- tors (e. g. , competitors can have different network sizes and network strengths) as well as the strate- gic interaction of the competing ? rms.
Thus, we follow the NEIO research approach. Using data from the home video game industry, we will esti- mate a structural econometric model assuming the data are equilibrium outcomes of the best-? tting noncooperative game in price and advertising. We consider a situation with two ? rms, each offer- ing its own proprietary and incompatible product technology. Each ? rm decides on the price and advertising expenditures for its product.
We model the direct effects of each ? rm’s customer network on its demand, as well as possible interactive net- work effects that may operate through price and advertising. Given the nature of the video game industry, we also consider the possible effects of Nintendo’s (incompatible) installed base of 8-bit systems and the possible effects due to ? rm differ- ences in software quality.
The demand model We consider a situation of two competing ? rms, each having a demand function of the follow- ing form: Qit =e? it P?? it it A? it it P? i jt A?? i jt i=N(intendo), S(ega) i? = j(1) Here, Qit =? rm i’s demand at time t,Pit =?rm i’s price at time t,andAit =? rm i’s advertising expenditures at time t. Further,? is the param- eter for brand-speci? c effects, ? and ? are the own price and advertising elasticities, and ? and ?are the cross-price and cross-advertising elastic- ities.
All the parameters are assumed to be non- negative. Consistent with prior research that ? nds asymmetric price and advertising elasticities across ?rms, we do not impose any constraints that these parameters must be equal across competitors. In Copyright ? 2002 John Wiley & Sons, Ltd. Strat. Mgmt. J. ,24: 375–384 (2003)
Research Notes and Commentaries 379 line with the published empirical literature, we also expect that there are diminishing marginal returns to advertising (? it <1, ? i<1) and the own price elasticity ? it is greater than one. In line with other empirical studies of network effects (e. g. , Majumdar and Venkataraman, 1998), we consider ? rm i’s 16-bit network size at time t, B16it , to be exogenously determined.
This simpli- fying assumption seems reasonable for exploring the role of network effects in a competitive sit- uation characterized by ? rms with short planning horizons (i. e. , this analysis should at least provide a lower bound on possible network effects). Further, we also consider the effects of ? rm i’s software quality at time t,Kit , since it is expected to in? u- ence ? rm i’s demand, as well as the effectiveness of its price and advertising. Finally, we control for the possible effects of Nintendo’s existing installed base of 8-bit systems at time t,B8Nt .
Letting ? 1ibe ? rm i’s direct network effect coef? cient, we incorporate the direct effects of a ?rm’s customer network through the exponential intercept term in the demand equation (1): ?Nt =? 0N+? 1NB16Nt +? 2NKNt +? 3NB8Nt ?St =? 0S+? 1SB16St +? 2SKSt (2) Here, ? 0icaptures possible brand-speci? c effects that are constant over time and not explicitly accounted for by the other variables. We also include appropriate terms in (2) for Kit and B8Nt . Following the established literature on network effects, we expect that ? 0i,? 1i,? 2iand ? 3Nare non-negative.
Our primary interest is in the param- eters ? 1Nand ? 1S, both of which we expect to be positive. The possible in? uence of a customer network on the effectiveness of a ? rm’s price decision is captured through its own elasticity: ?Nt =? 0N?? 1NB16Nt ?? 2NKNt ?? 3NB8Nt ?St =? 0S?? 1SB16St ?? 2SKSt (3) Similarly, the possible in? uence of a customer network on the effectiveness of a ? rm’s advertising decision is modeled as ?Nt =? 0N+? 1NB16Nt +? 2NKNt +? 3NB8Nt ?St =? 0S+? 1SB16St +? 2SKSt (4) Here, ? 0iand ? 0iare the own price and own adver- tising elasticities, respectively.
Our primary inter- est, however, is in ? 1i(? rm i’s price-network size coef? cient) and ? 1i(? rm i’s advertising-network size coef? cient). These coef? cients represent the interactive network effect of ? rm i’s customer base on its price and advertising effectiveness. Follow- ing the theoretical and empirical literature dealing with network effects, price sensitivity is expected to decrease as the network size increases (i. e. , ?1i? 0 or customers will be willing to pay more for a product technology supported by a large net- work of users due to an expected increase in com- plementary products; Brynjolfsson and Kemerer, 1996).
Similarly, advertising is likely to be more effective as the network size increases (i. e. , ? 1i? 0 or ? rms with a large network can maintain their demand with less advertising expenditures due to scale ef? ciencies associated with a larger ‘buzz factor’ around the expected increase in com- plementary products; Rosen, 2000).
As software quality increases, we also expect that price and advertising sensitivities will decrease and increase, respectively. Finally, a large installed base of prior product technology (B8Nt) is expected to be asso- ciated with lower price and higher advertising sensitivities for Nintendo. It follows that ? 0i>1 (since ? it >1and? 1i,? 2i,? 3N,Bit ? 0) and ? 0i< 1(since? it <1and? 1i,? 2i,? 3N,Bit ? 0). The competitive situation We view the duopolistic competition between ? rms as one of repeated games of strategic interaction.
Repeated games enable ? rms to enhance their posi- tions vis-`a-vis a one-shot game and can re?ect the long-term nature of strategic competition between ?rms. The unique equilibria in repeated games of ? nite duration are the same as those in a stage game played in every period (Fudenberg and Tirole, 1992). The pro? t for ? rm iat time tis ?it =(Pit ? ci)Qit ?
Ait ? Fi(5) where ciis the marginal cost and Fiis the ? xed cost of production for ? rm i. Since technologi- cal products have relatively short life cycles, we assume marginal costs are constant for each ? rm; this is also consistent with the NEIO approach. Copyright ? 2002 John Wiley & Sons, Ltd. Strat. Mgmt. J. ,24: 375–384 (2003) 380 V. Shankar and B. L.
Bayus Firms simultaneously maximize their pro? ts by choosing their own price and advertising expen- diture levels. We do not discuss other noncoop- erative games that we considered in the course of our research, including several Stackelberg leader– follower structures, since our video game data do not exhibit any strong leader–follower pat- terns. Based on the demand model (1) and the pro? t function (5), the Nash equilibria in price and advertising can be derived from the ? rst-order con- ditions. Details of the equations to be estimated, along with a discussion of the estimation approach, are in Shankar and Bayus (2002).
AN EMPIRICAL ANALYSIS Data The data available for estimation purposes include monthly time-series information between January 1993 and August 1995 for Nintendo and Sega 16- bit hardware sales (units), hardware price (dollars), advertising expenditures (millions of dollars), and installed customer base size (units) for the 16-bit systems (and Nintendo’s 8-bit system). A summary of this information is in Table 1. Data on sales of the top 10 software titles (i. e. , ‘killer’ games) for each system (millions of dollars) are used as our measure of software quality.
Sales and price information come from the NPD Group, a lead-ing organization that tracks this industry. The sales data are based on a sample of 17 leading U. S. retail chains that account for 65 percent of the video game systems sold. The average monthly price is computed by dividing the monthly dollar value of sales by the volume of units sold. Advertising information for the 16-bit systems comes from the Broadcast Advertising/Leading National Advertis- ers (BAR/LNA) reports published by Competitive Media Reporting.
To obtain monthly advertising ?gures, we divided the original quarterly values using a uniform distribution of spending. Since the Sega Genesis and Nintendo SNES systems were introduced before the start of our data series, the January 1993 value of each ? rm’s installed customer base was obtained from Brandenburger (1995). Given the wide range in values for the original data, natural logarithms of the network size variables (B16iand B8i) and software sales were used in the empirical analysis to stabilize the variation within these variables.
Also, dummy variables are included in the demand functions (via Equation 2) for November and December due to seasonal considerations. Finally, analysis of the correlations among the independent variables showed that multicollinearity was not a problem for these data. Estimation results We estimated the model using both 3SLS and GMM methods. Because a Glesjer (1969) test showed that heteroscedasticity is not a problem for our data, we only report the results for 3SLS in Table 2. From Table 2, the signs of the coef? cients are intuitive and reasonable.
The network effects asso- ciated with each ? rm’s 16-bit installed base are generally signi? cant. Software ‘quality’ has sig- ni? cant main effects, as well as signi? cant effects through price, for both Nintendo and Sega. With the exception of price, the effects of Nintendo’s prior product technology are insigni? cant. Own price and advertising elasticities are signi? cant for both ? rms, as are the cross-price elasticities. The signi? cant results for the November and December dummy variables are consistent with the seasonal nature of demand in this industry.
Importantly, the parameters associated with the 16-bit network effects (? 1i,? 1i,? 1i)are signi? – cant for at least one of the ? rms. These results indicate that the home video game industry does indeed exhibit network effects as proposed in the theoretical economics literature. As indicated by the third column in Table 2, the coef? cients relat- ing to the direct effect of network size (?1i)of Nintendo and Sega are not statistically different at the 0. 05 level, consistent with the assumption in most theoretical studies. However, the difference between the ? rms’ price-network size coef? cients (? 1i)is signi? cant at the 0. 05 level, as is the differ- ence between the ? rms’ advertising-network size coef? cients (? 1i).
These results re? ect asymmetry for the competitors in network strength through advertising and price; speci? cally, they are more favorable for Nintendo. From Table 1, it is clear that between 1993 and 1995 Sega maintained a substantially larger installed base of 16-bit systems than Nintendo.
In addition, both ? rms made different advertising and pricing decisions. This is particularly evident in 1995, when Nintendo spent almost twice as much in advertising than Sega, and had a higher price. Copyright ? 2002 John Wiley & Sons, Ltd. Strat. Mgmt. J. ,24: 375–384 (2003) Research Notes and Commentaries 381 Table 2. 3SLS estimation results Nintendo coef? cient estimates Sega coef? cient estimates Test of coef? cient difference Network effects Direct effect (? 1)1. 71 (0. 76)?
1. 93 (0. 78)?? Not signi? cant Price-network size interactive effect (–? 1)0. 10 (0. 04)? 0. 06 (0. 02)? Signi? cant?? Ad–network size interactive effect (?1)0. 08 (0. 03)? 0. 03 (0. 10) Signi? cant?? Control variables Firm-speci? c effects (? 0)6. 34 (2. 71)?? 6. 14 (4. 13) Signi? cant? Software quality (? 2)1. 65 (0. 54)?? 1. 76 (0. 51)?? Not signi? cant 8-bit network size (? 3N)0. 33 (0. 61) NA Not signi? cant November seasonal effects (? 4)0. 70 (0. 14)?? 0. 57 (0. 24)??
Not signi? cant December seasonal effects (? 5)1. 42 (0. 18)?? 1. 42 (0. 18)?? Not signi? cant Own price elasticity (–? 0)? 3. 23 (1. 34)?? 3. 46 (1. 45)? Not signi? cant Price–software quality (–? 2)0. 0025 (0. 001)? 0. 0031 (0. 001)? Not signi? cant Price–8-bit network size (–? 3N)0. 06 (0. 03)? NA Signi?cant? Own advertising elasticity (? 0)0. 13 (0. 04)? 0. 21 (0. 09)? Not signi? cant Ad–software quality (–? 2)0. 003 (0. 011) 0. 0013 (0. 0094) Not signi? cant Ad–8-bit network size (–? 3N)0. 08 (0. 12) NA Not signi? cant Cross-price elasticity (? ) 0. 28 (0. 11)? 0. 20 (0. 10)?
Signi? cant? Cross-advertising elasticity (–? )? 0. 03 (0. 13) ? 0. 05 (0. 11) Not signi? cant Marginal cost (kin U. S. dollars) 54. 67 (18. 12)?? 59. 32 (21. 27)?? Signi? cant? n=64; standard errors in parentheses; system-wide R2=0. 63 ?Signi? cant at 0. 05 level; ?? signi? cant at 0. 01 level Despite Sega’s initial lead in earned (estimated)gross pro? ts, Nintendo was able to just surpass Sega’s level of pro? ts during the months lead- ing to August 1995.
At the same time, Nintendo was able to pass Sega in unit sales during 1995 (see Table 1). The parameter estimates in Table 2 provide a possible explanation for this observed behavior. The demand parameters of Nintendo are either comparable with, or more favorable than, Sega’s parameters. In particular, Nintendo has stronger interactive network effects through price and advertising than Sega.
These strong network effects may have contributed to a decision to have higher equilibrium advertising expenditures and prices, which in turn enabled Nintendo to even- tually catch and surpass Sega in monthly demand. An interesting result is that the cross-price elas- ticities are signi? cant, but the cross-advertising elasticities are insigni? cant.
Given that the home video game industry is characterized by incom- patible hardware systems and unique game soft- ware (e. g. , Nintendo’s Super Mario Brothers and Donkey Kong vs. Sega’s Sonic the Hedgehog and Mortal Kombat), the advertising of each system appeals to its own consumer segment and ? rms primarily compete for new customers to the mar- ket. Due to the inherent nature of incompatible systems, the home video game industry seems to represent a setting in which demand is elastic with respect to its own price and advertising as well as competitive pricing, but is unresponsive to adver- tising of the closest substitute product.
In summary, after controlling for various possi- ble asymmetries between competitors, we ? nd that each ? rm’s 16-bit customer network has a direct effect on its own hardware demand and interactive effects through its own hardware price and adver- tising. More important, we ? nd strong evidence of asymmetric interactive network effects between the competitors.
In several industries for which network effects are important, a common situation is one in which there are multiple competing product technolo- gies that are incompatible (e. g. , Voortman, 1993). Today, for example, there are several compet- ing wireless communication standards, as well as several digital audio, video, and graphic formats. Copyright ? 2002 John Wiley & Sons, Ltd. Strat. Mgmt. J. ,24: 375–384 (2003) 382 V.
Shankar and B. L. Bayus When competing ? rms have incompatible and proprietary products, theory suggests that a com- petitive advantage accrues to the ? rm with largest customer network or installed base (e.g. , Katz and Shapiro, 1985; Farrell and Saloner, 1985). Of par- ticular interest is the theoretical result that in mar- kets with strong and symmetric network effects across competitors, situations of technology ‘lock- in’ can be obtained. In other words, once a par- ticular product technology gains any small lead over competing technologies in terms of its cus- tomer network size, there is a tendency for the technology with the larger network to become the industry standard (e. g. , Arthur, 1996).
This result implies that under some conditions an inferior product with a lead in establishing its own network will ultimately win out over a superior product (e. g. , David, 1985). As noted by Hill (1997), net- work externalities and the possibility of lock-in also suggests that ? rms with competing technology standards should attempt to build their installed customer bases as quickly as possible. Recall that we started out by asking three related research questions: Do network effects exist within each competing product technology? What is the nature of these network effects?
Are these network effects symmetric across ? rms? Consistent with most theoretical models’ assumption that the direct effect of a customer network on demand is sym-metric across competitors, we ? nd that the network size coef? cients of Nintendo and Sega are signi? – cant but not statistically different (see ? 1iestimates in Table 2). In agreement with the hedonic price models for computer spreadsheet software (e. g. , Brynjolfsson and Kemerer, 1996), we ? nd that price effectiveness of Nintendo and Sega is a func- tion of network size (see ? 1iestimates in Table 2).
In addition, we ? nd that each ? rm’s advertising effectiveness is in? uenced by its relative network size (see ? 1Nestimate in Table 2). Our empirical results also show that Nintendo and Sega have asymmetric network effects since the interactive network strength values through price and adver- tising are statistically different for the two ? rms (see ? 1and ? 1estimates in Table 2). These asym- metric network strength values may help explain why Nintendo was able to pass Sega in monthly sales of 16-bit home video game systems despite Sega’s larger installed base.
Our last research question concerned the implications of network effects on the outcome of competition among ? rms with incompatible technologies. Assuming that network effects are symmetric, the ? rm with the largest installed base of customers is generally thought to have an advantage over its competitors. However, our results indicate that network effects can depend on the size of the installed customer base and the network strength associated with its direct effect and with its interactive effects through price and advertising.
Moreover, the strength of each ?rm’s installed customer base can be different, leading to asymmetric network effects.