ABSTRACT

Theory suggests that the role of network externalities is critical to transforming new economy businesses from market entrant to market leadership positions. However, there has been relatively little empirical work investigating network externalities in e-commerce contexts. Building on the theory of networks and technology standards, this paper develops and tests three propositions related to the impact of network effects and the provision of free goods in web server markets. Hedonic pricing models, estimated on the basis of comprehensive time-series data are used to examine the factors that determine price and to gauge the impact of network effects and free goods. The market for Windows Web servers reflects the influence of network externalities, even after the entry of a viable, free-market alternative. The UNIX market, dominated by open-source and other free software products through the course of this study, does not exhibit network effects. The research reported here advances the theoretical literature by testing new propositions in a new unstudied context. Its empirical findings contribute to an understanding of the dynamics of the Web software market and thus future competition in e-commerce markets.

KEYWORDS

Free products, open-source software, network effects, network externalities, information systems economics, Web server software

INTRODUCTION

It has been suggested that competition in information technology (IT) industries is heavily influenced by positive network externalities–the notion that a product or service becomes more valuable as more users adopt the product (Katz and Shapiro, 1985). Research literature has suggested that positive network externalities exert influence in a number of IT markets including spreadsheet software markets [21, 6], shared electronic banking networks [31], and carrier reservation systems [7]. However, existing empirical research is limited in both context and in its consideration of the competitive dynamic of markets studied.

The need for further study is vital, as network externalities are likely to play an increasingly important role in electronic commerce markets (Downes and Mui, 1998; Shapiro and Varian, 1998). Positive network externalities fuel adoption cycles, raise barriers to market entry, generate switching costs, and allow dominant firms to exploit monopolistic rents. The ability to harness network externalities is thus vital to transforming startups into market leaders, as well as in helping otherwise successful firms establish influential roles new industries. It has been suggested that with the rise of a global information infrastructure, firms gain access to a means for creating network externalities through the low cost distribution of new products, increased access to consumers, and by generating exchange-based value (Hagel and Armstrong, 1997; Shapiro and Varian, 1998). However, without increased empirical investigation, the extent and impact of network externalities remains uncertain, clouding our understanding of both the managerial significance and theoretical underpinnings of this phenomenon.

The existence of network externalities has been demonstrated by establishing a positive relationship between product price and market share in the spreadsheet software market (Brynjolfsson and Kemerer, 1996). However, a gap in the literature exists relative to our understanding of the existence and extent of network externalities in other IT markets and in markets where free products are offered as credible and viable alternatives to priced commercial products. The importance of studying the impact of free products in network markets is critical for several reasons. First, it may be a strategic decision for firms to provide free products to capture the needed critical mass. Second, it may be logical for certain firms to offer free goods that support other product lines. Third, open-source software efforts have gained increased commercial and corporate support. Fourth, ubiquitous networking and the growth of e-commerce facilitate the rapid dissemination of digital products at near zero marginal costs, creating a market lubricant for the introduction of potentially disruptive free-products. All this means that the distribution of free products in software markets is likely to increase. Understanding the impact of such products on the competitive dynamic of markets is vital to all players involved.

The purpose of this study is to:

• review the literature on network externalities as it relates to e-commerce and software markets;
• develop propositions regarding the behavior of network markets under conditions when viable free goods exist as alternative to priced products; and
• test these propositions in similar markets that exhibit appropriate characteristics and interpret the results in light of the theoretical literature on networks and standards.

Network Externalities

Consumers value many products and services based not only on features, but also on the size of a product’s installed base of users (or network). Economists refer to these types of products as network goods, and the positive relationship between the perceived value of a product and its network size is attributed to positive consumption network externalities (Farrell and Saloner, 1985; Katz and Shapiro, 1985).

Research on networks has suggested a number of theoretical models that involve networks or standards. Models presented by Farrell and Saloner (1986), for example, conclude that network goods have a greater tendency towards monopoly and the strength of the network externalities created as a by-product of an existing installed base may lead to a bandwagon effect, resulting in choices of inferior technologies. Network externalities, thus, have strategic implications for technology adoption, predatory pricing, and product preannouncements. Katz and Shapiro (1986) suggest that the total benefit derived from a network product or service depends in part on the number of consumers who adopt compatible products in the future. Thus, consumers’ expectations may determine the outcome of competition in the network market.

The ability to harness network externalities is vital to startups seeking to pioneer new markets and to established firms seeking to transfer dominance from one market to the next. As such, network externalities represent a key force in business transformation as it relates to the establishment of market leadership. Several researchers have suggested that information systems (Bakos and Kemerer; 1992; Besen and Farrell, 1994) and e-commerce applications in particular (Downes and Mui, 1998; Shapiro and Varian, 1998) exhibit the characteristics of network goods and hence should be subject to network externalities. This positive association between consumer value and installed base is due primarily to the impact of three critical factors – exchange, stranding concerns, and extrinsic benefits. In terms of exchange, users are attracted to a technology that is compatible with a greater network of users, all else equal, because they can engage in a wider option of value-enhancing exchange (e.g. of information, money, programs). Secondly, users of IT are highly concerned about being stranded in an unsupported standard (Fichman and Kemerer 1993). Hence, users favor products that they believe will continue to dominate in the future. Also, the dominant product is likely to attract extrinsic benefits (Shurmer 1993). These may include add-on products, books and manuals, and skilled workers. The implied significance of stranding concerns and extrinsic benefits are particularly important for e-commerce firms, as they suggest that network externalities can be generated even in environments that support open standards.

The emergence of network externalities in the contemporary managerial zeitgeist, is suggested by the inclusion of the term in Wired’s Encyclopedia of the New Economy (Wired Staff, 1998), and broader referral to the concept as "Metcalf’s Law" in business bestsellers (Downes and Mui, 1998) and the trade press. However, despite a broad base of theoretical and analytical research (e.g. Oren and Smith, 1981; Farrell and Saloner, 1985 & 1986; Katz and Shapiro, 1985 & 1986; Economides, 1996), there has been relatively little empirical work investigating the existence and extent of the phenomenon in IT and e-commerce contexts. As such, the extent and impact of network externalities on inter-firm IT competition remains largely unknown (Liebowtiz and Margolis, 1994).

Pioneering empirical research on network externalities has focused on the context of the market for spreadsheet software (Gandal, 1994; Brynjolfsson and Kemerer, 1996). Gandal (1994) used hedonic pricing models (attribute variables regressed against price) to demonstrate that compatibility with a dominant standard yielded measurable value among spreadsheet products. In a later study (Gandal, 1995), this work was extended to examine standards compatibility across product categories – in this case database management systems and spreadsheets. The higher price premium placed on standards was taken as evidence of network externalities, since products complying with the standard could access a broader network. However, it is recognized that standards compliance is confounded as a measure of pure network benefit, since this can also enhance brand perception and act as a measure of intrinsic product quality. Brynjolfsson and Kemerer (1996) extended the hedonic test for network externalities, using market share as a proxy for the extent of network installed base. This work further supported the network externalities hypothesis by showing that firms with a larger market share (or network) exhibited standards and quality adjusted price premiums over competitors with smaller market shares.

Network Effects and the Provision of Free Products

Studies by Gandal (1994, 1995) and Brynjolfsson and Kemerer (1996) focus on software markets that are dominated by firms that charge a fee for the products offered and where there has been no credible threat from a free product offering. However, it is still unclear if network externalities hold up under alternate market conditions. Of particular interest is what happens when a market is impacted by a substantive free product offering.

Providing the market with a free product may, on the surface, seem irrational, especially when the market appears to be willing to pay for fairly priced goods. However, there are situations that may encourage organizations to offer free products in otherwise viable commercial markets. The critical mass theory suggests that network goods new to the market often experience start-up problems and that a network either has to obtain an installed base at least equal to the size of the critical mass or has to exit from the market (e.g., Rohlfs, 1974; Oren and Smith, 1981; Dybvig and Spatt, 1983). Vendors may find it optimal to subsidize adoption in the presence of strong network externalities (Economides, 1996). As an extreme form of subsidy, it may be logical for vendors to offer free products in an effort to transform their firms to a position of market leadership.

Another tactic is offering a free product to the market in hopes that the product will allow a firm to gain further influence over related markets (Besen and Farrell, 1994; Shapiro and Varian, 1998). The free product may persuade adopters to employ standards supported or complementary goods offered by the firm, hence fueling network externalities in related markets. It has also been recognized that technology adopters exhibit a fear of stranding (Fichman and Kemerer, 1993). Seeding the market with free goods creates additional benefits to the firm by reducing stranding concerns by lowering quality uncertainty (Rogers, 1995) and in generating switching costs that lock in an installed base (Nilssen, 1992). Microsoft has regularly offered free add-on products that increase the value of its Windows operating systems and applications, but that compete with priced products offered by third party firms. Some Internet firms also offer free value-added services that compete directly with existing product categories. As an example, the online store Garden Escape offers free landscaping applets and a library of garden plans, competing with commercial landscape software (Gurley, 1997). This tactic of distributing free software is most effectively used by firms offering multiple products or services. Free software is most likely to be deployed in cases where the future value to a firm that is derived from consumer adoption of complementary goods or services, or migration to the supported standard, is greater than the future value that could be derived from the sale of such goods at a fair market value. The practice of commercial firms seeding the market with free products has raised industry outcry among competitors and has prompted Justice Department investigation (U.S. Department of Justice, 1998). However, it remains to be seen how such a practice effects the economics of pricing in a given market.

A second area where viable, free products may be offered is through user-led consortia-developed products. Such consortia grow and become viable for several reasons. An innovation may be developed first on the grass-roots level, commercial firms may fail to adequately address niche markets, users may be dissatisfied with existing commercial offerings, or market participants may fear the dominance of a leading industry player (e.g. Katz and Shapiro, 1994; McHugh, 1998). Many free products have grown to dominate their respective markets. For example, as of this writing the Apache web server is the dominant Internet web server for UNIX operating systems and the Linux operating system is the leading UNIX variant for Intel-based platforms. Apache and Linux are offered as ‘open-source’ products, where the public is invited to submit product enhancements and updates for incorporation into future versions. As markets for these products expand, they are likely to be supported by firms that have a vested interest in challenging the existing market leader. Commercial firms have recently begun to embrace the open-source movement. IBM now offers fee-based support for Apache web servers, and Oracle, Informix, and Netscape have developed commercial products designed to run under Linux. As further endorsement of the open-source movement, Netscape released the source code for its Communicator product in early 1998 under an organization known as Mozlla.org. Mozilla.org is charged with maintaining updates and enhancements to the free Communicator product, with the expressed goal of migrating the bulk of development from Netscape developers to the broader Internet community.

Software products are particularly prone to free product offerings given the near zero values for both the marginal cost of duplication and distribution. The Internet is a catalyst fueling the viability of open source efforts, while the monopolistic dominance of key players encourages rival firms to support free offerings. While the trend is likely to continue, the impact of free software on fee-based commercial offerings is unknown. For purposes of our investigation, three propositions are developed, one citing the behavior in traditional network markets, one pertaining to changes to such markets when a viable free alternative is offered, and one where the market is dominated by a free offering.

PROPOSITIONS

The theory of network externalities suggests that the value of a product or service will increase as its installed base expands (Katz and Shapiro, 1985). Those firms that can establish a larger network are able to generate demand-side economies of scale that fuel further expansion. Firms that are able to exploit network externalities should be able to extract monopolist-like price premiums. The traditional approach to assessing consumer value is to use price as a proxy for the average consumer’s willingness to pay. This approach has been used in earlier network studies by Gandal (1994, 1995) and Brynjolfsson and Kemerer (1996) as well as in non-network models assessing the value of product features on computer hardware (e.g. Gordon, 1993; Rao and Lynch, 1993; Lynch, Rao, and Lin, 1994). Proposition 1 restates the network externalities hypothesis using price as a proxy for value.

When a well-known competitor enters a market and begins offering a product for free, this entry is likely to create a shock that impacts other market participants. The impact of this shock may vary according to the size of the given market participant’s installed base and the aggregate consumer perception of the product’s future viability. Shapiro and Varian (1998) suggest that this aura of inevitability, that is the perception by consumers that one product will eventually dominate the market, is critical to determining the eventual winners in competitive network markets. If consumers feel that a product is losing a war in a network market then a vicious cycle begins. The vicious cycle represents a turning point critical the transformation of firms from market entrant to market-dominant positions. Consumers will begin abandoning the loser, devaluing the product and migrating to the predicted eventual winner. Those firms with smaller market shares are likely to be most threatened by the new entrant. Such an entrant will polarize consumer choice between the established leader and the well-known challenger.

Thus, it is proposed that the arrival of a well-known challenger providing free goods will act as a catalyst for the vicious cycle, weakening the positive relationship between price and installed base and hastening the decline of market prices. The end result will be a weakening of priced-based exploitation of network effects. Average, quality-adjusted market prices are expected to decline, as the viable free alternative exerts downward price pressure on all market participants.

CONTEXT, MODEL AND DATA

Context

The context of this study is the market for World Wide Web server software. Figure 1 illustrates the rapid growth that this market achieved during the period of study. WWW server software represents the largest segment of the web software market, a market that International Data Corporation valued at $916 million 1996 and has predicted will balloon to $19 billion by the year 2000.

The market is worthy of study given that 1) the web is one of the most rapidly adopted information technologies of all time and impacts many constituencies, 2) the Internet and related e-commerce infrastructures support the diffusion of other emerging software technologies (e.g. audio, video, push), and 3) issues of competitiveness and market dominance in the WWW software market are currently fostering great debate in government and industry. The WWW software industry is also well suited for comparing markets dominated by priced vs. free products, given that during the time of this study, the Windows server market was dominated by priced servers and the UNIX market was dominated by free servers.

Model

The base model for this research is derived from pioneering work performed by Brynjolfsson and Kemerer (1996), whose work was in turn influenced by Gandal (1994). This model uses price as a proxy for the value that the average consumer places on a product. Using price as a value proxy has also been used in non-network analyses of the intrinsic value of features in computer workstations (Rao and Lynch, 1993) and microcomputers (Lynch, Rao, and Lin, 1994). The model is extended here to account for the potential price benefits associated with offering a trial version of a product. A trial version of a product should lower a user’s quality uncertainty and hence lower their risk premium and increase the price he or she is willing to pay for a product (Rogers 1995). Such an extension may be particularly important to note, as the Internet has allowed firms to begin wide-scale distribution of trial versions at near zero marginal cost. The extended model is expressed as:

P_it = f(N_it, S_it, F_it, T_it, O_it) (1)

where:

P_it = price of product

N_it = the installed base of the product

S_it = standards attributes of the product

F_it = product feature attributes of the product

T_it = free trial version offered for the product

O_it = other variables influencing price

for product i during time period t.

The general equation above (1) can be tested using a hedonic model. Hedonic models are designed to estimate the value that different product attributes contribute to a consumer’s utility. Such models view a particular product or service as consisting of various bundles of characteristics, each contributing to a consumer’s utility (Berndt, 1991). Hedonic models usually employ some form of product price (as a proxy for value) as the dependent variable and regress this value against variables that indicate the inclusion of features or other value-deriving attributes. The hedonic model has been used in many non-network studies examining price and value effects in the IT industry (e.g. Gordon, 1993; Rao and Lynch, 1993; Lynch, Rao, and Lin, 1994). This technique has also been used in earlier empirical work examining software-based network externalities (e.g. Gandal, 1994 & 1995; Brynjolfsson and Kemerer, 1996). It is particularly well suited for the study of network externalities, as the impact and significance of various intrinsic features (non-network benefits) can be considered along with likely sources of network externalities.

The semi-log specification has been chosen for model testing as this should provide a more natural expression of the characteristics studied. This is because the aggregate of various factors may have a multiplicative impact on consumer value, i.e., the whole is worth more than the sum of its parts (Fisher and Shell, 1971). This specification is also consistent with previous studies of network and non-network hedonic research (e.g., Rao and Lynch, 1993; Lynch, Rao, and Lin, 1994; Gandal, 1994 & 1995; Brynjolfsson and Kemerer, 1996).

The research model used is expressed as:

Data

Two data sets were gathered for this study, each consisting of a cross-sectional time series across 19 months from August 1995 through February 1997. Observations in these two data sets reflect non-shareware, commercial web server products available for the Windows and UNIX software platforms, respectively. In keeping with earlier studies and the restriction of the hedonic pricing method, only priced products were examined. The data yielded 321 observations of Windows products and 165 observations of UNIX products.

In order to conform with the model proposed above, data on relative installed base (share), standards, product features, and price were required. Monthly data on server market share was obtained from NetCraft Consulting. Since August 1995, NetCraft has run an automated monthly survey to determine the current installed base of various web server products. Figures are tabulated through a polling program that uses an HTTP command to collect product identification. The NetCraft survey represents the most widely quoted analysis of server software market share – during the time period of this study, the survey was cited over 50 times in the trade and business press. As such, the NetCraft study is likely the most representative proxy of user’s perception of a product’s network size, a key determinate for the strength of network externalities.

Standards and features variables were chosen 1) when they were employed by previous studies, or 2) when they consistently appeared across multiple product reviews from separate sources. A consistent set of feature definitions was suggested in material provided by MecklerMedia (publisher of industry magazines that provide server product reviews, and owner of the site Webcompare.com). Features mentioned in this material were then chosen for inclusion in the test model when a given feature was cited in multiple trade press articles (e.g. Airborne, 1996; InfoWorld, 1996; Wingfield, 1996) and in a popular text on web server deployment (Stein, 1995). Our choice of feature variables is also consistent with prior research on technology adoption and acceptance (e.g., Rogers, 1995; Davis et al., 1989). For example, the presence of a graphical interface for server maintenance and the ability to administer the server from a remote location both appear consistently in the product evaluation literature. They are considered ease-of-use features, and thus enhance consumers’ valuation and likelihood of acceptance. Similarly, our choice of standards variables is consistent with prior studies of network externalities in software industries (Brynjolfsson and Kemerer, 1996; Gandal, 1994 & 1995).

Once variables were selected, time-adjusted price, feature, standards, and trialability data were collected via a content analysis of company press releases, version update logs, and industry product reviews. The variety of sources consulted allowed for cross-checking and corroboration among multiple sources and, we believe, for a highly consistent and reliable data set. In the rare case where a discrepancy was noted among sources, the software developer was contacted for clarification. The use of such industry and company-supplied sources to acquire time-reflected data is similar to that used by Brynjolfsson and Kemerer (1996), Gandal (1994 & 1995), and Rao and Lynch (1993), among others.

All hedonic studies are limited by restrictions in the appropriate selection of feature variables and the richness of information conveyed by such measures. However, it is assumed that the depth of resources consulted, and the consistency of selected criteria with factors suggested in prior research, allow for a reflective set of time-adjusted feature and price data. Findings are detailed below, along with the noteworthy characteristics of each market explored. Secondary findings relating to the influence of standards, product features, and trialability on product price are also presented.

RESULTS AND INTERPRETATION

Three sets of results are presented in Table 2. The three models in the table, (a), (b), and (c), were arrived at by examining all possible variable combinations and refined by removing variables insignificant below the 90% confidence interval. The results presented in (a) reflect a refined model of the Windows web server market. This model is structurally similar to that employed by Brynjolfsson and Kemerer (1996), with the addition of the trialability factor (TRIAL). Model (b) attempts to gauge the impact of a well-known challenger providing free goods. This was done by including a dummy variable (one-zero indicator), postMSFT, valued at 1 for all observations after Microsoft entered the market. The results in (c) were derived from applying the test model to a sample representing a market dominated by a free product. Data on the UNIX web server market was used to test and report on these conditions. Multicollinearity was tested for in each model using both the VIF and Belsley-Kuh-Welsch diagnostics (Belsley, Kuh, Welsch, 1980), and in each model these results indicated that the independent variables were not significantly confounded with each other. (See Appendix A for further information on multicolinearity tests.)

Results in a Market Dominated by Priced Goods

Model (a) represents the traditional approach to investigating network externalities in software markets. The results from Model (a) indicate that the product’s market share (SHARE) is significantly related to the log of the product’s price. These results can be interpreted as being strongly supportive of the network externalities hypothesis. A rough interpretation of this suggests that with all other factors held constant, a one percent increase in the server’s installed base warrants a price that is slightly over 12 percent higher than its competitors (e^0.1153 = 1.122; or 12.2%). These findings also represent the first extension of Brynjolfsson and Kemerer’s (1996) model beyond the context of spreadsheet software. The findings are particularly noteworthy, given that they represent strong support for significant positive network externalities in a market that relies heavily on open standards.

Additionally, the positive coefficients of two of the three standards variables (DBLINK and SSL) also suggest qualified support for the existence of network externalities. Earlier studies have suggested that standards variables may be sources of network externalities, given that standards allow a network participant to engage in an exchange with more users (e.g. Gandal, 1994 & 1995). However, it is recognized that the positive benefits associated with standards may not be entirely due to network effects in that they may also increase consumers’ perception of the intrinsic quality of the product (Brynjolfsson and Kemerer 1996).

It is also interesting to note that the variable for the S-HTTP standard was not significant at the threshold level in any of the models presented. SSL and S-HTTP are competing, although not mutually exclusive security standards. Descriptive statistics show that the SSL standard (mean .3988) is much more popular than the S-HTTP standard (mean .1776). This result may also be interpreted as being supportive of the existence of network externalities in this market as witnessed via a significant premium being placed on the more popular standard (e.g. Katz and Shapiro, 1992).

The proxy for trialability (TRIAL) proved highly significant and yielded the largest coefficient. An interpretation of this is that firms that offer a trial version of their servers were (ceteris paribus) able to price their products roughly 1 1/3 times higher than were firms that did not offer a trial version. In the context studied, the web allows firms to reap advantages from distributing demonstration versions at a low, fixed cost. These results are supportive of Nejmeh’s (1994) suggestion that the Internet can be a strategic tool for the software enterprise. These results also support the notion that users recognize that there are risks associated with adoption (Rogers 1995) and that trial versions of products help reduce uncertainty.

The negative coefficient for DURATION suggests that, ceteris paribus, the longer a product has been on the market, the lower the price of the product is likely to be. Shapiro and Varian (1998) note that this pattern is common among technology innovators eager to exploit the willingness of early adopters to pay a premium to be part of the technology vanguard. This effect may be particularly strong in the software industry, given that digital products such as software benefit from a near-zero marginal cost of reproduction and distribution (Negroponte 1995). In this scenario, lower average prices result as the volume of consumers grow, as the relatively fixed cost of the initial software development is covered by early sales, and as more competitors exert downward pressure on the average price of server products.

The significance and positive coefficient of the TIME variable unexpectedly suggests that server prices have increased over time, all else being equal. These results are contrary to the quality-adjusted price declines over time identified in other studies of IT (Gordon, 1993; Rao and Lynch, 1993; Lynch, Rao, and Lin, 1994). One explanation for this effect is that as the Internet continued to grow, Internet software was recognized as being more valuable to an organization and new manufacturers coming to market reflected this perceived higher valuation in higher product prices (ceteris paribus).

The Impact of a Well-Known Challenger Providing Free Goods

While previous studies have focused on markets dominated by priced products, as noted earlier it is common for firms to introduce free products into markets that are dominated by priced goods (Besen and Farrell, 1994). Proposition 2 asserts that average market prices will fall with the arrival of a well-known free rival. Model (b) examines the Windows market as in (a), but adds an additional variable, postMSFT. This parameter is used to assess the impact of the entry of a free-producer by identifying observations that occurred after a viable challenger entered the market with a free offering. This variable is valued at 0 for observations during the first 8 months of the sample period when no Microsoft web server was offered, and 1 during the last 11 months of the sample period, during which Microsoft had introduced a free web server. The model (b) including this variable showed slightly more descriptive power than model (a) (adjusted r-squared of .779 vs. .769), while the direction, significance, and estimated coefficient of the variables suggests a consistency of results compared across the two models.

Results of the model suggest that during the last 11 months of the study, the average product price decreased by roughly 32 percent (e^-0.39 = 0.6771; or -32.2%). This supports the assertion in Proposition 2 that a well-known challenger offering a free product may exert negative price pressure on incumbent producers. Such a finding suggests that an analysis of likely entrants and their possible pricing strategies is critical for participants in network markets. A failure to do so may result in unforeseen price shocks that may impact the firm’s revenue stream as well as product profitability and viability in the market.

Product share was also shown as being positively associated with price. This suggests that significant network externalities can be derived even in markets impacted by the arrival of free-pricing newcomers. A comparison of the coefficient for the SHARE variable in models (a) vs. (b) suggest that the positive association between price and share was actually stronger when considered alongside the impact of the free-product offering entrant. A one percent difference in share yielded a roughly 13 percent price premium in model (b), vs. the 12 percent noted earlier in model (a). While admittedly small, the difference does suggest a slight strengthening of network effects when the shock of the new entrant is considered, as well as a consistency of the existence and significance of positive network externalities across both models.

The significant, positive relationship between price and market share in the model can be considered alongside the price drop occurring after the free challenger’s market entry. While it would seem that network-related price benefits were not lowered as Proposition 2 suggested, it is acknowledged that these benefits are likely to accrue to the firm with the most dominant market share. This is because network externalities are subject to positive feedback effect – growth begets growth. As such, it is likely that only the stronger incumbent firms would be able to maintain existing pricing. Such market share growth may be achieved by targeting customers migrating from weaker players. The established leader will be able to continue to derive network-based price premiums as long as consumers forecast the incumbents continued dominance with respect to the challenger.

Results in a Market Dominated by Free Goods (UNIX Market)

The UNIX market for web servers is dominated by freeware producers, which during the course of this study made up over 50% of that market’s installed base. This competitive dynamic challenges one of the assumptions of price-based investigations of network externalities - that suppliers price products as a reflection of consumer’s willingness to pay. All previous empirical studies of network externalities in IT have examined contexts where priced products dominate. However, as noted earlier, this condition can not be assumed in all contexts (Besen and Farrell, 1994).

The lack of prior empirical work under such conditions raises the following question - can a firm competing in a market dominated by free products still capture and benefit from network externalities? While one context is not likely to provide results that can be generalized in all cases, by beginning to look at such questions we develop a frame of reference for informing managerial decisions and for understanding the robustness and applicability of network externalities theory.

In order to test whether or not network externalities remain significant, even in markets where free products dominate, the UNIX market is examined as the representative sample. The sample of 165 observations of priced, commercial UNIX products was explored using all possible variable combinations, yielding the refined UNIX model (c). With an adjusted r-squared value of .724, this model yields explanatory power similar to the Windows market models.

However, the results from the UNIX market are significantly different from those of the Windows market. Most notably, the estimation results are not supportive of the network externalities hypothesis. The variable for server market share (SHARE) is significant, but it is negatively associated with price. Since only commercial firms that charge for web servers are included in the sample, this suggests that the UNIX market does not show demonstrable network benefits. One interpretation of the significant negative coefficient is that in the absence of pricing based on consumer value, network benefits fail and traditional economics take over. Firms that capture larger market share leverage their scale economies and lower their price to try to make their products seem as attractive alternatives to free competitors. These results would suggest that firms entering markets dominated by free products should expect to price their products only according to the quality-adjusted value that these products offer above and beyond the dominant free goods. Firms offering priced products must effectively exceed the free product’s quality and innovation schedule while pricing these innovations above their own fixed-costs. Those that can not may never see profitability.

The time trend in model (c), although small, is significant and yields a negative coefficient. Such negative time trends have been noted in other studies of information technology (e.g. Chow, 1967; Gordon, 1993). This is to be expected in the context studied given that, over time, it should become increasingly apparent to commercial participants that the free products are eroding the market share of for-fee products.

The significant feature variables (SCRIPT, SEARCH, and UDIR) all yield positive coefficients as would be expected, lending credibility to the overall estimates. Although an exact comparison of coefficients across the two markets is inappropriate, the consistency of significance and direction of influence suggests that feature selection yields reliable results across both WINDOWS and UNIX server markets. The ability to link to external databases is the only evidence of network externalities in the UNIX market and, as stated earlier, this measure is admittedly confounded with intrinsic quality benefits. Neither of the two security standards variables was significant.

The proxy for trialability (TRIAL) remained significant across all three models. The coefficient in model (c) suggests that, all else equal, firms offering trial versions can command prices that are roughly 63% greater than those of rivals that do not offer trial versions. When combined with the results from models (a) and (b), these results suggest strong support for trialability-related price premiums in software markets. This factor that had not been considered in any of the previous studies of network externalities in IT, yet firms planning to offer software products would be wise to consider the demonstrated positive relationship between trial versions and price.

CONCLUSION AND FUTURE RESEARCH OPPORTUNITIES

The role of network externalities in transforming firms and markets has been widely suggested. However, other than anecdotes and speculation, there has been relatively little empirical investigation of the existence and extent of network externalities in IT markets (Liebowitz and Margolis, 1994), and none considering e-commerce products or the role of free goods in network markets. Previous theoretical research has suggested that network externalities represent a key resource vital to transforming IT (e.g. Bakos and Kemerer, 1992) and electronic commerce (e.g. Shapiro and Varian, 1998) firms from market entrant to market dominant positions. Those that cannot leverage network externalities to seize a leadership position risk obsolescence. Recently, we have seen many Internet start-ups take advantage of market capitalization to position themselves as the market leaders by gobbling up other firms to boost their market share or installed base. These early efforts appear to be designed in part to alter consumers’ expectations regarding the eventual winner in the marketplace. For firms operating in markets of older technologies, establishing leadership in new markets may require a significant transformation of the firm’s product offerings, pricing strategy, and market focus (Christensen, 1997). The transition from market no-show to market leader may involve offering products for free in order to capture share, influence consumer perception of a firm’s eventual dominance, and gain control of standards.

It is hoped that this work contributes to the understanding of network effects and furthers development of network theory by examining a new, unstudied context, developing propositions based on prior theoretical assertions, and testing these propositions while extending existing modeling techniques. Although it would be inappropriate to suggest perfect generalizability of the results of this study, the early dynamics of the web software market may be similar to that of future competition in e-commerce markets, particularly for those products that support the e-commerce infrastructure. This investigation should provide a valuable reference example for both practitioners and researchers.

Certain limitations of this study are acknowledged. First, as with all hedonic pricing models, it is acknowledged that price may not fully proxy consumer’s perception of a product’s value. Second, the short time-series of observations used in our study can be a possible limitation. Markets may go through developmental stages that have not been adequately expressed by theory or captured using existing modeling techniques. Third, the results and interpretations may be subject to particular contexts and thus may not be generalized to other industry settings. Despite these limitations, we are confident that this work provides important implications for both researchers and managers. These implications and their significance are summarized in the following concluding remarks.

Primary Findings

By employing extensions of existing models in the context of the web server market, this work has provided much-needed support for the network externalities hypothesis in a new and previously unstudied context. The positive relationship noted between price and market share suggests that network externalities exist and are significant even in markets supporting so-called open standards. While it has been suggested that firms seek to impose their own standards on markets as a form of lock-in (Nilssen, 1992), these findings would suggest that dominant firms can still gain network benefits even if they embrace open standards.

While researchers have acknowledged offering free products as an appropriate tactic in competition among network goods (e.g. Shapiro and Varian, 1998), no prior empirical study has examined the market impact of the entry of a viable competitor offering a free-product. The negative price shock associated with the arrival of a viable free product suggests that market participants may benefit from environmental scanning and considering the pricing tactics and motivation of potential new entrants.

An unanticipated result associated with the entry of the free product is that the price / share relationship did not suffer and may have actually gotten stronger after the entry of the free provider. This result is particularly intriguing as it may demonstrate the hastening of bandwagon effects related to the arrival of the free provider. The vicious cycle identified by Shapiro and Varian (1998) may manifest itself as a migration away from smaller players to those firms with the largest installed base.

A positive price / share relationship was not demonstrated in markets where free products dominate. This suggests that firms seeking to penetrate markets dominated by free goods may have difficulty realizing price-based benefits from positive network externalities. Such firms may be resigned to aggressive competition on quality/features and may only be able to sway customers to the new offering if the value proposition offered by the priced products is clearly seen by consumers as being superior to the dominant, free alternative. These results are noteworthy given the penchant of e-commerce firms to begin offering free services that compete with priced products (Gurley, 1997) and the rise of open-source software products (Cortese, 1998).

Secondary Findings

In a market where more than one standard can be employed, products that support dominant standards were shown to exert a price premium. These results taken together with findings that support the existence of network externalities in open-standard markets, may be particularly interesting for firms that are faced with a choice between investing in an innovative standard vs. supporting a popular standard.

All models suggested that firms that offered a free trial version of their product enjoyed larger price premiums than competitors than did not offer trial versions. Earlier studies examining network externalities in software markets (Gandal, 1994 & 1995; Brynjolfsson and Kemerer, 1996) were conducted in the pre-Internet era and hence could not consider the impact of a global infrastructure for the low-cost distribution of digital products. This study implies that future hedonic models can be improved by considering trialability and that practitioners offering digital products and services would be well advised to consider offering trial versions online (e.g. Nejmeh, 1994).

Future Research

Future research may seek to track the impact and development of markets longitudinally after the entry of viable free alternatives. The economic theory of networks tends to suggest lower entry pricing (in some cases zero pricing), followed by above marginal cost pricing later. However, it remains to be seen what the long-term impact of viable free products are on firms that had previously been able to extract price premiums associated with the size of their installed base. The sources or factors determining when an entrant presents a credible threat are also poorly understood and worthy of future study. Additional modeling techniques may also be explored in order to identify and more accurately map the tipping point in network markets. Also, the viability of the open-source software movement as an alternative to priced-commercial products is worth considering over time. The precise conditions leading to the rise of such movements as well as the success and/or failure of efforts are currently poorly understood. Lastly, the modeling approach used in this research can be applied to other contexts that exhibit network externalities.
 

ENDNOTES

1.  Note that a free product offering differs from shareware products, where a version of the product is freely distributed, but a fee is requested for continued product use.

2. Although the semi-log specification is presented, both dual-log and un-logged models were also tested.  The semi-log form is chosen for presentation based on the appropriateness of the dependent variable distribution and for consistency with earlier research.  For brevity, the results of these additional explorations are not presented here, however results across all models were consistent in terms of significance and direction of influence.

3.  The direct database linking feature in a Web server was included in the test model in an effort to replicate Gandal's (1995) findings that database links are significant in determining consumer valuation among software products.  Security standards (SSL and S-HTTP) were considered given the implied value associated with exchanging files that support a consistent standard.  This sort of exchange is thought to be the same as file exchange, cited by Katz and Shapiro (1992) as a likely source of network externalities in the software market.  The ability to exchange secure data among networked products is contingent upon the use of a common standard, in the same way that the ability to exchange application data is contingent upon the use of a common file format.

4.  The size of the coefficient may be interpreted as a result of the lack of consolidation in the web server industry during the time of the study.

5.  It should be noted that the impact of TIME is negligible.  Further tests (Appendix A) demonstrate that the time indicator could be removed without significantly altering the significance or interpretation of the results.  TIME is included in the model for consistency with earlier research models and because its presence contributes to the overall explanatory power of the model without raising multicollinearity concerns.

6.  Ideally, Propositions 1 through 3 could be tested using time-varying samples from the same market.  However, the Windows web server market had not evolved to the point where a free product dominated.  This necessitated a consideration of a related, yet distinct market that meets the characteristics necessary for Proposition 3.  If Microsoft’s free web server continues growth, then this proposition may one day be tested with Windows market data.  However, it is believed that the markets for Windows and UNIX web servers are similar enough with respect to product characteristics as to warrant comparison of the results of separately modeled data sets.

REFERENCES

1. Airborne, M. What to Look for in a Web Server. SunWorld Online, May 1996.
2. Bakos, Y., and Kemerer, C. Recent Applications of Economic Theory in Information Technology Research. Decision Support Systems, 8, 5 (1992), 365-386.
3. Belsley, D.; Kuh, E.; and Welsch, E. Regression Diagnostics: Identifying Influential Data and Sources of Collinearity. New York, NY: John Wiley & Sons, (1980).
4. Berndt, E. The Practice of Econometrics, Classic and Contemporary. Reading, MA: Addison-Wesley, (1991).
5. Besen, S., and Farrell, J. Choosing How to Compete: Strategies and Tactics in Standardization. Journal of Economic Perspectives, 8, 2 (1994), 117-131.
6. Brynjolfsson, E., and Kemerer, C. Network Externalities in Microcomputer Software: An Econometric Analysis of the Spreadsheet Market. Management Science 42,12 (1996), 1627-1647.
7. Chismar, W.G. and Meier, J. "A Model of Competing Interorganizational Systems and its Application to Airline Reservation Systems," Decision Support Systems, 8, 5 (September 1992), 447-458.
8. Chow, G.C. Technological Change and the Demand for Computers. American Economic Review, 57, 7 (1967), 1117-1130.
9. Christensen, C., The Innovator’s Dilemma. Boston, MA: Harvard Business School Press, (1997).
10. Church, J., and Gandal, N. Complementary Network Externalities and Technological Adoption," International Journal of Industrial Organization, 11, 2 (1993), 239-260.
11. Church, J., and Gandal, N. Integration, Complementary Products, and Variety, Journal of Economics and Management Strategy, 1, 4 (1992), 651-675.
12. Cortese, A. Internet Server Software: A Feather in Apache’s Cap. BusinessWeek, (June 20, 1998).
13. Downes, L., and Mui, C. Unleashing the Killer App : Digital Strategies for Market Dominance. Boston, MA: Harvard Business School Press, (1998).
14. Dybvig, P. H., and Spatt, C. Adoption Externalities as Public Goods. Journal of Public Economics, 20 (1983), 231-247.
15. Economides, N. The Economics of Networks. International Journal of Industrial Organization, 14, 6 (1996), 673-693.
16. Farrell, J., and Saloner, G. Installed Base and Compatibility: Innovation, Product Preannoucements, and Predation. American Economic Review, 76, 5 (1986), 940-955.
17. Farrell, J., and Saloner, G. Standardization, Compatibility, and Innovation. Rand Journal of Economics, 16, 1, (1985), 442-445.
18. Fichman, R., and Kemerer, C. Adoption of Software Engineering Process Innovations: The Case of Object-Orientation. Sloan Management Review, 34, 2 (1993), 7-22.
19. Fisher, F.M., and Shell, M. Taste and Quality Change in the Pure Theory of the True Cost-of-Living Index. in Price Indexes and Quality Change, Griliches, Z. (ed.), Boston, MA: Harvard University Press, (1971), 16-54.
20. Gandal, N. Competing compatibility standards and network externalities in the PC software market. Review of Economics & Statistics, 77, 4 (1995), 599-608.
21. Gandal, N. Hedonic Price Indexes for Spreadsheets and an Empirical Test of the Network Externalities Hypothesis. RAND Journal of Economics, 25, 1 (1994), 160-170.
22. Gordon, R.J. The Measurement of Durable Goods Prices. Chicago, IL: University of Chicago Press, (1993).
23. Graybill, F., and Iyer, H. Regression Analysis: Concepts and Applications, Belmont, CA: Duxbury Press, (1994).
24. Gurley, J. When Software is More Than Just Software. Fortune, 136, 11 (1997), 222-226.
25. Hagel, J. and Armstrong, A. Net Gain, Boston, MA: Harvard Business School Press, (1997).
26. InfoWorld. Web Servers: Are You Being Served? 18, 15 (1996), April 8, 1996.
27. Katz, M., and Shapiro, C. System Competition and Network Effects, Journal of Economic Perspectives, 8, 2 (Spring, 1994), 93-115.
28. Katz, M. and Shapiro, C. Product Introduction with Network Externalities, The Journal of Industrial Economics, 40, 1 (1992), pp. 55-83.
29. Katz, M., and Shapiro, C. Technology Adoption in the Presence of Network Externalities. Journal of Political Economy, 94, 4 (1986), 822-841.
30. Katz, M., and Shapiro, C. Network Externalities, Competition and Compatibility. American Economic Review, 75, 3 (1985), 424-440.
31. Kauffman, R., and Wang, Y. (forthcoming). The Network Externalities Hypothesis and Competitive Network Growth. forthcoming in the Journal of Organizational Computing and Electronic Commerce.
32. Liebowitz, S.J., and Margolis, Steven E. Network Externality: An Uncommon Tragedy. Journal of Economic Perspectives, 8, 2 (1994), 133-150.
33. Lynch, B.; Rao, H.R.; and Lin, W. Economic Analysis of Microcomputer Hardware. Communications of the ACM, 23, 10 (1990), 119-129.
34. Matutes, C., and Regibeau, P. Compatibility and Bundling of Complementary Goods in a Duopoly. Journal of Industrial Economics, 40, 1 (1992), 37-54.
35. McHugh, J. For the Love of Hacking, Forbes, 162, 3, (August 10, 1998), 94-100.
36. Negroponte, N. Being Digital. New York, NY: Alfred A. Knopf, (1995).
37. Nejmeh, B.A. The Internet as a strategic tool for the software enterprise. Communications of the ACM, 37, 11 (1994), 23-27.
38. Nilssen, T. Two Kinds of Consumer Switching Costs. Rand Journal of Economics, 23, 4 (1992), 579-589.
39. Oren, S., and Smith, S. Critical Mass and Tariff Structure in Electronic Communications Markets. Bell Journal of Economics, 12, 2 (1981), 467- 487.
40. Rao, H.R., and Lynch, B.D. Hedonic Price Analysis of Workstation Attributes. Communications of the ACM, 36, 12 (1993), 95-102.
41. Rogers, E.M. Diffusion of Innovations: Fourth Edition, New York, NY: Free Press, (1995).
42. Rohlfs, J. A Theory of Interdependent Demand for a Communications Service, Bell Journal of Economics, 5, 1, (1974), 16-37.
43. Shapiro, C. and Varian, H. Information Rules: A Strategic Guide to the Network Economy, Boston, MA: Harvard Business School Press, (1998).
44. Shurmer, M. An Investigation Into Sources of Network Externalities in the Packaged PC Software Market. Information Economics and Policy, 5, 3 (1993), 231-251.
45. Stein, L.D. How to Set Up and Maintain a World Wide Web Site: The Guide for Information Providers, Reading, MA: Addison Wesley, (1995).
46. U.S. Department of Justice. U.S. vs. Microsoft, Civil Action 98-1232, (1998).
47. Wingfield, N. Netscape Fixes Apache in its Sights. C|net Central, (June 6, 1996).
48. Wired Staff. Encyclopedia of the New Economy, San Francisco, CA: Wired Ventures, Inc., (1998).

Each model presented in Table 2 has been checked for multicolinearity using the Variance Inflation Factor method (VIF), and using the Belsley-Kuh-Welsch diagnostics (Belsley, Kuh, Welsch, 1980), which are standard tests. Values of VIF greater than 10.0 may be considered large enough to suspect serious multicolinearity (Graybill and Iyer, 1994, p. 398). The VIF values for all variables in all models presented in Table 2 did not exceed the threshold of 10.0, nor did condition indexes in any of the models exceed the Belsley heuristic of 30. However, some concern over possible multicolinearity is acknowledged between variables DURATION (the duration in months that a product has been on market) and TIME (the month of a particular observation in the study, 0-19). As further test of the robustness of the models presented, the researchers have investigated the model’s results with the TIME trend removed.

It should be noted that in models (a) and (b), the TIME variable has the smallest coefficient and contributes only marginally to the model. It's removal maintains all significant variables and coefficient directions at an adjusted r-squared value of .740 with TIME removed vs. .769 with TIME included (see Table 3). While the primary variable of interest (SHARE) demonstrates a reduced coefficient (.0546), it remains a highly significant, positive contributor to the model. These results are particularly noteworthy in this context since the web market underwent tremendous growth during the period studied. If the impact of TIME had a strong distorting impact on the significance and direction of coefficients, one might question whether hedonic pricing can produce actuate measurements of market behavior at this phase of the industry's cycle. In this context, the TIME trend does not exert enough influence on the model to raise concern regarding support for the study's propositions. Since some level of significance is attached to TIME, and since it improves the overall fit of the model, the variable remains in the preferred refinement (inclusion of the time trend is also consistent with the approach by Brynjolfsson and Kemerer, 1996; and Gandal 1994).