Coordination versus differentiation in a standards war: 56K modems.

56K modems were introduced under two competing incompatible standards. We show the importance of competition between internet service providers in the adoption process. We show that ISPs were less likely to adopt the technology that more competitors adopted. This result is particularly striking given that industry participants expected coordination on one standard or the other. We speculate about the role of ISP differentiation in preventing the market from achieving standardization until a standard setting organization intervened.

1. Introduction

* We study the adoption of 56K modems by internet service providers (ISPs). Introduced in 1997, 56K modems allowed for data transfer off of the Internet at up to twice the speed of the previous technology at a time when the demand for large files such as graphics became increasingly important. Originally, there were two competing specifications for the standard from two competing consortia, one led by the equipment manufacturer US Robotics, the other by Rockwell. The technologies were functionally identical in the sense that they had the same performance characteristics. However, the technologies were incompatible. If a consumer used one technology and the consumer's ISP used the other, data transfer speed diminished to that of the previous technology, only 33K or 28K.

We focus on understanding the role of competition in adoption by an ISP. We show that ISPs differentiated across technologies rather than coordinating on one technology or the other. Specifically, we show that ISPs were less likely to adopt a technology as more of their local competitors adopted that technology. This differentiation is particularly important because, as we discuss below, it hindered coordination on a single standard technology, which could have provided important benefits.

Theories about standardization discuss the role of competitive choice between standards, but few prominent cases ever permit researchers to garner a close look at behavior during deployment, which we get here. Also, there is very little empirical research to examine how competitive behavior shapes demand for competing standards or vice versa. Data needs are the primary impediment. We rarely observe competition between two comparable technologies played out in more than one market. Even when that occurs, it is often difficult to disentangle the effects of competition from other important effects.

This study's setting is uniquely well suited to meet these requirements. An important feature of the ISP market is that consumers almost always connect to ISPs within their local telephone calling plan. This creates numerous geographically distinct or partially overlapping markets, which leads to geographically dispersed decision making and a variety of competitive interactions. We study over 2,200 ISPs in 2,300 calling areas. Thus, we are able to compare decisions across markets, where a variety of factors shape decision making, such as the competitive and demographic environment and ISP size.

We employ a series of empirical approaches. Simple statistics illustrate a prevalence toward "even splits" in local markets. That is, adopting ISPs were more likely to be evenly split between the two technologies than would be predicted by independent random choice. We also estimate a structurally motivated model of each ISP's choice over adoption of the two technologies as a function of the competitive environment, local demographics, ISP characteristics, and ISP decision making across multiple markets. We capture the influence of an ISP's adoption decision on its rivals in a discrete game of imperfect information as suggested in Seim (2004).

Estimating a discrete game has well-known problems with endogeneity and multiple equilibria. Fortunately, our data contain a great deal of useful variation. ISPs have different technological characteristics and face different demographic characteristics due to imperfectly overlapping coverage areas. We use this exogenous variation to predict the number of competitors that a given ISP faces. The great asymmetry in the data means that our models typically have a unique equilibrium. Throughout we conclude that an ISP is less likely to adopt a technology as more of its competitors do so. We are particularly sensitive to the robustness of our inferences to unobservable errors at ISPs or at locations, so we pursue a variety of strategies for estimation in the presence of such errors.

Understanding the deployment of 56K modems is also interesting because the "standards war" for 56K was well publicized. While many contemporary press reports discussed how modem makers competed fiercely for adoption by the earliest choosers, few had substantial data. Citing such accounts, Shapiro and Varian (1999, pp. 267-270) feature the case prominently in their discussion of strategic behavior and consortia development prior to deployment, but, again, they do not present any evidence about actual adoption. Similarly, contemporary press accounts tend to cover announcements from firms, not the deployment in each local area. No research has closely examined the deployment decision of service providers, as we do.

The events are also interesting because they end in intervention from the International Telecommunications Union, a quasi-government agency. Before ITU intervention, this experience appeared to be an example of "coordination failure." That is, there was a benefit to coordinating ISPs and consumers on a single standard as quickly as possible, but market actors failed to quickly standardize. Market participants expected that standardization would arise because it was in users' interests to do so. The popular standard would have more ISPs servicing it, which ensured consumers of high-quality, low-hassle, low-price service into the future. However, coordination did not arise in the first year of competition. Not only did the two technologies maintain relatively similar levels of ISP commitments, but overall sales to consumers and ISPs were well below what the market could have supported. Sales increased only after the ITU introduced a third incompatible standard as a new focal point. The new standard quickly gained market acceptance, and high industry sales followed.

Did competitive rivalry among ISPs contribute to the market's inability to coordinate on a standard unaided? In a final, more speculative, section of the article, we argue that the standards war was prolonged by the combination of the market's structure and the behavior it induced, i.e., the incentive to differentiate locally. This discussion directs further attention at issues not highlighted in the applied literature on standardization, such as the role of service provider competition in a standards war, and the role of standard-setting organizations when the specification for the potential standard is not fixed.

2. Related literature

* While user choice between alternatives plays a prominent role in many models of standardization, there are few empirical studies for characterizing its effect. Those that do so focused on either the decision of whether to adopt a standard or not, or the decision of which standard to adopt, but it has not dealt with decisions linking choice between two standards and nonadoption.

A few prior studies of competition in technology adoption provide us with general approaches for measuring competitive incentives. For instance, Klepper (2002) uses exit patterns, comparing very competitive and oligopolistic markets, to suggest a strong role for competition in cost-reducing technology adoption in a number of manufacturing industries. Genesove (1999) provides a study of the adoption of offset printing by newspapers and argues that firms in more competitive markets adopted earlier. Mulligan and Llinares (2003) show that ski lifts were less likely to adopt quality-enhancing technology when local competitors had done so.

We borrow broadly from the general approach of empirical studies of technology adoption in network industries. (1) For example, Saloner and Shepard (1995) show the existence of network effects in bank service by showing that banks with more consumers adopt ATM networks earlier. As in our article, they infer consumer behavior from observing decisions by firms in different locations. Gowrisankaran and Stavins (2004) and Ackerberg and Gowrisankaran (2006) look at the adoption of automated clearing house technology by banks. As we do here, they exploit overlapping local geographic markets for important variation. However, they use a very different structural model of adoption incentives.

There are a few articles that empirically model horizontal competition between two standards. For example, Dranove and Gandal (2003) argue that the introduction of the DIVX standard slowed down the acceptance of DVD technology. Park (2004) and Ohashi (2003) study the standards war between VHS and Beta in the VCR market. Using market-level data on quantities and prices, they focus on the role of installed base. Gandal, Kende, and Rob (2000) study network effects between producers of compact disks and producers of CD players. They too find evidence of the interaction between software and hardware.