Published by EH.NET (March 2003)


Peter Lewin, editor, The Economics of QWERTY: History, Theory, and Policy — Essays by Stan J. Liebowitz and Stephen E. Margolis. New York: New York University Press, 2002. xi + 266 pp. $50.00 (cloth), ISBN: 0-8147-5178-4.

Reviewed for EH.NET by Richard J. Sullivan, Banking Studies and Structure, Federal Reserve Bank of Kansas City.

A number of historical and theoretical studies published in the 1980s suggested that if network effects and economies of scale are significant, with information goods offered as an example, then free markets could lead to inferior technologies becoming entrenched. In the face of this possible market failure, some studies have suggested that government policy may improve social welfare. This book reprints research and essays published by Stan Liebowitz and Stephen Margolis over the course of the 1990s that challenges both the historical and theoretical grounds for these assertions.

Liebowitz and Margolis are masters of neoclassical theory. The book mixes theoretical and historical analysis, effectively confronting one with the other, and exposing both of their weaknesses and strengths. Much of their effort is aimed at Paul David’s work on path dependence. While Liebowitz and Margolis succeed in clarifying a number of issues regarding network economics, they never directly address David’s conception of path dependence. Despite this limitation, their work does pose many interesting questions regarding the role of history in economic analysis.

Economic historians will be most interested in Chapters 2 and 5. In Chapter 2, Liebowitz and Margolis attack Paul David’s claim that the QWERTY typewriter layout is an example of lock-in to technological inferiority (see, for example, David (1985)). David asserts that the Dvorak keyboard layout, developed after the QWERTY layout was well established, is technically superior but historical circumstances and market processes have prevented its adoption. Typists do not train on Dvorak because no employer uses it, and no employer uses it because there are no trained typists. The social benefit of switching standards may be greater than the social cost, but unless all users switch, the social benefit is unattainable. It is too costly to coordinate all users to make the switch, the result of an externality that causes market failure.

Liebowitz and Margolis question this reasoning. Theoretically, they argue that this is an externality of the type analyzed by Coase (1960). As such, interested parties can overcome the coordination failure. For example, an entrepreneur could patent a superior standard and market it in ways that overcome the lock-in of the inferior standard. Liebowitz and Margolis also attack David based on the historical record. David relies on tests conducted by the U.S. Navy that compared the QWERTY and Dvorak keyboard layout. Liebowitz and Margolis point to several flaws in the test methodology and note that Dvorak was involved in the tests, suggesting bias. Later, more sound testing showed limited or no advantage to the Dvorak keyboard. Liebowitz and Margolis also point to many events in the late 1800s that featured competition between various keyboard layouts. The prospective rewards for developing a superior keyboard were sufficiently high to make manufacturers actively seek the best layout.

In Chapter 5 Liebowitz and Margolis more directly criticize the concept of path dependence. According to them the ” ‘new’ positive feedback economics” argues that a “minor or fleeting advantage or . . . lead for some technology, product, or standard can have important and irreversible influences on the ultimate market allocation of resources, even in a world characterized by voluntary decisions and individually maximizing behavior.” Lock-in of inferior technology implies “marginal adjustments of individual agents may not offer the assurance of optimization or the revision of sub-optimal outcomes. In turn, this implies markets fail” (p. 96).

Liebowitz and Margolis identify access to information as a key element of whether path dependence is accurate and market failure important. They define three forms of path dependence distinguished by usable information about the future. First degree path dependence is where an optimal decision is made based on perfect foresight. Second degree path dependence is where a sub-optimal decision is made, but the mistake is due to imperfect foresight. The sub-optimal allocation of resources persists simply because switching costs are high.

Third-degree path dependence occurs when a sub-optimal decision is made despite having information that a superior choice is available. Some conditions, such as lack of a market, agents who are underrepresented, or coordination problems must be present for third-degree path dependence to occur. Coordination problems, for example, might cause us to be using the VHS video standard when the beta standard is superior because at the time when a choice had to be made it was too costly to know the preferences of all VCR consumers.

Only third-degree path dependence implies market failure because it is ex ante inefficient. However, Liebowitz and Margolis assert that there are no credible examples of third-degree path dependence. They argue that markets offer a number of responses to overcome lock-in (brand names, patents, or early market share commitments) and rearrange incentives so that someone captures the rents from the new technology and uses it to entice early adopters. With enough adoption, superiority of the new technology should become evident.

Chapters 3, 4 and 6 use more formal modeling to analyze network externalities. In Chapter 3 Liebowitz and Margolis distinguish between network effects and network externalities. Network effects occur when the number of consumers of a good influences the utility that an individual derives from a good. Network externalities occur when markets characterized by network effects fail to allocate resources properly. Liebowitz and Margolis argue that network effects may be common but network externalities are rare.

Chapter 4 establishes that network effects are not the cause of network externalities. Network effects are instead caused by peculiarities of production costs: if marginal costs are sufficiently low then there will be too little production of network services. Liebowitz and Margolis note that this implies that network externalities are not new. They involve conventional problems like natural monopoly and production externalities.

Chapter 6 presents a model of the choice of standards and draws conclusions regarding antitrust policy. The model has production technology (a supply curve for a product) and incorporates network effects into the demand for the product. By applying the same setup to two different standards, we can use the model to analyze the choice of standards. Depending on the relative slopes of demand and supply, outcomes can either be that one standard dominates or that both standards coexist. Importantly, it is not the presence of network effects that leads to this outcome. The element that determines whether one or both standards survive is the slope of the supply curve, that is, the underlying production technologies.

Liebowitz and Margolis go on to review three cases of standards competition (typewriter keyboards, VCR standards, and Mac versus IBM). In each case they argue that the superior technology won the competition. The implication is that antitrust should not address standards choice — the market can work to find the best technology. Antitrust could handicap an otherwise vibrant industry.

Chapters 7 and 8 are essays that define path dependence and network externalities, useful to those looking for a brief introduction to Liebowitz and Margolis’s approach to the subject. Much of the material is contained in other chapters, although the authors do place emphasis on remediability: third degree path dependence implies the existence of remediable inefficiencies.

Chapters 9 and 10 turn to more contemporary issues by analyzing Microsoft’s business practices. Chapter 9 looks in detail at Microsoft’s ability to leverage software sales off its dominance in operating systems, its strategy of bundling software products together, its effect on innovation, and its insistence on controlling icons placed on personal computers. Liebowitz and Margolis argue that in each case that Microsoft would not be likely to gain much competitive advantage. Chapter 10 looks at the record of sales for spreadsheets, personal finance software, and browsers. After a review of price, quality, market shares, and competition for software compatible on Apple computers, Liebowitz and Margolis find no reason to think that Microsoft took advantage of a dominance in operating systems to gain in other software markets.

The editor, Peter Lewin, contributes the first and last chapters. The first chapter is a useful overview, but the most original element of the chapter is a review of economic policy. Because of rapid changes in their market, products like information goods do not lend themselves well to concepts of static efficiency. They instead require a dynamic context at the level of economic institutions; for example, where we judge whether the process of competition is adequate rather than whether a particular price-quantity combination is momentarily efficient.

The final chapter summarizes the state of the debate in the economics of QWERTY. Lewin feels that the David/Liebowitz and Margolis debate is about policy. Lewin characterizes David’s historical examples as corresponding to second-degree path dependence, and his policy prescriptions involve using government intervention to slow markets from committing to a technology before information is available for the appropriate decision. The issue is what kind of policy action is best at different points of a market process.

But appropriate policy requires that government know when future knowledge will arrive. Or if economic agents have the information, they cannot coordinate in a way to exploit superior alternatives. Lewin feels that this “knowledge problem” is the crux of Liebowitz and Margolis’s arguments. They are skeptical that government would have superior information, and they believe that market processes can overcome coordination issues or other barriers that might prevent adoption of superior alternatives.

Where does Liebowitz and Margolis’s analysis leave the concept of path dependence for economic historians? With individual decision making in a free market framework, the neoclassical focus on institutional context, resource endowments, underlying technologies, and optimizing behavior, provides a powerful tool for understanding resource allocation. Yet for historical analysis, neoclassical theory falls short because it usually takes institutions, endowments, and technologies as given, which is unsuitable as the time frame for analysis expands.

An appealing feature of the concept of path dependence is that random events have been important to institutions and technology. Heroic individuals have played an important role in the design of governing institutions. Random discovery has contributed significantly to our basic knowledge. In other works, we live in a world where information reveals itself in an uncertain way. Optimizing behavior may help in directing society towards the best use of new information, but it does so imperfectly. The very existence of uncertainty affects behavior.

For example, path dependence has been useful to applied finance in understanding a puzzling tendency for managers to reject investment projects that have the highest perceived payoff. Instead, they often choose a project with a lower payoff, but one that allows for future flexibility. Real option theory helps to explain this decision. These managers preserve a real option rather than committing based solely on maximized present value. They know they have incomplete information at any point in time and desire flexibility because they know that new information will be forthcoming. For example, a firm might invest in a new production plant but choose to make it smaller than might be warranted based on projections of demand. But because the projections are uncertain, there is value in the option to build additional capacity in the future, and building a small plant preserves the option. Rather than committing investment over a long period of time, these managers make a sequence of decisions based on the information at hand but expecting future information to assist in making further decisions.

This example highlights the most important failing in Liebowitz and Margolis’s critique of Paul David’s work. Their critique never really addresses the central theoretical features of David’s conception of path dependence. David argues for the legitimacy of stochastic economic models with multiple equilibria (potential outcomes). Liebowitz and Margolis forcefully and effectively argue that economic processes can move an economy out of clearly undesirable situations. But underlying their analysis is a model with a single, global “best” outcome. The issue then becomes whether we have attained the best outcome, and if not, how can we get there.

David favors stochastic models because they allow for contingent processes, where outcomes are contingent upon certain historical factors. Economic change is like a branching process. At a point in time, we have several alternative choices to make. Once choices are made, we face a new set of alternatives (branches in a decision tree). Such a contingent process may or may not be inefficient and there is nothing that guarantees a particular outcome.

Under some conditions, path dependent processes can lead to outcomes that are inefficient. Feedback mechanisms, where conditions established in the past inform today’s decisions, occur for several reasons. Businesses invest in durable equipment and commit to particular technology. Consumers develop habits. Products are sometimes more useful if others use it, so that a consumer will most desire the ones that others have already purchased. These show up in self-fulfilling expectations (“I’ll get this because everyone else is”). As a result, the process systematically leads to an inferior outcome.

Some of the same forces that lead to market failure (positive economies of scale, network effects) also lead to path dependent processes. Thus while path dependence does not necessarily lead to inefficient outcomes, it can. David objects to Liebowitz and Margolis’s approach because outcome (efficient or not), rather than process, attains special significance. Moreover, their second-degree path dependence can occur because of mistakes (though informed mistakes). Yet the ahistorical approach they favor means that the nature of the mistake is never in question.

Liebowitz and Margolis do not dispute that economic theory can predict inefficient outcomes and correctly note that an important issue is whether these predictions are empirically relevant. The major problem with their empirical studies is that correctly specifying the appropriate counterfactual situation is very difficult. To properly measure how well off would we be if we had committed to another technology requires identification of advances that would take place in a technology that was never seriously pursued. In addition it must account for expenses incurred to mitigate problems that arise from the technology that was chosen. As a result, Liebowitz and Margolis’s efforts at debunking the QWERTY keyboard and other examples of third-degree path dependence have not been totally convincing.

Much of the debate surrounding path dependence has centered on efficiency. There has also been much discussion of the related issue of what to do about it. Before turning to economic policy, we need to consider the topic of allocation, which has been relatively neglected.

Neoclassical theory can show that trade can make people better off, but under general conditions many potential outcomes of a trading process could occur. Each outcome may be better than where everyone started, but exactly what outcome obtains is more difficult to predict. The question is the process of trading. Reasonable models that are path dependent might describe the process. The outcome may be efficient, but how we got there is an interesting question that path dependence can usefully describe.

This is an example of an allocation process, about which Liebowitz and Margolis have said little, but that David argues is an important subject of study. Self-reinforcing processes can lead to sustained advantages of opportunity and outcomes for some social groups. Understanding such a process may be useful if improving the welfare of a particular group is a social goal. Moreover, the process may lead to an efficient outcome, but there is no reason to necessarily think that we may approve of the particular allocation that obtains. Understanding the path dependent process may help to design ways to avoid undesirable allocations.

Many economists will be uncomfortable with David’s normative discussion of allocative outcomes because of the difficulty of making utility comparisons among individuals. David, in turn, objects to this impulse because static welfare analysis implicitly favors the status quo. Suppose the status quo is a result of a path dependent process that leads to some outcome that is in some sense objectionable. What justifies special consideration to those who could be harmed by a remedy? Welfare questions become much more complex if dynamic processes are path dependent even if there is little evidence of inefficiency.

Issues surrounding efficiency as well as allocation raise the question of economic policy. Can the government do something to improve outcomes where path dependence is an important element of the process?

Liebowitz and Margolis attack those who use lock-in as justifying a remedy where government chooses technological winners and losers. Liebowitz and Margolis have reason to be skeptical. Markets have self-correcting mechanisms and incentives can lead participants to adopt efficient procedures and develop superior substitute products. Poorly designed policies can delay corrective steps and create monopoly. The banking industry, for example, has a long history of government policies that may have impeded progress towards efficiency, such as interest rate ceilings, entry restrictions and product limitations. New technology or information goods may have some special qualities that lead to market failure. But if the market chooses poorly, why does it follow that the government can do better?

David has not recommended an economic policy that would pick a particular technology. Instead he favors government action that aids the market in making an appropriate choice. Examples I would point to include government-sponsored trials for the effectiveness of drugs or government funded basic research. Expanding these programs may make up for difficulties that the market has in identifying the best technology. But the potential pitfalls of poor government policy place the onus upon advocates of path dependence to bear a heavy burden of proof in proposing economic policy.

To make path dependence useful to economic analysis, and to help identify good policy options, some methodology must be specified to show how its outcome differs substantially from that of the neoclassical framework. David’s writings allow a description of path dependence and its appealing features. But the concept remains too sketchy to allow independent application. The real option example offered above attempts to operationalize the concept of path dependence in a manner that bridges the gap between David’s and Liebowitz and Margolis’s points of view. David may dismiss it, perhaps because it too closely resembles neoclassical optimization. But it is up to the defenders of path dependence to provide more detailed specification of the concept and an accessible methodology before further progress on its value can be made.

David’s emphasis on path dependence is part of a larger effort to develop a theory of economic change, with history as a central element. Economic historians therefore have a significant stake in the outcome. While Liebowitz and Margolis do not directly address David’s efforts, they do make important contributions regarding both the theoretical and empirical issues surrounding path dependence and network economics. This book brings their major writings on the topics conveniently together. It is useful to both economic historians as well as others interested in the “new” economy.


Ronald H. Coase, “The Problem of Social Cost,” Journal of Law and Economics, 1960, vol. 17, pp. 357-376.

Paul A. David, “Clio and the Economics of QWERTY,” American Economic Review, 1985, vol. 75, pp. 332-337.

Paul A. David, “Path Dependence and the Quest for Historical Economics: One More Chorus of the Ballad of QWERTY,” University of Oxford, Discussion Papers in Economic and Social History, Number 20, Nov.1997, (

Paul A. David, “At Last, a Remedy for Chronic QWERTY-skepticism!” Paper prepared for presentation at the European Summer School in Industrial Dynamics (ESSID), held at Institute d’Etudes Scientifique de Carg?se (Corse), France 5th – 12th September 1999, (

Richard J. Sullivan has published several articles on the history of patenting and technology. His latest work is entitled “A Guide to the ATM and Debit Card Industry,” which includes a brief history of the industry. It is available at