This paper, in many ways, is a response to Professor Radzicki's 2003 call in this journal for intensified use of systems dynamics and computer simulation modeling to rigorously address economic change as the alteration of social and economic structures themselves and not just changes within existing structures. In this vein, this paper aims to broaden the understanding of the process of technological change and standardization through an evolutionary approach, which avoids the determinism of conventional orthodox models of technology diffusion and standardization. A model is developed to explore how change between competing technological standards can be initiated endogenously within an industry. We do this by first focusing on the characteristics of challenging technological Innovations themselves. We then focus on the role of the changing preferences of agents adopting the technologies and how preference evolution can also induce technological change. Technological change at the industry level is important because it can influence the development of larger macroeconomic structures and institutions such as corporations, political regulatory structures, growth rates, and the accumulation and structure of capital. To achieve our goals we have developed an agent-based model (ABM)1 using distributed artificial intelligence (DAI) concepts drawn from the general methodology of social simulation.

The model also helps expand our understanding of path-dependent technological evolution, building upon Arthur 1988 and 1990, David 1985, and others' conceptualizations of increasing returns to adoption and technological lock-in. Our model intends to extend W. Brian Arthur's basic formulation beyond the one-time competition between two technologies to a vision of technological change as an ongoing evolutionary succession of technological standards over an infinite time horizon. Within this framework we examine hypotheses about how change is induced. The modeling allows us to recognize an apparent paradox in the increasing returns and lock-in conceptualization, mainly the fact that while increasing returns can drive an industry toward technological standardization, their presence always holds out the possibility of the emergence of a new technological alternative which can destabilize the lock-in. Technological lock-in can only be partially understood from the fundamentally new institutionalist perspective of increasing returns to scale. Explaining longer term technological stability in industries appears to require explanations beyond the new institutional arguments about decision making. Instead, the old institutional emphasis on ceremonial and learned behaviors and the emergence of durable institutional structures have greater explanatory power. While increasing returns modeling has advanced formal understanding, it is argued that the thinking of institutional and Post Keynesian authors like Thorstein Veblen, Gunnar Myrdal, John R. Commons, Clarence Ayres, and Nicholas Kaldor are still fundamental to any complete understanding of technological lock-in and stability.

Technology Change and Diffusion Modeling

This paper begins by extending the new institutionalist approach to technological competition in the presence of increasing return to scale. The main differences between the orthodox, or neoclassical, approach to the analysis of technological change and the approach adopted here basically arises from the objections of evolutionary and institutional economists to the way in which the (aggregate) production function is used by neoclassical economists and their apparent inability to explain the processes of technological change (Nelson and Winter 1974, 1977, 1982; Dosi 1982; Dosi et al. 1988). Thus, while the neoclassical approach portrays technological change as a simple change in the information available on the relationship between the economy's inputs and outputs (Stoneman 1983; Gomulka 1990), the evolutionary approach considers technological change to be the result of a self-referential process of evolution influenced by the prevailing economic, social, and political institutions. Technology, in this sense, is embedded within a physical infrastructure and set of social institutions that can generate complex "circular and cumulative causation" and path-dependent technological adoption patterns (Myrdal 1957). Accordingly, technological development should be understood as a process of evolution in which alternative technologies compete with one another and with the pre-existing dominant technology with considerable uncertainty at the outset about who the winners will be (Nelson and Winter 1982). Given that fundamental uncertainty is intrinsic to the process of technological decision making (Davidson 1982-1983, 1991), the processes are considered nonergodic and path dependent. Furthermore, the assumption of rational maximizing behavior must be replaced by a search for profit "in the dark" through heuristic search routines that emerge through complex interactions with existing technological infrastructure and social institutions. Decision making is furthermore influenced by ceremonial behavior, heuristics, custom, habit, and myth (Veblen 1994; Nelson and Winter 1982; Commons 1931) that can condition these search routines. As a result, there is no single welfare-maximizing equilibrium but rather a plurality of outcomes or possible equilibria. We evaluate the implications of these observations in the discussion section of the paper. Finally, in evolutionary models the structure, including institutions, is often made explicit so that its role in the process of technological change can be studied (Lipsey and Carlaw 1998).

Technology diffusion can be defined as a process whereby innovations (products, processes, and/or management techniques) propagate within and between economies (Stoneman 1986). Empirical research has shown that over time, the diffusion of new technologies follows a predictable pattern, represented graphically by an S-shaped curve (Griliches 1957; Mansfield 1961; Davies 1979; Gort and Keppler 1982), which traces the rate of adoption over time (Mansfield 1961, 1968). Diffusion research has historically focused on explaining the rate and the order in which innovations are adopted. Recent research into increasing returns to scale, or positive feedback, on the dynamics of technology diffusion has received growing attention in the last decade (Arthur 1989, 1990, 1994), although the fundamental concepts are deeply rooted in institutional thinking. Arthur has written extensively (1988,1990) about being influenced by institutional and Post Keynesian economists (including Kaldor 1981, Myrdal 1957, and David 1985) who emphasize the importance of positive feedback loops, increasing returns, and path dependency in explaining evolutionary economic behavior (Radzicki 2003).

While an S-shaped curve illustrates the rate of diffusion over time, it can also be used to describe how the performance of a technology improves relative to the effort put into its development and commercialization (Foster 1986, 96). This is illustrated in figure
1. As the figure shows, returns are not constant over the period of adoption and, after a point of inflection, the possible improvements in performance are progressively smaller (Moreau 1999,9; Laffond et al. 1999; Loch and Huberman 1999,12). In effect, technology improvement moves from a period of increasing returns in the lower and center segments of the S-curve to decreasing returns in the upper segment. Orthodox economic analysis tends to focus on the decreasing returns segment at the top of the curve (i.e., the long-term equilibrium), but the preceding period of increasing returns has been shown to have important impacts on the outcomes of technological competition (Arthur 1989; Schilling 1998). Ultimately, however, the increasing returns process is limited by the decreasing returns at the top of the S-curve, which provide the negative feedback that limits exponential growth as markets become saturated.

Recent work has focused on a handful of increasing returns mechanisms that include broad categories of scale, learning, network, and information economies or effects. Traditional scale economies arise from spreading large initial investments in research, development, and capital assets over increasing unit output. As greater production experience is acquired, manufacturers learn how to produce additional units more cheaply (learning by doing) (Arrow 1962). As greater experience is also acquired in the use of the technology, users' productivity increases (learning by using) (Sheshinsky 1967). Further positive externalities arise because the physical and informational networks in which technologies are embedded can grow more valuable to users as they increase in size (Katz and Shapiro 1985, 1986a, b; Farrell and Saloner 1986a, b; Economides 1996). Increased adoption furthermore lowers uncertainty and information search costs (Blackman 1999), reducing perceived risks to adoption, and can create bandwagon effects (Leibenstein 1950; Veblen 1994). These diverse increasing returns mechanisms combine to create powerful positive feedback loops that can dramatically alter technological performance as the adoption process proceeds.

Many authors have noted that these dynamics have important implications for economic outcomes (Young 1928; Arthur 1989). When increasing returns to adoption exist, the same distribution of technologies and user preferences can lead to different structures of results, depending on die initial conditions and the sequencing of adoption decisions (Economides 1996,26). Whereas some markets may be explained by current supply and demand, it is not possible to fully understand markets subject to positive feedback without knowing the historical pattern of technology adoption (Jaffe et al. 2000, 42; Economides 1996, 26), that is to say, "history matters" (David 1985, 1997). Again, these in many ways are modern rediscoveries of traditional institutionalist thinking which has long argued that increasing returns and concomitant institutional evolution strongly impacts future outcomes (Hodgson 1993; Buchanan and Yoon 1995). Given this, a technology's early superiority is no guarantee of long-term suitability (David 1989; Cowan 1990; Nelson 1994), and apparently inferior designs can become locked in through a path-dependent process in which circumstantial events can decide the winning alternative (David 1985, 1997). In these circumstances it has been argued that a "technology is not chosen because it is efficient, but becomes efficient because it has been chosen" (Rip and Kemp 1998, 353), reflecting the circular and cumulative causation endemic in technology adoption decisions.

The multiplicity of potential outcomes, or "stable attractors" in the vocabulary of systems analysis (Prigogine 1993), can be interpreted as spontaneous standards or dominant designs when expressed in terms of market share (Abernathy and Utterback 1978). The establishment of a standard can be understood as an end point or some advanced state in the diffusion process. The concepts of "dominant design" and "standard" are frequently used interchangeably in the literature (Afuah 1998; Schilling 1998), although the notion of dominant design tends to be broader and to some extent subsumes that of a standard (Suarez and Utterback 1995,417). A standard refers to a set of technical specifications to which the producer adheres, whether tacitly or as a result of a formal agreement (David 1987). The process by which a dominant design or standard emerges has often been treated as a "black box," in that a wide spectrum of factors which are difficult to identify and measure can interact (Lee et al. 1995; Suarez and Utterback 1995). In short, spontaneous standards emerge as a result of internal market processes and not as the result of coordinated action by market participants.

These dynamics are important because standards and/or dominant designs can become locked in for extended periods. The majority of technological lock-in models focus on the competition between two competing technologies and end with one technology capturing 100 percent of the market share (Arthur 1989).2 However, less attention has been given to the existence of previous standards or the conditions under which new technologies are able to displace old ones in a technological succession (Windrum and Birchenhall 2000). Most models of technological lock-in have portrayed the technology selection process as an "all or nothing" or "winner-takes-all" dynamic (Abrahamson and Rosenkopf 1997). It is rare, however, for a technology standard to achieve absolute diffusion, or 100 percent market share. Often multiple standard coexist, with technological alternatives surviving in market niches (Grubler 1990; Dalle 1995; Freeman 1996). This general shortcoming of both diffusion and lock-in models has been highlighted in various studies over the last decade (Schilling 2002,395; Jaffe et al. 2000, 41; Windrum and Birchenhall 1998, 112; David 1997, 36; Nelson 1994; David and Greenstein 1990,8). It seems clear that, rather than the deterministic view of conventional models of technology lock-in, an expanded formal view of the process of technology diffusion and standardization is needed which takes into account the possibility of a technological succession,3 understood as a series of replacements of old technologies by new technologies performing the same basic function (Grubler 1990, 1991).

Transitions between Technological Standards

Experience shows that no standard remains in place in an industry indefinitely (Abrahamson and Rosenkopf 1997; Ruttan 1997; Witt 1997; Grubler 1990; Ausubel 1989). On the contrary, if a long enough time horizon is taken, one sees a succession of standards, a dynamic of transition between multiple semi-stable equilibria. This succession has been recognized as a series of technology cycles (Anderson and Tushman 1990) which begin with a technological discontinuity, triggered by the emergence of a breakthrough innovation that significantly advances the technological state of the art. According to Richard Foster (1986) diis discontinuity can be represented as a "jump" between the two S-curves of the competing technologies.

In neoclassical economic formulations, technology is understood as a production function that defines the conversion of inputs into outputs (Sahal 1985). Alternative technologies simply produce different output sets and quantities from a given set of inputs. Competitive, fully informed firms operating in a perfect market can easily (cosdessly) identify any technological innovation that improves profitability; that is, any technology that produces more or better outputs from the given input set. Given this situation, rational agents will immediately adopt and install any superior innovation that enters the market. The neoclassical models thus envision a smooth transition from one technology to the next and tend to negate the possibility of technological lock-in and inefficient standardization (Arthur 1987).

If neoclassical mechanisms held, then there would be no possibility of technological lock-in. Empirical studies show, however, that normally an improvement in technology performance of substantial magnitude is required to displace an incumbent technological standard (Foster 1986). Practitioners have quantified a rule-of-thumb need of a ten-times or orderof-magnitude increase in performance (Grove 1996; Drucker 1993) to foster such a transition. Few empirical studies have tried to quantify the required performance increase in specific industries or historical circumstances, but one frequently cited example concluded that a 20-30 percent performance improvement was insufficient to cause a transition from the QWERTY typing standard to the Dvorak keyboard (David 1985).

One criticism of these works is the implication that only one performance variable, such as horsepower or processor speed, is of interest. The adoption decision thus focuses on the evaluation of a single criterion, frequently the expected return on investment. However, it is widely acknowledged that the performance of a technology is a multi-dimensional construct (Anderson and Tushman 1990, 627; Foray and GrObler 1990; Rogers 1995,206; Kemp 1997,88; Christensen 1997; Windrum and Birchenhall 1998, 114; Nelson 2000, 70). A formulation that recognizes technological products as an assemblage or package of performance attributes has been proposed as a more realistic alternative. The automobile, for example, is valued by users for many attributes including aesthetics, power, comfort, efficiency, and so on. By recognizing that new innovations can not only improve current performance but also can provide new functionality that the incumbent standard lacks opens new avenues for diversity and change (Christensen 1998). Cellular telephony, for example, is a case where a new technology provided traditional attributes of wire-line telephones but also offered the option of mobility that fixed telephony could not. The new functionality can potentially draw in enough adopters to drive the technology into self-sustaining, increasing returns-driven expansion, initiating a technological transition.

These approaches provide insight into how a technological transition may begin, but each dominantly addresses changes in the technologies themselves. Less effort has been focused on how changes in the large context, and specifically user preferences, can initiate a technological transition (Katz and Shapiro 1985, 426; Farrell and Saloner 1986, 941; David and Greenstein 1990, 7; Loch and Huberman 1999; Mulder et al. 1999, 9). If we accept that adopters' decisions are influenced by multiple preferences, and not just expected returns, and that preferences are in turn influenced by custom, routine, and institutions, then our understanding of technological change is incomplete without the potential for changes in preferences to initiate a transition.

A Technological Change Model

This section elaborates on a model that integrates the observations made in the previous sections. Unlike conventional models of technology diffusion and standardization, which focus on the rate at which one new technology diffuses until it is fully adopted, our model enables a broader approximation of the process of technological change by focusing on the extension of the diffusion of multiple alternative technologies and their possible standardization. The proposed model makes it possible to show in greater detail than conventional "all-or-nothing" models, which focus on the competition between just two alternatives, that industries are characteristically divided at all times between multiple available alternatives, all with different levels of adoption at each moment in time. To do so we rely on simulation as opposed to more traditional analytic modeling approaches.

A primary limitation of analytical models used in orthodox economic approaches, beyond their unrealistic simplifying assumptions, is that they solve for a single equilibrium solution. As discussed, however, the diffusion of technological innovations in a competitive environment characterized by increasing returns to scale provides for multiple outcomes or equilibria. The recognition that social and economic phenomena frequently exhibit characteristics typical of complex systems-significant nonlinearity among them-is a challenge to traditional research methods (Holland 1998; Epstein and Axtell 1996; Latané 1996; Gilbert 1995). Simulation is one of the only tools that allow the multiple "experiments" needed to illuminate the dynamics of a system. As a result, simulation has become a popular means of exploring complex natural and social systems in the last few years (Hannerman and Patrick 1997). Simulation has been put forward as a new way of conducting research, a "third scientific discipline" (Ilgen and Hulin 2000; Axelrod 1997) that complements and is built on the traditional methods of induction and deduction. An important branch of simulation in the social sciences is agent-based modeling (ABM), which is characterized by a number of autonomous agents which interact with among each other and with their environment with little or no central coordination (Conte et al. 1997; Epstein and Axtell 1996; Gilbert and Troitzsch 1999; Weiss 1999). Thus, the emergent properties of an ABM (i.e., the system's macro-behavior) are the results of bottom-up processes (arising from micro-level interactions between agents in the system) rather than of top-down processes. According to Wooldridge and Jennins 1995, in ABM the agents are computational processes that are characterized by (1) their autonomy, in that they control their own actions; (2) social abilities, that is, the agents interact with one another by means of some kind of "language"; (3) reactivity, in that the agents can perceive their environment and respond to it; and (4) proactivity, in that they are able to carry out actions in order to achieve an objective.

Clearly the adopters of technological innovations exhibit these characteristics to a large extent, making ABM an appropriate methodology for studying emergent phenomena in technology markets. It should be highlighted that although ABM uses simulation, its goal is not necessarily to represent a specific empirical application precisely but rather to give a more detailed understanding of the fundamental processes that may emerge in various applications. Given this aim, what is important is the simplicity of the assumptions and not a detailed representation of a particular reality (Axelrod 1997). Building on the ABM research methodology, this section describes the model put forward to achieve the aims and compare the hypotheses of the investigation. Its calibration, verification, and validation are also discussed.

Assumptions of the Model

To construct our model we expanded the scope of the so-called "informational cascade" models (Bannerjee 1992; De Vany and Walls 1996; Bikhchandani et al. 1992, 1998). First, it should be noted that new technologies appear on die market in a specific variety of forms, which, when they are sold directly to firms, may be interpreted as process innovations. Adopting a technology implies an initial choice of one of its multiple variants. When a number of "early adopters" decide to try out a variant of the technology instead of the established version, and they find it to be superior, the so-called "bandwagon" effect can take place (liebenstien 1950; Veblen 1994), whereby later adopters follow the early adopters' decision without having made the same investment in learning from experience. These "informational cascades" are defined as situations in which it is optimal for an individual, having observed the actions of those going before, to follow the behavior of the preceding individual without considering his own information. The concept is similar to that of replicator dynamics (Schuster and Sigmund 1983), a pattern which repeats itself in numerous evolutionary phenomena. Network externalities and other increasing returns to adoption reinforce this cumulative causation effect (Myrdal 1957).

Before deciding whether or not to adopt a technology alternative, the potential users have limited and imperfect information about its performance. Most models of technological change use the change in actual performance of the technologies in question as a fundamental axis of their dynamic (unless they consider it constant). The model proposed here takes an alternative and more realistic approach by focusing on the changing perceptions potential users have of the real performance of the technological alternatives. In the model, these perceptions of a technology improve with market experience and come closer to the actual performance as the number of users grows.

Numerical Simulations

As regards the choice of fifty iterations per simulation, this number was selected because it was considered a sufficiently distant horizon to enable the phenomenon of a series of successive standards to be studied. A more remote horizon did not yield different conclusions but required more computation time in the experiments.

As a sample, and for reasons of space, appendix 1 presents only the first ten simulations of all the standards recorded in each simulation. The table in the appendix shows the different characteristics of each of the standards, and the complete standardization process, recorded in each "history" (simulation) of this industry, according to the attributes described above (time, frequency, speed, duration, depth, susceptibility to changes in supply and in changes in demand). Appendix 2 shows the data for the first standard recorded in each simulation, filtered out from the general results of the 150 simulations. The characteristics of these standards will be examined in more detail below.

The set of graphs included in appendix 3 shows the results of a typical model simulation using the base scenario. The first graph shows the time course of the adoption (market shares as a unit proportion) of the technologies by the various producers present in each moment in the industry producing good X. Following Anderson and Tushman 1990, technologies that exceed a 50 percent share for at least three periods are identified as technology standards.12

The last two graphs show the evolution of the population of firms33 and the population of technologies, respectively. It is worth highlighting the striking visual isomorphism between the results of the model and the patterns of the phenomenon observed in real industries as, according to Marney and Tarbert 2000, this may be considered an indicator of the validity of die model. In a test of external validity, as understood by Kleijnen 1998, in appendix 4 it is possible to see that our model offers a representation of the phenomenon of technological succession consistent with the empirical evidence on the diffusion of technologies in industries as diverse as RAM chips, steel manufacturing, or power generation.

Examination of Hypotheses

It was argued above that in the presence of increasing returns to adoption the market may become locked in in favor of one of the competing alternatives due to historical circumstances. It is a recurring statement in the literature on this issue that under these conditions there is no guarantee that there are no alternatives to the winning technology that would have had lower unit costs if they had achieved an equivalent position of market dominance (Arthur 1989; David 1989; Cowan 1990; Metcalfe 1994). Moreover, throughout this paper we have maintained that the technology path followed by an industry is shaped by chance historical events (path dependency). Thus we can state that the distribution of market shares between the different technological alternatives at any given time, and the succession of possible technological standards, will depend on one hand on the chance order in which these alternatives are available for adoption and that in which the potential adopters enter and exit the industry. It will also depend on the characteristics of each of these heterogeneous "populations" (technologies and firms) at each time (performances and preferences, respectively).

Discussion

Orthodox models of technology diffusion and standardization have typically focused on the question of the rate of diffusion at which one new technology is fully adopted. The model presented here provides a broader approach to the process of technological change, from the perspective of the extension of the diffusion of multiple alternative technologies, and the related phenomenon of standardization. Furthermore the model treats the changes in either technological performance or adopter preferences needed to escape lock-in as endogenous. The model has allowed us to examine the hypothesis that in an industry characterized by increasing returns to adoption of technology, over a sufficiently long time horizon, we will witness a succession of multiple alternative equilibria (standards) which cannot be anticipated and are not necessarily Pareto-optimal.

Various authors have also suggested that it is necessary for there to be a significant improvement in the performance of an alternative to the standard to overcome the inertia of the system and initiate the transition from a locked-in technology to a new one (supply side). The model makes it possible to show with greater detail than conventional "all-or-nothing" models (which, moreover, usually focus on the competition between just two alternatives) the reality of a dynamic of transition between unstable equilibria in an industry divided at all times between multiple available alternatives, with different levels of adoption at each moment in time. The evidence also shows that die loss of dominant position by a technology standard is not always solely due to the emergence of an alternative technology offering significantly better performance. It may also be due to changes on the technology demand side. The proposed model has also allowed us to examine the hypothesis that either a sufficient improvement in the features of alternative technologies or a sufficient change in the preferences of potential adopters regarding the features of technologies can lead the cumulative endogenous phenomena in the industry, deriving from the presence of increasing returns to adoption, to break with the standard and trigger the transition toward the next standard, without the need for intervention from outside the industry.

These insights make a contribution to the formal understanding and simulation modeling of technological change, but they also point out an important paradox that neoclassical economic and new institutional approaches fail to convincingly explain. The presence of increasing returns in an industry means that the positive feedbacks are always present and thus provide the possibility for what Arthur has termed "perpetual novelty" (1997), meaning constant change and innovation in the industry. While this seems to accurately define technologically dynamic sectors like computers and information technology, other industries are far more quiescent and less prone to "perpetual novelty." Electricity generation and distribution, for example, has seen relatively few disruptive technological innovations, relying fundamentally on the same technology for more than fifty years. Standardization in the electricity sector seems to be a good example of technological lock-in, which is supposedly fostered by the presence of increasing returns to scale. But this observation then leads to the question "why haven't increasing returns in this industry led to dynamic change, to perpetual novelty?" Obviously there must be some countervailing negative feedbacks that are unaccounted for that can create longer term technological stability in industries.

In the model here, the negative feedback loop is generated by the exhaustion of increasing returns as markets become saturated. This, however, cannot create long-term stability as it merely sets the foundation for future substitution by a superior technology. Instead, as the modeling here demonstrates, it creates an opening for new innovations to emerge and exploit their own increasing returns-driven growth, displacing currently dominant technologies. Explaining long-term stability and technological lock-in must therefore depend on alternative causal forces. A logical explanation is that the emergence of a technological standard creates coupled changes in other social structures which subsequently foster lasting stability. The new institutional emphasis on decision making, reflected in the increasing returns literature, largely fails to address these coupled institutional changes. It is traditional (old) institutional theories and concepts (Veblen, Myrdal, Kaldor) that provide the most convincing explanations of long-term stability and lock-in.

These coupled changes wrought by new technologies are well documented and multiple. Corporations, for example, that have proprietary rights over the winning technology will create durable organizational structures to efficiently produce the technology. Their economic success allows them to become powerful members within an oligopolistic market structure created by the increasing returns standardization (Galbraith 1968). Similarly industrial users of the technology will alter their managerial routines and organizational structures to integrate the new technology into their activities (David 1985), causing coupled changes in workers' roles, facilities design, and so on. Commercial end users of technologies like the automobile also alter their behaviors, cultural practices, and fundamental lifestyles to incorporate the technology and the services it can provide (Unruh 2000). These coupled changes become part of Veblen's process of progressive habituation that creates durable customs, habits, and learned behaviors. Governments can also become involved in technological management for a variety of reasons including public safety, concerns for universal access, or even national security (Commons 1921). These actions often create durable political institutions such as the Tennessee Valley Authority or national utilities. The point is that our understanding of technological change and stability, as reflected in the technological lock-in literature, can be greatly enhanced by turning to the vast body of work by old institutionalist scholars, providing a useful approach to resolving the increasing returns-technological lock-in paradox.

Conclusion

This paper has pursued modeling that responds to Radzicki's 2003 JEI challenge to heterodox economists to use new modeling tools for institutional and Post Keynesian economic analysis. We have used agent-based economic computational modeling to expand the understanding of technological competition in the presence of increasing returns to adoption by extending the modeling effort beyond the one-time, "winner-takes-air approaches. It has shown that, over a long enough time horizon, an industry will see a succession of technological standards. Transitions between these standards can be fostered by the emergence of new technological alternatives that offer new sets of attributes. Transitions can also be induced by changes in the preference sets of potential adopters. These preferences changes are influenced by economic criteria as well as custom, habit, and organizational and other institutional factors. Finally it has also been argued that long-term technological stability or lock-in cannot be explained solely by the presence of increasing returns, as increasing returns can potentially generate perpetual novelty. Instead old institutionalist mechanisms are crucial to understanding longer term technological lock-in.