KNOWLEDGE DYNAMICS: Reconciling Competing Hypotheses from Economics And Sociology* Anne Marie Knott* The Wharton School of the University of Pennsylvania 2023 Steinberg Hall-Dietrich Hall Philadelphia, PA 19104-6370 Phone: (215) 573-9628 Fax: (215) 898-0401 knott@wharton.upenn.edu Bill McKelvey Anderson School at UCLA 110 Westwood Plaza Box 951481 Los Angeles, CA 90095-1418 Phone: (310) 825-7796 Fax: (310) 206-3337 mckelvey@anderson.ucla.edu June 22, 1999 * Please address all correspondence to the senior author, Anne Marie Knott. The authors would like to thank Dan Levinthal and participants at the Reginald Jones Center seminar series for helpful comments during development of this research. We would also like to acknowledge financial support from the Huntsman Center for Emerging Technologies. ii KNOWLEDGE DYNAMICS: Reconciling Competing Hypotheses from Economics And Sociology Abstract Our goal is to reconcile competing null hypotheses from sociology and economics with respect to knowledge flow—sociology assumes that knowledge flow is viscous, whereas economics assumes knowledge flows fluidly thereby discouraging investment in its creation. The concern is that the two fields’ attendant prescriptions might cancel one another. Our vehicle for reconciliation is a simulation of knowledge flow that embodies empirical regularities emerging from studies in both fields. Through simulation we find that both fields are partially correct. More importantly we find that the fields’ prescriptions for knowledge growth, rather than canceling one another, actually complement one another. 1 1. INTRODUCTION Both economics and sociology agree that innovation and human capital formation normally improves social welfare, but they differ in starting assumptions. In general, economics studies knowledge creation—How can public policy create incentives for firms to create new knowledge? Sociology, on the other hand, looks at knowledge diffusion—How can institutions get individuals to share their own knowledge and adopt that of others? Corresponding to these goals, the two fields have opposing null hypotheses regarding the flow of knowledge. In general, economists hold that knowledge flows freely—the challenge is to impede knowledge flow. In contrast, sociologists generally hold that knowledge is inert—the challenge is to facilitate knowledge flow. The null hypothesis that each field has adopted makes sense given its respective approach to innovation. Economists worry that firms will not invest in knowledge creation if they can not appropriate the returns from those investments. The greatest threat to appropriation is the relative ease with which knowledge can be transferred. Sociologists suspect that organizations will not share knowledge if they derive benefit from controlling it. They also see firms as comprised of boundaries around individuals and subunits that act as barriers to information flow. The problem with the opposing null hypotheses is that they may lead to contrary prescriptions that may potentially offset one another, such that neither innovation nor growth occurs. Thus, it is important to understand the relative correctness of each field’s null hypothesis. Are economists correct that knowledge flow is fluid, or are sociologists correct that knowledge is inert? We take a step toward resolving the debate by studying the impact of the driving assumptions, given various firm and industry conditions. Our approach begins with harvesting a set of stylized facts or known empirical regularities of knowledge flow from both literatures so as to develop a simple model of knowledge dynamics. We focus on empirical regularities spreading across 2 several fields of research and introduce a computational modeling approach so as to study the dynamics in between the extremes of the economists’ and sociologists’ assumptions—the range of dynamics more likely to be of use by managers interested in speeding up human capital appreciation within their firm, but worried about possible increased flow out to competitors. Thus, we try to answer questions regarding the inherent state of knowledge in a firm or an industry and the effects of various stimuli on knowledge growth in firms or in an industry. The method we use is an interacting agent model derived from the “spin glass” family of models frequently used in physics (Fischer and Hertz, 1993). In this article we: (1) highlight the debate; (2) resolve the debate such that research findings from the two literatures may be constructively integrated toward a theory of knowledge dynamics between the two extremes; and (3) provide some new insights that might guide public as well as managerial policy. We begin by reviewing the sociology of science, diffusion, technology economics, and management of technology literatures, setting out the empirical regularities as we go. Then we propose a formal model of knowledge dynamics that we evaluate through computer simulation. We validate the simulation relative to the empirical regularities and then utilize it to explore the dynamics affecting knowledge growth. Results and Discussion follow. 2. COMPARING THE PERSPECTIVES 2.1 Sociology of Science and Organization Sociological perspectives on knowledge flow are probably best exemplified by the sociology of science and the study of bureaucratic structure. Sociology of science is concerned with the interrelationship of science and society. How has science influenced values, education, class structure, ways of life, political decisions, and ways of looking at the world? How has society, in turn, influenced the development of science itself (Kaplan, 1964)? Organizational sociology has 3 uncovered significant deleterious effects of bureaucratic structure that inhibit the flow of information across organizational boundaries. Sociology of science emerged in the early twentieth century as an outgrowth of studies of the history of invention and technology (Ogburn, 1922; Usher, 1929; Gilfillan, 1933; Merton 1938). Sociology was ripe for a theory of science around the turn of the century, as technology formed a major force that threatened, or at least affected, society. Since its inception the sociology of science has come to be characterized as the “old” and “new” schools. Old sociology of science is best represented by the 240 or so published papers and classic book (1963) by Price and a second edition that includes nine later papers (1986; reissued by Columbia University Press with a foreword by Robert K. Merton and Eugene Garfield). Price’s papers document the quantitative study of science, what he called “scientometrics,” and particularly the study of citation indices. New sociology of science studies science as a social phenomenon, paying particular attention to the social construction aspects of scientific truth claims as characterized by the postmodernist literature (Mirskaya, 1990; Lynch,1993; Hilgartner and Brandt-Rauf, 1994, Pels, 1994; Murphy, 1994; Fuller, 1995a, b; Barnes, Bloor, and Henry, 1996). The impetus for new sociology of science was undoubtedly Kuhn’s 1962 book, The Structure of Scientific Revolutions, with the works of Hanson (1958) and Feyerabend (1970) also instrumental. Mulkay’s 1969 paper, “Some aspects of cultural growth in the natural sciences” sits as a dividing point between old and new— between the quantitative and functionalist views of Merton (1942) and the relativist/postmodernist views of post Kuhnian and postmodern sociology of science. Since we are interested in the causes of knowledge flow dynamics and whatever empirical regularities exist, needless to say, new sociology of science has little to offer once it moved into studying the social aspects of truth claims. This explains why most of the research we cite dates before new sociology of science. 4 Up until about the 20th century most people, while knowing that change occurs, tended to think of stability as more normal and preferable than change (Dubin, 1958: 117). Partly this was because U. S. society was rural dominated and more homogeneous and change occurs least readily in less heterogeneous and rural societies (Berelson and Steiner, (1964: 615–616). However the combined effects of technological development coupled with consequent social change stemming from the industrial revolution and the growth of factory employment in urban centers transformed all aspects of human life at ever more rapid speed and magnitude (1964: 615–617). The general public’s conception shifted to one in which change was more normal than stability, as well as more desirable. Further impetus came with World War II, as science attained political importance, both because it formed a major element in the national budget, and because it produced technology with substantial societal implications, e.g., the atomic bomb(Hiskes and Hiskes, 1986). Given technology as the major source of change and science at the heart of technology, a sociology of science emerged to deal with the patterns of change likely to affect society (Barber, 1952; Barber and Hirsch, 1962; Hiskes and Hiskes, 1986; Fuller 1993). Thus, understanding the rate of technological change became a significant goal of sociology of science. Another objective was to identify and promote factors that lead to change. 2.1.1 Knowledge Creation Process Sociology of science holds that knowledge accumulates. Exponential growth in several measures of scientific activity comprises the primary evidence supporting this view. Sociologists view invention as the cumulative synthesis of many individual items, though the magnitude of each item is small. Thus, scientific growth is largely a diffusion process: accretion of many small innovations each building upon prior innovations (Crane, 1972). Ogburn (1922) goes so far as to contend that when the necessary cultural base is in place, invention is inevitable—if one inventor 5 does not create the new device another will. He offers as supporting evidence the frequency of independent simultaneous invention. This is confirmed more recently by Kuhn (1959), and is evident in the account of the biologists’ search for the DNA “double helix” molecule (Watson, 1968). If growth in social welfare is the end, and invention the means, one merely needs to facilitate diffusion of the existing knowledge base to ensure efficient discovery of the inevitable invention. Diffusion mechanisms, thus, become the main focus of empirical studies in sociology of science. The science citation index has become a major data source for the investigation of the growth and the diffusion of knowledge since it provides a clear listing of prior innovations. A number of empirical regularities emerge from the examinations of scientific citations: 1. The growth of science as a whole follows a logistic curve, and while the total cost of research has increased by a factor of 4.5 since World War II, output has merely doubled. Further the growth of important contributions (heavily cited papers) has remained constant over the same period. “Thus we are multiplying lesser talents faster than the highest ones with half the scientific advance” (Price, 1963: 91). (Diminishing returns) 2. Output (complete papers per person) is highest for large groups and for solos (Price, 1963: 132). (U-shape productivity) 3. Per capita science activity in a country correlates with per capita wealth and level of economic development (Price, 1963: 43). (Wealth effect) 4. Competition (number of specialists attacking problem) increases when agreement about the importance of a field increases (Hagstrom, 1965). (Density dependence). 2.1.2 Stimulus to Knowledge Creation A parallel theme hinting that progress may not be inevitable, is the stimulus to invention—the extent to which invention is socially determined versus the inherent development of science. Ogburn (1922) acknowledges that necessity (social pull) plays a role in invention, but argues necessity is insufficient without a cultural base, and not necessary, as evidenced by frivolous invention. This view is not universally held. Price for example, holds that the greatest and most 6 useful advances in our technologies have come not from applied research, but from “basic research aimed at furthering understanding and curiosity” (1963: 155). Stein (1962) presents research findings indicating that personality factors also bear on knowledge creation, independent of culture or the basic/applied continuum. Why individual scientists do what they do is “little science” (Price, 1986). As the century ends, the concerns about the practices and social impact of “big science” have become more pronounced as people worry that big science has been stimulated beyond reason and to the exclusion of other factors affecting modern society—particularly the impact of the military/industrial symbiosis, hi-tech weapons and weapons of mass destruction, and chemical pollutants (Steneck, 1975; Reingold, 1979; Hiskes and Hiskes, 1986; Bell 1992). 2.1.3 Diffusion of Innovation A related literature dealing with the diffusion of innovation outside science draws conclusions similar to those of the sociology of science. Diffusion research grows out of the rural sociology studies of the 1940s that examined the diffusion of agricultural innovation—the most influential study being Ryan and Gross's (1943) investigation of the diffusion of hybrid corn seed. The research spans a number of disciplines including education (adoption of learning innovations), public health (adoption of health practices), communication (awareness of media events), marketing (the adoption of new products), and geography (role of spatial distance in technology adoption) (Rogers, 1995). The principal theme of this literature is that widespread adoption, even of an idea with obvious advantages, is often difficult. The research examines the factors affecting the rate of adoption, with a goal of devising policies that speed the adoption process. There are two important differences between the sociology of science and diffusion literatures. Sociology of science is fundamentally interested in the creation of new knowledge, whereas the 7 diffusion literature is concerned with the exploitation of existing knowledge. Therefore, because the sociology of science view is that creation is inevitable, given sufficient prior accumulation and diffusion, it follows that creation of new knowledge directly is isomorphic with the adoption of existing prior knowledge. Consequently, this distinction disappears—both see diffusion as critical. The second distinction is that while both literatures examine knowledge flow between individuals, sociology of science restricts attention to scientists embedded in the scientific institutional context, whereas the diffusion literature, perhaps because of its breadth, examines the implications of diffusion among individuals independent of any institutional context. Rogers (1995) develops a set of empirical regularities from review of approximately 4000 diffusion publications: 1. Diffusion follows a logistic curve, where the rapid growth phase is prompted by adoption of the innovation by opinion leaders (Tarde, 1903) (Logistic growth) 2. Imitation (adoption by one individual prompted by the adoption of another individual) is most frequent between individuals who share similar attributes (Technical proximity/homophily). 3. The very nature of diffusion requires that some heterophily exists, else there is nothing to diffuse (Heterophily) 4. Innovators and early adopters differ from later adopters in that they have higher education, social status, upward mobility, wealth, IQ, ability for abstract thought, ability for rational thought, and empathy. In addition, their communication patterns differ from later adopters. In particular, they have more exposure to mass media, engage in more information seeking, have more social ties, and are more cosmopolitan (Wealth effect) 5. Diffusion is a function of geographic distance (Hagerstrand, 1952) (Geographic proximity) 2.1.4 Boundary Effects While earlier diffusion findings were mostly about individuals independent of institutional context, now increasing numbers of individuals work in organizations. Thus, diffusion among individuals is more a function of diffusion across organizational boundaries. Weber’s (1947) ideal bureaucracy rests on division of labor, clear role definition, specialization, formalization, and centralization (with decentralization allowed if accompanied by appropriate rules). According to 8 “rational” organization design (Scott, 1998), the ideal bureaucracy is synonymous with clearly designed specialized roles and departments with accompanying rigid and impervious boundaries. In a comparative analysis of four countries Ben-David (1960) shows that the rate of innovation is hindered by strongly centralized organization structure and facilitated by loose and competitive structures. He also shows that innovation benefits from a diversity of roles. Mulkay (1969) concludes that though Kuhn’s (1962) normal science may progress adequately in bureaucratized structures, innovation is more apt to flourish when diverse roles are brought together, and when cross-fertilization is fostered by researchers occupying dual or multiple roles. Another recent view of bureaucratic effects is given by Pennings (1987: 208). He cites evidence that boundary spanning individuals (dual roles) are “crucial for the adoption of CAD/CAM in a production system. He also notes that high incidences of professionals and professional networks and a “loosely federated structure” aid the spread of innovations. Other innovation researchers also emphasize the importance of boundary-spanning gatekeepers (Pettigrew, 1973; Moch and Morse, 1977; and Ettlie, 1985). Boundary spanning is seen as essential to fostering innovation in Japanese firms (Nonaka, 1990), where information redundancy across departments as associated with a shared division of labor in which individuals are able to more broadly define their roles, thus partially negating the impermeable boundary effects of bureaucratic structure. The boundary spanners are the people who shore up homophily in firms where division of labor, specialized roles, and departmentalization all work to foster heterophily. Mohrman and Mohrman (1993) also report that bureaucratic structure and controls impede innovation and that rigid bureaucratic forms are antithetical to innovation. Boundaries negate the effects of physical or technical proximity. Leonard-Barton (1995) and Ashkenas et al. (1995) echo these results. Organizational elements that reduce or compensate for narrow role specializations, 9 reduce departmental boundary blocks to information flows, build interdepartmental networks, build role diversity and the interaction of employees playing multiple roles, and keep employees in close contact with external parties, that is, closer to competitive pressures and inputs from outside the organization, all work to enhance innovation (physical and technical proximity effects). 2.2 Technology Economics Economic perspectives on knowledge flow are best captured by economics of technology. The economics of technology literature was motivated by the link that (Solow, 1957) demonstrated between innovation and productivity growth. Its primary goal is to understand the conditions that facilitate innovation. The work is generally at the level of industries or the economy. The insights gained from work in this field are used to guide government policy, primarily through intellectual property law, antitrust regulation and national investment. The economics of technology literature focuses on two hypotheses, both of which are attributed to Schumpeter (1942). The first hypothesis is that large firms are more likely to innovate than small firms. This is essentially an hypothesis regarding firms’ capacity to innovate. The second hypothesis is that monopolists are more likely to innovate than firms without market power. This is essentially a hypothesis regarding firms’ incentives to innovate. 2.2.1 Firm Size There are number of factors tending to support Schumpeter’s first hypothesis. The advantages of large firms include the ability to hold a diversified portfolio of projects, R & D scale economies, cross fertilization of ideas, cheaper capital (through use of internal financing for risky products), complementary assets, stronger incentives for process improvements (due to larger base for spreading costs), serendipity (with more people there is greater likelihood that someone will recognize the value of an innovation), and more avenues for exploiting output. 10 There is an equally impressive set of factors suggesting that large firms are actually at a disadvantage in innovation. These, however, are organizational rather than economic issues, and thus may have received less attention in the early economics literature. These factors include the communication overhead of large size, a higher status of management jobs versus engineering jobs in large firms (thus drawing away the more experienced R & D professionals into management ranks), decision making filters, bias against imagination that drives away talent, and conservatism. Early challenges to the large firm hypothesis come from (Jewkes, Sawers and Stillerman, 1958) and (Hamberg, 1963). These studies indicate that less than 30% of major innovations come from large firms. This leads to a modified hypothesis by Hamberg: that large firms are likely to be minor sources of radical invention but major sources of improvement inventions. This view does not necessarily refute the firm size hypothesis—the cumulative effect of many small inventions may be greater than a smaller number of major innovations. Cohen and Levin (1989) develops a set of empirical regularities that summarize the conclusions from the firm size studies: 1. The likelihood of conducting R & D and the level of R&D spending increase monotonically with business unit size. (Business unit size explains approximately two thirds the variance of R&D expenditures.) (Wealth effect) 2. R & D productivity (output/expenditures) declines with firm size and, thus, smaller firms account for a disproportionately large share of invention relative to their size (Geographic and technical proximity).f The empirical regularities seem to imply that arguments in favor of both sides of the firm size hypothesis have merit. The arguments in favor of large size pertain to efficiency advantages. Large firms recognizing their efficiency advantages, are willing to spend more. This resembles the wealth hypothesis in the sociology and diffusion literatures. Arguments in favor of small size pertain to effectiveness advantages. Small firms provide environments more conducive to 11 invention, and therefore produce more invention per dollar than do their large firm counterparts. This finding resembles the proximity arguments in the sociology and diffusion literatures. 2.2.2 Monopoly Power The second Shumpeterian hypothesis is that monopoly power is conducive to invention. Monopoly power conveys two advantages for innovation. First, it insures appropriability for the investment a priori (incentive effect), and second, it provides profits posteriori to support further innovation (capacity effect). The counter-argument is that (1) once the monopolist is enjoying a rent stream, it has reduced its incentive to innovate, because it cannibalizes a portion of that rent stream (Arrow, 1962; Asher, 1964; Baldwin,1969; Demsetz, 1969); and (2) incentives to invest in innovation are reduced if the new ideas are easily expropriated by competitors. It is here more than anywhere that economists differ from sociologists. While sociologists see diffusion as “good” because it spreads good ideas throughout society, economists see diffusion negatively as expropriation of a firm’s ideas by competing firms. Therefore, though they agree with sociologists that diffusion aids broader social welfare, economists worry that expropriation discourages firms from investing, turning a “positive” into a “negative.” Thus, in what follows we often will appear to be interchanging “diffusion” with “expropriation”—same process, but different spin. Schumpeter’s second hypothesis has enjoyed much greater attention than his first. The economic concern is that because knowledge is a public good (is not consumed in use), and because its transfer is frictionless, the broader social returns to R&D may exceed the returns to private firms, reducing their incentive to continue investing in innovation. Thus, in the absence of intervention, R&D may fall below socially optimal levels (Arrow, 1962; Hirshleifer, 1971). Two options are open to policy makers under these circumstances. The first is patents (to increase the certainty that a firm may appropriate the full value of its investment); the second is enlightened 12 anti-trust policies (that recognize that the less efficient resource allocations under monopoly may be offset by the long term benefit of growth in economic welfare). Theoretical development of the second hypothesis examines competition among potential innovators engaged in an R&D race. The main questions are how profits, costs, and intensity of rivalry determine the speed of innovation (Kamien and Schwartz, 1982). The incentives to innovate consist of the “carrot” of innovation profits and the “stick” of lost profits in the event a rival innovates first. The main conclusions from the theoretical literature are: 1. A new good that does not supplant an existing one will appear more rapidly than a replacement good (Diminishing returns). 2. The greater the loss from rival precedence, the sooner development will take place (Wealth effect). 3. Rivalry decreases development time up to a threshold number of rivals, thereafter it increases development time (as added numbers of players tends to dissipate appropriability) (Density dependence). 4. Incumbents invest less than entrants when innovation is uncertain, yet they invest more than entrants when innovation is certain (Gilbert and Newberry, 1982; Reinganum, 1983, 1985; Salant, 1984; Katz and Shapiro, 1987). The factor inhibiting incumbent investment is the fact that they already enjoy a rent stream, thus they primarily invest only to protect it (Causal ambiguity). The general conclusions from the economic models of innovation are that rivalry and appropriability interact. Competition for an appropriable return leads to over-investment, while non-appropriability or absence of competition leads to under-investment. Empirical research tends to support the theoretical conclusions. Scherer (1980) finds that insulation from competitive pressure discourages innovation. Scherer (1965), Scott (1984), Levin et al. (1987) all found an inverted U relationship between innovation and rivalry. Scherer (1965) found maximum patent counts for a four-seller concentration ratio between 0.50 and 0.51— indicating that appropriability effects are U-shaped. However this relationship tends to disappear when additional variables are added to account for technological opportunity and other factors influencing innovation (Levin et al., 1985). 13 2.3 The Management of Technology Synthesis Our investigation shows that while the goals of both sociology of science and economics of technology are essentially the same—increased social welfare through innovation—the fields are markedly different in other respects. These differences are summarized in Table 1. The most notable differences for our purpose of reconciling views of knowledge flow, pertain to the patterns and stimuli of innovation. As to the pattern of innovation, sociology believes innovation is continuous (follows gradual accumulation)—thus to generate innovation, we merely need to diffuse prior knowledge. In contrast, economics believes innovation is discontinuous. Thus to generate innovation, we need to create incentives. With respect to the stimulus to innovation, sociology believes innovation is a byproduct of the accumulation of knowledge. Economics believes it is the result of deliberate strategies by key individuals to innovate. (Insert Table 1 about here) A synthesis of these two perspectives occurs to some extent, in the management of technology literature. Writers through the 1950s and 1960s have goals similar to economics of technology, but techniques more closely aligned with sociology of science. Management theorists are interested in problems of worker satisfaction and productivity, with of goal of enhancing both. They apply both qualitative and quantitative methods to the examination of product developments. Qualitative methods characterize communication patterns and other elements of project structure. Quantitative methods examine outcomes such as success rates and development times as a function of a number of project characteristics. (Gordon, Marquis et al., 1962, Kornhauser, 1953; Bennis, 1956; Shepard, 1956; Glaser, 1964; Allen, 1977). A similar synthesis of sociology and economics occurs in more recent work in management of technology with respect to notions of progress. Whereas sociology tends to see progress as 14 cumulative and incremental, and economics see progress as episodic, management of technology accounts for both in its life-cycle perspective. Technology life-cycles are punctuated by radical innovations (episodes), but progress through incremental innovation thereafter (Abernathy and Utterback, 1978; Abernathy and Clark, 1985; Anderson and Tushman, 1990). Thus, writers in the management of technology field have adopted a view wherein the radical innovations are viewed as acts of insight (by key individuals) involving the synthesis of items derived from prior acts of insight. Progress consists of acts of insight of differing degrees of importance converging toward a massive synthesis. While the intervening succession of events is orderly and logical, the punctuated intervals are indeterminate (Usher, 1929). Another, perhaps more powerful, synthesis of the sociological and economic perspectives occurs with the consolidation of empirical results from the three fields (including management of technology). Table 2 is an effort to summarize and classify these results by phenomenon (rows), level of analysis (columns) and source literature (font). The table clearly indicates that empirical regularities emerge despite the theoretical differences across the fields. The conclusions we draw from the table are first, that neither null hypothesis (knowledge inert vs. knowledge free-flowing) dominates the results. The second conclusion we draw is that the empircal regularities transcend levels of analysis. While this is what we would expect if knowledge is “atomistic”, as most of us would tend to believe, we now have some confidence that we have captured an entity for which it is possible to characterize fundamental flow properties. The third conclusion we draw is that the composite view given by the three fields offers us a multidimensional view of knowledge dynamics that is empirically grounded. These empirical findings pertain to the patterns of innovation (logistic growth within field, versus episodic growth across fields; bi-modal distributions of entities’ outputs), entities’ capacity to innovate (wealth effects, proximity effects), 15 as well as entities’ incentives to innovate (diminishing returns to innovative activity, density dependence of innovative output). (Insert Table 2 about here) A final conclusion we draw is that the joint effect of the nine principles is virtually impossible to discern via simple intuition—some causes are linear; some are nonlinear; and there are several of each. We have no theory governing how the various effects interact. Given this, How could a manager decide what the optimal level of investment is? How could a government policy advisor or an international development agency or and industry consortium work to assure an optimal level of knowledge in a given industry? How could an international alliance of global firms chart a course of investment in R&D for the alliance? To help unravel the countervailing effects of the nine principles, we employ the empirical regularities as specifications for a computational model of knowledge flow. For reasons outlined below we think a computational simulation is especially appropriate for better understanding knowledge flow dynamics. 3. TOWARD A THEORY OF KNOWLEDGE DYNAMICS 3.1 Why modeling? Logical positivism (Ayer, 1959) and logical empiricism (Kaplan, 1964) were abandoned by philosophers three decades ago (Suppe, 1977). Three replacement epistemologies have emerged since then: scientific realism (Aronson, Harre, and Way, 1994; de Regt, 1994), the semantic conception (Suppe, 1989, Thompson, 1989), and evolutionary epistemology (Hahlweg and Hooker, 1989; Rescher, 1990). Some essential features of this new view of epistemology are brought to light by (McKelvey, 1999c) under the label, “Campbellian Realism.” This term signifies Campbell’s life-long interest in evolutionary epistemology and scientific realism (Campbell, 1959; Campbell, 1974; Campbell, 1988). That Campbell the behavioral/social 16 scientist is deeply involved in this rereading of epistemology is critical for organization scientists because there is a tendency in some circles to eschew “normal science” epistmology because it is judged irrelevant to organization science (Perrow, 1994; Burrell, 1996; Chia, 1996). Drawing as it does on the three replacement epistemologies, Campbellian realism sets up a number of standards for effective science. Several of these assert a “model-centered science” in which the critical importance of the coevolutionary development of the theory–model link is highlighted. Though (Suppe, 1977) accepts that the semantic conception does not insist on “formal” mathematical or computational models as opposed to more qualitative formulations, the more effective sciences, Suppe, and most scientific realists (Bhaskar, 1975; Aronson, Harre, and Way, 1994; de Regt, 1994) and semantic conception adherents (Lloyd, 1988; Suppe, 1989; Thompson, 1989) accept the idea that theories benefit considerably if their nuances are developed in association with more formal modeling approaches. Pfeffer (1997: 195) offers an additional motive for a model-centered science—simplicity. Models have a way of forcing investigators toward more parsimonious theories. 3.2 Goals for the Model Given a Campbellian realist epistemology, we pursue the development of a theory of knowledge dynamics using a computational models. The primary goal for using this model is to conduct a “crucial experiment”(Stinchcombe, 1968) of the opposing null hypotheses of economics and sociology regarding knowledge flow. Does knowledge inherently flow freely (economics), or is it inherently inert (sociology)? We also hope to understand the links between the nine principles governing the input variables and the patterns of output, so that we can generate policy recommendations—what should the population configuration be if we want rapid knowledge diffusion (social welfare), and what is the 17 appropriate configuration if we want to increase a firm’s innovation rate, or inhibit diffusion (firm appropriability). At a minimum, the model should generate behavior consistent with the empirical regularities regarding knowledge flow—so that we can understand the system producing these behaviors. Given consistency with the empirical regularities, our level of confidence in the computational experiment is increased. Assuming the foregoing objective is reached—the model generates behavior consistent with the existing empirical regularities—we can then use the model to experiment with resource combinations that would be difficult to play around with in the real world. And a computational experiment platform also allows us to meet another standard of effective science, that is, to explore the “if this then that” aspects of our theory. Do the causal principles we identify have the effects— in our model at least—that we would predict? Finally, a computational agent-based model fits the assumption base of organization or innovation behavior. The actors/entities involved are NOT uniform; their idiosyncratic differences are stochastic and have nonlinear effects. Further, the number of variables we have in the model, coupled with their nonlinearity, makes a closed form mathematical solution extremely unlikely, giving the advantage to a computational simulation. 18 3.3 Modeling Approach According to a review by (Carley, 1995) a variety of computational models appear in organization science—nearly 100 model applications are reviewed. Mostly these models represent idealizations of individual or organizational decision processes designed to replicate the decision processes of specific individuals or firms. These are “thick” models, to adopt the parlance of Geertz (1971), in that they attempt to represent in considerable detail the target decision processes. More recently a different “thin” (agent-based) approach to modeling appears in the literature (Carley and Newell, 1994). Thin models make very few assumptions, or impose very few rules, about how individual agents decide. They simply expect agents to “stay the same” or “change”—a simple binary choice. As compared to thick, qualitative, in depth studies of a small sample of human behavior, thin studies use a few proxy variables across a large sample. Thick and thin models each have their advantages. Generally as sciences have (1) reached down toward the most basic particles, molecules, agents, or components (generally called microstates); (2) have accepted that microstates behave stochastically rather than uniformly; and (3) expect nonlinear dynamics, scientists have more frequently adopted “thin” agent-based models (McKelvey, 1997). Sometimes called “adaptive learning models,” agent-based models figure more prominently in the organizational learning literature, an early application being by (Cohen, March, and Olsen, 1972). Recently many more applications appear in this literature (Durfee, 1988; Masuch and LaPotin, 1989; March, 1991; Carley and Svoboda, 1996; Cohen, 1996; Warglien, 1996; Padgett, 1997; Prietula, Carley, and Gasser, 1998). Organizational adaptive learning also appears in applications of Kauffman’s NK model by Levinthal (1997) and McKelvey (1999a, b) that are rooted in the physicists’ spin glass models (Fischer and Hertz, 1993) and the computer scientists’ 19 cellular automata models (Weisbuch, 1993). Applications of the biologists’ genetic algorithm model (Mitchell, 1996) also now appear in the management literature (Bruderer and Singh, 1996). Our model is a derivative of the spin glass. We start with an n x n lattice of firms and their interconnections. Spin glasses model the rate at which the total energy level of a lattice declines as individual agents (vertices) attempt to reduce their energy in response to current states of their neighboring agents. This model type fits our needs since we are interested in the total knowledge growth of all firms comprising the lattice. Spin glass models are frequently used when a quantity such as energy and fitness (or knowledge) is expected to (respectively) decrease or increase. As you will see, the equations we use leave agents with essentially two choices at each of many time periods. At each time period each of our agents has a stochastically driven choice of expropriating knowledge from another firm (or not) or creating new knowledge on its own (or not). 3.3.1 Macro and Micro Variables and their Interdependence 3.3.1.1 Microvariables. We begin by defining a set of heterogeneous entities in a population (i = 1 to n). While these entities could be individuals, organizations, industries or even economies (as depicted in Table 2), and while our model intends to capture all such levels of analysis, here forward we will define the entities to be firms. Each firm is characterized by an initial endowment of knowledge, Ki, random normally distributed—thus some firms initially “know more” than other firms. This characterization is intended to capture wealth effects in the empirical regularities. Further, each firm has a location in physical space, Gi, as well as technological space, Ti. The location in physical space is intended to represent geographic location (e.g., Silicon Valley versus Buffalo). The technical location captures the location of the firm in technological space, e.g., pharmaceuticals are close to chemicals, but far from electronics. Some notion of physical space is necessary to incorporate the empirical regularity that transfers take place more efficiently over 20 short distances (geographic proximity). A notion of technical space is necessary to incorporate the empirical regularity that transfers take place more fluidly between firms that share a common knowledge base (technical proximity/homophily). “Tension” between firms dictates the extent and direction of knowledge flow between them. The tension is a function of physical proximity, Gij, and technological proximity, Tij, —the shorter the distance between two entities, the more likely it is that they will share information. Tension is also a function of the “knowledge differential” or differences in the amount of knowledge between the entities (Ki – Kj). Knowledge differential captures the intuition that if two entities have the same knowledge, there is no need for transfers to take place (requisite heterophily). 3.3.1.2 Actions Taken Each Period. Each period firms potentially gain knowledge in each of two ways. First they have the opportunity to invest in new knowledge—knowledge creation. Second, they have the opportunity to gain existing knowledge from other entities—knowledge expropriation. 3.3.1.2.1 Knowledge Creation. True investment behavior is potentially quite complex. Firms may invest merely to compensate for the obsolescence of the existing knowledge. In fact, a recent study indicates that almost all R&D investment in the pharmaceutical industry is that required to compensate for obsolescence (Knott and Bryce, 1998). Firms may invest because they can afford to, based on the profits from the prior period (capacity to invest). Alternatively, firms may invest because their future profits are threatened (incentive to invest). Moreover, as the technology economics literature indicates, these threats to future profits are multifaceted. We make two simplifying assumptions in the model’s investment rule. First, we consider only net investment, that is, we consider only investments exceeding those required to maintain the value of the existing asset stock (net investment = total investment – depreciation from prior period). Second, we 21 model only threat-induced investment, where threat is defined as a loss in the firm’s relative knowledge stock. Firms make net investments in an effort to preserve their share of the total knowledge in the industry. This behavioral assumption captures the stylized fact that innovation is an n-shaped (inverted u) function of industry concentration. We believe that the underlying mechanism producing the n-shaped function in concentrated industries is the zero-sum nature of competition—your gain comes at my expense and therefore I am inclined to invest in retaliation. Firms are always under tension to invest in creation as long as the knowledge stock is growing, because the growth in the stock in and of itself, causes them to lose share. Further, we believe there is causal ambiguity in knowledge creation. While firms may make investments, there is uncertainty regarding the amount of new knowledge these investments will yield. We thus model knowledge creation by firm i in period t as a random percentage,, of the lost share, where is Beta distributed with p = 25 and q = 25. The choice of Beta Distribution is driven by the need to bound the distribution at 0 and 1, but a desire to control the variance: K i t 1 K it Cit K i t 1 K it i 1,n i 1,n K it i 1,n (1) where: Cit = new knowledge created by firm i in period t = random number (from a Beta distribution) [25, 25] representing causal ambiguity Kit = knowledge stock of firm i in period t If equation 2 yields a negative value, which is the case if firm i enters period t having gained share, Cit is set to zero. 3.3.1.2.2 Knowledge Expropriation. Also in each period, firms have the opportunity to expropriate existing knowledge from neighbors (we do not distinguish between unintentional leakage and intentional sharing). There are two basic issues regarding implementation of 22 expropriation dynamics. The first issue is how many host firms a focal firm i, can expropriate from in a given period. The second issue is how firms choose hosts. With respect to the issue of number of host firms, we assume that firms have limited information capacity, and therefore restrict attention to a single firm at a time. With respect to the issue of host choice, there are two basic approaches. An omniscient approach assumes that firms are able to compute the expected expropriation from each of the potential hosts, then choose the host offering the maximum expropriation. This approach assumes that firms are able to rank order potential hosts in terms of their total knowledge, Kij. This is not unreasonable, since firms can compare products and make assumptions about the level of underlying technological expertise. Further, firms know which of their rivals in the industry are competing in the same arena, Tij, and they certainly know physical proximity, Gij. To be conservative, we nevertheless adopt a naïve approach. The logic driving our choice of the naïve approach is one of firms’ extractive opportunities. We assume that even if firms know which host has the greatest expropriation potential, they may not have an opportunity to extract knowledge from that host. We suppose instead that firms come in contact with rivals randomly, and that each contact presents an opportunity for expropriating rival knowledge. Thus expropriation for firm i in a given period is defined by random choice of firm j, with the knowledge differential between the two firms defined as Kij = Kj – Ki, and the attenuation crossing technical and geographic space, defined as squared Euclidean distance: (Gij2 + Tij2). Further we assume that there is causal ambiguity in the expropriation process. Firm i may not know which of the host’s knowledge is valuable, or may not be completely effective in expropriating the knowledge. Thus firm i is not necessarily able to extract the entire surplus of j’s knowledge. We 23 operationalize causal ambiguity with a random number , that like is Beta distributed, representing the share of differential knowledge that firm i is able to extract. Eit K jt Kit / gij2 tij2 (2) where: Eit = knowledge expropriated by firm i in period t =random number [25, 25] representing causal ambiguity tij = technical proximity between firm i and firm j gij = geographic proximity between firm i and firm j If equation 1 yields a negative value, true if i has more knowledge than j, Eit is set to 0. 3.3.1.3 Micro and Macro Outputs. For convenience, geographic location and technical location are held constant in the model. One can imagine that entities may migrate to areas with greater potential for knowledge expropriation, but we assume that plant location and technical specialty represent durable commitments. The primary variables that evolve over time, are the knowledge stocks of the individual entities, and the cumulative knowledge of the population. These primary variables generate derivative changes in the knowledge differential between neighbors, and in the propensity to invest in new knowledge. 3.3.1.4 Assumptions. Implicit in the model are a number of assumptions. First, we assume for the moment that there is unlimited capacity in the communication channel, that is, there is no upper bound on the flow rate of information between two firms. (Earlier we assumed some capacity constraint in restricting expropriation to a single host firm). Second, we assume that the number of firms in the industry is fixed—no entry or exit. Our other assumptions have been mentioned in the discussion of variables: investment is driven by loss of relative knowledge stock, expropriation is from a single firm, and firms do not migrate either geographically or technically. These are simplifying assumptions that may be relaxed in subsequent analyses. 24 3.3.2 Simulation Dynamics The simulation evaluates an industry of 100 firms random uniformly distributed in physical and technical space. Firms are endowed with initial knowledge stocks that are random normally distributed. Each firm’s knowledge stock is updated in each period t, for both creation Cit (equation 1) and expropriation Eit, (equation 2): K it 1 K C E it it it (3) The simulation is repeated for 50 periods, or until knowledge growth ceases. Frequently knowledge growth does not cease after 50 periods, but we believe two and a half generations is a duration consistent with the assumption that growth is driven by the creation/diffusion process rather than by some external shock. The entire model as described here and in the foregoing section is depicted in the flow diagram in Figure 1. Each simulation is seeded with a set of initial conditions. The initial conditions that the experimenter controls are the density of the geographic and technical space and heterophily of firms’ knowledge endowments. Population density is manipulated through specification of the geography/technology space. The nominal value for each is length 10. Thus, firms are populated in a 10 10 lattice. Heterophily is manipulated through specification of the standard deviation of firm knowledge. The nominal value for is 500, with of 2000 knowledge units (= 0.25 In addition to the experimental control of the initial conditions, randomness is introduced by the simulation’s draws for each firm’s location, Gi and Ti, and initial knowledge, Ki. 3.4 Testing There are two types of tests of the simulation. Validation testing to ensure simulation integrity and hypothesis testing to conduct the crucial experiment These are each discussed below. 25 3.4.1 Validation Validation requires that the simulation is well behaved and matches the empirical regularities it attempts to capture (Table 3). We test the basic function of the simulation by running a baseline. The baseline consists of a simulation seeded with the nominal values of each parameter as outlined above (technical and geographic space = 10 10, firm knowledge endowments are drawn from a population with = 0.25with causal ambiguity in both creation and expropriation (are nonzero))We examine single runs of the nominal case to verify the integrity of the growth and convergence paths. Next we examine that the simulation exhibits density dependence and heterophily effects. We test sensitivity to density by distributing the initial population of firms in technical and geographic space of varying density from 0.25 firms per unit area (Gi = Ti, = 20, so 100 firms are populated in a space of 400 square units) to 8.0 firms per unit area (Gi = Ti, = 3.53, so 100 firms are populated in a space of 12 square units). We test sensitivity to initial heterophily by varying the initial dispersion of population knowledge from = 62.5 (0.3125) to = 1500 (0.75 ). Tests of density dependence and heterophily effects are based on 100 runs for each specification of initial conditions. 3.4.2 Crucial experiments We conduct the crucial experiment of the opposing economics and sociology null hypotheses by examining shares of knowledge after 50 periods (or after reaching equilibrium, whichever occurs sooner) under a range of conditions. The range of conditions includes the validation cases as well as a full complement of interactive cases. The interactive cases combine three levels of density from validation with four levels heterophily from validation. Again, results for each case are based on 100 runs for each specification of initial conditions. 26 If knowledge amounts held by all firms in the lattice are equal after 50 periods (defined as zero variance of knowledge shares across firms), then economics is correct that knowledge flows freely—and therefore comes to an equilibrium of equal shares. If, however, there is variance after a reasonable time for the system to achieve equilibrium—defined as some firms retaining more knowledge than others—then sociology is correct that knowledge is friction-laden—some firms can keep other firms from expropriating what they have. Since the ultimate goal of both fields is maximum knowledge growth, and the null hypotheses are merely beliefs fueling prescriptions to achieve that end, an alternative crucial experiment examines the prescriptions. Here we compare the relative impact of expropriation—diffusion— mechanisms (sociological prescription) and creation—incentive—mechanisms (economic prescription) on knowledge growth. We do so by alternately deleting each mechanism from the baseline, to examine its marginal contribution to knowledge growth. 4. RESULTS 4.1 Validation Validation requires that the simulation is well behaved and matches the empirical regularities it attempts to capture (Table 3). 4.1.1 Baseline The first such validation is a functional test of a baseline with moderate density (1 firm per unit area), and moderate heterophily (= .25). Figure 2 presents results for a single run of the baseline simulation. The results indicate three things of interest. First, total industry knowledge grows by a factor of 10 over the 50 periods (an annual growth rate of 4.7%). While 4.7% growth is on the high side for real GDP growth (a plausible correlate for knowledge growth), it is within a reasonable range. Second, the growth while episodic, follows a logistic profile—growing slowly 27 at first, then more rapidly, before beginning to saturate. Finally, while there are outliers, there is a general trend toward convergence (50% of firms hold 0.5% market share). Examination of single runs over the parameter space indicate that the simulation is similarly well-behaved, thus we proceed with formal validation. 4.1.2 Density Dependence The main behaviors not explicitly designed into the simulation that it should exhibit are density dependence and heterophily effects. We validate the effects of density, by running simulations over a range of densities, with all other input conditions held at their nominal values. Comparison of these results indicates that the simulation behaves according to the empirical regularities. Figure 3 shows that knowledge grows more rapidly as industry density increases. Industry knowledge grows at an annual rate of 1.5% in the low density case of .25 firms per unit area. In contrast it grows at an annual rate of 57% for the high density case of 8 firms per unit area. Moreover, the relationship between growth and density is monotonic, as we expect. Thus the simulation appears to have captured density dependence. 4.1.3 Heterophily We validate another form of “proximity”—one in which proximity is defined by the amount of knowledge (initial share heterophily), rather than by the type of knowledge (technical distance). For example, even though all firms in the pharmaceutical industry utilize the same technical knowledge, some firms invest more heavily in R&D, and thus have more knowledge (and greater knowledge share). While there is the issue of requisite level of knowledge (similar to absorptive capacity), we do not model it in the simulation. The main effect of share heterophily in the simulation is that it increases the tension of one firm to extract knowledge from another firm, that 28 is, the “knowledge differential,” Kij, between firms. As the population becomes more heterophilous, the expected value of Kij, measuring the tendency toward expropriation, increases. We validate the effects of heterophily, by running simulations over a range of initial share dispersions, holding all other input conditions at their nominal values. Comparison of these results indicates that the simulation behaves according to the empirical regularities. Figure 4a shows that knowledge grows more rapidly as initial dispersion increases. Industry knowledge grows at an annual rate of 3.0 % in the low heterophily case of = 62.5 (= 0.3125). In contrast it grows at an annual rate of 9.2 % for the high heterophily case of = 1500 (= 0.75). Moreover, the relationship between growth and density is monotonic, as we expect. Thus the simulation appears to have captured heterophily effects. Note that while initial heterophily has an impact on growth, its effect is small relative to density, and it exhibits diminishing returns. Increases in initial heterophily beyond = 750 ( = 0.375) have little effect on growth. This is likely due to the fact that the dynamic process itself creates and destroys heterophily, thus the initial heterophily loses its significance except in extreme cases. 4.1.4 Interactive Cases As a final validation we examine the interactive effects of density and heterophily. We run simulations that simultaneously vary density and heterophily. This test goes beyond validation in that the empirical regularities have no prediction regarding the interaction. Figure 5 indicates that density is the dominant factor affecting knowledge growth. This is consistent with the findings from the separate tests of density and heterophily. In fact it appears that variance in growth attributed to heterophily is within the range of noise for the simulation. 29 4.1.5 Validation Summary The simulation design, the baseline results, as well as the combined results for density and heterophily tests, provide some confidence that we have developed a simulation that at least to a first approximation captures the empirical regularities of real world knowledge flow. Given confidence that the simulation represents the real world, we can proceed with the crucial experiments. 4.2 Crucial Experiments 4.2.1 Crucial Experiment of the Null Hypothesis We conduct a crucial experiment of the null hypothesis by examining final shares of the knowledge stocks for each of the conditions tested under simulation. These results are given in Figure 6. The results indicate that for ALL conditions, the knowledge stock fails to converge to an equilibrium of equal shares. This result supports the sociology null hypothesis in that the knowledge appears to exhibit friction—firms can keep other firms from expropriating all their knowledge. However, the prior results support the economics null hypothesis in that knowledge is clearly flowing among firms at a reasonable pace (as evidenced by its growth). This “paradox” of sorts may explain the persistence of the opposing null hypotheses. Beyond the basic result that shares fail to converge, there is value in examining the relative convergence under the various conditions. Figure 6a examines convergence across density; Figure 6b examines convergence across initial heterophily, Figure 6c examines convergence for the interactive cases. Each figure depicts the terminal dispersion (expressed as the standard deviation) of the shares of knowledge held by firms after 50 periods. In all cases the mean share after 50 periods was the expected value of 1% (total stock divided by 100 firms). 30 Figure 6a indicates that shares are more likely to converge under low density. While we might have expected the opposite result—high density/close proximity leading to wholesale imitation, it appears that what is driving behavior in close proximity is something more complex. Since knowledge flows so freely in close proximity (less loss over distance), shares change more rapidly, thereby increasing the pressure to create. This model interpretation is supported by theoretical arguments and empirical evidence regarding increased innovation in coevolutionary pockets, such as Silicon Valley, where there are strong proximity effects (Porter 1990). Figure 6b indicates that heterophily, as we have defined it (initial differences in knowledge held by firms), operates inversely to density. In some sense, both density and heterophily are capturing “closeness”. Density captures closeness in physical space and type of knowledge, heterophily captures closeness in the amount of knowledge. While Figure 6a indicated that closeness (in type of knowledge) prevented convergence, Figure 6b indicates that closeness (in amount of knowledge) facilitates convergence. It appears that when firms are close in the amount of knowledge there is little opportunity to gain from expropriation, thus there is little opportunity for consequent share changes that would stimulate creation. Figure 6c seems to indicate that density effects dominate heterophily effects. The basic result of the crucial test of the null hypothesis still holds—knowledge stock fails to converge to an equilibrium of equal shares—supporting the sociology null. However, knowledge is clearly flowing among firms at a reasonable pace (as evidenced by its growth)—supporting the economics null. Given the equivocal results, it becomes more important to conduct the comparative test of the two fields’ opposing prescriptions for knowledge growth. We do that next. 31 4.2.2 Comparative test of prescribed growth mechanisms The baseline simulation incorporates both fields’ prescriptions for growth—the economics prescription of endorsing incentives for firm investment in knowledge creation, and the sociology prescription of endorsing mechanisms that diffuse knowledge more rapidly. The economics prescription is embodied in the creation mechanism; the sociological prescription is embodied in the expropriation mechanism. We conduct a comparative test of the two mechanisms by examining the marginal effects of deleting each prescribed growth mechanism relative to the baseline. The test reveals that expropriation (the sociology prescription) is more effective in achieving industry knowledge growth than is creation (the economics prescription), given the way we have modeled each. A baseline simulation with both mechanisms (without causal ambiguity in either mechanism) yields annual knowledge growth of 29.1%. Deleting the creation mechanism (leaving only expropriation) yields 14.6% growth; deleting the expropriation mechanism (leaving only creation) yields only 8.1% growth (Figure 5). Given a goal of maximum knowledge growth, there is some support for the prescription of expropriation. Note however, that the knowledge that is “growing” with expropriation is the total knowledge held by firms, rather than new knowledge (unless we believe, as sociology of science has, that new knowledge is inevitable from widespread diffusion of existing knowledge). The most significant results in these tests are their magnitudes relative to the baseline. A baseline that incorporates both the economics and sociology mechanisms, but holds everything else constant, yields over twice the annual growth of either mechanism in isolation (and more than the sum of the two mechanisms by themselves). The higher growth from interactively applying both mechanisms (as in the baseline configuration) suggests that it is possible that there is an optimal combination of expropriation and creation that maximizes long term growth. 32 5. DISCUSSION Our goal in this study is to reconcile the competing null hypotheses of sociology and economics with respect to knowledge flow and knowledge growth—sociology assumes that knowledge flow (diffusion) is highly viscous whereas economics assumes knowledge flows too easily and therefore discourages firms from investing in its creation. The concern is to obviate the potential problem that the two fields’ prescriptions might cancel one another out. Our vehicle for reconciliation is a simulation of knowledge flow that embodies empirical regularities emerging from empirical studies in both fields (as well as from the integrative field, management of technology). The empirical regularities themselves are interesting because they are remarkably similar across the fields despite their dissimilar data bases. Our simulation results provide some insight into why the competing perspectives persist despite the fact that they appear countervailing in the respective economics and sociology literatures. The baseline simulation reveals that knowledge flows fluidly (supporting the economics null hypothesis), but that such flow does not lead to an equilibrium of homogenous firms with zero growth (supporting the sociology null hypothesis). Thus each field is partially correct. We probe these findings further by examining the two fields’ prescriptions for knowledge growth—sociology’s diffusion (expropriation) and economics’ creation. Because the prescriptions stem from opposing hypotheses regarding knowledge flow one might suspect they too oppose one another. However, our results suggest just the opposite. Rather than canceling one another, we find that the combination of both mechanisms generates more than twice the growth of either mechanism in isolation. This is a possible explanation for why the opposing null hypothesis have persisted. Each field assumes the presence of the other field’s mechanism, and thus focuses on the remaining challenge. 33 The reason the two mechanisms are mutually reinforcing is that each becomes exhausted in isolation. Expropriation relies on heterophily for knowledge growth, but each time a firm’s knowledge grows, the gap between its knowledge and that of the host firm closes. Ultimately, expropriation consumes all heterophily, leaving no suitable host firms from which to expropriate knowledge. However, each expropriation event leads to a change in all firms’ shares (either directly or by increasing the total knowledge stock, and therefore the share of any firm that has not gained knowledge). These share changes threaten firms with lost share and thereby stimulate creation. This creation then produces additional knowledge and heterophily to fuel further expropriation as well as creation. Thus, integration of the fields of economics and sociology seems to point to a growth solution that is superior to either field’s isolated solution. The economics prescription augments the dynamics the sociologists point to and the sociology prescription does the same for the dynamics concerning economists. Our results are necessarily sensitive to our particular implementation of the empirical regularities. In particular, we chose naïve expropriation (random choice of a single host). Our results would likely change if we increase the number of hosts or the manner in which they are selected. Similarly we choose threat induced creation (firms invest when their share of industry knowledge decreases). An alternative implementation is creation driven by capacity (size/profits). This too will affect our results. Other conditions also could be delineated. We began by translating a set of stylized facts into nine principles comprising a theory about knowledge flow dynamics. Through the use of a computational model we have taken a seemingly disparate set of stylized facts and principles stemming from decades of empirical research and shown that a constructive integration is possible. Our integration of the various linear and nonlinear principles suggests a way toward reconciling the contrary economics and sociology 34 perspectives regarding knowledge flow dynamics. Even though their null hypotheses about knowledge flows are diametrically opposed, the prescriptions from each field actually complement each other. We find that growth prescriptions taking advantage of both perspectives will dominate those of either field in isolation. Though our language has focussed on firms and industries, the empirical regularities indicate that knowledge phenomena are remarkably similar across several levels of analysis. Thus we think that the insights here, once confirmed, may be generalizable to phenomena described by all three literatures: diffusion among individuals, organizational knowledge creation and acquisition, and the use of knowledge at industry levels of competition among firms. We have demonstrated that a computational approach to formalized modeling can play a constructive role in organization science. In this instance we have focused on testing experimental adequacy (McKelvey 1999c)—using a model to test the causal nuances among variables that are essential to predictive theory. Given the number of stylized facts—nine—and their mixed linearity and nonlinearity, a computational model is particularly well suited to the phenomena. This said, however, the problem of ontological adequacy remains. Though the constituent “model structures” (McKelvey 1999c)—that is, the structural elements of our model such as the lattice, density dependence and proximity effects, and the rules governing agent behaviors in our model— rest on the empirical regularities we discovered in the literature, the integrated behavior of the elements depicted in our model have not been confirmed by real world behavior, or even via human experiments. We do not know for a fact that real world outcomes would appear as predicted by our simulation. Further research in this regard is called for. Knowledge dynamics and knowledge management are increasingly important in organization science, management, and strategy (Teece 1987; Leonard-Barton 1995; Ashkenas et al. 1995; 35 Myers 1996; Boisot 1998). Seemingly “old” sociology is has more to say about knowledge dynamics than “new” sociology. And it is clear from our Table 2 that old sociology matches up very well with much more recent work in economics. This suggests that Pfeffer’s (1993, 1995) worries about the inroads of economics are well taken but irrelevant. As consumers of stylized facts, we simply scoured them up from wherever we could find them—in this case old sociology and new economics. While new sociology might be interesting to sociologists, it is a distraction that increasingly makes much of that field irrelevant to organization science and strategy. Scientists select facts from wherever, as we have done. If economists supply facts and sociologists do not…… It does not take a rocket scientist to figure out why cites of economists are growing. As our model demonstrates, economics is winning these days because it creates and diffuses and doesn’t worry about expropriation. New sociology does not create and thus does not have to worry about expropriation—there is nothing there to expropriate—as we discovered. 36 REFERENCES Abernathy, W., and K. B. Clark 1985 “Innovation: mapping the winds of creative destruction.” Research Policy, 14: 3–22. Abernathy, W. J., and J. M. Utterback 1978 “Patterns of industrial innovation.” Technology Review, 80: 40–47. Adams, J. D., and A. B. Jaffe 1996 “Bounding the effects of R&D: An investigation using matched establishment-firm data.” RAND Journal of Economics, 27: 700–721. Allen, T. J. 1977 Managing the flow of technology. Cambridge, MA: MIT Press. Anderson, P., and M. Tushman 1990 “Technological discontinuites and dominant designs: A cyclical model of technological change.” Administrative Science Quarterly, 35: 604–633. Aronson, J. L., R. Harre, and E. C. Way 1994 Realism Rescued. London: Duckworth. Arrow, K. 1962 “The economic implications of learning by doing.” Review of Economic Studies, 155–173. Ashkenas, R., D. Ulrich, T. Jick, and S. Kerr 1995 The Boundaryless Organization: Breaking the Chains of Organizational Structure. San Francisco, CA: JosseyBass. Ayer, A. J. 1959 Logical Positivism. Glencoe, IL: The Free Press. Barber, B. 1952 Science and the Social Order. Glencoe, IL: Free Press. Barber, B., and W. Hirsch 1962 The Sociology of Science. New York: Free Press. Barnes, B., D. Bloor, and J. Henry 1996 Scientific Knowledge: A Sociological Analysis. Chicago, IL: University of Chicago Press. Bell, R. 1992 Inpure Science: Fraud, Compromise and Political Influence in Scientific Research. New York: Wiley. Bennis 1956 "Values and Organizations in a University Social Research Group," American Sociological Review, 21, 555-63. Berelson, B., and G. A. Steiner 1964 Human Behavior: An Inventory of Scientific Findings. New York: Harcourt, Brace & World. Bhaskar, R. 1975 A Realist Theory of Science. London: Leeds Books (2nd ed. published in 1997. London: Verso). Boisot, M. H. 37 1998 Knowledge Assets: Securing Competitive Advantage in the Information Economy. New York: Oxford University Press. Bruderer, E., and J. V. Singh 1996 “Organizational evolution, learning and selection: A genetic algorithm based model.” Academy of Management Journal, 39: 1322–1349. Burrell, G. 1996 Normal Science, Paradigms, Metaphors, Discourses and Genealogies of Analysis. In S. R. Clegg, C. Hardy, and W. R. Nord (eds.), Handbook of Organization Studies: 642–658. Thousand Oaks, CA: Sage. Burt, R. 1997 "The contingent value of social capital." Administrative Science Quarterly. 42(2): 339-365. Campbell, D. T. 1959 “Methodological suggestions from a comparative psychology of knowledge processes. ” Inquiry, 2: 152–182. Campbell, D. T. 1974 “Evolutionary epistemology. ”In P. A. Schilpp (ed.), The Philosophy of Karl R. Popper, Vol. 14 (1 & 2): 413– 463. LaSalle, IL: Open Court. Campbell, D. T. 1988 “Descriptive epistemology: Psychological, sociological, and evolutionary. ” In D. T. Campbell, Methodology and Epistemology for Social Science: Selected Papers, (E. S. Overman (ed.)): 435–486. Chicago: University of Chicago Press. Carley, K. M. 1995 “Computational and mathematical organization theory: Perspective and directions.” Computational and Mathematical Organization Theory, 1: 39–56. Carley, K. M., and A. Newell 1994 "The nature of the social agent.” Journal of Mathematical Sociology, 19: 221–262. Carley, K. M., and D. M. Svoboda 1996 “Modeling organizational adaptation as a simulated annealing process.” Sociological Methods and Research, 25: 138–168. Chia, R. 1996 Organizational Analysis as Deconstructive Practice. Berlin: Walter de Gruyter. Cohen, M. D. 1996 “Organizational learning of routines: A model from the garbage can family.” In M. Warglien and M. Masuch (eds.). The Logic of Organizational Disorder: 183–192. Berlin: Walter de Gruyter. Cohen, M. D., J. G. March, and J. P. Olsen 1972 “A garbage can model of organizational choice.” Administrative Science Quarterly, 17: 1–25. Cohen, W. M., and R. C. Levin 1989 “Empirical studies of innovation and market structure.” In R. Schmalensee and R. Willig (eds.), Handbook of Industrial Organization: 1059-1107. Amsterdam: North-Holland. 38 Cohen, W. M., and D. A. Levinthal 1990 “Absorptive capacity: A new perspective on learning and innovation.” Administrative Science Quarterly, 35: 128–152. Coleman, J. S., E. Katz and H. Menzel 1966 Medical Innovation: A Diffusion Study. Indianapolis: Bobbs-Merrill. Crane, D. 1972 Invisible Colleges: Diffusion of Knowledge in Scientific Communities. Chicago: University of Chicago Press. Darr, E. D., L. Argote, and D. Epple 1997 “The acquisition, transfer, and depreciation of knowledge in service organizations: Productivity in franchises.” Management Science, 41: 1750–1762. de Regt, C. D. G. 1994 Representing the World by Scientific Theories: The Case for Scientific Realism. Tilburg, The Netherlands: Tilburg University Press. Dubin, R. 1958 The World of Work: Industrial Society and Human Relations. Englewood Cliffs, NJ: Prentice-Hall. Durfee, E. H. 1988 Coordination of Distributed Problem Solvers. Boston: Kluwer. Ettlie, J. E. 1985 “The implementation of programmable manufacturing innovations.” In D. D. Davis (ed.), Dissemination and Implemention of Manufacturing Processes Feyerabend, P. 1970 “Against method: Outline of an anarchistic theory of knowledge.” In M. Radner and S. Winokur (eds.), Minnesota Studies in the Philosophy of Science, Vol. IV: 17–130. Minneapolis, MN: University of Minneapolis Press. Fischer, K. H., and J. A. Hertz 1993 Spin Glasses. New York: Cambridge University Press. Foster, R. N. 1986 Innovation: The Attacker's Advantage. New York: Summit Books. Frost, T. S. 1997 “Imitation to innovation: The dynamics of Korea's technological learning.” Journal of International Business Studies, 28: 868–872. Fuller, S. 1993 Philosophy, Rhetoric, and the End of Knowledge: The Coming of Science and Technology Studies. Madison, WI: University of Wisconsin Press. Fuller, S. 1995 “Is there life for sociological theory after the sociology of scientific knowledge.” Sociology, 29: 159–166. Fuller, S. 39 1995 “On the motives for the new sociology of science.” History of the Human Sciences, 8: 117–124. Galbraith, F. R., E. E. Lawler III, and Associates 1993 Organizing for the Future. San Francisco, CA: Jossey-Bass. Geertz, C. 1971 The Interpretation of Cultures. New York: Basic Books. Gilbert, R., and D. Newberry 1982 “Pre-emptive patenting and persistence of monopoly.” American Economic Review, 72: 514–526. Gilfillan, S. 1933 Sociology of Invention. Cambridge, MA: MIT Press. Glaser, B. 1964 Organizational Scientists: Their Professional Careers. Indianapolis: Bobbs-Merrill. Gordon, G., S. Marquis and O. W. Anderson. 1962 Freedom and Control in Four Types of Scientific Settings," American Behavioral Scientist, 6, 39-42. Hagerstrand, T 1952 The Propogation of Innovation Waves. Lund: Sweden. Hagstrom, W. 1965 The Scientific Community. New York: Basic Books. Hahlweg, K., and C. A. Hooker 1989 “Evolutionary epistemology and philosophy of science.” In K. Hahlweg and C. A. Hooker (eds.), Issues in Evolutionary Epistemology: 21–44. Albany, NY: State University of New York Press. Hamberg, D. 1963 “Invention in the industrial research laboratory.” Journal of Political Economy, 71: 95–115. Hanson, N. R. 1958 Patterns of Discovery. Cambridge, UK: Cambridge University Press. Hilgartner, S., and S. I. Brandt-Rauf 1994 “Data access, ownership, and control: Toward empirical studies of access practices.” Knowledge, 15: 355–372. Hirshleifer, J. 1971 “The private and social value of information and the reward to inventive activity.” American Economic Review, 61: 561–575. Hiskes, A. L., and R. P. Hiskes 1986 Science, Technology, and Policy Decisions. Boulder, CO: Westview Press. Holton, G. 1962 “Scientific research and scholarship: Notes toward the design of proper scales.” Daedalus, 91: 362–399. Jaffe, A. B. 1986 “Technological opportunity and spillovers of R&D.” American Economic Review, 76: 984–1001. Jaffe, A. B. 40 1993 “Geographic localization of knowledge spillovers as evidenced by patent citations.” Quarterly Journal of Economics, 108: 577–598. Jewkes, J., D. Sawers, and R. Stillerman 1958 The Sources of Invention. London: Macmillan and Company. Kamien, M., and N. Schwartz 1982 Market Structure and Innovation. Cambridge, MA: Cambridge University Press. Kaplan, A. 1964 “Sociology of science.” In R. E. L. Faris (ed), Handbook of Modern Sociology: 852–881. Chicago, IL: Rand McNally. Kaplan, A. 1964 The Conduct of Inquiry. New York: Chandler. Katz, M., and C. Shapiro 1987 “R & D rivalry with licensing or imitation.” American Economic Review, 77: 402–420. Knott, A. M., and D. Bryce 1998 “Reconciling entry with entry-detering levels of intangible assets.” Working paper, The Wharton School, University of Pennsylvania. Kornhauser, W. 1953 Scientists in Industry, Berkeley: University of California Press. Kuhn, T. S. 1962 The Structure of Scientific Revolutions. Chicago, IL: University of Chicago Press. Leonard-Barton, D. 1995 Wellsprings of Knowledge: Building and Sustaining the Sources of Innovation. Cambridge, MA: Harvard Business School Press. Levin, R., W. Cohen and D. Mowery 1985 “R & D appropriability, opportunity and market structure: New evidence on some Schumpeterian hypotheses.” American Economics Review Proceedings, 75: 20–24. Levin, R., A. Klevorick, R. Nelson and S. Winter 1987 “Appropriating the returns from industrial R & D.” Brookings Papers on Economic Activity: 783–820. Levinthal, D. A. 1997 “Adaptation on rugged landscapes.” Management Science, 43: 934–950. Lieberman, M. 1987 “Learning curve, diffusion and competitive strategy.” Strategic Management Journal, 8: 441–452. Lloyd, E. A. 1988 The Structure and Confirmation of Evolutionary Theory. Princeton, NJ: Princeton University Press. Lynch, M. 1993 Scientific Practice and Ordinary Action: Ethnomethodology and Social Studies of Science. Cambridge, UK: Cambridge University Press. 41 Madsen, T and B. McKelvey 1996 "Darwinian Dynamic Capability: Performance Effects of Balanced Intrafirm Selection Processes". Academy of Management Best Papers Proceedings. Mansfield, E. 1958 The Economics of Technological Change. New York: W.W. Norton and Company, Inc. March, J. G. 1991 “Exploration and exploitation in organizational learning.” Organization Science, 2: 71–87. Masuch, M., and P. LaPotin 1989 “Beyond garbage cans: An AI model of organizational choice.” Administrative Science Quarterly, 34: 38–67. McKelvey, B 1997 “Quasi-natural organization science.” Organization Science, 8: 351–380. McKelvey, B. 1999a “Avoiding complexity catastrophe in coevolutionary pockets: Strategies for rugged landscapes.” Organization Science 10: McKelvey, B. 1999b “Self-organization, complexity catastrophe, and microstate models at the edge of chaos.” In J. A. C. Baum and B. McKelvey (eds.),Variations in Organization Science: In Honor of Donald T. Campbell: 279–307. Thousand Oaks, CA: Sage. McKelvey, B. 1999c “Toward a Campbellian realist organization science.” In J. A. C. Baum and B. McKelvey (eds.), Variations in Organization Science: In Honor of Donald T. Campbell: 383–411. Thousand Oaks, CA, Sage. Merton, R. K. 1938 “Science, technology, and society in 17th century England.” Osiris, 4: 360–632. Merton, R. K. 1942 “Science and technology in a democratic order.” Journal of Legal and Political Sociology, 1: 115–126. Mirskaya, E. 1990 “Once again on the subject of sociology of science.” Science of Science, 10: Mitchell, M. 1996 An Introduction to Genetic Algorithms. Cambridge, MA: MIT Press. Moch, M. K. 1977 “Size, centralization, and organizational adoption of innovations.” American Sociological Review, 42: 716–725. Mohrman, S. A., and A. M. Mohrman Jr. 1993 “Organizational change and learning.” In J. R. Galbraith, E. E. Lawler III, and Associates, Organizing for the Future: The New Logic for Managing Complex Organizations: 87–108. San Francisco, CA: Jossey-Bass. Mulkay, M. 1969 “Some aspects of cultural growth in the natural sciences.” Social Research, 36: 22–52. Murphy, R. 42 1994 “The sociological construction of science without nature.” Sociology, 28: 957–974. Myers, P. S. 1996 Knowledge Management and Organizational Design. Boston, MA: Butterworth-Heinemann. Nelson, R 1989 “Capitalism as an engine of progress.” Research Policy, 19: 193–214. Nonaka, I. 1990 “Redundant, overlapping organization: A Japanese approach to managing the innovation process.” California Management Review 32: 27–38. Ogburn, W. F. 1922 Social Change. New York: Heubsch. Padgett, J. F. 1997 “The emergence of simple ecologies of skill.” In W. B. Arthur, N. Durlauf, and D. A. Lane (eds.), The Economy as an Evolving Complex System, Proceedings of the Santa Fe Institute XXVII: 199–221. Reading, MA: Addison-Wesley. Pels, D. 1994 “Karl Mannheim and the sociology of scientific knowledge: Toward a new agenda.” International Sociological Association (association paper). Pennings, J. 1987 “Technological innovations in manufacturing.” In J. Pennings and A. Buiterdam (eds.), New Technology as Organizational Innovation: 197–216. Cambridge, MA: Ballinger. Perrow, C. 1994 “Pfeffer slips.” Academy of Management Review, 19: 191–194. Pettigrew, A. M. 1973 The Politics of Organizational Decision Making. London: Tavistock. Pfeffer, J. 1993 "Barriers to the advancement of organizational science: Paradigm development as a dependent variable." Academy of Management Review, 18, 599–620. Pfeffer, J. 1995 "Mortality, reproducibility, and the persistence of styles of theory. Organization Science 6, 681–686. Pfeffer, J. 1997 New Directions for Organization Theory. New York: Oxford University Press. Price, D. J. de S. 1961 Science Since Babylon. New Haven, CT: Yale University Press. Price, D. J. de S. 1963 Little Science, Big Science. New York: Columbia University Press. Price, D. J. de S. 1986 Little Science, Big Science…and Beyond. New York: Columbia University Press. 43 Prietula, M. J., K. M. Carley, and L. Gasser 1998 Simulating Organizations: Computational Models of Institutions and Groups. Cambridge, MA: MIT Press. Reinganum, J. 1983 “Uncertainty, innovation and the persistence of monopoly.” American Economic Review, 73: 741–748. Reinganum, J. 1985 “Innovation and industry evolution.” Quarterly Journal of Economics, 99: 81–99. Reingold, N. 1979 The Sciences in the American Context: New Perspectives. Washington, DC: Smithsonian Institution Press. Rescher, N. 1990 Evolution, Cognition, and Realism: Studies in Evolutionary Epistemology. Lanham, MD: University Press of America. Roberts, E. 1968 “Entrepreneurship and technology.” Research Management, 11: 249–266. Rogers, E. M. 1995 Diffusion of Innovations, (4th ed.) New York: The Free Press. Romer, P. 1986 “Increasing returns and long-run growth.” Journal of Political Economy, 94: 1002–1037. Ryan, B., and N. C. Gross 1943 “The diffusion of hybrid seed corn in two Iowa communities.” Rural Sociology, 8: 15–24. Sah, R., and J. Stiglitz 1986 “The architecture of economic systems: Hierarchies and polyarchies.” American Economic Review, 76: 716– 727. Salant, S. W. 1984 “Pre-emptive patenting and the persistence of monopoly: Comment.” American Economic Review, 74: 247– 250. Scherer, F. 1965 “Firm size, market opportunity and output of patented invention.” American Economic Review, 55: 1097–1125. Scherer, F. 1980 Industrial Market Structure and Economic Performance. Chicago, IL: Rand McNally. Schumpeter, J. A. 1934 The Theory of Economic Development. Cambridge, MA: Harvard University Press. Schumpeter, J. A. 1942 Capitalism, Socialism and Democracy. New York: Harper and Row. Scott, J. 1984 Firm Versus Industry Variability in R & D Intensity: in Z. Griliches (ed) R & D, Patents and Productivity. Chicago, University of Chicago Press. Shepard, H. 44 1956 "Nine Dilemmas in Industrial Research" Administrative Science Quarterly, 1, 295-309. Solow, R. M. 1957 “Technical change and the aggregate production function.” Review of Economics and Statistics, 39: 312–320. Stein, M. K. 1962 “Creativity and the scientist.” In B. Barber and W. Hirsch (eds) The Sociology of Science: 329–343. New York: Free Press. Steneck, N. H. 1975 Science and Society: Past, Present, and Future. Ann Arbor, MI: Univeristy of Michigan Press. Stinchcombe, A. L. 1968 Constructing Social Theories. Chicago: University of Chicago Press. Suppe, F. 1977 The Structure of Scientific Theories. Chicago: University of Chicago Press. Suppe, F. 1989 The Semantic Conception of Theories and Scientific Realism. Champaign, IL: University of Illinois Press. Tarde, G. 1903 The Laws of Imitation. New York: Holt. Teece, D. 1987 “Profiting from technological innovation: Implications for integration, collaboration, licensing and public policy.” In D. Teece (ed.), The Competitive Challenge: Strategies for Industry Innovation and Renewal: 185–219. Cambridge, MA: Ballinger. Teece, D. 1987 The Competitive Challenge: Strategies for Industry Innovation and Renewal. Cambridge, MA: Ballinger. Thompson, P. 1989 The Structure of Biological Theories. Albany, NY: State University of New York Press. Usher, A. P. 1929 History of Mechanical Inventions. New York: McGraw Hill. Warglien, M. 1996 “Learning in a garbage can situation: A network model.” In M. Warglien and M. Masuch (eds), The Logic of Organizational Disorder: 163–182. Berlin: Walter de Gruyter. Watson, J. D. 1968 The Double Helix: A Personal Account of the Discovery of the Structure of DNA. New York: Atheneum. Weber, M. 1947 The Theory of Social and Economic Organization, A. H. Henderson and T. Parsons (eds.). Glencoe, IL: Free Press (first published in 1924). Weisbuch, G. 1993 Complex Systems Dynamics: An Introduction to Automata Networks. Reading, MA: Addison-Wesley. 45 46 Table 1. Contrasting the perspectives Goal Perspective Unit of analysis Unit of innovation Relevant environment Diffusion premise Incentive premise Sense of progress Stimulus to innovation The innovator Ownership Sociology of Science How and why does knowledge grow? Positive Scientist New knowledge Scientific field--Society Technology Economics How to increase innovation (to thereby economic growth) Normative Firm Invention Industry--Economy Necessary to facilitate growth of knowledge Scientists intrinsically motivated (seek recognition or priority) Cumulativism (incremental) Mechanistic determinism Response to necessity (market pull) Foci of interest determined by social forces Many inventors trying to solve problem, each contributing small piece Priority rights Suppresses incentives to innovate (by reducing appropriability) Innovation requires incentives (appropriability) Transcendentalism (episodic) Mystic determinism Arises from immanent development of science (technology push) Great man, act of insight Property rights 47 Table 2. Summary of empirical regularities across levels of analysis and academic fields LEVEL OF ANALYSIS PATERNS of GROWTH Logistic growth of innovation Individual (Tarde 1903) (Rogers 1962; Rogers 1995) Organization Episodic growth Bifurcation bi-modal equilibria (Cohen and Levinthal 1990) (Madsen ) (Knott and Bryce 1998) (Jewkes, D. et al. 1958; Hamberg 1963; Cohen and Levin 1989) Diminishing returns INCENTIVES CAPACITY TO INNOVATE CAUSAL FACTORS Density dependence Wealth effects Sharing/spillovers technical proximity /homophily heterophily /external ties geographic proximity Industry/Field (Mansfield 1958) (Price 1961;1963) (Foster 1986) (Crane 1972) (Holton 1962) (Abernathy and Utterback 1978) (Anderson and Tushman 1990) Country/Economy (Schumpeter 1934; Schumpeter 1942) (Romer 1986) (Price 1961;1963) (Scherer 1965; Scott 1984; Levin, Cohen et al. 1985) (Price 1961; 1963; Hagstrom 1965) (Rogers 1962; 1995) (Rogers 1962; 1995) (Coleman, Katz et al. 1966) (Price 1961; 1963) (Rogers 1962; 1995) Cohen, 1989 #26 (Lieberman 1987) (Jaffe 1986; Jaffe 1993; Adams and Jaffe 1996) (Price 1961; 1963) (Burt 1997) (Crane 1972) (Hagerstrand 1952) (Allen 1977) (Darr, Argote et al. 1997) (Jaffe 1986; 1993; Adams and Jaffe 1996) (Frost 1997) Economics literature Sociology literature Management literature Key to Phenomena: Logistic growth-Typically growth follows an S-curve under nominal conditions (slow initially, acceleration, then deceleration toward saturation). Episodic growth-Knowledge growth exhibits periods of incremental growth interspersed with discontinuities. While the discontinuity conditions are not well defined, they are most likely when fields shift technologies Bifurcation-Under some conditions (not well defined) populations segregate into clusters of high knowledge and low knowledge Diminishing returns-At some point, additional investments are less effective in producing new knowledge Density dependence-Knowledge growth appears to be an n-shaped function of density: with two few firms, knowledge grows slowly and appears linear (rather than logistic) because firms have little to learn from one another, and pose little threat to one another. Similarly with too many firms, firms have little incentive to produce new knowledge (because of diminishing returns) Wealth effects-Entities that are well-endowed make greater investments in knowledge Technical proximity-Entities that are similar are more likely to share knowledge Heterophily-However, if entities are too similar (i.e., they have identical knowledge) there is nothing to share Geographic proximity-Knowledge is more readily shared by entities in close physical proximity 48 Table 3. Treatment of each of the empirical regularities in the simulation Empirical regularity Treatment Logistic growth Tested in validation: Kt should take this form under the baseline Episodic growth Defer to follow-on: Requires multi-population study Bi-modal equilibria Defer to follow-on: Should emerge under some conditions (not yet known) Built into model design: Model assumes firms invest sufficiently to overcome diminishing returns Tested in validation: Vary technical and geographic space parametrically, expect higher growth in denser space Built into model design: Firms with greater share of knowledge create more absolute knowledge on average Built into model design: Built into design of expropriation mechanism-expropriation inverse function of distance Tested in validation: Vary distribution of knowledge endowment parametrically, expect higher growth with greater heterophily Diminishing returns Density dependence Wealth effect Proximity effects Heterophily 49 Figure 1. Simulation Flow Diagram Birth of population i=1 to n Gi=Ti=[25,25], Ti= B[0,t] Ki = N() Create new knowledge Cit = [(K i(t-1) / Ki(t-1)) – (K it / K it)] * ( K it) Appropriate existing knowledge Eit = [(Kjt- Kit) / [(gij2 + tij2)] Update stocks Kit+1 = Kit+ Eit + Cit Yes > 50 periods? Plot distribution of knowledge shares STOP Plot distribution of knowledge shares STOP No Share convergence? 50 Figure 2. Results from single run to show growth and share evolution Figure 2a. Growth in knowledge stock Knowledge stock Nominal Case 2.50E+06 2.00E+06 1.50E+06 1.00E+06 5.00E+05 0.00E+00 0 20 40 Years/Periods Figure 2b. Beginning histogram of knowledge shares 35 30 25 20 15 10 5 0.0015625 0.0046875 0.0078125 0.0109375 0.0140625 0.0171875 0.0203125 0.0234375 0.0265625 0.0296875 0.0328125 0.0359375 0.0390625 0.0421875 0.0453125 0.0484375 0.0515625 0.0546875 0.0578125 0.0609375 0.0640625 0.0671875 0.0703125 0.0734375 0.0765625 0.0796875 0.0828125 0.0859375 0.0890625 0.0921875 0.0953125 0.0984375 Figure 2c. Ending histogram of knowledge shares 50 40 30 20 10 0.0015625 0.0046875 0.0078125 0.0109375 0.0140625 0.0171875 0.0203125 0.0234375 0.0265625 0.0296875 0.0328125 0.0359375 0.0390625 0.0421875 0.0453125 0.0484375 0.0515625 0.0546875 0.0578125 0.0609375 0.0640625 0.0671875 0.0703125 0.0734375 0.0765625 0.0796875 0.0828125 0.0859375 0.0890625 0.0921875 0.0953125 0.0984375 60 51 Annual knowledge growth Figure 3. Effects of density on knowledge growth 0.60 0.50 0.40 mean 0.30 std dev 0.20 0.10 0.00 0 2 4 6 Density (firms per unit area) 8 10 52 Annual knowledge growth Figure 4. Effects of heterogeneity on growth 0.10 0.08 0.06 0.04 mean stdev 0.02 0.00 0 500 1000 1500 2000 Dispersion in initial knowledge (mean=2000) 53 Mean annual knowledge growth Figure 5. Interactive effects on knowledge growth 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 h=500 h=250 h=1500 0 5 Firms per square area 10 54 Figure 6a. Effects of density on share convergence* Share dispersion after 50 periods *Note mean share = .01 0.035 0.030 0.025 0.020 0.015 mean std dev 0.010 0.005 0.000 0 5 10 Density (firms per unit area) Share Dispersion after 50 Periods Figure 6b. Effects of heterogeneity on convergence 0.030 0.020 mean 0.010 stdev 0.000 0 500 1000 1500 2000 Initial dispersion in firm knowledge Final Dispersion (standard deviation) in knowledge shares Figure 6c. Interactive effects on share convergence 0.04 0.03 h=250 h=500 0.02 h=1000 h=1500 0.01 0 0 5 Firms per square area 10 55 Mean annual growth Figure 7a. Effects of prescriptions on growth 0.35 0.30 0.25 0.20 0.15 0.10 0.05 0.00 Unambiguous Expropriation Ambiguous Expropriation No Expropriation Unambig Causal ambiguity None Creation conditions Figure 7b. Effects of prescriptions on convergence Mean sigma share 0.05 0.04 0.03 Unambiguous Expropriation 0.02 Ambiguous Expropriation 0.01 No Expropriation 0.00 Unambig Causal ambiguity Creation conditions None