Measuring the Knowledge Base: A Program of Innovation Studies

M M e e a a s s u u r r i i n n g g t t h h e e K K n n o o w w l l e e d d g g e e B B a a s s e e A A P P r r o o g g r r a a m m o o f f I I n n n n o o v v a a t t i i o o n n S S t t u u d d i i e e s s Society, NGOs, etc. WEB based indicators Loet Leydesdorff Andrea Scharnhorst M EASURING THE K NOWLEDGE B ASE : A P ROGRAM OF I NNO VATION S TUDIES @ Loet Leydesdorff 1 & Andrea Scharnhorst 2 , 2003 1. Amsterdam School of Comm unications Re search (ASCoR), University of Amsterdam, http://ww.leydesdorff.net 2. Networked Research and Digital Inform ation, The Netherlands Institute for Scientific Inform ation Services, Royal Academ y of Arts and Sciences (KNAW), Amsterdam, http://ww.niwi.knaw.nl/nerdi This report was written for the “Förderinitiative Science Policy Studies ” of the German Bundesministerium für Bildung und Forschung. The program was organized by Rainer Hohlfeld of the Berlin-Brandenburgische Akademie der W issenschaften. The authors acknowledge the support of these organizations. Amsterdam, March 2003. ii Table of Contents S UMMARY ........................................................................................................................... iv 1. S YSTEMS OF I NNOVATION , “M ODE 2”, AND T HE T RIPLE H ELIX ........................................ 1 1.1 The Organization of Knowledge-Based Communications ................................... 1 1.2 The science system under the condition of globalisation...................................... 3 Co-existence, co-evolution, and lock-in ........................................................ 4 Globalization, stabiliza tion, and reflexivity .................................................. 6 1.3 From theories to empirical evaluations ................................................................. 7 Units of analysis versus units of operation .................................................... 7 The time dimension ........................................................................................ 8 Operationalization, data, and context ........................................................... 9 1.4 Statistical analysis, models, and simulations ...................................................... 11 Multidimensionality and the reduction of uncertainty ................................ 11 History, co-evolution, and the emergence of new systemness ..................... 11 Models and simulations ............................................................................... 15 1.5 Conclusion........................................................................................................... 17 2. C AN THE N EW M ODE OF K NOWLEDGE P RODU CTION BE M EASURED ? ............................ 20 2.1 The reflection of boundary-spanning mechanisms ............................................. 21 2.2 Methods and materials ........................................................................................ 23 2.3 Webometric data ................................................................................................. 25 2.4 Testing for Systemness ....................................................................................... 27 2.5 Conclusion........................................................................................................... 28 3. T HE C OMMUNICATIVE T URN IN THE S TUDY OF K NOW LEDGE -B ASED S YSTEMS ............. 30 3.1 The need for “reflexive indicator research” ........................................................ 31 The Organization of Reflexivity within the Systems .................................... 32 3.2 Indicators as representations of codified communications ................................. 35 Web Indicators ............................................................................................. 38 3.3 A Program of Innovations Studies ...................................................................... 39 3.4 Policies of Innovation: Innovation of Policies? .................................................. 41 References.......................................................................................................................... 43 iii S UMMAR Y In a reflection of new developmen ts in science and society, Gibbons et al . (1994) proposed to make a distinc tion between “Mode 2” and “Mode 1” types of knowledge production. Whereas “Mode 1” refers to the traditiona l shape of science, largely confined within institutional settings , “Mode 2” is comm unication driven. Organized knowledge production can then be considered as the codification of communication. Communications leave traces that can be studied a s indicators. Institu tions can be considered as retention mechanisms func tional for the reproduction of ever more complex, that is, scientific and knowledge-based, communications. From this perspective, “national systems of innovation” compete as niches in terms of their problem-solving capacities (Nelson & Wint er, 1982). In the Triple Helix m odel the institutions are analyzed in terms of university-industry-government relations. The network is continuously reshaped by knowle dge-based innovations that result from inventions at one level and feedback at other network levels. Knowledge-based communications and their re-combinations thus drive the institutional refo rm of the political economy into a knowledge-based ec onomy. The networks differentiate further while importing knowledge in th e form of innovations. As innovations take place at interfaces, the competitive advantages in a knowledge-bas ed economy can no longer be attributed to a single node in the network. The networks coordinate the sub-dynamics of (i) wealth production, (ii) organized novelty production, and (iii) private appropriation versus public control. Boundary- spanning mechanisms can be expected to ch ange systems by providing new options for innovation. The Internet can be considered as a boundary-spanning mechanism at the global level. It relates different parts of the knowledge pr oduction, diffusion, and control system to one another. Academic, econom ic, and public spheres can use the same media for the representations. The expectation of global exchange relations (“globalization”) changes the knowledgeable options within the lower-level systems by making new “variation” and other “selec tion” criteria possible. The focus on communication enables us to ope r ationalize the research q uestions in terms of indicators by using the ma thematical theory of comm unication. Fo r example, the systems of innovation can be measured in terms of interfaces among communications about new products and/or technological processes. Are the innovations under study incidental or systemic? The degree of syst emic behaviour can be tested correlating different types of data. For example, sys temness can be com pared with the historical development of time-series data. Are the em erging densities in network relations also reproduced? Two theories of communication provide the heuristics: (i) Luhm ann’s (1984 and 1990) sociological theory of co mmunication with its emphasis on functional differentiation provides hypotheses, and (ii) Shannon’s (1948) mathematical theory of communication can be used for the operationaliz ation. The com bination of thes e two theories with a very different status—i.e., a combination of theo ry and methods—enables us to update and inform empirical hypotheses about how the know ledge base transform s the institutional relations of an increasingly knowledge-based society. Policy implications are specified. iv Chapter One S YSTEMS OF I NNOV A TION , "M ODE 2", A ND T HE T RIPLE H ELIX In the period before the oil crises of the 1970s , that is, in the decades after World War II, social functions were deliber ately organized into instit utions on a one-to-one basis (Merton, 1942; Bush, 1945). Academic funding, for exampl e, was by and large based on internal processes of peer review (Mulka y, 1976). The oil crises of the 1970s, however, made clear that advanced industrial nations could outcompete low-wage countries only on the basis of the systematic exploitation of their respective knowledge bases (e.g., Nelson & Winter, 1977 and 1982; Freeman, 1982; cf. Freem an & Soete, 1997). The policy implications of this conclusi on were not simple. Innovation is based on knowledge flowing and recombining acros s interfaces (Kline & Rosenberg, 1986). Knowledge flows both within and across institutional boundaries. The crossing of institutional boundaries can be expected to imply transaction costs (Williamson, 1975), but it may also generate longer-term reve nues and synergies (e.g., Faulkner & Senker, 1995). The transaction costs can be consider ed as investments in establishing new relations of collaboration and competition. Thus, a dynamic view of a knowledge-based system can be generated in which instituti onal agents have to translate between short- term and longer-term optim izations using a variety of criteria. The trade-off between transaction costs and surplus value can becom e visible in the changing patterns of collabora tion. At the interfaces, new fo rms of organization can be invented. For example, the increase of co-author ed papers can be used as an indicator of the increasingly networked ch aracter of research (Wagner, 2002). The newly em erging structures can be expected to reconstruct the old ones. These reconstructions may take place at different levels of the complex system of knowledge production and control. 1.1 The Organization of Kn owledge-Based Communications After the second oil crisis of 1979, the te chno-sciences such as biotechnology, information technologies, and new material s rapidly became the top priorities for stimulation policies at the national level in the advanced industrial countries (OECD, 1980). These “platform sciences” (Langford & Langford, 2001) are based on the assumption that rearrangem ents across disciplina ry lines may generate comp etitive advantages, through synergies in the knowledge base, that can be exploited for economic development. Previous attempts for more dir ect m ission-oriented steering of the sciences had at that time already been evaluate d as less successful (e.g., Van den Daele et al ., 1977; Studer & Chubin, 1980). 1 The stimulation of university-industry relations becam e a second point of attention of S&T policy makers. Why had some countries b een m ore successful than others in the technological exploitation of their knowle dge base (Hauff & Scharpf, 1975; Irvine & Martin, 1984)? Why were certain sectors (e.g., chemistry, ai rcraft) within countries more successful than others in exploiting their respective knowle dge bases (Nelson et al. , 1982)? Could lessons be learne d from best practices acro ss sectors, and might such practices be transferable from one national context to another? It is far from obvious at which level one can stimulate a knowledge-based innovation system. Should one focus on optimization at the level of the institution al arrangements (e.g., Rothwell & Zegveld, 1981), or rather stimulate sp ecific science-technologies (Leydesdorff & Gauthier, 1996; cf. Li ssenburgh & Harding, 2000)? The uncertain definition of the unit of analysis for studyi ng a knowledge-b ased system of innovation in terms of nations, sectors, technologies, regions , etc., brings in new players as potentially important contributors. From the mid-1980s onwards, the European Un ion heavily experimented in a series of Framework Programs with policies for sc ience, technology, and innovation. Both transnational cooperation and c ooperation across sectors were systematically stim ulated. Within the newly emerging context of the European Union, regions tried to prom ote their position as a relevant level for systematic development of the knowledge infrastructure. In the U.S.A. the national system experim ented with granting rights to paten t to universities (the Bayh-Dole Act of 1980), along with systematic efforts to raise the level of knowledge-intensity within i ndustry, both at the level of the states and by stimulation programs at the level of the federal government (Etzkowitz, 1994). How successful have these attempts been? Has a European system of innovations emerged in competitio n to the underlying “n ational” system s? To which extent have European regions (e.g., Flanders and Catalonia) been successful in establishing their own systems of innovation (Leydesdorff, Cooke, & Olazaran, 2002)? Have sectors (e.g., ICT) been developed using patterns of innovation diff erent from those that were established in a previous cycle of industrial devel opment (Barras, 1992)? Have patterns of collaborations across national boundaries, secto rs, and disciplinary affiliations changed, and what have been the effects of these change s in te rms of the quali ty and quantity of the respective outputs? How can systems of know ledge-based innovation be assessed in terms of their relevant outputs? These empirical questions became even more pressing during the 1990s with the emergence of the Internet, which added a new dimension to the existing systems of innovation. The resultant global perspective makes another evaluati on possible. On the one hand, the Internet was expected to increase the chances for new partners to participate in knowledge production processes by providing almost free access to information sources worldwide. For example, South- and East-Asian countri es seemed initially better equipped than European nations for moving ahead in this new era of e-business, give n their specif ic mix of human resources and thei r flexibility in recomposing industrial structures and 2 knowledge infrastructures (Freeman & Perez, 1988). On the other hand, it has been noted that the Internet tends to reproduce the st ratification in the access to inform ation and perhaps even increases the barriers of entry to markets. How should European countries act and react? Would it be sufficient to stimulate ongoing processes of change, or should new framewor ks be proposed that enable collaborating partnerships to be developed? Which criteri a for the optimization should be used (e.g., national, transnational, sectoral)? Thus th e stage was set for a profound reformulation of S&T policy-making in the early 1990s. 1.2 The science system under the condition of globalisation The rise of the Internet and the global dime nsion raised a new quest ion in S&T policies. How do “internationalization” and “globalization” affect sy stems of organized knowledge production and control (Crawford, Shinn, and Sörlin 1993; Cozzens et al . 1990; Ziman 1994)? In a policy-oriented reflecti on of these developments, Gibbons et al. (1994) proposed to make a distinction between “Mode 2” and “Mode 1” t ypes of the production of scientific knowledge. Whereas “Mode 1” refers to the previ ous shape, largely confined within institutional settings, “Mode 2” is communication driven. Knowledge can then be considered as a codification of communications. Scientific knowledge can be contain ed within an institution or even an ind ividual agent as “tacit knowledge,” or it can be “published.” The dimensions of knowledge in private and public arenas resonate with the arrangement s of industry-government relations within political economies. The knowledge component adds a new dimension to the so-called “differential productivity growth puzzle” (Nel son & Winter, 1975) between sectors in the economy, and to the relations between public control and privat e appropriation of competitive advantages. The competitiv e advantages of nations are increasingly dependent upon scientific and technologica l progress (Krugm an, 1996). During the 1990s, knowledge-intensity thus becam e a driver of the reform of the political economies. Etzkowitz & Leydesdorff (1995) proposed to model the evolutionary dynamics of the knowledge-based economy as a “triple helix of university-industry-government relations” (cf. Leydesdorff & Etzkowitz, 1996 and 1998). According to the Tr iple-Helix model, three functions have to be fulfilled within a knowledge-based system of innovations: (i) wealth generation in the economy, (ii) novelty and innovation that ups et the equilibrium seeking mechanisms in (sem i-)market system s, and (iii) public control and private appropriation at the interfaces between ec onomic and scien tific production system s. The specific arrangements require interfaces between the three function systems to be institutionalized as a knowledge infrastructu re. However, the local s tabilizations and trajectories are under pressure fr om global developments. The latter can be considered as the prevailing regime. In the Trip le Helix mo del this ov erlay is operationaliz ed as the communication network between the institu tio nal partners. The knowledge-based regim e 3 of expectations guides the negotiations among the partner s in the Triple Helix as an overlay system of communicati on, negotiations, and programm ing. Under the condition of globalization, local nich es can gradually be dissolved because new horizons offer other options. As the relative we ights of relations in a network change by ongoing processes of collabora tion, appropriation, and compe tition, the new balances and inbalances can be expected to generate a feedback in the knowledge infrastructure at other ends. The (sub)systems can then be expe cted to re combine into new solutions with degrees of success. However, knowledge flows between systems can also be expected to be stabilized and further developed within the historical institution s that have served us hitherto. In this way, institution s may survive in a changing environm ent. The institutional arrangements provide the stabi lity that is necessary to access the ultra- stability of the globalized regimes (Luhmann, 2002, at p. 396). Note that the new options are locally im porte d from the global level as expectations, and therefore these reconstruc tions are knowledge-based. Knowledge-based innovation increasingly makes the innovate d systems also knowledge-based. The knowledge infrastructure is provided by networks am ong industries, academia, and governm ents. These three actors are interwoven as institutions in a network which carries the resulting knowledge base. The latter can be considered as a system of communications on top of the institutional carriers. W hile the institut ional networks integr ate, the communication systems can be expected to differentiate in term s of functions (Luhmann, 1984 and 1990). The reconstructed codifications enable both participants and observers to specify and change the systems under study inventively, that is, by proposing and codifying new combinations. This hypothesis will guide us here to map the system s under study. A knowledge-based cultural evol ution is thus envisaged which abstracts from and experminents with both the “natura l” and instit utional bases of the ca rriers at the level of the knowledge-infrastructure. The focus is on the overlay of communications. This knowledge base is meta-stabilized or globali zed as an operation on the stabilizations provided by the infrastructures. The natural a nd institutional bases can then be consid ered as givens and constraints on which the know ledge-based system operates by innovating both itself and its boundary conditions. Co-existence, co-evolution, and lock-in We emphasize that a systems-theoretical appro ach focusing on the network level allo ws for a specific perspective on the interactive m echanisms between the subsystem s. This perspective can be enriched with the results of institutional analyses at the lower levels. However, it adds to the latter by providing a persp ective on th e knowledge-based dimension that “catalyzes” the innovative proc esses of reorganizati on. This perspective, therefore, merits fu rther exploration. For example, the differently codified system s can develop co-evolutions when a coupling between two of them is m ade structural. Depe nding on the quality of the intera ction, that is whether the interaction is supportive or competitive, one can expect co-existence and/or 4 selection. Co-existence is the outcome of continuously generated stabilities between counteracting mechanisms within the overall system . The co-evolution then generates a process of “mutual shaping” between the co -evolving systems. W hen a third dynamic is added to such a co-evolutionary model, prev ious arrangements can be dissolved at a global level. The system can th erefore be expected to shape stable trajectories and global regimes endogenously (Leydesdorff & Van den Besselaar, 1998). The possible outcomes of the interplay among three subdynam ic s can be richer than in the case of two dynamics, as a new quality of interactions is introduced. Chaotic trajectories are also possible at this level. Wh ile there is no longer an essential solution or harmony in such “trialectics,” one can expect the “endless frontier” (B ush, 1945) to be replaced with an “endless transition” (Etzkowitz & Leydesdorff, 2000). The production of both new partners for interactions and new types of interactions is an endogenous feature of such complex dynamics. Each new dimension raises the number of possible realizations exponentially. 1 New institutional forms, for example, can serve as boundary-spanning mechanism s that enable the participants to specify new variations . These processes can be modeled. Although a prediction of specific variants with the h elp of these models rem ains principally impossible—because the model abstracts fro m the substantive content—the boundary conditions for successful variants can be test ed in simulatio ns (Ebeling & Scharnhorst, 2000). The systems of innovation can be expect ed to compete in their uphill search for new solutions and stabilizations (Kauffman, 1993; Frenken, 2000). For example, trajectories can be formed by chance p rocesses at interfaces when technologies are “locked-in” within indus tries (e.g., the QWERTY keyboard; David, 1985). Alternatively, scientific expertise and state apparatuses may begin to co-evolve such as in the energy and the health sect ors. The state and industry can also become “locked-in” like in the former Soviet-Union. In novation policies have to vary in terms of which “lock-ins” (between co-evolving subsyste ms) are prevalent, and o n the assessment of how these patterns can be systematically disturbed by a third dynamic. For exam ple, the market mechanism can reintroduce flexibil ities in the case of a bureaucratic lock-in, whereas, in the case of a tec hnological lock-in, government in terventions m ay be needed to break monopolistic tendencies. Thus the op tim ization of policies becomes increasingly dependent on the evolutionary assessment of the knowledge-based system. This dependency relationship tends to invert the cause-effect relation ship in political steering processes. The room available for steering is inc reasingly determined by the systems to be steered. However, the self-organ ization of the latter at the global level can be reflected and then made the subject of informed and knowledge-based policy-making. Whereas the lock-in phenomena can m ake the system robust agains t steering for long periods of time, this process of stabiliz ation is under perm anent pressure from an evolutionary perspective. The global pers pective destab ilizes local stabilizations. Systemic innovations are possible because of this destab ilization. However, destabilization can also lead to a collapse. The challenge for a complex system is to 1 Two dice provide 6 2 (=36) po ssible combina tions, while t hree dice provide 6 3 (= 216) combinations. 5 balance between the ability to innovate and stability. Innova tion policies can then reflect and/or disturb this balance, but without fu rther development of their own knowledge base the policy makers can be outcompeted by the knowledge bases of the (sub)systems to be steered. Globalization, stabiliza tion, and reflexivity A global regime results from closer in teractions among rela tively autonomous subsystems, for example, in terms of ne tworks of university -industry-government relations. The global regime is propelled as a complex dynamic am ong differently coded communication systems (e.g., the economy, sc ience, and policy-m aking). The network overlay emerges as a new unit of evoluti on. When this structural innovation can be temporarily stabilized, it may begin to coevolve with the subdynam ics upon which it builds. Given the selection pressure of the new dimension, old institution al arrangements m ay survive, but will probably have to adapt their fu nction, as well as their form, to the new environment. The hypothesis of a “global agent” can be formulated as the expectation of change because of the selection pressure on institutional arrangem ent s. The global agent, however, remains a network function and conseque ntly operates as a regime of uncertain expectations. It is not a steering age nt with a positive agenda, but a glo bal regime that exerts selection pre ssure by being pending. One should not reify this “global agent” as a metabiology or a supersystem. The various systems of expectations inte ract and produce an overlay within the syste m of interactions. This overlay globalizes the system by making other representations available com pared to those that could already be envisaged from the previously available pers pectives. These recombinations can then be attributed to a next-order or “global” system , but their possibility is only a result of an internal dynam ic that is added to the system as its “globalization.” This globaliza tion can be entertained reflex ively theref ore enriching the system. It provides a future-oriented knowledge-b ase that innovates th e historical systems with hindsight. The ability to innovate is based on inventing new codifications by reflexively rearranging at the borders. The dynamics of science and technology have induced a reflexive turn in other social systems. The effects of bei ng increasingly knowledge-based have first been reflected in science and technology studies (e.g., Whitley, 1984) . The “reflexive turn” in these stu dies (Woolgar & Ashmore, 1988) im plied that the idea of a single and universal yardstick—as searched for in the philosophy of science (e.g., Popper, 1935)—had to be given up in favour of codes that are continuously cons tructed and reconstructed. Unlike universal standards, asymmetry can be expected to prevail in exchange relations among systems and subsystems (Gilbert & Mulkay, 1984) because the systems exchange on the basis that they have different substances in stock. For example, the political system is initially interested in resu lts from the science system that inform decision making and policies without being unduly burdened with the 6 overwhelming uncertainties that are intrinsic to scientific inferenci ng. However, within the science system these uncertainties m ay ra ise new, and possibly fundamental research questions (Beck, 1986). Similarly, the science system can deve lop reflexively in relation to problems arising in industrial contexts (Rosenberg, 1976). Often new opportunities to patent arise unexpectedly within the research process. In other (e.g., industrial) contexts, scientific progress can sometimes be consider ed as an unintended side-product, with the intended focus being on problem-solution. Rosenberg (1982) raised the question “How ex ogenous is science?” Ever since, the non- linear dynamics in the science/ technology interactions has take n the lead in the research program of science, technology, and innovation studies (STI)—as we have now began to call this field of expertise (Wouters et al ., 1999). W hereas the sciences are developing along historical lines, i nnovation reorganizes the syst ems on which it builds at the interfaces. This continuous reorganization unde r the pressure of co mpetitive innovations has been institutionalized in advan ced indus trial systems since th e scientific-tech nical revolution of the period 1870-1910 (Braverma n, 1974). Since then, further development of technologies takes place at the interfaces of the scien ces, the economy, and the useful arts (Noble, 1977). In a later part of this study, we w ill distinguish between science indicators in terms of scie ntific com munication, technology i ndicators in terms of patents that map technological inventi ons, and innovation indicators that m ay also map market introduction, for example, at the Internet. We propose to recombine these indicators reflexively in a program of innovation studies. 1.3 From theories to empirical evaluations Units of analysis versus units of operation The analytical models provide us with heuris tics for the em pirical research. However. the knowledge component of systems cannot dir ectly be observed as organized knowledge acts as a different operator to the observable instantiations that it changes (Giddens, 1984). The knowledge systems studied operate dyna mically. A representation provides us only with a picture of the footpr ints of previous communications. Scientometric and webometric indicators tr ace functions of co mmunication. Functions can be attributed to institutions. For example, publications and citations span networks of communication, but one can also use them fo r ranking institutions in term s of their productivity. Note that we wish to alter the research focus: one can rank scientists, that is knowledge carriers, but scient ific communications devel op at the network level. Networks tend to develop in different directio ns with different qualities. Over tim e, the network dynamics may redefine w hat has be en a significant con tribution and in which respect. The dimensions (functions) of the ne tworks can be considered as increasingly orthogonal. Thus, the networ ks group the communications instead of ranking them. The significance of a contributio n is not an inherent proper ty of a contribution, but a construct “in the eye of the beholder” (Latour, 1987; Leydesdorff & Amsterdamska, 7 1990). The windows for studying subjects as intangible as knowledge production and communication, have to be carefully reflected as the order of co mmunications is not “naturally” given. We are constructing second-order constructs about knowledge-based constructs. Some authors have proposed the consideration of “national systems of innovation” as the appropriate unit of analysis for innova tion studies (Lundvall, 1992; Nelson, 1993). The choice for a national perspective allows for a dir ect link to the possib ilities and limitations of policy making by national governments. Furthe rmore, it enables the researcher to use national statistics (Lundvall, 1988). However, from a reflexiv e angle, each communality or dimension can be considered as a constr uct that can be more or less codified. For example, the notion of a national identity may be changing from a European perspective. The construction of a regional identity, for instance, has resounded in some regions because of linguistic differences, bu t in others, such as in France, regional authorities have been shaped in order to accommodate to European policies and harmonization. In other words, the units of an alysis and the system s of reference can be analytically considered as c onstructs that then tend to sh ape the analysis. The windows that we use provide us with a metaphor that can easily turn into a bias or a metonym. What can be considered as hi ghly relevant from one perspect ive, may be contextual from another. The categories in which science, technology, and innovation studies reconstruct bodies of knowledge have to remain hypothese s! It is precisely as hypotheses that the concepts invite to proceed to the ope rationalization and m easurement. The time dimension In contrast to a historical build up, the evolutionary dynamic continuously operates in the present and with hindsight, that is upon the in stantiations of the system s under study as its basis. Thus the global dimension tends to invert the historical time axis in the analysis. Whereas growth-rates, for example, are usually expressed with reference to a previous moment in time and time series are standard ized with reference to a historical moment (e.g., “1990 = 100”), the evolutionary perspective is policy releva nt because the analyst can take the present as the system of reference . The present s tate contains the analytical reconstructions as representations of its past. Thus the evolutionary analysis provides information from which one can develo p options with a greater or lesser degree of success, without pr escribing future behaviour in any sense. Hitherto, statisticians have had an in clination to build on their resour ces using a historical perspective. Sociologists interest ed in history may then be able to use th ese materials as illustrations in support of their narratives. However, the focus on knowledge-in tensive developments requires us to take a reflexive turn towards the data gathering process, both in the quantitative and in the qualitative dom ain. The program of innovation studies is anti-positivistic, as one begins with expecta tions instead of the observable “facts.” The facts mean different things at different sides of an interface. 8 From an evolutionary perspect ive on cultural phenom ena such as science and technology, the analyst first specifies which ass umptions went into the data collection and wh ether these assumptions are still va lid when, at a later stage, one raises new questions from the evolving science and tec hnology policy agenda. For exam ple, when studying the development of journal structures in “bio technology,” one has several options. If one fixes the journal set ex ante , one observes the developm ent of “biotechnology” as conceptualized at the beginni ng of the data collection (e.g., in 1985). If one defines the journal set dynamically, one studies the cha nging m eaning of “bio technology” in relation to other journals. If one fixes the journal set ex post, one refers to the understanding at the later moment in time (e.g., in 2003). The analysis of the historical strengths a nd weaknesses of a research portfolio does not itself suggest that one should “pick the winne rs” (Irvine & Martin, 1984) in order to strengthen one’s case globally, that is at the system ’s level. The “winners” may have been yesterday’s winners and one may have other reasons to strengt hen the hitherto relatively weak groupings or clusters (Porter, 1990). Th e empirical analysis informs us about the contingencies that can be expected. Howeve r, as the dynamics are complex, unintended consequences and unforseen externalities can also be expected. The formative evaluation during the process provides us with signals th at can then be made the subject of more systematic analysis. Operationalization, data, and context A crucial step in proceeding from the form ulation of analytical hypotheses to the collection of em pirical data is implied by the concept of operati onalization. How can one move from the analysis to th e indication of the importance of the concepts in a social reality? How can a reflexive an alyst make a convincing argu m ent when the notion of a system of reference can always be deconstr ucted, and the time line can be inverted in terms of what the historical account means for the present? As systems that contain knowledge should not be considered as given or immediately available for observation, one has to specify them analytically befo re they can be indicated or measured. In the end, the quant itative analysis depends on the qualitative hypotheses. For example, one can raise the qu estion of whether “Mode 2” has prevailed in the production of scientific knowledge. What would count as a demonstration of this prevalence and what would count as a counte rargument? Can instances be specified in which one would also be able to observe pr ocesses of tran sition between the two modes? What should one measure in which instances and why? While the qualitative analyst reduces the co mplexity by taking a perspective, quantitative analysis allows fo r the question of th e extent to which a perspectiv e highlights a relevant dimension. How much “Mode 2” is in the de velopm ent of biotechnology in Germany as compared with the developmen t of biotechnology in the Un ited States? A policy analyst is always able to indicate contingency, si milarities and differences, continuities and changes, but the quantitative analys is requires that these categories are specified as ex ante hypotheses in order that the expectations can be updated by the research efforts. The 9 empirical research should enable us to specify the pe rcentage of the vari ation that can be explained using one theoretical model or another. Whether “Mode 2” is “old wine in new bot tles” (Weingart, 1997) or new wine in old bottles depends on the definitions of the bot tles and the wines, and the pro cesses of change that are analytically outlined in the research design. The definition s with reference to a knowledge-intensive system are know ledge-intensive th emselves (Nowotny et al ., 2001). The observations and indicators are al so knowledge-intensive, as one can no longer assume that the data is readily at hand. The overwhelm ing availability of information in a knowledge-based society makes it necessary to reflect on the selection of data from a theoretical perspective. Established indicators were time-stamped in a previous period and historical evidence can retrospectively be recognized as anecdotic al. Systematic data collection, however, requires standards. The matching between the analytically relevant questions and the institutionalized routines asks for an info rmed trade-off between considera tions of a potentially very different nature. How does one define a baseline? How does one normalize? What is/are the relevant system(s ) of reference ? Scientometric indicators cannot simply be “applied” in another cont ext without generating terrible confusion. Scientometrics is a research effort in its own right, since the indicators have to be reflected. For example, the debate of “the decline of British science” (Irvine et al ., 1985) as measured in terms of publication perform a nce data was paradoxica lly possible because “British science” had been relatively stable (Braun et al ., 1991; Leydesdorff, 1991; Martin, 1991 and 1994). Thus, methodological decisi ons as to whether the analysts used an ex ante fixed journal set in order to make co mparisons along the tim e axis possible or followed the development of the dynamics of the jou rnal sets included in the Science Citation Index had an impact on whether one m easured decline or not. The further decision to attri bute each publicatio n with a British address to the UK with a full point or only pro ratio of the number of corporate addresse s in internationally coauthored publications (so-cal led “fractional” versus “int eger” counting) includes an effect of internationalizati on on the performance measuremen t th at can be expecte d to differ from nation to nation. Integer count ing, however, leads to problems in the normalization since the sum-total does no long er add to hundred percent (Leydesdorff, 1988; Anderson et al ., 1988; Braun et al ., 1989). These methodological problems reflect decisions that have to be taken on the basis of arguments. The theoretical grounds can be made relevant for the scientometric enterprise if they can be formulated as hypotheses that are operationalized reflexively before one is able to draw conclusions. The selection of data is necessarily specif ic and this specificity has to be reasoned. 10 1.4 Statistical analysis, models, and simulations Multidimensionality and the reduction of uncertainty In principle, data inform the hypotheses, but not by them selves and not necessarily in the positive. Data may also confuse us; particularly when they are so a bundantly available as nowadays. A variety of repres entations is always possible and this problem is further aggravated when databases are no longer subs tantively codified a nd dedicated, but when algorithmic search engines and m eta-crawlers be come widely available. The data provide us first with variation and th erefore uncertainty, and the pers pectives on the data m ay also be at variance. Specification of a reflexive perspective redu ces the uncertainty. This can also be expressed formally by using prob ability theo ry. First, a probability distribution of a variable can be hypothesized. Then, each furthe r specification can be considered as an additional condition to this pr obability distribution. The third law of probability calculus specifies that the likelihood of two probabiliti es (A and B) together is equal to the likelihood of A given B times the likeli hood of B. Or in formal language: p (A and B) = p (B) · p (A|B) Since all probabilities are sm aller or equal to on e, the uncertainty in the dis tribution A is reduced by our knowledge of the distribution B , unless the two distributions are completely independent. Whether the distributio ns can be considered as independent or not can be tested using signi ficance testing (e.g., chi-square ). If there is significant dependency, one is allowed to co nclude to a reduction of the uncertainty in the prediction. Therefore, multidimensionality not only enr iches the com plexity of the problem. It als o provides a way to reduce complexity. The specif ication of conditionality (that is, “contexts”) reduces the uncerta inty in the system under study (the “text”), unless the context was irrelevant. The dependence or independence of two variab les can be detected by analyzing their co- variation. If the two variables represent two different systems (or subsystems), these systems determine each other through th is window of co-variation or “mutual information,” but otherwise they only cond ition one another. Thus, the language of the quantitative analyst replaces expressions like “enabling and constraining” (Gid dens, 1984) with concepts of determination, reduc tion of uncertainty, and condition ality. The “mutual information” provides th e systems with windows upon each oth er. A co-variation when repeated over time may develop in to a co-evolution between two system s. History, co-evolution, and th e emergence of new systemness In many cases, one can build on existing defi nitions of system s—li ke “the research system”or “the patent system”—but in th e case of knowledge-bas ed systems one m ay also be interested in “emerging system ne ss”. Emergence can only be analyzed by 11 observing the interaction among systems over tim e. From this perspective one can analyze the evolution of each system along a trajectory. However, one can also focus on the interaction and potential co-evolution betwee n systems? Did the co-evolutions lead to a new system or did the (intended? ques tioned?) outcome fail to be realized? If there is co-variation over time and th en also co-evolution, one can expect the emergence of a degree of systemic developm ent. However, the question for the evaluation remains whether at a certain mo ment in time (e.g., today), system ness is prevailing over historical variation or not. These two dimensions —historical varia tion and systemness in the present—can be cons idered as analytically independent. The subsystems develop historical variati on along potentially differe nt trajectories, but the next-order system selects by weighing among the trajectories so that it can maintain its system’s order. 2 When a system evolves over time , one can ask how the state of the system at time t depends on the state of the sys tem at a previous moment ( t – 1). This relationship can also be turned around. Then, one asks how the state of the system at time t determines the state of the system in the future one time step ahead (that is, at t + 1). The Markov property states that the best prediction of the next stag e of a system is its current state (Yablonsky, 1986). “Markov systems” have no long-term m emory about historical orders at lower levels because the system is able to reorganize in the present using the relative weights of the various sub systems that it recom bines. In science and technology po licy the appearance of a next -order system is often a question more interesting than the expectation of stable development along a trajectory (Allen, 1994). For example, in the case of Europeanization one can raise the question of whether a European dimension of the publica tion system in a specific field can be discerned. One can test this hypothesis by me asuring the publication output for each of the individual nations histor ically and then make a pr ediction on the basis of the respective time series that can be co mpared with a prediction based on th e assumption of emerging systemness. 2 Systems with non-linear interactions exhibit addi tionally the capacity to develop different scenarios which may branch in time like a tree . This has also been called path-dependent development. Once, a trajectory is chosen, the s ystem is bounded to a path. One could say that a system once codified by this “lock-in” becomes a potential candidate for a next-order selectio n. 12 Country 1990 .... 2001 2002 2003 Country A a 1990 a 2001 a 2002 ? Country B b 2002 ? ... ? Country N ? Figure 1.1 Time series of rows versus syste mness over the c olumns For example (using Figure 1.1), one can make a historical predic tion of the publication performance of Country A in the year 2003 on th e basis of the values of the indicator a 1990 to a 2001 , and similarly for country B, etc. Th e alternative prediction would be that systemness has grown among these (European) na tions and that the European dimension would prevail. In that case, the situation in 2002 would provide us with a snapshot along the column dimension of Figure 1.1 of how far this system has developed. The best prediction of the situation in 2003 would then be based on the Ma rkov assumption that the current state (2002) would be maintained and reproduced as a dist ribution in the next year. As soon as one is able to measure the publication volume for the year 2003, the two predictions can be compared. One cannot reject a hypothesis on the basi s of a single measurement point, but the principle of testing two hypothe ses against each other may be clear. Th e two predictions above are analytically independent since ba sed on the rows and columns of the matrix, respectively. Therefore, the predicted values are different and they can be com pared with the measurement results. One can also hypothe size that the observed values are to a certain degree (e.g., 30%) predic ted by the one hypothesis and to a complementary degree (70%) by the other. Thus, one is able to specify the percentage of the va riation that can be explained by using one assumption or another. Using these methods, a European publication system could, for example, not yet been discerned in terms of the publ ication data included in the Science Citation Index (Leydesdorff, 2000a). In another study, Leyde sdorff & Oomes (1999) were able to show how the emerging European Monetary System (EMS) affected national sys tems in the monetary and economic domains, re spectively, during the period 1985-1995. Let us now proceed by generalizing the analyt ical independ ence of the prediction based on the rows versus the columns of a m atrix to all scientometric tables and spreadsheets that are thus designed. The two dimensions of an (asymmetrical) matrix refer to different dimensions of the system under study or—in other words—different system s of reference. For example, the scientometrician can count word-occurrences in documents. The documents are then considered as the cases and the words as the variables. The words provide us with an indicator of th e intellectual organization of the documents, whereas the documents can also be grouped in term s of their instit utional (e.g., national) 13 addresses. Thus, a matrix of words versus doc um ents provides us with information in the communicative dimension of the intellectua l exchange and the dimension of the institutional organization. These two dimens ions (subsystems) are coupled by the research design when specifying a hypothesis that can be tested. 3 words documents words documents words documents words documents y ear 0 y ear -1 y ear -2 y ear -3 Figure 1.2 Matrices of words versus doc uments in a time series When one constructs such a matr ix of words versus documents for each year in a series of years, one can place these matrices behind each other and then one would obtain a cube of information. Along the time axis one is able to ask whether the words (indicating the intellectual organization ) have grown into a system or whet her the institutes (represented as aggregates of documents) ha ve rearranged the ir relations. In each of these dimensions one can additionally ask the question to whic h extent systemness has become prevailing over historical variation (or not ). Finally, one can also raise questions of whether the co- variation in the matrices for each year ha s become increasing ly systemic over time. In Leydesdorff & Heimeriks (2001) this me thodology was applied on publication data in the field of biotechnology. The conclusion wa s that the American science system had been more self-organizing than the European system in the intellectual dim ensions as measured in terms of coherent word usages, perhaps because of the prevalence of national tendencies in word usage among the European nations. The European system, however, could be shown to have effects in the in stitutional dimension (L ewison & Cunningham, 1991). 14 3 In the case of a symmetrical co-o ccurrence matrix part of the information has been discarded by multiplying the original matrix with its transposed. Models and simulations The focus of our discussion has hitherto been on how self-organiza tion, co-evolution, and systemness can be analyzed by m easuring sets of variables at different moments in time. One level deeper, one can ask which dynamics can be expected to l ead to the tem poral and structural changes that one observes. Modeling and simulation focus on developing explanations for the observable patterns. Modeling efforts and simulation stud ies can be retrieved from the lite rature in science and technology studies, but these studies have been scarce (Goffman, 1966; Nowakowska, 1984; Kochen, 1983; Ebeling & Scharnhor st, 1986; Wagner-Döbler & Berg, 1993). Perhaps, models of stochastic processes have been an exception to this rule (e.g., Price 1976; Egghe & Rousseau, 1990). In evolutionary economics, however, models have been extensively used as tools fo r explaining technological develo pments (cf. Fisher & Pry, 1971; Silverberg, 1984; Arthur, 1989). In the 1990s a new type of m odel (so-called “agent-based ” m odels) gave modeling and simulation a further impetus in science and technology studies (Axt ell & Epstein, 1996). Because these models start from rules for i ndividual behavior, they are suitab le to model social processes that are based on assumpti ons about agency (Gilbert & Troitzsch, 1999). Processes in the sciences lik e citation and pub lication behavior, but also the diffusion of technologies and the appearance of the web, have since then been modeled from this perspective (Bruckner et al., 1990; Gilber t, 1997; Leydesdorff, 2001c; Scharnhorst, 1998 and 2001; Boudourides & Antypas, 2002) The modeling approach opens a different pers p ective on indicator research. On the basis of the conceptual framework of a m odel, new measurement requirem ents can be specified. Whereas the scientometric m easuremen t has focused on the historical cases that actually occurred, a model first specifies a real m of possible events. The actual events can then be considered as instantiatio ns. The instantiations are determ ined by parameter values. The parameter values have to be estimated on the basis of observations. Let us elaborate using an example. The growth of scientif ic disciplin es can be modeled as a process of competition. The growth curve of a single field can be considered as an outcome of the interplay between different fields (Bruckner et al ., 1990). In terms of population dynamics, the scientific fields are then defined as the units of evolution. These units, however, are not self-standing agents, bu t collectives of individual agencies. The higher-level units can be considered as pattern s of coherent behaviour of scientists who have grouped together at a lower level in “invisible colleges”. The interactions between different fields are caused by the underlying choices of the scientists working on certain topics and not on others. Scientists may move between fields by changing their research focus. By assu m ing the fields as the units of evolution in science, this movem ent can then be modelle d as a competition between the fields on the available scientists as human resources (Gilbert 1997). The scientists, however, are 15 steered both from the control level of the fi eld and, for example, in terms of available funding. Thus, the model develops at two levels at the same tim e and with feedback loops between these levels. In such a model, various processes at the mi cro level can be distinguished. For example, the educational process of scientists can be considered as an entry process to a field, the mobility between fields as an exchange process, and the career ending of scientific activities in a field as an exit proce ss. Thes e various processes have to be weighted by using parameters. The estimation of these parameters, however, can only be based on measurements. Confronted with the task to validate parame ters, one becom es aware that this information is not gathered by scientometri c indicator research in a way that can easily be transformed into the parameter values for sim ulations. The indicators tend to f ocus on specific processes, but not on the inte raction terms be tween different processes. For example, one can easily find data about the edu cation of sc ientists in d ifferent disciplin es, but data about the number of “newcome rs” in scientific specialtie s are far more difficult to retrieve. Migration pattern of scientists betw een specialties are seldom analysed and then not easily connected to the growth of sp ecialties (e.g., Mullins, 1972; Mulkay, 1977). In the case of scientom etric indicators, longitu dinal data are often difficult to obtain since the focus in scientometrics has been on wh at can be called “comparativ e statics.” How has the situation chang ed since a previous mo ment in tim e in terms of the observable data? Indicators then provide snapshots for different moments in time. The dynamic analysis is different from “comparative static s” since the latter a pproach does not aspire to analyze the processes underlying the observable changes. In summary, modelling efforts create a demand for new or ref ined measurement instruments. The estimation of parameters is oriented to the underl ying processes at a micro-level, while current indic ators tend to focus on the observable phenomena at a macro-level. Note that simulation models can also be used to produc e “virtual” indicators. Hypotheses about underlying processes of change in knowledge production can be turned into specific subroutines of a dynamic mode l. Simulations produce quantitative output that can be compared with what happened. The comparison between th e statistical and the “virtual” indicators can be further analysed, for example, for explaining what caused the difference. By comparing “virtual” indicato rs with empirical m easurements hypotheses behind a model can sometim es be tested. How can one compare the historically observed values with the evolutionary expected ones? In this study, we do not elaborate fu rther on modelling, but we focus on the measurement. However, we wish to emphasize that modelling and simulation can be understood as a part of indicator research as both research programs are interested in the quantification of the descrip tion and then also the explanatio n. The theoretical challenge consists of the creation of a link between the quantities of th e models and the quantities in the empirical observations. The se lections bo th in terms of relevant (empirical) data and in terms of assum ptions in the simulation mode ls have to be guided theoretically if we wish to relate these two domains as research programs in innovation studies. 16 1.5 Conclusion Quantitative indicator resear ch develops on the edge of evolutionary m odeling and historical observations. The hi storical observations of comm unication can be expected to contain uncertainty because the very con cept of communication implies an exchange. Thus, communication is distributed by its very nature. From this perspective, qualitative theorizing contributes by pr oviding hypotheses, i.e., unc ertain expectations. The challenge is to relate the specified expectations to obs ervable data. The hypotheses can first be used to distinguish structural uncer tainty from random fluctuations, error, and noise. The simulation model adds the interactio n of different proces ses to the hypotheses and it allows for experimentation with possi ble scenarios of syst em s development. The space of possible future developments can only be accessed algorith mically. Once a set of variables has been de fined, the algorithm describes the temporal changes of these variables by looking at the fluxes (dx/dt). To gain understanding, however, the analyst uses geometrical metaphors based on times series of variables or the analysis of the multi-variate com plexity as instantiations in the present. One would overstress one’s linguistic capacities by describi ng changes in the values and the meaning of variables in the same pass. Once the parameters are ch osen f or the representation, one is bound by a set of geometrical constraints on the repres entation. The qualitativ e appreciation in the narrative can therefore be considered as generating a metaphor or window on the complexities under study. If one tries to describe both change in the meaning of the variables and the value of the variables using a single design, the comprehension tends to become vague and confused. Luhmann, for example, invoked in such in stances the metaphor of a “paradox:” The algorithmic system can be expected to be m ore complex than the geometrical metaphor (“picture”) stabilized in a di scourse (Hesse, 1988). The disc ourse contains a perspective that can only be changed discursively. An addition al change in the meaning of the variables can then be formulated as a dynamic problem . The algorithmic formulation increases the com plexity to the extent that different perspectives can be entertained as competing for the explanation. The results of this process can discursively be appreciated as the update of the hypothesis. The ex po st picture can be different from the ex ante one. The representation is then “translated” (Callon et al ., 1986). However, does this imply that the repres ented system is also changed? Are we able to disti nguish dynamically the bias caused by our representations from the changes in the represented system s? In our opinion, quantification of these concurre nt processes of change is th e major research effort of quantitative studies in science, technology, and innovation studies following the reflexive turn in science, techno logy, and innovation studies. The set of variables or, in other words the de finition of the system, fixes an axis for the comprehension, but the codification of this system also generates a “blind spot.” 17 Empirically, the problem of a system with potentially chan ging taxonom ies re-appears in the choice of the units of analysis. However, can one change the unit of analysis “on the fly”? The algorithmic approach enabled us abov e to change to the specification of a unit of operation as different from a unit of an alysis. The focus on innovation studies makes this reformulation unavoidable because innov ation can only be defined as a unit of operation at an interface. In our opinion, this change of perspective to an algorithm ic approach—entailing the appreciation of narratives as heuristics— enables us to so lve some of the outstanding problems in the scientometrics program . For ex ample, the need for a historical baseline was signaled early in the scientometric en terprise (Studer & Chubin, 1980). Narin (1976) proposed to work with an ex ante fixed journal set as an analyt ical tool in order to make comparisons possible among time series data. As noted, this decisi on contributed to artifacts in scientometric representations of “t he decline of science in the U.K.” during the 1980s. Collins (1985) raised the question ab out the appropriate unit of analysis for scien ce policy evaluation. He argued against an institutional d e lineation of units of analysis in ord er to compare “like with like” (Mar tin & Irvine, 1983). Scientif ic developm ents, however, cannot be equated with the development of in stitutional units nor with fixed journal sets. The focus on flows of communication ma kes it necessary f irst to specify what the hypothesized (since not so easily observa ble) system of communications is communicating when it operates. The specifica tion of the unit of operati on extends the analysis with a specificati on in the relevant time dimens ion. Only after the hypothetical specification of the “what” of the communicatio n, can one a ddress the question of “how” this communication can be envisaged. The specification of the “how” of the operation then induces the specification of an indicator. For example, on the basis of the as sumption th at scientif ic specialti es are developed in terms of knowledge contributi ons, one can ask how knowledge is contributed. Scientific articles then become a prim e candidate for measurement. How are scientific articles related? Co-words among titles and citations can then be understood as indicators of the hypothesized exchange processes. The aggrega tion of citations (or other scientometric indicators) enables us to map the sciences under study at certain mom ents in time. This methodology can be contrasted with the use of citations (and other indicators) for the reconstruction of a hypothesized development over time. In other words, the theoreti cal specification constructs the (hypothetical) systems under study. For example, using a journal set provi des us with a focu s on the scientific publication system. Using patent data prov ides us with a focus on technological inventions. These two systems are differently codified and therefore can be expected to exhibit different dynamics. The st udy of scien tific citations in patent literature, and vice versa, provides us with a focus on the interf ace between these two literatures, th at is, patents and publications. However, there is no necessary relation between the two types of data. On the contrary, several studies (Narin & Noma, 1985; Na rin & Olivastro, 1992; Blauwhof, 1995; Schmoch, 1997; Meyer, 200 0a,b; cf. Grupp & Schmoch, 1999) have noted the narrowness of the window of co mmunication between these two systems. 18 Others have focused on institu tional rela tions between addresses in pa tent literature and scientific publications, but also, in this dimension, differentiation may prevail above integration (e.g., Riba-Vilanova & Leyde sdorff, 2000). The local integration communicates among systems that first have to be specified analyti cally. The expectation is that the communications can loc ally be in tegrated because they are differentiated in other dimensions. The evolutionary perspectiv e of innovation studies makes it furtherm ore necessary to delineate the systems of reference from th e perspective of hindsight. The hindsight approach generates a relation with future-ori ented policy perspectives as one informs the reader with reference to the present state of the systems under study. For exam ple, what we understand as “biotechnol ogy” nowadays is something completely different from what governments wanted to stimulate in the 1980s (Nederhof, 1988). Analogously, what industries subsume under “biotechnology” as a ca tegory at present, is different from the definition of “biotechnology” by research councils. A m odern society has many facets and is therefore differentiated in terms of its coordination m echanisms, codifications, and media of communication. 19 Chapter T w o C AN THE N EW M ODE OF K NOWLEDGE P RODUCTION BE M E ASURED ? We have argued hat a fundame ntal reformulation of the problems of Science, Technology, and Innovation Policies becam e urgent during the 1990s because the following developments reinforced each other: (1) The emergence, spread, and convergen ce of technological and communications paradigms such as the computer, mobile te lephony, and the Internet ; interaction itself has become more extensive am ong organizations, multi-layered, and therefor e relatively more important than the elaboration of perspectives within the walls of one’s own institution based on routines and tacit knowledge; (2) The interconnection between the laborat ory of knowledge-produc tion and users of research—at various levels—exemplified by the rapid growth of industry-university centers in which firms and academic researchers join tly set priorities; technolo gy transfer agencies within both universities a nd firms that negotiate with each other and move technologies in both directions; (3) The consequent transition from verti cal to lateral a nd multi-media modes of coordination, represented by the emergence of networks, on the one hand, and the pressure to shrink bureaucratic layers, on the other. The authors of the “Mode 2” thesis (Gibbons et al ., 1994) argued that this new configuration has led to a dedifferentiation of the relati ons between science, technology, and society. Internal codification mechanism s (like “truth-finding”) we re discarded as an “objectivity trap” (Nowotny et al. , 2001, at pp. 115 ff.). The epistemological core of science was declared not to be on ly uncerta in, but therefore (!) completely em pty. From this perspective, all scien tific and tec hnical communication boils down to communication that can be equated and compared with ot her communication from the perspective of science, technology, and innovation policies. In our opinion, the study of communication can be guided by available theories of communication. Two theories are then partic ularly important: Luhm ann’s sociological theory of communication (Luhmann, 1984 and 1990; Leydesdorff, 2001a) and the mathematical theory of communication (Shannon, 1948; Theil, 1972; Leydesdorff, 1995). The crucial point becom es how to relate these two theories with d ifferent (epistemological) statuses, so that the quantitative measurem ent enables us to update and inform the hypotheses based on substantive theorizing. 20 2.1 The reflection of boundary-spanning mechanisms University-Industry-Government relations can be considered as a boundary spanning mechanism in the knowledge infrastructure of societies. The operation of boundary spanning mechanisms indicates that the transa ctio n costs can be balanced by the expected surplus value of the collaboration. Intern ational coauthorship relatio ns, for example, provide another boundary spanning mech anism, but across national boundaries. Publications can also be rela ted intellect ually, for example, by being published in the same or similar journals. The Science Citation Index 2000 provided us with information about 3745 journals in which articles are published, usually including institutio nal and national identifiers alongside author names. This data enables us to study coauthorship relations that cross institutional boundaries. The 1,432,401 institutional and corp orate addresses contained in this data set were attributed with th e categories “univers ity,” “industry,” and “government” by using an automated routine. We were thus able to classify 86.6% of the addresses in these three cate gories. The identified affilia tions refer to 93.3% of the 778,446 records of unique documents in the database. 4 Once categorized, this data can be analyzed using the statistics of mutual inform ation. The mutual inform ation (or transmission) differs from co-variation, co-occurrence measures or correlation analysis because this m easure is also defined in three dimensions. 5 When the three subsystems (uni versity, industry, governm ent) are completely uncoupled, the mutual inform ation vanishes (T U-I-G = 0). When the three dynamics are mainly coupled by sharing a com munality in the variation (e.g., in the case of a hierarchical (e.g., ét atist) regime or perhaps in cor poratist arrangem ents), the value of this transmission is positive (Figure 2.1). Howe ver, when the three domains are liberally coupled through uncoordinated bi-lateral rela tions, this indicator can also become negative (Figure 2.2). Thus, the indicator provides us with a measure for the state of a Triple Helix system whenever the relevant relations can be counted. 4 These documents were written by 3,060,436 authors so that on average each document contains two addresses and four co-author names. 5 The transmission in three dimensions (x, y, z) can be defined as follows (Abramson, 196 3, at p. 129): T(xy z) = Σ xyz P(xyz) log {[P(xy).P(xz).P(yz)] / [P(x).P(y).P(z).P(xyz)]} Or in terms of the Shannon notation: T(xyz) = H(x) + H(y) + H(z) – H(xy ) – H(yz) – Hxz) + H(xyz) In the first formulation, P(x) stands for the pr obability of an event x and P(xy) for the probability that x and y occur together, etc. These probab ilities can be measured by counting frequencies of (cooccurences) of events as will be s hown in the empirical examples below. 21 Figure 2.1 Three subsystems with a center of coordination Figure 2.2 Three subsystems without center of integration i k j jk ik jj i k ijk j jk Conceptually, the potential ge neration of a negative entropy co rresponds with the idea of complexity that is contained or “self-organi zed” in a network of relations that lacks central coordination. The system then propels itself in an evolutionary mode (Figure 2.3). The reduction of the uncer tainty by this nega tive transmi ssion is a result of the network structure of bi-lateral relations. Note that the mutual inform ation in two dimensions contributes negatively to the uncertainty th at prevails, while the three-dimensional overlap increases the local entropy. ij ik j k i k j time Figure 2.3 Three subsystems with hypercyclic in tegration in a globalized dimension 22 The network structure operates globally by c onstraining and enabling local substructures. However, the overall structure cannot be comp letely perceived from any of the positions in the network since there is no center of c oordination. As this st ructu re operates in a virtual dimension, it remains la tent and cannot fully be obser ved locally. However, it can be hypothesized and then also measured. The theoretical specification of this virtual dimension reflects the evolving system. 2.2 Methods and materials As noted, the CD-Rom version of the SCI 2000 contains 1,432,401 corporate addresses. These addresses point to 725,354 records contai ned in this database on a total of 778,446. Only 53,092 records (3.7%) contain no address in formation. Our current research focuses on the international coau thorship relations in this data, but we will rep ort on that project elsewhere (Wagner & Leydesdorff, 2003). We focus on University-Industry-Government relations in this data set. The addresses were organized in terms of their attribution to university-industry- government relations. This was don e by a routine that first attri buted a univers ity label to addresses that contained the abbreviations “ UNIV” or “COLL” in the address field. The remaining addresses were thereafter subsequently labeled as “industrial” if they contained one of the following identifiers “CORP”, “INC ”, “LTD”, “SA” or “AG”. Thereafter, the file was scanned for the identif iers of pub lic research institut ions using “NATL”, “NACL”, “NAZL”, “GOVT”, “MINIST”, “AC AD”, “INS T”, “NIH”, “HOSP”, “HOP “, “EUROPEAN”, “US”, “CNRS”, “CERN”, “INR A”, and “BUNDES” as identifiers. The order is so that hits are removed when re trieved using these routin es. For example, addresses at the academy (“ACAD”) cannot be c onfused with a university address, s ince the latter addresses h ave th en already been removed. This relatively simple procedure enabled us to identify 1,239,848, th at is 86.6% of the total number of address records, in terms of their origin as “unive rsity,” “industry,” or “government.” The distribution are as follows: Number of addresses Percentage “University” 878,427 61.3 “Industry” 46,952 3.3 “Government” 314,469 22.0 – (not identified) 192,553 13.4 Total 1,432,401 100 Table 2.1 Addresses indicating university, industr y or government affiliations in the Science Citaiton Index 2000 23 These sets can now be combined with c ountry names. For exam ple, of the 251,458 records containing an address in the U.S.A ., 92.5 % (232,571) can be identified in terms of their origin in at least one of the three helices. More than 200,000 of these records (> 80%) contain at least one univers ity address (Godin & Gingras, 2000). number % ti T(uig) in mbits UI UG IG UIG Univers Industry Govern all 676511 93.3 -77.0 16270 108919 4359 5201 543123 41242 232096 USA 232571 92.5 -74.4 7200 37834 1782 2666 200149 18154 66416 EU 257376 93.0 -50.1 4455 52112 1485 2028 206747 11192 101545 JAPAN 6771 5 97.9 -92.1 4147 12492 954 1311 56534 9732 21664 UK 68404 93.1 -63 .1 1719 13098 394 690 5482 3 3970 26202 GERMANY 6101 7 94.7 -43.4 1028 14003 407 664 51283 2799 23701 FRANCE 41112 90.3 -52.1 439 11593 452 530 2 6133 1928 26595 SCAND 30939 95.8 -3 1.6 490 8477 162 371 26542 1263 13005 ITALY 28958 89.9 -29.4 362 7133 87 262 25633 905 10526 NETHERL 18357 95.3 -25.4 372 4482 106 259 16379 86 3 6593 RUSSIA 22767 98.6 -24.2 76 6315 162 138 11507 478 17611 INDIA 10916 89.2 -7 8.1 97 18 13 61 55 6099 407 6492 BRAZIL 9120 91.0 -22.4 137 1727 32 52 7968 267 2885 internat. coauthored 120086 98.9 -21.9 4550 47054 1349 2 545 1 07569 9422 61138 Table 2.2 University-Industry-Government a ddresses and relations in the Science Citation Index 2000 What does this table teach us? First, it confir ms that industry is not prominently present among the addresses of papers in the Science Citation Index. At the level of the database, industry is represented in appr . 6% of the papers. For the U. S.A. this figure is appr. 8%, and it is larger than 14% f or Japan. However, this percen tage is much lower for EU countries (4.3%). For example, this ratio is only 3.1% for Italy. The table shows that in France the numb er of papers with addresses of public research institutes is larger than those with university addresses. This contribu tes to the triple-h elix type of integration of the national system. T U-I-G is more negative for France than for Germany. The table shows that countries diffe r widely in term s of how the institutional arrangements operate among the main carriers of the knowledge infrastructure. The m ost negative value for the mutual information in th ree dimensions is found for Japan; the least negative for the internationally coauthored pa pers that are distingui shed as a separate category in the last line of Table 2.2. These results raise interesting questions that we will elaborate upon in another context. The main point here is that th e data is useful in raisin g sophisticated questions like whether, and how, to measure and evaluate tr iple helix configurations. This m easurement 24 can also be combined with specif ic journal sets indicating discipli nes and specialties. The mutual information provides us with an in tere sting indicator for the m easurement of configurations within these sets and for the relations among them. 2.3 Webometric data Since the mid-nineties, a growing body of litera ture has em erged about measuring science and technology activities on the Web using info rmetric, bibliom etric, and scientometric methods. In 1997 the name “webometrics” wa s introduced (Almind & Ingwersen, 1997). The journal Cybermetrics was also launched in 1997. Since then, the informetric community has taken up the investigation of the new electronic media, including the Internet (Larson, 1996; Rousseau, 1997; Ingwersen, 1998; Egghe, 2000; Thomas and Willet, 2000; Bar-Ilian, 2001; Bjöneborn and Ingwersen, 2001; Cronin, 2001; Thelwall, 2001). If scholarly and scientific re search and communication are more and more shaped by the Internet, analysis focussing on printed media may miss an im portant amount of research. In 1999, a first feasibility study granted by the European Commission stated “the opportunities for using informetric methods [o n the Web] are not yet well elaborated” (Boudourides, Sigrist et al ., 1999). Meanwhile, several article s have appeared which tried to define the main topics of webometric approaches. Qu estions are raised such as: methods for adequate data collection and th e use of search engines for that purpose (Snyder and Rosenbaum 1999); the problem of transferring terms lik e “citation” to the world of the Web (“sitations”; Rousseau, 1997) ; and the definition of impact factors for electronic journals (Ingwersen, 1998). Let us provide an example of the possible us e of Internet data by making a m easurement effort comparable to the scientometric one ou tlined in the previous section. In a study of university-industry-gov ernment relations, Leydesdorff & Curran (2000) previously measured the occurrences and co-occurrences of the words “university,” “industry”, and “government” on the Internet using the A ltaVista Advanced Search Engine. The advanced options of this search engine a llow for the searching of various countries and general top-level domains (e.g., .com, .edu, etc. ) in com bination with specific time-frames for the publication dates of the website s, as well as Boolean operators. Our previous study was replicated for different time periods, using various search engines by Bar-Ilan (2001). The author s howed, among other things, how sensitive the Internet is for the measurement at different times (Rou sseau, 1999). Here we will use the data only for the search terms “university”, “industr y”, and “governm ent”, during the period 1993- 2000. All measurements were perform ed on 13 November 2001, using the AltaVista Advanced Search Engine. The year 1993 was chos en as the first year of the time-series because web-based browsers based on hyperlinks were introduced at th at time (Abbate, 1999). 25 0 1000000 2000000 3000000 4000000 5000000 6000000 7000000 1 993 1 99 4 1 99 5 1 99 6 1 99 7 1 99 8 1 9 99 2 0 00 Uni v ers i t y In d u s t r y Gov er nment U-I U-G I-G U-I-G Figure 2.4 Results of searches using the AltaVista Advanced Search Engine The data is organized in a three dimensi onal array, usin g the three search term s as independent dimensions, for each year as follows: U-I-G industry univers ity U- G I-G gove r nmen t Figure 2.5 A representation of university -industry-government relations in a three dimensional array The values for T(uig) are always negative in the case of these Interne t data, but the cu rve further decreases linearly since 1995 (Figure 2.6). It has prev iously been noted that the Internet experienced commerc ialization from 1995 (Abbate, 1999) and that the behavior of curves changed dramatically from that year onwards (Leydesdorff, 2000b). This and/or the rapid growth obviously leads to a further differentiation of the sets containing the three keywords as retrieved by the AltaVista search engine. The decrease is remarkably steady (r 2 = 0.98). 26 R 2 = 0.98 -0.30 -0.25 -0.20 -0.15 -0.10 -0.05 0.00 1993 1994 1995 1996 1997 1998 1999 2000 T(u-i-g) Figure 2.6 Mutual information in three dimensions (“university,” “industry,” “government”) as measured using the AltaVista Advan ced Search Engine (at Nov. 13, 2001). 2.4 Testing for Systemness What does the above effect indicate in relation to the original data as exhibited in Figure 2.4 above? Does it really indicat e the self-organiza tion of a virtual dimension in the overlay of relations generated by the co-occurre nces of two words in bi-lateral relations? Is this an indication of increasing se lf-organization of an overlay system? For testing longitudinal data on whether the combined time series exhibit system ness in the data or not, a test was disc ussed in Chapter One. This test is bas ed on evaluation of the Markov hypothesis for the collection of data versus the individual time series. The application of the test on this data p rovided the following results: prediction of the value in 2000 7 categories (U, I, G, UI, UG, IG, UIG) four categories (UI, UG, IG, UIG) three categories (UI, UG, IG) on the basis of the univariate time series (1993-1999) 1.18 9.81 10.67 on the basis of the previous year (1999) (Markov property) 8.84 1.87 1.65 hypothesis of systemness - 7.76 (rejected) 7.44 9.02 Table 2.3 Testing the hypothesis of systemness in the Triple Heli x overlay of Un iversity-Industry- Government Relations. (All values ar e pr ovided in millibits of information .) When reading this table, one should keep in mind that an observation does not generate any probabilistic entropy when compared wi th a perfect predicti on. The value of the 27 indicator, therefore, correlates negatively w ith the quality of the prediction. The results then show that the prediction of the 2000 da ta, on the basis of the same data for the previous year, is inferior to the prediction on the basis of the time series of the various categories in the case of considering the whole system of seven categories. Thus the hypothesis that these seven categories develop as a system is rejected. 6 If the analysis is limited to the three bi-later al relations (right column of Table 2.3), the hypothesis of systemness in this data is st rongly corroborated. However, this prediction, is devalued by including the trilateral relations (middle column). So the conclusion is that this system of representations has been deve loping as a set of bila teral relations tha t contains a negative expected information valu e and, in this sense, self-organizes the complexity in the data using a vi rtual overlay of mutual relatio ns. 2.5 Conclusion In the first part of this chapter, we showed how the mutual inform ation of Triple Helix relations varies among world regions and countri es. Data and statisti cs were provided at the level of the comprehensive database, but al so more specifically for subsets indicating various countries (e.g., the USA, the UK, Russi a, Japan) and regional blocks (e.g., the EU). The results rais e interesting questions. For exam ple, one can wonder why Germany deviates from other countries in its res earch portfolio as exhibited in Table 2.2. Analogously, one can analyze in ternational co-authorship re lations as another boundary spanning mechanism, namely among nation st ates. Among other things, the subset of internationally co-authored papers was compar ed above with the datasets for individual countries. It could be shown that the internationally coau thored papers are far more homogenous in their pattern of cross-sectoral collaboration than any of the national sets (Wagner, 2002). These results bring us to a third set of questio ns—still to be investig ated at this mom ent in time—namely, the relationship between various boundary spanning m echanisms that can be analyzed within the data, for example: 1. international co-authorship data, 2. co-authorships across university-industry-gov ernment boundaries, and 3. co-publication in the same journal indi cating an intellectual boundary spanning mechanism. Note that boundary-spanning mechanism s ope rate by definition in a distributed and therefore uncertain mode. 6 The best predictions from the individual time series are based on the last two years only, indicating in itself the rapid development of the Internet that tends to overwrite previous historical data as it develops. This is reflected in the exponential growth rates visible in Fig ure 2.4. 28 This further analysis would enable us to specify the relative importance of the sectoral differentiation in Triple Helix patterns, the in tellectual organization in terms of journals, and the national subdivision of the system of international publica tions. Similarly, at the level of the Internet, one can repeat the above searches for specific domains like .nl for the Netherlands or .de for Germany. These various dimensions can also be studied in terms of the m utual information among them. The results can then be interp reted, enab ling us to raise further research questions. On the basis of previous projects we expect to find self-org aniza tion (that is, negative entropy in mutual information statistics) partic ularly when the dataset is restricted to addresses in the U.S.A., the European Uni on, and Japan, but not (or much less so) when we focus on relations among the 15 EU na tions (Leydesdorff & Heimeriks, 2001). As industry was poorly represented in the data from the Science Citation Index , the scientometric results were here above also compared with webometric data using the Altavista Advanced Search Engine. Our main purpose was to show th e practicality of the methodology in both static and dynamic designs. On the Internet, a strong developm ent of the Triple Helix could be demonstrated dur ing the period 1995-2000. Using this specific representation, it could be show n that the development was se lf-organ izing because of the bi-lateral relations between universities, industries, and governments. The unilateral and tri-lateral developments did not cont ribute to inte gration in this case. The various data used in this study are in teresting in them selves, but they remain statistical and, therefore, there are problem s with the measurement (for exam ple, when using AltaVista; Rousseau, 1999). We used the data sets above as examples of the sort of results that one is able to obtain when pe rforming empirical Triple Helix or “Mode 2” research. Our argument in this study is that one can easily obtain com plex data, but that these can only be analyzed with reference to a system that is hypothesized as being codified. In other words, the focus in this study was not on the measurement, but on the methodology, for analyzing data gathered for th e evaluation of Triple Helix developments that have been measured—qualitatively an d/or quantitatively—in term s of bi- and trilateral relations. A design fo r organizing this data and methods for its evaluation was specified. 29 Chapter Three T HE C OMMUNICA TI VE T URN IN THE S TUDY OF K NOWLEDGE -B ASED S YSTEMS In a complex and non-linear dynamic, each system remains under reconstruction and in evolutionary competition while reorgan izing complexity within its re levant environments. Since communication systems ar e increasingly knowledg e intensive, this reconstructive dynamics is continuously re inforced. The borders of the systems under study are then increasingly uncertain and theref ore a subject of theoretical reflection. New codifications reconstructed by the ongoing pr ocesses of innovation and translation may become more functional than the underlying ones in an evol utionary mode. The previous configurations can be translat ed and partially overwritten. The philosophy of science has b een responsive to these deve lopments in, and interfacing of, scientific communication. Firs t, the systematic use of scie nce in industry in the late 19 th century raised fundament al questions about the demarcation between science and non-science at the interfaces. This issue led to the so-called “linguistic turn” in the philosophy of science during the interbellum. While truth had previously been associated with ideas, a truth-value was, henceforth, attrib uted to statements, with so me being more likely to be true than others. The “communication turn” has changed the situ ation once more. The truth-value of a statement can increasing ly be considered as al so contextual. One has a degree of freedom to play with the centrality of concepts in terms of heuristics a nd puzzle-solving (Simon, 1969 and 1973). Kuhn (1962), for example, noted that the precise de finition of “atomic weight” differs between chem ical physics and physical chemistry, without creating confusion. Concepts have meaning within di scourses; meanings can be considered and reconstructed. Translations betw een discourses and reformulati ons can thus be considered as the carriers of knowledge-based developments (Leydesdorff, 2002). This implies neither ar bitrariness in what is true or not, nor a re lativistic position. It implies an empirical o rientation; communica tions leave traces that can be used as indicators. The various dimensions of a communication (includi ng its potential truth value) can be distinguished. Values in these different dimensions can be measured when the communication system is fully specified as the hypothesis. For example, scientific discourses can be e xpected to develop over time and thus may change in terms of what is considered to be true. Although the deline ation of a discursive system (e.g., the paradigm) rem a ins uncertain in te rm s of its boundaries, it can also be expected to be more certain in terms of its core. Codifications st ructure the discourses; translations enable the sk illed participants to comm unicate among them. All these 30 communicative acts can be observed and meas ured in terms of their in tensity and frequency. 3.1 The need for “reflexive indicator research” We argued that future indicator research should combine theoretical, historical, and empirical orientations. Indicators indicat e co mmunication processes that can (i) be analyzed substantively, (ii) modeled, and ( iii) measured in a variety of dimensions. Hitherto, the focus in most analyses ha s remained on how communications develop historically. Starting at a ce rtain point in time, paths of development can be traced following the arrow of time. The historical pe rspective dates the point for reference b ack in history. The evolutionary perspective pr ovokes the historical analysis by takin g the present state as its point of reference. The system under study develops in the present by communicating given its past. One can then disc uss how far one has to track back in order to understand the present state. The st udy of communication systems requires both historical and evolutionary oriented re search designs because the evolutionary “incursion” (Dubois, 1998) takes place in hi story. Giddens (1979) has called this “a double hermeneutics.” The understanding of the communication with hindsight feeds back on the historical understanding. In addition to the historical and evolutionary analyses, the structur al dimensions can be analyzed using sociological methods or th e evolutionary metaphor of “variation and selection.” This provides a snap shot of the complexity at one or another m oment in time, e.g., the present. Note that the historical anal ysis tends to use concepts such as “change and stabilization” along the time axis, whereas structural analyses focus on “variation and selections” at specific mom ents in time. The same events (“var iation”) can be provided with different semantics by using differe nt (orthogonal) axes for the reflection. If one considers the three perspectives—the forward, the structural, and the backward analysis—as independent dimensions of re flection one generates a picture in which unstable phases and the emergence of new systemness appear m uch clearer than when using only a single (e.g., historical) metaphor for the representation. Such a reflexive (re)combination additionally provides the pers pectiv e of prospective intervention and policy-making because the recombination includes the discourse of what the representations may mean in the present. In other words, one expects different pers pectives on communication processes to be possible because a communication system—t hat operates in terms of changing distributions—can be accessed from different angles. The historical description of why specific patterns of variation and stabilizat ion were shaped and reproduced, is one among these possible angles. The functions of communi cations, however, relate to the stru ctures prevailing in the present. These can be analyzed sociologically, economically or sociometrically. However, the network repr esentations may easily becom e too static. 31 They allow only for comparative static s. How can the underlying communication structures also be changed, for example, by policy and/or m anage ment interventions? How do options of change relate to structural patterns of ong oing processes of change (or drift)? An evolutionary perspe ctive that appreciates the dyna mics of stabilization with reference to globalization, can be added to the model. A dynamic is complex insofar as it can be d ecomposed in term s of various interacting subdynamics. The variety of perspectives can be combined intuitively (for exam ple, by a politician) and/or one can recombine the theo retical perspectives by using a m odel. The various perspectives have then to be formulat ed as manifestations of the comp lex system under specific conditions. One can try to capture this complexity by using a sim ulation model. However, the simulation model abst racts f rom the content in the unde rlying processes. It provides us with a formal representation that req uires substantive appreciation by a reflexive discourse. The social process is non-trivial: it makes each representation one knowledge claim among the possible representations . Latour (1988) has called th is “infra-reflexivity.” One cannot achieve a meta-position by further leg itim ating and/or delegitimating positions under study, for example, by using sophisticat ed mathematics. The form alization only enables us to evaluate the relative quality of the narratives as hypotheses explaining the phenomena and then the results may help to update the hypotheses for a next round of translations. The unforeseen side effects, unintended conseq uences, etc., can be expected to challenge the various discourses to update given the ye t unexplain ed and perhaps counter-intuitive findings of the quantitative evaluation. The re search process continuously improves on the quality of the representation in a com pe titive mode, but the r epresentations refer to systems that are developing at the same ti m e. The knowledge-based system remains in transition and the study of the system ther efore needs the furthe r development of its reflections. The Organization of Reflexi vity within the Systems The innovated systems absorb knowledge by being innovated. The observable arrangements therefore have an epistemo logical status beyond merely providing the analyst with one or another, as yet unreflex ive starting point for the narrative. The data can be used for informing ex ante —and for sometim es testing ex post —the theoretical expectations. Which layer operated with wh ich function, why, to which extent, and in which instances? This research program begi ns with expectations as different from observations: methodologically controlled obser vations can then info rm the theoretical expectations. The feedback layers of reflexive R&D manage ment, science policy, ci tation analy sis, etc. have in the meantime changed processes of knowledge production in science, technology, and society (Wouters, 1999). The nature of these changes is not unambiguous, but a number of hypotheses have been elaborated in the literature. However, there can be no 32 doubt that new communication processes ha ve been taking momentum since the introduction of the Internet and other in formation and communi cation technologies. Processes of communication can be expected to change basic mech anisms of knowledge production and communication. The communication of know ledge feeds back on its production by further codifying and pot entially changing previous knowledge. Indicator research so far ha s reacted sensitively on the changes in knowledge production. New indicators have been proposed regularl y. The growing field of webometrics has witnessed an “indicator flood” in an incr easingly information rich and knowledge-based environment. This creativity of indicator rese arch may turn into a weakness if no theoretical backing can be developed. Which indicator is a relevant one, for which process, at which moment in time? Only few studies have tried to include a reflexive level when a new indicator is proposed. What is indicated by the indicator and why is this indicator m ore suited to that purpose than comparable ones? There is an in trinsic ne ed of validation studies within the indicator domain that is reflexive on the dynamics of the system s that are indicated. Reflexivity gains a particular urgency in phases of de-stabilization, re-organization and the emergence of new (and potentially innova tive) structures. When the communication structures are developing at the same time, the starting points or the systems of reference have to be made as clear as possibl e so th at one can trace the cha nges that are under study in relation to the changes that are made visible and/or explai ned by the study. This reflexivity can be elabor ated in each of the thre e dimensions: theoretic ally, historic ally, and empirically. By raising first the substantive questi on of “what is communicated?”—e.g., economic expectations (in terms of profit and growth), theo retical expectations or perhaps scenarios of what can technologically b e realized given institu tional and geographic constra ints— the focus is firmly set on the spec ification of the media of communication. How are these communications related and converted into one another? W hy are these processes sometimes mutually attractive and reinforc ing one another, and under what conditions can the exchanges among them be sustained? In the historical dimension reflexivity stands for the introduction of a perspec tive that focuses on dynamic processes in contexts rath er than on historical results (e.g., Barnes & Edge, 1982; Latour, 1987). As noted, the evolu tionary perspective includes the time axis, but as a degree of freedom. The present is the relevant system of reference for p olicy analysis. However, the present is also historic al, that is, as a transient state towards new developments. Furthermore, the reflexive an alyst is aware of one’s own position in relation to previous lines of res earch and one’s social contexts. In the empirical dimension, reflexive indicato r research also comm unicates the expected boundaries of the methods and data used. Whic h were the selection criteria? What would count as unexpected events? Can the surprise value of the newly emerging developments be expressed in bits of information added to the system ? Thus char acterized “reflexive indicator research” is not a new paradigm. It reflects traditional standards of scientific analysis. But, given the drive by the data in indicator research and the developm ent 33 towards an algorithmic understanding, the stre ngthening of a theore tical approach m ay become a necessary condition for the further developm ent of quantitative approaches to the study of science, technology, and innovation. During the last two decades, the qualitative a nd the quantitative traditions in science and technology studies have grown increasingly apart (Leydesdorff & Besselaar, 1997; Van den Besselaar, 2000 and 2001). It is time for the pendulum to be turned given the urgent need to understand the effects of different forms of communication and their interaction in knowledge production. In our opinion, the grow ing diversification a nd specialization in the sciences, and the relations hips to their societal envi ronments, in a knowledge-based economy calls for integrative approaches with detailed appreci ation of the ongoing processes of differentiation. It is only on such a basi s that one can m ore precisely describe the options for m aking choices both in th e public and in the private dom ains (of enterprises, research groups, etc.). The need for knowledge-based sc ience-policy making to be ab le to m ake a distinction between what might make a difference and what might not, is reflected in the seriousness of the problem of integration and differen tiation in the theore tical description and explanation of the knowledge-b ased systems under study. Both qualitative theorizing and quantitative information are then needed. Our theoretical f ramework is neither exc lusive nor normative. On the one hand, we need the qualitative contribu tions because they generate hypotheses. On the other, indica tor resea rchers can pay attention to the elaboration of the theoretical frameworks implied in their research. As comm unication theoretical, systems theoretical, and evolution theoretical concepts are involved, this task of integration through reformulation cannot be considered as a sine cura (Luhmann, 1975). It is a widely held prejudice that quantit ative analysis is data-driven and poor in theorizing. In our opinion, this is a cultural misunderstanding. No measurem ent is purely technical; theoretical baselines are always involved. Each definition of a variable implies an image of a process that is represente d. The theoretical references are often not completely described in quantitative studi es. However, the problem is not “missed theory,” but “invisible theory.” The cure is discursive reflexiv ity. Indicator research is not a disciplin e with a single and commonly accepted theoretical background. It is an “in terdiscipline” with approaches as different as the disciplinary background of the researchers. 7 What can still be justified in the case of research results p resented to one scientif ic community of specia lists—when the theoretical foundations provide a common basis so that they do not have to be repeated—may loose its justification wh en crossing a (sub)disciplinary boundary. If the theoretical backgrounds in indicator research are no t sufficiently reflected, it becomes impossible to create “trading zones” (Nowotny et al., 2001). These trading zones 7 The observable data can be considered as “ phenotypical,” whereas the perspectives for th eir interpretation compete as “genotypes” that may be able to explain the observable variations (Langton, Taylor, Farmer, & Rasmussen, 199 2). 34 are needed in order to create a dialogue between different approaches inside the branch of quantitative analysis (e.g., between simulati on studies and measurement efforts) as well as towards qualitatively oriented scie nce, technology, and innovation studies. New research questions can then be formulated that appreciate the previously achieved results. 3.2 Indicators as representations of codified communications Our theoretical contribution in this study has been the use and operationalization of the communication-theoretical framework for st udying reflexively developing systems of knowledge production and control. This perspec tive enabled us to understand an indicator as a specific representation of a process of knowledge production and communication. The process that is represented can be specified as a theoretical hypothesis, and the indicator then provides us with observations that can be used to inform (enrich or sometimes reject) the hypothesis. In general, a STI indicator st ands for a social process in science and technology. The processes of communication can be made obser vable by using the indicator. However, since the processes of communication are dist ributed, the m easurement can be expected to contain an uncertainty. The observations therefore have to be interpreted. We use a communication theoretical and system s theoretical approach. Social processes become visible in the communi cations used in the social systems under study. Following Luhmann (1984) and others, social communication sys tems can additionally be expected to differentiate functionally. Accordingly, th e communications develop different systems of communications endogenous ly. We used Luhmann’s notion of “codification” to describe the different forms in which communications can be expressed, stored, and recalled. Indicator research is based on the as sum ption that it is possible to recall th e information more precisely by m ethodol ogically controlling the measurement instruments. The “literature m odel” has dominated the quantitative study of sc ientif ic communication in scientometrics for many decades. The scientif ic journal article has been considered as the core of this model and, as a result, codified communi cation was the basic form of communication under study. However, the ch anges in knowledge production and its embedding in communication, as we obser ve these phenomena nowadays, have consequences for the codification of scientific communication. For example, the increased use of info rmation and communication technologies in science (e.g., on-line publica tions, digital data production, and simulations) m ay already have challenged the model of the “journal article” as the pr evailing fo rm of scientific communication. The shift of attention from science, to science-based technology and innovation has led to systematic indicator research in patent databases. The embeddedness of science in technology, and vice versa, can be traced analyzing the differences and asymmetries in citation patterns between the dom ains of scientific literature and patenting (Schmoch, 1997; Grupp & Schmoch, 1999; Meyer, 2000a,b). 35 In order to produce a measurable indicator , the hypothesis of a process has to be operationalized. Variables are then defined. Measurement m ay lead to repeatable and reliable results (or not). The availabi lity of databases functions as a constra int. Databases are the backbone of indicator research. Both the availability of databases and the development of statistics provide inhe rent limitations to indic ator research. In this paper we m ainly focused on datab ases like the Science Cita tion Index . Medline and the European Patent Office database are other examples of data bases us ed in bibliometrics. We also discussed w eb data, e.g. using the Advanced Search Engine of AltaVista. In general, dedicated databases can be understood as representing specific types of codification in knowledge-bas ed communication. For example, the Science Citation Index is mainly useful for m apping the communication in science located in universities and public research institutions . Patent databases re present communications about technologies. The classification scheme outlined in Table 3.1 m ay be helpful in organizing the respective domains. Our empha sis on differentiation at interfaces as a condition for innovation has led us to focus on the combination of the different data sources. Table 3.1 Functional versus instit utional differentiation in the Internet age We consider the process of knowledge creatio n as a stepwise process from so-called “fundamental” knowledge towards market releva nt innovations, and vice versa, from the market into the knowledge pr oduction process. This proce ss contains feedback loops within each of the systems and among them. Each subsystem develops recursively and interactively. The feedback loops control th e forward mo vement of the process in an organized way (Kline & Rosenberg, 1986). The interaction of different knowledge networks links different phases in a heterogeneous process. Each heterogeneous pr ocess itself contains one rec onstruction of the historical 36 events among others possible, bu t the selection takes place fr om a hindsight perspective. It can be considered as an actualization (a state) of the system. The different phases and processes can be made visible as differences in the codific ations. The focus remains o n the emerging systems that result from this non-linear dynam ics. A linear combination of databa ses only in creases the comp lexi ty of the de scription by extending a relatively sim ple representation into the multidimension al perspective of interacting subdynamics. In order to handle the complexity, theory has to be introduced as an integrative and organizing element. The em erging system can be hypothesized. For example, one can consider the case of the creation and introduction of a pharmaceutical to exemplify how multidim ensionality can be bundled together in the description of the specific process of knowledge creation and innovation (Leydesdorff, 2001b). The innovation can be analyzed as a pe rformative act in history (Latour, 1987). Our perspective, however, is (neo-)evolutionary and syst ems theoretical: how are the coordination mechanisms between functi onal dom ains affected by innovation? The knowledge-based innovations can be expected to reconfigure the structures on which they build by reconstructing and recombining th em in term s of new representations. The uncertain definition of a system of innovation in term s of nations, sectors, technologies, regions, etc., br ings players other than the traditional ones into scope. Following upon the Bayh-Dole Act (1980), for ex ample, universities in the U.S.A. have been stimulated to subm it patent applicati ons. Does university research already play a strategic role in a domain of pa tentin g? Whereas this role can his torically be analyzed for innovations on a case-by-case basis, the delineation of a system of innovations is required for defining this role at the aggregated level. We propose to consider the st udy of “innovations” and the pote ntially systemic character of clusters of innovations as a third program of research in science, technology, and innovation studies . Science indicators have hither to focused on performance and scientific impact. Patent indica tors measure technical inv entiveness from a historical perspective (Sahal, 1981). I nnovation indicators turn the tables by using a hindsight, systems, and/or evolutionary perspective. I nnovation is per definition an emerging unit of analysis based on communication between diffe rent systems. The innovated systems can be changed significantly by an innovation. Innovations have been analyzed mainly under the aspect of technol ogical diffusion and technological forecasting. Note that the system of reference of such studies has been th e technological development under study: One then asks for the consequences or im pacts of new technologies, for example, in terms of technology assessm ent. In innovation studies, technological developments (or stagnations) them selves have to be explained. Under which conditions can further innovation be expected? U nlike bibliometric and patent analysis, modeling plays a more dominant role in this area because of the focus on new and emerging options. Empirical studies that trace th e growth of an innovation b ack have been relatively scarce in science and technology st udies and evolutionary econom ics (Von Hippel, 1988). One 37 reason may have been the lack of standardized “innovation databases” (Pavitt, 1984). Data gathering is often time-consumi ng, being tailor-m ade (Frenken, 2000 and 2001; Frenken & Leydesdorff, 2001). The Internet provides a new perspective on “data mining.” Systematic links between innovation (m arket), invention (patent) and scientific knowledge (literature) can now be constructed. A closer connection between science, te chnology, and innovation studies has also theoretical consequences. In innovation studies the focus is on the (re)constructed system as different from the historical constructi on. Models and simulations are introduced to explore the dynamical nature of the reco nstruction. From the perspective of the reconstructed system, innovation potentially restru ctures the history of the representations because the system continuously selects upon the variety of possible representations for its reconstruction. For example, when a national system of innovati on is assum ed as the system of reference, dimensions can be appreciated other than wh en one assumes a sector (e.g., chem istry) or a new technology (e.g., biotechnology) as the evolving system. One can always question these delineations as assum ptions; they can only be used as starting points for the reconstruction. Therefore, the definitions and delin eations have to be communicated together with the indicators proposed. Diffe rent perspectives can be expected, since innovations take place at the interfaces between systems. Reflexive indicator research becomes necessary when innovations are made the focus of research. On the basis of understanding the processes of knowledge creation as self-organizing and complex, we indicated above how one can te st for the hypothesized phenomena in term s of new structures potentially reproduced by coherent behavior. The case of the em ergence of a so-called European Research Area is such an example one would look for. The politically motivated proposal aims at an institutional in novation that should drive the European sciences into a phase transiti on. The emergence of new form s of self- organization can sometimes be tested (e.g., Leydesdorff, 2000a; Leydesdorff & Heimeriks, 2001). This approach, that is of testing a hypothesi s as a research question, can be extended to other theses. For example, the “Mode 2” th esis posits a change in the system of innovations in cognitive, social as w ell as institutional dim ensions. To which extent can hypotheses based on the “Mode 2” theory be tested empirically by using indicator research? Independently of the acceptance of the model as an explanation, one can entertain the thesis as a h ypothesis and ask for new social forms of knowledge production like virtual communities or collaboratories. W h at have been the effects of electronically mediated communications on scientific know ledge production and diffusion? In this context, web indicators are perhaps the most appropriate represen tations (Zelman, 2002). Web Indicators In the empirical part of this study, we drew attention to the possibi lities and limitations of web indicators. The Internet represents a medium for differently codified 38 communications. Traditional databases can nowadays be used on-line like the Web of Science— the on-line version of the Science Citation Index . Traditional communication channels like the scientific journals b ecome increasingly available on-line. In addition to these trends of digitalization, th e web offers the possibility to trace research activities that have not yet been documented. The rise of so-called Virtual Ethnography (Hine, 2000) is only one among a variety of new methodologies that have become available in science and technology studies . Cybermetrics or W ebometrics stand for quantitative approaches in this direction. Prob lems of reliable data and stability of the measurement over tim e are major methodologic al problem s when using web indicators. These problems do not have to surprise us gi ven the dynamic character of the Internet. Using search engine or meta-crawlers, one can compare the frequency of very different kind of communications (keyword s) as well as of institu tions (e.g., host extensions). However, further methodological research conc erning the stability of search engines m ay then increasingly be necessary. Different update frequencies can be expected in different domains. In summary, one can state th at automatized ways of data m ining have to be developed to use the “data flood” on the web for indicator purposes. The automation and consequent black-boxing of th eoretical assumptions into st andards, however, generates another tension that can drive new research processes both empirica lly and theoretic ally. The standards may have to be updated regularly. The standardization of “purchasing power pari ty” by the OECD can be considered as an early example of an evolutionary indicator be cause the values of th ese input indicators had regularly to be updated wi th reference to changes in th e exchange rates. This study concentrated on output indicators. However, we wish to draw attention to the effect of these further developments in output measurem ent on the study of i nput indicators. Input indicators like R&D expenditure, R&D pers onnel have mainly been developed by the OECD (e.g., 1976) and are often used in macr o analyses of science policy (e.g., OECD, 1980). Relatively less attention was drawn in this repor t to the ef fect of the ICT revolutions and the emerging focus on innovation on research of input indicators (OECD/Eurostat, 1997). For future indicator research, however, one may wish to raise questions about the need to match new (e.g., web-based) indicators with in put indicators. For example, ICT is often not classified as R&D. How is the efficiency of spending in new ar eas of tec hnoscience to be measured if interaction effects be tween “R&D” and “non- R&D” activities may become more important than the sum of the two efforts (Kaghan & Barnett, 1997)? 3.3 A Program of Innovations Studies A fundamental reformulation of the probl ems of Science, Technology, and Innovation Policies became urgent during the 1990s. Thr ee models have been central to the discussion about studying innovation systems: (i) the proposal to distinguish a “Mode 2” 39 type of knowledge production, (ii) the model of “national system s of innovation”, and (iii) the triple helix of univers ity-industry-governm ent relations. The authors of the “Mode 2” thesis (Gibbons et al ., 1994) have argued that the new configuration has led to a dedifferentiation of the relati ons between science, technology, and society. From the perspectiv e of thes e authors, all scientific and technical communication can be equated and compared with oth er communication from the perspective of science, tec hnology, and innovation policies. In our opinion, this model is based on a co nfusion of the representation with the represented system under study. The political or managerial repres entation provides us with a window that can be integrated because it uses a specific medium of communication. However, the represented syst ems are operationally expected to remain differentiated. If the integration is also successful in the systems under study (e.g., in the case of a reconstruction or i nnovation), the systems are in tegr ated at their interfaces and therefore they can be expected to restore al so their own orders by differentiating again after the integration. The inte gration means som ething differ ent for differently codified systems. Differentiation and integration do not exclude on e another, but rather assum e one another. The communication enables us to construct an integrated picture, but the underlying systems compete both in terms of their so cial realities and in term s of the representations that they enable us to cons truct at the interfaces. Systems of innovations solve the puzzle of how to interface different functions in th e communicatio n at the level of organization. Evolutionary economists have argued in favor of studying “national systems of innovation” as hitherto the most relevant leve l of integration. Indeed, they have provided strong arguments for this choice (Lundva ll, 1992; Nelson, 1993; Skolnikoff, 1993). However, these systems are continuously restructured under the drive of global differentiation of the expectations. Economie s are interwoven both at the level of the markets and in term s of multinational corporations, sciences are organized internationally, and governance is no longer limited within national boundaries. The most interesting innovations can be expected to involve boundary-spanning mechanisms. In other words, we agree with the “Mode 2”-model in assum ing a focus on communication as the driver of systems of knowledge produc tion and control. However, the problem of structural differences among the communications and the organization of interfaces remain crucial to the understandi ng of a global and knowledge-based economy. The wealth from knowledge and options for furt her developments have to be retained by reorganizing institutional arrangements with reference to the global horizons. The Triple Helix model of university-industry- government relations tries to capture both dynamics by introducing the notion of an ove rlay that feeds back on the institutional arrangements. Each of the helices develops in te rnally, but they also interact in terms of exchanges of both goods and services and in terms of knowledge-based expectations. The various dynamics have first to be distingui shed and operationalized, and then som etimes 40 they can also be measured. Throughout this report we have tried to show how the dynamics between the dimensions can then be reconstructed using indicator research. The strength of this research program is that it does not simply ge neralize on the basis of intuitions. The empirical results can be ex p ected to inform us. As the complexity increases, the results may often be counterintu itive. One m ay be able to appreciate them by innovating one’s theoretical assumptions. As the various subdynamics can better be understood, one may also be able to devel op simulation models on the basis of their reconstructions. There is an intimate connection between indi cator research and pa ramet er e stima tion in simulation studies when analyzing knowledge -based systems. Indicators study knowledge production and communication in terms of th e traces that communicatio ns leave behind, while simulations try to capture the operations and their interactio ns. The common assumption of indicator research and simu lation studies is that knowledge production, communication, and control are considered as operations that change the materials on which they operate. The unit of analysis is replaced with a unit of operation. The difficult relations between empirical stud ies and algorithm ic simu lations have to be guided by theorizing. Otherwise, the number of options explodes without quality control. What do the different pictures mean? Both theoretical specification and methodological control are needed. In our opinion, the study of communica tion and the interfacing can use available theories of communication. We have argued that two theories are then particularly important: (i) Luhmann’s soci ological theory of co mmunication w ith its emphasis on functional differentiation (Luhmann, 1984 and 1990; Leydesdorff, 2001a) and (ii) the mathem atical theory of communication that can be used for the operationalization (Shannon, 1948; Abramson, 1963; Theil, 1972; Leydesdorff, 1995). The combination of these two theories with a very different status—t hat is as th eory and methods—enabled us in the various chapters of this study to update and inform empirical hypotheses about how the knowledg e base transforms the inst itutional relations of an increasingly knowledge-based society. 3.4 Policies of Innovation: Innovation of Policies? The gradual transition from a political econom y to a knowledge-based economy potentially changes the cause-effect relations hips between the control systems and the systems to be steered. As the system to be steered becom es increasingly self-organizing— for example, in terms of containing “loc k-ins”—the options for steering become dependent on the windows that the system s leave for intervention. These windows can only be established on the basis of knowledge-based reconstructions. For example, in a political economy, the political system is inclined to steer a system like the scientific enterprise (or the national system of innovations) in term s of its institutional parameters (Spiegel-Rösing, 1973; Van de n Daele, Krohn, & Weingart, 1977). In a knowledge-based economy, institutional paramete rs tend to lose thei r relevance as the 41 institutions are under pressure of reorga nization. Networks of institutions shape university-industry-gover nment relationships in a non- linear dynam ic. The latter are driven by political and economic opportunitie s to grasp com petitive advantage. The knowledge-base of the economy is highly stru ctured by relevant interfaces that are continuously reproduced and yet differentiated within the system and its relations to different environments. “Validation boundaries ” can, for example, be considered as a knowledge-based equivalent of institutional b oundaries (Fujigaki, 1998). Validation boundaries are the result of codification pr ocesses that reconstitute institu tional delineations. A validation boundary can be expected to have an internal and external side with different characteristics. Fujiga ki & Leydesdorff (2000) have elaborated upon the concept of “validation boundaries ” for exchange and control processes at the societal interfaces of knowledge-based systems. Validation boundaries can also be considered as condensati ons and stabilizations of interacting communication processes. Each of the communication processes selects asymmetrically and asynchronically at releva nt interfaces, but some selections can be selected for stabilization ; some stabilizations can recursively be selected f or globalization. The institutional level provides the stab ility that is needed for participation in the globalization. Thus the perspective of the institu tional optimalization refers to the carrying capacity and the sustainabi lity of the network arrangem ents. In addition, the institutional environment provi des a trade-off between the m echanisms of a political economy such as pub lic control and private appr opriation. If one begins the political process only at the latt er end—for example, because th at is the tr aditional routine of producing policy decisions—one tends to lose precisely the knowledge-based dimension in the policy-making and/or the m anagerial processes. Knowledge-based development requires the policy and management control process itself to be innovated accordingly. For example, institutional interests have b een shaped by history. We have argued that evolutionary analysis requires the consideration of the historical formations ref lexively. The functionality of the institutions can continuously be discussed and analyzed. The discussion of the functionality of the delineations can be inf o rmed by indicator research. 42 R EFERENCES Abbate, Janet (1999). Inventing the Internet . Cambridge, MA: MIT Press. Abramson, N. (1963). Information Theory and Coding . New York, etc.: McGraw-Hill. Allen, Peter. M. (1994). “Coherence, Chaos and Evolution in the Social Context.” Futures 26(6): 583-597. Almind, T. and P. Ingwersen (1997). “Inform etric Analyses on the World Wide Web: Methodological Approaches to "Webometrics".” Journal of Documentation 53: 40 4-426. Anderson, J., P. M. D. Collins, J. Irvi ne, P. A. Isard, B. R. Martin, F. Narin, and K. Stevens (1988). “On-line approaches to measuring nati onal scientific output—A cautionar y tale,” Science and Public Policy 15, 153-61. Arthur, W. Brian (1989). “Competing Technol ogies, Increasing Returns, and Lock-In by Historical Events,” Economic Journal 16, 116-131. Bar-Ilan, Judith (2001), “Data collection methods on the Web for informetric purposes—A review and analysis,” Scientometrics 50(1), 7-32. Barnes, B., & Edge, D. (1982). Science in Context . Cambridge, MA: MIT Press. Barras, Richard (1990). “Interactive Innovation in financial and business services: The vanguard of the service revolution”. Research Policy 19, 215-37. Beck, Ulrich (1986). Die Risiko-Gesellschaft . Frankfurt a.M.: Suhrkamp. Beck, Ulrich (1992). Risk Society. Towards a New Modernity . London, etc.: Sage. Bjöneborn, L. and P. Ingwersen (2001) . “Perspectives of webometrics.” Scientometrics 50(1): 65- 82. Blauwhof, Gertrud (1995). The non-linear dynamics of technological developments: an exploration of telecommunications technology. Unpublished Ph.D. Thesis, University of Amsterdam . Bonitz, Manfred (Ed.) (2001) Memo rial issue of V. V. Nalim ov, Scientometrics 52(2), 101-36 0. Boudourides, Moses A., Beatrice Sigrist, & Phili ppos D. Alevizos (1999). “Webometrics and the Self-Organization of the European Information Society,” at < http://hyperion.math.upatras.gr/webom etrics/ > (26 October 1999). Boudourides, Moses A. & Gerasimos P. Antypas (2002) “A Simulation of the Structure of the World Wide Web.” at http://thalis.mat h.upatras.gr/~mboud our/articles/ssweb.pdf Braun, T., W. Glänzel, and A. Schubert (1989). “Assessing Assessments of British Science. Som e Facts and Figures to Accept or Decline,” Scientometrics 15, 165-70. Braun, T., W. Glänzel, and A. Schubert (1991). “The Bibliometric Assessment of UK Scientific Performance—Some Comm ents on Martin’s Reply,” Scientometrics 20, 359-62. Braverman, Harry (1974). Labor and Monopoly Capital . New York/London, Monthl y Review Press. Bruckner, Eberhard, Werner Ebeling & Andrea Scharnhorst (1990). “T he Application of Evolution Models in Scientom etrics.” Scientometrics 18 (1-2), 21–41. Bush, Vannevar (1945). The Endless Frontier: A Report to the President. Reprinted, New York: Arno Press, 1980. Callon, Michel, John Law, and Arie Rip (1986). Mapping the dynamics of science and technology . London: Macmillan. Collins, Harry M. (1985). “The Possibilities of Science Policy,” Social Studies of Science 15, 554-8. Cozzens, S.E., Healey, P., Rip, A., Ziman, J. (1990) The Research System in Transition . Kluwer Dordrecht Crawford E, Shinn T, Sörlin S (Eds) (1993) Denationalizing Science: The Contexts of International Scientific Practice. Sociology of Sciences Yearbooks , Vol. 16. Dordrecht: Kluwer 43 Cronin, B. (2001). “Bibliometrics and beyond: some thoughts o n web-based citation analysis.” Journal of Information Science 27(1): 1- 7. Dubois, D. M. (1998). “Computing Anticipatory Syst em s with Incursion and Hyperincursion.” In D. M. Dubois (Ed.), Computing Anticipatory Systems, CA SYS-First International Conference (Vol. 437, pp. 3-29). Woodbur y, NY: American Institute of Physics. Ebeling, Werner, and Andreas Scharnhorst (2 000). “Evolut ionary Models of Innovation Dynamics.” In: Traffic and Granular Flow ’99 – So cial, Traffic, and Granular Dynamics . Ed. by D. Helbing, H.J. Herrmann, M. Schreck enberg, D.E. Wolf. Berlin: Springer, pp. 43– 56. Egghe, Leo. (2000). “New informetric aspects of th e Internet: some reflections - m any problems.” Journal of Information Science 26(5): 3 29-335. Egghe, Leo & Ronald Rousseau (1990): Introduction to Informetrics: Quantitative Methods in Library, Documentation and Information Science . Amsterdam. Epstein, Joshua M. and Robert L. A xtell (1996) Growing Artificial Societies. Social Science from the Bottom Up . Washington: Brookings & MIT Press Etzkowitz, Henry & Loet Leydesdorff (2000). “T he Dynamics of Innovation: From National Systems and “Mode 2” to a Triple Helix of Univ ersity-Industry-Government Relations,” Research Policy 29, 109-123. Etzkowitz, Hen ry & Loet Leydesdorff, “The Triple Helix ---University-Industry-Gov ernment Relations: A Laboratory for Knowle dge Based Economic Develop ment,” EASST Review 14(1) (1995) 14-9. Etzkowitz, Henry (1994). “Technology Centers a nd Ind ustrial Policy: the Emergence of the Interventionist State in the USA,” Science and Public Policy 21 (2), 79-87 . Faulkner, W. and J. Senker (1995). Knowledge frontiers, public sector research and industrial innovation in biotechnology, engin eering cermaics, and parallel computing. Oxford/New York: Clarendon Press. Fisher, J.C. & R.H. Pry (1971) “A Simple Substitution M odel of Technological Change”, Technological Forecasting and Social Change 3, 75 -88 Freeman, Christopher (1982). The Economics of Industrial Innovation (Pinter, Londo n). Freeman, Chris, and Carlota Perez (1988). “Structu ral crises of adjustment, business cycles and investment behaviour,” in: Giovanni Dosi, Chistopher Freeman, Richard Nelson, Gerald Silverberg, and Luc Soete (Editors), Technical Change and Economic Theory . Londo n: Pinter, pp. 38-66. Freeman, Christopher and Luc Soete ( 3 1997 [1982] ). The Economics of Industrial Innovation London: Pinter. Frenken, K. (2000). “A Complexity Approach to In novation Networks. The Case of the Aircraft Industry (1909-1997),” Research Policy 29(2), 257-272. Frenken, K. (2001). Understanding Product Innovation using Complex Systems Theory. University of Amsterdam, Amsterdam. Frenken, K., & Leydesdorff, L. (2000). “Scaling Trajectories in Civil Aircraft (1913-1970),” Research Policy 29(3), 331-348. Gibbons, Michael, Camille Lim oges, Helga Nowotny, Simon Schwartzman, Peter Scott, and Martin Trow (1994). The new production of knowledge: the dynamics of science and research in contemporary societies . London: Sage. Giddens, Anthony (1979). Central Problems in Social Theory . London: Macmillan. Giddens, Anthony (1984). The Constitution of Society . Cambridge: Polit y Press. Gilbert, G. N . (1997) “A simulation of the structure of academic science.” Sociological Research Online; at http://www.socresonline.org.uk/socresonline/2/2/3.html , 1997, 2(2), 1-2. Gilbert, G. N., and M. J. Mulkay (1984). Opening Pandora’s Box. A Sociological Analysis of Scientists’ Discourse . Cambridge: Cambridge University Press. 44 Gilbert, N. and K. G. Troitzsch (1999). Simulation for the Social Scientist . Buckingham, Philadelphia: Open University Press. Godin, B. and Y. Gingras (2000). “The place of universities in the system of knowledge production,” Research Policy 29(2), 273-278. Goffman, W. (1966). “Mathem atical Approach to the Spread of Scientific Ideas – the History of Mast Cell Research.” Nature CCXII 5061, 449-452 Grupp, Hariolf, & Uwe Schmoch (1999). “Patent sta tistics in the age of globalisation: new le gal procedures, new analytical methods, new economic interpretation,” Research Policy 28, 377- 396. Hauff, V., and F. W. Scharpf (1975). Modernisierung der Volkswirtschaft. Technologiepolitik als Strukturpolitik . Frankfurt a.M.: EVA. Hesse, Mary (1988). “The Cognitive Claims of Metaphors.” The Journal of Speculative Philosophy. 2 (1): 1-16. Hine, Christine (2000). Virtual Ethnography . London: Sage. Ingwersen, P. (1998). “The calculation of Web Impact Factors.” Journal of Documentation 54(2): 236-243. Irvine, J., and B. R. Martin (1984). Foresight in Science: Picking the Winners . London: Pinter. Irvine, J., B. Martin, T. Peacock, R. Turner ( 1985). “Charting the decline of British science,” Nature 316, 587-90. Kaghan, W. N., & Barnett, G. B. (1997). “The De sktop Model of Innovation in Digital Media.” In H. Etzkowitz & L. Leydesdorff (Eds.), Universities and the Global Knowledge Economy . London and Washington: Pinter, pp. 71-81. Kauffman, Stuart A. (1993). The Origins of Order: Self-Organization and Selection in Evolution : New York: Oxford University Press. Kline, S., & Rosenberg, N. (1986). “An overvie w of innovati on.” In: R. Landau & N. Rosenberg (Eds.), The Positive Sum Strategy: Harnessing Technology for Economic Growth (pp. 275- 306). Washington, DC: National Academy Press. Kochen, Manfred (1983). “Mathematical Mode l for the Growth of Two Specialties.” Scie nce of Science 3(11), 199-217 Krugman, P. (1996). The Self-Organizing Economy . Malden, MA, and Oxford: Blackwell. Langford, Cooper & Martha Langford (2001). “F undin g Strategic University Research in Canada,” paper presented at th e Annual Meeting of the Society for the Social Studies of Science (4S), Boston, MA, November 2001. Larson, R. R. (1996). “Bibliometrics of the World Wide Web: An Exploratory Analysis of the Intellectual Structure of Cyberspace.” ASIS 96. Last accessed on August 4, 2001. See http://sherlock.berkele y.edu/asis96/asis96.htm l Latour, B. (1987). Science in Action . Milton Keynes: Open University Press. Latour, B. (1988). The Pasteurization of France. Cambridge, Mass./London: Harvard University Press). Lewison, G., & Cunningham, P. (19 91). “Biblio metric Studies for the Evaluation of Trans- National Research.” Scientometrics, 21, 223-244. Leydesdorff, Loet (1988). “Problems with the ‘m easurement’ of national sc ientific perform ance,” Science and Public Policy 15 (1988) 149-152. Leydesdorff, Loet (1991). “On the " Scientometric Declin e" of British Sci ence. On e Addition al Graph in Reply to Ben Martin,” Scient ometric s 20, 363-7. Leydesdorff, L. (1995). The Challenge of Scientometrics: the development, measurement, and self-organization of scientific communications . Leiden: DSWO Press, Leiden University. Leydesdorff, Loet (1998). “Theories of Citation?” Scientometrics, 43 (1998) 5-2 5. Leydesdorff, Loet (2000a). “I s the European Union Becoming a Single Publication System?” Scientometrics 47(2), 265-280. 45 Leydesdorff, Loet (2000b). “A Tri ple Hel ix of U niv ers it y-I ndus tr y-G ove rnm ent Re lat ion s,” The Journal of Science & Health Policy 1, 43-48. Leydesdorff, Loet (2001a). A Sociological Theory of Communication: The Self-Organization of the Knowledge-Based Society. Parkland, FL: Universal Publishers < http://www.upublish.com/books/leydesdorff.htm >. Leydesdorff, L. (2001b). “Indicators of I nnovation in a Knowledge-based Economy.” Cybermetrics , 5 (Issue 1), Paper 2, at http://www.cindoc.csic.es/cybermetrics/articles/v5i1p2.html . Leydesdorff, Loet (2001c). “Technology and Cultu re: The dissemination and the potential ' lock- in' of new technologies,” Journal of Artificial Societies and Social Simulation, Vol. 4, Issue 3, Paper 5, at < http://jasss.soc.surrey.ac.uk/4/3/5.html > Leydesdorff, L. (2002). “The Communication Turn in the Theory of Social Systems,” Systems Research and Behavioral Science, 19(2), 129-136. Leydesdorff, Loet, and Olga Amsterdamska (1990). “Dimensions of Citation Analy sis,” Scienc e, Technology and Human Va lues 15, 305-35. Leydesdorff, Loet & Michael Curran (2000). “Mapping University-Industry-Govern ment Relations on the Internet: An Exploration of Indicators for a Knowledge-Based Economy,” Cybermetrics 4 Issue 1, Paper 2 at < http://www.cindoc.csic.es/cybermetrics/articles/v4i1p2.html >. Leydesdorff, Loet & Henry Etzkowitz (1998). “The Triple Helix as a model for innovation studies,” Science and Public Policy 25(3), 195-203. Leydesdorff, Loet & Henry Etzkowitz (2001). “The Transfor mation of University-Industry- Government Relations,” Electronic Journal of So ciology 5 (4), at http://www.sociology.org/content/vol005.004/th.html Leydesdorff, Loet & Henr y Etz kowitz, “Emergence of a Triple Helix of University-Industry - Government Relations,” Science and Public Policy 23 (1996) 279-86. Leydesdorff, Loet & Élaine Gauthier (1996). “The Evaluation of National Performance in Selected Priority Areas using Scientometric Metho ds,” Research Poli cy 25, 431-50. Leydesdorff, Loet & Gaston Heimeriks (2001). “The Self-Organization of the European Information Society: The case of "biotechnology",” Journal of the American Society for Information Science & Technology JASIST 52(14), 1262-1274. Leydesdorff, Loet & Nienke Oo mes (1999). “Is th e European Monetary Sy stem Converging to Integration?” Social Science In formation 38 (1), 57-86 . Leydesdorff, L., & Peter van den Besselaar ( 1997). “Scientome trics and Communication Theory: Towards Theoretically Informed Indicators.” Scientometrics, 38 , 155-174. Leydesdorff, Loet, & Peter Van den Besselaar (1998). “Technological Development and Factor Substitution in a Non-linear Model,” Journal of Social and Evolutionary Systems 21, 173- 192. Leydesdorff, Loet, Philip Cooke and Mikel Olazar an (eds.), (2002). Special issue: “Regional Innovation Systems in Europe,” Journal of Technology Transfer 27 (1), 5-145. Lissenburgh, Stephen, & Rebecca Harding (2000) Knowledge Links: Innovatio n in university/business partnerships . London: IPPR. Luhmann, N. (1975). “Systemtheorie, Evol utionstheorie und Kommunikationstheorie,” Soziologische Aufklärung 2. Opladen: Westdeutscher Verlag, pp. 193-203. [Reprinted in: The Differentiation of Society , Translated by Stephen Holmes and Charles Larmore. New York: Columbia University Press, 1982, pp. 255-270.] ). Luhmann, Niklas (1984). Soziale Systeme. Grundriß einer allgemeinen Theorie . Suhrkamp, Frankfurt a. M. [ Social Systems (Stanford University Press, Stanford) 1995] . Luhmann, Niklas (1990). Die Wissenschaft der Gesellschaft . Frankfurt a.M.: Suhrkamp. Luhmann, Niklas (2002). Organization und Entscheidung . Opladen: Westdeutscher Verlag. 46 Lundvall, Bengt-Åke (1988). “Innovation as an interactive process: from user-producer interaction to the national system of innovation,” in: Giovanni Dosi, Chistopher Freeman, Richard Nelson, Gerald Silverberg, and Luc Soete (Editors), Technical Change and Economic Theory . London: Pinter, pp. 349-369. Lundvall, Bengt-Åke (Ed.) (1992). National Systems of Innovation . London: Pinter. Martin, B. R. (1991). “The Bibliom etric Assessment of UK Scientific Performance—A Reply to Braun, Glänzel and Schubert,” Scientometrics 20, 333-57. Martin, B. R. (1994). “British Science in th e 1980s—Has the Relative Decline Continued?” Scientometrics 29, 27-57. Martin, Ben R. & John Irvine (1983). “Assessing basic research: Some partial indicators of scientific progress in radio astronmy,” Research Policy , 12 (2), 61-90. Merton, R. K. (1942). “Science and Technology in a Dem ocratic Order.” Journal of Legal and Political Sociology, 1, 115-126. Meyer, Martin (2000a). “Does science push tec hnology? Patents citing scientific literature,” Research Policy 29, 409-34. Meyer, Martin (2000b). “What is special about pa tent citations? Differen ces between scientific and patent citations,” Scientometrics, 49, 93-123. Mulkay, Michael J. (1976). “The mediating role of th e scientific elite,” Social Studies of Science 6: 445-470. Mulkay, M. (1977). “Sociology of the Scientific Research Community.” In: I. Spiegel-Rösing & D. de Solla Price (Eds.), Science, Technology and Society (Vol. 93-148). Lond on: Sage. Mullins, Ncholas C. (1972). “The Development of a Scientific Specialty: The Phage Group and the Origins of Molecular Biology,” Minerva , 10: 51-82. Narin, Francis (1976). Evaluative Bibliometrics: The Use of Publication and Citation Analysis in the Evaluation of Scientific Activity , Washington, DC: National Science Foundation. Narin, Francis, & David Olivastro (1988). “Tec hnolog y Indicators Based on Patents and Patent Citations.” In: A.F.J. van Raan (Ed.), Handbook of Quantitative Studies of Science and Technology, Amsterdam: Elsevier, pp. 465-507. Narin, Francis, & David Olivastro (1992). “S tatus report: linkages between technology and science,” Research Policy 14, 237-249. Narin, Francis, and Elliott Noma (1985). “Is Technol ogy Becoming Science?” Scientometrics 7: 369-381. Nederhof, Anton J. (1988). “Changes in publica tion patterns of b iotechnologists: An evaluation of the impact of government stim ulati on program s in six industrial nations,” Scientometrics 14, 475-85. Nelson, Richard R. and Sidney G. Winter (1975). “Growth theory from an evolutionar y perspective: the differential productivity growth puzzle,” American Economic Review 65, 338-344. Nelson, R. R., and S. G. Winter (1977). “In Search of Useful Theory of Innovation,” Research Policy 6, 35-76. Nelson, R. R., and S. G. Winter (1982). An Evolutionary Theory of Economic Change . Cambridge, MA: Belknap Press of Harvard University Press. Nelson, R. R., ed. (1982). Government and Technical Progress: a cross-industry analysis . New York: Pergamon. Nelson, Richard R. (1994). “Economic Growth via the Coevolution o f Technology and Institutions.” In: Loet Leydesdorff a nd Peter Van den Besselaar (Eds.), Evolutionary Economics and Chaos Theory: New di rections in technology studies. London: Pinter, pp. 21- 32. Nelson, Richard R. (Ed.) (1993). Natio nal Innovation Systems: A comparative study . Oxford and New York: Oxford University Press. Noble, David (1977). America by Design . New York: Knopf. 47 Nowakowska, M. (1984). Theories of Research . Vol. I-II. Seaside: Intersy stems Publications. Nowotny, Helga, Peter Scott, and Michael Gibbons (2001). Re-Thin king Science: Knowledge and the Public in an Age of Uncertainty. Cambridge, etc.: Polity. OECD (1963, 3 1976). The Measurement of Scientific and Technical Activities: “Frascati Manual” . Paris: OECD. OECD (1980). Technical Change and Economic Policy . Paris: OECD. OECD/Eurostat. (1997). Proposed Guidelines for Collecting and Interpreting Innovation Data, "Oslo Manual" . Paris: OECD. Pavitt, Keith (1984). “Sectoral patterns of techni cal change: towards a theory and a taxonomy.” Research Policy 13, 343-73. Popper, Karl R. ([1935]/ 1959). The Logic of Scientific Discovery . London: Hutchinson. Porter, Michael E. (1990). The Competitive Advantage of Nations. London, Macmillan. Price, Derek J. de Solla (1976). “A General Theory of Bibliom etric and Other Cummulative Advantage Processes.”, Journal of the American Society for Information Science 27(1976)5, 292-306. Riba-Vilanova, Marta & Loet Leydesdorff (2001), “Why Catalonia cannot be considered as a Regional Innovation Sy stem,” Scient ometric s 50 (2), 215-240. Rosenberg, Nathan (1976). Perspectives on Technology . Cambridge: Cam bridge University Press. Rosenberg, Nathan (1982). “How ex ogenous is science?” pp. 141-59 in: Inside the Black Box: Technology and Economics . Cambridge: Cambridge University Press. Rothwell, R., & Zegveld, W. (1981). Industrial Innovation and Public Policy . London: Pinter. Rousseau, R. (1997). “Sitations: an exploratory study.” Cybermetrics 1(1). http://www.cindoc.csic.es/cybermetrics/articles/v1i1p1.html . Rousseau, Ronald (1999). “Daily time series of common single word searches in AltaVista and NorthernLight,” Cybermetrics 2/3, Paper 2 at http://www.cindoc.csic.es/cybermetrics/articles/v2i1p2.html . Sahal, Devendra (1981). “Alternativ e Conceptions of Technology,” Research Policy 10, 2-24. Scharnhorst, Andrea (1998). “Citation-Networks , Science Landscapes and Evoluti onary Strategies.” Scientometrics 43(1), 95-106. Scharnhorst, Andrea (2001). “Cons tructing Knowledge Landscapes within the Framework of Geometrically Oriented Evolutionary Theories.” In: Integrative Systems Approaches to Natural and Social Dynamics . M. Matthies, H. Malchow & J. Kriz (Eds.). Berlin: Springer Verlag, pp. 505-515. Schmoch, Uwe (1997). “Indicators and the Relations between Science and Technol ogy,” Scientometrics 38(1), 103-116. Science and Engineering Indicators 2000 (2000). National Science Board, Arlington, VA: National Science Foundation. Shannon, C. E. (1948). “A Mathematical Theory of Communication.” Bell System Technical Journal, 27, 379-423 and 623-356. Silverberg, Gerald (1984) “Embodied Technical Progress in a Dynam ic Economic Model: The Self-Organization Paradigm. In : R.M. Goodwin et al. (Eds) Non-linear Models of Fluctuating Growth. Berlin, etc.: Springer. Simon, H. A. (1969). The Sciences of the Artificial . Cambridge, MA/London: MIT Press. Simon, Herbert A. (1973). “The Organization of Complex System s.” In: Pattee, H. H. (Ed.), Hierarchy Theory: The Challe nge of Complex System s, New York: George Braziller Inc., pp. 1-27. Skolnikoff, Eugene B. (1993) The Elusive Transformation. Princeton, NJ: Princeton University Press. Snyder, H. and H. Rosenbaum (1999). “Can sear ch engines be used for web-link analysis? A critical review.” Journal of Documentation 55(4): 375-384. 48 49 Spiegel-Rösing, I. (1973). Wissenschaftsentwicklung und Wissensschaftssteuerung . Frankfurt a.M.: Athenaeum Verlag). Studer, K. E., and D. E. Chubin (1980). The Cancer Mission. Social Contexts of Biomedical Research . Beverly Hills, etc.: Sage. Theil, H. (1972). Statistical Decomposition Analysis . Am sterdam/ London: North-Holland. Thelwall, Mike (2001) “Extracting macr oscopic information from web links.” Journal of the American Society for Information Science and Technology , 52(13), 1157-1168. Thomas, O. and P. Willet (2000). “W ebometric an alysis of departments of librarianship and information science.” Journal of Information Science 26(6): 421-428. Towards a European Research Area - Key Fi gures 2001 - Special edition: Indicators for benchmarking of national research policies (2001). Luxembourg: Office for Publications of the European Communities. Van den Besselaar, P. (2000). “Communicati on between Science and Technology Studies journals.” Scientometrics, 47, 169-193. Van den Besselaar, P. (2001). “The cognitive and the social structure of science & technology studies.” Scientometrics, 51, 441-460. Van den Daele, Wolfgang, Wolfgang Krohn, and Peter Weingart (Eds.) (1977). Geplante Forschung: Vergleichende Studien über de n Einfluss politischer Programme auf die Wissenschaftsentwicklung. Frankfurt a.M.: Suhrkamp. Von Hippel, E. (1988). The Sources of Innovation . Oxford: Oxford University Press. Wagner, Caroline S. (2002). International linkage s: is collaboration creating a new dynamic for knowledge creation in science? Manuscript, Am sterdam School of Communications Research, University of Amsterdam. Wagner, Caroline S. & Loet Leydesdorff (2003). Mapping Global Science Using International Co-authorships: A Comparison of 1990 and 2000. Paper presented at the IXth Conference of the International Society of Scientometrics and Informetrics ISSI , Beijing, 25-29 August 2003. Wagner-Döbler, Roland & Jan Berg (1993). Mathematische Logik von 1847 bis zur Gegenwart. Berlin: de Gruyter Weingart, Peter (1997). “From “Finalization” to “Mode 2”: old wine in new bottles?” Social Science Information 36, 591-613. Weingart, Peter & Matthias Winterhagen (1984). Die Vermessung der Forschung . Frankfurt a.M.: Campus. Whitley, R. D. (1984). The Intellectual and Social Organizatio n of the Sciences . Oxford: Oxford University Press. Williamson, O. E. (1975). Markets and Hierarchies: Analysis and Antitrust Implications. A study in the Economics of Internal Organization. New York: Free Press. Woolgar, Steve, and Malcolm Ashmore (Eds.) (1988). Knowledge and Reflexivity: New frontiers in the Sociology of Knowledge . London: Sage. Wouters, Paul (1999). The Citation Culture . Ph. D. Thesis, University of Amsterdam. Wouters, Paul, Jan Annerste dt, and Loet Leydesdo rff, The European Guid e of Science, Technology, and Innovation Studies . Brussels: European Commission, DG XII, 1999. . Zelman, A. (2002). Mediated Communication and the Evolving Science System: M apping the Network Architecture of Knowledge Production. Ph.D. Thesis, University of Amsterdam, Amsterdam. Ziman J (1994). Prometheus Bound. Science in a dynamic steady state . Cam bridge: Cambridge University Press.