the determinants of national innovative...

Research Policy 31 (2002) 899–933

The determinants of national innovative capacity

Jeffrey L. Furmana,∗, Michael E. Porterb, Scott Sternca Boston University, School of Management, 595 Commonwealth Avenue, Room # 639a, Boston, MA 02215, USA

b Harvard Business School, Soldiers Field Road, Boston, MA 02163, USAc Kellogg Graduate School of Management, Leverone Hall, 2001 Sheridan Road, Evanston, IL 60208-2013, USA

Received 25 October 1999; received in revised form 14 June 2001; accepted 20 July 2001

Abstract

Motivated by differences in innovation intensity across advanced economies, this paper presents an empirical examinationof the determinants of country-level production of international patents. We introduce a novel framework based on the conceptof national innovative capacity. National innovative capacity is the ability of a country to produce and commercialize a flow ofinnovative technology over the long term. National innovative capacity depends on the strength of a nation’s common innova-tion infrastructure (cross-cutting factors which contribute broadly to innovativeness throughout the economy), the environmentfor innovation in a nation’s industrial clusters, and the strength of linkages between these two. We use this framework to guidean empirical exploration into the determinants of country-level differences in innovation intensity, examining the relationshipbetween international patenting (patenting by foreign countries in United States) and variables associated with the nationalinnovative capacity framework. While there are important measurement issues arising from the use of patent data, the resultssuggest that the production function for international patents is well-characterized by a small but nuanced set of observablefactors. We find that while a great deal of variation across countries is due to differences in the level of inputs devoted toinnovation (R&D manpower and spending), an extremely important role is played by factors associated with differences inR&D productivity (policy choices such as the extent of IP protection and openness to international trade, theshareof researchperformed by the academic sector and funded by the private sector, the degree of technological specialization, and eachindividual country’s knowledge “stock”). Further, national innovative capacity influences downstream commercialization,such as achieving a high market share of high-technology export markets. Finally, there has been convergence among OECDcountries in terms of the estimated level of innovative capacity over the past quarter century.Journal of Economic Literatureclassification: technological change (O3); technological change: choices and consequences (O33); economic growth andaggregate productivity: comparative studies (O57). © 2002 Published by Elsevier Science B.V.

Keywords:National innovative capacity; Endogenous technological change; National innovation systems; Patents; R&D productivity

1. Introduction

While R&D activity takes place in many coun-tries, the development and commercialization of“new-to-the-world” technologies has been concen-trated historically in relatively few countries. For

∗ Corresponding author.E-mail address:[email protected] (J.L. Furman).

example, during the 1970s and the early 1980s, twocountries, United States and Switzerland maintained aper capita “international” patenting rate well in excessof all other advanced economies. The variation amongadvanced economies in their ability to innovate atthe global frontier raises an empirical puzzle: if in-ventors draw on technological and scientific insightsfrom throughout the world, why does the intensity ofinnovation depend on location?

0048-7333/02/$ – see front matter © 2002 Published by Elsevier Science B.V.PII: S0048-7333(01)00152-4

900 J.L. Furman et al. / Research Policy 31 (2002) 899–933

This question is important for at least two rea-sons. First, though there is substantial agreement thattechnological innovation plays a central role in theprocess of long-run economic growth, there is debateabout the underlying drivers of the innovation pro-cess itself. International variation in the intensity ofinnovation presents an opportunity to examine vari-ous influences on the pace of technological change.Second, understanding international differences in theintensity of innovation informs public policy. Whilemost studies of innovation are set in agiven publicpolicy environment (Griliches, 1994, 1998), policyanalysis requires an evaluation of how innovationvaries with country-level policy differences.

This paper evaluates the sources of differencesamong countries in the production of visible innova-tive output. To do so, we introduce a novel frameworkbased on the concept ofnational innovative capac-ity. National innovative capacity is the ability of acountry—as both a political and economic entity—to produce and commercialize a flow of new-to-the-world technologies over the long term. Nationalinnovative capacity is not the realized level of inno-vative output per se but reflects more fundamentaldeterminants of the innovation process. Differences innational innovative capacity reflect variation in botheconomic geography (e.g. the level of spillovers be-tween local firms) as well as cross-country differencesin innovation policy (e.g. the level of public supportfor basic research or legal protection for intellectualproperty (IP)).

The national innovative capacity framework drawson three distinct areas of prior research: ideas-drivenendogenous growth theory (Romer, 1990), thecluster-based theory of national industrial competitiveadvantage (Porter, 1990), and research on nationalinnovation systems (Nelson, 1993).1 Each of theseperspectives identifies country-specific factors thatdetermine the flow of innovation. These theories sharea number of common analytical elements, but differ

1 While our framework is organized according to these threespecific formulations, we incorporate insights from related studiesin each research stream, including, among others, Jones (1995) andKortum (1997) in the ideas-driven growth literature, Rosenberg(1963), and Carlsson and Stankiewicz (1991) on the relationshipbetween innovation and industrial clusters and Mowery and Nelson(1999) on the linkages between industrial clusters and the nationalinnovation system.

with respect to their levels of abstraction and thefactors they emphasize. Whereas endogenous growththeory operates at a high level of abstraction, focusingon the economy-wide “knowledge stock” and the sizeof the R&D labor pool, the other two perspectivesemphasize more nuanced determinants. For example,Porter highlights the microeconomic underpinningsof innovation in national industrial clusters (includingthe interaction between input supply and local demandconditions, the presence and orientation of related andsupporting industries, and the nature and intensity oflocal rivalry), while the national innovation systemsliterature emphasizes the role of the overall nationalpolicy environment (e.g. IP or trade policy), highereducation, and country-specific institutions (e.g. thefunding approaches of specific agencies).

Our framework builds on these perspectives, char-acterizing national innovative capacity as the resultof three building blocks. First, national innovativecapacity depends on the presence of a strong com-mon innovation infrastructure: cross-cutting factorscontributing to innovativeness throughout the econ-omy. Among other things, the common innovationinfrastructure includes a country’s overall science andtechnology policy environment, the mechanisms inplace for supporting basic research and higher edu-cation, and the cumulative “stock” of technologicalknowledge upon which new ideas are developed andcommercialized. The common innovation infrastruc-ture therefore includes several of the elements high-lighted by the national innovation systems perspectiveand ideas-driven growth theory. Second, a country’sinnovative capacity depends on the more specificinnovation environments present in a country’s in-dustrial clusters. As emphasized by Porter (1990),whether firms invest and compete on the basis ofnew-to-the-world innovation depends on the microe-conomic environment in which they compete, whichwill vary in different fields. Third, national innova-tive capacity depends on the strength of the linkagesbetween the common innovation infrastructure andspecific clusters. For example, a given common in-novation infrastructure results in a more productiveflow of innovative output when there are mechanismsor institutions, such as a vibrant domestic universitysystem or established funding sources for new ven-tures, which encourage the commercialization of newtechnologies in particular clusters.

J.L. Furman et al. / Research Policy 31 (2002) 899–933 901

We use this framework to guide our empiricalevaluation of the sources of cross-country differencesin the production of innovative output. We do so byestimating the relationship between the production of“international” patents and observable measures ofnational innovative capacity. Our use of patent data toevaluate the rate of technological innovation is subjectto several important (and well-known) limitations, in-cluding differences in the propensity to patent acrossdifferent time periods, geographic regions, and tech-nological areas. We attempt to address these issuesby (a) using international patents; (b) establishingthe robustness of our results to controls through theuse of year and country dummies; and (c) carefullyinterpreting our findings in light of the potential formeasurement error.2 Also, since some elements ofnational innovative capacity (such as the environmentfor innovation in specific clusters) cannot be directlyobserved at the aggregate level, we employ measuresreflecting more aggregate outcomes associated withthe presence of these drivers.

Our results suggest that the production functionfor international patents is well-characterized by asmall number of observable factors that describe acountry’s national innovative capacity. We distinguishbetween scale-based differences across countries (aris-ing from differences in population or the level of in-puts devoted to innovation) and productivity-baseddifferences (i.e. differences in innovative output perunit of effort expended on the innovation process).While scale-related measures, such as total popula-tion, the size of the R&D workforce, or R&D spend-ing have important explanatory power, more nuancedfactors separately impact country-level R&D produc-tivity. We find robust and quantitatively important ev-idence that R&D productivity varies with aggregatepolicy choices such as the extent of IP protection and

2 In using international patenting data to understand the sourcesand consequences of innovation, this paper builds on Evenson(1984), Dosi et al. (1990), and recent work by Eaton and Kortum(1996, 1999). We extend these prior analyses by linking our resultsmore closely to a range of theories about the determinants of na-tional innovative capacity and by exploring a relatively long panelwhich allows us to incorporate both cross-sectional and time-seriesvariation. As well, we supplement the patent analysis by exam-ining alternative indicators of new-to-the-world innovative output,including scientific articles and export shares in “high-technology”industry segments.

openness to international trade, the shares of R&Dperformedby the academic sector andfundedby theprivate sector, the degree of specialization by technol-ogy area (a proxy for cluster specialization) and eachindividual country’s knowledge stock. We also findthat the estimated level of national innovative capac-ity affects total factor productivity (TFP) growth anda nation’s share of high-technology exports.

Our results provide evidence on the sources of dif-ferences in innovation intensity and R&D productivityacross countries and over time. We find that there hasbeen substantialconvergencein the level of per capitanational innovative capacity across the OECD sincethe mid-1970s. During the 1970s and early 1980s, thepredicted level of per capita international patenting byUnited States and Switzerland substantially exceededthat of other OECD members. Since that time, sev-eral countries (most notably Japan, some Scandina-vian countries, and Germany) have achieved levels ofpredicted per capita international patenting similar tothat of United States and Switzerland. Interestingly,there are exceptions to the convergence pattern; forexample, UK and France have shown little change intheir measured level of national innovative capacityover the past quarter century.

The paper proceeds by motivating and developingour theoretical framework, outlining the relationshipbetween “visible” innovative output and the elementsof national innovative capacity, describing data andmethods, and discussing our empirical results.

2. Theories of new-to-the-world innovationproduction

The national innovative capacity frameworkseeks to integrate three perspectives regarding thesources of innovation: ideas-driven growth theory,microeconomics-based models of national competi-tive advantage and industrial clusters, and research onnational innovation systems. While these perspectivescontain common elements, each highlights distinctdrivers of the innovation process at the national level.

Ideas-driven growth theory, the most abstract con-ceptualization, focuses at an aggregate level, empha-sizing the quantifiable relationships among a smallset of factors that determine the flow of new ideasin an economy. While the centrality of technological


innovation in economic growth has been appreciatedsince the seminal contributions of Solow (1956) andAmbramovitz (1956), it was only in the late 1980sthat technological change was treated endogenously.The Romer (1990) growth model articulates the eco-nomic foundations for a sustainable rate of technolog-ical progress (A) by introducing an ideas sector for theeconomy, which operates according the national ideasproduction function:

At = δHλA,tA

φt (1)

According to this structure, the rate of new ideasproduction is a function of the number of ideas work-ers (HA) and the stock of ideas available to theseresearchers (At ), making the rate of technologicalchange endogenous in two distinct ways. First, theshare of the economy devoted to the ideas sectoris a function of the R&D labor market (which de-terminesHA); allocation of resources to the ideassector depends on R&D productivity and the privateeconomic return to new ideas. Second, the produc-tivity of new ideas generation is sensitive to thestock of ideas discovered in the past. Whenφ > 0,prior research increases current R&D productivity(the so-called “standing on shoulders” effect); whenφ < 0, prior research has discovered the ideas whichare easiest to find, making new ideas discovery moredifficult (the “fishing out” hypothesis). Though thereis a sharp debate over the precise value of these pa-rameters (Jones, 1995; Porter and Stern, 2001)3 aswell as the precise form and equilibrium logic of themodel linking “ideas” production to economy-widelong-term productivity growth (Grossman and Help-man, 1991; Kortum, 1997; Silverberg et al., 1988),there is relatively broad agreement that these factorsare, indeed, crucial in explaining the realized level ofeconomy-wide innovation.

3 In Romer’s model of sustainable long-term growth from newideas,φ = λ = 1, implying that a given percentage increase in thestock of ideas results in a proportional increase in the productivityof the ideas sector. Under this assumption, the growth rate in ideasis a function of the level of effort devoted to ideas production(A/A = δHA), ensuring a sustainable rate of productivity growth.In contrast, Jones (1995) suggests thatφ andλ may be less than1, with the potential consequence of eliminating the possibilityof sustainable long-term growth. However, in a companion paper,we suggest that, at least for explaining the dynamic process ofproducing visible innovation (i.e. patents), the hypothesis thatφ =1 cannot be rejected (Porter and Stern, 2001).

Whereas ideas-driven growth theory focuses almostexclusively on this important but limited set of factors,a number of authors have emphasized the importanceof the microeconomic environment in mediating therelationship between competition, innovation, andrealized productivity growth. Building on importantstudies such as Rosenberg (1963), which identifiesinterdependencies between aspects of the microeco-nomic environment and the realized rate of technolog-ical innovation and economic growth, Porter (1990)developed a framework enumerating the characteris-tics of the environment in a nation’s industrial clustersthat shape the rate of private sector innovation, and ap-plied it in a sample of 10 countries over the post WorldWar II period. As several researchers have emphasized,it is important to recognize the dynamics of innovationwithin clusters, and particularly the role of dynamic in-teractions between clusters and specific institutions—from universities to public institutes—within givengeographic areas (Porter, 1990, 1998; Niosi, 1991;Carlsson and Stankiewicz, 1991; Audretsch andStephan, 1996; Mowery and Nelson, 1999).

The Porter framework encapsulates these forcesby identifying four key drivers (see Fig. 1). The firstis the availability of high-quality and specializedinnovation inputs. For example, while the overallavailability of trained scientists and engineers (empha-sized in ideas-driven growth theory) is important foreconomy-wide innovation potential, cluster-specificR&D productivity also depends on the availability ofR&D personnel who are specialized in cluster-relateddisciplines. A second determinant is the extent towhich the local competitive context is both intenseand rewards successful innovators. This depends ongeneral innovation incentives such as IP protection aswell as cluster-specific incentives such as regulationsaffecting particular products, consistent pressure fromintense local rivalry, and openness to internationalcompetition in the cluster (Sakakibara and Porter,2000). A third determinant of cluster-level innovationis the nature of domestic demand for cluster produc-ers and services. Innovation is stimulated by localdemand for advanced goods and the presence of asophisticated, quality-sensitive local customer base.Demanding customers encourage domestic firms tooffer best-in-the-world technologies, and raise the in-centives to pursue innovations that are globally novel.The final element in this framework is the availability,


Fig. 1. The innovation orientation of national industry clusters.

density and interconnectedness of vertically and hor-izontally related industries. These generate positiveexternalities both from knowledge spillovers, transac-tional efficiencies, and cluster-level scale economies,which are enhanced when clusters are concentratedgeographically. Overall, this framework suggests thatthe level of realized innovation in an economy de-pends upon the degree to which private R&D is fueledby innovation-based domestic competition.

The national innovation systems approach focuseson a textured description of the organization and pat-terns of activity that contribute to innovative behaviorin specific countries, and identifies those institutionsand actors who play a decisive role in particularindustries, emphasizing the diversity in national ap-proaches to innovation (see Nelson, 1993, for themost comprehensive account in this literature, as wellas Dosi, 1988, and Edquist, 1997).4 While both theideas-driven growth models and theories of national

4 This perspective is first articulated in the papers by Nelson(1988), Lundvall (1988) and Freeman (1988) in part five of Dosiet al. (1988).

industrial competitive advantage incorporate the roleof public policies in shaping the rate of innovation (atleast to some degree), the national innovation systemsliterature emphasizes the active role played by govern-ment policy and specific institutions. Particular institu-tional and policy choices highlighted by this literatureinclude the nature of the university system (Nelson andRosenberg, 1994), the extent of intellectual policy pro-tection (Merges and Nelson, 1990), the historical evo-lution of the organization of industrial R&D (Mowery,1984) and the division of labor between private indus-try, universities and government in R&D performanceand funding (Mowery and Rosenberg, 1998).5 Fig. 2highlights some of the key components and linkages

5 More generally, this literature builds on insights about thehistorical relationship between the resource endowments andgeographic structure of United States and the evolution of itsinstitutions and industries relative to that of UK and the rest ofEurope (Rosenberg, 1969, 1972; Nelson and Wright, 1992; Romer,1996). Recent research in this literature particularly emphasizesthe relationship between technological change, market structure,and institutions (see, especially, Mowery and Nelson, 1999).


Fig. 2. A comparison of some important elements of the national innovation systems of United States and Germany.

in the national innovation systems of United Statesand Germany, as described in Nelson (1993).

These perspectives share a number of insightsabout the innovation process. For example, all threeagree on the centrality of R&D manpower and the

need for a deep local technology base. Withoutskilled scientists and engineers operating in an en-vironment with access to cutting-edge technology, itis unlikely that a country will produce an apprecia-ble amount of new-to-the-world innovative output.


Beyond these common elements, Porter highlightsthe way the flow of innovation is shaped by special-ized inputs and knowledge, demand-side pressures,competitive dynamics, and externalities across re-lated firms and industries; in contrast, the nationalinnovation systems literature stresses the role playedby a nation’s scientific and innovation-oriented in-stitutions and policies. Whereas idea-driven growthand cluster theory focus on the economic impactof geography (i.e. localized spillovers), the nationalinnovation systems literature focuses more on thepolitical implications of geography (i.e. the impactof national policies and institutions). While all threeperspectives acknowledge the importance of bothpolitical and economic factors, prior work has not as-sessed how they interact in shaping the realized rateof innovation at the economy-wide level. This paperaims to contribute at this level, with an emphasison a connection between underlying microeconomicforces and models of endogenous technical change,in part responding to the call of Patel and Pavitt(1994) for analysis that articulates “the essentialproperties and determinants of national systems ofinnovation”.

3. National innovative capacity

National innovative capacity is defined as country’spotential—as both an economic and political entity—to produce a stream of commercially relevant inno-vations. Innovative capacity depends in part on theoverall technological sophistication of an economyand its labor force, but also on an array of investmentsand policy choices by both government and the pri-vate sector. Innovative capacity is related to but dis-tinct from scientific and technical advances per se,which do not necessarily involve the economic appli-cation of new technology. Innovative capacity is alsodistinct from current national industrial competitiveadvantage or productivity, which results from manyfactors (such as the skills of the local workforce andthe quality of physical infrastructure) that go beyondthose important to the development and commercial-ization of new technologies. Differences in nationalinnovative capacity reflect variation in both economicgeography (e.g. the impact of knowledge and inno-vation spillovers among proximate firms) and innova-

tion policy (e.g. the level of public support for basicresearch or protection for IP).

Technological opportunities are most likely to beexploited first in those countries whose environ-ments are most conducive to the development ofnew-to-the-world technology. Given the sustainedinvestment in innovative capacity in United Statesin the two decades after World War II, for example,it is not surprising that many of the most importantscientific and technological breakthroughs of thatera—including the transistor, the laser, electroniccomputing, and gene splicing—were developed andinitially exploited by American universities and com-panies. Even though serendipitous technical or sci-entific advances may occur in countries with lowerlevels of innovative capacity, the development andcommercialization of such advances is likely to takeplace in those countries with higher levels of innova-tive capacity. For example, while the British chemistPerkin was responsible for the initial discovery of ani-line dye, the chemical sector emerged most stronglyin Germany, at least in part due to Germany’s strongeruniversity–industry relationships and wider avail-ability of capital for technology-intensive ventures(Murmann, 1998; Arora et al., 1998).

We divide determinants of national innovative ca-pacity into three categories: the common pool ofinstitutions, resource commitments, and policies thatsupport innovation across the economy; the particu-lar innovation environment in the nation’s industrialclusters; and the linkages between them (see Fig. 3).The overall innovative performance of an economyresults from the interplay among all three.6

3.1. Common innovation infrastructure

Some of the most important investments and policychoices that support innovative activity have broad im-pact throughout an economy—these are the common

6 Our framework focuses on clusters (e.g. information technol-ogy) rather than individual industries (e.g. printers) as spilloversacross industrial segments connect both the competitiveness andrate of innovation towards this more aggregate unit of analysis.In addition, while the current discussion focuses at the countrylevel, we could also conduct our analysis at the regional level,with potentially important insights, as many of the most dynamicindustrial clusters seem to be quite local in nature (Porter, 1990,1998).


Fig. 3. National innovative capacity framework.

innovation infrastructure. The left-hand side of Fig. 3portrays some of its elements. Endogenous growththeory highlights two important determinations of theproduction of ideas in an economy: an economy’s ag-gregate level of technological sophistication (A) andthe size of the available pool of scientists and engi-neers who may be dedicated to the production of newtechnologies (HA). We expand on this conception toinclude other cross-cutting factors that impact inno-vative activity denoted byXINF, including the extentto which an economy invests in higher education andpublic policy choices such as patent and copyrightlaws, the extent of R&D tax credits, the nature of an-titrust laws, the rate of taxation of capital gains, andthe openness of the economy to international compe-tition. 7 It is important to note that the common in-novation infrastructure incorporates both the overallscale of innovation inputs within an economy (HA, thesize of R&D employment and spending) as well aseconomy-wide sources of R&D productivity (A andXINF).

7 While some of these policies have stronger impact on someindustries than others, e.g. intellectual property protection is es-pecially salient for pharmaceutical innovation, these policies gen-erally support innovation across a wide range of sectors in aneconomy.

3.2. The cluster-specific innovation environment

While the common innovation infrastructure setsthe general context for innovation in an economy, it isultimately firms, influenced by their microeconomicenvironment, that develop and commercialize innova-tion. Thus, national innovative capacity depends uponthe microeconomic environment present in a nation’sindustrial clusters (as highlighted in Fig. 1 and the di-amonds on the right-hand side of Fig. 3). A variety ofcluster-specific circumstances, investments, and poli-cies impact the extent to which a country’s industrialclusters compete on the basis of technological inno-vation (Porter, 1990). Innovation in particular pairs ofclusters may also be complementary to one another,both due to knowledge spillovers and other interrela-tionships (represented by lines connecting selected di-amonds in Fig. 3). Just as a strong cluster innovationenvironment can amplify the strengths of the com-mon innovation infrastructure, a weak one can stiflethem. For example, despite a strong infrastructure sup-porting scientific education and technical training inFrance, national regulatory policies towards pharma-ceuticals have limited innovation in the French phar-maceutical cluster through 1970s and 1980s (Thomas,1994).


3.3. The quality of linkages

The relationship between the common innovationinfrastructure and industrial clusters is reciprocal: fora given cluster innovation environment, innovativeoutput will tend to increase with the strength of thecommon innovation infrastructure (and vice versa).For example, while stringent local environmentalregulations and well-developed supporting industriesmay encourage innovation-oriented competition inthe environmental technologies cluster in Sweden,the ability of the Swedish cluster to generate andcommercialize environmental innovations also de-pends in part on the overall availability of trainedscientists and engineers, access to basic research,and overall policies which reward the developmentand commercialization of new technologies in theeconomy. The strength of linkages influences theextent to which the potential for innovation inducedby the common innovation infrastructure is trans-lated into specific innovative outputs in a nation’sindustrial clusters, thus shaping the realized rate ofnational R&D productivity. Linkages can be facili-tated by various types of institutions, ranging fromuniversities to cluster trade associations to informalalumni networks. In the absence of strong linkingmechanisms, upstream scientific and technical activ-ity may spill over to other countries more quicklythan opportunities can be exploited by domestic in-dustries. Germany took advantage of British discov-eries in chemistry, for example, while three Japanesefirms successfully commercialized VCR technologyinitially developed in United States (Rosenbloomand Cusumano, 1987). While the roles played bysome particular linking mechanisms have been stud-ied (from the Fraunhofer Institutes in Germany toMITI in Japan to the use of Cooperative Researchand Development Associations (CRADAs) in UnitedStates), there have been few attempts to evaluate theimpact of such institutions on international R&Dproductivity.

4. Modeling national innovative capacity

The national innovative capacity framework in-corporates a wide set of both the economic and pol-icy influences of national boundaries in explaining

cross-country differences in the intensity of inno-vation. We therefore integrate prior research thatfocuses on the impact of geography on knowledgespillovers and differential access to human cap-ital, 8 as well as the work that emphasizes howregional differences may be driven by differen-tial public policies and institutions (Nelson, 1993;Ziegler, 1997). Our framework embodies the pre-dictions of ideas-driven growth models while alsoincluding more nuanced factors, which have notbeen incorporated into formal models but may beimportant contributors to innovative output (suchas those related to industrial organization, the com-position of funding, and public policy). Finally,the framework highlights the potential importanceof the composition of research funding and per-formance and the degree of technological special-ization by a country’s R&D sector. For example,while public R&D spending adds to the innova-tive process by reinforcing the common innova-tion infrastructure, private R&D spending and thespecialization of a country’s technological out-puts also reflects the nation’s cluster innovationenvironment.

To estimate the relationship between the pro-duction of international patents and observablecontributors to national innovative capacity, weadopt the ideas production function of endoge-nous growth theory as a baseline (recall (1)). Thenational innovative capacity framework suggeststhat a broader set of influences determine innova-tive performance; hence, our framework suggestsa production function for new-to-the-world tech-nologies slightly more general than the Romerformulation:

Aj,t = δj,t (XINFj,t , Y CLUS

j,t , ZLINKj,t )HAλ

j,t Aφj,t (2)

As before, Aj,t is the flow of new-to-the-worldtechnologies from countryj in year t, HA

j,t the to-tal level of capital and labor resources devoted tothe ideas sector of the economy, andAj ,t is the to-tal stock of knowledge held by an economy at agiven point in time to drive future ideas production.

8 In addition to Porter (1990), see Jaffe et al. (1993), Krugman(1991), Saxenian (1994), Zucker, Darby and Brewer (1998),Audretsch and Stephan (1996), Glaeser et al. (1992), Bostic et al.(1996), Park (1995), Coe and Helpman (1995), and Keller (2000).


To these are addedXINF, which refers to the levelof cross-cutting resource commitments and policychoices that constitute the common innovation in-frastructure,YCLUS, which refers to the particularenvironments for innovation in a countries’ indus-trial clusters, andZLINK , which captures the strengthof linkages between the common infrastructure andthe nation’s industrial clusters. Under Eq. (2), weassume that the various elements of national innova-tive capacity are complementary in the sense that themarginal boost to ideas production from increasingone factor is increasing in the level of all of the otherfactors.

Deriving an empirical model from Eq. (2) re-quires addressing three issues: the source of statisticalidentification, the precise specification of the inno-vation output production function, and the sourceof the econometric error. Our choices with respectto each of these issues reflects our overarchinggoal of letting the data speak for itself as much aspossible.

First, the parameters associated with Eq. (2) areestimated using a panel dataset of 17 OECD countriesover 20 years. These estimates can therefore dependon cross-sectional variation, time-series variation, orboth. Choosing among the two potential sources ofidentification depends on the production relationshipsto be highlighted in our analysis. While compar-isons across countries can easily lead to problems ofunobserved heterogeneity, cross-sectional variationprovides the direct inter-country comparisons thatcan reveal the importance of specific determinantsof national innovative capacity. Time-series variationmay be subject to its own sources of endogeneity(e.g. shifts in a country’s fundamentals may reflectidiosyncratic circumstances in its environment), yettime-series variation provides insight into how acountry’s choices manifest themselves in terms ofobserved innovative output.

Recognizing the benefits (and pitfalls) associatedwith each identification strategy, our analysis explic-itly compares how estimates vary depending on thesource of identification. In most (but not all) of ouranalysis, we include either year dummies or a timetrend in order to account for the evolving differencesacross years in the overall level of innovative output.Much of our analysis also includes either country dum-mies or measures which control for aggregate differ-

ences in technological sophistication (e.g. as reflectedin GDP PER CAPITA).9 The analysis is organizedaround a log–log specification, except for qualitativevariables or variables expressed as a percentage. Theestimates thus have a natural interpretation in termsof elasticities, are less sensitive to outliers, and areconsistent with the majority of prior work in this area(Jones, 1998). Finally, with regard to the sources oferror, we assume that the observed difference from thepredicted value given by Eq. (2) (i.e. the disturbance)arises from an idiosyncratic country/year-specifictechnology shock unrelated to the fundamental deter-minants of national innovative capacity. Integratingthese choices and letting L denote the natural loga-rithm, our main specification takes the following form:

L Aj,t = δYEARYEARt + δCOUNTRYCj

+δINFL XINFj,t + δCLUSL Y CLUS

j,t

+δLINK L ZLINKj,t + λ L HA

j,t

+φL Aj,t + εj,t (3)

Conditional on a given level of R&D inputs (HA),variation in the production of innovation (A) reflectsR&D productivity differences across countries or time.For example, a positive coefficient on elements ofδINF, δCLUS or δLINK suggests that the productivityof R&D investment is increasing in the quality of thecommon innovation infrastructure, the innovation en-vironment in the nation’s industrial clusters, and thequality of linkages. AsAj,t , measured by the level ofinternational patenting, is only observed with delay,our empirical work imposes a 3-year lag between themeasures of innovative capacity and the observed re-alization of innovative output.

5. The data

Implementing Eq. (3) requires that each of the con-cepts underlying national innovative capacity be tied

9 By controlling for year and country effects in most of ouranalysis, we address some of the principal endogeneity and auto-correlation concerns. However, we have extensively checked thatthe results reported in Section 5 are robust to various forms ofautocorrelation (results available on request) and have investigatedthe potential for endogeneity more fully in related work (Porterand Stern, 2001).


to observable measures. We employ a novel dataset ofpatenting activity and its determinants in a sample of17 OECD countries from 1973 to 1996. Our results,then, describe the relationship between measured in-novative inputs and outputs in highly industrializedeconomies. Table 1 defines and provides sources forall variables; Table 2 reports the means and standarddeviations, and Appendix A lists all included coun-tries and reports pairwise correlations. We draw onseveral sources in constructing these data, includingUnited States patent and trademark office (USPTO),CHI research, the OECD basic science and technol-ogy statistics, NSF science and engineering indicators,the World Bank, the Penn world tables, and the IMSworld competitiveness report.10

5.1. The measurement of visible innovative output

Our analysis requires a consistent country-specificindicator of the level of commercially valuable in-novative output in a given year. We employ a vari-able based on the number of “international patents”(PATENTS), which is defined as the number of patentsgranted to inventors from a particular country otherthan United States by the USPTO in a given year. ForUnited States, PATENTS is equal to the number ofpatents granted to corporate or government establish-ments (this excludes individual inventors); to ensurethat this asymmetry between US and non-US patentsdoes not affect our results we include a US dummyvariable in our regressions.11

The pitfalls associated with equating patenting withthe level of innovative activity are widely recognized(Schmookler, 1966; Pavitt, 1982, 1988; Griliches,1984, 1990; Trajtenberg, 1990). As Griliches (1990,p. 1669) points out, “not all inventions are patentable,

10 To ensure comparability across countries and time, we sub-jected each measure to extensive analysis and cross-checking, in-cluding confirming that OECD data were consistent with com-parable data provided by individual national statistical agencies.When appropriate, we interpolated missing values for individualvariables. For example, most countries report educational expen-diture data only once every other year; our analysis employs theaverage of the years just preceding and following a missing year.Financial variables are in PPP-adjusted 1985 $US.11 We have also analyzed our results using the number of US

patents filed in at least one other international jurisdiction, andthere are no qualitative differences in any of our results (resultsavailable from the authors).

not all inventions are patented, and the inventionsthat are patented differ greatly in ‘quality’, in themagnitude of inventive output associated with them”.We address these limitations, in part, by constructingPATENTS to include only commercially significantinnovations at the world’s technical frontier and bycarefully testing the measure for robustness. First,since obtaining a “foreign” patent is a costly un-dertaking that is only worthwhile for organizationsanticipating a return in excess of these substantialcosts. Second, USPTO-granted “international” patent-ing (PATENTS) constitutes a measure of technologi-cally and economically significant innovations at theworld’s commercial technology frontier that shouldbe consistent across countries.12 Third, we are care-ful to accommodate the potential for differences inthe propensity to apply for patent protection acrosscountries and over time (as highlighted by Scherer,1983) by evaluating robustness of our results to year-and country-specific fixed effects.

Even with these checks, we recognize that that the“true” rate of technological innovation is unobservableand PATENTS is but an imperfect proxy for the levelof new-to-the-world innovation. With this in mind, weexplore the precision of the statistical relationship be-tween PATENTS and our measured drivers of nationalinnovative capacity (i.e. how well do the small num-ber of factors employed in our analysis explain this“noisy” measure of the innovation process?) and ourinterpretations take into account likely sources of biasarising from specific variables (i.e. are there reasonsthat our findings are driven by the PATENTS measurerather the level of innovation per se?). Ultimately, ourapproach is based on the assessment that patentingrates constitute “the only observable manifestationof inventive activity with a well-grounded claim foruniversality” (Trajtenberg, 1990, p. 183), a judgmentreflected in prior work employing international patentdata (Evenson, 1984; Dosi et al., 1990; Cockburn andHenderson, 1994; Eaton and Kortum, 1996, 1999;Kortum, 1997; Vertova, 1999).

Across the sample, the average number ofPATENTS produced by a country in a given year is3986 (with S.D. of 8220). PATENTS has increasedover time, reaching a peak in the final year of the

12 See Eaton and Kortum (1996, 1999) for a thorough discussionof the economics of international patenting.

910J.L

.F

urm

an

et

al./R

ese

arch

Po

licy3

1(2

00

2)

89

9–

93

3

J.L.

Fu

rma

ne

ta

l./Re

sea

rchP

olicy

31

(20

02

)8

99

–9

33

911


Table 2Mean and S.D.

Variable N Mean S.D.

Innovative outputPATENTS 378 3986.23 8219.89PATENTS/MILLION POP 378 3.73 1.02

Quality of the common innovation infrastructureA GDP/POP 357 18.66 5.10A PATENT STOCK 357 38016.59 98252.46HA POP 357 42.4 57.1HA FTE S&E 353 226344.60 407124.50HA R&D $ 355 12859.86 27930.46XINF ED SHARE 351 3.08 1.20XINF IP 162 6.87 0.97XINF OPENNESS 216 7.00 1.10XINF ANTITRUST 162 5.75 1.09

Cluster-specific innovation environmentYCLUS PRIVATE R&D FUNDING 355 48.60 12.88YCLUS SPECIALIZATION 357 0.02 0.03

The quality of linkagesZLINK UNIV R&D PERFORMANCE 355 21.50 6.20ZLINK VENTURE CAPITAL 214 5.45 1.32

Contributing and related outcome factorsJOURNALS 378 17446.39 28621.21GDP 357 772.44 1161.64LABOR 321 18.10 25.60CAPITAL 306 550.00 795.00TFP 304 1.42 0.21MARKET SHARE 357 5.88% 6.85%

sample (when the average level of PATENTS is equalto 5444). Fig. 4 reports two country-level measuresof differences in the intensity of innovation acrosscountries. The first panel explores the most aggregaterelationship, PATENTS PER CAPITA (in terms ofpopulation, in millions), while the second panel pro-vides a first glance of R&D productivity, the level ofPATENTS per unit of effortdevotedto innovation, asmeasured by the number of full-time equivalent sci-entists and engineers (PATENTS PER FTE S&E).13

While the latter corresponds more closely to a tra-ditional productivity measure, the former provides agreater sense of total innovation output relative tototal inputs whichcouldbe devoted to innovation.

13 Several alternative measures of productivity could be con-structed (e.g. PATENTS per unit of R&D expenditure) and areavailable upon request.

Three facts stand out. First, countries differmarkedly in their production of international patentsper capita and per unit of effort devoted to innova-tion. Second, at the beginning of the sample, the onlycountry with a per capita patenting rate similar to thatof United States is Switzerland, and the only addi-tional country with a similar level of PATENTS PERFTE S&E is Sweden.14 Third, there is noticeablenarrowing over time in the gap between countries.This convergenceis most apparent for Japan, butalso is evident for most (though not all) other OECDcountries.

We also explore several alternative output mea-sures to PATENTS that are less comparable acrosscountries and likely to be less closely linked to the

14 Recall that US patenting level is determined by the number ofpatents issued to US inventors associated with an institution suchas a company, governmental body, or university.


Fig. 4. (a) International patents per million persons; and (b) International patents per 10,000 FTE S&E.

level of new-to-the-world innovative output. The levelof publication in scientific journals (JOURNALS),although a product of inventive effort, is more anupstream indicator of scientific exploration than ofcommercially significant innovation. We examine twomore downstream measures of the impact of nationalinnovative capacity: the realized market share of acountry in “high-technology” industries (MARKETSHARE) and a measure of TFP, defined as the levelof GDP controlling for the levels of CAPITAL andLABOR. Both MARKET SHARE and TFP shoulddepend on national innovative capacity over thelong-run. In the near term, however, both measureswill be affected by other determinants of overall

competitiveness, including microeconomic factorsand macroeconomic factors that affect impact compet-itiveness but are only indirectly related to the rate ofinnovation.15

15 Several measures of technological output are neither availablenor usefully comparable across countries or time. For example,the level of technology licensing revenues realized by a countrycaptures activity in some technology areas, but in practice isnot nearly broad-based enough to have a well-founded claim forcomparability. While measures such as copyrights and trademarksare direct indicators of innovative output in certain industries (e.g.software), the lack of comparability of these data across countriesand time limits their usefulness for our analysis.


5.2. Measuring national innovative capacity

To estimate a model consistent with our frame-work, we require measures of a nation’s commoninnovation infrastructure, the innovation environmentin its industrial clusters, and the nature of the linkagesbetween these elements. For the common innovationinfrastructure, a number of relatively direct measuresare available; however, direct measures of the clusterinnovation environment and the quality of linkages arenot available for international data. We address thischallenge by employing intermediate measures orproxies that capture important economic outcomesassociated with strength in these areas.

The common innovation infrastructure consistsbroadly of a country’s knowledge stock, the overalllevel of human and capital resources devoted to in-novative activity, and other broad-based policies andresource commitments supporting innovation. Ouranalysis explores two distinct measures for knowl-edge stock at a given point in time (Aj ,t ): GDP PERCAPITA and the sum of PATENTS from the start ofthe sample until the year of observation (PATENTSTOCK).16 Although both measure the overall stateof a country’s technological development, they dif-fer in important ways. GDP PER CAPITA capturesthe ability of a country to translate its knowledgestock into a realized state of economic development(and so yields an aggregate control for a country’stechnological sophistication). In contrast, PATENTSTOCK constitutes a more direct measure of thecountry-specific pool of new-to-the-world technology.

We measure the level of capital and labor resourcesdevoted to the ideas-producing sector using eachcountry’s number of full-time-equivalent scientistsand engineers (FTE S&E) and gross expenditures onR&D (R&D $). While individual R&D and engineer-ing efforts will tend to be specialized in particulartechnical and scientific areas, the outputs of R&D im-pact a variety of economic sectors via direct applica-tion or as a basis for future efforts (Rosenberg, 1963;Glaeser et al., 1992). Hence, we classify the overallsupply of innovation-oriented labor and capital as keyelements of the common innovation infrastructure.We also include population (POP) inHA, since the

16 See Porter and Stern (2001) for a derivation and thoroughdiscussion of differences between these two measures.

total size of a country indicates the scale of resources(workers) potentially available for innovative activity.

There is evidence for convergence in the level ofresources devoted to R&D activity (Fig. 5). FTE S&Eper capita has generally increased among OECD coun-tries and growth has been higher for those countrieswhose initial levels were lower. Growth in FTE S&Eper capita over the sample period has been particularlyhigh in Japan and the Scandinavian countries, whileUK and United States actually experienced small de-clines in FTE S&E per capita. R&D $ also displaysa similar though less pronounced pattern of conver-gence.

The common innovation infrastructure also encom-passes national policies and other resource commit-ments that broadly affect innovation incentives andR&D productivity throughout the economy. We in-clude the fraction of GDP spent on secondary andtertiary education (ED SHARE) as a measure of theintensity of human capital investment. A high levelof ED SHARE creates a base of highly skilled per-sonnel upon which firms and other institutions acrossthe economy can draw, both for formal R&D activi-ties as well as other innovation-related activities. Wealso measure some policy choices that particularly af-fect the environment for innovative activity includingthe relative strength of IP protection, the relative strin-gency of country’s antitrust policies (ANTITRUST),and the relative openness of a country to internationaltrade and competition (OPENNESS). Controlling forresources, we expect the level of each of these threepolicy variables (measured on a 1–10 Likert scale)17

to be positively related to the productivity of inter-national patenting.18 Given that these variables aredrawn from an imperfect survey instrument, however,each captures the underlying concept only with in-evitable noise.

17 IP, OPENNESS, and ANTITRUST are average (1–10) Likertscore variables from the IMD World Competitiveness Report, anannual survey in which leading executives rank their perceptions ofcountries’ circumstances along a variety of dimensions relevant tointernational competitiveness. These variables become available inthe late 1980s (between 1986 and 1989 depending on the variable).The analysis corrects for missing values by including a dummyvariable which is equal to one in years where these measures areunavailable.18 Note, as well, that each of these policy measures may increase

the level of resources devoted towards innovation; however, ouranalysis focuses on the productivity effects of these policies.


Fig. 5. FTE S&E as a percentage of population selected countries (A–J) and (N–Z).

While the common innovation infrastructure isquite amenable to measurement, the environment forinnovation in a nation’s industrial clusters is difficultto measure because of the subtlety of the conceptsinvolved as well as the lack of systematic interna-tional data. Indeed, the aggregate framework offeredhere clearly cannot provide a test of the nuanced anddistinct implications of the national industrial clus-ter framework. For example, providing evidence thatinnovation is driven by the interactions among com-plementary industry segments versus rivalry withinindustrial segments requires a level of detail which

cannot be addressed in the current context.19 In-stead, we use the detailed qualitative and quantitativeinsights arising from prior research on industrialclusters to develop intermediate measures that areconsistent with the outcomes of innovation-orientedcluster-level dynamics.

First, we employ the intensity of privately financedR&D activity in an economy to measure the collectiveimportance of innovation-based competition acrossclusters. Conditional on the overall level of R&D

19 See, however, Porter and Stern (2001).


Fig. 6. Percent of R&D expenditure funded by industry.

investment in an economy, the fraction of total R&Dspending conducted by the private sector (PRIVATER&D FUNDING) provides a useful summary indica-tor of the vitality of the environment for innovation innational industrial clusters that is comparable acrosscountries and available for an international sample.20

This measure does not, of course, explicitly portraythe subtle implications of industrial clusters, but doesreflect a broader theme arising from that perspec-tive: the productivity of innovative investment willdepend on whether private firms within the economyare choosing to direct resources towards that end.

20 Other measures, such as the average private R&D-sales ratio,may be more ideal for this purpose, but comparable cross-countrydata are not available. Also, we do not include higher-order termssince we are focusing on the first-order impact of these measures intheir observed range rather than calculating their predicted impactoutside of the observed range or solving for the “optimal” levelof such measures.

Fig. 6 demonstrates the variance in this measure.In 1990, for example, companies financed between40 and 60% of R&D in most countries (Japan andSwitzerland are outliers with over 70%). Interestingly,the traditionally social market economy of Finlandis near the top of the range of this measure; thehigh private sector role in R&D is directly related toFinland’s globally competitive and innovation-driventelecommunications cluster.

Second, since individual clusters will tend to beassociated with technologies from specific techno-logical areas, we calculate a measure of the degreeof technological focus by a country (SPECIALIZA-TION) as a proxy for the intensity of innovation-basedcompetition in a nation’s clusters. SPECIALIZA-TION is a “relative” concentration index based on thedegree to which a given country’s USPTO-grantedpatents are concentrated across the three (relativelybroad) technology classes (chemical, electronics, and


Fig. 7. Composition of USPTO patent grants by technological class, 1995.

mechanical). While our measure of specialization istoo coarse to identify specific clusters and the roleof the mix of clusters in shaping R&D productivity,SPECIALIZATION is designed as a noisy (but un-biased) measure capturing an important consequenceof cluster dynamics, the relative specialization ofnational economies in specific technologies fields.Fig. 7 demonstrates that countries differ in the sharesof their patents associated with each area. We usethis classification system to develop a measure ofspecialization appropriate for our context. Specif-ically, traditional measures of specialization, suchas the Herfindahl (which in this context would beequal to

∑i=ELEC,CHEM,MECH(PATENTSi,t /TOTAL

PATENTSt )2 = ∑

i s2i,t ), ignore two issues important

for cross-country comparisons: (a) technology classesdiffer in terms of their average share across all coun-tries (e.g. the mechanical class has the highest averageshare); and (b) some countries have only a small num-

ber of patents overall.21 We overcome these concernsby using a methodology developed by Ellison andGlaeser (1997; hereafter, E–G). While the E–G indexwas developed and applied for measuring the special-ization of industries across geographic regions (Elli-son and Glaeser, 1997), it has been applied in severalother contexts, including the measurement of the de-gree of specialization of research output (Lim, 2000).In the present context, the E–G formula adjusts thecountry observed shares for each technology class toaccount for (a) the average share for that technologygroup (across the sample); and (b) for the total num-ber of patents in each “country–year” observation.

21 When a country has only a very small number of patents in agiven year, it is possible to overstate the degree of specialization.In the extreme, if a country only produces a single patent, itspatent “portfolio” will necessarily be confined to only a singletechnology class.


Allowing xi to be the average share of patent classiacross all country–years, the E–G index is

SPECIALIZATIONi,j,t =(

PATENTSi,j,t

PATENTSi,j,t − 1

)(∑i (si,j,t − xi)

2

1 −∑ix

2i

− 1

PATENTSi,j,t

)

Essentially, this measure captures the degree to whichPATENTS are specialized in countryi in yeart acrosseach technology classj (note that its mean is just above0, reflecting the fact that the “average” country, by con-struction, has an average level of concentration). Forexample, consistent with Japan’s strength in the elec-tronics cluster (see Fig. 7), its patenting is concentratedin electronics and its SPECIALIZATION measure(in 1995) is equal to 0.125, more than 3 S.D. abovethe mean.22 Like PRIVATE R&D FUNDING, SPE-CIALIZATION is a noisy but potentially useful andunbiased measure of the underlying strength and envi-ronment for innovation in a nation’s industrial clusters.

The final element of our framework, the strengthof linkages between industrial clusters and the com-mon innovation infrastructure and vice versa, is simi-larly difficult to quantify directly. The mechanisms forscientific and technological transfer vary both acrosscountries and over time. Our analysis considers twomechanisms that qualitative research suggests are rel-atively consistent contributors to the strength of link-ages: the importance of the university sector in R&Dperformance, measured by the share of total R&D ac-counted for by universities (UNIV R&D PERFOR-MANCE), and the availability of venture-backed (VC)financing. University research tends to be more acces-sible to researchers in industry than government lab-oratory research, and universities provide a forum forthe exchange of ideas between different R&D com-munities. The unique role that universities play intraining future industrial researchers suggests anotherway in which common resources for innovation (i.e.S&E graduate students) are mobilized in a nation’sindustrial clusters. VC (measured as a 1–10 Likertscale variable by the IMD Competitiveness Report,1989–1999) captures the degree to which risk capi-tal is available to translate scientific and technological

22 The data used in these calculations are from TechnologicalAssessment and Forecast Reports (USPTO, 2000a,b,c), which alsodefine the patent classes included in the chemical, electrical, andmechanical categories.

outputs into domestic opportunities for further inno-vation and commercialization.

6. Empirical results

Our empirical results are presented in threeparts. First, we present the paper’s primary results,which evaluate the determinants of the productionof new-to-the-world technologies. Second, we ex-plore the downstream consequences of national in-novative capacity through its impact on TFP andhigh-technology MARKET SHARE. The final sectiondiscusses the patterns in national innovative capacityimplied by the estimates over the sample period.

6.1. Determinants of the production ofnew-to-the-world technologies

Tables 3–7 present a broad range of results re-garding the relationship between innovative output(PATENTSt+3) 23 and the drivers of national inno-vative capacity (Aj,t , H

Aj,t , X

INFj,t , Y CLUS

j,t , ZLINKj,t ) that

highlight the main relationships in the data as wellas the source of variation underlying particular find-ings. Overall, we find a robust and relatively preciserelationship between PATENTS and measures asso-ciated with each component of national innovativecapacity: the common innovation infrastructure, thecluster-specific innovation environment, and the link-ages between these two areas.

Table 3 evaluates a series of pairwise relationshipsbetween PATENTS and several measures of countrysize and the total level of R&D inputs: POP, GDP, FTES&E, and R&D $. Since these measures are highlycorrelated with each other (see Appendix A), it is use-ful to establish the unconditional relationship betweeneach and PATENTS. Individually, each measure ex-plains between two-thirds and 90% of the overall vari-ation in PATENTS. The coefficients range from 0.98

23 Recall that we evaluate the relationship of PATENTS in yeart +3 to the level of the contributors to innovative capacity in yeartand that our main results are robust to changes in this lag structure.


Table 3Simple regressions on scale-dependent measures

Dependent variable= ln(PATENTS)j,t+3

Populationonly (3.1)

FTE S&Eonly (3.2)

R&D expendituresonly (3.3)

GDP only (3.4) L PATENT STOCKand YEAR (3.5)

L POP 1.159 (0.042)L FTE S&E 1.168 (0.022)L R&D $ 0.982 (0.027)L GDP 1.297 (0.033)L PATENT STOCK 0.971 (0.011)YEAR −0.100 (0.003)

R2 0.682 0.892 0.892 0.811 0.958

to 1.30, suggesting only a modest departure from con-stant returns-to-scale. These results demonstrate thata large fraction of the variance in PATENTS can beexplained by single measures of a country’s size or ef-fort devoted to R&D. However, these models providelittle intuition regarding the forces that drive nationalinnovative output. The question remains whether morenuanced factors have a separate and quantitatively im-portant impact on national innovative output distinctfrom these scale-dependent measures.

Table 4 begins with a specification similar to theformal model of the national ideas production func-tion suggested by Romer (1990) and Jones (1995).In the first regression (4.1), we see that PATENTS isincreasing in two variables suggested by endogenousgrowth theory, L GDP and L FTE S&E. Interpretingthe coefficients as elasticities, (4.1) implies that, ce-teris paribus, a 10% increase in GDP is associatedwith slightly more than a 10% rise in internationalpatent output. As suggested by proponents of endoge-nous growth, a country’s existing level of technolog-ical sophistication and the level of inputs devoted toR&D play key roles in determining innovative output.After controlling for the level of GDP and FTE S&E,POP has a negative coefficient; the lower the impliedlevel of GDP PER CAPITA, the lower the flow ofnew-to-the-world innovation.

We include the remainder of the measures ofHA

andXINF in (4.2). This specification (along with (4.3)and (4.4)) also includes year-specific fixed effects tocontrol for variation arising from changes in the rateof patent grants per year. With the exception of AN-TITRUST, each of the measures of the common in-novation infrastructure affects international patenting

significantly, with the expected sign, and with an eco-nomically important magnitude (note also that theseregressors have only a modest impact on the coeffi-cients associated with L GDP). Perhaps more interest-ingly, the sum of the coefficients on L R&D $ and LFTE S&E in (4.2) is quite similar to the single coeffi-cient on L FTE S&E in (4.1). In other words, the to-tal impact of R&D inputs (in terms of aggregate laborand capital) devoted to innovation is similar whetherwe focus on a single variable (e.g. FTE S&E in (4.1))or include both measures (e.g. FTE S&E and R&D $in (4.2)).

After controlling for L R&D $ and L FTE S&E(R&D inputs), the remaining coefficients can be in-terpreted as economy-wide factors affecting nationalR&D productivity. The elements ofXINF are expressedas Likert scale measures (IP, OPENNESS, and AN-TITRUST) and as shares of GDP (ED SHARE). Giventhe relative crudeness and limited availability of thesemeasures, it is striking that these variables enter signif-icantly, suggesting that policy variation has an impor-tant role to play in determining R&D productivity.24

The coefficients associated with the Likert scale mea-sures will measure the predicted percentage change inPATENTS which would result from a one unit changein that variable (e.g. from a value of 3–4). For ex-ample, the coefficients on IP (0.22) and OPENNESS(0.10) imply that a one-point increase in these Likert

24 Indeed, given the noisy nature of the Likert variables andthe fact that poorly designed antitrust policies can stifle ratherthan stimulate innovation (e.g. by expropriating the returns fromsuccessful R&D efforts), it is perhaps not surprising that the modeldoes not identify a separate impact for the ANTITRUST variable.

920J.L

.F

urm

an

et

al./R

ese

arch

Po

licy3

1(2

00

2)

89

9–

93

3


Table 5Determinants of the production of new-to-the-world technologies (using patent stock as knowledge stock)

Dependent variable= ln(PATENTS)j,t+3

Baseline ideas productionfunction (with year FTE)(5.1)

Common innovationinfrastructure (5.2)

National innovativecapacity: preferredmodel (5.3)

Quality of the common innovation infrastructureA L PATENT STOCK 1.174 (0.072) 0.605 (0.040) 0.478 (0.039)HA L FTE S&E 0.431 (0.062) 0.388 (0.050) 0.544 (0.051)HA L R&D $ 0.011 (0.040) 0.069 (0.037)XINF ED SHARE 0.034 (0.014) 0.073 (0.014)XINF IP 0.104 (0.044) 0.080 (0.033)XINF OPENNESS −0.011 (0.027) −0.001 (0.023)XINF ANTITRUST −0.010 (0.027)

Cluster-specific innovation environmentYCLUS PRIVATE R&D FUNDING 0.009 (0.002)YCLUS SPECIALIZATION 3.220 (0.555)

Quality of the linkagesZLINK UNIV R&D PERFORMANCE 0.006 (0.003)

ControlsCountry fixed effects SignificantYear fixed effects SignificantYEAR −0.099 (0.007) −0.081 (0.007)L GDP 1967 0.375 (0.085) 0.443 (0.083)US dummy −0.221 (0.081) −0.010 (0.076)R2 0.9960 0.9771 0.9819Observations 353 347 347

measures (roughly equal to 1 S.D. shift) is associatedwith a 22 and 10% increase in PATENTS, respectively.Similarly, a 1% point increase in ED SHARE (approx-imately 1 S.D.) is associated with an 11% increase inthe level of PATENTS.

We then turn to two specifications including mea-sures drawing upon insights emphasizing the impor-tance of the cluster-specific innovation environment(YCLUS) and the quality of linkages between the com-mon infrastructure and clusters (ZLINK ). 25 These ad-ditions neither affect the significance nor the apprecia-bly alter the coefficients of the measures included inprevious models. However, consistent with a perspec-tive that innovation-based competition in a country’sindustrial clusters raises the marginal productivity of

25 Note that the coefficients in (4.2) on L GDP and L POP areof roughly equal magnitude though opposite in sign, and so wesimply employ GDP PER CAPITA in most of the remainder ofour analysis.

R&D resources, SPECIALIZATION and PRIVATER&D FUNDING enter positively and significantly. A1 S.D. increase in SPECIALIZATION (0.03) is asso-ciated with an 8% increase in the rate of internationalpatenting by a country, while a 1% point increase inthe fraction of R&D fundedby the private sector isassociated with a 1.4% increase in PATENTS. Thisimplies that, for our sample, relatively higher levels oftechnological specialization and industry R&D fund-ing are associated with higher levels of R&D produc-tivity. While these findings in no way imply a dis-positive hypothesis test of the role that clusters andlinkages play in determining innovation, they suggestthat outcomes associated with the operation of thesemechanisms does play an important role in shapingnational R&D productivity.

The first of our indicators of the quality of linkagesbetween industrial clusters and the common innova-tion infrastructure, the fraction of R&Dperformedby

922J.L

.F

urm

an

et

al./R

ese

arch

Po

licy3

1(2

00

2)

89

9–

93

3

J.L.

Fu

rma

ne

ta

l./Re

sea

rchP

olicy

31

(20

02

)8

99

–9

33

923

924J.L

.F

urm

an

et

al./R

ese

arch

Po

licy3

1(2

00

2)

89

9–

93

3


universities enters into (4.3) with a positive sign andcoefficient significant at the 5% level. This impliesthat countries with a higher share of their R&D per-formance in the educational sector (as opposed to theprivate sector or in intramural government programs)have been able to achieve significantly higher patent-ing productivity. In light of the results for PRIVATER&D FUNDING, this finding provides evidence that,controlling for the level of R&D inputs, thecompo-sition of spending affects the level of realized inter-national patenting. The second indicator ofZLINK ,VENTURE CAPITAL, enters positively into the equa-tion, but is not significant. This may reflect the rela-tively low level of variation of VC across the worldoutside United States and the existence of non-VCsources of risk capital in many OECD countries.

Consistent with prior studies (Jones, 1995), thecoefficients of the year dummies decline over time,suggesting that average global R&D productivityhas declined since the mid-1970s. In other words,while PATENTS has been increasing over time asa result of increased investments in innovation andimprovements in the policy environment, countrieshave also tended to experience a “raising the bar”effect in which ever-more R&D resources must bedevoted in order to yield a constant flow of visibleinnovative output. We present our preferred model ofnational innovative capacity in (4.4), including onlythose contributors to national innovative capacitythat enter significantly in prior models (i.e. exclud-ing ANTITRUST and VENTURE CAPITAL). Eachof the included regressors has a quantitatively im-portant impact, consistent with our perspective thatrelatively nuanced influences on national innovativecapacity have an important impact on the level ofinternational patenting. In Table 5, we present regres-sions that employ the alternative measure ofAj ,t ,PATENT STOCK, rather than population-adjustedGDP.26 Whereas population-adjusted GDP is a

26 By including PATENT STOCK in the specification which in-cludes FTE S&E and controls for year and country effects, (5.1)serves as an empirical test of a key parametric restriction associatedwith ideas-driven growth models. In particular, in order for ideasproduction to be a sustainable source of equilibrium long-termgrowth, φ = 1 (a hypothesis which cannot be rejected in (5.1)).Such parametric restrictions for ideas-driven growth models areexplored much more extensively (and derived formally) in Porterand Stern (2001).

comprehensive, composite indicator of a country’stechnological sophistication, PATENT STOCK pro-vides a more direct measure of the knowledge stockupon which a country draws for technological inno-vation. This table also employs alternative structuresto control for year and country effects; (5.1) in-cludes dummies for every country and year, while(5.2) and (5.3) rely on GDP 1967 to control forthe “baseline” knowledge stock.27 By exploringdifferences in the measure of the knowledge stockas well as by varying the means of identification,Table 5 highlights the robustness of the results toalternative assumptions about the nature of hetero-geneity among countries and the specification for thecountry-specific level of technological knowledge.Though there are differences in the magnitude ofsome variables and the significance of others, the re-sults are similar to those obtained in Table 4. Perhapsthe most important difference is that the coefficienton FTE S&E is much smaller in the equations ofTable 5, suggesting greater concavity of PATENTSto FTE S&E. In other words, the impact of changesin FTE S&E on PATENTS is lower after controllingfor the realized level of prior international patenting.In addition, the coefficient on R&D $ is insignificant(though of similar magnitude to Table 4), implyingthat PATENT STOCK incorporates much of the sta-tistical information embedded in R&D $. Finally,OPENNESS becomes negative and insignificant, per-haps because of the small number of observationsfor this variable. Nonetheless, the basic elements ofour framework remain significant, suggesting that(a) the level of R&D inputs is a critical determinantof the level of realized innovation; and (b) morenuanced measures of the national environment for in-novation play an important role in determining R&Dproductivity.

We evaluate the relative role of these differentforces in explaining the overall dispersion of inno-vation in Table 6. As discussed earlier, even a singlemeasure of the size of the economy or the levelof R&D inputs can explain a substantial share ofthe overall variation in PATENTS. Therefore, even

27 This contrasts with our use of year-specific fixed effects fromTable 4. Note that because the Likert variables are not observeduntil the late 1980s, we include separate dummy variables todenote whether such variables are included in the regression.


though coefficients identify the statistical and quan-titative impact of the measures of national innova-tive capacity, it is useful to evaluate the degree towhich these additional measures explain the over-all variation in PATENTS. To do so, we undertakea series of two-stage regression procedures. In thefirst stage, we regress a year dummies and some(but not the full set) of measures associated withnational innovative capacity. For example, in (6.1),the first-stage regression includes year dummies,GDP PER CAPITA, FTE S&E and R&D $ (i.e. theelements ofA and HA). In the second stage, wethen regress theresiduals from the first stage onthe remaining measures (i.e. in (6.1), ED SHARE,IP, OPENNESS, PRIVATE R&D FUNDING, SPE-CIALIZATION, and UNIV R&D PERFORMANCE).The R2 of this regression provides a conservativeindication of the relative importance of measuresincluded in the second stage. Specifically, the sec-ond stage indicates the share of the variance inPATENTS unexplained by variables in the first stagethat can be explained by measures in the secondstage.

Our results are informative. Even after first con-trolling for factors associated with the endogenousgrowth literature (A andHA), over one-quarter of theremaining variance can be explained by relativelynoisy measures associated with the national innova-tive systems and national industrial clusters literatures(XINF, YCLUS andZCLUS). Similarly, our analysis sug-gests that differences in innovation across countriesare linked not only to the level of R&D inputs but todifferences in the drivers of R&D productivity. Specif-ically, when we includeonly R&D inputs in the firststage, nearly 50% of the idiosyncratic variance can betied to measures associated with R&D productivity inthe second-stage regression. In contrast, when we firstinclude the R&D productivity measures in the firststage and includeonly the R&D input measures in thesecond stage, a smaller share of the remaining variance(43%) is explained. Finally, even if we include all ofthe elements of the common innovation infrastructurein the first stage, the compositional variables (PRI-VATE R&D FUNDING, SPECIALIZATION, andUNIV R&D PERFORMANCE) explain over 14% ofthe remaining variance in the second-stage regression,a share consistent with the amount that can be ex-plained by the policy variables (IP and OPENNESS)

when they are included exclusively in the secondstage.

6.2. Robustness checks

Table 7 explores the robustness of the primarymodel to alternative specifications and sample sub-groups. Equation (7.1) demonstrates robustness tothe inclusion of scale-dependent measures (GDPand POP) alongside PATENT STOCK. Both GDPand PATENT STOCK are positive and significant.Despite being strongly correlated, each has a sepa-rate, quantitatively important impact on PATENTS.Also, as in (4.1) and (4.2), the negative coefficienton L POP suggests that, after controlling for thelevel of prosperity and technological sophistica-tion, increases in population (implying a decreasein GDP PER CAPITA) are associated with lessnew-to-the-world innovation. Together with our ear-lier results, we interpret this to suggest that bothPATENT STOCK and population-adjusted GDP pro-vide useful measures of a nation’s national innovationinfrastructure. Indeed, to the extent that PATENTSTOCK itself provides an index of the knowledgestock of an economy, the separate impact of GDPsuggests the role of economy-wide demand influ-ences in the rate of new-to-the-world innovation.High per capita income allows buyers to demandmore advanced products and services, encourag-ing the development of new-to-the-world technolo-gies.

To establish the precise role of cross-sectionalvariation in our results, equation (7.2) includescountry-specific fixed effects. Most of the resultsare robust to this modification: GDP PER CAPITA,R&D expenditure, FTE R&D, and higher educationspending remain significant and of the expected signin explaining patent output. It is interesting to notethat the magnitude of the coefficient on FTE S&E in-creases, suggesting that the level of innovative outputis sensitive tochangesin the level of the scientificworkforce within a given country. SPECIALIZATIONalso remains significant (and its coefficient increasesnearly 40%). However, the R&D composition vari-ables become insignificant and OPENNESS entersnegative and significant. The estimates of these regres-sors are sensitive to the inclusion of country-specificfixed effects because UNIV R&D PERFORMANCE,


PRIVATE R&D FUNDING change only slowly overtime and OPENNESS is observed for only a shorttime period (1989–1993).

We examine the robustness of our results to vari-ous sample subgroups in (7.3) and (7.4). Restrictingthe sample to post-1970s data in (7.3), we highlighta period of relatively higher macroeconomic stabil-ity in which the reliability of the data is most likelyimproved. All of the results remain significant, al-though the coefficients in this equation suggest thataggregate national prosperity (GDP PER CAPITA)is somewhat less important as a driver of R&D pro-ductivity in the last 15 years of our sample, whilethe importance of private sector funding (PRIVATER&D FUNDING) and university R&D performance(UNIV R&D PERFORMANCE) has increased. Inother words, in the second half of our sample pe-riod, more nuanced drivers of national innovativecapacity take on increased relative importance indetermining the flow of PATENTS. To explore re-gional differences in our data, we restrict attentionto European countries in (7.4). The impact of GDPPER CAPITA on innovative output is somewhatlower in these countries, while the relative influ-ence of ED SHARE is somewhat higher. Similarto earlier results, OPENNESS loses its significancein this sub-sample, suggesting that variation amongEuropean countries is not sufficient to establish thisresult.

We explore the impact of academic publicationon international patenting productivity by addingJOURNALS to our preferred specification in (7.5).Although this specification is obviously subject toendogeneity (and we leave separating out the sep-arate exogenous drivers of each to future work),the results suggest that while JOURNALS is sig-nificant and positively correlated with PATENTS,its inclusion does not substantially change ourearlier qualitative conclusions, except for reduc-ing OPENNESS to insignificance (consistent withsome of our other robustness checks), and some-what reducing the coefficients on FTE S&E and EDSHARE. In other words, the empirical relationshipbetween international patenting and key drivers ofcommercially-oriented innovative output is relativelyunaffected, at least in the short- to medium-term, bythe level of abstract scientific knowledge produced bya country.

6.3. The impact of national innovative capacity ondownstream competitiveness measures

Our analysis so far has focused exclusively on thesensitivity of PATENTS to measures of the strength ofnational innovative capacity. We extend our analysisto consider more downstream consequences of inno-vative capacity, and find that the elements of nationalinnovative capacity play an important role in shapingboth the level of TFP and MARKET SHARE.

Table 8 begins with an overall assessment of thesensitivity of TFP to cumulative ideas production(PATENT STOCK). Specifically, (8.1) evaluates thesensitivity of GDP to PATENT STOCK, conditionalon the level of LABOR and CAPITAL. The coeffi-cient on PATENT STOCK can be interpreted as acontributor to the level of TFP.28 As discussed morefully in Porter and Stern (2001), this relationship iscritical for establishing the role of “ideas production”in long-run economic growth. Theoretical growthmodels assume that there is a strong R&D productiv-ity advantage associated with technological sophisti-cation that translates into a proportional advantage inthe realized level of TFP. Equation (8.1) implies thatwhile the PATENT STOCK has a significant effect onTFP, this effect ismuchsmaller than proportionality(which would require that the coefficient be equal tounity). This modest relationship suggests that nationalinnovative capacity plays a significant role in shapingthe medium-term level of productivity, but raises thepossibility that the linkage between national inno-vative output and productivity growth may be moresubtle than commonly assumed.

We examine whether MARKET SHARE can beexplained using the national innovative capacityframework in equations of (8.2)–(8.4). First, (8.2)shows that (lagged) PATENT STOCK, GDP, POPand a time trend explain nearly 80% of the vari-ance in MARKET SHARE across countries. Thisdemonstrates that countries that accumulate ad-vanced knowledge stocks later achieve high shares in

28 Alternatively, we could have simply imposed factor shares onLABOR and CAPITAL and computed TFP directly; while weexperimented with this formulation, the less restrictive specificationin (9.1) is more consistent with the exploratory nature of theexercise in this paper (but see Porter and Stern, 2001, for furtherdetails).

928J.L

.F

urm

an

et

al./R

ese

arch

Po

licy3

1(2

00

2)

89

9–

93

3


worldwide high-technology markets. Equation (8.3)replaces PATENT STOCK with the predicted levelof PATENTS, where the weights are derived from(4.4). Not surprisingly, given how closely we char-acterize PATENTS in (4.4), these results are quitesimilar to those in (8.2). Finally, in (8.4), we in-clude the measures incorporated into our preferredmodel of national innovative capacity (4.4), alongwith JOURNALS. This exercise fits the MARKETSHARE data similarly well and each element of na-tional innovative capacity remains significant, withthe exception of IP, which is not estimated pre-cisely. Notably, JOURNALS is also insignificantin the regression, suggesting that scientific outputper se does not play an independent short-term rolein affecting national shares of world technologyexports.

6.4. Trends in national innovative capacity

Our framework allows us to analyze whether dif-ferences in the intensity of innovation and R&Dproductivity reported earlier (see Fig. 4) can be un-derstood in terms of the sources of national innovativecapacity. Since, at the most aggregate level, we areinterested in the potential production of innovativeoutput relative to national population, we measure in-novative capacity by calculating the level of expectedPATENTS (using the coefficients from our preferredmodel (4.4)) and dividing by the level of popula-tion (in millions).29 In effect, we are computing thepredicted value for a country’s level of internationalpatenting per capita based on its fundamental re-source and policy commitments, thereby providing auseful benchmark to compare the relative ability ofcountries to produce innovations at the internationalfrontier.

The result of this exercise is presented in Fig. 8.Consistent with the historical PATENTS PER CAPITAdata, United States and Switzerland are in the top tierthroughout the sample period. As a result of sustainedinvestments in fundamentals, such as increases in FTES&E and R&D $, improvement in IP protection andopenness, and a high share of R&D performed in

29 We have also completed this exercise dividing through byFTE S&E rather than POP. The results are quite similar (and areavailable upon request).

industry, Japan, Germany, and Sweden joined this topgroup over the course of the 1980s. A second set ofcountries, including the remaining Scandinavian coun-tries, France, and UK, comprise a “middle tier”, whilea third group, including Italy, New Zealand and Spain,lags behind the rest of the OECD over the full-timeperiod.30

The most striking finding of this analysis is theconvergencein measured innovative capacity amongOECD countries over the past quarter century. Notonly has the top tier expanded to include Japan,Germany, and Sweden, but some middle tier coun-tries, such as Denmark and Finland, have achievedsubstantial gains in innovative capacity. Moreover,convergence seems to be built on the fundamentalsof innovative capacity, rather than transient changes.This is exemplified by the case of Germany, in whichthe components of innovative capacity grew stronglythroughout the 1980s. Despite a drop-off resultingfrom reunification with the east beginning in 1990,Germany has maintained a relatively high level ofinnovative capacity throughout 1990s.

In general, there has been a slow but steady narrow-ing of the gap between the leaders in the OECD andnations with historically lower levels of innovative ca-pacity. It is important to note, however, that some ma-jor countries, most notably France and UK have seenerosion in their relative innovative capacity over thepast quarter century by investing less in common inno-vation infrastructure, providing less supportive clusterenvironments, and/or losing relative position in link-age mechanisms. While this approach does not providea forecast of the ability of a country to commercial-ize new technologies in the short-term, our results dosuggest that both Japan and Scandinavia have alreadyestablished themselves as important innovation cen-ters and that the nations that produce new-to-the-world

30 It is important to bear in mind that these results are in partaffected by the industrial composition of national economies interms of patenting propensity across industries. Our choice ofPATENTS implies that innovation in countries whose clusters areconcentrated in industries with low patent intensities (such as Italyin textiles) will be understated relative to those with clusters inpatent-intensive industries (such Switzerland in pharmaceuticals).However, it is useful to note that the analysis does control fordifferences captured in the scale of R&D effort per se, throughthe FTE S&E and R&D $ measures.


Fig. 8. Trends in national innovative capacity.

innovation are likely to become more diverse overtime.31

7. Concluding thoughts

This paper introduces the concept ofnational in-novative capacityto integrate previous perspectiveson the sources of differences in the intensity ofinnovation and R&D productivity across countriesand provides an empirical framework to distinguishamong alternative causes of these differences. Ourresults suggest that the empirical determinants ofinternational patenting activity are: (a) amenable tosystematic empirical analysis motivated by our frame-work and (b) more nuanced than the limited factorshighlighted by ideas-driven growth theory. We findthat a set of additional factors also plays an importantrole in realized R&D productivity. Further theoreticaland empirical research in growth theory may benefit

31 This part draws on Porter and Stern (1999).

from incorporating the role of industrial organizationand the national policy environment (e.g. the roleof the university system or incentives provided forinnovation).

Country-level R&D intensity and productivity seemto be amenable to quantitative analysis (though withsome caveats), a finding that should be of particularinterest to researchers in the tradition of the nationalinnovation systems literature. In particular, futureresearch can usefully distinguish between those phe-nomena that are reflected in observable measures ofinnovative output (such as the patenting activities weexamine here) and those with more subtle effects thatmay not be subject to direct observation (such asinstitutions or mechanisms encouraging non-patentedprocess innovations). At the very least, our resultssuggest that quantitative research can play a largerrole in distinguishing among alternative perspectivesin this field.

Our results suggest that public policy plays animportant role in shaping a country’s national inno-vative capacity. Beyond simply increasing the level of


R&D resources available to the economy, other policychoices shape human capital investment, innovationincentives, cluster circumstances, and the quality oflinkages. Each of the countries that have increasedtheir estimated level of innovative capacity overthe last quarter century—Japan, Sweden, Finland,Germany—have implemented policies that encouragehuman capital investment in science and engineering(e.g. by establishing and investing resources in tech-nical universities) as well as greater competition onthe basis of innovation (e.g. through the adoption ofR&D tax credits and the gradual opening of marketsto international competition).

Finally, our results suggest that United States, thedominant supplier of new technologies to the rest ofthe world since World War II, had been less proac-tive in its investment in innovative capacity in theearly 1990s than it was in the late 1980s. With theconclusion of the Cold War, the traditional Ameri-can rationale for investing in its national innovationinfrastructure became less clear, and US policy hadbecome less focused. In light of continuously in-creasing investments by an increasingly diverse set ofcountries, convergence in innovative capacity acrossthe OECD should not be surprising. This convergencesuggests that the commercial exploitation of emerg-ing technological opportunities (from biotechnologyto robotics to Internet technologies) may well be lessgeographically concentrated than was the case duringthe post World War II era.

Pairwise correlations

PATENTSt−3 POP GDP GDP PERCAPITA

PATENTSTOCK

FTES&E

R&D $ PRIVATER&DFUNDING

UNIVR&DPERFOR-MANCE

POPULATION 0.940∗GDP 0.973∗ 0.984∗GDP PER CAPITA 0.110∗ −0.063 0.044PATENT STOCK 0.920∗ 0.832∗ 0.906∗ 0.164∗FTE S&E 0.981∗ 0.974∗ 0.988∗ 0.051 0.897∗R&D $ 0.928∗ 0.844∗ 0.920∗ 0.175∗ 0.974∗ 0.903∗PRIVATE R&D

FUNDING0.226∗ 0.188∗ 0.216∗ 0.589∗ 0.218∗ 0.193∗ 0.252∗

UNIV R&DPERFORMANCE

−0.377∗ −0.488∗ −0.444∗ 0.110∗ −0.331∗ −0.437∗ −0.337∗ −0.216∗

SPECIALIZATION 0.052 0.015 0.027 −0.138∗ 0.004 0.0216 0.091 0.140∗ 0.003

Acknowledgements

We thank Joshua Gans, Chad Jones, Richard Lester,Paul Romer, Gary Pisano and two anonymous refer-ees for helpful suggestions and seminar participantsat the MIT Industrial Performance Seminar, IndustryCanada, and the Academy of Management for com-ments. Greg Bond, Chad Evans, and Propa Ghoshprovided excellent research assistance. The authorsalso gratefully acknowledge support provided by theCouncil on Competitiveness. Stern completed this pa-per while a Health and Aging Fellow at the NationalBureau of Economic Research, whose hospitality isgratefully appreciated. Responsibility for all errors andomissions remains our own.

Appendix A

Sample countries (1973–1995)

Australia France Netherlands UKAustria Germanya Norway United StatesCanada Italy SpainDenmark Japan SwedenFinland New Zealand Switzerland

a Prior to 1990, data for the Federal Republicof Germany include only the federal states of WestGermany; beginning in 1991, data for Germany incor-porate the new federal states of the former GermanDemocratic Republic.


References

Ambramovitz, M., 1956. Catching up, forging ahead and fallingbehind. Journal of Economic History 46, 385–406.

Arora, A., Landau, R., Rosenberg, N., 1998. Chemicals andLong-Term Economic Growth: Insights from the ChemicalIndustry. Wiley, New York.

Audretsch, D., Stephan, P., 1996. Company–scientist locationallinks: the case of biotechnology. American Economic Review86 (3), 641–652.

Bostic, R., Gans, J., Stern, S., 1996. Urban productivity andfactor growth in the late nineteenth century. Journal of UrbanEconomics 41 (1), 38–55.

Carlsson, B., Stankiewicz, J., 1991. On the nature, function, andcomposition of technological systems. Journal of EvolutionaryEconomics 1 (2), 93–118.

Cockburn, I., Henderson, R., 1994. Racing to invest? The dynamicsof competition in ethical drug discovery. Journal of Economicsand Management Strategy 3 (3), 481–519.

Coe, D., Helpman, E., 1995. International R&D spillovers.European Economic Review 39, 859–887.

Dosi, G. (Ed.), 1988. Technical Change and Economic Theory.Pinter Publishers, London, UK.

Dosi, G., Freeman, C., Silverberg, G., Soete, L., 1988. TechnicalChange and Economic Theory. Pinter Publishers, London, UK.

Dosi, G., Pavitt, K., Soete, L., 1990. The Economics of TechnicalChange and International Trade. Columbia University Press,New York.

Eaton, J., Kortum, S., 1996. Trade in ideas: patenting andproductivity in the OECD. Journal of International Economics40 (3/4), 251–278.

Eaton, J., Kortum, S., 1999. International technology diffusion:theory and measurement. International Economic Review 40 (3),537–570.

Edquist, C. (Ed.), 1997. Systems of Innovation: Technologies,Institutions, and Organizations. Pinter Publishers, London, UK.

Ellison, G., Glaeser, E., 1997. Geographic concentration in USmanufacturing industries: a Dartboard approach. Journal ofPolitical Economy 105 (5), 889–927.

Evenson, R., 1984. International invention: implications fortechnology market analysis. In: Griliches, Z. (Ed.), R&D,Patents, and Productivity. Chicago University Press, Chicago,IL, pp. 89–126.

Freeman, C., 1988. Japan: a new system of innovation. In: Dosi,G. (Ed.), Technical Change and Economic Theory. PinterPublishers, London, UK, pp. 330–348.

Glaeser, E., Kallal, H., Scheinkman, J., Shleifer, A., 1992. Growthin cities. Journal of Political Economy 100 (6), 1126–1152.

Griliches, Z. (Ed.), 1984. R&D, Patents and Productivity. ChicagoUniversity Press, Chicago, IL.

Griliches, Z., 1990. Patent statistics as economic indicators: asurvey. Journal of Economic Literature 92, 630–653.

Griliches, Z., 1994. Productivity, R&D, and the data constraint.American Economic Review 84, 1–23.

Griliches, Z., 1998. R&D and Productivity: The EconometricEvidence. Chicago University Press, Chicago, IL.

Grossman, G., Helpman, E., 1991. Innovation and Growth in theGlobal Economy. MIT Press, Cambridge, MA.

IMD Competitiveness Report, 1989–1999. World CompetitivenessYearbook. Lausanne, Switzerland.

Jaffe, A., Trajtenberg, M., Henderson, R., 1993. Geographiclocalization of knowledge spillovers as evidenced by patentcitations. The Quarterly Journal of Economics 434, 577–598.

Jones, C., 1995. R&D based models of economic growth. Journalof Political Economy 103, 739–784.

Jones, C., 1998. Introduction to Economic Growth. WW Norton& Co., New York.

Keller, W., 2000. Geographic localization of internationaltechnology diffusion. NBER Working Paper #7509.

Kortum, S., 1997. Research, patenting, and technological change.Econometrica 65 (6), 1389–1419.

Krugman, P., 1991. Geography and Trade. MIT Press, Cambridge,MA.

Lim, K., 2000. Basic research, applied research, and innovation inthe semiconductor and pharmaceutical industries. Unpublisheddoctoral dissertation, MIT Sloan School of Management.

Lundvall, B.A., 1988. Innovation as an interactive process: fromuser-producer interaction to the national system of innovation.In: Dosi, G. (Ed.), Technical Change and Economic Theory.Pinter Publishers, London, UK, pp. 349–369.

Merges, R.P., Nelson, R., 1990. On the complex economics ofpatent scope. Columbia Law Review 90, 839–916.

Mowery, D., 1984. Firm structure, government policy, and theorganization of industrial research: Great Britain and the UnitedStates, 1900–1950. Business History Review 58, 504–531.

Mowery, D., Rosenberg, N., 1998. Paths to Innovation. CambridgeUniversity Press, Cambridge, MA.

Mowery, D., Nelson, R.R. (Ed.), 1999. Sources of IndustrialLeadership: Studies of Seven Industries. Cambridge UniversityPress, Cambridge, MA.

Murmann, J.P., 1998. Knowledge and competitive advantage in thesynthetic dye industry, 1850–1914: the coevolution of firms,technology, and national institutions in Great Britain, Germany,and the United States. Unpublished doctoral dissertation,Columbia University.

Nelson, R.R., 1988. Institutions supporting technical change inthe united states. In: Dosi, G. (Ed.), Technical Change andEconomic Theory. Pinter Publishers, London, UK, pp. 312–329.

Nelson, R.R. (Ed.), 1993. National Innovation Systems: AComparative Analysis. Oxford University Press, New York.

Nelson, R.R., Wright, G., 1992. The rise and fall ofAmerican technological leadership: the postwar era in historicalperspective. Journal of Economic Literature 30 (4), 1931–1964.

Nelson, R.R., Rosenberg, R., 1994. American universities andtechnical advance in industry. Research Policy 23 (3), 323–348.

Niosi, J. (Ed.), 1991. Technology and National Competitiveness:Oligopoly, Technological Innovation, and InternationalCompetition. Mc-Gill-Queen’s University Press, Montreal.

Park, W.G., 1995. International R&D spillovers and OECDEconomic Growth. Economic Inquiry 33 (4), 571–591.

Patel, P., Pavitt, K., 1994. National innovation systems: why theyare important, and how they might be measured and compared.Economics of Innovation and New Technology 3 (1), 77–95.

Pavitt, K., 1982. R&D, patenting, and innovative activities: astatistical exploration. Research Policy 11 (1), 33–51.


Pavitt, K., 1988. Uses and abuses of patent statistics. In: van Raan,A. (Ed.), Handbook of Quantitative Studies of Science Policy.North-Holland, Amsterdam.

Porter, M.E., 1990. The Competitive Advantage of Nations. FreePress, New York.

Porter, M.E., 1998. Clusters and Competition: New Agendas forCompanies, Governments, and Institutions, on Competition.Harvard Business School Press, Boston, MA.

Porter, M.E., Stern, S., 1999. The New Challenge to America’sProsperity: Findings from the Innovation Index. Council onCompetitiveness, Washington, DC.

Porter, M.E., Stern, S., 2001. Measuring the ‘Ideas’ ProductionFunction: Evidence from International Patent Output. MITSloan School of Management, Mimeo.

Romer, P., 1990. Endogenous technological change. Journal ofPolitical Economy 98, S71–S102.

Romer, P., 1996. Why, indeed, in America? American EconomicReview 86 (2), 202–206.

Rosenberg, N., 1963. Technological change in the machine toolindustry, 1840–1910. Journal of Economic History 23, 414–443.

Rosenberg, N. (Ed.), 1969. The American System of Manufactures.Edinburgh University Press, Edinburgh.

Rosenberg, N., 1972. Technology and American Growth. Harper,New York.

Rosenbloom, R., Cusumano, M., 1987. Technological pioneeringand competitive advantage: the birth of the VCR industry.California Management Review 29 (4), 51–76.

Sakakibara, M., Porter, M.E., 2000. Competing at home to winabroad: evidence from Japanese industry. Review of Economicsand Statistics 53 (2), 310–322.

Saxenian, A., 1994. Regional Advantage: Culture and Competitionin Silicon Valley and Route 128. Harvard University Press,Cambridge, MA.

Scherer, F.M., 1983. The propensity to patent. International Journalof Industrial Organization 1, 107–128.

Schmookler, J., 1966. Invention and Economic Growth. HarvardUniversity Press, Cambridge, MA.

Silverberg, G., Dosi, G., Orsenigo, L., 1988. Innovation, diversity,and diffusion: a self organizing model. Economic Journal98 (393), 1032–1054.

Solow, R.M., 1956. A contribution to the theory of economicgrowth. Quarterly Journal of Economics 70, 65–94.

Thomas, L.G., 1994. Britain versus France in globalpharmaceuticals. Industrial and Corporate Change 3 (2), 451–489.

Trajtenberg, M., 1990. Patents as Indicators of Innovation,Economic Analysis of Product Innovation. Harvard UniversityPress, Cambridge, MA.

USPTO, 2000a. US Patent and Trademark Office, 2000.Technology Assessment and Forecast Report. Chemical Classes1977–December 1999. Information Products Division, TAFBranch, Washington, DC.

USPTO, 2000b. US Patent and Trademark Office, 2000.Technology Assessment and Forecast Report. Electrical Classes1977–December 1999. Information Products Division, TAFBranch, Washington, DC.

USPTO, 2000c. US Patent and Trademark Office, 2000.Technology Assessment and Forecast Report. MechanicalClasses 1977–December 1999. Information Products Division,TAF Branch, Washington, DC.

Vertova, G., 1999. Stability in national patterns of technologicalspecialisation: some historical evidence from patent data.Economics of Innovation and New Technology 8 (4), 331–354.

Ziegler, N., 1997. Governing Ideas: Strategies for Innovationin France and Germany. Cornell University Press, Cornell,NY.

Zucker, L., Darby, M., Brewer, M., 1998. Intellectual humancapital and the birth of US biotechnology enterprises. AmericanEconomic Review 88 (1), 290–306.

the determinants of national innovative...

Documents