The Technological Position of European Countries in Historical Perspective: Catching Up, Agglomeration and Path-Dependency

Transcription

1 The Technological Position of European Countries in Historical Perspective: Catching Up, Agglomeration and Path-Dependency Paper for the DIME workshop on Dynamics of Knowledge Accumulation, Competitiveness, Regional Cohesion and Economic Policies, 2 4 February 2006, WIIW, Vienna Odile E.M. Janne Department of Management, Birkbeck, University of London, o.janne@bbk.ac.uk Abstract This paper investigates whether there has been a long run tendency for technological activity to agglomerate or diffuse across European countries over the period The issue is discussed with reference to the literature on convergence in growth theory and the cumulative and differentiated nature of technological change in evolutionary economics. The paper evaluates to what extent path-dependent technological learning may have implied long-term international differentiation in national technological patterns, overall and at sectoral level; and whether the process may induce convergence and catching up, but also divergence, forging ahead and falling behind in relative technological position. The quantitative empirical study applies the Galtonian regression model using countries' patents granted in the US classified into four common groups of technological activity. A qualitative assessment of countries technological positions over time also takes into account social, cultural and institutional factors involved in supporting and directing technological change. The paper provides empirical support for the proposition that national technological positions can be established within sectors, tend to persist over time, and matter to explain the divergence and differentiation or convergence and catching up of countries. The results suggest a process of technological convergence between European countries, especially after the Second World War. However, the apparent convergence process may only embody some of the leading and intermediate European countries in particular sectors, and reflects the long run relative decline of the UK as a technological leader. Countries exhibit persistent different technological strengths and weaknesses at sectoral level, which relate to their historical conditions. This leads to considerations on the sources and development of national comparative advantage according to the ways institutions have supported particular technologies, and their significance as a factor explaining differences in growth performance. Governments would be constrained in their policies for the fostering of innovation and growth by their own prior accumulation of competencies, which tends to be sector-specific and influenced by multinational corporations. DIME is supported financially by the EU 6 th Framework Programme

2 The Technological Position of European Countries in Historical Perspective: Catching Up, Agglomeration and Path-Dependency. 1. Introduction Technology and Convergence Data and Methodology Agglomeration, convergence or catching up in technological activity across European countries National technological profiles and evolution across sectors...23 The Chemical Technological Group...23 The Electrical Technological Group...27 The Mechanical Technological Group...30 The Transport Technological Group...33 Discussion Conclusions

3 1. Introduction This paper investigates whether there has been a long-run tendency for technological activity to agglomerate or diffuse within Europe, using patents granted in the US between 1890 and The issue is discussed with reference to the literature on convergence in growth theory and the cumulative and differentiated nature of technological change in evolutionary economics. Drawn from the economic, business and technology history literature, a qualitative assessment of countries technological positions over time complements the evidence provided by the patent data to take into account the social, cultural and institutional factors involved in supporting and directing technological change. Technological progress has been identified by economists to be one of the main sources of economic growth. Economic historians, technologists, and advocates of the new growth and evolutionary theories all emphasise the importance of technology and innovation for understanding growth, development and convergence. In particular, it has been shown that convergence and divergence in economic performance owe much to differences in innovative activity (Dosi and Soete 1988, Bernard and Jones 1996, Fagerberg 1988, 1994, Fagerberg and Verspagen 2002). Changes in technological leadership have been argued to influence changes in economic leadership (Cantwell 1991a, Nelson and Wright 1994). Recently, there has been an increased interest in the subject of convergence, based on the observation that at least a restricted set of countries is following a converging trend of their levels of productivity and living standards (e.g. Baumol et al 1994, Durlauf 1996, Fagerberg et al 1999). One likely reason for this rebound of interest in economic growth is presumably its implications for the welfare of nations and for the design of policy, another may have been the challenge from the fast-growing countries of Asia, such as Japan, Korea, Taiwan and more recently China. Though often neglected in the literature, the explanation and analysis of international differences in technology are clearly important to both public and private policy-makers. This paper focuses on technological capabilities in their own rights. It identifies and discusses trends in the extent to which innovative activity has been geographically dispersed or concentrated amongst European countries during the twentieth century up to the early 1990s. The issue of geographical agglomeration or convergence and catching up of technological activity across countries over time is analysed by using the Galton model with countries' patents granted in the US classified into four common groups of technological activity, from 1890 to Results suggest, consistent with the convergence literature, that there 2

4 has been a process of technologies diffusion or catching up within European countries, especially after the Second World War. However, to conclude that there has been a broad European convergence may be fallacious, and no evidence is found of a general convergence trend. The apparent convergence process embodies only some of the leading and intermediate European countries in particular sectors, and reflects the long-run relative decline of the United Kingdom (UK) as a technological leader. Overall, the relative technological positions of countries stay significantly stable over the period. European countries exhibit different technological strengths and weaknesses at sectoral level, which relate to their historical conditions and tend to persist over time. The study shows that only a process of 'local' technological convergence can be expected within Europe, dominated by the convergence of some specific countries and without the elimination of a striking diversity of national technological performance across technological sectors. The literature is briefly reviewed in the next section. The third section introduces the data and methodology. The results and their discussion are presented in section four and five. The last section draws some conclusions. 2. Technology and Convergence The traditional neoclassical theory of economic growth, developed by Solow and others in the 1950s, predicts global convergence - that growth paths of different countries or regions will, at least in the long run, (more or less automatically) converge to each other. This result rests on the assumption of decreasing marginal returns to capital which, combined with initial differences in the capital-labour ratio, implies that countries with low initial per capita incomes tend to grow faster. This convergence hypothesis involves the notion of the advantage of 'relative backwardness' in levels of productivity, where followers tend to catch up faster if they are initially more backward. The catching up hypothesis implicitly assumes that technological knowledge is to some extent a public good, in the sense that its possible transfer costs would always be relatively small compared to the cost of creating new knowledge- i.e. innovating. Although technological progress is seen as an important engine of the growth process, it is regarded as a residual factor, created outside the economic system, as the part of economic growth that cannot be attributed to growth in physical factors of production (i.e. factor accumulation). Technological gaps would tend initially to occur from time to time essentially for accidental reasons and are transitory (Fagerberg 1988). 3

5 Though many empirical studies acknowledged the relevance of the catching up phenomenon, it seems that it is a local rather than a global one. While there is an agreement on the relevance of the convergence hypothesis for some countries (and perhaps for the OECD countries), it seems not to hold for the world as a whole, not only between countries but within them as well. The actual existence of convergence is still a controversial topic, mainly because of statistical difficulties with regard to estimation bias when using the Solow-Swann neoclassical model of growth (Lee et al 1997). Early studies based on the traditional convergence literature could only explain a small part of actual growth by the role of factor accumulation (Fagerberg 1988). The major share of actual growth had to be explained by exogenous technological progress and other unidentified sources. The development of growth accounting, which is based on the empirical analysis of growth performance and explanatory factors, led to a more eclectic view of growth. One of the most popular themes in the growth literature has been the issue of convergence of labour productivity and per capita income levels among the major industrialised countries over the last century and, especially since the Second World War. The process seems then to slow down after the mid-1970s. These results led to some controversy about how they should be interpreted, as to what extent the history of convergence over the last century should be seen as a process subject to some general tendencies, as opposed to a more specific interpretation of the post-war period? Although there was some doubt about the unambiguously beneficial position of backward countries, there was some evidence that the transfer of technology by several means permitted countries to catch up (Cornwall 1977, Nelson and Wright 1992, Baumol et al 1994, Bernard and Jones 1996). The most obvious explanation of observed productivity catch up was international technological spillover, or the ability of less advanced economies to imitate and copy the techniques of production used by more advanced ones (Baumol 1986, Abramovitz 1986, Dowrick 1992). Abramovitz (1986, 1994) and Perez and Soete (1988) warned that technological spillover may be effective only in the context of a 'favourable environment' comprising economic factors, social factors (level of education), socio-political factors (pressures for lobbies, social structures), historical, cultural and other factors. The potential for technological catching up for any country would consequently remain subject to some conditions, identified by Abramovitz as social capability and technological congruence. Nelson and Wright (1992, 1994) confirmed that convergence has occurred among countries with 4

6 modern education systems, strong internal scientific and engineering communities, and sophisticated industrial enterprises, while the others have tended to fall further and further behind. In this context, Baumol (1986, 1994) found evidence of catching up only between countries with similar systems and suggested different 'convergence clubs', as well as Fargerberg and Verspagen (1996), Galor (1996), Quah (1996) and Verspagen (1999). His analysis demonstrates that some of the poorest countries have failed to be part of the catching up process. Dowrick and Gemmel (1991) and De Long (1988) refuted the argument that there has been a tendency for poorer economies to catch up on the richer. Broadberry (1994, 1996) rejected the notion of global convergence within manufacturing in favour of local convergence within economies, eventually within Europe, having similar resource and factor endowments and demand conditions. Furthermore, Dowrick and Nguyen (1989) suggested that generally convergence takes place during periods of widescale expansion, while gaps widen during cyclical slow-downs such as the recession during the 1930s. Nelson, as reported in Fagerberg (1988), argues that the studies on growth accounts and the findings of possible explanatory variables do not really explain growth as they do not constitute a new tested theory on growth. They are rather descriptions based on a number of assumptions. To explain differences in growth, explanatory factors and how these interact have to be determined and explained, taking into account evolutionary and dynamic aspects of spatial development. The neoclassical perspective was argued not to explain satisfactorily the existence of persistent leaders and followers, nor changes in leadership, nor changes in the ranks of countries in their relative levels of productivity. These many issues remained to be researched in an alternative approach. In particular, technological issues, which appear to underlie the process of development, needed to be better understood and better taken into account. The neoclassical assumption of technology as an exogenous public good has not found much empirical support. As an alternative to the traditional perspective but still in the mainstream tradition, the more recent endogenous or new growth theory focuses on the importance of endogenising technological change in the new growth models (Romer 1986, 1990, Bernard and Jones 1996). A key element of the endogenous growth models is that technological knowledge is a non-rival good, being partly public and partly private, and would produce increasing returns. One of the findings of the new growth theories is that increasing returns to knowledge within a geographical area would explain spatial differences in growth rates (Romer 1986, 1990). Endogenous growth mechanisms may 5

7 generate divergence across economies and allow for convergence in some special circumstances conditional convergence. However, the endogenous theories and models have been criticised for not including the differing social, cultural and institutional factors in order to explain more realistically the uneven rate of development amongst countries (Patel and Pavitt 1996, Martin 1999). The evolutionary approach to economic growth represents a more radical departure from traditional analysis, which also provides an endogenous explanation to the process of technological change. Technology is defined as being partly tacit, specific and resulting from learning processes. Technological development is seen as cumulative, embedded in persons, firms, clusters of firms, regional and national networks (Dosi 1988, Nelson and Winter 1982, Rosenberg 1976). Geographical proximity is important to innovative activity because of the existence of knowledge spillovers and other externalities that are geographically bounded (Feldman 1993, Jaffe 1986, Jaffe et al 1993). The cumulative and path-dependent characteristics of technological change imply that a substantial, persistent and cumulative advantage in innovative activities would accrue to any country (or region) that gained a head start in any new technology since it would appropriate the available agglomeration economies. This may lead to self-reinforcing, virtuous circle type processes of economic growth. However, geographical areas may equally be locked in by historical events or by chance to different technologies and a path of technological change, which would be likely to evolve only gradually over time. It has been found that, from the early 1890s to the early 1910s, the sectoral distribution of technological activity of national groups of firms has become locked in to an established pattern (Cantwell 1991a). The notion of technological paradigm has been introduced, within an evolutionary approach, to understand better the long-term nature of technological change and provides a number of useful insights into the analysis of phenomena such as the persistence of technological differentials and changes of technological leadership. A technological paradigm defines not only the directions in which progress is possible, i.e. boundaries, but also the likely tactics to be used, i.e. the directions or trajectories to those boundaries (Dosi 1982, Nelson and Winter 1977). Periods of paradigm transition would allow for the existence of some temporary windows of opportunity for lagging countries as each new technological paradigm will require and generate new types of skills and new locational and infrastructural advantages (Perez and Soete 1988). Technological leadership may consequently change as, in the long run, there is a development in the types of technological activity that yield the greatest potential. Technological specialisation, and the 6

8 extent to which it matches the dominant paradigm, helps to understand why the ability to innovate varies among countries. Countries that had accumulated great advantages in the context of one technology system, as part of a more broadly defined technological paradigm, faced diseconomies in the transition period to a new technological paradigm. Periods of paradigm change may have allowed some lagging countries to catch up and even to surpass the previous leaders. As the process of innovation, technology diffusion and accumulation is geographically constrained, assumptions on the geographical reach of spillovers and other externalities become a crucial issue, because they can be defined at the regional, national or European level. The notion of National Systems of Innovation (NSI) puts major emphasis on the importance of national institutions, culture and history in the innovation and diffusion processes (Freeman 1995, Patel 1995, Patel and Pavitt 1994b, Nelson 1993, 1996). Nelson and Wright (1994) argue that national borders have largely defined the scope for technological progress. Language, terminology, institutional structures and objects of study produce national specificity in technological activities and are still relevant among European countries and regions despite modern communication techniques (Maurseth and Verspagen 1999). Hence, there is a growing interest in the implications for the role of policy for economic growth. Governments play an important role for both the public and private elements of the NSI (Patel and Pavitt 1994b, Cantwell 1998), and policies may have a long-term impact. Public policies can influence not only cultural and social conditions, the financial, educational and legal systems and the support of scientific and technological knowledge, but also the organisation of production and innovation within firms and knowledge diffusion through inter-companies links or between firms and research institutes. This strand of literature has important implications for the international diffusion of technology and the potential for technological catching up. In the European context, there are large differences between European countries in terms of technological competencies (Fagerberg et al 1999) and the geographical spread of innovative activity is fairly concentrated, especially at the regional level (Maurseth and Verspagen 1999, Cappelen et al 1999). The (international) diffusion of knowledge is argued not to be easy or 'automatic', so that international technological gaps have remained which, in turn, have led to international differences in economic performance. There is a strong presumption of the geographical concentration of the vast majority of new technologies within the countries that are technologically most advanced. This means that, with the economies of 7

9 agglomeration, while remaining within a paradigm, the most advanced countries would tend to maintain their position over time. In turn, there are national differences in how institutions support particular technologies or sectors. In this context, qualitative analyses by economic historians are especially relevant to provide a better perspective on the long run changing of socio-economic environment and technical change, within which the process of technological catching up may occur. Finally, in this evolutionary context, the technology gap approach of economic growth (Fagerberg and Verspagen 1996) concentrates on the importance of the creation and diffusion of technology for analysing differences in economic growth across countries (Fagerberg 1988) and regions (Fagerberg and Verspagen 1996, Cappelen et al 1999). Typically, this perspective focuses on a better understanding of the critical factors for national (or regional) growth and catch up. Results suggest that the potential for catch up by lagging countries (or regions) is there, but its impact depends on diverging factors (Fagerber and Verspagen 1996, Maurseth and Verspagen 1999, Verspagen 1991). The international process of innovation and diffusion may therefore generate a pattern where countries (or regions) follow converging as well as diverging trends. Sector- and country-specific conditions of technological learning and accumulation are important to explain the conditions of convergence or divergence in international technological activities. The main issue in this paper is the international differentiation and concentration in national technological activity overall, at the level of technological sectors, and their evolution over time. The paper investigates the proposition that national technological positions can be established, tend to persist over time, and matter to explain the differentiation and divergence, or convergence and catching up, of countries. 8

10 3. Data and Methodology Patent statistics are used in this context to investigate the historical pattern of technological convergence or agglomeration in Europe between 1890 and The limits as well as significance of patent statistics as an indicator of technological innovation, as well as their main applications, have been investigated in the literature and will not be repeated here (Soete and Wyatt 1983, Pavitt 1988, Acs and Audretsch 1989, Griliches 1990, Archibugi 1992). In particular, patent data remain one of the most accessible, detailed and historically reliable internationally comparable indicators of innovative activity across technological sectors. This paper draws upon a US patent database, compiled and developed at the University of Reading that covers patents granted in the US between 1890 and The patent records were collected manually from the US index of Patents and the US Patent Office Gazette between 1890 and From 1969 onwards, the US Patent and Trademark Office (USPTO) has computerised and regularly updated its patent records database. For the purpose of this paper, the US patent database distinguishes the year of grant, the location (or residence) of the research facility or individual responsible for the invention, and the type of technology being created. All European-owned patents granted in the US to all inventors, firms and other institutions (e.g. institutes or universities) have been grouped into four common types of technological activity and organised by country of origin. The US Patent Office provides a detailed and historically consistent technological classification for all patents into some 400 technological sectors. Each patent has subsequently been allocated to one of the four broadly defined sectoral groups of technological activities, consisting of chemical, electrical, mechanical and transport technologies (Annex Table 1). The variation in the propensity to patent the results of innovation activity amongst technological sectors will not influence the analysis since it is conducted at an intra-sectoral level. In each sector and for the total of all four sectors, patents have been allocated to sixteen European countries of origin. The patenting countries of origin comprise fourteen EU countries (all countries before the latest enlargement but Luxembourg 1 ), Switzerland, and Norway. Finally, patents have been calculated for ten consecutive decades, , , , etc. and for the most recent period to avoid random fluctuations. 1 Germany, UK, Italy, France, The Netherlands, Belgium, Sweden, Denmark, Ireland, Spain, Portugal, Greece, Austria, Finland. 9

11 For each sectoral group, the distribution of technological activities among European countries is analysed using both shares of US patenting and patents per capita. National shares of patenting in each period have been calculated to reflect the position of each country relative to all other European countries in the volume of technological activities in absolute terms. While the percentage distribution of patents reflects the size of each country, per capita national patenting points out the countries difference in technological intensity. Historical population figures have been taken from Maddison (1991) and completed for the most recent period with consistent data from OECD (1998) 2. Population data are nevertheless missing for Ireland 3, Spain, Greece and Portugal and those countries are excluded from per capita analyses. The analysis focuses on whether there is a tendency for the inequality of technological capabilities of European countries to increase or decrease throughout the period and how it has evolved. At different times, the degree of geographical concentration of technological activity can be measured by the variance (σ 2 ) of the cross-country distributions of patent shares (or patents per capita). When national patent shares (or patents per capita) are widely dispersed about their mean share, the degree of technological concentration across European countries will be relatively high, i.e. the variance of the cross-country distribution is large. In contrast, when the variance of the distribution of the European national shares of patenting (or patents per capita) is low, the geographical concentration of technological activity will be low, and the activity more widely dispersed across countries. The study of historical patterns of technological agglomeration or convergence requires a dynamic analysis of the cross-country distributions of patent shares (or patents per capita) that would reveal important changes in these distributions over time. For example, it would reveal how the patent shares of the less technologically advanced countries change relatively to those of the most technologically advanced, i.e. whether the tendency for technological activity to agglomerate geographically over time is stronger or weaker than the tendency towards technological convergence. It would also show more particularly the movement of countries up and down the patent shares distributions. The Galtonian regression model is a convenient and relevant methodology to examine those dynamic changes. A detailed discussion on the use of the Galtonian model, its development and its 2 All population figures are adjusted to refer to mid-year. They include all nationals present in the country, armed forces stationed abroad, and merchant seamen at sea. Aliens are included if they are permanently settled. 3 Ireland became independent from the UK in

12 justification can be found in Hart and Prais (1956), Hart (1976a, 1976b, 1994, 1995). Assuming the bivariate normal distribution of patent shares (or patents per capita) for each sector (and for the total of all sectors) at each relevant decade, the Galton model allows the study of changes in the variance of these distributions over time. This approach and the use of the variance to measure inequality on concentration were originally applied to economic problems in the different context of firm sizes (Hart and Prais 1956), incomes and labour productivity (Hart 1976a, 1976b, 1994, 1995), technological specialisation and leadership (Cantwell 1991a, 1991b, 1992, Cantwell and Andersen 1996) and exports specialisation (Dalum and Villumsen 1996). In the present context, the distributions were more closely lognormal than normal and a transformation to logarithms was performed 4. One of the reasons for discrepancies from normality is probably the significant difference among European countries in the levels of technological activities and national patent shares in the US (with a few large countries and a multitude of small ones). Hence, the variance of the logarithms of the European patent shares and patents per capita provides a better measure of the geographical technological concentration 5 and the well-known properties of bivariate normal distributions may be used (Hart and Prais 1956, Hart 1974 and 1976b). Long-term historical regressions are run by estimating cross-country regressions of national shares of patenting (and patents per capita) in the 1930s on equivalent in the 1900s, and in the 1980s on equivalent in the 1930s. The choice of these time periods is primarily dictated by the willingness to eliminate, as much as possible, any structural breaks in the distributions. A long-term analysis should allow for gradual or evolutionary changes in the geographical concentration structure. Simple cross-section regressions of the logarithms of European national shares of patenting at time t (one decade) on the equivalent logarithms of shares at the earlier time t-1 (the initial - earlier decade) are estimated, and similarly for the logarithms of European countries patents per capita. If S it denotes the logarithm of the national share of patenting in the US of the European country 4 To avoid the problem with zero values which occur in the logarithmic transformation, a constant of one was added to patent shares / patents per capita figures. Even with small discrepancies to the lognormal distributions, this will not invalidate the basic model in the dynamic analysis and the lognormal hypothesis could be accepted as a satisfactory approximation (Hart 1976a). 5 The use of the variance to measure inequality has some advantages even without assuming a lognormal distribution, but if lognormality may be assumed, it is the natural measure to use (Hart 1976b, 1974). The variance of the logarithms of the firm sizes has been discussed as a suitable index of business concentration (Hart and Prais 1956). The variance of the logarithms of income is used in Hart 1976a&b. The convergence of countries productivity over time may be measured by the variance of the logarithms of productivity (Hart 1995). 11

13 of origin i at time t, then the assumption of bivariate normality makes sure that the regression of S it on S it-1 is linear and homoscedastic, as in (1): S it = α 1 + β 1 S it-1 + ε it (1) Denoting P it the logarithm of the European country of origin i patents per capita at time t, we have similarly (2): P it = α 2 + β 2 P it-1 + ε it (2) with the stochastic disturbance ε it independent of S it-1 and P it-1 in (1) and (2) respectively. Regressions are estimated for each particular sector and for the total of all sectors. In the equations, the relation between the variance at the two dates is now given by (3): σ 2 t = β2 σ 2 t-1 + σ2 ε (3) The estimated value of the regression coefficients β (noted ˆ! ) could therefore be used to analyse the geographical concentration of technological activity over the two periods, distinguishing for each sector 6. When β=1, the national distribution of patent shares (patent per capita) remains unchanged overall over the period. Where 0<β<1, there is a regression towards the mean in Galton' s sense 7. Less technologically advanced countries on average catch up with the most advanced ones while the leading countries consisting of the major locations of technological concentration tend to slip back. The dispersion pattern of countries in technological activity moves towards the average. It is therefore possible (but not necessary) for the overall degree of technological concentration to decrease, i.e. for countries to converge technologically. The magnitude of (1-β) measures the size of the Galtonian 'regression effect' and will be estimated by (1- ˆ! ). If technologically small countries have significantly caught up with the technological leaders, the regression effect is expected to be significantly greater than zero. The expectation that β>0 relies on the prediction that the pattern of national technological competence and performance tends persist over time since the cumulative and pathdependent nature of technological change. As discussed earlier, countries may evolve and only gradually change their positions over time under certain conditions so that, for example, technologically backward countries may catch up with the leaders. In the case where β<0, the advantaged countries for technological activity in a sector would tend to become disadvantaged, and inversely for disadvantaged countries. Finally, if β>1 6 The intercept α measures and controls for the change in the absolute size of the patent variables. 12

14 technological agglomeration has increased over time and consequently disadvantaged countries with smaller patent shares (or patents per capita) on average tend to become still more disadvantaged (and inversely). A first hypothesis is that technological activity across countries retains at least the same dispersion in any sector, and that technological agglomeration tends to be reinforced over time. The alternative hypothesis is that the distribution of technological activity tends to broaden, reflecting that some countries that have not previously been technological centres are catching up the technological leaders, or more generally that there is a tendency of technological convergence amongst European countries. Two similar but different tests, using a Student's t distribution, were defined to examine this proposition. The first one is a two-sided test of whether β is equal to one, or that technologically advantaged countries do not on average become more advantaged, and inversely more disadvantaged for technologically disadvantaged countries. The second one is a composite hypothesis of whether β is significantly greater than or equal to one, β 1, as we may expect countries that have relatively high levels of technological activities to keep and develop their advantage, while smaller or comparatively disadvantaged countries may fall more and more behind. Over the long run, however, they may be some significant changes in the leadership structure of innovative activity amongst countries, if for example at least some relatively disadvantaged countries catch up with the leaders (if β 1 can be rejected so that β<1). The 95 per cent confidence intervals for β are also constructed to check for the estimates' accuracy. A relatively wide confidence interval would be indicative of a potentially greater degree of mobility between countries up and down the distributions. Therefore, the agglomeration hypothesis β 1 will be accepted if the point estimate β is itself around one or greater. Meanwhile, changes in the overall degree of technological concentration among countries can be measured by changes in the variance of the cross-country distributions over time. The proposition that technological activity has become more geographically concentrated in a few major countries (or alternatively that technological activity has diffused more broadly across countries) can be tested with another related procedure. With ρ denoting the Pearson s correlation coefficient between the logarithms of national patent shares (or 7 Galton (1989) found that the height of the children of unusually tall or unusually short parents tends to move towards the average height of the population. In his term, extreme values had to regress towards the mean height. 13

15 patents per capita) at times t and t-1 and defined as ρ 2 = 1 (σ 2 ε/σ 2 t), it will be seen that equation (3) can be reduced to (Hart and Prais 1956, Hart 1976b): σ 2 t /σ2 t-1 = β2 /ρ 2 (4) The variance, and therefore the degree of technological geographical agglomeration, will increase if the regression coefficient is greater than the correlation coefficient (β 2 >ρ 2). The agglomeration hypothesis here is that ˆ! must be greater than the estimate of the correlation coefficient, ˆ!, or similarly ˆ! / ˆ! >1. If there is convergence and the degree of technological concentration falls over time σ 2 t will be smaller than σ2 t-1 ( ˆ! < ˆ! ). The dispersion of the distributions is unchanged if ˆ! = ˆ!. ˆ! is an inverse measure of the degree of movement of countries up and down the patent shares or patents per capita distributions. The lower is ρ, the greater is the mobility of the relative technological position of countries. The magnitude of (1 ˆ! ) therefore measures the size of what is called the 'mobility effect'. If the correlation is strong, then existing leading countries have consolidated or at least preserved their position over the period, the mobility effect is weak. The phenomenon of geographical agglomeration, or convergence, of technological activity over time is analysed by decomposing it into a regression effect and a mobility effect. This decomposition and the interpretation of its underlying mechanisms has been compared with the concepts of β- versus σ-convergence introduced by Barro and Sala-i- Martin in the growth literature (Hart 1995, van Ark and Crafts 1996). Thus, a reduction in the dispersion among countries in terms of their technological performance (σconvergence) can be decomposed into a regression effect (β-convergence) where the primary concern is the relationships between the leading and other countries on average, and a mobility effect where the concern is more focussed on the degree of overtaking. It is possible for the variance to increase (β/ρ>1) while the regression effect suggests it is falling (β<1), by analogy to have a β-convergence and σ-divergence. The degree of technological concentration will increase if the measured mobility effect outweighs the estimated regression effect, i.e. (1 ˆ! ) > (1- ˆ! ). It may be that some countries have caught up but that the overall dispersion has widened. This may be discussed in relation to different notions of convergence where countries by large may converge towards the mean, while still some (leading) ones are leading their way ahead (Fargerberg and 14

16 Verspagen 2002). Inversely, it suffices the regression effect to exceed the size of the mobility effect to induce a decrease in the dispersion, even if the regression effect is weak. In other words, β-convergence is necessary, but not sufficient for σ-convergence. However, if β>1 inequality increases, since ρ 1. Finally, the use of a Galton model does not escape to the classical statistical problem of errors-in-variables. The least-squares logarithmic regression estimate of β is open to the objection that errors or transitory component procedures downward bias estimates (for a discussion see: De Long 1988, Hart 1976a, 1994 and 1995). A period of five years or more may be regarded as long enough to observe some regression taking place. However, Hart (1994, 1995) points out that the use of an alternative technique to ordinary least squares to estimate β does not necessarily provide better results and that the Galton model avoids regression fallacies or traps. Some limitation of the Galton model is however that it excludes from the analysis births, deaths, mergers and demergers of countries. Such changes in the population of our European countries could not be taken into account, even though they are acknowledged to be important and to have a likely significant effect on the statistics of patent shares. More significantly, there were the demerger of the former Austro-Hungarian Empire and the demerger/merger between West and East Germany. In spite of some limitations, the insights into changes in technological agglomeration provided by the simple Galtonian regression model have nevertheless been proved to be very useful. 15

17 4. Agglomeration, convergence or catching up in technological activity across European countries The extent of continuity in the concentration and structure of national technological activities in Europe is analysed. If the hypotheses cannot be rejected, it means that the locational concentration of technological activity has risen, and that there is no evidence in favour of the technological catching up (and convergence) hypothesis, at least within the countries sample. Variances of the distributions are first presented in Tables 1-2 for each decade from the 1890s to the 1980s and for the early 1990s. Overall technological concentration patterns are weaker, variances are smaller, in per capita distributions than in patent shares. The cross-country distributions in terms of patent shares are more unevenly spread between countries, promoting a more concentrated pattern, than in terms of patents per capita. Technological activities are highly concentrated in both Germany and the UK, on average accounting for more than 50 per cent of the total European share of US patenting. At the other extreme, many of the sample s countries are in the lower tail of the distributions, holding relatively small shares of patents in the US and staying relatively small over time. At the beginning of the century up to the late 1920s, there is a decline in variance so that technological activity diffused, although this is more ambiguous in terms of patents per capita and for the chemical sector. During the period of the Great Depression in the 1930s and the Second World War, there is some evidence of a relative widening of the average technological gaps between European countries. In the 1950s the variance declines steadily, supporting the hypothesis of technological diffusion or catching up among European countries. A moderate tendency towards convergence can be found in the 1960s, though some international divergence may be found in both the chemical and transport sectoral sectors. In the 1970s the variance declines still more steadily, supporting the hypothesis of technological diffusion among European countries. This technological convergence seems to slow down during the 1980s, but intensifies again in the early 1990s. [Tables 1-2] These observations may give some evidence to support the technological convergence hypothesis within the group of European countries over the twentieth century. However, if 16

18 there has been a convergence and catching up process among European countries, it has not been a continuous and smooth process as may have been predicted by the traditional neoclassical growth literature. The technological catch up process was certainly irregular as this trend was reversed by the effects of the two World Wars, the Great Depression and the ensuing political and financial troubles. In addition, the catching up has slowed during the 1980s that were also characterised by an economic recession and therefore turbulence and general slow-down in economic performance. Then, the mid-1980s corresponds to the slowdown in several countries of resources devoted to research and development (Archibugi and Pianta 1992). These observations are broadly consistent with earlier conclusions on the convergence of OECD productivity levels. The pattern of convergence of those relatively advanced industrialised countries during the twentieth century is usually divided into four periods (e.g. Abramovitz 1994, Baumol 1994). Between roughly 1870 and 1929 there was convergence among maybe the top eight economies in the group of the leading industrial countries. There was however little if any catch up towards the productivity leader, the US. During the period of the Great Depression and World War II (WWII) there was a sharp divergence which probably offset most of the previously acquired convergence. During the post-war growth boom there was convergence, a decrease in the variance of productivity levels, as well as catch up, a reduction of the gaps with the US. From 1973, at the end of the growth boom, there was a general slowdown of the convergence process. The regression results are reported in Table 3 for the cross-country distributions of the logarithms of European national shares of patents granted in the US, and Table 4 for the logarithms of patents per capita. Separate regressions were calculated for the four technological sectors and for the two periods investigated 8. Column (2) gives ˆ!, (3) ˆ! and (4) the standard error of the regression coefficient. Column (5) contains the t-test for the significance of the regression relation, and (6) the t-test used for the simple hypothesis that β=1, and composite hypothesis that β 1. The values of ˆ! / ˆ! are shown in the last column. The 95 per cent confidence intervals for β are reported in Tables 5-6. [Tables 3 and 5 around here] 8 Cross-country regressions were also run for the ten decades and separately. Compared with the longer-term periods, the short term values of and were all found to be at higher levels, all being closer to one with lower standard errors, supporting the view that the geographical structure of innovation would evolve only gradually over time. 17

19 The β values are significantly different from zero for all equations. In terms of patent shares (Table 3), the hypothesis of constant geographical structure of technological activity (β=1) could not be rejected. The test on the slope coefficient greater than or equal to one, β 1, is that of the reinforcement of geographical concentration of technological activity over time. This hypothesis could not be rejected at the 1 per cent level for all the regressions. Despite a point estimate ˆ! in the area of one, confidence intervals within which the true value of β may lie suggest some possible convergence, although the minimum values of the 95 per cent confidence intervals for ˆ! do not fall below 0.69 (Table 5). The values of ˆ! are generally high suggesting that the mobility effect is weak. However, following our second hypothesis, ˆ! exceeded ˆ! in all sectors during the second period, from the 1930s to the 1980s, so that ˆ! 2 decreased. Therefore, some convergence may have nevertheless occurred, although, as indicated by the high values of ˆ! / ˆ!, to a fairly moderate degree. [Tables 4 and 6 around here] The data on per capita national patenting in the US shows clearer trends (Table 4). From the beginning of the century until the 1930s, the proposition that β is one or greater cannot be rejected at the 1 per cent level of significance and the degree of geographical concentration increased for the chemical, electrical and transport technological groups. The standard errors of the regression coefficient estimates are nevertheless relatively high (Table 6). The exception is the mechanical sector for which there may be a significant regression effect as indicated by the β value of 0.57 which is significantly below one (Tables 2 and 4). Concerning the second long-term period until the 1980s, the β values were significantly below one, suggesting some tendency of catching up. The 95 per cent confidence intervals around ˆ! fall all below one. The regression effects (1-β) were all significantly greater than zero (around 0.45) to the extent that some technologically less important countries may have caught up with the technological leaders and/or certain leaders may have declined (towards the mean). The mobility effects were also generally higher compared with the patent share figures. However, the regression effects have outweighed the mobility effects so that there has been an overall convergence within the European geographical concentration structure. 18

20 The results for the second period in particular suggest a trend towards the overall convergence of national technological capabilities and a technological catch up process. Archibugi and Pianta (1992) also find national-level convergence among advanced countries for the 1970s-80s, using a variety of science and technology indicators (e.g. R&D as a percentage of GDP, published articles and citations). The decrease in variance is more pronounced in the pattern of patents per head (Table 2 and Table 4, col.(7)), where sectors have experienced decreases from 27% (in the transport sectoral group) up to 40% (in the electrical sectoral group), compared with that of patent shares (Tables 1 and 3: less than 10% for all sectors). However, high values of ˆ! and ˆ!, in particular for regressions on (the logarithms of) European shares of patenting, suggest a high degree of stability in the relative positions of countries over time as a result of technological cumulative mechanisms. The results are consistent with the findings of post WWII convergence and catch up, and subsequent weakening of this tendency (though usually discussed from an earlier period in the early/mid 1970s). The literature provides several interpretations of the particular or new factors that may have facilitated and encouraged convergence among the world s leading countries after WWII and particularly among European countries (Crafts and Toniolo 1996). According to Abramovitz (1986), the post WWII decades quite exceptionally gathered the main elements required for a catching up process. There were large technological gaps, enlarged social competencies and conditions favouring rapid realisation of potential. A process of recovery from the war, with the stimulus of the Marshall plan and reconstruction, has been a period of sustained investment and technological diffusion from the US as well as growth in international trade and macroeconomic stability (Maddison 1991, 1994). For those countries that are in position to benefit from technology transfer, increasing pace of the convergence process also arises from increases in the process of internationalisation and improvements in means of communication. This post-war boom has sometimes been described as the exploitation of once-for-all opportunities that would eventually be eroded (Maddison 1994). The observation that technological catching up may have weakened in the more recent period is consistent with the notion that the strength of catching up is proportional to the technological gaps between countries. However, some developments may persist in the future such as higher investments in R&D and education, the increasing 19