Technology, talent, diversity and the wealth of European regions Rutten, Roel; Gelissen, John

Tilburg University Technology, talent, diversity and the wealth of European regions Rutten, Roel; Gelissen, John Published in: European Planning Studies Document version: Early version, also known as pre-print Publication date: 2008 Link to publication Citation for published version (APA): Rutten, R. P. J. H., & Gelissen, J. P. T. M. (2008). Technology, talent, diversity and the wealth of European regions. European Planning Studies, 16(7), 989-1010. General rights Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. - Users may download and print one copy of any publication from the public portal for the purpose of private study or research - You may not further distribute the material or use it for any profit-making activity or commercial gain - You may freely distribute the URL identifying the publication in the public portal Take down policy If you believe that this document breaches copyright, please contact us providing details, and we will remove access to the work immediately and investigate your claim. Download date: 12. Feb. 2018

CEPS316544 Queries Roel Rutten & John Gelissen Dear Author Please address all the numbered queries on this page which are clearly identified on the proof for your convenience. Thank you for your cooperation Q1 Please check article title. Q2 Oerlemans et al., 2007. Please provide reference. Q3 1998 or 1990? Please check year. Q4 Cooke, 2002. Please provide reference. Q5 Please check spelling of author s name. Q6 Is a range required here? Q7 Please cite reference in the text. Q8 Please confirm citation of the tables in the text. Q9 Please provide the page range for the chapter. Q10 Please provide description for in Table 4.

CEPS316544 Techset Composition Ltd, Salisbury, U.K. 6/7/2008 European Planning Studies Vol. 16, No. 7, August 2008 5 EUROPEAN BRIEFING 10 Technology, Talent, Diversity and the Wealth of European Regions Q1 15 20 ROEL RUTTEN & JOHN GELISSEN Department of Organization Studies, Faculty of Social and Behavioural Sciences, Tilburg University, Tilburg, The Netherlands, Department of Methodology and Statistics, Faculty of Social and Behavioural Sciences, Tilburg University, Tilburg, The Netherlands (Received B; accepted B) 25 30 ABSTRACT In this article, we test the creativity and diversity hypothesis of Richard Florida for European regions. Florida argues that the level of regional economic development depends on the levels of technology, talent, and tolerance that regions harbour. Tolerance, in this case, is a measure for diversity of lifestyles, the creativity that results from it and population s openness towards non-traditional lifestyles. Using data for 94 European regions we investigate whether differences in creativity and diversity are a good predictor of differences in regional wealth in additive and multiplicative regression models. The results indicate that regional differences in diversity are directly related to differences in regional wealth. Moreover, we find that the synergetic effect of technology and talent on the level of regional wealth depends on the degree of diversity that resides within regions. Our findings support the idea that creativity and diversity deserve a more prominent place in economic geography. 35 40 Introduction One of the more thought-provoking contributions to the field of economic geography in recent years comes from Richard Florida. In his The Rise of the Creative Class (Florida, 2002), he argued that differences in economic development between regions should be explained not only in terms of their differences in innovation but that regional differences in creativity and diversity should be taken into account also. Although appealing to policy-makers, thus far, Florida s theory has met with little enthusiasm in the field of 45 Correspondence Address: Roel Rutten, Department of Organization Studies, Faculty of Social and Behavioural Sciences, Tilburg University, P.O. Box 90153, 5000 LE Tilburg, The Netherlands. Tel.: þ31 13 466 2164; Fax: þ31 13 466 3002. Email: R.P.J.H.Rutten@uvt.nl ISSN 0965-4313 Print=ISSN 1469-5944 Online=08=070989 22 DOI: 10.1080/09654310802163785 # 2008 Taylor & Francis

990 R. Rutten & J. Gelissen 50 55 60 65 70 75 economic geography (Gleaser, 2005). Part of the blame for this falls on Florida himself as his theory is little more than a coherent train of thought on the role of creativity and diversity in the economic development of regions. He has not provided the scientific community with well-defined variables, corresponding indicators, and testable hypotheses. On the contrary, Florida uses different variables and indicators to measure creativity and diversity in different publications (see, Florida, 2002, 2005; Florida & Tinagli, 2004). Nonetheless, Florida s train of thought is compelling and it may comfortably be seen as an elaboration of the milieu of innovation literature (e.g. Lagendijk, 2006; Lambooy, 2005; Oerlemans et al., 2007). Moreover, Florida s empirical work has provided some convincing evidence Q2 in support of his theoretical arguments. The objective of this paper is to provide some theoretical elaboration of Florida s argument and to put it to an empirical test in order to see what value Florida s work has for the field of economic geography. Our main research question is as follows: To what extent do regional differences in creativity and diversity explain differences in wealth between European regions? This article is structured as follows. In the next section, we outline the creativity thesis as developed by Richard Florida. This section provides the theoretical basis of the article where we argue that Florida s argument may be seen as an elaboration of the milieu of innovation theory. According to Oerlemans et al. (2007), the central argument of the milieu of innovation approach is to explain what external conditions contribute to the... adoption of innovations by existing enterprises. Innovative environments are seen as the breeding grounds of innovation and innovative entrepreneurs (p. 170). Essentially, Florida follows the same line of thinking, arguing that cosmopolitan, diverse, and tolerant regions are the best breeding grounds for creativity and, hence, innovation. Next, we discuss the causal mechanisms that link regional differences in Technology, Talent, and Diversity to regional differences in Wealth. In this section we also present our hypotheses. The fourth section presents the operationalizations of our variables, while the fifth section introduces the regions that are included in the empirical study. The sixth and seventh sections outline our empirical approach and method in relation to the sometimes imperfect data that we had to work with. The results of the empirical analyses are discussed in the eighth section. Finally, we reach our conclusions and discuss the implications of our findings for the field of economic geography. 80 85 90 The Creativity Thesis A key argument in Florida s work is that innovation is fuelled by creativity. Creativity is at the basis of new knowledge, new technologies, and their application in new products and services. In Florida s view, creative people are a crucial element of modern, innovative economies. The creativity argument adds to the existing literature on innovation and regional development in the following way. The existing literature focuses on innovation, both as a process and as an outcome, to explain regional development. In general, more innovative regions enjoy a higher level of economic development. This is because innovation is argued to contribute to competitiveness and competitiveness, in turn, has a strong influence on economic development (Porter, 1990; Best, 2001). Entering creativity Q3 takes this argument a step further. Florida (2002) asks himself: What is innovation and where does it come from? In a Schumpeterian way he argues that innovations result from making new combinations, more particular, new combinations of knowledge (cf. Lee et al., 2004; Thanawala, 1995). This making of new combinations, essentially, is a

Technology, Talent, Diversity and the Wealth of European Regions? 991 95 100 105 110 115 120 125 130 135 creative act. Whereas most scholars characterize the current stage of capitalism as knowledge based, or the knowledge economy (e.g. Cooke, 2002; Rutten, 2003), Florida calls it Q4 the creative economy. He sees human creativity as the defining feature of economic life.... [New] technologies, new industries, new wealth and all other good economic things flow from it (Florida, 2002, p. 21). Whereas mainstream economic geography literature speaks of knowledge and innovation in general terms, Florida is more specific about the underlying mechanism: human creativity. In particular he argues that: [Human] creativity is multifaceted and multidimensional. It is not limited to technological innovation or new business models. It is not something that can be kept in a box and trotted out when one arrives at the office. Creativity involves distinct kinds of thinking and habits that must be cultivated both in the individual and in the surrounding society. (Florida, 2002, p. 22) Put differently, in Florida s view successful economies are those economies that foster a creative ethos. It differs from the work ethos of the industrial economy in that it puts more emphasis on individuals whose job it is to think and act beyond the everyday routines of work and organization, i.e. to be creative (cf. Reich, 1992). Whereas the argument of the knowledge-based economy is that human capital in general has become crucial for innovation, Florida argues that one form of human capital in particular, i.e. creative capital, contributes most to innovation and economic development. Creative capital is an attribute of what Florida calls the creative class. This is a diverse class of consultants, lawyers, academics, artists, researchers, film makers, information technology (IT) specialists, and a score of other professions (Florida, 2002, pp. 327 329). The only thing that these people have in common professionally is that they get paid to solve problems of all sorts using their creativity. Furthermore they have in common that they value nontraditional and non-conformist lifestyles. That is, these people do not stop being creative after they leave their office. Creativity is part of their identity. Hence they enjoy cultural diversity and cultural amenities of all sorts, they are averse to bureaucracy, and they are not strict in separating work from leisure. The creative class is more cosmopolitan than the population average and it is usually concentrated in cities and regions that endorse diverse and cosmopolitan lifestyles. In the words of Florida: Regional economic growth is powered by creative people who prefer places that are diverse, tolerant and open to new ideas. Diversity increases the odds that a place will attract different types of creative people with different skill sets and ideas. Places with diverse mixes of creative people are more likely to generate new combinations. Furthermore, diversity and concentration work together to speed the flow of knowledge. Greater and more diverse concentrations of creative capital in turn lead to higher rates of innovation, high technology business formation, job generation and economic growth. (Florida, 2002, p. 249) One should note, though, that Florida seems to equal creativity with what he calls creative occupations and cosmopolitan and non-traditional lifestyles. It is beyond the scope of this paper to discuss the theoretical validity of this assertion but, arguably, it has several drawbacks. For example, cannot people in other occupations be creative as well? Do

992 R. Rutten & J. Gelissen 140 145 150 155 occupational data really reflect creativity, as certainly not all lawyers, consultants, and artists are blessed with creativity? In addition to creativity, Florida places a strong emphasis on the role of diversity as a driver of creativity. Diversity is a fundamentally social and cultural phenomenon. Those kinds of diversity make cities and regions attractive places for the creative class. Florida s argument with respect to creativity and diversity, thus, is an extension of the innovative milieu approach, which argues that certain conditions favour learning and innovation in regions. In sum, what is exotic about Florida s argument is not so much his approach. It seeks to explain differences in regional economic development on the bases of the differences in the characteristics of those regions, in particular with regard to their levels of creativity and diversity. As argued, this approach makes the region the level of analysis and it can be seen as an extension of the milieu of innovation approach. Florida s emphasis on creativity and diversity is new in the sense that it is missing from the mainstream literature on innovation and regions (e.g. Rutten & Boekema, 2007; Lagendijk, 2006). However, as Florida himself acknowledges, creativity and diversity have not gone completely unnoticed in the field of economic geography. The work of Jane Jacobs (e.g. Jacobs, 1961) has strongly advocated their role in regional economic development. It is where Florida attempts to measure creativity and diversity where his work gets a touch of the exotic. But before we turn to that, we must first take a closer look at the mechanisms behind regional economic development as Florida sees them. 160 165 170 175 180 The Causes for Differences in Regional Wealth Florida summarized the causes of differences in the level of regional economic development as the three Ts: Technology, Talent, and Tolerance. The three Ts capture the essence of Florida s theory, namely the extension of the mainstream literature s emphasis on technology and innovation to creative capital (talent) and diversity (tolerance) in the explanation of regional economic development. However, the T for tolerance, which features prominently in Florida s follow-up publications (Florida & Tinagli, 2004; Florida, 2005), does not entirely seem to cover the concept of diversity as he introduced it in his initial 2002 work. In this paper, we choose to stay as close as possible to Florida s original conceptualization of creativity and diversity, which means that, the appeal of the three Ts notwithstanding, tolerance will not be a separate variable in our study. For Florida, tolerance has always been an indicator for diversity. The argument runs as follows: Non-traditional cultures and lifestyles can only thrive in environments that are open to them. For Florida, tolerance is not an objective in itself, it is an indicator for the openness to non-traditional cultures and lifestyles of cities and regions. More tolerant places will attract more diverse people and, therefore, these places are better breeding grounds for creativity (cf. Florida, 2002, p. 250). In fact, Florida argues that of the three Ts, Tolerance is the most important variable explaining regional economic development in the creative age (Florida, 2005). To measure tolerance, or rather, diversity, Florida developed three indicators; the gay index, the bohemian index, and the melting pot index (Florida, 2002, pp. 333 334). Arguably, the gay index is the most exotic of all of Florida s indices. It measures the number of gays in the population on the assumption that gays represent a non-traditional lifestyle and (sub)culture. A large number of gays in the population signals this particular population s openness or tolerance to gays and, therefore, it may be expected to be open to other non-traditional lifestyles and subcultures,

Technology, Talent, Diversity and the Wealth of European Regions? 993 185 190 195 200 205 210 215 220 225 too (cf. Florida, 2005, p. 60). Despite the criticism that the gay index has provoked in certain (political) circles, from a methodological point of view, it has a compelling logic. Florida s second indicator for diversity concerns the bohemian values. These values represent how the creative class thinks about work, leisure, and society and they represent, in particular, a hedonist, non-conformist, and bureaucracy-averse lifestyle (see, Florida, 2002, pp. 193, 194, and 204). Again, the argument is that when more members of a population hold these bohemian values, the population will be more open to non-traditional lifestyles, since bohemian values themselves represent a non-traditional lifestyle. The role of values in economic development is not new. Earlier work by, for example, Granato et al. (1996) and Harrison and Huntington (2000) already demonstrated their importance. The novelty of Florida s argument lies in the kind of values he stresses, i.e. those captured by tolerance and bohemianism. The melting pot index is the third diversity indicator and it is based on the same logic as the previous two. The melting pot index is a measure for the number of foreign born people in the population. Arguably, these people are likely to have a culture and lifestyle that differs from the traditional culture and lifestyle of the population. More foreign borns, thus, may indicate that a population is more open to non-traditional lifestyles. Moreover, foreign borns directly contribute to diversity in the population. In his follow-up works, Florida focused more on values that reflect tolerance for nontraditional lifestyles than on tolerance and diversity itself. For example, in his 2004 publication he borrows from Inglehart s self-expression values (e.g. Ingleheart & Baker, 2000). Although Florida largely abandoned the earlier concepts, we prefer to keep the emphasis on diversity rather than on values that may reflect tolerance. Ultimately the actual diversity and tolerance in a population, rather than the underlying values, fuel creativity. Therefore, we substitute the T for Tolerance for the D for Diversity. Concerning the other two Ts, Technology and Talent, Florida uses different aggregate indices in different publications, but the underlying indicators remain more or less the same. Technology consistently reflects two things. On the one hand, it reflects the number of patents per capita and, on the other hand, the high-tech industrial inputs and outputs of an economy, such as investments in research and development (R&D). Both indicators are commonly used in the innovation and regions literature and they reflect how good or bad an economy performs with regard to technology development. Technology development, in turn, is widely recognized as an important competence in today s knowledge-based economy (e.g. European Commission, 2003). The final T, for Talent, too, reflects two underlying indicators. The first, human capital, is usually expressed as the percentage of the workforce with a bachelor s degree or higher. The second reflects the creative capital in a population, which Florida measures as the share of creative class occupations in the workforce. In sum, from Florida s initial work, we have derived three independent variables (Technology, Talent, and Diversity), as well as several indicators for each variable. The dependent variable, of, course, is the level of economic development, or Wealth, which can be expressed as the gross domestic product (GDP) per capita. Following the milieu of innovation approach, we assume a straightforward conceptual model where each independent variable has a direct effect on economic development. In addition, and in line with both the milieu of innovation and Florida s own arguments, the independent variables reinforce each other. That is, there is also an interaction effect of the three independent variables on Wealth (see Table 1). Q8

994 R. Rutten & J. Gelissen Table 1. Variables, indicators and data sources 230 235 240 245 250 255 Variable Economic development Technology Talent Diversity Measurements Wealth GDP per capita (source: Eurostat, 2001) Innovation Patents per million population (source: Eurostat, 3-year average 1999 2001) High-tech Private sector R&D expenditure (source: Eurostat, 4-year average 1998 2001 a Human capital Workforce with bachelor s degree or higher (source: Eurostat, 2002) Creative capital Share of KISs in total workforce (source: Eurostat, 2002 b ) Melting pot Percentage non-nationals in the population (source: Eurostat, 2000 c ) (In)tolerance (source: EVS, 1999) On this list are various groups of people. Could you please sort out any that you would not like to have as neighbours? People with a criminal record, people with a different race, left-wing extremists, heavy drinkers, right-wing extremists, people with large families, emotionally unstable people, Muslims, immigrants/foreign workers, people who have AIDS, drug addicts, homosexuals, Jews, Gypsies. Bohemian values (source: EVS, 1999) A composite index based on the following attitudes: How important is God in your life (1 ¼ not at all important to 10 ¼ very important Please tell me whether you think abortion can always be justified, never be justified, or somewhere in between (1 ¼ never to 10 ¼ always) How proud are you to be a [nationality] citizen? (1 ¼ very proud to 4 ¼ not at all proud) I d like you to tell me whether you have actually signed a petition, whether you might do it or would never, under any circumstances, do it Q6 Please tell me whether you think homosexuality can always be justified, never be justified, or somewhere in between (1 ¼ never to 10 ¼ always) Generally speaking, would you say that most people can be trusted, or that you can t be too careful in dealing with people? (1 ¼ most people can be trusted to 2 ¼ can t be too careful) 260 a 1999 for the Flemish provinces of Belgium. b 2000 for Sweden, 2001 for the Netherlands. c 2001 for Spain, 2003 for Germany. 265 270 The Variables The measurements of the variables come from two sources of data, i.e. Eurostat and European Value Studies (EVS). Eurostat is the statistical bureau of the European Union (EU) and maintains a large database on the social and economic situation in the countries and regions of the EU. Eurostat collect their data from the various national statistical bureaus in the EU but harmonizes them to enable comparison. The EVS data are periodically gathered through a large-scale survey and measure how the population of the various European countries think about a variety of social, economic, and cultural issues. The results are comparable between countries since all participating countries use the same

Technology, Talent, Diversity and the Wealth of European Regions? 995 275 280 285 290 295 300 305 310 315 questionnaire. Though collected at the national level, EVS data are available on the level of regions, too, and, fortunately, EVS uses the same regional subdivision as does Eurostat. Therefore, both databases are compatible. Once collected, we converted the data for all variables to indices with the EU average set at 100. This facilitates comparison between different variables and allows all measurements to be used in statistical analyses. Moreover, it is the same technique that Florida used in his initial 2002 publication. The dependent variable, the level of regional economic development, is expressed as the wealth of a region, under the assumption that more economically developed regions are also wealthier. A commonly used indicator for wealth is the GDP per capita. This figure is available from Eurostat. As to the independent variables, the two indicators for the first T, Technology, also come from Eurostat. In line with Florida (2002), the first technology indicator, innovation, was measured as the number of patents per million inhabitants for each region. The second technology indicator, high-tech, was measured as the investments in R&D of a region s private firms as a percentage of that region s GDP. This indicator ignores public R&D expenditure but has as benefit that it is available for the majority of EU regions. We constructed a Technology index which is the linear principal component of these two indicators. Cronbach s alpha for this index was 0.88 at the regional level. The second T, Talent, also has two indicators and they, too, were measured using Eurostat data. First, human capital, was measured in the same way as did Florida, that is, as the percentage of the workforce with a bachelor s degree or higher. Creative capital proved more difficult to measure. Whereas Florida measured creative capital as the share of creative occupations in the total workforce, similar data on occupations are unavailable for Europe. Instead, we used the share of knowledge-intensive services (KISs) occupations in the total workforce. KISs, on the one hand, include many more occupations in the services sector than does Florida s list of creative occupations. On the other hand, it excludes, for example, research professions in the manufacturing industry that Florida included. However, the KISs are the best available proxy for creative capital in European regions. On the basis of these two indicators, a Talent index was created using principal component analysis. Cronbach s alpha for this index was 0.74. We are aware that, for example, Gleaser (2005) argued that the effects of creativity and Q5 education should be separated analytically. However, in our sample education and creativity are relatively highly correlated (Pearson s r ¼ 0.623, p, 0.001). Although this finding might not seem to be too alarming at first sight, we also checked how these two variables behaved in a multiple regression analysis. Specifically, we regressed wealth on education, KISs, population density and the dummy variables for countries. Collinearity diagnostics indicated a square root of the Variance Inflation Factor of 3.446 for KISs. From this it is apparent that the precision of the regression estimate for this variable would clearly suffer from collinearity (Fox, 1991). Consequently, we cannot distinguish precisely between the effects of the two variables. To deal with this problem, we decided to combine the two measures in one construct. The third independent variable, Diversity, has three indicators, the first of which, Melting Pot, was measured using Eurostat data. Melting Pot expresses the percentage of non-nationals in the population, which is similar, though not identical, to Florida s foreign borns. For the second diversity indicator, tolerance, Florida used the number of gay people in the population. Unfortunately, such a figure is not available for Europe. However, the European Values Survey 1999 included measurements of how people feel

996 R. Rutten & J. Gelissen Table 2. Descriptive statistics of the research variables (N ¼ 94) Variable Mean Standard deviation Minimum Maximum 320 325 Regional wealth 97.160 27.154 53.52 194.15 Regional population density 2001 315.323 618.160 3.9 3827.5 Technology index 19.651 20.152 0 100 Talent index 43.908 21.867 0 100 Tolerance 58.286 18.883 0 100 Melting Pot index 87.449 72.463 7.99 438.75 Bohemianism index 47.783 21.687 0 100 330 335 340 345 350 about having as their neighbours gays, gypsies, Jews, Muslims, and several other subgroups in the population. To construct a Tolerance index, we first counted the number of times that each respondent agreed to having a member of these subgroups as their neighbour. Subsequently, we calculated the average score across regions. The third diversity indicator, bohemian values, presented the greatest challenge. Florida again resorts to occupation data, arguing that certain artistic professions are closely associated with bohemian values. As before, such data are unavailable for Europe but, again, the EVS data offer an alternative. Florida associates bohemian values with a hedonist, non-conformist, and bureaucracy-averse lifestyle. In the European Values Study, peoples attitudes towards the various components of Florida s bohemian values were measured: We made a composite measure of these attitudes by submitting them to a principal component analysis. Cronbach s alpha for this index of bohemianism was 0.76. We investigated the possibility of constructing an overall index for Diversity, but principal component analyses indicated that the indices of Melting Pot, Bohemianism and Tolerance were insufficiently correlated to form one principal component. Consequently, in the following analyses we include the indices for the dimensions of Diversity separately. The reason why these three indicators are not highly correlated may be explained as follows. Melting Pot is a purely demographical characteristic of a population and does not necessarily say something about the values that a population holds. These values, i.e. Bohemianism and Tolerance, may be held by the affluent and cosmopolitan part of the population only and my, therefore, not be reflected in a score for the population average. Moreover, Tolerance in this paper is actually measured as the absence of intolerance, which may not be the same as tolerance (see also the section Discussion and Conclusion ) (see Table 2). Q8 355 360 The Regions For practical reasons, the analysis in this paper is limited to the regions of the 15 old EU member states, since the availability of data is least problematic for these countries. Furthermore, the analysis is based on the statistical subdivision of the EU in NUTS-regions. 1 In his initial 2002-work Florida used agglomerations or city regions as his level of analysis. The NUTS-2 level regions come closest to this level, although one should keep in mind that considerably differences in size and population do exist between the NUTS-2 regions in the EU. For example, the whole country of Denmark is considered a NUTS-2 region, as is the city of Vienna. However, the population of the country Denmark is comparable to

Technology, Talent, Diversity and the Wealth of European Regions? 997 365 370 that of the NUTS-2 region Madrid. Moreover, Florida, too, had to content with varying scales of the US city regions. Finally, the reported effects were controlled for population density but no statistically significant findings emerged, which suggests that region size does not effect the outcomes of this study. Among them, the 15 old EU member states count 207 NUTS-2 regions. 2 We decided to include only those regions in the analysis for which we have data on all indicators for our four variables (see, Table 1). This reduced our sample to exactly 94 NUTS-2 regions. Most notably, no UK-regions are left in our sample since Technology data for the UK are available on the level of the larger NUTS-1 regions only. 375 380 385 390 Empirical Approach The purpose of the empirical analysis in this paper is to test whether Florida s key assumption is correct, which holds that differences in Wealth between regions are best explained on the basis of the differences in Technology, Talent, and Diversity between regions. In empirical terms this means that we can estimate three models that explain the differences in the level of economic development (Wealth) between regions. The first model accounts for the differences in Technology only. This model represents the traditional view in economic geography. The second additive model accounts for both Technology and Talent. It may be seen as an elaboration of the traditional model that few scholars in this field would disapprove of, since the innovative milieu literature is open to the role of human capital in economic development. The third additive model includes all three independent variables: Technology, Talent, and Diversity. This model represents Florida s theory and includes the elements that have aroused most controversy in the economic geography community. If Florida is correct, the third model should yield the best explanation of the differences in Wealth between the regions in our sample. That is, the third model should have the highest explained variance (R-squared). Finally, for exploratory purposes we also investigate the extent to which the three presumed determinants of differences in regional wealth in interaction relate to the dependent variable. 395 400 405 Method In order to investigate the extent to which Technology, Talent and Diversity independently or in combination contribute to the wealth of regions, we apply ordinary least squares (OLS) regression. In addition, we report jackknifed estimates of these models. The jackknife is one of several internal replication methods, next to bootstrapping and cross-validation. It involves omitting one case or a subset of cases of a fixed size at a time and conducting separate analyses for each configuration. Then, the variation in the estimators, denoted by (such as regression weights) is assessed. The specific steps in jackknifing are as follows (Mooney & Duval, 1993, pp. 23 24): (1) Divide the sample into g exhaustive and mutually exclusive subsamples of size h, such that gh¼n. (2) Drop out one subsample (usually of size h¼1) from the entire original sample. Calculate ^u 1 from that reduced sample of size (g21)h¼n2h.

998 R. Rutten & J. Gelissen (3) Calculate the pseudo value, ~u g, from this ^u 1 by weighting as follows: ~u g ¼ g ^u ðg 1Þ ^u 1 410 (a) Repeat steps 2 and 3 for all g subsamples, yielding a vector of g ~u g values. (b) Take the mean of these pseudo values to yield the jackknifed estimate of u, ~u: ~u ¼ g 1 X ~u g 415 420 425 430 We additionally report the jackknifed estimates for two reasons. In the first place, it allows assessing the stability of our regression results by determining confidence intervals about the estimator, because the jackknifed estimator is postulated to be normally distributed (Tukey, 1958). One can determine the confidence intervals by dividing the jackknifed estimator by its associated standard error to obtain a Student t value with df¼number of pseudo values 1. Secondly, because of the nature of our sample (European regions) we cannot rule out the possibility that specific regions substantially affect the results of the regression analyses by acting as influential data points. By omitting one observation (i.e. region) at a time and re-estimating the model, we can see the impact of any outliers in our analysis by inspecting the pseudo-values. Thus, the jackknife approach makes use of all of the data in a particular data set while eliminating bias related to the inclusion of atypical cases (Ang, 1998). 3 In particular, this procedure involved the following steps. First, we calculated the standard deviation of the pseudo-values of each independent variable on the basis of a Jackknifed regression model which included all independent variables. Next, we examined those cases of which the pseudo value was larger than two times the standard deviation of the pseudo values of the full data set. In other words, we examined which regions had a greater-than-average effect on the overall ^u. This approach showed that the omission of several regions would affect substantially the effects of the OLS regression analyses. 4 435 440 445 450 Results Direct Relations between Technology, Talent and Diversity and Regional Wealth The first question to be addressed is: to what extent do indicators of Technology, Talent and Diversity predict regional differences in Wealth, and what is the relative contribution of the individual indicators of Diversity to the explained variance in Wealth? To answer this question, we estimated three additive regression models, which are hierarchically ordered. This basic model (Model 1) predicts differences in regional wealth by Technology only. Model 2 includes our indicator of Talent while Model 3 represents Florida s complete model and also includes the three Diversity indicators, Tolerance, Melting Pot and Bohemianism (non-conformism). The reported effects of all variables are controlled for regional differences in population density and country differences. The results of these models are presented in Table 3. To allow closer inspection of the regression results Q8 we also provide the reader with so-called added-variable plots (see Figures 1a 1e). These plots have slopes equal to the corresponding partial OLS regression coefficients of Model 3; the numbers in the plots represent region codes (see Appendix).

Technology, Talent, Diversity and the Wealth of European Regions? 999 455 460 465 470 475 480 485 490 495 Table 3. Summary of hierarchical jackknifed regression analysis for variables predicting regional wealth (N ¼ 94) Model 1 Model 2 Model 3 OLS Jackknife OLS Jackknife OLS Jackknife Technology index 0.663 0.630 0.524 0.452 0.300 0.248 (0.349 0.977) (0.045 1.126) (0.220 0.828) (20.002 0.906) (0.018 0.583) (20.080 0.575) Talent index 0.787 0.896 0.228 0.259 (0.344 1.230) (0.397 1.394) (20.202 0.659) (20.253 0.771) Tolerance 0.042.065 (20.177 0.262) (20.210 0.341) Melting Pot index 0.273 0.295 (0.161 0.385) (0.128 0.461) Bohemianism index 0.668 0.670 (0.282 1.053) (0.163 1.176) Constant 70.320 70.490 61.446 60.581 10.569 6.558 (52.911 87.729) (56.156 84.824) (44.431 78.461) (38.721 82.442) (239.755 52.605) (230.627 43.742) R-squared 0.560 0.561 0.621 0.620 0.744 0.750 R-squared adjusted 0.482 0.482 0.548 0.546 0.683 0.689 Note: Reported effects are controlled for country differences and differences in regional population density in 2001; estimates of the control variables are available from the authors upon request. R-square difference (OLS) Model 2 Model 1 ¼ 0.061 F(1,78) ¼ 12.523, p, 0.01. R-square difference (OLS) Model 3 Model 2 ¼ 0.124 F(3,75) ¼ 12.107, p, 0.01. Significant at 5%; 95% confidence intervals in parentheses.

1000 R. Rutten & J. Gelissen 500 505 510 515 520 525 530 535 540 Figure 1. Added variable plot of regression of regional wealth on (a) Technology index, (b) Talent index, (c) Tolerance index, (d) Melting Pot index and (e) Bohemianism (see Appendix for region codes)

Technology, Talent, Diversity and the Wealth of European Regions? 1001 545 550 555 560 565 570 Figure 1. Continued 575 580 585 Model 1 shows that regional differences in the level of technological development are directly related to regional differences in wealth: as expected, the higher the level of technology that resides in a region, the better it performs economically. This effect is corroborated by both the OLS estimates and the jackknifed estimates. Turning to the results of Model 2, in which we control for regional differences in the degree of Talent (human capital and creative capital), we see that the OLS estimates indicate that both Talent and Technology are directly related to regional differences in Wealth. The jackknifed estimates, however, indicate that only Talent is directly related to regional differences in Wealth with regional technological differences held constant. An interesting finding is that when we hold constant for the indicators of Diversity (see Model 3) Tolerance, Melting Pot and Bohemianism the latter two indicators exhibit relatively strong direct effects on Wealth. In particular, the results indicate that as the

1002 R. Rutten & J. Gelissen 590 595 600 605 percentage of non-nationals in a region is higher, wealth is higher, holding the other indicators constant. Moreover, the more regions can be characterized as non-traditional (Bohemian), the better they perform economically, holding other indicators constant. Comparison of standardized coefficients from Model 3 (not reported here) showed that the percentage of non-nationals in the regional population is the strongest predictor of regional wealth. Finally, adding the soft indicators for Diversity significantly increases the predictability of Wealth: R-squared increases from 62% to 75% (R-squared change ¼ 0.124, p, 0.01). Note furthermore that the previously found direct relationships between Technology and Talent on the one hand, and Wealth on the other hand are markedly reduced in Model 3. In particular, the OLS estimate of the effect of Technology drops from 0.524 to 0.300 and the effect of Talent drops from 0.787 to 0.228 and only the effect of Technology now reaches the conventional level of significance. According to the jackknifed model neither Technology nor Talent is directly related to Wealth, once the indicators for Diversity are controlled. The fact that once the effects of Diversity differences are controlled, the effects of Technology and Talent on Wealth partially (OLS-estimate) or completely (Jackknifed estimate) disappear suggesting that the latter associations are largely spurious: Presumably, Technology and Talent are related to the level of regional economic development because higher levels of diversity engender more Technological advancement and the nurturing of Talent in a region on the one hand, and directly improve regional economic development, on the other hand. 5 610 615 620 625 630 Interactions Effects between Technology, Talent and Diversity on Regional Wealth The second question we seek to answer in this contribution is whether Technology, Talent and Diversity in combination will significantly increase the predictability of Wealth in comparison to the traditional model, which predicts Wealth on the basis of Technology and Talent only. In other words, here we are concerned with the interaction or synergetic effects of Technology, Talent and the indicators of Diversity on Wealth. To examine these interaction hypotheses, we estimated three models, where each model refers to the interaction between Technology and Talent, on the one hand, and the separate indicators for Diversity: Tolerance, Melting Pot and Bohemianism, on the other hand. Every model includes the main effects, all two-way interactions and the three-way interaction between the variables concerned. Again, we report both OLS and Jackknifed results of these models, which are reported in Table 4. Q8 We are mainly concerned with the existence of significant three-way interactions between Technology, Talent and each dimension of Diversity. The findings indicate that the three-way interaction effect between Technology, Talent and Melting Pot (Model 5) is most consistent, because the estimate of this effect is significant according to both the OLS and the jackknifed approach. The estimates of Models 4 are less consistent (Model 4: three-way interaction only significant according to the OLS model) and the three-way interaction in Model 6 is clearly not significant. Therefore we refrain from further discussion of these models, and limit our discussion to the findings of Model 5. As an aid to the interpretation of the significant three-way interaction of Model 5, we provide two (three-way) interaction plots (see Figure 2a and 2b) illustrating the slopes of Wealth on Melting Pot, at high and low values of Technology and Talent. The low and high values are defined as one standard deviation below the sample mean and

Technology, Talent, Diversity and the Wealth of European Regions? 1003 635 640 645 650 655 660 665 670 675 Table 4. Summary of OLS and jackknifed regression analyses predicting regional wealth with two- and three-way interaction effects (N ¼ 94) Model 4 Model 5 Model 6 Independent variables OLS Jackknife OLS Jackknife OLS Jackknife Main effects Technology index 0.831 0.860 0.572 0.457 0.746 0.473 (0.410 1.252) (0.308 1.411) (0.220 0.925) (0.082 0.833) (0.358 1.135) (20.225 1.170) Talent index 0.508 0.499 0.476 0.535 0.397 0.639 (0.087 0.928) (20.028 1.026) (0.066 0.886) (0.058 1.011) (20.120 0.914) (20.245 1.523) Tolerance 0.130 0.288 (20.243 0.504) (20.240 0.815) Melting Pot index 0.232 0.249 (0.114 0.350) (0.061 0.437) Bohemianism index 0.452 0.355 (20.079 0.982) (20.770 1.479) Two-way interactions Technology Talent 20.021 20.026 20.018 20.015 20.004 0.011 (20.035 to 20.008) (20.046 to 20.006) (20.036 ;0.001) (20.037 0.008) (20.026 0.018) (20.022 0.044) Technology Tolerance 20.009 20.002 (20.031 0.013) (20.031 0.028) Talent Tolerance 0.029 0.027 (0.013 0.044) (0.002 0.053) Technology Melting Pot 20.008 20.006 (20.013 to 20.002) (20.012 to 20.001) Talent Melting Pot 0.003 0.003 (20.002 0.009) (20.004 0.011) Technology Bohemianism 20.014 0.025 (20.036 0.008) (20.071 0.021) Talent Bohemianism 0.008 0.008 (20.000 0.001) (20.017 0.033) Three-way interactions Technology Talent Tolerance 0.001 0.001 (0.001 0.002) (20.000 0.002) Technology Talent Melting Pot 0.149 a 0.184 a (0.030 0.269 ) (0.000 0.364) Q10 Technology Talent Bohemianism 0.000 0.001 Continued

1004 R. Rutten & J. Gelissen 680 685 690 695 700 705 710 715 720 Table 4. Continued Model 4 Model 5 Model 6 Independent variables OLS Jackknife OLS Jackknife OLS Jackknife (20.000 0.001) (20.001 0.002) Constant 119.100 122.924 96.865 92.825 108.493 106.287 (97.077 141.123) (95.391 150.457) (72.508 121.223) (66.594 119.055) (82.426 134.560) (72.223 140.352) R-squared 0.730 0.735 0.746 0.755 0.673 0.673 R-squared adjusted 0.655 0.662 0.677 0.686 0.583 0.582 Note: Independent variables are mean-centred. Reported effects are controlled for country differences and differences in regional population density in 2001; estimates of the control variables are available from the authors upon request. R-square difference (OLS) Model 4 two-way interaction effects model with Tolerance þ control variables ¼ 0.040 F(1,73) ¼ 10.847, p, 0.01. R- square difference (OLS) Model 5 two-way interaction effects model with Melting Pot þ control variables ¼ 0.022 F(1,73) ¼ 6.221, p, 0.05. R-square difference (OLS) Model 6 two-way interaction effects model with Bohemianism þ control variables ¼ 0.002 F(1,73) ¼ 0.377, p. 0.10. a Original coefficient 1000. Significant at 5%; 95% confidence intervals in parentheses.

Technology, Talent, Diversity and the Wealth of European Regions? 1005 725 730 Figure 2. Three-way interaction plots of regression slopes of regional wealth on Melting Pot, at high and low values of Technology and Talent 735 740 745 750 755 760 765 one standard deviation above the sample mean of Technology, Talent and Melting Pot, respectively. 6 From Figure 2(a), we see that for regions harbouring a low degree of Talent, differences in Wealth are related to differences in Diversity as measured by the Melting Pot index, but this association depends on the degree of technological development of a region. In particular, when a region is low on Technology, the Melting Pot index has a positive effect. However, when a region is high on Technology, there is hardly any effect of Diversity. Thus, regions characterized by low Talent and low Technology show a stronger benefit from a higher percentage of non-nationals in the population than regions characterized by low Talent and high Technology. Turning to the interaction effect for regions which are characterized by a high degree of Talent (Figure 2b), we see that, in these regions, Melting Pot is positively related to Wealth, but also that this association depends on the degree of Technology in the region: when a region is low on Technology, the effect of Melting Pot on Wealth is stronger than when the region is high on Technology. In other words, regions high on Talent and high on Technology benefit less from a more diverse population than regions high on Talent and low on Technology. Thus, there is evidence for a salutary effect of a higher percentage of non-nationals in the population for the economic development of a region under specific conditions. Discussion and Conclusion The goal of this paper was to assess to what extent creativity and diversity explain differences in the level of regional economic development (Wealth) between European regions. In our models, we have measured creativity and diversity with a Diversity variable that had three indicators: Tolerance, Melting Pot, and Bohemianism. Since Bohemianism was operationalized in terms of the opinions of a region s population on several issues, it does not measure creativity in the same way as Florida does. Florida measures Bohemianism in terms of occupations since they reflect best what people actually do. The same goes for Talent, that is, for our proxy of creative capital. Although, like Florida, we do measure this in terms of occupations, i.e. the KISs, Florida would probably argue that this measure says something about the amount of knowledge that people use in their work but not necessarily about how creative they are in doing so, since our KIS and his

1006 R. Rutten & J. Gelissen 770 775 780 785 790 795 800 805 810 creative class do not overlap. Finally, Tolerance, too, is measured in terms of the opinions of a region s population on various subgroups in that population rather than as the actual presence of these subgroups (in particular, gays) in the population. Consequently, given the data limitations our research is not an EU carbon copy of Florida s US research, which, to some degree, compromises the comparability between both researches. This also brings us to the statistical difficulties related to our data. The number of regions in our sample is relatively small. Relevant Eurostat data are nor available for all regions on NUTS-2 level and EVS sometimes reported too few respondents on NUTS-2 level regions to admit them to our sample. In response, we have used statistical techniques that allowed us to deal more adequately with the intricacies of small samples such as in the current study. Still, some caution must be exercised when interpreting our empirical findings. For example, we refrained from including too many control variables (e.g. investments). Thus, our findings might suffer from omitted variable bias. Another limitation is that we were not able to estimate more elaborate models. For example, Florida and Tinagli (2004) propose a model which contains considerably more variables than our model. The size of our sample is not large enough to reliably estimate such a model or an even more elaborate path model. The latter would allow investigating indirect effects and causal mechanisms. We note that such a model would put very strict demands on the design of the study in order to be causally valid. However, in spite of these shortcomings we are confident that we have found important positive evidence for the hypothesis that soft factors (our Diversity indicators) matter. Looking at the direct effects of the three Diversity indicators in our model, we found no support for Tolerance, but Melting Pot and Bohemianism have a strong effect on Wealth. In fact, our analyses suggest that the direct effect of Technology and Talent on Wealth may be a spurious relation and that Melting Pot and Bohemianism explain the variance in both Wealth and in Technology and Talent. These outcomes beg the question why Melting Pot and Bohemianism produce strong effects but Tolerance none at all. This is counterintuitive since, if Melting Pot and Bohemianism are important, it is difficult to see how they can thrive in a climate of intolerance. The way we measured Tolerance may explain the absence of statistically significant results. Our Tolerance indicator says something about the opinions of the population as a whole. It is conceivable that, even if this climate were negative, Tolerance would still be high in those parts of society and the economy where tolerance matters for economic development. Intolerance has become rare in the business community partially because of equal rights legislation and ethnically and culturally mixed parts of a city usually show substantial levels of tolerance. Intolerance tends to be high in neighbourhoods of socially deprived, lower class White residents. If this is true, it means that population averages concerning, for example, tolerance, are of limited value in social and economic analysis because such analyses rarely pertain to the socio-economic situation as a whole but instead focus on specific subdivisions, as in the case of explaining Wealth from Diversity. The finding that our measure of Tolerance was not substantively related to the other indicators of diversity supports this argument. That being said, we can draw a number of conclusions from the effects of Diversity that we found in our analysis. In the first place, Florida s theory is largely confirmed for the EU regions in our sample. The predicted associations between, on the one hand, Technology, Talent, and Diversity, and, on the other hand, Wealth, were found. Also in line with Florida s argument, the effect of Diversity, in particular, was very strong in our model. Of course, critics may argue (and, in fact, have argued, cf. Florida, 2005), that this may