Measuring regional innovation: A critical inspection of the ability of single indicators to shape technological change

Technological Forecasting & Social Change xxx (xxxx) xxx–xxx

Contents lists available at ScienceDirect

Technological Forecasting & Social Change journal homepage: www.elsevier.com/locate/techfore

Measuring regional innovation: A critical inspection of the ability of single indicators to shape technological change ⁎

Christoph Hausera, Matthias Sillera,b, , Thomas Schatzera,b, Janette Waldec, Gottfried Tappeinera a

Department of Economics, University of Innsbruck, Universitaetstr. 15, 6020 Innsbruck, Austria Institute for Economic Research, Chamber of Commerce of Bolzano/Bozen, Suedtirolerstr. 60, 39100 Bolzano, Italy c Department of Statistics, University of Innsbruck, Universitaetstr. 15, 6020 Innsbruck, Austria b

A R T I C L E I N F O

A B S T R A C T

JEL classification: O11 O31 O2 R11

The disparities in regional innovation are often illustrated in both scientific research and politics by a single innovation indicator or a composite index. Do such undeniably catchy approaches really convey a better understanding of regional innovation? A composite index can only be employed for an effective innovation policy if the various innovation indicators are highly correlated and affected similarly by the same drivers. The paper investigates driving forces for three composite innovation indices and six innovation indicators covering various aspects of innovation. The analyses demonstrate that the effects of the drivers differ substantially with regard to the investigated aspects of innovation. Knowledge about relevant drivers of innovation is indispensable for the design of an efficient innovation policy. Therefore using only a composite index in order to predict or to influence the innovation dynamic of a territory is highly problematic because of the loss of important parts of the underlying transmission mechanism from innovation policies to innovation outcome. Concentrating on one innovation indicator signifies investigating a specific aspect of regional innovation. Provided these limitations are intended the application of a single indicator may be more appropriate.

Keywords: Regional innovation Innovation dimensions Patent applications Community Innovation Survey Innovation drivers Methodological study

1. Introduction Although innovation is a key subject in regional economics (Asheim et al., 2011; Cooke et al., 1997; Doloreux and Porto Gomez, 2017) and at the center of attention in current discussions of economic policy (European Commission, 2016, 2010; European Council, 2015), there is no standard method for operationalizing and measuring regional innovation performance. This is even more surprising given the extensive scientific debate on suitable innovation indicators at various observational levels (Archibugi, 1988; Becheikh et al., 2006; Griliches, 1990; Janger et al., 2017; Kleinknecht et al., 2002; Smith, 2005). The importance of the regional dimension in innovation economics is widely acknowledged (Asheim et al., 2011; Carayannis et al., 2017; Cooke, 1992; Crevoisier, 2004; Morgan, 1997; Tödtling and Trippl, 2005). Reasoning is diverse. Firstly, regional differences with regard to innovation patterns and performance in an industrialized area such as the EU are sizeable and potential pitfalls on the road to further economic cohesion (Beugelsdijk et al., 2017; Camagni and Capello, 2013). Secondly, although knowledge spillovers are present, there is evidence that these spillovers are spatially bounded (Bottazzi and Peri, 2003), and, thirdly, an effective innovation policy necessitates an

institutional carrier, which is typically a politically delimited region (OECD, 2011). Considerable research has been devoted to identify political channels for bridging the gap in innovation differentials between European regions. However, clear conceptualization of innovation and identification of the related driving forces is essential in order to design a regional innovation policy and to assess its impact. Accordingly, appropriate innovation indicators are needed. It is of paramount importance to understand the basic relationships between driving forces and outcomes in regional innovation processes in order to highlight best practice approaches and illustrate effective measures for lagging regions. Current empirical literature on regional innovation intensity primarily focuses on three measurement approaches. The first and probably most frequently applied method is to quantify innovation using a single indicator. Patent statistics (Bilbao-Osorio and Rodríguez-Pose, 2004; Bottazzi and Peri, 2003; Di Cagno et al., 2016; Hauser et al., 2007; Moreno et al., 2006, 2005) and indicators derived from such data, e.g. patent citations (Maurseth and Verspagen, 2002; Paci and Usai, 2009), dominate this group. The second approach to analyze regional innovation employs an extensive set of indicators. Countries or regions are clustered based on

⁎

Corresponding author at: Institute for Economic Research, Chamber of Commerce of Bolzano/Bozen, Suedtirolerstr. 60, 39100 Bolzano, Italy. E-mail addresses: [email protected] (C. Hauser), [email protected] (M. Siller), [email protected] (T. Schatzer), [email protected] (J. Walde), [email protected] (G. Tappeiner). https://doi.org/10.1016/j.techfore.2017.10.019 Received 16 January 2017; Received in revised form 11 October 2017; Accepted 24 October 2017 0040-1625/ © 2017 Elsevier Inc. All rights reserved.

Please cite this article as: Hauser, C., Technological Forecasting & Social Change (2017), https://doi.org/10.1016/j.techfore.2017.10.019

Technological Forecasting & Social Change xxx (xxxx) xxx–xxx

C. Hauser et al.

scores on goods innovations, while the Austrian region Burgenland or Thüringen in Germany show the reverse picture. The right panel of Fig. 1 compares the indicator for service innovations with the RII. Looking on service innovative regions like Northern Ireland or Alentejo (Portugal) both have above average values, however, with respect to the RII both regions are clearly below the average. On the other hand, the regions Sjælland (Denmark) and Bremen (Germany) show below average values for service innovation but score above average on the RII. The figure shows that regions can be identified that score high on ‘service innovation’ and distinctively lower on ‘goods innovation’ and vice versa. Consequently, focusing only on one of the innovation indicators does not reveal the entire picture of innovative performance of a region. The figure also shows that the regional performance on ‘service innovations’ is not necessarily reflected in the composite index, and thus even the RII, comprising these two highly correlated indicators, is not capable to capture the above feature of innovation. In this paper we focus on three potential problems of using a composite index:

these indicators with the aim to identify various types of innovation systems (Capello and Lenzi, 2013; Navarro et al., 2009; Pinto, 2009). Accordingly, these studies emanate from a multidimensional innovation design allowing different typologies of innovation processes (for details on territorial innovation approaches, see Asheim et al., 2011; Camagni, 1995; Cooke et al., 1997; Crevoisier, 2004). The third approach combines a multitude of innovation indicators directly or stepwise to form a composite index. There are various examples for composite innovation indices. The best known examples of national innovation indices are the Bloomberg Innovation Index,1 the Global Innovation Index (Dutta et al., 2015) and the European Innovation Scoreboard (Hollanders et al., 2016a). The Statsamerica Innovation Index2 compares rates of innovation activities in US states. The most important innovation index at the regional level is the European Regional Innovation Scoreboard (RIS) whose 7th edition was released in summer 2016 (Hollanders et al., 2016b). This periodic exercise comparatively assesses the innovation performance of European regions through the RIS regional innovation index (RII) based on indicators referring to three pillars: enablers, firm activities and outputs. The main objective is to provide a monitoring system tracking regional innovation results. Due to the high attention the European Regional Innovation Scoreboard gets at various levels (EU Commission, national institutions, and media), political decision makers may rely on this tool to evaluate the position of their respective territory as well as to deduce potential pathways to get a higher ranking with respect to RII. However, all three employed approaches show clear shortcomings. In the literature these issues are particularly discussed for patent statistics. A key limitation of patent statistics is that they primarily cover inventions and not commercial innovations (Smith, 2005), while reflecting innovative activities of various sectors very differently (Brouwer and Kleinknecht, 1999; Cohen et al., 2000). The ability of patent data to reflect service innovations (Blind et al., 2003; Hipp and Grupp, 2005) and process innovations (Arundel and Kabla, 1998; Blind et al., 2003; Brouwer and Kleinknecht, 1999) is also limited. In contrast, identification of innovation regimes attempts to do justice to the variety and complexity of innovation activities in different sectors and territories. These analyses take a very broad view of innovation processes and mostly do not distinguish between innovation drivers and outcomes of such processes. Such a mixture of cause and effect prohibits that the approach is suitable for foresighted policy recommendations, it is rather appropriate to depict the current situation. An index such as the RII attempts to combine the informational density of a single indicator with the broad coverage of regime patterns by subsuming data from a multitude of different sources. However, considering the multidimensionality of the innovation concept via an index requires at least two prerequisites: High correlation of the adopted indicators and similar drivers affecting innovation outcomes. The aggregation of different, not highly correlated innovation indicators blurs or rather eliminates information. Even if the innovation indicators highly correlate they may not be influenced by the same driving forces and therefore the analysis may lead to distorted results. Such a misleading finding is illustrated using two regional innovation indicators, namely the percentage of firms with goods innovations and with service innovations respectively, and the RII. The RII comprises the regional share of technological innovators, a combination of the two aforementioned innovation indicators, and also the introduction of process innovations. The left panel of Fig. 1 clearly shows that, although the two innovation indicators show a coefficient of determination of 0.46, several regions score high only on either one of the innovation indicators. Examples are the metropolitan centers of London and Prague with high values on service innovations and below average

• A mix of drivers and outcomes of innovation activities in one index

•

•

impedes the prediction and analysis of effects of driving forces on outcomes and thereby complicates or impedes the identification of effective policy measures. For example, one prominent driver of innovation namely ‘Non R&D innovation expenditures’ is included in the RII. If policy programs want to evaluate whether this innovation driver influences the degree of innovation as measured by the innovation index RII, such a practice is precluded because via definition ‘Non R&D innovation expenditure’ affects the innovation index RII. However, if innovation drivers and innovation outcome indicators are clearly separated from each other one can attempt to answer that question. One of our (preliminary) results demonstrates that especially for this driver no evidence was found for a statistical significant impact on various innovation indicators and indices, except of course for RII. The selection of indicators and the adopted weights strongly affect the ultimate value of an innovation index without being yet stringently established by theoretical considerations from innovation economics. The index value of a region (and consequently also the size and significance of drivers) is the end product of a selection of different aspects of innovation and their combination via a specific weighting scheme. Without being aware of the specific composition of the index, a rank of a region with respect to such an index is not interpretable (and is often not interpretable even with a profound knowledge of the composition of the index) and therefore useless for policy recommendations. This problem is demonstrated using innovation indices with differently obtained aggregation weights (by experts or by a data driven method) and showing the consequences thereof for example with respect to innovation driving forces. Even very sophisticated indices based on multiple aspects of innovation could enable or even induce policy makers to focus on positions in the final or subordinated rankings and thereby transform the analysis into a beauty contest or a pretext for advancing a political agenda. Adopting data of appropriate innovation indicators for various aspects of innovation may prevent political decision-makers from over-simplifying and may draw a more accurate picture of a region's innovation performance.

These problems are investigated by analyzing the effects of a comprehensive set of innovation drivers (inferred from literature) on:

• a widely used composite index of regional innovation (RII), • a modified version of the former, eliminating all components representing innovation drivers, • a composite index with endogenous weighting obtained by a principal component analysis (PCA), • a set of single innovation indicators derived from the Community

1 For Bloomberg Innovation Index, see: http://www.bloomberg.com/graphics/2015innovative-countries/. 2 For Statsamerica Innovation Index, see: http://www.statsamerica.org/innovation/ innovation_index/region-select.html.

2

Technological Forecasting & Social Change xxx (xxxx) xxx–xxx

C. Hauser et al.

3

3

Alentejo

Prague

Prague London

1

Northern Ireland 0

Thüringen −1

Sjælland

Northern Ireland 0

Thüringen Sjælland

−1

Burgenland

Burgenland −2

−3

−3

−1

London

1

−2

−2

Alentejo

2

Service innovators

Service innovators

2

0

1

2

−2

−1

Goods innovators

0

1

2

RII index

Fig. 1. The problem of capturing various innovation aspects using an innovation index illustrated with the innovation indicators ‘goods innovators’ and ‘service innovators’ and the innovation index RII. The index aggregates these two innovation indicators. (Standardized values are used.)

•

Innovation Survey (CIS) along theoretical principles of the OECD known as the ‘Oslo Manual’ (OECD, 2005), and on the number of patents per million inhabitants as it is used extensively in literature as an innovation indicator.

Table 1 Classification of innovation aspects numbered 1 to 12, as suggested by the Oslo Manual (OECD, 2005). Firm novelty

The collected innovation indicators reflect different aspects of the multifaceted innovation concept in varying degrees. The systematic conception of the outcome of innovation processes with the construction of apposite indicators prevents an arbitrary adoption of available measures. All innovation indicators are subsequently processed by a PCA to study their dimensionality. The discretionary application of weights in the aggregation process (as employed for the RII) is thus replaced by the endogenous weighting technique of the PCA. If and only if all innovation indicators are affected by the same driving forces in the same direction and in a comparably quantitative manner, the use of a single indicator/index is sufficient to design and monitor innovation policy. In all other cases, a more differentiated approach is indispensable. The main objective of the paper is to give a qualitative impression of lost information by concentrating on only one innovation index or on a ranking based on a single innovation indicator. It is demonstrated that the concentration on a single indicator impedes a robust prediction of the effects of a differentiated innovation policy. The article is organized as follows. In Section 2 the frame of the analyses is illustrated. In Section 3 the data with its transformation processes and the methods employed are described. The empirical results are presented in Section 4, and in Section 5 the findings are discussed and conclusions are proposed.

Products Processes

Goods Services

Market novelty

No success

Success

No success

Success

1 5 9

2 6 10

3 7 11

4 8 12

commercial success (invention with no apparent commercial success compared to genuine innovations successfully introduced to the market). The systematization of these dimensions provides the innovation classification framework illustrated in Table 1, clearly comprising twelve different aspects for the analysis of innovation. Data at the regional level representing the different aspects of innovation along the classification of Table 1 can be obtained from the European CIS. Table 2 shows indicators selected for the present analyses attempting to capture all aspects of Table 1. Among the innovation indicators both indicators referring to the turnover share of newly introduced innovations have an exceptional position. By capturing the relative sales revenues of innovations (compared to traditional products) they also reflect their commercial impact and thus go beyond measures primarily emphasizing inventive activity such as e.g. patents. In order to analyze whether the eight innovation indicators are influenced by the same driving forces in a comparable way, a set of potential drivers of innovation is needed. Our selection of innovation drivers is largely based on the existing literature, particularly referring to the concept of the knowledge production function (Griliches, 1979). The following groups of drivers are considered: human capital (Dakhli and De Clercq, 2004; Fritsch and Slavtchev, 2007; Lund Vinding, 2006; Marrocu et al., 2013), expenditures for R&D efforts (Buesa et al., 2010; Crescenzi et al., 2013; Jiao et al., 2016; Rodríguez-Pose and Crescenzi, 2008), quality of institutions (Barbosa and Faria, 2011; Rodríguez-Pose, 2013; Rodrik et al., 2004), social capital (Crescenzi et al., 2013; Hauser et al., 2007; Landry et al., 2002), culture (Herbig and Dunphy, 1998; Kaasa and Vadi, 2010) as well as agglomeration forces in the form of Marshall-Arrow-Romer externalities or Jacobian-type spillovers

2. Frame of the analyses The Oslo Manual (OECD, 2005) provides a comprehensive framework for the classification of innovation outcomes according to three principal dimensions: type of technology3 (process vs. product innovations with the latter categorized in goods vs. service innovations), level of novelty (innovation new to the firm or new to the market), and 3 It should be noted that besides technological innovation the literature also identifies important effects from non-technological innovation types, such as marketing or organizational innovations (Evangelista and Vezzani, 2010; OECD, 2005; Sapprasert and Clausen, 2012).

3

Technological Forecasting & Social Change xxx (xxxx) xxx–xxx

C. Hauser et al.

Table 2 Regional innovation indicators derived from the Community Innovation Survey and Eurostat Regio Database. Name

Source

Description (aggregated indicatora)

Innovation aspects (Table 1)

Goods innovators Service innovators Process innovators New-to-firm product innovators New-to-market product innovators Turnover share of new-to-firm product innovations Turnover share of new-to-market product innovations Patent applications

CIS2008 CIS2008 CIS2008 CIS2008 CIS2008 CIS2008 CIS2008 Eurostat

% of goods innovators % of service innovators % of process innovators % of new-to-firm product innovators % of new-to-market product innovators Mean turnover share of new-to-firm product innovations Mean turnover share of new-to-market product innovations Patent applications per million inhabitants

1, 5, 9, 1, 3, 2, 4, 3,

2, 3, 4 6, 7, 8 10, 11, 12 2, 5, 6 4, 7, 8 6 8 4, 7, 8, 11, 12

a Aggregation method: Binary CIS indicators result in regional share (percentage) of firms that introduced the respective innovation type (or novelty degree of innovations); turnover share variables are aggregated to the mean turnover share of innovative products over all firms in a region.

level are reported in Table A.1 in Appendix A.5 Since CIS data with appropriate NUTS codification are not available as a central database, these data are collected separately from national statistics offices with varying access procedures. The restricted access to regional data is arguably a major reason why CIS data play practically no role in regional innovation research with the exception of the indicators employed by RIS. The eight indicators reflecting regional innovation are shown in Table 2. All CIS-based indicators are surveyed on the firm level and then aggregated on the regional level (see Table A.2 in Appendix A). The regional innovation indicators based on CIS data are computed as percentages and patent data are standardized per million inhabitants of a region. Five CIS indicators are calculated as the percentage of firms giving positive answers to the respective questions. The two indicators referring to the realized turnover share of innovative products are included as regional unweighted average percentages. The regional CIS percentages also include data on large firms (with the exception of Bulgaria and Ireland where large firms are not surveyed) but in an unweighted fashion.6 Consequently, the elaborated indicators are comparable to the ones proposed by the RIS which exclusively relies on data from small and medium enterprises.7 The second part of the present analysis investigates potential driving forces of innovation. As benchmark to the aforementioned innovation indicators the composite innovation index from the RIS is replicated along with an index that only uses the innovation indicators from the RIS approach. The next subsection describes the indices based on RIS data.

Fig. 2. A conceptional graphic representation of relevant groups of innovation drivers for the two levels of regional innovation outcome is shown. The figure depicts the various driving forces of innovation grouped with regards to contents and the used innovation indicators. In Section 3 the various innovation drivers are specified in detail.

(Bilbao-Osorio and Rodríguez-Pose, 2004; Greunz, 2004; Li, 2015). Although a wide range of drivers are captured, it is not the aim of this paper to exhaustively model the relations to innovation outcome. The approach employed here aims to assess whether or not the drivers affect the innovation indicators in a similar way. Fig. 2 illustrates the proposed relationships. According to the framework illustrated in Fig. 2 the number of independent outcome dimensions is examined applying PCA to the innovation indicators. How the arising components can be interpreted and how efficiently the PCA uses the information incorporated in the components are investigated. Afterwards regression analysis is used to study whether or not the individual innovation indicators, the derived component resulting from the PCA and the two different RIS indices are influenced by the same driving forces. A spatial econometric approach is employed to account for the presence of spatial effects.

3.2. Regional Innovation Scoreboard indices Data is taken from Annex 5 of the RIS 2012 report (Hollanders et al., 2012), which lists the normalized indicators for three different time periods. Since the CIS data used in the construction of the aforementioned innovation indicators refer to the CIS2008 wave, the corresponding values are extracted. The RIS Regional Innovation Index (‘RII’) is constructed by aggregating seven innovation drivers and five innovation indicators applying a set of weights specified in the report8 and described in Table 3. In order to analyze the consequences of the

3. Data 3.1. Innovation indicators All but one innovation indicator are constructed from the CIS. The exception is the patent indicator (average yearly number of patent applications per million inhabitants during the period 2006 to 2008) that is derived from Eurostat's regional database. The CIS is a periodic survey and collects input and output information on the innovation activities of European firms over a three-year period. The harmonized survey methodology follows the Oslo Manual (OECD, 2005).4 The data is obtained from the sixth wave (CIS2008), which covers the observation period from 2006 to 2008. Given that the CIS data are available only for a subset of European regions, the final dataset comprises 101 regions. Information on the regions used in this analysis as well as the corresponding NUTS

5 The dataset includes regions at different NUTS levels, which, however, for most countries correspond to the regional coverage within the RIS. 6 To be precise for Denmark and Poland only weighted data was available. In order to check the sensitivity of the results with regard to this inconsistency the regions of both nations were also excluded from the analyses but the findings did qualitatively not change. 7 Since the CIS assigns the innovation activities of multi-establishment enterprises to the region where the headquarters are located, the presence of large firms could possibly cause distortions and over-estimate innovation activities in central places. For this reason the RIS only uses data from SMEs in indicator construction and therefore completely disregards innovation in large firms (see footnote 2 on page 8 in the RIS 2014 report). 8 For the NUTS2 regions of Austria only RIS data at the NUTS1 level is available. Consequently, the NUTS1 values have been used in order to calculate the RIS index for the corresponding NUTS2 regions. Similarly, for Spain, Ireland, Slovakia and Slovenia, the average of the RIS values for NUTS2 regions have been used in order to obtain data at the corresponding NUTS1 level.

4 For metadata and methodological issues see: http://ec.europa.eu/eurostat/cache/ metadata/en/inn_cis2_esms.htm.

4

Technological Forecasting & Social Change xxx (xxxx) xxx–xxx

C. Hauser et al.

Table 3 Indicators and drivers used in RIS 2012 composite innovation index.

Name

Typea

Percentage population aged 25–64 having completed tertiary education R&D expenditure in the public sector as % of regional GDP R&D expenditure in the business sector as % of regional GDP Non-R&D innovation expenditures as % of turnover SMEs innovating in-house as % of SMEs Innovative SMEs collaborating with others as % of SMEs Public-private co-publications per million population PCT patents applications per billion regional GDP (in PPS ) SMEs introducing product or process innovations as % of SMEs SMEs introducing marketing or organizational innovations as % of SMEs Employment in knowledge-intensive services + employment in mediumhigh/high-tech manufacturing as % of total workforce Sales of new to market and new to firm innovations as % of turnover

Driver Driver Driver Driver Indicator Driver Driver/indicator Indicator Indicator Indicator Driver

Weights RII 12.5% 12.5% 8.33% 6.25% 6.25% 6.25% 8.33% 8.33% 6.25% 6.25% 12.5%

18.25%

Indicator

6.25%

18.25%

RII outcome

25% 18.25% 18.25%

The classification of ‘driver’ and ‘indicator’ should be differentiated from the classification into ‘enablers’, ‘firm activities’ and ‘outputs’ as used within the RIS.

a

Finally all additional innovation drivers from the RIS are considered for the regression analysis. These are ‘Innovative SMEs collaborating with others as percentage of SMEs’, ‘Number of public-private co-authored research publications per million population’, and ‘Non-R&D innovation expenditures as percentage of turnover’ (for detailed description see Table 3). ‘Innovative SMEs collaborating’ is not included in the regression analyses as just innovative SMEs are counted and therefore this indicator cannot be classified clearly as an innovation driver. Additionally, this indicator cannot be classified in one of the innovation areas defined in Table 1 and is therefore also not considered as innovation outcome. As the indicator ‘Public-private co-publications’ cannot be classified without a doubt into innovation driver or innovation indicator this characteristic is not used. The high degree of multicollinearity makes it impossible to analyze the impact that the selected driving forces have on innovation using all considered drivers simultaneously. This would lead to extremely unstable estimates and inflated standard errors. In order to handle this problem two different approaches are applied with the aim to include as much independent information as possible within and between the considered groups of drivers. Where multicollinearity is caused by bivariate correlations of two variables, one of these variables is excluded from the dataset (guided by theoretical considerations and an exclusion process described in Table A.6). For the strongly correlated Social Capital dimension ‘Weak Ties and Social Trust’ and the variable EQI a different approach is applied. The component ‘Weak Ties and Social Trust’ is integrated in the model and also the residual effect exerted by institutions, by inserting the residual values from a regression of EQI on ‘Weak Ties and Social Trust’. Additionally a dummy for Portugal is included in the regression analyses in order to capture appropriately the remarkably high values of the CIS innovation indicators of the Portuguese regions. Both Germany and Austria feature a uniquely popular vocational education and training system providing education at a level comparable to tertiary education. This dual education system results in high shares of population with educational attainment at the secondary and postsecondary levels and thus in lower rates of tertiary educated population (OECD, 2010). To consider this characteristic another dummy for these two countries is included. Table A.5 in Appendix A describes the independent variables used in the final regression models, and Table A.6 presents the excluded variables with the corresponding reasons for exclusion. With regard to the observation unit the innovation indicators are collected at the firm level, some innovation drivers like the ones concerning Social Capital or Culture are collected at the individual level and others like ‘Private R&D expenditures’ at the firm level. However, all innovation drivers and innovation indicators are ultimately aggregated to the analyzed level of interest, i.e. the regional level.

use of innovation drivers in the creation of the summary innovation index a second index based only on the five outcome indicators used in the RIS is constructed. This index is denominated ‘RII outcome’. 3.3. Innovation drivers Data for each of the six groups of driving forces are collected, i.e. R&D, Human Capital, Institutions, Social Capital, Culture, and Agglomeration Forces. These data are retrieved predominantly from Eurostat. Exceptions are the European Quality of Government Index (EQI) developed by Charron et al. (2014) and the data for the construction of Social Capital and Culture. Social Capital and Culture are both seen as latent concepts and operationalized based on household surveys applying factor-analytical methods. Social Capital proxies are developed from questions on attitudes toward and networks of social interaction in the European Values Study (EVS). In order to obtain stable estimates the responses from three waves of the survey (1990, 1999, 2008) are pooled. Given that Social Capital is formed over centuries (Putnam et al., 1993) and is thus highly persistent over time (Guiso et al., 2008), such a pooling appears admissible. The methods adopted by Hauser et al. (2007) and Bjørnskov (2006) are applied. The various dimensions are identified by applying a PCA with a Varimax rotation to eight items in order to account for the multidimensionality of Social Capital. The PCA extracts three uncorrelated components. The first component is termed ‘Strong Ties’ since this component is most strongly associated with the importance of close relationships such as family or friends, whereas the second component is labeled ‘Weak Ties and Social Trust’ and describes engagement in associational activities as well as general trust. The third component describes ‘Political Interest’ as an indicator for engagement with civil society. However, since theory presumes an impact on the exchange of knowledge only for the former two social capital components, ‘Political Interest’ is excluded from the subsequent analysis. Following Inglehart and Baker (2000) and Inglehart and Welzel (2010) the dimensions of the latent concept Culture are also obtained by applying a PCA with Varimax rotation to eight regionally aggregated items from the pooled waves of the EVS data. The PCA extracts two uncorrelated components termed ‘Traditional vs. Secular-rational Values’ and ‘Survival vs. Self-expression Values’. The first dimension describes changes linked with the transition from agrarian to industrial society, associated with bureaucratization, rationalization, and secularization. The latter refers to polarization between emphasis on order, economic security, and conformity and emphasis on self-expression, participation, subjective well-being, trust, tolerance, and quality of life concerns (Inglehart and Welzel, 2010). Tables A.3 and A.4 in Appendix A provide information on the EVS items used for construction of Social Capital and the dimensions of Culture and on the achieved statistical quality criteria. 5

Technological Forecasting & Social Change xxx (xxxx) xxx–xxx

C. Hauser et al.

Table 4 Communalities and loading matrix from principal component analysis on regional innovation indicators are shown.

Innovation indicator

Communalities

New-to-firm product innovators Goods innovators New-to-market product innovators Process innovators Service innovators Patent applications Turnover share of new-to-firm product innovations Turnover share of new-to-market product innovations Eigenvalue % of explained variance (cumulative)

0.940 0.915 0.910 0.829 0.646 0.564 0.842 0.802

Components 1 0.963 0.948 0.943 0.868 0.790 0.693 −0.040 0.281 4.65 58.2

2 0.117 0.125 0.144 0.273 0.147 −0.289 0.917 0.850 1.79 80.6

Notes: PCA with Varimax rotation. Kaiser-Meyer-Olkin measure of sampling adequacy: 0.744; number of observations: 101. Grey background denotes factor loadings exceeding 0.5 in absolute value.

regarded as statistical noise, this finding suggests the existence of innovation aspects that are exclusively (with regard to the considered innovation indicators) captured by patents. The conspicuous findings with regard to the two indicators for the turnover share of innovative products indicate that ‘new-to-firm sales’ and ‘new-to-market sales’ may be characterized by qualitative or conceptual problems. The additional investigation of the stability of these indicators over time provides further evidence for this conclusion. The stability is analyzed using the RIS (Hollanders et al., 2014) database that includes data for four observation years (equivalent to four CIS waves). The RIS uses the CIS variables referring to the turnover shares of innovative products combined in a single indicator in its analyses. The inspection shows that this variable exhibits substantial variation over time. Whereas an indicator such as regional patenting activity is almost completely stable over the four waves of the seven-year time frame, regional data for the combined sales variables differ strongly in their correlations. Because of the serious problems involved with the two sales variables, they are not further analyzed and instead additional research is suggested to be conducted with regard to the validity of these innovation indicators. Consequently, the PCA has been carried out again using the remaining innovation indicators which results in a single component with an overall explained variance of 77.3%. This endogenously weighted component is denoted as ‘EWC’. The communalities and the component matrix are shown in Table A.7 in Appendix A. In the following section the impact of the considered driving forces on the individual innovation indicators and on the ‘EWC’ as well as on the RIS-based indices is analyzed.

4. Results 4.1. Aggregation of innovation indicators PCA is computed with all eight innovation indicators and a Varimaxtype rotation method is applied. According to the Kaiser criterion, the PCA extracts two uncorrelated components with eigenvalues greater than unity with an overall explained variance of 80.6%. Bartlett's Test of Sphericity produces highly significant results and the Kaiser-Meyer-Olkin criterion of sampling adequacy is 0.744, suggesting general suitability of the indicators to be used in PCA. The communalities and the rotated component matrix of the PCA are shown in Table 4. The obtained results suggest that the information condensation with PCA satisfies all established statistical quality criteria and propose that innovation is not one-dimensional as it exhibits with regard to the employed innovation indicators two independent dimensions. The two resulting components clearly separate the two levels of innovation outcome. The first component is characterized by high loadings of all indicators reflecting the generation of innovations, namely the CIS indicators referring to the percentage of firms with the introduction of goods, services, and process innovations as well as both indicators describing the degree of novelty of introduced product innovations. In addition, the patent indicator exhibits a high loading on this component. The second component shows high loadings for both indicators for the market success of introduced innovations and thus reflects the second level of innovation outcome. However, the results are not a perfect fit with established theory. In particular, two issues stand out. Firstly, although more than one innovation dimension was expected, a separation of first and second degree of innovation outcome is surprising, because indicators corresponding to the same causal chain should exhibit high loadings on the same components (as illustrated in Fig. 2). However, since an orthogonal factor rotation is applied, component 1 and component 2 by definition show no correlation.9 This means that direct innovation summarized in the first component is linearly unrelated to the turnover share from sales of innovative products. Time lags between innovation activities and market success of these introduced innovations may indeed be conceivable, complete missing of correlation, however, does not seem plausible. Secondly, patent applications show a negative loading on the second component. Even if the absolute value is less than 0.5, it must be taken into account. A negative association between patents and the turnover share of newly introduced innovations is in any case counterintuitive. Thirdly, the indicator for patent statistics shows a remarkably low communality of 0.564. Considering that patents are reflected by the first component, a loading of 0.693 implies that less than 50% (0.6932 < 0.5) of the variation of the indicator is used in the first component. If the remaining 50% are not

4.2. Drivers of innovation The impact of the driving forces on regional innovation is estimated by regression analysis. A spatial error model is applied since the residuals resulting from OLS models show significant spatial autocorrelation, and thus estimation via OLS would produce inconsistent standard errors.10 In order to assess model fit two statistics are employed: Pseudo-R2 gives the squared correlation between observed values and the estimates and pseudo-R2spatial considers additionally the effect of the spatial correlation (Anselin, 1988). In order to keep the Type I error at a reasonable size findings are considered as statistically significant at a 1 percent significance level. The results of the estimation are shown in Table 5. A first inspection of the results shows that the examined drivers can explain the considered innovation indicators and indices with very different degrees of success. The pseudo-R2 ranges from 60% for ‘service innovations’ up to 92% for ‘RII’. This is not surprising, as “weak” innovations like ‘service’ and ‘process innovations’ are more difficult to explain as hard indicators like ‘patents’. But the result is also a first indication that soft innovations may be driven by different drivers than

9 A PCA with an oblique rotation method qualitatively produces the same results and shows no significant correlations between the two resulting components. These results are not reported here, but can be made available by the authors upon request.

10

6

The spatial weighting matrix is a row-standardized Queen matrix.

Technological Forecasting & Social Change xxx (xxxx) xxx–xxx

C. Hauser et al.

Table 5 Estimation results from spatial error regression models (all displayed coefficients are standardized coefficients).

Private R&D expenditures Government R&D expenditures Higher education R&D expenditures Tertiary education Strong Ties Weak Ties and Social Trust Traditional vs. Secular-rational Values EQI Residuals Employment Manufacturing Urban Population Non R&D innovation expenditure Dummy Portugal Dummy Germany and Austria Spatial correlation coefficient Pseudo-R² Pseudo-R²spatial Observations

goods

service

process

new-to-market

new-to-firm

log(patent)

EWC

RII

RII outcome

0.149 (0.064) 0.105 (0.072) 0.009 (0.069) -0.044 (0.092) -0.041 (0.052) 0.445*** (0.073) -0.054 (0.061) 0.268*** (0.073) 0.361*** (0.066) 0.000 (0.056) -0.046 (0.054) 2.260*** (0.253) 0.717*** (0.183) 0.196 0.816 0.823 101

0.165** (0.058) 0.000 (0.073) 0.093 (0.059) 0.097 (0.096) -0.187** (0.060) 0.167 (0.103) -0.266*** (0.078) 0.006 (0.082) -0.054 (0.070) 0.086 (0.054) -0.032 (0.055) 3.136*** (0.326) 0.042 (0.250) 0.726*** 0.597 0.839 101

0.071 (0.063) 0.142 (0.078) -0.001 (0.065) 0.050 (0.103) -0.230*** (0.064) 0.245 (0.102) -0.044 (0.081) 0.042 (0.086) 0.128 (0.075) -0.029 (0.059) 0.040 (0.059) 2.111*** (0.338) 0.837*** (0.254) 0.657*** 0.630 0.814 101

0.292*** (0.073) 0.228** (0.080) 0.051 (0.079) -0.147 (0.102) 0.066 (0.057) 0.400*** (0.080) 0.009 (0.066) 0.272*** (0.081) 0.344*** (0.073) 0.066 (0.063) -0.044 (0.060) 1.716*** (0.274) 0.500 (0.199) 0.121 0.771 0.775 101

0.133 (0.059) 0.099 (0.072) 0.103 (0.061) 0.065 (0.095) -0.070 (0.058) 0.295** (0.091) -0.139 (0.073) 0.141 (0.079) 0.214** (0.068) -0.005 (0.054) 0.000 (0.054) 2.090*** (0.305) 0.695** (0.228) 0.616*** 0.725 0.840 101

0.095** (0.033) -0.023 (0.043) -0.081 (0.034) 0.325*** (0.056) -0.117** (0.036) 0.249*** (0.064) -0.083 (0.047) 0.005 (0.049) 0.088 (0.041) 0.044 (0.031) -0.046 (0.032) 0.523** (0.196) 0.848*** (0.152) 0.779*** 0.817 0.947 101

0.195*** (0.051) 0.105 (0.062) 0.032 (0.053) 0.082 (0.083) -0.129 (0.050) 0.295*** (0.078) -0.095 (0.063) 0.118 (0.068) 0.218*** (0.059) 0.004 (0.047) -0.014 (0.047) 2.303*** (0.263) 0.761*** (0.196) 0.603*** 0.800 0.879 101

0.264*** (0.043) 0.191*** (0.047) -0.094* (0.048) 0.157** (0.060) -0.056 (0.033) 0.421*** (0.047) -0.150*** (0.038) 0.207*** (0.047) 0.076 (0.043) 0.158*** (0.037) 0.096** (0.035) 0.750*** (0.158) 0.325** (0.114) 0.052 0.920 0.921 101

0.230*** (0.048) 0.059 (0.054) -0.116 (0.050) 0.038 (0.070) -0.152*** (0.040) 0.379*** (0.057) -0.248*** (0.048) 0.166** (0.056) 0.033 (0.050) 0.090 (0.042) 0.079 (0.041) 1.620*** (0.199) 0.673*** (0.144) 0.304** 0.892 0.902 101

Notes: Dependent variables: ‘Goods innovators’ reflect the regional percentage of firms with the introduction of goods innovations. The same definition applies to ‘service’, ‘process’, ‘new-to-market’ and ‘new-to-firm’. The patent indicator (number of patents per million inhabitants) is included as a ‘log patent’ in order to account for the skewed distribution of this variable. ‘EWC’ denotes the component resulting from PCA of innovation indicators. ‘RII’ is the replicated summary innovation index from the RIS and ‘RII outcome’ is based on outcome indicators from the RIS only; standard errors are in parentheses. ⁎⁎ p < 0.01. ⁎⁎⁎ p < 0.001.

indices, show the same sign, the two indices additionally exhibit different significant drivers (traditional vs. secular values, EQI, and employment in manufacturing). One cannot decide which of the two reveals the “correct” drivers but these differences show that the weighting scheme matters (the weighting of ‘RII outcome’ is exogenously and that of ‘EWC’ endogenously given). Using just one index indicates driver importance, but not for which aspect of innovation. This circumstance becomes obvious by looking simultaneously at the other indicators for innovation. The significance of a driver for an innovation index does not denote that this driver is significant for all single innovation indicators. The most prominent example in our case is the impact of traditional values on ‘RII outcome’ but not on ‘log patents’. At the same time the impact of tertiary education on ‘log patents’ is highly positive but not on the composite indices. This demonstrates that there exist kinds of driving forces with different impact signs on different aspects of innovation. All interpretations regarding only one aspect are misleading. But even in the case when both innovation indices show a significant impact of a driver, it is necessary for practitioners designing an innovation policy to identify the aspects of innovation influenced by an innovation driver. Private R&D impacts many innovation aspects but not for example ‘process innovations’ or government R&D only significantly impacts ‘new-tomarket innovations’. Also social capital shows opposed impacts (in the form of strong ties at one side and weak ties on the other). This characteristic becomes evident when investigating ‘new-to-firm innovations’ and ‘log patents’ separately and not just the composite index ‘EWC’ for example.

innovations for goods. On the other hand, the high pseudo-R2 for ‘RII’ is simply attributable to the fact, that ‘RII’ incorporates many of the driving forces per definition. It is quite obvious, that the ‘RII’ has substantial weaknesses if applied for the search of innovation drivers: the mixing of indicators of cause and effect is the cardinal point. The fact, that it is the only characteristic showing a significant sign for non R&D innovation expenditure is the consequence of this driver-indicator-mixture. This mixing makes ‘RII’ inapplicable for the search of innovation drivers and therefore the ‘RII’ in its original form is not be regarded further in the analyses. For a plausibility check of the results, the focus of the analyses is put on the innovation indicator ‘log patents’, which is profoundly investigated in the literature. The identified drivers are private R&D expenditures, tertiary education, strong ties, and weak ties and social trust. All of these variables have a statistically significant impact on ‘log patents’, being in line with the results reported in the literature (Bottazzi and Peri, 2003; Crescenzi and Rodríguez-Pose, 2013; Greunz, 2004; Moreno et al., 2005; Varsakelis, 2006). Furthermore, the social capital component weak ties and social trust shows a prominent impact, indicating that a trustful society combined with the high quality of institutions favors regional innovation in terms of strong patent activity (Akçomak and ter Weel, 2009; Crescenzi et al., 2013; Hauser et al., 2007; Kaasa, 2009). The findings are in accordance with the literature and therefore the chosen approach demonstrates to be appropriate to investigate the drivers' impacts on innovation outcomes more generally. The results for the two indices (‘EWC’ and ‘RII outcome’) are complex: Although all coefficients, that are statistically significant for both 7

Technological Forecasting & Social Change xxx (xxxx) xxx–xxx

C. Hauser et al.

5. Discussion and conclusions

productivity) of the intended policy measure. Obviously such a differentiated view has the potential to improve the efficiency of an innovation policy. Considering this, the well-known innovation indices and the derived rankings of countries or regions are important as a warning signal and they may be an interesting marketing instrument to establish the interest for research and innovation in the broad public, but for the technical work, preparing a program for innovation policy, the challenge of more and more diverse indicators should be accepted. The OECD did an important task with the Oslo Manual in considering innovation as a multidimensional concept and the EU followed by implementing the idea within the CIS. Unfortunately – in contrast to other important surveys – the CIS microdata together with regional identifiers are not or at least not easily accessible to researchers. This impedes the validation of the data and puts obstacles to the qualitative development of the survey. This has to be changed in order to allow scholars to work on a comprehensive empirical concept to measure and monitor innovation.

The objective of the analyses was to investigate whether it is sufficient to concentrate on an innovation indicator or innovation index or if important value is added by using various indicators/indices for different aspects of innovation. The results show that in order to comprehensively analyze the multidimensional concept of innovation, a multitude of indicators have to be considered separately. It is possible to use a single indicator when the focus relies on a specific aspect of innovation only, e.g. goods innovations. However, to analyze the relevant drivers of the innovation performance of a region multiple indicators need to be considered. A straightforward finding still worth mentioning is that composite indices mixing driving forces and innovation indicators are not constructive, neither to analyze nor to design measures for an innovation policy. This follows by simple logical reasoning and is corroborated by the achieved results. However, as composite indices are still used in the literature and for regional rankings (not only for innovation but also for competitiveness to make a further example) a warning in this respect is appropriate. Innovation indices are frequently reported in the general media. Such publications rely mainly on interpretation of ranking positions. On the contrary scholarly articles adopt a more critical stance toward composition of an index (in particular with regard to the adopted weighting scheme among the few examples are Schibany and Streicher (2008), Grupp and Schubert (2010), and Makkonen and van der Have (2013)) and only rarely use the indices as input in further research. We add to the results of these studies that weighting schemes have not only an important impact on the resulting index but also on the identifiable driving forces of the index and consequently of the operationalized concept therewith. Therefore a composite index must be deeply grounded on theory and even then the problem remains that a complex phenomenon like innovation can simply not be represented by a one-dimensional index. The estimated relationships between drivers and innovation indicators vary considerably both regarding size and significance. Analyzing the impact of different measures of an innovation policy on different aspects of innovation is necessary to predict the outcomes of such a policy and to forecast indirect effects (e.g. employment,

Acknowledgements This work contains statistical data from ONS, which is Crown Copyright. Use of the ONS statistical data in this work does not imply endorsement by the ONS with regard to the interpretation or analysis of the statistical data. This work uses research datasets that may not exactly reproduce National Statistics aggregates. The authors thank the ONS (Office for National Statistics, London), ZEW (Centre for European Economic Research, Mannheim), Statistics Austria (Vienna), the Czech Statistical Office (Prague), the National Statistical Institute of Bulgaria (Sofia), Statistics Denmark (Copenhagen), Statistics Finland (Helsinki), the INSEE (National Institute of Statistics and Economic Studies, Paris), the GUS (Central Statistical Office of Poland, Warsaw), the Direção-Geral de Estatísticas da Educação e Ciênciafor (Lisbon), the National Institute of Statistics of Romania (Bucharest), the INE (National Statistics Institute of Spain, Madrid) for their kind assistance in supplying regional CIS data. The authors are also grateful to the editor and anonymous referees for their comments.

Appendix A Table A.1 Regions used in the analysis. Country

NUTS 1

AT BG CZ DK FI DE IE FR PL PT RO SK SI ES UK

Austria Bulgaria Czech Republic Denmark Finland Germany Ireland France Poland Portugal Romania Slovakia Slovenia Spain United Kingdom

Regions 2 9

2 8 5 4 16 2 8 15 5 8 1 1 6 12

AT11, AT12, AT13, AT21, AT22, AT31, AT32, AT33, AT34 BG3, BG4 CZ01, CZ02, CZ03, CZ04, CZ05, CZ06, CZ07, CZ08 DK01, DK02, DK03, DK04, DK05 FI13, FI18, FI19, FI1A DE1, DE2, DE3, DE4, DE5, DE6, DE7, DE8, DE9, DEA, DEB, DEC, DED, DEE, DEF, DEG IE01, IE02 FR1, FR2, FR3, FR3 FR4, FR5, FR6, FR7, FR8 PL11, PL12, PL21, PL22, PL31, PL32, PL33, PL41, PL42, PL43, PL51, PL52, PL61, PL62, PL63 PT11, PT15, PT16, PT17, PT18 RO11, RO12, RO21, RO22, RO31, RO32, RO41, RO42 SK0 SI0 ES1, ES2 ES3, ES4, ES5, ES6 UKC, UKD, UKE, UKF, UKG UKH, UKI, UK UKJ, UKK, UKL, UKM, UKN

Notes: The CIS data for Germany, the Czech Republic, Denmark, France, Spain, Portugal, and the United Kingdom are based on the NACE, Rev. 2, divisions for CIS core and additional coverage, whereas the data for the remaining territorial units include only the CIS core-coverage industries. According to the CIS2008 methodological recommendations the following NACE, Rev. 2, divisions refer to CIS core coverage: 09, 10–33, 35, 36–39, 46, 49–53, 58, 61, 62, 63, 64–66, 71. Additional coverage includes the following NACE, Rev. 2, divisions: 41–43, 45, 47, 55–56, 59–60, 68, 69, 70, 72, 73, 74, 75, 77, 78, 79, 80, 81, 82. Agriculture and forestry and fishing (NACE, Rev. 2, divisions: 01–03) are excluded from the present analysis. The stability of the results with respect to limitation to core-coverage industries has been checked. The findings are qualitatively very similar when using data for only core industries and for all available industries.

8

Technological Forecasting & Social Change xxx (xxxx) xxx–xxx

C. Hauser et al.

Table A.2 CIS indicator construction (firm level). Innovation indicator

Sectiona Question

Codification

Goods innovators

2.1

Service innovators

2.1

Process innovators

3.1

Yes: 1 No: 0 Yes: 1 No: 0 Yes: 1 No: 0

New-to-firm innovators

2.3

New-to-market innovators

2.3

Turnover share of new-tofirm innovations Turnover share of new-tomarket innovations

2.3

a b

2.3

During the three years 2006 to 2008 did you introduce: new or significantly improved goods? During the three years 2006 to 2008 did you introduce: new or significantly improved services? During the three years 2006 to 2008 did you introduce (min. 1): new or significantly improved methods of manufacturing or producing goods or services; new or significantly improved logistics, delivery or distribution methods for inputs, goods, or services; new or significantly improved supporting activities for processes, such as maintenance systems or operations for purchasing, accounting, or computing?b Were any of your product innovations (goods or services) during the three years 2006 to 2008: only new to the firm? Were any of your product innovations (goods or services) during the three years 2006 to 2008: new to your market? Turnover share in 2008 of goods or services innovations introduced from 2006 to 2008 that were only new to the firm. Turnover share in 2008 of goods or services innovations introduced from 2006 to 2008 that were new to the market.

Yes: 1 No: 0 Yes: 1 No: 0 Share Share

Refers to the section number of the relevant question in the CIS2008 Eurostat-harmonized questionnaire. The CIS in the United Kingdom includes only a single question asking whether process innovations were introduced in the observation period under consideration or not.

Table A.3 Communalities and loading matrix from principal component analysis on social capital items from European Values Study are shown. EVS item

Communalities

How important in your life: friends? How important in your life: politics? How important in your life: family? Number of groups in which you hold membership Number of groups for which you do volunteer work Generalized trust Number of groups of people that you do not want to have as neighbors How often do you discuss politics with your friends? Eigenvalue % of explained variance (cumulative)

0.978 0.943 0.982 0.917 0.743 0.704 0.736 0.880

Components Strong Weak Ties & Ties Social Trust 0.969 −0.054 0.965 −0.104 0.956 −0.223 0.911 −0.179 0.228 0.809 0.763 −0.350 0.671 −0.343

Political Interest −0.188 0.040 −0.135 0.235 0.191 0.017 −0.410

−0.216 3.16 39.4

0.894 1.11 86.0

0.186 2.61 72.1

Notes: PCA with Varimax rotation. Kaiser-Meyer-Olkin measure of sampling adequacy: 0.657. Grey background denotes component loadings exceeding 0.5 in absolute value.

Table A.4 Communalities and loading matrix from principal component analysis on cultural items from European Values Study are shown. EVS item

Importance of God Children Obedience Faith Justification Abortion National Pride Feeling Happiness Economic Physical Security Disapproval Homosexuality Never Sign Petition Eigenvalue % of explained variance (cumulative)

Communalities

0.858 0.726 0.727 0.486 0.680 0.740 0.860 0.715

Components Traditional vs. Secularrational Values 0.868 0.840 0.810 0.684 0.092 0.145 0.502 0.358 2.99 37.4

Survival vs. Selfexpression Values 0.323 0.142 0.268 0.137 0.856 0.812 0.779 0.766 2.80 72.4

Notes: PCA with Varimax rotation. Kaiser-Meyer-Olkin measure of sampling adequacy: 0.736. Grey background denotes component loadings exceeding 0.5 in absolute value.

9

Technological Forecasting & Social Change xxx (xxxx) xxx–xxx

C. Hauser et al.

Table A.5 Innovation drivers used in the final regression models. Group

Innovation driver

Description and calculation

Source

Research and development

Social capital

Strong Ties

Culture

Weak Ties and Social Trust Traditional vs. Secularrational Values EQI residuals

Average yearly expenditures for R&D efforts in the business enterprise sector from 2006 to 2008 (million euros in ppp per inhabitant) Average yearly expenditures for R&D efforts in the government sector from 2006 to 2008 (million euros in ppp per inhabitant) Average yearly expenditures for R&D efforts in the higher education sector from 2006 to 2008 (million euros in ppp per inhabitant) Percentage population aged 25–64 having completed tertiary education (average 2006 to 2008) Relationships with family and friends (resulting from PCA on eight regionally aggregated variables) Associational activity and confidence in other humans (resulting from PCA on eight regionally aggregated variables) Resulting from PCA on eight regionally aggregated EVS variables

Eurostat

Human capital

Private sector R&D expenditures Government sector R&D expenditures Higher education sector R&D expenditures Tertiary education

Quality of governance Agglomeration forces Additional RIS variable Dummy variables

Employment manufacturing Population in urban areas Non R&D innovation expenditures Dummy Portugal Dummy Germany and Austria

Eurostat Eurostat Eurostat

Residuals from the regression of EQI (European Quality of Government Index) on Social Capital dimension ‘Weak Ties and Social Trust’ Manufacturing employment as share of total employment (average 2006 to 2008)

EVS; own calculation EVS; own calculation EVS; own calculation Own calculationa Eurostat

Share of population living in NUTS 3 regions classified as urban areas (in 2008)b Non R&D innovation expenditures as a percentage of turnover

Eurostat RIS

Binary variable coded 1 for regions in Portugal and 0 otherwise Binary variable coded 1 for regions in Germany and Austria and 0 otherwise

a

Charron et al. (2014). This indicator was calculated on the basis of the European Union urban-rural typology. This typology classifies regions as ‘predominantly rural’, ‘intermediate’ or ‘predominantly urban’ (Eurostat, 2010). b

Table A.6 Innovation drivers excluded either by a PCA computed in the first step or based on the multicollinearity diagnosis for the regression models in the second step. Group

Innovation driver

Research and development

Private sector Average share of R&D personnel in the business R&D enterprise sector as a share of total population from 2006 to 2008 personnel Government sector R&D personnel Higher education sector R&D personnel Private sector researchers Government sector researchers Higher education sector researchers

Description and calculation

Source

Reason for exclusion

Eurostat

Step 1: Three uncorrelated PCA components are obtained and for each component the representative variable indicated in Table A.5 (with the highest communality) is employed.

Average share of R&D personnel in the Eurostat government sector as a share of total population from 2006 to 2008 Average share of R&D personnel in the higher Eurostat education sector as a share of total population from 2006 to 2008 Average share of researchers in the business Eurostat enterprise sector as a share of total population from 2006 to 2008 Average share of researchers in the government Eurostat sector as a share of total population from 2006 to 2008 Eurostat Average share of researchers in the higher education sector as a share of total population from 2006 to 2008

10

Technological Forecasting & Social Change xxx (xxxx) xxx–xxx

C. Hauser et al.

Human capital

HRST

HRST-O HRST-E HRST-C

Culture

Quality of governance

Agglomeration forces

Survival vs. Selfexpression Values Trust in national institutions Trust in international institutions GDP per capita

Employment services Employment KIS + MHTC

a

Persons with tertiary education (ISCED) and/or employed in science and technology (as share of total population) Persons employed in science and technology (as share of total population) Persons with tertiary education (ISCED) (as share of total population) Persons with tertiary education (ISCED) and employed in science and technology (as share of total population) Resulting from PCA of eight regionally aggregated EVS variables

Eurostat

Resulting from PCA applied to seven EVS variables

EVS; own Step 1: A single PCA component is identified. calculation Therefore EQI as a representative variable is employed further. EVS; own calculation

Resulting from PCA applied to seven EVS variables Gross domestic product per capita (in million euros in ppp; average 2006 to 2008)

Step 1: A single PCA component is identified. Therefore, only one variable, e.g. tertiary education, is used.

Eurostat Eurostat Eurostat

EVS; own Step 2: Due to the strong correlation with EQI calculation this component is excluded from further analysis.

Eurostat

Step 1: A single PCA component is obtained. Due to the strong correlations between GDP per capita and R&D expenditures, between employment services and HRSTO and between employment KIS + MHTC and tertiary education, the two variables indicated in Table A.5 are used in the regression analysis. Both are used as the communalities are not excessively high.

Employment in service industries as share of total Eurostat employment (average 2006 to 2008) Employment in knowledge-intensive services and Eurostat medium-high and high-tech manufacturing industries as a share of total workforce (average 2006 to 2008)a

According to Eurostat, the following NACE, Rev. 2, divisions are classified as knowledge-intensive services (KIS): 50, 51, 58 to 63, 64 to 66, 69 to 75, 78, 80, 84 to 93.

Table A.7 Communalities and loading matrix from principal component analysis on innovation indicators are shown. Innovation indicator

Communalities

Component EWC

New-to-firm product innovators Goods innovators New-to-market product innovators Process innovators Service innovators Patent applications Eigenvalue % of explained variance (cumulative)

0.949 0.917 0.903 0.810 0.649 0.412

0.974 0.958 0.950 0.900 0.805 0.642 4.64 77.3

Notes: Kaiser-Meyer-Olkin measure of sampling adequacy: 0.840; number of observations: 101.

Chang. 34, 253–277. http://dx.doi.org/10.1016/0040-1625(88)90071-6. Arundel, A., Kabla, I., 1998. What percentage of innovations are patented? Empirical estimates for European firms. Res. Policy 27, 127–141. http://dx.doi.org/10.1016/ S0048-7333(98)00033-X. Asheim, B.T., Smith, H.L., Oughton, C., 2011. Regional innovation systems: theory, empirics and policy. Reg. Stud. 45, 875–891. http://dx.doi.org/10.1080/00343404. 2011.596701. Barbosa, N., Faria, A.P., 2011. Innovation across Europe: how important are institutional differences? Res. Policy 40, 1157–1169. http://dx.doi.org/10.1016/j.respol.2011.05. 017.

References Akçomak, I.S., ter Weel, B., 2009. Social capital, innovation and growth: evidence from Europe. Eur. Econ. Rev. 53, 544–567. http://dx.doi.org/10.1016/j.euroecorev.2008. 10.001. Anselin, L., 1988. Spatial Econometrics: Methods and Models. Kluwer Academic, Dordrecht. http://dx.doi.org/10.1007/978-94-015-7799-1. Archibugi, D., 1988. In search of a useful measure of technological innovation (to make economists happy without discontenting technologists). Technol. Forecast. Soc.

11

Technological Forecasting & Social Change xxx (xxxx) xxx–xxx

C. Hauser et al.

Evol. Econ. 14, 563–592. http://dx.doi.org/10.1007/s00191-004-0234-8. Griliches, Z., 1979. Issues in assessing the contribution of research and development to productivity growth. Bell J. Econ. 10, 92–116. http://dx.doi.org/10.2307/3003321. Griliches, Z., 1990. Patent statistics as economic indicators: a survey. J. Econ. Lit. 28, 1661–1707. Grupp, H., Schubert, T., 2010. Review and new evidence on composite innovation indicators for evaluating national performance. Res. Policy 39, 67–78. http://dx.doi. org/10.1016/j.respol.2009.10.002. Guiso, L., Sapienza, P., Zingales, L., 2008. Long term persistence. In: NBER Working Paper No. 14278. National Bureau of Economic Research, Cambridge, MA. http://dx.doi. org/10.3386/w14278. http://www.nber.org/papers/w14278. Hauser, C., Tappeiner, G., Walde, J., 2007. The learning region: the impact of social capital and weak ties on innovation. Reg. Stud. 41, 75–88. http://dx.doi.org/10. 1080/00343400600928368. Herbig, P., Dunphy, S., 1998. Culture and innovation. Cross Cult. Manag. 5, 13–21. http://dx.doi.org/10.1108/13527609810796844. Hipp, C., Grupp, H., 2005. Innovation in the service sector: the demand for service-specific innovation measurement concepts and typologies. Res. Policy 34, 517–535. http://dx.doi.org/10.1016/j.respol.2005.03.002. Hollanders, H., Rivera Léon, L., Roman, L., 2012. Regional Innovation Scoreboard 2012. European Commission, DG Enterprise and Industry, Brussels. Hollanders, H., Es-Sadki, N., Buligescu, B., Rivera Léon, L., Griniece, E., Roman, L., 2014. Regional Innovation Scoreboard 2014. European Commission, DG Enterprise and Industry, Brussels. Hollanders, H., Es-Sadki, N., Kanerva, M., 2016a. European Innovation Scoreboard 2016. European Commission, DG Internal Market, Industry, Entrepreneurship and SMEs, Brussels. Hollanders, H., Es-Sadki, N., Kanerva, M., 2016b. Regional Innovation Scoreboard 2016. European Commission, DG Internal Market, Industry, Entrepreneurship and SMEs, Brussels. Inglehart, R., Baker, W.E., 2000. Modernization, cultural change, and the persistence of traditional values. Am. Sociol. Rev. 65, 19–51. http://dx.doi.org/10.2307/2657288. Inglehart, R., Welzel, C., 2010. Changing mass priorities: the link between modernization and democracy. Perspect. Polit. 8, 551–567. http://dx.doi.org/10.1017/ S1537592710001258. Janger, J., Schubert, T., Andries, P., Rammer, C., Hoskens, M., 2017. The EU 2020 innovation indicator: a step forward in measuring innovation outputs and outcomes? Res. Policy 46, 30–42. http://dx.doi.org/10.1016/j.respol.2016.10.001. Jiao, H., Zhou, J., Gao, T., Liu, X., 2016. The more interactions the better? The moderating effect of the interaction between local producers and users of knowledge on the relationship between R&D investment and regional innovation systems. Technol. Forecast. Soc. Chang. 110, 13–20. http://dx.doi.org/10.1016/j.techfore.2016.03. 025. Kaasa, A., 2009. Effects of different dimensions of social capital on innovative activity: evidence from Europe at the regional level. Technovation 29, 218–233. http://dx.doi. org/10.1016/j.technovation.2008.01.003. Kaasa, A., Vadi, M., 2010. How does culture contribute to innovation? Evidence from European countries. Econ. Innov. New Technol. 19, 583–604. http://dx.doi.org/10. 1080/10438590902987222. Kleinknecht, A., Van Montfort, K., Brouwer, E., 2002. The non-trivial choice between innovation indicators. Econ. Innov. New Technol. 11, 109–121. http://dx.doi.org/10. 1080/10438590210899. Landry, R., Amara, N., Lamari, M., 2002. Does social capital determine innovation? To what extent? Technol. Forecast. Soc. Chang. 69, 681–701. http://dx.doi.org/10. 1016/S0040-1625(01)00170-6. Li, X., 2015. Specialization, institutions and innovation within China's regional innovation systems. Technol. Forecast. Soc. Chang. 100, 130–139. http://dx.doi.org/10. 1016/j.techfore.2015.06.032. Lund Vinding, A., 2006. Absorptive capacity and innovative performance: a human capital approach. Econ. Innov. New Technol. 15, 507–517. http://dx.doi.org/10.1080/ 10438590500513057. Makkonen, T., van der Have, R.P., 2013. Benchmarking regional innovative performance: composite measures and direct innovation counts. Scientometrics 94, 247–262. http://dx.doi.org/10.1007/s11192-012-0753-2. Marrocu, E., Paci, R., Usai, S., 2013. Proximity, networking and knowledge production in Europe: what lessons for innovation policy? Technol. Forecast. Soc. Chang. 80, 1484–1498. http://dx.doi.org/10.1016/j.techfore.2013.03.004. Maurseth, P.B., Verspagen, B., 2002. Knowledge spillovers in Europe: a patent citations analysis. Scand. J. Econ. 104, 531–545. http://dx.doi.org/10.1111/1467-9442. 00300. Moreno, R., Paci, R., Usai, S., 2005. Spatial spillovers and innovation activity in European regions. Environ. Plan. A 37, 1793–1812. http://dx.doi.org/10.1068/a37341. Moreno, R., Paci, R., Usai, S., 2006. Innovation clusters in the European regions. Eur. Plan. Stud. 14, 1235–1263. http://dx.doi.org/10.1080/09654310600933330. Morgan, K., 1997. The learning region: institutions, innovation and regional renewal. Reg. Stud. 31, 491–503. http://dx.doi.org/10.1080/00343409750132289. Navarro, M., Gibaja, J.J., Bilbao-Osorio, B., Aguado, R., 2009. Patterns of innovation in EU-25 regions: a typology and policy recommendations. Eviron. Plann. C. Gov. Policy 27, 815–840. http://dx.doi.org/10.1068/c0884r. OECD, 2005. Oslo Manual: Guidelines for Collecting and Interpreting Innovation Data. OECD Publishing, Paris. http://dx.doi.org/10.1787/9789264013100-en. OECD, 2010. Education at a Glance 2010. OECD Publishing, Paris. http://dx.doi.org/10. 1787/eag-2010-en. OECD, 2011. Regions and Innovation Policy. OECD Publishing, Paris. http://dx.doi.org/ 10.1787/9789264097803-en. Paci, R., Usai, S., 2009. Knowledge flows across European regions. Ann. Reg. Sci. 43,

Becheikh, N., Landry, R., Amara, N., 2006. Lessons from innovation empirical studies in the manufacturing sector: a systematic review of the literature from 1993–2003. Technovation 26, 644–664. http://dx.doi.org/10.1016/j.technovation.2005.06.016. Beugelsdijk, S., Klasing, M.J., Milionis, P., 2017. Regional economic development in Europe: the role of total factor productivity. Reg. Stud. http://dx.doi.org/10.1080/ 00343404.2017.1334118. Bilbao-Osorio, B., Rodríguez-Pose, A., 2004. From R&D to innovation and economic growth in the EU. Growth Chang. 35, 434–455. http://dx.doi.org/10.1111/j.14682257.2004.00256.x. Bjørnskov, C., 2006. The multiple facets of social capital. Eur. J. Polit. Econ. 22, 22–40. http://dx.doi.org/10.1016/j.ejpoleco.2005.05.006. Blind, K., Edler, J., Schmoch, U., Anderson, B., Howells, J., Miles, I., Roberts, J., Green, L., Evangelista, R., Hipp, C., Herstatt, C., 2003. Patents in the Service Industries. Final Report Prepared for the European Commission. FhG-ISI, Karlsruhe. Bottazzi, L., Peri, G., 2003. Innovation and spillovers in regions: evidence from European patent data. Eur. Econ. Rev. 47, 687–710. http://dx.doi.org/10.1016/S00142921(02)00307-0. Brouwer, E., Kleinknecht, A., 1999. Innovative output, and a firm's propensity to patent. An exploration of CIS micro data. Res. Policy 28, 615–624. http://dx.doi.org/10. 1016/S0048-7333(99)00003-7. Buesa, M., Heijs, J., Baumert, T., 2010. The determinants of regional innovation in Europe: a combined factorial and regression knowledge production function approach. Res. Policy 39, 722–735. http://dx.doi.org/10.1016/j.respol.2010.02.016. Camagni, R.P., 1995. The concept of innovative milieu and its relevance for public policies in European lagging regions. Pap. Reg. Sci. 74, 317–340. http://dx.doi.org/10. 1111/j.1435-5597.1995.tb00644.x. Camagni, R., Capello, R., 2013. Regional innovation patterns and the EU regional policy reform: toward smart innovation policies. Growth Chang. 44, 355–389. http://dx.doi. org/10.1111/grow.12012. Capello, R., Lenzi, C., 2013. Territorial patterns of innovation: a taxonomy of innovative regions in Europe. Ann. Reg. Sci. 51, 119–154. http://dx.doi.org/10.1007/s00168012-0539-8. Carayannis, E.G., Goletsis, Y., Grigoroudis, E., 2017. Composite innovation metrics: MCDA and the Quadruple Innovation Helix framework. Technol. Forecast. Soc. Chang. http://dx.doi.org/10.1016/j.techfore.2017.03.008. Charron, N., Dijkstra, L., Lapuente, V., 2014. Regional governance matters: quality of government within European Union member states. Reg. Stud. 48, 68–90. http://dx. doi.org/10.1080/00343404.2013.770141. Cohen, W.M., Nelson, R.R., Walsh, J.P., 2000. Protecting their intellectual assets: appropriability conditions and why U.S. manufacturing firms patent (or not). In: NBER Working Paper No. 7552. National Bureau of Economic Research, Cambridge, MA. http://dx.doi.org/10.3386/w7552. http://www.nber.org/papers/w7552. Cooke, P., 1992. Regional innovation systems: competitive regulation in the new Europe. Geoforum 23, 365–382. http://dx.doi.org/10.1016/0016-7185(92)90048-9. Cooke, P., Gomez Uranga, M., Etxebarria, G., 1997. Regional innovation systems: institutional and organisational dimensions. Res. Policy 26, 475–491. http://dx.doi. org/10.1016/S0048-7333(97)00025-5. Crescenzi, R., Rodríguez-Pose, A., 2013. R&D, socio-economic conditions, and regional innovation in the U.S. Growth Chang. 44, 287–320. http://dx.doi.org/10.1111/grow. 12011. Crescenzi, R., Gagliardi, L., Percoco, M., 2013. Social capital and the innovative performance of Italian provinces. Environ. Plan. A 45, 908–929. http://dx.doi.org/10. 1068/a45221. Crevoisier, O., 2004. The innovative milieus approach: toward a territorialized understanding of the economy? Econ. Geogr. 80, 367–379. http://dx.doi.org/10.1111/j. 1944-8287.2004.tb00243.x. Dakhli, M., De Clercq, D., 2004. Human capital, social capital, and innovation: a multicountry study. Entrep. Reg. Dev. 16, 107–128. http://dx.doi.org/10.1080/ 08985620410001677835. Di Cagno, D., Fabrizi, A., Meliciani, V., Wanzenböck, I., 2016. The impact of relational spillovers from joint research projects on knowledge creation across European regions. Technol. Forecast. Soc. Chang. 108, 83–94. http://dx.doi.org/10.1016/j. techfore.2016.04.021. Doloreux, D., Porto Gomez, I., 2017. A review of (almost) 20 years of regional innovation systems research. Eur. Plan. Stud. 25, 371–387. http://dx.doi.org/10.1080/ 09654313.2016.1244516. Dutta, S., Lanvin, B., Wunsch-Vincent, S., 2015. The Global Innovation Index 2015: Effective Innovation Policies for Development. Cornell University, INSEAD, and WIPO, Fontainebleau, Ithaca, and Geneva. European Commission, 2010. Europe 2020: a strategy for smart, sustainable and inclusive growth. Communication from the Commission, COM(2010) 2020. European Commission, Brussels. European Commission, 2016. Better regulations for innovation-driven investment at EU level. Commission Staff Working Document. Publications Office of the European Union, Luxembourg. http://dx.doi.org/10.2777/987880. European Council, 2015. Recommendations on Broad Guidelines for the Economic Policies of the Member States and of the European Union, Council Recommendation (EU) 2015/1184. Eurostat, 2010. Chapter 15: a revised urban rural typology. In: Eurostat Regional Yearbook 2010. Publications Office of the European Union, Luxembourg, pp. 239–253. Evangelista, R., Vezzani, A., 2010. The economic impact of technological and organizational innovations. A firm-level analysis. Res. Policy 39, 1253–1263. Fritsch, M., Slavtchev, V., 2007. Universities and innovation in space. Ind. Innov. 14, 201–218. http://dx.doi.org/10.1080/13662710701253466. Greunz, L., 2004. Industrial structure and innovation - evidence from European regions. J.

12

Technological Forecasting & Social Change xxx (xxxx) xxx–xxx

C. Hauser et al.

highlights include the impact of weak ties on regional innovation, persistence of social norms rooted in collective memory and impact of social and institutional trust on work attitudes. His approach is mainly empirical and combines information from geographical information systems with spatial analyses.

669–690. http://dx.doi.org/10.1007/s00168-008-0256-5. Pinto, H., 2009. The diversity of innovation in the European Union: mapping latent dimensions and regional profiles. Eur. Plan. Stud. 17, 303–326. http://dx.doi.org/10. 1080/09654310802553571. Putnam, R.D., Leonardi, R., Nanetti, R.Y., 1993. Making Democracy Work: Civic Traditions in Modern Italy. Princeton University Press, Princeton. Rodríguez-Pose, A., 2013. Do institutions matter for regional development? Reg. Stud. 47, 1034–1047. http://dx.doi.org/10.1080/00343404.2012.748978. Rodríguez-Pose, A., Crescenzi, R., 2008. Research and development, spillovers, innovation systems, and the genesis of regional growth in Europe. Reg. Stud. 42, 51–67. http://dx.doi.org/10.1080/00343400701654186. Rodrik, D., Subramanian, A., Trebbi, F., 2004. Institutions rule: the primacy of institutions over geography and integration in economic development. J. Econ. Growth 9, 131–165. http://dx.doi.org/10.1023/B:JOEG.0000031425.72248.85. Sapprasert, K., Clausen, T.H., 2012. Organizational innovation and its effects. Ind. Corp. Chang. 21, 1283–1305. http://dx.doi.org/10.1093/icc/dts023. Schibany, A., Streicher, G., 2008. The European Innovation Scoreboard: drowning by numbers? Sci. Public Policy 35, 717–732. http://dx.doi.org/10.3152/ 030234208X398512. Smith, K., 2005. Measuring innovation. In: Fagerberg, J., Mowery, D.C., Nelson, R.R. (Eds.), The Oxford Handbook of Innovation. Oxford University Press, Oxford, pp. 148–177. Tödtling, F., Trippl, M., 2005. One size fits all? Towards a differentiated regional innovation policy approach. Res. Policy 34, 1203–1219. http://dx.doi.org/10.1016/j. respol.2005.01.018. Varsakelis, N.C., 2006. Education, political institutions and innovative activity: a crosscountry empirical investigation. Res. Policy 35, 1083–1090. http://dx.doi.org/10. 1016/j.respol.2006.06.002.

Matthias Siller is a PhD candidate at the Department of Economics, University of Innsbruck, Austria. He also works as a researcher at the Institute for Economic Research of the Chamber of Commerce of Bolzano, Italy. His research interests concern regional economics and regional innovation systems. Thomas Schatzer is a PhD candidate at the Department of Economics, University of Innsbruck, Austria and researcher at the Institute for Economic Research of the Chamber of Commerce of Bolzano, Italy. His research interests include regional competitiveness, productivity of regions and regional economics in general. Janette Walde is Associate Professor of Statistics and Econometrics at the Department of Statistics, University of Innsbruck, Austria. She is also speaker of the research platform ‘Empirical and Experimental Economics’ at the University of Innsbruck and Associate Editor of the journals ‘Advances in Statistical Analysis’ and ‘Environmental Modelling and Assessment’. Her research activities concern the development and application of statistical and computational methods to address problems in Economics and Biology. From a statistical point of view her research focus is on methods capturing appropriately nonlinear relationships with spatiotemporally correlated observation units. Gottfried Tappeiner is a full Professor of Economics at the Department of Economics of the University of Innsbruck, Austria. He is Dean of Studies of the Department of Economics. Research interests: Regional economics, environmental economics, effects of social capital. A special focus is the impact of the quality of input data (errors, aggregation problems, data transformation, preliminary estimation methods) on the results of the applied tools of analyses.

Christoph Hauser is Assistant Professor at the Department of Economics, University of Innsbruck. His research interests include regional innovation systems, sources and effects of social capital as well as regional growth and development processes. Research

13