Science and Public policy February 2010 0302-3427/10/010063-16 US$12. 00 Beech tree Publishing 2010 63 Science and Public policy, 37 (1 february 2010, pages 63 78 DOI: 10.3152/030234210x491623; http://www. ingentaconnect. com/content/beech/spp Impact of Swiss technology policy on firm innovation performance: an evaluation based on a matching approach Spyros Arvanitis, Laurent Donzé and Nora Sydow This paper investigates the impact of the promotional activities of The swiss Commission of Technology and Innovation (CTI) on the innovation performance of the supported firms based on a matched-pairs analysis of 199 firms supported by the CTI in the period 2000 2002. CTI's promotional activities significantly improved the innovation performance of the firms that they supported with respect to six different measures of innovation performance. This could be shown by four different matching methods. A further finding was that the magnitude of the impact correlated positively with the relative size of the financial support as measured by the quotient of the volume of financial support to the volume of a supported firm's own research and development expenditures. HE IMPACT OF THE INNOVATION promottio policy of the‘Commission of Technollog and Innovation'(CTI), which is the most important government agency for the promotiio of innovation in Switzerland, was investigated in this study. The CTI mainly supports research and development (R&d) co-operation projects from all scientific fields by funding the public partner (a univerrsit or a public research institution) in such a cooperaation the private partner being an enterprise that agrees to contribute to this project at its own expeens by at least the amount of funds offered by the CTI (private contribution of at least 50%;%the‘bottoomup'principle of support. The projects to be subsidized are selected by committees of experts that evaluate the applications by some criteria of excellence. There have also been some recent prograamme for the promotion of specific technologies (e g. Medtech, Topnano21), but this type of speciifi support has always been of minor importance. The principle of indirect R&d support of good projeccts which are proposed jointly by a private and a public partner, is fundamental to The swiss technoloog policy and, as a main promotional policy, to our knowledge, is unique in Europe. Our main hypothesis was that: on average enterpriise that were supported by the CTI would show a significantly higher innovation performance, measurre through six innovation measures (e g. sales, share of innovative products), than‘structurally similar'firms without such activities. To show this, we used matched-pairs analysis for a set of firms supported by CTI and the corresponding control groups for the period 2002 2004. Matching methods based on direct comparisons of participating and nonparticipating agents, which were used first in labour market evaluations, have also been applied to evaluate the technology programmes of European countries (see Almus and Czarnitzki, 2003; Czarnitzki and Fier, 2002 (for Germany; Pointner and Rammer, 2005 (for Austria; Görg and Strobl, 2007 (for Ireland. 1 A major advantage of the matching methods rather than the regression approach is that the matching is nonparametric. As such, it T Spyros Arvanitis is at KOF Swiss Economic Institute, ETH Zurich, 8092 Zurich, Switzerland; Email: arvanitis@kof. ethz. ch; Tel:++41 44 632 51 68; Fax:++41 44 632 13 52. Laurent Donzé is at the Faculty of economics and Social sciences, University of Fribourg, 1700 Friboug, Switzerland; Email: laurent. donze@unifr. ch; Tel:++41 26 300 82 75; Fax:++41 26 300 97 81. Nora Sydow is at the Economic Research Department, Credit suisse, 8070 Zurich, Switzerland; Email: nora. sydow@creditsuiissecom; Tel:++41 44 333 45 10; Fax:++41 44 333 56 79. This study was supported financially by The swiss Federal office for Professional Education and Technology. Impact of technology policy on innovation by firms Science and Public policy February 2010 64 avoids the functional form restrictions implicit in running a regression of some type. A brief description of the approach pursued in this paper is as follows: we identified the subsidized firms in the period 2000 2002 from the CTI database. We collected innovation data for the promoted firms similla to those already existing for a sample of innovating firms of The swiss Innovation Survey 2002 (Arvanitis et al. 2004). ) We estimated the propensity scores with respect to the likelihood of receiving a CTI subsidy. We then applied four different matching methods in order to find the structurally similar‘twin'firms for every subsidized firm. We tested the statistical signifiicanc of the difference of the means of six differeen innovation measures of the subsidized firms and the non-subsidized firms of the matched control group. We constructed a subsidy quotient: the amount of R&d promotion divided by the R&d budget of the firm in the same period. We were able to distinguish between firms with a high (higher than the median) and a low subsidy quotient (lower than the median), and carry out a statistical test on the difference of the differences of the means of the innovation variables of the subsidized firms and the matched nonsubsiidize firms. For the period 2002 2004 we found that (with one exception), for all six innovation measures and for all four matching methods applied, the innovation performmanc of CTI-subsidized firms was on average significantly higher than that of the non-subsidized firms in the matched control group. Further it was shown that the promotion effect was (with one exceptiion dependent on the magnitude of the promotion ratti (as measured by the ratio of R&d subsidies by CTI to a firm's own R&d expenditure. The new elements in our analysis were: the use of innovation data for the subsidized firms, collected by means of a survey; the use of four different matching methods that alloowe us to test the robustness of our results; and the investigation of the effect of promotion ratio as measured by the ratio of R&d subsidies by CTI to a firm's own R&d expenditure. The paper is structured as follows: first, we present the conceptual framework of the study; secondly, we give an overview of similar studies. Thirdly, we deal with the data sources; fourthly, we present some informmatio on the patterns of CTI promotion in the reference period. Fifthly, we provide a detailed discusssio of our methodology for estimating the impaac of CTI subsidies on the innovation performance of firms. We then discuss the results and provide a summary and some implications for technology policy. Conceptual framework Technology policy: public fiscal policies to support innovation Most OECD countries use large amounts of public funds to support activities that are intended to enhaanc innovation in the business sector. These funds are used often to provide direct support for private sector research and innovation. A further way of supporting private investment in innovation is through tax incentives for R&d expenditures (see Jaumotte and Pain, 2005 for a survey of the main fiscal policies to support innovation. The underlyyin justification for public policies to support innovaatio is provided by the economic argument that otherwise the private sector would invest less in innovaativ activities than is socially desirable. The reasons for such‘market failure'that leads to underinvesstmen in innovative activities could be: informatiiona imperfections, informational externalities due to knowledge spillovers, financial market failurre or shortages of highly qualified personnel (Nelsoon 1959; Arrow, 1962. Thus, public fiscal policies to support innovation are designed to alleviate particcula forms of market failure that would lead to under-investment. For example, programmes offeriin financial support for small or young firms are intennde to stimulate additional R&d and innovation in firms that would otherwise have difficulty funding themselves in the capital market. In practice, identifyyin the firm or project categories that should be subsidized requires difficult judgements to be made. Swiss technology policy There is a long tradition in Switzerland of refraining from directly funding business firms for innovation activities. In a comparison of industrialized countries only Japan and Luxembourg show a comparably low percentage of government financing for R&d (OECD 2007). ) This tradition is based on a wide Spyros Arvanitis is a senior researcher at the KOF Swiss Economic Institute and a lecturer in economics at the ETH Zurich. He is head of the research section on innovation economics at the KOF. Dr Arvanitis holds a doctorate in economics from the University of Zurich and a doctorate in chemistry from the ETH Zurich. He has published extensivvel on the economics of innovation, technology diffusion, determinants of the performance of firms, and market dynamics. Luarent Donzé has been an associate professor at the Univerrsit of Fribourg, Switzerland since 2002. Before this appoinntmen he was a senior researcher at the KOF Swiss Economic Institute at the ETH Zurich. He teaches and researrche on statistics and econometrics, especially on measures of economic inequality, the construction and maintenance of panels of firms, and matching methods. Nora Sydow has been at the Economic Research Departmeen of Credit suisse since 2008. Before this appointment she was a researcher at the KOF Swiss Economic Institute at the ETH Zurich and responsible for the KOF Enterprise Panel. She holds a master's degree from the University of St gallen, Switzerland. Impact of technology policy on innovation by firms Science and Public policy February 2010 65 consensus not only among political actors but also among organizations representing business interests. According to the results of The swiss Economic Survve (Arvanitis et al. 2007), less than 10%of Swiss firms perceive a lack of public R&d promotion to be a strong, or very strong, obstacle to their innovation activities; this percentage has remained practically constant since 1990. As a consequence, only a few fiscal initiatives to support research and innovation at firm level have been launched in recent years. CTI is the government agency through which public funds are poured into the business sector. Besides the promotion of entrepreneurship through CTI's start-up funding programme plus a mobilization initiaativ called Venturelab, most of CTI funds are direccte to financing‘bottom-up'-initiated R&d projects from all scientific fields, CTI supporting the academic partner of the project. There have also been programmes for the promotion of specific technologies (e g. Medtech Topnano21) but this kind of specific support has always been of minor importance. The principle of indirect R&d support of good projects, which are proposed jointly by a private and a public partner, is fundamental to Swiss technology policy. To the best of our knowledge, it is unique in Europe as a main promotional policy. 2 Methods of evaluation of measures of technology policy Evaluating the outcomes of subsidized projects is difficult, both because of the difficulties in estimatiin the wider social benefits that they generate and because of the difficulties in assessing what the‘counterfactual'would have been in the absence of public support. Typically evaluations of outcomes, i e. estimations of the impact of policy, proceed by means of an ex post assessment of the activities of the firms that have received subsidies. Such evaluatiion can be subject to selection-bias problems becaaus subsidized firms are not a random group. They are selected mostly because of the high quality of the proposed projects, that is, those projects that are the best candidates for funding are also the projects that would have expected the largest output in the abseenc of funding. There are several empirical strategies for mitigatiin selection bias in the ex post evaluations e g. regresssio with controls for unobserved effects; regression with fixed effects or‘difference in differencces'selection models and matching methods based on direct comparisons of the participating and nonparticipating agents, i e. on matched samples of treated and untreated entities (Klette et al.,2000; Jaffe, 2002; Arvanitis and Keilbach, 2002. In this study we apply matching methods to evaluate the impact of R&d subsidies on the innovatiio performance of subsidized firms. A major advanntag of the matching methods over the regression approach is that the matching is nonparametric. As such, it avoids the functional form restrictions impliici in running a regression of some Kind of course, this method also has shortcomings. First, a close similarity with respect to all observable characterristic that are believed to be correlated with the likelihood that a firm or a project would be selected for subsidies may fail to control fully for any selectiio bias, given that in most cases only a restricted dataset of firm characteristics is available. Secondly, due to a lack of information, potential knowledge spillovers are taken not into consideration (this also happens when regressions are run). Theoretically, the only setup for a support measure in order to avoid ex ante selection bias would be to undertake an evaluation by awarding grants (subsidies) randooml within a pool of actors who are judged suitabbl for funding (Jaffe, 2002). But such a random mechanism for distributing subsidies also raises the issue of whether or not the social welfare would be lower if some projects with a high potential go withoou funding. Empirical evidence on the effectiveness of technology policy Recent overviews of the empirical literature suggest that the empirical evidence as to the effectiveness of subsidies is not homogeneous (David et al. 2000; Hall and Van Reenen, 2000; Klette et al. 2000; Jaumotte and Pain, 2005), a finding also confirmed by the meta-analysis by Garcia-Quevado (2004) which was based on the results of 39 studies of the effectivenees of public subsidies. All overviews emphasize the importance of the control variables included in any empirical assessment and the level of aggregation at which a study is conducted. Differences with respect to these two factors seem to explain a large part of the differences found between empirical studies. Thus, for the assessment of a study, it is necessary to take these two factors into consideration. Summary of similar studies In this section of the paper we review studies at firm level that aim to measure the impact of public fiscal support on some performance measure and apply The principle of indirect R&d support of good projects, which are proposed jointly by a private and a public partner, is fundamental to Swiss technology policy. To the best of our knowledge, it is unique in Europe as a main promotional policy Impact of technology policy on innovation by firms Science and Public policy February 2010 66 either matching approaches (as in this paper) or selecctio correction approaches. Most studies use contemporraneou data on the states of subsidized and non-subsidized firms (as in this paper. Table 1 preseent a summary of such studies. Seven of them reffe to European countries (Austria, Germany, Ireland, Spain and Switzerland), six of them apply matching approaches and one of them only uses a selection correction approach. Moreover, the study for Ireland combines selection correction approach and matching method, that for Austria uses both approaaches Finally, three of the non-European studies (USA, Japan and Israel) use versions of the selection correction method, while the Canadian study is based on a matching approach and is the only study that compares the impact of two different policy instrumments Six out of ten studies use R&d intensity, R&d expenditure or R&d personnel as the target variables of the promotional measures. For one study the target variable is innovation expenditure. The Canadian study uses eight different outputorieente innovation measures as target variables. Finallly in three studies some technology diffusion measure is chosen as the goal variable. Most studies Table 1. Summary of selected empirical studies Study/country Policy instrument being evaluated Number of firms Approach Impact on target variable Sakakibara (1997), Japan Government-sponsored cooperrativ R&d projects organized by Ministry of International Trade and Industry (1983 1989) 226 Selection correction: Two-equation system (participation eqn. R&d effort eqn. R&d spending:++Patents:++Busom (2000), Spain R&d subsidy programme 1988 154 Selection correction: Two-equation system (participation eqn.:R&d effort eqn. patent eqn. R&d expenditures: R&d personnel, R&d expenditures/sales, R&d personnel/employment:++Wallsten (2000), USA Small Business Innovation research (SBIR) Programme (1990 1992) 81 Selection correction: Three-equation system (two different participation eqns.:R&d spending eqn. employment eqn. R&d spending 1992: -employment 1993: no effect Arvanitis et al. 2002), Switzerland Programme of promoting use of Computer Integrated Manufacturing Technologies (CIMT)( CIM Programme, 1990 1996) 463 Selection correction: Two-equation system (participation eqn. CIMT adoption eqn. Change in CIMT intensity (1990 1996:++for firms with less than 200 employees+for firms adopting CIMT for first time Donzé (2002), Switzerland Programme of promoting use of CIMT (CIM Programme, 1990 1996) 463 Matched-pair analysis (several alternative methods) Change in CIMT intensity (1990 1996):++for firms with less than 200 employees+for firms adopting CIMT for first time Lach (2002), Israel R&d grants from Office of Chief Scientist at Ministry of Industry and Trade (1990 1995) 325 Difference -indifference estimator R&d spending:++for small firms no effect for large firms Czarnitzki and Fier (2002), Germany Public innovation subsidies in German service sector 210 Matched-pairs analysis (nearest neighbour matching) Innovation expenditure: innovation expenditure/sales:++Almus and Czarnitzki (2003), Germany R&d subsidies to East german firms (1994,1996, 1999) 622 Matched-pairs analysis (calliper matching) R&d intensity:++Pointner and Rammer (2005), Austria Programme of promoting use of CIMT (Flexcim Programme, 1991 1996) 301 (a) Selection correction: Two-equation system (participation eqn.:CIMTADOPPTIO eqn. b) matched-pair analysis Change in CIMT intensity (1992 1998:++for firms with less than 200 employees+for firms with low intensity of CIMT use Görg andstrobl (2007), Ireland R&d grants from (Industrial Development Agency (IDA) Ireland and Forbairt (1999 2002) 828 Combination of matching approach and difference-indiffeerenc estimator R&d spending; R&d spending per employee: small domestic firms:++medium domestic firms: no effect; large domestic effects: -all size classes of foreign firms: no effect Bérubé and Mohnen (2007), Canada R&d tax credits versus R&d tax credits+R&d grants 584 Matched-pairs analysis (nearest neighbour matching) Firms with tax credits +R&d grants are more innovative than firms with only tax credits for 6 out of 8 innovation indicators Notes:+(+-positive (negative) and statistically significant effect at 10%test level Impact of technology policy on innovation by firms Science and Public policy February 2010 67 find a positive policy effect but in some cases only for small firms. The USA study is the only one, which finds a negative effect for R&d spending, meaning that subsidies were crowding out private R&d spending. Although all the studies in Table 1 refer to the firm as the analytical entity, a closer comparison of the results of these studies is not possiibl due to large differences with respect to the variables taken into consideration in order to control for selection bias. Database Our information sources were: a list of the firm projects that were subsidized by the CTI in the period 2000 2002; additional information on the firms whose projeect were subsidized that was collected through a survey of the subsidized firms based on a shortenne version of the questionnaire used in The swiss Innovation Survey 2002; and the data for firms that reported the introduction of innovations in the period 2000 2002 in The swiss Innovation Survey 2002. The CTI database contained information on 634 subsiddize R&d projects that were finished between 1 january 2000 and 31 december 2002. There was informmatio on the scientific field of the project, the amount of the subsidy granted, and the name and address of the enterprises that conducted the subsidiize projects. These firms made up our sample of subsidized firms. Start-ups, nonprofit organizations and mergers were excluded from this sample becaaus their specific characteristics could be not identiffie in our pool of control firms. Further firms that had ceased to exist by December 2003 were also remoove from the sample. The final sample contained 307 subsidized firms. These firms received a shortenne version of the questionnaire of The swiss Innovattio Survey 2002.3 185 firms completed the questionnaire (see Table A1 in the Appendix to this paper for information on the response rates by scientiifi field. A further 14 subsidized firms were identiffie among the participants of The swiss Innovation Survey 2002. Hence, the sample we used for the study contained data on 199 firms (64.8%of the subsidized firms. Additional information on the determiinant of the propensity scores (see section on Method) was collected through a telephone survey of the 122 subsidized firms that did not complete the postal survey. This additional information allowed us to estimate the propensity scores based on data for all 307 subsidized firms. The 996 firms that participated in The swiss Innovattio Survey 2002 and reported the introduction of innovations in the period 2000 2002 built the pool of non-subsidized firms from which a control group was constructed (KOF panel database). For the firms that finished their projects subsidiize by the CTI during the first half of the period 2000 2002, i e. until the middle of 2001, we reckon that they would still have had one -and-a-half years until the end of the reference period to realize some impact of these projects on their innovation performmanc (e g. introduce new products); one-and-ahaal years is an adequate time lag between R&d and realization of R&d outcomes for most industries and for incremental innovations. For the firms that compleete their subsidized R&d during the second half of the reference period, particularly in the year 2002, it is questionable, whether or not they would have had enough time until the end of 2002 to realize any additional innovation gains. 53%of projects were finished by the middle of 2001,78%by the end of 2001. Hence, for the large majority of the projects there was enough time to have a measurable impact of R&d on their innovation performance. For the remaining 28%of the firms it is possible that only part of the impact could be realized before the end of 2002. In this sense our estimations of the impact of CTI promotion would thus represent a lower bound on the possible effects. Patterns of CTI promotion in period 2000 2002 As already mentioned, in the period 2000 2002 634 R&d projects were supported by the CTI. Table 2 shows the scientific fields in which these projects were located and the amount of the subsidies granted by scientific field. The projects in the fields of machiiner and apparatus construction as well as informattio technology (software) amounted to about 33%of all projects and also received about 33%of the total subsidies. In general the subsidies were distributed rather broadly among several scientific fields, which was in accordance with the general promotion policy of the CTI, based mainly on the‘bottom-up'principle of support. So-called futureorieente technologies such as biotechnology (3. 6%of projects, 4. 5%of subsidies) and nanotechnology (5. 7%of projects, 3. 8%of subsidies) do not seem to have been promoted particularly. In total, about 120 million Swiss francs (CHF) were invested in projects promoted by the CTI, i e. CHF60 million per annum. The mean subsidy per project was CHF190, 000. The mean amounts among scientific fields varied betwwee CHF167 000 for information technology and CHF267, 000 for microelectronics. This means that including the firms'contribution of at least the same amount as the CTI subsidy, about CHF400, 000 was invested per project. Table 3 shows the distribution of subsidies among firms by scientific field. Enterprises with more than one project were classified by the scientific field of the project with the highest subsidy. The share of firms with projects in machinery, apparatus construuctio and information technology is about 22, %Impact of technology policy on innovation by firms Science and Public policy February 2010 68 significantly lower than the respective share of projeect of these scientific fields. In contrast material sciences are represented better among firms (about 24%)than among projects (about 12%.%The subsidized firms are characterized further by the industry affiliation and the number of employees in full-time equivalents (firm size. 52%of promoote firms belonged to mechanical and electrical machinery, electronics and instruments. This was the dominant group among subsidized firms in accordaanc with the importance of these capital goods industtrie for Swiss manufacturing with respect to generated value added, employment and innovativeneess even if it is represented rather over. Chemical and pharmaceutical firms, which are on average the most innovative Swiss firms, are quite underrepreesente among the subsidized firms (4%),reflecctin the strong tendency of this branch of aboveaveerag investment in R&d. With the exception of wholesale trade the service sector is represented in the sample of the subsidized firms only by business services (computer services engineering, business consulting, etc. about 21%.%Small firms with up to 50 employees have a share of about 55%,firms with more than 200 employees a share of only about 25%,firms with more than 500 employees a share of about 10%.%Both the distribution among industries and among firm size classes seem to be in accordance with the policy pursued by the CTI of promoting mainly small-and medium-sized enterprises in all sections of the economy; there is even a tendency to promote small-rather than medium-sized firms. Method Our main hypothesis is that the CTI support, particulaarl through co-financed research projects in cooperratio with universities, would show on average a significantly higher innovation performance, as measured by output innovation measures (e g. sales share of innovative products), than‘structural similaar firms without such activities. We used several matching methods to demonstrate this. In order to measure appropriately the influence of CTI subsidies on a firm's innovation performannc(‘treatment effect')4 we should be able to measure the performance difference of the two‘states'of a firm (subsidized by the CTI(‘treated')/non-subsidized by the CTI (‘non-treated'keeping all other things equal. In a cross-sectional framewoork usually only one of these two possible states is observable: either a firm is subsidized or it is subsidized not. Thus, in most cases it is not possible to make a proper comparison of these states. Heckman et al. 1998) developed a methodology to approximate this non-observable(‘counterfactual')state of a certain firm with the observable same state of another firm which is‘structurally similar'to the first one according to a series of firm characteriistic formally defined by a vector X. Thus, besiide the group of firms, which are subsidized by CTI in a certain time period, we need a pool of Table 2. Subsidized projects and volume of subsidy by scientific field 2000 2002 Scientific field Number of projects Percentage CTI subsidy (in CHF (Swiss francs)) Percentage CTI subsidy per project (in CHF) Construction technology 27 4. 3 3, 801,686 3. 1 140,803 Biology 23 3. 6 5, 462,365 4. 5 237,494 Electrical machinery/electronics 32 5. 0 6 , 477,776 5. 4 202,431 Information technology 103 16.2 17,235, 837 14.3 167,338 Machinery, construction of apparatus 105 16.6 22,735, 819 18.8 216,532 Material sciences 56 8. 8 13,992, 873 11.6 249,873 Microelectronics 48 7. 6 12,810, 767 10.6 266,891 Nanotechnology 36 5 . 7 4, 537,160 3. 8 126,032 Process engineering 41 6. 5 8 761,137 7. 2 213,686 Production/management concepts 51 8. 0 8, 406,303 7. 0 164,829 Other 112 17.7 16,631, 768 13.8 148,498 Total 634 100.0 120,853, 491 100.0 190,621 Source: CTI database, authors'calculations Table 3. Subsidized enterprises by scientific field 2000 2002 Scientific field Number of firms Percentage Construction technology 11 5. 5 Biology 7 3. 5 Electrical machinery/electronics 12 6. 0 Information technology 21 10.6 Machinery, construction of apparatus 23 11.6 Material sciences 48 24.1 Microelectronics 21 10.6 Nanotechnology 6 3. 0 Process engineering 16 8. 0 Production/management concepts 14 7. 0 Other 20 10.1 Total 199 100.0 Notes: Enterprises with more than one project were classified by scientific field of project with highest subsidy Source: CTI database, authors'calculations Impact of technology policy on innovation by firms Science and Public policy February 2010 69 firms which are subsidized not out of which‘structurrall similar'firms are selected according to a‘proximity'criterion (control group). The comparisso of the two states for subsidized and nonsubsiidize firms is performed by comparing the means of the innovation performance variables for the‘treated'firms and the‘twin'‘non-treated'firms matched to the‘treated'ones according to a proximity criterion. The multi-dimensionality of the matching problem (matching with respect to each single element of a vector X of firm characteristtics can be reduced under certain conditions (Rosenbaum and Rubin, 1983) to a monodimennsiona (scalar) propensity score which comprehhend the entire information of all relevant characteristics. 5 The state of a firm belonging to the group of the‘treated'firms is described by d=1, the state of a‘non-treated'firm by d=0. If Y1i is a vector of innovattio measures for the treated firm i i (d=1) and Y0i the corresponding vector for a firm j belonging to the control group j (d=0), which is the‘twin'firm to firm i, then the performance difference betwwee the two firms is defined as: Y=Y1i-Y0i (1) In a first step we estimated by a probit model the propensity scores P (X i e. we estimated the probabiilit of a firm having a research project subsidized by the CTI as a function of a vector X of firm characterristic As independent variables X we used: a variable characterizing a firm's R&d activities (continnuou vs. occasional), the degree of exposure to internaationa competition (export activities yes/no), age of firm(‘firm founded before 1996'),size of firm (dummy variables for six size classes), industry affiliation (dummy variables for three sub-sectors), geographical location (dummy variables for six geograpphica regions) and language of the questionnaire (see Table A2 in the Appendix to this paper for the results of the probit estimates). In a second step all firms were distributed to adjusttmen cells according to the quintiles of the propennsit scores estimated by the equation in Table A2. The search for a‘twin'firm is restricted then only to the firms of the same adjustment cell, i e. the quintile of propensity scores. In a third step the‘structurally similar'firm inside an adjustment cell was identified for each treated firm. In order to test the robustness of our results, we used four different matching methods to identify the structurally similar firms out of the pool of the nontreeate firms. According to the first method used in this study nearest neighbour matching, the‘twin'firm j to firm i is one fulfilling the condition: minij Pi Pj (2) where Pi, and Pj are propensity scores for the firms i and j, respectively. The treated firm can have a higher or a lower propensity score than the nontreeate one, therefore the absolute value of the differrenc of the two propensity scores has to be considered. The second method used in this study, calliper matching, is based on the same proximity measure as the nearest neighbour method which in this case is restricted up to a certain value e (maximum admissibbl difference of the propensity scores): Pi Pj<e (3) Different adjustment cells can have different e values. The e values are dependent on the distributiio of the propensity scores inside an adjustment cell. According to the third method, kernel matching, a weighted sum of all available control group firms inside an adjustment cell, not a single‘twin'firm as in the other two methods, is ascribed to every treated firm. The performance difference between the treated and the non-treated firms is defined now as:(4) where ij w is the weighting factor({}0 1; 0 1, ij ij j d w w d===The weighting factor in equation (4) is defined as:(5) where is the kernel6 at the point is the bandwidth of the kernel The bandwidth was set specifically for every adjusttmen cell. Also in this case the choice of the bandwidth was dependent on the distribution of the propensity scores in the adjustment cells. The fourth and last method, the local linear regresssio matching, is based on the same concept as kernel matching. In this case all available observatiion of the control group are given also a specific weight. This weight is high for small‘distances'betwwee a pair of firms, low for large‘distances'and also contains a linear term. The weighting factor is defined as follows:{ {1 0 0 i ij j j d Y Y wy} ==-0 ij ij ik k d G w G==0()i k N P P a-0()i k ik N P P G g a-=0 N a Impact of technology policy on innovation by firms Science and Public policy February 2010 70 (6) where and is the kernel7 at the point In a fifth step, the means of the variables measuring innovation performance of the group of the treated firms and the group of the‘twin'non-treated firms were compared. We used six innovation variabble covering the output side of the innovation process: an ordinal measure of the technical importance of the introduced product and process innovations; 8 an ordinal measure of the economic importance of the introduced product and process innovations; 9 percentage reduction of average variable productiio costs due to process innovation; sales of new products new to the firm or to the market as a percentage of total sales; sales of significantly improved or modified (alreead existing) products as a percentage of total sales; and sales of products new to the market worldwide. We use several innovation indicators in order to test the robustness of our results given that innovation is a latent phenomenon and every single indicator measures only partly aspects of this complex phenomenon. In a sixth and last step we calculated a subsidy quotient for every subsidized firm by dividing the amount of the granted subsidy by the total R&d expendiiture in the period 2000 2002. This subsidy quotient measured the relative magnitude of the subsiidy10 We divided the subsidized firms into two groups: one group with firms with a subsidy quotient higher than the median(‘high-subsidy'firms) and a second one with firms with a subsidy quotient lower than the median(‘low-subsidy'firms. Then, we calcullate the difference of the means between subsidiize and non-subsidized firms separately for the‘high-subsidy'and the‘low-subsidy'firms. We tested if the difference in the former case was significcantl larger than the difference in the latter case. If this was the case, we interpreted this result as empirrica evidence that the impact of the CTI subsidies was correlated positively to the magnitude of the subsidy quotient. Hence,‘high-subsidy'firms would show a larger impact than the‘low-subsidy'ones. Results of the matched-pairs analysis Comparison of the innovation performance of subsidized firms depending on the subsidy quotient Table 4 provides a qualitative summary of the resuult of the comparison of the innovation performannce as measured by six different indicators, of the subsidized and the non-subsidized firms for four different matching methods. We calculated the differrenc of the means of the two categories of firms (subsidized, non-subsidized) for six innovation variables and four matching methods, i e. for 24 differren cases. With one exception(‘importance of introduced innovations from an economic point of view';‘'‘nearest neighbour'method) we found that the subsidized firms showed a significantly higher 0 1(,)N N A b W i j C D)} -=-2 0 0 ij ik k i k d ij j i ik k i k d A g G P P B G P P G P P===2 0 0 2 0 ij ik k i j d k d ik k i k D c G g P P D G P P===k i ik N P P G g a-=0()i k N P P a-Table 4. Summary of results with respect to receiving a subsidy for various matching methods Variable Significantly higher means of subsidized than of nonsubsiidize firms (after matching) Nearest neighbour Calliper Kernel Local linear regression Importance of introduced innovations from a technical point of view*Yes Yes Yes Yes Importance of introduced innovations from an economic point of view*No Yes Yes Yes Percentage reduction of average variable production costs due to process innovation Yes Yes Yes Yes Sales of significantly improved or modified (already existing) products as a percentage of total sales Yes Yes Yes Yes Sales of products new to firm or to market as a percentage of total sales Yes Yes Yes Yes Sales of products new to market worldwide as a percentage of total sales Yes Yes Yes Yes Notes:**Originally ordinal variable measured separately for product and process innovations on a five-point Likert scale (1=very small, 5=very high. Mean values are used for product and process innovations. Statistical significance: 5%test level Impact of technology policy on innovation by firms Science and Public policy February 2010 71 innovation performance than non-subsidized firms (at the 5%test level. Hence, these results seem to be quite robust across various methods and innovatiio indicators. Having controlled for the size and age of the firms, sector affiliation, region, export propensity, and the existence of continuous R&d activiitie in the propensities equation, these performannc differences have to be traced with good reason to the main difference between the two groups of firms, namely having or not having received subsidiie from the CTI in the reference period. For the effectiivenes of CTI promotion policy is the result for the six output-oriented innovation indicators of particular interest. Subsidized firms show a significanntl higher innovation performance than structuralll similar non-subsidized enterprises. The detailed results in terms of figures for each innovation measure and each method are found in Tables A3 A6 in the Appendix. For example, coluum 1 in Table A3 shows the mean value (score) for every innovation indicator for all available non-subsidized firms before matching. Column 2 presents the mean values for the matched non-subsidized firms, i e. those firms that were seleccte (out of the pool of non-subsidized firms) by the matching method used (in this case:‘‘nearest neighbour'method) as‘similar'to the subsidized ones. The figures in the latter case are systematically larger than in the former case, reflecting the fact that firms with a high innovation performance are seleccte by the applied method to match subsidized firms that are expected to be highly innovative in ordde to obtain grants. Column 3 shows the corresponndin figures for the subsidized firms. Column 4 shows the difference between the mean values for the subsidized firms (column 3) and the mean values of the matched non-subsidized firms (column 2). Finally, column 5 presents the results of tests of the statistical significance of the differences in column 4. These results show that there are substantial differeence in innovation performance. For the outputorieente indicators the differences vary significantly between only 9 11%for the qualitative selfassesssmen of the technical importance of the innovattion introduced and a threefold to fivefold larger magnitude in the case of sales of products new to the market. A further interesting point, particularly for policy-makers, is subsidized that firms seem to be significantly more innovative especially in terms of new products than non-subsidized ones. Comparison of the innovation performance of high subsidy'and‘low subsidy'firms Table 5 contains a qualitative summary of the resuult of the comparison of the differences of the innovation performance of high-subsidy 'and It is interesting to note, particularly for policy-makers, that subsidized firms seem to be significantly more innovative, especially in terms of new products, than non-subsidized ones Table 5. Summary of results with respect to the magnitude of the subsidy quotient for various matching methods Variable Significantly higher differences of differences of means of subsidized and non-subsidized firms (after matching) for subsidized firms with a subsidy quote>median than for subsidized firms with subsidy quotient<median‘Nearest neighbour'‘Calliper'‘Kernel'‘Local linear regression'Importance of introduced innovations from a technical point of view*Yes Yes Yes Yes Importance of introduced innovations from an economic point of view*No No No No Percentage reduction of average variable production costs due to process innovation Yes Yes Yes Yes Sales of significantly improved or modified (already existing) products as a percentage of total sales Yes Yes Yes Yes Sales of products new to the firm or to the market as a percentage of total sales Yes Yes Yes Yes Sales of products new to the market worldwide as a percentage of total sales Yes Yes Yes Yes Notes:**Originally ordinal variable measured separately for product and process innovations on a five-point Likert scale (1=very small, 5=very high. Mean values are used for product and process innovations. Statistical significance: 5%test level Impact of technology policy on innovation by firms Science and Public policy February 2010 72‘low-subsidy'firms from that of the respective groups of non-subsidized firms. For five innovation indicators we found that the difference of the means of the‘high-subsidy 'and the non-subsidized firms is significantly higher (at the 10%level) for all four matching methods than the respective differeence for the‘low-subsidy'firms (i e. significanntl positive difference of the differences). Hence, for these cases we have some empirical evidence that the impact on innovation performance is dependent on the relative magnitude of the subsiid granted. The larger the amount of the subsidy relative to a firm's own R&d investment, the stronger is the impulse for the innovation performannc of a firm. For one innovation variable(‘importtanc of introduced innovations from an economic point of view')we could not find any significant effect, meaning that relatively larger subsidies do not necessarily result in a stronger tendeenc by subsidized as compared to non-subsidized firms to introduce innovations that are economicaall important. It appears that larger subsidies resuul in more technologically important innovations in subsidized firms than in non-subsidized firms. This is understandable given that all subsidized collaborration are between firms and universities that provide cooperating firms with knowledge that is primarily of high technological value. This does not mean that higher subsidies cannot generate (additioonal economic success: according to our results the larger the subsidy (in relative terms), the larger the impact effect for a series of indicators that measure the economic success of innovation (sales shares of products with different grades of innovativeeness reduction in costs. More detailed results in terms for figures for each innovation measure and each method can be found in Tables A7 A10 in the Appendix. For example, column 1 in Table A7 shows the differences betwwee subsidized firms with subsidy quotients smaller than the median and the corresponding matched non-subsidized firms. Column 2 presents the results with respect to the statistical significance of these differences. Columns 3 and 4 show the differeence between subsidized firms with subsidy quotieent larger than the median, column 4 refers to the statistical significance of these differences. Finally, column 5 reports on the results of tests of the statistiica significance of the difference of the differences of the means. As we can see the difference between subsidized and non-subsidized firms, for example, for the sales shares of products that are new worldwiid for firms with small subsidy quotient increases from 7. 10 percentage points to 12.60 percentage points for firms with large subsidy quotients. The respecctiv increase for the sales shares of new producct (either new to the firm or new to the market) amounting to 18.20-8. 00=10.20 percentage points as well as for significantly improved products (amounting to 14.90-7. 60=7. 30 percentage points) are even larger. Conclusion Based on a matched-pairs analysis of 199 firms supporrte by the CTI in the period 2000 2002 and a control group of 996 firms that were supported not by the CTI, we found that the CTI promotion significcantl improved the innovation performance of supported firms with respect to six different measurre of innovation performance. This could be shown by four different matching methods (with the exception of the nearest neighbour method for the indicator‘importance of introduced innovations from an economic point of view'.'A further finding was that the magnitude of the impact correlated positively with the relative size of financial support as measured by the quotient of the volume of financial support to the volume of a supporrte firm's own R&d expenditures. The present analysis yields some information on three policyrellate issues: the type of enterprises that received subsidies from the CTI; the effectiveness of CTI promotion policy; and the relationship between subsidy quotient and poliic effectiveness. The results of the study show a positive picture of the CTI's promotion policy. Subsidized firms are mainly small-and medium-sized enterprises (perhaps too many micro-firms among them) whose promotion is an explicit goal of CTI policy, the technological orientaatio of subsidized projects is quite broad, also covering currently fashionable fields such as biotechnnolog and nanotechnology. Further, subsidized firms represent a wide spectrum of manufacturing firms, the concentration on firms for machinery, electroonic and instruments reflecting the current structure of Swiss manufacturing. The‘bottom-up'principle applied by the CTI for allocating funds seems to be quite effective. An additional positive element is that policy is not just effective but it becomes more effectiiv if the financial support is raised. All this is also in accordance with the general principles of The swiss technology policy tending to be‘non-activist',providiin primarily for the improvement of framework condittion for private innovation activities. Even if a policy measure is successful from a microecconomi point of view, it still remains an open question whether or not this policy measure is also relevant in macroeconomic terms. In the case of the CTI policy investigated in this paper, it is questionabbl if an amount of about CHF60 million in 2004 (meanwhile CHF100 150 million of additional R&d support per annum) could have a discernible impact on an economy that invested about CHF19 billion in R&d in 2004. A further open question is of course, if some kind of‘functional equivalent'of this policy at a broader base, e g. R&d tax incentiive would do better, but such a discussion would be beyond the scope of this empirical paper. Impact of technology policy on innovation by firms Science and Public policy February 2010 73 Appendix Table A1. Survey of subsidized enterprises: structure of answering enterprises by scientific field Scientific field Number of addressed enterprises Number of answering enterprises Percentage share of answering enterprises Construction technology 16 11 68.8 Biology 13 7 53.8 Electrical machinery/electronics 18 12 66.7 Information technology 38 20 52.6 Machinery 70 46 65.7 Material sciences 33 20 60.6 Microelectronics 27 16 59.3 Nanotechnology 6 5 83.3 Process engineering 29 15 51.7 Production/management concepts 23 14 60.9 Other 34 19 55.9 Total 307 185 60.3 Table A2. Propensity of having a research project subsidized by CTI as function of various firm characteristics (probit estimation; dependent variable: research project subsidized by CTI in period 2000 2002, yes/no) Firm characteristics Test level 5%Firm characteristics Test level 5%Firm size: Sector: 20 49 employees-0. 31 Traditional manufacturing-0. 54 (0. 11)( 0. 10) 50 99 employees-0. 52 Traditional service industries -1. 23 (0. 13)( 0. 23) 100 199 employees-0. 45 Modern service industries (0. 12) Region: 200 499 employees Region of Lake Geneva 500 999 employees Midlands region 1000 employees and over North western Switzerland-0. 30 Other characteristics:(0. 14) Continuous R&d activities 0. 40 Eastern Switzerland (0. 10) Central Switzerland Export activities 0. 43 Ticino (0. 11) Language of questionnaire: Firm founded before 1996-0. 86 French 0. 56 (0. 14)( 0. 10) German N 1317 Adj. Mcfadden-R2 0. 14 %concordance 76.50 Notes: Only coefficients of variables that were significant at the 5%level are reported All variables in table are dummy variables Reference group for firm size: up to 19 employees Reference sector: high-tech manufacturing; definition: high-tech manufacturing: chemistry, plastics, machinery, electrical machinery, electronics/instruments; modern service industries: banking/insurance, computer services; other business services; traditional manufacturing: food/beverage/tobacco, textiles, clothing/leather; wood processing, paper, printing, glass/stone/clay, metal, metalworking, watches, other manufacturing, energy; traditional service industries: wholesale trade, retail trade, transport/telecommunication, hotels/catering, personal services Reference region: Zurich Reference language: Italian (continued) Impact of technology policy on innovation by firms Science and Public policy February 2010 74appendix (continued) Table A3. Comparison of subsidized/non-subsidized enterprises, matched by‘nearest neighbour'method Measures of innovation performance All non-active firms before matching Non-active firms after matching (control group) Active firms Difference in means of active firms/non-active firms (column 3 column 2) Means Statistical significance (test level 5%)Importance of introduced innovations from a technical point of view*3. 34 (0. 03) 3. 44 (0. 05) 3. 75 (0. 06) 0. 31 (0. 08) Yes Importance of introduced innovations from an economic point of view*3. 36 (0. 03) 3. 60 (0. 06) 3. 65 (0. 06) 0. 005 (0. 081) No Percentage reduction of average variable production costs due to process innovation 4. 98 (0. 29) 3. 59 (0. 43) 8. 61 (1. 24) 5. 02 (1 . 32) Yes Sales of significantly improved or modified (already existing) products as percentage of total sales 33.73 (0. 84) 36.60 (1. 61) 48.36 (2. 39) 11.76 (3. 01) Yes Sales of products new to firm or to market as percentage of total sales 15.73 (0. 57) 17.24 (1. 39) 27.46 (2. 27) 10.22 (2. 73) Yes Sales of products new to market worldwide as a percentage of total sales 4. 44 (0. 39) 3. 01 (0. 36) 15.58 (2. 10) 12.57 (2. 10) Yes Notes:**Originally ordinal variable measured separately for product and process innovations on a five-point Likert scale (1=very small, 5=very high) Mean values are used for product and process innovations Number of non-subsidized firms=996; number of subsidized firms=199 Standard errors are in brackets under the means Two-tailed t-test used for difference of means Table A4. Comparison of subsidized/non-subsidized enterprises, matched by‘calliper'method Measures of innovation performance All non-active firms before matching Non-active firms after matching (control group Active firms Difference of means of active firms/non-active firms (column 3 column 2) Means Statistical significance (test level 5%)Importance of introduced innovations from a technical point of view*3. 34 (0. 03) 3. 36 (0. 02) 3. 75 (0. 06) 0. 39 (0. 06) Yes Importance of introduced innovations from an economic point of view*3. 36 (0. 03) 3. 43 (0. 01) 3. 65 (0. 06) 0. 22 (0. 06 Yes Percentage reduction of average variable production costs due to process innovation 33.73 (0. 84) 36.32 (0. 43) 48.36 (2. 39) 12.04 (2. 47) Yes Sales of significantly improved or modified (already existing) products as a percentage of total sales 4. 98 (0. 29) 5. 71 (0. 12) 8. 61 (1. 24) 2. 90 (1 . 24) Yes Sales of products new to firm or to market as a percentage of total sales 15.73 (0. 57) 17.28 (0. 27) 27.46 (2. 27) 10.18 (2. 34) Yes Sales of products new to market worldwide as a percentage of total sales 4. 44 (0. 39) 5. 94 (0. 18) 15.58 (2. 10) 9. 64 (2. 01) Yes Notes:**See footnotes to Table A3 for key (continued) Impact of technology policy on innovation by firms Science and Public policy February 2010 75 Appendix (continued) Table A5. Comparison of subsidized/non-subsidized enterprises, matched by‘kernel'method Measures of innovation performance All non-active firms before matching Non-active firms after matching (control group) Active firms Difference of means of active firms/non-active firms (column 3 column 2) Means Statistical significance (test level 5%)Importance of introduced innovations from a technical point of view*3. 34 (0. 03) 3. 39 (0. 02) 3. 75 (0. 06) 0. 36 (0. 06) Yes Importance of introduced innovations from an economic point of view*3. 36 (0. 03) 3. 46 (0. 01) 3. 65 (0. 06) 0. 19 0. 06) Yes Percentage reduction of average variable production costs due to process innovation 4. 98 (0. 29) 5. 85 (0. 11) 8. 61 (1. 24) 2. 76 (1 . 22) Yes Sales of significantly improved or modified (already existing) products as percentage of total sales 33.73 (0. 84) 36.01 (0. 39) 48.36 (2. 39) 12.35 (2. 44) Yes Sales of products new to firm or to market as ercentage of total sales 15.73 (0. 57) 16.94 (0. 30) 27.46 (2. 27) 10.52 (2. 36) Yes Sales of products new to market worldwide as percentage of total sales 4. 44 (0. 39) 5. 82 (0. 17) 15.58 (2. 10) 9. 76 (2. 10 ) Yes Notes:**See footnotes to Table A3 for key Table A6. Comparison of subsidized/non-subsidized enterprises, matched by‘local linear regression'method Measures of innovation performance All non-active firms before matching Non-active firms after matching (control group ) Active firms Difference of means of active firms/non-active firms (column 3 column 2) Means Statistical significance (test level 5%)Importance of introduced innovations from a technical point of view*3. 34 (0. 03) 3. 39 (0. 02) 3. 75 (0. 06) 0. 36 (0. 06) Yes Importance of introduced innovations from an economic point of view*3. 36 (0. 03) 3. 46 (0. 01) 3. 65 (0. 06) 0. 19 (0. 06) Yes Percentage reduction of average variable production costs due to process innovation 4 98 (0. 29) 5. 85 (0. 11) 8. 61 (1. 24) 2. 76 (1. 22) Yes Sales of significantly improved or modified (already existing) products as percentage of total sales 33.73 (0. 84) 36.01 (0. 39) 48.36 (2. 39) 12.35 (2. 44) Yes Sales of products new to firm or to market as percentage of total sales 15.73 (0. 57) 16.94 (0. 30) 27.46 (2. 27) 10.52 (2. 36) Yes Sales of products new to market worldwide as a percentage of total sales 4. 44 (0. 39) 5. 82 (0. 17) 15.58 (2. 10) 9. 76 (2. 10 ) Yes Notes:**See footnotes to Table A3 for key (continued) Impact of technology policy on innovation by firms Science and Public policy February 2010 76 Appendix (continued) Table A7. Results with respect to magnitude of subsidy quotient for 2000 2002, calculated using‘nearest neighbour'method Measures of innovation performance Subsidized firms: subsidy quotient>median Subsidized firms: subsidy quotient<median Difference of means of subsidized/nonsubsiidize firms Statist. signif. test level 10%)Difference of means of subsidized/nonsubsiidize firms Statist. signif. test level 10%)Difference of the difference of the means (column 3-column 2) Importance of introduced innovations from a technical point of view*0. 42 Yes 0. 18 Yes Yes Importance of introduced innovations from an economic point of view*0. 05 No 0. 03 No No Percentage reduction of average variable production costs due to process innovation 6. 80 Yes 3. 80 Yes Yes Sales of significantly improved or modified (already existing) products as percentage of total sales 13.90 Yes 8. 20 Yes Yes Sales of products new to firm or to market as percentage of total sales 17.90 Yes 7. 10 Yes Yes Sales of products new to market worldwide as percentage of total sales 15.50 Yes 9. 80 Yes Yes Notes:**See footnotes to Table A3 for key Table A8. Results with respect to magnitude of subsidy quotient (2000 2002) using‘calliper'method Measures of innovation performance Subsidized firms: subsidy quotient>median Subsidized firms: subsidy quotient<median Difference of means of subsidized/nonsubsiidize firms Statist. signif. test level 10%)Difference of means of subsidized/nonsubsiidize firms Statist. signif. test level 10%)Difference of difference of means (column 3-column 2) Importance of introduced innovations from a technical point of view*0. 46 Yes 0. 33 Yes Yes Importance of introduced innovations from an economic point of view*0. 13 No 0. 26 Yes No Percentage reduction of average variable production costs due to process innovation 4. 10 Yes 1. 90 Yes Yes Sales of significantly improved or modified (already existing) products as percentage of total sales 14.10 Yes 7. 20 Yes Yes Sales of products new to firm or to market as percentage of total sales 17.90 Yes 7. 70 Yes Yes Sales of products new to market worldwide as percentage of total sales 12.60 Yes 7. 20 Yes Yes Notes:**See footnotes to Table A3 for key (continued) Impact of technology policy on innovation by firms Science and Public policy February 2010 77 Notes 1. See Bozeman (2000); Georghiou and Roessner (2000; and Feller (2007) for recent reviews of the central issues related to the evaluation of the effectiveness of technology programmes. See also Science and Public policy (34 (10), 679 752) dedicaate to‘New frontiers in evaluation'.'Finally, see OECD (2006a) for an analysis more from the point of view of the policy-maker; Polt et al. 2001) for the role of framework conditions for the evaluation of industry university collaboratioons and Polt and Streicher (2005) for the evaluation of large programmes such as the Framework programmes of the Europeea Union. 2. For overviews of Swiss technology policy see OECD (2006b) and European commission (2008. Lepori (2006) gives a longteer analysis of public research policy primarily with respect to universities and public research organizations. Griessen and Braun (2006) deal with the problems of political coordination of innovation policies in Switzerland. Appendix (continued) Table A. 9. Results with respect to magnitude of subsidy quotient (2000 2002) using‘kernel'method Measures of innovation performance Subsidized firms: subsidy quotient>median Subsidized firms: subsidy quotient<median Difference of means of subsidized/nonsubsiidize firms Statist. signif. test level 10%)Difference of means of subsidized/nonsubsiidize firms Statist. signif. test level 10%)Difference of the difference of means (column 3-column 2) Importance of introduced innovations from a technical point of view*0. 39 Yes 0. 30 Yes Yes Importance of introduced innovations from an economic point of view*0. 08 No 0. 24 No No Percentage reduction of average variable production costs due to process innovation 3. 60 Yes 1. 70 Yes Yes Sales of significantly improved or modified (already existing) products as percentage of total sales 14.40 Yes 7. 60 Yes Yes Sales of products new to firm or to market as percentage of total sales 18.10 Yes 8. 10 Yes Yes Sales of products new to market worldwide as percentage of total sales 13.10 Yes 7. 20 Yes Yes Notes:**See footnotes to Table A3 for key Table A. 10. Results with respect to magnitude of subsidy quotient (2000 2002) using‘local linear regression'method Measures of innovation performance Subsidized firms: subsidy quotient>median Subsidized firms: subsidy quotient<median Difference of means of subsidized/nonsubsiidize firms Statist. signif. test level 10%)Difference of means of subsidized/nonsubsiidize firms Statist. signif. test level 10%)Difference of difference of means (column 3-column 2) Importance of introduced innovations from a technical point of view*0. 40 Yes 0. 31 Yes Yes Importance of introduced innovations from an economic point of view*0. 09 No 0. 24 No No Percentage reduction of average variable production costs due to process innovation 3. 80 Yes 1. 90 Yes Yes Sales of significantly improved or modified (already existing) products as a percentage of total sales 14.90 Yes 7. 60 Yes Yes Sales of products new to firm or to market as a percentage of total sales 18.20 Yes 8. 00 Yes Yes Sales of products new to market worldwide as a percentage of total sales 12.60 Yes 7 10 Yes Yes Notes:**See footnotes to Table A3 for key Impact of technology policy on innovation by firms Science and Public policy February 2010 78 3. The questionnaire may be obtained from the authors. It is available in German, French and Italian. 4. The expression‘treatment effect'comes from labour market research, where individuals are treated'via a concrete policy measure. It is used here analogously for firms subsidized by the CTI. 5. See Heckman et al. 1999) for a survey on various matching procedures. Caliendo and Huber (2005) and Caliendo and Kopeinig (2005) give overviews of recent developments with respect to matching methods. 6. We used a‘biweight kernel'(quartic kernel) for the function G(.)It is defined as follows: The results are sensitive, not to the kernel function used, but to the choice of the bandwidth. 7. We also used the‘biweight kernel'here (see Note 6). The bandwidth was determined as follows (Silverman, 1986): an=2. 7768 (H/1. 34) N-1/5 where N is the number of observations of the control group or the group of treated firms, and H is the distance between the quintiles. For the adjustment cell 5 we used a bandwidth of 0. 15.8. The ordinal variable was measured originally separately for product and process innovations on a five-point Likert scale (1=very small, 5=very high; here we use the mean values for the product and process innovations. 9. See Note 8. 10. There is some measurement error in this calculation due to the time incongruence between subsidies granted before the beginning of 2000 and R&d expenditures strictly referring to the period 2000 2002 that unfortunately cannot be quantified and corrected. In order to minimize the influence of this error we distinguish only two‘crude'groups of subsidized firms. References Almus, M, and D Czarnitzki 2003. The effects of public R&d subsiddie on firms'innovation activities: the case of Eastern Germaany Journal of Business and Economic Statistics, 21 (2), 226 236. Arvanitis, S and M Keilbach 2002. Evaluation methodologies, econometric models: microeconometric models. In RTD Evaluatiio Tool Box: Socioeconomic Evaluation of Public RTD Policies (EPUB), W Polt, J. Rojo, A Tübke, G Fahrenkrog and K Zinöcker (eds. pp 101 118. Vienna: European commission. Arvanitis, S, H Hollenstein and S Lenz 2002. The effectiveness of government promotion of advanced manufacturing technologiie (AMT: an economic analysis based on Swiss micro data. Small Business Economics, 19 (4), 321 340. Arvanitis, S, H Hollenstein, N Sydow and M Wörter 2007. Innovationsakttivitäte in der Schweizer Wirtschaf t Eine Analyse der Ergebnisse der Innovationserhebung 2005. Bern: Staatssekretarria für Wirtschaft. Arvanitis, S, J Von Arx, H Hollenstein and N Sydow 2004. Innovationsakttivitäte in der Schweizer Wirtschaft-Eine Analyse der Ergebnisse der Innovationserhebung 2002. Bern: Staatssekretarria für Wirtschaft. Arrow, K J 1962. Economic welfare and the allocation of resources to invention. In The Rate and Direction of Inventive Activity, R Nelson (ed.).Princeton, NJ: Princeton university Press. Bérubé, C and P Mohnen 2007. Are firms that received R&d subsiddie more innovative? UNU-MERIT Working Paper No. 2007-015. Maastricht: UNU-MERIT. Bozeman, B 2000. Technology transfer and public policy: a review of research and theory. Research policy, 29 (4/5), 627 655. Busom, I 2000. An evaluation of the effects of R&d subsidies. Economics of Innovation and New technology, 9 (2), 111 148. Caliendo, M and R Huber 2005. The microeconometric estimation of treatment effects: an overview. IZA Discussion paper No. 1653. Bonn, Germany: IZA (Institute for the Study of Labour. Caliendo M. and S Kopeinig 2005. Some practical guidance for the implementation of propensity score matching. DIW Discussiio Paper No. 485. Berlin: DIW (German Institute for Econoomi Research. Czarnitzki, D and A Fier 2002. Do innovation subsidies crowd out private investment? Evidence from the German service sector, ZEW Discussion paper No. 02-04, Mannheim, Germany: ZEW. David, P, B Hall and A Toole 2000. Is public R&d a complement for private R&d? Research policy, 29 (4/5), 497 529. Donzé, L 2002. Matched-pair analysis based on business survey data to evaluate the policy of supporting the adoption of advannce manufacturing technologies by Swiss firms, KOF Working Paper No. 65, July 2002. Zurich: KOF. European commission 2008. Annual innovation policy trends and appraisal report: Switzerland 2007 2008 European trend chart on innovation. Brussels: European commission. Feller, I 2007. Mapping the frontiers of evaluation of public-sector R&d programmes. Science and Public policy, 34 (10), 681 690. Garcia-Quevado, G 2004. Do public subsidies complement businees R&d? A meta-analysis of the econometric evidence. Kykloos 57 (1), 87 102. Georghiou, L and D Roessner 2000. Evaluating technology programmes: tools and methods. Research policy, 29 (4/5), 657 678. Görg, H and E Strobl 2007. The effect of R&d subsidies on privaat R&d. Economica, 74 (2), 215 234. Griessen, T and D Braun 2008. The political coordination of knowledge and innovation policies in Switzerland. Science and Public policy, 35 (4), 277 288. Hall, B and J Van Reenen 2000. How effective are fiscal incentiive for R&d? A review of the evidence. Research policy, 29 (4/5), 449 469. Heckman, J, H Ichimura, J Smith, J. and P Todd 1998. Characterizzin selection bias using experimental data. Econometrica, 66 (5), 1017 1098. Heckman, J J, R J Lalonde, and J A Smith 1999. The economics and econometrics of active labour market programmes. In Handbook of Labour Economics, A Ashenfelter and P E Todd (eds. pp 1865 2097. Amsterdam: Elsevier. Jaffe, A b 2002. Building programme evaluation into the design of public research-support programmes. Oxford Review of Econoomi Policy, 18 (1), 22 34. Jaumotte, F and N Pain 2005. An overview of public policies to support innovation, OECD Economics department Working papers No. 456. Paris: OECD. Klette, T J, J Moen, and Z Griliches 2000. Do subsidies to commerrcia R&d reduce market failures? Microeconometric evaluation studies. Research policy, 29 (4/5), 471 495. Lach, S 2002. Do R&d subsidies stimulate or displace private R&d? Journal of International Business Studies, 50 (4), 369 390. Lepori B 2006. Public research funding and research policy: a long-term analysis for The swiss case. Science and Public Policcy 33 (3), 205 216. Nelson, R R 1959. The simple economics of basic scientific reseaarch Journal of Political economy, 67 (3), 297 306. OECD 2006a. Government R&d funding and Company Behavioour Measuring Behavioural Additionality. OECD: Paris. OECD 2006b. OECD Reviews of Innovation policy: Switzerland. Paris: OECD. OECD 2007. Science, Technology and Industry Board Innovatiio and Performance in the Global economy. Paris: OECD. Pointner, W and C Rammer 2005. Wirkungsanalyse. In Diffusionsorienntiert Technologiepolitik Eine vergleichende Wirkungsanaalys für Österreich, die Schweiz, Deutschland und die USA, W Pointner and W Polt (eds..Vienna: Leykam. Polt, W c Rammer, H Gassler, A Scibany and D Schartinger 2001. Benchmarking industry science relations: the role of framework conditions. Science and Public policy, 28 (4), 247 258. Polt, W and G Streicher 2005. Trying to capture additionality in Framework programme 5: main findings. Science and Public policy, 32 (5), 367 373. Rosenbaum, B R and D B Rubin 1983. The central role of the propensity score in observational studies for causal effects. Biometrika, 70 (1), 41 55. Sakakibara, M 1997. Evaluation of government-sponsored R&d consortia in Japan. In Policy Evaluation in Innovation and Technology Towards Best Practices, pp 225 253. OECD: Paris. Silverman, R 1986. Density Estimation. London: Chapman and Hall. Wallsten, S C 2000. The effects of government industry R&d programmes on private R&d: The case of the Small Business Innovation research Programme. Rand Journal of Economics, 31 (1), 82 100.0 2 2 15/16 1-()i k N P P a

< Back - Next >

Overtext Web Module V3.0 Alpha
Copyright Semantic-Knowledge, 1994-2011