Master thesis
The regional innovation paradox:
R&D subsidy application behaviour based on technological intensity and region
University of Groningen Faculty of Economics and Business
By:
Niels Reidsma - s3398625 Strategic Innovation Management
Supervisor: Dr. Florian Noseleit Co-assessor: Dr. Charlie Carroll
Date: 21-01-2019
Words: 10.277
Abstract
The regional innovation paradox describes the contradiction between the greater need of lagging regions for R&D subsidies and their incapacity to apply for those funds, and the attribution of those subsidies to more technologically advanced, urban regions. We are contributing to a gap in the literature by trying to measure what the influence of the technological intensity and the region is on the total subsidy applied for in the WBSO, a Dutch R&D subsidy program. The findings of this paper have allowed us to conclude that the technological intensity and the region account for different results regarding the applications for R&D subsidies. The three hypotheses of this paper were supported. Firms who are high- tech or are located in urban regions do apply on average for higher amounts of R&D subsidies, and high-tech firms in urban regions do apply for even more. The model explains up to 27 percent of the overall variance in subsidy applications for the WBSO. Both the technological intensity of firms and the region thus accounts for different results with regard to the application to the WBSO, policy makers have to consider this as it has strong implications on the innovation policy. They should account for it to make sure that there is a fair distribution of R&D support.
Keywords: R&D subsidy, WBSO, application behaviour, technological intensity, region
Introduction
Innovation activity is not randomly distributed but geographically clustered in a small number of major cities (Asheim and Gertler, 2005). The more technological advanced the innovation activity, the more it is geographically clustered. This despite the many attempts of other places to attract the innovation activity or generate their own. Over the years, scholars have found that this spatial concentration of innovation activities has grown (Cortright and Mayer, 2002;
Feldman, 2001). Public agencies are aiming to technologically upgrade low-tech firms and stimulate declining regions through their Research and Development policies (Blanes and Busom, 2003; Clausen, 2007). Governmental funding of R&D activities is considered as one of the most important tools for stimulating economic growth. Governments fund business R&D through subsidies or grants and offer fiscal incentives to firms who engage in R&D activities.
Governmental funding of business R&D accounts for 9% of total performed business R&D in the OECD (OECD, 2017).
Most of the research on governmental funding is focused on the effectiveness of tax incentive schemes. Scholars have discovered mixed results on the firms’ R&D intensity, some report crowding-out effects (Busom, 2000; Wallsten, 2000) and others reject that public R&D substitutes private R&D investment (Aerts and Schmidt, 2008; Almus and Czarnitzki, 2003).
A common result within innovation research is that innovation levels and the effectiveness of R&D subsidies differ between industries and regions (Castellacci and Lie, 2015; Herrera and Nieto, 2008). There has been found that high-technological industries have a higher probability of receiving a R&D subsidy than low-technological industries (Busom, 2000; Herrera and Nieto, 2008). Rural regions have been proven to be less innovative compared to urban regions, their R&D intensity is lower and are more concerned with incremental and process innovations (Audretsch and Feldman, 1996; Fritsch, 2000; Tödtling and Trippl, 2005).
Public funding of R&D is justified by economic theories and is based on the argument that firms underinvest in R&D because the social return is higher than the private returns, and thus facilitates knowledge spillovers (e.g. David, Hall and Toole, 2000; Feldman and Kelly, 2006).
However, knowledge spillovers have geographical boundaries and mainly can be observed in high-technological industrial clusters and urban innovative regions (Bottazzi and Peri, 2003;
Tödtling and Trippl, 2005). They examined innovation policies and advocate for differentiated
innovation policies based on industry- and regional characteristics. Many scholars have thus
examined the effectiveness of R&D support schemes, but surprisingly enough there has been
rarely looked at the application behaviour of firms based on their technological intensity and
the region of location. We therefore focus on the influence of these factors on the R&D subsidy application behaviour and aim to answer the following research question:
How does the technological intensity and the region explain variations in R&D subsidy application behaviour?
According to David et al. (2000) there is a selection bias in the allocation of R&D funding schemes, and before assessing the effectiveness of these R&D funding schemes there must be examined what characteristics influence the probability of applying for a subsidy. In addition, Lee (2011) and Silva, Silva and Carneiro (2017) argue that it is important to investigate the firm and industry characteristics before testing the efficiency of R&D funding schemes, because this could improve future research. By answering the research question, we aim to contribute to a more extensive framework of the industrial and geographic characteristics of the firms that apply for R&D subsidies. These characteristics can be implemented by scholars in their evaluation of R&D funding schemes.
R&D is considered as an important driver of economic growth and therefore it is important that firms apply for the R&D funding schemes. This paper uses a data set from the Netherlands, to evaluate the R&D subsidy application behaviour. In the Netherlands, the WBSO is the most important tool for stimulating business- and academic performed R&D. The latest number of firms that applied for the WBSO in 2016 is 4.9% (Rijksdienst voor Ondernemend Nederland, 2017b). This paper will give policymakers the possibility to evaluate their innovation policies and review if they are successful in achieving their goals.
Applying for the public R&D subsidies is seen as a difficult and resource consuming task (Aerts
and Schmidt, 2008; Busom, 2000). Therefore, firms in the Netherlands are increasingly using
an intermediary, whom applies for the WBSO subsidy in the name of the firm (Rijksdienst voor
Ondernemend Nederland, 2017b). This paper used the data set of such an intermediary and the
results are therefore also useful for subsidy-consultants and in particular the company who
provided the data set. With the results they will be able to look at the geographical distribution
of the applications and the insights from the literature on the most important geographical
characteristics that are of interest could provide them with additional knowledge. Therefore, it
could help them improve their subsidy allocation rate and could deliver new insights for
promotional activities along with identifying possible clients and target markets.
This paper is structured as follows: first the theoretical framework behind the research question
is developed. This framework includes the reasoning towards the hypotheses. In the next
chapter the background and preparation of the data is discussed, along with the methodological
choices for analysing it. This is followed by the results of the statistical analysis. Furthermore,
the results are discussed, along with the possible theoretical and managerial implications
regarding the results. At last the limitations of this research are stated and there are
recommendations made for future research.
Theoretical framework
This chapter further elaborates on the research question of this paper: “How does the technological intensity and the region explain variations in R&D subsidy application behaviour?”. The reasoning behind the governmental support of R&D incentive schemes is first discussed, followed by a contextual explanation of the Dutch setting. The conceptual model is then developed from the theoretically identified influencers of R&D subsidy application behaviour.
Literature review
Governments fund business R&D through subsidies or grants and offer fiscal incentives to firms who engage in R&D activities. Subsidies and grants raise the private marginal rate of return to R&D investment, tax incentives reduce the marginal cost of R&D (Hall & van Reenen, 2000).
The government aims to overcome underinvestment in R&D and stimulate knowledge spillovers, and thereby hopes to stimulate sustainable economic development (e.g. Czarnitzki and Hussinger 2004). According to them there are market failures, thus private expenditures on R&D are below the socially optimal level and public intervention is desired. Blanes and Busom (2003) and Clausen (2009) have evaluated the goals of public agencies regarding R&D subsidies. They argue that the public agencies: firstly, want to correct market failures by funding R&D projects that would not be carried out without subsidy; secondly, to technologically upgrade firms in low-tech industries and declining regions; and thirdly, to stimulate the so called ‘national champions’.
Only in the last couple of years, scholars have concerned themselves with the issue of firms’
participation in R&D incentive schemes. While most of the studies are concerned with the effectiveness of such schemes, just a few studies review the determinants of firms’ participation (e.g. Blanes and Busom, 2004; Aschhoff, 2010). Due to differences in methods and samples, the actual effectiveness of R&D subsidies is still not clear (Busom, Corchuelo and Martínez, 2014). Scholars did find that the probability of receiving a subsidy and the effectiveness is influenced by a number of firm-level characteristics, for example firm size, past experience in R&D, and number of R&D employees (e.g. Almus and Czarnitzki, 2003; Busom, 2000;
Czarnitzki and Licht, 2006; Wallsten, 2000). However, there is still limited knowledge about
firm heterogeneity and the characteristics that influences the probability of receiving subsidies
(David et al., 2000).
The innovation literature has become more aware of the fact that R&D activities generate important ‘spillovers’ of knowledge, and that the external assimilated knowledge has a substantial influence on the internal innovation processes and development. Scholars have theoretically substantiated and empirically found that knowledge spillovers are geographically close to their source (e.g. Ansel in, Virga and Aces, 1997; Audretsch, 1998). Therefore, it is assumable that knowledge spillovers are important in shaping the regional characteristics of innovation activities and has that this has to be recognized by policy makers. Knowledge spillovers stimulate localised learning processes and lead to economic benefits of firms that are geographically close to the source (Asheim and Isaksen, 2002; Asheim and Gertler, 2005). The innovation activity therefore tends to geographically cluster, creating high-tech industries in urban regions and regions that fall behind (Asheim and Gertler, 2005). This research aims to explain variations in R&D subsidy application behaviour of firms, by looking at the geographical clustering of innovation activities which creates high-tech urban regions.
Contextual setting
The promotion of innovative R&D activities in the Netherlands is arranged in the legal Act WBSO, which supports research and development of firms (Rijksdienst voor Ondernemend Nederland, 2017a). The Dutch government compensates with this Act a part of the costs of firms for their R&D activities. The WBSO was first introduced in 1994 and has had an enormous transformation due to its growth in budget and numbers of projects (Rijksdienst voor Ondernemend Nederland, 2017b). In the Netherlands the stimulation of R&D by the government happens through direct government expenditures or by fiscal discounts (Rijksdienst voor Ondernemend Nederland, 2017a). Fiscal discounting is aimed at stimulating private R&D expenditures by discounts for taxes on personnel-costs of R&D activities and a deduction of non-personnel costs for R&D activities and is arranged in the legal Act WBSO (Hall & van Reenen, 2000; Rijksdienst voor Ondernemend Nederland, 2017a).
For this research a data set from an intermediary is used, the applications of this firm account
for roughly ten percent of all application and should therefore be a correct representation of the
actual situation. The discussion below on the application behaviour of industries and regions
can be viewed from a firm perspective, but also form a government perspective. Governments
makes policies which fit their agenda and will probably try to stimulate certain areas and
industries to stimulate economic growth. Therefore, both perspectives are taken into account in
developing the hypotheses and further on in this research.
Hypotheses development Technological intensity
Industries face different technological opportunities and the expected growth in demand varies considerably, therefore it is expected that the innovation production functions are rather different (Blanes and Busom, 2004; Silva et al., 2017). The technological opportunity, which is defined in the innovation literature as the advancement in scientific and technological understanding about the own or another industry (Klevorick, Levin, Nelson and Winter, 1995), is likely to be the same among firms within the same industry. However, many scholars argue that it most probably different across industries (e.g. Cohen, 1995; Klevorick et al., 1995). In the line of these findings, Lim (2003) provides support for differences in two industries. He argues that the innovation production functions and processes in the semiconductor and pharmaceutical industries are different, the latter is more related to basic research. These differences result in deviant set-up costs of R&D across the industries and diverge parameters for economies of scope and scale.
Due to these differences between industries it is likely that firms in low-tech industries, who are thus expected to have a lower R&D intensity on average than firms in high-tech sectors, have less difficulty increasing their total R&D spending when utilizing a tax incentive scheme.
This is in line with several scholars, who argue that firms within different industries and branches follow completely different innovation strategies (Castellacci, 2008; Malerba, 2005).
According to them the development of new and breakthrough technologies is not a dominant strategy in low-tech industries. Firms in those industries rely instead on more incremental innovations, such as investment in process-development or development of new production technologies. Firms in high-tech industries are expected to achieve a lower additionality effect, as they already have a high R&D intensity which will decrease the effect of utilizing a tax incentive scheme (Castellaci and Lie, 2015). However, firms in high-tech industries receive higher amounts of subsidy (e.g. Busom, 2000; Silva et al., 2017). This could be due to their technological knowledge and expertise regarding R&D activities, which clearly influences their application behaviour for the R&D tax incentive schemes (Aerts and Schmidt, 2008; Almus and Czarnitzki, 2003).a
To continue, the technological intensity of a given industry is thus a reflection of the R&D intensity within that industry. There has been found that the R&D intensity of an industry has a positive correlation with the probability of project approval in tax incentive schemes (e.g.
Almus and Czarnitzki, 2003; Busom, 2000; Silva et al., 2017; Vence, Gutín and Rodil, 2000).
Next to this, there is empirical evidence that the experience in R&D and the regularity of the R&D activities, thus the technological capability, increase the probability of receiving a subsidy and the amount of subsidy, both at the firm and industry level (e.g.; Aschhoff, Fier and Löhlein, 2006; Herrera and Nieto, 2008). Silva et al. (2017) found evidence in line of these findings, but also found that the food and beverages and the textiles and clothing industries, which are classified as low-tech industries, where among the industries with the highest probability of approval. They argue that this could be because these industries also perform high- technological advanced R&D projects, but the main activities of the firms are still low- technological. Over all, these scholars claim that high-tech industries have a higher probability of receiving subsidies and the total amount of the subsidies is higher, and are more present in tax incentive schemes, while the additionality effect for those industries is lower (e.g. Huergo and Moreno, 2017).
These arguments lead to the following hypothesis:
H1: Firms in high-technological industries apply on average for higher amounts of total subsidies.
Region
The development of technologies and related R&D activities does often not happen in isolation according to the innovation literature, but several external sources attribute to this process:
universities; public research institutions; and other firms in the same or other industries (e.g.
Audretsch and Feldman, 1996). The innovative capabilities of firms are influenced by the spatial proximity to external sources, therefore it is important to take into consideration the regional aspects and search for differences between geographic areas (Lee, 2011; Silva et al., 2017). A high population density, such as in cities, has been proven to encourage knowledge spillovers as it creates a good environment to contact between persons and firms (Carlino, 2001).
As is common in innovation studies that take into account the regional conditions, this paper
has adopted the simple regional urban-rural scheme. Urban regions benefit on average from a
higher development, more innovation resources and innovative solutions. Rural regions on the
other side have a lower level of development, due to the meagre innovation culture which is
characterized by the firms’ inability to recognize the importance to innovate (e.g. Buesa, Heijs,
Pellitero and Baumert, 2006). This research does however not focus on the discussion which
regions should be supported by the R&D incentive schemes, as already many scholars have
contributed to this discussion. But it can be very valuable for policy makers to know the outcomes of this paper before designing new policies.
Previous research has shown that the innovation activity and the R&D intensity is lower in rural regions than in urban regions (Audretsch and Feldman, 1996; Fritsch, 2000; Tödtling and Trippl, 2005). They conclude that when the region is being considered, the innovation activity of firms is different in several ways; which leads to the suggestion that the R&D policy makers should consider the regional differences. Several other studies did also investigate the distribution of R&D subsidies between urban and rural regions, and concluded that the urban regions show a higher tendency to receive R&D subsidies and that these subsidies are higher (Czarnitzki and Fier, 2002; González, Jaumandreu and Pazó, 2005). For the reasons argued above, it is interesting to investigate if urban regions apply more often and for higher amounts of R&D subsides. Policy makers can learn from this, as urban regions are thus already higher developed, and rural regions could benefit from tax incentive scheme to stimulate their economic development and catch up on innovation levels with urban regions.
These arguments lead to the following hypothesis:
H2: Firms in urban regions apply on average for higher amounts of total subsidies.
Regional innovation paradox
The regional innovation paradox describes the contradiction between the greater need of
lagging, rural, regions for R&D subsidies and their incapacity to apply for those funds, and the
attribution of those subsidies to more technologically advanced, urban regions (Oughton,
Landabaso and Morgan, 2002). According to many scholars, innovation is a systemic process
that is for a large part shaped by the properties of the location and the region where the processes
take place (Cooke, 2001; Gertler, Wolfe and Garkut, 2000; Tödtling and Trippl, 2005). In the
light of this, Asheim and Gertler (2005) reviewed the literature regarding the clustering of
innovative activities in technologically advanced urban regions. They have come up with three
geographically bounded properties that have influence on innovative activities. The first is the
tacitness of the knowledge base, which means that innovation and learning requires sources to
be geographically close to each other to transfer knowledge. The second is that public sources
of technological opportunities need to be present to provide a strong incentive for, especially
high-tech, firms to locate in a particular region. The third is that urban regions are better in
attracting resources (e.g. highly educated employees) necessary to achieve economic success
and thus stay technological advanced.
Scholars in the innovation literature argue that high-technological industries have a stronger knowledge spill over effect throughout the economy than low-technological industries (Castellacci, 2008). These industries play an important role in the economic system. As already argued, spatial proximity to the external knowledge source is thus important for the knowledge spillover effect. Scholars and policy makers should then observe the geographic concentration, especially in the high-technological industries where knowledge spillovers play a more important role (Audretsch and Feldman, 1996; Castellacci, 2008). There has already given a lot innovation literature attention to location. Technologically advanced regions can be seen as a strategic source because it has knowledge spillovers, public sources of technological opportunities and the its attractiveness for highly qualified people, which impacts the innovative capabilities and helps establishing a competitive advantage (Asheim and Gertler, 2005; Cooke, 2001; Gertler et al., 2000; Tödtling and Trippl, 2005).
Herrera and Nieto (2008) found that the urban regions and the high-technological industries in Spain apply for more subsidies, and that the effect of the subsidies is higher in the urban regions than in the rural regions. They conclude with a recommendation to include the geographical location of firms in further evaluation studies of R&D incentive schemes. As argued above, knowledge spillovers are more important in high-technological industries and the spatial proximity to the source is crucial to transfer the knowledge, thus technologically advanced firms tend to cluster in major cities (Asheim and Gertler. 2005; Audretsch and Feldman, 1996;
Castellacci, 2008). Public agencies tend to sustain the performance of these technological advanced, successful, urban regions which is commonly referred to as picking the winners (Heijs and Herrera, 2004; Oughton et al., 2002). However, the drawback of these R&D policies is that the successful regions are favoured and the rural regions will not be able to catch up, which will increase the differences within countries. Given these arguments and findings, it is interesting to test the following hypotheses:
H3: Firms in urban regions who are high-technological intensive apply on average for higher
amounts of total subsidies.
Control variables
In our models we use several control variables. The control variables included are described and discussed below.
Size
Firms’ participation in R&D incentive schemes has been studied for several years, and several variables have been proven to significantly influence the probability of a firm to apply for R&D subsidies. One of the most important determinants that influence the probability is firm size, the literature states that usually larger firms have a higher probability of receiving a subsidy (e.g. Czarnitzki and Fier, 2002; Almus and Czarnitzki, 2003; Herrera and Heijs, 2004; Silva et al., 2017). There are however some scholars who found that smaller firms have a higher chance of receiving subsidies (Busom, 2000; Fier et al., 2006). These scholars argue that this might be influenced by the public agency’s goals, because in some countries smaller firms are the main target to receive R&D subsidies.
Age
The age of firms is another variable that has been found a significant determinant of subsidy application behaviour. Empirical studies have found that the age of the firms positively influences the application behaviour, thus older firms have a higher probability of applying (Busom, 2000; Czarnitzki and Hussinger, 2004; Silva et al., 2017). However, some scholars have found that the age of firms does not influence the probability of applying (Herrera and Heijs, 2004; Almus and Czarnitzki, 2003).
Type of project
Another control variable will be added to the analysis and that is the type of project involved.
The WBSO distinguishes between three different types of R&D projects wherefore firms can apply: 1) product 2) process and 3) programming. Scholars have found that cooperation projects have a higher probability of receiving subsidies, which could be due to the public agencies goals to generate knowledge flows between firms and that stimulates knowledge spillovers (Feldman and Kelley, 2006; Dumont, 2013).
Sector
Scholars in the innovation literature often argue that manufacturing firms have a higher R&D intensity, apply more often to R&D incentive schemes and receive higher amounts (e.g.
Castellaci and Lie, 2015; Sliva et al., 2017). There has been found that the tax incentive schemes
in Portugal are aimed mainly at R&D projects by manufacturing firms, in both high- and low- technology industries (Silva et al., 2017). As argued in the development of hypotheses 1, high- technological industries apply on average for higher amounts of total subsidies. However, Silva et al. (2017) found that the low-tech industries: food and beverages and the textiles and clothing industries, also belong to the group with the highest probability of approval. For this paper it could thus be interesting to control whether manufacturing industries indeed do apply for higher amounts of subsidies, despite belonging to the low-tech classified industries.
Next to this, Castellaci and Lie (2015) argue that in the literature service industries have a lower propensity to apply to a R&D incentive scheme. For service industries, R&D is not a dominant strategy to develop new technologies, just as in low-tech industries they rely for a large part on incremental innovations, such as investment in process-development or development of new production technologies. They however found the opposite; service industries have a higher propensity to apply for R&D incentive schemes. It is thus not been tested well in the literature if either the manufacturing industries apply for higher amounts of subsidies, or the service industries. Therefore, also this variable is identified as an interesting control variable.
Figure 1. Conceptual Model
Methodology Data set description
The aim of this paper is to analyse if the technological intensity and the region affect the average amount of subsidy applied for in the Dutch WBSO act. In order to do this, a Dutch intermediary has provided a data set. This includes information on all the project applications, regardless of they have been rejected or approved, from 2975 firms for the period 2012 till 2018. Not all applications from 2018 are processed in the data set, which is why this year is dropped for the final analysis. Before performing the analyses, the data set is enriched by the intermediary with additional data from the Netherlands Enterprise Agency (RVO), the Dutch Chamber of Commerce and another internal system of the intermediary. Finally, the data set is cleaned by two student researchers from internal mistakes and incomplete applications, which is done in cooperation with the data expert of the intermediary.
A large number of applications missed some values. Data imputation is used to solve this problem and prevent listwise deletion, this method involves replacing the missing values by imputed values. At first is the number of employees imputed, based on the firm age, number of R&D employees and the turnover. Then the turnover is estimated based on the firm age, number of R&D employees and number of employees. The data is cross sectional and longitudinal, which is why panel data analysis performed. The panel data analysis is done in the statistical software program STATA SE. The data set contains firms who in some cases have applied for subsidies multiple times per year, therefore the data is collapsed to get the summary statistics per year by individual firm. The new data set now consists of the means of firm age and size, and the total of project type, wage-hours and cost-and-expenditure per application year. The data set is unbalanced, which means that not for every year all firms have applied for subsidy.
In table 1 the distribution over the years can be seen. A total of 623 firms (21.93%) have applied
for subsidies all the consecutive years from 2012 till 2017. Next to this, have 811 firms
(28.58%) applied for subsidy for several years, but with a pause between them. The rest of the
firms (71.42%) have applied for one or more years, but without a pause of one or more years.
Table 1. Distribution of applications
Distribution of T_i: min 5% 25% 50% 75% 95% max
1 1 1 3 5 6 6
Descriptive statistics
The applications that are not submitted, incomplete or have incorrect values are deleted out of the data set, because those are not useful for analysing the empirical framework of this paper.
This drops the total of applications from approximately 22000 applications to 17000 applications of 2975 firms. The collapsed data set which will be analysed has 8843 observations from 2841 firms and the total amount of subsidy applied for is 4.27 billion euro.
Of all the applications for the WBSO in the Netherlands in 2017, 81.5 percent were granted.
But that the subsidies are granted does not necessarily mean that they are used, as firms not always use the granted subsidies (WBSO, 2018). This can be the case for example when they do not have enough R&D employees to use the full amount of granted subsidy or decide to not carry out the project. The intermediary accounts for approximately 10 percent of the applications for the WBSO per year and has nine different offices fairly distributed across the country.
The classification of technological intensity of industries is according to the taxonomy of the OECD, as well as the division of industries belonging to either the manufacturing industry or service sector (Galindo-Rueda and Verger, 2016). Due to several historical changes in the definition of urban areas in the Netherlands, urban is defined in this publication as
Freq. Percent Cum. Pattern
623 21.93 21.93 111111
283 9.96 31.89 . . . 1
232 8.17 40.06 . . . . 11
230 8.10 48.15 1. . .
182 6.41 54.56 . . . 111
149 5.24 59.80 11 . . . .
119 4.19 63.99 111 . . .
119 4.15 68.15 1111 . .
93 3.27 71.42 . . 1111
811 28.58 100 (other patterns)
2841 100 XXXXX
municipalities with 75.000 inhabitants or more. This definition is based on the publication of Dijkstra and Poelman (2014).
In table 2 the distribution of subsidy application is shown. As the individual applications are collapsed by year, the average subsidy per application is the sum of the applications of a given firm in 1 year. The table also states the total number of applications and the total amount of subsidy applied for. The high-tech firms account for 33,76% of all applications and they applied for 2.14 billion euro R&D subsidies in the period of 2012-2017. Low-tech firms have applied for 66,24% of all applications, worth of 2.13 billion euro. The technological intensity and the region of the firms seems to influence the height of the average subsidy applied for strongly.
Firms in rural areas who are low-tech apply on average per application 236.920,67 euro (45,07%) less than high-tech firms in the same area. The same counts for firms in urban areas, where low-tech firms apply on average per application 448.565,16 euro (48,14%) less than high-tech firms in the same area.
Table 2. Descriptives subsidy distribution
Area No. of applications Total subsidy
(bln) € Avg. subsidy per application €
Low-tech & rural 3602 1.04 288.728,48
Low-tech & urban 2256 1.09 483.156,03
High-tech & rural 1579 0.83 525.649,15
High-tech & urban 1406 1.31 931.721,19
Total 8843 4.27
According to the evidence in the literature, there is expected a positive relationship between the two independent variables, the technological intensity (being high-tech) and the region (being located in an urban region), with the dependent variable the total subsidy applied for. There is also expected that the interaction of the independent variables, high-tech firms located in urban regions has a positive relationship on the total subsidy applied for. In table 3 it is described how the variables, including the control variables, are measured in the data set. For the firm age and firm size, the log + 1 is taken to assure that the data is normally distributed. This is an often used method in comparable studies (e.g. Wanzenböck et al., 2013). The total subsidy is the sum of the applied for hours multiplied by the wage and plus the amount of cost expenditure, by a firm in a year (WBSO, 2018). The log + 1 is also taken for this variable to make it normally distributed. The technological intensity is according to the taxonomy of the OECD, and is a binary variable in the data set as it has a 0 if the firm is low-tech and a 1 if the firm is high-tech.
The region is also a binary variable and is classified as a 0 for rural areas, with less than 75.000
inhabitants, and a 1 for urban areas with 75.000 and more inhabitants. The binary variable sector is also in line with the taxonomy of the OECD, if it has the value 0 the firm belongs to the service sector and if it has a 1 it belongs to the manufacturing sector. The WBSO distinguishes between three different kinds of R&D projects wherefore firms can apply: programming, process and product projects. The sum is taken per project in a given year per firm, to control for the project specific effects.
Table 3. Description of variables
Variable Description
Total subsidy Log + 1 of the sum of the hours*wage and the amount of cost expenditure applied for by a firm in a year Technological intensity 1 if the firm is high-tech
Region 1 if the firm is urban
Firm age Log + 1 of the nonlinear function of the firm in the year of application (in years)
Firm size Log + 1 is taken of the total number of employees
# of projects programming The number of applications in a year for programming projects
# of projects process The number of applications in a year for process projects
# of projects product
The number of applications in a year for product projects
Sector 1 if the firm belongs to the manufacturing industry
To have a better understanding about the variables in the data set, the descriptives of the variables that are analysed are listed in table 4. The table shows that the average amount of subsidy applied for of a firm is 482.614 per year. The mean of the technological intensity can be interpreted as 34% of the firms being high-tech. The same counts for the region and sector, 41% of the firms are located in urban regions and 29% belong manufacturing sector. In the data set, the firms that have applied for R&D subsidies have on average 177 employees and exist for 18 years.
Table 4. Descriptive values variables
Variable Obs. Mean Std. Dev. Min Max
Total Subsidy 8843 482.614 3.012.309 0 163.000.000
Technological intensity 8843 .34 .47 0 1 Region 8843 .41 .49 0 1 Firm size 8841 177 730 0 26161
Firm age 8761 18 17.5 0 187
# of projects programming
# of projects process
# of projects product Sector
8843 8843 8843 8802
6.1 2.5 1.5 .29
11.5 8.6 11.5 .45
0 0 0 0
74 106 229 1
Model specification
Due to the cross sectional and longitudinal nature of the data, panel data analysis is performed in this paper to test the empirical model. The variables technological intensity, region and sector are time-invariant in the data set, which rules out the use of fixed effects models because in the fixed effects model these variables are absorbed by the intercept. This paper will therefore use a random effects model (equation 1). There is not chosen to include the between effects model as this averages out the time component and thus completely discards the time variation, while the random effects also uses the time variation in the data set (Cameron and Trivedi, 2009;
Woolridge, 2010). The Breusch-Pagan Lagrange multiplier test showed that the random effects test is more appropriate than performing an OLS regression (appendix 1), therefore this paper will perform random effects analyses.
Equation 1. Y it = α + βX it + u it ; i = 1, 2, …, N; t = 1, 2, …, T.
u it = μ i + v it
A random effects model does not only predict the change over time, but does also explain the
differences between the units of analysis. The coefficients thus simultaneously explain the
change over time and the cross-sectional differences between the units of analysis. It assumes
that a one unit increase in X is associated with a certain change in Y. Next to this, many authors
agree that the random effects model, compared to a fixed effects model, better allows to
generalize the findings outside of the used sample (e.g. Cameron and Trivedi, 2009; Woolridge,
2010). In the analysis will be tested for serial correlation and heteroskedasticity, and after the
random effects regression the variance inflation factors will be tested. If these tests will be
negative, the variables are tested in further analysis and excluded if necessary.
Results
The variables included in the empirical framework are tested in a Pearson’s correlation matrix (table 5) to assess the relationships between the variables. The variables are all small to medium correlated and statistically significant, except the correlation of number of projects process with number of projects programming (-.007, p > .05) and the correlation of firm size with region (.003 , p > .05) are not statistically significant. The independent variable technological intensity has a small positive correlation with the total subsidy, r = .195, p < .001, the technological intensity explains 3,8% of the variation in total subsidy. There is also a small positive correlation between region and total subsidy, r = .082, p < .001, the region explains 0,7% of the variation in total subsidy.
Table 5. Pearson’s correlation matrix
Total Subsidy
(Ln)
Tech
intensity Region Firma size
(Ln) Firm age
(Ln)
# of projects program ming
# of projects process
# of projects product
Sector
Total Subsidy (Ln) 1
Technological intensity 0.195*** 1
Region 0.082*** 0.083*** 1
Firm size (Ln) 0.207*** -0.093*** 0.003 1
Firm age (Ln) 0.231*** 0.04*** -
0.159*** 0.172*** 1
# of projects programming 0.227*** 0.136*** 0.188*** 0.093*** -0.035*** 1
# of projects process 0.178*** -0.044*** 0.097*** 0.074*** 0.048*** -0.007 1
# of projects product 0.356*** -0.027* -
0.055*** 0.138*** 0.152*** -0.052*** -0.364*** 1
Sector 0.188*** 0.404*** -
0.171*** 0.139*** 0.324*** -0.213*** 0.059*** 0.135*** 1
* p < 0.05, ** p < 0.01, *** p < 0.001
