H
OW FIRM CHARACTERISTICS EXPLAIN THE AMOUNT
OF SUBSIDY APPLIED FOR
MS
C
.
T
HESIS
University of Groningen
Faculty of Economics and Business
by
Gerben Wiersema – S2300311
Strategic Innovation Management
Date
25-06-2018
Word Count: 8775
Abstract
This thesis researches the firm characteristics that explain the amount of subsidy firms apply for. The focus is on the Dutch tax incentive called the WBSO specifically. Based on the literature on subsidies and tax incentives relevant firm characteristics that explain subsidy application behavior are identified. A dataset collected by a Dutch consultancy firm specialized in the application of subsidies is analyzed. Using regression analysis, the amount of subsidy applied for is explained and the hypotheses are tested. We find support for the positive effects of the firm characteristics size and R&D intensity. For firm age no significant non-linear relationship is found. The models explain up to 44.6 percent of the between variance and 8 percent of the within variance. This thesis thereby contributes to the current literature on subsidies and knowledge about the WBSO. Moreover, the results provide the consultancy firm with insights on; the characteristics that explain subsidy applications, for what applications consultants under- or over-apply, and interindustry comparisons.
Introduction
Research and development (R&D) spending is seen by economists as a main factor for sustainable growth in industrialized countries (Czarnitzki & Hussinger 2004). Government policies contribute to stimulating firm innovativeness and are therefore of interest from a scientific and practical perspective (Wanzenböck, Scherngell, & Fischer, 2013).
A first question in evaluating policies is how subsidies are allocated among firms and projects (Blanes and Busom, 2004). A similar argument is brought forward by Wanzenböck et al. (2013, p. 66) who state: “it is of central interest to gain a deeper understanding on the interrelations
between effects of public funding and the particular responsiveness of firms to these funds, especially with regard to distinct firm-specific characteristics.” In their study, Wanzenböck et
al. (2013) find different outcomes for different characteristics.
For policy makers it is therefore interesting to know what type of firms apply for subsidies and for what amount they apply. Understanding firm specific characteristics could help to stimulate firm behaviour in the heterogeneous innovation process (Wanzenböck et al., 2013). Moreover, Lee (2011) argues that the direction and magnitude of the subsidy will vary for different firm characteristics. Understanding this provides insights in how policies are in line with goals set out by the policy maker.
It is acknowledged that there is limited knowledge about the heterogeneity of firms applying for- and receiving subsidies and there have been limited attempts to solve this (Lee, 2011). Blanes and Busom (2004) state that there is “very little” research on the allocation process. In addition, Clausen (2009) argues that research on subsidies has been limited by the lack of understanding regarding the amount of subsidy firms get access to. Based on the problem of firm heterogeneity, the aim of this paper is to increase the knowledge in this field. This thesis addresses part of this problem by focusing on application behavior. This is researched on the firm level and therefore the research question of this thesis is:
RQ: How do firm characteristics explain the amount of subsidy applied for?
Currently there is limited knowledge about what characteristics explain these differences. This study can provide policy makers with insights on which policies can be based, as subsidies have different effects for different types of firms (e.g. Busom, Corchuelo and Martínez-Ros (2017). In this thesis the database of a consultancy firm is used. This consultancy firm is specialized in the application of subsidies and has over 30 years of experience in the field. This thesis aims to provide relevant insights for practitioners in the subsidy field. Subsidy consultants can learn from this research what firm characteristics predict subsidy application and firm behavior. From the findings of this thesis a model can be derived that explains the amount of subsidies applied for. This model could in particular help the consultancy firm that provides the data. The results could increase their application success rate, improve evaluations and decision making or contribute to identifying new clients or market segments.
This thesis aims to identify the relevant characteristics of firms applying for subsidies in the existing literature. The identified characteristics are tested and introduced into a model to increase the understanding of how characteristics explain the allocation of subsidies. This is tested based on the available panel data of the consultancy firm from 2012 to 2017. Support is found for the hypothesis that Size is related positively and significantly to the total subsidy. Furthermore, no significant non-linear effect for firm age on the total subsidy is found. R&D
intensity relates positively and significantly to the total subsidy. Finally, the results provide
relevant insights for practitioners in the subsidy consulting field by comparing predicted applications to actual applications.
Literature Review
This thesis will continue with a literature review that is organized as follows. First, the rationale for subsidies and tax incentives will be discussed from a theoretical perspective. The literature review further elaborates on the current gaps in the subsidy literature. Second, the institutional setting will be explained along with the firm characteristics that influence the amount of subsidy applied for. Third, the conceptual model and hypotheses are introduced.
State of the art
Public support is found to significantly contribute to firm spending on R&D in several studies (e.g. Almus and Czarnitzki, 2003; Clausen, 2009; Falk, 2006). However, there remains debate on the effectiveness. Some academics find that the relationship between public support and private R&D investment is supplementary whereas others conclude that the relationship is complementary (David, Hall and Toole, 2000). Supplementary means that firms use the subsidy instead of private funds, whereas complementary implies that firms invest additional funds. Academics, more recently, started studying the effects of R&D support by researching the
additionality of government incentives (e.g. González and Pazó, 2008; Lee, 2011). Behavioural additionality measures the extent to which a received subsidy is invested in additional R&D
activities, rather than in R&D activities that would have taken place anyway (Wanzenböck, Scherngell, & Fischer, 2013).
According to Peneder (2008), governments have two options for increasing firm incentives to invest: fiscal incentives and the direct funding of targeted expenditures (i.e. subsidies). These are found to be substitutes by; Guellec and Van Pottelsberghe De La Potterie (2003) and
Peneder (2008). Guellec and Van Pottelsberghe De La Potterie (2003), in a meta study on OECD countries, find that in general both subsidies and tax incentives have positive effects on privately financed projects.
In comparison to subsidies, tax incentives are seen as more neutral and non-interfering (Busom, Corchuelo and Martínez-Ros, 2014). With direct support, or subsidies, governments can target specific projects. With tax incentives the firms generally possess more freedom to decide for themselves for what projects to apply (Peneder, 2008).
Institutional setting
In the Netherlands, innovation is promoted by the Wage Tax and Social Insurance Act. The WBSO1 started in 1994 and developed into the most important subsidy for the promotion of R&D in the Netherlands (Poot, Hertog, Grosfeld, and Brouwer, 2003). Dutch firms received a total of 6.7 billion euro from the WBSO subsidy in 2016 (Ministerie van Economische Zaken, 2017). Both in terms of scope and budget the WBSO thereby contributes significantly to R&D spending. From the firms that received a subsidy, 97% was SME (Ministerie van Economische Zaken, 2017). Lokshin and Mohnen (2012), found empirically that the WBSO is effective for stimulating R&D spending, but that partial crowding out effects exist for larger firms.
The WBSO is considered as a relatively generous, accessible, and easy to apply for subsidy
(Keijzers & Bos-Brouwers 2005). According to a report about the WBSO and RDA (Research
& Development Aftrek wet), published by the Netherlands Enterprise Agency, 4.9% of the
Dutch firms made use of the WBSO subsidy in 2016 (RVO 2016). Nonetheless, not all applications are successful. In the period 2006-2010, 28.110 out of 145.230 (19.35%) requests were retreated (Verhoeven, Van Stel & Timmermans, 2012).
The importance of studying firm determinants
As discussed in the introduction, the problem of heterogeneity of firms is acknowledged in the subsidy literature (e.g. Heijs & Herrera, 2004). Firm heterogeneity leads to biases in econometric papers (David et al., 2000; Lichtenberg, 1984). The importance of understanding firm characteristics as a first step is remarked by David et al., (2000) and Silva, Silva and Carneiro (2017). According to Lee (2011), there are limited theoretical models or frameworks, preventing the theory to advance. More knowledge about the relevant firm characteristics can help to overcome the problems found in econometric papers. As is stated by Clausen (2009);
“it is far from simple to predict why some firms get access to a higher subsidy amount compared to other firms.” Hence, it is important to research the firm determinants in more depth to
overcome this gap in the literature.
One of the ongoing discussions is to research the effects of heterogeneity of firms applying, and the selection bias of policy makers (David et al., 2000). Some prior research to the amount of subsidy received has been conducted on the firm level. For example, Wallsten (2000) and Lach (2002) look at this as part of their studies to explain effects of subsidies. They take into account a few variables, such as R&D spending, age and size. However, only limited variance could be explained by these variables.
Knowledge on firm heterogeneity can improve the relevance of future studies. David et al. (2000) found that research on subsidies is dispersed in terms of the level of analysis, and that results therefore can differ. Aerts and Czarnitzki (2004), state that the findings of macro-level studies often cannot be confirmed on the Micro-level (i.e. the firm level).
Scholars can use matching methods or other statistical methods to account for the fact that subsidies are not distributed randomly (e.g. Almus and Czarnitzki, 2003). With this method, researchers compare the results of subsidized firms to a group of firms with similar characteristics without the subsidy. The accuracy of these models can be increased when knowledge on what variables to include is improved.
At last, knowledge on firm determinants is relevant for policy makers. Knowledge on firm characteristics and firm behavior can provide policy makers with the tools to adapt their policies accordingly and thereby maximize the effect of the public support.
Conceptual Part
The conceptual model is constituted from firm characteristics that are identified in the literature and are found to be relevant in explaining subsidy variance across firms. The conceptual model is reported at the end of this section (figure 1).
As the application behaviour of firms is heterogeneous, it is meaningful to consider the firm perspective for understanding the variance in total subsidies applied for (Busom et al., 2017). The firm-perspective is seen as the most relevant perspective.
In addition, the government policies will likely affect firm application behaviour, and therefore this perspective is also important to study. The government perspective is therefore seen as complementary to the firm-level perspective. For the hypothesis development the firm perspective is first discussed for every variable, followed with a small discussion on the government perspective. According to Aerts and Czarnitzki (2004), earlier studies are criticized to not take into account the selection bias of governments.
Size
An additional reason that large firms can apply for a higher amount, is that these firms have formal R&D centres, laboratories or R&D managers. These firms are said to have more slack resources. Slack resources allow a firm to undertake R&D related projects (Majumdar, 2000). Previous literature has examined the relationship between size and subsidies. Almus and Czarnitzki (2003) have found that size is positively related to the probability of receiving a subsidy for Eastern German firms. They have found that a 10% increase in size leads to a 2.2% increase in the probability of receiving a subsidy. They ascribe this effect to information advantages, higher R&D capabilities and more capacity or staff members to apply for funds (Almus and Charnitzki, 2003). Similarly, Duguet (2004) has pointed out a positive effect for size for the French context, while Heijs and Herrera (2004) have found this for the Spanish case. Aerts and Czarnitzki (2004), have found that subsidized firms are larger, and that this was one of the most important predictive characteristics. Wallsten (2000) did find a significant correlation between receiving R&D subsidies and the number of employees for the SBIR subsidy in the US. Busom et al. (2017) have found a positive relationship between firms with over 200 employees and participation in tax incentives for Spanish manufacturing firms. For larger firms it is easier to generate funding internally (Clausen, 2009). Larger firms can benefit from spill-over effects within their organization and profit from economies of scale and scope (Clausen, 2009). As policy makers look to reach the highest levels of grant effectiveness it is likely that they favour projects from larger firms. This is referred to as pick-the-winner strategies (Clausen, 2009). From a government perspective it is therefore likely that size has a positive effect on the total subsidy applied for.
Moreover, it is likely that larger firms are more formally organized and thereby fulfil requirements set by policy makers. As public institutions are likely to look for clearly specified projects this could increase the chances for larger firms to receive subsidies.
To conclude, being of larger size is found to make it easier to apply for subsidies in previous studies. Several scholars have found correlations between size and firms participating in- or receiving subsidies. Therefore, it is offered:
H1: Firm size will be positively and significantly related to total subsidy Firm Age
the total amount of subsidy applied for. Prior research has found that older firms have a higher probability to receive subsidies (Almus and Czarnitzki, 2003). Clausen (2009), in an empirical study, did find that age is positively correlated to R&D spending.
Moreover, from a firm perspective, as age increases, the experience and capabilities of the management team are expected to increase (Heijs & Herrera, 2004). Management team capabilities and experience are expected to influence the firms approach to innovation, as well as the approach to subsidy applications.
Most findings indicate that the older the firm, the more likely the firm is to receive a subsidy (e.g. Busom, 2000; Czarnitzki and Hussinger, 2004). Czarnitzki and Hussinger (2004) state that it is likely that policy makers are looking for entrepreneurs that are innovative and successful, to maximize the positive effects of the subsidy.
However, not all academics agree that age is positively related to the amount of subsidy received. For example, Almus and Czarnitzki (2003) argue that old firms are often regarded as being reluctant to innovate, and that young firms therefore are expected to apply more for innovation subsidies. Segarra-Blasco and Teruel (2016), have looked at the role of age on the success of applications for R&D subsidies for Catalonian firms. The authors were unable to find a significant relationship to application propensity. However, they did find that younger firms have a higher propensity to receive a subsidy. Furthermore, Schneider and Veugelers (2010) argue that EU governments aim to support younger firms.
For the WBSO, it is expected that the policy makers are (more) tolerant towards starters. A starter receives forty percent reduction of the first €350.000 spend on R&D, whereas for non-starters this is 32 percent (WBSO manual 2017). A starter is defined as someone that has not received a WBSO for more than 3 years in the first five years of the company having employees (WBSO manual 2018). This indicates that the intention of the subsidy is to be beneficial for starters. It is therefore possible that selection criteria for starters are less strict. Verhoeven et al. (2012) did find that 28% percent of the users of the WBSO are starters. From the firms that used the WBSO in 2010 the average years of making use of the WBSO was around 4-5 years (Verhoeven et al., 2012).
al., 2013). From a government perspective it is therefore expected that the intention is to favour starter and stimulate R&D for these firms.
Based on this it is expected that a non-linear relationship exists for firm age. The literature suggest that very young firms and starters apply for more subsidies than moderately old firms. Thereafter, the relationship is expected to be positive. It is therefore offered that a U-shape relationship exists for age and the total subsidy. Therefore, it is offered:
H2: Age has a U-Shaped relationship with the total subsidy R&D intensity
R&D intensity is often related to size. This stems from the Schumpeterian theory that suggested that larger firms have disproportional higher levels of R&D (Clausen, 2009; Levin, Cohen & Mowery, 1985). R&D intensity is however not the same, as R&D intensity can differ per firm and industry and the measurement is different. The number of R&D employees is seen as a proxy for R&D intensity in this study.
R&D intensity is likely to increase the level of received subsidies as this is found to effectively support R&D spending (Wallsten, 2000; Clarysse, Wrigth & Mustar; 2009). Wallsten (2000) did find that the level of R&D spending of a firm is a relevant predictor for firms receiving subsidies, while more subsidies did not lead to significant increases in the number of employees. From a firm perspective it is likely that higher levels of R&D intensity lead to a higher probability that a firm applies for a subsidy. Reason for this is that the incentive to apply becomes bigger as more R&D-employees are eligible for a tax reduction. Moreover, an increased number of R&D employees makes it more likely that staff in the organization is aware of the possibility to apply. At last, the more the organization is oriented towards research, the more likely that distinct research projects are defined.
favor high R&D spending firms over other firms, as they look for the highest impact of a subsidy. Therefore, it is offered:
H3: R&D intensity will be positively and significantly related to the total subsidy. Control variables
In the models presented several control variables are included. The control variables included are described and discussed in the following section.
Fixed Sum
The Fixed Sum2, option was introduced in 2012 and is part of the Cost-expenditure option
(WBSO Manual, 2018). Fixed Sum indicates that the organization can apply for 1800 labor hours at a rate of €10 per hour, and additional hours at €4 per hour. For the consulting firm it is relevant to see if this option leads to different outcomes compared to the cost/expenditure option. With the cost/expenditure option the cost and expenditure are precisely calculated (e.g. not fixed).
Industry
In some industries it is likely more common to apply for a subsidy than in other industries. For example; a knowledge intensive industry is more likely to have high levels of R&D. Some industries have (more) work that qualifies for receiving the subsidy. The ICT industry is for example more software intensive compared to more traditional industries. Busom et al. (2014) have found a positive relationship between high-tech industries and the participation of firms in tax incentives and firms that participate in both tax incentives and subsidies.
Over the years the WBSO has become more favorable to software projects. In the ICT and Architect/Engineer sectors the reach and awareness of the WBSO increased significantly after the WBSO definition became broader (Verhoeven et al., 2012). Moreover, the WBSO is likely to be allocated in certain industries more than in others. It is probable that firms in an industry that is shrinking would receive less subsidies as policy makers are expected to look for projects that benefit society (Heijs & Herrera, 2004). On the other hand, it could be that the government looks to subsidize those industries to keep them competitive (Summers, 1988).
Type of project
The WBSO identifies different types of projects and can be divided in three subcategories: 1) projects product 2) projects process and 3) projects programming. What kind of a project a firm applies for could have impact on the chances of getting a subsidy granted. Moreover, the
requirements differ per type of project (WBSO Manual, 2018). Poot et al. (2003) include different projects in their method to account for the specific characteristics of the WBSO. Therefore, the number of projects per category are included in the model.
Figure 1. Conceptual Model – Total Subsidy
Methodology
MeasurementThe different variables are identified based on the literature review and the possibilities the data provide. As the dataset is from an external party the different variables in the next section are thoroughly described in the next section. A short overview is provided the appendix (appendix 1), including references to literature that support the use of the variables.
Independent variables
For Size the number of employees is used as a proxy variable. In additional analysis we look at the turnover as an alternative proxy of Size. The number of employees has “1” values for 3235 observations, which are missing data points. This means, in most cases, that the consultant did not know the number or did not fill it in. Data amputation is used for the firms that have a “1”. This is further described later in the methodology sector, under the dataset section. The natural log + 1 is taken to make the data less skewed. This approach is common in this line of research (e.g. Wanzenböck et al., 2013).
For R&D intensity the total number of R&D employees is included as a proxy variable. This is the average number of R&D employees per submission year. For this the log + 1 of R&D employees is taken. Unfortunately, the data did not allow to measure this by the R&D expenditure divided by the total sales, as is common practice (Wanzenböck et al., 2013). Dependent variables
The dependent variable is the total subsidy applied for. The applied for subsidy consists of Wage Hours and Cost Expenditure, as this is how the WBSO is subdivided. From this the log function plus one is taken to make the data normally distributed. This dependent variable is unique for this type of research, as most authors look at firm participation (e.g. Busom et al., 2017) or the propensity to receive a subsidy (e.g. Segarra-Blasco and Teruel, 2016).
Total Subsidy = LN((Wage Hours + Cost Expenditure) + 1)
In additional analysis the subsidy per employee is used as an alternative dependent variable. The advantage of this is that it allows for better comparison between firms. Therefore, the total subsidy is divided by the number of employees. Again, the log + 1 is taken.
Control variables
Several control variables are added to the model. In the analysis three variables are included to control for the number of projects. This is subdivided in three different types of projects;
product, process and programming. These are the types of subsidy that consultants can apply
for. Product relates to physical innovations and process relates to process innovations. When the project concerns the development of software it is qualified as programming.
Next to this a variable is added to control whether the application is applied for as a fixed sum. This is to control for differences in the total subsidy that results from applying for a fixed sum or calculated cost-expenditure. This is a binary variable that is 1 if the subsidy is applied for as fixed sum.
Empirical strategy
For panel data analysis the unobserved effects model can be used (Equation 1). In this the y and the x are observable random variables and c denotes the unobserved random variable, that is assumed to be constant over time. The uit represents the idiosyncratic errors and is assumed to be uncorrelated with the x. T is the time variable. In this model ci can be treated as a fixed effect or a random effect. It is argued that random effects model is appropriate in most cases as the main goal is to find out whether ci is correlated with the independent variables xit, t = 1, 2, ...,
T (Wooldridge, 2010). The random effects model is commonly referred to as the firm specific
effects model and this is the approach taken in this thesis. In later analysis, the Hausmann test is employed to test whether fixed effects or random effects are appropriate.
Equation 1:
Linear regressions analysis is used to test the hypotheses. The vce, cluster option is used to account for serial correlation and heteroskedasticity. This option allows for intergroup correlation and robust variance estimators (Stata Guide, 2017, p. 2892). Vce, cluster relaxes the assumption of independence of observations (Stata Guide, 2017, p. 323).
In further analysis the data will be tested for variance inflation factors by doing a VIF test after running regression analyses. In case variables have a high VIF (i.e. >10 indicates multicollinearity), subsequent analysis is conducted to test the effect of these variables in models in which those variables are excluded (Hair, Babin and Anderson, 2010).
Dataset
For this research a dataset from a Dutch consultancy firm is used. This dataset incorporates the data of WBSO applications from 2975 firms in the period 2012-2018. The data is collected from the output of software in which consultants of the consulting firm fill in applications for the WBSO. This dataset is enriched with data from the Dutch Chamber of Commerce, the Netherlands Enterprise Agency (RVO) and the CRM-system of the consulting firm.
entries without any data were deleted. This resulted in a sample of roughly 17000 applications from 2975 firms.
An additional measure taken to improve the dataset is imputation. For the number of employees and turnover the missing values are estimated. For the number of employees this is based on the firm age, R&D intensity and the turnover. For sole-ownership firms the estimated values are than replaced back to 1. For turnover the estimates are based on the firm age, R&D intensity and firm Size.
After this step the data is collapsed, meaning that a dataset is made of summary statistics per year. For this: Firm age, Size and R&D intensity are collapsed based on the average of the application year. The sum of the different types of projects, the Wage-hours and the
Cost-and-Expenditure is taken. The data set is collapsed by organization ID and Start Year.
At last, the applications from 2018 are dropped as not all applications are submitted and therefore this year is not relevant to include in the analysis. This results in a final dataset with 8841 observations from 2840 unique firms and a total of 3.8 billion euro in subsidies applied for. One observation covers the applications from one firm in one year. In the process, four applications years from one firm are dropped due to extremely high values. For both turnover and the number of employees a case of an extreme and unrealistic high value is replaced to a missing value.
The subsidy is applied for in two forms; as wage-hours (or wage-cost) and cost-expenditure (Figure 2)3. Wage hours are the tax-deductible hours per employee. Cost-expenditure means that specific investments in for example machinery or ICT-tools are tax-deductible. This can be applied for as a “fixed rate” or a calculated sum called “cost-expenditure” (WBSO Manual, 2018). The fixed rate means that the entrepreneur that applies receives a tax reduction of €10,- per R&D hour for the first 1800 hours he or she applies for, and €4,- per R&D hour for the remaining hours (WBSO Manual, 2018).
Figure 2: Distribution of WBSO-subsidy per year
In the following table the descriptive values and standard deviation per variable are depicted (Table 1). The turnover variable is only used in subsequent analysis.
Table 1: Descriptive values main variables
Variable Obs. Mean Std. Dev. Min Max
Firm Age 8759 18 17 0 187 Size 8838 144 730 0 26,161 R&D Intensity 8742 19 92 0 3,486 Turnover 8790 66,100,000 402,000,000 15 23,800,000,000 Total Subsidy 8841 432,625 1,726,765 0 51,500,000 # of projects product # of projects process # of projects programming 8841 8841 8841 6.1 2.5 1.5 11.5 8.6 3.8 0 0 0 229 106 74
In the data set firms from different industries are analysed. For the industries the SBI-codes are used (CBS). The 19 SBI main sectors are reduced to 12 different industries4. This approach is similar to Galindo-Rueda and Verger (2016). The: The production and distribution of and trade
in electricity, natural gas, steam and cooled air with Extraction and distribution of water and
waste and wastewater management and remediation categories are added together to create the Electricity, gas and water supply, waste management and remediation industry dummy. The Other Industry variable consists of all industries with less than 100 observations. These are: Mining and quarrying, Accommodation and food service activities, Real estate activities, Education and Culture and sports and recreation.
The applied for subsidies provide a rough estimate of the total allocated subsidy. In 2017 81.5 percent of the applied for subsidy was allocated (WBSO, 2018). The subsidies are distributed among all provinces in the Netherlands. The consultancy firm is responsible for roughly ten percent of the application in the Netherlands and has nine different offices across the country. This gives an indication of how the data set is generalizable across the Netherlands.
The distribution of applications and firms per industry is shown in Table 2. The table further includes the total subsidy in millions per industry, the total number of employees per industry and the average subsidy applied for per employee. It should be noted that the number of employees is the sum of all observations. Meaning that if firm X has thousand employees and applies in five years, the total number of employees reported is five thousand.
Table 2: Distribution per Industry
Industry Number of applications Number of firms Total subsidy in million € Total Number of Employees Avg. subsidy per employee in € Industry 1 2573 699 1710 389650 4339 Industry 2 1716 555 949 227507 4171 Industry 3 1394 481 271 167696 1616 Industry 4 1378 474 476 53231 8942 Industry 5 897 293 168 81171 2027 Industry 6 197 70 50.2 78996 636 Industry 7 186 34 50.3 39210 1282 Industry 8 114 46 38 28713 1323 Industry 9 100 39 29.7 46204 643 Industry 10 97 33 25.1 38835 646 Industry 11 79 29 14.6 2196 6649 Industry 12 73 30 16.1 102336 157 Total 8804 2783 3798 1.255.745 Avg. 2702
present. A “1” value for Pattern means that an observation exists for a firm in a certain year. For 623 (21.93%) firms all observation for all years are present in the dataset. This means that for 623 firms the consultancy firm consequently applied for a subsidy from 2012-2017. For 283 firms only one observation is present in 2017, for 232 firms only in 2016 and 2017, and so on. For 28.55 % of the observations it cannot be said with certainty that no gaps exist. A gap would imply that these firms for a certain year did not apply, applied independently or applied via a competitor of the consultancy firm.
Table 3. Distribution of data over the years
Distribution of T_i: min 5% 25% 50% 75% 95% max
1 1 1 3 5 6 6
Freq. Percent Cum. Pattern
Results
The correlation matrix of the variables used in the regression analysis is provided in Table 4. This indicates no problems of multicollinearity. Only the R&D Intensity variable has a relatively high correlation with the Total Subsidy. Values below .7 are not seen as problematic (Hair et al., 2010). Because the value is close to .7, VIF tests are conducted in later analysis.
Table 4. Correlation matrix
Total Subsidy (Ln) Firm Age (Ln) Size (employees) (Ln) R&D Employee s (Ln) Turnover (Ln) # of projects product # of projects process # of projects programmin g Total Subsidy (Ln) 1 Firm Age (Ln) 0.220*** 1 Size (Ln) 0.243*** 0.161*** 1 R&D Intensity (Ln) 0.675*** 0.346*** 0.296*** 1 Turnover (Ln) 0.329*** 0.158*** 0.404*** 0.469*** 1 # of projects product 0.327*** 0.148*** 0.128*** 0.317*** 0.117*** 1 # of projects process 0.165*** 0.0462*** 0.0675*** 0.173*** 0.0617*** 0.362*** 1 # of projects programming 0.208*** -0.0369*** 0.0895*** 0.253*** 0.0186 -0.0516*** -0.00656 1 * p < 0.05, ** p < 0.01, *** p < 0.001
Table 5. Model 1-3
Model: (1) (2) (3)
Total Subsidy (Ln) Total Subsidy (Ln) Total Subsidy (Ln)
# of projects product 0.028*** 0.028*** 0.023*** (0.003) (0.003) (0.003) # of projects process 0.016*** 0.016*** 0.011*** (0.003) (0.003) (0.002) # of projects programming 0.0755*** 0.0764*** 0.0577*** (0.010) (0.010) (0.007)
Fixed Sum (dummy) 0.194 0.222 -0.176
(0.135) (0.137) (0.093)
Industry Dummiesa Included Included Included
2013b -0.029 -0.012 (0.021) (0.021) 2014 -0.077* -0.056* (0.030) (0.027) 2015 -0.116*** -0.084** (0.029) (0.028) 2016 -0.109*** -0.108*** (0.033) (0.032) 2017 -0.090** -0.114*** (0.033) (0.032) Firm Age (Ln) 0.248 (0.242) Firm Age2 (Ln) -0.098 (0.100) Size (Ln) 0.0340* (0.014) R&D Intensity (Ln) 0.623*** (0.037) _cons 10.18*** 10.23*** 9.388*** (0.302) (0.302) (0.234) Number of observations 8800 8800 8457 Number of Firms 2812 2812 2693 R2 Within 0.057 0.060 0.080 R2 Between 0.144 0.145 0.446 R2 Overall 0.158 0.159 0.470
Note a: Industry dummies not presented for the sake of brevity (see appendix 2 for full model) Noteb: 2012 is base year
Standard errors in parentheses
* p < 0.05, ** p < 0.01, *** p < 0.001
that Size is positively and significantly related to the total subsidy (ß = 0.034, p=0.014). Therefore, hypothesis 1 is supported at the 0.05 significance level and it can be concluded that
Size is positively and significantly related to the total subsidy. This is consistent with similar
findings by Busom et al. (2017), who looked at firm participation. For Firm Age and Firm Age2
no significant effects are found. Therefore, hypothesis 2 is rejected (ß = -0.098) at the 0.05 significance level. No U-shaped relationship between firm age and total subsidy exists for our model. For R&D Intensity a significant effect is found. Hypothesis 3 is fully supported, and it is found that R&D Intensity (ß = 0.62) is significant at the 0.001 significance level (p=0.000). This is contradicting to Busom et al. (2017), who did not find a direct correlation between R&D intensity and firm participation. This could be due to that Busom et al. (2017) did look at firm participation and not the total amount of subsidy. The VIF values, except for firm age and firm
age 2, are <10, indicating that the level of multicollinearity is acceptable (Hair, Anderson,
Tatham and Black, 1995). The relationship between firm age and the total subsidy is further examined in subsequent analysis. The scatter plot of the predicted values from model 3 is depicted below (figure 4).
Additional Analysis
Fixed Effects ModelIn panel data analysis two types of regression models are common: random effects and fixed effects models. As is discussed in the methodology, a random effects model is the preferred option for this thesis based on the theory. The Hausmann’s test, that is used to differentiate between both effects, rejects the null hypothesis: difference in coefficients not systematic (appendix 3). Hence, the test results suggest a fixed effects model is appropriate. As this contradicts the expectations based on the theory, the fixed effects model is tested and reported to provide the results that would have been found under a fixed effects model (Table 6).
Table 6. Model 4 Fixed Effects Model
Model 4 Total Subsidy (Ln)
Year variable (Continuous) -0.028**
(0.01) # of projects product 0.019*** (0.002) # of projects process 0.015*** (0.002) # of projects programming 0.054*** (0.005) Firm age(Ln) -0.362 (0.462) Firm age2 (Ln) 0.136 (0.155) Size (Ln) -0.028* (0.014) R&D Intensity (Ln) 0.300*** (0.025) _cons 67.43** (20.50) Number of observations 8482 Number of firms 2710 R2 0.092
Standard Errors in parentheses
* p < 0.05, ** p < 0.01, *** p < 0.001
variable is significant at the 0.01 level (ß=-0.028, p=0.008). Interpretation is that the Total
Subsidy (Ln) is expected to decrease by -0.028*2018 for the upcoming year.
Post-Hoc Analysis
Because no support for hypothesis 2 is found, a robustness check with age as a linear variable is conducted. For this, the firm age2 variable is dropped from the analysis. Testing this in model 5 results in finding support for hypothesis 2 at the 0.05 significance level (appendix 4). The coefficient for this is 0.057 and the hypothesis is accepted at the 0.01 confidence interval (p=0.024). All VIF values are below ten in this model. It can be concluded that firm age is positively and significantly related to the total subsidy. This differs from the hypothesis that the relationship is non-linear, as expected based on the literature analysis.
Moreover, the data allows to take turnover(ln) as an alternative proxy for Size. One theoretical argument to use turnover is that an increase in subsidy received could increase the number of employees hired by the firm. This would cause some endogeneity among firm Size and the receipt of public funding (Almus and Carnitzcki, 2003). The results from this model are reported in model 6 (appendix 4). Model 6 shows that turnover is significantly and positively related to the total subsidy (ß=0.024, p=0.001). This result is similar to using the number of employees as proxy for the Size variable. Moreover, the R-squared does not increase using this variable. It can therefore be concluded that based on this employees and turnover explain the
total subsidy equally. However, analysis shows that turnover has a VIF of 35.17. Using
employees as proxy for Size is therefore recommended.
Subsidy Per Employee
In further analysis the subsidy per employee is used as an alternative dependent variable. The advantage of testing with this variable is that it allows for better comparison between firms. Hence, testing with subsidy per employee is more suited to derive practical implications for the consultancy firm. The results from this analysis are reported in appendix 5 (Model 7) and the scatter plot in appendix 6. What is particularly interesting is that the results differ substantially compared to the models with total subsidy as dependent variable. For Firm Age (ß=-0.641),
Size (ß=-0.011) and R&D intensity (ß=-0.002) the coefficients are negative. Only Firm Age is
Discussion & Conclusion
In this thesis the knowledge on how firm characteristics explain the amount of subsidies applied for is increased. By analysing the data of roughly 2700 firms the results give a close approximation of the situation in the Netherlands. This study, to the best of our knowledge, is the first to look at firm characteristics and how they explain the applied for subsidy in case of the Dutch tax incentive WBSO.
As is discussed in the results section, support is found for the hypotheses 1 and 3. The first hypothesis, that predicts that Size will positively relate to total subsidy is supported. Moreover, it can be concluded that the R&D intensity significantly influences the total subsidy applied for. The main model is significant at the 0.001 significance level. It can therefore be concluded that
R&D intensity and Size, together with the control variables, explain a substantial part of the
answer to the question; how do firm characteristics explain the total subsidy applied for? Another interesting result is that the non-linear function of firm age does not significantly relate to the total subsidy applied for. This is contradicting to what is expected from the literature review. In additional analysis, a positive and linear effect is found. The results contradict with the findings of Segarra-Blasco and Teruel (2016), that did not discover significant relationships for applying propensity and firm age for Catalonian firms. This indicates that the relationship can differ per institutional setting.
Moreover, a strong and negative relationship between firm age and the total subsidy per
employee is found. From this result, it seems that the WBSO is effective in stimulating younger
Practical Implications
With the developed models part of the variance in applications can be predicted. This is relevant for the practical implications of the thesis. With the predictions from the model, the consultancy firm can compare the current applications with the expected applications. This allows for comparisons across industries, firms and offices of the consultancy firm.
To allow for comparison the following procedure is performed. The results from the regression with subsidy per employee as dependent variable (model 7) are used to predict the subsidy per employee for every observation in the dataset. The predicted value is then compared to the actual applied for subsidy per employee. Next, the difference between the predicted and actual subsidy per employee is calculated. Based on the percentiles of the difference a score from 1 to 4 is assigned to each observation. A “1” means that the observation is among the 25 % percent of observations that differ the most from the predicted score in a negative sense. A score of “4” indicates that the observation is among the 25 % with the largest difference in a positive sense. Observations with a score of “1” provide the consultancy firm with an indication that for this particular observation a higher subsidy per employee might be possible.
TABLE 7: Distribution per industry (p 50) industry 1 2 3 4 Total Industry 1 20 14 25 20 79 % 25.32 17.72 31.65 25.32 100 Industry 2 647 712 582 628 2,569 % 25.18 27.72 22.65 24.45 100 Industry 3 30 25 18 27 100 % 30.00 25.00 18.00 27.00 100 Industry 4 85 34 27 51 197 % 43.15 17.26 13.71 25.89 100 Industry 5 247 198 215 237 897 % 27.54 22.07 23.97 26.42 100 Industry 6 26 15 26 30 97 % 26.8 15.46 26.8 30.93 100 Industry 7 268 370 373 367 1,378 % 19.45 26.85 27.07 26.63 100 Industry 8 309 361 396 328 1,394 % 22.17 25.9 28.41 23.53 100 Industry 9 422 366 417 511 1,716 % 24.59 21.33 24.3 29.78 100 Industry 10 44 31 42 69 186 % 23.66 16.67 22.58 37.1 100 Industry 11 17 5 11 40 73 % 23.29 6.85 15.07 54.79 100 Industry 12 36 21 19 38 114 % 31.58 18.42 16.67 33.33 100 Total 2,151 2,152 2,151 2,346 8,800 % 24.44 24.45 24.44 26.66 100
Limitations & Future Research
The results and conclusions of this thesis are limited in that they are based on a specific institutional setting. The results are therefore not generalizable across countries or even between different subsidies and tax incentives in the Netherlands. To exemplify, Busom et al. (2017) have concluded that differences exist between characteristics that explain tax incentive participation and subsidy participation in the Spanish manufacturing industry.
Moreover, this thesis has several limitations that result from the data. First of all, the reliability of the data depends on how accurate the consultants have filled in the data. It is impossible to control if all data is filled in consistently and accurately. For the consultants there is limited incentive to fill in the data as accurate as possible, as the characteristics are for descriptive purposes rather than part of the application. Moreover, the consultant can fill in data for the application that is a minimum of one year and a maximum of five years prior to the application date. A second limitation to the data is that it is unknown if a firm decides to go to a competitor or apply independently for a certain year or a certain project. It is possible that firms do apply for subsidies in certain years, without making use of the services from the consultancy firm. Third, the used variables explain limited variance. The data did not allow to measure the characteristics of the entrepreneur or the quality of the application of a specific project. This is some unobserved effect that would be interesting to study in future studies. The dataset did provide variables that are to some extend related to each other (Age, Size and R&D intensity) and more variation in variables would have been interesting. Furthermore, the models did not control for the hourly wage of employees per firm and not for whether or not a firm cooperates with other firms for a certain project. These control variables are recommended to consider in future research. A forth limitation is that the Hausmann’s test suggested a fixed effects model, where theory suggests a random effects model. A last limitation of the dataset is that it is does not provide information on whether firms actually realize the subsidy applied for and if the subsidy is granted. This limits the usability for further research and the implications that the results have. It is therefore recommended for future researchers to enrich the data with the realized results.
References
Aerts, K. (2008). Who Writes the Pay Slip? Do R&D Subsidies Merely Increase Researcher
Wages? K.U.Leuven - Faculty of Business and Economics; Leuven, Belgium
Akcigit, U., Hanley, D., & Serrano-Velarde, N. (2013). Back to basics: Basic research
spillovers, innovation policy and growth (No. w19473). National Bureau of Economic
Research.
Almus, M., & Czarnitzki, D. (2003). The effects of public R&D subsidies on firms' innovation activities: the case of Eastern Germany. Journal of Business & Economic
Statistics, 21(2), 226-236.
Blanes, J. V., & Busom, I. (2004). Who participates in R&D subsidy programs? The case of Spanish manufacturing firms. Research policy, 33(10), 1459-1476.
Busom, I., Corchuelo, B., & Martínez-Ros, E. (2014). Tax incentives… or subsidies for business R&D?. Small Business Economics, 43(3), 571-596.
Busom, I., Corchuelo, B., & Martínez-Ros, E. (2014). Tax incentives… or subsidies for business R&D?. Small Business Economics, 43(3), 571-596.
CBS. (2017). Standaard Bedrijfsindeling 2008. Kamer Van Koophandel. Retrieved June 19,
2018, from https://www.kvk.nl/download/SBI_code_NL_2018_dec2017_tcm109-451912.pdf.
Clarysse, B., Wright, M., & Mustar, P. (2009). Behavioural additionality of R&D subsidies: A learning perspective. Research Policy, 38(10), 1517-1533.
Clausen, T. H. (2009). Do subsidies have positive impacts on R&D and innovation activities at the firm level? Structural change and economic dynamics, 20(4), 239-253.
Czarnitzki, D., & Hussinger, K. (2004). The link between R&D subsidies, R&D spending and technological performance.
David, P. A., Hall, B. H., & Toole, A. A. (2000). Is public R&D a complement or substitute for private R&D? A review of the econometric evidence. Research policy, 29(4-5), 497-529. Duguet, E. (2004). Are R&D subsidies a substitute or a complement to privately funded R&D? Revue d'économie politique, 114(2), 245-274.
Falk, M. (2006). What drives business Research and Development (R&D) intensity across Organisation for Economic Co-operation and Development (OECD) countries? Applied Economics, 38(5), 533-547.
Galindo-Rueda, F., & Verger, F. (2016). OECD taxonomy of economic activities based on R&D intensity.
Hair, J. F. Jr., Anderson, R. E., Tatham, R. L. & Black, W. C. (1995). Multivariate Data
Analysis (3rd ed). New York: Macmillan.
Hair, J., Black, W. C., Babin, B. J., & Anderson, R. E. (2010). Multivariate data analysis (7th
ed.). Upper saddle River, New Jersey: Pearson Education International
Huergo, E., & Jaumandreu, J. (2004). How does probability of innovation change with firm age?. Small Business Economics, 22(3-4), 193-207.
Huergo, E., & Trenado, M. (2010). The application for and the awarding of low-interest credits to finance R&D projects. Review of Industrial Organization, 37(3), 237-259. Keijzers, G., & Bos-Brouwers, H. E. J. (2005). Innovaties voor duurzaamheid in het MKB; naar nieuwe aangrijpingspunten voor beleid.
Lach, S. (2002). Do R&D subsidies stimulate or displace private R&D? Evidence from Israel.
The journal of industrial economics, 50(4), 369-390.
Levin, R. C., Cohen, W. M., & Mowery, D. C. (1985). R & D appropriability, opportunity, and market structure: new evidence on some Schumpeterian hypotheses. The American
Economic Review, 75(2), 20-24.
Lichtenberg, F. R. (1984). The relationship between federal contract R&D and company R&D. The American Economic Review, 74(2), 73-78.
Lokshin, B., & Mohnen, P. (2012). How effective are level-based R&D tax credits? Evidence from the Netherlands. Applied Economics, 44(12), 1527-1538.
Majumdar, S. K. (2000). With a little help from my friends? Cross-subsidy and installed-base quality in the US telecommunications industry. International Journal of Industrial
Organization, 18(3), 445-470.
Schneider, C., & Veugelers, R. (2010). On young highly innovative companies: why they matter and how (not) to policy support them. Industrial and Corporate change, 19(4), 969-1007.
Segarra-Blasco, A., & Teruel, M. (2016). Application and success of R&D subsidies: what is the role of firm age? Industry and Innovation, 23(8), 713-733.
Silva, A. M., Silva, S. T., & Carneiro, A. (2017). Determinants of grant decisions in R&D subsidy programmes: Evidence from firms and S&T organisations in Portugal. Science and
Public Policy,44(5), 683-697. doi:10.1093/scipol/scx002
STATA Base Reference Manual. Stata Press (2017) https://www.stata.com/bookstore/base-reference-manual/, release 15
Peneder, M. (2008). The problem of private under-investment in innovation: A policy mind map. Technovation, 28(8), 518-530.
Poot, T., den Hertog, P., Grosfeld, T., & Brouwer, E. (2003). Evaluation of a major Dutch Tax Credit Scheme (WBSO) aimed at promoting R&D. In FTEVAL Conference on the
Evaluation of Government Funded R&D, Vienna.
Verhoeven, W.H.J., Van Stel, A.J., & Timmermans, N.G.L. (2012). Evaluatie WBSO 2006-2010. Effecten, doelgroepbereik en uitvoering; Rapport in opdracht van het Ministerie van Economische Zaken.
Verhoeven, W.H.J., Van Stel, A.J. (2013) 'Hoe effectief is de WBSO-regeling door de jaren heen?', Beleidsonderzoek Online juni 2013, DOI: 10.5553/Beleidsonderzoek.000023
Wallsten, S. J. (2000). The effects of government-industry R&D programs on private R&D: the case of the Small Business Innovation Research program. The RAND Journal of
Economics, 82-100.
Wanzenböck, I., Scherngell, T., & Fischer, M. M. (2013). How do firm characteristics affect behavioural additionalities of public R&D subsidies? Evidence for the Austrian transport sector. Technovation, 33(2-3), 66-77.
Appendix
Appendix 1
Table 8. Summary of variables
Variable Description Notes Literature reference(s)
Dependent variables
Total Subsidy The subsidy applied for per firm per year
The total subsidy consists of Log+1 of the sum of the hours*wage and the amount of cost expenditure applied for by a firm in a year Total Subsidy Per Employee (Only for additional analysis)
Ratio to measure the applied for subsidy
The total subsidy per employee consists of the Log+1 of the Total Subsidy divided by the number of employees Independent variables
Size Total Number of
employees
Log+1 is taken of the total number of employees. Moreover, turnover is used as an alternative proxy to include size effects in the model
Aerts (2008); Almus and Czarnitzki (2003); Busom et al., 2017); Clausen (2009); Duguet (2004); Heijs & Herrera (2004);
Wanzenböck et al., (2013)
Firm Age Firm Age2
Age of the firm and Age of the firm to the power of 2
Log +1 of the nonlinear function is used. In later
analysis the firms are divided in four categories to account for multicollinearity. The
information for this is retrieved from the Dutch Chamber of Commerce
Busom (2000); Czarnitzki and Hussinger (2004); Heijs & Herrera (2004);
Wanzenböck et al. (2013); Huergo, E., & Trenado, M. (2010).
R&D Intensity Number of R&D employees
For this the Log+1 of the number of R&D employees allocated to the projects in a year, is taken as a proxy.
Busom et al., (2017); Clarysse, Wrigth & Mustar (2009); Wallsten (2000); Wanzenböck et al., (2013)
Control variables
Type of project Project, process or programming
The WBSO considers different types of projects. This control variable is used to account for project specific effects
Poot, Hertog, Grosfeld, and Brouwer (2003); Silva, Silva, & Carneiro, (2017)
Fixed Sum If the application was
submitted as Fixed Sum
A dummy to account for the application being fixed sum. This means that the
firm/consultant can apply for 1800 labor hours at a rate of €10 per hour, and additional hours at €4 per hour instead of a
calculated rate.
Industry The industry that the firm belongs to
For this the SBI codes are used. Aims to control for industry specific effects.
Akcigit, Hanley & Nicolas Serrano-Velarde (2012); Silva, Silva, & Carneiro, (2017)
Year variable Dummies for
2013-2017
This accounts for year specific effects that might occur, for example from differences in WBSO regulations.
Aerts (2008); Silva, Silva, & Carneiro, (2017)
Appendix 2
Table 9. Model 1-3 including industries
(1) (2) (3)
Total Subsidy (Ln) Total Subsidy (Ln) Total Subsidy (Ln)
# of projects product 0.028*** 0.028*** 0.023*** (0.003) (0.003) (0.003) # of projects process 0.016*** 0.016*** 0.011*** (0.003) (0.003) (0.002) # of projects programming 0.0755 *** 0.0764*** 0.0577*** (0.010) (0.010) (0.007)
Fixed Sum (dummy) 0.194 0.222 -0.176
2016 -0.109*** -0.108*** (0.033) (0.032) 2017 -0.090** -0.114*** (0.033) (0.032) Firm Age (Ln) 0.248 (0.242) Firm Age (2)(Ln) -0.098 (0.100) Size (Ln) 0.0340* (0.014) R&D Intensity (Ln) 0.623*** (0.037) _cons 10.18*** 10.23*** 9.388*** (0.302) (0.302) (0.234) Number of observations 8800 8800 8457 Number of Firms 2812 2812 2693 R2 Within 0.057 0.060 0.080 R2 Between 0.144 0.145 0.446 R2 Overall 0.158 0.159 0.470
Note: 2012 is base year and other industries is base industry Standard errors in parentheses
* p < 0.05, ** p < 0.01, *** p < 0.001
Appendix 3
Table 10. Hausmann Test
(b) (B) (b-B) sqrt(diag(V_b-V_B)) FE RE Difference S.E. Firm Age (Ln) -0.357 0.248 -0.604 0.429 Firm Age 2 (Ln) 0.131 -0.098 0.229 0.136 Size (Ln) -0.029 0.034 -0.063 0.010 R&D Intensity (Ln) 0.300 0.623 -0.323 0.017 2013 -0.031 -0.012 -0.020 0.010 2014 -0.082 -0.056 -0.026 0.019 2015 -0.094 -0.084 -0.010 0.028 2016 -0.129 -0.108 -0.021 0.037 2017 -0.133 -0.114 -0.019 0.046 # of projects product 0.019 0.023 -0.004 0.001 # of projects process 0.016 0.011 0.005 0.001 # of projects programming 0.056 0.058 -0.002 0.003
b = consistent under Ho and Ha; obtained from xtreg
B = inconsistent under Ha, efficient under Ho; obtained from xtreg Test: Ho: difference in coefficients not systematic
chi2(12) = (b-B)'[(V_b-V_B)^(-1)](b-B) = 449.70
Appendix 4
Table 11. Additional Analysis
Model: (5) (6)
Total Subsidy Total Subsidy
# of projects product 0.0235*** 0.024*** (0.003) (0.003) # of projects process 0.0102*** 0.010*** (0.003) (0.002) # of projects programming 0.0598*** 0.061*** (0.007) (0.007)
Fixed Sum (dummy) -0.172 -0.165
(0.093) (0.093)
Industry Dummiesa Included Included
2013b -0.0110 -0.014 (0.021) (0.021) 2014 -0.0611* -0.064* (0.027) (0.027) 2015 -0.0889** -0.085** (0.028) (0.028) 2016 -0.111*** -0.117*** (0.032) (0.032) 2017 -0.117*** -0.123*** (0.032) (0.032) Firm Age (Ln) 0.0568* 0.0678** (0.024) (0.024) Size (employees) (Ln) 0.0341* - (0.013) - Turnover (Ln) - 0.0239** - (0.007) R&D Intensity (Ln) 0.608*** 0.604*** (0.037) (0.038) _cons 9.447*** 9.185*** (0.194) (0.224) Number of observations 8645 8644 Number of firms 2752 2752 R2 Within 0.080 0.080 R2 Between 0.441 0.440 R2 Overall 0.467 0.467
Note a: Industry dummies not presented for the sake of brevity Noteb: 2012 is base year
Standard errors in parentheses
* p < 0.05, ** p < 0.01, *** p < 0.001
Appendix 5 Table 12. Model 7
Model: (7)
Firm Age (Ln) -0.641*** (0.034) Size (Ln) -0.011 (0.023) R&D Intensity (Ln) -0.002 (0.04) # of projects product 0.023*** (0.002) # of projects process 0.011*** (0.003) # of projects programming 0.045*** (0.008)
Fixed Sum (dummy) 0.180
(0.097)
Industry dummiesa Included
2013b 0.061* (0.025) 2014 0.0173 (0.031) 2015 0.005 (0.034) 2016 0.029 (0.036) 2017 -0.009 (0.038) _cons 9.537*** (0.293) Number of observations 8631 Number of firms 2745 R2 Within 0.061 R2 Between 0.277 R2 Overall 0.257
Note a: Industry dummies not reported for the sake of brevity Note b: 2012 is base year
Standard errors in parentheses
* p < 0.05, ** p < 0.01, *** p < 0.001
Appendix 7: Scores per office
Table 13. Distribution of scores between offices
Office 1 2 3 4 Total Office 1 499 511 485 447 1,942 26% 26% 25% 23% 100% Office 2 391 395 369 326 1,481 26% 27% 25% 22% 100% Office 3 328 311 344 442 1,425 23% 22% 24% 31% 100% Office 4 257 255 295 303 1,110 23% 23% 27% 27% 100% Office 5 221 253 266 260 1,000 22% 25% 27% 26% 100% Office 6 215 213 178 209 815 26% 26% 22% 26% 100% Office 7 51 57 68 148 324 16% 18% 21% 46% 100% Office 8 88 54 47 94 283 31% 19% 17% 33% 100% Office 9 36 38 39 49 162 22% 23% 24% 30% 100% Other 64 65 60 109 298 21% 22% 20% 37% 100% Total 2,150 2,152 2,151 2,387 8,840
