The moderating effect of Human Capital on the Innovation
and Employment growth relationship
By Friso Paul Verhoef S2696142
f.p.verhoef@student.rug.nl 20-June-2017 University of Groningen Faculty of Economics and Business
Msc Thesis
Supervisor: Prof. Raquel Ortega-Argilés Co-assessor: Prof. Stephan Klasen
Keywords: Innovation, Employment, Human capital
Abstract
2
Table of contents
1. Introduction ... 3
2. Literature review ... 4
2.1 Theoretical Precursors ... 5
2.2 Innovation and employment growth ... 5
2.3 Human capital and employment growth ... 8
3. Empirical model ... 11
4. Data description ... 16
4.1 Data, sample and variables ... 16
3
1. Introduction
In 2010 the European Commission set a growth strategy for the next 10 year, called Europe 2020 (European Commission, 2010). In this document a bolt statement was made: “In a
changing world, we want the EU to become a smart, sustainable and inclusive economy”.
Concretely, this means that the EU has set five ambitious objectives – on employment, innovation, education, social inclusion and climate/energy – to be reached by 2020. For example the EU wants an increase in R&D spending of 3% by 2020, achieve a 75% employment rate for workers aged 20-64, and a competing third level education (European Commission, 2015). One of the main flagship initiatives taken by the EU to reach these EU 2020 goals is innovation (European Commission 2012). Innovation is seen as a main engine for generating growth and jobs, increasing the competitiveness of the EU and solving challenges related to population ageing (European Commission, 2014).
A legitimate question that could be raised by these policy suggestions is what the effects of long-term competitive policies, such as innovation, on employment will be. This question however is difficult to answer, which is pointed out in the vast literature on the subject. The relationship between technical progress or innovation and employment is a topic that is debated for some time and dates back to classical economists like Ricardo (1929), Marx (1906) and Schumpeter (1936). Two views have emerged in the debate. On the one hand, the optimistic view, this view stresses the capacity of innovation to generate new job opportunities. On the other hand, the pessimistic view, this view underlines market failures which can limit the compensation effects of innovation and emphasises the negative impact that technical change has on employment (Spiezia and Vivarelli, 2002). These opposing views make a theoretical analysis of innovation complicated.
4 Therefore, the research question for this paper is twofold:
First: Does innovation positively influence employment growth at the regional level in the EU? Second: Does human capital positively moderates the relationship between innovation and employment growth?
Being able to answer this twofold research question provides supportive evidence to the new Economic Growth strategy of the European Commission as well as adds to existing literature in three ways. First, it gives additional empirical evidence concerning the impact of innovation on employment. Second, it fills up the void concerning the moderating effect of human capital at the regional level. Last, it stresses the importance of distinguishing between quality and quantity of human capital.
Using data from 289 regions located in Europe during the period 2000-2012 this paper finds evidence that innovation has a positive effect on employment growth. However, this effect can only be found when innovation is combined with skills, emphasizing the difference between the quality and quantity of human capital. For education levels, this paper finds that only high levels of education have a significant and positive effect on employment growth. For low levels of education, the effect on employment growth is negative, while for medium levels of education the effects are not significant. Additional analyses confirm these findings, but also show that depending on the sector, different levels of education and skills have different effect on the employment share of that sector, suggesting that different policies are needed for different sectors.
The remainder of the paper is as follows. In the next section, relevant literature concerning innovation and human capital will be reviewed. In the third section, the method used to analyse the research question will be discussed. Then in the fourth section the data used in the analysis will be presented. After which in the fifth section, the results of the analysis and the tests will be shown and elaborated on. Finally, the paper will end with some conclusions and policy implications derived from the findings as well as with suggestions for further research.
2. Literature review
5
2.1 Theoretical Precursors
Defining theoretical precursors will give an understanding of where the research comes from and on which assumptions the empirical model in section three is based. This section will therefore look at theoretical precursors concerning innovation and human capital. Early on, Romer (1990) developed an endogenous growth theory, where investments in innovation lead to an increase in output and thereby to economic growth. The crucial feature of Romer’s model is that knowledge enables the development and production of new goods, as well as it increases the total stock of knowledge thereby increasing the productivity. Looking at the theoretical precursors of human capital, Nelson and Phelps (1966) argued that a more educated labour force would imitate frontier technology faster, where states that are further away from the frontier have greater catching-up benefits. Expending on their work, Benhabib and Spiegel (1994) argued that a more educated labour force would also innovate faster. Aghion et al. (2009) build on the theoretical model of Acemoglu et al. (2002) and argue that innovation makes intensive use of highly educated workers while imitation relies more on combining physical capital with less educated labour. Even though these theories use economic growth as determinant by extension employment growth can also be used.
2.2 Innovation and employment growth
This section will first present evidence on the effect of R&D and patents, afterwards presenting evidence on the effect of different outcomes of innovation. Unfortunately, there is no dataset that allows to compare different outcomes of innovation at a regional level for the whole EU. However, the research is relevant in providing evidence and support for the research question raised in this paper.
6 of research; research which aggregates product and process innovation, and research that disaggregate product and process innovation.
Using EU scoreboard data from 2004 until 2009 and a reduced form equation in which Research and Development (R&D) can account for both product and process innovation Bogliacino (2014) finds that R&D has a positive impact on employment growth. More interestingly he finds that there is a non-linear effect of R&D on employment which he finds to be positive. However, Griliches (1990) points out that, although patent data and R&D data are often chosen to individually represent the same phenomenon, these exists a statistical discrepancy in that there is typically a great randomness in patent series, whereas R&D values are much more smooth. Because of this, the variable of interest could be measured with noise if one takes either innovative input or innovative output. He therefore argues that both innovation input and output have to be taken into account.
Coad and Rao (2011) use patents and R&D to account for both innovation input and output. They create an original database for the patents by matching the NBER patents database with the Compustat file database for four two-digit manufacturing industries that are known for their high patenting activity. They find that depending on the size of the company, R&D has no or a positive effect on employment growth in high-tech US manufacturing industries. The results they find for the effect of patents on employment is similar to that of R&D, patens are positively associated with subsequent employment growth.
This is confirmed by Buerger et al. (2012) who also look at R&D and patents. They use regional data of German firms and focus on four different industries between 1999 and 2005. For the transport industry they find no significant correlation between patent applications and employment growth, however they argue that this is in line with the expectations. For the other three industries (Chemistry, Electrics and Manufacturing) they do find a positive effect of patents on employment growth; for R&D they find that there is a so called “success-breeds-success” effect, meaning that R&D results in employment growth but that employment growth also leads to growth in R&D.
Agovino et al. (2016) go even a step further by investigating the impact of patents/R&D ratio and patents/spillover ration on employment levels. For this they used two sources of data; the R&D data was collected by using EU R&D investment scoreboard, while the patent data was collected by using the OECD’s REGPAT database, which covers firms’ patent applications to the European Patent Office (EPO). They find that R&D and patents/R&D ratio always have a positive effect on employment, in addition R&D measured as R&D expenditures is more significant and has a bigger impact on employment than patents/R&D ratio.
7 a larger effect for normal employment. Second, they look at the number of companies in R&D and Computer related activities as innovation input and find that the number of companies in R&D has no effect on employment but the number of companies in Computer and related activities do.
The second stream of papers take into consideration the distinction between different types of innovation. As already mentioned, theory distinguishes between two forms of innovation output: product innovation and process innovation. Antonucci and Pianta (2002) argue that both of them have different effects, while process innovation leads to improvements in efficiency of production of particular goods and services, product innovation increases the quality and variety of goods and may even open up new markets. Besides the different ways of innovation, the impact on employment is also different.
Using a micro-dataset from West German manufacturing firms from 1980 until 1992 Smolny (1998) finds positive effects for both product and process innovation on employment growth. His empirical results even show that innovations change market behaviour. In sectors with a large share of product innovators, firms more often change employment and less often change prices. He concludes this is evidence that product innovations reduce price competition. With the same token but at a regional level, Capello and Lenzi (2012) analyse that different effects of process and product innovation on employment growth. To capture the different types of innovation, they use various questions of the Community Innovation Survey (CIS) and in particular those concerning the share of firms introducing only product innovations and the share of firms introducing only process innovation. To be able to use CIS data, which is at country level, for regional comparison they construct an estimation methodology using weights to redistribute the country level data to the regional level data. What they find is that the relationship between innovation and employment growth cannot be studied at the regional level without considering the regional aspects, which may mitigate or magnify the direct effect of innovation on employment. When the direct effects are moderated by regional specialization they find that product innovation has a positive effect on employment, while process innovation has a negative effect on employment growth when moderated by regional settlement structures. Going a step further, Lachenmaier and Rottman (2007) combine innovation input and output with process and product innovation using a unique long panel data set of German manufacturing firms from 1982 to 2003. They do this by using the IFO innovation survey’s which contain different measures of innovation making it possible to distinguish between product and process innovation. This way they have R&D product and process innovation as well as patent product and process innovation, which they aggregate into product and process innovation. They find that both product and process innovation have a positive and significant effect on employment growth. Moreover, in general the impact of process innovation is higher than that of product innovation.
8 where product and process innovation are distinguished. Because CIS surveys measure innovations over a three year period without specifying the exact year in which the innovation occurred. Meriküll computes an innovation variable that takes a value of 1 for each year within the three-year period that the firm was innovating, and 0 for each year within the three year period that the firm was not innovating. He finds a positive effect of both types of innovation on employment, however the effect of process innovation on employment is bigger than that of product innovation. He says that is because demand expansion compensate for the factor saving effect, which indicates a high price elasticity of demand.
The findings of both streams of empirical research show, that in general innovation has a positive effect on employment growth. Theory argues that it is important to use both innovation input and output measures, as well as distinguishing between process and product innovation. However, empirical analysis shows that the overall impact of innovation on employment growth is positive even if only input or output measures are taken or when product and process innovation are combined. Following these results we can formulate the following hypothesis:
H1: Innovation has a positive effect on employment growth
2.3 Human capital and employment growth
The third theoretical aspect in this paper looks at the moderating effect of human capital on the relationship between innovation and employment growth. This section will make the distinction between quality and quantity of human capital, by first examining the effect of skill-levels after which the effects of education are examined.
The relationship between human capital and employment growth has been widely studied. However, Sasso and Ritzen (2016) argue that it is important to look at the measure that is taken for human capital; human capital can be measured by taking input measures (school enrolment and school attainment) or output measures (attained knowledge and skills). According to them most research has measured human capital through input variables however these do not capture the actual knowledge and skills that human capital provides individuals. This has led to contrasting results on the role of human capital on productivity and growth because, they measure only the knowledge that students have and do not represent the human capital that is embodied in the workforce. The skills of the workers operating in the various industries are shaped not only by the formal education that they received, but also by the experience and training that they have gone through in their working life.
9 looking at the moderating effect of human capital it is important to distinguish between input and output measures, or in other words, quantity and quality of human capital
Output measures of human capital, or the quality of human capital are usually measured by skill-levels. In their 2003 paper, Autor et al. formulate a theoretical model to conceptualize work from a machine’s eye view and ask the question: Which tasks can be performed by a computer? Autor et al. (2003) distinguishes tasks into; analytic interactive, routine and non-routine tasks, and manual non-routine and non-non-routine tasks. (A table with examples can be found in the appendix). They find when looking at analytic and interactive tasks that routine tasks, which are typically low-skilled activities such as calculations and record-keeping are subjected to substantial substitution. However, non-routine tasks, which are typically high-skilled such as managing others or medical diagnosis have strong complementarities with innovation. For manual tasks Autor et al. (2003) find that routine tasks, which are typically low-skilled, such as picking or sorting are again subjected to substantial substitution. While for non-routine tasks, which are typically low-skilled, such as janitorial services they find that there are limited opportunities for substitution or even complementarity with innovation. The theoretical model of Autor et al. (2003) points out that when innovation is present, the amount of jobs that require high-skilled workers increase while the amount of jobs that require low-skilled workers decrease.
Using Italian Community Innovation Survey data (CIS2) from 1993-1995, Evangelista and Savona (2003) empirically confirm the findings of Autor’s theoretical model. Evangelista and Savona (2003) look at the employment impact of innovation in the heterogeneous universe of services with a survey that asks firms if the introduction of innovation has led to: (i) an increase, (ii) a decrease, or (iii) had no effect on the use of labour. Unlike other countries involved in CIS2, Italian firms have also been asked about the impact innovation has on labour with different levels of qualification (high, medium, low). They find that innovative firms are more likely to indicate a positive impact of innovation on total employment. More importantly, the positive impact that innovation has on employment is limited to high skilled labour.
10 employability. The indirect effects are twofold. Berry and Flaeser (2005) argue that highly educated workers attract other workers and that this external effect produces and increases regional employment, particularly in regions with higher shares of highly skilled people (Glaeser et al., 1995; Simon, 1998). The second indirect effect education has, comes from the fact that education leads to labour market and social mobility for the poorest society (Di Cataldo and Rodriguez-Pose, 2017).
Empirically analysing the effects of education on employment growth; Morrison et al. (2001) use data on U.S. manufacturing sectors from 1959 till 1989 and a dynamic cost function framework to assess the impact of trade, technology, and outsourcing on shifts in labour demand. They find that the demand for labour is in favour of highly educated people, while reducing the demand for workers without a college degree. This is confirmed by Ciccone and Papaioannou (2009) who find while using input measures of human capital - school attainment and school enrolments ratios - that countries with higher levels of human capital have higher value-added growth and higher growth in skill-intensive industries.
With the same token but using two measures of education - university education and research institutes - Schlump and Brenner (2010) empirically analyse the effect education has on employment. To do this they use data on the number of graduates in diverse university subjects in all German labour market areas as well as the number of employees in various subjects in research institutes of the four main German research societies from 1999 to 2006. They find that both variables - university education and research institutes - affect local employment growth and that the impact of the two is very similar. They state that university education as well as public research is important because it stimulates employment as well as it can foster regional economic growth.
Using only one measure of education but having a dataset of individuals residing in 283 U.S. metropolitan areas from 1980 until 2000 Winters (2013) investigates educational externalities and employment with a pooled cross-section analysis. He finds that higher levels of education result into higher employment due to increases in labour market participation. Furthermore, he finds that workers sort into the labour market that gives them the highest utility, which confirms the importance of regional characteristics argued by Capello and Lenzi (2012). Capello and Lenzi (2012) argued that regional aspects need to be considered because they may mitigate or magnify the effect innovation has on employment.
On another note but as important Di Cataldo and Rodriguez-Pose (2017) argue there are several externalities facilitating labour market inclusion that stem from education. Higher education rates decreases criminal activities (Lochener and Moretti, 2004), increases civic participation (Milligan et al., 2004) and quality of life (Shapiro, 2006). To measure the moderating effect of education Di Cataldo and Rodriguez-Pose (2017) take the percentage of total students in higher education and find an overall positive effect on employment growth.
11 human capital, and education-levels the quantity of human capital. Overall it is found that even though there is an important distinction between the two, they both have a positive impact on employment growth. Skill-levels increase the employment due to the “skill-biased technological change” hypothesis proposed by Griliches (1979). While education-levels increase employment due the fact that educated people are more mobile, and have a bigger probability of obtaining a job. From this we can propose two hypotheses concerning human capital:
H2: Skill-levels positively influence the effect that innovation has on employment growth
H3: Education positively moderates the effect that innovation has on employment
With the three hypotheses proposed in this section, the paper aims to answer the twofold research question stated in the introduction. Being able to answer this research question adds to existing literature by giving additional empirical evidence to the effect of innovation on employment growth, which has been researched for a long period however no definitive prove has been provided up to this point. Furthermore, even though the relationship between human capital and employment has been researched intensively, the moderating effect of it on the relationship between innovation and employment growth has not. Therefore being able to answer the second part of the research question will be an important contribution to existing literature. This because it has implications for the development of policies that use investments in innovation and human capital as means to increase employment and economic growth. In the next section an empirical model that makes it possible to test the hypotheses and to answer the research question will be presented.
3. Empirical model
12 The empirical model that is used in this paper is designed first of all, to test if innovation stimulates employment growth and second, to test if human capital positively moderates the effect that innovation has on employment growth. The empirical model in this paper uses the model developed by Di Cataldo and Rodriguez-Pose (2017) as a basis and is then extended by variables based on multiple other papers, which allows for assessing the employment performance of different EU regions based on their specific structural characteristics. The advantage of the model is that it is not overly complicated, and according to Di Cataldo and Rodriguez-Pose (2017) solves the endogeneity problem that could arise when analysing employment growth. The empirical model is the following one:
∆𝐿i,t= 𝛽0+ 𝛽1 𝐺𝑟𝑜𝑤𝑡ℎ 𝑓𝑎𝑐𝑡𝑜𝑟𝑠𝑖,𝑡+ 𝛽2∆ 𝑙𝑜𝑔 𝐺𝐷𝑃 𝑝/𝑐𝑖,𝑡+ 𝛽3𝐶𝑜𝑛𝑡𝑟𝑜𝑙𝑠𝑖,𝑡 + 𝜑𝑖+ 𝜏𝑡+ 𝜀𝑖,𝑡 (1)
The dependent variable, ∆𝐿i,t is the annual growth in regional employment where i stands for the specific region in year t. The annual employment growth is the difference between employment in year t compared to the employment in year t-1 expressed in percentage. This paper however will use a different proxy for the employment growth measure. Following Criscuolo et al. (2014) OECD paper, the model in this paper will use the Job Creation Rate (JCR) as a proxy for employment growth. The JCR is the Job Creation (JC) over average employment in a 2-year period:
𝐽𝐶𝑅𝑖,𝑡 = 𝐽𝐶𝑖,𝑡
0.5 ∗ (𝐸𝑖,𝑡+1 + 𝐸𝑖,𝑡) (2)
Where Ei,t stands for the amount of people employed in region i at year t. Using the JCR has
several advantages; first it has the advantage that growth rates will display a relatedly normal distribution; secondly, following from the Harrod-Domar and Solow model (Solow, 1956) investments take time to result into growth and therefor in employment growth. However, due to the fact that the JCR measures the average growth in jobs over a two year period, independent variables do not have to be lagged, which also has the advantage of having no reverse causality or endogeneity problems.
Di-Cataldo and Rodriguez-Pose (2017) distinguish between four growth determinants which influence employment growth: human capital, innovation capacity, transport infrastructure stock, and quality of government institutions. Due to the scope of this paper only human capital and innovation capacity will be considered as growth factors, transport infrastructure stock and quality of government institutions will be considered as control variables. The proxy variables that are used in the empirical model for these two growth factors are discussed next.
13 (2017) uses one measure for education however this paper distinguishes between three levels of education: (i) Primary and lower secondary, (ii) upper secondary and post-secondary non-tertiary education, and (iii) non-tertiary education. By distinguishing between the different levels of education, this paper can examine possible differences in the impact of the different levels on innovation and employment growth. The theory argues that education has a positive effect on employment, therefore it is expected that a higher educational attainment will have a bigger positive effect on employment growth.
To measure the quality of human capital, skill-levels of workers are used as a proxy. Most studies use education or PISA scores as a proxy for skill-levels (Vivarelli, 2014). However, education is used as a proxy for the quantity of human capital and unfortunately, PISA scores are not available for the entire period of analysis. Therefore, this paper uses the participation rate in education and training as a proxy for the quality of human capital. The participation rate in education and training describes the lifelong learning of workers, which entails all learning activities undertaken throughout life (after the end of initial education) with the aim of improving knowledge, skills and competences (Eurostat ESMS, 2017). Theory argues that skill-levels have a positive effect on employment growth, it is therefore expected that skills will have a positive effect on employment growth.
For the innovation capacity the proxy that Di-Cataldo and Rodriguez-Pose (2017) use is the amount of patent applications per million inhabitants in the region. From Griliches (1990) it follows that the variable of interest is measured with noise if one takes either R&D data or patent data. However, Griliches (1990) also argues that patents are an imperfect proxy for innovation performance, because not all inventions are patented and patenting propensity differs across sectors. Together with potential multicollinearity problems when patent data and R&D data are both added, the model in this paper will only use the growth rate in R&D expenditures as a proxy for innovations. In addition, this proxy will have the same normal distribution advantage as with the JCR, and by using R&D the model will be able to capture innovative activities that do not result into patents. Theory argues that innovation capacity has a positive influence on employment growth, therefore the higher the innovation capacity is in a region, the higher the probability that the coefficient is positive and bigger.
Perugini and Signorelli (2007) argue that when investigating employment, Okun’s law1 and economic cycles must be accounted for. In our model, this is done by including the logarithm of the change in GDP per capita for a region and time. Again, this will have the same normality advantages as with the JCR and the R&D growth rate. A positive correlation is expected between GDP per capita and employment change.
To control for possible substitution effects of investments, this paper will include physical capital and infrastructure. The proxy used for physical capital is the capital formation, capital formation describes the net capital accumulation in a specific region in a specific year and refers
1 Okun’s law pertains to the relationship between the changes in unemployment rate and the growth in GDP. It
14 to additions in the capital stock (Lequiler, 2006). King and Rebelo (1990) find that the presence of physical capital influences the economic growth of a country. Therefore, a positive relationship is expected between capital formation and employment growth. The model controls for two types of infrastructure. The first one is digital infrastructure, this to control for the potential spillover effects of knowledge through the use of internet across countries (Choi and Hoon Yi, 2009). Choi and Hoon Yi (2009) hypothesize that the increase in the use of the internet in a country has a positive impact on economic growth. It is therefore expected that internet has a positive effect on employment growth. The proxy used for the digital infrastructure is the access to internet, which is measured as the percentage of inhabitants in a specific region in a specific year that have internet access.
The second type of infrastructure variable is transport infrastructure. Transport infrastructure is included in the model because it is used to create employment, which is done through the construction, operation and maintenance of roads, highways and railways (OECD, 2002). Transport infrastructure is also capable of attracting private investment and generating jobs by increases in the demand for labour (Venables, et al., 2014; Di-cataldo and Rodriguez- Pose, 2017). Therefore, it is expected that transport infrastructure has a positive effect on employment growth. The proxy that will be used for transport infrastructure stock is the number of kilometres of roads per square kilometre. This proxy is chosen because it is a widespread measure of transport infrastructure density and because it gives a better representation of the transport network in a region than kilometres of motorways or railways (Di-Cataldo and Rodriguez-Pose, 2017).
Knack (1999), argues that the quality of government institutions has an effects on labour demand, wages and employment, as well as on the administrative capacity and the quality of legislation. The model in this paper therefore includes the quality of government institutions and does this by using two proxies. First, labour unions, the majority of research finds that unions increase wages of workers (Nickell, 1997; Faini, 1999; Koeniger et al., 2007). The proxy used for unions is the Trade Union Density of the OECD, which unfortunately is only available at the national level. The second proxy used for the quality of institutions is a proxy for if the region is regionally elected. It is argued that the basic political structure of a country is an important factor in understanding the scope for regional action (OECD, 2011). Regions may play a passive role, such as stages or implementers, however they may also play an active role, such as partners or independent policy makers. If a region is regionally elected, this will enhance the probability of the region playing a more active role. This in turn influences employment growth because it is found that active participation influences the return of innovation on employment (Perry and May, 2007). Combining these two proxies it is expected that the quality of government institutions has a positive impact on employment growth, the better the quality the higher employment growth is
15 markets, the scope for specialization, and create the required demand for innovation (Klasen and Nestmann, 2006; Shabani et al. 2012), all of which promote the creation and diffusion of new technologies and economic growth. Therefore, the expected relationship between population density and employment growth is positive. Lastly due to the fact that regional panel data is used, the model in this paper will accounted for regional and time fixed effects by incorporating region (𝜑𝑖) and time (𝜏𝑡) dummies. These region and time dummies control for any regional characteristics and time factors that could influence employment growth. The new estimated model therefore becomes:
𝐽𝐶𝑅𝑖,𝑡 = 𝛽0+ 𝛽1 𝑅&𝐷 𝐺𝑟𝑜𝑤𝑡ℎ𝑖,𝑡+ 𝛽2 𝐻𝑢𝑚𝑎𝑛 𝐶𝑎𝑝𝑖𝑡𝑎𝑙𝑖,𝑡 + 𝛽3 𝐺𝐷𝑃 𝑝/𝑐 𝐺𝑟𝑜𝑤𝑡ℎ 𝑖,𝑡
+ 𝛽4 𝑃ℎ𝑦𝑠𝑖𝑐𝑎𝑙 𝐶𝑎𝑝𝑖𝑡𝑎𝑙𝑖,𝑡+ 𝛽5 𝐼𝑛𝑓𝑟𝑎𝑠𝑡𝑟𝑢𝑐𝑡𝑢𝑟𝑒𝑖,𝑡+ 𝛽6 𝐼𝑛𝑠𝑡𝑖𝑡𝑢𝑡𝑖𝑜𝑛𝑠𝑖,𝑡
+ 𝛽7 𝑃𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛𝑖,𝑡+ 𝜑𝑖+ 𝜏𝑡+ 𝜀𝑖,𝑡 (3)
This model makes it possible to test the first hypothesis, Innovation has a positive effect on
employment growth. This hypothesis will be confirmed if 𝛽1 𝑅&𝐷 𝐺𝑟𝑜𝑤𝑡ℎ𝑖,𝑡 is positive and
significant. However, hypotheses two and three cannot be tested by this model. To be able to test hypotheses two; Skill-levels positively influence the effect innovation has on employment
growth, an interaction variable between skill-levels and R&D growth needs to be added to
model (3). Theory argues that skill-levels have a positive effect on the relationship between innovation and employment, it is therefore expected that the interaction variable skills * R&D growth will be positive.
To be able to test hypotheses three, Education positively moderates the effect innovation has
on employment, an interaction variable between education and R&D growth need to be added
to model (3). The same proxies for education as in model (2) will be used. Meaning that three interaction terms will be added to the model, one for each level of education. Theory argues that education levels have a positive effect on the relationship between innovation and employment. Therefore, it is expected that all three interaction variables will be positive. However as in model (2) by adding interaction terms of all three education levels it makes it possible to examine the difference in education levels. Therefore, it is expected the impact of the interaction variable becomes bigger the higher the populations education attainment in a region is. The final model of this paper therefore becomes:
𝐽𝐶𝑅𝑖,𝑡 = 𝛽0+ 𝛽1 𝑅&𝐷 𝐺𝑟𝑜𝑤𝑡ℎ𝑖,𝑡+ 𝛽2 𝐻𝑢𝑚𝑎𝑛 𝐶𝑎𝑝𝑖𝑡𝑎𝑙𝑖,𝑡 + 𝛽3 𝐺𝐷𝑃 𝑝/𝑐 𝐺𝑟𝑜𝑤𝑡ℎ 𝑖,𝑡
+ 𝛽4 𝑃ℎ𝑦𝑠𝑖𝑐𝑎𝑙 𝐶𝑎𝑝𝑖𝑡𝑎𝑙𝑖,𝑡+ 𝛽5 𝐼𝑛𝑓𝑟𝑎𝑠𝑡𝑟𝑢𝑐𝑡𝑢𝑟𝑒𝑖,𝑡+ 𝛽6 𝐼𝑛𝑠𝑡𝑖𝑡𝑢𝑡𝑖𝑜𝑛𝑠𝑖,𝑡
+ 𝛽7 𝑃𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛𝑖,𝑡+ 𝛽8 𝐻𝑢𝑚𝑎𝑛 𝐶𝑎𝑝𝑖𝑡𝑎𝑙𝑖,𝑡∗ 𝑅&𝐷 𝐺𝑟𝑜𝑤𝑡ℎ𝑖,𝑡 𝜑𝑖+ 𝜏𝑡
+ 𝜀𝑖,𝑡 (4)
This model makes it possible to test hypotheses two and three. Hypothesis two, Skill-levels
positively influence the effect innovation has on employment growth, will be confirmed if
𝛽8 𝑆𝑘𝑖𝑙𝑙𝑠𝑖,𝑡∗ 𝑅&𝐷 𝐺𝑟𝑜𝑤𝑡ℎ𝑖,𝑡 is positive and significant. Hypothesis three, Education positively moderates the effect innovation has on employment, will be confirmed if one of the three
16 Before the analysis can be performed, this paper includes different tests to ensure that the model fits the data. This paper will test for potential outliers in the data, possible multicollinearity, heteroscedastic errors, autocorrelation, and normality. Furthermore, due to the use of panel data this paper will test if a random or a fixed effects model fit the data better, as well as if time fixed effects should be used. Fortunately, due to the construction of the model, there is no potential for reverse causality, therefore this paper will not control for reverse causality. The test, their outcome, and the resulting action are described at the end of the next section, after the description of the data.
4. Data description
4.1 Data, sample and variables
The dataset analysed in this study consists of regional panel data for 289 regions belonging to 33 different countries covering the period 2000-2012 (a list of countries used in the analysis can be found in the appendix). The panel structure of the data is unbalanced but ensures minimum sample sizes for all time period. Raw data is mainly extracted from Eurostat and
OECD.
From a geographical point of view, the dataset includes data on regions from 28 EU countries and Former Yugoslav Republic of Macedonia, Iceland, Liechtenstein, and Norway. Regions from the EU represent the largest proportion in the sample, accounting for more than two thirds of the observations.
In order to analyse the effect of innovation on employment, employment is operationalised by the job creation rate using information from Eurostat (2017) at the NUTS2 level. Employment is the amount of people between the age of 25 and 64 who are being employed expressed per 1000 employees. The job creation rate uses the employment data to generate the average employment growth over a two year period for a specific region.
With respect to the independent variables, we operationalise regional innovation capacity using growth in R&D investments. R&D investment data is collected form Eurostat database and it is defined as amount of Euro’s per inhabitant invested in all sectors per region. R&D growth rate uses this data to calculate the growth of R&D expenditures in a specific year for a specific region.
17 (Eurostat, 2017). The ISCED 1997 categories for education attainment at 1-digit level can be found in the appendix. The ISCED 1997 distinguishes between three levels of education: (i) Less than, primary and lower secondary education, levels 0,1 and 2; (ii) Upper secondary and post-secondary non-tertiary education, levels 3 and 4; and (iii) Tertiary education, levels 5 and 6. Data for different levels of education are expressed as the percentage of the population between 25 and 64 that has attained that level. To measure the quality of human capital, this paper uses skills. Skills data is collected from Eurostat, it is measured as the participation rate in education and training four weeks prior to the survey and is expressed in percentage. These percentages are calculated as annual averages of quarterly EU Labour Force Survey data (EU-LFS).
As control variables, our analysis considers some drivers of regional productivity found in previous literature among economic development or industrial structure. An increase in GDP per capita expressed in euros per inhabitant is positively correlated with employment generation following Okun’s law. Data on GDP per capita is collected from Eurostat, after which the growth of GDP per capita in year t for region i is calculated.
To control for the substitution effect of investments between R&D, physical capital, infrastructure and human capital on job creation our analysis includes capital formation to capture physical capital. This variable captures the net additions to physical capital stock of region i in year t expressed in millions of Euro’s, and is collected from Eurostat. Additionally, in order to capture infrastructure stock, the analysis includes information of the access to internet, which is collected from Eurostat. The variable is described as the percentage of inhabitants in a specific region in a specific year that has internet access. Unfortunately, this data is only available from 2006 till 2012 and only for a small group of regions. General infrastructure data is collected from Eurostat, and describes the number of highway per thousand square kilometres a region has in a specific year.
As part of the control variables, our analysis would like to understand the regional endowments with regards to economies of scale and investment. Regarding the size of the region, our analysis includes population data from Eurostat. It is described as the number of inhabitants per square kilometre of region i in year t.
18 For the robustness analysis, data on three additional variables are collected; employment shares in the primary, industry, and high-tech sector. Data for these variables are collected from Eurostat. The variables are described as the percentage of workers between the age of 25 and 64 who are active in that sector. The share of employment in the primary sector entails the number of workers active in agriculture and fishing compared to overall employment. The share of employment in the industry sector entails the number of workers active in the industry sector except construction compared to overall employment. The share of employment in the high-tech sector entails the number of workers active in the high-high-technology manufacturing and knowledge-intensive high-technology services sectors compared to overall employment. The variable Access to internet has the least amount of observations this is due to the fact that this data is only available for the period 2006-2012 and not for all regions. To keep observations to a maximum, the variable Access to internet is not included. The rest of the variables have observations in all years from 2000 till 2012 but unfortunately not for all regions. For the main variables (Job creation rate, R&D growth, Capital formation, Primary education, Secondary education, Tertiary education and Skills) there are only a limited amount of observations missing because data was not available for every region in every year. Due to data availability, Union density is the same for every region in a specific country and specific year.
4.2 Descriptive statistics
Now that the data sources are described, this paper will look at the descriptive statistics. As defined in the empirical model section, this paper will test for the presence of outliers in the data. The test used in this paper to find outliers is multiplying the standard deviation by three and then adding this outcome to the mean to find the upper boundary, and by subtracting the outcome of the mean to find the lower boundary (Osborne, 2004). If the value of the observation is outside of these boundaries it is an outlier. There are outliers in Job creation rate, R&D growth, Capital formation, Primary education, Tertiary education, GDP growth, Employment share high-tech sector, employment share primary sector, Population density, Infrastructure, and Union density. After outlier detection the final sample is displayed in table 1.
Table 1, Descriptive statistics final sample
Variables Observations Mean Std. Dev Min Max Job creation rate (%) 3,572 0.00689 0.0307 -0.215 0.182 R&D growth (%) 2,659 6.8935 16.71 -59.82 96.09 Capital formation (million Euro) 3,075 8,163 7,053 123.8 41,423 Primary education (%) 3,528 29.30 15.16 3 81.90 Secondary education (%) 3,561 46.99 15.30 6.900 80.30 Tertiary education (%) 3,556 23.16 8.694 3.700 50.80 Skills (%) 3,441 10.31 7.789 0.500 36.10 GDP growth (%) 3,201 3.332 5.902 -16.67 24.82 Population density (inhabitants per km2) 3,481 247.8 377.9 2.800 3,064
Infrastructure (km per thousand km2) 2,292 30.00 24.07 1 113
19 Besides the elimination of outliers, this paper also takes the natural logarithm of certain values to deal with extreme values and to generate close to normal distributions for variables which have only positive values. Variables of which the natural logarithm is taken are: Capital formation, Primary education, Secondary education, Tertiary education, Skills, Population density, and Infrastructure.
This paper looks at 289 regions in 33 countries for a time period of 13 years. What can be seen from table 1, is that most variables have around the 3000 observations, with some variables having 3500 observations.
20
Table 2, Descriptive statistics Southern Europe
Variables Observations Mean Std. Dev. Min Max Job creation rate (%) 780 0.00886 0.0340 -0.130 0.132 R&D growth (%) 628 6.810 17.11 -59.82 92.31 Capital formation (million Euro) 533 7,906 8,440 423.5 33,666 Primary education (%) 751 52.13 10.22 22.50 81.90 Secondary education (%) 783 28.49 10.63 7.300 49.60 Tertiary education (%) 783 18.12 8.150 3.700 46
Skills (%) 737 5.962 3.361 0.500 15.80
GDP growth (%) 732 2.230 4.257 -13.09 17.92
Population density (inhabitants per km2) 753 188.0 246.9 22 1,327
Infrastructure (km per thousand km2) 468 28.63 17.03 3 98
Union (%) 780 24.83 7.796 14.30 36.90
Employment share primary sector (%) 752 8.833 7.321 0.195 36.69 Employment share industry sector (%) 759 16.65 7.080 2.925 33.76 Employment share high-tech sector (%) 632 2.688 1.536 0.500 9
Table 3, Descriptive statistics Northern Europe
Variables Observations Mean Std. Dev. Min Max Job creation rate (%) 772 0.00611 0.0263 -0.0900 0.182 R&D growth (%) 496 3.189 15.85 -56.49 82.35 Capital formation (million Euro) 753 9,053 5,913 123.8 38,188 Primary education (%) 787 25.71 7.201 9.900 50 Secondary education (%) 787 43.39 6.348 22 61.20 Tertiary education (%) 784 30.80 6.140 15.90 50.80
Skills (%) 779 21.00 5.341 4.500 35.80
GDP growth (%) 674 1.842 6.658 -15.73 21.74
Population density (inhabitants per km2) 790 299.8 511.1 2.800 3,064
Infrastructure (km per thousand km2) 196 13.16 12.61 1 61
Union (%) 756 45.99 20.46 25.80 92.50
Employment share primary sector (%) 588 2.870 2.018 0.219 13.22 Employment share industry sector (%) 772 15.98 4.625 2.897 30.19 Employment share high-tech sector (%) 736 4.813 1.959 1.400 12.80
Table 4, Descriptive statistics Eastern Europe
Variables Observations Mean Std. Dev. Min Max Job creation rate (%) 705 0.00318 0.0370 -0.165 0.117 R&D growth (%) 622 12.821 22.65 -59.5 87.5 Capital formation (million Euro) 714 3,256 2,728 150.6 23,119 Primary education (%) 714 18.61 8.064 3.300 42.40 Secondary education (%) 714 64.38 8.423 34.90 80.30 Tertiary education (%) 714 17.02 7.009 6.500 42.40
Skills (%) 672 4.387 3.263 0.500 18.20
GDP growth (%) 652 7.431 8.262 -16.67 24.82
Population density (inhabitants per km2) 696 174.7 359.4 30.50 2,564
Infrastructure (km per thousand km2) 468 10.78 11.12 1 56
Union (%) 495 18.10 5.689 6.100 44.70
Employment share primary sector (%) 714 12.22 10.58 0.118 52.10 Employment share industry sector (%) 714 25.49 6.465 9.496 39.73 Employment share high-tech sector (%) 624 3.386 1.827 0.600 8.600
Table 5, Descriptive statistics Western Europe
Variables Observations Mean Std. Dev. Min Max Job creation rate (%) 1,315 0.00815 0.0268 -0.215 0.176 R&D growth (%) 913 4.9248 9.57 -44.49 96.09 Capital formation (million Euro) 1,075 10,927 7,309 1,047 41,423 Primary education (%) 1,276 24.07 9.998 3 68.60 Secondary education (%) 1,277 50.83 10.84 6.900 71.40 Tertiary education (%) 1,275 25.00 6.584 11 49.50
Skills (%) 1,253 9.384 5.891 1.700 36.10
GDP growth (%) 1,143 2.580 2.952 -11.37 20.14 Population density (inhabitants per km2) 1,242 292.0 341.0 2.900 2,358
Infrastructure (km per thousand km2) 1,160 41.15 25.08 1 113
Union (%) 1,352 21.92 13.51 7.600 56.30
21
4.3 Preliminary tests
The descriptive statistics of both the total sample and the four European macro-region samples showed some interesting differences. This section will therefore perform preliminary tests to insure that the right form of the model is used for analysis.
The first test that was performed was to control or outliers, which is done in the previous section. The second test this paper performs is for multicollinearity. The results of this test can be found in the appendix. The cut-off point in this paper is 0.5, values above 0,5 and below -0,5 will indicate that there is a case of multicollinearity. The results show that the variables primary- and secondary-education show signs of multicollinearity. Also, tertiary education and skills show signs of multicollinearity. The rest of the variables have values that are significant below the cut-off point. Besides a correlation matrix this paper also looked at the Variance Inflation Factor (VIF), a VIF value above 10 would indicate multicollinearity presence. Given that, the value of the test is 9.84, the test gives indication of weaker multicollinearity problems. However, this test is performed for the regression where all variables are added, even the variables that show signs of multicollinearity. The VIF will therefore be lower for the other regressions, eliminating any problems of multicollinearity.
Two tests are performed to determine the type of regression that fits the data best. First the Hausman test is performed, this test shows that there are systematic differences between coefficients, therefore a random effects model should be used. The second test is the Breusch and Pagan Lagrangian multiplier test for random effects, which shows that the there are no random effects. The results of these tests combined show that there is no evidence of significant differences across regions which would suggests a simple OLS regression. However due to the fact that panel data is used there could still be time-fixed effects. Also by looking at the descriptive statistics it could be possible that there are European macro-region- or country-fixed effects. Tests show that there are time-fixed effects and that both European macro-region- and country-fixed effects are present. This means that different countries, European macro-regions and years influence the job creation rate. Therefore two models will be used, one with country dummies to control for the country-fixed effects, and one with European macro-region dummies to control for the European macro-region-fixed effects. However, because this papers initial level of analysis was with regional-fixed effect, country-fixed effects will be used as the main regression. This because, country-fixed effects are more similar to regional-fixed effects than European macro-region-fixed effects.
22 the kernel density estimation shows that the residuals have a relatively normal distribution. Also, the Central Limit Theory argues that the distributions of the sum of a large number of variables – more than 50 - will be approximately normal, regardless of the underlying distribution (Hill, 2007). The analysis performed has more than 50 observation, therefore it is assumed that the residuals of the regression are normally distributed. The residuals do however show signs of heteroscedasticity, the results of the Breusch-Pagan / Cook-Weisberg test for heteroscedasticity support the suspicion of heteroscedasticity. To control for heteroscedasticity this paper uses robust standard errors to minimize the effect of heteroscedasticity. Lastly this paper tests for autocorrelation, to test for autocorrelation this paper uses the Wooldridge test for autocorrelation in panel data. The Woolridge test shows that there is autocorrelation at the 10% level, to control for this clustered robust standard errors are used. After the preliminary tests, the final dataset will comprise of approximately 1390 observations.
5. Analysis
The empirical model in part three in combination with the preliminary tests in the previous section have resulted in the final model used for the analysis in this paper. This section will use this final model to test the three hypotheses proposed in section 2, after which a robustness analysis is performed to strengthen and confirm the results of the main analysis
5.1 Results
The results of the main model of this paper – with country-fixed effects - are presented in table 6. The results of the model with European macro-region-fixed effects are presented in table 7. Regression 1-5 show the effect of innovation measured in terms of R&D growth on the job creation rate and the direct effect of human capital on the job creation rate measured by different education levels and skills. Regression 6-10 show the interaction variables of the different education levels and skills with R&D growth and their effects on the job creation rate.
23 Table 6, Main regression with Country dummies: the effect of innovation and education on employment
24 Table 7, Main regression with European macro-region dummies: the effect of innovation and education on employment with Country dummies
25 Even though there are some differences between table 6 and 7, the analyses show some interesting results. Looking at R&D growth and its effect on the Job Creation Rate (JCR) - average growth in employment over two years – in the regressions where R&D is not combined with human capital; it is found that R&D growth as an overall positive effect on JCR. This is in line with a lot of research, Vivarelli (2014) found in his meta-study an overall positive and significant effect of innovation on employment growth. Also, Bogliacino (2014) found that R&D has a positive effect on employment. However, when examining tables 6 and 7 in more detail it is found that only in table 7, R&D growth has a significant and positive direct effect on JCR if it is not combined with human capital. In table 6 R&D growth is only significant when it is combined with skill-levels. If all other variables stay the same, a one percentage point increase in R&D growth in model 7 will increase JCR by 0.000140 units.
When looking at the regressions where human capital is combined with R&D growth, both tables offer similar results. Namely R&D growth has only a positive and significant effect on employment growth when it is combined with skill-levels. This is despite the negative effect of the interaction term skills and R&D growth, the negative effect of this interaction term does not outweigh the positive direct effect of R&D growth on JCR. When looking at the impact of R&D growth in table 6 model 10 - when all the other variables stay the same - if R&D growth increases with one percentage point JCR increases by 0.000326 units while in table 7 this is 0.000258 units. The fact that if human capital is combined with innovation, innovation is only positive and significant when it is combined with skill-levels is also found by Evangelista and Savona (2003). They find that the positive impact innovation has on employment is limited to high skilled labour, the explanation that they give is the skill-biasedness of innovations. Innovations now a days are mostly ICT related, which require skills to operate, they therefore argue that only high skilled workers can take advantage of innovations. Also, Hanushek and Woessmann (2012) look at the difference between education and skills, where skills are measured by PISA scores; they find that skill-levels are more important for economic growth than education; confirming again the findings of table 6 and 7.
26 capital to have lower spillover effects. It could therefore be possible that by taking European macro-regions the spillover effects of capital formation on employment growth cannot be measured. Even though the effect of capital formation on the JCR differs per regression, keeping all other variables the same, on average an increase of one percentage point of the capital formation in table 6 will increase the JCR by 0.00003 units.
For the human capital measures two similar results are found in table 6 and 7. First of all, in regression two and seven of both tables it can be seen that primary education has a significant and negative effect on the JCR. On average the negative direct effect of a one percentage point increase of primary education on the JCR, given that other variables remain constant, in table 6 is -0.0000610 units and in table 7 is -0.0000383 units. Secondly, of the four measures of human capital in both tables only tertiary education has a positive and significant direct effect on the JCR. The direct effect of one percentage point increase in tertiary education on the JCR in table 6 is 0.000078 units while in table 7 0.000048 units. These findings are in alignment with the findings of Morrison et al. (2001); they find that the demand for labour was in favour of highly educated people, while reducing the demand for workers without a college degree. Also, Winters (2013) finds that higher levels of education result into higher employment and argues that this is due to increases in labour market participation. Di Cataldo and Rodriguez-Pose (2017) explain these findings by stating that higher education directly increases the probability for a job by having more available jobs to work at. These results therefore indicate that higher levels of education directly contribute to the increase of employment, while low levels of education negatively contribute to employment creation.
27 These findings show that hypotheses two and three concerning the moderating effect education and skills have on the relationship between innovation and employment growth cannot be confirmed. For education, only the direct effect of tertiary education is significant and positive, the interaction effects of all education levels are insignificant. For skills, no direct effect is significant however the interaction variable is. The interaction variable for skills is significant at the one percent level in table 6 and the twenty percent level in table 7. Unfortunately, the interaction variable is negative, which is the opposite of what would have been expected. This could be due to the proxy that is used for skills – lifelong learning of workers – that does not measure the actual skill-level of the worker. However due to these findings hypotheses two and three cannot be confirmed.
When looking at the control variables, first of all two variables are dropped in table 6 due to the use of a Pooled OLS regression with country dummies; this because the variables Regionally
Elected and Union density are the same for every region in a country and therefore are already
captured by the country dummies (To keep the same construction these variables are also dropped in table 7). Concerning the other control variables it can be seen that Okun’s law is confirmed, GDP growth is significant and positive in every regression. An increase in GDP growth today will lead to an increase in the average employment over the next two years. This confirms the argument made by Perugini and Signorelli (2007) that when investigating employment, Okun’s law and economic cycles must be accounted for. Besides GDP growth, the variable Crisis is another control variable that is highly significant, the variable Crisis is a dummy variable that is one during the financial crisis that started in 2007. It can be seen that the impact of this variable is negative for all regressions and highly significant for most. Which is also found by Campello et al. (2010), they found that the financial crisis had a twofold negative effect on employment growth. First of all, the crisis resulted into plans of deeper cuts in employment, and second, the crisis also postponed investments that would have resulted into employment growth.
The other control variables seem to be not significant; infrastructure has a negative impact on the average employment growth in the next two year in both tables but is not significant. This is opposite from what the OECD (2002) finds, they find that infrastructure has a positive effect on employment growth. This however can be due to fact that the proxy used in this paper did not include all forms of infrastructure, it only included the number of kilometres of road per square kilometre. Also, the control variable Population density is not significant, but interestingly table 6 and 7 have contrary results; where it is positive in table 7 and negative in table 6. Shabani et al. (2012) argue that population has a positive effect due to agglomeration effects. Even though the variables are not significant, the potential impact is different when country-fixed effects are used compared to when European macro-region-fixed effects are used, this can be due to different agglomeration aspects.
28 where Eastern Europe is used as the base category, it is found that both Western- and Southern Europe are significant.
5.2 Robustness
To add robustness to the results, this paper disaggregates the JCR into employment shares of three different sectors: primary, industry, and high-tech. This is because the impact of innovation on employment growth can significantly differ between sectors (OECD, 2015). Following the same reasoning, human capital may also have a significant different impact between sectors. Therefore, knowing how human capital and innovation contribute to the employment shares of the different sectors can be of importance for economic growth and potential political policies.
Because the dependent variable is the employment share of a sector in region i year t several independent variables are lagged. This is because it takes time for inputs to result in employment growth. Furthermore, tests show that due to the disaggregation of the JCR in to three different sectors, the model does have regional fixed effects. However, to be able to compare the results to the main regression in table 6, the robustness model will also be a Pooled OLS with country and year dummies (for completeness the regressions with regional fixed effects can be found in the appendix). For simplicity, the results of the robustness analysis will only be compared with table 6 and not with table 7.
If we compare the robustness regressions with the findings of the main regression, it can be seen that table 9, which illustrates the employment shares of the industry sector, confirm the conclusion of our main analysis that innovation is only significant when it is combined with skills. If all other variables stay the same, a one percent increase in R&D growth will increase the employment share of the industry sector by 0.0392 units. Again, the interaction variable skills and R&D growth is negative but because the interpretation of the interaction variable is according to a linear-log function the total effect of innovation on the share of employment in the industry sector is positive. For both the primary industry and the high-tech industry it is found that innovation, expressed in R&D growth, has no positive and significant effect on the employment shares even if innovation is combined with different measures of human capital. This suggests that the positive effect of innovation on employment is limited to the industry sector.
29
30
31
32 Looking at the interaction terms of table 7-9, there are some differences compared to the main regression analysis in table 6. For the primary sector it is found that combining R&D with primary education is significant and increases the employment share. Even though the direct effect of R&D stays negative it is interesting to see that combining innovation with primary education contributes to the growth in the employment share of the primary sector. The results of the industry sector best resemble the results of the main analysis; only the interaction term of skills and R&D is significant, and again negative. But as already said the total effect of innovation on the employment share in the industry sector is positive. For the employment share of the high-tech sector it is found that none of the interaction terms are significant.
Furthermore, when looking at the robustness regressions it can be seen that the human capital variables have increased in their significance. While in the main model (table 6) capital formation, GDP growth and crisis were capturing all the significance, in the robustness model the significance of the model is distributed evenly among the different explanatory variables. These models allow disentangling the effect of education and R&D on Employment growth. In general, the robustness model shows that education and skills seem to play an important role. For the primary sector it is found that the direct effect of primary education is significant and positive also when the interaction term is added. While higher levels of education and skills seem to have an overall significant and negative direct effect on the employment shares of the primary sector.
For the industry sector, it is found that primary- and secondary education have an overall significant and positive direct impact. The direct effect of tertiary education and skills on the employment share in the industry sector are significant but negative, even when the interaction term is added. When looking at the high-tech sector, it is found that the direct effect of tertiary education and skills is significant and positive for both the regression with and without the interaction term. The direct effect of secondary education seems to be not significant both with and without the interaction term. Primary education does have a significant impact on the employment share in the high-tech sector, but this impact is negative and stays negative when the interaction term is added.
33 Combining the results from table 6 with 8-10 we can argue that there is evidence that innovation has a positive effect on employment growth but that this is only true for the industry sector, for the other sectors innovation is not significant. The direct effect human capital on employment growth, interestingly seem to depend on the sector. Concerning the moderating effect of human capital, education and skills, on the relationship between innovation and employment, the interaction terms are not significant or negative. However, when innovation is combined with skills the overall effect of innovation on employment growth becomes significant and positive, confirming the findings in the main analysis.
6. Conclusion
In 2010, the European Commission announced its strategy for 2020. For this strategy the EU had set five ambitious objectives, among which an employment rate of 75% for workers between the ages of 20 and 64, as well as having a competing third level education. One of the main initiatives to reach these ambitious goals is innovation, which has been seen as the main driver of generating jobs and increasing competitiveness. However, from theory the relationship between innovation, human capital and employment growth is ambiguous. This paper therefore contributes to this discussion by looking at the relationship between innovation, human capital and employment growth. But also by justifying the role that investments in innovation and education may have in increasing growth and competitiveness. To test this relationship this paper tries to answer the following twofold research question: First: Does innovation positively influence employment growth on a regional level? Second: Does human capital positively influence the relationship between innovation and employment growth?
This paper looked at the employment growth of 289 regions located in Europe from 2000 until 2012 using two Pooled OLS models, one with country-fixed effects and one with European macro-region-fixed effects. To measure employment growth, the job creation rate of a region was taken, this is an innovative measure of employment growth developed by the OECD which gives the average employment growth over a two year period. While most studies use patents as a measure of innovation, this paper took R&D growth as a measure of innovativeness to also include innovations that do not lead to patents. More importantly, the novelty of this paper comes from the measurements that are taken for human capital, in this study human capital is disaggregated into four different proxies; primary education, secondary education, tertiary education, and skill-levels. This made it possible to perform an analysis where a distinction was made between the effect of quality and quantity of human capital on employment. Furthermore, in the robustness section a distinction is made between employment shares of three different sectors, which gives the opportunity to see how innovation and human capital influences different sectors.
