The rise of the self-employment level in the European Union: A panel data analysis

The rise of the

self-employment level in the

European Union

A panel data analysis

MASTER THESIS ECONOMICS & GOVERNANCE

Rachelle Zieleman

Supervisor Prof. Dr. Olaf van Vliet

March 15th 2021 Word count: 9698

Rachelle Zieleman

Foreword

After studying at the University of Leiden for a number of years, the time has come to hand in my master thesis and thereby complete my master’s degree of Economics & Governance. While following this master programme, I studied policy issues at the intersection between economics and public administration, I developed the skills and I have acquired knowledge on how to use data and perform a statistical analysis. When deciding upon the subject for this research, I decided to formulate an approach that would incorporate elements of the two courses that I found the most enjoyable and interesting during my degree course – Political Economy in International Perspective and Economics of Regulation. Therefore, I chose the subject of self-employment. I have always found this an interesting construction in the Netherlands and I believe this has become more and more important on today’s and tomorrow’s labour market.

The level of self-employment in Europe is the central theme of this research. Although initially it was not a subject that I was particularly familiar with from a theoretical or academic standpoint, throughout the course of my research I have managed to develop a much deeper understanding of the subject and how it can be influenced. In this

research I studied how technological development, the level of education and economic development can influence the level of self-employment in 21 European countries. This subject is still topical and relevant, this is evident from the recently issued report by the “Commissie Regulering van Werk” (Borstlap, 2020).

Graduating has been an educational process in which I was able to apply knowledge and skills that I gained during my studies in this research. I have also developed new skills. Learning how to work with the statistical software of Stata was not an easy assignment, but proved to be very instructive and challenging. I did not want to shy away from this challenge and this process proved to be most enlightening.

I am grateful to all of my lecturers at Leiden University for broadening my knowledge of economics and governance issues in a way that was both enjoyable and stimulated critical thought, and especially to my thesis supervisor Dr. Olaf van Vliet for his guidance, patience and advice throughout the whole research project.

I hope you enjoy reading my research.

Contents p. 6

Section I | Introduction p. 8

Section II | Theoretical framework p. 11

2.1 History of the labour market p. 11

2.2 The level of self-employment p. 14

2.3 Indicators of self-employment p. 15

2.3.1 Indicator 1: Technological development p. 17

2.3.2 Indicator 2: The level of education p. 18

2.3.3 Indicator 3: Economic development p. 19

Section III | Methodology p. 21

3.1 Panel data p. 21

3.2 Random effects model p. 22

3.3 Heteroskedasticity and serial correlation p. 23

3.4 Data collection p. 24

Section IV | Descriptive evidence p. 26

4.1 Self-employment rate p. 26

4.2 Indicator 1: Technological development p. 29

4.3 Indicator 2: The level of education p. 31

4.4 Indicator 3: Economic development p. 32

Section VI | Conclusion and discussion p. 39

6.1 Conclusion p. 39

6.2 Discussion p. 40

Section I | Introduction

The flexibility of the labour market has raised a lot of scholarly interest in the recent years. Given the concerns of analysts and policy-makers, attention is mainly focused on political and economic consequences, looking at the labour market and the

unemployment level.

In the European Union the group of self-employed people is relatively big, but has received less attention. It is quite common in research to look at the entire labour

market, and not to control specifically for self-employed people (Nowotny, Mooslechner & Ritzberger-Grunwald, 2009). This assumes that flexibility is not important. But this assumption is wrong. The unexplored topic of self-employment is in need of more research. The number of self-employed people has increased dramatically in the last decades. This flexible labour market, as with any form of labour, “is shaped” by technological development, the level of education and economic development in the world. The composition of the labour market has changed and labour policies need to adapt to this new situation. They no longer fit the new situation with many forms of labour.

It is important to understand the economic well-being of self-employment to use its power and adapt to the needs. What will be the explanation for the rise of

self-employment in the European Union? This research seeks to find out what accounts for the increase of the self-employment rate. What moves people to become self-employed? What part do technological developments play in this decision? What will the

distribution in the labour market be in a decade? How will the distribution of the labour market change due to technological development? Do we feel that we can make enough money for ourselves due to new technologies? All of these questions should be relevant to domestic policymakers. An evidence-based approach is necessary to answer these questions, in order to respond to the changing labour market in the future. This

provides the motivation to this research, which seeks to answer the following research question:

Why did the number of self-employed people increase between 1985-2017?

This research will be conducted as a comparative study. The goal of this prospective research is to estimate the effect of technological developments, the increase in education and economic development in the period from 1985 until 2017 on

self-employment in the European Union. The method used to try and ascertain the effects of technological development, the education level and economic development on the level of self-employment is to use pooled cross-sectional time-series data for 21 European countries for the period of 1985-2017. Using this data, I performed a panel data regression with random effects.

One of the most pressing social issues these days is solving the problems of

unemployment, where technological development plays a big part. We are currently in a situation which requires a lot of technological resources in everyday life. The rise of concern for self-employment is applicable in both the economics and sociology

literature. My findings are therefore relevant for both students of Public Administration, but also to politicians and policy-makers because nowadays the focus on regulation around self-employment is heavily underexposed. Current research mainly looks at the effect technological progress has on employment and income levels, but fails to look at the huge increase in the various forms of non-standard employment, while this can have a relevant effect. According to various research, the level of self-employed people in Europe will keep growing and the quality of work will keep improving, due to

technological developments (OECD1, 2020). But, there is a lack of empirical evidence on the types of employment and all the outcomes a change in this field can have.

I argue in this research that the self-employment level in Europe is influenced by various factors in the world. In particular technological development, the increase in tertiary education and economic development. I believe that these factors accelerate the self-employment rate. I support this claim by first, descriptive evidence and second, a quantitative analysis. The structure of the research is as follows. In section II I provide a framework for understanding the increase in self-employment in the European Union. The literature on the labour market is reviewed and introduces mechanisms that affect

level and economic development have a positive effect on the level of self-employment. In section III the methodology of the research will be explained, which includes looking at the research design and data collection and analysis. Section IV provides descriptive evidence that three different indicators can have an influence on the level of

self-employment. Section V provides a panel data analysis and the results. Subsequently, the last section, section VI, draws conclusions and speculates about the future of the

Section II | Theoretical framework

In this section, I provide a framework for understanding the increase in

self-employment in Europe over the last couple of decades. I start by reviewing the labour market as a whole and then zoom in on the self-employment level and the indicators of self-employment.

2.1. History of the labour market

After WWII the labour market has changed significantly. Not only the type of work changed, but also the workforce, the institutional rules and the structure of wages. The society started to recover with rapid industrialization and the emergence of the welfare state. The leading sectors of employment in the 1960s used to be agriculture and

industry. But the next decades the employment in this sector dropped by half (Iversen & Crusack, 2000). Governments responded to this by promoting labour in the tertiary service sector and welfare-state spending like issuing social security income transfers and government consumption. This was internally driven by employment losses in the traditional sector (Iversen & Crusack, 2000). The emergence of the welfare state is important to understand how the welfare state is likely to change in the future.

With the shift from the primary sector to the tertiary sector as the main source of employment, another determining change occurred. Somewhere people moved from working for their families by making things from start to finish, to working as paid employees and becoming more specialized in their skills. With this change productivity of labour went up, standards of living rose and people had access to more goods thanks to mass production. The service economy began to shape with the creation of a system for social security, labour law and tax rules. As a result of this change in the post-war period, labour unions began to grow in membership and achieved an institutional role (Ebbinghaus & Visser, 1998). These organizations of workers sought to improve wages and working conditions through collective action, strikes, and negotiations. The

organizations negotiated agreements for engaged employers and employees that applied to certain sectors. Especially public-sector employees formed the base of this

movement. This unionization increased until the early 1980s in most OECD countries (Kwon & Pontussen, 2010).

But nowadays we observe a decline in union membership due to social, economic and political changes (Ebbinghaus& Visser, 1998). At the moment most unions are for public sector jobs, like teachers. So what do jobs in this new society look like? There have been a lot of changes from the postwar period till today. A major change has happened in the demography. Life expectancy has increased in all industrial countries, while birth rates have declined. As a result, the working population has decreased. Other changes like changes in family structure and the structure of jobs also have a considerable impact on the working population. If present policies continue like this, spending on pensions and health care will double on some countries, which do not have the income to make up for it. Another factor that has a great impact on the world is technological development. It promotes globalization all around the world. Also, nations are becoming wealthier and the secondary education rate has increased in the last decades (World Bank, 2020).

Alongside these long-term trends the financial crisis started in the years after 2007. The financial crisis led to enormous political and economic complications for the European Union. This crisis highlighted the issue of high levels of public debt and the potential unsustainability of running budget deficits to fund public expenditures (Neaime, 2015). The recession led to a period of economic stagnation and widespread unemployment, which is one of the main challenges for the changing European labour market. As illustrated in figure 1 below, in all European states, the unemployment level rose tremendously after 2008 (OECD4, 2020). In some countries, the unemployment levels rose very sharp. In Greece the unemployment rate rose from 7.8% in 2008 to 27.0% in 2013 and the youth employment rate in Spain rose from 18.1% to 55.5% over the same period (OECD4, 2020). These numbers are alarming.

Figure 1: The unemployment rate as % of the labour force, 2008-2020.

Source: OECD4, Unemployment rate (2020)

The financial crisis put extra emphasis on the importance of the unemployment issue. Particularly in the European states burdened by national debt. High unemployment levels are dreadful to a country’s economy. When people are not working, the country’s tax income is lower and the welfare expenditures increase. This can “bring about a sharp drop in living standards, a weakening of social life, and marginalization with respect to those in work – effects which can become cumulative and lead to a situation of intense poverty and, at the least, of social rupture” (Gallie and Paugam, 2000). Unemployment is a waste of human capital. Therefore, one of the key issues for policymakers is unemployment, and how to boost the economy to achieve the best possible labour market. The states in the European Union must adapt its policy to the flexibility that currently prevails on the labour market.

2.2. The level of self-employment

But despite the economic crisis, the self-employment level kept rising in the European Union. I use the Netherlands as an example. The employment relationship in the Netherlands used to be characterized by wage-dependent work under the explicit authority of the employer. People are employed when they work under an authority relationship for an employer, receive wages and perform personal labour for the

employer (Commissie Borstlap, 2020). But, the old employment contracts are no longer functional and the Dutch legislator decided that so much legal uncertainty had arisen that it was time to renew the employment contract. The first time the term

“self-employed” was introduced was in 1967 and in the eighties the term was first introduced by the Dutch tax authorities (Wilthagen, 2019). Ever since the form of labour was

introduced, many organizations have found it difficult to give this form of work a place in their policy and regulations. By then self-employed people were mainly construction workers doing temporary jobs, but it was still difficult. The social security systems were designed for people who are employed, so that they automatically contribute to all kinds of taxes. Insurance agencies and the tax authorities were not set up for the arrival of self-employed persons. That was going to be a problem, not only in the Netherlands but in all of Europe.

The Netherlands is an entrepreneurial country in which everyone actively participates in the realization of their own and common prosperity. Everyone receives basic

protection services (Commissie Borstlap, 2020). From an international perspective, the level of self-employment in the Dutch labour force is very high, which leads to

fragmentation of production and a higher volatility of the economy (Commissie Borstlap, 2020). But these social services generally do not apply to the self-employed. Self-employed people are expected to be able to negotiate their own terms and not need the protection from the government. This creates a clear dividing line between the self-employed and employees (Commissie Borstlap, 2020).

But, where the governmental protection disappears, the choice to become self-employed is encouraged in the Netherlands. There are a considerable number of tax benefits (start-up benefits) for entrepreneurs. For example the self-employed tax

deduction for self-employed persons, which is considered as compensation for the high inflation and unfavourable income development (Commissie Borstlap, 2020). This tax-reduction is even higher in the first five years when you are starting to be

self-employed. Another tax incentive to become self-employed is the SME profit exemption. This reduces the taxable profit which reduces the tax to pay (Belastingdienst, 2020). Other tax benefits are the indemnity from payroll tax, co-working allowance for your partner and suspension relief for self-employed persons who quit (zzp-Nederland, 2020). These arrangements are clearly attractive to a lot of people, because between 2003 and 2019 the number of self-employed grew by almost 500,000 (Commissie Borstlap, 2020). Tax deduction for self-employed people is a very interesting factor to include in an analysis. Unfortunately, taxes are imposed nationally, and therefore they are different for every state in the European Union. Therefore, in this research, tax deductions and other benefits are not included. This, however, does not make this point any less relevant.

We can conclude that the labour market in Europe has changed in the last decades. Worker mobility has increased and more and more people do not have a standard employment contract. The institutions have not been able to keep up with changes in the labour market and are discouraging mobility. There is an important task for labour market economists to think about what institutions should look like, in such a way that they are better suited to the current and future labour market (van der Klaauw, 2013).

2.3 Indicators of self-employment

The awareness of the self-employment level in the European Union is growing. But how is the self-employment rate, as defined by the OECD, influenced on the labour market? The self-employment rate in this research is defined according to the definition of OECD statistics. The OECD considers that the self-employment rate is “the employment of employers, workers who work for themselves, members of producers' co-operatives, and unpaid family workers” (OECD1, 2020). The last category is particularly important in the agricultural sector. Unpaid family workers “lack a formal contract to receive a fixed amount of income at regular intervals, but share the income generated by the

employment rate. Most researchers who study the level of self-employment remove the agricultural sector from their data, or look at the sector separately (Paul & Thomas, 2020). I deliberately chose not to remove this sector, especially because I look at nations and not at specific regions. It is important to factor in the agricultural sector, because nowadays, not everyone needs to work in agriculture to provide for food. I believe everyone currently has the choice to work in agriculture or to get an office job. The OECD also states that self-employment “may be seen as a survival strategy for those who cannot find any other means of earning an income or as evidence of

entrepreneurial spirit and a desire to be one’s own boss” (OECD1, 2020). This is exactly how this research looks at self-employment. I want to look into the reason that creates this strategy, entrepreneurial spirit and desire.

The OECD defines the labour force as “employed people those are aged 15 or over who report that they have worked in gainful employment for at least one hour in the

previous week or who had a job but were absent from work during the reference week” (OECD1, 2020). This is, for example, also the case in the Netherlands. At the last quarter of 2017 the employed labour force in the Netherlands consisted of 8,651 million people, and the unemployed labour force 0,391 million people (CBS, 2020). The gross

employment rate was therefore 70.1%, which is not bad at all. But if you imagine that 16.7% of the labour force in the Netherlands in 2017 was self-employed, this is a relatively big part (OECD1, 2020).

In this research, I look at three possible indicators which can take part in explaining the rise in the self-employment rate. The first indicator is technological development. I believe the rise in technological development tends to increase the self-employment rate. The second indicator is the level of education. I expect that the rise in the education level contributes to the rise of the self-employment rate. The third and last indicator in this research is economic development. I believe that economic development also fuels the self-employment rate.

2.3.1. Indicator 1: Technological development

The first indicator in this research that is linked to the self-employment level in Europe is technological development in the world. I do not think I still need to explain that there has been a massive increase in technological development in the last 50 years.

Nowadays, a smartphone and a stable internet connection almost seem like the basic necessities of life. The level of communication and the exchange of information in the world has never been higher. This is all due to the invention of computer chips and information technology, but also through artificial intelligence and biotechnologies. One single invention or development can influence the whole world. Technology is not limited by borders anymore. Technological developments contribute to the increase of globalization in the world. Globalization is the increasing process of economic, social and political international integration. This is also generated by technological

development, so the effects of globalization and technological developments are difficult to distinguish.

Technological development has taken the form of a decline in the costs of computerizing routine tasks, such as auditing, administrative work and monitoring activities (Author, Dorn & Hanson, 2015). Hereby, technological development is replacing the workers performing these tasks, which is often the goal of technological development. For

example, the computer is designed to make calculations faster, and with fewer mistakes, than people do. Another sector where the consequences of technological development are clearly visible is the financial sector. The payment system proceeds almost

completely through apps and less and less through human hands. Automation and robotization lead to increased efficiency, so that we can do more and more with less input from people: technological aids ensure that together we create more value per hour worked (Commissie Borstlap, 2020). This has greatly affected the labour market, and with that, effects for the level of self-employment.

There is a large body of literature relating to the correlation between technological development and self-employment, and although not entirely consistent, most studies do find that a higher level of technological development leads to a higher

self-technological development as the level of patent applications. I expect that the rise in technological development tends to increase the self-employment rate.

Hypothesis 1: Increased patent applications will be positively correlated with the level of self-employment.

2.3.2. Indicator 2: The level of education

A second factor that is linked to the self-employment level in Europe is the education level. The level of higher education in the world is increasing rapidly (Goldwin & Katz, 2008). The OECD acknowledges this and states that “as globalization and technology continue to re-shape the needs of labour markets worldwide, the demand for

individuals with a broader knowledge base and more specialized skills continues to rise” (OECD2, 2020). To measure the indicator of education level I will use the OECD database “Population with tertiary education”. The education level in this research that has been used is the percentage of tertiary education. Tertiary education is defined according to the definition of the OECD. The OECD states that the population with tertiary education are “those having completed the highest level of education, this includes both theoretical programmes leading to advanced research or high skill professions such as medicine and more vocational programmes leading to the labour market” (OECD2, 2020).

Research from Lucas’ found out that the increase in higher education can affect the level of self-employment in many states. He states that education enhances an individual’s managerial ability, which therefore increases an individuals’ tendency to become self-employed (Le, 1999). Other researchers also acknowledge that a more educated person has a higher feasibility of becoming self-employed (Le, 1999). I believe this is the case in Europe. In other parts of the world that are developed, this can be different. In less-developed countries self-employed people are overall workers on the street or farmers. So there, the increase in higher education may discourage an individual to become self-employed, compared to Europe, where I believe it would encourage an individual to become self-employed. Therefore, I think it is likely that a rise in the tertiary education level will lead to an increase in the self-employment rate in Europe.

Hypothesis 2: A higher level of tertiary education will be positively correlated with the level of self-employment.

2.3.3. Indicator 3: Economic development

The third factor that is linked to the self-employment level in Europe is the level of economic development. Over the last couple of decades, the economic development in Europe rose tremendously (OECD3, 2020). The average wealth of states increased and the standards of living increased with it. To measure the indicator of economic

development, I will look at the GDP per capita of the European states. I will use the “Gross Domestic Product (GDP)” database from the OECD. The OECD defines the gross domestic product (GDP) as “the standard measure of the value added through the

production of goods and services in a country during a certain period” (OECD3, 2020). It measures the income earned from that production. OECD also states that GDP is the most important indicator to capture economic activity.

According to the International Growth Centre, GDP per capita does not measure the full concept of development, but it is an adequate scale to measure the economic growth of a state (Sili, 2019). The economic development in Europe spiked with the reduction of international trade barriers. Nowadays, import duties hardly have any effect because states are opening up their borders so goods and services can move easily. This process of economic integration is part of the globalization process that is happening in the world the last couple of decades. Globalization and the European internal market lead to an increase in value creation at a macro level due to rising labour productivity.

However, both developments also ensure a greater supply of workers at the base of the labour market, who are sometimes also prepared to achieve work (Commissie Borstlap, 2020).

I believe this increase in GDP per capita in Europe, can increase the level of

self-employment. Robson and Parker suggest that the level of self-employment is related to the level of GDP per capita (Robson & Parker, 2004). Higher GDP per capita might indicate resilient demand conditions within countries, which might improperly benefit

development comes increasing demand, which leads to more jobs, and therefore also more self-employment. Therefore, I predict a positive relationship between higher economic development and the level of self-employment.

Hypothesis 3: Higher economic development is positively correlated with the level of self-employment.

Section III | Methodology

Section III presents the methodology that is used in this research. The goal of this prospective research is to estimate the effect of technological developments, the

increase in education and economic development in the period from 1985 until 2017 on self-employment. The research tries to answer the following research question: Why did

the number of self-employed people increase between 1985 and 2017?

3.1. Panel data

The method used to try and ascertain the effects of technological development, the education level and economic development on the level of self-employment is to use pooled cross-sectional time-series data for 21 European countries for the period of 1985-2017. Using this data, I performed a panel data regression with random effects. One of the main reasons for using this method is the large number of observations in these datasets due to the multidimensional character. Panel data controls for variables you cannot observe like cultural factors or national policies (Torres-Reyna, 2007). National time-series data lack power and reliability because of the short spans of data, panel data is a better method for this research because it looks at more than one country and one time period.

According to Baltagi (2005), panel data has a lot of advantages. The first advantage that is listed is “controlling for individual heterogeneity”. “Panel data suggests that

individuals, firms, states or countries are heterogeneous” (Baltagi, 2005). In this research I look at the data of 21 different countries. These countries differ in their institutions, history and political regimes. All these country-specific variables have an effect on the level of self-employment. Not accounting for this country heterogeneity causes wrong assumptions. The second advantage that Baltagi states is that “panel data give more informative data, more variability, less collinearity among the variables, more degrees of freedom and more efficiency” (Baltagi, 2005). In a panel data analysis is a lot of variation, namely variation between states and within states (Baltagi, 2005). These cross-state dimensions add variability, which leads to more accurate assumptions.

adjustment” (Baltagi, 2005). In panel data you can observe changes in other periods. In this research it is possible to see how the level of self-employment changes in the process between 1985-2017.

The fourth advantage of panel data that Baltagi explains is that panel data “are better able to identify and measure effects that are simply not detectable in pure cross-section or pure time-series data” (Baltagi, 2005). Panel data filters out the right effects which leads to accurate conclusions. The last advantages Baltagi mentions are that panel data allows to “construct and test more complicated models” and that panel data analysis reduces and eliminates “biases resulting from aggregation over firms or individuals” (Baltagi, 2005). According to Baltagi’s extensive argumentation for panel data, I believe this is the best research design for this research.

3.2. Random effects model

When using panel data, you need to decide between a fixed effects model or a random effects model. The test to decide which model you should use is the Hausman test. The Hausman test was performed to check if random or fixed effects are appropriate within this analysis. This test turned out not to be significant 1_{. So, the random effect model is} appropriate for the data. In the random effect model the alteration between items is assumed to be random and uncorrelated with the independent variables included in the model. According to Baltagi (2005), a random effects model is appropriate, because in this research N is large. A fixed effect model would lead to an enormous loss of degrees of freedom and allows for endogeneity of the regressors (Baltagi, 2005). In contrast, the random effect model allows for exogeneity of the regressors.

The random effect model is the following:

𝑦_𝑖𝑡 = 𝛼 + β(𝑋_𝑖𝑡) + 𝜀_𝑖𝑡

In this equation 𝑦_𝑖𝑡 indicates the dependent variable, which is the level of

self-employment. The parameter 𝛼 indicates the constant. The parameter β(𝑋_𝑖𝑡) indicates the effect of the primary explanatory variable of interest, which will either be

technological development, education level or economic development, depending on what hypothesis is being tested. And at last 𝜀_𝑖𝑡 is denoting the equation’s error term. In the random effect model it is assumed that the error term is not correlated with the predictors (Torres-Reyna, 2007).

3.3. Heteroskedasticity and serial correlation

In this research heteroscedasticity consistent (robust) standard errors will be used. Assuming homoscedasticity will lead to restrictive assumptions (Baltagi, 2005). “Cross-sectional units may be of varying size and as a result may exhibit different variation” (Baltagi, 2005). Therefore, robust standard errors will be used in this research that will correct for possible heteroskedasticity. To test for serial correlation in the data, I used the Wooldridge test2_{. It turned out that there was some serial correlation present. This} occurs when panel data sets are unequally spaced. It was no surprise that there was some serial correlation in the dataset, because panel data cannot be collected every period in this large time frame. According to Baltagi (2005), it is likely to come across serial correlation when using data on countries or states, because in certain years, the data is just not there. In some years the data might not be recorded, hard to obtain or is missing (Baltagi, 2005). Therefore, I needed to modify the data. I did this using an unequally spaced panel data regression model with AR(1) remainder disturbances (Baltagi & Li, 1991).

3.4. Data collection

The data used in this study is predominantly obtained from OECD databases, in

particular the “OECD employment rate”, “Population with tertiary education” and “Gross domestic product (GDP)”. One dataset is retrieved from the World Bank database,

namely the “Patent applications” dataset. This data is pooled and this leads to 423 observations over 21 countries in the period of 1985-2017. It is assumed that the OECD database provides reliable datasets. The OECD usually sources its data either in

conjunction with Eurostat and other UN agencies, or from published national accounts from agencies such as the Office of National Statistics in the United Kingdom and is therefore seen as a reliable source of data (OECD, n.d). I chose to use the OECD database mainly because of the large size and availability of the database. There is data available for most of the existing states and for very diverse categories. The data is considered to be of high quality. According to the Guardian, the OECD database is in the Top 10 of best sources of data for research (Holden, 2016). Also, the OECD data is consistent and uniform. Every variable is measured in the same way for each country which leads to a higher validity of the data. This also applies for the World Bank database, which is the other source of data that has been used. The World Bank database is also in the Guardian’s top 10 of best sources of data for research (Holden, 2016).

I use the time frame of 1985 until 2017. This time frame was chosen for three reasons. The first reason is that this time period is large enough to observe a trend. The second reason is data availability. The year 1985 was the first year and 2017 was the last year that consisted data for most of the countries. The last reason is that, according to the theory, is the beginning of the postwar period, in which a lot of changes are observed. I use the data of 21 countries in the European Union. I chose European countries because I believe European countries are alike in many ways. They share the same market and most of the countries also share the same currency. Also, countries in Europe have likely norms and values, and comply with the same rules. European states also score almost the same on the Human Development Index and have many more similarities (Statista, 2021). In the research, I excluded Bulgaria, Cyprus, Croatia, Malta, Romania,

Luxembourg and the Slovak Republic because there was not enough data available for these countries.

Prior to running regressions and tests on the panel data, the datasets have been structured and merged in SPSS Statistics. All the regressions and tests have thereafter been conducted using Stata 16.1.

Section IV | Descriptive evidence

Section IV presents descriptive evidence that technological development, education level and economic development have an influence on the level of self-employment in Europe.

4.1. Self-employment rate

The independent variable in this research is the percentage of self-employment. This data is retrieved from the “OECD employment rate” (OECD, 2020). This data is

presented yearly, as a percentage of the total employment level, therefore it was not necessary to create a natural logarithm. Figure 2 shows a scatter plot to explore this data. Table 1 shows the self-employment level in 2017 of the European countries that are used. In figure 2 you observe one line that is inconsistent with the trend. This is the state of Greece. The self-employment rate in Greece is relatively high and inconsistent with the other countries in Europe. A possible reason for this can be that in Greece a very big share of the population works in agriculture (OECD5, 2021). People working in agriculture are often employed. This also explains the fact that the level of self-employment in Greece is declining, because it is no longer necessary for so many people to work in agriculture to provide for their own food.

Figure 2: Scatter plot of level of self-employment 1985-2017.

Table 1: Self-employment rates (percentages) 2017.

Country Self-employment rate in 2017 (%)

Austria 12.38 Belgium 14.29 Czech Republic 17.12 Germany 10.19 Denmark 8.34 Spain 16.48 Finland 13.17 France 11.61 United Kingdom 15.36 Greece 34.07 Hungary 10.34 Ireland 15.46 Italy 23.20 Lithuania 12.01 Latvia 12.66 Netherlands 16.74 Poland 20.14 Portugal 16.98 Slovenia 14.58 Sweden 9.86

Note: self-employment as a percentage of total employment Source: OECD1 (2020).

4.2. Indicator 1: Technological development

The dependent variables are factors that could influence the level of self-employed people in a state. There are three indicators from which the effect is measured on the percentage of self-employment. The first is technological development. The data that will be used for this indicator is the data on patent application. This data is retrieved from the “Patent applications” dataset from the World Bank database. This data is

presented in absolute numbers, therefore a natural logarithm has been created. Figure 3 provides an overview of all the available patent information on granted patents. You can observe the massive rise in patent applications in the last decades. In this research I use patents from all over the world, and not just a specific country, because I believe that patents on inventions and innovations do not have to be domestic to have an influence in another country. For example, if a company in Japan invents a new computer chip, it does not mean that we cannot buy and use this chip in the Netherlands. The approach is that the number of patent applications, including their nation of origin, is a measure of technological progress (Faust, 1990).

Figure 3: Scatter plot of LN patents 1985-2017.

Finding a good way to measure technological development is a challenge. Coccia (2008) states that “measurement demands some one-one relation between the numbers and magnitudes in question - a relation which may be direct or indirect, important or trivial, according to the circumstances” (Coccia, 2008). This is hard to find for technological development, but I believe an adequate method to measure technological development is to use data on patent application (Basberg, 1987). In this research I will use the World Bank database “Patents applications, residents (World Bank, 2020). The World Bank defines patent applications as the “worldwide patents application filed through the Patent Cooperation Treaty procedure or with a national patent office for exclusive rights for an invention - a product or process that provides a new way of doing something or offers a new technical solution to a problem” (World Bank, 2020).

Patent data includes information about new technologies and patents are the first sources of information available about new technologies. According to Basberg “the use of patent statistics rests on the assumption that they are reflecting inventive activity and innovation” (Basberg, 1987). I agree that patents reflect the level of innovation in the world and are therefore a valuable indicator for technological development. This is a valuable source of information for technological change. Data on patent application also perfectly lends itself for comparisons across time and countries because it is very consistent (World Bank, 2020).

4.3. Indicator 2: The level of education

The second independent variable that has been used in this research is the education level. The data that is used to measure the indicator of education level is data on tertiary education. This data is retrieved from the OECD database “Population with tertiary education”. This data is presented in percentages of the same age group. Because the data is presented in percentages, it was not necessary to create a natural logarithm. Figure 4 provides an overview of the available data on tertiary education in the age group of 25-34 year olds. You can observe the massive increase in tertiary education.

Figure 4: Scatter plot of tertiary education in the age group of 25-34 year olds 1985-2017.

Source: OECD2 (2020)

In this research, I look specifically at the age group of 25-34 year olds because I believe this is the age period in which one chooses their career path. I also chose this age group because I believe there are not many people who after this age, still change their

education level by pursuing a tertiary degree. In 2010, over 75% of students in

(Statista, 2010). Therefore, I believe the age group of 25-34 is appropriate for this research.

4.4. Indicator 3: Economic development

The last dependent variable that has been used in this research is the level of economic development. The data that is used to measure this is data on the Gross Domestic Product per capita. This data is retrieved from the “Gross Domestic Product (GDP)” database from the OECD. This data is presented in absolute numbers, namely in US dollars, therefore a natural logarithm has been created. Figure 5 provides an overview of the available data of the GDP in Europe and shows the increasing GDP levels in Europe. .

Figure 5: Scatter plot of LN of GDP per capita 1985-2017.

Section V | Regression analysis and results

By using the method laid out in chapter IV, this section will provide the results of the analysis aimed at answering the following research question: Why did the number of

self-employed people increase between 1985 and 2017? The chapter provides the

answers to the questions of whether the hypotheses conducted in section III will be confirmed or rejected. The results are shown in table 2, table 3, and table 4. Table 2 presents the coefficient results and the robust standard errors from the panel data analysis. Table 3 presents the results when Greece is removed from the analysis and table 4 presents the results when a distinction is made between the time period before and after the financial crisis of 2008. The coefficients have been rounded off to two decimal places. The Chi-square statistics are reported with the Pearson chi-square value and the significance level. The robust standard errors are displayed in parentheses next to the coefficients. Coefficients with an * are statistically significant. The *, **, *** marks denote the significance test at the 10 percent, 5 percent and 1 percent level,

respectively. In advance of looking at the results, it should certainly be made clear that this panel regression model is statistically significant in terms of explaining the

research. Wald χ2 _{= 89.05***, the results of the Wald chi-square test indicate that the} model as a whole is statistically significant at the .001 level.

Table 2: Regression output for the dependent variables.

Independent variables Dependent variable: self-employment rate (%)

Constant 48.93 (6.45)*** LN Patents .98 (.45)* EDU 25-34 (%) -.40 (.03) LN GDP per capita -4.61 (.45)*** N observations 423 N countries 21 Wald χ2 89.05***

Note: Robust standard errors in parentheses. *p<.05, **p<.01, ***p<.001

But how do these results fit the theoretical expectations? Table 2 shows the results of the regression analysis when using the self-employment rate as the independent

variable, therefore the coefficients tell us about the strength of the relationship between the various indicators and the self-employment rate. Not all indicators can confirm the expectations and findings from before. Not only do two out of the three dependent variables present a statistically significant relation. Also, two out of the three models prove to be negatively related to the level of self-employment, while I expected them to be positively related. Table 2 shows that technological development, in terms of patent application is significantly and positively correlated with the level of self-employment.

These results are in line with the expectations made before. The first hypothesis, namely, “increased patent applications will be positively correlated with the level of self-employment”, can be accepted based on the datasets used in this research. There is a statistically significant relationship at the .05 level, so the level of patent applications have had a statistically significant, positive influence on the level of self-employment.

There are some surprises in the other two variables. Economic development, in terms of GDP per capita, is statistically significant at the .001 level. But unexpectedly, it is

negatively correlated with the level of self-employment. The hypothesis “higher economic development is positively correlated with the level of self-employment” cannot be accepted based on the datasets used in this research. There is a statistically significant relationship, so the GDP per capita has had a statistically significant, but negative influence on the level of employment. I expected that the level of self-employment was related to the level of GDP per capita. This assumption is proven, since there is a statistically significant correlation at the .001 level. But I expected this

correlation to be positive. I expected that a higher GDP per capita would provide for better conditions which would increase the self-employment level. But this is not the case in the context of this research. There is a possible explanation for this result. A higher GDP per capita may lead to better conditions on the labour market, and more demand which leads to more jobs. But this does not necessarily have to lead to a higher level of employment. It actually may discourage the likelihood of becoming self-employed because it makes the alternative, working for a set wage, more attractive.

The other surprising result is that the level of education, in terms of tertiary education in the age group of 25-34 year olds, is negatively correlated with the level of self-employment, and that this relation is not statistically significant. The hypothesis “a higher level of tertiary education will be positively correlated with the level of self-employment” cannot be accepted based on the datasets used in this research. There are multiple explanations for these results. I expected that education enhances an

individual’s managerial ability, which therefore increases an individual’s tendency to become self-employed. Therefore, I expected it to be likely that a rise in the tertiary education level in the age group 25-34 year olds would lead to an increase in the

there is no proven correlation between the education level on the level of self-employment. It is feasible that a higher level of education makes entry into the wage sector more possible and can lead to higher wages. This may discourage the likelihood of business ownership (Le, 1999). Le (1999) also states that there was a lot of

conflicting evidence in empirical studies and that also a lot of other variables can play a part, like immigration status or gender.

As observed in Figure 2 in Section IV, Greece is an outlier when it comes to the level of self-employment in Europe. The self-employment rate in Greece is relatively high and inconsistent with the other countries in Europe. Therefore, I also looked at the results, when the data of Greece was removed. Table 3 shows the results of the regression analysis when Greece is removed from the dataset. The coefficients tell us about the strength of the relationship between the various indicators and the self-employment rate. We see that there are no big differences between the results with and without Greece. Technological development and is still significantly and positively correlated with the level of self-employment. Also, the level of education, in terms of tertiary

education in the age group of 25-34 year olds, is still negatively correlated with the level of self-employment. And GDP per capita is still statistically significant and negatively correlated with the level of self-employment.

Table 3: Regression output for the dependent variables with and without Greece Independent

variables

Dependent variable: self-employment rate (%)

Dependent variable: self-employment rate (%) excluding Greece Constant 48.93 (6.45)*** 41.23 (6.10)*** LN Patents .98 (.45)* 1.07 (.45)* EDU 25-34 (%) -.40 (.03) -.04 (.02) LN GDP per capita -4.61 (.45)*** -4.11 (.68)*** N observations 423 392 N countries 21 20 Wald χ2 89.05*** 73.70***

Note: Robust standard errors in parentheses. *p<.05, **p<.01, ***p<.001

Another trend that could influence the results of the analysis is the financial crisis of 2008. The financial crisis was a period of economic stagnation and rising

unemployment. Therefore, I also looked at the difference between the pre-crisis period and the post-crisis period. Table 4 shows the results of the regression analysis before the financial crisis of 2008, and after the financial crisis. The coefficients are, also in this comparison, not very different. The indicators demonstrate the same relationships and the same level of statistical significance before and after the financial crisis of 2008.

Table 4: Regression output for the dependent variables 1985-2007 and 2007-2017. Independent

variables

Dependent variable: self-employment rate (%) 1985-2007

Dependent variable: self-employment rate (%) 2007-2017 Constant 47.16 (11.21)*** 45.31 (8.21)*** LN Patents 1.18 (.97) .73(.52) EDU 25-34 (%) -.04 (.04) -.03 (.04) LN GDP per capita -4.76 (1.06)*** -3.85 (.98)*** N observations 224 199 N countries 21 20 Wald χ2 53.78*** 20.93***

Section VI | Conclusion and discussion

This section will provide a general conclusion of the research and a brief discussion on possible points of improvement.

6.1. Conclusion

Through this analysis I have attempted to find reasons for the rise of the level of self-employment in Europe through a panel data analysis. First, I provided a framework to understand the increase in self-employment in the European Union. I argued that the self-employment level is influenced by various factors, namely technological

development, the increase in tertiary education and economic development. I supported this claim by, first, descriptive evidence and second, a quantitative analysis. In section III the methodology of the research was explained and section V provided the results of the panel data analysis.

The first expectation that rose was that increased technological development, measured by patent applications, was positively correlated with the level of self-employment. The results showed that this expectation can be accepted based on the datasets used in this research. The second hypothesis was that a higher level of tertiary education was positively correlated with the level of self-employment. This hypothesis could not be accepted because the results showed that there is no proven correlation between the education level on the level of self-employment. The last expectation was that a higher level of economic development, as measured by GDP per capita, was positively

correlated with the level of self-employment. This hypothesis could not be accepted because the relationship proved to be negative. Also, the results proved to be consistent when removing the state of Greece from the analysis and the indicators demonstrate the same relationships and the same level of statistical significance before and after the financial crisis of 2008.

Not all indicators can confirm the expectations and findings from before. Although these results were not completely what I expected to find, they have important implications

highlighted some startling results, it is unreasonable to formulate a complete answer to the research question. Many factors can influence the level of self-employment

positively or negatively. The answer goes beyond the scope of this research.

6.2. Discussion

In any research there remains room for further improvement. The first point that could be in need of improvement is the research design that is used. Because although panel data has a lot of advantages, there are also some limitations. The first limitation of a panel data analysis that Baltagi states in his book is the problem of “design and data collection” (Baltagi, 2005). This is mainly about the extent of the data or the reference period. You might question whether the period of 1985 till 2017 captures the

phenomena I wanted to demonstrate. The second limitation of a panel data analysis that could apply to my research is the limitation of already existing data. When you use already existing data, instead of collecting the data yourself, you have no control over the quality of the data. But on the other hand, it proposes the chance that you can use high quality data, without going through the trouble of collecting it yourself. I do not believe the data in this research has any problems, because the data from the OECD and the World Bank are very reliable, but of course this can never be ruled out.

Another point of improvement is the operationalization of the indicator of technological development. There are some questions around the use of patent data to measure technological development. Patent data are very practical as an indicator of technological change. But it is debatable whether every patent truly leads to an innovation (Basberg, 1987). Patents do not make a distinction between basic innovations and those that lead to great technological improvements. It is also debatable whether all patent institutions can be compared. Overall, this research provides a basis for other researchers to study the phenomena of self-employment, but to formulate a final answer to the research question, more quantitative and qualitative research is needed.

Literature

Autor, D. H., Dorn, D. & Hanson, G. H. (2010). Untangling trade and technology: evidence from local labour markets. The Economic Journal, 125(1), p. 621-646.

Baltagi, B. H. (2005). Econometric Analysis of Panel Data (3rd ed.). Wiley.

Baltagi, B. H. & Li, Q. (1991). A joint test for serial correlation and random individual effects. Statistics & Probability Letters, 11(3), p. 277-280.

Basberg, B. L. (1987). Patents and the measurement of technological change: A survey of the literature. Research Policy, 16(1), p. 131-141.

Belastingdienst. (2020). Aangifte inkomstenbelasting voor ondernemers. Retreived on 10 december 2020, from

https://www.belastingdienst.nl/wps/wcm/connect/bldcontentnl/belastingdienst/zake lijk/winst/inkomstenbelasting/inkomstenbelasting_voor_ondernemers/

van den Berge, W., Erken, H., de Graaf-Zijl, M., & van Loon, E. (2014). The Dutch labour market during the Great Recession. CPB Netherlands Bureau for WEconomic Policy

Analysis, p. 1-54.

Button, K., (2004). The rationale for road pricing: standard theory and latest advances.

Research in Transportation Economics, 9(1), p. 3-25.

Centraal Bureau voor de Statistiek. (2020). Beroepsbevolking. Retrieved on December 15, 2020, from

https://www.cbs.nl/nl-nl/visualisaties/dashboard-beroepsbevolking/beroepsbevolking

Coccia, M. (2018). Measurement of the evolution of technology: A new perspective.

CocciaLAB Working Paper, 29(1), p. 1-41.

Commissie Borstlap. (2020). In wat voor land willen wij werken? The Hague, The Netherlands: Commissie Regulering van Werk.

Dachs, B. (2017). The impact of new technologies on the labour market and the social economy. European Parliament Research Service, Scientific Foresight Unit (STOA), p. 1-61.

Ebbinghaus, B,. & Visser, J. (1998). When Institutions Matter: Union Growth and Decline in Western Europe, 1950-95. European Sociological Review, 15(2), p. 135-158.

Faust, K. (1990). Early identification of technological advances on the basis of patent data. Scientometrics, 19(1), p. 473–480.

Goldwin, C. & Katz, L. (2008). The Race between Education and Technology. Harvard University press.

Holden, J. (2018). The top 10 sources of data for international development research. The Guardian. Retrieved from:

https://www.theguardian.com/global-development- professionals-network/2016/mar/16/the-top-10-sources-of-data-for-international-development-research

van der Klaauw, B. (2013). Naar een modernere arbeidsmarkt. TPEdigitaal, 7(2), p. 54-59.

Knotter, A., (1986). Stedelijke economie en arbeidsmarkt. Amsterdam in de eerste helft van de negentiende eeuw. BMGN - Low Countries Historical Review, 101(4), p.551–580.

Kwon, H. Y., & Pontusson, J. (2010). Globalization, labour power and partisan politics revisited. Socio-Economic Review, 8(2), p. 251–281.

Le, A. T. (1999). Empirical Studies of Self-Employment. Journal of Economic Surveys,

13(4), p. 381–416.

Nas, T. F. (2026) Cost-Benefit Analysis: Theory and Application. London, United Kingdom: Lexington books.

Neaime, S. (2015). Sustainability of budget deficits and public debts in selected European Union countries. Journal of Economic Asymmetries, 12(1), p. 1-21.

Nowotny, E., Mooslechner, P., & Ritzberger-Grunwald, D. (2009). The Integration of

European Labour Markets. Edward Elgar Pub.

OECD1 (2020) Self-employment rate. Retrieved on November 28, 2020, from: doi: 10.1787/fb58715e-en

OECD2 (2020), Population with tertiary education. Retrieved on November, 28 2020, from: doi: 10.1787/0b8f90e9-en

OECD3 (2020), Gross domestic product (GDP). Retrieved on November 28, 2020, from doi: 10.1787/dc2f7aec-en

OECD4 (2020), Unemployment rate, Retrieved on November 15, 2020 from https://data.oecd.org/unemp/unemployment-rate.htm

OECD5 (2021), Employment by activity, Retrieved on February 13, 2021 from https://data.oecd.org/emp/employment-by-activity.htm

Parker, S. C., & Robson, M. T. (2004). Explaining International Variations in Self-Employment: Evidence from a Panel of OECD Countries. Southern Economic Journal,

71(2), p. 287–301.

Paul, S. & Thomas, L. (2020). The agricultural productivity gap and self-employment bias

in the labor income share. IZA Institute of Labor Economics.

Plümper, T., Troeger, V. E. & Manow, P. (2005). Panel data analysis in comparative politics: Linking method to theory. European Journal of Political Research, 44(1), p. 327-254.

Sili, L. (2019, February 4). Is GDP an adequate measure of development? - IGC Blog. IGC. Retrieved from:

https://www.theigc.org/blog/is-gdp-an-adequate-measure-of-development/

Statista. (2021, January 11). Human development index of European countries 2019. Retrieved from: https://www.statista.com/statistics/933977/human-development-index-of-european-countries/

Torres-Reyna, O. (2007). Getting Started in Stata and R - DSS at Princeton University. Retrieved from: https://dss.princeton.edu/training/

Wilthagen, T. (2019). Geef de zzp’er de ruimte en fiscaliseer de sociale zekerheid. Retreived on 2 december 2020, from https://fd.nl/opinie/1321068/geef-de-zzp-er-de-ruimte-en-fiscaliseer-de-sociale-zekerheid

World Bank (2020), Patent applications, residents, Retrieved on November 28, 2020, from https://data.worldbank.org/indicator/IP.PAT.RESD?end=2017&start=1985

WTO. (2017), Impact of technology on labour market outcomes, World Trade Report 2017: Trade, Technology and Jobs, WTO, Geneva.

Vrijburg, H., & Geilenkirchen, G. (2019). Road Pricing by design: the Dutch case.

Vickerman, R., Spiekermann, K., & Wegener, M. (1999). Accessibility and Economic Development in Europe. Regional Studies, 33(1), p. 1–15.

Vijayamohanan, P. N. (2016). Panel data analysis with Stata part 1: Fixed effects and

random effects models. ResearchGate. Retrieved on January 10, 2021 from:

https://www.researchgate.net/publication/316888692_Panel_Data_Analysis_with_Stat a_Part_1_Fixed_Effects_and_Random_Effects_Models

Zzp-Nederland (2020). Belastingvoordeel: alle startersvoordelen op een rij. Retrieved on 10 december 2020, from