Does entrepreneurship pay? :

(1)

Master‟s Thesis Msc Entrepreneurship 2015:

Does Entrepreneurship Pay?

Student: Umer Saqib

UvA ID 10827196/ VU ID 2566920

Supervised by Dr. Philipp Koellinger

Date of completion: 30

th

June 2015

Program: Msc in Entrepreneurship

(2)

1

Preface

I am really grateful to my professor Dr. Philipp Koellinger for supervising and helping me throughout my thesis and be of assistance whenever it was required. I would also thank my group members who have gone through the same path as me and we can really relate to each other for that. Last but not the least I would love to dedicate my master‟s thesis to none other but my parents who are my backbone and the reason why I had the opportunity to have my education from these prestigious institutions in the world, thousand miles away from home. Who were my support in all thick and thin days and motivated me through my hard times studying abroad. Thank you.

(3)

2

Abstract

This intention of this thesis is to replicate the study done on United States data by Hamilton in 2000 with the aim to providing an answer to the question, “Does entrepreneurship pay?” In order to delve deeper and provide an answer, I utilized data from United Kingdom. The entire dataset spanned from 2000 to 2004 and consists of 16,098 observations over the time span, among which 13,895 are wage employees and 2,203 are self-employed. The data included individuals who are white males and between age 18-65 years. In my thesis I intend to find the possible explanations for the wage differentials between the self-employed and wage workers. The dependent variable taken into analysis was the gross weekly earnings of the individual from the main job. The independent variables taken into analysis were potential labour market experience, tenure in the current job, years of education, college degree or lower and marital status. Their effect on the dependent variables were seen over the years observed. There were multiple regressions applied on the dataset that included Ordinary Least Square & Quantile regressions and Random Effects Estimator. The empirical results suggest that the entrepreneurs have lower initial earnings and have lower earnings growth as well unless they do not belong to the upper quartile of the distribution. Those experienced and successful entrepreneurs earn more when they have been in the same business for more than 5 years and fall in the 75th percentile or higher. Furthermore, the results suggest that the earnings of entrepreneurs never overtake the earnings of the wage workers at mean, 25th and 50th quantiles. Despite all these findings individuals choose to opt for self-employment due to non-pecuniary benefits that are substantial, such as „being your own boss‟.

(4)

3

Introduction

The field of entrepreneurship is trending and several famous researchers are making efforts to discover more about it. A variety of academics, companies and educational institutions are digging into the phenomenon of entrepreneurship, trying to understand how it works. However, the central focus of the debate is whether or not entrepreneurship pays more as compared to wage employment? In other words, this is asking whether you earn more if you have your own business or do you earn more if you are a wage worker? What are the determinants influence whether individuals choose self-employment and what are the pecuniary and non-pecuniary factors that play part in this decision? This is the question that still does not have one accurate answer. Different researchers working on this topic have various answers and opinions on it. Hamilton (2000) conducted a study on US data and the objective of the study was to determine the extent to which the behaviour of the individuals who are choosing to enter entrepreneurship or remain as an entrepreneur can be explained by the models of entrepreneurship and labour market. The models he used in his paper were 1) Investment and agency models (Lazer and Moore 1984), 2) Matching and Learning models (Roy 1951; Jovanovic 1982), and 3) Super star models (Rosen, 1981).

The data utilized for the study was from the Survey of Income and Program Participation (SIPP) from 1984. He concluded that choosing self-employment as your career or staying self-employed pays less than working as a wage employee except if you are in the upper quartile of the self-employment income distribution. On the other hand, people who are in the 25th and 50th percentile of the self-employed earn less than the wage employees. Hamilton (2000) also mentioned that if you have been self-employed for ten years, you earn 35% less income than a wage worker who has been wage employee for the same time period. However, people still opt for entrepreneurship and choose to stay self-employed due to non-pecuniary benefits, such as „being your own boss‟. Besides this there are researchers who negate this view by using different data sets and methods, such as Fairlie (2005) and Mcmanus (2005), who suggested that entrepreneurs have better incentives and options to evade taxes by underreporting their income to the tax authorities (Parker 2009) and there are many other factors that determine the choice of entering the entrepreneurship and staying self-employed. Evans and Leighton (1989) mention that people who prefer greater autonomy will choose to be self-employed and Blachflower and Oswald (1992) found in their study that self-employed people enjoy greater job satisfaction as compared to the wage employees.

(6)

5

The purpose of my thesis is to enter into a debate that has been simmering within academic for quite long, the central question of which focuses on whether or not entrepreneurship pays more than wage employment. Furthermore, I intend to determine if Hamilton‟s conclusion is generalizable, thus applying to the dataset I utilized on Great Britain. To answer the aforementioned question, I will replicate Hamilton‟s study (2000) on the United States and apply it to the British dataset, the details of which will be discussed later in the methodological section of this thesis. I will apply the same control variables and will take into consideration to utilize a sample that is as similar as possible to the one used by Hamilton in his study, and run the same tests and analyses. However, there will be additional analysis and tests carried out that are not in the Hamilton‟s paper in order to analyse the phenomenon in depth and detail. Hamilton (2000) carried out his research on white males residing in US and they are between the age of 18 to 65 years. These are the controlled sample that I will be using for my study as well and I will be engaging in a comparison between the self-employed and wage workers in order to see who earns more and which factors and variables are influencing this. In a lot of cases researchers have seen that even if the entrepreneurs are earning less than the wage workers, they still choose to stick with entrepreneurship. This shows that money is not the only incentive or motivation for entrepreneurs, as there a lot of other non-pecuniary factors influencing that choice as well. The main purpose of this thesis is to determine whether entrepreneurship is better and pays more than the wage employment. Or, in contrast, is the opposite true, or is there no significant difference between the both choices of profession? This thesis will consist of five main sections. These are as following: 1) Literature review, where I will discuss the different opinions and perspectives of the researchers who have conducted research on related topics to my thesis, 2) Data description that describes the dataset used for the study, including how I procured the data sample being used, 3) Methods section, which will explain the tests and analyses that I ran on the datasets and why I chose those tests and analyses, 4) Results section, which will describe and interpret the results from the tests and analyses, and 5)Discussion & Conclusion, which will summarize the discussions, results, and literature, thus concluding with an analysis of my findings.

(7)

6

Literature Review

What evidence exists on the relative earnings of entrepreneurs compared to wage workers?

Several researchers have conducted studies that compare the earnings differentials between

the entrepreneurs and the wage workers (Astebro, 2012; Hamilton, 2000). However, comparatively there is not much we know about the pecuniary and non-pecuniary factors influence on the decision of becoming an entrepreneur and casual factors of earnings of self-employed. There are three methods to measure the income of the entrepreneurs:

1) Net profit: The profit from the business after taxes. Entrepreneurs may under-report net profit by mentioning lesser income or higher expenses.

2) Draw: It is the money that a business generates for the owner or the amount of money that the entrepreneur draws from the business.

3) Equity Adjusted Draw (EAD): It represents the draw + adjusted change in the value of business equity over time t and t+1.

It has been an ambiguous task to calculate the exact income of the entrepreneurs. One of the reasons for this is that there are no accurate or complete methods present to measure the income of the entrepreneurs and they are quite often biased and reluctant in mentioning their real income as well. However, there are different opinions regarding the earnings differentials in the existing literature. The empirical studies tell us that entrepreneurs have higher initial earnings growth in a new business as compared to the paid employees in a new job on average and the potential wages of entrepreneurs do not have a significant difference from the wages of the wage workers (Brock and Evans, 1986; Rees and Shah, 1986; Borjas and Bronars, 1989; Evans and Leighton, 1989). However, even then people choose to opt for self-employment and remain self-employed. This is due to the non-pecuniary reasons, such as being your own boss (Hamilton, 2000). The studies show that the mean and median income of entrepreneurs is less than that of the wage workers (Levine & Rubinstein, 2013; Astebro, 2000). The research carried out by Hamilton (2000) states that the entrepreneurs on average earn less than the wage workers. The study stated that the entrepreneurs falling in the 25th and 50th percentile earn less than the wage workers no matter what method is used to measure the income of the entrepreneurs. However, when it comes to the 75th percentile the entrepreneurs earn more than the wage workers when the EAD method is used to measure the income of the entrepreneur. This is because there are few successful entrepreneurs who have really high

(8)

7

earnings as compared to the wage workers and they are referred as “superstars” (Rosen 1981). This is related to the superstar models that suggest that the mean of comparisons between the earnings of wage workers and the self-employed will be influenced by a small number of very high income earnings superstar entrepreneurs. On the other hand, entrepreneurs who lie in the median, earn 25% less than the wage workers on average and would be better off if they shift to the wage employment at any time, no matter how long they have owned that business (Hamilton, 2000).

In his paper Astebro (2012) suggests that 75% of all the self-employed would be earning more if they would have not entered entrepreneurship as their income in wage jobs would be higher. He also stated that when EAD is not used the entrepreneurs will earn less than the wage workers at all the levels except the 99th percentile where it could be seen that self-employed income is 50% more than the wage workers income. There is higher variance in the earnings distribution of the entrepreneurs and higher positive skew as compared to the wage workers. Earning profiles for the entrepreneurs are lower in the beginning and are generally flatter than the earnings profiles of the wage workers (Astebro, 2012). Average decline could be seen in the income of the wage workers who shift to the self-employment and remaining there for an average duration. The literature also suggests that the earning distribution among the entrepreneurs and the wage workers is also dependent on the industry they fall in. 75% of technological entrepreneurs which are patent or invention holders Astebro, (2003) would be better off if they would stay as wage workers. 97% of all the investors who invest in the new ventures would gain more if they would put their money in the banks at even zero interest. Still people invest and choose to opt for entrepreneurship, major reason being non-pecuniary benefits such as autonomy and freedom. Secondly, there is a chance that you can be a superstar entrepreneur and fall in the 0.5% who earn at the rate of above 1400% (Astebro, 2012). However there are researchers who concluded that entrepreneurs have higher mean and median earnings compared to the wage workers, if properly controlled for the variables like education and personal characteristics (Rosen and Willen, 2002). Similarly Daly (2009) finds that men who are self-employed tend to have similar earnings as compared to the members of matched controlled group in short and medium term but entrepreneurs have higher income in the longer term.

(9)

8

What are the potential reasons for the reported results?

We can find vast literature and several methods when it comes to measuring the income of wageworkers; on the other, measuring the income of an entrepreneur presents a difficult task. In contrast to the vast literature examining the differential among the wage workers, entrepreneurs are usually excluded from the labour market studies (Hamilton, 2000). This is because of the complicated and non-accurate measures for the income of the entrepreneurs. Net profit is the most common used measure of self-employed income and the alternative to it is the Draw (Parker, 2009) and Equity Adjusted Draw (Hamilton, 2000; Astebro, 2012). The research carried out in the past suggests that measuring the income of wage workers is easier than measuring the income of the entrepreneurs. This is because entrepreneurs show biasness while mentioning their accurate income to evade the income tax. Entrepreneurs have the liberty to report their income to the tax authorities themselves; they find it easy to evade tax form the tax authorities by mentioning low income (Parker, 2009). Hence when these entrepreneurs are surveyed, they under report their income and avoid mentioning their real income.

The question that researchers seek answers for is that why do entrepreneurs choose to become an entrepreneur and stick with self-employment even they have lower earnings and would be better off if they would have sticked with wage jobs? The literature suggests that the individuals choose entrepreneurship for several non-pecuniary reasons such as being your own boss (Hamilton, 2000), freedom, autonomy, sill utilization, proving your worth (Astebro, 2012). Schumpeter (1947) highlighted the non-pecuniary motivation to become an entrepreneur, highlighting the „...will to found a private kingdom, impulse to fight, to succeed for the sake, to prove oneself superior, not of the fruits of success, but of success itself‟. The present empirical literature indicates that the self-employment does not give returns economically and it is motivated by non-pecuniary factors such as job satisfaction and pursuit for autonomy and freedom (Van Praag and Versloot, 2007). Hamilton (2000) gives evidence in his paper that many individuals are willing to sacrifice higher earnings as employees in order to stay as an entrepreneur. Similarly Santarelli and Vivarelli (2007) argued that the results from their questionnaire invariably show that the desire to exploit your own skills and search for independence and autonomy are ranked higher among the determinants of forming a new firm than pursuing pecuniary rewards in wage employment. There are two types of entrepreneurship entries: proactive and reactive (Berglaan, 2009). The proactive ones leave the wage job in order to become an entrepreneur and the reactive ones enter entrepreneurship in order to escape unemployment. A study carried out on Norwegian data by Berglaan (2009)

(10)

9

showed that the transition rate from wage jobs to self-employment is on average for men is 1.5% (as compared to 2% - 3.5% in the U.S (Parker, 2004), while transition from entrepreneurship to wage employment is 8.7% and the average rate form unemployment to self-employment is 3.2%. Lastly, only 1% of self-employed become unemployed as compared to 2.8% of wage workers becoming unemployed. In sum, people without appropriate skills choose entrepreneurship in order to get rid of unemployment or take entrepreneurship as the last resort when they could not do well in wage employment sector, resulting in incompetent entrepreneurs and small business that do not survive long (MacDonald, 1988).

Astebro (2012) mentioned the star and misfit theory that details how some entrepreneurs earn way higher than the wageworkers who lie in the upper most quartile of entrepreneurs and then there are misfits who change their occupations because they do not fit in one particular occupation they are more likely to choose self-employment as last option. These are possible reasons for the wage differentials between wage workers and self-employed. Lastly, the literature mentions that entrepreneurs are happier and more satisfied with their jobs and enjoy the freedom and autonomy (Bradley & Roberts, 2004). Despite the lower earnings, that explains that they seek other non-pecuniary rewards than higher income levels.

What challenges do the researchers have to face when trying to estimate the relative income of the entrepreneurs compared to the wage workers?

Measuring income of an entrepreneur has been a challenge for researchers as compared to the wage workers, and that is one of the reasons that self-employed are excluded from the labour market studies quite often (Hamilton, 2000). The reason why it is a challenge is because the entrepreneurs in most of the countries have the responsibility to report their income to the tax authorities themselves. So entrepreneurs get the liberty to understate their income and overstate their expenses so that they can evade the income tax whereas the wageworkers do not get this opportunity (Peterson, 2009). Due to this reason entrepreneurs are reluctant to mention their real incomes in the survey questionnaires and fill in biased values usually, that are not the real representative of their incomes which results in distorting and biasing the results. A part from this, entrepreneurs can also hide their earnings by not taking them out of their firm. They tend to do this in order to save for their healthier retirement or to increase the value of the firm so they can pass it on to their potential successors in a good shape (Tergiman, 2009). Secondly major issue arises when we talk about the fringe benefits and to see what is their monetary value. The gaps in earnings are usually understated given that

(11)

10

wage workers benefit from the fringe benefits such as health insurance and paid vacations. The wageworkers get fringe benefits from their employers such as vacations, health insurance, bonuses and company vehicles etc. whereas in the case of the self-employed they cannot enjoy these benefits as they do not have an employer; they have to pay for all these expenses, necessities and luxuries themselves. These benefits account for a significant part of the income for both wage workers and the entrepreneurs but it is difficult to give these benefits an accurate monetary value to calculate the wage differentials (Moskowitz and Vissing-Jorgensen, 2002).

Moreover one another challenge that is faced by the researchers is to which income measure method to take into account while measuring the income of the entrepreneurs. As mentioned above in the literature that there are three different methods to measure entrepreneurial income i.e. Net Profit, Draw and EAD. All three methods yield different incomes for the entrepreneurs that lead to different results when they are compared to the income of the wage workers. Lastly in most of the surveys and datasets all these measures of income are not present for the entrepreneurs and usually the income earned is given, same case with my dataset. Hence it is a challenge for the researchers to make sure that the measure they are using is credible and valid to determine the earnings of the self-employed and compare it with the earnings of the wageworkers.

(12)

11

Data Description

The data I used for my thesis to see the wage differentials between self-employed and wage employees is from the British General Lifestyle Survey (GLF) formerly the General Household Survey (GHS) that ran from 1972-2011. It was a continuous annual national survey of people living in Britain and was conducted by the Office for National Statistics (ONS). The main aim of the survey was to collect data on a range of core topics, covering household, family and individual information. The UK Data Archive holds the GLF/GHS data from 1972-2011 but the standard access End User License data is only available from 1972-2006 and for the rest a Special License is required. The GHS was conducted annually except for the breaks in 1997-1998 when the survey was reviewed, 1999-2000 when the survey was redeveloped and 2005-2006 when the design of the survey was changed. So that is the reason that taking the whole dataset into account was not possible as variables that were needed were changed and their methods of measurement were different for different time periods. For example in some years net income is mentioned and in the other gross income. The data could not be taken till 2011 due to the Special License requirement. Hence the period taken into account turns out to be the best fit for my thesis due to availability of the required variables and consistency of the method used to conduct the survey is similar in this period. The data involved the general household information of the individuals that included age, sex, ethnicity, type of employment, gross weekly earnings, years of education, marital status, time in current job and number of hours worked per week etc. There were control variables applied to the dataset that were age, sex and ethnicity to have the correct sample for the analysis, that gave us the dataset of individuals that are white males, from age 18 to 65 years, who that are working full time and are either wage workers or self-employed. The whole dataset form year 2000 to 2004 consists of 16,098 observations over the time span, among which 13,895 were wage employees and 2,203 were self-employed. Table 1 below gives the summary of the variables used in the dataset.

TABLE 1: Variable names and their description

Variable name Variable Description YPserial Personal serial number

Year Year survey was conducted

(13)

12

Agesq Age squared

Sex Sex of respondent

Degree Have a college degree or lower

Grossearning Gross weekly earnings form main job(Pounds)

Ethnic Ethnicity of respondent

Workhrs Total hours worked per week

Eduage Age when left full time education

Stat Employment status: Employed or self employed

Marital Marital status

Jobexper Labour market experience

Jobexpersq Labour market experience squared

Eduyears Years of education

Jobtime Tenure in the current job

The independent variables in the dataset are the ‘marital’, ‘eduage’, ‘jobexper’, ‘jobexpersq’, ‘age’, ‘agesq’, ‘stat’, ‘degree’, ‘jobtime’ and ‘workhrs’. Among the independent variables „marital, degree and jobtime’ are categorical variables. They are split into dummies that are mentioned below:

marital:1= Married, 2= Cohabiting, 3= Single, 4= Widowed, 5= Divorced, 6= Separated and 7= Same sex couple

jobtime: 1= 1 month or less, 2= 2 to 3 months, 3= 4 to 6 months,4= 7 to 11 moths, 5= 1 year, 6= 2 to 4 years, 7= 5 to 9 years and 8= 10 years or more.

degree: 0= other or no qualification and 1= bachelor degree or higher qualification

The purpose is to see the effect of these independent variables over time on the dependent variable in the dataset that is the ‘grossearning= gross weekly earnings form the main job’. However the controlled variables applied to each of the required sample were the sex = males, age= 18 to 65 years and ethnicity= white. The reason to choose this sample was to take a sample with characteristics that are similar to the one taken by (Hamilton, 2000) in his study, so the replication would be as close as possible. Secondly it‟s a big dataset that

(14)

13

includes more than 20,000 observations each year before applying the control variables. This helps to make sure that there are enough observations in the sample that is required for my thesis, so it will yield better results and analysis. In further process to refine the data even more I only took the people from the population who are either employed or self-employed as their full time and main job because taking people into account with wage employment or self-employment as their second job distort the earnings profile of the individual and have effect on the results. Two new variables were created from the variables available in the dataset. The variable available in the dataset was eduage= age when individual left full time education. Subtracting the age when individual left their fulltime education from 6 (that is assumed when individuals start their education as taken by (Hamilton, 2000)) i.e. eduage – 6 gives us the eduyears= years of education the individual had. Once we have years of education the individual had we can calculate the labour market experience in the similar way Hamilton (2000) did in his paper i.e. “labour market experience = age of the individual – years of education – 6” gives you the new variable „jobexper’.

Moreover all the missing values were eliminated from the data set as they are responsible for distorting the data. In total there were 161 missing values that were eliminated from all the variables together. Similarly all the abnormal observations and responses were also excluded from the data set that would have effect the accuracy of the results. Superstars who were earning substantially high income as compared to the rest of the individual are the example of the abnormal values/observations.

Table 2 below shows the number of observations each year and the frequency percentage for each year as well.

TABLE 2: Number of observations and their percentage each year.

Years Freq. Percentage Cum.

2000 3,426 21.22 21.22 2001 3,603 22.21 43.43 2002 1,823 11.56 54.99 2003 3,937 24.45 79.44 2004 3,309 20.56 100.00 Total 16,098 100.00

(15)

14

Table 3 below shows the variable description and the summary statistics that represents the demographic characteristics and the productivity of the self-employed and wage workers in the dataset. Taking the mean of the independent variables in the dataset, it tells us that the entrepreneurs tend to have more labour market experience as compared to the wage workers. However wage workers are more likely to be married and are more likely to have college degree and better educated than the self-employed which goes against the findings of (Hamilton, 2000).

TABLE 3: Variable descriptions and summary statistics, by employment sector Variable Description Paid Employees Self-Employed

Mean jobexper Potential Labour

market experience: age-education-6

23.157 27.802

jobtime Tenure in current job 6.321 6.893

marital Married, spouse present

1.842 1.740

eduyears Years of education 13.047 11.583

degree Went to college .213 .196

Observations 13895 2203

The table 4 below shows the results from Levene‟s test and the t-test that we conducted to see if the variance among the groups is equal or not. The results below show that except for marital status, the variance among the groups is not equal because the p-values are significant that enable us to reject the null hypothesis of the Levene‟s test i.e. Ho: „The variance in the variables, among the self-employed and the wage workers is equal‟. So except marital status we rejected the null hypothesis and conducted t-test for all the other variables with unequal variance.

(16)

15

TABLE 4: Independent samples Levene’s test and t-test Levene’s Test for equality of

variance

t-test for equality of means Variables frequency p-value t p-value jobexper 2.072 0.000 -21.112 0.000

jobtime 1.080 0.019 -14.919 0.000

marital 0.943 0.066 6.118 0.000

degree 1.070 0.040 2.038 0.042

eduyears 6.757 0.000 11.296 0.000

The Table 4 above states the t-test values for the unequal variance in variables among the self-employed and the wage workers. We conducted the t-test in order to see if the means of the variables among self-employed and the wage workers are equal or not. For the t-test our null hypothesis was Ho: „The means of the variables among self-employed and the wage workers are equal‟. So from seeing the significant p-values of the t-test we can reject the null hypothesis and we know that the mean of the variables between two groups are not equal. The t values show the variance among the groups for a particular variable and the p-values show that if they are significant or not. In the case above we can see that there is variance among the distribution of self-employed and the wage workers in all the explanatory variables taken into account.

The table 5 below shows the differences in wage workers and entrepreneurs weekly earnings distributions in the year 2004. This year was taken as the reference year due to appropriate number of observations and being the most recent year of the survey. Unlike (Hamilton, 2000) I used only the weekly earnings from the main job that were available in the dataset instead of using net profit, draw or EAD for different measures of entrepreneur‟s income.

TABLE 5: Summary statistics of weekly earnings for self-employed and wage workers for year 2004

Statistics Wage employed income Self-employed Income

Mean 524.92 507.81 Standard deviation 484.53 487.65 25th percentile 301.28 230.76 50th percentile 426.92 384.61 75th percentile 600.01 623.61 Observations 2,892 480

(17)

16

Mean gross weekly earnings for self-employed are 3.36% less than the gross weekly earnings for wage workers but the difference is not significant. This is against the finding of (Hamilton, 2000) when he used net profit and draw as the measure of earnings but the results in my finding are similar to his findings for the 25th, 50th and 75th quantile in the case where EAD is used as the measure of income. Self-employed has a greater variance in the earnings than the employees, as stated by Borjas and Bronas (1989) in their study. The findings for the lower quartile and the median show that the entrepreneurs earn 23.4% and 10.1% less than the wage workers respectively in those quartiles. Lastly the results also show that the entrepreneurs who fall in the upper quartile earn around 3.8% more than the wage workers.. Figure 1 below shows the density graph of the hourly earnings of self-employed and the wage workers.

Figure 1: Empirical distribution of hourly earnings for self-employed and wage workers

0 .0 2 .0 4 .0 6 .0 8 .1 kd e n si ty h o u rl ye a rn in g 0 50 100 150 200 Hourly earnings

(18)

17

The graph indicates that the density of entrepreneurs and wage workers who are earning certain amount of earnings per hour. A large number of entrepreneurs are earnings below 25 pounds an hour and even larger number of wage workers fall in the same hourly earnings in the lower quartile as well. However for the median and upper quartile we can see that there are entrepreneurs who have hourly earnings of more than 100 pounds an hour and in the last 99th percentile there are entrepreneurs who have hourly earnings of almost 200 pounds. This tells us that self-employed who are in the 75th and the 99th percentiles of the distribution have higher earnings as compared to the wage workers who fall in the same percentiles.

(19)

18

Methods

This section will explain the methods, models and the strategies that were used and applied for the analysis of the data in my thesis. The methods used in this paper are similar to quite an extent to what Hamilton (2000)has applied in his paper. However there are additional models and tests used in this paper unlike Hamilton‟s. The additional model and test that were used are the random effects regression and the Hausman test that has been explained below.

OLS and Quantile regressions:

Firstly the regression I used is the Ordinary Least Square (OLS) that has been also applied by Hamilton (2000) in his paper. OLS regression estimates the unknown parameters in a linear regression model with an attempt to find a function that closely approximates the data by a „line of best fit‟. In technical terms, OLS fits a straight line through the scatter plot of the data points that minimise the sum of residuals, which are the squared vertical distances from the line.

The equation for the OLS model is mentioned below: Y = α + β1X1 + β2X2 + β3X3

It is a multiple explanatory variables equation where Y is the dependent variable and X1,2&3 are the independent variables and α is the Y-intercept of the equation. The assumptions for the multiple OLS regression are that. 1) The relationship between Y and independent variables is linear, 2) The data is a random sample form the population, 3) In the data none of the independent variables are constant and there is no perfect collinearity among the explanatory variables, 4) The value of the independent variables should not contain any influence from the mean of the unobserved variables, 5) There is heteroskedisctic variance from the line of best fit of the regression. Apart from the mean OLS regression, for OLS quantile regressions I divided my dataset into quartiles of 25th percentile, 50th percentile and 75th percentile. After dividing them into quartiles I ran the OLS regression on Stata, separately for self-Employed and wageworkers and for each quartile as well. That gave me four OLS regressions results for mean, 25th, 50th and 75th percentile. The dependent variable in the regression was „grossearning’ and the independent variables were „jobexper’, ‘jobexpersq’, ‘martial’, ‘degree’, ‘jobtime’, ‘eduyears’ and ‘year’(the description for the variables is mentioned above). For the OLS and quantile regressions, the categorical variables

(20)

19

are split into dummies as well, in order to see the effect of the dummies on the dependent variable.

Hausman test:

Secondly, I ran the Hausman‟s test in order to see that which model, either random effects or the fixed effects model fits the best to my panel data. This test compares between the random effects model and the fixed effects model, to determine which model is appropriate. For that purpose I ran both fixed effects and random effects regression on my data. The test is based upon the Prob>chi2 value that tells if we can reject the null hypothesis or not. Where null hypothesis is Ho: Random effect is appropriate and alternative hypothesis is Ha: Fixed effect is appropriate. The test results are mentioned in the results section below.

Random effects model:

The regression model that I used for the analysis is the random effects model that has not been used by Hamilton (2000) in his paper and that is one of the drawbacks of his study as he only used Pooled and quartile OLS regressions. The reason to use random effects model and not using the fixed effects model is determined from the Hausman‟s test results that indicated that random effects model is appropriate for my dataset. The rationale behind the random effects model is that unlike fixed effects model, it is assumed that the variations across individuals are random and uncorrelated with the other independent variables in the model. So we assume that all individuals have same innate abilities and are not superior or inferior to the others. We also assume that the population is infinite and the individuals are randomly selected. In panel data the data has been collected from series of years and varied individuals so it is unlikely for all the years and individuals to be equivalent, so we assume that there is no common effect size and in this case random effect is more easily justified. This model also helps to control the unobserved heterogeneity when it is correlated with independent variables and constant over time. The reason we use random effect is that if you believe that the difference across individuals and entities have some kind of effect on your dependent variable. The equation for random effects is mentioned below:

Yit = βXit + α + ε

Here Yit is the dependent variable, β is the slope parameter, Xit is the independent variable, α is the individual specific effect and ε is the error term. The random effects model assumes

(21)

20

that the individual specific effects α are distributed independently of the independent variables and are not correlated with them. We also assume that ε the errors are correlated within each unit but uncorrelated across units. Random effects model omits the fixed effects and makes up for the omission, by modelling the error structure but condition being that the omitted fixed effects are not correlated with the independent variables. Because these omitted variables can lead to biased coefficient estimates. The regression was run on Stata where dependent variable in the regression was „grossearning’ and the independent variables were „jobexper’, ‘jobexpersq’, ‘martial’, ‘degree’, ‘jobtime’, ‘eduyears’ and ‘year’. The results from the regression will be discussed in the results section.

(22)

21

Results

OLS and Quantile Regressions:

The tables below shows the OLS and quantile regressions run on self-employed and wage workers separately. For each sector four OLS regression were run for mean, 25th, 50th and 75th percentile to see the results for each part of the distribution.

Table 6: OLS and Quantile regressions for Self-employed

Regression # of observations R2

OLS 2,203 0.0890

25th 555 0.0692

50th 1,103 0.0860

75th 1,717 0.0690

Variable OLS P-value QR_25 p-value QR_50 p-value QR_75 p-value

jobexper 7.529 (3.957) 0.057 -0.022 (0.861) 0.981 -1.288 (1.083) 0.234 2.508 (1.416) 0.077 jobexpersq -0.219 (0.066) 0.001** 0.005 (0.015) 0.729 0.003 (0.018) 0.829 -0.077 (0.023) 0.001** degree Went to college 260.795 (45.713) 0.000*** -13.404 (7.908) 0.091 -1.892 (9.998) 0.850 8.816 (13.802) 0.523 marital Cohabiting -96.785 (34.473) 0.005** 6.2507 (7.499) 0.405 -0.227 (9.072) 0.980 -9.592 (11.393) 0.400 Single -77.837 (32.059) 0.015** 8.651 (6.697) 0.197 -8.298 (8.974) 0.355 -21.683 (12.556) 0.084 Widowed 133.241 (90.168) 0.140 34.346 (15.364) 0.026** -1.944 (13.977) 0.889 -1.076 (30.882) 0.972 Divorced 37.204 (79.726) 0.641 -3.407 (11.624) 0.770 6.654 (12.480) 0.594 -5.562 (16.431) 0.735 Separated 207.076 (202.484) 0.307 11.654 (20.007) 0.560 0.196 (23.507) 0.993 -7.292 (28.245) 0.796 Same sex couple -205.277 (136.827) 0.134 65.938 (12.966) 0.000*** 5.419 (34.614) 0.876 -24.867 (67.205) 0.711 jobtime 2-3 months -192.832 (111.804) 0.303 -0.054 (17.039) 0.997 7.989 (24.740) 0.747 -12.898 (36.789) 0.726 4-6 months 171.753 (166.578) 0.399 -3.063 (17.403) 0.860 43.796 (24.942) 0.079 8.581 (35.802) 0.811 7-11 months -98.174 (116.405) 0.773 9.831 (13.989) 0.482 35.399 (20.363) 0.082 19.064 (31.455) 0.545 1-2 years -34.519 (119.649) 0.759 15.536 (13.880) 0.264 46.383 (20.834) 0.026 79.181 (33.054) 0.017** 2-4 years 35.758 0.650 15.659 0.203 73.411 0.000*** 94.081 0.001**

(23)

22 (116.301) (12.290) (18.486) (28.536) 5-9 years 52.316 (115.398) 0.374 29.780 (12.014) 0.013** 96.276 (17.986) 0.000*** 97.389 (27.994) 0.001** 10 years or more 103.041 (115.908) 0.001** 28.339 (11.572) 0.015** 103.919 (17.261) 0.000*** 125.517 (27.355) 0.000*** eduyears 16.508 (4.798) 0.001** 0.314 (1.028) 0.76 -1.964 (1.1530) 0.089 -0.645 (1.5562) 0.678 year 2001 41.211 (35.902) 0.251 -11.547 (7.574) 0.128 -24.661 (8.5189) 0.004** -22.741 (11.403) 0.046** 2002 93.515 (57.259) 0.103 -15.571 (8.757) 0.076 -18.715 (10.888) 0.086 -16.205 (14.706) 0.271 2003 (77.421 (33.875) 0.022** 1.010 (7.556) 0.894 1.490 (8.4338) 0.860 -1.514 (10.762) 0.888 2004 86.152 (30.306) 0.005** 4.308 (8.223) 0.601 -1.684 (8.9115) 0.850 16.926 (11.439) 0.139 _cons 124.733 (126.719) 0.325 84.984 (18.702) 0.000*** 172.239 (26.290) 0.000*** 205.703 (35.719) 0.000***

Table 7: OLS and Quantile regressions for the Wage Workers

Regression # of observations R2

OLS 13,895 0.1203

25th 3,474 0.0883

50th 6,949 0.1374

75th 10,422 0.1990

Variable OLS p-value QR_25 p-value QR_50 p-value QR_75 p-value

jobexper 15.894 (1.144) 0.000*** 1.118 (0.333) 0.001** 2.648 (0.326) 0.000*** 5.518 (0.375) 0.000*** jobexpers q -0.319 (0.021) 0.000*** -0.034 (0.006) 0.000*** -0.068 (0.006) 0.000*** -0.124 (0.007) 0.000*** degree Went to college 152.250 (15.326) 0.000*** -0.917 (4.329) 0.832 10.263 (4.061) 0.012** 33.728 (4.051) 0.000*** marital Cohabitin g -44.204 (13.432) 0.001** 3.144 (3.392) 0.354 -8.494 (2.979) 0.004** -15.002 (3.337) 0.000*** Single -105.443 (10.954) 0.000*** -8.031 (3.226) 0.013** -26.614 (2.920) 0.000*** -44.572 (3.310) 0.000*** Widowed -64.595 (49.647) 0.193 -14.893 (11.598) 0.199 1.071 (10.957) 0.922 -24.664 (12.553) 0.049** Divorced -67.843 0.000*** -4.093 0.456 -13.070 0.011** -23.454 0.000***

(24)

23 (13.991) (5.488) (5.122) (5.712) Separated -20.107 (25.349) 0.428 5.118 (8.717) 0.557 -5.735 (7.355) 0.436 -5.735 (8.277) 0.488 Same sex couple 11.504 (68.993) 0.868 -40.857 (24.923) 0.101 -27.551 (19.357) 0.155 -36.185 (22.699) 0.111 jobtime 2-3 month -7.790 (24.184) 0.747 0.391 (8.573) 0.964 11.901 (9.228) 0.197 7.616 (11.354) 0.502 4-6 months -4.017 (21.193) 0.850 12.086 (7.041) 0.086 22.256 (7.159) 0.002** 13.756 (8.784) 0.117 7-11 months 39.917 (23.416) 0.088 7.722 (6.677) 0.248 22.933 (6.817) 0.001** 22.776 (8.389) 0.007** 1-2 years 61.709 (21.119) 0.003** 16.232 (6.073) 0.008** 31.021 (6.175) 0.000*** 34.481 (7.689) 0.000*** 2-4 years 77.566 (19.468) 0.000*** 21.942 (5.872) 0.000*** 43.217 (5.958) 0.000*** 54.054 (7.416) 0.000*** 5-9 years 119.495 (20.657) 0.000*** 26.847 (6.213) 0.000*** 55.048 (6.109) 0.000*** 72.615 (7.595) 0.000*** 10 years or more 140.315 (19.625) 0.000*** 34.687 (6.095) 0.000*** 69.554 (5.973) 0.000*** 99.801 (7.461) 0.000*** eduyears 28.864 (2.208) 0.000*** -1.256 (0.689) 0.068 0.777 (0.649) 0.231 6.836 (0.643) 0.000*** year 2001 10.528 (11.513) 0.360 7.647 (3.221) 0.018** 10.181 (2.856) 0.000*** 13.727 (3.188) 0.000*** 2002 -58.553 (15.149) 0.000*** -33.309 (3.789) 0.000*** -43.093 (3.827) 0.000*** -45.002 (4.417) 0.000*** 2003 37.991 (12.019) 0.002** 6.361 (3.257) 0.050** 14.116 (2.795) 0.000*** 24.173 (3.127) 0.000*** 2004 53.528 (12.875) 0.000*** 3.103 (3.576) 0.386 17.178 (3.055) 0.000*** 29.254 (3.354) 0.000*** _cons -112.648 (32.809) 0.001** 187.719 (10.765) 0.000*** 199.605 (10.448) 0.000*** 150.044 (11.694) 0.000***

The table 6 and table 7 above show the four OLS regressions that were run for the self-employed and wage workers. The regression was run for the mean, 25th, 50th and the 75th percentile for both employment statuses. There is heteroskedasticity in the distribution of data, it means that the variation of the dependent variable is not equal across independent variable‟s values. The dataset was divided into quartiles in order to see the results at each quartile of the distribution. The same procedure has been done by Hamilton (2000) in his paper. The R2 values and the number of observations are mentioned for each quartile and sector as well. For the self-employed on average the R2 indicates that the independent variables have 8.9% effect or variation on the dependent variable and in the case of wage workers it is 12% effect or variation on average which is higher than the self-employed. The

(25)

24

bigger the R2 the better it is for explaining the effect of explanatory variables on the dependent variable.

According to the OLS regressions we can tell that labour market experience has a positive relation with the earnings of the wage workers at the mean and increase the earnings by approximately 16 pounds for an additional unit of labour market experience. However we cannot conclude the same for the entrepreneurs due to insignificant p-values among all the quartiles for the self-employed. One of the reasons for this is the smaller sample size of entrepreneurs and most of the entrepreneurs are biased while filling in the survey questionnaires and most of the times abstain from filling in the right income to evade income taxes. On the other hand the labour market experience for the wageworkers has positive effect on the employees earnings and the relationship is significant for all the quartiles which represent that for wage workers the labour market experience matters more and most for the ones in the upper quartile, as larger effect could be seen there. Secondly the results for the OLS indicate that having a college degree increases the earnings for both sectors. Though in OLS entrepreneurs have higher reward as compared to the wage workers if they have a college degree or higher. However, in the median and upper quartile the wage workers have bigger effect on their earnings profile by having a college degree unlike entrepreneurs that do not have a significant relationship in the quartile regressions.

As explained above marital status is a categorical variable that have been added as dummies and the results in the table show that entrepreneurs and wage workers who are married and have a spouse earn more as compared to the individuals who are not married and do not have a spouse. The dummies for categorical variables are mentioned in the tables to see relationship of each marital status with the earnings. For both wage workers and the entrepreneurs the relationship between earnings and being single and cohabiting is negative and significant, indicating that individuals who are single and do not have a spouse earn less as compared to the ones with a spouse. Same conclusion stands for the widowed and separated employees as well. Furthermore the explanatory variable tenure in current job explains higher and positive earnings for both the wage workers and entrepreneurs as their time in current job increases. For entrepreneurs the results show that they will lose money if they are new in the business but if they manage to survive in the business for more than two years their earnings start increasing with respect to the increase in their tenure. The earnings of wage workers on average are higher than the self-employed if they are in the same job for more than 1 year and will always stay higher as the tenure in the current job increases.

(26)

25

However there are entrepreneurs in the median and upper quartile who have higher earnings than the wage workers if they manage to survive in the business for five years or more. In the OLS regression the years of education increase the earnings for both the entrepreneurs and the wage workers. However on average, for the wageworkers it has greater increase in the earnings as compared to the entrepreneurs. It is consistent with our result above that having a college degree or higher education has greater impact on the earnings of the wage workers. The same trend could be seen in (Hamilton, 2000) study that the education years matter more for the wage workers even though the entrepreneurs are more educated on average. Lastly the results show that earnings for entrepreneurs and wage workers have increased over the years while keeping year 2000 as the base year. However for year 2002 the earnings were in negative for both the sectors. The possible reason for this could be a significantly smaller number of observations in this year and due to any unobserved macro incident.

(27)

26

Hausman’s test:

Figure 2 below shows the results from the Hausman‟s test.

Figure 2: Hausman’s Test

To conduct this test we have to run regressions for the fixed effect and the random effects so the test can determine for us the best model that fits our dataset. The Prob>chi2 value of the Hausman‟s test is insignificant that tells us to use random effects model for our data set. This is because our null hypothesis was that the Ho: The random effects model is appropriate and the alternative hypothesis for the test was Ha: The fixed effect model is appropriate. Hence

Prob>chi2 = 0.2537 = 24.85

chi2(21) = (b-B)'[(V_b-V_B)^(-1)](b-B) Test: Ho: difference in coefficients not systematic

B = inconsistent under Ha, efficient under Ho; obtained from xtreg b = consistent under Ho and Ha; obtained from xtreg 2004.year 33.32814 56.0484 -22.72025 17.22429 2003.year 34.0098 41.07408 -7.064275 17.04653 2002.year -64.28805 -40.43939 -23.84866 21.18203 2001bn.year 24.73122 13.15943 11.57179 16.23847 8.jobtime 142.2869 139.4356 2.851372 49.15222 7.jobtime 84.58839 114.2829 -29.69448 50.20881 6.jobtime 71.41176 74.78708 -3.375316 50.13263 5.jobtime 39.76639 57.31161 -17.54523 51.79235 4.jobtime 12.81254 28.49101 -15.67848 56.20261 3.jobtime 33.58329 15.9627 17.6206 61.77015 2bn.jobtime 11.48581 -28.12701 39.61283 79.71584 1.degree 117.8125 165.8887 -48.07619 19.33777 7.marital -225.5496 -8.183822 -217.3658 155.636 6.marital 4.639688 12.19783 -7.558143 47.07614 5.marital -33.86379 -50.45724 16.59345 32.02403 4.marital -55.57106 -26.36712 -29.20394 68.64502 3.marital -99.06514 -104.1105 5.045373 23.27064 2bn.marital -48.40014 -51.09984 2.699706 21.25376 eduyears 25.03214 26.91249 -1.880343 2.87433 jobexpersq -.275445 -.3134666 .0380216 .0475335 jobexper 12.67176 15.14817 -2.476401 2.684391 stat -24.03423 -20.41887 -3.615352 19.16909 fixed random Difference S.E.

(b) (B) (b-B) sqrt(diag(V_b-V_B)) Coefficients

(28)

27

the Prob>chi2 value is below greater than 0.05 so we cannot reject the null hypothesis and will apply the random effects model in this paper.

Random effects model:

According to the results from the Hausman‟s test I used the random effects model on my dataset with the assumption that there is heteroskedasticity, meaning that the variance in the variables is unequal across the regression line and secondly individual specific effects are uncorrelated with the independent variables. This means that for example the individual innate ability and characteristics are not correlated with the explanatory variables such as labour market experience, marital status and job tenure etc. The test was run for all the observation of self-employed and wage workers combined. Table 8 below shows the results for the regression:

Table 8: Regression table for Random Effects Estimator

R-sq: within = 0.0795, between = 0.1190, overall = 0.1103

grossearning Coef. Std. Err. z P>z p-value [95% Conf. Interval]

stat -20.41887 10.71643 -1.91 0.057 -41.4227 .5849502 jobexper 15.14817 1.26104 12.01 0.000*** 12.67657 17.61976 jobexpersq -.3134666 .0234925 -13.34 0.000*** -.3595111 -.267422 eduyears 26.91249 1.631129 16.50 0.000*** 23.71553 30.10944 marital cohabiting -51.09984 11.45767 -4.46 0.000*** -73.55646 -28.64322 single -104.1105 11.3984 -9.13 0.000*** -126.451 -81.77005 widowed -26.36712 45.05029 -0.59 0.558 -114.6641 61.92984 divorced -50.45724 18.47574 -2.73 0.006** -86.66902 -14.24545 separated 12.19783 27.95937 0.44 0.663 -42.60153 66.99719 same sex couple -8.183822 61.12966 -0.13 0.894 -127.9958 111.6281 1.degree 165.8887 10.93792 15.17 0.000*** 144.4508 187.3266 jobtime 2-3 months -28.12701 39.07918 -0.72 0.472 -104.7208 48.46677 4-6 months 15.9627 30.71591 0.52 0.603 -44.23939 76.16478 7-11 months 28.49101 28.27664 1.01 0.314 -26.93018 83.9122 1-2 years 57.31161 26.04591 2.20 0.028** 6.262566 108.3607 2-4 years 74.78708 24.95259 3.00 0.003** 25.88091 123.6933 5-9 years 114.2829 25.26865 4.52 0.000*** 64.75723 163.8085 10 years or more 139.4356 24.66009 5.65 0.000*** 91.10267 187.7684 year 2001 13.15943 11.15038 1.18 0.238 -8.694908 35.01376 2002 -40.43939 13.47327 -3.00 0.003** -66.84652 -14.03226

(29)

28 2003 41.07408 10.8904 3.77 0.000*** 19.7293 62.41886 2004 56.0484 11.35373 4.94 0.000*** 33.7955 78.30129 _cons -82.1444 35.69954 -2.30 0.021** -152.1142 -12.17458 sigma_u 0 sigma_e 464.15654

rho 0 (fraction of variance due to u_i)

From the results above we cannot determine if the entrepreneurs earn less or more than the wageworkers. This is due to the insignificant p-value of the „stat’ variable coefficient that represents the self-employed. However labour market experience and the years of education gained have a positive effect on the earnings of the wage workers and entrepreneurs. From the table above it means that each year increase in the labour market experience leads to approximately 15 pounds more in earnings per week and an additional year of education effects earnings positively up to 27 pounds per week. We can deduce from the table above that education have major effect on the earnings of both entrepreneurs and paid employees. Higher the level of education attained, more the earnings for the individual would be Praag & Witteloostuijn (2009). The results show that individuals with college degree earn approximately 166 British pounds more than the ones with no college degree or some other lower qualification. The current job tenure variable has categories which are mentioned above in the table. The results from the regression state that individuals who stick to the same job for greater tenure their earnings get higher and increase by time. If an individual stick with the same job for about two years or more his earnings increase. There is a positive relationship between the earnings and job tenure of the individuals and there is a significant increase in the earnings if an individual is in the same job for more than 5 years. This shows us that larger the tenure in the same job, higher the earnings are for the entrepreneurs and the wage workers.

The random effects estimator results show that the people who are married and have their spouse present earn more as compared to the non-married and single people. Aforementioned in my data the variable for marital status has categorical variable and they were added as dummies in the regression to see what marital status has what effect on the earnings and it turns out to be that people who are single, divorced and cohabiting earn less than the married individuals with their spouse present (Hamilton 2000). Moreover I intend to see how the wages of the individuals have changed over the timespan in the data. The regression took year 2000 as the base year and the results are consistent with OLS regression and have the

(30)

29

same conclusion that the income has increased over the years except for year 2002 where there is a decrease. This could be due to lesser number of observations in that year or some unobserved external incident.

Lastly the R2 represents the variation for the dependent variable and the regressors. The „R2 within‟ tells us that there is 7.9% variation in earnings within individuals over time. The „R2 between‟ represent that there is 11.9% variation in gross earnings between the individuals regardless of time and „R2 overall‟ indicates that there is 11% variation in the earnings over time and individual in the data. The rho in the table represents the interclass correlation of the error which is the fraction of the variance in the error due to individual specific effect. In my case it is zero that tells us that the idiosyncratic error in the dataset is really high and there is no individual specific effect that the regression detected.

(31)

30

Discussion & Conclusion:

In this section I will be summarising my results and discuss how they relate to my research question and existing literature. There were several factors that were considered to explain the earnings differential between the self-employed and the wage workers. In my case the results are concluded to be a bit similar to what Hamilton (2000) proposed but still there are quite some different findings in my paper due to different dataset and measures of income used. For most of the self-employed the earnings at 75th percentile statistics show that their earnings are higher than the wage workers in most of the cases. However for the mean, 25th quantile and the median entrepreneurs have lower earnings than the wage workers. The variation in the earnings profile of the self-employed is much higher than the wage workers which is consistent with the results of Hamilton (2000) when Equity Adjusted Draw is used as the measure of income but that is not the case in my data as I used only gross weekly earnings from the main job and only the earnings at 75th percentile for the self-employed overtakes wage workers earnings rather than mean, median and 25th quartile. The reason for the variation is that entrepreneurs who are earning substantially high earnings and fall in the 75th and the 99th percentile of the self-employed distribution. As Parker (2009) mentioned that there is high proportions of the entrepreneurs earn both really low and really high earnings as compared to paid employees. However this is not the entire conclusion there are several other empirical results that tell a different story. Unlike (Hamilton, 2000) I used random effects estimator in my paper as well in addition to the OLS and quantile regressions. In my dataset the entrepreneurs on average have more labour market experience than the wage workers but it has a positive relationship with both sectors explaining higher labour market experience leads to higher earnings. From the empirical evidence we can conclude that it pays more to be a paid employee with higher labour market experience because for wage workers the labour market experience has higher degree of effect on average, median and upper quartile as well. This is against the investment and learning model that says that the experience in self-employment will overtake the alternative paid employment wages at one point (Roy, 1951; Jovaovic 1982) and supports the agency model (Lazar, 1981; Lazear and Moore 1984) that argues that the paid employees earnings would be steeper to avoid shirking which is not an option for entrepreneurs as there are no agency problems in entrepreneurship. Wage workers are tend to be more educated as compared to the self-employed and are more likely to have a college degree. Education proved to be the most

(32)

31

important explanatory variable in the analysis. The empirical results show that education has a positive and significant relationship with the earnings of both paid employees and the self-employed. Moreover it is the explanatory variable that affects earnings the most and plays the major role in varying the earnings in both sectors. Though the wage workers are more educated but on average education pays off more to the entrepreneurs, however in the median and upper quartiles wage workers benefit more from higher level of education.

Individual who are married and living with their spouse have higher earnings in the case for both entrepreneurs and the wage workers. The statistics show that the wage workers are more likely to be married as compared to the self-employed and this explains one of the reason as well why the wage workers have higher earnings on average. Individuals who are single, cohabiting, divorced, widowed or same sex couple have varied earnings among them but they all earn less than the married individuals at all quantiles. The results are consistent in OLS and random effect estimators for this explanatory variable. The evidence from the empirical results show that entrepreneurs and wage workers have positive increase in their earnings as their tenure grows in the same job. However in the beginning of the tenure both entrepreneurs and wage workers tend to have lower earnings and in case of entrepreneurs they tend to lose money when they start the business. If they manage to survive in the business for more than 2 years their earnings start increasing with respect to increase in tenure. This is against the argument of Astebro, (2012) who states that entrepreneurs earn more in the beginning of their tenure and start losing after 10 years in the same business. On the other hand the self-employed earnings never overtake the earnings of the wage workers at the mean distribution, this supports the argument form MacDonald, (1988) that argues that individuals who realise that they are not “rising stars” will quit the employment and eventually the self-employment sector will consist of a number of immature and inexperienced entrepreneurs and a few experienced and very successful ones. This argument is also consistent with the findings that the successful and experienced entrepreneurs in the upper quartile, who manage to survive in the business for more than 5 years earn higher than the wage workers and even higher if they are in the current business for 10 years or more.

With respect to what we have seen in the literature, results and discussion we can conclude that there are a lot of non-innate and non-pecuniary factors that determine people choosing self-employment and to see whether being self-employed pays more than wage workers or not? To answer the research question “Does entrepreneurship pays?” the findings tell us yes it does pay to be an entrepreneur and it also does not as well. It depends where does one fall in

(33)

32

the self-employed distribution. Being an entrepreneur only pays if one is in the upper quartile of the self-employed distribution. The rest of the entrepreneurs would be better off if they have wage jobs because according to the variables taken into account and assuming ceteris paribus, wage workers earn more than the self-employed in all the other cases except the 75th percentile of the distribution. Entrepreneurs who are successful, experienced and have been in the same business for more than 5 years only tend to earn more than the wage workers but there are really few of them in numbers and they fall in the upper most quantiles of the entrepreneurship distribution. On the other hand if we take the average of the earnings of self-employed, they are lower than the wage workers and they never overtake the earnings of the paid employees, regardless of job tenure and labour market experience. The differentials in the earnings between wage workers and self-employed would be even higher if we consider the fringe benefits that the paid employees enjoy from their employers such as paid vacations, insurance and company vehicle. These benefits may add up to 20% of the paid workers compensations (Hamilton, 2000). It is ambiguous to give them a monetary value hence they were not included in the analysis however they add up to the opportunity cost of not being a paid employee as an entrepreneur.

Beside all this above, individuals still want to enter into entrepreneurship and choose to stick with it despite lower earnings and lower earnings growth if you are a mediocre entrepreneur. This suggests that this behaviour is derived due to substantial non-pecuniary benefits such as being your own boss. Astebro (2012) mentioned that the entrepreneurs are more satisfied with their job than the wage workers because their job provides them more flexibility, autonomy, skill utilization and greater job security. Doing what you love to do and greater freedom are mostly responsible for the difference in job satisfaction between employees and self-employed. People entering into self-employment are happier and the ones who keep shifting between self-employment and paid employment are even more satisfied and happier (Astebro, 2012).

However apart from this discussion and conclusion there are limitations to this study. First, the superstar entrepreneurs are excluded from the dataset who were earning substantially high incomes that it was biasing the earnings distribution of the entrepreneurs. Secondly, there was no focus given to the difficulty of the work between self-employment and the wage employment. Thirdly, the economic situations and the macro effects were not taken into account that affects the earnings from the businesses but not the salaries of the wage workers. Moreover the compensations that are enjoyed by the paid employees are not taken into

(34)

33

account. Lastly the probability of survival in the business and a paid job is not taken into account. Future researches would be more valuable and would add more to this domain if they take these factors into account and taking a broader view of factors that could affect the earning differentials between entrepreneur and wage workers.

(35)

34

References

Astebro, T. (2012). Returns to entrepreneurship. The Oxford Handbook of Entrepreneurial Finance. Edt: Douglas Cumming, Oxford: Oxford University Press, pp. 45-108.

Berglann, H. (2009). Entrepreneurship: Origins and returns.Centre for Economic Policy Research. London:

Borjas, George J., and Bronars, Stephen G. (1989) „„Consumer Discrimination and Self-Employment.’’J.P.E. 97:pp 581–605.

Bradley, D., & Roberts, J. (2004). Self-Employment and Job Satisfaction: Investigating the Role of Self-Efficacy, Depression, and Seniority :Journal of Small Business Management (1st ed., Vol. 42, pp. 37-58). Blackwell Publishing.

Daly M., (2009). Another Look at the Returns to Trying Self-Employment. Doctoral Thesis, Chapter 1. Boston University.

Hamilton, B.H., (2000). Does entrepreneurship pay? An empirical analysis of the returns to self-employment. Journal of Political Economy 108 (3), 604–631.

Jovanovic, Boyan. (1982) „„Selection and the Evolution of Industry.’’Econometrica 50 :pp 649–70.

Lazear, Edward P. (1981)„„Agency, Earnings Profiles, Productivity, and Hours Restrictions.‟‟A.E.R.71:pp 606–20.

Lazear, Edward P., and Moore, Robert L. (1984) „„Incentives, Productivity, and Labor Contracts.‟‟Q.J.E.99 : 275–96.

Moskowitz, T.J. and Vissing-Jorgensen, A. (2002) The Returns to Entrepreneurial

Investment: A Private Equity Premium Puzzle, American Economic Review 92(4):pp 745-778.

Does entrepreneurship pay? :

Master‟s Thesis Msc Entrepreneurship 2015: