Self-employment and Education

(1)

1

Self-employment and Education

An analysis on the link between entrepreneurship, human capital and its other determinants

Rense F. Deen ~ 2059630 University of Groningen

January 2016

Supervisor: prof. dr. J.P. Elhorst Master Thesis

Msc. Economics

Abstract

Are highly educated more likely to become self-employed? Do smart people start their own business? This paper analyzes how the choice of individuals on whether to become self-employed or wage-employed depends on the education level and other determinants.

Past studies have so far find very contradicting evidence: support for all three possibilities (a positive link, a negative link, no link at all) has been found. Using American data from the General Social Survey between 2002 and 2014 the impact of general education on the propensity to become an entrepreneur is investigated in a binary setting. Unlike the majority of previous research, the possible endogeneity of education is accounted for. The empirical results show an insignificant ambiguous relation in most standard specifications, but once endogeneity is corrected for a significant relationship between the two is found for females. Additional, intergenerational links, marital status and race are found to be important determinants of self-employment.

JEL classification: D12, I26, I29, J19, J24, L26

Keywords: Self-employment, entrepreneurship, education, human capital, occupational

choice.

(2)

2

1. Introduction

Self-employment, or entrepreneurship, has long been considered one of the main drivers of our modern economy. Many studies have underlined its importance in the process of job-creation, innovation and economic growth. Already a century ago Schumpeter wrote about the ‘’creative destruction’’ caused by entrepreneurship (The Theory of Economic Development, 1934) and the new production processes, goods, et cetera it is accompanied by. This notion has led to many governments trying to stimulate self-employment through various policies and laws. Not only because of the general benefits for the economy, but also because it is viewed as a way out of unemployment and poverty.

There are quite a few different views regarding the questions ‘’why become self- employed?’’ and ‘’which factors does one consider when deciding on this matter?’’. The most straightforward viewpoint is to just look at it from a utility-maximizing perspective.

An individual simply follows a standard random utility model in which his choice in the labour market is based on the highest expected utility from different occupations. The individual will become self-employed if his expected utility in that situation is higher compared to the other options: becoming employed, or maybe even unemployed. At first glance one might expect the major determinant of the outcome to this problem to be the earnings-differential between the wage-sector and self-employment, but this is not true.

Through the years many determinants, sources of utility, have been identified through research. These range from standard individual characteristics to entrepreneurial related variables: the business environment, one’s desire for flexibility, existence of capital constraints, preference for autonomy and independence, the level of risk aversion, the support from family and friends (financially or otherwise), and so forth. Determinants like these, among many others, can influence how potential entrepreneurs perceive the returns to self-employment, and subsequently the probability that they will attempt it.

A particular concept that is often linked to entrepreneurship and which is well-

researched in the business-environment is human capital. The question often asked is

how important it is in determining the success of a business. How important is job-

training, or creativeness? Or a manager’s instincts, his inherent competitive edge? Once

that has been answered the next topic becomes to analyze how many of these qualities

can actually be taught to people; which factors are intrinsic to oneself and which can be

learned. Taking it another step further, not only business-specific specific training is

considered, but also general education. While this relationship between education and

(3)

3

entrepreneurial success is extensively researched, this paper will look at this from another angle: not self-employment success is studied, but the entry into self- employment.

The answer to this question is not a straightforward as one might think at first glance. One of the reasons of this are two popular but very divergent views of the self- employed. The first viewpoint, typically encountered in business literature, is one that sees the self-employed as the visionaries or mavericks. Independent entrepreneurs who drop out of school to grab an opportunity on the market. Ambitious workers with a certain ability or nick for trade who accept risk in return for a greater (potential) reward. The alternative view however has a more negative (or perhaps just more sober) perspective.

The self-employed may instead be discouraged wage workers who finds his or her offered wages dissatisfactory or his employment in the wage sector too sporadic, irregular. In the most extreme case self-employment is simply a way out of unemployment, if the individual is unable to find a job in the wage sector. As such, in the first theory mentioned above entrepreneurship is the optimal outcome of a person’s maximization problem, but in the second theory self-employment is the second-best solution. Understanding the differences in implications between these views is important. Rissman (2003) for example argues that policy-makers and researchers alike make no distinction between the two, and that they both claim all self-employed generate job growth, promote upward mobility and foment technological change. Consequently entire institutions and many tax codes have been implemented to encourage self-employment and entrepreneurship and increase the positive benefits they create. However, as he states: ‘’these alleged benefits may, in fact, be true for the entrepreneurs of the economy. However, for those self- employed who are discouraged wage workers, the benefits may not be as clear’’. Tax breaks and less restrictive capital constraints aimed at supplementing self-employment may be a much less than optimal solution for individuals to escape poverty than, for example, increasing their human capital or implementing policies that reduce job search costs.

In this paper the relationship between self-employment and education will be investigated in three ways: a short univariate analysis, a logit analysis with multiple specifications, and finally an instrumental variables approach will be employed to

correct for possible endogeneity. The two main hypothesis that we will try to answer are

as follows:

(4)

4

Hypothesis 1: ‘’the self-employed report a higher level of education than those who are wage-employed.’’

Hypothesis 2: ‘’a higher number of years of formal education will increase the probability of becoming self-employed compared to wage-employed.’’

The rest of the paper has the following structure: in section 2 past literature and findings regarding this topic is reviewed, section 3 summarizes and discusses the prevailing theories on the link between entrepreneurship and education, section 4 is about the exact methodology of the paper, section 5 elaborates on the (data) set being used, under section 6 the findings are reported, section 7 summarizes the conclusions and section 8 discusses some shortcomings and potential avenues for future research.

2. Literature Review

The relation between education and self-employment is often researched in the context of the return to investment in education. How much extra income is generated from an additional year of schooling or how or it influences the success of someone’s business.

However, other methods of researching this relationship are significantly scarcer. This includes the method adopted in this paper: comparing the two in the context of entry in a self-employment status. In this section I will discuss some of the papers that have nonetheless done so, among others that have not, and found very mixing results.

The literature on the returns to schooling and the longevity of newly-formed small businesses are quite unanimous: most find a positive relation between education and earnings/success. In general an increase of 7-13% in earnings for every extra year of education is found; given that someone is self-employed being more educated seems beneficial. The research in this area is a tricky subject and many contradicting results have been obtained. All three possible options (a positive effect of education on self- employment, a negative effect, or no significant effect) are all supported by various papers. As said, this is contrary to research on self-employment success.

Wu & Wu (2014) look at education and self-employment in Britain using a

maximum likelihood estimation. After constructing probit models, including industrial

and regional controls, and disaggregating the sample by male and female they find that

high levels of human capital (as measured by the degree the individual has attained) is

significantly positive for females, but completely insignificant for men. They explain that

(5)

5

this finding, female entrepreneurs being higher educated than their male counterparts, could ‘’reflect their increased self-assurance and, perhaps, ability to withstand gender discrimination’’.

Bates (1995) reports similar results. By stressing the importance of the difference between industries when it comes to self-employment and looking at this in detail in logit regressions, he also finds a striking difference between males and females on this subject.

In addition, he discovers a revealing difference between various industries. The impact of higher education in identifying self-employment entrants is highly positive in his ‘skilled services’ sub-sample, but highly negative for ‘construction’. Probability of entry into skilled services self-employment rises substantially at each of the higher levels of college education, contrary to construction where high school dropouts much more like to enter.

He concludes that this might be the reason why many other papers find no significant influence of education when looking at self-employment entry generally.

Goetz & Rupansingha (2014), using American data, research the growth in self- employment in rural and urban areas using mostly the ‘New Growth Theory’ (explained in the next section). They split their regional (dummy) variable not according to geographical location, but to agglomeration instead. Nine different subsamples are utilized, ranging from ‘’metropolis with more than one million inhabitants’’ to ‘’completely rural, far from any urban area’’. After controlling for a very extensive list of variables they find that a larger college-educated population share actually deters self-employment in small metropolitan areas. According to the authors, this could be due to the presence of university or college campuses where wage-and-salary employment tends to be more common. Moreover, having a college-educated population is apparently important for self-employment growth in the nonadjacent rural counties, but not for those that are adjacent to metropolitan areas. Being located next to a metro area may reduce the need to have college graduates in stimulating self-employment growth, simply because college graduates can more readily be hired from metropolitan areas in the non-metro adjacent areas. The authors stress the importance of the policy implications this result has. ‘’relying on or attracting more highly educated populations to induce self-employment may work as a strategy in metro non-adjacent areas, but not in those located next to urban labour markets’’.

Differing from the aforementioned literature, Dawson, Henley & Latreille (2009)

investigate the motivations behind people choosing for self-employment. Distinguishing

(6)

6

between ‘positive’ (or ‘prosperity-pull’) and ‘negative’ (or ‘poverty-push’) reasons they conduct a multivariate regression analysis using UK Labour Force data. They find a positive relationship between earning ‘A-level qualification’ (post-compulsory education, but below a university qualification). This is consistent with the finding that self- employment is higher amongst individuals who have undertaken some post-compulsory schooling, but who preferred to pursue vocational rather than professional skills/careers.

In addition, the paper provides more insight into the specific motivations behind individuals choosing self-employment, looking at it through various educational levels.

Those with high degree-level qualifications are more likely to report that self- employment was chosen because of ‘better working conditions’, ‘independence’, and

‘nature of their occupation’, while those without any formal qualifications are more likely to become entrepreneurs due to ‘no jobs available locally’, ‘joined family business’, and

‘more money’. These are considered the expected, straightforward results, with the unskilled being more likely to encounter greater problems finding work, and less likely subsequently to command a (relatively) high wage. The distinct effects of family business in the economy are also found here: those who entered self-employment because of a family business are much more likely to lack any form of formal education. However, the writers note though that the downside of their research is that only those who are self- employed are considered in this study, leading to a potential self-selection bias.

Taniguchi (2002) researches the determinants of self-employment of women specifically, adopting a view based on the Resource Theory of self-employment (this is expanded upon in the next section). She states that the effect of educational attainment on self-employment entry of women remains unclear, theoretically as well as empirically.

The theory suggests that education might aid in female’s business ownership to some

extent by providing contact information (for example because of attendance in small

business courses). However, ‘’formal education might matter less for entry into self-

employment than it does for transitions into the wage/salary sector where minimum

education requirements are more likely to play a key role. She also accentuates the

importance of work experience in this research, arguing that other studies have weak

measures of this variable, as it is likely to be highly correlated with education. Once this

is fully taken into account, formal schooling ‘’…will not affect the process of self-

employment as much as that of the wage/salary sector employment’’. Using data from the

National Longitudinal Survey of Youth (USA) she finds evidence that supports this.

(7)

7

Rees & Shah (1986) take a slightly different approach than the previously mentioned articles. They first estimate an earnings equation using the General Household Survey (UK) and subsequently use the differential of this equation between wage- and self-employment as the main determinant in explaining the differences in entry among these two groups. In their constructed structural probit equation they find that a higher education has a greater impact on employee earnings than on self-employment earnings, but nevertheless raises self-employment probability. This could mean education embodies two aspects of human capital, one of which is to increase productivity at work, and the other is to reduce variance of self-employment earnings. Unfortunately, providing evidence for this is beyond the scope of their analysis.

Kim & Cho (2009) are one of the many papers that find a negative relation between education and entrepreneurship. Utilizing South-Korean data in a cross-sectional Logit analysis they find a clear significant negative effect of the educational level on self- employment entry, thus supporting the push-theory of self-employment. They add that even if this is the case, conditional on someone being self-employed, their chances of survival are lower than that of well-educated people.

Hout & Rosen (2000) focus in their analysis more on intergenerational links and race and its effects on entrepreneurship, but also consider education; using a handful of dummy variables like the majority of other papers. After controlling for all other effects in a multivariate setting they find no evidence in any shape or form of the significance of education. However, they do find a very significant impact of the self-employment status of an individual’s father, on that individual. This result will be utilized in the regressions used in this paper.

Rissman (2003), employing a conditional fixed effects logit model using National Longitudal Survey of Youth (USA) panel data, identifies a difference of the influence of education between whites and non-whites. The latter are significantly less likely to choose self-employment, for each higher level of education they obtain. The author interprets this finding in the following way: ‘’as education increases and presumably opportunities available to Nonwhites become more available, the likelihood of entering self- employment declines’’.

As stated earlier, the research on self-employment success is far greater than that

on the actual decision of entrepreneurship. Nonetheless, part of the articles analyzing the

former also research the latter to some extent. This includes Robinson & Sexton (1994).

(8)

8

In their probit regression, using data from the U. S. Census of Population sample, they conclude that higher levels of education increase both the probability of becoming self- employed and the success of individuals in that sector in terms of earnings. They add however that a major shortcoming of their study, from which many others suffer as well, was the inability to study the effects of specific types of education/educational programs such as entrepreneurship or business school programs, as opposed to simple general levels of education. Similarly, Kangasharju & Pekkala (2001) focus on entrepreneurship success rather than entry, but still provide some insights on the matter. Using Finnish data in a logit model, their main result is that the exit probability is lower for the firms run by highly educated in an economic downturn, but it is actually higher in an economic upturn.

They attribute this to two specific reasons. First, self-employment is apparently not as attractive as wage employment for the highly educated, due to lower (and less secure stream of) earnings prospects. Second, the external demand for the labour faced by the highly educated is than those faced by the lesser educated in an economic boom.

Grilo & Thurik (2008), van der Zwan et al. (2010) and van der Zwan et al. (2007) conduct similar analyses on the determinants of self-employment, and how recent entrepreneurial exit relates to entrepreneurial entry. They provide a novel viewpoint by splitting up self-employment into respectively seven and six different stages of

‘’engagement levels’’, rather than just two (either someone is self-employed or not),

ranging from ‘’never thought about starting a business’’ to ‘’having an older, well-

established business’’ and ‘’no longer being an entrepreneur’’. The first papers finds that

education matters in triggering at least the thought of starting a business, even if the

thought is later abandoned. The second paper has a comparable conclusion: ‘’educational

attainment mainly distinguishes individuals without entrepreneurial engagement from

those having the potential to engage in entrepreneurship’’, but fails to find any significant

effect on actual realized self-employment. The third paper investigates the relation

between latent (the desire for) entrepreneurship and actual self-employment by

estimating a pair of probit equations simultaneously. In one actual entrepreneurship is

explained by latent entrepreneurship and explanatory variables, while in the other latent

entrepreneurship is explained by actual entrepreneurship (and a nearly identical set of

explanatory variables). They conclude that the level of education does not significantly

influence the willingness to be, or become, self-employed.

(9)

9

Summarizing, even though the importance of education is generally accepted the exact role it plays in the development of entrepreneurs is still a tricky subject, and no universally agreed upon conclusion has been reached. The contradicting evidence makes it so that merely depending on what method of analysis is used, evidence supporting all different views (a positive effect of education on self-employment, a negative effect, or no effect at all) could be found

¹

. This very much confirms the need for more research on the subject, so the specific relationship can be uncovered.

3. Background theory on self-employment, education & other determinants Since the research on the relationship between educational attainment and self- employment selection is so conflicting, many different theoretical arguments have been constructed to support the various differing results. In this section I will mention and briefly explain some of the more prevalent ones.

Utility maximization and the theory of income choice are two of the oldest and straightforward theories (at least partially) explaining why individuals would choose to become self-employed. In short, people become self-employed if it guarantees a more credible benefit rather than other labour market alternatives; the utility associated with the returns accruing from the two types of activity is higher for either of the two options.

Also known as occupational choice theory, this often boils down to comparing the net present value of the future earnings of each possible choice. According to Lucas’ (1978) old general equilibrium model, a person chooses either paid employment or management (entrepreneurship) based on his management skills. He showed that the most talented individuals, those with the highest innate entrepreneurial ability, completed their careers establishing successful enterprises, while the less talented remained wage employed.

Human capital theory of course expanded upon this: investing in an individual’s human capital, raising his/her ability, should then lead to a higher tendency to become self- employed. Many scientists have later questioned this notion greatly: in what sense can managerial abilities actually be taught? How great is the role that intrinsic, natural talent (the entrepreneurial edge) which cannot be taught plays? If one assumes it can, the next question becomes what kind of education makes this possible. Close to all research and

1A few other papers not mentioned that find a positive relationship are Livanos (2009), and Boden (1996); others that find no relationship at all include Allen & Curington (2014), Taylor (1996), and a last paper that reports a negative link are van Es & van Vuuren (2011).

(10)

10

theories focus on standard formal education, but one can argue that it could potentially only be specialized education, such as business courses and programs, that has any sort of influence. Van der Sluis et al. (2008) summarize the ambiguous nature the quite well:

''Not only may [a higher education make] enterprise profit rise, but the entrepreneur may also enjoy success in other dimensions: a larger firm, richer perquisites, higher socioeconomic status. In all, this increases the expected utility attached to entrepreneurship and thereby favors this occupational choice. Education now has a positive effect on both the choice of and the performance in entrepreneurship. Of course, if education is also allowed to raise the productivity of the individual as an employee, the effect of education on the choice again becomes ambiguous.''

A counter-argument here is provided by Lu (1999), who mentions the outside option value in this situation. It is argued that formal education matters less for entry into self-employment than it does for transitions into the wage/salary sector where minimum education requirements are more likely to play a key role. Startienè et al. (2010) supplement this view when looking at the differences between urban and rural areas in particular. They state that in more rural/agricultural areas with higher illiteracy/lower education levels those workers who are educated are scarcer, meaning education opens up more opportunities in wage-employment; i. e. the outside option value compared to self-employment increases.

As Kim & Cho (2009) note as well, the theory of income choice makes some bold but conflicting predictions. For example, the theory suggests that an economic depression will lead to an increase in start-up activity on the grounds that the opportunity cost of starting a firm as decreased, since finding wage employment becomes more difficult.

However, it is a fact that the unemployed tend to possess lower endowments of human

capital and entrepreneurial ability that are needed to set-up and sustain new businesses,

leading to believe an economic downturn may actually may imply a lower degree of

entrepreneurial activity. Evans & Leighton (1989) have expanded upon this by

introducing liquidity constraints and looking at personal wealth in more detail. According

to them economic depressions lead to lower levels of wealth, which reduce the probability

of becoming self-employed. The assumption that starting a new business requires a

certain level of capital and that less wealthy individuals may face constraints in acquiring

this needed capital seems logical, but may not be completely true. Hurst & Lusardi (2004)

show in their extensive research that the relationship between household wealth and

(11)

11

propensity to become an entrepreneur is highly non-linear. Most start-ups require low starting capital, and they subsequently find that only for the top 5% of Americans there exists a positive relationship between the two.

Another line of theories focusses more on the importance of risk. They simply see the entrepreneur as a person with a high risk-taking level. Individuals make the decision between the risky option of managing an enterprise, with higher expected/potential earning, or the less risky option of paid work in hired employment, with corresponding lower earnings. This is sometimes related to the portfolio choice theory, where it is assumed a person can freely combine self- and hired employment. In this setting entrepreneurship is also considered more risky, and persons are able to choose the most personally acceptable level or risk and acquire income by combining the two types of professions accordingly.

However, the finding of many studies that highly educated persons earn more as employees than they would do as self-employed (Kangasharju & Pekkala [2001] for example) is completely at odds with the aforementioned theories. Instead they emphasize the importance of the -currently very popular views- ‘push’ and ‘pull’ theories that I have briefly mentioned at the start of this paper. Although these do overlap with a couple of the other theories, they are still quite distinct. These schools of thought argue that personal professional choices are primarily determined by external uncontrollable forces from the outside, such as unemployment, business cycles or unattractive workplace conditions.

According to the ‘push’ theory individuals are pushed into self-employment due to unemployment (in a macro-setting: the unemployment rate in an economy). Rising unemployment reduces the opportunity for hired work as well as the expected income from hired employment, thus ‘pushing’ someone into self-employment so he or she avoids unemployment. In this situation it is simply the lesser of two evils, or a second-best solution until a good job-opportunity in hired employment presents itself. Expectedly, the

‘pull’ theory argues the opposite, and sees entrepreneurs instead as ambitious

opportunists who start their venture voluntarily. Prosperity, economic expansion,

improving market conditions and more easily obtainable credit makes individuals more

likely to start a new firm. Unemployment in this case is expected to have a negative

relationship with entrepreneurship. As expanded upon in the previous section, a plethora

of support for both of these views has been found.

(12)

12

In addition to all the views mentioned above, there are a few more specific, specialized/smaller, theories that tend to focus on only a single aspect or determinant of self-employment. These include Lazear’s (2002) jack-of-all-trades theory, sociological- psychological, intergenerational pick-up rate theory, New Home Economic models and Resource theory. Home Economic Models assume that individuals allocate their waking hours between familial and income-generating activities depending on the relative time spent in these areas of production. Taniguchi (2002) uses this to look at women specifically and argues that ‘’the boundaries between home time and market time may be less rigidly demarcated for the self-employed. […] self-employment would then be a more appealing form of work for those who desire more time flexibility in balancing family and career’’. The Resource theory predicts that those who have or have obtained more

‘resources’ relevant to entrepreneurship, such as cumulative work experience and business-related knowledge, are more likely to become self-employed. The jack-of-all- trades theory on the other hand claims that people with more than one skill or ability are more like to become an entrepreneur than a normal employee. In order to start a business one needs a wider range of knowledge than hired employees, and as such those with more diversified talents are more likely to do so. The intergenerational pick-up theory on the other hand argues that the tendency for people to become entrepreneurs is mostly inherited from their parents. People whose parents are self-employed (run a family business for example), are significantly more likely to become self-employed themselves.

Sociological-psychological theories are a last group of views that try to explain self- employment from a less economical perspective, and can generally be divided in ‘positive’

and ‘negative’ theories. The positive theories interpret self-employment as the need for self-expression, independence, status, pecuniary advantage, cultural expression, or simply a way to achieve someone’s personal goals. The negative views on the other hand see entrepreneurship is the outcome of discrimination, dissatisfaction with hired employment conditions, psychological discomfort, limited opportunities in the labour market (for minorities for example).

Contrary to all the contradicting predictions from the theories related to self- employment entry, almost every single theory discussing education and self-employment success/performance predicts a positive relation between the two. Kangasharju &

Pekkala (2001) however provide an important criticism to these, specifically regarding

the research done in this area. They agree with the statement that, in general, higher

(13)

13

education of self-employed people should improve the growth opportunities of their firms, since it improves the ability to understand market prospects and changes, allowing for better exploitation of the demand on the market. However, it is important to note that higher education does not necessarily increase the rate of survival. The authors point out that it is important to distinguish between the success of a firm, and the success of the entrepreneur operating the firm. Even though the individual cares about the success of his or her firm, he/she cares more about his/her own labour market success, whether as self-employed or as an employee, it does not matter. This means that the closing of a firm is not always a failure, but simply the result of a better job market offer for the owner.

There is an underlying assumption here though, namely that wage work is more attractive than self-employment. If it would not be, then higher survival rates for firms run by highly educated should be observed in every phase of the business cycle, but if the assumption is correct, then the survival rate would not necessarily be higher; for the aforementioned reason.

In addition, Kangasharju & Pekkala (2001) name another mechanism that can cause biased results in analyses on entrepreneurial entry and exit. They theorize that if it is assumed that firms run by highly educated people grow faster than those run by the less educated, there is a higher chance that these educated entrepreneurs start receiving their earning as wages instead of entrepreneurial profit/income; i. e. they become employees in the firm they own. In some datasets used by other researcher these individuals will disappear from the pool of self-employed, leading to an upward bias in the failure rates of highly educated entrepreneurs compared to others.

In summary, the theoretical and empirical literature are quite alike in the fact that

they both provide ample evidence and support for practically all viewpoints regarding

education and self-employment entry, unlike the fairly unanimous conclusion on self-

employment performance/success and education. This only further underlines the need

for additional research, to test which theories actually play a significant role and which

do not.

(14)

14

4. Methodology

Several methods to understand the relationship between education and the propensity to become an entrepreneur will be employed in this analysis, which I will briefly discuss in this section. They were chosen due to them being the most appropriate considering the dataset and necessary for examining the previously mentioned relationship. These methods and models are very common, especially in this area of research, and as such will not be discussed extensively, only in the context of how they are used in this specific situation. Using American cross-sectional data from the General Social Survey (which will be elaborated upon in the next section) I will start off with a univariate analysis, only looking at self-employment entry and education, to get some preliminary insights into the relation between. Next up multiple logit

²

models will be estimated. This is because, as previously mentioned, not self-employment success is examined (which is the more popular area of research) but self-employment entry. As such the dependent variable will be binary in nature, contrary to continuous. Lastly an IV-probit model will be estimated to account for the potential endogeneity of the education variable in the regression. Below I will expand upon these three ‘phases’ of analysis.

Preliminary analysis

In the univariate ‘phase’ primarily a few simple means and variance comparison tests will be performed to get an idea of the characteristics of the variables of interest. It should also allow us to give an answer to the first hypothesis. In addition, the dataset makes it possible to differentiate between genders and race, so it will be impossible to compare outcomes on these grounds as well. These tests include the Kolmogorov- Smirnov test, the Mann-Whitney test, the Brown-Forsythe-augmented variance ratio test, and the two-sample t-test. They will be discussed in the ‘’Empirical findings’’ section.

Logit estimation

Secondly, and more interestingly, logit regressions will be run to estimate the effect of education on the chance to become self-employed. In general logistic models are argued to have more straightforward interpretations of their coefficients (logit models have

2 In the literature both probit and logit models are utilized. As a robustness check I will also estimate the probit model. But, as is not unusual, the results differ only very slightly between the two.

(15)

15

logistically distributed error terms, while those of probit ones are normally distributed), but in the literature both are used; very often obtaining similar results. Additionally, later on I will estimate an instrumental variables probit function, whose results are obviously a little bit more possible to be compared to another probit, rather than logit, model. The full logit equation, which will be estimated using the maximum likelihood estimation, has the following (standard reduced) form:

𝑆𝑒𝑙𝑓𝐸𝑚𝑝

_𝑖

= 𝛽

₀

+ 𝛽

₁

𝐺𝑒𝑛𝑑𝑒𝑟

_𝑖

+ 𝛽

₂

𝐴𝑔𝑒

_𝑖

+ 𝛽

₃

𝐴𝑔𝑒²

_𝑖

+ 𝛽

₄

𝐸𝑑𝑢𝑐

_𝑖

+ 𝛽

₅

𝐸𝑑𝑢𝑐²

_𝑖

+ 𝛽

₆

𝑅𝑎𝑐𝑒

_𝑖

+ 𝛽

₇

𝑀𝑎𝑟𝑖𝑡𝑎𝑙

_𝑖

+ 𝛽

₈

𝐻𝑒𝑎𝑙𝑡ℎ

_𝑖

+ 𝛽

₉

𝐶ℎ𝑖𝑙𝑑

_𝑖

+ 𝛽

₁₀

𝑅𝑒𝑔𝑖𝑜𝑛

_𝑖

+ 𝛽

₁₁

𝑌𝑒𝑎𝑟

_𝑖

+ 𝛽

₁₂

𝑈𝑛𝑒𝑚𝑝

_𝑖

+ 𝛽

₁₃

𝑃𝑎𝑆𝑒𝑙𝑓𝑒𝑚𝑝

_𝑖

+ 𝛽

₁₄

𝑀𝑎𝑆𝑒𝑙𝑓𝐸𝑚𝑝

_𝑖

+ 𝜀

_𝑖

Where 𝛽

₀

is the constant, 𝜀

_𝑖

the error term, SelfEmp stands for whether someone is self- employed or not, Gender for gender, Age for age, Age² is age²/100, Educ is the years of completed education, Educ² is educ²/100, Race makes the distinction between whites and non-whites, Marital stands for if someone is married or not, Health encompasses if someone is in good or poor health, Region is a set of nine dummy variables for each of the USA’s nine major regions (east-north central, west-north central, east-south central, west- south central, middle Atlantic, south Atlantic, mountain states, Pacific states, New England area), Year are the group of year dummies, Child is the number of children of the individual, Unemp is the regional unemployment rate divided by the regional rate of job vacancies, PaSelfEmp measures if person’s father is or used to be self-employed, and MaSelfEmp does the same for the individual his or her mother. It should be noted that this full model will only be estimated at the end, other models which do not include all the variables will be estimated first. Model number 1 will include only the variables up to and including Child and excluding the quadratic terms for age and education; in Model 2 the regional and year dummies will be added; Model 3 adds the quadratic terms for age and education; Model 4 adds the U/V-ratio and Model 5 adds MaSelfEmp and PaSelfEmp.

Additionally, every model will be estimated separately for males and females. Only Model 1 (the base model) will be estimated with the full sample, every other one will drop Gender and thus be estimated twice.

Some other alternative specifications will also be used. Model 6 will model

education differently: instead of a continuous variable it will instead be a group of dummy

variables corresponding to each completed level of education (whether primary-school

(16)

16

was completed or not, high-school was completed, college or university was completed, et cetera). Model 7 will look at the importance of race in more detail: rather than just differentiating between males and females, the same will be done between white people and people of colour, this should give even more insight in the possible heterogeneity of the influence of education. Lastly, Model 8 estimates the full logit model.

Explanatory variables

The decision of the inclusion of the aforementioned variables is mainly based on past research and the limitations of the dataset. In the upcoming paragraph(s) I will shortly elaborate on this.

Age is used as a determinant in practically all previous literature, with various different argumentations. Although the sign of its coefficient differs among papers, its significance is undeniable. Some attribute a negative sign to the more reckless and less risk-averse nature of younger people which makes entrepreneurship more attractive, simultaneously arguing that older people might apparently prefer (work) stability more.

Contrary, other researchers instead relate age to experience, stating younger people prefer to first gain some experience in their field of work before starting a business themselves

³

. Additionally, persons close to retirement who got laid off often have a hard time finding a new job. For these people going into self-employment (even if just for a few years) becomes increasingly attractive. To capture this relationship more precisely recent literature occasionally includes a quadratic term for age. Allen & Curington (2014), Vejsiu (2011) and Taylor (1996) for example allow for this non-linear relationship in their regressions, and subsequently find a maximum at around age 45-50, meaning the propensity to become self-employed first rises with age, reaches a maximum and then decreases again.

Gender is another important determinant of self-employment. Firstly, the data shows entrepreneurship is a predominantly male-dominated affair and secondly, females appear to enter into self-employment because of significantly differing motivations than men. As such, following the advice of Allen & Curington (2014), Boden (1996), Robinson

& Sexton (1994) and many others I will split every regression based on gender. To name a few possible reasons: Taniguchi (2002) and Allen & Curington (2014) find that men

3 For example, this is quite common for doctors, dentists and similar professions, who begin their careers in wage employment and start their own practice at a relatively late age.

(17)

17

more likely to become self-employed due to their desire for wealth, independence and risk-attitude while females do so because of their preference for flexible hours or,

‘’…women with limited earning potentials will be constrained to a greater extent in their ability to balance career and family tasks because the cost of quality child care is more likely to exceed their pay’’; and thus opt for (part-time) self-employment instead. The former group of reasons seem to generally dominate the latter one.

Related to this are the Marital and Child variables. Generally it is found that being married and having children increases the probability to become self-employed. Having children often increases the preference for working from home and flexible working hours, while being married (and having another person in the household providing income) allows a person to enter in the more ‘risky’ self-employment and pursue his or her own interests more directly (rather than going into wage employment). Both these variable are thus expected to have a positive influence on the dependent variable. Lastly, Henley (2007) argues that lenders are more likely to lend money (for start-ups) to married individuals as they tend to have higher financial stability.

A variable for general healthiness is also included in the analysis. According to Rees

& Shah (1994): ‘’the longer hours and the greater responsibility associated with self- employment mean that less healthy people are likely to find it a more demanding’’. Wu &

Wu (2014) add that having a health problem that limits the ability of the individual to work considerably reduces the probability of the person entering self-employment.

Region is included as standard control variables to account for geographical differences in self-employment. Plenty of research find significant differences in entrepreneurship between various areas, and between rural and urban areas.

Unfortunately the dataset does not contain enough information to deal with geographical differences sufficiently, but the best that is available will be used.

Race is also included as a common control variable. It is found that in countries all

over the world, including the USA, ethnic minorities have a harder time entering into

certain occupations. However, it is not really clear if they find it harder to obtain paid

employment in certain areas or occupations relative to setting up a business, but it

appears worth taking it into account. Research using American data often distinguishes

between whites, blacks, Latinos and ‘other’, such as Houten & Rosen (2000) and Livanos

(2009), but in general no significant difference between the different ethnic minorities is

found. Hence, I will limit the variable to merely white and non-white.

(18)

18

The self-employment statuses of the individual’s parent(s) are included to test for intergenerational links. Wu & Wu (2014) find that whether or not someone is self- employed has a significant influence on the entrepreneurship decision of their offspring.

Especially father-son and mother-daughter links seem to be rather strong, which supports the view of family businesses being passed down from parent to child. Moreover, aside from physically inheriting a business children gain experience in the field beforehand and obtain the knowledge to set up and run their own business from living with parent(s) that are already self-employed.

The time dummies are included to capture the aggregate changes over time which, considering the total sample spans 12 years, could be considerate in size.

Finally, the potential influence of unemployment on self-employment is well- documented and already discussed earlier in this paper, so that will not be repeated. It is however important to note that, unlike the majority of the research on this topic, it is not merely the unemployment rate itself that will be used as an explanatory variable. Instead, Taylor (1996) his procedure will be followed, who argues that the unemployment rate in itself is not precise enough to give a good indication of the possibility and ability of individuals to find suitable (wage) employment. ‘’clearly the number of other individuals looking for work is important, but the amount of actual jobs available is just as relevant’’.

Dividing the number of unemployed by the amount of job openings/vacancies in the region should provide a better proxy for the competition for paid employment an average individual can expect to face in the job market; fewer job openings in the area given a certain amount of people looking for work should make becoming self-employed a more attractive option. Since this type of data is available for the United States, this variable will be constructed and added to the analysis.

Instrumental Variables approach

Grilo & Thurik (2008) note that, in the setting of entrepreneurial choice (i. e. self-

employment) and education endogeneity problem might arise. The world of self-

employment seems to be known for its endogeneity problems, or as Verbeek (2004) for

example states: ‘’it is often argued that many explanatory variables are potentially

endogenous, including education level’’. Surprisingly however, only very few articles take

this into account, with only Grilo & Thurik (2010) and van der Zwan et al. (2008) as

(19)

19

notable exceptions. As such, in the final part of the analysis in this paper an IV-probit model will be estimated as an attempt to resolve this (potential) problem.

There are a few possible reasons to suspect endogeneity, such as omitted variable bias. Ability is often mentioned when it comes to education: the unmeasured ability of a child could impact both the education the child receives and later in life could impact his/her career choice. The effect previously attributed to education could actually be because of this unmeasured ability. Although a bit more far-fetched, reverse causality could also be a problem. Being self-employed (or not) might increase the prospect of education in the future just as education might influence the self-employment decision, according to Bascle (2008). Similar to this is the notion that education could have a potential option value attached to it. It has been argued that more education actually

‘creates’ value by providing the option of even more (higher/advanced) education. In this situation standard models produce biased results as well, increasing the need for something such as an instrumental variables approach.

The variables used as instruments are the education level of the mother and father of the individual. Family background variables like these are popular in the research regarding the returns to schooling (where usually future income is the dependent variable). The argument goes that children’s ‘choices’ (which sometimes are not choices) are highly correlated with the characteristics of their parents. The social class of the parents influences the choice for education, but does not influence their future earnings directly. I deem it not too far of a stretch to extend this reasoning to the choice for entrepreneurship as well, and hence see the previously mentioned variables as good candidates for instruments. Still, as is to be expected, the appropriate tests will be carried out to test if they actually are relevant, exogenous and valid.

5. Data

The data used in the analysis comes from the General Social Survey (GSS). It’s an American

dataset which gathers data by conducting face-to-face interviews with a representative

sample of about 3000 adults living in the USA. It is a project conducted by the National

Opinion Research Center (NORC) at the University of Chicago and used to be repeated

every year until 1994, after which it (due to insufficient funding) changed to only be

conducted in even-numbered years.

(20)

20

The GSS has a few attractive properties to it compared to other (popular) microeconomic datasets. For one, it has an average response rate around 75 to 80 percent, which is considerably higher than most other social and political surveys achieve.

Contrary to other popular surveys such as the Current Population Survey and Census the GSS has relatively detailed characterization of the family background of the individuals, allowing us for example to use the self-employment and education choice of the father and mother in our analysis. One specific thing to keep in mind though is the potential ‘bias’

using only American data might lead to (unlike analyzing multiple countries at once, such as the OECD group), because According to Startiène et al. (2010): ''the effect of college graduation on the probability of selection into an entrepreneurial position is higher in the USA than elsewhere, implying either an educational environment that is more conducive for entrepreneurship development or better business conditions that attract more highly educated individuals''. This should be kept in mind.

Although the GSS can provide us with the microeconomic data, for the unemployment/vacancies-ratio variable macroeconomic data is needed. The United States Bureau of Labor Statistics is luckily able to provide this: both the data on unemployment and on job vacancies. The Local Area Unemployment Statistics (LAUS) programme produces monthly and annual employment, unemployment, and labor force data for regions and divisions, States, counties, metropolitan areas, and many cities, by place of residence. The Job Openings and Labor Turnover Survey on the other hand produces data on job openings, hires, and separations. Both these datasets make it possible to distinguish between farm and non-farm areas (allowing all the farm-related data to be excluded) and between the various regions of the U. S. This makes the data perfectly compatible with the data extracted from the GSS.

Unfortunately, not all variables of interest have data associated with them for

every year (mainly because not every single type of questions is asked every year in the

GSS), forcing me to drop a considerable amount of them. In addition, it has been argued

that the famous Dotcom bubble of the 2000’s has also had a very significant impact on

self-employment, causing a structural break in the data. This was mainly due to the fact

that way less private funding (for new business) and self-employment in general was

observed after the bubble burst (Rybczynsky 2015). As a result of the above

considerations and dropping observations with missing data the sample will be limited to

the interval 2002-2014. In addition all observations of individuals related to agriculture

(21)

21

are dropped. This is common practice in this field or research, as agriculture is viewed as a special case: the self-employment dynamics of farmers and related professions are seen as substantially different as those of others, and as a result are often left out of the analysis.

I will follow this viewpoint and drop them as well. In the end this all leaves me with 14.716 observations in total, which should still be more than plenty for a sufficient analysis.

Unfortunately the same individuals are not followed over time, so I am limited to a pooled cross-section instead of panel data.

In the upcoming paragraphs I will shortly specify the used variables more precisely. Every word or sentence in quotation marks is/are the literal questions or possible answers from the survey.

The definition of self-employment

So far I have avoided defining the self-employed, but I will do so here. Regrettably, there is no unanimously agreed upon definition for entrepreneurship. As said at the start of this paper, there are mainly two views regarding entrepreneurship: the occupational notion, where entrepreneurship is simply the creation of a new business that is owned by the creator (self-employment), and the behavioural notion, that sees entrepreneurship as the seizing of an economic opportunity. In this last view the entrepreneur is not required to be the owner of a business, but merely to be an individual that observes and acts upon perceived opportunities, as a form of arbitrage. As has probably become clear from me using ‘self-employment’ and ‘entrepreneurship’ interchangeably, I use the former definition.

In the GSS sample respondents are asked the following question: ‘’Are/Were you self-employed or do/did you work for someone else?’’. Those who are self-employed are coded as 1, those who are wage-employed are coded as 0.

Education

Educ measures the highest year of education completed, which is a number between 0

and 20. Educ²/100 is education-squared divided by 100, to try to capture the non-linear

effect education might have. As a robustness check we use an alternative measure of

education, since aside from the number of completed years the GSS also asks for the

specific acquired degrees. This is modeled as a set of five dummy variables, which take

the value of 1 if the degree in question is the highest the respondent has obtained, and 0

(22)

22

if it is not. The five categories are ‘less than high school’, ‘high school’, ‘junior college’,

‘bachelor’, and ‘graduate’.

Age

As with education, to account for possible non-linearities both the individual’s age and the square of his/her age divided by one hundred are used as explanatory variables. As is usual, only people aged 18-65 are considered, the rest is ignored. Sometimes it is argued people above 65 years of age should be considered as well, but this is only done for countries without a sufficient social security system for the elderly to rely on; in which case becoming self-employed actually becomes an attractive option (read: necessary). For the USA however, this effect is deemed irrelevant.

Gender

Only used in the very first model, gender is set to 0 for males and 1 for females.

Race

As discussed above, ethnic minorities are grouped together as ‘Non-White’ and set equal to 0. ‘Whites’ are set to 1.

Marital

Similarly to Race, the possibilities other than ‘married’ are grouped together. ‘divorced’,

‘widowed’, ‘separated’ and ‘never married’ are set to 0, while ‘married’ is coded as 1.

Health

Another dichotomous variable, set to 0 if the person is in ‘normal’, ‘good’ or ‘exceptional’

condition, and set to 1 if it is ‘poor’.

Child

The number of children the respondent has.

(23)

23

Region

A set of 9 geographical dummy variables, one for every regional division as used by the United States Census Bureau. They are ‘New England’, ‘Middle-Atlantic’, ‘East-North Central’, ‘West-North Central’, ‘South-Atlantic’, ‘East-South Central’, ‘West-South Central’,

‘Mountain’ and ‘Pacific’. A table of which states each group is exactly comprised of can be found in appendix A.

Year

7 dummy variables corresponding to the years the surveys were conducted: 2002, 2004, 2006, 2008, 2010, 2012 and 2014.

Parent employment status

Identical to the variable for self-employment of the individual himself. Set equal to 1 if the mother/father is self-employment, 0 if wage-employed.

U/V-ratio

Combining the data from the American LAUS and JOLTS programmes makes it possible to manually construct this variable for the 2002-2014 period. The unemployment ratio is monthly data available per state. Following Taylor (1996) the states are grouped together in the same way as the regional dummies, and for every year used in the regression model (2002, 2004, 2006, 2008, 2010, 2012 & 2014) the corresponding unemployment rate is calculated as the average value of the five years before said year. The same is done for the job vacancy rate. Defined as the number of non-farm job openings on the last business day of the month as a percent of total employment plus job openings it is divided by region and averaged (by months) over the five years prior the year in question. The U/V-ratio is then simply the calculated unemployment rate divided by the calculated job vacancy rate.

Table 1 summarizes the most important descriptive statistics of the data. The nine

regional and seven year dummies are left out since we are not really interested in the

values or means of those variable. The table speaks mostly for itself, but there are a couple

of noteworthy things. For one, unlike many other papers there actually are more females

than males being analyzed. Often females are dropped from the sample completely

(‘’because self-employment is a male-dominated affair’’, [Robinson & Sexton 1994]) or

(24)

24

miss-represented because of self-selectivity issues. This does not appear to be the case here. In addition, although there are more men than women self-employed, the difference between the sexes seems much smaller here than for the previous generation: the gap between self-employed fathers and mothers is much larger. This by itself does not say much however, and will be investigated further in the next section.

Table 1. Summary statistics

All Male Female

Variables Mean Std Dev Mean Std Dev Mean Std Dev

Self-employed 0.112 0.316 0.139 0.346 0.090 0.286

Age 42.62 12.62 42.33 12.70 42.04 12.55

Education 13.71 2.947 13.662 3.054 13.760 2.853

Gender 0.540 0.498 - -

White 0.746 0.435 0.764 0.425 0.731 0.444

Married 0.490 0.500 0.495 0.500 0.485 0.500

Health 0.031 0.174 0.030 0.170 0.033 0.178

Children 1.662 1.525 1.523 1.557 1.781 1.487

Mother Self-Emp. 0.067 0.250 0.064 0.244 0.070 0.255

Father Self-Emp. 0.170 0.375 0.174 0.379 0.166 0.372

U/V-ratio 2.152 0.701 2.155 0.704 2.150 0.699

Regions

New England 0.042 0.200 0.044 0.205 0.40 0.196

Middle-Atlantic 0.127 0.334 0.122 0.327 0.132 0.339

East-North Central 0.167 0.375 0.166 0.372 0.171 0.377

West-North Central 0.066 0.247 0.066 0.249 0.065 0.246

South-Atlantic 0.210 0.407 0.205 0.404 0.214 0.410

East-South Central 0.060 0.238 0.060 0.238 0.060 0.238

West-South Central 0.107 0.309 0.107 0.309 0.107 0.309

Mountain 0.079 0.269 0.078 0.268 0.079 0.270

Pacific 0.141 0.348 0.152 0.359 0.132 0.338

Education ~ binary

<High school 0.113 0.316 0.124 0.330 0.103 0.304

High school 0.511 0.500 0.505 0.500 0.516 0.500

Junior college 0.088 0.284 0.075 0.264 0.099 0.299

Bachelor degree 0.188 0.391 0.191 0.393 0.185 0.389

Graduate degree 0.100 0.300 0.104 0.306 0.096 0.295

Observations 14,716 6,776 7,940

Education mother 11.91 3.568 12.01 3.59 11.82 3.55

Education father 11.99 4.036 12.00 4.11 11.98 3.97

Observations 10,177 4,756 5,421

Notes: the values of the dummy variables for Region and Education (alternative) may not add up to 1 due to rounding. Unfortunately, data about the education of the parents is not available for every observation.

(25)

25

6. Empirical findings Univariate results

First off, self-employment and education are analyzed in a simple univariate framework.

Table 2 shows the results of the two-sample Kolmogorov-Smirnov test. This is used to test the equality of distribution functions of the two groups of self-employment regarding education. It is a decent starting point because it does not make an assumption about the distribution of data (since it is not completely certain the data is perfectly normally distributed). At the 1% significance level the hypothesis that the distributions of the two groups are equal is rejected, suggesting a significant difference between the two. Although by no means it indicates a direct effect of education on the two, it does suggest a relationship between the two.

Similarly, the two-sample Wilcoxon rank-sum (or the Mann-Whitney U-) test is employed to test for (in)equality in the medians of education between self-employed and not self-employed individuals. Table 3 shows the results of this test, which rejects the null- hypothesis of equality of medians of education at the 1% significance level. This suggests a probable relationship between the two, although again the nature of this relationship cannot be inferred.

To get an idea of the variances of the variables of interest the Brown-Forsythe statistic is computed. This test is a variation of the standard F-test that is usually used in this type of comparison in the sense that its results are more robust than the normal F- test since it makes less strict assumptions on the shape of the underlying distribution.

Table 4 shows the results of this test. At the 5% significance level the null-hypothesis that the variance ratio of the two groups regarding education is equal to 1 is rejected. This result leads into the final test, which is the standard two-sample t-test. Because the variances were shown to be different the t-test is performed while accounting for this.

The result can be seen in table 5. This test rejects at the 5% significance level the null- hypothesis that the means of education of both groups are the same.

The results of all four tests used so far all point in the same direction and do not

contradict each other: according to the data the self-employed and wage-employed do

differ significantly from each other when it comes to education. In fact, the test shows that

at the 99% confidence level the mean education of the self-employed is higher than that

one of the wage-employed, which confirms our hypothesis that ‘’the self-employed report

a higher level of education than the wage-employed’’. Still, this is only the beginning, in

(26)

26

the upcoming section(s) there will be looked at the precise relationship between education and entrepreneurship in more detail.

Table 2. Two-sample Kolmogorov-Smirnov test

Group D p-value

Wage-employed 0.046 0.002

Self-employed -0.010 0.754

Combined K-S 0.046 0.004

Table 3. Two-sample Wilcoxon rank-sum test Value Wage-employed observations 13061 Self-employed observations 1655

z-value -3.049

Prob > | z | 0.009

Table 4. Robust equal variance test (Brown-Forsythe)

Value Pr > F

Wage-employed observations 13061 -

Self-employed observations 1655 -

Alternative estimator 1 5.437 0.020

Alternative estimator 2 6.447 0.011

Degrees of freedom (1, 14714) -

Notes: the first alternative estimator replaces the mean that is used in the standard test with the median, while alternative estimator two replaces it with the 10%

trimmed mean

Table 5. Two-sample t-test with unequal variances

Group Obs Mean Std. Error Std. Dev.

Wage-employed 13,061 13.69 0.256 2.926

Self-employed 1,655 13.88 0.076 3.105

Combined 14,716 13.71 0.024 2.947

t-value -2.341

p-value 0.019

Notes: H

0

: difference = 0, H

alt

: difference ≠ 0

(27)

27

Regression results

Table 6 reports the findings of the base model, calculated using Maximum Likelihood Estimation (as specified earlier in section 4). The first thing of notice for the pooled regression is that gender has a large impact on self-employment. Boden’s (1996), Taniguchi’s (2002) and Allen & Curington’s (2014) findings that men and women differ strongly in their (self-)employment choice is supported. The general view of entrepreneurship being a mainly male-dominated affair is reflected greatly in the data, as males are much more likely to become self-employed than females. Thus, splitting the sample by sex for further regressions seems to be a well-supported decision.

The main variable of interest, education, shows a positive relationship between schooling and the tendency to be self-employed at the 5% significance-level, although only for males, not females. This supports the view(s) that the more educated individuals are those that have acquired (more of) the knowledge associated with starting a business, see the variance of potential earnings of self-employment reduced (Rees & Shah [1986]) and/or are better able to take advantage of arbitrage/business opportunities in the market; and are thus more likely to become self-employed. For females however we are unable to find the same result.

Age is found to be an important factor for self-employment as well. For all three groups it is positive and significant at the 1%-level, indicating that the older someone gets the greater the probability of self-employment. It would be interesting and probably more accurate to incorporate the age someone actually become self-employed, rather than just looking at whether someone is self-employed at a certain age or not, but unfortunately the data does not allow for this. Nonetheless, it indicates that the theory of entrepreneurs being young opportunistic visionaries/mavericks might be flawed.