Employment: Evidence From Switzerland
by
A.R. Weijling
A thesis submitted in partial fulfillment of the requirements for the degree of
MSc. Entrepreneurship
Authors argue that entrepreneurship is a career choice that does not pay, perhaps except for the entrepreneurial superstars. This study aims to replicate Hamilton (2000) by investigating earnings differentials between male, working-age self-employed and wageworkers using fifteen years of panel data from Switzerland. The main findings from OLS and quantile estimators confirm existing evidence. The results suggest entrepreneurs to suffer an earnings discount in comparison to their salaried counterparts at the median and .25 quantile. Controlling for unobserved effects such as entrepreneurial ability, self-employed are found to earn CHF5,350 less annually. The empirical evidence is commonly associated with non-pecuniary benefits, a preference for the right tail of the positively skewed earnings distribution and potential overconfidence.
VU University Amsterdam
&University of Amsterdam
2015Supervisor: Philipp Koellinger Student no.: 2542745 (VU); 5933811 (UvA)
TABLE OF CONTENTS
1. INTRODUCTION ... 3 2. LITERATURE ... 5 2.1MEASUREMENTISSUES ... 8 3. DATA ... 9 3.1MODEL ... 12 3.2VARIABLES ... 13 3.3SUMMARYSTATISTICS ... 14 4. EMPIRICAL RESULTS ... 174.1OLSANDQUANTILEREGRESSION ... 17
4.2FIXEDEFFECTSREGRESSION ... 22
5. DISCUSSION ... 26 5.1CONCLUSION ... 27 5.2IMPLICATIONS ... 28 5.3LIMITATIONS ... 29 REFERENCES ... 30 APPENDIX ... 33 ACKNOWLEDGEMENTS
This study was carried out as a final fulfillment for achieving my master’s degree in Entrepreneurship. I thank Philipp Koellinger, my thesis supervisor, for his excellent comments and helpful thesis meetings. Philipp inspired me to make the most out of my final academic piece of work by demanding high-quality and promoting independent thinking. I also thank my fellow thesis students for the insightful discussions we had, the pleasant collaboration and how we motivated one another when the deadline seemed to approach all too fast. In addition, I thank João Santos Silva for pointing me in the right direction of conducting quantile regressions.
1. INTRODUCTION
Earning an income is arguably an important reason why people choose a certain occupation. Since the 1970s, the choice for an entrepreneurial occupation is one made by an increasing number of people in most of todays developed economies (Wennekers, van Stel, Carree, & Thurik, 2010). One would therefore expect self-employed to earn at least the same as wageworkers do, given they face risks in developing their ventures employees don’t face in their working careers (Levine & Rubinstein, 2013). What is puzzling however is that several studies conclude entrepreneurs don’t get rewarded for those risks, in that their earnings are lower than that of salaried wageworkers. For example, Hamilton (2000) conducted an extensive study to investigate the returns to entrepreneurship using 1984 panel data from the Survey of Income and Program Participation (SIPP). Hamilton (2000) uses three measures of earnings in self-employment, finding that mean hourly earnings for entrepreneurs are lower than that of wageworkers. Earnings profiles of self-employed never reached the same level and were generally flatter than those of employees, except for the 75th percentile (Astebro, 2012). Evans and Leighton (1989) found similar results in their study, using data from the 1966 National Longitudinal Survey of Youth (NLSY). In addition to reporting self-employed to have lower returns than employees, Evans and Leighton (1989) show prior wage work experience to be of more value in wage work than in self-employment. In contrast, other researchers have found opposing evidence while using different data. For instance, Fairlie (2005) finds entrepreneurship to be a better career choice than wage work for disadvantaged males, having both higher mean and median earnings than employees. While most authors don’t include women in their sample for labor market participation reasons, Fairlie (2005) finds young women to earn less than wageworkers. A study by Tergiman (2009), replicating Hamilton’s (2000) study shows entrepreneurs don’t earn less than wageworkers, but their earnings profile is in fact U-shaped. Tergiman (2009) finds self-employed yield higher returns for the first few years of tenure, after which wageworker returns take over. Ultimately though, returns to entrepreneurship again exceed that of employees, which according to Astebro (2012) is consistent with the 75th percentile Hamilton (2000) shows is earning more than wageworkers. Despite mixed evidence on the relative self-employment income, the general consensus for the U.S. is that entrepreneurs earn less than their paid-employment counterparts. Given that entrepreneurs on average work longer hours, their income position might be even worse than reported here – according to Aronson (1991) their income was at a 70% disadvantage in comparison to wageworkers. Other international evidence is also inconclusive, pointing in the direction of a premium for self-employed in some countries while others show findings similar to the U.S. For instance, self-employed individuals in the Netherlands
and Germany are found to earn more than those in wage work (Blumberg & de Graaf, 2004; Constant & Shachmurove, 2006). This is in contrast with empirical evidence from the U.K., where mean and median earnings of self-employed are below that of employed people, as reported by Taylor (1996). It appears no clear evidence can be found to show entrepreneurs earn less than wageworkers, or the other way around, even while sometimes claimed by authors. Earnings differentials between wageworkers and self-employed appear to be country-specific and highly dependent on measurement methods used. Also, due to several measurement issues results from income comparisons might be biased. For instance, entrepreneurs are known to under-report their income and over-report expenditures for tax reasons. Pissarides and Weber (1989) developed a method to estimate this under-reporting and found entrepreneurial households to underreport income by 35%. There are other important issues regarding measurement which will be dealt with in the next section. Also, incorrectly dealing with cases that report income both from self-employment and wage work may cause additional difficulties in analysis (Astebro, 2012; Parker, 2009).
In summary, a multitude of studies reporting empirical evidence on earnings differentials between wageworkers and self-employed can be found in the literature. Given the fact that many of these studies focus on U.S. data, this study aims to replicate the research done by Hamilton (2000) in a different setting of time and space, using data from the Swiss Household Panel (SHP). While there is European evidence on self-employment earnings (Clark & Drinkwater, 1998; Robson, 1997), it’s not always consistent with results from the U.S. For instance, Clark and Drinkwater (1998) report median income for self-employed in the U.K. to be above median wage income, albeit this holds only for an ethnic minority of Chinese. Robson (1997) reports similar evidence, finding an average premium for self-employment of 1.6% over paid employment, also using U.K. data. This study therefore aims to update and add to the existing empirical results in the field of entrepreneurship research. Similar comparisons as those made by Hamilton (2000) will be made, including pooled OLS and quantile regressions. In addition, 15 years of data from the SHP dataset, which contains panel data from 1999 onwards, will be analyzed while the main contribution of Hamilton (2000), using multiple income measures, couldn’t be replicated due to data limitations. Nevertheless, the empirical results may add to the understanding in both scientific and practical terms, as according to the author’s best knowledge the returns to entrepreneurship haven’t been studied before in Switzerland. Furthermore, the educational system in Switzerland is unlike that of many other countries in that a significant proportion of the population chooses for a vocational education, while education is often regarded as an important explanatory variable. Since little empirical evidence shows an earning advantage for entrepreneurs, this paper aims to extend current
knowledge about this topic, by investigating whether generally accepted claims also hold in another setting. Moreover, while most authors, including Hamilton (2000), do not examine fixed effects regression models this study accounts for unobserved heterogeneity in estimating earnings differentials between self-employed and wageworkers. The goal of this study therefore is to add to the state of knowledge of the supposed earnings differential between wageworkers and self-employed. As many authors note, it is surprising to recognize many people choose for an entrepreneurial career while apparently they would be better off to remain in or enter paid employment. However, as Parker (2009) correctly states, the evidence for this claim is not unambiguous, although neither is the claim entrepreneurs earnings-wise systematically fare better. Therefore, this study replicates Hamilton (2000) using panel data from Switzerland to investigate whether consistent evidence can be found in another country and in doing so, enrich the understanding of this topic.
The remainder of this paper continues as follows. The next section contains a literature review that describes existing evidence on the earnings differential of entrepreneurs compared to wageworkers as well as measurement issues related to this topic. Section 3 presents the data source used for this study, construction of the income measures and explanatory variables, the statistical methods applied and summary statistics. Section 4 provides empirical results of the analysis and relates that to existing theory, discussing potentially contrasting or confirming findings. Section 5 concludes and discusses limitations and implications for both science and practice.
2. LITERATURE
Self-employment takes up an increasing share of the labor force in many of today’s industrialized economies, a trend that started around the 1970s (Bernhardt, 1994; Blau, 1987). While rates of self-employment are increasing, an important question in the entrepreneurship literature concerns the extent to which the income of self-employed is comparative to that of wageworkers. The well-known study by Hamilton (2000) shows that individuals are better off choosing for an occupation as employee instead of becoming an entrepreneur, except for the 75th percentile. Using 1984 SIPP data, Hamilton (2000) analyzed 8,771 male individuals of the working age (18-65), excluding both farmers and professionals such as lawyers as the income of the former is dependent on subsidy programs while income of the latter is mostly top-coded. In addition, women are excluded to prevent possible labor market participation issues, leaving Hamilton (2000) with male small business owners. Using three methods of measuring income for self-employed, Hamilton finds mean and median hourly earnings to be lower for self-employed than paid
employees. For only one of the three measures the top 25% of the self-employed income distribution earn more than wageworkers. Not only were median and mean earnings lower, the self-employed in addition showed lower earnings growth, ultimately creating a negative 35% earnings differential for entrepreneurs who were in business for ten years (Parker, 2009). While Hamilton (2000) focuses on white males, Borjas and Bronars (1989) present racial income differences of self-employed, finding the returns of black entrepreneurs are lower than that of white entrepreneurs. A limitation of this study however is the fact that it only uses 1980 cross-sectional data, which doesn’t allow for time-series analysis. Using longitudinal data from 12 National Longitudinal Survey of Youth (NLSY) waves between 1966 and 1981, Evans and Leighton (1989) show self-employed to have lower log earnings than wageworkers. The main finding of their study however is that individuals having wage experience are better off in wage work than in self-employment, as it is rewarded higher in wage work. Some caution should be taken in comparing their results to others however, as their sample is limited to men aged 39 years old at the maximum. Farmers and professionals are not excluded but controlled for in the OLS regressions (Astebro, 2012).
Not all research points in this direction – using an updated NLSY sample, Kawaguchi (2003) shows that while the earnings differential between self-employed and wageworkers is positive at first. At zero years of tenure and experience self-employed earn around 35% more than their salaried counterparts but this quickly decreases and ultimately turns negative after 10 years of experience and 10 years of tenure. Interestingly, Tergiman (2009) has opposing results, showing a U-shaped earnings differential with self-employed actually having higher hourly earnings after 40 years of tenure, contrasting Kawaguchi (2003). Women and professionals were dropped from the 2001 SIPP sample as well as individuals reporting an occupation both as entrepreneur and employee. Being self-employed people earn more for the first 6 years of tenure, when the earnings for employees rise above those of the self-employed. After around four decades, entrepreneurial income rises sharply above that of employees again – comparable to the top 25% of the income distribution by Hamilton (2000). Evidence also exists of entrepreneurs earning more than wageworkers, as Fairlie (2005) reports higher mean and median income for the former. Fairlie, 2005 focuses on disadvantaged families using NLSY data, excluding military, students and those who report spending less than 1400 hours per year on their business. According to Fairlie’s (2005) research, young men from disadvantaged families – defined as those in which both parents have less than a high school education – earn more than employees do both at the mean and median, although this doesn’t hold for young women from disadvantaged families. Similar results can be found in a more recent study by Levine and Rubinstein (2013), also reporting an income premium for entrepreneurs over salaried workers. Levine and Rubinstein (2013) argue that even while the
traits and abilities of self-employed are the same as wageworkers, a large part of literature remarkably indicates they earn less. By differentiating the sample of NLSY and Current Population Survey (CPS) data into incorporated and unincorporated entrepreneurs, they are able to distinguish “true entrepreneurs” from the casual business owner like the hot dog vendor (Glaeser, 2007). Using this proxy for entrepreneurship they report average and median hourly earnings of incorporated self-employed to be 48% and 28% higher, respectively. In turn, the unincorporated entrepreneurs earn less per hour than salaried employees. Levine and Rubinstein (2013) conclude the incorporated have distinct characteristics and “tend to be male, white, better-educated, and more likely to come from high-earning, two-parent families” (p. 31) that account for their higher returns in self-employment.
While the previous findings were based on US data, several other studies present international evidence from European or other industrialized countries, which will be discussed in the following section. Using data from the British Household Panel Survey (BHPS), Taylor (1996) aims to estimate the combined effect of three reasons to become an entrepreneur, one of which is earnings. Taylor (1996) states mean earnings to be lower for the self-employed than for the employed, while Ajayi-Obe and Parker (2005), using the same BHPS dataset, report self-employed males and females to earn more than wageworkers, with exception of self-employed males that have no employees. Blanchflower and Shadforth (2007) find similar results as Taylor (1996) does, finding median weekly income of self-employed is £84 lower than that of employees. It should be noted however this is one of the few studies that hasn’t performed a regression estimation to draw comparisons. Evidence from other industrialized countries also shows mixed results: in Canada, average self-employment earnings were up to 53% higher than mean wage income in 1981 (Bernhardt, 1994), similar to how Australian (Kidd, 1993) and Spanish entrepreneurs earn less at the median than their salaried counterparts do, as found by Albarran, Carrasco, and Martinez-Granado (2009) using 1985-1997 panel data from the Spanish Family Expenditure Survey (ECPF). Substantial variation is also found within OECD countries: while people are mostly better off in employment, earnings-wise (OECD, 1986), Dutch and German entrepreneurs apparently fare better than wageworkers, although this differs between ethnic groups (Constant & Shachmurove, 2006; Constant, Shachmurove, & Zimmermann, 2007).
In summary, most studies find empirical evidence that self-employed tend to earn less than wageworkers, although differences can be observed between datasets. While most U.S. data points in the direction of a discount for entrepreneurs, other international studies also provides evidence for a premium in entrepreneurship, for example in Germany, Canada and the Netherlands (OECD, 1986). In general though, mean, median and 25th percentile entrepreneurs have lower income than
wageworkers, whereas from the third quartile onwards the self-employed overtake paid employment income. This may reflect the existence of superstar entrepreneurs in the higher ends of the income distribution, which is positively skewed across almost all datasets. Based on these findings, becoming an entrepreneur apparently is a choice that doesn’t pay, although there are several non-pecuniary reasons to become self-employed as well. These may include higher levels of autonomy and flexibility or a preference for the skew, where individuals aim for the rewards found in the right tail (Hundley, 2001).
2.1 MEASUREMENT ISSUES
There are several challenges related to comparing estimates of earnings from entrepreneurship to earnings from paid employment. One well-recognized difficulty is that of potential underreporting of income by self-employed, who are apparently wary of reporting their income to third parties for tax evasion reasons. According to Pissarides and Weber (1989) entrepreneurs underreport household income by around 35%, found using income and household expenses data from the Family Expenditure Survey. Hurst, Li, and Pugsley (2014) find evidence along the same lines, stating even when they are assured of confidentiality entrepreneurs still underreport their income by 25% on average. Using the log-linear Engel curve on two datasets (the Consumer Expenditure Survey and the Panel Study of Income Dynamics), Hurst, Li, and Pugsley (2014) identify differences in food expenditures by self-employed and salary workers, which serves as potential evidence for income underreporting by entrepreneurs. While there are multiple studies showing how entrepreneurs underreport income to tax authorities as well as in household surveys (see for instance Andreoni, Erard, and Feinstein (1998), or more recently Feldman and Slemrod (2007)) and Parker (2004) emphasizes it to be one of the most serious problems in data from employment earnings, according to Cumming (2012) the effect on earnings distributions of self-employed is negligible.
Other difficulties in analyzing differences may arise from data collection and manipulation efforts, starting with deciding whom to include in the sample of entrepreneurs. Although Hamilton, (2000) excludes farmers as their income might be dependent on government subsidies, Evans and Leighton (1989) retain people in the agricultural sector, professionals and military personnel, finding self-employed to earn less than wageworkers. Fairlie (2005), in studying self-employment earnings of both men and women from disadvantages families, reports this only holds for men while Borjas and Bronars (1989) studies racial differences, observing income from non-whites to be lower than whites. Differences in sampling methods might affect these disparities in earnings profile comparisons across studies, as Astebro (2012) suggests. In addition to excluding farmers,
Hamilton (2000) excluded highly paid professionals such as lawyers whose income might suffer from top-coding, Top-coding, or setting a maximum number for earnings is another measurement issue that might impact mean estimates. Related to top-coding is bottom-coding, in which case negative income is set to zero or minimum values are used. Both practices cause bias in the means and Parker (2009) therefore recommends researchers not to apply these coding principles. Dealing with income reported by survey respondents might also become problematic in those cases when entrepreneurs report income both from wage work and self-employment, when self-employed only report income from employment or vice versa. Dropping these observations from the analysis may seriously impact sample size, as is evident from Fairlie (2005), where more than half of the panel reported inconsistent income. A third measurement issue lies in the distinction of self-employed into incorporated and unincorporated entrepreneurs. Contrary to most studies, Levine and Rubinstein (2013) disaggregate self-employed into these two categories and find incorporated entrepreneurs to be better educated and earn more than both wageworkers and unincorporated business owners. The problem however resides in correctly regarding their employment status: as acknowledged by Blau (1987) and Evans and Leighton (1989), incorporated business owners are “frequently classified with paid workers” (Bernhardt, 1994, p. 277), as is the case for the US Current Population Survey.
3. DATA
The data used for this study comes from the Swiss Household Panel (SHP), a yearly panel study that is focused on observing changes in the social characteristics of the Swiss population and their representation. Data collection for the study started in 1999 and has been conducted yearly among a randomly drawn sample of households and household members aged 14 and over. The SHP consists of three samples, with the first one drawn in 1999, the second one in 2004 and the most recent one in 2013. For the first sample, 5,074 households representing a total of 12,931 individuals were interviewed. The second sample added 2,538 households and 6,569 individuals while the last one added another 4,093 households and 9,945 individuals to the sample (FORS, 2014). The questionnaire covers a broad array of topics, among others including household composition, tenancy, standard of living, health, education, employment and financial situation (Voorpostel, et al., 2014). With a few exceptions, for instance due to temporal differences, the questions in the panel study are the same each year. Any changes in the panel questions did not pose any major problems for this study, at most requiring recoding or merging of some variables. For this paper, data from the year 1999 up to and including 2013 was used, only including men
aged between 18 and 65 years old. Women were excluded from the sample to avoid any labor market participation problems. Where Hamilton, 2000 and others (e.g., Evans and Leighton (1989), Bernhardt (1994) focused on white males, no race data is available in the SHP dataset, making it impossible to exclude blacks and other non-whites1. Non-whites are often excluded for labor force participation reasons, as Borjas and Bronars (1989) show in their study of racial differences regarding rates of entrepreneurship and earnings.
Respondents were asked in their first interview whether they were an employee in a public or private firm or self-employed, and were asked whether any changes occurred in their occupational status in subsequent interviews. A number of respondents indicated income from both paid employment as well as self-employment (1,320 respondents). Depending on the occupational group they answered to be belonging to (either employees or self-employed), the income from the corresponding group was taken into account in the analysis. The resulting sample consists of 24,726 observations of employees and 7,153 observations of self-employed, or 28.9%. Demographic characteristics of both wageworkers and business owners are given in table 1, showing entrepreneurs on average are older, are more often married and have a higher level of potential labor market experience when compared to employees.
Table 1: Summary statistics for the year 20132
Mean
Variable Employees; Self-Employed; t-statistic
name Description N = 1,499 N = 277 (two-tailed)
age Age of person 42.40 47.90 -7.816 ***
civstatus Marital status 0.55 0.64 -2.887 **
potexp Potential labor market experience 26.60 30.10 -5.152 ***
educat Highest level of education achieved
(0) Lower than vocational schooling 0.09 0.06 2.451 **
(1) Vocational education & training 0.34 0.37 -1.012
(2) High school 0.23 0.26 -1.370
(3) University of Applied Sciences 0.14 0.12 0.824
(4) University 0.21 0.19 0.528
Note: * = p < 0.05; ** = p < 0.01; *** = p < 0.001.
1 The reasoning to exclude non-whites from the analysis mainly stems from the fact they often have other incentives to move into self-employment, also linked to labor force participation issues. As these differences between white and non-white entrepreneurs can distort the results and analyzing the inequalities stands too far from the main objective, the original idea was to exclude them. Unfortunately, the SHP does not collect information on race.
2 These summary statistics are for the most recent year (2013) in the SHP only, to prevent any bias that might arise from different numbers of observations for each individual. While some might have participated in every wave of the panel, some might have participated only once. In table 1 in the appendixsummary statistics across all years can be found.
Independent t-tests were run to compare mean differences across the employment sector groups, of which the t-statistics are presented in the last column of table 1. Table 2 shows results of both the Levene’s test for equality of variances, which was conducted first, as well as the t-test results for every variable. With exception of education, the patterns found are similar to those as reported by Hamilton (2000), with entrepreneurs more likely to be married and having a higher level of labor market experience. The fraction of entrepreneurs in the SHP dataset having an education lower than vocational schooling or at high school level is significantly different from the fraction of paid employees, while it shows no statistically significant difference for the other levels3. In other words, wageworkers and self-employed in the SHP are roughly equally educated, while Hamilton found entrepreneurs to be better educated. Given these results, potential earnings differences between wageworkers and self-employed might at least partially be affected by the fact that these two groups have different traits. In addition, what is noticeable is that the fraction of people having a vocational schooling as their highest education seems relatively high. However, this is typical for the Swiss education system, where a substantial number of people opt for vocational education and training (VET), often a beneficial choice in terms of earnings and unemployment rates (Fazekas & Field, 2013; Bachmann, 2012).
Table 2: Independent t-test results between employment sectors for the year 2013 Levene’s test for equality of variances T-test
t-statistic
Variable name F-statistic p-value (two-tailed) p-value
age 38.688 p < 0.001 -7.816 p < 0.001 civstatus 57.177 p < 0.001 -2.887 0.004 potexp 9.433 0.002 -5.152 p < 0.001 educat (0) 18.648 p < 0.001 2.451 0.015 (1) 3.556 0.059 -1.012 0.312 (2) 7.331 0.007 -1.370 0.171 (3) 2.800 0.094 0.824 0.411 (4) 1.151 0.283 0.528 0.598
3 While tables 1 and 2 may suggest only the fraction of entrepreneurs having an education lower than vocational schooling is significantly different from wageworkers, this also holds for the high school education level. The reason this isn’t evident from table 1 or 2, is due to the fact these are statistics for the year 2013, so to remain consistent with only reporting the most recent year. However, given these demographics aren’t expected to vary much over the years, t-tests for comparing means were also conducted across all panel years (tables 1 and 2 in the appendix). In this case, the fraction of high school educated individuals was significantly different between employment sectors, and hence is reported as such in the text above.
3.1 MODEL
The main analyses are done using a model that takes the usual form of a set of demographic characteristics and variables relating to labor market experience to estimate the dependent earnings variable. Let 𝑌!"# represent the earnings of individual 𝑖 with employment status 𝑗 at time 𝑡, given by
𝑌!"# = 𝑋!"𝛽!+ 𝑓! 𝐸𝑋𝑃𝑅!"# + 𝜖!"#, 𝑗 = 0, 1 (1)
where 𝑋!"𝛽! is a vector of observed individual demographic characteristics, 𝑓! 𝐸𝑋𝑃𝑅!"# is a vector of job experience variables and 𝜖!"# is an observed random error term. Employment status 𝑗 indicates whether someone is a wageworker (𝑗 = 0) or self-employed (𝑗 = 1). The individual demographic characteristics taken into account are age, age squared, a dummy for civil status and educational attainment. The vector of labor market experience includes potential labor market experience and potential labor market experience squared, as opposed to others (for instance, Braguinsky & Ohyama (2007) and Tergiman (2009)) that include both experience and tenure variables. Adding the tenure variable however resulted in non-significant regression coefficients, which is why it was left out of the earnings equation. Squared terms were included for both age and experience to test for a quadratic relationship between earnings and these specific variables. While people tend to have greater earnings in proportion to their increasing age, this effect might become non-linear from a certain age-level onwards, and negative thereafter. As people grow older, health limitations may start to impact their working skills and their earnings may reach an inflection point, a point at which earnings start to decrease, typically toward someone’s retirement age. The same principle applies to potential labor market experience, as workers might become stuck in the same patterns, which inhibits them from adapting to changing conditions, affecting their earnings growth. While the terms of a regression are generally to be interpreted independent of other terms, this is not the case for quadratic terms, where both the non-quadratic and quadratic terms are based on the same variable. In this case, a change in the age variable is accompanied by a change in age squared, which allows for calculating the instantaneous rate of change to examine the proposed curvilinearity. In order to obtain robust test results from the earnings regressions, clustered standard errors are applied to the OLS, quantile (Parente & Silva Santos, 2016) and fixed effects estimators. Given panel data is used in this study, each cross-section of a respondent is seen as a cluster of observations, with multiple observations per individual over time. Since the standard errors will be independent between groups of observations but correlated within, the standard errors have to be clustered to correct for this. Applying these clustered standard errors helps in meeting the homoscedasticity assumption.
3.2 VARIABLES
Efforts were made in selecting and creating the explanatory variables to follow Hamilton (2000) as closely as possible. In creating the potential labor market experience variable however, a different approach had to be taken, as Hamilton sets it equal to age – education – 6, where 6 is the generally used compulsory schooling age. However, due to the nature of the Swiss education system those already in the labor force often choose to extend their skills through vocational training. While the SHP contains data on when individuals completed their highest education, this disallowed computing a proper length of education, causing heavily overestimated education lengths and underestimated potential labor market experience. For instance, some individuals indicated to have finished their highest level of education at the age of 50. As compulsory schooling in Switzerland starts at age 6, this would imply 44 years of education, as the SHP provides no information on education starting years. To overcome this problem, potential labor market experience was created using information on the number of years spent in paid work and regular job starting age. First, the number of years working was computed by subtracting the age at which one started a regular job from the respondent’s current age. The age of 15 years assumed to be the lower limit at which someone can start a job, given compulsory schooling in Switzerland takes 9 years according to the UNESCO International Standard Classification of Education (UNESCO, 1997). Respondents could also indicate the number of years they spent in paid work. Whenever the difference between this amount and the number of years worked was between -10 and 10 years, it was replaced with the average of both measures. Outside this interval, the differences became too large and were deemed unrealistic given age or employment status.
A tenure variable representing the length of time in the current job or business had to be created as well. While the panel data allowed for construction of this variable, there was only information regarding tenure available for a limited number of respondents. As such, plugging the tenure variable into any of the regressions decreased the sample size for self-employed so much that the achieved results became meaningless and not statistically significant. Tenure was therefore left out of further analysis.
Data limitations hindered construction of multiple income measures, like Hamilton (2000) did, since the SHP does not contain data on business profits, draw or equity. The available information on income from dependent (wage) and independent (self-employment) work was used to construct both hourly and yearly earnings variables, of which summary statistics can be found in tables 2 and 3, respectively. Since income was reported both on a yearly and monthly basis, the monthly income was used to construct hourly earnings, since it’s not dependent on the number of
months per year the income has been received and therefore allows for better comparisons. Monthly income was firstly divided by four to arrive at a weekly salary, assuming 4 weeks of holidays per year (in this, SHP guidelines and Swiss statutory requirements (Ray, Sanes, & Schmitt, 2013) were followed), and next by the number of contractual hours. In addition to the number of contractual hours, respondents indicated the number of hours worked per week, which was used to compute hourly earnings when the number of contractual hours was unavailable. A check was performed on those who indicated to be working full-time yet reported contractual hours below 37 hours per week (the hour minimum for full-time work), as it caused hourly earnings to fluctuate heavily. For these people, the actual number of hours worked per week was used instead, which fitted better within yearly patterns. Next, those for whom hourly earnings computed using contractual versus actual hours worked differentiated negatively were verified and adjusted where necessary. Positive differences were not corrected, as this indicated actual hours to be lower than contractual hours and was deemed irrelevant since contractual hours was taken as starting point4. Given the size of the dataset, the top 0.1% (over CHF407 per hour; 57 observations) of hourly earnings was manually verified.
3.3 SUMMARY STATISTICS
The results shown in table 3 show the differences in hourly earnings of employees and wageworkers for the most recent year of the SHP. The rationale for this is the same as that of table 1. Hourly earnings statistics across all years can be found in table 4 in the appendix, although they show the same pattern as the one presented here, with the mean being higher for the self-employed while the .25quantile and median earnings are lower compared to employees. At the mean level, self-employment returns are 9.1% higher than mean paid employment salary. These characteristics are also displayed in the distribution of earnings shown in figure 1, which reveals the self-employment earning distribution to be more skewed than that of the wageworkers’. This is emphasized by the 25th and 50th percentile being 37 and 13 percent lower than those of the employees. The majority of the self-employed earn less than wageworkers do, while there is a long right tail indicating there are at least some entrepreneurs that are rewarded better than employees. While the maximum hourly salary for paid employees is around CHF230, the top hourly income for self-employed goes well beyond that and is almost three times higher. While the earnings distribution across all years show similar results, it is less skewed and the earnings at the 75th
4 In addition, in most cases this meant actual hours to be only 1 hour less than contractual hours or actual hours to be around 1 hour, while contractual hours was around 40. While the former would have only minimal impact hourly earnings, the latter was disproportionate after comparing to relevant observations.
percentile are almost equal (-CHF0.30 for self-employed), while for the most recent year entrepreneurs earn a premium of 8.5% at the .75th percentile.
Table 3: Hourly earnings for the year 2013
Hourly Earnings
Statistic Employees; N = 1,568 Self-employed; N = 247
Mean 44.62 48.71
Standard deviation 20.18 68.33
25th percentile 33.32 20.85
50th percentile 42.82 36.90
75th percentile 54.79 59.45
Figure 1: Kernel density plot of hourly earnings for the year 2013.
0 .005 .01 .015 .02 .025 D e n si ty 0 50 100 150 200 250 Hourly Earnings Wageworkers Self-Employed
Table 4: Annual earnings for the year 2013
Annual Earnings
Statistic Employees; N = 1,589 Self-employed; N = 229
Mean 95,857 102,943
Standard deviation 53,452 92,880
25th percentile 65,000 51,000
50th percentile 91,900 83,200
75th percentile 121,100 132,900
Figure 2: Kernel density plot of yearly earnings for the year 2013.
These findings support the results found by Hamilton (2000), reporting comparable earnings distributions. While the numbers vary slightly across the three measures of self-employment earnings used by Hamilton (2000), the overall pattern is the same with the first and median quartiles being lower for self-employed. The third quartile is higher when the EAD measure is used, although for the net profit and draw measure the 75th percentile is lower for entrepreneurs. It is important to note the results presented here might not be entirely comparable to the EAD and net profit measures reported by Hamilton (2000), while the draw is a more equivalent earnings measure as it is the income the owner extracts from it’s business (Astebro, 2012). In contrast to the mean and .75 quantile of the draw measure, being lower for self-employed than for wage income, the findings
0 2.0e-06 4.0e-06 6.0e-06 8.0e-06 .00001 D e n si ty 0 150,000 300,000 450,000 600,000 Yearly Earnings Wageworkers Self-Employed
reported in table 3 show higher mean and third quantile earnings. The dispersion shown by the relatively large standard deviation, .25 quantile and median are comparable to the draw measure.
In addition to hourly earnings, the SHP also provides data on yearly income, for which the same analysis was conducted. The results found when using yearly income data are qualitatively identical to those when hourly earnings were used. Figure 2 plots the distributions of yearly earnings for self-employed and paid employees while summary statistics for both are presented in table 4. Both display the same distribution when compared to the hourly earnings. Mean and .75th percentile annual earnings are higher for entrepreneurs whereas first quartile and median annual earnings are distinctly higher for the wageworkers. The same patterns can be observed in the analysis of the PSID (Panel Study of Income Dynamics) and KLIPS (Korean Labor and Income Panel Study) by Astebro (2012): the self-employed earn a premium compared to wageworkers at the mean, earn less at the percentiles in the lower ends of the income distribution and earn more at the percentiles in the higher ends of the income distribution.
4. EMPIRICAL RESULTS
4.1 OLS AND QUANTILE REGRESSION
The parameter estimates from OLS and quantile regressions using annual and hourly earnings are presented in tables 5 and 6 respectively. The quantile regressions allow for comparing the effects of the explanatory variables across the quantiles. Given the skewness of the income distributions, the regression coefficients of the quartiles are expected to vary from the mean, represented by the OLS in the second column. As expected, the magnitude of the coefficients increases toward the upper quartiles for the annual earnings, except for age, with the change being stronger for the self-employed (found in panel B). Interestingly, the coefficients tend to decrease for the hourly earnings estimates as they progress toward the higher end of the distribution of the wageworkers, while this doesn’t seem to apply for self-employed (panel A and panel B of table 6). Even while there is a tendency for a decrease, it is small. Moreover, overall the parameter estimates are found to vary less across quantiles for the salaried employees than for their self-employed counterparts, although this mainly holds for the educational attainment variables. This is suggested by the roughly equal estimates when comparing OLS and quantile regression results. It is line with expectations and similar to what Hamilton (2000) presents, as it reflects “the relative lack of skewness of the wage distribution” (p. 615). Although insignificant, the coefficients for the constants are considerably higher for entrepreneurs than for wageworkers for both yearly and
hourly measures, meaning that, ceteris paribus, self-employed start out at a higher mean income level than paid employees do. This is in contrast to the results of Hamilton (2000), who finds that it differentiates across the self-employed earnings measure used, with the wage constant being higher only for the EAD and net profit measure. With regard to the draw measure however, which is the closest indicator to entrepreneurial earnings, the results presented in tables 5 and 6 confirm Hamilton (2000). This inconsistency might therefore be attributed to the different earnings measures used and the skewness of the earnings distributions, which already indicated differences at the mean. The effect of labor market experience is shown by Hamilton (2000) to be greater for those in paid employment than for business owners and of similar size across both OLS and quantiles regressions. Similar to Tergiman (2009), a disparity to these results are exhibited in the coefficients presented here and labor market experience is found to have a greater effect on yearly earnings of self-employed instead. While the difference is positive in favor of employees for hourly earnings, the magnitude is not similar across the OLS and quantile regressions. The educational attainment variables support the results presented by Hamilton (2000), Evans and Leighton (1989) and Fairlie (2005), having in general the same sign. The quantile regressions however suggest a flatter increase for those in paid employment than for entrepreneurs, both across quantiles and educational attainment levels. This is interesting given the contrast to Hamilton (2000), reporting a significant difference for employed workers. However, this might be explained by the nature of the Swiss education system and labor market, where labor market entry is relatively easy through apprenticeships and vocational trainings (Hanushek, Woessmann, & Zhang, 2011). Employers can trust vocational credentials to represent well-skilled workers, as companies often participate in defining training curricula (Shavit & Müller, 2000), as opposed to the U.S. where the schooling system is characterized by high schools and colleges. The majority of the sample respondents that indicate it to be their highest level of education achieved, which is 39% for wageworkers and 40% for self-employed, confirms the popularity of the vocational schooling. The disparity in earnings, commonly found to be negative for entrepreneurs, is also evidenced from the empirical results presented, albeit perhaps to a lesser extent. The considerable number of people choosing for an entrepreneurial career, despite scholarly findings that in general the returns to entrepreneurship are lower in self-employment than in paid employment (Astebro, Braunerhjelm, & Broström, 2013; Hartog, Van Praag, & Van der Sluis, 2010), has therefore also been explained by non-pecuniary benefits. A multitude of studies have identified these non-pecuniary benefits such as flexibility in work schedules, autonomy or independence; (Aghion & Bolton, 1992; Blanchflower & Oswald, 1998; Duncan, 1976). For instance, Moskowitz and Vissing-Jorgensen (2002) state that over one fifth of a 1992 census survey among business owners reported “being one’s own boss” as the key
reason to enter into entrepreneurship. Being independent as an entrepreneur provides them with higher degrees of self-determination and the freedom to make their own decisions, whereas on the opposite side people in paid employment are obligated to do what their manager asks from them. Frey, Benz, and Stutzer (2004) introduce the notion of procedural utility to be a non-pecuniary benefit of being self-employed. Procedural utility refers to how people value the procedures, conditions and processes that lead to outcomes, and not only put value on the actual outcome (Benz, 2007). While often neglected (Benz & Frey, 2008), the idea proposed is that persons in independence (self-employed) achieve higher levels of procedural utility than those in hierarchical organizations (employees). Using panel data from the U.K., Germany and Switzerland, Benz and Frey (2008) find greater levels of job satisfaction, which is used as a proxy for procedural utility, among entrepreneurs than among wageworkers after controlling for pay, work hours, tenure and other factors. Their results confirm independent employment is regarded as being a source of job satisfaction, with scores of across the three countries being highest in Switzerland5. Indeed, among others, Hamilton (2000) allocates the difference in earnings between wageworkers and self-employed to non-pecuniary benefits. The empirical results presented are in support of this notion, providing evidence for the rationale why self-employed remain in business despite the apparent gains that are to be made when they would switch to paid employment6.
5 A score of 8.10 on a scale of 0 – 10, with a standard deviation of 1.72.
6 Hamilton (2000) finds these gains to be dependent on business tenure and potential labor market experience, even though the alternative starting wage is always higher than earnings in entrepreneurship. No such distinction is made in this study due to the lack of a tenure variable.
Table 5: Parameter estimates of Annual Earnings Regression Variable OLS .25 .50 .75 A. Wageworkers (N = 18,653; R2 = 0.2892) Constant -21,040 -17,204 -10,312 13,153 (12,098) (11,110) (7,834) (14,555) age 1,544 1,207 887 50 (865) (824) (574) (1,017) age2 1 -5 9 28* (11) (10) (7) (14) civstatus Married 10,821** 9,017** 7,863** 8,947** (1,608) (1,325) (1,139) (1,498) educat Vocational 11,502** 11,646** 18,505** 14,510** (2,068) (1,966) (1,417) (3,075) High school 25,644** 21,994** 32,524** 31,547** (2,406) (2,225) (1,728) (3,332) UoAS 36,051** 33,242** 41,984** 41,493** (2,832) (2,538) (2,208) (3,731) University 51,322** 38,903** 52,099** 58,422** (3,680) (2,973) (2,460) (4,202) potexp 2,729** 2,594** 2,308** 2,579** (521) (490) (372) (517) potexp2 -65** -52** -59 -75** (10) (10) (7) (13) B. Self-Employed (N = 5,213; R2 = 0.1065) Constant 65,932 -34,877 -6,144 35,457 (101,053) (32,022) (25,312) (35,557) age -4,681 2,749 719 -1,736 (7,298) (2,362) (1,863) (2,533) age2 67 -39 1 36 (85) (29) (24) (32) civstatus Married 19,488** 13,513** 12,947** 16,577** (4,858) (3,550) (2,811) (3,606) educat Vocational 3,692 6,947 14,076** 16,674* (5,740) (5,258) (3,549) (6,795) High school 19,937** 11,347 26,164** 34,904** (6,754) (6,491) (4,860) (7,257) UoAS 27,240** 24,391** 36,929** 47,422** (7,029) (7,930) (5,609) (7,107) University 91,502** 35,578** 60,388** 103,399** (24,122) (12,580) (7,603) (15,825) potexp 6,290 1,824 2,500* 3,952** (4,065) (1,471) (1,038) (1,420) potexp2 -129 -24 -53* -85** (77) (30) (23) (31)
Table 6: Parameter estimates of Hourly Earnings Regression OLS .25 .50 .75 Variable A. Wageworkers (N = 17,592; R2 = 0.2355) Constant 8.32 -2.73 1.55 11.26* (5.93) (3.55) (4.01) (4.66) age -0.06 0.32 0.29 -0.00 (0.42) (0.26) (0.27) (0.32) age2 0.01* 0.00** 0.01 0.01** (0.01) (0.00) (0.00) (0.00) civstatus Married 2.87** 2.01** 2.47 2.93** (0.65) (0.45) (0.44) (0.56) educat Vocational 6.59** 9.59** 8.55 5.60** (0.89) (0.80) (0.99) (1.08) High school 12.77** 13.51** 13.53 12.19** (1.02) (0.83) (1.05) (1.23) UoAS 17.54** 18.60** 18.52 17.22** (1.15) (0.90) (1.15) (1.30) University 23.32** 21.03** 22.80 22.66** (1.49) (1.16) (1.28) (1.71) potexp 0.88** 0.83** 0.78 0.86** (0.22) (0.15) (0.17) (0.21) potexp2 -0.03** -.02** -0.02 -0.03** (0.00) (0.00) (0.00) (0.00) B. Self-Employed (N = 4,584; R2 = 0.0876) Constant 39.67 -13.10 -1.58 14.43 (44.18) (7.85) (9.96) (17.55) age -1.74 1.29* 0.64 -0.11 (3.10) (0.56) (0.73) (1.33) age2 0.03 -0.01 -0.00 0.01 (0.04) (0.01) (0.01) (0.02) civstatus Married 7.62** 3.85* 3.55** 4.47** (2.72) (1.60) (1.30) (1.47) educat Vocational 0.81 8.49** 8.77** 6.28** (5.67) (1.90) (2.44) (2.25) High school 9.76 9.49** 13.25** 14.50** (6.14) (2.18) (2.74) (2.67) UoAS 10.26 15.47** 18.86** 18.41** (6.26) (2.93) (2.88) (2.78) University 33.64** 23.59** 28.72** 38.75** (10.90) (2.72) (3.33) (6.60) potexp 0.62 -0.25 0.19 0.57 (1.62) (0.34) (0.42) (0.75) potexp2 -0.03 0.00 -0.01 -0.02 (0.03) (0.01) (0.01) (0.02)
4.2 FIXED EFFECTS REGRESSION
In addition to OLS and quantile regression, a fixed effects regression model is conducted to account for unobserved heterogeneity, or any unobserved factors in the error term that are fixed over time and within clusters of individuals, but affect 𝑌!"#. These time-constant factors are represented by 𝛼! and equation (1) can be rewritten to:
𝑌!"# = 𝑋!"𝛽! + 𝑓! 𝐸𝑋𝑃𝑅!"# + 𝛼! + 𝜖!"#, 𝑗 = 0, 1 (2)
These unobserved factors may for instance include risk preferences, autonomy or ability and while they are constant for the same individual over time, they may be different from one individual to the next (Wooldridge, 2013). The Hausman test (Hausman, 1978) is used to determine whether the time-varying coefficients achieved through either a random effects or fixed effects model show any statistically significant difference. The results of the Hausman test are presented in table 7.
Table 7: Hausman Test Results
Coefficients Difference
Variable βFE βRE ( βFE - βRE ) Standard Error
age -925 2,338 -3,263 1,018 age2 -17 -1 -16 5 Employment status Self-Employed -5,350 -5,495 145 215 civstatus Married 3,084 4,565 -1,481 439 educat Vocational 12,736 11,631 1,105 1,255 High school 9,869 16,590 -6,721 1,639 UoAS 15,512 24,160 -8,648 1,852 University 25,919 41,192 -15,273 2,271 potexp 7,554 2,963 4,592 1,007 potexp2 -69 -78 9 4
The Hausman test statistic is given by:
𝐻 = (𝛽!" − 𝛽!")! 𝑣𝑎𝑟(𝛽!"− 𝛽!")
Under the null hypothesis, the Hausman test statistic follows a χ2-distribution with𝑘 degrees of freedom, where 𝑘 is the number of endogenous variables. This allows equation (3) to be written as:
𝜒! = (𝛽
!"− 𝛽!")′[𝑣𝑎𝑟(𝛽!" − 𝛽!")]!!(𝛽!"− 𝛽!") (4) If the null hypothesis can’t be rejected, the difference between the coefficients of the fixed effects and random effects models will be relatively small, since both will be consistent estimators. In addition, under the null hypothesis the variance of the random effects model will be smaller than the variance of the fixed effects one. As a mathematical consequence, equation (3) will have a small numerator and a large denominator, resulting in a small test statistic. In contrast, whenever the difference between both estimators’ coefficients is large, so is the numerator in equation (3). This implies the test statistic 𝐻 will be large as well and thus the null hypothesis might be rejected. The test results indicate the assumption for the random effects model are violated, rejecting the null hypothesis of equal coefficients, which is why the fixed effects model is used (χ2 = 110.29; p < 0.001)7. This implies there is covariance between the fixed effect 𝛼! and 𝑋!" that is unequal to zero. Given the test results presented in table 7, this most probably applies for the variables age, educational attainment and potential labor market experience.
The results of the fixed effects regression model are presented in table 8. Firstly, these results will be compared across the employment sectors, after which a comparison will be drawn with respect to the pooled OLS regression. The most interesting coefficient is the one indicating employment status, which, given the fixed effect estimator uses time-demeaned data shows the effect of a change in employment status on earnings. In other words, it illustrates those moving from wage work into self-employment would earn CHF5,350 annually less in comparison to those that don’t change, all else being equal. Further regarding the yearly earnings estimates, the coefficients belonging to the different levels of educational attainment show a pattern similar to the OLS regressions. Respondents having a vocational level of education apparently fare better than those having a high school degree, both statistically significant at p < 0.01. Analogous to the OLS and quantile regression results, these findings might be attributed to the nature of the Swiss vocational education system. Having a higher degree (from a university of applied sciences or a university) increases yearly earnings substantially: moving from a vocational degree to a university
7One of the assumptions of the Hausman test is that one of the estimators needs to an efficient one. This assumption is violated when the observations are clustered, as is the case. Therefore, the Hausman test was conducted without applying clustered standard errors. After the test indicated the fixed effects model was appropriate, this estimator was run with clustered standard errors. Although this affected the standard errors, the effect was relatively small: the only variables notably affected were age-squared and civil status, which became insignificant.
degree doubles the income. This, in combination with the higher earnings enjoyed by paid employees, may be explained by the employer screening (Hamilton, 2000) refers to or the screening hypothesis as proposed by Lazear (1981), one of the two main paradigms in the field that studies the relationship between education and productivity (Brown & Sessions, 1999). As for this relationship, studying the returns to investments in education, Psacharopoulos (1994) finds education to be an attractive opportunity for workers that enhances future earnings prospects, independent of the research stream. According to Lazear (1981), who follows the screening view, employers apply screening to create a compensation scheme for their workers to encourage productivity, as employers are unaware of their potential competence characteristics. Given the existence of information asymmetry in that firms prior to employing don’t know in what interests potential employees act and the costs associated with training workers, employers use wages as a “screening-device” (Salop & Salop, 1976). As for self-employed however, this agency problem does not exist, the assumption being they are both principal and agent and any educational investments must be reflected in true returns to their business, which is similar to what Wolpin (1977) implies. Therefore, earnings profiles are generally flatter for self-employed than wageworkers’ earnings, of which the empirical results presented are confirming evidence, although this mainly seems to hold for yearly earnings and less for hourly earnings. One reason for this may be the nature of data collection on income in the SHP, where 13th or 14th months, bonuses and gratifications are included with the annual earnings.
Earnings differentials between self-employed and wageworkers may also be related to selection effects associated with unobserved fixed characteristics or abilities, specific to a certain sector. Standard economic theory proposes individuals are aware of their abilities and will choose for a career in the sector in which their expected earnings or utility is highest. The formal starting point for this theory is Roy’s (1951) study, in which underlying factors affecting workers of making an optimized choice are discussed. Roy (1951) suggests workers to be competent in multiple sectors but are only able to apply their skills in one, motivating their choice for the sector with highest expected pays off. There may be certain time-invariant differences between wageworkers and entrepreneurs that drive these decisions. The learning mechanism employed by Jovanovic (1979) takes on a slightly different point of view, assuming workers will make rational decisions upon learning about the payoffs. They will remain on the job that is supposed to yield highest rewards and for instance move to self-employment if one finds expected income to be higher there (Evans & Jovanovic, 1989).
In summary, what becomes apparent from this fixed effects estimator is that the discount entrepreneurs experienced, as found in the OLS regression estimates, remains unchanged in sign
when controlling for fixed effects. While these findings appear to hold for yearly earnings, the results for hourly earnings are suggestive (t = -1.93; p = 0.054) but insignificant, which creates an opportunity for future research. Therefore, after controlling for fixed effects, the conclusions regarding the earnings differential between paid employees and self-employed does not change. The empirical evidence indicates entrepreneurs to earn less than their salaried counterparts, which is consistent with the general conclusion drawn by Astebro (2012) that “the entrepreneurial discount remains even after controlling for individual fixed effects (p.46).
Table 8: Fixed Effect Parameter estimates of Annual and Hourly Earnings Regression
Yearly Earnings Hourly Earnings
Variable N = 23,866; R2 = 0.0850 N = 22,176; R2 = 0.0239 Constant 23,676 0.427 -4.87 0.823 (29,782) (21.79) Employment status Self-Employed -5,350** < 0.001 -1.62 0.054 (1,221) (0.84) age -925 0.617 0.48 0.702 (1,851) (1.27) age2 -17 0.309 0.02 0.121 (17) (0.01) civstatus Married 3,084 0.094 1.63 0.075 (1,843) (0.92) educat Vocational 12,736** < 0.001 12.46** < 0.001 (1,932) (1.19) High school 9,869** < 0.001 15.90** < 0.001 (2,564) (2.12) UoAS 15,512** < 0.001 13.40** < 0.001 (3,778) (2.06) University 25,919** < 0.001 16.54** < 0.001 (4,867) (2.75) potexp 7,554** < 0.001 0.13 0.914 (1,549) (1.22) potexp2 -69** < 0.001 -0.03** 0.003 (17) (0.00)
5. DISCUSSION
The goal of this thesis was to analyze the earnings differentials between self-employed and wageworkers. Empirical studies in general show a discount for entrepreneurs at most or all levels of the income distribution. In order to replicate Hamilton (2000) as closely as possible, this study uses Swiss household panel data from 1999 to 2013, investigating male entrepreneurs and wageworkers aged between 18 and 65 years old. The earnings of self-employed and their salaried counterparts are explored through OLS regressions, quantile regressions and a fixed effect estimator, therefore replicating in a different setting, time and location-wise. The main findings of this paper are consistent with those found by Hamilton (2000), who shows that mean and median entrepreneur are doing worse than wageworkers in terms of earnings. Hamilton (2000) reports the earnings distribution for self-employed to be more skewed than for paid employees and only at the 75th percentile their earnings overtake those of employees. In addition, the earnings profiles for entrepreneurs generally demonstrate slower earnings growth and tenure-earnings profiles are flatter. The alternative entry wage available to self-employed is higher across the entire distribution than entrepreneurial earnings. Similar results are found by Tergiman (2009) and Evans & Leighton (1989). The empirical evidence found supports these claims and shows that entrepreneurs are the median or at the first quartile have lower income than paid employees do, while at the 75th percentile earnings rise above that of wageworkers. At the mean self-employed enjoy an earnings premium of 9.2% and 7.4% for hourly and yearly earnings respectively. The variance employed by the self-employed earnings distribution is much higher as the standard deviation is also between 2 and 3 times greater, denoting a heavily skewed distribution with a long right tail. The parameter estimates from the OLS and quantile regressions indicate a similar pattern where the coefficients for the constants are higher for self-employed and the coefficients for the wageworkers show less variation across quartiles. In contrast to Hamilton (2000), the empirical results show potential labor market experience to have a greater positive effect on self-employed than on wageworkers, comparable to the study conducted by Tergiman (2009), who finds greater earnings for self-employed who just started their business. The results presented here are consistent with that, with the coefficient for yearly earnings for entrepreneurs over twice of that of wageworkers’. The effect grows stronger toward the higher percentiles of the income distribution. The estimates for educational attainment also rise faster for self-employed across both educational level and quantiles, whereas Hamilton (2000) finds a significant difference when compared to those in paid employment. Given the fact that entrepreneurs overall earn less than their salaried counterparts, authors are having trouble matching this to the number of people entering into self-employment,
especially given the fact they have on average similar traits and education levels (Levine & Rubinstein, 2013). To account for this, non-pecuniary benefits not found in paid employment or the fat right tail of the self-employed earnings distribution may be of critical importance for people to become an entrepreneur (Van Praag, Van Witteloostuijn, & Van der Sluis, 2009). However, the demographic characteristics presented in table 1 in contrast show wageworkers and self-employed to be different from one another, statistically significant.
To account for any unobserved heterogeneity, a fixed effects model is estimated as self-employed and wageworkers might differ from one another in terms of risk preferences for instance. Empirical results show self-employed to earn less than wageworkers do, which amounts to CHF5,350 yearly. The results also indicate hourly earnings to be lower, but the coefficient is insignificant. Therefore, after controlling for fixed effects entrepreneurial annual earnings are found to be significantly lower than that of wageworkers. This is consistent with the reported OLS and quantile regressions results as well as with the largest part of the earnings distributions. In addition, it is similar to Fairlie (2005), also finding no difference compared to earlier earnings regressions after performing a fixed effects regression to account for unobserved differences.
5.1 CONCLUSION
Based on international empirical evidence authors generally claim individuals in self-employment suffer from a negative earnings differential in comparison to those in paid employment. While having similar traits (Levine & Rubinstein, 2013; Constant, Shachmurove, & Zimmermann, 2007), researchers find entrepreneurs to earn less both at the mean and median level. Hamilton (2000), having conducted a detailed study using U.S. panel data, reports median income of entrepreneurs to be up to 35% lower than that of wageworkers, robust across different earnings measures. Except for those in the .75th quantile, Hamilton (2000) estimates entrepreneurs to have earned more and experienced greater earnings growth in case they had switched to paid employment. Similar results are presented by Evans and Leighton (1989), who report log mean earnings to be lower for self-employed. In contrast, evidence also exists of entrepreneurs having a higher income than their salaried counterparts. For instance, Fairlie (2005) concludes male self-employed from disadvantages families to have a higher income at the mean and median. Other international studies also report mixed results.
The aim of this study was therefore to enhance the state of knowledge by investigating the earnings from self-employed in a new setting to examine whether any significant differences exist in comparison to wageworkers. As an update to Hamilton (2000), this study uses Swiss Household Panel data covering the year 1999 up to and including 2013, restricted to males between the ages of
