The Entrepreneurial Spawning of Scientists and Engineers: Stars, Slugs, and the Small Firm Effect
Daniel W. Elfenbein, Barton H. Hamilton, and Todd R. Zenger
Olin Business School, Washington University in St. Louis, Campus Box 1133, One Brookings Drive, St. Louis, MO 63130-4899; [email protected], [email protected], [email protected]
Using data from a broad sample of US scientists and engineers, we examine the relationship be-tween work experience in firms of different sizes and worker ability on decisions to enter entre-preneurship. We find that workers in small firms are far more likely than others to leave paid employment to found new ventures, and that prior experience in small firms predicts a number of positive performance outcomes in the early stages of entrepreneurship. Additionally, we show that entrepreneurs come disproportionately from the high and low end of the paid-employment ability distribution as reflected in their pay; and we find that small firms are disproportionately populated with such workers. Together our results suggest that heterogeneous preferences alone are not responsible for the relationship between firm size and spawning. Rather the labor market plays a role in sorting those for whom the pecuniary returns to entrepreneurship are likely to be highest into small firms, and it does the same for “misfits”, whose only work alternative may be self-employment. Moreover our results suggest that small firms not only employ individuals who are more likely to become entrepreneurs, they also make these workers better entrepreneurs, par-ticularly those of high ability.
____________________________________________________________________________________
1. Introduction
Entrepreneurship has been lauded by numerous observers as a driving force behind economic
growth and technological change. Not surprisingly, therefore, a large body of research has focused on the
determinants of entrepreneurship. Much of this research has focused on how individual characteristics
predict entrepreneurial activity and has explored the role of such factors as gender, race, education, credit
constraints, preferences, and cognitive differences on individual decisions to found entrepreneurial ven-
tures (e.g., Evans and Leighton, 1989, Borjas and Bronars, 1989; Busenitz & Barney, 1997; Blanchflower
and Oswald, 1998, Hamilton, 2000; Hurst and Lusardi, 2004). A growing literature has also emphasized
the role that the broader economic and social context plays in spawning entrepreneurship (Saxenian,
1994). Of particular interest is how the characteristics of an entrepreneur’s prior employer affect entre-
preneurial activity. In this vein, recent research highlights the role of a prior employer’s size in the proc-
ess of entrepreneurial spawning (Gompers, Lerner, and Sharfstein, 2005; Drobev and Barnett, 2005;
Sorensen, 2007).
- 1 -
We begin by presenting new data that highlights the dominant role small firms play in spawning
entrepreneurship among scientists and engineers in the United States. Table 1 highlights the relationship
between employer size and self-employment entry using panel data on scientists and engineers covering
1995-2001 from the National Science Foundation’s Scientist and Engineers Statistical Data System
(SESTAT).1 Roughly two-thirds of all entrepreneurial ventures started by members of this group during
the period were founded by individuals employed immediately prior in firms of fewer than 100 employ-
ees. Moreover, among scientists and engineers working in small (1-25 employees) firms, 16.7% moved
to self-employment in next two years, compared to only 1.0% of those at firms with 5000 or more em-
ployees. In this paper we explore why firm size has such a profound effect on transitions to entrepreneur-
ship.
Explanations for transitions to entrepreneurship generally fall into two broad categories. Either
future entrepreneurs are born with attributes that uniquely attract them to entrepreneurship, or accumu-
lated experiences enable effective entrepreneurship and thereby encourage transitions to self-employment
(Shane, 2003). This distinction suggests two types of explanations for the small firm effect. One type of
explanation highlights the capacity of small firms to systematically attract (or select) those with prefer-
ences or ability for entrepreneurship. Employment in small firm may be more similar to self-
employment than employment in large firms, and as a consequence small firms may simply attract those
most likely to become entrepreneurs. For instance, small firms may entice workers who value the auton-
omy offered by being one’s own boss or who are more able to bear risk (Hamilton, 2000; Parker, 2006;
Astebro and Thompson, 2007). Or, small firms may systematically attract individuals with higher abil-
ity—individuals who are then lured to self-employment as a means of capturing greater value from their
ability—in a manner described in the ability-sorting models of the type developed by Roy (1951) and oth-
ers. Thus, small firms may employ (temporarily) a higher share of those with the ability profile that will
make them likely to (rationally) expect higher earnings in entrepreneurship.
1 To our knowledge, reports of this relationship in the literature for the US are limited to Boden (1996), who exam-ines a sample of all workers from the Current Population Survey. Wagner (2004), and Sorenson (2007) examine
- 2 -
The alternative category of explanations highlights the unique context of small firms as a vehicle
that encourages entrepreneurship or transforms employees into entrepreneurs. One possibility is that sys-
tematically lower pay in small firms creates a lower opportunity cost of transitioning to self-employment.
The rewards anticipated when moving to entrepreneurship thus look particularly attractive to those em-
ployed in small firms. More interestingly, small firms may play an important role in transforming em-
ployees into effective entrepreneurs. Thus, employment in small firms may provide the opportunity to
develop a broad set of skills (Lazear, 2005) or access to valuable networks and / or knowledge valuable
for entrepreneurship (Gompers, Lerner, & Scharfstein, 2005; Stuart & Ding, 2006).
The explanations discussed above differ in their implications for the performance of entrepreneu-
rial ventures spawned from small firms. If the “small firm effect” is due solely to the allure of autonomy
in small firms, then the new ventures founded by employees of small firms should perform similarly to
new ventures founded by former large firm employees. If small firms disproportionately promote entre-
preneurship among low-performing “slugs” rather than high-performing “stars,” then the expected per-
formance of new ventures founded by former small firm employees would likely be worse than new ven-
tures founded by their large firm counterparts. Alternatively, if entrepreneurial skill is developed within
small firms, or if working in small firms provides critical information and access to networks, then the
performance of the resulting spawned ventures should be enhanced.2
We explore evidence of and explanations for the small firm effect using a new dataset of science
and engineering graduates from American universities between 1947 and 2001 (SESTAT) that contains
extensive information on individuals’ education, job experience, and demographic characteristics. The
SESTAT is especially suited for our analysis because it has longitudinal information from 1995 – 2001
for a large number of individuals. The large sample size is necessary to provide sufficiently many obser-
vations to analyze infrequent transitions such as moving from a large firm to self-employment. The data
we examine are distinct from those used in other studies of entrepreneurship and self-employment. Prior
German and Danish workers, respectively, from a broad set of occupations and industries.
- 3 -
studies focus either on broad national samples, which attempt to represent the entire working population
(e.g., Evans & Leighton 1989; Sorensen 2007) or on narrow data sets of venture-backed start-ups (e.g.,
Gompers et. al, 2005). Studies of the former type may overstate the importance, from an economic
standpoint, of small proprietorships such as barbershops, caterers, and convenience stores founded by
those with limited education. Studies of the latter type, while highly valuable, draw conclusions based on
examination of an elite group whose members may not be responsive to the same considerations as those
of the broader population. The data in this paper occupy a middle ground between these types of studies.
The individuals we examine have all achieved at least a bachelor’s degree in a science and engineering
field, and in many cases have received PhD’s. Our sample embodies those who are most likely to be the
targets of policy-makers concerned with entrepreneurship as a force of economic growth—individuals
with high levels of human capital in dynamic, knowledge-intensive fields.
In addition to documenting the striking “small firm effect” within a focused set of high human-
capital individuals, we present novel analyses that explore competing explanations. Our empirical analy-
sis suggests that employment in small firms does more than merely prompt entrepreneurship. New ven-
tures founded by employees of small firms outperform those founded by the employees of large firms.
Ventures founded by individuals previously employed in small firms persist longer, are more likely to
emerge as incorporated firms, are larger in size, and for the subset of high-ability workers, deliver higher
initial returns to the founders. We also find evidence that the small firm effect at least partly involves a
form of substantive transformation that occurs while employed within small firms—a transformation that
directly affects individuals’ performance in self employment. Employees in small firms perform a
broader set of commercial activities (e.g. sales and marketing) than employees in large firms, and this
breadth appears to contribute to transitions into self-employment. We also test whether differences in pay
levels between small and large firms account for much of the “small firm effect” and find limited support
for any simple relationship. Our results do, however, show that the future self-employed are significantly
2 Another possibility is that employees in small firms receive more accurate signals about their entrepreneurial abil-ity. This, too, might lead the resulting entrepreneurial ventures to be more successful on average.
- 4 -
more likely to come from the top and bottom ends of the paid employment distribution, consistent both
with prior findings that misfits or low ability “slugs” enter self employment (perhaps due to lower oppor-
tunity costs), and with selection-based models that suggest that the most able workers become self em-
ployed to maximize the returns to their human capital. However, a sorting process occurs into small
firms, as well. Small firms are populated disproportionately by high ability stars previously employed in
larger firms and perhaps by slugs screened away from large firms.
Collectively, our findings suggest that the small firm effect is not driven solely by worker sorting
according to (unobservable) preferences for risk or autonomy. Rather, they suggest that some of the small
firm effect is due to sorting on ability across firms of different sizes and that employees in small firm ac-
quire entrepreneurial skill. This is a key result, as potential entrepreneurs, managers, and policy-makers
alike may make different decisions depending on whether they view the dynamics of entry into entrepre-
neurship as being driven primarily by preferences or by factors that relate directly to productivity.
We proceed with the paper as follows. Section 2 outlines the theory that our research builds
upon. Section 3 describes the data. Section 4 examines entrepreneurial entry and sorting by ability. Sec-
tion 5 explores performance differences between entrepreneurs coming from small and large firms, and
Section 6 concludes.
2. Theory
Several arguments may be proposed to explain the greater rate at which small firms spawn new
ventures. The arguments fall into two broad categories: explanations that focus on the small firm context
and explanations that focus on individual attributes of those who select into and out of small firms.
2.1 The Small Firm Context
The employment context of small firms may contribute to a greater propensity for entrepreneur-
ship through a range of mechanisms. Employees of small firms may have lower wages and thus transi-
tion to self-employment due to lower opportunity costs. Moreover, those in small firms may confront
poorly developed internal labor markets, leaving them with limited opportunities for internal promotion or
increased pay (Sorensen, 2007). While these explanations may indeed explain a greater propensity of
- 5 -
small firms to spawn entrepreneurs, they cannot explain any enhanced performance within these ventures.
We therefore focus on the type and range of skills and resources that individuals may obtain through em-
ployment in small firms.
Employees of small firms may have increased access to skill development opportunities, knowl-
edge, and outside networks and resources critical to entrepreneurial success, as well as broader exposure
to more heterogeneous information and contacts outside the firm (Dobrev and Barnett, 2005). The in-
creased diversity of information and broader network access may promote greater capacity for recogniz-
ing entrepreneurial opportunities. If the discovery of entrepreneurial opportunities involves combining
broad and diverse knowledge, then the broad exposure to various functions, tasks, and external buyers and
suppliers provided in small firms may promote individuals’ capability in entrepreneurship. Consistent
with this logic, Lazear (2005) argues that entrepreneurship demands a diverse set of skills including both
application knowledge and a wide range of management skills. Entrepreneurs not only require an entre-
preneurial idea, but they require a more balanced, jack-of-all-trades set of skills. Arguably, employment
in a small firm requires the employee to acquire a range of skills that will be valuable in subsequent en-
trepreneurial ventures. Sorensen (2007) uses these arguments among others to explain his findings of a
small firm entry effect. Similarly, Gompers, Lerner, Scharfstein (2005) suggest that those employed in
small entrepreneurial firms gain access to valuable networks critical to entrepreneurship. Within these
firms, future entrepreneurs learn essential steps in founding a firm. Finally, Stuart and Ding (2006) find
that movement into entrepreneurship is more likely when colleagues and co-authors have prior experience
in entrepreneurship. Thus, small firms may provide important context in which workers acquire human
capital that will increase their chance of success in entrepreneurship, and thus promote a higher probabil-
ity of entrepreneurial spawning.
2.2 Sorting Explanations for Entrepreneurial Entry
The processes of employee sorting into and then out of small firms into self-employment provides
an alternative explanation for the small firm effect. Small firms may simply attract individuals who have
a greater propensity for entrepreneurship. For instance, by offering greater levels of autonomy, small
- 6 -
firms may attract those who find the bureaucracy of large firms unappealing and value independence of
entrepreneurship (Hamilton, 2000; Halaby, 2003; Astebro and Thompson, 2007; Sorensen, 2007).3 If,
however, the small firm entry effect results simply from sorting on preferences for autonomy, then we
should find no empirical relationship between prior small firm employment and subsequent entrepreneu-
rial performance.
Sorting may occur, however, on a dimension that is directly related to entrepreneurial productiv-
ity. In particular, sorting may reflect ability, where ability influences both the decision to enter self-
employment and subsequent performance. In exploring the role of ability and prior employment on en-
trepreneurship decisions, we build on the matching logic of Roy (1951) and Jovanovic (1979), where in-
dividuals with differing levels of sector-specific abilities choose the employment or entrepreneurship state
that yields the highest level of utility.4 A number of theories argue that high ability workers sort into
large firms. For instance, there may be complementarities between ability and capital that advantage
large firms (Lucas, 1978); complementarities between highly skilled managers and highly skilled workers
(Oi, 1983); or fixed costs associated with hiring high ability workers that large firms bear more easily
(Kremer, 1993). Consistent with this logic, empirical studies show that large firms on average offer
greater pay than small firms (Brown and Medoff, 1989; Troske, 1999). Consequently, employees of
small firms may have a lower opportunity cost of transitioning to self-employment, relative to employees
of similar ability in large firms. Employees of large firms may simply be reluctant to forego higher pay.
Other scholars have emphasized that large, bureaucratic firms are less able to directly link pay to
performance because of higher measurement costs (Garen, 1985) or higher costs involved in addressing
the perceived inequity that often accompanies performance-based pay (Zenger, 1994; Nickerson and Zen-
3 Sorensen’s (2007) analysis seeks to demonstrate that the small firm effect is independent of precisely this type of sorting. Astebro and Thompson (2007) argue that entrepreneurs (and by extension individuals joining small firms) have tastes for variety and receive non-pecuniary benefits from being a “jack of all trades.” 4 Braguinsky and Ohyama (2007), discussed below, develop a model of job-matching similar to Jovanovic (1979) in which workers learn about their ability over time. An attractive feature of their model is that it predicts that entre-preneurs coming from the upper part of the paid wage distribution will differ in the types of firms they found. How-ever, they do not directly analyze the role that small firms play in spawning entrepreneurship or subsequent entre-preneurial performance.
- 7 -
ger, 2008). As a consequence of these higher costs, large firms may link pay to easily observed measures
such as schooling, while small firms, by more effectively rewarding individual performance, attract the
more able employees from large firms (Zenger, 1994). Thus, the most talented workers at large firms
may choose to migrate to small firms or self-employment where they can more fully capture the returns to
their ability.5
How might such arguments contribute to explaining the small firm effect? If small firms are dis-
proportionately stocked with stars, then a greater propensity for entrepreneurship among employees of
small firm may reflect the abundance of high ability employees searching for higher compensation. If, by
contrast, small firms are disproportionately stocked with low ability individuals, then the small firm effect
may simply reflect the lower opportunity cost of self-employment that these low ability employees face.
While the latter argument might help explain a greater propensity for self employment among employees
of small firms, only the former argument would help explain why prior employment in small firms is as-
sociated with enhanced entrepreneurial performance.
3. Data
We construct a sample of individuals with science and engineering degrees using data from the
Scientists and Engineers Statistical Data System (SESTAT). To reduce undesirable heterogeneity, we
limit attention to males between the ages of 22 and 65 who were employed full-time in the United States
and who did not possess a professional degree such as a law or medical degree. A detailed discussion of
the construction of the sample and the choices made to limit heterogeneity can be found in the online ap-
pendix, which also defines the key variables and provides summary statistics for the sample.
Table 2 compares the means (and, for salary, the median) of several of the key explanatory vari-
ables across different employer types. In this table, we include all self-employed, not just those who tran-
sition into self-employment while under observation. The average self-employed individual is signifi-
5 A related literature, typified by the work of Zucker, Darby, and Brewer (1998) explores the role of superstar scien-tists in the genesis of high tech spinoffs. Hellmann (2007) explores the role of firms’ innovation policies, namely whether to commit ex ante, (not) to commercializing innovations not directly related to the firm’s core business, play in the decisions by innovative employees to start their own firms.
- 8 -
cantly older than the average employed person in our sample; he earns more than employees in small
firms, but less than employees in the largest firms. As we are interested primarily on the transition from
paid employment to self employment as it relates to firm size, we focus mainly on the differences in indi-
vidual responses across firm size categories. A handful of notable differences emerge. First, as men-
tioned above, median salary increases with firm size. Second, average job tenure is longer in large firms
than in small firms. Third, large firm workers are more likely to be engineers, perform a modestly
broader set of R&D activities,6 and are more likely to be primarily engaged in R&D than small firm
workers. Finally, small firm workers seem to be engaged in a significantly broader set of commercial
activities than those who work in large firms.
Appendix B presents additional analyses about the differences between small firms and large
firms, and shows that the same individual performs a broader range of commercial activities when em-
ployed in a small firm. Additionally, it provides evidence that workers leaving large firms to join small
firms are pulled disproportionately from the high end of the large-firm pay distribution, consistent with
the idea that high-ability workers join small firms to capture a greater share of returns to their ability.
4. Entrepreneurial Entry and Sorting
4.1 Examining Transitions into Self Employment
In the samples we construct above, nearly half of all movement into self-employment comes from
firms of less than 25 employees and just under two-thirds comes from firms with fewer than 100 employ-
ees. Given that such a disproportionate share of all movement into self-employment comes from small
firms, a critical empirical question is explaining this simple fact. We begin by examining the factors that
are correlated with individuals’ transitions from paid employment to self-employment. In particular, we
are interested in understanding the degree to which the strong relationship we observe between firm size
at time t and the likelihood of being self-employed at time t+2 can be explained by (a) heterogeneity
across individuals on observables such as education, race, location, etc. which could be correlated with
6 The count of research activities and the count of commercial activities used in Sections 4 and 5 below are both assembled using the data on individual activities on the job reported in Table 2.
- 9 -
firm size; (b) opportunity cost differences in leaving self-employment for paid employment, which could
be relevant if small firms pay less than large firms; (c) differences in paid-employment ability as reflected
in pay, which could explain a firm size effect if large and small firms are populated by workers of differ-
ent ability; (d) differences in activities on the job across small and large firms; or (e) differences in the
frequency with which employees of small and large firms change jobs. To explore these issues, we esti-
mate the following model:
PR(SEi,t+2 = 1 | SE it = 0) = α + βXi + γZit + µ d(i)t + εit+2 (1)
In this equation, SEit equals 1 if individual i is self employed in year t and 0 otherwise. The vector Xi is a
set of time-invariant individual characteristics such as race and type and field of highest degree7, and the
vector Zit is a vector of all potentially time-varying individual characteristics, such as marital status, num-
ber of children in the household, and location, as well as all characteristics of the individual’s employer,
job activities, and the individual’s job tenure and pay at his particular employer at time t. Employer char-
acteristics within Zit include firm size and location (generally region), job activity variables include
dummy variables indicating whether the employee’s principle job is research and development, a measure
of the diversity of activities pursued on the job, as well as fourteen dummy variables that reflect the ac-
tivities on which the individual reported spending 10% or more of his time in any given week. Differ-
ences in the average rate of transitioning into self-employment over time are captured by µd(i),t, which we
allow to vary by the type of highest degree held by the individual (d(i)), and εit represents the idiosyn-
cratic error. We estimate equation (1) only for those who are paid employees at time t; i.e., self-employed
individuals are excluded from the estimation. The estimated coefficients can be interpreted as the likeli-
hood of transitioning into self employment at t+2 as functions of Xi and Zit, rather than the likelihood of
being self employed given Xi and Zit.
7 In principle, the highest degree and the field of the highest degree could vary across years in the non-PhD sample, but since we exclude those who are not working full time from the sample, in practice there is no case in which these variable change during the course of our panel.
- 10 -
In Table 3, we report estimates of equation (1) using a probit specification. To facilitate interpre-
tation, we display marginal effects associated with the estimated coefficients. Column (1) serves as a
baseline for considering the importance of employer size at time t in explaining self-employment at t+2,
controlling only for year effects. The results reflect the patterns evident in Table 1. Employees of smaller
firms transition into self-employment much more frequently than those working in larger firms. The like-
lihood of transition declines monotonically with our firm size categories. The differences in transition
rates across firms are economically significant, with individuals in firms of size 1 – 25 employees transi-
tioning into self-employment at a rate that is nearly six times the average rate in the sample. Individual
and joint tests of equality across the firm size coefficients reject at the p < .001 level.
Column (2) adds a number of individual characteristics, the employer’s location, and the individ-
ual’s tenure in the current job. The estimated marginal effects on the firm size dummy variables decline
in magnitude by 10 to 15%, but they remain economically and statistically significant. Surprisingly,
quadratic functions of age and job tenure are not significant in this regression or in subsequent specifica-
tions that employ a richer set of covariates. Columns (3) and (4) check the robustness of the small firm
effect to controls for industry and firm age. Because of data availability, these regressions are limited to
examining only transitions from 1997 to 1999 and 1999 to 2001. Column (3) provides the baseline esti-
mates for these years, employing the controls used in column (2). In column (4), we incorporate industry
controls and a dummy variable that equals 1 if the individual worked for a firm that was less than 5 years
old at time t. Incorporating controls for industry and firm age reduces the magnitude of the firm size co-
efficients by an additional 15 to 18%. Industry controls are jointly significant, and the coefficient on the
new firm dummy is positive and significantly different from zero. While the new firm dummy has some
explanatory power in predicting movement into self-employment, we note that its explanatory power is
modest when compared to the firm size coefficients.
We next explore opportunity cost explanations for the differences in transitions into entrepreneur-
ship. The summary statistics in Table 2 showed that small firm employees earn lower wages than those in
larger firms. The discussion in Section 2.2 argued that because they pay less, small firms may spawn
- 11 -
more entrepreneurs due to the lower opportunity cost of self-employment entry for these employees. In-
clusion of the paid employment (log of) weekly wages in equation (1) should therefore reduce the impact
of the firm size variables if the small firm effect on entry reflects differential opportunity costs.8 Com-
parison of the firm size coefficients in column (5) when wages are included in equation (1) with those in
column (2) where pay is excluded indicate that differences in pay by firm size cannot explain the small
firm effect. Moreover, the coefficient on log(weekly wageit) is positive rather than negative, although it is
not significantly different from zero at conventional levels. These findings are inconsistent with the view
that the small firm effect is generated by a lower opportunity cost of self-employment for workers in
small firms.9
Columns (6) and (7) of Table 3 begin to explore ability sorting explanations by further examining
the actual relationship between compensation in paid employment and the transition to self employment.
In section 2 above, we hypothesized that—for different reasons—the highest and lowest ability workers
would be the most likely to transition into self-employment. We test these hypotheses using a crude
measure of relative ability—the position of a given individual within the pay distribution10 in a given year
among individuals with the same highest degree. Thus, we construct a percentile rank in the pay distribu-
tion separately for BAs, MAs, and PhDs in each year. In column (6), we include pay quintile into the
transition equation and find that employees in the highest and lowest pay quintile are more likely to enter
into entrepreneurship in the subsequent period. While the inclusion of this measure of pay does not ap-
pear to explain a large portion of the small firm effect, the “U-shaped” relationship between self-
8 To avoid estimating a supply response to wage rates, we employ the respondent’s (log of) weekly wage (reported annual salary divided by reported weeks worked) as the measure of pay. 9 We also estimated the model in column (5) including the expected weekly wage in t+2, rather than the actual wage in period t, and an estimate of expected self-employment earnings in t+2 using the approach of Willis and Rosen (1979). We find that larger expected pay differentials between self-employment and paid employment are associ-ated with a higher rate of self-employment entry, but the effect is not statistically significant at the 10% level. Moreover, the coefficient estimates for the firm size indicators are virtually identical to those presented in column (5). Results available from the authors upon request. 14 Thus, those whose highest degree is a BA are compared with other BAs, MAs are compared with other MAs, and PhDs are compared to PhDs.
- 12 -
employment entry and prior wage indicates that both low wage and high wage employees are more likely
to become entrepreneurs. To provide an indication of the magnitude of this effect, in column (7) we in-
clude pay percentile and pay percentile squared into the transition equation. The estimated coefficients
indicate that an individual at the 10th percentile is approximately 26% more likely than an individual at
median of the distribution to enter self-employment in the following period. Similarly, an individual
earning at the 90th percentile is approximately 28% more likely to enter self-employment in the subse-
quent period. The relatively high rate of entrepreneurial entry among workers in the lowest quintile of the
wage distribution is not particularly surprising: this is consistent with the view that these workers have the
lowest opportunity cost of becoming self-employed. However, this argument cannot explain why em-
ployees at the top of the pay distribution also are more likely to become entrepreneurs. These individuals
may be drawn by relatively high returns in entrepreneurship. We investigate this phenomenon more fully
in Sections 4.2 and 5.
Given that the small firm entry effect cannot be explained solely by differences in observed
worker characteristics or differential opportunity costs, we next examine the “jack of all trades” hypothe-
sis that small firms provide a greater opportunity to accumulate the broad array of skills that are valuable
in self-employment. Beginning with our baseline specification of the self-employment entry model in
column (1) of Table 4 (derived from column (7) of Table 3), we introduce measures of the breadth of ac-
tivities of the individual in the firm: a count of the number of commercial activities that the individual
reported engaging in and a count of the number of research activities. Column (2) shows that the coeffi-
cient on the count of commercial activities is positive and significantly different from zero, consistent
with idea that those with a broader range of skills are more likely to become entrepreneurs. On the other
hand, the coefficient on the count of R&D activities is negative and significant, suggesting that perhaps
those who engage in R&D are more likely to need the complementary resources of an existing firm to be
productive. In column (3), we allow the count of commercial and R&D activities to vary non-
monotonically; zero is the omitted category in both cases. The estimates show that those who engage in
the broadest range of commercial activities are most likely to transition into self-employment in the sub-
- 13 -
sequent period, while those who perform a broader range of R&D activities have a similar, reduced pro-
pensity to become self-employed as those who perform only one R&D activity. In columns (4) and (5),
we include, respectively, dummy variables for each of the 14 activities reported and activity dummy vari-
ables interacted with a PhD dummy variable. Incorporating these significantly improves the fit of the
model and reduces the magnitude of the small firm coefficients by an additional 12 to 16%. In sum, this
analysis has two main implications: First, similar to the theory of Lazear (2005), we find that performing
a broader range of commercial activities in one’s current job increases the likelihood of subsequently be-
coming an entrepreneur, although the same cannot be said for R&D activities. Second, we find that a por-
tion of the small firm effect can be explained by the fact that small firm employees perform a broader set
of activities than employees in large firms and are less likely to be engaged in R&D activities.
Finally, we investigate the degree to which the observed relationship between firm size and entry
into self-employment is a function of the increased rate at which employees from smaller businesses
change employers. If all employees who separate from a given employer are equally likely to become
self-employed, the higher rates of transition we observe when we estimate equation (1) may result from
the fact that employees at small firms are simply more likely to leave their jobs than those leaving larger
firms. To examine this possibility we estimate a multinomial logit model for those who are not self-
employed at time t with the following choices between period t and t+2:
1. Remain with current employer in both periods
2. Change jobs, but do not become self-employed
3. Become self employed in t+2
If employees changing jobs have a constant rate of entering self-employment, then we expect the ratio of
the resulting coefficients to be approximately equal across the two equations. Table 5 presents the esti-
mates for this model. The omitted decision in these estimations is choice 1, remaining with the current
employer. Comparison of column (1a) and (1b) shows that the rate at which entrepreneurship increases
given a decrease in firm size is significantly higher than the rate at which changing jobs increases; how-
ever, this higher rate of job separation alone is insufficient to explain the small firm effect. Columns (2a)
- 14 -
and (2b) indicate that workers at the bottom end of the pay distribution are also more likely to leave their
firm, either to change jobs or to become self-employed. The opportunity costs for these workers of leav-
ing their employer appears to be low, regardless of their destination. In contrast, highly paid employees
are not more likely to switch employers than the median-paid worker, but these “stars” are significantly
more likely to become entrepreneurs. Overall, the results of this section indicate that a portion of the
small firm entry effect can be explained by differences in the observed characteristics of workers in small
vs. large firms. Small firm workers are also more likely to be engaged in commercial activities similar to
those performed by entrepreneurs. We find little support for the view that all small firm employees have
lower opportunity costs of self-employment, although the finding that workers in the lowest quintile of
the pay distributions are more likely to enter self-employment is consistent with the opportunity cost ex-
planation. Finally, star employees are more likely to become entrepreneurs. We now investigate this
phenomenon in greater depth.
4.2 Examining Ability Sorting by Investigating Differences in Paid Employment Wages
In Table 3 above, we find preliminary evidence that both high- and low-ability workers are more
likely to transition into self-employment. While low ability workers (“slugs”) may have a lower opportu-
nity cost of self-employment, this argument cannot explain why high ability workers (“stars”) are more
likely to become entrepreneurs. In this section, we analyze the relationship between ability and entry into
self-employment in greater detail, and we further investigate differences in ability of future entrepreneurs
by size of their current employer. The ability-sorting models of Roy (1951), Jovanovic (1982), and others
suggest that stars may choose to become entrepreneurs in order to more fully the capture returns to their
ability. To the extent that the link between pay and performance differs by firm size (Garen, 1985; Ras-
musen & Zenger, 1990), we may observe substantial differences in the sorting of workers into entrepre-
neurship from small vs. large firms. To simplify the analysis below, we define small firms as firms with
100 or fewer employees and larger firms as firms with 100 or more employees.
We begin by examining the distribution of pay across several categories of employers and re-
spondents. Figure 1a illustrates that the distribution of pay differs significantly in large firms, small
- 15 -
firms, and self-employment in ways that are consistent with the prior literature on the relationship be-
tween pay and firm size (Garen, 1985; Rasmusen & Zenger, 1990). The variance of self employment pay
exceeds the variance of pay in small firms which, in turn, exceeds the variance of pay in large firms. The
figure also shows a higher average pay in large firms relative to small firms, again consistent with prior
literature. Figure 1b compares the paid employment wages at time t of those who become self-employed
by time t+2 and those who remain in paid employment. Consistent with the theory that suggests that stars
(and slugs) are more likely to leave paid employment to become self employed, we see greater density at
high (and low) pay levels in paid employment of the future self-employed relative to those who remain in
paid employment. Figures 1c and 1d compare the current pay of future entrepreneurs with that of “stay-
ers” in small and large firms, respectively. Overall, these figures provide further evidence that those en-
tering self-employment are more likely to be either “stars” or “slugs” than those remaining in self-
employment.11 Both small and large firms appear to be losing their stars (and slugs) to entrepreneurship.
To examine this relationship in further detail accounting for worker and firm characteristics, we
estimate “pre-program” regressions (Heckman and Hotz, 1989) of the following form for paid employees:
PAY it = α + βXi + γZit + δSEi,t+2 + µt + εit, (2)
where Xi, Zit, and µt, are, as before, time-invariant individual characteristics, time-varying individual (and
individual-job) characteristics, and time dummy variables. PAYit is the individual i’s pay at time t, and is
measured as the log the weekly wage for individual i at time t. Because we are estimating a continuous
variable, Zit contains a richer set of covariates (such as the U.S. state in which the employee works) than
in the transition regressions above. The vector SEi,t+2 indicates the individual i’s future self employment
status. In the baseline analysis it is simply a dichotomous variable that equals 1 if the individual i is self
employed by period t+2 and is 0 otherwise. In subsequent regressions we also interact this future self-
employment status variable with the type of employer at time t and whether the individual is in an R&D
11 In an online appendix ((http://www.olin.wustl.edu/faculty/elfenbein/SmallFirmOnlineAppendix.pdf) we plot these distributions separately for PhDs and non-PhDs.
- 16 -
track job at time t. In examining the relationship described in equation (2) we estimate an OLS regression
and censored quantile regressions at the 10th, 25th, 50th, 75th, and 90th percentiles.
Similar to the interpretation of the estimates of equation (1) above, we interpret an individual’s
pay in the present as a reflection of the ability of the worker in paid employment. Thus, the coefficient δ
in the OLS regression indicates whether, on average, those who will become self employed are “better”
employees than those who remain in paid employment, controlling for other major predictors of pay. The
interpretation of δ in the quantile regressions is similar. A positive and significant δ in the 90th percentile
regression, for example, indicates that the 90th percentile of the pay distribution for the future self-
employed is higher than that for those who remain in paid employment, controlling for other factors.
Table 6 presents the results of these estimations on the pooled sample. The pay variable is the log
of (salary/number of weeks worked). All coefficients other than δ are suppressed. In the OLS regres-
sions, robust standard errors are displayed. In the quantile regressions, we present bootstrap standard er-
rors. In analysis series (1), SEi,t+2 is a single dichotomous variable indicating whether the individual be-
comes self-employed in the next period. These analyses confirm that the differences in the distributions
of pay illustrated in Figure 1b are statistically significant: the median of the pay distribution for the future
self-employed is 4.6% higher than those remaining in paid employment, while movers earned 12.2%
more (p < .001) at the 90th percentile and 14.9% (p < .001) less at the 10th percentile compared to stayers.
In analysis series (2), we interact the future self-employment variable with firm size. These analyses sug-
gest the dispersion of pay among those leaving small firm employment to start their own firms is even
greater than that among those leaving large firms; the effect is particularly pronounced for low-paid small
firm workers.
We further subdivide the future self-employed according to whether their primary activity in the
paid employment was R&D or non-R&D, and interact this categorization with firm size. Analysis series
(3) presents the results of these estimations. For both small and large firms, future entrepreneurs who are
focused on R&D seem to be drawn from a similar pay distribution as those who remain in paid employ-
- 17 -
ment. Rather, these regressions show that the increased pay dispersion of future entrepreneurs is coming
mainly from non-R&D track employees. While the pay dispersion of non-R&D future entrepreneurs in
small firms is modestly greater than for the same set of future entrepreneurs in large firms, the difference
is somewhat less pronounced than the differences observed in analysis series 2, suggesting that the com-
position of future entrepreneurs by job type might differ across firm size. Overall, the results are consis-
tent with the predictions from sorting models (e.g., Roy, 1951) that star employees will be more likely to
choose entrepreneurship because employers are unable to tightly link pay to ability/performance. It is
also clear that stars enter self-employment from small as well as large firms. This effect appears to be
particularly pronounced for individuals primarily engaged in non-R&D activities, which is not surprising
given the finding from Table 2 that entrepreneurs undertake a wider variety of commercial tasks. The
finding that employee “slugs” also disproportionately choose entrepreneurship is consistent with the no-
tion that these individuals face lower opportunity costs of self-employment.
5. Small Firm Experience and Entrepreneurial Performance
In this section, we investigate whether those with experience working in smaller firms perform
better in self-employment. If the entry effect described above reflects the greater opportunity that small
firms provide to accumulate human capital that is valuable in entrepreneurship, or small firms attract
higher quality “latent” entrepreneurs, then small firms should spawn better performing entrepreneurs.
Conversely, if small firms simply attract individuals with preferences for independence who then become
entrepreneurs, we will not necessarily observe a positive relationship between prior employment in small
firms and performance. To distinguish among these predictions, we estimate entrepreneurial performance
relationships of the form:
PERF it = α + βXi + γZit-2 + ρFSIZEi,t-2 + θlog(wageit-2)+ νit,, (3)
where PERFit is the measure of entrepreneurial performance, Xi and Zit-2 are as defined above, and
FSIZEi,t-2 is a vector indicating the size of the firm employing individual i in period t-2, prior to self-
employment entry. The inclusion of the weekly wage in period t-2 in equation (3) accounts for the role of
ability (in paid employment) for initial entrepreneurial success. We consider two specifications of the
- 18 -
firm size vector. First, we employ a single dummy variable for firms of size 1 – 25. Roughly half of all
entrants into self-employment come from firms of this size. Second, we employ multiple dummies for
firm size to explore potentially non-linear effects.
We focus on four measures of initial entrepreneurial performance (PERFit): (a) persistence in
self-employment; (b) whether or not the self-employed individual incorporates the new venture; (c) the
number of direct reports in the first period of self-employment; and (d) total pay in the first period of self-
employment. We interpret persistence in self-employment as a proxy for survival of the new enterprise,
although other interpretations are possible. We interpret the decision to incorporate as a proxy for the size
and growth potential of the new venture. Prior research indicates that unincorporated business owners
may be less innovative and less likely to undertake risks, and often have slower growth trajectories than
incorporated ventures (Ribstein, 2004). Incorporated businesses are also more likely to be able to attract
outside capital (Mackie-Mason and Gordon, 1997). Similarly, we interpret the number of direct reports
for the self-employed individual as proxies for the size of the new venture. Finally, we examine pay in
the first period of self-employment as a measure of the monetary returns from the enterprise.
Table 7 reports the results of analyses for three of the performance measures. In columns (1) - (3)
we investigate persistence in self employment using a probit analysis. For these regressions, the depend-
ent variable is equal to 1 if the individual reported being self-employed in both t and t+2 and 0 otherwise.
For this analysis we include only those who reported being employed in a firm in t-2.12 We report mar-
ginal effects. Columns (1) and (2) examine different parameterizations of prior employer size, and col-
umn (3) adds additional controls for breadth of activities in the prior job into Zit-2. The results are consis-
tent across each of the specifications: those entrepreneurs coming from small firms survived in self-
employment longer than those who entered self-employment from large firms. Somewhat surprisingly,
persistence in self-employment is unrelated to the prior wage, although prior research activities have a
significant and negative impact on survival. Next we turn to the decision to incorporate, and we report
12 We are limited to 735 observations (out of the 1522 individuals entering self-employment between periods t-2 and t) because we require information self-employment status in both period t and t+2.
- 19 -
these results in columns (4) - (6) of Table 7. PERFit in this case equals 1 if the self-employed individual
reported being incorporated in period t, and 0 otherwise. We restrict the sample to those who transitioned
into self-employment in the current period, and employ a probit specification. Each set of estimates
paints a similar story. Those coming from smaller firms are more likely to enter self-employment as in-
corporated entities. Unlike in the analysis of persistence, this choice seems highly correlated with the
prior wage. Given that incorporated firms have higher rates of survival and greater revenue potential,
these results indicate that small firms spawn entrepreneurial ventures of greater size and stability.
We further our investigation into the initial size of new ventures by examining responses to a sur-
vey question in which individuals were asked the number of direct reports they had in their current job.
Because the number of employees supervised is reported as free response (i.e., 0, 1, 2, 3, etc), this meas-
ure allows us to distinguish between entrepreneurs who have employees and those who do not.13 For
first-period self employed, the median and mean of the number of direct reports are 0 and 2.5, respec-
tively. Given the preponderance of zeros in the distribution of the dependent variable and the skewness of
the distribution, we estimate equation (3) as an ordered logit, in which the ordered choices reflect the
number of direct reports.14 In columns (7) - (9) of Table 8 we show that prior employment in small firms
is associated with an increasing number of direct reports in the new firm, although the coefficient esti-
mates suggest that the effect might not be entirely monotonic: prior employment in a firm of size 26-100
has the greatest impact on the number of direct reports in the new venture. Positive and significant coef-
ficients on lagged weekly wage and prior breadth of commercial experience suggest that ability and ac-
quired “general management” skill may also be important determinants of the initial size of the entrepre-
neurial enterprise.
13 We could not infer this from the firm size category responses since “1” is included in the smallest category of firm size, 1-10 employees. 14 In unreported analyses, we corroborate these results using negative binomial specifications, finding identical pat-terns to those reported below.
- 20 -
The results in Table 7 show that small firms spawn entrepreneurial ventures that have more em-
ployees and are more likely to survive and incorporate. In Table 8, we re-estimate equation (3) using the
(log of) annualized pay in the entrepreneurial venture in period t as the measure of initial performance.
We employ a censored-normal regression to account for the top-coding of pay for some sample members
(and limited bottom coding as well). Columns (1) and (2) indicate that lagged firm size does not appear
to significantly impact initial self-employment income. The primary determinant of initial entrepreneurial
pay is the lagged paid employment wage, suggesting that more able employees form more financially
successful businesses.
The analysis in Section 4 suggested that entrepreneurship attracts both stars and slugs from paid
employment. To investigate the possibility that small firm experience has a different impact on high abil-
ity versus low ability workers, we divide the sample into two groups—those earning more than the me-
dian for their education type in year t-2 and those earning less than the median—and repeat the analyses
on these subsamples. The contrast between the estimates produced by the two subsamples is striking. As
illustrated in columns (3) and (4), for those entrepreneurs who were among the top-half of wage earners,
prior experience in small firms is associated with significantly higher pay than prior experience in large
firms, controlling for a variety of characteristics.15 In addition, the coefficient on the lagged wage sug-
gests that individuals who were stars in paid employment are also star entrepreneurs, at least in terms of
initial returns. By contrast, columns (5) and (6) show that entrepreneurs coming from the bottom half of
the income distribution display no such benefit from prior experience in a small firm. Additionally, there
is very little relationship between these employees’ prior earnings in paid employment and their earnings
in self-employment. This pattern in the data suggests two interpretations. First, low-earning small firm
employees may be constrained either by their position in the firm or by their ability from acquiring the
15 Given the fact that we include prior pay as a control variable, this might be interpreted as the impact of small firm experience on a change in pay. In unreported regressions we eliminate prior pay as a control variable. In these analyses, the estimated coefficients on self-employment pay rise in magnitude and significance. Additionally, we run a regression on the difference in pay in self-employment at time t and paid employment at time t-2 and find nearly identical results, with, and without including the level of prior pay as a dependent variable.
- 21 -
benefits of small firm experience. Second, low-pay workers who become entrepreneurs may be doing so
because their current pay does not accurately reflect their ability.
The results in Table 8 imply that part of the reason that small firms spawn more entrepreneurs is
because some of these individuals experience higher initial returns in their new ventures. We now at-
tempt to distinguish between two alternative explanations discussed in Section 2 for the positive effect of
small firm experience on entrepreneurial performance observed in Tables 7 and 8: (a) individuals acquire
human capital by working in small firms that make them more successful entrepreneurs; (b) small firms
attract individuals with higher levels of “latent” entrepreneurial ability (i.e., higher values of νit, in equa-
tion (3)).16 To account for the potential non-random selection into small firms implied by explanation (b),
we re-estimate the performance equation (3) adopting the inverse propensity score weighting methods
discussed in Hirano and Imbens (2002) and Wooldridge (2007).17 These methods allow us to more fully
capture non-random selection based on observed characteristics. While an obvious instrument is not
available that would allow us to account for selection on unobservables, we are able to condition on the
lagged wage. If entrepreneurial ability is strongly correlated with ability in paid employment (as appears
to be the case from the strongly positive estimates of θ in Table 8, at least for above-median workers), the
lagged wage variable should incorporate some of the effect implied by explanation (b). Considering the
“treatment” as having worked in a small firm (1 – 25 employees) in t-2, we construct both the average
treatment effect and the treatment effect for the treated. The former measures the impact of having
worked in a small firm on entrepreneurial success for the average entrepreneur; the latter measures the
treatment effect for the set of entrepreneurs who worked in a small firm prior to starting their venture.
16 Explanation (b) implies cov(FSIZEi,t-2,νit,) > 0 in equation (3). 17 The propensity score weighting approach is very similar to matching on propensity scores. In the first step, we estimate a logit model for the probability that an individual worked in a small firm in t-2, including Xi and Zit-2 as covariates. In the second step, equation (3) is re-estimated via weighted least squares (or logit) using the inverse of the predicted propensity scores from step 1 as weights. The form of the weights depends on whether the average treatment effect or treatment effect for the treated is being estimated. Hirano and Imbens (2002) provide a clear introduction to these methods.
- 22 -
The results from our treatment effect estimates for various measures of performance are reported
in Table 10. In row (1), we examine persistence in self employment using a weighted linear probability
model to facilitate comparison with the marginal effects probit regressions reported in Table 7. Similarly,
in row (2) we examine the choice to enter as an incorporated entity also using a weighted linear probabil-
ity model. In row (3) we report the propensity score adjusted coefficients on entry size using weighted
OLS. In each case we corroborate the analyses using weighted logit and ordered logit specifications in
unreported regressions and find coefficients of the same sign and significance levels as those reported in
Table 9. In rows (4)–(6) we examine self-employment pay for the entire sample and for the subsample of
those coming from the top and bottom halves of their respective salary distributions at t-2, respectively,
using OLS (unfortunately the appropriate weighted regressions that incorporated censoring were not
available). Adjusting for the non-random selection based on observed characteristics, we continue to find
significant positive effect of small firm experience on self-employment survival, likelihood of incorpora-
tion, entry size, and entrepreneurial returns (again for the top half of the lagged wage distribution only).
In addition, row (5) column (2) shows that the impact on initial entrepreneurial earnings is larger for the
subsample of individuals who chose to work in a small firm in t-2. While we cannot completely rule out
explanation (b) in the absence instrumental variables accounting for selection on unobservables, the find-
ings in Table 9 support the view that the positive impact of small firm experience on initial entrepreneu-
rial performance results from skills that nascent entrepreneurs acquire while working in these firms.
To summarize the findings from this section, small firms spawn larger, more stable ventures that
are initially more likely to survive. Moreover, small firm experience is associated with higher initial en-
trepreneurial returns, at least for high ability workers. Some of the small firm effect appears to reflect the
accumulation of human capital that is valuable once the individual starts his business, although sorting
explanations cannot be ruled out. The positive small firm performance effect implies that the greater pro-
pensity to enter self-employment from small firms observed in Section 4 does not only reflect preferences
for autonomy. Small firms spawn more entrepreneurs in part because workers from these firms earn
higher returns in self-employment.
- 23 -
6. Conclusion
Our objective has been to both present and explain evidence of a “small firm effect” in patterns of
entrepreneurial spawning. We find evidence that prior employment in small firms plays a surprisingly
large role in explaining transitions to entrepreneurship. Moreover, we find evidence that this greater pro-
pensity to spawn is not merely a reflection of small firms attracting those who either prefer the non-
pecuniary reward of self-employment or a simple result of small firms awakening a greater desire for en-
trepreneurship among their employers. Instead, we find evidence that prior employment in small firms
appears to play some substantive role in transforming individuals such that the ventures they found enjoy
higher performance. Thus, a portion of the greater propensity of small firms to spawn entrepreneurship
may reflect the higher performance that those employed in small firms anticipate when choosing to enter
self-employment.
Our results also suggest explanations for the origin of this small firm effect. We find evidence
consistent with the idea that employment in small firms provides access to knowledge, networks, or hu-
man capital valuable in founding an entrepreneurial venture. After controlling for the attributes of those
who select into small firms and for regional differences, we still find strong evidence that employment in
a small firm dramatically enhances the probability of entering self-employment. This small firm effect
appears to partly reflect the fact that small firms push engineers and scientists to perform a much broader
set of activities, particularly activities beyond research. Moreover, consistent with Lazear’s argument that
entrepreneurs must be jacks of all trades, performing a broad array of activities does predict movement
into self-employment. Employment in small firms appears to play a critical role in providing this broad
training valuable in entrepreneurship.
Our results also shed light on other explanations for the surprisingly large role that small firms
play in spawning entrepreneurship. We have empirically examined three categories of explanation: abil-
ity sorting, human capital accumulation, and non-pecuniary rewards. Our results provide clear evidence
of a pattern of sorting by ability into self-employment—a pattern that may promote the performance of
entrepreneurial ventures spawned by small firms. We find evidence that those who transition into self
- 24 -
employment from small firms are disproportionately drawn from the upper end of the pay distribution,
particularly the upper tail. We also find evidence that those entering self-employment from small firms
are disproportionately drawn from the lower tail of the distribution. However this appears to simply re-
flect the fact that lower tail employees have lower opportunity costs and are changing jobs more fre-
quently. Thus, while the lower tail in small firms appears equally likely to depart for either self-
employment or to paid employment in large or small firms, star workers are significantly more likely to
seek self employment upon exit. We interpret these results as suggesting that the more able are drawn
into self-employment by the opportunity for greater pay, while the less able in small firms, who earn
rather low pay and thus face a rather low opportunity cost in changing jobs, simply depart small firms
more frequently and move into firms of varying size and into self-employment.
In summary, our results suggest that small firms may play several important roles in promoting
successful entrepreneurship. Small firms appear to provide important access to the broad skills and net-
works required to facilitate entrepreneurship. We speculate the small firms may also provide an arena in
which individuals self-discover their capacity for entrepreneurship. Those who experience success in
paid employment within these small firms and receive high pay essentially recognize that they possess a
capacity to found a new venture. In part, this self-recognition could reflect the knowledge accumulated
while employed within small firms, and it may enable potential entrepreneurs in small firms to make
more accurate assessments of their likely performance when making the leap to entrepreneurship. While
our study has made important headway in documenting and explaining the small firm effect, there is
clearly much that remains unexplored.
- 25 -
7. References
Astebro, Thomas and Peter Thompson. 2007. “Does it Pay to be a Jack of All Trades?” Mimeo, Univer-sity of Toronto.
Blanchflower, David G. and A. Oswald. 1998. “What Makes an Entrepreneur?” Journal of Labor Eco-nomics 16(1): 29-60.
Boden, Richard J. 1996. “Gender and self-employment selection: An empirical assessment.” Journal of Socio-Economics 25(6): 671-682.
Borjas, George J. and Stephen G. Bronars. 1989. “Consumer Discrimination and Self-Employment.” Journal of Political Economy 108(3): 604- 631.
Braguinsky, Serguey, and Atsushi Ohyama,. 2007. “Where Does Entrepreneurship Pay?” Working Pa-per, State University of New York, Buffalo.
Brown, Charles and James Medoff. 1989. “The Employer Size-Wage Effect.” Journal of Political Econ-omy 97(4): 1027-1059.
Busenitz, Lowell W. and Jay B. Barney. 1997. “Differences between Entrepreneurs and Managers in Large Organizations.” Journal of Business Venturing 12(1): 9-30.
Dobrev, Stanislav D. and William P. Barnett. 2005. “Organizational Roles and Transition to Entrepre-neurship.” Academy of Management Journal 48(3):433-449.
Evans, David S. and Leighton, Linda S. 1989. “Some Empirical Aspects of Entrepreneurship.” Ameri-can Economic Review 79(3): 519-535.
Garen, John E. 1985. “Worker Heterogeneity, Job Screening, and Firm Size.” Journal of Political Econ-omy 93(4): 715-740.
Gompers, Paul, Lerner, Josh, and David Scharfstein. 2005. “Entrepreneurial Spawning: Public Corpora-tions and the Genesis of New Ventures, 1986 to 1999.” Journal of Finance 60(2): 577-614.
Halaby, Charles N. 2003. “Where Job Values Come From: Family and Schooling Background, Cogni-tive Ability, and Gender.” American Sociological Review 68(2): 251-278.
Hamilton, Barton H. 2000. “Does Entrepreneurship Pay? An Empirical Analysis of the Returns to Self-Employment.” Journal of Political Economy 108(3): 604- 631.
Heckman, James J. and V. Joseph Hotz. 1989. “Choosing Among Alternative Nonexperimental Methods of Estimating the Impact of Social Programs: The Case of Manpower Training.” Journal of the American Statistical Association 84(408): 862-875.
Hellman, Thomas. 2007. “When Do Employees Become Entrepreneurs?” Management Science 53(6): 919-933.
Hurst, Erik and Lusardi, Anna Maria. 2004. “Liquidity Constraints, Household Wealth and Entrepre-neurship.” Journal of Political Economy 112(2): 319-347.
Hirano Keisuke and Imbens, Guido W. 2002. “Estimating Causal Effects using Propensity Score Weighting: An Application to Data on Right Heart Catheterization.” Working Paper, the University of Miami and University of California, Berkeley.
Jovanovic, Boyan. 1979. “Job Matching and the Theory of Turnover.” Journal of Political Economy 87(5): 972-991.
Jovanovic, Boyan. 1982. “Selection and the Evolution of Industry.” Econometrica 50: 53-75.
- 26 -
- 27 -
Kremer, Michael. 1993. “The O-Ring Theory of Economic Development.” Quarterly Journal of Eco-nomics 108(3): 551-575.
Lazear, Edward P. 2005. “Entrepreneurship.” Journal of Labor Economics 23(4): 649-680.
Lucas, Robert E. 1978. “On the Size Distribution of Business Firms.” The Bell Journal of Economics 9(2): 508-523.
Mackie-Mason, Jeffry and Gordon, Roger. 1997. “How Much do Taxes Discourage Incorporation?” Journal of Finance 52(2): 477-505.
Nickerson, Jackson A. and Todd R. Zenger. 2008. “Comparison Costs, and the Economic Theory of the Firm," Strategic Management Journal, forthcoming.
Oi, Walter Y. 1983. “The Fixed Employment Cost of Specialized Labor,” in Jack E. Triplett (ed.), The Measurement of Labor Costs. Chicago: University of Chicago Press.
Parker, Simon C. 2006. “A Selection-Based Theory of the Transition from Employment to Entrepre-neurship: The Role of Employer Size.” IZA Discussion Paper #2071.
Rasumusen, Eric and Zenger, Todd R. 1990. “Diseconomies of Scale in Employment Contracts.” Jour-nal of Law, Economics, and Organization 6(1): 65-92.
Ribstein, Larry E. 2004. Unincorporated Business Entities, 3rd. ed. Newark, NJ: LexisNexs.
Roy, Andrew D. 1951. “Some Thoughts on the Distribution of Earnings.” Oxford Economic Papers 3: 135-146.
Saxenian, AnnaLee. 1994. Regional Advantage: Culture and Competition in Silicon Valley and Route 128. Cambridge, MA: Harvard University Press.
Shane, Scott. A General Theory of Entrepreneurship. Northhampton, MA: Edward Elgar Press.
Sorensen, Jesper B. 2007. “Bureaucracy and Entrepreneurship: Workplace Effects on Entrepreneurial Entry.” Administrative Science Quarterly 52(3): 387-412.
Stuart, Toby and Ding, Waverly W. 2006. “When Do Scientists Become Entrepreneurs? The Social Structural Antecedents of Commercial Activity in the Academic Life Sciences.” American Journal of Sociology 112(1): 97-144.
Troske, Kenneth R. 1999. “Evidence on the Employer Size-Wage Premium From Worker-Establishment Matched Data.” Review of Economics and Statistics 81(1): 15-26.
Wagner, Joachim. 2004. “Are Young and Small Firms Hothouses for Nascent Entrepreneurs? Evidence from German Micro Data.” IZA Discussion Paper #989.
Willis, Robert J. and Sherwin Rosen. 1979. “Education and Self-Selection.” Journal of Political Econ-omy 87(5): 7-39.
Wooldridge, Jeffrey M. 2007. “Inverse Probability Weighted Estimation for General Missing Data Prob-lems.” Journal of Econometrics 141(2): 1281-1301.
Zenger, Todd R. 1994. “Explaining Organizational Diseconomies of Scale in R&D: Agency Problems and the Allocation of Engineering Talent, Ideas, and Effort by Firm Size.” Management Science 40(6): 708-729.
Zucker, Lynne G. , Michael R Darby, and Marilynn B. Brewer. 1998. “Intellectual Human Capital and the Birth of U.S. Biotechnology Enterprises” The American Economic Review 88(1): 290-306.
-
0.2
.4.6
.81
Kern
el D
ensi
ty
5 6 7 8 9Log of (Weekly Wage + 10)
Paid Employee at t+2 Self Employed at t+2
Pooled SampleComparison of Current Paid-Employment Wages of Future Self-Employed and Future Paid Employees
0.2
.4.6
.81
Kern
el D
ensi
ty
5 6 7 8 9Log of (Weekly Wage + 10)
Large firm at t, Not SE at t+2 Large firm at t, SE at t+2
Pooled Sample Large Firms OnlyComparison of Current Paid-Employment Wages of Future Self-Employed and Future Paid Employees
28 -
Figure 1a. Distributions of Weekly Wages in Large Firms, Small Firms, and Self-Employed, Pooled Sample
0.2
.4.6
.81
Kern
el D
ensi
ty
5 6 7 8 9Log of (Weekly Wage + 10)
Large Firm Small FirmSelf Employed
Pooled SampleComparison of Current Weekly Wages of By Firm Size and SE Status
Figure 1c. Distribution of Weekly Wages for Future Self-Employed, Compared with Distribution of Those Remaining in Paid Employment in Small Firms,
Pooled Sample
0.2
.4.6
.8Ke
rnel
Den
sity
5 6 7 8 9Log of (Weekly Wage + 10)
Small firm at t, Not SE at t+2 Small firm at t, SE at t+2
Pooled Sample Small Firms OnlyComparison of Current Paid-Employment Wages of Future Self-Employed and Future Paid Employees
Figure 1b. Distribution of Weekly Wages for Future Self-Employed, Com-pared with Distribution of Those Remaining in Paid Employment, Pooled
Sample
Figure 1d. Distribution of Weekly Wages for Future Self-Employed, Com-pared with Distribution of Those Remaining in Paid Employment in Large
Firms, Pooled Sample
- 29 -
Table 1. Comparison of Job Separation and Transitions into Self-Employment by type of employment in prior survey period. The sample consists of males whose responses are included in the SESTAT restricted file in 1995, 1997, 1999, and the SDR in 2001 and who were at least 22 in 1995 and not more than 65 in 2001. Individuals who were not in the labor force in all relevant periods are eliminated from the sample. Individuals whose highest degrees were not in a science or engineering field are also eliminated from the sample, as are all individuals who reported working fewer than 30 hours per week on average and fewer than 30 weeks per year.
Fraction of Employees in: 1997 1999 2001 All Years
TurnoverSelf-
Employed TurnoverSelf-
Employed TurnoverSelf-
Employed TurnoverSelf-
EmployedEmployer Type in Prior Survey Episode
Bus: 1 - 25 37.2% 15.7% 40.8% 17.4% 36.0% 17.9% 38.5% 16.7%Bus: 26 - 100 29.7% 5.3% 33.6% 5.8% 27.1% 4.2% 31.0% 5.4%Bus: 101 - 1000 26.1% 3.2% 30.1% 3.2% 28.1% 2.9% 27.9% 2.9%Bus: 1001 - 5000 20.3% 1.6% 23.7% 1.9% 23.2% 1.8% 22.0% 1.7%Bus: 5000 + 14.4% 0.9% 16.1% 1.3% 18.0% 0.9% 15.6% 1.0%Government 8.2% 0.6% 9.5% 0.8% 9.9% 0.6% 8.9% 0.7%Secondary Ed. 12.2% 1.1% 12.8% 0.9% 9.0% 1.4% 12.0% 1.1%University / Research Institute 9.6% 0.4% 11.3% 0.6% 10.6% 0.4% 10.4% 0.4%
Notes: Prior survey episode occurred two years earlier, e.g., for 1997 the prior survey episode was in 1995. Turn-over includes those who report changing jobs and those whose status changes from paid employment to self em-ployment. Table 2. Summary Statistics by Self- Employment Status and Firm Size. The sample consists of males whose responses are included in the SESTAT restricted file in 1995, 1997, 1999, and the SDR in 2001 and who were at least 22 in 1995 and not more than 65 in 2001. Individuals who were not in the labor force in all relevant periods are eliminated from the sample. Individuals whose highest degrees were not in a science or engineering field are also eliminated from the sample, as are all individuals who reported working fewer than 30 hours per week on aver-age and fewer than 30 weeks per year. Self-Employed 1-25 26-100 100-1000 1000-5000 5000+
Age 46.3 40.4 38.6 38.3 39.1 39.7Year 1997.2 1997.2 1997.2 1997.2 1997.1 1997.3Years in Current Job 8.6 5.1 4.4 4.5 5.6 7.0Hours worked 49.2 47.8 47.5 47.0 46.6 46.9Weeks Worked 50.6 51.4 51.7 51.6 51.8 51.8Salary (median) 60,000 52,000 57,550 58,000 62,000 69,000HD: Bachelors' .424 .474 .496 .508 .485 .408HD: Masters' .136 .175 .192 .201 .201 .207HD: PhD. .439 .351 .311 .290 .314 .385HDF Computer .077 .104 .114 .119 .128 .131HDF Life Science .201 .161 .144 .131 .122 .096HDF Phys Science .116 .141 .157 .150 .162 .166HDF Soc Science .314 .191 .134 .122 .102 .083HDF Engineering .292 .401 .451 .478 .486 .523White .814 .777 .760 .743 .746 .735Commercial Activity Count 2.27 2.13 1.98 1.82 1.67 1.49Research Activity Count 1.24 1.71 1.87 1.92 1.97 2.18Primary Activity is R&D .167 .273 .318 .338 .368 .432N 8,846 8,505 7,975 15,379 11,622 34,078
-
Table 3. Probit Analysis of Transition into Self-Employment from Paid Employment at For-Profit Firms (marginal effects, pooled sample). The sample consists of males whose responses are included in the SESTAT restricted file in 1995, 1997, 1999, and the SDR in 2001 and who were at least 22 in 1995 and not more than 65 in 2001. Individuals who were not in the labor force in all relevant periods are eliminated from the sample. Individuals whose highest degrees were not in a science or engineering field are also eliminated from the sample, as are all individuals who reported working fewer than 30 hours per week on aver-age and fewer than 30 weeks per year. The dependent variable is SELF-EMPLOYEDt+2. All regressions include only those who were not self employed at time t. All covariates are at time t+2, unless otherwise specified. Standard errors, clustered on individuals are in brackets.
Description:
Baseline
Add Individual
Observables
Robustness: 1997-
2001
Robustness: In-dustry & Firm Age
Opportunity Cost
High & Low Abil-
ity?
High & Low Abil-
ity? Column: (1) (2) (3) (4) (5) (6) (7)
Bus: 1 – 25t ***.1985 [.0086] ***.1707 [.0085] ***.1714 [.0112] ***.1405 [.0109] ***.1725 [.0086] ***.1699 [.0087] ***.1670 [.0086] Bus: 26 – 100t ***.0674 [.0061] ***.0604 [.0060] ***.0584 [.0079] ***.0490 [.0075] ***.0608 [.0060] ***.0603 [.0060] ***.0591 [.0059] Bus: 101 – 1000t ***.0297 [.0035] ***.0263 [.0034] ***.0226 [.0046] ***.0189 [.0043] ***.0264 [.0034] ***.0263 [.0034] ***.0259 [.0034] Bus: 1001 – 5000t ***.0131 [.0035] ***.0113 [.0033]
**.0123 [.0046]
**.0103 [.0043]
***.0121 [.0033]
***.0112 [.0033]
***.0111 [.0032]
Bus: 5000 +t omitted omitted omitted omitted omitted omitted omitted
Age .0003 [.0003] .0001 [.0005] .0002 [.0005] .0005 [.0004] .0001 [.0004] .0000 [.0004] Age Squared * 100 .0010 [.0007] .0013 [.0015] .0011 [.0010] .0003 [.0007] †.0013 [.0007] †.0013 [.0007] Job Tenure t .0004 [.0003] -.0000 [.0004] -.0002 [.0004] -.0000 [.0003] .0004 [.0004] .0004 [.0003] Job Tenure Squaredt* 100 -.0011 [.0011] .0002 [.0015] .0007 [.0015] .0007 [.0017] -.0013 [.0016] -.0013 [.0011] White †-.0034 [.0018] -.0027 [.0024] -.0025 [.0036] -.0045 [.0026] *-.0038 [.0018] *-.0041 [.0018] Spouse works full time .0031 [.0021] †.0054 [.0029] †.0052 [.0029] .0029 [.0021] .0032 [.0020] .0034 [.0020] Spouse works part time .0001 [.0038] -.0024 [.0036] -.0023 [.0036] .0001 [.0025] -.0003 [.0024] -.0001 [.0025] Spouse does not work †-.0042 [.0021] -.0036 [.0030] -.0034 [.0030] -.0046 [.0032] *-.0048 [.0021] -.0049 [.0021] Children in the Household *.0015 [.0006] .0015 [.0009]
.0013 [.0009]
.0013 [.0009]
*.0015 [.0063]
*.0016 [.0006]
Year * Degree **Y **Y *Y *Y **Y **Y **YHDF * Degree N ***Y ***Y ***Y ***Y ***Y ***YRegional Dummiest N *Y *Y *Y *Y *Y *YIndustry Dummies N N ***Y ***Y N N NEmployer < 5 years old t **.0178 [.0079]
Log Weekly Waget .0020 [.0015]Weekly Wage Quintilet == 1 *.0058 [.0031] Weekly Wage Quintilet == 2 -.0026 [.0023] Weekly Wage Quintilet == 4 -.0012 [.0021] Weekly Wage Quintilet == 5 ***.0072 [.0022] Weekly Waget Percentile / 100 ***-.0593 [.0106] Weekly Waget Pctle. / 100 Squared
***.0601 [.0094]
Obs P. .0356 .0356 .0370 .0370 .0356 .0356 .0356N 37,642 37,628 20,134 20,134 37,607 37,628 37,628Log Pseudolikelihood
-4967.9 -4823.3 -2640.1 -2599.9 -4822.4 -4807.7 -4799.4
Pseudo-R2 .1418 .1667 .1706 .1834 .1667 .1694 .1708
30 -
*** = significant at p ≤ 0.001; ** = significant at p ≤ 0.01; * = significant at p ≤ 0.05; † = significant at p ≤ 0.1 (two-sided test) Note: For firm size category variables, the omitted variable is more than 5000 employees. Non-married is the omitted category for marital status.
Table 4. Probit Analysis of Transition into Self-Employment incorporating Job Activities in Paid Employ-ment (marginal effects, pooled sample). The sample consists of males whose responses are included in the SES-TAT restricted file in 1995, 1997, 1999, and the SDR in 2001 and who were at least 22 in 1995 and not more than 65 in 2001. Individuals who were not in the labor force in all relevant periods are eliminated from the sample. In-dividuals whose highest degrees were not in a science or engineering field are also eliminated from the sample, as are all individuals who reported working fewer than 30 hours per week on average and fewer than 30 weeks per year. The dependent variable is SELF-EMPLOYEDt+2. Coefficient for age, age squared, job tenure, job tenure squared, marital status, race, and number of children are suppressed. All regressions include only those who were not self employed at time t. All covariates are at time t+2, unless otherwise specified. Standard errors, clustered on individuals are in brackets.
Baseline Activity Count Activity Count --Non monotonic
Full activity vec-tor
Activities X De-grees
Column: (1) (2) (3) (4) (5) Bus: 1 – 25t ***.1670 [.0086] ***.1590 [.0084] ***.1580 [.0084] ***.1439 [.0082] ***.1416 [.0081] Bus: 26 – 100t ***.0591 [.0059] ***.0552 [.0057] ***.0550 [.0057] ***.0513 [.0056] ***.0508 [.0055] Bus: 101 – 1000t ***.0259 [.0034] ***.0241 [.0033] ***.0241 [.0033] ***.0224 [.0032] ***.0221 [.0032] Bus: 1001 – 5000t ***.0111 [.0032] ***.0101 [.0032] ***.0099 [.0032] ***.0091 [.0031] ***.0089 [.0030] Bus: 5000 +t omitted omitted omitted omitted omitted Year * Degree **Y *Y *Y **Y **Y HDF * Degree ***Y ***Y ***Y ***Y ***Y Regional Dummiest *Y *Y *Y *Y *Y Commercial Activities
Count **.0015 [.0005] 1 .0014 [.0022] 2 .0010 [.0023] 3 .0018 [.0022] 4 *.0060 [.0023] 5 or more **.0122 [.0048]
Activity Dummies ***Y ***Y Act. Dummies X PhD. ***Y
R&D Activities Count ***-.0025 [.0006]
1 ***-.0082 [.0016] 2 ***-.0110 [.0017] 3 or more ***-.0110 [.0018]
Activity Dummies ***Y **Y Act. Dummies X PhD. **Y
Weekly Wage Percentilet / 100 ***-.0593 [.0106] ***-.0504 [.0106] ***-.0459 [.0106] ***-.0482 [.0104] ***-.0461 [.0103] Weekly Wage Percentilet / 100 Squared
***.0601 [.0094] ***.0518 [.0094] ***.0485 [.0094] ***.0503 [.0092] ***.0486 [.0092]
Obs. P. .0356 .0356 .0356 .0356 .0356 N 37,628 37,628 37,628 37,628 37,628 Log Likelihood -4799.4 -4872.7 -4769.8 -4740.3 -4719.5 Pseudo-R2 .1708 .1737 .1759 .1810 .1846 *** = significant at p ≤ 0.001; ** = significant at p ≤ 0.01; * = significant at p ≤ 0.05 (two-sided test) Note: For firm size category variables, the omitted variable is more than 5000 employees.
- 31 -
- 32 -
Table 5. Multinomial Logit Analysis of Likelihood of Entering Self-Employment or Changing Jobs in Paid Employment. The sample consists of males whose responses are included in the SESTAT restricted file in 1995, 1997, 1999, and the SDR in 2001 and who were at least 22 in 1995 and not more than 65 in 2001. Individuals who were not in the labor force in all relevant periods are eliminated from the sample. Individuals whose highest degrees were not in a science or engineering field are also eliminated from the sample, as are all individuals who reported working fewer than 30 hours per week on average and fewer than 30 weeks per year. The dependent variable con-sists of three choices, (1) the individual does not change employers, (2) the individual changes employer but does not become self employed, and (3) the individual leaves paid employment and becomes self employed. In the re-sults reported below, (1) is the omitted choice. All regressions include only those who were not self employed at time t. All covariates are at time t+2, unless otherwise specified. Standard errors, clustered on individuals are in brackets.
Choice:
Change Jobs, Not Self Employed
Self-Employed Change Jobs, Not Self Employed
Self-Employed
Column: (1a) (1b) (2a) (2b) Bus: 1 – 25t ***.5269 [.0485] ***2.7080 [.0972] ***.5009 [.0489] ***2.7129 [.0979] Bus: 26 – 100t ***.5709 [.0456] ***1.6321 [.1104] ***.5539 [.0459] ***1.6462 [.1129] Bus: 101 – 1000t ***.4982 [.0367] ***1.0366 [.1284] ***.4835 [.0370] ***1.0479 [.1085] Bus: 1001 – 5000t ***.2986 [.0408] *.5071 [.1283] ***.2904 [.0411] *.5109 [.1296] Bus: 5000 +t Omitted omitted Omitted omitted Age ***-.0257 [.0073] .0001 [.0157] ***-.0241 [.0075] -.0058 [.0158] Age Squared * 100 .0088 [.0167] .0533 [.0320] .0045 [.0167] .0593 [.0320] Job Tenure t ***-.1226 [.0067] -.0086 [.0123] ***-.1201 [.0068] -.0063 [.0124] Job Tenure Squaredt* 100 ***.2790 [.0286] .0234 [.0467] ***.2698 [.0291] .0128 [.0473] White ***-.1432 [.0334] *-.1897 [.0767] ***-.1416 [.0335] *-.1993 [.0767] Spouse works full time -.0079 [.0389] .0928 [.0859] .0021 [.0392] .1100 [.0864] Spouse works part time *-.1258 [.0511] -.0486 [.1072] *-.1167 [.0512] -.0532 [.1072] Spouse does not work .0225 [.0450] *-.2197 [.1014] .0302 [.0451] *-.2355 [.1014] Children in the Household *-.0305 [.0141] .0524 [.0276] *-.0493 [.0142] *.0548 [.0275] Year * Degree ***Y *Y ***Y *Y HDF * Degree ***Y ***Y ***Y ***Y Regional Dummiest ***Y **Y ***Y *Y Research Activity Dummies **Y *Y ***Y *Y Commercial Activity Dummies **Y ***Y **Y ***Y Log Weekly Waget ***-.1881 [.0284] .0098 [.0642] Weekly Wage Quintilet == 1 ***.4459 [.0545] **.3236 [.1145] Weekly Wage Quintilet == 2 **.1303 [.0448] -.0685 [.1115] Weekly Wage Quintilet == 4 .0018 [.0407] -.0167 [.1005] Weekly Wage Quintilet == 5 -.0203 [.0441] ***.3095 [.0940] N 37,628 37,628 Log pseudo-likelihood -21606.8 -21561.4 Pseudo R2 .1017 .1029
*** = significant at p ≤ 0.001; ** = significant at p ≤ 0.01; * = significant at p ≤ 0.05 (two-sided test)
Table 6. Prospective analysis of current salary on likelihood of moving into self-employment in subsequent periods for pooled sample. The sample con-sists of males whose responses are included in the SESTAT restricted file in 1995, 1997, 1999, and the SDR in 2001 and who were at least 22 in 1995 and not more than 65 in 2001. Individuals who were not in the labor force in all relevant periods are eliminated from the sample. Individuals whose highest degrees were not in a science or engineering field are also eliminated from the sample, as are all individuals who reported working fewer than 30 hours per week on aver-age and fewer than 30 weeks per year. The dependent variable is the log of the weekly wage at time t. All regressions include only those who were not self em-ployed at time t. The coefficients in the wage equation are suppressed. These include the state in which the employee works, the employee’s highest degree, the field of the highest degree, age and age squared, job tenure and job tenure squared, race dummy variables, a small firm dummy variable, and year dummies. For the OLS specification, robust standard errors are in brackets. For the quantile regressions, we present bootstrap standard errors with 100 repetitions. N = 37,607. Analysis
Series
Key Coefficients
OLS
Quantile: 10%
Quantile: 25%
Quantile: 50%
Quantile: 75%
Quantile: 90%
(1) SEt+2 -.0186 [0196] ***-.1485 [.0355]
*-.0431 [.0207]
*.0457 [.0158]
***.0779 [.0165]
***.1215 [.0168]
(2) SMALLFIRMt *SEt+2 -.0421 [.0286] **-.2009 [.0665] *-.0799 [.0348] *.0637 [.0227] ***.1188 [.0259] ***.1255 [.0257] LARGEFIRMt *SEt+2 .0190 [.0219] *-.0833 [.0385]
-.0192 [.0182]
.0283 [.0219]
***.0609 [.0171]
**.0858 [.0297]
(3) SMALLFIRMt *RNDTRACKt* SEt+2 -.0835 [.0618] -.1381 [.1211] -.0215 [.0794] -.0209 [.0472] .0583 [.0499] .0424 [.0453]
SMALLFIRMt *NOTRNDTRACKt* SEt+2
-.0307 [.0321] **-.2219 [.0741] -.0845 [.0296] ***.0751 [.0219] ***.1279 [.0295] ***.1571 [.0284]
LARGEFIRMt *RNDTRACKt* SEt+2 .0414 [.0256] .0272 [.0450] .0241 [.0290] .0200 [.0266] .0287 [.0311] .0274 [.0372] LARGEFIRMt *NOTRNDTRACKt*
SEt+2
.0119 [.0276] **-.1161 [.0401] †-.0417 [.0245] .0367 [.0315] **.0721 [.0253] **.1296 [.0428]
*** = significant at p ≤ 0.001; ** = significant at p ≤ 0.01; * = significant at p ≤ 0.05; † = significant at p ≤ 0.1 (two-sided test)
-33-
-3
Table 7. Performance in Self-Employment among newly self-employed by size of previous employer. The sample consists of all members of the pooled sample who moved from employment in a for-profit business to self-employment. For the probit analysis, the coefficients presented are marginal effects. The independent variable “Comm. Activities” is the count of commercial activities performed by the individual in his job prior to entering self-employment, and “Res. Activities,” similarly is the count of research activities in the prior job. The categories for the ordered probit analysis are (1) 0 employees, (2) 1-4 direct reports, (3) 5-16 direct reports, (4) 17-64 direct reports, and (5) 65 or more direct reports. Standard errors are in brackets.
Dependent Variable:
Persistence in Self Employment
Entry as Incorporated Entity
Number of Direct Reports in New Firm
Specification: Probit (marginal effects)
Probit (marginal effects)
Ordered Tobit
Column:
(1) (2)
(3)
(4) (5)
(6)
(7)
(8)
(9)
Firm Size: 1-25 t-2 *.1023 [.0385]
*.0972 [.0476]
*.0862 [.0481]
*.0599 [.0262]
***.1049 [.0318]
***.1119 [.0321]
***.2096 [.0606]
***.2723 [.0748]
*.1715 [0763]
Firm Size: 26-100t-2
-.0484 [.0639]
-.0480 [.0656]
*.1021[.0406]
*.1029 [.0409]
***.3396 [.0979]
**.2548 [.0994]
Firm Size: 5000+t-2 .0387 [.0669]
.0700 [.0669]
.0685 [.0432]
.0503 [.0432]
-.1712[.1073]
-.1357 [.1094]
Year = 1999 .0605 [.0614]
.0603 [.0615]
.0557 [.0630]
-.0281 [.0279]
-.0304 [.0280]
-.0285 [.0281]
.0993 [.0638]
.0975 [.0640]
*.1308 [.0649]
Year = 2001 ***.2798 [.0541]
***.2805 [.0542]
***.2847 [.0547]
-.0633 [.0469]
-.0622 [.0469]
-.0643 [.0471]
†.1890 [.1092]
†.1963 [.1093]
†.2581 [.1108]
HD: Master’s .0492 [.0649]
.0512 [.0650]
.0667 [.0638]
-.0102 [.0374]
-.0118 [.0374]
-..0251 [.0380]
**-.2223 [.0856]
**-.2265 [.0858]
-.1385 [.0876]
HD: PhD -.0633 [.0605]
-.0643 [.0614]
-.0371 [.0644]
†-.0625 [.0343]
†-.0647 [.0344]
*-.0799 [.0351]
***-.3832 [.0795]
***-.3822 [.0797]
***-.2760 [.0820]
Age *.0208 [.0106]
†.0203 [.0106]
†.0193 [.0108]
†.0100 [.0060]
†.0108 [.0060]
*.126 [.0061]
.0203 [.0137]
.0208 [.0138]
.0088 [.0141]
Age Squared * 100 -.0318 [.0212]
-.0307 [.0212]
-.0311 [.0215]
†-.0215 [.0118]
†-.0229 [.0119]
*-.0249 [.0120]
*-.0584 [.0277]
*-.0601 [.0278]
-.0282 [.0278]
White ***.1757 ***.1775 [.0464] [.0467]
***.1721 [.0480]
-.0367 [.0320]
-.0375 [.0321]
-.0263 [.0323]
*-.1566 [.0735]
*-.1752 [.0736]
*-.1830 [.0750]
Log Salaryt-2 .0101 [.0184]
.0095 [.0184]
.0072 [.0189]
***.0608 [.0152]
***.0611 [.0152]
***.0586 [.0153]
***.1245 [.0355]
***.1297 [.0355]
***.1033 [.0360]
Comm. Activities t-2 .0120 [.0132]
-.0012[.0092]
***.2591[.0216]
Res. Activities t-2 ***-.0503[.0138]
***.0435[.0096]
*.0581[.0221]
N 735 735 735 1522 1522 1522 1522 1522 1522Observed P. .5497 .5497 .5497 .5591 .5591 .5591 62.77 83.84 236.82Log Likelihood -472.0 -471.9 -464.4 -1029.4 -1026.1 -1015.9 -1584.9 -1574.4 -1497.8Pseudo R2 .0669 .0683 .0818 .0142 .0174 .0272 .0194 .0259 .0733
4-
*** = significant at p ≤ 0.001; ** = significant at p ≤ 0.01; * = significant at p ≤ 0.05; † = significant at p ≤ 0.1 (two-sided test)
Table 8. Censored normal regression analysis of first period self-employment earnings by size of previous employer. The sample consists of all members of the pooled sample who moved from employment in a for-profit business to self-employment. The dependent variable is the log of salary in the first period of self-employment. The dependent variable is considered top-censored if salary is greater than or equal to 150,000, and it is considered bot-tom-censored if salary equals 0. Firm size and salary variables refer to the individual’s employer immediately prior to transitioning into self-employment and are measured at t-2. State dummy variables (e.g., AK, AR, AZ, etc.), race dummy variables (African-American, Asian, and Hispanic), and dummy variables for the field of the individual’s highest degree (computer science, physical science, life science, social science, and engineering) are included in the regressions below but are not reported. Standard errors are in brackets. Subset:
Entire Sample
Top Half of Wage Earnerst-2
Bottom Half of Wage
Earnerst-2 (1) (2) (3) (4) (5) (6) Firm Size: 1-25 t-2 .0479
[.0596] .0255
[.0723] *.1498
[.0744] .1385
[.0882] -.0861
[.0946] -.0869
[.1210] Firm Size: 26-100 t-2 -.0329
[.0962] -.0014
[.1178] -.0290
[.1550] Firm Size: 5000+ t-2 -.0592
[.1015] -.0458
[.1173] .0531
[.1900] Year = 1999 -.0736
[.0624] -.0723
[.0625] -.0221
[.0772] -.0216
[.0772] †-.1742 [.0975]
†-.1749 [.0980]
Year = 2001 -.0499 [.1076]
-.0508 [.1076]
.1622 [.1396]
.1622 [.1396]
*-.3554 [.1686]
*-.3546 [.1688]
HD: Master’s .0730 [.0678]
.0780 [.0847]
.0705 [.1025]
.0707 [.1025]
-.1282 [.1442]
.-.1244 [.1444]
HD: PhD *.2997 [.1187]
*.3225 [.1485]
-.2565 [.2022]
-.2511 [.2028]
***.8308 [.2119]
***.8302 [.2120]
Age .0109 [.0171]
.0103 [.0171]
-.0111 [.0230]
-.0116 [.0231]
-.0121 [.0242]
-.0120 [.0242]
Age Squared * 100 -.0346 [.0317]
-.0335 [.0318]
-.0010 [.0421]
.0019 [.0423]
-.0126 [.0461]
-.0127 [.0461]
Log Weekly Waget-2 ***.5100 [.0387]
***.5105 [.0388]
***1.1833 [.1214]
***1.1833 [.1215]
*.1070 [.0543]
*.1063 [.0543]
N 1520 1520 987 987 533 533 Log Likelihood -2164.8 -2164.6 -1335.5 -1335.4 -736.3 -736.2 Pseudo R2 .0622 .0622 .0722 .0720 .0576 .0577 *** = significant at p ≤ 0.001; ** = significant at p ≤ 0.01; * = significant at p ≤ 0.05; † = significant at p ≤ 0.1 (two-sided test) Note: In column 4 the difference between the coefficients on Firm Size: 1-25 t-2 and Firm Size: 5000+ t-2 are statistically different at the p < .1 level.
- 35 -
Table 9. Inverse Propensity Score Weighted Analysis of Performance. The sample consists of all respondents who moved from employment in a for-profit business to self-employment. In row (1), the dependent variable is dichotomous, equal to 1 if the individual transitioning into self employment was self-employed at both time t and time t+2 and equal to 0 if the individual was self-employed at time t but reported working in another job at time t+2. In row (2), the dependent variable is equal to 1 if the individual entered self employment as an incorporated entity and 0 otherwise. In row (3) , the dependent variable is equal to 1 if the respondent had 0 direct reports, 2 if the re-spondent had 1-4 direct reports, and 3, 4, and 5 if the respondent had 5-16, 17-64, or 65 or more direct reports, re-spectively. In rows (4)-(6), the dependent variable is the log of the salary reported in the first period of self em-ployment, top-coded at 150,000. Firm size and salary variables refer to the individual’s employer immediately prior to transitioning into self-employment and are measured at t-2. Propensity scores (for being in the Firm Size: 1-25t-2 category) are estimated using variables from t-2 including age, education, gender, job tenure, salary, and location variables. Standard errors are in brackets.
Average Treatment
Treatment on Treated
Row Dependent Variable Column: (1) (2)
(1) Persistence in Self-Employment (Linear Probability)
Coeff: †.0752 [.0402] †.0794 [.0435]
N 735 735 F-statistic †3.33 †2.16
(2) Entry as Incorporated Entity (Linear Probability)
Coeff: *.0628 [.0269] *.0712 [.0288]
N 1522 1522 F-statistic *5.44 *6.12
(3) Entry Size (OLS)
Coeff: *.0990 [.0434] *.0956 [.0479]
N 1522 1522 F-statistic *4.66 *3.99
(4) Log(Salaryt) (OLS)
Coeff: .0341 [.0573] .0379 [.0650]
Entire Sample N 1522 1522 F-statistic 0.35 .034
(5) Log(Salaryt) (OLS)
Coeff: *.1719 [.0709] **.2207 [.0782]
Top Half of Wage Earnerst-2 N 988 988 F-statistic *5.88 *7.95
(6) Log(Salaryt) (OLS)
Coeff: -.1005 [.0849] -.1256 [.1021]
Bottom Half of Wage Earnerst-2 N 534 534 F-statistic 1.40 1.51
*** = significant at p ≤ 0.001; ** = significant at p ≤ 0.01; * = significant at p ≤ 0.05; † = significant at p ≤ 0.1 (two-sided test)
- 36 -
Appendix. Exploring Differences between Employees in Small and Large Firms
Table A1 below presents the results of a regression of the activity count on individual fixed ef-fects, firm size dummies, age, and job tenure. The coefficients on firm size in Table B1 above are identi-fied by the movements of individuals between firms of different sizes (or size changes in the firms in which they remained employed). The results indicate that in moving from the largest firm category to a small firm, an individual is likely to perform significantly more commercial activities and is significantly more likely to be a supervisor (unless moving to the smallest firms). The individual is no more or less likely to be engaged in research activities. Figure A1 illustrates the current distribution of wages within large firms, comparing those who remain employed in the large firms and those who move to small firms. The figure indicates that those who move to small firms come disproportionately from the tails of the current pay distribution. We at-tempt to corroborate this finding using a prospective regression explaining current log weekly wage as a function of observable individual characteristics and a dummy variable SMALL-FIRMt+2 which indicates whether the employee was working in a firm of size 100 or less at time t+2. We display the results of quantile regressions with bootstrap standard errors, calculated with 100 repetitions, in Table B2. The es-timates indicate that after controlling for individual characteristics, the 90th percentile of pay for those who move from large to small firms is significantly higher than those who stay. We interpret these result as indicating the following: (1) low pay workers in large firms join small firms at a disproportionate rate, although this may be driven by compositional effects, (2) high-pay workers in large firms join small firms at a disproportionate rate, consistent with stronger pay-performance links in small firms. Figure A1. Comparison of wage distribution in large firms between those who leave to join small firms and those who stay in large firms.
0.5
1Ke
rnel
Den
sity
5 6 7 8 9Log of (Weekly Wage + 10)
Stay in large firms Move to small firms
Pooled SampleCurrent Paid-Employment Wages of Large-Firm Employees Who Stay vs. Leave to Join Small Firms
- 37 -
Table A1. Fixed effect regression of activities on firm size. The sample consists of males whose responses are included in the SESTAT restricted file in 1995, 1997, 1999, and the SDR in 2001 and who were at least 22 in 1995 and not more than 65 in 2001. Individuals who were not in the labor force in all relevant periods are eliminated from the sample. Individuals whose highest degrees were not in a science or engineering field are also eliminated from the sample, as are all individuals who reported working fewer than 30 hours per week on average and fewer than 30 weeks per year.
Dependent Variable
Number of Com-mercial Activities
Number of Research
Activities
Supervisor
Column: (1) (2) (3) Bus: 1 – 25t ***.1278 [.0287] .0011 [.0275] .0152 [.0114] Bus: 26 – 100t ***.1278 [.0254] -.0210 [.0243] ***.0377 [.0101] Bus: 101 – 1000t ***.0903 [.0198] -.0312 [.0190] **.0206 [.0078] Bus: 1001 – 5000t .0288 [.0185] *-.0482 [.0177] .0033 [.0074] Bus: 5000 +t omitted omitted omitted Age ***.1015 [.0057] ***-.0832 [.0054] ***.0445 [.0023] Age Sq ***-.0018 [.0001] ***-.0010 [.0001] ***-.0008 [.0000] Job Tenure **.0069 [.0027] ***.0133 [.0027] ***.0107 [.0011] Job Tenure Sq **-.0003 [.0001] **-.0003 [.0001] ***-.0003 [.0001] Individuals 40,770 40,770 40,770 Obs / Individual 1.9 1.9 1.9 F 50.16 56.65 74.18
*** = significant at p ≤ 0.001; ** = significant at p ≤ 0.01; * = significant at p ≤ 0.05; † = significant at p ≤ 0.1 (two-sided test) Table A2. Prospective analysis of current salary of large firm employees, comparing those who stay in small firms with those who leave to join small firms. The sample consists of males whose responses are included in the SESTAT restricted file in 1995, 1997, 1999, and the SDR in 2001 and who were at least 22 in 1995 and not more than 65 in 2001. Individuals who were not in the labor force in all relevant periods are eliminated from the sample. Individuals whose highest degrees were not in a science or engineering field are also eliminated from the sample, as are all individuals who reported working fewer than 30 hours per week on average and fewer than 30 weeks per year. The dependent variable is the log of the weekly wage at time t. All regressions include only those who were not self employed at time t, and all individuals who leave large firms to become entrepreneurs are excluded from the analysis as well. The coefficients in the wage equation are suppressed. These include the state in which the em-ployee works, the employee’s highest degree, the field of the highest degree, age and age squared, job tenure and job tenure squared, race dummy variables, a small firm dummy variable, and year dummies. N = 23,849.
Quantile:
10%
25%
50%
75%
90%
SMALL-FIRMt+2 -.0265
[.0194] -.0184
[.0125] .0037
[.0140] †.0191
[.0100] **.0572 [.0191]
*** = significant at p ≤ 0.001; ** = significant at p ≤ 0.01; * = significant at p ≤ 0.05; † = significant at p ≤ 0.1 (two-sided test)
- 38 -
ec1 e-companion to Author: Tis a Butter Place
This page is intentionally blank. Proper e-companion title page, with
INFORMS branding and exact metadata of the main paper, will be pro-
duced by the INFORMS office when the issue is being assembled.
1
ec2 e-companion to Elfenbein, Hamilton, Zenger: The Entrepreneurial Spawning of Scientists and Engineers: Stars, Slugs and the Small Firm Effect
Sample Construction and Data Description
We construct a sample of individuals with science and engineering degrees using data from the
Scientists and Engineers Statistical Data System (SESTAT). This data file is comprised of responses to
three separate surveys—the National Survey of Recent College Graduates (NSRCG), the National Survey
of College Graduates (NSCG), and the Survey of Doctoral Recipients (SDR). All survey responses in
SESTAT are restricted to respondents who earned a science or engineering degree (S&E). The sampling
methodologies vary widely across each of these three surveys.18 For example, the sample population for
the 1993 NSCG was chosen by the Bureau of the Census to be representative of all college graduates in
all fields as of 1990; SESTAT reports a sub-sample of these respondents who received S&E degrees or
were employed in an S&E field. The NSRCG sampled S&E degree recipients from the prior two-year
window, and the SDR defined as its sample population all people who had received an S&E doctorate
from a U.S. institution by the year preceding the survey. For each of these survey programs, individuals
responded to multiple survey episodes, allowing us to track their behavior across time. We combine data
from all three surveys in 1995, 1997, and 1999 and augment it with data from the SDR in 2001.19 We
make the following additional restrictions to eliminate sources of undesirable heterogeneity:
• To avoid problems of retirement, full-time education, and other choices about whether to enter or
remain in the labor force, we eliminate all of those who are not in the labor force in each year be-
tween 1995 and 2001 and further eliminate all of those under age 22 or above age 65 in any year
between 1995 and 2001. Together these eliminations reduce the sample by roughly 10%.
• To avoid some of the impact of family choices on self-employment, we focus only on males (who
comprise roughly 70% of survey responses).
18 For details, see http://sestat.nsf.gov/. 19 Although 2003 data are available for the SDR, we do not include these data in the present analysis. A change in the survey design in 2003 appears to generate a sharp spike in the number of respondents listing themselves as in-corporated self-employed; to eliminate this potential source of error, we discard the 2003 data. In future work, we can establish which of the respondents listed this choice mistakenly, based upon their responses to other survey questions.
ec3 e-companion to Elfenbein, Hamilton, Zenger: The Entrepreneurial Spawning of Scientists and Engineers: Stars, Slugs and the Small Firm Effect
• Since we use measures derived from annual salary in our analysis below, we wish to avoid con-
founding total pay with choices about working part-time vs. full time. Therefore, we eliminate
from the sample all those who report working fewer than 30 hours per week and all those who re-
port working fewer than 30 weeks per year (approximately 6% of survey responses).
• Because we want to focus exclusively on scientists and engineers, we eliminate all those whose
highest degree was not in a science and engineering field and further, we drop from our analysis
any individual who also holds a professional degree (such as an M.D., J.D., DVM., etc.).20 This
group comprises approximately 8.5% of survey responses.
• To avoid confounds due to currency differences, all respondents working outside the United
States are excluded from the sample. This group represents less than .01% of survey responses.
We use all survey responses meeting the criteria described above to generate Table 1. As dis-
cussed in the Introduction, Table 1 illustrates the likelihood that an individual working for an employer of
a given type in year t-2 has either changed jobs, labeled “turnover” in the table, or has become self-
employed by year t. The turnover category includes transitions to self-employment, and may include
some instances of individuals who have become owners in the firms for which they now work.
Table A1 provides summary statistics about individual demographics, such as age, race, and
marital status; individual-job characteristics, such as job tenure, reports for hours and weeks worked, sal-
ary, and a set of characteristics about the individual’s activities on the job; educational attainment and the
field of the individual’s highest degree; employer characteristics, such as the size21 or age of the employ-
ing firm if the individual is in paid employment, and indicator variables about whether the individual is
self-employed; a set of characteristics about the individual’s activities on the job; the main industry of the
20 Masters in Business Administration (MBA) degrees are not considered by the NSF to be professional degrees, so PhDs who hold MBAs are included in the sample. 21 To simplify the reporting we collapse size categories 1-10 and 11-25 employees into one size category, and we do the same for size categories 101-500 and 501-1000. In unreported analyses we test the robustness of our analyses by using all of the size categories available and also eliminating the 1-10 category. These robustness checks yield iden-tical results to the ones we report below.
ec4 e-companion to Elfenbein, Hamilton, Zenger: The Entrepreneurial Spawning of Scientists and Engineers: Stars, Slugs and the Small Firm Effect individual in self-employment or the main industry of the employer firm; and location. Salary data is
generally top-coded at $150,000 for the NRCG and NSRCG surveys, but not for the SDR survey. We
top-code the data from the SDR survey at $150,000 and use only the top-coded salary in the analysis in
the main section of the paper. 22 Data about firm age and industry were collected beginning in 1997, and
consequently the number of individual observations with these data is substantially smaller. We report
fourteen “activities on the job,” which are responses to a series of survey questions asking whether the
individual spent more than 10% of a typical work-week on the activity in question. Additionally, we re-
port a dummy variable indicating whether the individual claimed that his primary activity in employment
was R&D.
22 SESTAT provides two salary figures, each of which was constructed by the surveying agencies by combining information on weekly / monthly earnings and weeks / months worked. In the public use data they provide a salary measure that is top-coded at 150,000. In the restricted data file, they provide a salary measure that is top coded at 999,996. We are skeptical that this measure reports large salary figures accurately for respondents to the NSRCG and NSCG surveys given the large mass of observations at the $150,000 level; a similar mass of observations on $150,000 is not a characteristic of the SDR responses. In the online appendix, we break the sample into two parts: those whose highest degree is a PhD.s (and whose data come from the SDR) and those whose highest degree is an MA or BA. For the PhD sample, we repeat the analysis in Section 4 using the salary provided in the restricted data file, which is not top-coded at $150,000. This analysis shows the same pattern as in the pooled data, i.e. a dispropor-tionate share of the highest-earning PhDs transition into self-employment.
ec5 e-companion to Elfenbein, Hamilton, Zenger: The Entrepreneurial Spawning of Scientists and Engineers: Stars, Slugs and the Small Firm Effect Table EC1. Summary Statistics for Scientists and Engineers Working in For-Profit Enterprise. The sample consists of males whose responses are included in the SESTAT restricted file in 1995, 1997, 1999, and the SDR in 2001 and who were at least 22 in 1995 and not more than 65 in 2001. Individuals who were not in the labor force in all relevant periods are eliminated from the sample. Individuals whose highest degrees were not in a science or en-gineering field are also eliminated from the sample, as are all individuals who reported working fewer than 30 hours per week on average and fewer than 30 weeks per year. Workers in government, university / research institutes, secondary or primary education, defense, and other non-profits are excluded.
Obs Mean Median Std Dev Min Max Age 86,405 40.0 39 10.4 22 65 Year 86,405 1997.2 1997 1.92 1995 2001 Years in Current Job 86,405 6.1 3.25 6.9 0 44.9 Hours worked in primary job 86,405 47.3 45 8.2 30 80 Weeks worked in primary job 86,405 51.6 52 1.7 30 52 Salary 86,405 69,774 63,000 46,601 0 999,996 Salary, top-coded as 150,000 86,405 67,495 63,000 32,695 0 150,000 Highest Degree: Bachelor's 86,405 .453 0 .498 0 1 Highest Degree: Master's 86,405 .193 0 .395 0 1 Highest Degree: Ph.D. 86,405 .354 0 .478 0 1 Highest Degree Field: Computer 86,405 .119 0 .324 0 1 Highest Degree Field: Life Science 86,405 .128 0 .333 0 1 Highest Degree Field: Physical Science 86,405 .154 0 .361 0 1 Highest Degree Field: Social Science 86,405 .132 0 .338 0 1 Highest Degree Field: Engineering 86,405 .468 0 .499 0 1 White 86,405 .753 1 .432 0 1 Married 86,405 .727 1 .445 0 1 Has spouse who works full time 86,405 .327 0 .469 0 1 Has spouse who works part-time 86,405 .145 0 .352 0 1 Has spouse who does not work 86,405 .261 0 .439 0 1 Children Living in Household 86,405 .98 1 1.18 0 17 Employer:
Self-Employed 86,405 .102 0 .303 0 1 Self-Employed, Incorporated 86,405 .046 0 .209 0 1 Self-Employed, Not Inc. 86,405 .056 0 .231 0 1 Business, 1-25 employees 86,405 .098 0 .298 0 1 Business, 26-100 employees 86,405 .092 0 .289 0 1 Business, 101-1000 employees 86,405 .178 0 .383 0 1 Business, 1000 – 5000 emp. 86,405 .135 0 .341 0 1 Business, 5000+ emp. 86,405 .394 0 .489 0 1 Business is 5 years old or less 58,250 .128 0 .333 0 1 Turnover 86,405 .192 0 .394 0 1
Activities on the Job: Accounting, Finance, Contracts 86,405 .264 0 .441 0 1 Applied Research 86,405 .409 1 .491 0 1 Basic Research 86,405 .161 0 .368 0 1 Computer Applications 86,405 .494 0 .500 0 1 Development 86,405 .419 1 .493 0 1 Design 86,405 .452 0 .497 0 1 Employee Relations 86,405 .309 0 .462 0 1 Managing or Supervising People 86,405 .533 1 .499 0 1 Other 86,405 .052 0 .222 0 1 Production, Operations, and Mainte-nance 86,405 .114 0 .318 0 1
Quality or Productivity Management 86,405 .282 0 .450 0 1 Sales, Purchasing, or Marketing 86,405 .314 0 .464 0 1 Professional Services 86,405 .153 0 .360 0 1 Teaching 86,405 .087 0 .282 0 1 Primary Activity is R&D 86,405 .353 0 .478 0 1
ec6 e-companion to Elfenbein, Hamilton, Zenger: The Entrepreneurial Spawning of Scientists and Engineers: Stars, Slugs and the Small Firm Effect
Obs Mean Median Std Dev Min Max Employer Main Business
Agriculture, Forestry, or Fishing 58,250 .022 0 .146 0 1 Biotechnology 58,250 .033 0 .179 0 1 Construction or Mining 58,250 .040 0 .197 0 1 Education / Public Admin. / Gov't 58,250 .005 0 .070 0 1 Finance, insurance or real estate 58,250 .052 0 .223 0 1 Health Services 58,250 .050 0 .217 0 1 Information technology 58,250 .177 0 .382 0 1 All other services 58,250 .056 0 .230 0 1 Manufacturing 58,250 .253 0 .435 0 1 Research 58,250 .094 0 .291 0 1 Transportation Services, Utilities, etc. 58,250 .058 0 .058 0 1 Wholesale or retail trade 58,250 .038 0 .191 0 1 Other 58,250 .121 0 .326 0 1
Location: New England 86,350 .074 0 .262 0 1 Mid Atlantic 86,350 .159 0 .365 0 1 South Atlantic 86,350 .154 0 .361 0 1 East North Central 86,350 .147 0 .353 0 1 West North Central 86,350 .062 0 .242 0 1 East South Central 86,350 .031 0 .174 0 1 West South Central 86,350 .100 0 .300 0 1 Mountain 86,350 .065 0 .247 0 1 Pacific 86,350 .207 0 .405 0 1