Mobility and
Entrepreneurship:
Evaluating the Scope of
Knowledge-Based
Theories of
Entrepreneurship
Lars Frederiksen
Karl Wennberg
Chanchal Balachandran
Knowledge-based theories of entrepreneurship infer transfer of knowledge from the
effect of labor mobility on entrepreneurial entry. Yet, simple selection or situational
mechanisms that do not imply knowledge transfer may influence entrepreneurial entry in
similar ways. We argue that the extent to which such alternative mechanisms operate,
labor mobility predicts entry but not subsequent performance for entrepreneurs. Analyses
of matched employee–employer data from Sweden suggest that high rates of geographical
and industry mobility increase individuals’ likelihood of entrepreneurial entry
but have no effects on their entrepreneurial performance. This indicates that the relationship
between labor mobility and entrepreneurial entry do not necessarily imply knowledge
transfer.
The flow of people between firms, industries and locations has been proposed as an
important mechanism for transferring knowledge1 and is argued as important for innovation,
firm demography, and entrepreneurship (e.g., Agarwal, Echambadi, Franco, &
Sarkar, 2004; Audretsch & Keilback, 2007). For entrepreneurship research, individual
mobility is central since it invokes the creation, transfer, and destruction of resources and
capabilities in different markets (Wright, 2011). While knowledge-based theories propose
knowledge flow as the key mechanism that connects labor mobility to entrepreneurial
entry and performance (Agarwal, Audretsch, & Sarkar, 2007; Helfat & Lieberman,
2002), they are rarely challenged by alternative selection-based mechanisms such as individual
“taste for variety” or “labor market misfits” (A˚stebro, Chen, & Thompson, 2011;
Please send correspondence to: Karl Wennberg, tel.: 146-705-10 53 66; email: karl.wennberg@liu.se.
1. Following theories of knowledge spillover, we refer to “knowledge” as an appropriable resource that is
distinct from individual “skills” (Audretsch & Keilback, 2007).
March, 2016 359
DOI: 10.1111/etap.12223
1042-2587
VC 2016 The Authors. Entrepreneurship Theory and Practice published
by Wiley Periodicals, Inc. on behalf of SAGE Publications Inc.
This is an open access article under the terms of the Creative
Commons Attribution-NonCommercial-NoDerivs License, which permits
use and distribution in any medium, provided the original work is
properly cited, the use is non-commercial and no modifications or
adaptations are made.
A˚stebro & Thompson, 2011) and situational mechanisms2 such as individual opportunity
cost (Amit, Muller, & Cockburn, 1995; Sørensen & Fassiotto, 2011). To the extent selection
or situational factors explain the relationship between labor market mobility and
entrepreneurship, knowledge-based theories that infer a treatment effect of knowledge
transfer on entrepreneurship entry may be overestimated. While we expect to find a positive
effect of labor market mobility on entrepreneurial entry through selection or situational
mechanisms, we also expect to find an adverse or no effect on entrepreneurial
performance if knowledge transfer is not the key mechanism underlying entrepreneurial
entry. The basis of our argument is the assumption that knowledge is a key predictor of
entrepreneurial performance (Agarwal et al., 2004; Dencker, Gruber, & Shah, 2009;
Helfat & Lieberman; Klepper & Sleeper, 2005). We test our arguments by analyzing the
effect of different types of mobility on entrepreneurial entry and entrepreneurial performance
relative to post-entry earnings.
We posit two research questions. The first question regards pre-founding characteristics
of entrepreneurs: To what extent does prior industry and geographical mobility influence
individuals’ propensity to found a new firm? We argue that certain individual
characteristics in combination with the opportunity structure they confront may explain
entry into entrepreneurship. First, highly mobile individuals have a taste for variety or are
misfits in the labor market that position them in constant search for new jobs. As jobs in
established organizations tend to be standardized, individuals with a taste for variety are
likely to enter entrepreneurship in search for different job experiences. Second, interindustry
mobility provides opportunity for information arbitrage or results in low average
industry specific experience, reducing the opportunity cost of switching from employment
to entrepreneurship. Thus, a combination of selection and situational mechanisms
may drive mobility into entrepreneurship without necessarily being accompanied by
knowledge transfer.
To examine our claim that entry into entrepreneurship of highly mobile individuals
may be driven by selection and situational mechanisms, we test the relationship between
mobility and entrepreneurial performance. Knowledge-based theory of entrepreneurship
posits individual mobility as a focal mechanism through which knowledge created in
established organizations flows to newly founded firms (Agarwal et al., 2007; Franco &
Filson, 2006; Saxenian, 2006). If the flow of valuable knowledge is what drives entrepreneurship
through employee mobility, then we should also expect systematic difference in
the performance of entrepreneurs established by highly mobile individuals and those who
experienced less mobility across jobs and industries. Alternatively, individuals who enter
entrepreneurship simply because of their innate characteristics or the low opportunity
cost in leaving paid employment do not necessarily transfer valuable knowledge predictive
of entrepreneurial performance (Amit et al., 1995). These alternative explanations
suggest that while mobile individuals are more likely to enter entrepreneurship, they do
not necessarily exhibit better performance as entrepreneurs. Hence, our second question
is: How does prior job mobility of firm founders affect entrepreneurial performance?
To investigate these questions, we use matched employee–employer data that follows
individuals and firms in Sweden over 11 years. This enables us to pinpoint pre-founding
effects on individuals arriving into entrepreneurship (e.g., spawning firm characteristics)
and their career histories, allowing us to construct detailed measures of individual
2. Situational mechanisms pertain to the structure of opportunities an individual faces while selecting from
a set of choices (Hedstr€om & Swedberg, 1998). As such, we do not assume any changes in the individual,
for example, through learning, as a result of mobility.
360 ENTREPRENEURSHIP THEORY and PRACTICE
mobility (A˚stebro & Thompson, 2011). As we are interested in evaluating the scope of
knowledge-based theories we zoom in on a subsample of industries where knowledge is
of vital importance: knowledge-intensive industries.3 These industries serve as an ideal
research setting to test our argument regarding knowledge flows, entrepreneurial entry,
and performance. First, since knowledge is central to innovation and imitation in products
and services as well as a key determinant of entrepreneurship (Delmar & Wennberg,
2010), knowledge-intensive industry is a suitable setting in which to test our arguments.
Second, prior studies on knowledge-based theories of entrepreneurship mostly relied on
data from knowledge-intensive industries such as laser (Klepper & Sleeper, 2005) or
disk-drive industries (Agarwal et al., 2004) etc., whereby these industries serve as a natural
setting for comparison. Third, focusing on this setting minimizes the unobserved heterogeneity
issues surrounding industry differences in studying the rates of
entrepreneurship (Eckhardt & Shane, 2011).4
Our work seeks to bring clarity into the mechanisms that tie mobility to entrepreneurship.
While knowledge-based theories of entrepreneurship posit knowledge transfer as a
key mechanism that drives entrepreneurship (Acs, Braunerhjelm, Audretsch, & Carlsson,
2009; Agarwal et al., 2007), we show that it may not necessarily always be true. We found
that pre-founding job mobility affects the prevalence of entrepreneurial entry but not the
earnings of entrepreneurs. Our results on differential effects of mobility on entrepreneurial
entry and earnings provide a note of caution to researchers: Resources and knowledge
may not be automatically inferred to flow from mobility without considering simple counterfactuals
such as individuals’ innate taste for variety (A˚stebro & Thompson, 2011) or
the opportunity structure they confront (Sørensen & Sharkey, 2014). Without attending to
simple explanations related to selection or situational effects, models that attribute knowledge
transfer as a mechanism that drive mobility into entrepreneurship will risk overestimation.
As a corollary, our research also speaks to the expanding research on “firm
spawning” (e.g., Klepper & Sleeper, 2005; Sørensen & Phillips, 2011), by showing that
the “small firm spawning effect” documented in prior studies exercises differential effects
on entrepreneurial entry and earnings in knowledge-intensive industries.
Theory and Hypotheses
We offer five hypotheses to probe different mechanisms related to the effect of labor
market mobility of individuals on entrepreneurial entry and performance. We structure
the hypotheses development in a way that highlights the limitations of inferring knowledge
flows from the effect of mobility on entrepreneurship. We do this by contrasting
knowledge-based explanations with alternative explanations based on selection and situational
mechanisms, such as taste for variety, labor market misfits, and opportunity cost.
Mobility and Entry into Entrepreneurship
Why are individuals with a high degree of industry mobility more likely to found a
firm than individuals with less mobility in their career history? Typical arguments refer to
knowledge accumulation and spillover effects as mechanisms by which highly mobile
3. Please see the Methods section and the Appendix for definitions and descriptions of these industries.
4. While the data allow us to investigate instances where prospective entrepreneurs venture into other industries
(Agarwal et al., 2004), our results are not exhaustive to entrepreneurship in all potential industries.
March, 2016 361
individuals enter entrepreneurship (e.g., Agarwal et al., 2004). Yet, part of the reason can
be attributed to the possibility that certain individual characteristics such as taste for
variety are correlated with their propensity for mobility. Ghiselli (1974, p. 81) suggests
the idea of the hobo effect as “...the periodic itch to move from a job in one place to some
other job in some other place.”5 If individuals with a high degree of mobility are primarily
driven by a taste for variety, they may have a psychological bias against staying in regular
jobs for long periods of time (A˚stebro & Thompson, 2011). As a result of constant search
for varied experience, they also exhibit a certain degree of misfit in the labor market.
Such underlying individual characteristics could affect individuals’ choice for entrepreneurial
entry, working independently from a change in their knowledge pool. Hence,
selection on unobserved traits may explain both mobility and entry into entrepreneurship
without any treatment effect of knowledge transfer.
Mobility and entry into entrepreneurship may also be codetermined by situational
mechanisms such as individual opportunity structure. First, individuals who transfer
between industries are more likely to identify niche opportunities in the market as well as
to engage in arbitrage of information, which is asymmetrically distributed in society
(Shane, 2000). Aldrich and Kenworthy (1999, p. 20) argue that “... innovative start-ups
are often the result of creative experimentation with new ideas by outsiders to an industry.
Experience guides the choice of a domain for exploration, but indifference to industry
routines and norms gives an outsider the freedom to break free of the cognitive constraints
on incumbents.” As mobility helps individuals to identify opportunities, we can expect a
positive effect of mobility on entrepreneurial entry without actually having any changes
in their knowledge pool.
Second, highly mobile individuals may lower opportunity cost in entering entrepreneurship
(Amit et al., 1995). While moving from one industry to another, individuals
forego industry-specific skills that would otherwise contribute to their human capital. The
loss of human capital associated with switching industries reduces their seniority for
which they may incur a wage penalty in the labor market. As a result, highly mobile individuals
have less incentive to search for jobs in established firms, resulting in increased
mobility into entrepreneurship.
Third, situational mechanisms (e.g., opportunity for information arbitrage, opportunity
cost) may interact with selection mechanisms (e.g., taste for variety, labor market
misfits) in driving entry into entrepreneurship without any knowledge diffusion. For
instance, opportunity for arbitrage is more likely to be utilized by individuals with a taste
for variety. They are less likely to be satisfied with accumulating skills and competences
related to a particular industry. Continuous search for different experiences would predispose
them to experiment with entrepreneurship because the set of available jobs in an
industry may not remain attractive long enough for such individuals. The innate taste for
varied experience will likely make them misfits in any labor market that operates with
institutionally defined roles and knowledge structures (e.g., professional boundaries),
making them more susceptible to respond to opportunity structures. Taken together we
posit:
5. Ghiselli (1974, p. 81) continues: “This urge to move seems not to result from organized or logical
thought, but rather would appear more akin to raw, surging, internal impulses, perhaps not unlike those that
cause birds to migrate. Floaters readily provide socially acceptable explanations for their peripatetic activity,
but under careful examination these explanations turn out to be little more than rationalizations. The simple
fact is that after being in one place for a matter of months, or perhaps a year or so, depending on the strength
and periodicity of his itch, the individual is impelled to pack up and move to another place and another job.”
362 ENTREPRENEURSHIP THEORY and PRACTICE
Hypothesis 1: The number of different industries within which an individual has
held a job is positively related to an individual’s likelihood of entrepreneurial
entry.
To further investigate the effect of mobility on entrepreneurial entry, we hypothesize
another dimension of mobility—across geographical regions. Geographical mobility as a
mechanism for knowledge spillovers in entrepreneurship is primarily studied from the
network or internationalization perspectives: for example, how entrepreneurs starting
businesses in their local environment can draw on local resources and relationships
(Stuart & Sørenson, 2003) or how immigrants and expatriates who move from, or back to,
their home country set up businesses in new regions (Saxenian, 2006). As such, we know
less about more general types of geographic mobility—that of individuals within national
boundaries—as a mechanism regulating the propensity for entrepreneurial entry. Out-ofarea
moves entail disruption to social resources and roles, altering the locations of life
and work. Such moves are often provoked by dissatisfaction with current work or life
states, or by the lure of more attractive work or life situations elsewhere. This typically
means that while some individuals move to start businesses in remote regions where
industry conditions are fertile, most entrepreneurs choose to start businesses in the region
in which their social network is embedded (Cooper, 1985).
The social network perspective—frequent in studies of transnational entrepreneurship
as well as on entrepreneurial opportunity discovery—suggests two types of mechanisms
whereby geographical mobility may trigger entrepreneurship: First, moving out of
their original location exposes individuals to new types of opportunities, which they may
be able to recognize and develop. The movement from far away may enhance the possibility
for bringing new skills and ideas into a context that may value such “foreign” and
different products or services (Davidsson, 2004).
Second, a diverse geographical background with time spent in a number of locations
provides opportunities for building an expanded social network in terms of “weak ties” to
a variety of potential customers, suppliers, and other stakeholder. Newcomers in a region
that have a significant range of weak ties are ideally positioned for information arbitrage
and thereby the discovery of entrepreneurial opportunities (Kim & Aldrich, 2007). A high
degree of geographical mobility may thus enhance individuals’ network resources as well
as their cognitive base for identifying opportunities (Saxenian, 2006), leading us to
suggest:
Hypothesis 2a: The number of moves between different geographic areas is positively
related to an individual’s likelihood of entrepreneurial entry.
Hypothesis 2b: The distance of geographic moves is positively related to an
individual’s likelihood of entrepreneurial entry.
Mobility and Entrepreneurial Performance
If mobility between different industries or locations leads to accumulation of knowledge,
mobility should not only influence the likelihood of entrepreneurial entry but could
also influence entrepreneurial performance. Given the counterfactual possibility that reasons
for mobility are driven by situational or selection mechanisms, examining the effects
March, 2016 363
of mobility on entrepreneurial performance provides an interesting litmus test of
knowledge-based arguments. If mobility is strongly associated with some other background
variable that does not affect entrepreneurial success, we would expect mobility to
exhibit a positive effect on entrepreneurial entry but exhibit no effect—or even an adverse
effect—on entrepreneurial earning.
If movement across different industries and locations is prompted by an individual’s
taste for variety, this may, however, have detrimental effects on entrepreneurial capabilities
from too frequent job shifts (Elfenbein, Hamilton, & Zenger, 2010). In fact, this may
be detrimental for human capital accumulation, since a fair length of tenure is required to
learn any job and thus accumulate knowledge. Better performance in entrepreneurship
usually requires in-depth knowledge regarding the needs of the market and meeting those
demands with sufficient quality in technical or product-related competences, skills that
are often acquired through education or accumulated industry experience (Dahl &
Reichstein, 2007). For instance, in a study of the invention of three-dimensional (3D)
printing, Shane (2000) argues that exploitation of entrepreneurial opportunity in 3D printing
is influenced by the extent to which entrepreneurs acquired prior knowledge related to
the technology and their familiarity with market needs and potential customers.
Yet another alternative explanation is that mobile individuals may have difficulties in
finding a good “match” in either paid employment or entrepreneurship (A˚stebro et al.,
2011). They simply may have abilities that are ill-suited for employment or have less ability
to work well with others. Both alternative explanations (i.e., taste for variety and labor
market misfits) suggest that if knowledge is not driving the mobility–entrepreneurship
relationship, we would observe mobile individuals to be more likely to enter entrepreneurship
but not exhibit higher performance:
Hypothesis 3a: The number of different industries within which an entrepreneur
has held a job has a negative or zero association with entrepreneurial performance.
Hypothesis 3b: The number of moves between different geographic areas made by
an entrepreneur has a negative or zero association with entrepreneurial performance.
Data and Methods
Responding to pleas in entrepreneurship research to use longitudinal data that does
not depend on success or recall bias we employ an extract from a set of two matched longitudinal
data sources on the Swedish labor market to test our hypotheses. Our data come
from governmental registers maintained for research purposes by Statistics Sweden. The
first source is LISA—a database on all legal residents of Sweden over the age of 16 from
1989 onwards. It contains demographic and financial information from a number of sources,
including individual tax statements, financial records, birthplace registries, and
school records. The second source is RAMS—an annual mandatory survey of the total
population of all legal residents matched to firms having at least one employee or earning
a profit. This source offers information on employment, industrial structures, and also
tracks flows in the labor market.
The sample frame of our database is the complete set of individuals and firms in
knowledge-intensive industries, according to Eurostat and OECD classifications of such
sectors (G€otzfried, 2004). In total, twenty-two 5-digit industry codes equivalent to the
364 ENTREPRENEURSHIP THEORY and PRACTICE
U.S. Standard Industrial Classification (SIC) system are included in the sample, comprising
33% of the Swedish economy but over 40% of GDP (see the Appendix). While only
individuals active in these industries at any time from 1989 to 2002 are included, we have
complete labor market history on these individuals regardless of their prior employment
industry, and we continue to observe individuals that move to other industries. Hence,
while our inferences with regards to entrepreneurial entry and earnings are limited to ventures
in knowledge-intensive industries, the measures of mobility are unbiased. These
matched databases enable us to observe transitions in labor status and the characteristics
of firms in which individuals have been employed and/or founded. Important for our purposes,
the data permit us to observe the career trajectory of individuals transcending from
employment into entrepreneurship.6
The construction of the sample for analysis was based on a set of restrictions from the
LISA data set to eliminate sources of heterogeneity. First, to avoid potential biases from
situations where gender may impact family choices on entrepreneurship entry and performance,
we included only men in our sample. Second, we exclude those with any experience
in entrepreneurship between 1989 and 1994, since the dynamics of serial
entrepreneurship is known to be distinct from that of novice entrepreneurs (Ucbasaran,
Westhead, & Wright, 2009), which could bias our estimates of mobility. Third, to develop
measures of pre-entry labor market experience we eliminate observations prior to 1994
(Sørensen & Phillips, 2011). Finally, to avoid problems of retirement and full-time education,
we eliminate all individuals who exit the labor force any time between 1995 and
2001, and further eliminate all below age 23 or above age 65 in any of the years 1994 to
2001.
In total, 230,362 individuals fulfill our sample criteria. We follow these from 1994
until 2002 and include their labor market history between 1989 and 1994 to construct predictor
and control variables. The long time period of the data addresses the challenge in
studying career histories and mobility over time (Carroll & Mosakowski, 1987).
Dependent Variables
We model two key outcomes in entrepreneurship research: entrepreneurial entry in
the form of creating a new legal organization (Shane, 2009) and entrepreneurial performance
measured as firm founder’s income from entrepreneurship.
Following earlier research, the first dependent variable, entry, is a dichotomous indicator
of whether an individual in our population was active in entrepreneurship during
year t, but not in the preceding year (Sørensen & Phillips, 2011). We define an entrepreneur
as an individual who reports residual income from a company in which she works
full time. To exclude the prevalence of part-time or “hybrid” entrepreneurs (Folta,
Delmar, & Wennberg, 2010) we include only firms where the individual also holds the
majority ownership stake.
The second dependent variable is Entrepreneurial performance. We measure performance
using entrepreneurial earnings. Prior studies have exhibited difficulties in measuring
firm performance since performance in entrepreneurship is difficult to collect for
discontinued ventures. Performance has often been ignored or measured indirectly as
“financial leverage” (Bates, 1990) or “money taken out of the business” (Gimeno, Folta,
Cooper, & Woo, 1997). These are imperfect measures since entrepreneurs often choose to
6. A somewhat similar approach is taken by papers investigating related issues using generalizable longitudinal
data on the U.S. (Ozcan & Reichstein, 2009) or Danish labor markets (Nanda & Sørensen, 2010). €
March, 2016 365
forego current benefits in preference of reinvesting the money. We overcome this problem
by using the multi-level nature of the data, where firm-level performance variables
can be constructed according to Hamilton’s (2000) definition of entrepreneurial earnings
as (revenues – expenses 5 money taken out 1 retained earnings), where performance is
measured as the sum of money taken out 1 retained earnings. The measure was highly
skewed and we therefore transformed it into log form.
Job Switch. To account for the possibility that mobility may be related to any type of
job switch (including entrepreneurial entry) we also model Job switch, defined as leaving
paid employment and moving to another employer between year t and the preceding year.
This provides a counterfactual to the test of our mobility variables for entrepreneurial
entry.
Independent Variables
Industry Mobility. Movement across industrial sectors exposes the individual to a wider
range of tasks and organizational models, thus extending the individual’s experience in
terms of scope. Using LISA, we count the number of job moves between different industry
domains (i.e., across 5-digit SIC codes) an individual has experienced during the past
5 years.7
LISA includes information on individuals’ county of residence and how often
addresses are updated. We used this information to create two variables. Geographical
mobility is an ordinal-scaled variable counting the number of times an individual has
moved between municipalities during the past 5 years.8 Geographical mobility distance
measures the accumulated distance that an individual has moved during the past 5 years,
using a variant of the geographic dispersion index developed by Sørenson and Audia
(2000). A higher value of this measure implies high geographical mobility; lower values
imply low mobility.
Control Variables
Our argument that mobility matters for entrepreneurship hinges on the proposition
that it is not exclusively either: (1) the traditional time-invariant variables like gender or
age, or (2) human and social capital parameters such as education levels, experience of
previous entrepreneurial endeavors, having entrepreneurial parents, or exposure to particular
peers or even (3) the life-course variables of marriage and children—that best predict
the potential for entrepreneurship. To address concerns that our results are caused by
7. We included a squared term to investigate the potential nonlinear effects of industry mobility to further
probe the “taste for variety” argument. This variable was not statistically significant but highly collinear
with the industry mobility variable. We therefore report models without the nonlinear effect of industry
mobility as not to overfit our model and induce risks of multicolinearity. As a robustness test, we categorized
industry mobility only when shifts occurred in two-digit SIC codes. This decreases the variable’s mean
value from 1.36 to 1.03 and the maximum value from 11 to 7 but in our Cox models it was still a positively
associated with entrepreneurial entry (p > .001) but not significantly related to performance (p < .10).
8. Using industry and geographic mobility as independent variables has an additional advantage in disentangling
entrepreneurship entry from being a random outcome of mobility in general. While industry switches
will alter the opportunity cost of entrepreneurship entry because of the loss of industry-specific human capital,
geographic mobility is weakly correlated with job mobility in our setting (Korpi, Clark, & Malmberg, 2011).
366 ENTREPRENEURSHIP THEORY and PRACTICE
other effects we include a number of additional control variables, which have been shown
in previous studies to influence entrepreneurial entry and performance.
Employer size is found to influence the probability of entrepreneurial entry (Sørensen,
2007a). We measure size of prior employer organization with five dummy variables
where the first indicates employer size 1–25, the second indicates employer size 26–100,
the third employer size 101–1,000, the fourth employer size 1,001–5,000, and the fifth
(omitted as baseline in regression) for employer size 5,0001. Employer age is measured
with two dummy variables since our data is left censored at 1989. The first denotes
employer age 0–2 and the second denotes employer age 3–6. Work tenure is captured by a
variable measuring the number of years of employment since entering “time at risk”—
i.e., 1994. By nature of the sample set-up, this key variable is not truncated. Hence, the
variable takes the value between 1 and 5. We also include its squared effect: work tenure
squared, to control for potential nonlinear effects.9 We include a variable Years of education
to operationalize general human capital (Arum & M€uller, 2004; Br€uderl, Preisend€orfer,
& Ziegler, 1992). To remedy the risk of our mobility variable being confounded with
the accumulation of specific human capital, we created the variable Industry experience,
measured as number of years of experience in a focal industry (SIC-3 level equivalent)
during the 5 most recent years of an individuals’ career. The variable was thus truncated
above 5. Truncation of independent variables risks underestimating the effect of the variance
in the variable at the positive side of the distribution and increases the likelihood of
type-two errors. However, only 3% of the sample had 5 years or more industry experience,
indicating low risk of systematic bias.10
The choice to set up a new firm may be affected by parental heritage; that is, a child
following the career footsteps of his or her parents (Evans & Leighton, 1989). Parental
heritage is known to affect entrepreneurial entry through transfer of human capital, exposure
to role models, and inheritance of genetic disposition (Aldrich & Ruef, 2006;
Sørensen, 2007b). To control for this we include a dummy variable, Parents’ entrepreneurship,
measuring exposure to entrepreneurship during upbringing. This variable is
derived from Statistics Sweden’s cross-generation database that provides data on the
labor market activities of all Swedish residents living in the same household from 1960 to
2002.
Children: A time-variant ordinal variable measures number of children in the household.
Married: We used information on household composition from LISA to construct a
dummy variable that equals 1 if the individual is married or cohabiting. Ethnic background:
Immigrants are known to have higher probabilities of entering entrepreneurship
(Arum & M€uller, 2004). We measure immigration with a dummy that takes the value 0
for Nordic (i.e., Scandinavian) and 15 other. Wage as employee is measured in logarithmic
form to control for the opportunity costs of employment (Ozcan & Reichstein, 2009). €
Spouse wage is measured in logarithmic form to control for the fact that household
resources are often pooled in entrepreneurial firms (Folta et al., 2010). Household wealth
is a dummy variable taking the value “1” if a person has assets (including a house)
exceeding $110,000, which controls for the potential of liquidity affecting likelihood of
entry. All individuals living in Sweden receive a personal identification number based on
their date of birth. We used this information to calculate individual’s Age (in years).
9. The squared term of work tenure is highly collinear with the constitutive term. In unreported models—
available on request—we therefore included only the constitutive term, with results unchanged.
10. As robustness checks we fitted unreported models including a dummy variable for individuals with 61
years of experience. This slightly decreased effects sizes but significance levels were still well below 5%.
March, 2016 367
Our theory posits an alternative to knowledge spillover argument for departures from
wage work to entrepreneurship, yet we cannot distinguish between departures driven by a
“taste for variety” and low opportunity costs in terms of low/stagnating wages. To proxy for
the potential of necessity-driven departures from employment to entrepreneurship by
employees, we include the control variable Wage trend, coded as an accumulated 3-year
change in wage. For most individuals, this variable takes a positive value but for some it
takes a negative value, indicating falling wages. This variable approximates for the opportunity
costs of leaving paid employment for entrepreneurship. Inverse Mills Ratio; In our
data, 2,549 individuals engage in entrepreneurship at any time during the period 1995 to
2002. Our analysis of entrepreneurial performance is based solely on these individuals, who
constitute a subsample of the overall group of potential entrepreneurs. It is possible that this
subsample exhibits some specific characteristics that simultaneously affect their likelihood
of engaging in entrepreneurship and their performance as entrepreneurs, raising a risk that
coefficients that show a statistically significant relationship with subsequent performance
will be biased. We approached that aspect of the subsample by using a two-step Heckman
process where the estimates of entrepreneurial entry in Table 3 are used as a self-selection
control for entry when estimating the performance equation. We used Lee’s (1983) generalization
of the Heckman selection model to create a selection-correction variable (Inverse
Mills Ratio) from the predicted estimates of entrepreneurial entry in Table 3. As instrument
for the selection equation, we excluded the variables: Wage trend, Wage, and Spouse wage,
all of which were significantly related to likelihood of entry but only very weakly related to
entrepreneurial performance. Introducing the Inverse Mills Ratio in Table 4 lowers the risk
of observing spurious results based on sample selection bias. All predictor and control variables
were lagged 1 year to avoid simultaneity bias.
Estimation Strategy
Our first set of analyses uses competing hazard rate analysis to estimate models of
entrepreneurial entry, compared to taking a new job (Br€uderl et al., 1992). The flexible Cox
specification is used, which does not require specific assumptions in regards to the distribution
of the hazard rate (Lancaster, 1979).11 Cox regression models time (t) as a function of
an underlying hazard h and a set of exponentiated beta coefficients (bij) and covariates (x)
for individual at year. The baseline hazard h corresponds to the case where all covariates
(x) equals 0, and is shifted up or down proportionally with changes in the covariates:
h tð Þ5h0ð Þt exp bijxij   (1)
The standard Cox model does not account for the potential of unobservables in the data.
The presence of unobserved heterogeneity is conceivable in entrepreneurship due to the
unobservable nature of entrepreneurial skills, which may be uncorrelated with observable
skills (Parker, 2005; Shane, 2000).12 The problem of unobserved heterogeneity—called
11. We plotted the data to ensure the proportional-hazards assumptions were met. We also plotted baseline
Kaplan–Meier survival curves for all hypothesized variables and compared these with the curves predicted
by the Cox models to ensure that predicted effects did not deviate abnormally from observed values.
12. Our sample is designed to restrict heterogeneity in, e.g., prior entrepreneurial experiences and gender to
test the theoretical variables of mobility and prior employment. We still cannot rule out the presence of
unobserved heterogeneity. A common remedy to the problem of unobserved heterogeneity is to estimate
models with individual fixed effects. However, by the nature of our design individuals may enter entrepreneurship
only once and as a consequence, fixed effects cannot be utilized (Sørensen & Phillips, 2011).
368 ENTREPRENEURSHIP THEORY and PRACTICE
“frailty” in hazard analysis—tends to stem from incomplete model specification
where models with no frailty are at risk overestimating negative time dependence in
the hazard rate, as well as underestimating the proportionate effect on the hazard
from a change in the predictors (Lancaster, 1979). Hazard models can be extended to
account for heterogeneity by incorporating frailty as a latent random effect that enters
multiplicatively on the hazard function. We try to accommodate the existence of
unobserved heterogeneity (frailty) by estimating a version of the Cox model which
incorporates a gamma distributed random effects term with mean zero to summarize
unobserved frailty connected to each individual spell. The random effects are speci-
fied to arise randomly but stratified according to some observable group characteristics
that may be correlated with unobservable characteristics. Entrepreneurial ability
is generally believed to be correlated with education and other human capital variables.
We, therefore, stratified individuals according to years of education (17 groups)
since educational level is known to be correlated with both IQ and unobserved ability
(Van Praag & Cramer, 2001).13 In the model estimated, the random effects term (v)
describes unexplained heterogeneity as the influence of unobserved risk factors in the
model as:
hijð Þt 5h0ð Þt exp bijxij1vi
  (2)
In the regression outputs, all coefficients are displayed as hazard rates (HR), which
ease interpretation of these as marginal effects. These we attend to for each hypothesis. A
coefficient of 1.01 indicates that a one-unit increase in covariate x increases the likelihood
of the outcome variable (entry or exit) with 1%, while .99 indicate that a one-unit increase
in covariate x decreases the likelihood of the outcome variable with 1%.
In the second set of analyses, we apply linear regression models to examine the
effects of theoretically motivated mobility variables for entrepreneurial performance.
These models are estimated using two different specifications, pooled OLS, and panel
data models estimated with random effects. Both models were estimated with standard
errors clustered around individuals to accommodate for autocorrelation.