Where are the women entrepreneurs? Business ownership growth by gender across the American urban landscape.
Conroy, Tessa ; Weiler, Stephan
This study identifies the determinants of growth for male and
female business ownership in a subset of U.S. counties. The results
indicate that there are important characteristic and behavioral
differences between the male and female populations in each county that
affect regional changes in business ownership for each gender. In
particular, the education level of males and females as well as the
local family structure impact the propensity for firms owned by each
gender differently. A Blinder-Oaxaca type decomposition, a novel
approach in the context of regional outcomes, demonstrates that although
the effect of characteristic differences is larger, the behavioral
differences are key to narrowing the gender disparity in business
ownership. (JEL L26, R2, R3)
I. INTRODUCTION
Small business is of growing importance in economic development
strategies implemented by both the public and private sectors.
Subsidies, tax breaks, and other incentives have been used at all levels
of government in the United States with the goal of increasing the
number of proprietors in the workforce. Yet, entrepreneurial activity in
the United States varies dramatically across space. Regional studies of
entrepreneurship show that such spatial variation is not random, but
seems systematically related to specific factors associated with
particular locations. Several studies, going back as far as Bartik
(1985) and more recently by Goetz and Rupasingha (2009), try to identify
the location-specific characteristics that explain the spatial variation
in entrepreneurship. Although these studies link entrepreneurial
activity to several important regional characteristics, none consider
how the impact of these characteristics might vary by gender.
Existing studies establish several key relationships, but do so by
pooling the entrepreneurial activity of men and women together, using
firm births or the self-employment rate, for example. Yet, men and women
do warrant separate study. Knowledge, resources, and constraints are
distributed differently across members of society, and certainly across
gender. Women are systematically different from men in their skills,
social responsibilities, and opportunities. They can also be expected to
assess the local market, value regional characteristics, and respond to
their communities differently from men. This of course has implications
for the spatial distribution of entrepreneurial activity for each
gender.
In U.S. counties, the number of firms relative to the labor force
is 19% on average but ranges from higher than 50% in some counties to
less than 5% in others. However, measures of entrepreneurship that
aggregate across gender conceal significant differences between men and
women. In some rural southern counties the number of female-owned firms
relative to the female labor force is as low as 2%, but is greater than
25% in several Colorado and Massachusetts counties, and for men, it
ranges from 9% to over 70%. (1) Clearly, the propensity for female-owned
firms is higher in some counties than in others and likewise for men,
but on average the propensity for female-owned firms is much lower than
that for males.
In this study, we consider two possible sources of the gender
disparity in business ownership. First, the male and female populations
may be characteristically different at the mean, in education attainment
for example. Second, male- and female-owned firm formation may result
from gender-specific local behavioral patterns, indicated by different
coefficients in gendered empirical models. This is the first known study
using U.S. data to examine both the characteristic and behavioral
differences in relation to the gender disparity in firm ownership across
regional outcomes. This article applies a Blinder-Oaxaca type
decomposition to establish whether the characteristic or behavioral
differences are key to closing the gender gap.
First, we identify characteristic differences between the male and
female populations across counties by evaluating mean differences. Then
the empirical analysis identifies behavioral differences between the
male and female populations as they relate to their respective
entrepreneurial outcomes. We include a number of explanatory variables
to control for both place- and people-based characteristics. Place-based
characteristics include features of the location such as the industrial
composition and level of natural amenities. People-based
characteristics, or demographics, consist of countywide average values,
so that each variable proxies for the average characteristics of the
pool of workers/potential business owners.
We focus on explanatory variables related to education and family
to better understand gender differences in firm ownership. For example
we ask, what is the expectation for growth in female business ownership
conditional on the education attainment of the pool of potential female
business owners? One way to measure education attainment is with the
share of the female population at each major degree level and then
include the shares as explanatory variables in an empirical model. We
follow this strategy for each gender and draw comparisons. As is typical
when studying labor market outcomes for various groups, we use a
decomposition to determine the relative importance of characteristic and
behavioral differences between men and women in explaining the gender
gap in business ownership.
The results indicate that there are characteristic and behavioral
differences between the county-level male and female populations that
drive the gender disparity in business ownership. We focus on these
gender differences with regard to education attainment and the local
family structure. Growth in the propensity for both male- and
female-owned firms is higher in counties with a large share of males and
females with bachelor's degrees, respectively, but the effect is
much stronger for men. The propensity for female-owned firms lags in
regions with a large share of the least and most highly educated women.
Although the share of married adults has little to no effect, the number
of children per adult is negatively associated with the growth in the
propensity for firms owned by either gender, but again the effect is
much larger for men.
The concluding decomposition shows that although the effect of
having different mean characteristics is larger in absolute value, the
behavioral differences are key to alleviating the gender disparity in
business ownership, if that is indeed the goal of local policymakers.
Many prior regional entrepreneurship studies have implicitly focused on
males, potentially resulting in genderblind policy implications. The
decomposition makes clear that policies aimed at enhancing
entrepreneurship affect men and women differently. Understanding the
gender differences is an important aspect of informed policy aimed at
equitably enhancing entrepreneurship.
II. MEASURING ENTREPRENEURSHIP
In this study, we measure entrepreneurship with the propensity for
female (male)-owned firms calculated as the number of female (male)owned
firms relative to the female (male) labor force. Gendered data on
business ownership by county comes from the 2002 and 2007 Survey of
Business Owners (SBO) administered by the U.S. Census Bureau. Firms are
classified as female-owned if women hold 51 % or more of the stock or
equity in the company. Firms are male-owned if women hold less than 51%
of the stock or equity in the company.
The term entrepreneur has been used broadly, and consequently, has
taken on a variety of meanings depending on the context. Sometimes
"entrepreneur" refers simply to someone who is self-employed.
At other times it implies specific functions such as risk-bearer or
innovator. Unquestionably, entrepreneurs take on several varying roles,
blurring the definition. Still, describing someone as an entrepreneur
does identify that person as having unique qualities apart from others
in the business sector. The term at least implies the most fundamental
role of business owner or manager and the right to extract excess
revenue above costs.
The multifaceted nature of entrepreneurship makes it difficult to
measure. Precisely because of these difficulties, Low (2009) argues that
economics is beginning to focus on a functional definition of
entrepreneurship. The emphasis is now on, "what entrepreneurs do
rather than who they are" (Low 2009, 5). The three main functions
she identifies are (1) ownership or operation of a firm, (2) risk- and
uncertainty-bearing, and (3) innovation or the reallocation of resources
(Low 2009). Hence, key aspects of entrepreneurship are difficult to
quantify, count, and measure. Data that entirely satisfy all of the
common concepts of an entrepreneur are nonexistent. Where ideal data are
nonexistent, research on entrepreneurship has had to resort to what is
available. Although the number of firms relative to the labor force does
not entirely capture the essence of entrepreneurship, it is available
and gender disaggregated.
III. REGIONAL ENTREPRENEURSHIP AND GENDER
A. Regional Explanations of Entrepreneurship
The spatial variation in entrepreneurial activity has been linked
to the characteristics of regions and the communities that reside in
them. Previous studies show that the industrial composition, policy
environment, and labor market all influence local levels of
entrepreneurship. In addition, the demographic profile of some regions
is more conducive to new business. That is, some locations have a
population of people that is more entrepreneurial than others. Counties
with more attractive natural amenities also tend to lure new businesses
and host more entrepreneurs (Florida 2002).
To the extent that the local labor market determines the relative
returns to self-employment, it has an important role in establishing the
incentives for entrepreneurship. Goetz and Rupasingha (2009) find that
proprietor earnings have a positive and significant impact on the
growing density of proprietors, whereas wage-and-salary income has a
negative impact, suggesting that individuals do in fact make their
employment choice according to relative returns. Similarly, Low and
Weiler (2012) found that in regions where wageand-salary employment is
more volatile or riskier, the self-employment rate is higher.
Reasonably, the level of local joblessness also factors into the
wage-and-salary option, although the unemployment rate can have a
spurious relationship to measures of entrepreneurship (Storey 1991).
The local industrial composition explains much of the variation in
entrepreneurship (Glaeser 2009). Some industries are more conducive to
entrepreneurship, and some locations are more conducive to certain
industries. The regional industrial mix will influence the opportunities
that potential entrepreneurs are likely to see and exploit in a
particular place. For example, Goetz and Rupasingha (2009) find that
entrepreneurial activity is higher in counties with industry
concentrations in construction and services. Conversely, mining and
utilities do not support high levels self-employment (Glaeser 2009). New
firms in particular industries also require certain inputs. Glaeser
(2009) finds that concentrations of industry suppliers have a strong
positive effect on self-employment rates.
Human capital has long been considered an important driver of
entrepreneurship both at the individual and regional levels. Higher
self-employment rates are generally found in locations with older and
more educated populations, yet various measures of human capital and
entrepreneurship yield mixed results (Acs and Armington 2004; Glaeser
2009; Goetz and Freshwater 2001; Goetz and Rupasingha 2009; Lee,
Florida, and Acs 2004; Low, Henderson, and Weiler 2005). For example,
Low, Henderson, and Weiler (2005) find that entrepreneurial depth, the
value added by business owners, is higher in counties with higher
college education attainment, but entrepreneurial breadth, the size, and
quantity of small businesses, is unaffected.
A recent study in the microeconomic literature found that the
relationship between education and entrepreneurship is curvilinear.
Entrepreneurship is most strongly associated with education attainment
at the bachelor's degree level, whereas both low and very high
levels of formal education have a relatively weak or negative
relationship to self-employment perhaps indicative of the evolving
opportunity cost of self-employment across levels of education
attainment (Kim, Aldrich, and Keister 2006). At relatively low levels of
education, even low-wage employment could be more lucrative than the
income potential of self-employment. At high levels of education, the
return in the wage-and-salary labor market may well exceed the return in
self-employment. Consequently, the propensity for entrepreneurship may
be highest among those with mid-level education attainment. It is quite
possible that a similar pattern exists at the regional level. A large
population of men and women with a bachelor's degree may be most
conducive to local entrepreneurial activity.
A few regional studies include both the shares of high school
graduates and college graduates as determinants of entrepreneurial
activity. Acs and Armington (2006) found that the share of college
graduates and, unexpectedly, the share of high school dropouts are both
positively associated with higher entrepreneurial activity. They explain
the unexpected effect of high school dropouts in terms of labor supply:
entrepreneurs may benefit from abundantly available low-skill labor. In
some cases, after controlling for age as a proxy for work experience,
which is positive and statistically significant, the share of college
graduates has no effect (Bartik 1989; Goetz and Rupasingha 2009). The
mixed results across regional studies suggest that the relationship of
local human capital accumulation to entrepreneurial activity is still
somewhat unclear.
B. Gender Considerations: Human Capital and Family Structure
Regional factors such as human capital, labor market conditions,
industrial composition, and natural amenities seem to drive
entrepreneurship, but none of these factors have been considered in
relation to gender. One recent study by Rosenthal and Strange (2012)
focuses on women entrepreneurs and the importance of knowledge
spillovers and agglomeration in their business location decision. They
develop an analytical model where females are less networked than their
male counterparts, and as a consequence, have limited access to
knowledge spillovers. Empirically, the authors demonstrate that women
are in fact located further from agglomerated areas. Yet even this
recent study provides only a limited picture of female entrepreneurs in
a regional context.
Previous studies of entrepreneurship in the microeconomic
literature indicate that there are systematic differences between men
and women business owners and those differences likely have implications
for their respective local entrepreneurial activity. Women use processes
different from men to identify opportunities (DeTienne and Chandler
2007) and make systematically different workforce management choices
(Matsa and Miller 2014). Women entrepreneurs generally have less work
experience in business and management, but more formal education than
their male counterparts (Cowling and Taylor 2001). Compared to men,
women with less education are more aware of their knowledge
deficiencies, more likely to perceive certain obstacles, and ultimately,
less likely to become entrepreneurs (Huarng, Mas-Tur, and Yu 2012;
Kourilsky and Walstad 1998). Additionally, because of their lack of
previous work experience, those women who do choose to start their
business often fail to remain self-employed (Rosti and Chelli 2005).
Human capital is not the only factor that enters into the
self-employment decision differently for men and women. Family and
children also influence men and women differently in their decision to
become self-employed (Boden 1996, 1999a; Georgellis and Wall 2005;
Hundley 2000). Women still have primary responsibility for family and
children, which means they have less flexibility in their daily lives
(OECD 2004). For women, self-employment may allow the flexibility to
stay at home and meet the demands of being a spouse and mother (Hundley
2000). Women with small children in the household are more likely to
enter self-employment (Boden 1996; Bruce 1999) yet, the presence of
young children had no significant impact for men (Boden 1999b). As
further evidence of the demands of child rearing, women, especially
women with young children, cite reasons related to family and schedule
flexibility as their primary motivation for becoming self-employed
(Boden 1999b). However, it seems that both men and women considering
selfemployment benefit from the support of a spouse (Boden 1999b;
Taniguchi 2002).
C. Regional Implications of Gender in Entrepreneurship
The gender differences in entrepreneurship have mostly been
determined in the microeconomic literature (i.e., Boden 1996; Hundley
2000; Georgellis and Wall 2005), which tends to consider
entrepreneurship in an occupational choice framework drawing on Evans
and Leighton (1995) and Blanchflower and Oswald (1998). In this
framework, each person rationally chooses between entrepreneurship and
wage-and-salary employment based on the utility maximizing principle.
Both pecuniary benefits, which are largely a function of human capital,
and nonpecuniary factors, such as family responsibilities enter into the
equation. In an efficient equilibrium allocation, only those
entrepreneurial opportunities that are most lucrative will motivate
departure from the competitive labor market. Hence, the extent of
entrepreneurship in a given region represents the share of individuals
with a self-employment opportunity that is better than their
wage-and-salary option.
The gender differences previously discussed suggest that human
capital and family structure, in particular, are considered in a
consistently different way between men and women, and as a result, men
and women make systematically different occupational decisions.
Extrapolating this model out to the regional level implies that the
local human capital profile and family composition would relate
differently to the local propensity for male- and female-owned firms.
While occupational choice theory focuses on individual characteristics,
in this study, as in Goetz and Rupasingha (2009), county characteristics
are used as proxies for the average characteristics of the population
pool from which entrepreneurs are drawn, and to reflect the local
environment in which they make their employment decisions. We use these
factors to explain the propensity for male-owned firms and the
propensity for female-owned firms, then draw gender comparisons.
IV. HYPOTHESES AND EMPIRICAL MODEL
A. Hypotheses
To develop the first hypothesis we consult recent studies of
entrepreneurship that focus on human capital. Kim, Aldrich, and Keister
(2006) suggest that entrepreneurial propensity is not a strictly
increasing function of human capital. Individuals with low and very high
levels of human capital are less likely to become entrepreneurs compared
to those with a college degree who are most likely to become
entrepreneurs. Men and women with only a high school diploma may lack
the skills and resources necessary to earn high returns as an
entrepreneur, and consequently, are more likely to choose
wage-and-salary employment. Doctorates may also coincide with a lower
propensity for entrepreneurship, but in contrast to those with high
school education, because the highly educated generally have lucrative
wage-and-salary options. Those with a college degree are likely well
suited for a relatively profitable entrepreneurial option compared to
wage-and-salary employment. In light of these findings, we hypothesize
the regional parallel: the propensity for female (male)-owned firms is
higher in regions with a large share of females (males) with
bachelor's degrees, and lower in regions with larger shares of less
and highly educated females (males). Hence, the relationship between
local education attainment and the propensity for firms owned by either
gender forms an inverted "U."
A secondary, but important consideration in a gendered study of
entrepreneurship, is the impact of a spouse and children. Family
structure and the demands of household production likely impact the
entrepreneurial propensity of men and women very differently. While it
may be the case that the demands of caring for children impede
entrepreneurship, previous literature suggests children have a positive
impact on the self-employment propensity for women as they seek a
flexible professional life that can accommodate their family life.
Children, however, have no impact on male self-employment. These gender
differences suggest that women still bear the primary responsibilities
of child rearing. It is reasonable to think that gender roles will have
cumulative implications reflected in the local propensity for male- and
female-owned firms. We hypothesize that the number of children per adult
woman will positively impact the propensity for female-owned firms and
the number of children per adult man will have no impact on the
propensity for male-owned firms. With regard to marriage, we hypothesize
that the propensity for firm ownership is higher where larger shares of
men and women are married.
B. Empirical Model
Regional studies of entrepreneurship that focus on gender are
sparse, hence there is little guidance for developing a gendered
empirical model. As highlighted above, there are gender-specific
considerations that motivate occupational choice. It seems that the
utility maximizing solutions are systematically different by gender,
primarily because men and women consider their human capital and family
situation differently. The regional drivers of male- and female-owned
firms are also likely to be systematically different making it
appropriate to use a gendered empirical model. The model developed here
is applied separately, yet in parallel, to the propensity for male- and
female-owned firms. The explanatory variables were selected based on the
regional characteristics previously shown to drive entrepreneurship
based on the work of Goetz and Rupasingha (2009), Glaeser (2009), Acs
and Armington (2006), and Acs and Armington (2004), as well as aggregate
measures of the educational and demographic characteristics shown
significant in the regional and micro-literature based on the work of
Kim, Aldrich, and Keister (2006), Boden (1999b), and Taniguchi (2002).
Considering each gender separately has the advantage of allowing us to
compare coefficients across models and gain insight into how education
and family structure factor differently into entrepreneurship for men
and women.
We test the hypotheses discussed previously using a model that
includes a vector of control variables with an additional vector of
human capital variables and measures of marriage and children. In large
part, we follow Goetz and Rupasingha (2009) in their choice of
explanatory variables, so that we can focus the analysis on the human
capital variables h, and measures of family structure f. The model can
be described generally as follows, where e is the propensity for either
male- or female-owned firms.
(1) [DELTA]e = [delta]h + yf + [zeta]r + e.
For example, for women [DELTA]e is equal to the change in the
propensity for female-owned firms, h contains variables measuring the
education of the female population, f measures marriage and children in
relation to the female population, and r is a set of regional control
variables that are the same for the male and female models.
Whereas studies focused on the individual's entrepreneurial
choice may use dummy variables for education attainment, here in a
regional context we consider a slightly different question and empirical
strategy. We ask, what is the expectation for growth in female (male)
business ownership conditional on the education attainment of the pool
of potential female (male) business owners? One way to measure education
attainment is with the share of the female (male) population at each
major degree level and then include the shares as explanatory variables
in the empirical model. In the U.S. Census, education is measured by the
highest level of education attainment for each person. We use the number
of people at a level of education attainment aggregated and normalized
into rates of education attainment for each county by gender. As in
other regional studies of entrepreneurship we consider multiple levels
of education. Acs and Armington (2004) and Goetz and Rupasingha (2009)
include the share of adults with a high school degree as well as the
share of college graduates. For this study, the four most advanced
levels of attainment are considered, namely, a high school diploma,
bachelor's degree, master's degree, and doctorate degree. To
avoid collinearity, only 4 of 16 possible measures of education
attainment are included. Levels below a high school diploma are excluded
as are professional degrees, such as a JD or MD.
The variables that measure family structure focus on marriage and
children separately. In the model of the propensity for female-owned
firms, for example, we include married women as a share of women over
the age of 15. We also include the number of children (age 17 or under)
per female over the age of 16. Similarly, we include married males as a
share of males over the age of 15 and the number of children per male in
the model of the propensity for male-owned firms.
The explanatory variables include a number of demographic and
regional characteristics widely used in regional models of
entrepreneurship. This group of variables includes controls for the
local labor market conditions, industry shares, and characteristics of
the local community. In addition to the measures of education
attainment, we also include the median age of the population of each
gender to capture the typical amount of work experience in the local
population. Labor market conditions are measured by the
employment-population ratio for each gender, with the expectation that
as employment increases the relative return to self-employment likely
decreases as does the incentive to own a firm.
Proprietor earnings per job and wage-and-salary earnings per job
are included to account for the relative incentives to each type of
employment, expecting that as proprietor earnings fall, firm ownership
will decrease and vice versa for wage-and-salary earnings. Wealth, as a
form of collateral, is important to potential entrepreneurs who may seek
loan financing. Owning a home and higher home value improve the
prospects of securing the loan financing for a new venture. We include
the share of owner-occupied homes and median value of homes. To control
for economic growth, we include the growth rate of income per capita
during the 5-year period preceding the business ownership measure.
Services, retail trade, and construction industries are included to
control for the local industrial mix and the growth patterns of
different sectors, measured as a share of total establishments (Malecki
1994). Population density controls for agglomeration and spillover
effects. Last, the natural amenities score is included with the
expectation that entrepreneurs who are more footloose will locate in
more scenic areas. Variable descriptions and sources can be found in the
Appendix.
V. DATA
Counties are becoming a common unit of analysis in studies of
entrepreneurship (Goetz and Rupasingha 2009; Rupasingha and Goetz 2011).
Arguably, metro areas are favorable for this analysis because they
capture cities which are an intuitive economic unit. Similarly,
commuting zones are a natural choice for regional analyses, as they link
metro areas to the labor supply from surrounding counties. Counties too
are a sensible unit of analysis for a regional study of
entrepreneurship. They are generally centered on a large city, often the
county seat, which anchors local labor and consumer markets. The county
seat typically hosts a number of local government agencies that attract
private businesses and residents. While commuting activity may blur
county boundaries, people generally prefer to live close to their
workplace and will choose to reside near the employment center in their
county. Counties thus have the advantage of being a smaller geographic
unit, within which there is reasonably cohesive economic activity.
A. Sources
County-level business ownership data come from the SBO, which is
administered by the U.S. Census Bureau every 5 years (specifically,
years ending in 2 and 7). The U.S. Census Bureau maintains a list of all
nonfarm firms with and without paid employees operating during the year
of the survey with receipts greater than $ 1,000 based on tax return
data. A sample of those firms is questioned on their employment,
payroll, and receipts. The resulting data are reviewed, edited and
tabulated, then made available to the public by geographic area. In our
analysis we use the publicly available estimates of the sum of employer
and nonemployer firms for U.S. counties. As of 2007, 79% of firms were
nonemployers, an increase from 76% in 2002.
Data from the SBO is withheld for many counties because the
estimates do not meet publication standards, by having a relative
standard error that is too high for example. Other data is withheld to
avoid disclosing data for individual companies. Given the criteria for
excluding an observation, there may be certain counties that are
systematically absent from the full sample. Rural and sparsely populated
counties where there are fewer businesses are more likely missing
because of a higher variance in the data and the risk of exposing
specific firms.
The regression analysis to follow analyzes changes in
entrepreneurship as a function of prior conditions. Results from the
U.S. Panel of Entrepreneurial Dynamics indicate that the median time for
a new firm birth from conception to the start of business is 19-24
months (Reynolds 2007, 55-56). In light of these findings, data from the
SBO is likely tied to factors from 1 to 2 years prior. With this
reasoning, we match the change in business ownership data between 2002
and 2007 with demographic data from the 2000 Decennial Census.
Calculating the dependent variable requires using county-level
labor force estimates for each gender. For the early time period,
gendered labor force estimates are available from the 2000 Decennial
Census. For the later period, labor force estimates are available from
the 2005 American Community Survey (ACS). The 1-and 3-year estimates are
limited by area size and consequently do not include all counties. Only
areas with a population greater than 65,000 people are estimated
annually.
Whereas the availability of labor force data is limited at the
county level, data describing industry, wages, and income are more
readily available. To be consistent with the demographic data, we use
measures of employment and the local industrial mix from 2000. The wage
and income data are from the Bureau of Economic Analysis Regional Data
Center. The industry shares are calculated as a share of total
establishments based on data available from the County Business
Patterns. Data on natural amenities come from the U.S. Department of
Agriculture (USDA)-Economic Research Service (ERS). The ERS calculates a
natural amenities score for each county based on topography and climate,
ranging from roughly -7 to 12, which is assumed constant over time.
The resulting dataset is a cross-section of 646 counties. The
counties included are those with a population greater than 65,000 in
2005. The counties are limited in this way to take advantage of the ACS
2005 annual estimates, which are produced only for a subset of counties
above the 65,000 population threshold. This restriction is the most
limiting factor in data availability, as it reduces the number of
possible counties included in the analysis to fewer than 700. Missing
values in the ACS and SBO require that additional counties be dropped,
further limiting the cross-section analyzed here.
[FIGURE 1 OMITTED]
Truncating the counties included in the analysis by population as
described above limits the analysis to a subset of primarily metro- and
micropolitan counties (Figure 1). The counties included are 85%
metropolitan counties with an urban core greater than 50,000,10%
micropolitan counties with an urban core between 10,000 and 50,000, and
5% of counties are nonmetro/microreflecting the possibility for a county
to fall above our population threshold but not be part of a core-based
metro area. The primarily metropolitan character of the remaining
counties may limit the generalizability of the results, as less
populated or rural areas may feature different entrepreneurial behavior
(Figueroa-Armijos and Johnson 2013).
B. Summary Statistics
The regional gender differences in the propensity for male- and
female-owned firms may be a function of characteristic differences
and/or behavioral differences. We hypothesize that the populations of
men and women are different in both ways: characteristically (a
difference in means [bar.x]s) and behaviorally (a difference in
coefficients [beta]s). The regression analysis to follow describes the
behavioral differences in detail. First, we examine the differences in
characteristics shown by the descriptive statistics.
The propensity for female-owned firms is much lower than that for
male-owned firms and the gap is persistent over time. On average the
propensity for female-owned firms was 9.6% in 2002, less than half of
the propensity for male-owned firms at 22.7%. The propensity for
female-owned firms increased in 2007 to 10.7%, but the gender gap
remained relatively constant as the propensity for male-owned firms also
increased to 24.6%. The male propensity is not only higher on average
but spread across a much wider range from approximately 12.1% to more
than 47.7%, whereas the propensity for female-owned firms is as low as
4.5% and near 25.1% at the highest.
Table 1 shows that education attainment at all levels differs
between genders. In 2000, 31.1% of women and 29.1% men held a high
school diploma as their highest degree. At all higher levels of
education attainment the share of women is smaller than that for men.
However, the gender differences are slim with close to 15% of the
population holding a bachelor's degree at the highest for both
genders and close to 6% holding a master's degree for both genders.
The largest gender difference is at the doctorate level; 1.4% of men
have a doctorate, nearly three times the share of women with a
doctorate.
A larger share of men than women were married in 2000 and there
were also more children per male than per female. The employment
population ratio was much higher for men at 67% than for women at 55%.
The median age of women is slightly higher for women, which is
consistent with the higher life expectancy for women.
Proprietor income per job was less than wage-and-salary income per
job, suggesting an incentive for traditional employment at the mean or
indicating an equilibrium given the likely non-pecuniary benefits of
self-employment, particularly for women (Lombard 2001). Income per
capita grew nearly 20% between 1997 and 2002. The share of service
establishments is the largest of those included, followed by retail
trade, and construction. Nearly 69% of homes were owner-occupied, and
the median housing value was just over $121,000.
The usual measures of population density may be heavily influenced
by the nonurban area of counties which, as pointed out by Bunten et al.
(2014), can be quite heterogeneous. Alternatively, we use tract-weighted
population density. For the year 2000, tract population density is
weighted with the tract population and summed by county.
C. Difference in Means
The difference in the mean propensity for male- and female-owned
firms is clear from Table 1. However, the differences between genders in
education attainment, the share of married adults, and children per
adult may seem quite small. Table 2 shows that these differences are
actually statistically significant. So, even if the behavioral
differences are small or nonexistent, it is still the case that by
county the population of females is systematically different from males
in ways that may explain regional variation in the propensity for male-
and female-owned firms.
VI. ANALYSIS
The empirical model explains the change in the propensity for firm
ownership as a function of initial conditions given by the lagged
regional characteristics. The model can be written as:
(2) [e.sub.git,t-[tau]] = [[beta].sub.0] +
[delta][h.sub.gi,t-[tau]]] + [gamma][f.sub.gi, t-[tau]] +
[zeta][r.sub.i,t- [tau]] = [[epsilon].sub.gi,t-[tau]]
where the subscript indicates the gender of focus g, at time t, in
county i. e is the change in the propensity for female (male)-owned
firms between time t and t - [tau], h is a vector of human capital
variables, f is vector of family structure variables, and r is a vector
of regional control variables. [delta], [gamma], and [zeta] are the
parameters to be estimated.
We estimate the male and female models using ordinary least squares
(OLS) and then combine the results using seemingly unrelated estimation,
which applies the Eicker-Huber-White sandwich covariance estimator. The
coefficients will be the same between the OLS and seemingly unrelated
estimation but the standard errors are smaller in the latter estimation
because it uses a larger number of observations to estimate the
simultaneous (co)variance matrix. The standard errors are valid
regardless of cross-equation correlation or heteroskedasticity.
Estimating the model in this way allows for cross-model hypotheses that
are useful for making gender comparisons. (2)
Selected results are presented in Table 3; the complete results are
reported in the Appendix. Clearly, human capital and family structure do
matter in determining changes in the propensity for both male- and
female-owned firms. The results for female human capital accumulation
are entirely consistent with our hypothesis. Growth in the
propensity for female-owned firms is lower in regions with a larger
share of females with only a high school diploma. Also consistent with
the hypothesis is the strong positive effect of the share of females
with a bachelor's degree. The coefficient on the share of females
with a master's degree is also positive but only marginally
significant. The relationship between human capital and growth in the
propensity for female-owned firms turns strongly negative again at the
doctorate level. These results are consistent with the descriptive
analysis by Fairlie and Robb (2009) who found that compared to male
business owners, a lower percentage of women business owners were high
school dropouts and also that a lower percentage had graduate degrees.
It seems the relationship of human capital accumulation to growth in the
propensity for female-owned firms forms an inverted "U."
Ownership growth is lower in counties with large shares of females with
either a doctorate or high school diploma at the highest. Ownership
growth is higher in counties with a large share of females with a
bachelor's degree and, to a lesser extent, a large share of females
with a master's degree.
The regional human capital profile that is most conducive to
increasing the propensity for male-owned firms is simpler than that for
women. The relationship of human capital to growth in the propensity for
male-owned firms is concentrated entirely on the strong positive
relationship to the share of males with a college degree. The inverted
"U" relationship observed for females is only true for males
to the extent that the college-educated seem to have the greatest
potential for firm ownership.
Marriage has a weakly positive relationship to changes in the
propensity for male-owned firms. Children, however, have strong negative
effect on changes in the propensity for male- and female-owned firms,
contrary to our hypothesis. It seems that a local demographic with many
young children would also feature a lower propensity for both male- and
female-owned firms. (3)
The results are substantively consistent when including state fixed
effects or a metropolitan statistical area (MSA) effect as reported in
the Appendix. In this sample of counties, many MSAs have just one county
ruling out a conventional fixed effect for each MSA. Instead we
construct an MSA effect equal to one if there is at least one other
county in the same MSA. We expect that there is a benefit to having
neighboring counties via agglomeration effects. The results, reported in
the Appendix, show a positive effect associated with having at least one
MSA neighbor but the effect is only weakly significant in the female
model and insignificant in the male model. The key results are generally
consistent with the results presented in Table 3, but some cases show
larger gender differences.
As a first step toward understanding gender differences, we compare
the coefficients on human capital, marriage, and children from each
model shown in Table 4. Only the coefficients on the share of male and
female high school graduates and those with a doctorate are
statistically different. Keeping in mind that the high school and PhD
educated share of the population are significant only in the female
regression, there is little evidence of behavioral differences between
genders. The college-educated population share and children per adult
are the only variables that are statistically significant in both
regressions, but there is no statistical difference between
coefficients.
Comparing the coefficients as in Table 4 is informative, but
incomplete for identifying gender differences. To get a more accurate
sense of the gender differences it is necessary to consider the
coefficients with respect to the data. Table 5 shows the impulse
response to each variable, the percentage point change in the growth of
the propensity for male- and female-owned firms that results from a 1
standard deviation change in each explanatory variable. Even though the
coefficients are not statistically different in most cases, there are
large gender differences as measured by the impulse responses shown in
Table 5.
To simplify the discussion, we focus only on the impulse responses
for statistically significant coefficients from the regression models.
The largest positive impulse response is at the bachelor's degree
across genders, though the effect for men is more than double the size
of the effect for women. A I standard deviation increase in the share of
college-educated males corresponds to an almost 0.7 percentage point
increase to the change in the propensity for male-owned firms,
equivalent to just over one-third of the average change in the
propensity for male-owned firms. A 1 standard deviation increase in the
share of females with a bachelor's degree corresponds to a 0.34
percentage point increase to the change in the propensity for
female-owned firms equivalent to just under one-third of the average
change in the propensity for female-owned firms. The share of females
with a master's degree also corresponds to an increase in the
change in the propensity for female-owned firms of 0.2 percentage
points. A 1 standard deviation increase in the share of females with
high school graduates and the share of females with doctorates
correspond to a 0.6 and 0.4 percentage point decrease, respectively, to
the change in the propensity for female-owned firms.
The impulse response to a 1 standard deviation change in the number
of children per adult is larger for men. An increase of children per
adult of 0.09, 1 standard deviation, corresponds to a roughly 0.6
percentage point decrease in the change in the propensity for
female-owned firms relative to the labor force and a 0.8 percentage
point decrease in the change in the propensity for male-owned firms.
The behavioral differences between men and women are relatively
weak when we consider coefficients alone. It may be easy to conclude
that behavioral differences between men and women do little to explain
the gender disparity in business ownership. However, the impulse
responses, which combine for each gender the behavioral component given
by the coefficient with the characteristic component given by the data,
show that the gender differences are substantial. In addition, it is
also useful to consider the behavioral differences in aggregate.
Combining the slight behavioral differences across the coefficients
results in a cumulative effect that explains a significant portion of
the gender gap as shown in the following section.
VI. BLINDER-OAXACA DECOMPOSITION
As in many studies that focus on the difference in labor market
outcomes between groups, we decompose the mean differences in the change
in the propensity for male- and female-owned firms based on the above
linear regression models following the Blinder-Oaxaca Decomposition
(Blinder 1973; Oaxaca 1973). Whereas the difference in means,
coefficients, and impulse responses have been broken down for each
variable of interest in the previous section, a Blinder-Oaxaca
decomposition usefully summarizes the importance of the behavioral
differences taken together versus characteristic differences taken
together. As in Jann (2008), the question is how much of the mean
outcome difference is accounted for by group differences in the
explanatory variables between the male and female populations across
counties.
The decomposition described here is formulated from the viewpoint
of females. So, the group differences in the predictors are weighted by
the coefficients from the female model. The characteristic component
measures the expected adjustment to the change in the propensity for
female-owned firms if the average characteristics of the female
populations were the same as the average characteristics of the male
populations in our sample of counties. Similarly, the coefficient
component measures the expected adjustment to the change in the
propensity for female-owned firms, if the female population behaved as
the male population, and therefore, the coefficients between the male
and female model were equal.
The top panel of Table 6 reports the mean predictions of the change
in the propensity for male-and female-owned firms and the difference,
while the decomposition is in the bottom panel. In this sample, the mean
change in the propensity for male-owned firms is 0.0189, close to 2
percentage points. The mean change in the propensity for female-owned
firms is 0.0109, close to 1 percentage point. The differential of 0.008
is divided into three components. The largest component, the
"characteristic effect" shows the importance of characteristic
differences between the male and female populations across counties. It
reflects the adjustment at the mean to the change in the propensity for
female-owned firms that we might expect if the mean characteristics of
the female populations across our sample of counties were the same as
the mean characteristics of male populations. The decrease of 0.0554
indicates that if the differences in average education attainment,
family structure, and other explanatory variables were eliminated, the
change in the growth of propensity for female-owned firms would actually
be negative. At the mean, the propensity for female-owned firms would
have fallen from close to 9.6% in 2002 to roughly 5.1% in 2007. In the
typical county in the sample beginning with approximately 8,100
female-owned firms in 2002, the negative change is equivalent to losing
roughly 3,900 female-owned firms during the period from 2002 to 2007.
The second term of 0.013 measures the adjustment to the change in
the propensity for female-owned firms if behavioral coefficients from
the male model were applied to the data on female populations in our
sample of counties. Adding the increase of 0.013 to the actual mean
change in the propensity for female-owned firms of 0.0109, shows that if
women behaved as men, the change in the propensity for female-owned
firms would be more than twice as high. Under this scenario, the
propensity for female-owned firms would have increased from close to
9.6% in 2002 to 12% in 2007, at the mean. This result implies an
increase of approximately 1,800 female-owned firms from 2002 to 2007 for
the typical county in the sample. The third component is the interaction
term that measures the simultaneous effect of differences in
characteristics and coefficients. In this case, it captures the
offsetting effects of the difference in characteristics and the
differences coefficients.
The implications of these results are important as they relate to
policy. A policy that results in the female population acquiring
characteristics more similar to the male population may actually result
in an increased gender disparity in business ownership. As long as women
behave differently, equalizing characteristics may do little to narrow
the gender disparity in firm ownership. For example, if highly educated
women are consistently more likely to enter wage-and-salary employment
relative to their male counterparts, increasing the share of highly
educated women to match the share of highly educated men will likely
only reduce female entrepreneurship.
Rather, a locality interested in increasing the propensity for
female-owned firms may be better served by a policy that instead focuses
on behavioral differences. The female population does not behave as the
male population. Policies aimed at increasing the change in the
propensity of female-owned firms must recognize these gender-specific
behaviors. For example, clearly the populations of men and women who
hold a bachelor's degree have the greatest potential for business
ownership. Yet, the propensity for college-educated males to act in
terms of entrepreneurship is roughly twice that for females as indicated
by the coefficients shown in Table 4 and the impulse responses shown in
Table 5. A policy that either incentivizes college-educated women to
choose entrepreneurship or relaxes the constraints they face could
effectively change the behavior of women in a way that increases
female-owned firms resulting in greater equality in business ownership.
VIII. CONCLUSION
This study indicates that the determinants of growth in the
propensity for male- and female-owned firms are different, particularly
with regard to local education attainment. Only the share of males with
a bachelor's degree explains changes in male firm ownership whereas
all four measures of female education attainment explain changes in
female firm ownership. The share of females with a doctorate and the
share with only a high school diploma are both negatively related to
changes in the propensity for female-owned firms, demonstrating that
female-owned firms may not necessarily increase as the female population
becomes more educated. The negative effect of large shares of less and
very highly educated females combined with the positive effect at the
bachelor's and master's degree level, suggests that the
relationship between female entrepreneurial activity and human capital
accumulation forms an inverted "U." The same relationship is
true for men only to the extent that there is a strong positive effect
at the college level. Family structure is also an important determining
factor in the growth of male- and female-owned firms. In contrast to
some previous studies that suggest that children have a positive
relationship to entrepreneurship, our regional results show that the
effect of children per adult has a highly significant negative effect on
changes in the propensity for both male- and female-owned firms.
The article's decomposition shows that even though the effect
from characteristic differences between the male and female population
is much larger in absolute value, the behavioral differences are crucial
for closing the gender gap in business ownership. If the female
population had the same education attainment, marriage rate, average
number of children, median age, and employment population ratio as the
male population, the gender gap in the propensity for firm ownership may
actually be much wider. Conversely, although the effect is smaller, if
women exhibited the same behavior as males but were still
characteristically different from men, the increase in the propensity
for female-owned firms would be higher, resulting in a narrower gender
gap. In contrast to most past considerations of regional
entrepreneurship policy that have implicitly focused on males, the
decomposition makes clear that policies aimed at enhancing
entrepreneurship will affect men and women differently.
APPENDIX
DATA DEFINITIONS
TABLE A1
Data Sources and Descriptions
Variable Source
Propensity for female- Survey of Business Owners 2002
(male-)owned firms and 2007, Census 2000,
American Community Survey
(ACS) 2005 Estimates
High school graduates, as % of Census 2000
male/female adult population
College graduates, as % of Census 2000
male/female adult population
Persons with MA degree, as % of Census 2000
male/female adult population
Persons with PhD, as % of Census 2000
male/female adult population
Employment population ratio Census 2000
Median age Census 2000
Proprietor income per job Bureau of Economic Analysis
Wage-and-salary income per job Bureau of Economic Analysis
Median housing value Census 2000
Owner-occupied homes Census 2000
Growth rate of income per capita Bureau of Economic Analysis
Construction establishments County Business Patterns 2000
Service establishments County Business Patterns 2000
Retail trade establishments County Business Patterns 2000
Natural amenities scale United States Department of
Agriculture. Economic
Research Service
Population density Census 2000
Variable Description
Propensity for female- The ratio of female-owned
(male-)owned firms (nonfemale owned) firms to the
female (male) labor force.
High school graduates, as % of The ratio of female (male) high
male/female adult population school graduates to the female
(male) population age 25 or
older. Education level
determined by highest degree
attained.
College graduates, as % of The ratio of female (male)
male/female adult population college graduates to the female
(male) population age 25 or
older.
Persons with MA degree, as % of The ratio of female (male) MA
male/female adult population graduates to the female (male)
population age 25 or older.
Persons with PhD, as % of The ratio of female (male) PhD
male/female adult population graduates to the female (male)
population age 25 or older.
Employment population ratio The ratio employed females
(males) to the female (male)
population.
Median age As reported.
Proprietor income per job The ratio of proprietor income to
proprietor employment.
Wage-and-salary income per job The ratio of wage-and-salary
disbursements to
wage-and-salary employment.
Median housing value As reported.
Owner-occupied homes The ratio of owner-occupied
homes to total.
Growth rate of income per capita Growth rate of per capita income
for the 5-year period ending in
the year of the measure of firm
ownership (i.e., 1997-2002).
Construction establishments The ratio of construction
establishments to total.
Service establishments The ratio of service
establishments to total.
Retail trade establishments The ratio of retail trade
establishments to total.
Natural amenities scale The natural amenities of a
location based are based on
topography and climate. A high
natural amenity score for a
county is associated with warm,
sunny winters, low-humidity
summers, and mountainous or
otherwise scenic terrain.
Population density Tract population density is
weighted with the tract
population and summed by
county.
RESULTS
TABLE A2
OLS Results
Number of Observations = 646
[R.sup.2] = 0.2602
F = 13.33
[DELTA] Propensity for
Female-Owned Firms
Variable Coef. Robust SE
Propensity for female-owned -.3400 ** 00.0440
firms, 2002
Female high school graduates, as -.0847 ** 00.0167
% of female adult population
Female college graduates, as % .0645 * 00.0294
of female adult population
Females with MA degree, as % .0938 00.0561
of female adult population
Females with PhD, as % of -.6392 ** 00.2441
female adult population
Median age -.0010 *** 00.0003
Female employment population -.0762 ** 00.0164
ratio
Married females as a share of the .0177 00.0217
adult female population
Children per female over 16 -.0612 ** 00.0129
Proprietor income per job .0000 00.0001
Wage-and-salary income per job .0000 00.0001
Median housing value .0000 00.0000
Owner-occupied homes, % of .0276 * 00.0128
total
Growth rate of income per capita .0001 00.0001
Service establishments, % of total -.0037 00.0240
Retail trade establishments, % of -.0034 00.0272
total
Construction establishments, % .0199 00.0300
of total
Natural amenities score .0006 00.0003
Population density .0003 00.0002
Constant .1430 *** 00.0236
Number of Observations = 646
[R.sup.2] = 0.1527
F = 5.44
[DELTA] Propensity for
Male-Owned Firms
Variable Coef. Robust SE
Propensity for male-owned firms, -.0856 *** 00.0285
2002
Male high school graduates, as % of -.0075 00.0257
female adult population
Male college graduates, as % of male .1231 ** 00.0451
adult population
Males with MA degree, as % of male -.0607 00.0951
adult population
Males with PhD, as % of male adult .0013 00.1131
population
Median age -.0022 ** 00.0005
Male employment population ratio .0093 00.0220
Married males as a share of the adult .0552 00.0325
male population
Children per male over 16 -.0906 ** 00.0176
Proprietor income per job -.0001 00.0001
Wage-and-salary income per job -.0002 00.0002
Median housing value .0000 ** 00.0000
Owner-occupied homes, % of total .0455 * 00.0205
Growth rate of income per capita .0002 00.0002
Service establishments, % of total -.0233 00.0392
Retail trade establishments, % of -.0330 00.0392
total
Construction establishments, % of -.0853 * 00.0400
total
Natural amenities score .0012 ** 00.0005
Population density .0007 ** 00.0002
Constant .1215 00.0352
Significance at the 1%, 5%, and 10% level shown by ***, **, and *,
respectively.
TABLE A3
Seemingly Unrelated Regressions
Observations [R.sup.2] [chi square] P
[DELTA] Propensity 646 0.2597 237.63 .00
for female-owned
firms
[DELTA] Propensity 646 0.1519 122.51 .00
for male-owned
firms
Breusch Pagan test of p = .0448
independent
residuals p value
Variable Coef. SE
[DELTA] Propensity for female-owned firms
Propensity for female-owned firms, 2002 -.3601 *** 0.0342
Female high school graduates, as % of female -.0855 *** 0.0171
adult population
Female college graduates, as % of female adult .0683 *** 0.0270
population
Females with MA degree, as % of female adult .0893 * 0.0533
population
Females with PhD, as % of female adult -.6037 *** 0.2137
population
Mamed females as a share of the adult female .0189 0.0199
population
Children per female over 16 -.0590 *** 0.0128
Female employment population ratio -.0774 *** 0.0160
Median age -.0009 *** 0.0003
Proprietor income per job .0000 0.0001
Wage-and-salary income per job .0000 0.0001
Median housing value .0000 0.0000
Owner-occupied homes, % of total .0279 ** 0.0140
Growth rate of income per capita .0001 0.0001
Service establishments, % of total -.0032 0.0239
Retail trade establishments, % of total -.0036 0.0233
Construction establishments, % of total .0198 0.0267
Natural amenities score .0007 ** 0.0003
Population density .0003 * 0.0002
Constant .1408 *** 0.0224
[DELTA] Propensity for male-owned firms
Propensity for male-owned firms, 2002 -.1018 *** 0.0219
Male high school graduates, as % of male adult -.0076 0.0239
population
Male college graduates, as % of male adult .1263 *** 0.0373
population
Males with MA degree, as % of male adult -.0468 0.0803
population
Males with PhD, as % of male adult population .0043 0.1055
Mamed males as a share of the adult male .0569 ** 0.0289
population
Children per male over 16 -.0909 *** 0.0152
Male employment population ratio .0064 0.0180
Median age -.0021 *** 0.0005
Proprietor income per job -0.0001 0.0001
Wage-and-salary income per job -.0003 * 0.0002
Median housing value .0000 ** 0.0000
Owner-occupied homes, % of total 0.0464 ** 0.0205
Per capita income growth 0.0002 0.0002
Service establishments, % of total -0.0329 0.0365
Retail trade establishments, % of total -0.0377 0.0346
Construction establishments, % of total -0.0876 ** 0.0369
Natural amenities score 0.0012 *** 0.0004
Population density 0.0007 *** 0.0002
Constant 0 1272 0.0310
Significance at the 1%, 5%, and 10% level shown by ***, **, and *,
respectively.
TABLE A4
Seemingly Unrelated Estimation
Number of Observations = 646
Variable Coef. Robust SEs
[DELTA] Propensity for female-owned firms
Propensity for female-owned firms, 2002 -.3400 *** 0.0433
Female high school graduates, as % of -.0847 *** 0.0165
female adult population
Female college graduates, as % of female .0645 ** 0.0289
adult population
Females with MA degree, as % of female .0938 * 0.0553
adult population
Females with PhD, as % of female adult -.6392 *** 0.2405
population
Married females as a share of the adult .0177 0.0214
female population
Children per female over 16 -.0612 *** 0.0127
Female employment population ratio -.0762 *** 0.0162
Median age -.0010 *** 0.0003
Proprietor income per job .0000 0.0001
Wage-and-salary income per job .0000 0.0001
Median housing value .0000 0.0000
Owner-occupied homes, % of total .0276 0.0126
Growth rate of income per capita .0001 0.0001
Service establishments, % of total -.0037 0.0236
Retail trade establishments, % of total -.0034 0.0268
Construction establishments, % of total .0199 0.0296
Natural amenities score .0006 ** 0.0003
Population density .0003 0.0002
Constant .1430 *** 0.0233
[DELTA] Propensity for male-owned firms
Propensity for male-owned firms, 2002 -.0856 *** 0.0281
Male high school graduates, as % of male -.0075 0.0253
adult population
Male college graduates, as % of male adult .1231 *** 0.0445
population
Males with MA degree, as % of male adult -.0607 0.0937
population
Males with PhD, as % of male adult .0013 0.1114
population
Married males as a share of the adult male .0552 * 0.0320
population
Children per male over 16 -.0906 0.0173
Male employment population ratio .0093 0.0217
Median age -.0022 *** 0.0005
Proprietor income per job -.0001 0.0001
Wage-and-salary income per job -.0002 0.0002
Median housing value .0000 *** 0.0000
Owner-occupied homes, % of total .0455 ** 0.0202
Per capita income growth .0002 0.0002
Service establishments, % of total -.0233 0.0386
Retail trade establishments, % of total -.0330 0.0387
Construction establishments, % of total -.0853 ** 0.0394
Natural amenities score .0012 *** 0.0005
Population density .0007 *** 0.0002
Constant .1215 *** 0.0346
Significance at the 1%, 5%, and 10% level shown by ***, **, and *,
respectively.
TABLE A5
Seemingly Unrelated Estimation with State Fixed Effects
Number of Observations = 644
Variable Coef. Robust SEs
[DELTA] Propensity for female-owned firms
Propensity for female-owned firms, 2002 -.4589 *** 0.0452
Female high school graduates, as % of -.0142 0.0278
female adult population
Female college graduates, as % of female .1074 *** 0.0316
adult population
Females with MA degree, as % of female .1790 ** 0.0826
adult population
Females with PhD, as % of female adult -.4926 ** 0.2504
population
Married females as a share of the adult -0.0306 0.0238
female population
Children per female over 16 0.0071 0.0153
Female employment population ratio -.0622 *** 0.0180
Median age .0004 0.0004
Proprietor income per job .0000 0.0001
Wage-and-salary income per job -.0002 0.0001
Median housing value 5.70E-08 ** 0.0000
Owner-occupied homes, % of total -.0119 0.0164
Growth rate of income per capita .0004 *** 0.0001
Service establishments, % of total -.0447 * 0.0235
Retail trade establishments, % of total .0009 0.0264
Construction establishments, % of total .0388 0.0304
Natural amenities score .0010 ** 0.0004
Population density .0001 0.0002
Constant .0859 *** 0.0274
[DELTA] Propensity for male-owned firms
Propensity for male-owned firms, 2002 -.1393 *** 0.0336
Male high school graduates, as % of male .0293 0.0387
adult population
Male college graduates, as % of male .1204 ** 0.0483
adult population
Males with MA degree, as % of male adult .0950 0.0984
population
Males with PhD, as % of male adult .0288 0.1116
population
Married males as a share of the adult .0107 0.0342
male population
Children per male over 16 -.0413 ** 0.0190
Male employment population ratio -.0148 0.0221
Median age -.0007 0.0006
Proprietor income per job -.0001 0.0001
Wage-and-salary income per job -0.0003 * 0.0001
Median housing value -1.01E-08 0.0000
Owner-occupied homes, % of total .0121 0.0245
Per capita income growth .0004 * 0.0002
Service establishments, % of total -.0646 0.0412
Retail trade establishments, % of total -.0503 0.0399
Construction establishments, % of total -.0705 0.0479
Natural amenities score .0025 *** 0.0007
Population density .0004 ** 0.0002
Constant .1228 *** 0.0368
Significance at the 1%, 5%, and 10% level shown by ***, **, and *,
respectively. Vermont and Wyoming not included because they each
only have one county in the sample.
TABLE A6
Difference in Coefficients with State Fixed Effects
Female Male [chi square] p Value
High school graduates, -0.0142 0.0293 1.03 .31
as % of adult
population
College graduates, as % 0.1074 0.1204 0.06 .80
of adult population
Adults with MA degree, 0.1790 0.0950 0.46 .50
as % of adult
population
Adults with PhD, as % of -0.4926 0.0288 4.12 .04
adult population
Married adults as a -0.0306 0.0107 1.03 .31
share of the adults
population
Children per adult 0.0071 -0.0413 4.13 .04
TABLE A7
Impulse Responses with State Fixed Effects (in Percentage Points)
Female (%) Male (%)
High school graduates, as % of adult -0.09 0.21
population
College graduates, as % of adult population 0.57 0.67
Adults with MA degree, as % of adult 0.46 0.26
population
Adults with PhD, as % of adult population -0.27 0.04
Married adults as a share of the adults -0.17 0.06
population
Children per adult 0.07 -0.37
TABLE A8
Seemingly Unrelated Estimation with MSA Effect
Number of Observations = 646
Variable Coef. Robust SEs
[DELTA] Propensity for female-owned firms
Propensity for female-owned firms, 2002 -.3397 *** 0.0431
Female high school graduates, as % of -.0840 *** 0.0165
female adult population
Female college graduates, as % of female 0.0644 ** 0.0287
adult population
Females with MA degree, as % of female 0.0870 0.0559
adult population
Females with PhD, as % of female adult -.6033 0.2391
population
Married females as a share of the adult 0.0188 0.0212
female population
Children per female over 16 -.0607 *** 0.0126
Female employment population ratio -.0776 *** 0.0160
Median age -.0010 *** 0.0003
Proprietor income per job 0.0000 0.0001
Wage-and-salary income per job 0.0000 0.0001
Median housing value 0.0000 0.0000
Owner-occupied homes, % of total 0.0239 * 0.0127
Growth rate of income per capita 0.0001 0.0001
Service establishments, % of total -0.0090 0.0236
Retail trade establishments, % of total 0.0046 0.0272
Construction establishments, % of total 0.0157 0.0294
Natural amenities score 0.0006 0.0003
Population density 0.0002 0.0002
MSA neighbors 0.0022 * 0.0012
Constant 0.1434 *** 0.0232
[DELTA] Propensity for male-owned firms
Propensity for male-owned firms, 2002 -.0834 *** 0.0284
Male high school graduates, as % of male -.0057 0.0252
adult population
Male college graduates, as % of male adult 0.1214 *** 0.0446
population
Males with MA degree, as % of male adult -.0607 0.0936
population
Males with PhD, as % of male adult 0.0122 0.1113
population
Married males as a share of the adult male 0.0565 0.0320
population
Children per male over 16 -.0904 *** 0.0173
Male employment population ratio 0.0081 0.0216
Median age -.0022 *** 0.0005
Proprietor income per job -.0001 0.0001
Wage-and-salary income per job -.0002 0.0002
Median housing value 0.0000 ** 0.0000
Owner-occupied homes, % of total 0.0432 ** 0.0201
Per capita income growth 0.0002 0.0002
Service establishments, % of total -.0277 0.0390
Retail trade establishments, % of total -.0288 0.0390
Construction establishments, % of total -.0887 ** 0.0399
Natural amenities score 0.0012 *** 0.0005
Population density 0.0006 *** 0.0002
MSA neighbors 0.0016 0.0017
Constant 0.1223 *** 0.0346
Notes: The variable "MSA neighbors" is a dummy variable equal to 1
if a county has at least one other "neighbor" county in their MSA
included in the sample.
Significance at the 1%, 5%, and 10% level shown by ***, **, and *,
respectively.
TABLE A9
Difference in Coefficients with MSA Effect
Female Male [chi square] p Value
High school graduates, -0.0142 0.0293 1.03 .31
as % of adult
population
College graduates, as % 0.1074 0.1204 0.06 .80
of adult population
Adults with MA degree, 0.1790 0.0950 0.46 .50
as % of adult
population
Adults with PhD, as % of -0.4926 0.0288 4.12 .04
adult population
Married adults as a -0.0306 0.0107 1.03 .31
share of the adults
population
Children per adult 0.0071 -0.0413 4.13 .04
TABLE A10
Impulse Responses with MSA Effect (in Percentage Points)
Female (%) Male (%)
High school graduates, as % of adult -0.09 0.21
population
College graduates, as % of adult population 0.57 0.67
Adults with MA degree, as % of adult 0.46 0.26
population
Adults with PhD, as % of adult population -0.27 0.04
Married adults as a share of the adults -0.17 0.06
population
Children per adult 0.07 -0.37
BLINDER-OAXACA DECOMPOSITION
As in Jann (2008), the question is how much of the mean outcome
difference,
(A1) D = E ([e.sub.mit,t-[tau]]) - E [[e.sub.fit,t-[tau]])
where [e.sub.git,t-[tau]] denotes the expected value of the change
in propensity for male- or female-owned firms between time t and t -
[tau] in county is accounted for by group differences in the explanatory
variables. In our case, the female population in a county i is an
observation. The female populations across counties form a
"group." The structure for the male "group" follows
the same logic.
Based on the linear model
(A2) [e.sub.git,t-[tau]] = [X'.sub.gi,t-[tau]] +
[[beta].sub.g] [e.sub.git], E ([e.sub.git]) = 0 g member (m,f)
where h, f, and r are summarized by X,[beta] summarizes the
estimated parameters, and e is the error term, the mean differential can
be expressed as the difference in the group-specific means of the
regressors as follows
(A3) D = E([e.sub.mi,t-[tau])-E(efit-t-[tau]) =
E([X.sub.mi,t-[tau]]' [[beta].sub.m] [E.sub.(Xfi,t-[tau]]'
[beta]f
because
(A4)
[e.sub.git-t-[tau]] + = E ([X'.sub.gi,t-[tau])][[beta].sub.g]
E ([[epsilon].sub.git]) = E([X.sub.git-[tau])'] [[beta].sub.g]
where E([[beta].sub.git]) = [[beta].sub.git] and
[E.sub.([epsilon]gil)] = 0 by assumption.
To identify the contribution of group differences in the
explanatory variables to the outcome difference (A4) can be rearranged
and expressed as follows.
(A5) D=(E ([X.sub.mi,t-[tau]]) - E ([X.sub.fi,t-[tau]]))'
[[beta].sub.f] + E ([X.sub.fi,t-[tau]]))' ([[beta].sub.m] -
[[beta].sub.f]) + (E([X.sub.mit-[tau])-E(([X.sub.fi,t-[tau]]))'([[beta].sub.m] - [[beta].sub.f]).
The decomposition is expressed in three parts.
(A6) D = E + C + I.
The first part,
(A7) E=(E([X.sub.mi. t-[tau]]) - E(([X.sub.fi,t-[tau]]))'
[[beta].sub.f]
captures the component of the differential that is due to
characteristic differences between the male and female population, also
called the "endowments effect." The second part,
(A8) C = E(([X.sub.fi,t-[tau]]))'([beta].sub.m],-
[[beta].sub.f])
measures the part of the differential attributable to behavioral
differences measured by the differences in coefficients. Last,
(A9) I = (E ([X.sub.m,t-[tau]])) - E ([X.sub.fi,t-[tau]]))'
([[beta].sub.m] - [[beta].sub.f])
is an interaction term that captures the fact that differences in
endowments and coefficients exist simultaneously between men and women.
ABBREVIATIONS
ACS: American Community Survey
MSA: Metropolitan Statistical Area
OLS: Ordinary Least Squares
SBO: Survey of Business Owners
doi: 10.1111/ecin.12224
REFERENCES
Acs, Z. J., and C. Armington. "Employment Growth and
Entrepreneurial Activity in Cities." Regional Studies, 38(8),
2004,911-27.
--. Entrepreneurship, Geography, and American Economic Growth. New
York: Cambridge University Press, 2006.
Bartik, T. J. "Business Location Decisions in the United
States: Estimates of the Effects of Unionization, Taxes, and Other
Characteristics of States." Journal of Business & Economic
Statistics, 3(1), 1985, 14-22.
--. "Small Business Start-Ups in the United States: Estimates
of the Effects of Characteristics of States." Southern Economic
Journal 55, 1989, 1004-18.
Blanchflower, D. G., and A. J. Oswald. "What Makes an
Entrepreneur?" Journal of Labor Economics, 16(1), 1998,26-60.
Blinder, A. S. "Wage Discrimination: Reduced Form and
Structural Estimates." Journal of Human Resources, 8(4),
1973,436-55.
Boden, R. J. "Gender and Self-Employment Selection: An
Empirical-Assessment." Journal of Socio-Economics, 25(6), 1996,
671-82.
--. "Flexible Working Hours, Family Responsibilities, and
Female Self-Employment Selection." American Journal of Economics
and Sociology 58(1), 1999a, 71-83.
--. "Gender Inequality in Wage Earnings and Female
Self-Employment Selection." Journal of Socio-Economics 28, 1999b,
351-64.
Bruce, D. "Do Husbands Matter? Married Women Entering
Self-Employment." Small Business Economics, 13(4), 1999,317-29.
Bunten, D., S. Weiler, E. Thompson, and S. Zahran.
"Entrepreneurship, Information, and Growth." Journal of
Regional Science, 2014. doi: 10. Ill 1/jors. 12157 [Epub ahead of
print].
Cowling, M., and M. Taylor. "Entrepreneurial Women and Men:
Two Different Species?" Small Business Economics, 16(3), 2001,
167-75.
DeTienne, D., and G. Chandler. "The Role of Gender in
Opportunity Identification." Entrepreneurship: Theory and Practice,
31(3), 2007, 365-86.
Evans, D. S., and L. S. Leighton. "Some Empirical Aspects of
Entrepreneurship." American Economic Review, 79(3), 1995,519-35.
Fairlie, R. W., and A. M. Robb. "Gender Differences in
Business Performance: Evidence from the Characteristics of Business
Owners Survey." Small Business Economics, 33(3), 2009, 375-95.
Figueroa-Armijos, M., and T. G. Johnson. "Entrepreneurship in
Rural America Across Typologies, Gender and Motivation." Journal of
Developmental Entrepreneurship, 18(2), 2013, 1350014.
Florida, R. "The Rise of the Creative Class." Washington
Monthly, 34(5), 2002, 15-25.
Georgellis, Y., and H. J. Wall. "Gender Differences in
SelfEmployment." International Review of Applied Economics, 19(3),
2005, 321-42.
Glaeser, E. L. "Entrepreneurship and the City," in
Entrepreneurship and Openness: Theory and Evidence, edited by D.
Audretsch, R. Litan, and R. Strom. Northampton: Edward Elgar Publishing,
2009, 131 -80.
Goetz, S. J., and D. Freshwater. "State-Level Determinants of
Entrepreneurship and a Preliminary Measure of Entrepreneurial
Climate." Economic Development Quarterly, 15(1), 2001,58-70.
Goetz, S. J., and A. Rupasingha. "Determinants of Growth in
Non-Farm Proprietor Densities in the US, 1990-2000." Small Business
Economics, 32(4), 2009, 425-38.
Huarng, K.-H., A. Mas-Tur, and T. H.-K. Yu. "Factors Affecting
the Success of Women Entrepreneurs." International Entrepreneurship
and Management Journal, 8(4), 2012,487-97.
Hundley, G. "Male/Female Earnings Differences in
Self-Employment: The Effects of Marriage, Children, and the Household
Division of Labor." Industrial and Labor Relations Review, 54(1),
2000, 95-114.
Jann, B. "The Blinder-Oaxaca Decomposition for Linear
Regression Models." The Stata Journal, 8(4), 2008, 453-79.
Kim, R H., H. E. Aldrich, and L. A. Keister. "Access (Not)
Denied: The Impact of Financial, Human, and Cultural Capital on
Entrepreneurial Entry in the United States." Small Business
Economics, 27(2), 2006, 5-22.
Kourilsky, M. L., and W. B. Walstad. "Entrepreneurship and
Female Youth: Knowledge, Attitudes, Gender Differences, and Educational
Practices." Journal of Business Venturing, 13, 1998, 77-88.
Lee, S. Y., R. Florida, and Z. J. Acs. "Creativity and
Entrepreneurship: A Regional Analysis of New Firm Formation."
Regional Studies, 38(8), 2004, 879-91.
Lombard, K. "Female Self-Employment and Demand for Flexible,
Nonstandard Work Schedules." Economic Inquiry, 39(2), 2001, 214-37.
Low, S. A. "Defining and Measuring Entrepreneurship for
Regional Research: A New Approach." PhD thesis, University of
Illinois Urbana-Champaign. 2009
Low, S. A., and S. Weiler. "Employment Risk, Returns, and
Entrepreneurship." Economic Development Quarterly, 26(3), 2012,
238-51.
Low, S. A., J. Henderson, and S. Weiler. "Gauging a
Region's Entrepreneurial Potential." Economic Review, 90,2005,
61-89.
Malecki, E. J. "Entrepreneurship in Regional and Local
Development." International Regional Science Review, 16(1-2), 1994,
119-53.
Matsa, D. A., and A. R. Miller. "Workforce Reductions at
Women-Owned Businesses in the United States." Industrial and Labor
Relations Review, 67(2), 2014,422-52.
Oaxaca. R. "Male-Female Wage Differentials in Urban Labor
Markets." International Economic Review, 14(3), 1973, 693-709.
OECD. "Women's Entrepreneurship: Issues and
Policies." Paper presented at the 2nd OECD Conference of Ministers
Responsible for Small and Medium-Sized Enterprises (SMEs), Promoting
Entrepreneurship and Innovative SMEs in a Global Economy: Towards a More
Responsible and Inclusive Globalisation, Istanbul, Turkey 3-5 June,
2004. Accessed April 21, 2015.
http://www.oecd.Org/cfe/smes/31919215.pdf.
Reynolds, P D. "New Firm Creation in the United States: A PSED
I Overview." Foundations and Trends in Entrepreneurship, 3(1),
2007, 1-150.
Rosenthal, S. S., and W. C. Strange. "Female Entrepreneurship,
Agglomeration, and a New Spatial Mismatch." Review of Economics and
Statistics, 94, 2012, 764-88.
Rosti, L., and F. Chelli. "Gender Discrimination,
Entrepreneurial Talent and Self-Employment." Small Business
Economics, 24(2), 2005, 131-42.
Rupasingha, A., and S. J. Goetz. "Self-Employment and Local
Economic Performance: Evidence from US Counties." Papers in
Regional Science, 92(1), 2011, 141-61.
Storey, D. J. "The Birth of New Firms-Does Unemployment
Matter? A Review of the Evidence." Small Business Economics, 3,
1991, 167-78.
Taniguchi, H. "Determinants of Women's Entry into
SelfEmployment." Social Science Quarterly, 83(3), 2002, 875-93.
(1.) Prior to 2007, male-owned firms were not tabulated by the
Census. Only female-owned firms and the total number of firms were
counted. The number of male-owned firms is calculated as the total
number of firms less female-owned firms.
(2.) Employment decisions for men and women are likely determined
to some extent by the household and for that reason it may not be
appropriate to estimate the regressions separately. Instead, the
propensity for male- and female-owned firms may be correlated via
household decision-making. Consequently, the models may be related via
correlated error terms. If the two error terms are correlated, then
estimating the equations jointly is a more efficient alternative to
estimating them equation-by-equation. If the errors are uncorrelated
across equations, then the estimates will be identical to the OLS
estimates of each equation, which are reported in the Appendix. Though
the results of seemingly unrelated regressions (SUR) are reported in the
Appendix and differences are slight, we focus on seemingly unrelated
estimation, which is robust to cross-model correlation and
heteroskedasticity.
(3.) This study uses all children under the age of 17 to calculate
children per adult. However, it may be the case that the relationship
between children per adult and the propensity for firms depends on the
age distribution of children. As children age and become more
independent, women in particular may be better able to commit themselves
to a business venture or, as the case may be, pursue wage-and-salary
employment. Using the National Longitudinal Survey of Youth, Taniguchi
(2002) finds that having small children has no effect on women's
entry into self-employment, though it has a negative effect on
transitions into wage-and-salary employment. However, having older
children who are more self-sufficient does positively impact transition
in self-employment. Future research would benefit from decomposing the
measure of children per adult into smaller age groups. Access to health
insurance also likely interacts with the responsibility of raising
children. Parents with insurance through their employer may be reluctant
to transition to self-employment if it means reducing or giving up their
coverage entirely. Unfortunately, insurance coverage is beyond the scope
of this county-level study, but should be considered in future research.
TESSA CONROY and STEPHAN WEILER *
* Alexandra Bernasek, David Mushinski, and Dawn Thilmany provided
appreciated input on the dissertation that formed the basis for part of
this paper. We owe particular thanks to Eric Thompson for his insight to
suggest a decomposition for the analysis as well as the editor for his
supportive guidance of the manuscript to publication and the referees
for their constructive suggestions.
Conroy: Research Associate, Department of Agricultural and Applied
Economics, University of Wisconsin-Madison, Madison, WI53706. Phone
608-265-4327, Fax 608-2624376, E-mail tconroy2@wisc.edu
Weiler: Professor and Research Associate Dean. Department of
Economics, Colorado State University, Fort Collins, CO 80523. Phone
970-491-5538, Fax 970-4910528, E-mail Stephan.Weiler@colostate.edu
TABLE 1
Summary Statistics
Number of Mean SD Min. Max.
Observations = 646
Variable
Propensity for female- 9.57% 2.31% 4.44% 20.87%
owned firms, 2002
Propensity for female- 10.66% 2.53% 5.63% 25.14%
owned firms, 2007
Propensity for male- 22.66% 4.73% 12.06% 45.20%
owned firms, 2002
Propensity for male- 24.55% 4.94% 13.79% 47.68%
owned firms, 2007
Year = 2000
Female HS graduates, as
% of adult females 31.11% 6.67% 13.11% 52.17%
Female college graduates,
as % of adult females 14.76% 5.26% 5.63% 34.52%
Females with MA degree,
as % of adult females 5.65% 2.56% 1.65% 18.75%
Females with PhD,
as % of adult females 0.56% 0.55% 0.04% 4.75%
Married females as %
of females age 15+ 55.87% 5.64% 34.10% 73.24%
Children per female over 16 0.58 0.09 0.27 0.95
Female emp. pop. ratio 55.35% 6.09% 35.26% 72.01%
Female median age 36.77 3.64 23.10 55.40
Male HS graduates, as
% of adult males 29.14% 7.08% 10.13% 51.41%
Male college graduates,
as % of adult males 15.84% 5.57% 6.06% 38.64%
Males with MA degree,
as % of adult males 5.80% 2.78% 1.74% 20.54%
Males with PhD,
as % of adult males 1.42% 1.34% 0.09% 11.57%
Married males as %
of males age 15+ 59.85% 5.26% 39.65% 74.75%
Children per male over 16 0.62 0.09 0.26 0.96
Male emp. pop. ratio 66.64% 7.43% 36.38% 86.35%
Male median age 34.40 3.41 23.40 52.70
Proprietor income per
job ($ 1000s) 24.53 12.14 6.25 144.34
Wage-and-salary income
per job ($ 1000s) 31.19 8.06 21.20 143.89
Growth rate of 19.95% 5.46% 2.51% 93.95%
income per capita
Service estabs, % of total 37.86% 3.96% 25.07% 59.36%
Retail trade estabs, 24.61% 3.49% 16.01% 41.39%
% of total
Construction estabs, 11.35% 3.22% 1.83% 24.12%
% of total
Owner-occupied homes, 68.99% 8.90% 19.54% 88.08%
% of total
Median housing value 121,379 62,962 47,700 1,000.000
Natural amenities scale 0.65 2.64 -5.01 11.17
Tract-weighted population
density (thousands) 2.91 6.42 0.05 113.53
TABLE 2
Difference in Means
Female Male
Variable Mean Mean t-Statistic p Value
Female (male)-owned firms 9.57% 22.66% -89.98 .000
relative to the female
(male) labor force, 2002
Female (male)-owned firms 10.66% 24.55% -99.82 .000
relative to the female
(male) labor force, 2007
Year = 2000
High school graduates, as % 31.11% 29.14% 25.47 .000
of adult population
College graduates, as % of 14.76% 15.84% -19.00 .000
adult population
Persons with MA degree, as 5.65% 5.80% -3.26 .001
% of adult population
Persons with PhD, as % of 0.56% 1.42% -25.88 .000
adult population
Married persons as a share 55.87% 59.85% 45.47 .000
of the adult population
Children per person age 16 0.58 0.62 30.87 .000
and over
TABLE 3
Seemingly Unrelated Estimation
Number of Observations = 646
Variable Coef. Robust SEs
[DELTA] Propensity for female-owned firms
Female high school graduates, as % of -.0847 *** 00.0165
female adult population
Female college graduates, as % of female .0645 ** 00.0289
adult population
Females with MA degree, as % of female .0938 * 00.0553
adult population
Females with PhD, as % of female adult -.6392 *** 00.2405
population
Married females as a share of the adult .0177 00.0214
female population
Children per female over 16 -.0612 *** 00.0127
[DELTA] Propensity for male-owned firms
Male high school graduates, as % of male -.0075 00.0253
adult population
Male college graduates, as % of male adult .1231 *** 00.0445
population
Males with MA degree, as % of male adult -.0607 0.0937
population
Males with PhD, as % of male adult .0013 0.1114
population
Married males as a share of the adult male .0552 * 0.0320
population
Children per male over 16 -.0906 *** 0.0173
Significance at the 1%, 5%, and 10% level shown by ***, **, and *,
respectively.
TABLE 4
Difference in Coefficients
Female Male [chi square] p Value
High school graduates, -0.0847 -0.0075 6.34 .01
as % of adult
population
College graduates, as % 0.0645 0.1231 1.27 .26
of adult population
Adults with MA degree, 0.0938 -0.0607 2.07 .15
as % of adult
population
Adults with PhD, as % of -0.6392 0.0013 7.10 .01
adult population
Married adults as a 0.0177 0.0552 1.01 .31
share of the adults
population
Children per adult -0.0612 -0.0906 2.02 .15
TABLE 5
Impulse Responses (in Percentage Points)
Female Male
High school graduates, as % of adult -0.56 -0.05
population College graduates, as % of adult 0.34 0.69
population Adults with MA degree, as % of 0.24 -0.17
adult population Adults with PhD, as % of adult -0.35 0.002
population Married adults as a share of the adult 0.10 0.29
population Children per adult -0.57 -0.82
TABLE 6
Decomposition
Coef. Robust SE
Change in the propensity for .0189 *** 0.0008
male-owned firms
Change in the propensity for .0109 *** 0.0006
female-owned firms
Difference .0080 *** 0.0010
Characteristics -.0554 *** 0.0066
Behavior .0130 ** 0.0050
Interaction .0505 *** 0.0083
Significance at the 1 % and 5% level shown by *** and **,
respectively.
