The Determinants of Self-Employment Growth

Abstract

The sustained surge in self-employment since 2000 has largely gone unnoticed by policy makers and economic developers. Here the authors document this surge and identify variables associated with expanding self-employment. Results provide mixed evidence about the importance of capital access to self-employment growth, but reveal that different approaches are needed in different county types depending on their proximity to metro areas and population size, if the goal is to increase future rural self-employment rates. In all county types, the initial share of self-employed predicts self-employment growth, underscoring the importance of a culture favoring entrepreneurship and path dependence. Self-employment earnings and educational attainment also play significant roles, as does the ethnic diversity of the population. Population density matters in both rural and urban counties, but less so within individual rural–urban continuum code categories. State policy, especially labor market freedom, has important effects on self-employment in most county types and periods studied.

Keywords

entrepreneurship self-employment policy agglomeration

Introduction

Over the period 2000-2009, the ratio of rural¹ self-employed to wage-and-salaried workers surged from just less than 24% to more than 30%. The additional one million rural workers who were self-employed in 2009 represent an almost one-quarter increase over the year 2000. In contrast, there were more than half a million (568,000) fewer rural wage-and-salary jobs in 2009 than in 2000.² In the “jobless recovery” from the recession of 2000, self-employment has clearly been vital to the economic survival of many rural workers, households, and the communities in which they reside. Recent research also suggests that individuals in rural areas were drawn into self-employment by following opportunities before the recession of 2007 (e.g., brought about by technological change), but that they have been forced into such activities out of necessity since then due to the lackluster employment recovery (e.g., Figueroa-Armijos, Dabson, & Johnson, 2012). Despite the common and valid perception that self-employment itself is low paying (see, e.g., Goetz, Rupasingha, & Fleming, 2012, Fig. 2) and only a last resort for many workers, a growing literature suggests that self-employment has tangible positive impacts on local income and employment growth, and that it is also associated with reduced poverty rates at the county level (Fleming & Goetz, 2011; Goetz et al., 2012; Henderson & Weiler, 2010, provide a review; Rupasingha & Goetz, 2013).

Thus, understanding the determinants of growth in self-employment is important not only because the self-employed create jobs for themselves but also because they stimulate job creation elsewhere in the local economy. Given the fact that monthly wage-and-salary job creation rates are still below the level needed to absorb new entrants into the workforce, let alone allow a recovery of the jobs lost in the 2007-2008 recession, a focus on the self-employed and identifying possible ways of supporting them is timely and urgent. For example, a common perception in the media is that lack of access to credit at the community level prevents more individuals from working for themselves. Even though banks are reportedly “flush with cash,” credit barriers such as lack of collateral may make it difficult for potential entrants to access these funds. Using individual-level data from the 1996 and 2001 Survey of Income and Program participation waves, Bates, Lofstrom, and Servon (2010) find no evidence to suggest that capital access barriers prevent small business starts in general, but Bates and Robb (2013) suggest that limited access to capital prevents minority-owned businesses from contributing more fully to local economic development.

In addition, other locally varying predetermined variables have likely influenced changes in self-employment rates over time. Here we focus on these variables as potential policy levers. Other policy constraints, operating at the national level, also have been cited in the literature as restricting self-employment or entrepreneurship. These include the lack of a national health care program and disincentives that render workers who want to start their own business ineligible to receive unemployment compensation.

The plan of this article is as follows. In the next section, we motivate a regression model to explain rural self-employment growth, drawing on the existing literature. Then we discuss and define our data and present summary statistics. This is followed by the regression results, along with a discussion. Although the primary focus is on rural areas because entrepreneurship there is relatively more important and yet these areas have been studied to a lesser degree, as we note in note 1, we also present corresponding results and interpretations for urban areas (see Acs et al., 2008; Glaeser, 2007; Glaeser, Rosenthal, & Strange, 2010) for examples of urban-focused studies).³ Thus, one contribution of the article is that we study the impact of regressors on self-employment or proprietorship formations across different types of rural and urban areas, as measured by the U.S. Department of Agriculture’s (USDA) 2003 Rural–Urban Continuum Code (RUCC-03), and we also examine the dynamics across rural areas for different years corresponding to the peaks and valleys of the most recent business cycles (i.e., annually, starting in 2003).

Literature Review and Regression Model

While there is a growing literature on the individual-level and geographic area–level determinants of entrepreneurship (e.g., Acs & Armington, 2006; Bates, 1990; Doms, Lewis, & Robb, 2010; Goetz & Rupasingha, 2009; Henderson, Low, & Weiler, 2007; Michelacci & Silva, 2007), few systematic and rigorous studies explore the causes of self-employment or entrepreneurship at the level of U.S. county types or labor market areas, and none focus explicitly on the RUCCs (defined below). Because rural areas are not homogeneous but differ in terms of key characteristics such as population density and access or proximity to cities, we maintain that such an analysis is important to fully understand the process of self-employment growth in those areas (see also Walzer, Athiyaman, & Hamm, 2007, p. 68, who mention the RUCC as one proxy measuring access to metro areas). And, no prior study examines relationships in the current economic downturn specifically at the rural county level.

To motivate our regression model, we draw primarily on two relatively recent studies by Acs and Armington (2006) and Goetz and Rupasingha (2009). The book Entrepreneurship, Geography, and American Economic Growth by Acs and Armington (2006) marked one of the first applications of new growth theory (NGT) concepts to understanding the determinants of new establishment formations and sectors across 394 spatial units (labor market areas or LMAs) with varying economic characteristics. The dependent variable was calculated over the years 1995 and 1996, with regressors generally measured in levels in 1994, or changes over the period 1992-1994. More specifically, Acs and Armington calculate new firm formation rates across LMAs as new establishments in 1995 and 1996 per 1994 worker. They were able to distinguish among six sectors and model growth as a function of firm size, sector specialization, proprietor shares, educational attainment, recent growth in income and population, and the unemployment rate within the LMA. In general, their coefficient estimates have expected signs (discussed next) and the adjusted R² values are high for these types of studies—generally higher than 63%.

In the Acs and Armington (2006) study (p. 65 ff.), a locality in which large firms (more employees per establishment) dominate has fewer new firm formations because knowledge is developed and applied within the firm, rather than being allowed to spill over into the community. Furthermore, larger firms tend to crowd out smaller, more competitive firms that may be more entrepreneurial. Counties with more specialized industries, as measured by the number of establishments per 1,000 population, on the other hand, provide greater exposure for potential entrepreneurs to different management and technical production practices, which may in turn translate into new business ideas.

Along these lines, Acs and Armington also suggest that a higher share of existing self-employed workers in the community, and fewer high school dropouts and more college graduates, are all associated with a higher rate of new establishment formation. This reflects both a more conducive existing entrepreneurial climate and more potential for innovation that builds on human capital spillovers within a locality. In addition, the (lagged, compound) population and income growth rates from 1992 to 1994 are included to control for the desirability of a community for migrants and opportunities for selling products, respectively. The unemployment rate, finally, measures the degree to which individuals are driven into self-employment by a lack of alternative work opportunities.

These authors caution that even though their independent variables are measured 1 or 2 years before the period over which establishment growth occurs, the regressors may not be strictly exogenous, and that the results therefore need to be interpreted with caution. They also note (note 11, p. 68) the exclusion of financial variables, which are important factors in new firm formation and “which [they] hope to take into account in subsequent research.” In this article, we introduce a number of candidate variables to capture the potential effect of access to financing, as explained below.

In addition, we draw on Goetz and Rupasingha (2009) for additional variables to include in the regression analysis. Chief among these are the relative financial returns to potential entrepreneurship, the riskiness of those returns, homeownership characteristics and basic banking variables as proxies for access to capital, income within the community as a measure of demand (beyond recent income growth), an ethnic fractionalization index, basic socioeconomic and demographic variables, natural amenities, and economic policy variables measured at the state level (see Appendix A). In the present study, we expand this vector by including measures of liquidity available and competition among, or availability of, bank branch offices. The Goetz and Rupasingha article is based on a utility-maximizing choice between wage-and-salary- and self-employment, reflected in relative earnings, and uses 1990 as the base year, and models the change in the self-employment rate between 1990 and 2000.

We stress that our model and analysis are based primarily on economic factors influencing self-employment growth and that other factors have been examined in the literature (e.g., Hustedde, 2007 on culture; also see Lyons, Alter, & Audretsch, 2012). Although this may seem to be a serious shortcoming at first glance, we do control for important cultural influences by including the initial share of self-employed in the workforce in 2000 as a regressor. This captures cultural influences in that year, which are likely path dependent. One related factor that we are unable to measure independently at the county level is the innate ability of individuals or the presence of entrepreneurship education programs that influence entrepreneurial outcomes (e.g., Kayne, 2007; Kutzhanova, Lyons, & Lichtenstein, 2009); again, however, these would be reflected in the initial share of self-employed at the beginning of the period over which growth is calculated. In addition, an ethnicity variable in the model helps to capture the fact that individuals from certain cultures have an inherently higher or lower propensity to engage in entrepreneurship. The same is accomplished by including the manufacturing employment share, as regions with a legacy in this sector tend to be less entrepreneurial: Because of significant economies of scale, manufacturing tends to crowd out entrepreneurship (Glaeser, Kerr, & Kerr, 2012).

It is important to note the problem of endogeneity or reverse causation in the regressors. For example, areas with more highly educated populations may provide a larger population pool of qualified self-employed workers, and at the same time the self-employed may be attracted to communities with more highly educated workers, thus causing biased parameter estimates. However, because we are using time lags (growth in self-employment occurs subsequent to the initial period during which regressors are measured), we can claim quasi-exogeneity for our results; furthermore, we are following precedents in the literature by using these time lags.

Recognizing that a given variable can belong to different groupings, we attempt to classify our variables to organize and streamline the discussion of results (see Athiyaman & Walzer, 2007, for a different classification scheme). These groups represent the following variables, as summarized in Table 2, along with the hypothesized directions of their effects. Based on this review, we posit the following categories of determinants of self-employment growth over time: (1) self-employment preconditions, (2) NGT variables (from Acs & Armington), (3) financial access variables, (4) other predetermined controls (from Goetz & Rupasingha), and (5) state-level policy shifters (also from Goetz and Rupasingha, but listed separately because they are subject to policy change). We discuss each of these in turn, along with the expected signs of the regressors.

Self-Employment Preconditions

As in Acs and Armington, the initial share of self-employed workers per wage-and-salary worker captures a more conducive climate toward entrepreneurship, as well as a higher intrinsic leaning toward such activity among the resident population; more importantly, this creates a path dependency. Perhaps most significantly, Goetz and Rupasingha find that the self-employed respond rationally to economic incentives, at least over the period 1990-2000. In their study, higher returns to self-employment and lower wage-and-salary earnings and lower earnings risks associated with self-employment all lead to higher rates of self-employment over time. In a recent extension, Low and Weiler (2012) find that the relative risk of and return to wage-and-salary employment also influence self-employment determination. The unemployment rate, for which we allow a nonlinear effect as in Goetz and Rupasingha, is associated with a greater push into self-employment as job prospects otherwise deteriorate in a county; here, we expect to observe a U-shaped effect on the dependent variable. For age, likewise, we expect a nonlinear effect because prior on-the-job experience in the labor force is important to the self-employment effort, and yet fewer older workers are likely to be willing to cope with the stresses of working for themselves (on the other hand, they could face more age discrimination from employers).

The ethnic fractionalization index is based on Alesina, Baqir, and Easterly (1999) and captures the ethnic diversity of a county. A higher index means greater diversity in terms of ethnic groups, and our calculation takes into account all of the major races reported in the U.S. Bureau of the Census classification; the index has a correlation of about 90% with African American (Black) presence in counties, and thus, in part, captures potential opportunities—or the lack thereof—for minority and otherwise underserved populations. In the entrepreneurship literature, ethnic minorities and (to a lesser degree) females have been found to prefer working for themselves to avoid being discriminated against. Finally, a higher female labor force participation rate is hypothesized to have a negative effect, as females generally have been found to be less likely to start their own firms (Kelley et al., 2012), despite the possibility of labor market limitations such as “glass ceilings.” We use these variables as proxies for the types of populations from which the self-employed are likely (or not) to emerge.

New Growth Theory Variables (From Acs & Armington)

Following Acs and Armington, we include average firm size as a regressor, with the expectation that larger firms conduct more entrepreneurial activity internally (thus internalizing spillovers), thereby crowding out self-employment; here, we also allow for nonlinearities. Educational attainment has consistently been found to influence entrepreneurship and self-employment, and here we include both the lower end (dropouts) and high achievers (college graduates, see also Figueroa-Armijos et al., 2012).

Recent research suggests that rural areas have lower-firm entry rates relative to urban areas, ceteris paribus, because of fewer economic spillovers and because lower salvage values of rural capital require higher expected profits (Yu, Orazem, & Jolly, 2011; also Johnson & Quance, 1972). Unlike Goetz and Rupasingha, who include all U.S. counties, we cannot use a rural indicator variable given that we stratify our regression analyses by (rural and urban) county types. Instead, we add population density to capture the presence of agglomeration economies, or the lack thereof. Note that these agglomeration benefits can be offset by higher factor costs associated with density or congestion, including for land and labor (Moretti, 2004). Last, following Acs and Armington (as well as Goetz and Rupasingha), we also include lagged compound growth rates in per capita income and in population in the local county, both of which are hypothesized to have positive effects. Growth in population may create new self-employment opportunities associated with non-Schumpeterian types of entrepreneurship (Goetz & Freshwater, 2001).

Financial Access Variables

The percent of homes that are owner-occupied, along with the median home value in 2000, are included to serve as basic measures of collateral availability (considering that homes are Americans’ largest source of wealth)—these were both statistically significant (and positive) in Goetz and Rupasingha. Two other variables available at the county level are the value of bank deposits in the county and the number of bank branch offices. We include measures for these from the year 2000, normalized by population. As a novel and alternative measure to bank deposits per capita (which had a sign counter to expectations in Goetz and Rupasingha), we consider dividend, rent, and interest payments (DRIPs) into the county, also per capita. Conceptually, this is a pool of funds potentially available to local businesses and we expect greater availability of local capital to be associated with more self-employment creation or proprietorship starts.

Other Predetermined Controls (From Goetz & Rupasingha)

Our socioeconomic variables (from Goetz & Rupasingha) include per capita personal income net of DRIPs and a vector of employment shares by major industry. These include construction, manufacturing, retail trade, and finance, insurance, and real estate (FIRE). Briefly, construction workers are more likely to be self-employed, whereas for communities with manufacturing dominance the opposite is often found to be the case; likewise, with the rise of big-box retail, retail workers are less likely to be self-employed. On the other hand, FIRE workers (e.g., realtors, financial advisors) are more likely to be self-employed. Again, we use these variables as sector controls or as proxies for the types of populations from which the self-employed are likely (or not) to emerge.⁴

In recent years, there has been an explosion of interest in the role of natural amenities in driving rural economic development (the seminal article is Deller, Tsai, Marcouiller, & English, 2001). Although this variable from the USDA’s Economic Research Service (McGranahan, 1999) is not amenable to policy change, it is an important control variable. In brief, higher amenities have been associated with both faster income and population growth, and also faster increases in the rate of self-employment over time, beyond the income and population growth controls already included (e.g., this captures higher-end tourism of second homes development, which would not necessarily be reflected in the two growth variables).

State-Level Policy Shifters

Last, we include state-level economic freedom measures that are explicit policy variables. As described in Appendix A, they capture the relative size of government in the state’s economy, takings and discriminatory taxation, and relative freedom in the labor market. Higher values of these variables indicate more freedom. Greater economic freedom is hypothesized to lead to higher self-employment growth rates, although a competing hypothesis suggests that the self-employed are not driven away by higher or more confiscatory taxes if as a consequence they also believe that there is a lower likelihood of taxes subsequently rising. Alternatively, as Goetz and Rupasingha suggest, higher taxes can encourage self-employment because they provide incentives for tax avoidance, which is easier when individuals work for themselves.⁵

Data Definitions, Preliminary Analysis, and Summary Statistics

Figure 1 shows the self-employment rate for the years 1969-2011 for metro and nonmetro (rural) counties; this is the self-employment number divided by contemporaneous wage-and-salary employment (se_t/ws_t). The structural break in the rate of change in the ratio in 2001 in both metro and nonmetro counties is noteworthy. Before this year, the rural self-employment rate on average increased by 0.20 percentage points annually; since 2001, it has grown at a robust pace of 0.72 percentage points annually. This pace was surpassed or matched only twice historically (over the period shown), in early 1980 and late 1980/early 1990, but then only for shorter durations of two or three consecutive years.

Figure 1.

Ratio of self- to wage-and-salary employment (se_t/ws_t), 1969-2011.

The dependent variable in our regression model is defined as the simple change in the ratio of self-employed to wage-and-salary jobs over time, as discussed above: [(se_t_+δ − se_t)/ws_t]_I, where se is the number of self-employed workers, ws the number of wage-and-salary workers, i indexes the county, t is the base year (2000), and δ is the increment or lag in time over which the change is calculated (e.g., when δ = 9, the rate of change is between the years 2000 and 2009). Of the different calculations available for expressing self-employment change, this one most closely follows that used by Acs and Armington (2006).⁶ This variable is shown for the period 2003-2011 in Figure 2.

Figure 2.

Growth in self-employment rate (Acs & Armington), 2003-2011.

This self-employment change measure differs from that used in Goetz and Rupasingha (2009), who use the change in the ratio (se_t/ws_t − se_t₋₁/ws_t₋₁) for each year and county, and that we also used in our earlier discussion as well as our graphical analysis below. The ratio used by Goetz and Rupasingha increases when se rises more rapidly than ws or when ws declines more rapidly than se. Likewise, the Acs and Armington ratio is positive as long as se_t_+δ > se_t, and it is increasing so long as (se_t_+δ − se_t) > ws_t. These two different ways of calculating the self-employment rate and its changes produce similar results, albeit on different scales.

The relatively sharp increase in the self-employment rate, or the change in self-employment relative to 2000 wage-and-salary employment over the years 2000-2009, did not occur evenly across rural America. In this context, a variable useful for stratifying nonmetro counties is the RUCC, measured in 2003 (RUCC-03).⁷ The definitions of these different counties are provided in Table 1 and basically capture two dimensions: population size and concentration (urban population of over 20,000; urban population of between 2,500 and 20,000; and no urbanized populations) and adjacency or nonadjacency to metro areas. Other studies have shown that adjacency to metro areas can convey important benefits to a rural county (Partridge & Rickman, 2008), including access to labor markets and services. There are about 48 million rural residents, a number that is close to the total population of senior citizens in this country. The other three county types (1-3) are metro counties with successively smaller population totals.

Table 1.

Selected 2000 Population Statistics by RUCC.^a

RUCC-03	Total population	Mean population	Population per square mile	Counties (n)	Definition
1	148,023,564	374,743.2	1,256.0	395	Metro, population 1 million or more
2	54,357,459	169,867.1	255.8	320	Metro, population 1 million to 250,000
3	27,035,638	81,188.1	131.1	333	Metro, population less than 250,000
4	14,259,827	66,947.5	101.1	213	Urban, population 20,000+ adjacent
5	5,207,328	51,557.7	64.5	101	Urban, population 20,000+ not adjacent
6	14,997,680	25,121.7	41.9	597	Urban, population 2,500-19,999 adjacent
7	8,248,109	18,874.4	30.8	437	Urban, population 2,500-19,999 not adjacent
8	2,405,935	10,415.3	21.7	231	Completely rural, adjacent
9	2,741,321	6,637.6	12.6	413	Completely rural, not adjacent
All	282,171,957	90,730.5	244.9	3,040

RUCC is the rural–urban continuum code provided by the Economic Research Service, U.S. Department of Agriculture. We interchangeably refer to metro counties as urban and all nonmetro counties as rural. In the text, the following abbreviations are used: 1 = large urban; 2 = medium urban; 3 = small urban; 4 = large rural, adjacent; 5 = large rural, not adjacent; 6 = medium rural, adjacent; 7 = medium rural, not adjacent; 8 = small rural, adjacent; 9 = small rural, not adjacent.

Source. Authors’ calculations using U.S. Census data.

Figure 3 shows that 2000 self-employment shares decline smoothly across the first five county types (from 1 to 5) and then increase for the smaller rural counties, with the respective adjacent county showing a higher share (e.g., 4 vs. 5). At the same time, and Figures 4 and 5 show that larger increases in self-employment rates occurred in the metro-adjacent rural counties, across the three different size classifications, using both the Acs and Armington and Goetz and Rupasingha measures of self-employment. This likely reflects greater access to suppliers—and market outlets—for self-employed workers as they responded to the economic challenges of the 2000-09 period by switching to self-employment.

Figure 3.

Self-employment ratio in 2000 by rural–urban continuum code (RUCC-03).

Figure 4.

Change in self-employment ratio (Goetz & Rupasingha) 2000-2009 by rural–urban continuum code (RUCC-03).

Figure 5.

Change in Self-Employment Ratio (Acs & Armington) 2000-2009 by rural–urban continuum code (RUCC-03).

The basic variables used in our regressions are defined in Table 2, which also provides hypothesized directions of the effects of variables and summary statistics. We provide separate statistics for all rural and urban counties for which valid data are available across these variables. The data are compiled from two primary sources, USA Counties (http://censtats.census.gov/usa/usa.shtml) for basic Census Data and the Bureau of Economic Analysis’ (BEA) Regional Economic Information System (REIS) for data on self-employment numbers, total full- and part-time employment, and dividends, rents and interest payments, and intercensal year population estimates. Furthermore, both the RUCC-03 code and the Amenities Index are from the USDA’s Economic Research Service, and the state-level policy measures are from the www.freetheworld.com website.

Table 2.

Variables, Expected Signs, Definitions, and Summary Statistics: Rural and Urban.

			Rural		Urban
Variable	E [b]	Definition	M	SD	M	SD
Dependent variables, Acs and Armington definition (see text)
ACSSHG03	+	Self-employment change, 2000-2003	0.017	0.063	0.030	0.090
ACSSHG04	+	Self-employment change, 2000-2004	0.024	0.066	0.045	0.098
ACSSHG05	+	etc.	0.037	0.074	0.062	0.110
ACSSHG06	+	etc.	0.046	0.078	0.075	0.120
ACSSHG07	+	etc.	0.060	0.087	0.093	0.132
ACSSHG08	+	etc.	0.061	0.089	0.094	0.134
ACSSHG09	+	etc.	0.066	0.095	0.097	0.137
ACSSHG10	+	etc.	0.099	0.172	0.105	0.157
ACSSHG11	+	Self-employment change, 2000-2011	0.106	0.181	0.110	0.161
Independent variables
1. Self-employment preconditions
SHARE00	+	Self/wage-and-salary employed	0.197	0.064	0.185	0.071
NFPINC00^a	+	Self-employment earnings ($/self-employed)	17.6	5.5	22.3	10.7
WSE2000^a	−	Wage-and-salary earnings ($/worker)	23.2	3.8	28.6	6.6
CV9605	−	Coefficient of variation of self-employment earnings, 1995-2000	0.163	0.080	0.168	0.080
URATE00	−,+	Unemployment rate (%)	6.014	2.891	5.266	1.986
MEDAGE00	+,−	Median age of population (years)	38.2	4.0	35.8	3.3
ETHNIC00	+	Ethnic diversity index (see text)	0.196	0.169	0.258	0.172
FLF00	−	Female labor force participation (%)	45.8	2.3	46.5	2.0
2. New growth theory variables (Acs & Armington)
ESTSZE00	−,+	Firm size (workers/establishment)	11.0	4.3	14.1	5.1
DROPOUT	−	Adults without high school degree (%)	24.1	8.9	20.2	9.2
EDU68500	+	Adults with college degree or + (%)	14.4	5.7	20.4	9.3
POPDEN	+	Population density (population/square mile)	38.3	39.7	593.5	2689.2
CMPDINCO	+	Compound per cap. inc. growth, 1995-2000	1.049	0.018	1.050	0.013
CMPDPOPU	+	Compound population growth, 1995-2000	1.004	0.013	1.014	0.015
3. Financial access variables
PCOWNOCC	+	Owner-occupied homes (%)	75.1	6.0	72.5	9.3
MEDHVALU^b	+	Median home value ($10,000)	7.0	3.6	10.9	5.5
DRIPPC00	+	Dividend, rent, interest ($/capita)	4.479	1.975	4.790	2.012
BNCHPC00	+	Bank branches (no./1,000)	0.535	0.305	0.346	0.168
DEPOPC00	+	Bank deposits ($/capita)	12.4	7.1	11.3	9.1
4. Other predetermined controls (Goetz & Rupasingha)
INCWODRP^a	+	Income without DRIPS ($/capita)	21.9	4.3	26.9	6.9
CONS00	+	Employment share, construction (%)	7.6	2.4	7.9	2.3
MANU00	−	Employment share, manufacturing (%)	15.9	9.9	15.8	7.2
RTRADE00	−	Employment share in retail trade (%)	11.2	2.2	11.9	1.6
FIRE00	+	Employment share in finance, insurance, real estate (%)	3.890	1.210	5.861	2.182
AMNSCALE	+	Amenity scale (see text)	−0.06	2.252	0.265	2.360
5. State policy controls
AREA1	+	Government size (see text), 2002	8.004	0.785	8.036	0.743
AREA2	+	Takings and discriminatory taxation, 2002	7.021	0.511	7.090	0.521
AREA3	+	Labor market freedom, 2002	7.056	0.681	7.156	0.629

Note. Data are measured in 2000 except where indicated.

Source: Authors.

Scaled (divided) by 1,000.

Scaled by 10,000.

The self-employment numbers are developed by the BEA from Federal Tax Schedule C, Form 1040 filings, and include unincorporated workers who work for themselves. They are also referred to as nonfarm proprietor(ship)s. The numbers include full- and part-time workers and it is possible that the same worker files multiple schedules, and a worker may be both self-employed and also work for someone else on a wage payroll (and therefore be counted in the ES 202 Unemployment Insurance series). In addition, underreporting of self-employment is likely to be widespread. Thus, these numbers are not without problems and they need to be viewed with caution.

Estimation Strategy and Results

We start with two regressions that approximate the Acs and Armington and Goetz and Rupasingha studies; our results are different from those obtained in these earlier works, but considering differences in time periods, geographic units, and measures of the dependent variables (compared with Acs & Armington), this is not unexpected. These results are reported in Appendix B. Our primary regressions draw on both these earlier articles. In the case of Acs and Armington, we dropped the Industry Specialization variable because it consistently produced the unexpected sign. Instead, we substituted a vector of employment shares by industry, thus following the Goetz and Rupasingha paper more closely. From the Acs and Armington article, we retained the compound growth rates in population and income leading up to the growth period, as well as the educational attainment measures and the initial self-employment share and the unemployment rate. The remaining regressors are based on Goetz and Rupasingha, except that we add additional financial access proxies.

Results for the 2000-2009 Time Period

Table 3 shows core regression results for changes in self-employment between 2000 and 2003 as well as the six ensuing years, relative to 2000 wage-and-salary employment, as a function of 2000 baseline regressors for rural areas (Table 4 has results for urban areas). Recognizing the limitations of such an analysis, we propose that the results in Table 3 potentially reveal changes in the effects of the regressors over the business cycle and, more specifically, in the period leading up to and including the housing collapse that started in late 2007. Effects of most of the variables are robust over this period, regardless of the period chosen (i.e., δ). We report standardized estimates to avoid scaling issues in interpreting the parameter estimates. In this manner, we are able to assess the effects of the baseline year regressors as the first decade of the new century progressed. Of course these baseline conditions also changed as the years went by and we do not capture these changes in our analysis.⁸ Even so, the gradual increase in the adjusted R² value as more time lapses since 2000 is noteworthy.

Table 3.

Rural Results, Standardized Beta Coefficients (β).

Dependent variable is growth in self-employment
Variable	2000-2003	2000-2004	2000-2005	2000-2006	2000-2007	2000-2008	2000-2009
1. Self-employment preconditions
SHARE00	0.089*	0.157*	0.241*	0.287*	0.341*	0.356*	0.393*
NFPINC00	0.056*	0.048*	0.036^†	0.031^†	0.023	0.036^†	0.055*
WSE2000	0.041	0.041	0.033	0.022^†	0.019	0.035	0.055*
CV9605	−0.095*	−0.120*	−0.108*	−0.098*	−0.085*	−0.075*	−0.059*
URATE00	−0.168*	−0.141*	−0.154*	−0.155*	−0.169*	−0.192*	−0.211*
URATESQ	0.142*	0.119*	0.125*	0.125*	0.133*	0.154*	0.174*
MEDAGE00	0.502^†	0.524^†	0.448^†	0.400^†	0.366^†	0.408	0.470^†
MEDAGESQ	−0.482^†	−0.529^†	−0.452^†	−0.437^†	−0.410^†	−0.464^†	−0.537*
ETHNIC00	0.096*	0.107*	0.119	0.137*	0.150*	0.171*	0.182*
FLF00	−0.086*	−0.046^†	−0.039	−0.040	−0.039	−0.057*	−0.079*
2. New growth theory variables (Acs & Armington)
ESTSZE00	−0.291*	−0.212*	−0.188*	−0.177^†	−0.171^†	−0.165^†	−0.145^†
ESTSZESQ	0.193*	0.153^†	0.140^†	0.136^†	0.133^†	0.127^†	0.114^†
DROPOUT	0.057^†	0.090*	0.089*	0.085*	0.090*	0.092*	0.087*
EDU68500	0.138*	0.140*	0.126*	0.130*	0.135*	0.161*	0.178*
POPDEN	0.087*	0.081*	0.081*	0.082*	0.079*	0.081*	0.081*
CMPDINCO	0.106*	0.121*	0.125*	0.121*	0.119*	0.108*	0.093*
CMPDPOPU	0.146*	0.178*	0.192*	0.197*	0.198*	0.188*	0.168*
3. Financial access variables
PCOWNOCC	−0.156*	−0.163*	−0.161*	−0.148*	−0.142*	−0.099*	−0.047
MEDHVALU	−0.609*	−0.771*	−0.793*	−0.827*	−0.808*	−0.713*	−0.556*
OWNOCCMV	0.604*	0.768*	0.796*	0.826*	0.806*	0.649*	0.429*
DRIPPC00	0.082	0.093^†	0.086	0.096^†	0.083	0.110^†	0.129*
BNCHPC00	0.050	0.045	0.051	0.050	0.050	0.037	0.025
DEPOPC00	−0.039	−0.041	−0.047	−0.041	−0.032	−0.027	−0.019
4. Other predetermined controls (Goetz & Rupasingha)
INCWODRP	−0.190*	−0.159*	−0.145*	−0.139*	−0.122*	−0.116*	−0.114*
CONS00	0.023	0.045	0.045	0.052^†	0.056^†	0.046^†	0.033
MANU00	−0.018	−0.022	−0.024	−0.020	−0.006	−0.030	−0.052^†
RTRADE00	−0.100*	−0.086*	−0.084*	−0.075*	−0.072*	−0.076*	−0.077*
FIRE00	−0.088*	−0.071*	−0.072*	−0.067*	−0.060*	−0.050*	−0.034^†
AMNSCALE	−0.032	0.000	0.004	−0.001	0.010	0.013	0.016
5. State policy controls
AREA1	0.032	0.045	0.048	0.021	−0.003	−0.018	−0.025
AREA2	−0.042^†	−0.044^†	−0.048^†	−0.042^†	−0.040^†	−0.022	−0.013
AREA3	0.058*	0.071*	0.080*	0.083*	0.099*	0.102*	0.094*
Adjusted R²	0.120	0.167	0.215	0.242	0.283	0.280	0.282

Source. Authors.

†

Significant in a one-tailed test. *Significant at 10% or lower.

Table 4.

Urban Results: Standardized Beta Coefficients (β).

Dependent variable is growth in self-employment
Variable	2000-2003	2000-2004	2000-2005	2000-2006	2000-2007	2000-2008	2000-2009
1. Self-employment preconditions
SHARE00	0.250*	0.304*	0.364*	0.411*	0.464*	0.465*	0.478*
NFPINC00	0.019	0.021	0.019	0.025	0.022	0.022	0.026
WSE2000	0.012	0.002	−0.002	−0.002	−0.005	−0.001	0.004
CV9605	0.018	0.001	0.004	0.011	0.014	0.016	0.018
URATE00	0.036	0.015	−0.001	−0.006	−0.041	−0.040	−0.044
URATESQ	−0.113	−0.099	−0.087	−0.085	−0.051	−0.055	−0.053
MEDAGE00	−0.905*	−0.899*	−0.970*	−1.047*	−1.089*	−1.079*	−1.051*
MEDAGESQ	0.871*	0.876*	0.945*	1.006*	1.054*	1.041*	1.014*
ETHNIC00	0.112^†	0.129*	0.132*	0.151*	0.173*	0.184**	0.193*
FLF00	0.055	0.072	0.072	0.066	0.064	0.051	0.035
2. New growth theory variables (Acs & Armington)
ESTSZE00	−0.128	−0.107	−0.099	−0.071	−0.048	−0.052	−0.056
ESTSZESQ	0.059	0.052	0.049	0.036	0.027	0.030	0.033
DROPOUT	0.127^†	0.142*	0.145*	0.140*	0.133*	0.123*	0.106^†
EDU68500	−0.075	−0.045	−0.049	−0.035	−0.013	−0.011	−0.012
POPDEN	0.047	0.058^†	0.065*	0.073*	0.077*	0.079*	0.077*
CMPDINCO	−0.020	−0.021	−0.026	−0.033	−0.032	−0.034	−0.033
CMPDPOPU	0.188*	0.220*	0.233*	0.224*	0.220*	0.215*	0.208*
3. Financial access variables
PCOWNOCC	−0.028	−0.018	−0.020	0.002	0.002	0.016	0.030
MEDHVALU	−0.175*	−0.193*	−0.210*	−0.212*	−0.217*	−0.223*	−0.223*
OWNOCCMV	0.004	0.019	0.036	0.033	0.042	0.025	0.000
DRIPPC00	0.002	−0.008	−0.015	−0.012	−0.018	−0.018	−0.012
BNCHPC00	−0.107	−0.115*	−0.114*	−0.104*	−0.094^†	−0.099*	−0.100*
DEPOPC00	−0.030	−0.026	−0.022	−0.022	−0.021	−0.019	−0.018
4. Other predetermined controls (Gotez & Rupasingha)
INCWODRP	0.277*	0.277*	0.282*	0.270*	0.259*	0.274*	0.275*
CONS00	0.001	0.017	0.023	0.035	0.048	0.030	0.007
MANU00	0.012	0.018	0.015	0.012	0.015	−0.003	−0.020
RTRADE00	0.005	0.020	0.023	0.028	0.029	0.027	0.023
FIRE00	−0.065	−0.064	−0.063	−0.059	−0.055	−0.056	−0.056
AMNSCALE	−0.135*	−0.120*	−0.114*	−0.115*	−0.108*	−0.099*	−0.085*
5. State policy controls
AREA1	−0.026	−0.017	−0.009	−0.018	−0.035	−0.035	−0.033
AREA2	−0.077*	−0.080*	−0.080*	−0.077*	−0.065*	−0.057*	−0.048
AREA3	0.135*	0.132*	0.129*	0.130*	0.133*	0.131*	0.122*
Adjusted R²	0.132	0.191	0.246	0.279	0.322	0.318	0.319

Source. Author.

†

Significant in a one-tailed test. *Significant at 10% or lower.

Focusing on the last column in Table 3 (2000-2009), it is apparent that the block of variables representing self-employment preconditions performs very well, as does the block of NGT variables from Acs and Armington. The initial self-employment share has the highest standardized beta coefficients (except for median housing value), and the other preconditions variables generally perform as hypothesized. In particular, greater earnings risk and female labor force participation both reduce self-employment, whereas higher returns to self-employment and greater minority shares are associated with more such employment.

For the unemployment rate, we also follow Goetz and Rupasingha and include a quadratic term that yields a U-shaped effect: A rising unemployment rate initially lowers self-employment, but beyond a threshold it leads to higher self-employment as workers give up on finding wage-and-salary jobs. Our median age measure is statistically different from zero only when we allow for a quadratic effect (and then only in a two-tailed test, as reported in Table 3).

This suggests an increasing and then declining effect of age; that is, an inverse U-shaped effect whereby self-employment initially rises but then tapers off with more experience of the workforce.

As already noted, the NGT variables are remarkably robust and most have the expected signs. The establishment size variable is negative but not statistically different from zero when we use only the linear term. A sensitivity test, however, shows that including a squared term yields a U-shaped effect on relative self-employment changes, and both terms are statistically significant.⁹ As establishment size increases, the self-employment growth rate initially drops as more innovation occurs within firms and self-employment is crowded out, but eventually this kind of activity expands as firms become larger, and perhaps even contract out opportunities to the self-employed. Both compound income and population growth have the expected positive signs, as does the college-educated population share. Note that the positive effect of high school dropouts is counter to expectations but consistent with the findings of Acs and Armington. Thus, with the exception of the sector specialization variable (see note) and the nuance of a nonlinear relationship with establishment size, our results support those obtained by Acs and Armington even though our dependent variable is different, as is our time period of analysis. The beta coefficient estimates for the self-employment preconditions and NGT variables are plotted in Figure 6A and B over time.

Figure 6.

Key regression results for rural areas: (A) Part 1 and (B) part 2.

The financial access variables generally provide mixed results. The direct (uninteracted) effects of our two housing-related variables are consistently negative, in contrast to the Goetz and Rupasingha findings. However, when we allow for an interaction term (reported in Table 3), that term is positive and statistically different from zero. Thus, having both a higher share of owner-occupied homes and higher median home values in a county appears to increase access to capital in support of self-employment, and this effect was weakened only slightly in the Great Recession of 2008. Dividend, rents, and interest payments for the most part also are statistically significant but at a weaker level (two-tailed test), except in the last year shown where the effect is different from zero even in a one-tailed test.

For the other predetermined controls (from Goetz & Rupasingha), income per capita has an unexpected negative effect, while retail trade and FIRE have statistically significant negative effects over the entire period. Whereas the effect is not unexpected for retail trade (with the shift to big box and other chain stores driving out mom-and-pop type retail businesses), the effect of the finance, insurance, and real estate sector leading up to the housing bubble is perhaps counter to expectations. Although we are controlling for population growth in the years 1995-2000, we would have expected more self-employment increases in the FIRE sector as the housing sector expanded across the nation, even in rural areas. On the other hand, it is plausible that these kinds of workers entered into employment contracts with new firms started in the run-up to the housing bubble of 2008, rather than working for themselves.¹⁰ For construction, there is a statistically weaker effect in 2008 and immediately before and after the crash. For manufacturing, the effect is negative throughout the period, as expected, but (weakly) significant only in the last year reported.

Among state policy controls, greater labor market freedom is associated with more self-employment activity, conceivably because it is easier to hire workers. The negative sign on takings and discriminatory taxation is not entirely unexpected: Less freedom on this variable can drive individuals into self-employment because such activity is easier to shield from tax authorities than wage-and-salary earnings (Goetz and Rupasingha obtain the same negative effect for this variable, and it has been documented elsewhere).

Table 4 presents comparable regressions for the metro areas. In general, even though the adjusted R² variables are slightly higher than for the rural regressions, these equations do not perform as well. Among the results that stand out are the importance of initial self-employment shares as well as the median age and ethnic variables. For median age, however, the effect is opposite to that expected: declining self-employment rates with rising age of the county population and an eventual increase beyond a turning point. The effect of wages, as an opportunity cost to self-employment, is negative in four of the seven periods examined, but these do not rise to the level of statistical significance.

For the NGT variables, the statistical significance of the dropout rate, population density, and compound population growth in urban areas stand out in important ways. Median home values and availability of bank branch offices have unexpected negative effects. The interaction effect between homeownership rate and median home value is not statistically significant. It is not clear how much should be read into the result for the number of branch offices, but it may be worthy of further investigation. For example, did bank branch offices “push” out loans to questionable self-employed borrowers in the years leading to the peak of the housing bubble? On the other hand, income per capita has the expected positive effect (whereas lagged compound income growth does not), while amenities have an unexpectedly significant negative effect. Two of the three state policies exert statistically strong influences on self-employment growth, with the same signs as in rural counties. Figure 7 shows standardized coefficient estimates for key variables in urban areas.

Figure 7.

Key regression results for urban areas.

Results by Rural–Urban Continuum Code

As noted, a key variable of interest, although it does not appear among the regressors, is the RUCC03. As part of the analysis, we estimated separate regressions for each of the RUCC codes to assess whether or not different policies may be needed for different types of rural counties across the nation for the period 2003-2009. Results are reported in Tables 5 and 6 and provide insights into the independent effects of proximity and population size. As noted earlier, the RUCC code captures both population size (declining with higher codes) and metro-adjacency, with the odd number codes (5, 7, and 9) designating nonadjacent and possibly more remote, less accessible counties. A Chow test indicates that these RUCC models are structurally different so that the various processes generating self-employment vary across the county types.¹¹ In the remaining discussion we refer to Code 1 counties as large urban, Code 2 as medium urban, and Code 3 as small urban. Code 4 and 5 counties are large rural adjacent and nonadjacent, respectively. Codes 6 and 7 refer to medium-sized rural adjacent and nonadjacent. Finally, Code 8 and 9 counties are adjacent and nonadjacent small rural.

Table 5.

Detailed Results by Rural–Urban Continuum Code for 2003-2009, Standardized Beta Coefficients (β).

Dependent variable is growth in self-employment
Variable	1 Large urban	2 Medium urban	3 Small urban	4 Large rural adjacent	5 Large rural not adjacent	6 Medium rural adjacent	7 Medium rural not adjacent	8 Small rural adjacent	9 Small rural not adjacent
1. Self-employment preconditions
SHARE00	0.470*	0.610*	0.416*	0.409*	0.668*	0.524*	0.290*	0.620*	0.148*
NFPINC00	0.039	0.111*	0.006	0.064	0.062	0.130*	0.070*	0.101^†	0.070^†
WSE2000	−0.114^†	0.096	0.098	−0.063	−0.131	0.113*	0.021	0.084	0.119*
CV9605	−0.012	−0.064^†	0.070	−0.064	−0.035	0.001	−0.327*	0.195*	−0.097^†
URATE00	−0.204	0.076	0.185	0.401*	−0.503	−0.001	−0.266*	−0.912*	−0.236
URATESQ	0.039	−0.132	−0.326	−0.561*	0.588*	−0.023	0.291*	0.936*	0.188
MEDAGE00	−1.586	−0.787*	−0.786*	−1.900*	2.913*	0.142	2.149*	2.046*	−0.466
MEDAGESQ	1.673	0.702*	0.632*	1.704*	−3.168*	−0.341	−2.340*	−2.028*	0.403
ETHNIC00	0.353*	0.262*	−0.014	0.242*	0.262*	0.245*	0.190*	0.159*	0.078^†
FLF00	−0.016	0.032	0.065	0.051	0.047	0.034	−0.008	0.078	−0.134*
2. New growth theory variables (Acs & Armington)
ESTSZE00	−0.119	0.317	−0.296	0.367	−0.164	−0.142	−0.318*	0.084	−0.296
ESTSZESQ	0.103	−0.210	0.140	−0.289	−0.177	0.214	0.260*	−0.119	0.164
DROPOUT	−0.070	0.103	0.252*	0.211*	0.167^†	0.076	0.178*	0.241*	−0.049
EDU68500	−0.177	0.219*	−0.185*	0.043	0.274*	0.050	0.227*	0.180*	0.194*
POPDEN	0.141*	−0.015	0.025	−0.051	0.194*	0.002	0.012	−0.045	0.172*
CMPDINCO	−0.082	0.129*	−0.061	0.135*	0.077	0.131*	0.027	−0.069	0.029
CMPDPOPU	0.232*	0.041	0.360*	0.259*	0.254*	0.191*	0.091^†	0.210	0.015
3. Financial access variables
PCOWNOCC	−0.115	−0.084	−0.066	−0.069	0.262	0.133	0.034	−0.070	−0.055
MEDHVALU	−0.391*	−1.059*	−0.119	−0.648	−0.832	0.414	−1.622*	−1.639*	−0.489*
OWNOCCMV	0.039	0.906*	−0.059	0.718	0.515	−0.491	1.429*	1.268*	0.316
DRIPPC00	−0.173*	0.106	0.112	0.152	0.025	0.069	0.475*	−0.127	0.073
BNCHPC00	0.051	−0.067	−0.203*	−0.122*	0.111^†	−0.143*	−0.105	0.177*	0.076
DEPOPC00	0.011	0.007	−0.131*	0.033	−0.135	0.188	0.091^†	−0.244*	−0.162*
4. Other predetermined controls (Goetz & Rupasingha)
INCWODRP	0.666*	−0.139	0.230^†	0.102	0.551*	−0.027	−0.275*	0.362*	−0.081
CONS00	−0.019	0.145^†	−0.104	0.136*	−0.295*	0.006	0.068	0.005	0.074
MANU00	0.008	0.021	−0.072	0.032	0.050	−0.107^†	−0.073^†	0.054	−0.009
RTRADE00	0.030	0.014	0.000	0.029	0.042	−0.028	−0.037	0.026	−0.172*
FIRE00	−0.148*	−0.002	0.030	0.091	−0.057	−0.013	−0.041	0.049	0.027
AMNSCALE	−0.068	−0.095	−0.129*	−0.216*	0.017	0.050	0.025	0.216*	0.185*
5. State policy controls
AREA1	−0.089	−0.037	0.005	−0.063	−0.233*	0.010	−0.138*	0.000	−0.026
AREA2	0.007	−0.047	−0.141*	−0.104^†	0.002	−0.090*	0.154*	−0.060	−0.079
AREA3	0.108*	0.125*	0.125*	0.161*	0.228*	0.180*	0.106*	0.059^†	0.076^†
Adjusted R²	0.253	0.394	0.408	0.537	0.571	0.362	0.435	0.452	0.248

Source. Authors.

†

Significant in a one-tailed test. *Significant at 10% or lower.

Table 6.

Summary of Key Variables Affecting Self-Employment by Category of Determinant and Rural–Urban Continuum Code.

	Category of self-employment determinant
	1. Self-employment preconditions	2. New growth theory (Acs & Armington)	3. Financial access variables	4. Other predetermined controls (Goetz & Rupasingha)	5. State policy controls
Code 1Large urban	Initial share wage-and-salary earningsEthnicity	Population densityPopulation growth	Home value DRIP payments	Income/capitaFIRE jobs	Labor market freedom
Code 2Medium urban	Initial share“Self-employment earnings”RiskinessExperienceEthnicity	College graduatesIncome growth	Home value	Construction	Labor market freedom
Code 3Small urban	Initial shareExperience	High school dropoutsCollege graduatesPopulation growth	Bank branches Bank deposits	Income/capitaAmenities	TaxationLabor market freedom
Code 4Large rural, adjacent	Initial shareUnemploymentExperienceEthnicity	High school dropoutsIncome growthPopulation growth	Bank branches	ConstructionAmenities	TaxationLabor market freedom
Code 5Large rural, not adjacent	Initial shareExperienceEthnicity	High school dropoutsCollege graduatesPopulation densityPopulation growth	Bank branches	IncomeConstruction	Smaller governmentLabor market freedom
Code 6Medium rural, adjacent	Initial share “Self-employment earnings”Wage-and-salary earningsEthnicity	Income growthPopulation growth	Bank branches	Manufacturing	TaxationLabor market freedom
Code 7Medium rural, not adjacent	Initial share “Self-employment earnings”RiskinessUnemploymentExperienceEthnicity	Firm sizeHigh school dropoutsCollege graduatesPopulation growth	Home valueDRIPS	Income/capitaManufacturing	Smaller governmentTaxationLabor market freedom
Code 8Small rural, adjacent	Initial share “Self-employment earnings”RiskinessUnemploymentExperienceEthnicity	High school dropoutsCollege graduatesPopulation growth	Home valueBank branchesBank deposits	Income/capitaAmenities	Labor market freedom
Code 9Small rural, not adjacent	Initial share “Self-employment earnings”Wage-and-salary earningsRiskinessFemales in labor force	College graduatesPopulation growth	Home value Bank deposits	Retail tradeAmenities	Labor market freedom

Note. Italics indicate the effect is in a direction opposite to that expected. DRIPS = dividend, rent, and interest payments; FIRE = finance, insurance, and real estate.

Source. Authors (Table 5).

A first important result that again stands out in Table 5 is the strong, positive effect of the initial share of self-employed, as well as the ethnicity variable—even though the effect of the latter declines gradually over space and is nonexistent in small urban counties (Code 3 counties, with metro populations of less than 250,000; see also Figure 8A). Regardless of where a county is located on the rural–urban continuum, prior self-employment history or path dependence and ethnic composition matter in important ways. The effect of self-employment returns is statistically significant in just over half of the county types, while wage-and-salary earnings have the expected negative effect only in large urban counties with populations of one million or more, and the effect is weak statistically. At the same time, the effect of wage-and-salary earnings is positive and robust in adjacent medium rural counties (urban population of 2,500-19,999, metro-adjacent) and in small nonadjacent rural counties, which could indicate entrepreneurship of opportunity rather than necessity in these two county types.¹² The riskiness of self-employment has the biggest negative effect in the medium and small rural counties that are nonadjacent (Codes 7 and 9), and the sign is actually in the wrong direction for counties that are small rural and adjacent (Code 8): Here, greater risk attracts more self-employed. This may be related to commodity-driven boom-bust cycles rather than irrational decision making.

Figure 8.

Key regression results for different rural–urban continuum codes (RUCCs), 2003-2009: (A) Part 1 and (B) part 2.

A higher unemployment rate is associated with eventually declining self-employment only in large rural adjacent counties (urban population of 20,000 and adjacent, or Code 4) in the 2003-2009 period. Here the fact that the self-employed lose unemployment benefits once they explore opportunities to work for themselves may reduce the self-employment rate as more individuals qualify for unemployment benefits. This effect would be worth exploring in future research because it has potentially important policy implications.

For median age and median age squared, the flipping of signs for large rural metro-adjacent counties (Code 4 with urban populations of 20,000 or more and adjacent) and those that are not adjacent (Code 5 with urban populations of 20,000 or more, not adjacent) is noteworthy: We obtain a U shape for adjacent counties and an inverse U for those not adjacent in the case of the larger urbanized nonmetro counties. Youth entrepreneurship educational programs (see Schroeder, 2007, for an example of youth engagement systems) may be especially effective in counties with metro-adjacent cities, where the tendency is for self-employment to pick up only among older residents while younger ones shy away from such activity. Similarly, the shape of the unemployment rate effect also flips between large rural adjacent counties (Code 4 with urban population of 20,000 or more and adjacent) and medium-sized rural nonadjacent counties (Code 7, urban population of 2,500-19,999, not adjacent), and the small rural adjacent counties (Code 8). The ethnicity variable has the expected positive effect in all but one of the county types, as already noted: small urban counties (Code 3). For the other self-employment preconditions variables, the effects are at best mixed, with the shape of the relationship with respect to unemployment and age reversing itself across the county types.

Among the NGT variables, the effect of education is generally consistent with the results reported for urban and rural areas (for 2003-2009) at the lower end (high school dropouts), while a larger college-educated population share actually deters self-employment in small metros (see Figure 8B; the effect is also negative in large metros, but not statistically significant). This could reflect the presence of college or university campuses where wage-and-salary employment tends to be more common. The finding may also be related to that of the ethnicity variable in this county type. It is similarly noteworthy that having a college-educated population is important for self-employment growth in the nonadjacent rural county types (Codes 5, 7, and 9) regardless of size, but not in those that are adjacent to metro counties (Codes 4 and 6), although the Code 8 counties are an exception here. The status of being located next to a metro area may lessen the need to have college graduates to stimulate self-employment growth (although this is not true for small rural adjacent or Code 8 counties among the adjacent set). Perhaps this is so because college grads can more readily be hired from metro areas in the nonmetro adjacent counties.

The population density variable, which has a positive effect across rural and urban areas consistent with the agglomeration literature, was found to matter only in large metros, large rural nonadjacent, and small rural nonadjacent counties (Codes 1, 5, and 9). Perhaps remarkably, this important variable matters at the very extremes of the RUCC03 and exactly in the middle. Lagged compound population growth again generally had the expected positive effect, likely capturing self-employment of opportunity. Especially in large- and medium-sized rural counties (Codes 4 and 6), the positive effect likely reflects spillover growth from adjacent metro areas.

Among the financial access variables, positive effects of the DRIP payments are observed in medium-sized rural counties that are not adjacent (Code 7), and the number of bank branches per capita has a positive effect in large rural nonadjacent counties (Code 5 counties with possibly thinner financial markets) and in small rural adjacent (Code 8) counties. In contrast, in small urban counties (Code 3) and large- and medium-sized metro-adjacent rural counties (Codes 4 and 6 counties), higher levels of bank branches per capita actually deter self-employment formation according to these results; it is not clear what role the metro-adjacency status of large and medium rural counties (Codes 4 and 6) may play in this context Table 5?

For the predetermined controls from Goetz and Rupasingha, income per capita without DRIPs matters positively in large urban counties (Code 1), large rural nonadjacent counties (Code 5), and in small adjacent rural counties (Code 8), but negatively in medium-sized rural counties that are not adjacent (Code 7). The share of workers in construction has opposite effects in the two large rural county types (Codes 4 and 5) depending on adjacency status, possibly reflecting relative demands for new housing and other construction in these two county types. Higher shares of manufacturing employment have the expected negative effects in both the adjacent and nonadjacent medium-sized rural counties (Code 6 and 7 counties); that is, in the smaller nonmetros regardless of adjacency status. For FIRE, a statistically significant effect is detected only in large urban counties (Code 1), and the effect is unexpectedly negative.

And finally, amenities matter in a positive manner only in the smallest rural counties regardless of adjacency status (Codes 8 and 9), and they do not matter in other rural counties except that they depress self-employment growth in large metro-adjacent rural counties (Code 4), which is similar to their effect in small urban counties. The effect of state policies is again similar to that found in the separate rural–urban regressions over time, although the flipping of signs for policy area 2 (takings and discriminatory taxation) between small medium-sized rural counties that are and are not adjacent (Code 6 and 7 counties) stands out. Perhaps even more significant is the finding that each of the three state government policy types considered here has a statistically significant effect in the medium-sized rural counties that are nonadjacent to metro areas (Code 7). Here, the results indicate that a relatively smaller role of government in the local economy (in terms of higher consumption expenditures, transfers, and subsidies and social security payments relative to GDP, at the state level), is associated with less growth in self-employment over time. On the other hand, lower levels of takings and discriminatory taxation are associated with more self-employment growth and greater labor market freedom (less restrictive minimum wages, less union activity, and proportionally less government employment) are both associated with more self-employment growth. The latter result holds across each of the county types across the rural-urban continuum. Higher government taxation is associated with more self-employment growth in small urban counties, large rural adjacent, and medium rural adjacent counties (Codes 3, 4 and 6). In the literature this has been explained by a lower expectation of future tax increases, which in turn encourages business formation.

Conclusion

Our results confirm that all rural and urban county types are not the same and that different types of policy interventions are required depending on the type, if the goal is to more uniformly increase self-employment activity across the nation. For example, relying on or attracting more highly educated populations to induce self-employment may work as a strategy in metro nonadjacent counties, but not in those located next to urban labor markets. On the other hand, increasing the number of bank branches per capita or finding other ways of increasing capital availability would appear to be an effective strategy only in the smallest metro-adjacent rural counties (Code 8). Otherwise, lack of access to capital does not appear consistently to constrain expanded self-employment. More generally, our results underscore the importance of the effect of metro adjacency on information search and related transactions costs in accessing markets.

Much has been made recently within the economics literature of the importance of density and agglomeration benefits. In this context, it is noteworthy that the detailed regressions by RUCC reveal that population density plays a statistically significant role only in large urban counties (Code 1) as well as in the largest and smallest nonadjacent rural counties (Codes 5 and 9), where the latter two are both disadvantaged by lack of access to a nearby metropolitan county. In these types of counties it may be necessary to encourage greater population concentration (i.e., consolidation of communities) if the goal is to support higher self-employment rates. Although they cannot change their stocks of natural amenities, the smaller rural counties (Codes 8 and 9) could expand the marketing and promotion of such amenities, if they wish to attract new self-employed workers from elsewhere.

Perhaps most important, our results reveal the importance of existing self-employment shares in predicting future self-employment growth in a county, which in turn may suggest a more entrepreneurial culture, stronger business networks, and local communities that are more supportive of the self-employed. Furthermore, the self-employed respond rationally to economic signals, including the returns to and the risks of self-employment, as measured at the beginning of the period over which self-employment growth is measured. Even so, there are important structural differences across the various county types.

Footnotes

Appendix

Appendix B

Auxiliary Regression Results (Replication of Earlier Studies)

Acs–Armington model			Goetz–Rupasingha model
	Standard coefficient	t		Standard coefficient	t
(Constant)		−7.27	(Constant)		−2.37
EstabSize2000	−0.072	−2.93	NEWdropout	0.074	2.01
SectorSpec2000	−0.124	−4.28	EDU685200D	0.175	4.39
NEWdropout	0.175	6.11	CmpdIncome	−0.010	−0.419
EDU685200D	0.244	8.17	share00	0.127	4.47
CmpdIncome	0.051	2.35	urate00	−0.025	−0.37
CmpdPopuln	0.173	7.82	uratesquare	0.069	1.13
share00	0.342	13.9	nfpinc00	0.042	1.76
urate00	−0.047	−2.04	CV96_05	−0.171	−8.04
Adjusted R²: 0.222	N = 1,991		PCOWNOCC	0.033	1.09
Note: Sectorspec, urate00: wrong signs.			MEDHVALU	0.002	0.06
Adjusted R² lower than Acs and Armington			medage00	0.072	2.23
			ethnic00	0.179	6.03
			flf00	0.044	1.47
			agr00	0.304	6.56
			cons00	0.114	3.72
			manu00	0.330	8.34
			rtrade00	−0.050	−1.89
			prserv00	0.060	2.00
			DEPOPC00	−0.006	−0.26
			amnscale	0.013	0.42
			area1	0.030	0.93
			area2	−0.034	−1.06
			area3	0.154	6.06
			pcinc00	−0.040	−1.00
			ws_inc00	0.040	1.40
			adjusted R² = 0.147	N = 1,991

Acknowledgements

The authors thank their anonymous reviewers for valuable comments.

Authors’ Note

The views expressed in this article are the views of the authors and do not necessarily reflect the views or policies of the Federal Reserve Bank of Atlanta, the Federal Reserve System, the U.S. Department of Agriculture, or Pennsylvania State University.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Stephan J. Goetz gratefully acknowledges financial support under USDA NIFA grant no. 2011-51150-19609, as well as support from the Pennsylvania State University Agricultural Experiment Station.

Notes

Author Biographies

Stephan J. Goetz is director of the Northeast Regional Center for Rural Development and professor of Agricultural and Regional Economics at Pennsylvania State University. He has published widely in the field of economic development, with a recent emphasis on self-employment determinants and impacts.

Anil Rupasingha is community economic development research advisor and economist at the Federal Reserve Bank of Atlanta. His areas of interest include the economics of entrepreneurship and small business development.

References

Acs

Z. J.

Armington

(2006). Entrepreneurship, geography, and American economic growth. New York, NY: Cambridge University Press.

Acs

Z. J.

Glaeser

Litan

Fleming

Goetz

S. J.

Kerr

. . . Strange

(2008). Entrepreneurship and urban success: Toward a policy consensus. Kansas City, MO: Ewing Marion Kauffman Foundation.

Alesina

Baqir

Easterly

(1999). Public goods and ethnic divisions. Quarterly Journal of Economics, 114, 1243-1284.

Athiyaman

Walzer

(2007). Entrepreneurship growth in a cross section of Midwestern counties (Working Paper). Illinois Institute for Rural Affairs, Western Illinois University, Macomb.

Bates

(1990). Entrepreneur human capital inputs and small business longevity. Review of Economics and Statistics, 72, 551-559.

Bates

Lofstrom

Servon

(2010). Why have learning programs targeting disadvantaged small-business borrowers achieved so little success in the United States? (IZA Discussion Paper No. 5212). Retrieved from http://ftp.iza.org/dp5212.pdf

Bates

Robb

(2013). Greater access to capital is needed to unleash the local economic development potential of minority-owned businesses. Economic Development Quarterly, 27, 250-259. doi:10.1177/0891242413477188

Deller

S. C.

Tsai

T.-H.

Marcouiller

D. W.

English

D. B. K.

(2001). The role of amenities and quality of life in rural economic growth. American Journal of Agricultural Economics, 83, 352-365.

Doms

Lewis

Robb

(2010). Local labor force education, new business characteristics, and firm performance. Journal of Urban Economics, 67, 61-77.

10.

Fleming

Goetz

S. J.

(2011). Does local firm ownership matter? Economic Development Quarterly, 25, 277-281. doi:10.1177/0891242411407312.

11.

Figueroa-Armijos

Dabson

Johnson

T. G.

(2012). Rural entrepreneurship in a time of recession. Entrepreneurship Research Journal. Advance online publication. doi:10.2202/2157-5665.1044

12.

Glaeser

E. L.

(2007). Entrepreneurship and the city (Discussion Paper No. 2140). Cambridge, MA: Harvard Institute of Economic Research.

13.

Glaeser

E. L.

Kerr

S. P.

Kerr

W. R.

(2012). Entrepreneurship and urban growth: An empirical assessment with historical mines (NBER Working Paper 18333). Cambridge, MA: National Bureau of Economic Research. Retrieved from http://www.nber.org/papers/w18333

14.

Glaeser

E. L.

Rosenthal

S. S.

Strange

W. C.

(2010). Urban economics and entrepreneurship. Journal of Urban Economics, 67, 1-14.

15.

Goetz

S. J.

Freshwater

(2001). State-level measures of entrepreneurship and a preliminary measure of entrepreneurial climate. Economic Development Quarterly, 15, 58-70.

16.

Goetz

S. J.

Rupasingha

(2009). Determinants of growth in non-farm proprietorship densities: 1990-2000. Small Business Economics, 32, 425-438.

17.

Goetz

S. J.

Rupasingha

Fleming

(2012). The economic impacts of self-employment: A review of the evidence. Journal of Agricultural and Applied Economics, 44, 315-221.

18.

Henderson

Low

S. A.

Weiler

(2007). The drivers of regional entrepreneurship in rural and metro areas. In Walzer

(Ed.), Entrepreneurship and local economic development (pp. 81-102). Lanham, MD: Lexington Books.

19.

Henderson

Weiler

(2010). Entrepreneurs and job growth: Probing boundaries of time and space. Economic Development Quarterly, 24, 23-32.

20.

Hustedde

(2007). What’s culture got to do with it? Strategies for strengthening an entrepreneurial culture. In Walzer

(Ed.), Entrepreneurship and local economic development (pp. 39-58). Lanham, MD: Lexington Books.

21.

Johnson

G. L.

Quance

C. L.

(1972). The overproduction trap in U.S. agriculture. Baltimore, MD/London, England: Johns Hopkins University Press/Resources for the Future.

22.

Kayne

(2007). Effective entrepreneurship education programs. In Walzer

(Ed.), Entrepreneurship and local economic development (pp. 165-185). Lanham, MD: Lexington Books.

23.

Kelley

D. J.

Ali

Brush

Corbett

A. C.

Majbouri

Rogoff

E. G.

(2012). 2012 United States report. Global Entrepreneurship Monitor, National Entrepreneurial Assessment for the United States of America, Babson College and Baruch College, New York, NY.

24.

Kutzhanova

Lyons

T. S.

Lichtenstein

G. A.

(2009). Skill-based development of entrepreneurs and the role of personal and peer group coaching in enterprise development. Economic Development Quarterly, 23, 193-210.

25.

Low

Weiler

(2012). Employment risk, returns, and entrepreneurship. Economic Development Quarterly, 26, 238-251.

26.

Lyons

Thomas S.

Alter

Theodore R.

Audretsch

David

Augustine

Darline

. “Entrepreneurship and Community: The Next Frontier of Entrepreneurship Inquiry.” Entrepreneurship Research Journal 2, no. 1 (2012): 1-26.

27.

McGranahan

(1999). Natural amenities drive rural population change (Agricultural Economic Report No. AER781). Washington, DC: U.S. Department of Agriculture.

28.

Michelacci

Silva

(2007). Why so many local entrepreneurs. Review of Economics and Statistics, 89, 615-633.

29.

Moretti

(2004). Workers’ education, spillovers and productivity: evidence from plant-level production functions. American Economic Review, 94, 656-690.

30.

Partridge

M. D.

Rickman

D. S.

(2008). Distance from urban agglomeration economies and rural poverty. Journal of Regional Science, 48, 285-310.

31.

Rupasingha

Goetz

S. J.

(2013). Self-employment and local economic performance: evidence from US counties. Papers in Regional Science, 92, 141-161. doi:10.1111/j.1435-5957.2011.00396.x

32.

Schroeder

(2007). The difference makers: Entrepreneurial young people. In Walzer

(Ed.), Entrepreneurship and local economic development (pp. 145-163). Lanham, MD: Lexington Books.

33.

Walzer

Athiyaman

Hamm

G. F.

(2007). Entrepreneurship and small business growth. In Walzer

(Ed.), Entrepreneurship and local economic development (pp. 59-79). Lanham, MD: Lexington Books.

34.

Orazem

P. F.

Jolly

R. W.

(2011). Why do rural firms live longer? American Journal of Agricultural Economics, 93, 673-692.