Innovation,Entrepreneurship,and Metropolitan Economic Performance

Abstract

The literature suggests that technological advance is the major driver of economic growth, yet how new knowledge translates into superior economic performance is not described by the growth theories. Two recently proposed frameworks, the missing link hypothesis and the knowledge spillover theory of entrepreneurship, describe a mechanism of the relationship between knowledge creation and regional economic performance through entrepreneurs. This study empirically tests these frameworks using the data on professional, scientific, and technical services in U.S. metropolitan areas from 2001 to 2005. The results indicate an intervening role of entrepreneurs in the relationship between patenting activity and job-creating behavior of incumbent companies, thus lending partial support to the missing link hypothesis. The knowledge spillover theory of entrepreneurship is not supported, as greater local knowledge generation does not translate into increased firm formation.

Keywords

knowledge spillover theory of entrepreneurship innovative environment job creation firm formation

Why do some regions grow continuously for many years whereas others stagnate? Why do some regions grow faster than others? The theoretical breakthrough in answering these questions started by Solow (1956) and Romer (1990) has lost its momentum, leaving some important questions unanswered. Following the neoclassical growth and endogenous growth theories, technological advance is believed to be the major driver of economic growth, yet how exactly new knowledge translates into superior economic performance by regions was neither described by the growth theories nor found unequivocal empirical explanation. Empirical studies, lacking theoretical underpinnings, looked into networks (Wal & Boschma, 2009), labor mobility (Almeida & Kogut, 1999), and other potential facilitators of spillovers.

In the past few years, however, researchers have increasingly paid attention to the link between innovation, entrepreneurship, and regional economic outcomes. “Entrepreneurial approach” builds on a vision of entrepreneurs as the ones who translate knowledge into economically useful knowledge or penetrate the “knowledge filter” (Acs & Plummer, 2005); therefore, more entrepreneurial regions with greater stock of knowledge should grow faster. The research efforts in this field culminated in proposition of two related but slightly different theories on the role of entrepreneurs in knowledge utilization. The first one, the so-called missing link hypothesis (MLH), suggests that entrepreneurship is the conduit necessary for the research and development (R&D) results to find their ways to profitable economic applications. Without entrepreneurs, accumulated knowledge would not influence regional economic performance (Audretsch & Keilbach, 2008).¹ The second framework, the knowledge spillover theory of entrepreneurship (KSTE), postulates that greater accumulated knowledge begets more entrepreneurs because the more knowledge is produced and remains unused, the greater regional entrepreneurial opportunities exist (Acs, Braunerhjelm, Audretsch, & Carlsson, 2009).²

The MLH and the KSTE have been mostly tested using data on European regions. The scarcity of empirical attention to these theories in the context of U.S. regions is surprising, given immediate policy implications of the findings. As policy makers strive to revive and boost economic performance of their jurisdictions, general knowledge of what factors are likely to have the greatest impact in the long run is important. If the MLH is supported, efforts by private and public entities to advance innovation should be supplemented by local furtherance of entrepreneurship in order for regions to tap into the synergies suggested by the MLH. If empirical evidence backs the knowledge spillover theory of entrepreneurship, economic development programs should strive to facilitate knowledge creation in the first place.

This research contributes to the literature by testing empirically both theories in the context of the U.S. metropolitan statistical areas (MSAs). It estimates the effects of innovation and entrepreneurship on job and firm creation, and firm growth in professional, scientific, and technical services (North American Industry Classification System [NAICS] 54) during the 2001 to 2005 period. The results lend partial support to the MLH, whereas the KSTE is not supported. The findings point to the important role played by NAICS54 entrepreneurs, as they both mediate and moderate the effect of metropolitan innovation on job creation by incumbent firms and on the number of growing incumbent companies. The effect of economy-wide entrepreneurship on overall job creation and the number of expanding incumbent firms is negligible.

This study briefly reviews the classical growth models and presents the MLH and the KSTE. The latter two frameworks depict different mechanisms of the relationship between knowledge creation and regional economic performance. The study uses these differences to test the MLH and the KSTE and describes econometric estimation approach and estimation results for both.

Endogenous Growth Theory and Entrepreneurial Approach

Conditioning economic growth on technological progress and innovation in empirical research is rather a convention now. There is a solid theoretical base for this. According to the endogenous growth theory, innovative efforts by individual firms promote growth. In the basic setup of endogenous growth models, a profit-maximizing firm purposefully invests resources in new knowledge creation to ensure its competitive advantage. A superior performance by individual firms translates into faster economic growth of a region (Romer, 1990). This story implies that firms exist exogenously. They create knowledge endogenously, hence, the name of the theory. In a framework related to the endogenous growth theory, Griliches (1979) proposes a knowledge production function that specifies economic knowledge as an input into innovative activity by firms. Because of partial excludability, new knowledge cannot be monopolized entirely and “spills over” to other companies not engaged in R&D. The growth is then reinforced by intratemporal knowledge spillovers that make increasing returns to scale possible. The effects of accumulated knowledge, however, are not uniform across space. The new economic geography contends that knowledge gets accumulated in the locations it has been produced in, and explains the importance of spatial proximity for local knowledge spillovers. These spillovers result in greater productivity of firms both engaged in R&D and not engaged. Regional growth rates, then, depend on the local accumulated stock of knowledge and the opportunity for spillovers.

Local knowledge spillovers are extremely important for explanation of growth because a production function with the possibility of spillovers allows for increasing returns to scale. Increasing returns are able to support observed empirically continuous growth in some countries, despite decreasing marginal returns on capital and labor. Perhaps for this reason many studies assumed presence of local knowledge spillovers for the analysis of regional economic performance. Yet, some researchers stress insufficient understanding of the actual spillover mechanisms (Breschi & Lissoni, 2001). Indeed, during two decades after the endogenous growth theory was formulated, no convincing theory has been proposed that would explain how exactly new knowledge affects regional growth rates.

The MLH and the KSTE fill this gap by introducing entrepreneurship as a mechanism of spillovers in the endogenous growth theory framework (Acs, Audretsch, Braunerhjelm, & Carlsson, 2012; Acs, Braunerhjelm, et al., 2009; Braunerhjelm, Acs, Audretsch, & Carlsson, 2010). The entrepreneurial view starts with the premise that knowledge, once generated, needs to be turned into economically useful knowledge (the one that finds profitable market applications) to promote economic growth. If the new knowledge is not used in the market, it is useless for the purposes of immediate growth (Acs, Plummer, & Sutter, 2009). Entrepreneurs are hypothesized to “penetrate the knowledge filter” and put new knowledge to work³; therefore, entrepreneurs (in addition to technological innovation) play a central role in economic growth (Acs & Sanders, 2012).

According to a stylized entrepreneurial approach story, an incumbent firm produces new knowledge (develops technological innovations) and implements the resulting products in the market deriving monopolistic profits on its inventions. The process of knowledge creation, however, is not a straightforward one. In the course of trials and errors, a fraction of new knowledge is discarded as useless for the marketing purposes. Because of information asymmetries naturally embedded in all new knowledge (Arrow, 1962), people engaged in research and development, or those who have access to research activities, may disagree with the incumbent firm with regard to the market prospects of a newly created technology. If this happens, a knowledge entrepreneur (a person willing to bring a new technology, abandoned by an incumbent firm, to market) would set up a business with the goal to capitalize on the dormant ideas. This framework reverses the traditional argument on the endogenously created knowledge by exogenously existing firms. In contrast to the endogenous growth theory, entrepreneurial view implies that firms are endogenously created to use exogenously existing knowledge (Audretsch, 1995). Firm creation, in turn, should result in superior economic performance by regions with greater stock of knowledge.

The story about starting a new business with the purpose of using dormant knowledge gives rise to at least two related hypotheses explaining the role of entrepreneurs in the modern knowledge economy. According to the MLH, entrepreneurship is a conduit that helps translate new knowledge into economically useful knowledge (Braunerhjelm et al., 2010). This way, technological innovation is a necessary but insufficient condition for economic growth. In order for the new knowledge to affect regional economic performance, entrepreneurs are needed (Audretsch & Keilbach, 2008). They have to be present in a locality a priori to provide a channel for the new knowledge to the market. Both technological innovation and entrepreneurs are necessary for economic growth. Combined together, they become a sufficient condition for the purpose.

The KSTE is a more straightforward extension of the endogenous firm creation story. The theory seeks to explain the level of entrepreneurship in a region implicitly assuming its positive role in economic growth (Audretsch, 2007). According to the KSTE, some knowledge inevitably remains unused because of inherent informational asymmetries with regard to the market potential of every new idea. The more knowledge remains dormant this way, the more opportunities for knowledge entrepreneurs exist in an area. Greater knowledge creation (or more ideas being unused) should lead to greater firm formation rates and greater entrepreneurial capital in geographical regions (Acs, Braunerhjelm, et al., 2009).

These two theories imply different mechanisms of the relationship between new knowledge, entrepreneurship, and regional economic performance. In the missing link framework, entrepreneurs make the effect of innovation on economic outcomes possible. This relationship is intratemporal. In the knowledge spillover theory of entrepreneurship, past level of innovation should predict current level of firm formation, at least in knowledge-intensive industries. This article uses these differences to test the MLH and the KSTE in a context of one U.S. industry.

Data and Variables

This research focuses on professional, scientific, and technical services (NAICS54) during the 2001 to 2005 period. This industry is characterized by a high level of expertise and training of its employees. Some occupations in the industry, such as scientific research or computer systems design, are likely to be particularly perceptive to the innovative environment. The time span of the study is determined by data availability. Although such a short period may limit generalizability of the findings, it is hoped that the results are useful as preliminary evidence on the mechanism of the relationship between knowledge creation and metropolitan economic performance in the industry of interest.⁴

Dependent Variables

This study estimates the effects of innovation on three outcomes when testing the MLH and on one outcome (measured in two different ways) when testing the KSTE. The MLH implies that entrepreneurial capital of a region would ensure that local invention translates into innovation, thus enhancing regional economic performance. The literature does not provide a uniform measure of this important statistic of regional well-being; actually, considerable disagreement exists as to what constitutes economic growth and economic development (Partridge & Rickman, 2003). The extant studies focus on per capita gross domestic product (Well, 2007), increase in gross domestic product (Rodriguez-Pose & Ezcurra, 2011; van Stel, Carree, & Thurik, 2005), increase in total regional income or earnings (Aldrich & Kushmin, 1997; Monchuk, Miranowski, Hayes, & Babcock, 2007), and increase in employment and productivity (Frenken, Oort, & Verburg, 2007), to name just a few.

Arguably, firm formation and employment growth are economic indicators that are highly relevant for the purposes of policy making. To test the MLH, these regional characteristics are used to operationalize economic development of MSAs in the United States. New jobs can be added either by existing businesses or by new start-ups. Both sources of employment are important, although empirical literature suggests a leading role played by newly established companies (Acs & Armington, 2004). Two variables, the number of new jobs in expanding firms (EmpExpand) and the number of expanding firms (FirmsExpand), capture the contribution of existing businesses to regional economic performance. The number of jobs created by start-ups (EmpNew) does the same for new companies. All dependent variables are population adjusted. They are calculated for professional, scientific, and technical services in the U.S. metropolitan areas by aggregating firm-level data into metropolitan indicators.

The KSTE suggests that the new knowledge generated in a region should stimulate firm formation. It appears logical to test the KSTE by modeling the number of start-ups as a function of local knowledge generation. The theory does not specifically indicate in which industries the proposed relationship should be present, although knowledge-intensive industries are more likely to benefit from knowledge generation. The number of new firms per 1,000 residents in the NAICS54 industry (StUps) is a dependent variable that depicts the specifics of a knowledge-intensive industry, whereas the number of new firms in all industries per 1,000 residents (StUpsAll) is an alternative (more general) dependent variable. Table 1 presents summary statistics for the dependent variables.

Table 1.

Descriptive Statistics for the Dependent Variables, by Year.

Statistics	2001	2002	2003	2004	2005
EmpExpand
Mean	1.775	1.888	1.223	1.521	1.174
Standard deviation	4.294	2.563	1.172	2.796	1.144
Minimum	0.083	0.072	0.080	0.037	0.129
Maximum	68.920	34.169	9.431	41.252	11.624
FirmsExpand
Mean	0.176	0.228	0.209	0.189	0.206
Standard deviation	0.105	0.131	0.113	0.108	0.110
Minimum	0.038	0.048	0.046	0.024	0.050
Maximum	1.011	1.527	1.010	1.222	1.335
EmpNew
Mean	3.045	3.687	2.173	1.671	1.705
Standard deviation	3.052	2.801	2.763	1.548	1.826
Minimum	0.354	0.512	0.328	0.228	0.184
Maximum	23.398	22.232	40.620	12.229	27.281
StUps
Mean	0.537	0.841	0.802	0.453	0.585
Standard deviation	0.291	0.483	0.496	0.252	0.344
Minimum	0.154	0.134	0.195	0.113	0.132
Maximum	2.546	4.007	2.766	2.224	3.146
StUpsAll
Mean	26.5	23.4	12.3	17.2	23.7
Standard deviation	77.7	76.2	35.0	48.0	61.2
Minimum	0.2	0.1	0.1	0.1	0.1
Maximum	675.2	796.9	332.3	490.7	569.4

Note. The number of observations (metropolitan statistical areas [MSAs]) is 358 for each variable and for every year.

The National Establishment Time Series (NETS) Database is the data source for all response variables and is created by Walls and Associates from the Dun and Bradstreet data. The database consists of yearly snapshots of the U.S. economy (all firms recoded by Dun and Bradstreet to be active) performed every January from year 1990. It contains names, location, years of operation, industrial classification, estimated number of employees, and other characteristics. To create the response variables in this study, establishments that added new jobs since the previous year and newly created establishments in professional, scientific, and technical services, as well as new establishments in all industries, are identified. The indicators of interest are then aggregated for each year and each MSA. Assuming that the NETS Database is a census of active establishments in a given year, such aggregation should produce accurate statistics intended to capture regional economic performance.

Independent Variables

There are two explanatory variables in the models testing the MLH and one explanatory variable in the models testing the KSTE. According to the MLH, regional economic performance is conditional on both new knowledge production and entrepreneurship in a region. The literature usually uses R&D expenditures, share of employees in knowledge-intensive industries, and patent counts to operationalize new knowledge production (innovation). Population-adjusted patent counts in MSAs, Patents, are used to approximate regional innovative activity. To compute the variable for each metropolitan area, all patents granted to inventors residing in an MSA are added together. If more than one inventor is listed on a patent, a corresponding fraction is assigned to the metropolitan areas of inventors’ residence. To capture the effects of the knowledge stock, not patents per se, patents are attributed to the year when a patent application is submitted. It means that patents granted in 2008 with an application date of 2004 appear in the data set under year 2004, whereas patents granted in 2004 with an application date of 1999 do not appear. The U.S. Patent and Trademark Office website (http://patft.uspto.gov/) is the data source.

One has to keep in mind that patent count as an indicator of regional innovative activity has a number of limitations. This variable accounts only for inventions that have been assessed and granted a patent by the U.S. Patent and Trademark Office. Innovations that go “unnoticed” by this governmental authority and innovations that are denied a patent cannot be captured by the patent count variable. In addition, the economic value of each patent (and thus its usefulness) differs greatly (Griliches, 1979; Pakes & Griliches, 1980), especially after a relatively recent surge in U.S. patenting because of the patent reform in this country (Gallini, 2002). Despite this fact, patent count is perhaps the best readily available indicator of underlying inventive activity in a metropolitan area⁵ (Acs, Anselin, & Varga, 2002; Feser, 2002; Griliches, 1990).

Another important source of regional well-being, as suggested by the MLH, is entrepreneurship. The empirical literature traditionally approximates the level of entrepreneurship in a region by per capita number of establishments below a certain size (in employment, sales, stocks, or other indicators), younger than certain age (usually 5-7 years), a per capita or aggregate number of new entrants as a fraction of the total number of incumbent firms (Praag & Versloot, 2007), and a share of self-employed in the economy (Acs, Braunerhjelm, et al., 2009). Braunerhjelm et al. (2010) note that per capita number of start-ups is an optimal measure of entrepreneurship in an area.⁶

The research that links regional economic performance to knowledge creation via entrepreneurship refers to entrepreneurship in general and does not specify in which industries this mechanism is more likely to manifest itself. This study employs the number of start-ups per 1,000 residents, StUpsAll, as a measure of entrepreneurship.⁷ The variable is calculated by aggregating the total number of start-ups in all industries into metropolitan variables from the NETS Database.⁸ This indicator is able to capture the institutional environment that promotes or hinders entrepreneurship, entrepreneurial attitudes in the society, opportunities for entrepreneurship in a region, as well as the ability of current or potential entrepreneurs to use these opportunities. One needs to remember, however, that opportunity for entrepreneurship is likely to differ across industries, whereas the nature of the human capital accumulated in a region determines the extent to which such opportunities in each industry are used. Because the dependent variables in the models testing the MLH capture only professional, scientific, and technical services, using a measure of entrepreneurship that would encapsulate the specifics of entrepreneurship in this industry is justified. In addition to StartUpsAll, models in this study include population-adjusted number of start-ups in the NAICS54 industry (StUps) as an alternative measure of entrepreneurship that is specifically related to the entrepreneurial opportunities in professional, scientific, and technical services and the ability of metropolitan population to capitalize on them. StUps is calculated from the NETS Database following the methodology used to calculate StUpsAll.

The KSTE postulates that innovative activity in a region should fuel firm formation because of greater entrepreneurial opportunities stemming from new knowledge generation. A population-adjusted patent counts, Patents, is used as a measure of metropolitan innovative activity in a test of knowledge spillover theory of entrepreneurship. Because it may take extended time for the new knowledge to make its way into the market, up to 10 yearly lags of the explanatory variable are introduced in the models. The data source and calculation procedure are described above. Table 2 gives summary statistics for the main explanatory variables in this study.

Table 2.

Descriptive Statistics for Explanatory Variables, by Year.

Patents
Year	1992	1993	1994	1995	1996
Mean	0.197	0.207	0.234	0.264	0.255
Standard deviation	0.188	0.200	0.234	0.299	0.300
Minimum	0.000	0.005	0.004	0.005	0.000
Maximum	1.476	1.625	2.081	2.935	3.230
Year	1997	1998	1999	2000	2001
Mean	0.290	0.289	0.302	0.320	0.323
Standard deviation	0.366	0.372	0.397	0.433	0.475
Minimum	0.000	0.000	0.000	0.011	0.000
Maximum	4.180	3.975	4.443	4.927	5.121
Year	2002	2003	2004	2005
Mean	0.316	0.310	0.284	0.263
Standard deviation	0.464	0.467	0.446	0.422
Minimum	0.000	0.000	0.000	0.000
Maximum	5.301	4.912	4.857	4.584
StUps
Year	2001	2002	2003	2004	2005
Mean	0.537	0.841	0.802	0.453	0.585
Standard deviation	0.291	0.483	0.496	0.252	0.344
Minimum	0.154	0.134	0.195	0.113	0.132
Maximum	2.546	4.007	2.766	2.224	3.146
StUpsAll
Year	2001	2002	2003	2004	2005
Mean	26.5	23.4	12.3	17.2	23.7
Standard deviation	77.7	76.2	35.0	48.0	61.2
Minimum	0.2	0.1	0.1	0.1	0.1
Maximum	675.2	796.9	332.3	490.7	569.4

Note. The number of observations (metropolitan statistical areas [MSAs]) is 358 for each variable and for every year.

Control Variables

The list of regional characteristics that are important for firm formation and job creation, and may potentially intervene into the relationship between the dependent and explanatory variables in this study, is very broad. The empirical literature consistently points out the role played by labor pool quality, agglomeration economies, regional economic conditions, average income, and other factors (Armington & Acs, 2002; Knoben, Ponds, & Oort, 2011; Qian, Acs, & Stough, 2013). For this study, a level of educational attainment, gross metropolitan product (GMP), job density, population density, industrial diversity, and per capita income are calculated. Because of multicollinearity problems, some of these variables are omitted from the analysis. The models testing the MLH include measures of educational attainment, GMP, industrial diversity, and job density as controls. All these variables except for diversity are included in the KSTE testing. All models include time dummies for years 2002 to 2005, with 2001 as a reference category.

The argument tested in this study maintains that the new knowledge should affect regional performance because entrepreneurs use this new knowledge. As Qian and Acs (2013) argue, knowledge utilization depends on absorptive capacity of a region, which is related to the ability of local entrepreneurs “to understand new knowledge, to recognize its value, and to commercialize it by creating a new firm” (Qian & Acs, 2013, p. 185). I use educational level in the U.S. MSAs, Education, measured by percentage of adults with a bachelor’s degree or higher, as a proxy for human capital that is an essential component of the regional absorptive capacity. The U.S. Census Bureau is the data source for this variable.

GMP reflects the size of a metropolitan economy. In general, greater GMP is associated with more resources available to firms and should be positively related to economic performance. The variable GMP is retrieved from the Bureau of Economic Analysis’ data (http://www.bea.gov/regional/). It is calculated per capita and is adjusted for inflation.

Another important factor to be included in the models is the level of industrial agglomeration. The empirical literature shows that agglomerated economies enjoy higher productivity and greater levels of firm formation and growth (Martin & Ottaviano, 2001; Melo, Graham, & Noland, 2009; Rosenthal & Strange, 2003, 2004). Businesses can benefit from agglomeration because of deeper markets for the products, reduced transaction costs, and plentiful specialized suppliers (Marshall, 1890/1920). On the other hand, high rental prices and congestion associated with urban agglomerations might hinder firm formation and job creation in a region. Possible ways to measure agglomerated economies include population, firm, or employment densities. These measurements, however, are usually highly correlated and cannot be included in a model simultaneously. This study approximates the level of agglomeration by job density, Density, which is calculated as the number of jobs in all sectors (in thousands) divided by the land area of an MSA. The NETS Database and the U.S. Census Bureau are the data sources for this variable.

Industrial diversity promotes recombination of ideas and innovation (Feldman & Audretsch, 1999; Jacobs, 1969), can alleviate negative economic trends and promote spillover effects (van Stel & Nieuwenhuijsen, 2004). The variable Diversity controls for this possibility. Following a number of studies (Attaran, 1986; Bishop & Gripaios, 2007), the models in this study include entropy measure as an approximation for the diversification of the economy in each metropolitan area. As suggested by Bishop and Gripaios (2007), total diversity can be expressed as follows:

T D = \sum_{1}^{n} S_{i} \ln (\frac{1}{s_{i}}),

where S_i stands for the share of a three-digit NAICS category in total MSA employment and there are n such categories. The total diversity index is zero if all employment is concentrated in one sector; it is maximized if employment is distributed evenly among the sectors. The measure is also dependent on the total number of sectors with the share of each sector weighted by the logarithmic function. The diversity index is calculated from the NETS Database employing the NAICS three-digit classification (all industries with nonzero employment in an MSA) and aggregated employment statistics from the same source.

The Missing Link Hypothesis

Empirical Testing

The MLH implies that the influence of new knowledge on regional economic performance is conditional on the presence of a sufficient number of entrepreneurs. For the purpose of empirical testing, one can model entrepreneurship as an intervening variable in the relationship between innovation and regional economic outcomes. The literature suggests two ways of testing the effects of major and intervening variables on a dependent variable, mediation, and moderation.⁹ If mediation is present, the main explanatory variable has both direct and indirect (via intervening variable) effects on the outcome. If moderation is present, the effect of the main explanatory variable on the outcome depends on the level of another variable. Both these possibilities are tested to establish if entrepreneurship is a mediator, a moderator, both, or neither.

The mediation effect is tested by following a three-step procedure suggested by Baron and Kenny (1986) and represented graphically in Figure 1.

Figure 1.

Mediated relationship.

In the first step, the effect of the main explanatory variable on the outcome is tested. In the second step, the effect of the main explanatory variable on the intervening variable is estimated. Finally, both variables are included in the model to determine if they are significant predictors. These three steps (provided there is a theoretical reason to expect the relationship being tested) are sufficient to establish mediation. The significance of the main independent variables in the third step shows if there is a complete or a partial mediation.

The following equations for each dependent variable are estimated using two approximations for the level of entrepreneurship in a region, StUpsAll and StUps:

Step 1 : D e p V a r_{i t} = β_{1} + β_{2} P a t e n t s_{i t} + β_{3} X_{i t} + e_{i t}

Step 2 : S t U p s A l l_{i t} = α_{1} + α_{2} P a t e n t s_{i t} + α_{3} X_{i t} + e_{i t}

S t U p s_{i t} = α_{1} + α_{2} P a t e n t s_{i t} + α_{3} X_{i t} + e_{i t}

Step 3 : D e p V a r_{i t} = γ_{1} + γ_{2} P a t e n t s_{i t} + γ_{3} S t U p s A l l_{i t} + γ_{4} X_{i t} + e_{i t}

D e p V a r_{i t} = γ_{1} + γ_{2} P a t e n t s_{i t} + γ_{3} S t U p s_{i t} + γ_{4} X_{i t} + e_{i t}

where β, α, γ are regression coefficients; subscript i denotes an MSA; and subscript t denotes a year. X _it is a matrix of control variables (Education, Diversity, Density, and GMP) and year dummies.

In a moderated statistical relationship, the effect of one explanatory variable depends on the level of another explanatory variable (Figure 2).

Figure 2.

Moderated relationship.

To test for moderation, interaction term Interaction is calculated by multiplying corresponding values of Patents and StUpsAll, and Patents and StUps. Significance of the interaction regression coefficient and its direction indicate presence of moderation. Interaction is included in the equations below, together with other explanatory and control variables. The meaning of the equation components is identical to the previous equations.

D e p V a r_{i t} = γ_{1} + γ_{2} P a t e n t s_{i t} + γ_{3} S t U p s A l l_{i t} + γ_{4} I n t e r a c t i o n_{i t} + γ_{5} X_{i t} + e_{i t}

D e p V a r_{i t} = γ_{1} + γ_{2} P a t e n t s_{i t} + γ_{3} S t U p s_{i t} + γ_{4} I n t e r a c t i o n_{i t} + γ_{5} X_{i t} + e_{i t}

The nature of the data dictates the econometric approach to be used to estimate Equations (1) to (4a). Heteroskedasticity of the data makes parameter estimates inefficient and hypotheses tests unreliable. In addition, inertia in several variables introduces autocorrelation in the error terms. Feasible generalized least squares approach with panel specific first-order autoregressive process and heterogeneous panels are used to resolve the problems related to heteroskedasticity and autocorrelation. The next subsection presents estimation results and discussion.

Estimation Results and Discussion

Estimation results are presented for each dependent variable separately. The tests for both intervening mechanisms (mediation and moderation) are reported in each table of results. Tables 3 to 5 test the MLH using StUpsAll as a measure of entrepreneurship, whereas Tables 6 to 8 employ the variable StUps.

Table 3.

Estimation Results for the MLH.

	Mediation test			Moderation test
	(1)	(2)	(3)	(4)
Variable	Dependent variable EmpExpand	Dependent variable StUpsAll	Dependent variable EmpExpand	Dependent variable EmpExpand
Patents	0.105* (0.054)	9.542*** (1.142)	0.105* (0.055)	0.182*** (0.056)
StUpsAll			0.000 (0.000)	0.001** (0.000)
Interaction				−0.003*** (0.001)
Education	7.372*** (0.314)	−21.478*** (4.978)	7.355*** (0.315)	7.310*** (0.314)
Diversity	−0.020 (0.108)	32.362*** (1.628)	−0.031 (0.108)	0.023 (0.109)
Density	0.992*** (0.218)	−22.343*** (2.688)	0.100*** (0.217)	1.027*** (0.218)
GMP	2.891*** (0.515)	−20.881*** (6.524)	2.935*** (0.510)	2.856*** (0.517)
Year2002	0.131*** (0.030)	6.610*** (0.288)	0.129*** (0.030)	0.117*** (0.030)
Year2003	−0.294*** (0.036)	4.592*** (0.385)	−0.296*** (0.036)	−0.315*** (0.0363603)
Year2004	−0.180*** (0.038)	−1.234*** (0.452)	−0.180*** (0.038)	−0.189*** (0.0381)
Year2005	−0.427*** (0.041)	2.047*** (0.514)	−0.427*** (0.041)	−0.432*** (0.041)
Constant	−0.365 (0.417)	−101.360 (5.991)	−0.327 (0.418)	−0.521 (0.420)
Observations	1,790
Groups	358
Wald χ²	1,809	1,624	1,816	1,867
Prob > χ²	0.0000	0.0000	0.0000	0.0000

Note. MLH = missing link hypothesis; GMP = gross metropolitan product; DV = dependent variable. DV is the number of new jobs in incumbent firms; entrepreneurship is measured by population-adjusted number of start-ups in all industries.

Significant at .10 level. **Significant at .05 level. ***Significant at .01 level.

Table 4.

Estimation Results for the MLH.

	Mediation test			Moderation test
	(1)	(2)	(3)	(4)
Variable	Dependent variable FirmsExpand	Dependent variable StUpsAll	Dependent variable FirmsExpand	Dependent variable FirmsExpand
Patents	0.024*** (0.003)	9.542*** (1.142)	0.024*** (0.003)	0.028*** (0.003)
StUpsAll			0.000 (0.000)	0.000*** (0.000)
Interaction				−0.000*** (0.000)
Education	0.659*** (0.0139)	−21.478*** (4.978)	0.659*** (0.014)	0.654*** (0.014)
Diversity	0.025*** (0.005)	32.362*** (1.628)	0.023*** (0.005)	0.024 (0.005)
Density	0.024** (0.010)	−22.343*** (2.688)	0.024** (0.010)	0.023** (0.010)
GMP	0.098*** (0.023)	−20.881*** (6.524)	0.098*** (0.023)	0.097*** (0.022)
Year2002	0.048*** (0.001)	6.610*** (0.288)	0.048*** (0.001)	0.047*** (0.001)
Year2003	0.026*** (0.001)	4.592*** (0.385)	0.026*** (0.001)	0.025*** (0.001)
Year2004	0.006*** (0.001)	−1.234*** (0.452)	0.006*** (0.001)	0.005*** (0.001411)
Year2005	0.018*** (0.002)	2.047*** (0.514)	0.018*** (0.002)	0.018*** (0.002)
Constant	−0.084 (0.018)	−101.360 (5.991)	−0.077 (0.018)	−0.080 (0.018)
Observations	1,790
Groups	358
Wald χ²	6,760	1,624	6,864	6,817
Prob > χ²	0.0000	0.0000	0.0000	0.0000

Significant at .10 level. **Significant at .05 level. ***Significant at .01 level.

Table 5.

Estimation Results for the MLH.

	Mediation test			Moderation test
	(1)	(2)	(3)	(4)
Variable	Dependent variable EmpNew	Dependent variable StUpsAll	Dependent variable EmpNew	Dependent variable EmpNew
Patents	0.060 (0.071)	9.542*** (1.142)	0.045 (0.073)	0.047 (0.080)
StUpsAll			0.001* (0.001)	0.001 (0.001)
Interaction				0.0003 (0.002)
Education	9.841*** (0.476)	−21.478*** (4.978)	9.864*** (0.476)	9.855*** (0.476)
Diversity	0.658*** (0.131)	32.362*** (1.628)	0.605*** (0.134)	0.606*** (0.134)
Density	1.035*** (0.372)	−22.343*** (2.688)	1.068*** (0.000)	1.064*** (0.373)
GMP	6.276*** (0.001)	−20.881*** (6.524)	6.462*** (0.001)	6.516*** (1.066)
Year2002	0.656*** (0.042)	6.610*** (0.288)	0.646*** (0.042)	0.645*** (0.042)
Year2003	−0.710*** (0.050)	4.592*** (0.385)	−0.714*** (0.050)	−0.713*** (0.050)
Year2004	−1.169*** (0.053)	−1.234*** (0.452)	−1.158*** (0.053)	−1.157*** (0.0534)
Year2005	−1.207*** (0.057)	2.047*** (0.514)	−1.201*** (0.057)	−1.200*** (0.057)
Constant	−2.213 (0.497)	−101.360 (5.991)	−2.052*** (0.508)	−2.054*** (0.507)
Observations	1,790
Groups	358
Wald χ²	2,583	1,624	2,570	2,572
Prob > χ²	0.0000	0.0000	0.0000	0.0000

Significant at .10 level. **Significant at .05 level. ***Significant at .01 level.

Table 6.

Estimation Results for the MLH.

	Mediation test			Moderation test
	(1)	(2a)	(3a)	(4a)
Variable	Dependent variable EmpExpand	Dependent variable StUps	Dependent variable EmpExpand	Dependent variable EmpExpand
Patents	0.105* (0.054)	0.029*** (0.011)	0.016 (0.044)	−0.322*** (0.085)
StUps			0.936*** (0.068)	0.799*** (0.072)
Interaction				0.391*** (0.112)
Education	7.372*** (0.314)	2.072*** (0.069)	5.339*** (0.301)	5.394*** (0.295)
Diversity	−0.020 (0.108)	0.068*** (0.017)	0.010 (0.095)	−0.009 (0.088)
Density	0.992*** (0.218)	0.002 (0.041)	0.791*** (0.192)	0.772*** (0.188)
GMP	2.891*** (0.515)	0.818*** (0.141)	2.340*** (0.372)	2.461*** (0.367)
Year2002	0.131*** (0.030)	0.254*** (0.006)	−0.109*** (0.036)	−0.123*** (0.036)
Year2003	−0.294*** (0.036)	0.189*** (0.007)	−0.529*** (0.040)	−0.552*** (0.038)
Year2004	−0.180*** (0.038)	−0.123*** (0.007)	−0.074* (0.038)	−0.107*** (0.036)
Year2005	−0.427*** (0.041)	−0.021** (0.008)	−0.437*** (0.040)	−0.467*** (0.038)
Constant	−0.365 (0.417)	−0.205 (0.063)	−0.459 (0.362)	−0.276 (0.333)
Observations	1,790
Groups	358
Wald χ²	1,809	6,574	2,845	3,266
Prob > χ²	0.0000	0.0000	0.0000	0.0000

Significant at .10 level. **Significant at .05 level. ***Significant at .01 level.

Table 7.

Estimation Results for the MLH.

	Mediation test			Moderation test
	(1)	(2a)	(3a)	(4a)
Variable	Dependent variable FirmsExpand	Dependent variable StUps	Dependent variable FirmsExpand	Dependent variable FirmsExpand
Patents	0.024*** (0.003)	0.029*** (0.011)	0.014*** (0.003)	−0.028*** (0.005)
StUps			0.127*** (0.003)	0.110*** (0.004)
Interaction				0.044*** (0.005)
Education	0.659*** (0.014)	2.072*** (0.069)	0.408*** (0.014)	0.409*** (0.014)
Diversity	0.025*** (0.005)	0.068*** (0.017)	0.022*** (0.003)	0.024*** (0.003)
Density	0.024** (0.010)	0.002 (0.040)	0.033*** (0.008)	0.037*** (0.008)
GMP	0.098*** (0.023)	0.818*** (0.141)	−0.045*** (0.014)	−0.047*** (0.014)
Year2002	0.048*** (0.001)	0.253*** (0.006)	0.013*** (0.002)	0.014*** (0.002)
Year2003	0.026*** (0.001)	0.189*** (0.007)	−0.004** (0.002)	−0.003 (0.002)
Year 2004	0.006*** (0.001)	−0.123*** (0.007)	0.019*** (0.002)	0.019*** (0.002)
Year 2005	0.018*** (0.002)	−0.021*** (0.008)	0.014*** (0.002)	0.014*** (0.002)
Constant	−0.084 (0.018)	−0.205 (0.063)	−0.079 (0.013)	−0.076 (0.013)
Observations	1,790
Groups	358
Wald χ²	6,760	6,574	8,438	7,455
Prob > χ²	0.0000	0.0000	0.0000	0.0000

Note. MLH = missing link hypothesis; GMP = gross metropolitan product; DV = dependent variable. DV is the number of expanding firms; entrepreneurship is measured by population adjusted number of start-ups in NAICS54.

Significant at .10 level. **Significant at .05 level. ***Significant at .01 level.

Table 8.

Estimation results for the MHL.

	Mediation test			Moderation test
	(1)	(2a)	(3a)	(4a)
Variable	Dependent variable EmpNew	Dependent variable StUps	Dependent variable EmpNew	Dependent variable EmpNew
Patents	0.060 (0.071)	0.029*** (0.011)	0.009 (0.051)	−0.254*** (0.096)
StUps			2.830*** (0.082)	2.669*** (0.089)
Interaction				0.348*** (0.110)
Education	9.841*** (0.476)	2.072*** (0.069)	3.360*** (0.373)	3.428*** (0.376)
Diversity	0.658*** (0.131)	0.068*** (0.017)	0.314*** (0.110)	0.316*** (0.110)
Density	1.035*** (0.372)	0.002 (0.040)	0.918*** (0.297)	0.943*** (0.296)
GMP	6.276*** (0.001)	0.818*** (0.141)	4.405*** (0.811)	4.580*** (0.830)
Year2002	0.656*** (0.042)	0.253*** (0.006)	−0.090** (0.046)	−0.077* (0.046)
Year2003	−0.710*** (0.050)	0.189*** (0.007)	−1.351*** (0.049)	−1.335*** (0.050)
Year2004	−1.169*** (0.053)	−0.123*** (0.007)	−0.862*** (0.049)	−0.870*** (0.049)
Year2005	−1.207*** (0.057)	−0.021*** (0.008)	−1.183*** (0.051)	−1.187*** (0.051)
Constant	−2.213 (0.497)	−0.205 (0.063)	−0.998 (0.418)	−0.931 (0.422)
Observations	1,790
Groups	358
Wald χ²	2,583	6,574	4,537	4,494
Prob> χ²	0.0000	0.0000	0.0000	0.0000

Note. MLH = missing link hypothesis; GMP = gross metropolitan product; DV = dependent variable. DV is the number of jobs created by start-ups; entrepreneurship is measured by population adjusted number of start-ups in NAICS54.

Significant at .10 level. **Significant at .05 level. ***Significant at .01 level.

Table 3 shows the effects of Patents and StUpsAll on the number of existing firms that created new jobs in years 2001 to 2005. The results suggest that incumbent firms tend to create more jobs in metropolitan areas with higher innovative activity. Innovation is also positively related to firm creation as follows from Equation (2). Simultaneous inclusion of innovation and entrepreneurship measures in Equation (3) does not change the results obtained in Equation (1). This implies that entrepreneurship (measured by the total number of start-ups) does not mediate the effect of innovation on incumbents’ expansion. There is, however, evidence of statistical moderation (Equation 4). In this specification, both entrepreneurship and innovation are positively related to job creation by the existing firms, but the effect of one variable decreases as the level of another one increases.

The coefficients of the control variables reveal a rather interesting picture. First, diversity does not affect job creation by existing firms but it does promote entrepreneurship. More diverse MSAs tend to set up more new firms. Education is positively related to the likelihood of incumbent firms to create new jobs, but it is negatively related to the population-adjusted number of start-ups in a region. This indirectly supports the perspective proposed by Shane (2008). According to this view, entrepreneurship in general is a response to the lack of economic opportunities in a region or the result of inability to use such opportunities because of low educational level and other factors.¹⁰ Estimation results in Table 3 go in line with such reasoning. Size of a metro economy is positively related to job creation by existing firms and is negatively related to firm creation. The effects of job density uncover the same pattern.

Table 4 shows estimation results for the population-adjusted number of expanding firms as a dependent variable. In general, the effects of innovation and entrepreneurship are identical to Table 3. Companies are more likely to add new jobs in professional, scientific, and technical services if they locate in more innovative MSAs. The effect of entrepreneurship in a mediation test is statistically insignificant. In a moderation test, both innovation and entrepreneurship appear as significant predictors, but the coefficient of the latter is tiny. Equation (4) implies interaction between Patents and StartUpsAll. The positive effect of innovation on the number of expanding firms somewhat decreases in metropolitan areas with higher levels of entrepreneurship, but the magnitude of the decrease is negligible.

MSAs with higher levels of educational attainment, more diversified economies, greater numbers of workers per square mile, and greater GMP tend to have more expanding firms per 1,000 residents. Regions with the same attributes except for diversification create less new businesses. Diversity appears to promote firm formation in this sample.

The evidence in favor of the missing link hypothesis is even weaker if the number of jobs in business start-ups approximates regional economic performance (Table 5). In a test for mediation, innovation emerges as a strong predictor of firm creation in all industries (Equation 2). Firm creation, in turn, is somewhat related to job creation by new firms (Equation 3), although the coefficient is very small and significant only at the .1 level. A test for moderation does not support the MLH. Neither innovation, nor entrepreneurship is a significant predictor of the dependent variable. It is not surprising that the interaction term is also insignificant. Weakness or lack of a statistically meaningful relationship may be explained by the resource constraints usually faced by newly established firms. Looking for new knowledge and implementing it is a costly process. During the first years of operation, young businesses mainly struggle to survive (Geroski, 1995) and are likely to remain unreceptive to the opportunities their regional environment has to offer. After several years, provided a firm survives, it grows in its ability to capitalize on the knowledge available locally. Results reported in Table 5 implicitly support this supposition. The effect of the control variables is identical to Table 4.

This study suggests that overall level of entrepreneurship in an MSA plays a limited role in job creation by incumbent firms. Tables 3 and 4 imply that StUpsAll moderates the effect of knowledge creation on the number of jobs added by existing businesses and on the number of such companies. In the tests for moderation reported in these tables, entrepreneurship appears as a predictor of the dependent variables, although the coefficients are small. In addition, the negative interaction term suggests that contribution of entrepreneurship to job creation by the existing firms diminishes in MSAs with higher levels of knowledge creation. There is no evidence that StUpsAll mediate the effect of Patents on EmpExpand and FirmsExpand. The MLH is not supported when start-ups’ job-creating behavior is considered. Table 5 shows that the number of jobs added by newly established companies is not sensitive to the metropolitan innovative environment. The positive association between StUpsAll and EmpNew is significant at the .1 level only and the coefficient is infinitesimal. Both entrepreneurship and innovation are insignificant in a test for moderation. Their interaction is also insignificant. The explanation is likely to lie in the specific characteristics of the new firms. Such firms often face resource constraints that disable their capability to benefit from the knowledge environment in which businesses locate.

An alternative measure of entrepreneurship, the population-adjusted number of start-ups in NAICS54, may be a more valid operationalization for the purposes of testing the MLH in this industry. After all, entrepreneurs not related to professional, scientific, and technical services might have inadequate ability to recognize and use new knowledge relevant for job and firm creation in this industry. StUps is used as an approximation for entrepreneurship in Tables 6 to 8.

Table 6 reports estimation results for the models testing mediating and moderating role of entrepreneurship in the relationship between innovation and the number of new jobs in existing companies. The results are different from the ones presented in Table 3. Entrepreneurship appears to fully mediate the effect of innovation on the dependent variable, thus lending empirical support to the MLH. Patents is positively related to both incumbent firm expansion (Equation 1) and business formation in professional, scientific, and technical services (Equation 2a). Unlike Table 3, the coefficient of innovation in Equation (3a) becomes insignificant, whereas entrepreneurship emerges as a very strong predictor of firm expansion. Moderation effect is also present but estimation results suggest a different pattern of the relationship between the dependent and explanatory variables. Equation (4a) confirms the important role played by entrepreneurship in job creation by incumbent businesses. Innovation, to the contrary, has a negative effect that decreases as the level of entrepreneurship in a metropolitan area increases. The magnitude of the interaction term and the coefficient of Patents are approximately equal. It suggests that entrepreneurship effectively eliminates the possible negative effect of innovation on firm expansion.

Of all control variables only Diversity has the same effect in all four equations as the one reported by Table 3. It is positively related to firm formation in professional, scientific, and technical services, although the coefficient is remarkably smaller (Equation 2a). In all other equations this variable is not significant. The remaining controls usually have positive and statistically significant coefficients in all equations presented in Table 6. Educational attainment in an MSA is a strong predictor of job creation by incumbent companies. This confirms the results of Table 3. When it comes to firm formation in NAICS54, however, education has a positive effect, unlike the one reported for entrepreneurship in all industries. This result seems intuitive because of the nature of the NAICS54 industry, which relies on more educated workforce. Business density promotes firm expansion but is not statistically related to firm formation in professional, scientific, and technical services. Metropolitan areas with greater GMP tend to enjoy higher job creation by incumbent firms and greater entrepreneurship in NAICS54.

Table 7 suggests partial mediation of the relationship between metropolitan innovation and the number of expanding firms by the level of entrepreneurship in NAICS54 industry. According to the mediation test, Patents has a positive effect on the dependent variable in all three equations. This effect decreases in magnitude but preserves its statistical significance when StUps is included in the analysis. Similar to Table 6, entrepreneurship is also a moderator (Equation 4a). In this specification, innovation is negatively related to the number of expanding firms, but this effect rapidly decreases in more entrepreneurial areas. Estimation results for the control variables are consistent with Table 6. Education, industrial diversity, job density, and GMP of an MSA appear to increase the number of expanding businesses in the NAICS54 industry.

Table 8 exhibits results for the population-adjusted number of start-ups. In this case, patenting intensity in an MSA is positively related to firm formation in the industry, and firm formation predictably determines the number of new jobs added by newly established firms per 1,000 residents. One must be cautious in interpreting the estimated relationship as one of a mediating nature—it may be an artifact of the way the dependent and independent variables are operationalized in the model. The test for moderation depicts a pattern similar to Table 6. The number of start-ups has a strong positive effect on the number of jobs created by new firms. Innovation, to the contrary, is negatively related to the dependent variable. This negative effect is counteracted by the level of new firm formation in NAICS54, as the positive coefficient of the interaction term is larger than the negative coefficient of Patents (Equation 4a). The coefficients and significance level of the control variables are identical to Table 7.

This study shows that empirical evidence in favor of the MLH is stronger when the focus is on existing firms and when the number of start-ups in professional, scientific, and technical services approximates entrepreneurship. The table data in this study indicate that innovation influences the number of expanding firms and the number of jobs such firms create, as well as entrepreneurial level in the industry. The effect of Patents disappears or becomes smaller in magnitude when StUps is included in the models. It means that entrepreneurship fully or partially mediates the effect of innovation on job creation by incumbent firms. There is also evidence of moderation in the relationship between Patents, StUps, and the first two dependent variables.

Knowledge Spillover Theory of Entrepreneurship

Empirical Testing

The logic of the KSTE differs from that of the MLH. According to the KSTE, knowledge produced but unused in an area leads to increased firm formation in the knowledge-intensive industries or industries with some R&D component. In other words, assuming a constant share of new knowledge is left dormant, past values of innovation should lead to a greater number of new firms in the future. For the purpose of empirical testing, an ideal scenario would be to include as many lagged innovation measures as seems reasonable and see if the relationship between firm formation and innovation changes over time. Distributed lag models do exactly this. Unfortunately, this approach cannot be used because of the inertia exhibited by the metropolitan innovation measures over time. If lagged values of innovation level are included in the same equation, the regression fails to distinguish the effect of these variables and is likely to produce biased coefficients. To get around this problem, separate models are estimated for each lagged value of Patents. A change in the significance of the estimates as well as a change in their magnitude should indicate a support, or a lack of support, for the KSTE.

Equation (5) tests if the KSTE holds in the case of professional, scientific, and technical services in the U.S. metropolitan statistical areas during the 2001 to 2005 time period:

D e p V a r_{i t} = β_{1} + β_{2} P a t e n t s_{i t - T} + β_{3} X_{i t} + e_{i t}; T \in [0; 9],

where βs are regression coefficients; subscript i denotes an MSA, subscript t denotes a year, and T is the number of lags of the explanatory variable. X _it is a matrix of control variables (Education, Density, and GMP) and year dummies. The explanatory variable is a population-adjusted patent count in an MSA during a 5-year period. Ten such periods, from 1992 to 1996 to 2001 to 2005, are included in the models. If the KSTE holds, 10 years should be sufficient for the hypothesized effect to manifest itself in the data. Again, I use feasible generalized least square approach with panel-specific first-order autoregressive process and heterogeneous panels to estimate Equation (5).

Estimation Results and Discussion

The estimation results suggest that past patenting activity in the U.S. MSAs is not statistically related to the number of start-ups in professional, scientific, and technical services (Table 9). There is no evidence of a change in the relationship over time. This means that the KSTE is not supported in this study. This result goes somewhat in line with the findings of Knoben et al. (2011), who report a minor effect of the local knowledge base on firm formation in the Netherlands and show that this effect may be significantly overestimated by the analysis if it does not account properly for agglomeration.

Table 9.

Estimation Results for the KSTE.

Variable	2001-2005	2000-2004	1999-2003	1998-2002	1997-2001	1996-2000	1995-1999	1994-1998	1993-1997	1992-1996
Patents	−0.011 (0.011)	−0.003 (0.010)	−0.005 (0.011)	−0.002 (0.011)	−0.002 (0.011)	−0.002 (0.012)	−0.014 (0.013)	−0.011 (0.0137)	−0.011 (0.014)	−0.026 (0.016)
Education	2.134*** (0.066)	2.121*** (0.066)	2.132*** (0.066)	2.128*** (0.066)	2.122*** (0.066)	2.126*** (0.066)	2.136*** (0.066)	2.139*** (0.066)	2.141*** (0.066)	2.139*** (0.066)
GMP	0.001*** (0.000)	0.001*** (0.000)	0.001*** (0.000)	0.001*** (0.000)	0.001*** (0.000)	0.001*** (0.001)	0.001*** (0.000)	0.001*** (0.000)	0.001*** (0.000)	0.001*** (0.000)
Density	−0.010 (0.040)	−0.007 (0.040)	−0.008 (0.040)	−0.007 (0.040)	−0.007 (0.040)	−0.006 (0.040)	−0.010 (0.040)	−0.009 (0.040)	−0.010 (0.040)	−0.012 (0.040)
Year2002	0.253*** (0.006)	0.253*** (0.006)	0.253*** (0.006)	0.252*** (0.006)	0.253*** (0.006)	0.252*** (0.006)	0.253*** (0.006)	0.253*** (0.006)	0.253*** (0.006)	0.253*** (0.006)
Year2003	0.187*** (0.007)	0.187*** (0.007)	0.187*** (0.007)	0.187*** (0.007)	0.187*** (0.007)	0.187*** (0.007)	0.187*** (0.007)	0.187*** (0.007)	0.187*** (0.007)	0.188*** (0.007)
Year2004	−0.125*** (0.007)	−0.124*** (0.007)	−0.124*** (0.007)	−0.124*** (0.007)	−0.124*** (0.007)	−0.124*** (0.007)	−0.125*** (0.007)	−0.124*** (0.007)	−0.125*** (0.007)	−0.124*** (0.007)
Year2005	−0.024*** (0.008)	−0.023*** (0.008)	−0.024*** (0.008)	−0.024*** (0.008)	−0.023*** (0.008)	−0.024*** (0.008)	−0.024*** (0.008)	−0.024*** (0.008)	−0.024*** (0.008)	−0.023*** (0.008)
Constant	0.050 (0.015)	0.050 (0.015)	0.049 (0.015)	0.049 (0.015)	0.050 (0.015)	0.049 (0.015)	0.050 (0.015)	0.047 (0.015)	0.048 (0.015)	0.050 (0.015)
n	1,790
Groups	358
Wald χ²	6,402	6,373	6,372	6,346	6,349	6,327	6,435	6,445	6,454	6,535
Prob > χ²	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000

Note. KSTE = knowledge spillover theory of entrepreneurship; GMP = gross metropolitan product; DV = dependent variable. Time periods indicated refer to the variable Patents only; the 2001 to 2005 values of the DV and all control variables are used to estimate all equations. DV is the number of new NAICS54 firms per 1,000 residents.

***

Significant at .001 level. **Significant at .01 level. *Significant at .05 level.

Among the control variables, Education is positive and statistically significant. New firms belonging to the NAICS54 industry are more likely to be started in metropolitan areas with high levels of education, which seems logical. There is some evidence of a negative influence of industrial agglomeration, as job density is negatively related to the number of new firms, although the magnitude of the coefficient is negligible. The positive effect of GMP on the number of start-ups in the industry of interest is also negligible.

The fact that the KSTE should hold for the knowledge-intensive industries (professional, scientific, and technical services, in particular) follows from the stylized story behind this theory. It is nowhere explicitly stated in the theory itself. Acs et al. (2009) talk about firm formation in general. Table 10 presents an estimation of results identical to the previous model with the number of start-ups in all industries per 1,000 residents as the dependent variable.

Table 10.

Estimation Results for the KSTE.

Variable	2001-2005	2000-2004	1999-2003	1998-2002	1997-2001	1996-2000	1995-1999	1994-1998	1993-1997	1992-1996
Patents	8.071*** (1.291)	9.830*** (1.174)	8.833*** (1.359)	9.232*** (1.337)	13.820*** (1.581)	13.244*** (1.589)	12.381*** (1.529)	15.484*** (1.849)	14.660*** (1.661)	20.275*** (1.752)
Education	−35.827*** (5.247)	−32.966*** (4.930)	−37.487*** (5.249)	−37.410*** (5.141)	−43.471*** (5.383)	−42.634*** (5.085)	−39.128*** (5.257)	−43.315*** (5.411)	−45.707*** (5.377)	−42.368*** (5.330)
GMP	−0.003 (0.006)	−0.006 (0.006)	−0.005 (0.006)	−0.005 (0.006)	−0.004 (0.007)	−0.007 (0.007)	−0.009 (0.007)	−0.007 (0.007)	−0.010 (0.006)	−0.010 (0.006)
Density	−30.202*** (0.003)	−30.278*** (3.034)	−29.304*** (3.064)	−30.184*** (2.971)	−30.893*** (2.984)	−0.029*** (0.003)	−27.235*** (2.963)	−25.853*** (0.8508)	−0.024*** (0.003)	−0.027*** (0.003)
Year2002	−1.953*** (0.320)	−2.192*** (0.319)	−2.262*** (0.320)	−2.243*** (0.319)	−1.954*** (0.324)	−2.453*** (0.316)	−1.937*** (0.325)	−2.436*** (0.329)	−2.479*** (0.316)	−2.983*** (0.254)
Year2003	−6.876*** (0.428)	−7.111*** (0.426)	−7.293*** (0.428)	−7.404*** (0.4264)	−6.738*** (0.430)	−7.488*** (0.418)	−7.231*** (0.437)	−7.262*** (0.434)	−8.180*** (0.437)	−11.359*** (0.347)
Year2004	−4.080*** (0.504)	−4.409*** (0.499)	−4.480*** (0.499)	−4.797*** (0.497)	−4.277*** (0.501)	−4.890*** (0.485)	−4.442*** (0.506)	−4.923*** (0.507)	−5.024*** (0.491)	−8.687*** (0.414)
Year2005	0.709 (0.575)	0.535 (0.567)	0.314 (0.563)	0.133 (0.559)	0.273 (0.564)	−0.272 (0.548)	0.143 (0.570)	−0.266 (0.570)	−0.275 (0.560)	−1.945*** (0.478)
Constant	29.639 (1.268)	28.986 (1.210)	30.025 (1.221)	30.375 (1.192)	30.539 (1.202)	30.746 (1.106)	29.270 (1.225)	29.588 (1.176)	30.672 (1.114)	31.235 (1.175)
n	1,790
Groups	358
Wald χ²	840	919	896	932	909	991	839	868	1,021	2,233
Prob > χ²	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000	0.0000

Significant at .10 level. **Significant at .05 level. ***Significant at .01 level.

In stark contrast to the results reported in Table 9, Patents is a significant predictor of the total number of new firms in an MSA. The coefficient of the past values of explanatory variable is significant; moreover, it increases in magnitude as more distant past is brought in the model. It is debatable if this evidence supports the KSTE, which implies knowledge entrepreneurs. Most likely, the effect of knowledge creation on firm formation differs among industries, and the theory holds for some industries but not for others.

The coefficients of the control variables are noteworthy. The size of a metropolitan economy is negatively related to the population-adjusted total number of start-ups, but the magnitude of the coefficient is very small. Job density is also negatively related to the level of entrepreneurship. Perhaps the most interesting result of Table 10 is a large and negative coefficient of education, which indirectly supports Shane’s (2008) argument mentioned earlier. The argument states that the new firms in general are predominantly set up by people with low levels of educational attainment in the areas that lack alternative economic opportunities, such as corporate employment prospects.

In summary, this study does not find empirical support for the KSTE in the case of professional, scientific, and technical services in U.S. MSAs. Further research employing other industries should reveal the instances when the theory is supported. This information will be useful to delineate the boundaries of the KSTE in the U.S. context.

Conclusions

The purpose of this article is to test the MLH of entrepreneurship and the KSTE in the context of the U.S. MSAs. The MLH emphasizes the central role of entrepreneurs in the nexus between new knowledge creation and regional economic performance. The KSTE ascribes new firm formation in a region to the stock of unused knowledge produced locally. As policy implications of these two models differ, the empirical tests performed in this study should provide preliminary evidence useful for educated policy design.

The study results are mixed. Overall, the MLH is supported with some reservations. The analysis indicates an intervening role of entrepreneurs in the effect of metropolitan innovation on job-creating behavior of incumbent NAICS54 companies when entrepreneurship is measured by the number of start-ups in this industry. If the total number of new firms in all industries is used to quantify entrepreneurship, support for the missing link hypothesis is limited. There is no empirical backing for the missing link hypothesis when the number of new jobs in NAICS54 start-ups is the focus of the analysis. In this case, if entrepreneurship appears to intervene in the association between innovation and the dependent variable, these results are likely to be driven by the operationalization of the model.

This study fails to empirically verify the knowledge spillover theory of entrepreneurship. The number of start-ups in professional, scientific, and technical services in U.S. MSAs during the 2001 to 2005 period is not statistically related to the level of innovative activity going 10 years back. Contrary to the logic presented by the KSTE, prior levels of innovation do predict firm formation in general, and this effect increases in magnitude as the time gap between Patents and StUpsAll widens. Industry-specific differences in the relationship between innovation and business formation are likely to determine this result, inviting further empirical tests that involve other industries and time periods.

To summarize, the analysis arrives at the following conclusions. First, during the 2001 to 2005 period, incumbent companies in professional, scientific, and technical services were more likely to hire new employees and created more jobs if they located in metropolitan areas characterized by intensive patenting and high levels of NAICS54 entrepreneurship. Moreover, the moderation and mediation test results indicate that entrepreneurs in this industry were needed for the local knowledge to find its way into market applications as postulated by the missing link hypothesis. In other words, regions with favorable entrepreneurial climates tended to benefit more (in terms of new jobs added by incumbent NAICS54 firms) from locally generated knowledge. Second, higher levels of metropolitan innovation did not translate into higher levels of firm formation in professional, scientific, and technical services during the study period. This suggests that new knowledge per se is not sufficient to stimulate active firm entry in this industry. Third, the results of this study depend on the metric used to measure entrepreneurship. The study finds that job creation in the NAICS54 industry was sensitive to the presence of entrepreneurs in this industry, whereas no relationship between the dependent variables and entrepreneurship in general was established. Finally, an unexpected and important finding of this study is indirect corroboration of the skeptical view of American entrepreneurs.¹¹ Shane (2008) contends that an average entrepreneur in this country is a person who wants to work for himself, and contributes to neither growth nor innovation. According to this logic, geographic regions¹² with a less educated population and fewer employment opportunities should be more entrepreneurial because setting up a private firm might be the only viable option for many people to support their families (Storey, 1991). Notably, less educated MSAs created significantly more new firms in total but significantly less firms in NAICS54, professional, scientific, and technical services during the study period.

A preliminary policy take-away from this research is that innovative regions require entrepreneurs to create more jobs in the NAICS54 industry. Policies seeking to expand employment should promote entrepreneurial environment. Favorable entrepreneurial milieu may help incumbent firms to capitalize on the locally generated knowledge and to add new jobs. Further research spanning longer time periods and utilizing more resent data is needed, however, to establish generalizability of these conclusions and recommendations.

Footnotes

Acknowledgements

I sincerely thank Professor Deborah A. Strumsky for sharing her data and discussing early ideas for this research.

Declaration of Conflicting Interests

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

Funding

The author received no financial support for the research, authorship, and/or publication of this article.

Notes

Author Biography

Alexandra Tsvetkova is a postdoctoral researcher at the Department of Agricultural, Environmental, and Development Economics, The Ohio State University, where she studies the effects of energy sector on economic performance of urban and rural areas. Her broad research interests focus on the determinants of regional economic growth.

References

Acs

Z. J.

Anselin

Varga

(2002). Patents and innovation counts as measures of regional production of new knowledge. Research Policy, 31, 1069-1085.

Acs

Z. J.

Armington

(2004). Employment growth and entrepreneurial activity in cities. Regional Studies, 38, 911-927.

Acs

Z. J.

Audretsch

Braunerhjelm

Carlsson

(2012). Growth and entrepreneurship. Small Business Economics, 39, 289-300.

Acs

Z. J.

Autio

Szerb

(2012). National systems of entrepreneurship: Measurement issues and policy implications (George Mason University School of Public Policy Research Paper No. 2012-08). Retrieved from http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2008160

Acs

Z. J.

Braunerhjelm

Audretsch

Carlsson

(2009). The knowledge spillover theory of entrepreneurship. Small Business Economics, 32, 15-30.

Acs

Z. J.

Parsons

Tracy

(2008). High-impact firms: Gazelles revisited. Washington, DC: Small Business Administration Office of Advocacy.

Acs

Z. J.

Plummer

(2005). Penetrating the “knowledge filter” in regional economies. Annals of Regional Science, 39, 439-456.

Acs

Z. J.

Plummer

Sutter

(2009). Penetrating the knowledge filter in “rust belt” economies. Annals of Regional Science, 43, 989-1012.

Acs

Z. J.

Sanders

(2012). Patents, knowledge spillovers, and entrepreneurship. Small Business Economics, 39, 801-817.

10.

Acs

Z. J.

Szerb

(2012). Global Entrepreneurship and Development Index 2012. Chelthnam, England: Edward Elgar.

11.

Acs

Z. J.

Szerb

Ortega-Argiles

Coduras

Aidis

(2012). The Regional Entrepreneurship and Development Index (REDI): The case of Spain (George Mason University School of Public Policy Research Paper No. 2012-04 [revised]). Retrieved from http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1970009

12.

Aldrich

Kushmin

(1997). Rural economic development: What makes rural communities grow? Washington, DC: ERS.

13.

Almeida

Kogut

(1999). Localization of knowledge and the mobility of engineers in regional networks. Management Science, 45, 905-917.

14.

Armington

Acs

Z. J.

(2002). The determinants of regional variation in new firm formation. Regional Studies, 36(1), 33-45.

15.

Arrow

K. J.

(1962). Economic welfare and the allocation of resources for invention. In Nelson

R. R.

(Ed.), The rate and direction of inventive activity (pp. 609-626). Princeton, NJ: Princeton University Press.

16.

Attaran

(1986). Industrial diversity and economic performance in U.S. areas. Annals of Regional Science, 22(2), 44-54.

17.

Audretsch

(1995). Innovation and industry evolution. Cambridge: MIT Press.

18.

Audretsch

(2007). Entrepreneurship capital and economic growth. Oxford Review of Economic Policy, 23(1), 63-78.

19.

Audretsch

Keilbach

(2008). Resolving the knowledge paradox: Knowledge-spillover entrepreneurship and economic growth. Research Policy, 37, 1697-1705.

20.

Baron

R. M.

Kenny

D. A.

(1986). The moderator-mediator variable distinction in social psychological research: Conceptual, strategic and statistical considerations. Journal of Personality and Social Psychology, 51, 1173-1182.

21.

Bishop

Gripaios

(2007). Explaining spatial patterns of industrial diversity: An analysis of sub-regions in Great Britain. Urban Studies, 44, 1739-1757.

22.

Braunerhjelm

Acs

Z. J.

Audretsch

Carlsson

(2010). The missing link: Knowledge diffusion and entrepreneurship in endogenous growth. Small Business Economics, 34, 105-125.

23.

Breschi

Lissoni

(2001). Knowledge spillovers and local innovation systems: A critical survey. Industrial and Corporate Change, 10, 975-1005.

24.

Feldman

M. P.

Audretsch

(1999). Innovation in cities: Science-based diversity, specialization and localized competition. European Economic Review, 43, 409-429.

25.

Feser

E. J.

(2002). Tracing the sources of local external economies. Urban Studies, 39, 2485-2506.

26.

Frenken

Oort

F. V.

Verburg

(2007). Related variety, unrelated variety and regional economic growth. Regional Studies, 41, 685-697.

27.

Gallini

(2002). The economics of patents: Lessons from recent U.S. patent reform. Journal of Economic Perspectives, 16, 131-154.

28.

Geroski

P. A.

(1995). What do we know about entry? International Journal of Industrial Organization, 13, 421-440.

29.

Griliches

(1979). Issues in assessing the contribution of R&D to productivity growth. Bell Journal of Economics, 10, 92-116.

30.

Griliches

(1990). Patent statistics as economic indicators: A survey. Journal of Economic Literature, 28, 1661-1707.

31.

Jacobs

(1969). The economy of cities. New York, NY: Random House.

32.

Knoben

Ponds

Oort

F. V.

(2011). Employment from new firm formation in the Netherlands: Agglomeration economies and the knowledge spillover theory of entrepreneurship. Entrepreneurship & Regional Development: An International Journal, 23, 135-157.

33.

Marshall

(1920). Principles of economics. London, England: Macmillan. (Original work published 1890)

34.

Martin

Ottaviano

(2001). Growth and agglomeration. International Economic Review, 42, 947-968.

35.

Melo

Graham

Noland

(2009). A meta-analysis of estimates of urban agglomeration economies. Regional Science and Urban Economics, 39, 332-342.

36.

Monchuk

Miranowski

Hayes

Babcock

(2007). An analysis of regional economic growth in the U.S. Midwest. Review of Agricultural Economics, 29(1), 17-39.

37.

Pakes

Griliches

(1980). Patents and R&D at the firm level: A first report. Economics Letters, 5, 377-381.

38.

Partridge

Rickman

(2003). Do we know economic development when we see it? Review of Regional Studies, 33(1), 17-39.

39.

Praag

C. M. V.

Versloot

P. H.

(2007). What is the value of entrepreneurship? A review of recent research. Small Business Economics, 29, 351-382.

40.

Qian

Acs

Z. J.

(2013). An absorptive capacity theory of knowledge spillover entrepreneurship. Small Business Economics, 40, 185-197.

41.

Qian

Acs

Z. J.

Stough

(2013). Regional systems of entrepreneurship: The nexus of human capital, knowledge and new firm formation. Journal of Economic Geography, 13(4) 559-587. doi:10.1093/jeg/lbs009

42.

Rodriguez-Pose

Ezcurra

(2011). Is fiscal decentralization harmful for economic growth? Evidence from the OECD countries. Journal of Economic Geography, 11, 619-643.

43.

Romer

P. M.

(1990). Endogenous technological change. Journal of Political Economy, 98(5), S71-S102.

44.

Rosenthal

S. S.

Strange

(2003). Geography, industrial organization, and agglomeration. Review of Economic and Statistics, 85, 377-393.

45.

Rosenthal

S. S.

Strange

(2004). Evidence on the nature and sources of agglomerated economies. In Henderson

J. V.

Thisse

J.-F.

(Eds.), Handbook of regional and urban economics, Volume 4: Cities and geography (pp. 2119-2166). Amsterdam, Netherlands: Elsevier.

46.

Shane

S. A.

(2008). The illusions of entrepreneurship: The costly myths that entrepreneurs, investors, and policy makers live by. New Haven, CT: Yale University Press.

47.

Solow

R. M.

(1956). A contribution to the theory of economic growth. Quarterly Journal of Economics, 70(1), 65-94.

48.

Storey

D. J.

(1991). The birth of new firms: Does unemployment matter? A review of the evidence. Small Business Economics, 3, 167-178.

49.

van Stel

A. J.

Carree

Thurik

(2005). The effect of entrepreneurial activity on national economic growth. Small Business Economics, 24, 311-321.

50.

van Stel

A. J.

Nieuwenhuijsen

H. R.

(2004). Knowledge spillovers and economic growth: An analysis using data of Dutch regions in the period 1987-1995. Regional Studies, 38, 393-407.

51.

Wal

A. T.

Boschma

(2009). Applying social network analysis in economic geography: Framing some key analytic issues. Annals of Regional Science, 43, 739-756.

52.

Well

(2007). Accounting for the effects of health on economic growth. Quarterly Journal of Economics, 122, 1265-1306.