Professional Documents
Culture Documents
An International Journal
To cite this article: Brian E. Whitacre, Devon Meadowcroft & Roberto Gallardo (2019)
Firm and regional economic outcomes associated with a new, broad measure of
business innovation, Entrepreneurship & Regional Development, 31:9-10, 930-952, DOI:
10.1080/08985626.2019.1630486
1. Introduction
Innovation has long been hailed as an important component of business success (Heunks 1996;
Srinivandasan et al. 2009). Measurable differences in innovation have been found between large and
small businesses (Acs and Audretsch 1988) and rural and urban firms (Keeble 1993). Additionally,
a growing body of evidence has examined whether innovative activity ‘spills over’ to nearby regions,
and to what extent such activity is associated with economic growth (Fritsch and Franke 2004;
Rodriguez-Pose and Crescenzi 2008; McCann and Simonen 2005). Although most empirical studies
focused on innovation use data from the European Union (EU), U.S. federal policy efforts have
embraced the idea that innovation can improve economic outcomes. For example, the department
of Housing and Urban Development’s (HUD) Rural Innovation Fund, the Economic Development
Administration’s (EDA) Regional Innovation Strategies, and the now defunct Department of
Commerce’s National Innovation Marketplace all promote innovation as a way to increase job oppor-
tunities and improve quality of life. However, there are no commonly accepted criteria for being
‘innovative,’ and little U.S.-based research has explored the regional-level economic outcomes asso-
ciated with business innovation. Further, most studies in this area use a diluted measure of innovation –
focused on patent activity or research dollars spent – and potentially ignore the role of more home-
grown solutions or opportunity-recognizing business activity. This paper contributes to the existing
body of evidence in this field by developing a new, broader measure of innovation and assessing its
relationship with both firm and regional economic activity in the U.S. Such a measure is likely more
relevant to the many small businesses across the country that lack a formal research and development
agenda.
The objectives of this study are: (1) to determine what business-level outcomes are associated with
a new, broad measure of innovation and (2) to explore regional-level economic outcomes that may be
associated with this updated measure of business innovation. The paper primarily uses data from the
restricted-access Rural Establishment Innovation Survey (REIS), which was organized and distributed by
the USDA’s Economic Research Service (ERS) in 2014. The REIS contains a sample of over 10,000 businesses
with at least 5 employees, with about 74% coming from non-metropolitan counties. The survey asks
detailed questions about business practices, including many activities that might be considered innova-
tive but that have not been available to use in prior innovation metrics (for example, how often business
processes are changed in order to address customer complaints). This study uses factor analysis on 15
innovation-related survey questions to create a broad ‘innovation index’ that should more accurately
capture the variety of tools used by U.S. businesses to introduce new goods or services.
For business-level outcomes, our approach follows a simple linear model where innovation is
directly related to output. Regression is used to determine the effect of the new measure of firm
innovation on company sales, hourly wages paid, and benefits offered. Control variables include
the size of the firm, industry category (by 2-digit North American Industry Classification System
(NAICS) code), regional location (by Census region), and participation in government programs
such as technology assistance or worker training efforts. The independent variable of interest in
these models is the innovation index, and an interaction term is used to assess whether the impact
of innovation varies by metropolitan status.
The regional analysis follows the recent literature on knowledge spillovers, and constructs aggregate
innovation indices for newly defined commuting zones (CZs), which were designed by ERS to represent
local economies and typically contain 3–6 counties. County-level economic variables such as unemploy-
ment rates, median household income, creative class employment, and poverty rates were also aggre-
gated to the CZ level using population weights by county. Multiple techniques are used to assess the
relationship between the innovation index and economic variables, including simple correlations, bivari-
ate Moran’s I measures, and spatial regressions that allow estimation of spillover effects.
The paper proceeds with a literature review focused on linkages between innovation, small businesses,
and regional development. The following section details the data and methodology used, including
construction of the innovation index that serves as the independent variable of interest in the business-
and regional-level regressions. This section also introduces the CZs and presents the rationale for their use.
The results section then provides descriptive statistics for all variables in the analysis and interprets the
findings of our primary specifications. The paper concludes with a discussion of the limitations of this
analysis, areas for future research, and policy implications of the results.
2. Literature review
2.1. A brief overview of innovation and its measurement
Innovation is a complex process with a long history in the economic literature. Perhaps the most
famous conceptualization of economic innovation is Schumpeter’s ‘creative destruction’ model,
which argued that the engine of capitalism was constant change through new commodities, new
technologies, new sources of supply, and new types of organizations that incessantly replace the
old (Schumpeter 1942). This work built on Schumpeter’s earlier (1934) treatise on economic
development that differentiated between product innovation (defined as the introduction of
a new good) and process innovation (defined as the introduction of a new method of production).
Many studies since then have compared and contrasted these two categories of innovations
(Utterback and Abernathy 1975; Fritsch and Meschede 2001; Becker and Egger 2013); though
Simonetti, Archibugi, and Evangelista (1995) argue that over 95% of all innovations fall into
a grey zone depicting a combination of the two groups.
The economics literature has also delved into how innovation happens, and has constructed
several competing theories for how innovation contributes to firm-level outputs and larger-scale,
932 B. E. WHITACRE ET AL.
regional development. Early work posited a simple linear model with a direct relationship between
applied research, innovations, and greater growth (Maclaurin 1953; Godin 2005). Along this line,
Roper, Du, and Love (2008) use data from Ireland to model the ‘innovation value chain,’ and find
strong complementarity between public and private (i.e. firm-level) sources of knowledge. They
find that both forms (public and private) contribute to product and process innovation, but
determine that public sources have only an indirect impact. Alternatively, the ‘chain-linked
model’ by Kline and Rosenburg (1986) emphasizes the numerous feedback mechanisms associated
with innovation and distinguish between early and late-stage research. Other theoretical models
attempting to uncover the impact of innovation on broader economic outcomes include regional
innovation systems (Asheim and Isaksen 2002) and knowledge spillovers (Jaffe 1986; Audretsch and
Feldman 1996). These are reviewed in more detail in section 2.2, including a discussion of how the
REIS data used here fits in.
One of the most challenging aspects of studying innovation is measuring it. The OECD (1992,
1996, 2005) has now constructed three editions of the ‘Oslo Manual’ with guidelines for gathering
and using innovation data. Similar to other studies on the topic, this manual attempts to distin-
guish between product and process innovations (and other categories, such as marketing or
organizational innovations). While several pages of the manual are focused on the components
and coverage of innovation activities, it recognizes that ‘it is generally difficult to ask for quanti-
tative measures of the effects of innovations in surveys (p. 109).’ The Oslo manual emphasizes
breaking down innovation expenditures by type of activity, and utilizing appropriate survey
techniques to collect data on innovation-related activities. The UK developed its own innovation
index based by building on the recommendations of the Oslo manual and other efforts (National
Endowment for Science, Technology and the Arts (NESTA) 2009). Their quantification broke
innovation investment into seven categories, and found that the aggregate investment was
responsible for two-thirds of productivity growth between 2000 and 2007.
Traditionally, many studies have proxied for innovation by using data on patent applications or
research and development expenditures by firms or regions. However, Wojan and Parker (2017)
argue that by relying heavily on this type of measurement, rural firms may be penalized. They
contend that grassroots or user-developed innovation may be more important to rural firms, but
note the historical difficulty in developing a more comprehensive measure via surveys. This
limitation led to the development of the survey used in this paper, the 2014 REIS, which captures
self-reported innovation among firms in rural areas of the United States.
(2011) generally find that innovativeness is directly related to firm value. Likewise, management
scholars argue that innovation capability is the most important determinant of firm performance
(Mone, McKinley, and Barker 1998). When the discussion turns to innovation’s impacts on larger
regions, however, the findings are more nuanced. Studies using both the regional innovation
systems approach and the knowledge spillover approach have found links between innovation and
advancements in regional economic development and expansion (Ahlstrom 2010; Fagerberg and
Srholec 2008; Cornett 2009; Pezeshkan et al. 2016; Albert 2017; Asheim and Coenen 2005). Findings
of this nature have been tied to arguments calling for regional policy to promote innovation
through fostering entrepreneurship and R&D (Landabaso 1997; Michael and Pearce 2009).
However, such findings are not universal. Shearmur and Bonnet (2011) use Canadian data to
argue that there is essentially no link between local innovation (measured by patent applications)
and local development. Similarly, Gonzalez, Jung, and Pena (2015) find only a very limited
connection between knowledge and innovation-driven entrepreneurship in less developed
countries.
One additional area of interest for our study is the discrepancy in innovation between rural and urban
areas. Keeble (1993) found that firms in metropolitan areas have higher growth rates and produce more
original innovative products than their non-metropolitan counterparts. Likewise, Isaksen and Onsager
(2010) show that new firms that are involved in knowledge intensive, innovative industries are more likely
to form in urban areas over rural locations. Funding for innovation projects in rural areas is also a problem.
One study in Europe (Bonfiglio et al. 2017) discovered that funds allocated for revitalizing rural develop-
ment through innovation oftentimes did not make it to their intended audiences due to policy disputes.
This finding is even more troubling in light of the finding in Esparcia (2014), who emphasized that plans to
encourage business innovation in non-metropolitan locations will likely not succeed if outside funding is
not made available. Wojan and Parker (2017), who were the first to use the REIS data built upon here, also
find that ‘substantive innovators’ were more likely to exist in urban areas. However, Wojan and Parker
demonstrate that innovation rates are actually very similar between urban and rural manufacturing
establishments, but are notably lower for rural firms in the service sector.
Most of the above studies use European data, and there is little empirical work on this topic for
U.S. firms or areas. The current paper addresses this gap, and assesses linkages between innovation
and firm/regional outcomes using specific theoretical models discussed above. In particular, our
firm-level approach follows the simple linear model that assumes a direct relationship between
innovation and firm outcomes, while our regional model uses a knowledge spillover framework to
assess the extent to which high-innovating regions impact their neighbors. Wojan and Parker
(2017) provide the only analysis of this topic for the rural U.S., and find that innovation intensity
positively impacted rural employment and wage growth during the economic recovery of
2010–2014. However, their methods for defining innovation and assessing its regional impact
vary from the approach used here. Further, while the recent literature has focused heavily on
whether spillover effects from innovation exist – that is, whether surrounding areas benefit from
high levels of innovation – most of these studies use data from Europe (Fritsch and Franke 2004;
McCann and Simonen 2005; Asheim and Coenen 2005). Notably, such studies were likely influential
in the EU’s development of the ‘Smart Specialisation Agenda,’ which emphasizes targeted innova-
tion support and developing region-specific strategies (European Commission 2014). Putting the
focus solely on the U.S., and incorporating a broader measure of innovation, will aid in making
meaningful policy recommendations that can be tailored to domestic businesses and regions. This
includes exploring whether innovation is more or less important for businesses in rural locations,
and the extent to which spillovers exist for areas surrounding an innovation center. The existence
of such spillovers in a U.S. context would provide ammunition for policy advocates who might
argue that policies promoting innovation impact more than just the areas receiving innovation-
related programs or funding.
934 B. E. WHITACRE ET AL.
representing more innovative businesses. A benefit of this index is that businesses across different
industries and geographies can be compared to one another, despite significant characteristic
differences. Another benefit is that the index does not use an arbitrary cutoff point to define which
businesses are innovative; rather, it is a continuous measure.
Hourlywage is a continuous variable denoting the average wage paid to all employees
(wages over $300/hour were excluded, since some respondents apparently entered annual
salaries as opposed to hourly wages). Growingmarket is a binary variable that takes the value
1 if the current market for the business’s goods/services is growing and 0 otherwise. Finally,
Employeebenefitssum is an ordinal variable from 0 to 6 based on the types and number of
benefits provided to employees. Such benefits include retirement plans, whether businesses
cover the costs of worker training, and paid maternity leave. While none of these variables is
arguably ‘better’ than more traditional measures of success such as profit, they do provide
a variety of views into how the business is being run. All three variables are designated as the
dependent variables in regression analyses with a series of business characteristics serving as
the independent variables. We follow a traditional linear model to estimate this relationship,
with the hypothesis that the innovation index has a direct impact on the outcome measures.
Thus, the basic regressions take the form:
yi ¼ Xi β þ Ii γ þ ðNM Xi Þθ þ ðNM Ii Þδ þ εi (1)
where yi is one of the three business-level economic outcome measures listed above; Xi is a vector of
control variables such as number or type of employees, NAICS code, and self-reported economic vitality
of the local community; Ii is the primary variable of interest (the innovation index); NM is a dummy variable
for whether or not the business is located in a non-metropolitan county; β; γ; θ; and δ are associated
parameter vectors; and εi is the associated error term, all for business i. A key feature of these models is
that the interaction terms created (NM Xi , NM Ii ) will allow for testing of whether the specific business
characteristics that affect business’ economic outcomes change in rural locations. Given the varying
nature of the dependent variables, the final specifications will include simple ordinary least squares
(hourlywage), logit (growingmarket), and ordered logit (employeebenefitssum) models.
We are up front that our specification does not firmly establish a direction of causality. It is
conceivable (and even probable) that higher wages or growing markets may drive firms to be more
innovative, rather than vice-versa. However, the survey does not provide a suitable measure that
could serve as an instrument for the innovation index, and thus our analysis is intended to function
as preliminary evidence regarding whether the relationship is positive or negative.
observations to construct a representative index score for a county would be highly susceptible to
sampling error. Thus, the counties were mapped to substate-level commuting zones (CZs) com-
piled by ERS, which were created ‘to more closely reflect the local economy where people live and
work’ (Economic Research Service (ERS) 2016).6 This achieves the goal of finding a functional
geography that allows for aggregating innovation scores. Each CZ contains a core city, and REIS
respondents likely draw from CZ assets as they run their businesses (i.e. for labor, input purchases,
and sales territory). CZs with fewer than 4 observations were removed out of concern that the
resulting estimates would be highly susceptible to sampling bias. There are 706 CZs in the
continental U.S., and aggregate innovation indices can be compiled in 581 of them (82%).
To model the relationship between innovation and regional-level economic outcomes, county-
level socioeconomic variables were compiled from the U.S. Census.7 County population weights
were then used to construct aggregate-level CZ measures (i.e. by weighted average). Variables
compiled include unemployment rates, median household income, poverty rates, rural-urban
continuum code (RUCC), natural amenity rankings, % white/Hispanic/African-American, % with
bachelors or graduate degrees, farming/manufacturing/mining dependency, % in specific age
categories, % employed in the creative class, and % with access to 25 megabits per second
broadband (as defined by the 2014 National Broadband Map). Simple correlations and bivariate
Moran’s I8 calculations are used to demonstrate whether innovation and four distinct regional
economic outcome variables (median household income, unemployment rates, poverty rates, and
% creative class) are positively or negatively related. These variables were chosen based on their
importance in documenting the economic health of a region (World Bank 2014), and are commonly
used in the regional science literature (Florida 2014; McGranahan, Wojan, and Lambert 2011;
Partridge, Rickman, and Li 2008; Partridge and Rickman 2003).
Moving beyond these simple bivariate tools, a spatial econometric approach is used to model
the relationship between innovation and the four regional-level outcome variables. In particular,
a spatial Durbin error model (SDEM) controls for spatial heterogeneity in the data while also
allowing the user to uncover direct and spillover effects of specific variables of interest (innovation,
in this case). This approach follows the knowledge spillover framework discussed in the literature
review. The formal model is:
yi ¼ Xi β þ WXi θ þ ui (2)
ui ¼ λWui þ εi (3)
where yi is one of the four regional-level economic outcome measures listed above; Xi is a vector of
control variables such as the percentage of residents in specific age categories, rates of education
attainment, racial and ethnic composition, degree of rurality (as measured by the rural-urban
continuum code), various types of economic dependence as defined by ERS (government, manu-
facturing, mining, or farming),9 broadband availability, a ranking of natural amenities, and the
innovation index; W is a spatial weights matrix (a queen contiguity matrix is used in practice); β; θ;
and λ are associated parameter vectors; and εi is the associated zero-mean error term, all for
commuting zone i.
Several recent papers have noted the benefits of this form of spatial specification, in particular that the
direct (β) and spillover (θ) effects of each variable are easy to identify (Vega and Elhorst 2013; LeSage 2014).
Further, while other specifications such as the spatial lag or combined lag/error might allow for a more
generalized spatial approach, there is concern about overparameterization and the difficult justification of
global (as opposed to local) spillovers required for spatial lag and combined lag/error models (Anselin
2003; Lacombe and LeSage 2012). The SDEM approach, alternatively, only includes local spillovers which
are more appropriate for most regional development variables of interest (LeSage 2014). Again, we note
that this approach cannot determine the direction of causality, but does allow for isolation of the
relationship between regional-level innovation and specific economic outcome measures. We also note
ENTREPRENEURSHIP & REGIONAL DEVELOPMENT 937
that restricting our analysis to CZs with at least 4 observations results in a spatial weight matrix with
‘holes’ – i.e. missing neighbors for some zones. We argue that this does not significantly impact the results
since only a relatively small portion (18%) of CZs lack innovation measures (Bivand and Portnov 2004), but
perform a robustness check using distance-based spatial weights (and ensuring all CZs have neighbors).
4. Results
4.1. Factor analysis results and subsequent innovation index
As section 3.1 discussed, factor analysis was used to reduce the large number of variables potentially
related to innovation into more workable groups that deal with a specific feature. As such, the technique
identified the underlying variable(s) that quantified a specific aspect of innovation. Four specific factors
were identified, consisting of underlying variables ranging from one to six in number (Table 1). Each of
the factors is given a name that relates to the underlying variables it includes. Variables with less than 0.3
factor loadings were excluded from a factor, leaving only the variables shown.10
Factor 1 is named ‘Innovative Accomplishments’ and contains six variables that relate to
businesses actively producing new goods and services and engaging in innovation-related activ-
ities. Factor 2, ‘Intellectual Property,’ contains two variables that focus on the business’ participation
in patents, copyrights, trademarks, and non-disclosure agreements. Factor 3 has only a single factor
relating to the sources that businesses gather their information from, and is named ‘Innovation
Sources.’ Finally, factor 4 is named ‘Customer Focus’ and contains two variables that describe the
effect that customer feedback has on businesses practices.
Once the four factors were constructed and estimated for each observation, they were com-
bined to create the innovation index. To make the analysis more straight-forward, the innovation
index was normalized so that it would take values from 0 to 1. Thus, a higher innovation index
score indicated a higher level of general innovation. Building off of our simple linear innovation
model, we view each of the four factors as being components of basic research. As such, each of
the four factors was weighted evenly. Figure 1 shows that most (92%) of the businesses in the
sample had an innovation index ranking of less than 0.6. The average value was 0.426, and a score
of 0.545 would move a business into the 75th percentile.
The variables that made up the four factors were examined to assess how the individual
component means varied by innovation index quintile. The findings are presented in Table 2. In
general, as the innovation index increased, the component variable means increased. For example,
the percentage of firms who sold new goods earlier than their competitors (sellnewgoodearly) was
6000
5,064
(47%)
5000
4,355
(41%)
Number of Respondents
4000
3000
2000
717
1000
527 (7%)
(5%)
75
(1%)
0
Less than 0.2 0.2 ≤ I.I. < 0.4 0.4 ≤ I.I. < 0.6 0.6 ≤ I.I. < 0.8 0.8 ≤ I.I. ≤ 1.0
Ranges for Innovation Index
only 6% for firms with scores in the lowest quintile but rose to 94% for firms with scores in the
highest quintile. Sales from new goods (newgoods%sales) accounted for only 2.3% of all sales for
firms in the lowest innovation index quintile, but were over 34% for firms in the highest. All
variables included in the factor analysis monotonically increase across the quintiles, indicating that
the index captures higher levels of each underlying contributor and providing reassurance that it is
in fact measuring overall ‘innovation.’
Another item of interest is how the innovation index breaks out across industries (2-digit NAICS
codes). Table 3 shows that only the information industry has statistically higher levels of innovation
when compared to the national average, while others (mining, transportation) score significantly
lower. Table 4 takes the discrepancy between metro and non-metro a step further and displays
average innovation scores across the 9-category rural-urban continuum codes (or Beale codes). The
results reinforce that innovativeness is negatively associated with the degree of rurality. In particular,
a sizable drop-off is noted between categories 3 and 4 (note that codes 1–3 are for metropolitan
counties, and 4–9 and for non-metropolitan). While none of the three metropolitan codes has an
index score that is statistically different from the national average, each of the six non-metropolitan
codes is statistically lower. These findings generally agree with those in Wojan and Parker (2017).
Table 5. (Continued).
Hourlywage Growingmarket Employeebenefitssum
Dependent Variable (1) (2) (3) (4) (5) (6)
R-squared 0.14 0.11
/cut1 0.09 0.69 **
/cut2 1.37 *** 1.92 ***
/cut3 2.56 *** 3.10 ***
/cut4 3.96 *** 4.40 ***
/cut5 5.55 *** 6.04 ***
/cut6 7.16 *** 7.73 ***
*, **, and *** represent statistical significance at the p < 0.10, 0.05, and 0.01 levels, respectively.
wages are smaller in non-metro areas. Nm*i-index was positive and significant for logit model (4),
indicating that the impact of a high innovation index score is larger for non-metropolitan businesses –
that is, the probability of a firm being involved in a growing market is higher for a specific innovation index
score if the firm is located in a non-metro area.
The other parameter results generally follow expectations, such as the positive impact on wages
for the percentage of employees in management (pmgmt) and the negative impact for the percent
employed in production (pprod). Wages for firms in the Northeast and West were higher than the
default region of the Midwest, and the number of benefits offered was lower for firms in the South.
Several NAICS categories were shown to impact wages (relative to the default of NAICS 71 – arts
and entertainment), although only NAICS 54 (Professional/technical service) was consistently
positive across the three main dependent variables. Interestingly, area attractiveness (areaattract)
and the economic vitality (economyvit) of a region are only significant for the growingmarket
specifications – and are negative. Note, however, that the non-metro interaction term for area
attractiveness is a positive shift that counteracts the original parameters. There is a marginally
statistically significant impact on participation in government programs (govprograms1) for the
number of benefits offered; otherwise no impact of government programs is recorded.
These findings suggest that innovation is highly important for the firm-level outcomes being con-
sidered. The marginal effect of having a high level of innovation was among the highest of all factors
considered for each of the three dependent variables. It is also true that the impact of innovation varies by
non-metropolitan status: the effect on hourly wages is lower, but it increases the likelihood of being in
a growing market.
Moran’s I = 0.597
B. E. WHITACRE ET AL.
MHI Poverty
Moran’s I = 0.466 Moran’s I = 0.491
Figure 2. Maps and Moran’s I values for innovation index, unemployment, median household income, and poverty.
ENTREPRENEURSHIP & REGIONAL DEVELOPMENT 943
(a) Innovation Index & % Creative Class (b) Innovation Index & Unemployment Rates
r = .284 r = -.084
(c) Innovation Index and Median Household Income (d) Innovation Index & Poverty Rates
r = .259 r = -.221
Figure 3. Correlation coefficients and scatterplots – Innovation Index and regional economic variables of interest(a) Innovation index &
% creative class (b) Innovation index & unemployment rates (c) Innovation index and median household income (d) Innovation index &
poverty rates.
p-values less than 0.01.13 However, these simple statistics do not control for any other variables
such as education or industrial composition, so the analysis moves to spatial regressions.
The SDEM regression results in Table 6 indicate that after controlling for other potentially
influential factors, high levels of innovation have a direct relationship with three of the economic
outcome variables of interest: median household income, poverty levels, and creative class employ-
ment. The income finding is consistent with Wojan and Parker (2017), who found that wage growth
in zones with high levels of nominal innovators were 3% higher during 2010–2014 when compared
to similar zones with non-innovators. However, the only metric that displays evidence of spillovers
from innovation is creative class employment. The W*i-index parameter is only statistically sig-
nificant here, suggesting that CZs can benefit from high levels of innovation in neighboring areas.
Other control variables in the four models generally behave as the literature would suggest. This
includes higher education levels being associated with greater income and creative class employ-
ment (but lower poverty and unemployment rates); higher proportions of residents age 40–60
implying higher income levels but lower poverty levels; and percentage of African-Americans being
associated with lower incomes but higher levels of poverty and unemployment. Economic depen-
dency categories (as defined by ERS) vary in importance across the outcome measures, with CZs
containing counties that are dependent on farming, mining, and manufacturing all seeing higher
income levels and lower poverty rates. Interestingly, broadband availability seems to be important
for 3 of the 4 measures, with access to speeds of at least 25 megabytes per second (MBPS) being
associated with greater income and creative class employment, and lower levels of poverty. The
944 B. E. WHITACRE ET AL.
(a) Innovation Index & % Creative Class (b) Innovation Index & Unemployment Rates
I = .139 I = .010
<.001 =.332
(c) Innovation Index and Median Household Income (d) Innovation Index & Poverty Rates
I = .125
I = -.118
<.001
<.001
Figure 4. Bivariate Moran’s I scatterplots(a) Innovation index & % creative class(b) Innovation index & unemployment rates (c)
Innovation index and median household income(d) Innovation index & poverty rates.
spatial error parameter (λ) is always significant at the p < 0.01 level, and maps/Moran’s I values of
the residuals suggests that the SDEM specification appropriately dealt with the spatial nature of the
data (i.e. there is little evidence of spatial clustering in the residuals). Additional regressions using
nearest neighbor or distance-based spatial weights (to remove observations with no neighbors)
showed no meaningful variation from the results shown here.
5. Conclusion
The aim of this study was to determine if a new, broad measure of business innovation has meaningful
relationships with the economic well-being of U.S. firms and regions. Using a dataset composed of mainly
the Rural Establishment Innovation Survey (REIS), an innovation index is created using factor analysis and
then incorporated into regression models to assess the connection between innovation and firm- and
regional-level outcomes. At the firm level, regression models building on the simple linear theory find that
innovation has a positive relationship with the wages paid to employees, the growth of the market for the
firm’s primary product, and the number of benefits offered to employees. These results provide some
backing for the linear model, particularly since our new measure of innovation is significantly broader than
most previous studies. This is consistent with Balconi, Brusoni, and Orsenigo (2010) argument that the
linear model is still relevant. When an interaction term between the innovation index and a binary variable
ENTREPRENEURSHIP & REGIONAL DEVELOPMENT 945
indicating non-metropolitan status is included, both negative and positive effects are found. Innovation’s
relationship with the hourly wage paid to employees is smaller in non-metropolitan counties (although
still positive overall); conversely, innovation’s effect on the probability of a firm’s products being in
a growing market is greater for firms in non-metropolitan counties.
With respect to the CZ level regional analyses, simple correlations show that innovation is positively
associated with several measures of economic outcomes; the relationship holds after control variables are
included. It is important to note that some degree of spatial autocorrelation appears to exist for
innovation: a map of the mean innovation index scores across CZs demonstrates a statistically significant
Moran’s I coefficient. Correlations are seen between innovation and the four measures of regional
economic prosperity, and are confirmed with bivariate Moran’s I estimates. The SDEM models (which
follow a knowledge spillover framework) find that innovation has a positive relationship with the median
household income and the percentage of employees in the creative class, and a negative relationship with
poverty. There is evidence that spillovers can exist from innovation, as CZs that are neighbors to areas with
high levels of innovation appear to experience higher levels of creative class employees than would be
expected. This result adds a U.S. context to the existing body of literature examining innovation spillovers
(Asheim and Coenen 2005; Cornett 2009; Albert 2017). Community and regional developers should be
aware of the wide variety of positive economic outcomes associated with a broad measure of firm-level
innovation.
There are several limitations worth mentioning. The primary one, noted throughout, is that no firm
statements can be made about the direction of causality. The REIS data is purely cross-sectional, so teasing
out causal relationships in the firm-level models by following specific firms over time was not possible.
Future research should test whether innovation is truly an exogenous explanatory variable, or if firm/
economic well-being instead act as a driver of innovation. To deal with this possible endogeneity, a two-
946 B. E. WHITACRE ET AL.
stage least squares (2SLS) approach might be implemented using an appropriate instrument. A proper
instrument in this case would be one that is associated with innovation, but not the economic outcome
measures. Additionally, while our construction of the index assumes that the four underlying factors
represent distinct approaches to innovation, an alternative framework might consider that one or more
factors serve as precursors to another. We leave this as an avenue for future research. A final limitation is
that the firm-level models only deal with three specific output measures available in the REIS – all of which
are self-reported. Other outcome measures, such as overall profitability or employee satisfaction, would be
useful.
Even with these limitations, the analysis has important policy implications across two general
dimensions. First, the types of innovation supported by federal, state, and local initiatives should
not focus solely on traditional avenues such as formal research and development or patent-
oriented programs. Formal U.S. innovation policy is quite diverse, but major programs such as
the Small Business Innovation Research (SBIR) dominate expenditures at the national level (Shapira
and Youtie 2010). Such programs often emphasize technology development and use, which is
commonly tied to customary types of innovation activity. This research has shown that defining
innovation in a broader context (with activities more typical of many small businesses) is still
associated with positive impacts for both firm-level and regional-level economic measures. Thus,
future policies should consider how to promote several of the specific factors in the ‘innovation
index.’ For example, incentivizing firms to develop new sources of information (Factor 3 in Table 1),
or implement methods for assessing/using customer satisfaction (Factor 4) are two explicit ways
that could promote the non-traditional version of innovation espoused here. Regional policy
efforts, including state-level initiatives, may be more effective at helping businesses develop
these more comprehensive innovation-oriented procedures.
Second, the research here provides insights about policy efforts in rural versus urban areas. The firm-
level results suggest that urban areas benefit more from innovation in terms of wages paid (similar to the
finding in Rodriguez-Pose and Crescenzi 2008); however, a rural advantage exists for market growth.
Therefore, regional innovation policies that are geared towards improving wages may find more success
in more urban locations, while those focused on growing a specific market might accomplish more in
relatively peripheral areas. This is not to suggest that rural or urban innovation policies should prioritize
one outcome over another, rather that policymakers should be aware of the potentially differing impacts
across geographies. Finally, the regional-level analysis suggests that innovation spillovers exist for creative
employment, which has been shown to be particularly relevant for rural economic growth (McGranahan,
Wojan, and Lambert 2011). Thus, innovation investments in core locations may see benefits accrue to the
surrounding rural areas. However, our results suggest that expectations for such spillover effects should be
limited, given the lack of meaningful relationships with nearby income levels, poverty rates, or unemploy-
ment rates.
Notes
1. Tradable industries are defined as those in the mining (NAICS code 21), manufacturing (31–33), wholesale
trade (42), transportation and warehousing (48), information (51), finance and insurance (52), professional/
technical services (54), management of businesses (55), and arts/entertainment (71).
2. This attempt at measuring innovation with the REIS survey differs from the methodology used by Wojan and
Parker (2017), who differentiate between 3 categories of innovators (non/nominal/substantive) by using latent
class analysis on 8 specific REIS questions.
3. A reviewer notes concern that some factors may be considered consequences of, rather than inputs to,
innovation. However, the wording of each specific question generally connote the business’ approach to
innovation. Further, we view the specific items noted in each question as potential inputs to the simple linear
innovation model, where they are seen as components of basic research. As such, we believe that they are
appropriate to use as variables in the factor analysis.
4. As noted in Kaiser (1974), a KMO score in the 0.90’s is deemed ‘marvelous’, a score in the 0.80’s is ‘meritorious’,
a score in the 0.70’s is ‘middling’, a score in the 0.60’s is ‘mediocre’, a score in the 0.50’s is ‘miserable’, and
a score below 0.50 is ‘unacceptable.’
ENTREPRENEURSHIP & REGIONAL DEVELOPMENT 947
5. The ‘predict’ command in STATA performs this task. Although most of our variables are ordinal, we did not use
a polychoric correlation matrix to smooth our variables. This is because the bias associated with using ordinal
variables in PCA has been shown to be small with symmetric and unimodal data (Olsson 1979; Rigdon and
Ferguson 1991). 12 of our 15 variables have skewness coefficients between −0.5 and 0.5. Our large sample size
(10,000 observations) also reduces concerns about bias.
6. Wojan and Parker (2017) also use commuting zones for their analysis.
7. 2011–2015 5-year American Community Survey (ACS) data was used to match the 2014 Business Innovation
Survey.
8. The bivariate Moran’s I calculates the relationship between one variable and neighboring values for a different
variable. It also ranges from −1 to 1, with a positive and significant relationship implying that high values of
variable 1 are surrounded by high values for variable 2.
9. ERS defines these types of counties based on the industry’s percentage of total earnings or employment. The
percentages vary by industry (ERS, 2017).
10. Low factor loadings indicate that the variable is not highly correlated with that particular factor. Note that
all of the factor loadings in Table 2 are > 0.3. The four questions listed in Appendix A that were excluded
from the final factors were Q28 (ABANDONED), Q34 (EXTRACASH), Q38 (RECESSION) and Q39
(INCREASE13).
11. Moran’s I is a measure from −1 to 1 with 0 representing a purely random spatial distribution and 1
representing complete spatial autocorrelation (where a county’s value depends entirely on its neighbors).
12. USDA-ERS (2014) defines the creative class as those employed in ‘creative’ occupations, specifically occupa-
tions ‘developing, designing, or creating new applications, ideas, relationships, systems, or products, including
artistic contributions.’
13. See Footnote 8 for a definition of the bivariate Moran’s I. The positive and significant values here suggest that
regions with high (low) values of variable 1 are surrounded by high (low) values of variable 2.
Disclosure statement
No potential conflict of interest was reported by the authors.
Funding
This work was supported by the Economic Research Service [5464-MSU-USDA-0095].
References
Acs, Z., and D. Audretsch. 1988. “Innovation in Large and Small Firms: An Empirical Analysis.” The American Economic
Review 78 (4): 678–690.
Ahlstrom, D. 2010. “Innovation and Growth: How Business Contributes to Society.” Academyof Management Journal 24
(3): 11–24.
Albert, M. 2017. “Entrepreneurship, Innovation, and Regional Performance: Application for the Spanish Regions.”
Entrepreneurship & Regional Development 29 (3–4): 271–291. doi:10.1080/08985626.2016.1267805.
Anselin, L. 2003. “Spatial Externalities, Spatial Multipliers, and Spatial Econometrics.” International Regional Science
Review 26 (2): 153–166. doi:10.1177/0160017602250972.
Asheim, B. T., and A. Isaksen. 2002. “Regional Innovation Systems: The Integration of Local ‘Sticky’and Global
‘Ubiquitous’ Knowledge.” The Journal of Technology Transfer 27 (1): 77–86. doi:10.1023/A:1013100704794.
Asheim, B. T., and L. Coenen. 2005. “Knowledge Bases and Regional Innovation Systems: Comparing Nordic Clusters.”
Research Policy 24: 1173–1190. doi:10.1016/j.respol.2005.03.013.
Audretsch, D. B., and M. P. Feldman. 1996. “R&D Spillovers and the Geography of Innovation and Production.” The
American Economic Review 86 (3): 630–640.
Balconi, M., S. Brusoni, and L. Orsenigo. 2010. “In Defence of the Linear Model: An Essay.” Research Policy 39 (1): 1–13.
doi:10.1016/j.respol.2009.09.013.
Becker, S. O., and P. H. Egger. 2013. “Endogenous Product versus Process Innovation and a Firm’s Propensity to
Export.” Empirical Economics 44 (1): 329–354. doi:10.1007/s00181-009-0322-6.
Bivand, R., and B. Portnov. 2004. “Exploring Spatial Data Techniques Using R: The Case of Observations with No
Neighbours.” In Advances in Spatial Econometrics: Methodology, Tools, and Applications, (pp. 121-142). Springer,
Berlin, Heidelberg..
948 B. E. WHITACRE ET AL.
Bonfiglio, A., B. Camaioni, S. Coderoni, R. Esposti, F. Pagliacci, and F. Sotte. 2017. “Are Rural Regions Prioritizing
Knowledge Transfer and Innovation? Evidence from Rural Development Policy Expenditure across the EU Space.”
Journal of Rural Studies 53: 78–87. doi:10.1016/j.jrurstud.2017.05.005.
Cornett, A. P. 2009. “Aims and Strategies in Regional Innovation and Growth Policy: A Danish Perspective.”
Entrepreneurship & Regional Development 21 (4): 399–420. doi:10.1080/08985620903020078.
Crowley, F., and J. Bourke. 2018. “The Influence of the Manager on Firm Innovation in Emerging Economies.”
International Journal of Innovation Management 22 (3): 1850028. doi:10.1142/S1363919618500287.
ERS (Economic Research Service). 2016. “Commuting Zones and Labor Market Areas.” https://www.ers.usda.gov/data-
products/commuting-zones-and-labor-market-areas/
Esparcia, J. 2014. “Innovation and Networks in Rural Areas. An Analysis from EuropeanInnovative Projects.” Journal of
Rural Studies 34: 1–14. doi:10.1016/j.jrurstud.2013.12.004.
European Commission. 2014. Smart Specialisation and Europe’s Growth Agenda. Publications Office of the European
Union. Publications Office of the European Union. http://ec.europa.eu/regional_policy/sources/docgener/studies/
pdf/smart_spec_growth_agenda.pdf
Fagerberg, J., and M. Srholec. 2008. “National Innovation Systems, Capabilities, and EconomicDevelopment.” Research
Policy 37: 1417–1435. doi:10.1016/j.respol.2008.06.003.
Florida, R. 2014. “The Creative Class and Economic Development.” Economic Development Quarterly 28 (3): 196–205.
doi:10.1177/0891242414541693.
Fritsch, M., and G. Franke. 2004. “Innovation, Regional Knowledge Spillovers and R&D Cooperation.” Research Policy 33:
245–255. doi:10.1016/S0048-7333(03)00123-9.
Fritsch, M., and M. Meschede. 2001. “Product Innovation, Process Innovation, and Size.” Review of Industrial
Organization 19 (3): 335–350. doi:10.1023/A:1011856020135.
Fruchter, B. 1954. Introduction to Factor Analysis. NY: D. Van Nostrand .
Godin, B. 2005. “The Linear Model of Innovation: The Historical Construction of an Analytical Framework.” Working
paper. http://www.csiic.ca/PDF/Godin_30.pdf
Gonzalez, J. L., A. Jung, and I. Pena. 2015. “Innovation-Driven Entrepreneurship in Developing Economies.”
Entrepreneurship & Regional Development 27 (9–10): 555–573. doi:10.1080/08985626.2015.1075602.
Heunks, F. J. 1996. “Innovation, Creativity, and Success.” Small Business Economics 10: 263–272. doi:10.1023/
A:1007968217565.
Isaksen, A., and K. Onsager. 2010. “Regions, Networks, and Innovative Performance: The Case of Knowledge-Intensive
Industries in Norway.” European Urban and Regional Studies 17 (3): 227–243. doi:10.1177/0969776409356217.
Jaffe, A. 1986. “Technological Opportunity and Spillovers of R&D: Evidence from Firms’ Patents, Profits, and Market
Share.” American Economic Review 76: 984–1001.
Kaiser, H. 1974. “An Index of Factor Simplicity.” Psychometrika 39: 31–36. doi:10.1007/BF02291575.
Keeble, D. 1993. “Small Firm Creation, Innovation, and Growth and the Urban-Rural Shift.” In Small Firms in Urban and
Rural Locations, edited by J. Curran and D. Storey, 54–78. London: Routledge.
Kline, S.J., and N. Rosenberg, . 1986. The Positive Sum Strategy: Harnessing Technology for Economic Growth. USA: The
National Academy of Science.
Lacombe, D., and J. LeSage. 2012. “Using Bayesian Posterior Model Probabilities to Identify Omitted Variables in
Spatial Regression Models.” Available at SSRN: http://ssrn.com/abstract=2174610
Landabaso, M. 1997. “The Promotion of Innovation in Regional Policy: Proposals for a Regional Innovation Strategy.”
Entrepreneurship & Regional Development 9 (1): 1–24. doi:10.1080/08985629700000001.
LeSage, J. 2014. “What Regional Scientists Need to Know about Spatial Econometrics.” The Review of Regional Studies
44 (1): 13–32.
Maclaurin, W. 1953. “The Sequence from Invention to Innovation and Its Relation to Economic Growth.” Quarterly
Journal of Economics 67: 97–111. doi:10.2307/1884150.
McCann, P., and J. Simonen. 2005. ““Innovation, Knowledge Spillovers and Local Labor Markets.” Papers in Regional
Science 84 (3): 465–485. doi:10.1111/j.1435-5957.2005.00036.x.
McGranahan, D., D. Wojan, and D. Lambert. 2011. “The Rural Growth Trifecta: Outdoor Amenities, Creative Class, and
Entrepreneurial Context.” Journal of Economic Geography 11 (3): 529–557. doi:10.1093/jeg/lbq007.
Michael, S. C., and J. A. Pearce II. 2009. “The Need for Innovation as a Rationale for Government Involvement in
Entrepreneurship.” Entrpreneurship & RegionalDevelopment 21 (3): 285–302.
Mone, M. A., W. McKinley, and V. L. Barker. 1998. “Organizational Decline and Innovation: A Contingency Framework.”
Academy of Management Review 23: 115–132. doi:10.5465/amr.1998.192965.
Moore, D., K. Miller, and D. Vakoch. 2015. “National Survey of Business Competitiveness.” Washington State University
Social & Economic Sciences Research Center Data Report, 14–56.
Morgan, K. 1997. “The Learning Region: Institutions, Innovation, and Regional Renewal.” Regional Studies 31: 491–503.
doi:10.1080/00343409750132289.
NESTA(National Endowment for Science, Technology and the Arts). 2009. “The Innovation Index: Measuring the UK’s
Investment in Innovation and Its Effects.” https://www.nesta.org.uk/sites/default/files/innovation_index_2009.pdf
ENTREPRENEURSHIP & REGIONAL DEVELOPMENT 949
OECD (Organization for Economic Cooperation and Development). 1992, 1996, 2005. Oslo Manual: Guidelines for
Collecting and Interpreting Innovation Data.
Olsson, U. 1979. “On the Robustness of Factor Analysis against Crude Classification of the Observations.” Mutivariate
Behavioral Research 14: 485–500. doi:10.1207/s15327906mbr1404_7.
Partridge, M., and D. Rickman. 2003. “Do We Know Economic Development When We See It?.” The Review of Regional
Studies 33 (1): 17–39.
Partridge, M., D. Rickman, and H. Li. 2008. “Who Wins from Local Economic Development? A Supply Decomposition of
U.S. County Employment Growth.” Economic Development Quarterly 23 (1): 13–27. doi:10.1177/0891242408322314.
Pezeshkan, A., A. Smith, S. Fainshmidt, and A. Sedeh. 2016. “National Business Systems and FirmInnovation: A Study of
Developing Economies.” Journal of Business Research 69: 5413–5418. doi:10.1016/j.jbusres.2016.04.147.
Rigdon, E., and C. Ferguson. 1991. “The Performance of the Polychoric Correlation Coefficient and Selected Fitting
Functions in Confirmatory Factor Analysis with Ordinal Data.” Journal of Marketing Research 28 (4): 491–497.
doi:10.1177/002224379102800412.
Rodriguez-Pose, A., and R. Crescenzi. 2008. “Research and Development, Spillovers, Innovation Systems, and the
Genesis of Regional Growth in Europe.” Regional Studies 42 (1): 51–67. doi:10.1080/00343400701654186.
Roper, S., C. Hales, J. Bryson, and J. Love 2009. “Measuring Sectoral Innovation Capability in Nine Areas of the UK
Economy.” Report for NESTA Innovation Index project. https://www.nesta.org.uk/sites/default/files/measuring-
sectoral-innovation.pdf
Roper, S., J. Du, and J. Love. 2008. “Modelling the Innovation Value Chain.” Research Policy 37 (6–7): 961–977.
doi:10.1016/j.respol.2008.04.005.
Rosenbusch, N., J. Brinckmann, and A. Bausch. 2011. “Is Innovation Always Beneficial? A Meta-Analysis of the
Relationship between Innovation and Performance in SMEs.” Journal of Business Venturing 26 (4): 441–457.
doi:10.1016/j.jbusvent.2009.12.002.
Rubera, G., and A. H. Kirca. 2012. “Firm Innovativeness and Its Performance Outcomes: A Meta-Analytic Review and
Theoretical Integration.” Journal of Marketing 76 (3): 130–147. doi:10.1509/jm.10.0494.
Schumpeter, J. A. 1934. The Theory of Economic Development. Cambridge, MA: Harvard University Press. Original
German Edition 1911.
Schumpeter, J. A. 1942. Capitalism, Socialism and Democracy. London: Routledge.
Shapira, P., and J. Youtie. 2010. “The Innovation System and Innovation Policy in the United States.” In Competing for
Global Innovation Leadership: Innovation Systems and Policies in the USA, EU, and Asia, edited by Frietsch and
Schuller, 5–29. Stuttgart: Fraunhofer IRB Verlag.
Shearmur, R., and N. Bonnet. 2011. “Does Local Technological Innovation Lead to Local Development? A Policy
Perspective.” Regional Science Policy and Practice 3 (3): 249–270. doi:10.1111/j.1757-7802.2011.01040.x.
Simonetti, R., D. Archibugi, and R. Evangelista. 1995. “Product and Process Innovations: How are They Defined? How
are They Quantified?.” Scientometrics 32 (1): 77–89. doi:10.1007/BF02020190.
Sonn, J., and M. Storper. 2008. “The Increasing Importance of Geographical Proximity in Technological Innovation: An
Analysis of U.S. Patent Citations 1975-1997.” Environment and Planning A 40 (5): 1020–1039. doi:10.1068/a3930.
Srinivandasan, S., K. Pauwels, J. Silva-Risso, and D. M. Hanssens. 2009. “Product Innovations, Advertising, and Stock
Returns.” Journal of Marketing 73 (1): 24–43. doi:10.1509/jmkg.73.1.24.
U.S. Department of Agriculture. 2014 Creative Class County Codes. Economic Research Service, October 2017. https://
www.ers.usda.gov/data-products/creative-class-county-codes/documentation/
Utterback, J. M., and W. J. Abernathy. 1975. “A Dynamic Model of Process and Product Innovation.” Omega 3 (6):
639–656. doi:10.1016/0305-0483(75)90068-7.
Vega, S., and J. P. Elhorst. 2013. “On Spatial Econometric Models, Spillover Effects, and W.” Paper presented at the 53rd
ERSA Conference, Palermo.
Wojan, T., and T. Parker. 2017. Innovation in the Rural Nonfarm Economy: Its Effect on Job and Earnings Growth, 2010-
2014. ERR-238, U.S. Department of Agriculture, Economic Research Service.
World Bank. 2014. World Development Indicators. Washington, DC: World Bank. doi:10.1596/978- 1-4648-0163-1
950 B. E. WHITACRE ET AL.
Means
Index Component Variable Name Mean Min Max Metro Non-Metro
Sources of new information Newinfo 9.22 0 18 9.29 8.82 ***
Monitor customer satisfaction Satisfaction 1.18 0 2 1.19 1.11 ***
Impact of customer complaints on business practices Complaintimpact 1.49 0 2 1.49 1.46 *
New goods or services in the past three years Newgoods 3.25 0 6 3.33 2.82 ***
Abandoned or incomplete innovation activities Abandoned 0.56 0 2 0.58 0.44 ***
Improved goods or services in 2013 Improvedgoods 2.58 0 5 2.61 2.37 ***
Sell new goods early or before competitors Sellnewgoodearly 0.55 0 1 0.56 0.50 ***
Percentage of sales through new or improved goods Newgoods%sales 14.71 −5 100 15.02 12.67 ***
Engage in innovation-related activities Innovactivities 3.71 0 10 3.80 3.15 ***
If extra cash is spent on innovation or investments Extracash 2.32 0 4 2.32 2.30
If business has participated in patent applications Patentapp 0.08 0 1 0.08 0.05 ***
If this business has any trademarks, copyrights, or NDA’s Trademarks 0.62 0 4 0.66 0.35 ***
Innovation during economic recession Recession −0.15 −1 1 −0.14 −0.17
Has innovation increased since 2013 Increase13 0.23 −1 1 0.24 0.16 ***
Growth in the past three years Improv3yrs 4.08 0 9 4.13 3.80 ***
*, **, and *** denote statistical differences between the metro and non-metro means at the p < .10, .05, and .01 levels,
respectively.