-
An Examination of the Geography of Innovation
MARYANN P. FELDMAN (Economics and Management, Goucher College,
1021 Dulaney Valley Road,
Baltimore, MD 21204, USA)
This research examines spatial patterns of manufacturing product
innovation and provides a model of factors contributing to
geographic concentration of innovation. The model maintains that
innovation is related to concentrations of innovative inputs,
including: university R&D, industrial R&D, the presence of
related industry and the presence of specialized business services.
Empirical estimation reveals evidence of the importance of
geographically mediated spillovers.
1. Introduction
While there is general agreement that the rate of technical
change is important in determining an economy's rate of growth, we
have a limited understand-ing of the sources of technical progress
and of why the pace of progress varies so significantly over time
and space (Lucas, 1993). One finding of the new growth theories is
that divergence in growth rates may be a result of
'" increasing returns to knowledge (Romer, 1986). This is
relevant for innova-~ tive activity because perhaps more than other
economic activities, innovation '" depends on knowledge. The
ability to produce commercially viable product ] e . innovation
relies on scientific and technical knowledge coupled with knowl-i
edge of the market. As a result, geographic concentrations of
knowledge are N likely to create higher levels of innovation than
would otherwise be achieved. ] o This paper provides an empirical
investigation of the location of innova-> ~ tion, and seeks to
contribute to our knowledge of product innovation along .. j two
dimensions. The first purpose is to develop a model of the spatial
\)
~ dimensions of innovation within a broader institutional
understanding of the S innovation process. This conceptualization
relates the location of innovation l to the presence of the various
knowledge bases which contribute to the 1 process of
commercializing technology. The second purpose is to test empiri--§
..!S © Oxford University Press 1993 ------------ 451
at University of North Carolina at Chapel Hill on July 30,
2014http://icc.oxfordjournals.org/Downloaded from
http://icc.oxfordjournals.org/
-
------ An Examination of the Geography of Innovation
cally this conceptualization introducing a measure of innovative
output to the discussion of the location of innovative activity.
The data are introduced and the spatial patterns of innovation are
discussed. The estimation of an innovation production function
reveals that innovation is found to cluster geographically in areas
which contain concentrations of specialized resources which enhance
and facilitate the innovation process.
2 . Why Location Matters for Innovative Activity
Innovation is expected to exhibit strong geographic clustering
because the process of bringing new products to market relies on
specific scientific and technical knowledge. Innovation has a
cumulative character which builds on existing expertise. Early in
the literature, Machlup (1962, pp. 161-163) pointed out that the
process of solving a technical problem often raises new research
questions and positions existing innovators to continue to
innovate. An area with innovative activity will develop a set of
specialized resources which provide comparative advantage for the
next round of innovation. This process is defined by Arthur (1990)
as self-reinforcing expertise and gives rise to the geographic
clustering of innovative activity.
The ability to produce innovative output is determined by a set
of knowledge capabilities or inputs which are not confined to the
organizational boundaries of the firm (Nelson and Winter, 1982).
Knowledge inputs can be embodied in human, institutional and
facility form, and will be relatively immobile (Tassey, 1991). As
knowledge in a product category or industry develops, it becomes
cumulative and non-transferable or place specific (SPRU, 1972;
Thomas, 1985; Dosi, 1988a; Lundvall, 1988; Grossman and Helpman,
1989).
Commercialization requires firms to organize the diverse types
of comple-mentary knowledge which facilitate the innovation process
(Teece, 1986). Firms may be expected to internalize knowledge
sources up to the point at which external transactions become
advantageous. If knowledge sources are too costly, too specialized
or otherwise constrained from becoming part of the firm, then
external transaction may be preferred. Geographic concen-trations
of resources, known as agglomeration economies, may be
concep-tualized as similar to the economies of scope realized by
complementary activities in large organizations. Proximity of
complementary activities promotes information transfer which lower
the cost and reduce the risks associated with innovation. In other
words, innovative activity benefits from the presence of a variety
of resource and knowledge inputs embedded in a socially-constructed
and spatially-delimited setting.
Dosi provides five 'stylized facts' or characteristics of the
innovation
-------------- 452 --------------
at University of North Carolina at Chapel Hill on July 30,
2014http://icc.oxfordjournals.org/Downloaded from
http://icc.oxfordjournals.org/
-
------ An Examination of the Geography of Innovation ------
process which are instructive in considering why location may
benefit inno-vative activity (Dosi, 1988a, pp. 222-223). The
stylized facts are: the uncertainty of the innovation process, the
reliance on university research, the complexity of the innovation
process, the importance of learning by doing, and the cumulative
character of innovative activity. Each of these characteris-tics is
considered in turn.
Innovation can be viewed as highly uncertain. This uncertainty
extends beyond the lack of information about anticipated events to
include the existence of previously undefined scientific and
technical problems. One means to reduce uncertainty in the
innovation process is to participate in information exchanges to
keep a company at the cutting edge of a technology and to
facilitate problem solving tasks. Innovative networks can be
interpreted as the formation of research communities in which firms
join to exploit new developments in an industry in a timely manner
(Nelson, 1990). To the extent that location promotes timely
information exchange, innovation will be enhanced. The cost of
membership in innovative networks is a reciprocal sharing of
information which creates a de facto market for these transactions.
The importance of networking for innovation in specific industries
within geographic areas has been documented by Saxanien (1990) for
Silicon Valley and Powell (1989) for the biotechnology
industryl.
The uncertainty involved in using a new technology provides an
incentive for firms to locate together (Lundvall, 1988, p. 355).
When technology is standardized and reasonably stable, information
exchange may be translated into standard codes. In this case, long
distance transmission of information can take place at low costs.
On the other hand, when technology is complex and evolving rapidly,
long distance standardized transmission is not possible. Therefore,
location close to the source of the technology allows firms to
translate information into useable knowledge, creating an incentive
for firms using complex and dynamic technologies to locate near
knowledge sources.
A geographic concentration of rival firms may provide a
knowledge resource to reduce innovation's uncertainty. Von Hippel
(1988) finds that reciprocal information trading between rival
firms provides an important innovative input. Allen (1983) suggests
evidence of the geographic nature of information trading among
rival firms in the nineteenth century English steel industry. A
geographic concentration of rival firms appears to facilitate
networking and problem solving, and advance the state of knowledge
in the industry (Porter, 1990). As the presence of an industry
expands in a given location, firms can specialize in the production
of complementary products, and provide expertise to enhance
solution searches and reduce uncertainty.
1 Freeman (1991) provides a review of studies on the importance
of networking for innovarive activity.
------------ 453
at University of North Carolina at Chapel Hill on July 30,
2014http://icc.oxfordjournals.org/Downloaded from
http://icc.oxfordjournals.org/
-
------ An Examination of the Geography of Innovation ------
Universities have become important to the innovation process.
Universities emphasize the free exchange and flow of information
and their existence in an area creates a sort of intellectual
commons. In contrast to the notion that knowledge is a public good
easily transferred, via publications, gaining commercial control
over a new technology requires access to individuals who can turn
information into knowledge (Nelson, 1989). An example of the
importance of face-to-face interaction is provided by a survey of
biotechnology researchers by Grefsheim et al. (1991). This work
uncovered that the most important and timely information comes from
personal communications which provide information far in advance of
printed sources. Interviewed researchers also felt that the
stylistic limitations of formal papers limited their substantive
usefulness. Specifically, 'Formal papers do not contain the
experimenter's strategies and perspectives, nor can they convey
what the experimenter thinks the work means and how it dovetails
with or contradicts other work' (Grefsheim et al., 1991, p. 41).
While it cannot be disputed that academic conferences and
long-distance consulting arrangements provide a means for
information dissemination, such contact is less frequent, more
costly and qualitatively different.
The complexity of innovative activity increases the scope of the
activities needed to complete the innovation process. To manage
this complexity, innovators must conduct intricate search
procedures across a variety of disciplines to find specific
information. Within each discipline searched, the source of
information will be highly specialized. The limited usefulness of
this information on a day-to-day basis favors external
transactions.
The increased scope of innovative activity is suggested by the
increased prominence of business services. These services provide
information about consumer demand and help shepherd new product
innovation through a maze of regulations and product
specifications. The specialized services of patent attorneys,
market research and feasibility studies, and commercial testing
laboratories are beyond the means of all except the largest
corpora-tions to internalize. Survey work by MacPherson (1991)
found that the intensity of the usage of external producer services
correlates highly with realized product innovations in medical and
chemical firms. Most importantly, since producer services exist
solely to supply information, these firms tend to locate near their
clients (Coffey and Polese, 1987).
There is a certain element of serendipity in the search for
relevant informa-tion. Shimshoni (1966) argues that the larger the
number of skills and interests represented in a given geographical
area, the greater the probability of encounters which may lead to
fruitful information exchanges. Firms located in areas with a range
of information sources to enhance the innovation process, will
realize lower search costs in obtaining relevant information.
------------- 454 ------------
at University of North Carolina at Chapel Hill on July 30,
2014http://icc.oxfordjournals.org/Downloaded from
http://icc.oxfordjournals.org/
-
------ An Examination of the Geography of Innovation ------
Some aspects of knowledge have a tacit nature which cannot be
completely codified and transferred through blueprints and
instructions. This knowl-edge is learned through practice and
practical example (Nelson and Winter, 1982). Experimentation in the
form of learning by, doing and learning by using play an important
role in innovation. Such expertise can come from a variety of
sources in related industries. It may be generated by product
buyers as they provide information about their needs to enhance
product design and development (Von Hippel, 1988). This expertise
may be facili-tated by input suppliers who disseminate technical
information which, in turn, facilitates new product innovation
(Cohen et at., 1987). In addition, competitors, who face the same
obstacles and bottlenecks, can be an impor-tant source of tacit
information (Von Hippel, 1988; Porter, 1990).
Recent work by Carlson and Jacobsson (1991) suggests that the
market for new technologies is primarily regional. The development
of technologically complex products requires close collaboration
between suppliers and custom-ers. Until a product becomes
standardized, constant specification and design changes make it too
costly for suppliers to get involved with distant customers.
The five characteristics presented here characterize innovation
as a process reliant on timely flows of external information. The
uncertainty of the innovation process suggests that one way in
which firms can reduce uncer-tainty is by engaging in reciprocal
sharing of information or networking with related firms. The
prominence of university research argues for proximity to this
innovative input to stay at the cutting edge of technology. The
increased complexity of innovation suggests that other sources of
information such as related industry presence and specialized
business services are key to innova-tive success. These specialized
information sources tend to locate near their client markets.
Finally, the cumulative nature of innovative activity suggests that
areas with demonstrated innovative success have assembled
information to facilitate the next round of innovation. Firms
located in areas with limited access to information inputs must
rely on their own internal efforts and will face higher costs in
acquiring information (Davelaar and Nijkamp, 1989; Brody and
Florida, 1991). Firms may attempt to internalize these knowledge
sources by hiring skilled individuals with relevant expertise but
external transactions may be less costly. For the above reasons,
innovation is expected to exhibit pronounced geographic clustering.
This will be explored in the next section.
3. Spatial Patterns of Innovation
A fundamental obstacle to more systematic analysis of the effect
of location on innovative activity has been the lack of a good
measure of innovative ------------ 455 ------------
at University of North Carolina at Chapel Hill on July 30,
2014http://icc.oxfordjournals.org/Downloaded from
http://icc.oxfordjournals.org/
-
An Examination of the Geography of Innovation
output2 . In the absense of direct measures of innovation,
alternatives such as the amount of inputs used in the innovative
process, for example, R&D expenditures (Malecki, 1985), or
output measures such as the quantity of patented inventions Oaffe,
1989) have been used. R&D expenditures indicate resources
budgeted towards producing innovative output, not the success of
these efforts. The reliability of patent data has been questioned
because not all patented inventions prove to be commercially viable
and many successful innovations are not patented (Mansfield,
1991).
This research uses a direct measure of innovative output to
explore the location of innovative activity. In 1982, the Small
Business Administration (SBA) conducted a census of innovation
citations from over 100 scientific and trade journals. The SBA data
captures innovations which, by nature of the citation, added new
economically useful knowledge to a product category3. In contrast
to patent data, which marks the certification of a new invention,
innovation citations announce the market introduction of a
commercially viable product4 • The SBA data contains a total of
4200 manufacturing product innovations with information on the
location of the enterprise which introduced the innovations.
Innovations are attributed to states in which the establishment
responsible for the major development of the innovation was
located6 .
2 Griliches (1990, p. 1669) states that 'The dream of getting
hold of an output indicator of inventive activity is one of the
strong motivating forces for economic research in this area.'
, This data was analyzed in considerable detail by Acs and
Audretsch (1988, 1990). 4 The innovation citation data are highly
correlated with other measures of innovative inputs. R&D
expenditures are from the National Science Foundation as
reported by Jaffe (1989). Patent counts by state are also from
Jaffe (1989) and represent the average annual number of patents
received in 29 states over an 8 year period. High-technology
employment data are from the US Office of Technology Assessment
(1984) for the year 1982.
Correlations Among Measures of Innovative Activity
Innovation PatentJ R&D Employment
Innovation 1.0000 Patents 0.9344 1.0000 R&D 0.8551 0.8804
1.0000 Employment 0.9737 0.9888 0.7013 1.000
5 Information as to the economic significance or the revenue
generated by each innovation is not available. An example of the
demands of this type of determination is ptovided by Trajtenberg
(1990) in an extensive study of CT scanners.
6 Since headquarters may announce new product innovations, the
SBA data differentiates between the location of the innovating
establishment and the location of the innovating entity (Edwards
and Gordon, 1984). The parent company or headquarters is known as
the entity. For example, Intel Corporation introduced a 16-Bit
Micro-Controller (Model Number 8096). The major development was
done by a division of Intel in Arizona. Intel is headquartered in
California. In this case, the state of the establish-ment is
Arizona and the state of the entity is California: the innovation
was attributed to the state of Arizona. The site of the major
development of the innovation is known as the establishment. The
state identifier of the establishment is used to investigate the
spatial patterns of innovation.
----------- 456
at University of North Carolina at Chapel Hill on July 30,
2014http://icc.oxfordjournals.org/Downloaded from
http://icc.oxfordjournals.org/
-
------ An Examination of the Geography of Innovation ------
There are some limitations and potential sources of bias to
note. The data are only available as a cross-section for the year
1982. The innovation citations which appear in publications may be
biased towards unusual or specific items which the editors consider
to be of special interest. Large firms with public relations
departments may have greater rapport with journal editors resulting
in an over representation of their innovative activity. This is
perhaps offset by the fact that small firms aggressively pursue new
product announcements as inexpensive advertising.
States are not an entirely satisfactory unit of observation to
use in this analysis. In contrast to the micro-economics literature
in which the unit of analysis is generally accepted to be the firm,
there is no such consensus in geography7. The analysis of geography
is further handicapped by the lack of data on an ideal unit of
observation. Theoretically, spatial processes occur within the
boundaries of geographic areas characterized by functional
link-ages and dependencies (Czmanski and Ablas, 1979). Geographic
data, however, are collected and defined within the boundaries of
political jurisdic-tions. While ideally we would like data at a
sub-state level of aggregation, no such data currently exists.
Table 1 presents the geographic distribution of innovations for
the leading states. There is noticeable spatial concentration in
the distribution of manu-facturing product innovation. Eleven
states accounted for 81 % of the 4200 innovations. In order to
normalize for differences in the state size, the number of
innovations per 100 000 manufacturing employees is provided. New
Jersey had the highest rate of innovation followed by Massachusetts
and California. These states innovated more than twice the national
rate of 20. 34 innovations per 100 000 manufacturing workers. Seven
other states, New Hampshire, New York, Minnesota, Connecticut,
Arizona, Colorado and Delaware were also more innovative than the
national average.
Table 2 presents the distribution of innovations by two-digit
industries within states. The first row presents the national
totals. For example, SIC 20: food and kindred products, accounted
for 110 innovations introduced to the market in 1982. The second
column presents the total number of innova-tions attributed to each
state. Each cell in this Table corresponds to a state and industry
combination. For example, Arizona was the source of 21 innovations
for SIC 36: electronic and other electrical equipment and
compo-nents.
A plausible hypothesis about the distribution of innovations
among states and industries is that the proportion of innovations
attributed to each
7 Agglomerations have been measured in rhe lirerarure using a
variery of geographic unirs, ranging from rhe mulri-srare
Manufacruring Belr (Norron and Rees, 1979) co rhe Merropoliran
Sratisrical Area (Oakey, 1985).
------------ 457
at University of North Carolina at Chapel Hill on July 30,
2014http://icc.oxfordjournals.org/Downloaded from
http://icc.oxfordjournals.org/
-
------ 'An Examination of the Geography of Innovation
TABLE 1. Which States are the Most Innovative?
Innovations Innovations per 100000 State manufacturing
workers
New Jersey 426 52.33 Massachusetts 360 51.87 California 974
46.94 New Hampshire 33 30.84 New York 456 29.48 Minnesota 110 28.65
Connecticut 132 28.51 Arizona 41 27.70 Colorado 42 22.46 Delaware
15 21.13 National 4200 20.34 Rhode Island 24 18.46 Pennsylvania 245
18.28 Illinois 231 18.16 Texas 169 16.14 Wisconsin 86 15.61 Ohio
188 15.00 Florida 66 14.60
SOURCE: Number of innovations is from the SBA innovation data.
Number of manufacturing workers is from the 1982 CemUJ of
Manufacturer! .
industry and state is exactly determined by the probability of
innovations attributed to the industry and to the state. For
example, the proportion of innovations in electronics in California
may be determined by the probability of an innovation occurring in
the state of California and the probability of an innovation
occurring in the electronics industry. [If s references the state
and i references the industry, Psi = Ps. p;J. If this were true the
concentration of innovations within a state and industry would
simply reflect the propensity of the state and the industry to
innovate with no interaction between the two influences. In such a
case, the distribution of innovations would be deter-ministic and
geographic factors would not affect the location of innovation. A
chi-squared test of the independence of state location and
two-digit indus-tries' innovations rejects this hypothesis at the
0.99 confidence level. The number of innovations in a state and
industry are not determined strictly by the state and industry
propensity to innovate.
Another hypothesis about the location of innovation is that the
geographic distribution of innovation in an industry is determined
by the location of that industry. Innovation may be geographically
concentrated because indus-tries are geographically concentrated
due to considerations of cost factors (Isard, 1956). The location
of industries may determine the location of product innovation. In
this case, the number of innovations in an industry
--------------458 --------------
at University of North Carolina at Chapel Hill on July 30,
2014http://icc.oxfordjournals.org/Downloaded from
http://icc.oxfordjournals.org/
-
An Examination of the Geography of Innovation
TABLE 2. Number of Innovations by State and Industry
Selected Two-Digit SIC Code InduJtries
State Total 20 28 34 35 36 37 38
National total 4200 110 292 201 1392 840 66 1042 Arizona 41 0 0
2 8 21 0 9 California 974 8 18 26 398 288 18 187 Connecticut 132 3
11 5 47 24 3 33 Georgia 53 12 2 3 12 8 1 11 Illinois 231 5 15 18 81
31 1 56 Mass. 360 13 22 15 102 0 3 120 Michigan 112 1 11 7 29 9 21
0 Minnesota 110 6 5 8 38 14 0 28 New Jersey 426 6 75 19 114 58 0
131 New York 456 20 39 20 111 0 2 147 Ohio 188 3 16 21 51 27 3 44
Pennsylvania 245 14 27 16 78 30 0 66 Texas 169 4 14 10 58 35 2 37
Wisconsin 86 2 6 28 17 29
SIC 20: food and kindred products; SIC 28: chemicals and allied
products; SIC 34: fabricated metal products; SIC 35: industrial and
commercial machinery and computer equipment; SIC 36: electronic and
other electrical equipment; SIC 37: transportation equipment; SIC
38: measuring. analyzing and controlling instruments.
and state would be highly correlated with industrial presence.
The correlation of the count of innovation and total state
manufacturing value added was 0.236. For two digit industries and
states, the correlation between innovation and value added was
0.4202. Although innovation is positively correlated with industry
presence, the relationship is far from deterministic.
The geographic concentration of innovation is even more
pronounced when greater industry detail is considered. Table 3
presents the distribution of innovation by state for a subset of
the most innovative three digit industries. The industries
presented in this Table are ranked by the number of innovations
attributed to the product category. For example, there were 954
product innovations in computers (SIC 357) and 668 product
innova-tions in measuring and controlling instruments (SIC 382).
Column 3 lists the number of innovations attributed to each
state.
For each three digit industry, the two states which accounted
for the highest number of innovations is listed. For example, the
state of California provided 365 innovations for SIC 357: on
average, one innovation in the computing machinery industry in
California for every day of the year. This was 38.3 % of the
innovations in the computing machinery industry as indicated by
column 4. The state of California had a grand total of 974
innovations and column 5 reports that the computing machinery
industry
------------------------459 ------------------------
at University of North Carolina at Chapel Hill on July 30,
2014http://icc.oxfordjournals.org/Downloaded from
http://icc.oxfordjournals.org/
-
An Examination of the Geography of Innovation
TABLE 3. Distribution of Three Digit Industry by State
(1) (2) ( 3) (4) (5 ) (6) % of % of Location
Product State Count innovations state quotient
Computers SIC 357 n = 954 California 365 38.3 37.6 167.84
Massachusetts 82 8.6 22.8 100.44
Measuring and controlling instruments SIC 382 n = 668 California
134 29.1 13.8 126.42 Massachusetts 94 14.1 26.1 164.15
Communication equipment SIC 366 n = 376 California 116 30.9 11.9
132.22 New York 45 12.0 9.9 110.00
Electrical components SIC 367 n = 261 California 128 49.0 13.2
211.29 Massachusetts 26 10.0 7.2 116.13
Medical instruments SIC 384 n = 228 New Jersey 57 25.0 13.5
248.15 New York 51 22.4 11.2 207.41
General industrial machinery and equipment SIC 356 n = 164
Pennsylvania 25 15.2 10.2 261.54 New Jersey 18 11.0 4.2 107.69
Drugs SIC 283 n = 133 New Jersey 52 39.1 12.3 381.25 New York 18
13.5 4.0 121.88
represented 37.6 % of the state innovations. We also calculated
innovation location quotients for these industries. The innovation
quotients were calcu-lated as the percentage of innovation in a
state accounted for by an industry divided by the percentage of
national innovations accounted for by that industry. The ratio is
then multiplied by 100. An innovation quotient of 100 indicates
that the innovations are equally represented in the state and
national economies. An innovation quotient greater than 100
indicates that the region can be regarded as relatively specialized
in that activity. For the 13 most innovative three digit
industries, the average innovation quotient was 239.5. This
indicates significant regional specialization of innovative
activity for these industries.
The analysis presented here demonstrates that product
innovations exhibit a strong tendency to cluster geographically and
this tendency is more pronounced when individual industries are
considered. Geographic cluster-ing of innovation has been noted in
the 19th century examples of cotton textiles in Lowell,
Massachusetts, machinery in Connecticut, and paper in
------------ 460 ------------
at University of North Carolina at Chapel Hill on July 30,
2014http://icc.oxfordjournals.org/Downloaded from
http://icc.oxfordjournals.org/
-
------ An Examination of the Geography of Innovation ------
western Massachusetts as well as the more contemporary examples
of auto-mobiles in Detroit, semi-conductors and personal computers
in Silicon Valley and aeronautical products in Los Angeles (Hall
and Markusen, 1985; Rosenberg, 1962; Scott, 1989). The data
presented here suggests that geo-graphic clustering is a more
general phenomenon. Theoretical models indi-cate that the
geographic clustering of innovative activity is due to increasing
returns to non-transferable activities (Arthur, 1990; Grossman and
Help-man, 1989). Based on the innovation location quotients, it
appears that certain states have developed a comparative advantage
for innovation in specific industries. The next section will
consider the empirical estimation of innovative production
functions which link the location of innovative output to the
presence of specialized resources to facilitate the innovation
process.
4. An Empirical Model of Innovation and Location
Emphasis in the industrial organization literature has shifted
towards under-standing the process of achieving a commercially
viable innovation (Mowery and Rosenberg, 1989). Kline and Rosenberg
(1987) suggest four key know-ledge bases or innovative inputs to
the innovation process: university re-search, industrial R&D,
related industry presence, and specialized business services.
Innovative output, INNis> the count of innovation for an
industry i, in a geographic area, s, is modeled as a function of
four innovative inputs:
INNis = f (UNIVIs' INDISI RELPRESIs' BSERVIs) (1)
where UNIVIs represents university research at the level of the
academic department. INDlJ represents industrial R&D
expenditures. RELPRESIs represents related industry presence,
including firms using related technolo-gies and downstream users of
a technology who may disseminate information relevant to
innovation. BSERVIs represents the presence of specialized
busi-ness services related to innovative activity within the
industry. The industry group, I, is used to measure the more
encompassing technological area across which informational
spillovers are expected to occur. Implicit in the innova-tion
process is a time lag involved in translating an invention into a
commercial product innovation. The exact time lapse between
discovery and subsequent commercial product innovation is highly
variable and difficult to specify. For example, Mansfield (1991)
estimated that the average time lag between an academic research
finding and the first related commercial introduction of a new
product was, on average, 7 years, with a standard deviation of 2
years. To model the expenditures on the innovative inputs, we use
the average annual expenditures in the 10 year period prior to
the
-~------------------461 -------------------------
at University of North Carolina at Chapel Hill on July 30,
2014http://icc.oxfordjournals.org/Downloaded from
http://icc.oxfordjournals.org/
-
------ An Examination of the Geography of Innovation ------
introduction of the innovation. This measure attempts to capture
the stock of knowledge embodied in a geographic area and provides a
more appropriate measure of regional capacity than a single year's
expenditures.
The dependent variable in the innovation equation is the number
of innovations for an industry, i, which originated in state, S8.
The independent variables in the model are defined as follows.
University R&D is expenditure for research performed by
academic departments and is from the National Science Foundation's
Survey of Science Resources. Related industry presence is measured
as value added for the major sector of the industry under
consideration. As an example, SIC 283, drugs, would benefit from
the presence of related activity in the industrial group SIC 28,
chemicals and allied products.
There are a variety of business services which provide knowledge
of the market and the commercialization process. For example, the
services of patent attorneys may be a critical input to the
innovative process. Unfortun-ately, data on the presence of this
input do not exist. All legal services are grouped together in SIC
8111 without any finer detail. Of all the producer services
available to support innovative activity, the only category which
is specifically and solely targeted ro the introduction of new
innovations is commercial testing laboratories, SIC 7397. This is
used as a rough proxy measure for business services related to
innovative activity. To measure BSERV.s> the annual average
receipts of commercial testing laboratories are used.
Industrial R&D is measured as expenditure for industrial
R&D performed within companies as reported by the National
Science Foundation's Science Resources Survey. These data do not
include the cost of R&D contracted to outside organizations
such as universities and colleges, nonprofit organiza-tions,
research institutions and other companies. The data represent the
10 year average of R&D expenditures. NSF confidentiality
requirements mean that data is unavailable for some locations.
Industrial R&D expenditures are available for 29 states. The
available state data contain over 92 % of the total innovations
introduced in the US in 1982. The sample for estimation accounts
for 78% of the US population and 81 % of the university research
expenditures in 1982. Conversely, states for which R&D data is
unavailable account for 325 innovations or 7.7 % of the total
innovations. The estimation
• The estimation is based on the innovation data for the 13 most
innovative three-digit industries. These 13 industries account for
80% of the total innovations. There were a total of 95 three-digit
industries which contain at least one innovation citation. When the
data are stratified by state, a large number of zeto cells result.
In order to proceed with the estimation, it is necessary to limit
the sample. Each industry included in the estimation contains a
minimum of 50 innovation citations for the year 1982. In the sample
used for estimation, there are 140 state and industry observations,
or 37 %, with zero innovations.
-------------- 462 -------------
at University of North Carolina at Chapel Hill on July 30,
2014http://icc.oxfordjournals.org/Downloaded from
http://icc.oxfordjournals.org/
-
------ An Examination of the Geography of Innovation
uses the resulting combination of 29 states and 13 industries
which yields 377 observations.
Using state level data to model a more local phenomenon may bias
the parameter estimates. Two additional variables are added to
control for aggregation bias. State population, POP.s> is
included as a control for state size and to facilitate cross state
comparisons. A geographic concentration variable, CONC.s> is
added to control for within state variation. This variable measures
the degree to which manufacturing activity is concentrated within
states and is added to the equation to compensate for the use of
states as the unit of observation. The numerator of the
concentration index is the value of manufacturing shipments for the
largest MSA in the state and the denomi-nator is the value of
manufacturing shipments for the entire state from the Census of
Manufacturers9 . Industry sales, SALES i . are included as a
control for product demand which may affect the quantity of
innovations generated within an industry.
The model is estimated for two variations of the industrial
R&D variable to consider the empirical question of the
technological area across which 'spill-overs' associated with
industrial R&D occur. The first variation measures total
industrial R&D for the state, IND. s:
log(INNs) = 13 1log(UNIV/s) + I3Jog(lND.J + 133log(RELPRES 1s )
+ 134 log(BSERV.J + 135CONC.s + 136log(POP.s)
(2)
The second variation restricts industrial R&D to a given set
of related industries within the state, IND/s:
log(lNNis ) = 13 1log(UNIV1J + 132log(lND1s) + 133 log(RELPRES1J
+ 134log(BSERV.J + 135 CONC .s + 136log(POP s )
(3)
The second specification requires detail on industrial R&D
expenditures at the two digit industry level. Unfortunately, before
1987, NSF only provides data on total industrial R&D
expenditures at the state level and to proceed it is necessary to
estimate industry level expenditures. Data from the 1987 National
Science Foundation, Survey of Industrial Research and Development
was used to allocate the average state R&D expenditures to two
digit industries within a state, IND1s' Expenditures are classified
under the industry of the
9 For the estimation of the innovation equation the log of the
geographic concentration variable is not taken. There is no strong
a priori functional specification and the estimation of the
innovation equation with a log transformation of this variable
yielded similar results.
------------ 463 -------------
at University of North Carolina at Chapel Hill on July 30,
2014http://icc.oxfordjournals.org/Downloaded from
http://icc.oxfordjournals.org/
-
------ An Examination of the Geography of Innovation ------
performing company. The allocation method described above
assumes that the percentage of industrial R&D originating
within an industry and state has remained relatively stable. There
is no evidence that the within state allocation of R&D
expenditures has changed. Malecki and Bradbury (1991) found that
the geographic location of corporate R&D facilities in the US
has remained consistent since 1970. Glasmeier (1988) noted a
movement of scientists and engineers to the South and West during
this time period, attributing this migration to the establishment
of technological branch plants which locate R&D activity nearer
the locus of production. Howells (1990, p. 138) found that these
units are more oriented towards process innovation and incremental
improvements and are fundamentally different from central R&D
laboratories likely to be involved with product innova-tion.
Nationally, the percentage of R&D expenditure allocated to the
two digit industries considered here was relatively stable from
1978 to 1987 and the within state R&D expenditures allocation
is assumed to be similarly stable.
There are several estimation issues to consider. The dependent
variable, the number of innovations by state and industry, is a
censored dependent variable. The number of innovations will either
be zero or some positive integer. Cases with no innovations provide
information about how innova-tive locations differ from
non-innovative locations. For these reasons, the Tobit model is
used. An additional econometric concern always suspect with
geographic cross-sectional data is multicolinearity. With state
data, it is highly likely that the independent variables may be
affected by some com-mon trend or underlying state characteristics.
10 Multicollinearity may result in less statistically significant
coefficient estimates than expected (see Table 4).
IND., IND[s
UNIV I ,
RELPRESS Is BSERV.,
TABLE 4. Correlation Matrix for Innovative Inputs
IND., IND[s
1.00 n.a. 1.00 0.68 0.51 0.63 0.58 0.73 0.53
UNIVIs
1.00 0.39 0.56
RELPRESIs BSERV"
1.00 0.53 1.00
NOTE: Reporred correlarions are for rhe log values of each of
rhe variables. n.a. indicares rhe variables do nor appear in rhe
model rogerher.
10 No conclusive presence of sparial-correlarion was found and
no correcrion was made. Geographic fixed effecr models were
esrimared and rhe resulrs were robusr. Addirionally, Breusch-Pagan
resrs revealed no hereroscedasriciry in rhe innovarion equarion
specificarion.
------------- 464 ------------
at University of North Carolina at Chapel Hill on July 30,
2014http://icc.oxfordjournals.org/Downloaded from
http://icc.oxfordjournals.org/
-
------ An Examination of the Geography of Innovation ------
5. Empirical Results
The empirical results confirm that innovative activity within
states is related to the presence of a broader technological
infrastrucrure within the state. Table 5 presents the results of
the maximum likelihood estimation of the Tobit specification of the
innovation equation. The innovative inputs of university R&D,
related industry presence, and the presence of specialized business
services, are positive and statistically significant. Results are
pre-sented for two variations of the industrial R&D variable:
model A uses total industrial R&D expenditures, IND.,; model B
uses estimates of R&D for the two digit industry, IND[so This
allows an empirical testing of the extent of the technological
neighborhood across ~hich R&D spill-overs occur. For example,
do R&D spill-overs occur within a narrow band of related
in-dustries or does R&D spill-over cross the entire spectrum of
manufacturing industries? A comparison of the two models suggests
that model A, using total industrial R&D expenditures, provides
a better model of the innovation equation than the alternative
model B. In as far as any inferences can be drawn, model A exhibits
a better statistical fit. Total industrial R&D expenditures has
a larger and more statistically significant coefficient than
TABLE 5. Tobit Estimation of Innovation Equation
Variable Model A Model B
Log(IND.s) 0.190a (0.054)
Log(lND js ) 0.049" (0.034)
Log(lNIVjs) 0.123a 0.176a (0.044) (0.045)
Log(RELP RES Is) 0.296a 0.35S" (0.046) (0.050)
Log(BSERV.S> O.llsa 0.212a (0.057) (0.057)
Log(POP.s) 0.103a 0.131 a (0.030) (0.033)
Log(SALES j,) -0.230 -0.260 (0.117) (0. 12S)
CONC.s 1.020a 1. 113" (0.195) (0.214)
-
------ An Examination of the Geography of Innovation
industrial R&D expenditures estimated at the industry level.
The elasticity of innovative output with respect to total
industrial R&D expenditures is three times the elasticity of
industry specific R&D expenditures. The higher coefficient on
total R&D expenditures may well reflect spill-overs across
industries but still within state boundaries. Furthermore, research
and development expenditures are usually undertaken by large
integrated cor-porations with interests in several product
categories. The lower coefficient for INDlJ may reflect the fact
that industry assignment is based on the industry code of company
performing the R&D, not on the industry towards which the
research efforts and expenditures are directed. The results
indicate that the presence of research expenditure undertaken by
universities within the state also increase innovative output.
Since the university research ex-penditure variable, UNIVIs> is
measured at the departmental level, the results indicate that the
strength of the research expenditure at the academic departments
relevant to the innovative industry will have a statistically
significant effect on innovative output within the state ll. The
finding that university research expenditure translates into
increased innovative activity in a state is supported by recent
work by Jaffe (1989), Mansfield (1991), and Acs et al. (1992).
Related industry presence, RELPRESIs> is statistically
significantly associ-ated with innovative activity in a state. This
indicates that experience with a technology as embodied in the
manufacturing process, translates into in-creased innovative output
in a state. Related industry presence was measured as value added
in the larger industry group to which the innovation is attributed.
This variable has the largest elasticity of innovative output with
respect to an innovative input. This result supports the idea that
learning by doing and learning by using are important inputs to the
innovation process. The magnitude of this coefficient raises
questions about the relationship of industry presence to the other
innovative inputs. A strong industry presence in a region may
reflect higher industrial and university R&D expenditures in
related technical areas.
The final innovative input is specialized business services,
BSERV.s' Specialized business services ease the innovation process
by providing infor-mation about diverse items such as next round
capital financing or regula-tions regarding product testing and
standardization. Their presence in an area may facilitate
contractual arrangements to incorporate this knowledge
II An alrernative specification of the innovation equation using
tOtal university research expenditure, similar to the specification
presented using two variations of industrial R&D, was
estimated. Total state university research expenditure was
statistically significantly related to innovations within the
state, however, the overall fit of the model declined.
Pragmatically, this is understandable because it is less plausible
that basic research in biology would spill-over to innovations in
an unrelared field such as electronics.
------------ 466 -----------
at University of North Carolina at Chapel Hill on July 30,
2014http://icc.oxfordjournals.org/Downloaded from
http://icc.oxfordjournals.org/
-
------ An Examination of the Geography of Innovation ------
into the innovation process. Specialized business services are
statistically significantly related to innovative output at the
state level. This finding is consistent with MacPherson (1988)
which suggests that localized external sources of specialized
business services contribute to realized innovative output.
Three control variables were added to the model: SALESi., POP.s
and CONC. s • Industry sales, SALESi ., provides a control for the
three digit industry to which the innovation belongs. As discussed
earlier, all the industries used in the analysis are highly
innovative opportunity industries. This variable provides a control
for the total national industry sales which is related to the
demand for the industry's products. This variable was not
statistically significant. Population size, POP.s and the
geographic co-incidence variable, CONC." are included to mitigate
the effect of aggrega-tion bias caused by using states as the unit
of observation. As discussed earlier, states are a less than
satisfactory unit of observation. It is difficult to consider that
economic spill-overs occur equally in California and in Rhode
Island. For this reason, POP.s provides a control for the size of
the state and facilitates cross state comparisons.
The geographic coincidence index, CONC.s , provides a control
for within state variation in the concentration of. manufacturing
activity. We would expect that the more concentrated manufacturing
activity is within a state, the more likely that innovative inputs
would spill-over into increased in-novative output. The geographic
coincidence variable is of the hypothesized sign and is
statistically significant. This result confirms the view that
within state concentrations of manufacturing activity are also
important to innova-tive output.
The results suggest that the clustering of product innovation at
the state level is related to the presence of innovative inputs.
These findings are consistent with the view of innovation as a
process facilitated by diverse types of expertise and knowledge
(Kline and Rosenberg, 1987).
6. Summary and Discussion
The evidence presented here suggests that the process of
introducing new products to the market is facilitated by location.
Product innovations exhibit a pronounced tendency to cluster in
states which contain concentrations of innovative inputs.
These states appear to have developed a comparative advantage
for innova-tive industries which does not rest on natural resource
endowments or the availability of low cost labor. The comparative
advantage for innovative
------------ 467
at University of North Carolina at Chapel Hill on July 30,
2014http://icc.oxfordjournals.org/Downloaded from
http://icc.oxfordjournals.org/
-
An Examination of the Geography of Innovation
activity is provided by the location of specialized knowledge
resources which enhance the innovation process.
By many indicators, US research capabilities are the envy of the
world. This lead appears to be lost in the translation of
scientific knowledge into new commercial innovations. American
firms suffer from longer develop-ment cycles for new products and
there is evidence that a wide variety of US industries have failed
to commercialize new innovations effectively and rapidly 12. This
evidence, coupled with America's declining trade perform-ance, has
provoked wide debate over the need to find new ways to organize
innovative activity. The results presented here suggest that
geographic location and the geographic complementarity of
innovative inputs may warrant further study and consideration.
Acknowledgements
The author would like to thank Zoltan Acs, David Audretsch,
Wesley Cohen, Giovanni Dosi, Richard Florida, Michael Fritsch, Mark
Kamlet, Edward Malecki, and Paula Stephan, the participants at the
WZB seminar on the Geography of Innovation, and the three anonymous
referees for comments. Cynthia Brandt provided research
assistance.
References
Acs, Z. J. and D. B. Audretsch (1988), 'Innovation in Large and
Small Firms: An Empirical Analysis,' American Economics Review, 78,
678-690.
Acs, Z. J. and D. B. Audretsch (1990), Innovation and Small
Firms, MIT Press: Cambridge.
Acs, Z. J., D. B. Audretsch and M. P. Feldman, (1992), 'The Real
Effects of Academic Research: A Comment' American Economics Review,
82, 363-367.
Allen, R. C. (1983), 'Collective Innovation,' Journal 0/
Economic Behavior and Organization, 4, 1-24. Arthur, B. W. (1990),
'Positive Feedbacks in the Economy,' Scientific American, February,
92-99.
Brody, D. and R. Florida (1991), 'Falling Through the Cracks:
The U.S. Failure in Active Matrix Display Technology,' Working
Paper 91-10, School of Urban and Public Affairs, Carnegie Mellon
University.
Carlson, B. and S. Jacobsson (1991), 'What Makes the Automation
Industry Strategic?' Economics 0/ Innovation and New Technology, I,
93-118.
Coffey, W. J. and M. Polese (1987), 'Trade and Location of
Producer Services: A Canadian perspective,' Environment and
Planning A, 19, 597-611.
Cohen, W. M., R. C. Levin and D. C. Mowery (1987), 'Firm Size
and R&D Intensity: A re-examination,' in T. F. Bresnaham and R.
Schmalensee, (eds) The Empirical Renaissance in Industrial
Economics, Basil Blackwell: New York.
Czmanski, S. and L. A. de Q. Ablas (1979), 'Identification of
Industrial Clusters and Complexes', Urban Studies, 16, 61-80.
12 Derrouzos et al. (1988); and Florida and Smith (1990) provide
comprehensive reviews.
------------- 468
at University of North Carolina at Chapel Hill on July 30,
2014http://icc.oxfordjournals.org/Downloaded from
http://icc.oxfordjournals.org/
-
An Examination of the Geography of Innovation
Davelaar, E. and P. Nijkamp (1989), 'The Role of the
Metropolitan Milieu as an Incubaror Center for Technological
Innovations: A Dutch Case Study,' Urban Studies, 26, 516-29.
Dosi, G. (1988a), 'Sources, Procedures and Microeconomic Effects
of Innovation,' journal of Economic Literature, 36, 1120-1171.
Edwards, K. 1. and T. J. Gordon, (1984), 'Characterization
ofInnovations Introduced on the US Matket in 1982,' The Futures
Group, US Small Business Administration, Contract No.
SBA-6050-0A-82.
Florida, R. and D. F. Smith (1990), 'Finance Capital and Spatial
Organization: An Analysis of the Venture Capital Industry,' SUPA
Working Paper Series, Carnegie-Mellon University, Working Paper
90-21.
Freeman, C. (1991), 'Networks of Innovarors: A Synthesis of
Research Issues,' Research Policy, 20, 499-514.
Glasmeier, A. K. (1988), 'Facrors Governing the Development of
High Tech Industty Agglomerations: A Tale of Three Cities',
Regional Studies, 22, 287-301.
Gtefsheim, S., J. Franklin, and D. Cunningham (1991),
'Biotechnology Awareness Study, Part 1: Where Scientists get their
Information.' Bulletin of the Medical Library Association, 79,
36-44.
Hall, P. and A. Matkusen (1985), Silicon Landscapes. Allen and
Unwin, Bosron.
Griliches, Z. (/990), 'Patent Statistics as Economic Indicator:
A survey,'journal of Economic Literature, 28, 1161-1707.
Grossman, G. M. and E. Helpman (1992), Inn()tJation and Growth
in the Global Economy. The MIT Press: Cambridge, MA.
Howells, J. (/990), 'The Location and Organization of Research
and Development: New Horizons,' Research Policy, 19, 133-146.
Isard, W. (/956), Location and SPace-economy. The MIT Ptess:
Cambridge, MA.
Jaffe, A. B. (1989), 'Real Effects of Academic Research,'
American Economics Review, 79, 957-970.
Kline, S. J. and N. Rosenberg (1987), 'An Overview
ofInnovation,' in R. Landau and N. Rosenberg, (eds) The Positive
Sum Strategy, National Academy Press: Washington, D.C.
Lucas, R. E. Jr. (1993), 'Making a Miracle.' Econometrica, March
1993, 61, 251-72.
Lundvall, B.-A. (1988), 'Innovation as an Interactive Process:
User Producer Relations,' in G. Dosi et al. (eds), Technical Change
and Economic Theory, Francis Pinter: London.
Machlup, F. (1962), 'The Supply of Inventors and Inventions,' in
R. Nelson, (eds) The Rate and Direction of Inventive Activity,
Princeton University Press: Princeton.
MacPherson, A. (1988), 'New Product Development among Small
Toronto Manufacturers: Empirical Evidence on the Role of Technical
Service Linkages,' Economic Geography, 61-75.
MacPherson, A. (/99 I), 'Interfirm Information Linkages in
Economically Disadvantaged Regions: An Empirical Perspective from
Metropolitan Buffalo,' Environment and Planning, A, 591-606.
Malecki, E. J. (1985), 'Industrial Location and Corporate
Organization in High Technology Industries,' Economic Geography,
345-369.
Malecki, E. J. (1985b), 'Public Sector Research and Development
and Regional Economic Performance in the United States', in The
Regional Economic Impact of Technological Change, A. T. Thwaites
and R. P. Oakley (eds). Pinter: London.
Malecki, E., and S. Bradbury (1991), 'Locational Decision-making
of R&D Faciliries and Professional Labor', Regional
Studies.
Mansfield, E. J. (1991), 'Academic Research and Industrial
Innovation', Research Policy, 20, 1-12. Mowery D. and N. Rosenberg
(1989), Technology and the Pursuit of Economic Growth. Cambridge
Universi-ty Press: New York.
Nelson, R. (1989), 'US Technological Leadership: Where did it
come ftom and where did it go?', 1-45.
Nelson, R. (1990), 'Capitalism as an Engine of Progress,'
Research Policy, 19, 193-214.
Nelson, R. and S. Winter (1982), An Evolutionary Theory of
Economic Change. Harvard University Press: Cambridge, MA.
----------- 469
at University of North Carolina at Chapel Hill on July 30,
2014http://icc.oxfordjournals.org/Downloaded from
http://icc.oxfordjournals.org/
-
An Examination of the Geography of Innovation
Oakey, R. (1985), ·High Technology Industries and Agglomeration
Economies,' in P. Hall and A. Markusen (eds.), Silicon Landscapes.
Allen and Unwin: Bosron.
Porter, M. (1990), The Competitive Advantage 0/ Natiom, Free
Press: New York. Romer, P. M. (1986), 'Increasing Returns and Long
Run Growth,' Journal 0/ Political Economy, 94, 1002-37.
Rosenberg, N. (1982), Imide the Black Box: Technology and
Economics, Cambridge University Press: New York.
Saxanien, A. (1990), 'The Origins and Dynamics of Production
Networks in Silicon Valley', paper presented at the International
Workshop on Networks of Innovators, Montreal.
Scott, A. J. (1989), Metropolis: From the Division 0/ Labor to
Urban Form. University of California Press: Los Angeles.
Shimshoni, D. (1966), 'Spin-Off Enterprises from a Large
Government-Sponsored Laboraroty,' unpub-lished docroraI
dissertation, Sloan School, MIT, Cambridge.
Science Policy Research Unit (1972), Success and Failure in
Industrial Innovation. Center for the Study of Industrial
Innovation: London.
Tassey, G. (1991), 'The Functions of Technology Infrastructure
in a Competitive Economy, Research Policy, 20, 329-343.
Teece, D. J. (1986), 'Profiting from Technological Innovation:
Implications for Integration, Collabora-tion, Licensing and Public
Policy,' Research Policy, IS, 285-305.
Thomas, M. (1985), 'Regional Economic Development and the Role
of Innovation and Technological Change,' in A. T. Thwaites and R.
P. Oakley (eds), The Regional Economic Impact o/Technological
Change, St. Martin's Press: New York.
US Office of Technology Assessment (1984), Technology,
Innovation and Regional Economic Develop-ment, GPO: Washington,
DC.
Von Hippel, E. (1988), The Sources 0/ Innovation, Oxford
University Press: New York.
------------ 470
at University of North Carolina at Chapel Hill on July 30,
2014http://icc.oxfordjournals.org/Downloaded from
http://icc.oxfordjournals.org/