The Geographic Sources of Innovation: Technological Infrastructure and Product

Innovation in the United States

Author(s): Maryann P. Feldman and Richard Florida
Source: Annals of the Association of American Geographers , Jun., 1994, Vol. 84, No. 2
(Jun., 1994), pp. 210-229
Published by: Taylor & Francis, Ltd. on behalf of the Association of American
Geographers

Stable URL: https://www.jstor.org/stable/2563394

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

Taylor & Francis, Ltd. and Association of American Geographers are collaborating with JSTOR
to digitize, preserve and extend access to Annals of the Association of American Geographers

This content downloaded from

201.150.98.4 on Thu, 22 Jul 2021 16:55:32 UTC
All use subject to https://about.jstor.org/terms
 The Geographic Sources of Innovation:
Technological Infrastructure and Product
Innovation in the United States
Maryann P. Feldman* and Richard Florida**

*Department of Economics and Management, Goucher College

**Center for Economic Development, H. John Heinz III School of Public Policy and
Management, Carnegie Mellon University

Since the sixteenth century, and perhaps clustering or agglomeration of economic activ-
before, the fortunes of regions and na- ity creates scale economies, facilitates face-to-
tions have depended on new ideas and face interaction, and shortens interaction dis-
new products that energized these places and tances. The interaction of all of these factors
facilitated their economic growth. If innovation lends itself to innovation in economic pro-
is one of the keys to prosperity, then precisely cesses and products. In contemporary devel-
how this happens-how a region breaks with oped or mature economies, product innova-
convention and introduces new products-is a tion is also linked with the composition of firms
question of some significance. Not surprisingly, and activities in these clusters. Especially im-
a variety of scholars have tried to find the keys portant in this regard is the technological infra-
to innovation, to unlock the doors of the inno- structure within these clusters. Building upon
vation process, and to render what is inside recent reconceptualizations in economic ge-
less mysterious and more accessible to less ography and economics, we suggest that inno-
fortunate regions and states (Kline and Rosen- vation in the late twentieth century is unusually
berg 1987; Landes 1969; Malecki 1991; Mokyr dependent on an area's underlying technologi-
1990; Rosenberg 1972; 1982). In his classic cal infrastructure. Having burst beyond the
work on innovation and capitalism, Joseph confines of the organizational boundaries of an
Schumpeter (1954) argues powerfully that eco- individual firm, innovation is increasingly de-
nomic growth requires innovation-the gen- pendent on a geographically defined infra-
eration of higher quality products at lower unit structure that is capable of mobilizing technical
costs than had previously been obtainable. resources, knowledge, and other inputs essen-
What then do we know about this key vari- tial to the innovation process. This infrastruc-
able for economic growth? The literature on ture consists of sources of knowledge: net-
innovation, as one might expect, is daunting; works of firms that provide expertise and tech-
it ranges from heroic accounts of inventors and nical knowledge; concentrations of research
innovators to the more prosaic accent on the and development (R&D) that enhance oppor-
factors of land, labor, and capital. A full appre- tunities for innovation by providing knowledge
ciation of innovation, of course, requires both of new scientific discoveries and applications;
of these approaches, and various others in be- and business services with expertise in product
tween; but in this paper we have space to deal positioning and the intricacies of new product
with only a portion of the problem-the geo- commercialization. Once in place, these geo-
graphic dimensions of innovation and their graphic concentrations of infrastructure en-
structural conditions in one place, the United hance the capacity for innovation as their re-
States, and at one time, the early 1980s. spective regions develop and specialize in par-
Geographers and economists have often ticular technologies and industrial sectors. Ge-
noted the congruent clustering of economic ography, in other words, serves as the vessel
activity and innovation. While much about this in which entrepreneurs, venture capitalists,
congruence remains unclear, we know that the and other agents of innovation, organize an

Annals of the Association of American Geographers, 84(2), 1994, pp. 210-229

C Copyright 1994 by Association of American Geographers
Published by Blackwell Publishers, 238 Main Street, Cambridge, MA 02142, and 108 Cowley Road, Oxford, OX4 tIF, UK.

This content downloaded from

201.150.98.4 on Thu, 22 Jul 2021 16:55:32 UTC
All use subject to https://about.jstor.org/terms
 The Geographic Sources of Innovation 211

infrastructure that brings together the crucial worth considering sub-national or regional sys-
resources and inputs for the innovation pro- tems of innovation.
cess. A second stream of research attempts to do
The empirical model of the geography of this, providing richly detailed case studies of
innovation presented here tests the hypothesis the origins and development of "regional inno-
that innovation is concentrated in places that vation complexes" (Stohr 1986). Case studies
possess a well-developed technological infra- of Route 128 (Dorfman 1983), Silicon Valley
structure. The latter is defined in terms of the (Saxenian 1985), and Orange County (Scott
agglomerations of four indicators: 1) firms in 1988), among others, suggest that innovation
related industries; 2) university R&D; 3) indus- is a complex geographic process with multiple
trial R&D; and 4) business-service firms. Our spatial determinants. The focus on individual
analysis confirms and extends this hypothesis. case studies, while richly informative, does not
We demonstrate that not only do innovations yield the kind of general findings which would
cluster geographically in areas that contain permit a broad conceptualization of the geo-
concentrations of specialized resources indica- graphic dimensions of innovation. The case
tive of technological infrastructure, but also that study literature encourages scholars to shift fo-
these spatial concentrations of specialized re- cus from the firm-level to a consideration of
sources mutually, and positively, reinforce a re- innovation as a social process reliant on exter-
gion's capacity to innovate. nal, geographically based sources of knowl-
edge (Dosi 1988).
A third stream of literature considers the role
of geographic agglomeration in technological
Innovation in Geographic Theory innovation and economic development
(Oakey 1985; Thomas 1985). Storper and
Recent research on the geographic, organ- Walker's (1989) theory of "geographic industri-
izational, and economic dimensions of innova- alization" captures the spatial nature of the
tion divides into three streams. The first of process of technological change and industrial
these deals with the location of R&D inputs and development. Other studies of the role of ag-
technology-based industries. Malecki's (1981; glomeration economies focus upon the con-
1985; 1986; 1990) documentation of the loca- centrations of key resources and organizational
tion of R&D activities has been followed up by networks (DeBresson and Amesse 1992). Flor-
research on the spatial distribution of high- ida and Kenney (1988), for example, report
technology industry and employment. that innovation is a product of an underlying
Markusen, Hall, and Glasmeier (1986) observe, social structure of innovation which is geo-
for example, that high-technology industries graphically based. Jaffe (1989) and Acs,
are associated with higher wage rates and Audretsch, and Feldman (1992) note the pro-
higher levels of unionization. Some scholars ductivity effects associated with the proximity
note that these clusterings of innovative capac- of industrial and academic R&D. According to
ity are less the result of planning and of con- this line of thinking, the regional specialization
scious strategy than of chance, serendipity, or of industrial activity is an important facet of
"historical accidents" (Arthur 1990b; Scott and advanced industrial economies (Krugman
Storper 1990). It is difficult, however, to explain 1991a; 1991b; David and Rosenbloom 1990).
exactly why some regions are able to capture Locational clusters of economic activity and in-
the consequent benefits of serendipity, while novation are, in turn, the product of historical
other regions are not and their fortunes lan- processes or "path-dependence" (Arthur
guish. A number of scholars have noted the 1988; 1990a). As regions develop, certain ca-
increasing importance of innovation to the pacities are "locked-in" as resources are tai-
economic restructuring of advanced capitalist lored to the innovative activity of specific tech-
economies (Florida and Kenney 1990; Harvey nologies or industries. Historical processes re-
1989; Storper and Walker 1989). And there is inforce the regional specialization of innovative
a growing literature on national innovation sys- capabilities. Innovation thus benefits from the
tems (Nelson 1993). However, given what is congruent clustering of related institutions and
known about the innovation process, it is the synergies created by embedded networks

This content downloaded from

201.150.98.4 on Thu, 22 Jul 2021 16:55:32 UTC
All use subject to https://about.jstor.org/terms
212 Feldman and Florida

of individuals and institutions-relationships The Geographic Distribution of

not adequately incorporated into existing
Innovation in the United States
models.
in 1982
Drawing upon recent advances in geo-
graphic theory, particularly the concept of
We begin our analysis with an overview of
"geographic industrialization" (Storper and
the geographic distribution of new product in-
Walker 1989), we suggest that geography in-
novations in the United States in 1982. Our
deed plays a most fundamental role in the in-
source of data on commercial product innova-
novation process. Innovations are less the
tion is the 1982 census of innovation citations
product of individual firms than of the assem-
from trade journals and business publications
bled resources, knowledge, and other inputs
conducted by the Small Business Administra-
and capabilities that agglomerate in specific
tion (SBA). Unlike patent data which certify
places. Innovating firms and organizations har-
new inventions, the SBA innovation census re-
ness the institutions and the resources that
ports on the market introduction of new prod-
constitute the technological infrastructures of
uct innovations.1 The SBA innovation census
specific places. This infrastructural perspective
lists 4,476 product innovations; of these, 4,200
on innovation differs sharply with the prevail-
innovations contain information on the loca-
ing "location scanning" perspective. The latter
tion of the establishment that introduced the
suggests that individual firms freely scan the
innovation.
environment and select particular locations in
As Figure 1 shows, the geographic distribu-
accordance with the functional requirements
tion of product innovations is highly concen-
of firms. The former suggests instead that in-
trated among states. Eleven states account for
novation depends on a technological infra-
81 percent of the 4,200 innovations (Table 1).
structure of various resources and institutions-
When the absolute distribution of innovations
the indigenous manufacturing capabilities of
is converted to a rate of innovations per
networks of firms, the R&D efforts and capa-
100,000 manufacturing employees, geographi-
bilities of private enterprises and universities,
cal concentration persists (Figure 2). The rates
the concentrations of specialized commerciali-
of product innovation in New Jersey, Massa-
zation support services-that develop over
chusetts, and California are double the national
time. In time, as it were, specific places de-
rate; and seven other states-New Hampshire,
velop differential technological capabilities and
New York, Minnesota, Connecticut, Arizona,
capacities for innovation. Geography thus
Colorado, and Delaware-exceed the national
plays a fundamental role in the innovation
average.
process; it constitutes the spatial locus wherein
Table 2 sheds additional light on the relation-
the various elements of technological infra-
ship between innovation and other commonly
structure are organized.
used measures of innovative activity. Simple
Furthermore, given this general perspective
correlations between the 1982 SBA innovation
on innovation processes, we do not believe
data and patent counts, R&D expenditure, and
that serendipitous events "spark" economic
high-technology employment by state2 suggest
development: the spark of innovation and the
a close association. While R&D is considered
ability of an area to capture the benefits of
an input to innovation, patents and high-tech-
serendipity are rather the products of a well-
nology employment are often used as proxies
rounded technological infrastructure. The re-
of innovative output. The geographic distribu-
gional capacity to sustain innovation is thus
tion of patents and high-technology under-
embodied in institutions and resources that
score the patterns of concentration revealed in
reflect significant investments over time. In this
the SBA data. As Figure 3 demonstrates, pat-
sense, an area's underlying technological infra-
ents are geographically concentrated in Califor-
structure enhances the potential for innovation
nia and on the east coast in Massachusetts,
and shapes the locational choices of individual
New York, and New Jersey. Patents, however,
firms. Simply put, locational advantage and in-
exhibit more geographic dispersal than prod-
novative capacity stem from, and are embod-
uct innovations. Griliches (1990), Mansfield
ied in, the technological infrastructure of a
(1984), and Scherer (1983) all warn that the
place.

This content downloaded from

201.150.98.4 on Thu, 22 Jul 2021 16:55:32 UTC
All use subject to https://about.jstor.org/terms
 The Geographic Sources of Innovation 213

Counts of Innovation by State

1000 500 250 100 50 25 10 2

Source: U.S. Small Business 4
Administration Innovation -
Number of Innovations Citation Data, 1982.

Figure 1. Number of innovations by state in 1982.

number of patented inventions is not directly The geographic concentration of product in-
equivalent to a measure of innovative output novation is even more pronounced among
as many patented inventions never become particular industries. Table 3 shows consider-
commercially viable products while many suc- able specialization of innovative activity at the
cessful products are never patented. There is state level. California, for example, specializes
a higher incidence of patenting in the states of in electronics-related innovation. Indeed, it is
Ohio, Pennsylvania, Illinois, and Michigan-the the most innovative state in five electronics-re-
traditional manufacturing belt. This finding may lated sectors: computers, measuring instru-
reflect industry differentials in the propensity ments, communications equipment, electronic
to patent. For example, Scherer (1983) finds a equipment, and electronic industrial machin-
higher incidence of patenting in traditional in- ery. California's advantage reflects, in large
dustries such as industrial and residential measure, the broad infrastructure for electron-
equipment; stone, clay, and glass products; ics-related innovation that has grown up in
and household appliances. In addition, firms California's Silicon Valley over the past three
may opt not to patent in rapidly changing tech- decades (Saxenian 1985). Similarly, New Jersey,
nological fields such as advanced electronics with its world-class pharmaceutical and chemi-
because the technical detail required in patent cal complexes, leads in innovations related to
applications release proprietary design details drugs and medicine (Feldman and Schreuder
which can be easily exploited by competitors 1993). New York State, meanwhile, is a center
(Mansfield 1984). for innovations in photographic equipment-a

This content downloaded from

201.150.98.4 on Thu, 22 Jul 2021 16:55:32 UTC
All use subject to https://about.jstor.org/terms
214 Feldman and Florida

Table 1. Distribution of Innovation leader in innovations in metal fabrication.

by State. Pennsylvania, with a similarly long history of

Innovations heavy manufacturing in steel, electronic-power

per 100,000 equipment, and other sectors, accounts for the
Manufacturing lion's share of innovation in the field of general
State Innovations Workers
industrial machinery. And lastly, Illinois with its
New Jersey 426 52.33 massive industrial complexes around Chicago
Massachusetts 360 51.87 (Markusen and McCurdy 1989) and elsewhere
California 974 46.94
accounts for the largest share of innovation in
New Hampshire 33 30.84
New York 456 29.48 the domain of special industrial machinery.
Minnesota 110 28.65 The locational specialization of innovation is
Connecticut 132 28.51 further highlighted in the location quotients re-
Arizona 41 27.70
ported in Table 3.3 The average location quo-
Colorado 42 22.46
tient of 239.5 for the thirteen most innovative
Delaware 15 21.13
National 4200 20.34 industries offers a clear indication of a sig-
Rhode Island 24 18.46 nificant specialization in innovative activity by
Pennsylvania 245 18.28 state. Generally speaking, then, the product
Illinois 231 18.16
innovation data convey considerable regional
Texas 169 16.14
Wisconsin 86 15.61 specialization in innovative activity in states
Washington 48 15.38 that have developed specialized capacities for
Ohio 188 15.00 innovation in particular technologies and in-
Florida 66 14.60
dustrial activities.
Oregon 32 14.48

Source: Numbers of innovations are from the SBA innova-

tion data. Numbers of manufacturing workers are from the
1982 Census of Manufacturers (U.S. Bureau of the Census
1986).
A Geographic Model of
Innovation

fact which is not altogether surprising given the The geographic distribution of innovation is,
opto-electronics complex around Rochester- we believe, a function of an area's underlying
comprised of companies such as Kodak, technological infrastructure. As noted above,
Xerox, and Bausch and Lomb, among others this technological infrastructure consists of:
(Sternberg 1991). Innovation in metal fabrica- concentrations of industrial and university R&D
tion and industrial machinery are concentrated that enhance new product ideas and inven-
in the heavy manufacturing states of the indus- tions by providing sources of technological op-
trial Midwest. Ohio, long a center for steel pro- portunity; agglomerations of manufacturing
duction and metal-working for heavy manufac- firms in related industries that provide addi-
turing and consumer durable goods, is the tional sources of expertise and tacit knowl-
edge, particularly the capacity to translate new
ideas into actual commercial products; and
networks of business-service providers that
Table 2. Correlation Analysis of Alterna-
support the overall product innovation process
tive Measures of Innovation.
by supplying knowledge and information on
Employ- technological trends and product markets. The
Innovation Patents R&D ment
congruent clustering of these several inputs
Innovation 1.0000 creates scale economies, facilitates knowledge-
Patents .9344 1.0000 sharing and cross-fertilization of ideas, and
R&D .8551 .8804 1.0000 promotes face-to-face interactions of the sort
Employment that enhance effective technology transfer. The
.9737 .9888 .7013 1.0000 geographic proximity of these inputs promotes
Sources: Patent counts by state are from Jaffe (1989). High- information transfer and spill-overs that lower
technology employment data are from the U.S. Office of
the costs and reduce the risks associated with
Technology Assessment (1984). R&D expenditures are
from the National Science Foundation as reported by Jaffe innovation. Furthermore, the clustering of
(1989). these regional "stocks" of innovative capabili-

This content downloaded from

201.150.98.4 on Thu, 22 Jul 2021 16:55:32 UTC
All use subject to https://about.jstor.org/terms
 The Geographic Sources of Innovation 215

Innovations per
_ ~~~100,000 Manufacturing Workers

Number of Innovations per 100,000 Manufacturing Workers

U >28.0 U 7.1 - 14.0 01J -3.5

* 14.1- 28.0 3.6-7.0 D Sources: U.S. Small Business Administration

Innovation Citation Data & 1982
Census of Manufacturers.

Figure 2. Innovations per 100,000 manufacturing workers in 1982.

ties and resources are embodied in human and graphic locations of commercial innovation.
institutional forms and interrelations that reflect These data represent an advance on previous
a cumulative history of investments made in research that measures innovation with the
specific places over long periods of time proxies of patents (Jaffe 1989), high-technology
(Sweeney 1987; Tassey 1991; Storper and firms and employment (Markusen, Hall, and
Walker 1989). Glasmeier 1986), R&D expenditures (Malecki
1983), or R&D personnel.
Although the SBA data offer a more direct
Product Innovations: The Dependent measure of product innovations, our source
Variable has its limitations and potential biases. First, the
SBA innovation citations are compiled from a
The dependent variable in this model, wide variety of industry announcements and
namely the commercial product-innovation ci- trade publications, and these may be biased
tations compiled by SBA in 1982, is based on toward unusual or special-interest products.
the only data currently available on the geo- Second, the SBA data are compiled by states.

This content downloaded from

201.150.98.4 on Thu, 22 Jul 2021 16:55:32 UTC
All use subject to https://about.jstor.org/terms
216 Feldman and Florida

< ~~~~~~Location of Patents

Aube o f Paet

Source: 1982 data from Commisioner

**of Patents and Trademarks
5000 3000 1500 1000 500 300 100 50 Annual Report, 1985.

Figure 3. Number of patents by state in 1982.

Using the state as the unit of analysis inevitably generally regarded as ones of considerable in-
obscures spatial processes that occur within a novation (U.S. Office of Technology Assess-
state or across state boundaries. While we ment 1984). With the emergence of new high-
would prefer to use sub-state units of analysis technology industries such as personal com-
and then aggregate in accordance with func- puters, computer work-stations, software and
tional linkages and dependencies (Czmanski biotechnology, innovation occurred in en-
and Ablas 1979), such data are simply unavail- trepreneurial start-up companies as well as in
able. Since the SBA data are the best that we the larger, more established firms such as IBM
have, we attempt, alternatively, to minimize and DuPont. In this regard, 1982 provides a
potential sources of aggregation bias by intro- particularly useful vantage point on the geog-
ducing an index of geographic concentration raphy of innovation. The SBA data, moreover,
as a control variable. Third, the SBA data are distinguish between the location of the estab-
only available for one year, 1982. While cross- lishment responsible for the major develop-
sectional data such as these preclude consid- ment of an innovation and the location of the
eration of subsequent technological and indus- corporate headquarters or parent company.
trial restructuring, the SBA's selection of 1982 We, of course, use the establishment location
has some compensating virtues. The early in our analysis.
1 980s is a particularly useful time to explore the The dependent variable in our model, inno-
geography of innovation, since these years are vative output (INNis), is the number of innova-

This content downloaded from

201.150.98.4 on Thu, 22 Jul 2021 16:55:32 UTC
All use subject to https://about.jstor.org/terms
 The Geographic Sources of Innovation 217

Table 3. State Competitive Advantage in Innovative Industries.

Industry N Leading State n Location Quotient

Computers 954 California 356 167.8

Measuring Instruments 668 California 134 126.4
Communications Equipment 376 California 116 132.2
Electronic Equipment 261 California 128 211.3

Medical Instruments and Supplies 228 New Jersey 57 248.2

General Industrial Machinery 164 Pennsylvania 25 261.5

Drugs 133 New Jersey 52 381.3

Special Industrial Machinery 116 Illinois 11 171.4
Misc. Fabricated Metal Products 105 Ohio 18 384.0
Electronic Industrial Machinery 74 California 17 94.4
Photographic Equipment 61 New York 18 260.0
Plastic and Synthetic Materials 51 Texas 10 491.7
Cleaning Preparations 50 New York 10 183.3

N indicates the total number of innovations for an industry; and n indicates the number of innovations for a state.

tions for an industry i in a state s in 1982. When the market in 1982 would benefit from the
these data are stratified by state and by indus- stock of infrastructural resources that had been
try, a large number of zero cells result. In order in existence for the preceding decade. Accord-
to proceed with the estimation, we confine our ingly, we measure the stocks of the four inno-
analysis to the thirteen most innovative three- vative inputs as the average annual expendi-
digit Standard Industrial Classification (SIC) tures for each in the ten years prior to the 1982
code industries. Each of these industries ac- introduction of innovations into the commer-
counts for 50 or more innovations, and the cial market.
thirteen industries as a whole account for 80 University research provides basic knowl-
percent of total innovations in our sample (Ta- edge that may be critical for the innovation
ble 3). The remaining 82 industries with one to process. Nelson (1986) and Mansfield (1991)
50 innovations account for just 20 percent of note that university R&D enhances the stock
all innovations in 1982. of basic knowledge, generates increased tech-
nological opportunities across a wide range of
industrial fields, and increases the potential
Technological Infrastructure: The productivity of private industrial R&D. Overall
Independent Variables then, university R&D has a positive effect on
commercial innovation and generates a sig-
The four independent variables in the model nificant social rate of return-in excess of 25
are indicators of technological infrastructure. percent according to one recent study (Mans-
They are: (1) firms in related manufacturing field 1991). Figure 4 presents the distribution
industries, (2) industry R&D, (3) university of university-research expenditures. While the
R&D, and (4) business services. These are sup- concentration of university R&D in a few major
plemented by a series of control variables. Be- clusters, for example Boston-Cambridge and
cause innovation is a process, it is charac- the San Francisco Bay area, is well known,
terized by a time lapse between the first stages other states, notably New York, Texas, and
of invention and the final stages of commer- Maryland also have high expenditures on uni-
cialization. The length of this lag is difficult to versity research. The presence of university re-
specify, however. A recent study estimates that search simply may not be sufficient by itself to
the lag between an academic research finding guarantee innovation and technology-based
and the commercial introduction of a new spin-offs (Feldman 1994b). There is some evi-
product averages seven years, with a standard dence that the co-location of university and
deviation of two years (Mansfield 1991). We industrial R&D at the state level tends to exert
thus assume that innovations introduced into positive impacts on the generation of patents

This content downloaded from

201.150.98.4 on Thu, 22 Jul 2021 16:55:32 UTC
All use subject to https://about.jstor.org/terms
218 Feldman and Florida

2000 . .... . .... .....

.......
....
......
.....
..... ... ..
.... ....

1500.0
!airlines:

0
1600,
0
400-
0 0 ..... .....
1200. .... .......

0 1000.0
..... .....

. ..... . . .......

0 800,0 ..... ......

.. . .... ....
.... .. ....

0 ....... ... . ....

....... ...... ...
. .... . .. ..

600.
....... ... ....

. ..... .......

. .. .... . ... .... . .... !!!AS CI I Is .

...... .... ...... . ..... . .... ......

400.0, ...... .... ...

. .. ...... ......

.. . .... ..... ......

.... . ......

2 0 0 0 . .... . .... .. ......

...... ....
...... . . ... .......... .....

Oj . ...... 1800.0
. ... .. ..... .... . .

;African ..... ..... Africa n! . .... .....

...... ...... . . .

.... . ...... .
...... ..... ....... ..... . . .... .......
..... ..... ....... ......

...... ..... Tiii] 400 0

... . ......
.... ...... ...

...... .....

...... . ..... ... . . ....

.......
...........
...... ....... ......
..... ..... ... . .. ...... ....

..
...
...... ......
.
. ........
......
... ..... .. .
... ... ... . ..... ...... ....

...... ......
.. ...... ......
.. ..... ... . ......

... .... ....... ... ... .....

. ..... . ........

... ...... ....... . .....

..... ...... ... . ..

z - -
. ... ...... ...

.. ... . ... ....

0.0
Source: National Science Board,
Science & Eggineefing
Indicato , Tenth Edition,
1991.

Figure 4. Distribution of university R&D expenditures in 1989.

and innovations (Jaffe 1989; Acs, Audretsch, required for new product development. With
and Feldman 1992). But the evidence is not the notable exceptions of firms such as ATT
unequivocal; Markusen, Hall, and Glasmeier and IBM that conduct basic research in labora-
(1986) report that university-research expendi- tories, industrial R&D laboratories tend to spe-
tures may have a negative effect on the loca- cialize in market-oriented R&D, and more spe-
tion of high-technology industry.4 cifically, in the translation of scientific and tech-
Because of the vast differences in the scope nical information into new innovations. Geog-
and commercial applicability of university re- raphers, in particular, have highlighted the
search, our measure of university R&D (UNIV1s) regional concentration of industrial R&D and its
is based on funding at the level of academic important role in the innovation process and
department. Using data from the National Sci- in the formation of regional innovation com-
ence Foundation's (NSF) Survey of Science Re- plexes (Stohr 1986; Tassey 1991). Malecki's
sources, we assign academic departments to (1983) study of the geographic distribution of
relevant industries at the level of the two-digit R&D noted the regional specialization of R&D
SIC code. Innovations in industry SIC 283, activity, in general, and the marked concentra-
Drugs, for example, are linked to research in tions on the East and West Coasts, in particular.
the academic departments of medicine, biol- Figure 5 demonstrates the continuation of this
ogy, chemistry, and chemical engineering pattern with a proportional representation of
(Feldman 1994a). state industrial R&D expenditures. Here it is
Industrial R&D laboratories also serve as important to note that, as we found with uni-
sources of scientific and technical knowledge versity research, the geographic distribution of

This content downloaded from

201.150.98.4 on Thu, 22 Jul 2021 16:55:32 UTC
All use subject to https://about.jstor.org/terms
 The Geographic Sources of Innovation 219

Industrial R & D Expenditures

~~~H ~

No data for:
DE, ID, MT
NH, ND, SD,
WV, WY.

Source: National Science Board, Science

and Engineering Indicators, Tent
Edition, 199 1. Data are for the year
1989 as geographic coverage for
earlier years was subject to greater
suppression for states. Data are in
...... .... ...............m illions of current dollars.

20,000 10,000 5,000 2,500 1,000 500 200 100

Millions of Dollars

Figure 5. Industrial R&D expenditures in 1989.

a single component of the technological infra- ments prevent data disclosure in 21 of the 50
structure does not mirror precisely the geo- states. But these omissions are not so serious
graphic distribution of innovations. The loca- as might first appear. The 21 states for which
tion of industrial R&D more closely resembles R&D data are unavailable account for just 325
the location of patents. This incongruence is innovations or 7.7 percent of all innovations.
not surprising from our point of view since Conversely, we have industrial R&D expendi-
multiple resources are required to create the tures for 29 states, and they account for 92
technological infrastructure on which innova- percent of all SBA-reported innovations and 81
tion depends. percent of university-research expenditures in
Industrial R&D (INDiS) is measured as the ex- 1982. Our model, thus, uses the innovation
penditures for in-company industrial R&D as data for the 29 states and 13 industries that
reported to the NSF Science Resources Survey. together account for the overwhelming num-
This report has limitations, however. It does not ber of product innovations in 1982.
include the cost of R&D contracted to univer- Turning next to our third independent vari-
sities and colleges, nor to nonprofit organiza- able, we note that geographers and econo-
tions, research institutions, and other compa- mists have often pointed to the role of prox-
nies. In addition, NSF's confidentiality require- imity in the innovation process. Cities and re-

This content downloaded from

201.150.98.4 on Thu, 22 Jul 2021 16:55:32 UTC
All use subject to https://about.jstor.org/terms
220 Feldman and Florida

gions serve as "incubators" of innovations tion process, as for example when innovations
(Thompson 1962). More specifically, in the in semiconductors spill over into electrical,
case of high-technology regions, networks of consumer electronics, and computer indus-
manufacturing firms are particularly crucial for tries.
new ideas and sources of knowledge for inno- The presence of concentrations of firms in
vation (Stohr 1986; Storper and Walker 1989; related manufacturing industries (RELPRESs) is
Sayer and Walker 1993). Concentrations or ag- measured as value-added for the major indus-
glomerations of firms in related industries pro- try two-digit group that encompasses the
vide a pool of technical knowledge and exper- three-digit industry under consideration (Figure
tise and a potential base of suppliers and users 6). Returning to our example of the drug indus-
of innovations. These networks play an espe- try (SIC 283), we use the value added in the
cially important role when technological related industrial group of Chemicals and Allied
knowledge is informal or "tacit" in nature, Products-SIC 28.
when knowledge and ideas are hard to codify, Business-service providers constitute our
and when "practical mastery" plays a large role fourth and final independent variable. These
(Storper and Walker 1989). Suppliers and end- providers play key roles in regional innovation
users of a technology also provide an impor- complexes (Stohr 1986) such as Silicon Valley
tant source of additional knowledge and ideas (Saxenian 1985) and Route 128 (Dorfman
(Von Hippel 1988). Concentrations of these 1983). Providers such as commercial-testing
firms foster important synergies in the innova- laboratories, market-research firms, and patent

ooo00 :

6?~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~..... ...- _s
500000

4 ;00w

o i E100200E;010 . . . iE:i2i5::1EE;1i . . i~ii0i:1 :!0ESE IN }6

r l000000 t t<tictur

30~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
O - - O t o . _ U . ; ; :fI- - - 1 :1:; 1 : l-L-:01Ei02i01E00:i~ii:0ili0020: i~ii1000:;E 21ii~i:1E 1:: 1- - 1 1; 1:: E 0i . . . . . . . .

000.0~~~~~~~~~~~~~~~~~~~~~~~~~~~ .. ....... . . ... . ..... .

| Source: 1977 Census of Manufacturing -,-e2 d$,O o}

Figure 6. Distribution of manufacturing value added in 1977.

This content downloaded from

201.150.98.4 on Thu, 22 Jul 2021 16:55:32 UTC
All use subject to https://about.jstor.org/terms
 The Geographic Sources of Innovation 221

attorneys offer important sources of informa- product innovation. Only one SIC category-
tion on technological and product opportuni- SIC 7397, Commercial Testing Laboratories-is
ties, marketing and sales trends, regulations clearly linked with the introduction of new in-
and standards, the law, and financing of the novations. In the absence of other data, we use
sort required to bring innovations to market value-added in SIC 7397 as a surrogate indica-
and to effectively position new product offer- tor for specialized business services (BSERV S).
ings. MacPherson (1991) finds a strong corre- The geographic distribution of these services is
lation between the usage of external producer presented in Figure 7. Once again, we observe
services and new products; other research in- geographic concentration in California, New
dicates that specialized producer services tend York, New Jersey, Massachusetts, Pennsylva-
to locate near their clients (Coffey and Polese nia, and Texas. Of all of our independent vari-
1987). ables, the location of specialized business serv-
Business services comprise a wide range of ices most closely resembles the location of
activities. Although a variety of firms provide product innovation. However, these mappings
knowledge of the market and the commerciali- are not perfect, as some states, such as Florida,
zation process, the Census classification sys- Louisiana, Connecticut, and Rhode Island have
tem makes it difficult to isolate those services a relatively higher representation of business
that directly support the innovation process. services when compared to innovation. But
For example, the census grouping of all legal that is to be expected since our thesis suggests
services under one SIC classification (SIC 8111) that it is the interactions and synergies of these
makes it impossible to isolate the critical serv- elements of the technological infrastructure
ices of patent attorneys for the process of that provide the key to explaining the geo-

Distribution of Producer Services

Receipts of Commercial Testing Lab

(millions of dollars)

Source: The date represent the ten year

average receipts for commercial
testing labs and are compiled for the
Census of Service Industries.
125,000 75,000 50,000 25,000 10,000 5,000 1,000

Figure 7. Distribution of producer services as measured by receipts

This content downloaded from

201.150.98.4 on Thu, 22 Jul 2021 16:55:32 UTC
All use subject to https://about.jstor.org/terms
222 Feldman and Florida

graphic distribution of innovation, that is, these R&D (Mowery and Rosenberg 1989) and that
elements in space work interactively rather industrial R&D expenditures would be allo-
than in isolation. cated to regions with strong and related uni-
versity research programs (Jaffe 1989; Mans-
field 1991). The industrial R&D equation ac-
Implementing a Geographic Model of cordingly regards that variable as a function of
Innovation: Specification and Estimation university research, firms in related industries,
and corporate headquarters, (HDQRT.s)-an in-
Our geographic model of innovation regards dicator of Fortune 500 corporate headquar-
innovation as a function of four classes of in- ters.6 The equation is as follows:
novative inputs: networks of firms in related
manufacturing industries, concentrations of
log(/ND.,) = co1log(UNIV.5) + co2log(RELPRES.,)
university R&D, concentrations of industrial + (o3log(HDQRT5) + (o410g(POP.5) + as (2)
R&D, and concentrations of business-service
providers. In formal terms, innovative output The third equation in the model examines
(INNjs), or the number of innovations for indus-
the determinants of the location of university
try i in a geographic area s, is a function of: R&D. Given that firms tend to use research at
university research (UN!V1s), industrial R&D nearby universities and the strong tendency for
(INDis), networks of related firms (RELPRES1s), the co-location of industrial and university
and specialized business services (BSERVis): R&D, the third equation explores the interac-
tion between university research, industrial
Log(INNi5) = jIog(UNIV15) + 210og(IND/,) + R&D, and firms in related industries. This equa-

W3Log(RELPRES15) + P41og(BSERV 5) + tion also includes a variable for federally

P5CONC., + P610g(POP.,) + funded research and development centers

f710g(SALESj.) + ?js (1) (FFRDC's)-an indicator of the university's
ceptiveness to participate in technology trans-
fer with industry. Equation 3 is specified as fol-
Because we expect knowledge spill-overs lows:
across related technological fields, the sub-
script refers to industries that use similar tech-
nology. Two variables are added to control for
log(UNIV/s) = y1log(XjND/s) +
y2log(>RELPRES1s) + y3FFRDC.s +
aggregation bias. State population (POP.s) con-
Y410g(POP.S) + VIS (3)
trols for variable state sizes and thus facilitates
cross-state comparisons. An index of geo-
graphic concentration (CONC.S) controls for The fourth and final equation in the system
within-state variation and compensates for the examines the determinants of the location of
use of states as the unit of observation.5 This specialized business services. The presence of
index measures the share of the state's value such services is expected to reflect the client
of manufacturing shipments held by the state's base on which these services depend, for ex-
largest SMSA. A third variable, industry sales ample, industrial R&D laboratories (Coffey and
(SALESj ), controls for variable demands for in- Polese 1987). We also expect that specialized
novations generated within an industry. The producer services, such as commercial testing
model further specifies innovation as a recur- laboratories, are a function of the overall base
sive system consisting of four equations and a of business services, hence our inclusion of a
series of individual equations that isolate the variable (TOTALBSERV.S) that represents the
determinants of industrial R&D, university stock of receipts for general management and
R&D, and business services. consulting services (SIC code 7392). Equation
The second equation in the model examines 4 is thus specified as follows:
the determinants of the location of industrial
R&D. We expect that R&D laboratories will log(BSERV.s) = c1log(IND.s) +
tend to locate either near production facilities cX21og(TOTALBSERV.s) +
or near firm headquarters (Malecki 1985; cC310g(POP.s) + i., (4)
1990), that is to say, the locational patterns will
reflect the historical development of industrial The four equations form a recursive system

This content downloaded from

201.150.98.4 on Thu, 22 Jul 2021 16:55:32 UTC
All use subject to https://about.jstor.org/terms
 The Geographic Sources of Innovation 223

for testing the effects of technological infra- The new dependent variable, Log[1 0(1 +
structure on commercial product innovation INNi,)], eliminates zero values yet preserves
and for isolating the determinants of the com- the relative ranking of innovative observations.
ponents of that infrastructure. The system is A second statistical issue is the likelihood of
recursive in the sense that there is no direct disturbance. Owing to spatial autocorrelation,
feedback from the second, third, and fourth the structure of the error terms is often a prob-
equations to the first equation. And because of lem with regional cross-sectional data. A ran-
the time lag in translating successful innovative dom shock affecting economic activity in one
output into a new round of expenditures on state may, for example affect economic activity
innovative inputs, the system is not simultane- in adjacent states when the several states ex-
ous in the usual sense. Summary statistics for hibit economic linkage. In this case, distur-
the system appear in Table 4. bance terms in contiguous or related states
Before presenting the results, however, we will be related and the parameter estimates will
need to attend to three statistical issues: the not be efficient. The results of Durbin-Watson
censored nature of the dependent variable, the tests for autocorrelation on various orderings
likelihood of disturbances, and the prospect of of the observations are inconclusive, however;
multicollinearity in the independent variables. hence we make no corrections for spatial auto-
In the first instance, the dependent variable is correlation. A second concern is the possibility
censored in the sense that the number of in- of heteroscedasticity of the error term. The
novations by state and industry will either be Breusch-Pagan test for heteroscedasticity in the
zero or some positive integer. Because the es- error term reveals none in the innovation
timation of a production function such as equa- equation specification.
tion (1) relies on a log-log transformation, ob- The third statistical issue is the presence of
servations with a value of zero present a prob- multicollinearity in the independent variables.
lem. In order to estimate the innovation equa- This is particularly a problem with cross-sec-
tion, it is necessary to transform the dependent tional geographic data since these data may be
variable, INNi,, and thus eliminate zero values.affected by some common trend or underlying

Table 4. Summary Statistics.

Variable Mean St. Dev. Min Max

Innovations (INNis)a 7.72 24.49 0.00 365.00

University Research 32.52 59.41 0.30 380.60
(UNIVs)b
Industry R&D (IND.s)b 582.90 818.51 9.00 3883.00
Related Industry Presence 903.06 1062.60 4.04 4404.00
(RELPRESis)b
Business Services 13.88 17.02 0.50 89.52
(BSERVs)b
Geographic Concentra- 0.41 0.23 1.10 0.94
tion Index (CONC.s)b
Industry Sales (SALESi.)b 9.82 3.48 3.73 16.24
Sales of Fortune 500 Firms 30,281.00 55,249.00 100.00 271,700.00
(HDQRTS)b
FFRDC (FFRDC.s)b 0.59 1.00 0.00 4.00
General Business Services 443.21 545.71 40.71 2440.00
(TOTALBSERV.s)b
>jNDisc 84.49 120.35 0.49 603.90
>RELPRESisc 1732.10 1901.10 59.17 8381.00
State Population (pops)d 5919.07 4905.33 955.00 22,350.00
aMeasured as integer counts of innovations.
bUniversity research expenditures, industrial R&D expenditures, r
ness services and total industry sales are in millions of 1972 dollars.
CThe summation operator on industrial R&D expenditures (INDIs) and value added and RELPRESIs indicates a total for
industries relevant to an academic department (Feldman 1994a).
dPopulation is measured in thousands.

This content downloaded from

201.150.98.4 on Thu, 22 Jul 2021 16:55:32 UTC
All use subject to https://about.jstor.org/terms
224 Feldman and Florida

Table 5. Correlation Matrix for Inde- Table 6. Model Results.

pendent Variables.
Dependent Variable: Log[1 O(INNi, + 1)]
IND.s UNIVis RELPRESIs BSERVs Log(IND.s) 0.241 a (0.054)
Log(UNIVjs) 0.1 55a (0.043)
IND.s 1.00 Log(RELPRES1s) 0.1 44a (0.045)
UNIVis 0.68 1.00 Log(BSERV.s) 0.272a (0.055)
RELPRESIs 0.63 0.39 1.00 Log(POP.s) 0.054b (0.030)
BSERV.s 0.73 0.56 0.53 1.00 Log(SALESi.) -0.236a (0.113)
Note: Reported correlations are for the log values of each CONC.s 1.021 a (0.189)
of the variables.
Dependent Variable:
Log(UNIV.s) 0.566a (0.074)
Log(RELPRES.s) 0.466a (0.089)
Log(HDQRT.s) 0.1 80a (0.040)
Log(POP.s) 0.0486b (0.026)
state characteristics. The correlation matrix of
the innovative inputs indicates some evidence Dependent Variable: Log(U
of multicollinearity (see Table 5), and this inter- Log(YJNDjs) 0.256a (0.039)
dependence may result in higher variances in Log(>RELPRES1s) 0.338a (0.047)
FFRDC.s 0.539a (0.054)
the parameter estimates and lesser statistical
Log(POP.s) -0.11 8a (0.031)
significance in the coefficients.
Given these concerns, the four equations are Dependent Variable: Log(BSERV.s)
estimated using Three-Stage Least-Squares Log(IND.s) 0.1 09a (0.040)
(3SLS) Regression. We also use instrumental Log(TOTALBSERV.s) 0.707a (0.057)
Log(POP.s) -0.295a (0.215)
variables to correct for correlation across the
Note: The instruments used include all of the exogenous
equations. In the model, these instruments in-
variables appearing on the right-hand side of the equa-
clude all of the exogenous variables that appear tions in the model. Standard errors are in parentheses.
on the right-hand side of the equations. The The number of observations is equal to 377.
asignificance of at least .95.
first stage of estimation uses the contempora- bsignificance at .90.
neous values of the variables and then adjusts
for the covariance matrix of the residuals. The
interrelationships between the variables and
the equations indicate that the efficiency of the
parameter estimates increase with this type of
tive activity within states is indeed related to
estimation.7
the factors that comprise their underlying tech-
nological infrastructure.
Consider first the innovation equation. The
Geographic Sources of coefficient estimates for all of the independent
Innovation: The Empirical Results variables are positive and significant. The size
of the industrial R&D coefficient suggests that
Table 6 presents the results of the estimation it plays a key role in the innovation process.
of the innovation modei. Generally speaking, The coefficient of university R&D is similarly
the model performed well. The coefficients for positive and significant. The latter confirms the
all four components of the technological infra- results of Jaffe (1989) and Acs, Audretsch, and
structure-industrial R&D, university research, Feldman (1992) as it demurs from the findings
related industries, and business services-are of Markusen, Hall, and Glasmeier (1986) who
positive and statistically significant. Our find that university R&D is negatively related to
findings are also robust, judging from runs of a high-technology industry and employment.
number of permutations of the model. The first The coefficient for the presence of related
run measures all variables on a per-capita basis, manufacturing industries is likewise positive
and the basic results (signs and significance) and significant, that is, concentrations of and
remain the same. The second run tests the synergies among firms in related industries
robustness of the results for states with more tend to foster innovation. Furthermore, the
than one large manufacturing center. Once size of the coefficient for business services in-
again, the basic results are unchanged. Simply dicates that the presence of these services has
put, the empirical results suggest that innova- a particularly positive effect on the innovation

This content downloaded from

201.150.98.4 on Thu, 22 Jul 2021 16:55:32 UTC
All use subject to https://about.jstor.org/terms
 The Geographic Sources of Innovation 225

process, perhaps because these services are The findings of the business services sub-
critical in the concluding marketing phase of equation indicate that the presence of special-
the commercialization process. ized business services is positively related to
Turning now to the model's sub-equations, industrial R&D and to the overall business-serv-
the coefficients for the industrial R&D equation ices sector. In other words, specialized busi-
are all positive and statistically significant. In- ness services, in this case commercial testing
dustrial R&D is related to university R&D ex- laboratories, are co-located with their principal
penditures, concentrations of firms in related clientele, the R&D laboratories. In addition,
manufacturing industries, and the presence of concentrations of specialized producer serv-
corporate headquarters. The magnitude of the ices are related to a large overall business serv-
coefficient of university R&D expenditures, as ice sector.
expected, suggests an especially strong rela- Taken together, these findings provide the
tionship between university and industrial beginnings of an explanation for the dynamics
R&D. This result affirms that university R&D of an area's technological infrastructure. The
increases technological opportunities available factors that comprise the region's technological
in a state or region and provides incentives to infrastructure work together to create an over-
invest in private industrial R&D to exploit the all capacity that is conducive to innovation.
stock of basic scientific knowledge (Nelson Each of the components must be in place for
1986). The coefficient for the presence of re- innovation to occur; however it is the interac-
lated industries suggests a fairly strong relation- tion and synergy among these components
ship between industrial R&D and the broader that accounts for a higher propensity to inno-
industrial base. This is not surprising since in- vate in particular places. Above all, our findings
dustrial R&D tends to feed off, as well as sup- indicate the mutual reinforcement of the four
port, clusters of manufacturing activity. major components of technological infrastruc-
The empirical findings for the university R&D ture: industrial R&D, university R&D, firms in
sub-equation indicate a close association be- related industries, and business services. The
tween it and industrial R&D and related indus- synergies among these four components yield
tries. The relationship between university R&D a technological infrastructure with a high pro-
and the presence of corporate headquarters, pensity for product innovation. The innovative
however, is somewhat weaker, though still sig- capacity of an area-in this case, states-hinges
nificant. The coefficient for federally funded on this underlying technological infrastructure.
research centers is positive and statistically sig- Thus, our empirical results confirm that ag-
nificant. In sum, the association between uni- glomeration, and geography more broadly,
versity R&D expenditures and both industrial play significant and important roles in the or-
expenditures on R&D and industrial activity ganization and mobilization of knowledge in
and industrial R&D in related fields reaffirms behalf of commercial product innovation.
Mansfield's (1991) finding that private firms
utilize research findings generated from nearby
universities. On the whole, it appears that uni-
versity R&D has a greater effect on industry
Some Thoughts on the
R&D than vice-versa. Indeed, the impact of
university R&D on industrial R&D has twice the
Geography of Product
magnitude of the impact of industrial R&D on Innovation: Today and
university R&D. University R&D may therefore Tomorrow, Here and Abroad
play a critical role in the innovation process by
attracting industrial R&D and by leveraging in- We have explored the geography of innova-
dustrial activities. However, the relatively tion, suggesting that it is unusually dependent
smaller effect of industrial R&D on the univer- on an area's technological infrastructure. We
sity R&D may be explained by the fact that a have oriented our analysis around an empirical
large proportion of total university R&D, almost model of the innovation process which intro-
two-thirds, is provided by the federal govern- duces a new, and previously unavailable,
ment and as such it may be less responsive to measure of innovative output at the state level.
state industrial priorities and concerns (Na- Our model presumes that innovation is a func-
tional Science Board 1989). tion of an area's underlying technological infra-

This content downloaded from

201.150.98.4 on Thu, 22 Jul 2021 16:55:32 UTC
All use subject to https://about.jstor.org/terms
226 Feldman and Florida

structure, which, in turn, consists of: university capacity to innovate is the product of com-
R&D, industrial R&D, agglomerations of related plexes of enterprises and R&D; networks of
industry, and specialized business services. institutions and institutional resources; concen-
The model is formulated as a recursive system trations of human talent, knowledge, and skill;
in order to improve our understanding of the and a legacy of sustained investment in an
interrelationships between the four innovative area's technological capability. Our findings,
inputs noted above. therefore, reinforce Storper and Walker's con-
The findings of the model confirm the hy- cept of geographic industrialization (1989)
pothesis that innovation is a function of an while adding to it the related notion that the
area's technical infrastructure. Innovation is re- sources of innovation that propel the contem-
lated to the geographic concentrations of in- porary processes of industrialization are them-
dustrial R&D, university R&D, related indus- selves geographic in nature. Our results also
tries, and business services. Our results imply redirect attention to the ways in which particu-
significant synergy and mutual reinforcement lar places have acquired a comparative advan-
among the factors that comprise the techno- tage for innovation and economic develop-
logical infrastructure. ment. If we are correct, locational advantage
Our findings further suggest that there is would seem to reflect cumulative investments
considerable geographic specialization in the in human and technological capability in spe-
technological infrastructures of various places. cific places, more so than the conventional
The capacity to innovate is very much the his- natural advantages of land, labor, and capital.
torical legacy of specialized concentrations of In the modern economy, locational advantage
R&D, industrial activity, and support services in the capacity to innovate is ever more de-
that build up in particular places over time. In pendent on the agglomeration of specialized
other words, different places are the sources skills, knowledge, institutions, and resources
of particular types of innovation. California- that make up an underlying technological in-
with its clusters of high-technology electronic frastructure.
producers, suppliers, business service provid- At a broader level, our findings provide a
ers, and venture capitalists in places like Silicon deeper understanding of innovation as a geo-
Valley and elsewhere-specializes in innova- graphic as well as an economic process. For
students
tions related to electronics. New Jersey, with its of capitalist development from Adam
massive chemical and pharmaceutical com- Smith to Karl Marx and Joseph Schumpeter,
plexes, is the center for innovations in drugs, innovation has been regarded as a primary
medicines, and medical equipment. Innova- source-if not "the" primary source-of eco-
tions related to photographic equipment and nomic growth and development. For such
opto-electronics are concentrated in New theorists, innovation is the product of individ-
York, most notably in the Rochester-area's ual capitalist firms, entrepreneurs, and organi-
opto-electronics complex composed of Xerox, zations which function to organize and har-
Kodak, Bausch and Lomb, and related suppli- ness the various technological and organiza-
ers. The industrial Midwest, with its history of tional inputs required for innovation, profit,
manufacturing infrastructure in steel, automo- and growth. But, as we have seen, the capacity
tive, appliance, and consumer durable produc- for innovation extends far beyond the bounda-
tion, is the primary source of innovations in ries of the individual firm. In the United States
metal fabrication and industrial machinery. Yet today, innovation is no longer the province of
while each of these complexes specializes in a the inventor, the risk-taking entrepreneur, the
specific type of innovations, they all depend insightful venture capitalist, or the large re-
upon a set of underlying factors that comprise source-rich corporation. Innovation instead has
a technological infrastructure for generating its sources in a broader social and spatial struc-
new ideas and bringing them to the market. ture-a landscape of agglomerated and syner-
Our findings thus suggest that not only does gistic social and economic institutions welded
geography play a central role in the innovation into a technological infrastructure for innova-
process, but further that innovation is itself a tion. It is in this fundamental sense that geog-
geographic process. Geography, in an integral raphy organizes the innovation process and
sense, organizes and advances innovation. The helps sustain the spatially uneven growth and

This content downloaded from

201.150.98.4 on Thu, 22 Jul 2021 16:55:32 UTC
All use subject to https://about.jstor.org/terms
 The Geographic Sources of Innovation 227

progress of advanced technological econo- did not use the log of the geographic concentra-
tion variable because there is no strong a priori
m ies.
functional specification and because the estima-
tion of the innovation equation with a log transfor-
mation of this variable yields similar results.
6. No simultaneity between an area's innovative suc-
Acknowledgments
cess and industrial R&D allocation is estimated be-
cause of the time lag in introducing an innovation
Research funding from the U.S. Economic Devel-
to the market. Industrial R&D typically precedes
opment Administration, U.S. Small Business Admini-
market introduction by four to five years and thus
stration, and the Duquesne Light Fund for Economic
a firm's response to successful innovative outputs
Development at Carnegie Mellon is gratefully ac-
will reflect that time delay.
knowledged. The authors would like to thank Zoltan
7. The system of equations is estimated using Seem-
Acs, David Audretsch, Wesley Cohen, Mark Kamlet,
ingly Unrelated Regression which essentially
Edward Malecki, Donald F. Smith, Paula Stephan, and
amounts to the first two stages of three-stage least
the anonymous referees for thoughtful comments.
squares. A comparison of the parameter estimates
Appreciation is expressed to the Center for Regional
indicates a gain of efficiency with three-stage least-
Economic Issues at Case Western Reserve University
squares regression owing to the presence of
and Marshall Feldman at the University of Rhode Is-
cross-equation correlation.
land for assistance with various data. We would also
like to thank Tom Rabenhorst and the cartography
group at the University of Maryland Baltimore County
for their assistance. Cynthia Brandt and Gail Cohen
Shaivitz provided research assistance. The title of this
paper is respectfully and unabashedly borrowed References
from the work of Eric Von Hipple.

Acs, Z. J., Audretsch, D. B., and Feldman, M. P.

1992. Real Effects of Academic Research: Com-

Notes ment. American Economic Review 81:363-367.

Arthur, W. B. 1988. Urban Systems and Historical

1. While precise data on the economic significance Path Dependence. In Cities and Their Vital Sys-

of each innovation are not available, the SBA pro- tems, ed. J. Ausubel and R. Herman, pp. 85-97.
vided ratings of their economic significance; less Washington, D.C.: National Academy Press.
than 1 percent represented the first type of prod- * 1990a. Silicon Valley Locational Clusters:
uct in a category, 14 percent represented a sig- When Do Increasing Returns Imply Monopoly?
nificant improvement over existing technology, Mathematical Social Sciences 19:235-251.
and 85 percent represented a modest improve- *1 990b. Positive Feedbacks in the Economy.
ment to an existing product. Information on the
Scientific American 2:92-99.
revenue generated by each innovation is not avail-
Coffey, W., and Polese, M. 1987. Trade and Loca-
able.
2. R&D expenditures are from the National Science tion of Producer Services: A Canadian Perspec-

Foundation as reported by Jaffe (1989). Patent tive. Environment and Planning A 19:597-611.
counts by state are also from Jaffe (1989) and rep- Czmanski, S., and Ablas, L. A. de Q. 1979. Iden-
resent the average annual number of patents re- tification of Industrial Clusters and Complexes.
ceived in 29 states over an eight-year period. Urban Studies 16:61-80.
High-technology employment data are from the David, P., and Rosenbloom, J. 1990. Marshallian
U.S. Office of Technology Assessment (1984) for Factor Market Externalities and the Dynamics of
the year 1982.
Industrial Localization. Journal of Urban Econom-
3. Innovation quotients are calculated as the percent-
ics 28:349-370.
age of industry innovations in a state divided by
DeBresson, C., and Amesse, F. 1992. Networks of
the percentage of national innovations in that in-
dustry. The quotient is then multiplied by 100. An Innovators: A Review and Introduction to the
innovation quotient of 100 indicates that the inno- Issues. Research Policy 20:363-380.
vations in a particular industry are equally distrib- Dorfman, N. 1983. Route 128: The Development
uted in the state and national economies. of a Regional High Technology Economy. Re-
4. The result may be attributable to the measurement search Policy 12:299-316.
of university research as federal funding for a sin- Dosi, G. 1988. Sources, Procedures and Microe-
gle year. Federal funds, on average, account for
conomic Effects of Innovation. Journal of Eco-
only 65 percent of total university research expen-
nomic Literature 36:1120-1171.
ditures, albeit with sizable variations between in-
Feldman, M. P. 1994a. The Geography of Innova-
stitutions. Furthermore, a one-year flow of funds
may not capture the stock of university research tion. Boston: Kluwer Academic Press.
that defines innovative capacity in an area. * 1994b. The University and High-Technol-
5. In the estimation of the innovation equation, we ogy Start-ups: The Case of Johns Hopkins Uni-

This content downloaded from

201.150.98.4 on Thu, 22 Jul 2021 16:55:32 UTC
All use subject to https://about.jstor.org/terms
228 Feldman and Florida

versity and Baltimore. The Economic Develop- gions and States, ed. J. Schmandt and R. Wilson,
ment Quarterly 8. pp. 97-119. Boston: Unwin Hyman.
Feldman, M. P., and Schreuder, Y. 1993. Pharma- 1991. Technology and Economic Develop-
ceutical Compound: An Historical Investigation ment. Essex: Longman Scientific and Technical.
of the Geographic Concentration of the Pharma- Mansfield, E. J. 1984. Comment on Using Linked
ceutical Industry in the Mid-Atlantic Region. Patent and R&D Data to Measure Interindustry
Working paper, Hagley Museum and Library, Technology Flows. In R&D, Patents, and Produc-
Wilmington, Delaware. tivity, ed. Z. Griliches, pp. 461-464. Chicago:
Florida, Richard, and Kenney, Martin. 1988. Ven- University of Chicago Press.
ture Capital and Technological Innovation in the . 1991. Academic Research and Industrial
U.S. Research Policy 17:119-137. Innovation. Research Policy 20:1-12.
, and . 1990. The Breakthrough Illu- Markusen, A., Hall, P., and Glasmeier, A. 1986.
sion: Corporate America's Failure to Move from High Tech America: The What, How, Where and
Innovation to Mass Production. New York: Basic Why of Sunrise Industries. Boston: Allen and Un-
Books. win.
Griliches, Z. 1990. Patent Statistics as Economic Markusen, A., and McCurdy, K. 1989. Chicago's
Indicator: A Survey. Journal of Economic Litera- Defense Based High Technology: A Case Study
ture 28:1 661-1 707. of the 'Seedbeds of Innovation' Hypothesis. Eco-
Harvey, D. 1989. The Condition of Postmodernity. nomic Development Quarterly 3:15-31.
Cambridge, Massachusetts: Blackwell Publish- Mokyr, Joel. 1990. The Lever of Riches: Techno-
ers. logical Creativity and Economic Progress. Ox-
Jaffe, A. B. 1989. Real Effects of Academic Re- ford: Oxford University Press.
search. American Economic Review 79:957- Mowery, D., and Rosenberg, N. 1989. Technology
970. and the Pursuit of Economic Growth. New York:
Kline, S. J ., and Rosenberg, N. 1987. An Overview Cambridge University Press.
of Innovation. In The Positive Sum Strategy, ed. National Science Board. 1989. Science and Engi-
R. Landau and N. Rosenberg, pp. 275-305. neering Indicators. Washington, D.C.: U.S. Gov-
Washington, D.C.: National Academy Press. ernment Printing Office.
Krugman, P. 1991a. Geography and Trade. Cam- Nelson, R. 1986. Institutions Supporting Technical
bridge, Massachusetts: MIT Press. Advance in Industry. American Economic Re-
. 1991b. Increasing Returns and Economic view 76:186-189.
Geography. Journal of Political Economy 31:483- 1993. National Innovation Systems: A
499. Comparative Analysis. New York: Oxford Uni-
Landes, D. S. 1969. The Unbound Prometheus. versity Press.
Cambridge: Cambridge University Press. Oakey, R. 1985. High Technology Industries and
MacPherson, A. 1991. Interfirm Information Link- Agglomeration Economies. In Silicon Land-
ages in an Economically Disadvantaged Region: scapes, ed. P. Hall and A. Markusen, pp. 94-117.
An Empirical Perspective from Metropolitan Boston: Allen and Unwin.
Buffalo. Environment and Planning A 23:591- Rosenberg, N. 1972. Technology and American
606. Economic Growth. New York: Harper and Row.
Malecki, E. 1981. Science, Technology, and Re- . 1982. Inside the Black Box: Technology
gional Economic Development: Review and and Economics. New York: Cambridge Univer-
Prospects. Research Policy 10:312-334. sity Press.
. 1983. Federal and Industrial R&D: Loca- Saxenian, A. 1985. Silicon Valley and Route 128:
tional Structures, Economic Effects and Interrela- Regional Prototypes or Historical Exceptions? In
tionships. Final Report to the National Science High Technology, Space and Society, ed. M. Cas-
Foundation, Division of Policy Research and tells, pp. 81-115. Beverly Hills: Sage.
Analysis. Washington, D.C. Sayer, A., and Walker R. 1993. The New Social
. 1985. Industrial Location and Corporate Economy. Oxford: Basil Blackwell.
Organization in High Technology Industries. Eco- Scherer, F. M. 1983. The Propensity to Patent. In-
nomic Geography 75:345-369. ternational Journal of Industrial Organization
* 1986. Research and Development and the 1:1 07-1 28.
Geography of High-Technology Complexes. In Scott, A. J. 1988. New Industrial Spaces. London:
Technology, Regions and Policy, ed. John Rees, Pion.
pp. 51-74. Totowa, New Jersey: Rowman and * 1990. Metropolis; From the Division of La-
Littlefield. bor to Urban Form. Los Angeles: University of
. 1990. Technological Innovation and Paths California Press.
to Regional Economic Growth. In Growth Policy Scott, A. J., and Storper, M. 1990. Work Organi-
in the Age of High Technology: The Role of Re- zation and Local Labor Markets in an Era of Flex-

This content downloaded from

201.150.98.4 on Thu, 22 Jul 2021 16:55:32 UTC
All use subject to https://about.jstor.org/terms
 The Geographic Sources of Innovation 229

ible Production. International Social Science Jour- Technological Change, ed. A. T. Thwaites and
nal 129:573-591. R. P. Oakey, pp. 13-33. New York: St. Martin's
Schumpeter, Joseph. 1954. Capitalism, Socialism Press.
and Democracy. 3rd ed. New York: Harper and Thompson, W. 1962. Locational Differences in In-
Row. ventive Effort and Their Determinants. In The
Sternberg, Ernest. 1991. Sectoral Clusters in Eco- Rate and Direction of Inventive Activity, ed. R.
nomic Development: Lessons from Rochester Nelson, pp. 253-271. Princeton: Princeton Uni-
and Buffalo, New York. Economic Development versity Press.
Quarterly 5:342-356. U.S. Department of Commerce, Bureau of the Cen-
Stohr, W. 1986. Regional Innovation Complexes. sus. 1981. 1977 Census of Manufacturers.
Papers of the Regional Science Association Washington, D.C.: U.S. Government Printing
59:29-44. Office.
Storper, M., and Walker, R. 1989. The Capitalist 1986. 1982 Census of Manufacturers.
Imperative: Territory, Technology and Industrial Washington, D.C.: U.S. Government Printing
Growth. Oxford: Basil Blackwell. Office.
Sweeney, G. P. 1987. Innovation, Entrepreneurs U.S. Office of Technology Assessment. 1984.
and Regional Development. New York: St. Mar- Technology, Innovation and Regional Economic
tin's Press. Development. Washington, D.C.: Government
Tassey, G. 1991. The Functions of Technology In- Printing Office.
frastructure in a Competitive Economy. Research Von Hippel, E. 1988. The Sources of Innovation.
Policy 20:329-343. New York: Oxford University Press.
Thomas, M. 1985. Regional Economic Develop-
ment and the Role of Innovation and Technologi- Submitted 10/92, Revised 5/93, 9/93, Accepted
11/93.
cal Change. In The Regional Economic Impact of

Feldman, Maryann P., and Florida, Richard. 1994. The Geographic Sources of Innovation: Tech-
nological Infrastructure and Product Innovation in the United States. Annals of the Association
of American Geographers 84(2):210-229. Abstract.

The fate of regions and of nations increasingly depends upon ideas and innovations to facilitate
growth. In recent years, geographers have made fundamental contributions to our understanding
of the innovation process by exploring the diffusion of innovation, the location of R&D, and the
geography of high-technology industry. This paper examines the geographic sources of inno-
vation, focusing specifically on the relationship between product innovation and the underlying
"technological infrastructure" of particular places. This infrastructure is comprised of agglomera-
tions of firms in related manufacturing industries, geographic concentrations of industrial R&D,
concentrations of university R&D, and business-service firms. Once in place, these geographic
concentrations of infrastructure enhance the capacity for innovation, as regions come to spe-
cialize in particular technologies and industrial sectors. Geography organizes this infrastructure
by bringing together the crucial resources and inputs for the innovation process in particular
places. Using a direct measure of commercial product innovation, an empirical model of the
geography is presented. The model tests the hypothesis that innovation is concentrated in places
that possess a well-developed technological infrastructure. The analysis confirms this hypothesis;
innovations cluster geographically in areas that contain geographic concentrations of specialized
resources indicative of technological infrastructure. The spatial concentration of these resources,
furthermore, reinforces their capacity to innovate. Key Words: industry R&D, innovation, knowl-
edge-base, regional capacity, technological infrastructure, university R&D.

This content downloaded from

201.150.98.4 on Thu, 22 Jul 2021 16:55:32 UTC
All use subject to https://about.jstor.org/terms