Download as pdf or txt
Download as pdf or txt
You are on page 1of 19

This article was downloaded by: [Universite De Paris 1]

On: 18 November 2012, At: 21:50


Publisher: Routledge
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer
House, 37-41 Mortimer Street, London W1T 3JH, UK

Economics of Innovation and New Technology


Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/gein20

The Geography Of Knowledge Spillovers And


Technological Proximity
a
Corinne Autant-Bernard
a
Université Jean Monnet Creuset, 6 rue Basse des Rives, 42 023Saint-Etienne
Cedex 2
Version of record first published: 28 Jul 2006.

To cite this article: Corinne Autant-Bernard (2001): The Geography Of Knowledge Spillovers And Technological Proximity,
Economics of Innovation and New Technology, 10:4, 237-254

To link to this article: http://dx.doi.org/10.1080/10438590100000010

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions

This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to
anyone is expressly forbidden.

The publisher does not give any warranty express or implied or make any representation that the contents
will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses
should be independently verified with primary sources. The publisher shall not be liable for any loss, actions,
claims, proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or
indirectly in connection with or arising out of the use of this material.
u Tlchn.. 2001. W.10. pp. 237-254
E m I ~ o New 0 2001 OPA (Owrseas Pubiihen Association) N.V.
Rcprinu availabk d i m l y fm the plblishn Published by license under
Phommpying pamined by licmcc only the Harwmd Acadank Publishen Imprint.
pail of 'lhe Gorda,and B w h Publishing Gmup.

THE GEOGRAPHY OF KNOWLEDGE


SPILLOVERS AND TECHNOLOGICAL
PROXIMITY
Downloaded by [Universite De Paris 1] at 21:50 18 November 2012

UniversitkJean Monnet Cretkset, 6 rue Base des Rives,


42 023 Saint-Etienne Cedex 2

(Received October 22, 1999; Infinal form May 04.2000)

This paper tests the presence of technological spillovers for the French case and studies why
they occur. Based on a knowledge production function, spillovers are introduced as an exter-
nal stock of knowledge. Two dimensions are improved: A geographical and a technological
effect. The results indicate that technological externalities m u r . Spillovers are conditional to
technological proximity and, to a lesser extent, to geographical distance. However, externali-
ties are not as generalized as they could be. They do not stem from the whole stock of exter-
nal knowledge They flow only through human capital. People thus appear as an essential
channel for the diffusion of knowledge.

Keywora's: geography of innovation;externalities; knowledge spillovers

First introduced by A. Marshall in 1906, technological externalities are


frequently mentioned today in economic analysis. They are stressed in the
understanding of spatial concentration phenomena (Krugman [1991],
Arthur [1986]). They play a key role in endogenous growth theories as
well. This concept of externality is often associated with the idea of a geo-
graphic limitation of such spillovers. Indeed, in geographic economics, the
local dimension of externalities leads to the spatial concentration of eco-
nomic activity. In a more implicit way, most endogenous growth theories
* Corresponding Author: e-mail: autant@univ-stetienne.fr
238 CORINNE AUTANT-BERNARD

assume that externalities are geographically bounded, at least within


national boundaries.
Consequently, empirical studies of spillover phenomena have developed
over the past ten years.* There have been many attempts to test the local
dimension of externalities generated by innovative activity. Four
approaches can be distinguished: the use of patent citations as a paper trail
of spillovers, the study of innovative activity concentration, the measure of
geographic coincidence and the observation of local interactions.+ The
Downloaded by [Universite De Paris 1] at 21:50 18 November 2012

majority of those studies deal with the American case. They conclude that
there is a significant localization of spillovers. However, this result may be
strongly linked to the American institutional system. Besides, it is difficult
for econometric studies to model externalities on the one hand and their
geographic dimension on the other. Studies that model externalitiescannot
give a clear indication of the spatial dimension. Conversely, the measure
of a geographic dimension is carried out at the expense of a precise meas-
ure of spillovers.
The production function of innovation used here tries to overcome these
difficulties. I suggest a new model that accounts for technological exter-
nalities and their geographic dimension at the same time. Besides mode-
ling externalities, the study puts forward a method to test the impact of
spatial dimension, by confronting distinct geographic levels.
A detailed account of the model is given in section H. The results confirm
the presence of technological externalities in France and the fact that they
are geographically bounded. These results are analyzed in the second sec-
tion. However, the model has to deal with a difficulty that is common to all
models of geography of innovation which is the uneven spatial distribution
of economic activities. This is not neutral and needs to be taken into
account if we do not wish to award geographic proximity with an effect
that in fact arises from the sectoral polarization of economic activity.
Therefore, an indicator of technological proximity is used (section 3). This
leads to a more precise measurement of the geographic dimension of spill-
overs. The conclusion (section 4) summarizes h e main results and gives
some directions for further research.
* Cf. especially A. JafTe [1989]. Z. Acs, D. Audretsch, M.Feldman [1991]. A. Jaffe, M.
Trajtenberg, R. Henderson (19931, M. Feldman [1994], D. Audremh and M. Feldman [1996a
and 19%b], L. Anselin, A. Varga and Z. Acs [1997].
t Cf. C. Autant-Bernard and N. Massard [I9991 for a detailed survey of the econometric
literature on geographical spillovers.
GEOGRAPHY 239

1. THE MODEL
The aim is to test the existence of a localization of spillovers. Two difficul-
ties need to be overcome: the measuring of externalities, on the one hand,
and the integration of the geographic dimension, on the other. Now, as has
already been mentioned,* externalities (and especially their geographic
dimension) are, by definition, an intangible phenomenon. There is no real
indicator of externalities. Their measure can therefore only be indirect.?
Downloaded by [Universite De Paris 1] at 21:50 18 November 2012

The method usually employed consists in inserting an external stock of


knowledge into the production (or innovation) function.* Externalities are
considered as the impact of research activities carried out by anyone
research unit on the output of other research units.

1.1. Modeling technological spillovers


In such a view, spillovers are introduced here in a production function of
innovation as an external stock of knowledge. Consequently, I test the
presence of technological spillovers by looking at the relation between the
innovative output of a group of firms and the research camed out by other
firms. But there are several ways of measuring this stock of knowledge.
First of all, we may assume that it is the level of knowledge that prevails.
There may be more innovation when R&D expenditure is high and R&D
staff numerous. To account for this relation, an indicator of R&D level is
used. It is measured by R&D expenditure. However, some studies empha-
size the human capital factorq Consequently, the percentage of the popula-
tion devoted to research becomes the determining variable of innovative
output, and no longer the level of R&D. Indeed, the level of innovation is
probably linked to the proportion of researchers.
This observation of both research intensity and R&D levels does not
seem antagonistic. We could be led .to believe that a qualitative effect
P. Krugman [1991].
t Even patent citations used by A. Jaffe. M. Trajtenberg and R. Henderson [I9931 and then
by P. Almeida and B. Kogut [1997a and 1997bI are very indirect indicators o f spillovers.
They did not account for all spillovers and conversely, they cover a wider field ihan just spill-
overs.
$ Cf. for instance Z. Griliches [1991]. C. Antonelli (19941, and L. Anselin. A. Varga and Z.
Acs, (19971 for econometric studies. See also B. Verspagen (19921 for an overview of techno-
logical externalities modeling i n endogenous growth models and J. Mairesse and P.Mohnen
(19941 and Cheynet and Fadairo (19981 for an overview o f technological externalities rnode-
ling.
1 In endogenous growth models (R. Luca. [1998]) as in geography o f innovation (L.
Zucker, M . Darby and J. Armstrong (19941).
240 CORINNE AUTANT-BERNARD

occurs in addition to a quantitative one. It seems therefore relevant to test


the impact of both dimensions on innovation capacity. In the first case,
externalities result from a stock of knowledge. Bn the second case, they
result from the structure of human capital. The human capital indicator
used here is the ratio between the number of researchers and the total
research staff. So, the human capital variable (noted KH) represents the
proportion of researchers relatively to the total research staff.
For innovative output, the indicator is the patent. Patents are quite a
Downloaded by [Universite De Paris 1] at 21:50 18 November 2012

good indicator of innovation because of their features. Firstly, there is a


strong link between patents and inventions." Secondly, patents are regis-
tered administratively. This generates exhaustive data and reduces the risk
of statistical error, in so far as they are clearly defined. However, they
involve limits that must not be underestimated. All iainovations are not pat-
entedt Conversely, every patent does not lead to an innovation. Only 50%
of patents represent a real innovation (cf. D. Guellec and B. van Pot-
telsberghe [1999]).We must therefore bear in mind the gap between pat-
ents and innovations when we interpret the results.

1.2. Modeling the spatial dimension of spillovers


The second point of the study relies on the local dimension of technologi-
cal externalities. We may wonder whether geographic proximity promotes
externalities. Dealing with this question requires a particular form of mod-
eling. A local stock of knowledge can be taken into account. It is the
method used by several American and Italian studies.$ But such a bound-
ing of the geographic area in which spillovers can occur does not seem
fully satisfactory. At the most, it allows us to determine if spillovers are
geographically bounded. But it does not prove that their diffusion is con-
strained by distance. To demonstrate the impact of geographic distance,
we must be able to affirm that an agent is more affected by his neighbors'
activity than by the activity of agents that are physically distant. This is
During the last two centuries. there has been no major invention that has not been pat-
ented (D. Guellec and B. van Pottelsberghe [1999]). Besides, we can assume that patents are
a good indicator of innovation since they have a cost. The cost is accepted in so far as the
application is expected to produce returns that exceed the cost.
t There are two reasons for this. On the one hand. innovators may prefer to protect their
innovations by keeping them secret. Indeed, patenting requires divulging information, which
is not always desirable. On the other hand, some inventions cannot be patented. The SESSl
investigationon innovation concludes that only 30% of innovationslead to a patent grant.
$ C. Antonelli [1994]. L. Anselin. A. Varga and Z. Acs [1997]. but also L. Zucker. M.
Darby and J. Armstrong [1994].
GEOGRAPHY

why it would seem better to do a comparative analysis of different geo-


graphic levels. The localization of spillovers could then be tested by com-
paring the impact of the close neighborhood with the impact of a more
distant neighborhood.
In this respect, the geographic level used is the French administrative
"departement". Spillovers are studied in testing the eventual impact of the
research of bordering "departements" on the innovative output of a given
"departement". But this geographic unit is not necessarily the most rele-
Downloaded by [Universite De Paris 1] at 21:50 18 November 2012

vant to account for local externalities. If there are local spillovers, some of
them probably occur on a smaller scale than the departement. Neverthe-
less, the departement is an acceptable geographic level. It is the smallest
administrative unit for which data are available. It is also a rather consist-
ent level, in so far as departements generally include one large urban
center and its agglomeration. So, this scale offers a relative homogeneity.

of the model
1.3. ~~ecifu:ation
The general equation is as follows:
(1) Wig) = a +Pi ln(RDg) + P-2 ln(RDv(g))+ P3 ln(RDvl(g))'
+ P4 ln(KHg) + P 5 In(KHv(g,)+ P6 ln(KHvr(g)) + P
g is the geographic area considered, v is the close neighborhood of this
area and v' is a more distant neighborhood. I is an indicator of innovation
output, RD measures the stock of knowledge and KH the human capital.*
a is a constant and p is a random disturbance.+
More precisely, I use the following data: I is the number of patents, RD
is measured in R&D expenditure and KH is the number of researchers
divided by the total research staff. The geographic area g is the French
administrative "departement", v represents all the bordering "departe-
ments" of g and v', the bordering "departements" of v. In this way, I may
observe the relation between the production of innovation for each "depar-
tement" and the research effort carried out in its periphery, defining con-
centric areas around each "departement". Innovative activities produce
-- - - --- - - -
* Data come from the R&D inquiry of the French Ministry of National Education.
Research and Technology and from the Observatoire des Sciences et Techniques (O.S.T.).
t The model is expressed in logarithms. In addition to the interest it implies for the inter-
pretation of the results, this functional form appears to be the more suitable. The Bera and
McAleer test suggests a preference for the semi-log form to the linear one. Then. comparing
explanatory power of log-log form and semi-log form shows a superiority of the log-log one.
Consequently, it is relevant to specify the model in logarithm%.
242 CORINNE AUTANT-BERNARD

local spillovers if the innovation output (Ig) is influenced more by the


research canied out in a close neighborhood than the research carried out
in a distant neighborhood.
The data lased concern all the French departements, except the
DOM-TOMand Corsica, for which the geographic dimension cannot be
studied by the method used here. As patents for the year 1996 are in fact
an average over three years (1994-1995 and 1996), the explanatory varia-
bles of the model are the R&D expenditure and h e human capital for
Downloaded by [Universite De Paris 1] at 21:50 18 November 2012

1993. The level and the intensity of the research carried out in 1993 would
explain the innovation output of 1994,1995 and 1996. It means we assume
there is a lag between the moment research is actually carried out and the
moment when it materializes as an invention. It takes time for investment
in R&D (whether physical investment or investment in human capital) to
materialize, for example as a patent. Such a lag has not been studied in
depth here. Data does not cover a long enough time Therefore, to
generalize this result, it should be inferred that, for other years of patent-
ing, the same lag occurs.

2. TECHNOLOGICAL SPILLOVERS AND GEOGRAPHIC


PROXIMITY
Results tend to confirm the hypothesis that there is a localization of tech-
nological spillovers, even if this is not a major effect. The variables RD
and KH give interesting information at infra-"departemental" level as to
spillovers coming from the surrounding neighborhoods. Before analyzing
each variable, some general results are given.

2.1. General results


Since the dependent variable is normally distributed (cf. appendix), we can
use an ordinary least square instead of a tobit model which is frequently
used when innovation is the dependent variable.
But there may be heteroscedasticity. Actually, departements are very dif-
ferent as regards their economic weight. This could affect the variance of
the residuals. For this reason, weighted least squares are used. The weight-
ing is made using the share of each departement in the French Gross
* I have only estimated function ( I ) with the patents for 1996 and several years for R&D
and human capital. The highest R* is for the research canied out in 1993.
GEOGRAPHY 243

National ~roduct.*It accounts for a size effect. Then, the results of the
White test confirm the homoscedastic characteristic of the model. It seems
therefore that the eventual heteroscedasticity has been efficiently elimi-
nated, thanks to the weighting. Consequently, estimations wers able to be
made without White correction.
The Pregibon specification-test was also made. It appears that the
squared predicted value is not significant. There is no problem for the
specification of the model.+ Moreover, it indicates that there is no endog-
Downloaded by [Universite De Paris 1] at 21:50 18 November 2012

eneity.
Globally, the model is well explained. The R ~ like
, the adjusted-R~,is
high and the F-test is significant. Independent variables give results that
are consistent with expectations.

TABLE I
Observations = 94 R2: 0.896 ~djusted-R2: Standard Error: Weighting variable:

Dep. Variable = lg 0.889 4.9308-02 PartPIB

Indep. Variables fl Standard Emr T Signif.

(constant) 0.725 0.778 0.932 0.254


RDs 0.638 0.025 25,775 0,000

2.2. Local variables


Not surprisingly, we notice the strong impact of R&D on the number of
patents, at an infra-"departemental" scale. Indeed, not only RDg has a sig-
nificant effect, but also the parameter pl is relatively high. An elasticity of
* The weighting variable, noted PartPlB is measured by dividing the product of each
"departement"by the Gross National Product.
t This result confirms the Ben and McAlcer test (cf. Note p. 5).
244 CORINNE AUTANT-BERNARD

0.638 is observed. This link between R&D and patents is consistent with
expectations and with the results of other empirical studies.*
The human capital effect on innovation is all the more noticeable. Patent
elasticity regarding the human capital level is close to one. The more
researchers there are relative to the total number of research staff, the
higher the number of patents. So, as we may believe intuitively, the level
of R&D alone is not enough in order to understand the innovative capabil-
ity of a geographic area. The way this effort is distributed is at least as
Downloaded by [Universite De Paris 1] at 21:50 18 November 2012

decisive as its level. If the number of researchers is low, R&D expenditure


is not really productive. An area with a high level of research expenditure
and a small number of researchers may have a lower innovation level than
an area with less R&D expenditure but with a higher percentage of
researchers.

2.3. The geographic dimension of technological spillovers


As for externalities, several remarks can be made. The research level done
in a closed neighborhood (RDv(g)) does not significantly affect the level of
innovation. The R&D of bordering departements does not seem to play a
significant part and the use of a "decreasing" method leads to the elimina-
tion of this variable.
This result tends to show the weakness of the hypothesis of a spillovers
localization. However, such a result is not obtained for KHv(g). On the
contrary, the latter is significant at a 5% threshold. It influences the
number of patents positively. It means that spillover effects occur. There-
fore, the research level of the neighbors matters less than the intensity of
their research effort.
Human capital seems to play a great part and workers appear as an
essential vector of local innovation. Looking at variables KHg and KHv(g),
it seems that human capital is a source of externalities not only because
knowledge flows thanks to people (KHv(g)) but also because local skills
can be used to capture externalities and to make them as available knowl-
edge for the departement. In this way, the strong significant effect of inter-
nal human capital (KHg) can be considered as a complement to benefit
from technological externalities.
* For instance, B. Hall. Z. Griliches and J. Hausman [I9861 observed a low difference
between the level of R&D and the number of patents registered. D. Guellec and B. van Pot-
telsberghe [I9991 obtained the same result looking at growth rates.
GEOGRAPHY 245

The innovative level of a geographic area is linked to the research car-


ried out in its periphery. So, there are technological externalities. On the
other hand, the research carried out further away (RD,e(g), KHvyg))does
not produce a significant effect on Ig. Therefore, the idea that externalities
spread only inside geographic boundaries is validated.
Several reasons could explain such a phenomenon. Firstly, the fact of
being in a same place allows more frequent meetings and may lead to
interactions that would not have occured with geographic remoteness. This
Downloaded by [Universite De Paris 1] at 21:50 18 November 2012

interaction may take various forms: R&D cooperation agreements, busi-


ness relations, informal exchanges, etc.
Knowledge tacitness is also a reason of the local dimension of spillo-
vers. Tacit knowledge would require geographic proximity whereas codi-
fied knowledge would spread even from a distance.* In this respect,
externalities of tacit knowledge should be localized because they are
incorporated into people. Studies of P. Almeida and B. Kogut [1997a and
1997b1, C. Antonelli [1994], E. Mansfield [I9951 and, above all, the work
of L. Zucker, M. Darby et J. Armstrong [1994] show that externalities
occur thanks to worker mobility or inter-personal relations. Now, labor is
strongly localized. This phenomenon, observed for the American case, is
probably all the more true in France where geographic mobility is low.
This analysis is supported by the positive influence of KH,(g). The number
of patents increases with the proportion of researchers in the closed neigh-
borhood. So, a share of externalities would be localized through tacit
knowledge.
Finally, this first regression emphasizes two important results. In France,
technological externalities occur and they have a geographic dimension.
But those spillovers are not very strong. It is not local research in general
that matters, but only human capital.

3. EXTERNALITIES AND THE POLARIZATION


OF TECHNOLOGICAL ACTIVITIES
Economic activities are unevenly distributed over the national territory.
This polarization is at once quantitative and qualitative. In fact, not only
do some departements have higher activity, but there is also a geographic
specialization. As a consequence, bordering departements could be spe-
* Cf. especially C. Antonelli (19991 and A. Rallet [1993].
246 CORINNE AUTANT-BERNARD

cialized in the same kind of activities. Then, the results obtained on the
basis of equation (1) can be affected by an uneven technological dishibu-
tion. A departement can benefit from its neighbor's research activity not
because they are geographically close, but because their technological
fields are similar. In this case, B5 may be significant not because the
research is done in the periphery of g, but because v and g have a strong
technological proximity. So, it is important to control that the spatial effect
is not only the result of a technological polarization of economic activities.
Downloaded by [Universite De Paris 1] at 21:50 18 November 2012

We have to distinguish spatial and technological effect.

n ~ technological proximity
3.1. ~ c c o u n t ifor
In this perspective, I insert a new variable (P) into function (1). It measures
the technological closeness between each departement and its neighbors.
This variable is constructed like Jaffe's indicator of SMSA technological
proximity (A. Jaffe [1986]). Vectors of technological position are defined
for each departement. The position of a departement is measured looking
at the number of patents registered in each innovative field. Then, the indi-
cator of technological proximity (P) measures the "likeness" between the
technological position of a departement and the technological position of
its neighborsg . If vectors of technological position are the same, then,
Pp=l. Conversely, the more P is close to zero, the more departement g
gv.
is technologically different from ~ t neighbors.
s
Several interesting results proceed from gross data. Firstly, there are
considerable disparities between departements. Some are technologically
very close to their neighbors whereas some have a specific innovative
activity. Generally, departements with less affinity with their neighbors
have a low level of innovation. This is not very surprising since the vectors
of technological position of such low innovative departements can be radi-
cally altered by the registration of a single patent.
Conversely, departements that are technologically close to their whole
neighborhood are the more innovative ones. However, this may depend on
a specific effect for the Paris region rather than on a characteristic of every
high innovative departements. Actually, the five departements with the
* The position of the departement is compared with the technological position of its neigh-
borhood. If the vector of technological position of departement g is noted Fg and the vector of
.
its neighbors is noted Fv( then the indicator of technological proximity Pgv is:
Pgv= F ~ F ~ ~ ( [ F ~ F ~ F ~ ~ Only
, F ~technological
~ ~ $ ' ~ . proximity with bordehg departements
is measure slnce t ere s no externalitiesstemming from farther areas.
GEOGRAPHY 247

highest technological proximity coefficients are in the "Ile de France"


region. This supports the standard established fact that there is a polariza-
tion of innovative activities in this area. There is both a concentration of
research activities*and a specialization of these activities in the same tech-
nological fields. This phenomenon should be considered as an effect that is
unique to the "Ile de France" region, in so far as it is not observed for all
other departements with high level patenting. For instance, departements
like RhBne or Gironde, in spite of a high level of innovation, are not signif-
Downloaded by [Universite De Paris 1] at 21:50 18 November 2012

icantly technologically close to their neighbors.


By and large, the level of innovation of a geographic area does not seem
independent from the degree of technological proximity with neighbors.
Then, KHv(g) might cover both a spatial and a sectoral dimension, since,
for some departements, the neighbors in geographic space are also the
neighbors in technological space.

3.2. Technologicalproximity and geographic proximity


Some externalities may not be linked to a location effect, but to the polari-
zation of economic activities. So, the percentage of externalities flowing
through technological proximity needs to be distinguished from externali-
ties spreading through geographic proximity. Thissis the aim of the intro-
duction of variable P into the model. By introducing this new variable, the
technological proximity effect can be neutralized and a unique spatial
effect is highlighted.
The model becomes:

The introduction of P produces some noticeable results (Cf. table 11).


First, it results in a rise of adjusted-R~.It means that technological proxim-
ity contributes to explain the variability in patents. Student-tests confirm
this result since they lead to reject the null hypothesis for PI7. The positive
sign of r7 shows that the innovative capability of a departement is posi-
tively influenced by the technological proximity between a departement
* About 50% of the French innovative activity are concentrated in the region "Ile de
France" (49% of researchers. 46% of total re-search staff, 48% o f R&D expenditure and 41%
o f patents in 1996). This concentration is significantly higher than for other economic activi-
ties since about 30%of the French National Product are done in "lle de France". So, there's a
strong specialization of Paris region in innovative activities.
248 CORINNE AUTANT-BERNARD

and its neighbors. So, a positive impact would result from an agglomera-
tion of closed activities.

TABLE I1

Obsetvations= 94 R? ~djusted-R~ Standard Emr Weighting variable:


Dep. variable = Bg 0,912 0.905 4551E-02 PartPIB
Indep. variables StandardEmr Signif.
Downloaded by [Universite De Paris 1] at 21:50 18 November 2012

(constant) 0216 0.729 0296 0.768

RD8 0,560 0.030 18,628 0.000

Working in a same technological field, cooperation and information


exchanges are probably more numerous. It is obviously easier for instance
for an area to benefit from its neighborhood research in the pharmaceutical
field if the area includes a large pharmaceitical sector. If this is not the
case, results from external research cannot be used directly and require
modifications to fit other fields.
However, literature on this question is debated considerably. According
to A. Jaffe [1986], A. Jaffe, M. Trajtenberg and R. Henderson [1993], D.
Audretsch and M. Feldman [I995 and 19991, the largest share of spillovers
comes from firms that belong to different technological areas. On the other
hand, for A. Jaffe [1989], academic research has a higher impact within a
same field. Spillovers are linked to one field and research does not have a
general spread effect.
The results obtained here for the French case give a general outline of
the impact of technological proximity. But it is not enough to say if it is
more efficient to have an agglomeration of close activities or, on the other
hand, if diversity produces positive effects. This prospect would require an
analysis with data by field.
GEOGRAPHY 249

Secondly, the introduction of P modifies the model. We have to study to


what extent other variables are affected. Internal variables (RDg and KHg)
remain significant and parameter values are not strongly altered. Variables
relative to geographically far areas (RD,P(~) and KHvt(g))also r d n the
same. They are still non-significant. RDv(g)is still without influence on
innovation output too.
Nevertheless, introducing technological proximity in the model is not
neutral. It modifies the parameter of KHv(g).When technological proxim-
Downloaded by [Universite De Paris 1] at 21:50 18 November 2012

ity is taken into account, it reduces the impact of geographic externalities


on innovation. This result supports the hypothesis of an overlap between
geographic and technological externalities. KHv(g) covered these two
effects first. With the technological proximity indicator, geographic and
technological effects are dissociated. Introducing -P, the technological
effect is neutralized and KHv(g)represents a pure geographic effect.
So, results summarized in table I1 shade the hypothesis of local spillo-
vers. Then, if we do not account for the sectoral effect P, we overevaluate
the geographic effect.
Nevertheless, KHv(g) remains significant at a 5% threshold, which
means that local externalities keep an influence at the innovation level.
This result can express two phenomena that are not necessarily inconsist-
ent with each other. Firstly, it may indicate that there are two distinct
effects leading to externalities: technological proximity on the one hand
and geographic proximity on the other. Secondly, we may think that tech-
nological and geographic proximity are not alternatives. They may com-
plement one another. Indeed, these two dimensions are closely linked. The
polarization of economic activity may originate, for a part, from spillovers.
So the technological proximity between geographically close areas may
result from local spillovers. More basically, geographic vicinity may be
required to catch externalities stemming from a technological proximity.

4. CONCLUSION
At the end of this study, five main results stand out. First, the innovative
output of a geographic area is not only linked to internal factors of produc-
tion. It is also linked to external research activity. This supports the idea of
technological externalities.
Secondly, these spillovers do not occur in the same way from one area to
one another. They are geographically bounded. Actually, an area benefits
250 CORINNE AUTANT-BERNARD

only from the research activities of its close neighbors, and not from the
research of more distant areas.
Thirdly, these spillovers and their geographic dimension are not very
strong. There is a significant impact from the percentage of researchers but
not from the level of research expenditure. Therefore, the level of R&D
does not seem to produce externalities. Only human capital matters.
. Fourthly, the innovative output of an area is linked positively to the tech-
nological proximity between this area and its neighbors. The fact that an
Downloaded by [Universite De Paris 1] at 21:50 18 November 2012

area has the same technological profile as its neighbors increases the inno-
vative capability.
Finally, the introduction of technological proximity reduces local spillo-
vers. The human capital of the neighborhood has a smaller effect on inno-
vation when technological proximity is neutralized.
This last result opens three ways for research. First, technological prox-
imity and geographic proximity may not be alternative channels of knowl-
edge diffusion. On the contrary, they may overlap. Testing such a
complementarity requires further processing. These two dimensions may
be accounted for as a join effect. Externalities between technologically
close but geographically distant "departements" may also be studied.
Furthermore, the positive effect of technological proximity on innova-
tive output indicates that the cluster of similar activities benefits innova-
tion. However, the measure used here does not really account for
specialization. It measures the likeness of technological profile. Two areas
may then be similar if they specialize in the same technological field, but
also if they are both unspecialized. In this case, it is diversity that has a
positive effect. It should then be useful to study this result thoroughly by
means of a sectoral analysis. It would answer the question of infra and
inter-sectoral externalities.
Finally, these results on geographic dimension of spillovers do not fit
exactly with the idea that knowledge is a public good. The human capital
significance indicates that the number of researchers matters more than
R&D expenditure. This underlines the part played by human capital. Peo-
ple seem to be a channel of geographic spillovers. It is therefore necessary
to pursue the analysis in this direction, in order to understand the potential
sources of externalities and the channels of knowledge transfers.
GEOGRAPHY 25 1

References
ACS, Zoltan J.. AUDRFTSCH. David B. and FELDMAN, Maryann P. (199i). "Real effects
of academic research: comment." The American E c o M ~ ~Review, C vol. 82. no. I,
March, p. 363-367.
ALMEIDA, Paul and KOGUT, Bmce (1997a). "The localization of ideas and the mobility of
engineers in regional networks," Working Paper, June, p. 45.
ALMEIDA. Paul and KOGUT, Bmce (f997b). ''The exploration of technological diversity
and the geographic localization of innovation," S d l Business Economics, no. 9, p. 21-
31.
ANSELIN. Luc, VARGA, Attila and ACS. Zoltan (1997), "Local geographic spillovers
Downloaded by [Universite De Paris 1] at 21:50 18 November 2012

between university research and high technology innovations." Journal of Urban Eco-
nomics. no. 42. p. 422448.
ANTONELLI. Cristiano (1994), "Technological districts localized spillovers and produaiv-
ity growth. The Italian evidence on technological externalities in the core regions."
International Review of Applied Economics, p. 18-30.
ANTONELLI, Cristiano (1999). "Communication and innovation: The evidence within tech-
nological districts." Working Paper. May.
ARGHIBUGI. Daniele (1992). "Patenting as an indicator of technological innovation: a
review." Science and Public Policy, vol. 19, no.6. December, p. 357-368.
ARTHUR. W. Brian (1994). Increasing rehtnu and path dependence in the economy. The
University of Michigan Press,coll. Economics, Cognition, and Society, p. 201.
AUDRETSCH. David B. and FELDMAN. Maryann P. (1996 a). "Innovative clusters and the
industry life cycle." Review i~lndusrrialOrganization, no. 11, p. 253-273.
AUDRETSCH. David B. and FELDMAN. Maryann P. (1996 b). "R&D spillovers and the
geography of innovation and production." The Atnericun Economic Review. vol86, no.
3, June. p. 630640.
AUDRETSCH, David B. and VIVARELLI. Marco (1994). "Small firms and R&D spillovers:
Evidence from Italy," Revue d'Econotnie Industrielle. no. 67, p. 225-237.
AUTANT-BERNARD. Corinne and MASSARD. Nadine (1999). "Econodtrie des extemal-
it& technologiques locales et gkographie de I'innovation: une analyse critique." Pro-
gramme CNRS: les enjeux dconomiques de I'innovation. k s Cahiers de I'lnnovation,
no. 99025.
CARRINCAZEAUX, Christophe, LUNG, Yannick and RALLET, Alain (1997). "De la local-
isation h I'organisation spatiale des activitks de recherched6veloppement des entre-
prises. Hypoth&ses thkoriques et dsultats empiriques dans le cas de la France,"
CotntnunicationprJsent/c au XLV?"'~ congrhs de I'AFSE. September. Paris p. 16.
CHEYNET. Pascal and FADAIRO, Muriel (1 998). "Les m6thodes de mesure des externalit&
technologiques. Un apequ des travaux kcono~triques."Working Paper for Programme
CNRS Les enjeux economiques de I'innovation.
DORMONT, Brigitte (1999). Intmducrii~nd l'iconomdtrie. Montchrestien. Paris, p. 414.
FELDMAN. Maryann P. (1994). 77w geography of innovariun, Economics of Science, Tech-
nology and Innovation, vol 2, Kluwer Academic Publishers. Dordrecht, Boston. Lon-
don, p. 155.
FELDMAN. Maryann P. (1998). "The new economics of innovation, spillovers and agglom-
eration: a review of empirical studies." Working Paper. March, p. 34.
GLAESER. Edward L.. KALLAL, Hedi D., SCHEINKMAN, J0.d A. and SHLEIFER.
Andrei (1992). "Growth in cities." Juurnal of Political Economy, vol. 100, no. 6. p.
1126-1152.
GRILICHES, Zvi (1979). "Issues in assessing the contribution of research and development
to productivity growth." 77w Bell Journal of Ecmotnics, vol. 10, no. I. p. 92-1 16.
GRILICHES. Zvi (199 I), 'VKsearch for R&D spillovers." NBER Working Papers Series.
Working Paper no. 3768, July, NBER, Cambridge, p. 38.
252 CORINNE AUTANT-BERNARD

GUELLEC, Dominique and van POTTELSBERGHE. Bruno (1999), "Les brevets comm
indicateurs de I'innovation," Programme CNRS: Les enjeux konomiques de I'imova-
tion. Les Cahiers de llnnovation. no. 99024.
HALL, Bronwyn H., GRILICHES, Zvi and HAUSMAN. Jeny A. (1986). "Patents and
R&D: Is there a lag?," InternationalE c o M ~ ~Review,
C vol. 27.no.2, p. 265-283.
JAFFE. Adam B. (1986). Technological opportunity and spillovers of R&D: evidence from
firm's patents, profits and market value," The American Economic Review, vol. 76, no.
5. December, p. 98k1001.
JAFFE. Adam B. (1989). "Real effects of academic research." The American Economic
Review, vol. 79, no. 5, December, p. 957-970.
JAFFE. Adam B., TRAJTENBERG, Manuel and HENDERSON. Rebecca (1993). "Geo-
Downloaded by [Universite De Paris 1] at 21:50 18 November 2012

graphic localization of knowledge spillovers as evidenced by patent citations," The


Quarterly Journal of Economics. August, p. 577-598.
KRUGMAN, Paul (1991a). Geography and trade. Leuven University Press. Leuven. Bel-
gium. The MIT Press, Cambridge, Massachusetts. London, England, 4' edition, p. 136.
KRUGMAN. Paul (1991b). "Increasing retwns and economic geography," Journal of Politi-
cat Economy, vol. 99, no. 3, p. 483-499.
LUCAS, Robert E. (1988). ''On the mechanics of economic development." Journal of Mone-
tary Economics. 110.22. p.3-42.
MADDALA, G.S. (1992). Introduction to econometrics. b n t i c e Hall, Englewood Cliffs,
2*edition (1'' edition 1988). p. 631.
MAIRESSE, Jacqws and MOHNEN, Pierre (1990). "Recherchedeveloppement et produc-
tivite. Un survol de la litt6rature konodtrique," EconoPnie et Statistique. I.N.S.E.E.,
no. 237-238. p. 99-108.
MANSFIELD, Edwin (1995). "Academic research underlying industrial innovations:
sources. characteristics, and financing," The Review of Economic and Statistics, vol.
LXXVII, no. 1, February, p. 55-65.
MANSFIELD. Edwin (1998). "Academic research and industrial innovation: an update of
empirical findings," Research Policy, vol. 26, no. 7-8. April, p. 773-776.
MAUREL. Fran~oiseand SEDILLOT. Beatrice (1997). "La concentration ghgraphique des
industries franpises." Economie et Prbision, no. 131. October-December.
MARSHALL. Alfred (1906). Principes d'4conomiepolitique, Gordon & Breach. Paris, Lon-
don, New Yo&, reprint 1971, vol. 2, p. 576.
M.E.N.E.S.R.T. (1987-96). EnquPte R&D, Ministere de I'Education Nationale de I'Enseigne-
ment Su@rieur, de la Recherche et de la Technologie.
MOHNEN. P. (1991). 'Survol de la linkrature sur les extemalith technologiques," in DE
BANDT, J. and FORAY, D. (eds.), L'ivaluatinn iconomique de la recherche et du
chnngement technique, Editions du CNRS. Paris, First part, chap. 1.
O.S.T. (1994). (19%). (1998). Sciences er Technologie. Indicateurs, Rapport de I0.S.T.. Eco-
nomica.
O.S.T. (1997), Les chlffres clef de la science et de la technologic, Economics.
RALLET, Alain (1993). "Choix de proximite et processus d'innovation technologique,"
Revue d'Economie Rigionale et Urbaine, no. 3 (no. sp4cial). p. 365-386.
ROMER. Paul (1991), "Progrk technique endoghne," Annales d'Econotnieet de Statistiques,
no. 22, p. 32.
van LEUVEN. J.W.M. (1996). "Patents statistics as indicators for innovation," Ecnnomitrie
de I'innovation: Le brevet. Association d'Econodtrie AppliquCe. L I X ' ~Conference,
Luxembourg, November 28-29. p. 277-292.
VERSPAGEN, Bart (1992). "Endogenous innovation in neo-classical growth models: a sur-
vey:' Journal ofMacroeconomics.vol. 14, no.4, p. 6 3 \ 4 2 .
ZUCKER. Lynne G.. DARBY, Michael R. and ARMSTRONG, Jeff (1994). "intellectual
capital and the fm:The technology of geographically localized knowledge spillovers,"
NBER Working Paper Series, Working Paper no. 4946, NBER, Cambridge, p. 59.
GEOGRAPHY

APPENDIX: STATISTICS OF PATENTS

Depanetnents Patents Depanements Patents

LOZERE SOMME
CREUSE SAVOlE
CANTAL CHER
Downloaded by [Universite De Paris 1] at 21:50 18 November 2012

GERS PAS-DE-CALAIS
ARIEGE VIENNE
DORDOGNE MARNE
LOT DROME
MEUSE PY RENEES-ATLANTIQUES
' HAUTES-ALPES INDRE-ET-LOIRE
AVEYRON EURE-ET-LOIR
LOT-ET-GARONNE MEURTIIE-ET-MOSELLE
HAUTE-LOIRE VAR
TARN-ET-GARONNE MORBIHAN
LANDES CALVADOS
PYRENEES-ORIENTALES AIN
ALPES-DE-HAUTE-PROVENCE COTES-D'ARMOR
AUDE LOlRE
AUBE SAONE-ET-LOIRE
CORREZE DOUBS
HAUTE-SAONE HERAULT
MAYENNE SEINE-MARITIME
HAUTE-MARNE LOIRE-ATLANTIQUE
ALLIER PUY-DE-DOME
HAUTES-PYRENEES ILLE-ET-VILAINE
lNDRE COTE-D'OR
YONNE EURE
AlSNE LOIRET
254 CORINM AUTANT-BERNARD

Departements Patents Departernents Patents

DEUX-SEVW 12,617 HAUT-RHIN 61,007


NWRE 12.669 GIRONDB
TERRITOIRE DE BELFORT 12,964 MOSELLE
VOSGES 15.154 OlSE 87,453
JURA 15,348 BOUCHES-DU-RHONE 97.01
Downloaded by [Universite De Paris 1] at 21:50 18 November 2012

VENDEE 15.356 NORD 102.216

ARDECHE 16,262 SEINE-ET-MARNE 113,18


FINISTERE 16568 HAUTE-GARONNE 1 13,244
MANCHE 16.608 ALPES-MARITIMES 118.722
CHARENTE-MARITIME 16,637 HAUTE-SAVOIE 119,262
VAUCLUSE 16,948 BAS-RHIN 126,503
ORNE 17,939 VAL-D'OISE 147.9 14
LOIR-ET-CHER 19.055 VAL-DPMARNE 182.856
ARDENNES 19.461 ESSONNE 215.998
SARTHE 21,385 RHONH 264.74 1
TARN 21.674 ISERE 296.668
GARD 21,729 YVELINES 3 18,342
CHARENTE 23.171 HAUTS-DE-SEINE 553.661
MAINE-ET-LOIRE 25.069 PARIS 580.977

You might also like