DP
RIETI Discussion Paper Series 15-E-108
Fresh Brain Power and Quality of Innovation in Cities:
Evidence from the Japanese patent database
HAMAGUCHI Nobuaki
RIETI
KONDO Keisuke
RIETI
The Research Institute of Economy, Trade and Industry
http://www.rieti.go.jp/en/
RIETI Discussion Paper Series 15-E-108
September 2015
Fresh Brain Power and Quality of Innovation in Cities:
Evidence from the Japanese patent database*
HAMAGUCHI Nobuaki†
KONDO Keisuke ‡
RIEB, Kobe University
RIETI
RIETI
Abstract
This paper analyzes whether freshness of knowledge increases the quality of innovation by
using the Japanese patent database. Agglomeration is generally believed to foster the
creation of new knowledge through knowledge spillover, such as active face-to-face
communication; however, expansion of common knowledge within research communities
may discourage high-quality innovation. Taking this into consideration, we attempt to
examine the turnover effects of knowledge workers across cities by looking at the
interregional migration of university graduates. We find that the quality of innovation as
measured by the number of patent citations tends to be higher in cities with bigger
migration flows of university graduates. More importantly, we find that metabolizing
agglomeration plays an important role for high-quality innovative activities.
JEL classification: O31, O34, R12, R23
Keywords: Innovation, Patent, Migration, Agglomeration
RIETI Discussion Papers Series aims at widely disseminating research results in the form of
professional papers, thereby stimulating lively discussion. The views expressed in the papers are
solely those of the author(s), and neither represent those of the organization to which the author(s)
belong(s) nor the Research Institute of Economy, Trade and Industry.
We thank Yoshiyuki Arata, Masahisa Fujita, Arata Ito, Tomoya Mori, Masayuki Morikawa, Kentaro Nakajima,
Toshihiro Okubo, Yukiko Umeno Saito, Isamu Yamauchi, and the participants of the RIETI Discussion Paper Seminar
for their useful comments and suggestions. Naturally, any remaining errors are our own. We are grateful to the
Ministry of Internal Affairs and Communications for providing the micro-data of the Population Census, and we truly
thank Yuki Ishizuka for her excellent assistance in applying for the micro-data. This research was conducted under the
project “Restoration from Earthquake Damage and Growth Strategies of the Japanese Regional Economy” of the
Research Institute of Economy, Trade and Industry.
†Research Institute for Economics and Business Administration (RIEB), Kobe University: 2-1 Rokkodai-cho, Nada-ku,
Kobe-shi, Hyogo, 657-8501, Japan. (E-mail: hamaguchi@rieb.kobe-u.ac.jp)
‡Research Institute of Economy, Trade and Industry (RIETI): 1-3-1 Kasumigaseki, Chiyoda-ku, Tokyo, 100-8901, Japan.
(E-mail: kondo-keisuke@rieti.go.jp)
*
2
1 Introduction
Are large cities more innovative? Existing studies have shown keen interest in the significance
of cities in innovative activities. Lucas (1988) pointed out that externalities on human capital
are most likely manifested in cities since their dense concentration of people is best suited for
exploiting them. Perhaps the most prominent feature of cities is diversity in human resource.
Feldman and Audretsch (1999) found that diversity across complementary economic activities in
large cities sharing a common science base is conducive to innovation.
As a stylized fact, densely populated larger cities have more diversified industry and talent
(e.g., Bacolod et al., 2009). Knowledge exchanged with others will be important input for the
creation of new knowledge. Following the seminal work of Jaffe et al. (1993), many studies
have found that knowledge transfer and exchanges are constrained by distance (Criscuolo and
Verspagen, 2008; Inoue et al., 2013; Murata et al., 2014). Therefore, big cities are more suitable for
research and development, as evidences were already found that knowledge creation is related
to the size of city population (Bettencourt et al., 2007) or its density (Carlino et al., 2007).1 More
comprehensive literature review on agglomeration and innovation is provided by Carlino and
Kerr (2015).
Figure 1 presents the geographical distribution of patents in Japan at the municipality level.
Consistent with previous studies, big cities such as Tokyo, Osaka, and Nagoya tend to produce
more patents in both 1980 and 2000. However, in order to answer the opening question, information based on simple patent count data is unsatisfactory because not all patents represent
innovation with high value.2 In this regard, it is essential to introduce quality measurement of
patents to evaluate the degree of innovation.
[Figure 1]
Another issue is the relevance, if any, of population size or density to innovation. It is often
assumed that proximity to a greater number of people facilitates face-to-face communication
and fosters innovation. However, frequent interactions would increase common knowledge
and reduce diversity, which limit opportunities for learning from each other. In fact, Huber
(2012) indicates that technological knowledge spillover effects within the Cambridge Information
1
See also Audretsch and Feldman (1996) and Audretsch (1998).
Nagaoka et al. (2014, p. 1086) point out that “not all patents represent innovation, nor are all innovations are
patented.” Although this is beyond the perspective of this study, we should keep in mind that uncodified tacit
localized knowledge plays an important role in higher productivity in industrial clusters (Audretsch, 1998).
2
3
Technology Cluster are very weak. We therefore attempt to introduce an explicit notion of an
innovative city where knowledge exchange interaction is maintained at a high level.
Few studies have been done on the mechanism of knowledge exchange and creation. Analyzing the model based on a trade-off between the necessity of building common knowledge to
facilitate communication and the benefit of maintaining the exclusive knowledge of each worker,
Berliant and Fujita (2012) made a seminal contribution to show that the diversity of knowledge
workers matters, in addition to the number of these workers, in determining the production of
new knowledge, thus leading to higher long run economic growth. Importantly, Berliant and Fujita (2012) point out that just having a sufficiently large size of knowledge workers is not enough
to promote innovation because research cooperation combinations tend to be locked-in, i.e., it
tends to increase common knowledge while the diversity of exclusive knowledge is reduced.
Given a set of the population of knowledge workers, their findings call for higher mobility among
institutions.
In this study, we argue that migration flow brings in fresh brain power and sheds standardized
ideas; it also maintains the diversity of brain power at a city-level. For example, Faggian and
McCann (2009) criticized the existing literature on geography of innovation and said that it tends
to ignore the role played by the mobility of human capital. Their analysis demonstrated the
simultaneous significance of university graduate human capital inflows and regional innovation
performance.
Our study differs from Faggian and McCann (2009) in two aspects. First, we pay attention
not only to university graduate inflow but also to outflow. In a related study, Duranton and Puga
(2001) presented “a nursery city model” where diversified cities act as nurseries for firms to find
their ideal production process and later exit to seek for lower costs elsewhere. Such outflow is
essential to reduce congestion and uphold high productivity in innovation in diversified cities.
Second, we evaluate cities’ innovation productivity not by a simple count data of patents but by
citations that represent the value of each patent.
Our study uses the Institute of Intellectual Property (IIP) patent database, which contains
patent citation information. We measure the quality of innovation by the number of forward
patent citations by examiners.3 In the Japanese patent system, examiners of the Japan Patent
Office refer to prior art patents when they reject patent applications. As such, patents frequently
cited by the examiners can be viewed as an indicator of innovation quality.4 The patent citation
3
4
See Griliches (1990) and Nagaoka et al. (2014) for indicators of patent quality.
See also Jaffe and Trajtenberg (1999), Alcácer and Gittelman (2006), Singh (2008), Alcácer et al. (2009), and Cotropia
4
data on examiner rejection covers relatively a long period, which enables us to explore how the
turnover effects of knowledge workers have changed over time.
To sum up, our research question is to address whether the higher mobility of university
graduate human capital, both inflow and outflow, that sets the metabolism of a city’s brain power,
has an impact on higher value of created knowledge. We find that the quality of innovation
measured by the number of patent citation tends to be higher in cities with bigger migration flows
of university graduates. Furthermore, we show that the turnover effects of university graduates
are highly heterogeneous among different periods and locations.
The remainder of this paper will proceed as follows. Section 2 sets the empirical framework.
Section 3 describes the data. Section 4 discusses the estimation results. Finally, Section 5 concludes
the discussion.
2 Empirical Framework
2.1 Testing Impacts of Fresh Brain Power on Quality of Innovation
In order to address the research question mentioned in the previous section, we estimate the
following linear regression model:5
CITEDit = α log(GMrt ) + β log(PDrt ) + γEDrt + δDIrt + X irt η + τt + ψk + uit ,
(1)
where CITEDit is the number of forward citations of patent i applied in year t; GMrt is the average
gross migration flows of university graduates (i.e., sum of university graduate in-migrants and
out-migrants) in municipalities r where inventors are registered in patent i in year t; PDrt is the
average population density in municipalities r where inventors are registered in patent i in year t;
EDrt is the average share of university graduates in municipalities r where inventors are registered
in patent i in year t; DIrt is the average industrial diversity index in municipalities r where inventors
are registered in patent i in year t; X irt is the vector of control variables (municipality area, team
size, dummy of invention network, cross term of the team size and the dummy of invention
network, and dummies of technological classes); τt is the application year effect; ψk is the fixed
et al. (2013) for issues on self-citations of inventors and differences between inventor and examiner citations.
5
Although the Poisson regression may be a better framework for count data analysis, we use a linear regression
model in this study due to the difficulty of numerical maximization with the large size of firm fixed dummies.
5
effect of firm k that has the right of patent i; and uit is an error term.6
Parameter α captures the turnover effect of knowledge workers controlled for the effects of
agglomeration (PD), share of university graduates (ED), and industrial diversity (DI). We expect
that the parameter α takes the positive sign, indicating that the active metabolism of cities enhances
the quality of innovation.
Controlling for firm heterogeneity ψk is important to avoid an omitted variable bias for estimate
α. The quality of innovation will be higher for firms that have bigger investment for R&D
activity. Those firms attract more knowledge workers and therefore the migration variable may
be correlated with firm characteristics.
It may be criticized that the migration variable, log(GMrt ), does not consider the directions
of migration. To deal with this issue, we divided gross migration into in-migration and outmigration. The regression model (1) is modified as follows:
CITEDit = α1 log(IMrt ) + α2 log(OMrt ) + α3 log(IMrt ) · log(OMrt )
(2)
+ β log(PDrt ) + γEDrt + δDIrt + X irt η + τt + ψk + uit ,
where IMrt and OMrt respectively refer the average in-migration and out-migration flows of
university graduates in municipalities r where inventors are registered in patent i in year t.
Decomposing the turnover effect into three parameters α1 , α2 , and α3 , we consider the flows of
migration as well as the directions of migration. Looking at parameter α3 , we emphasize the
importance of the turnover of university graduates and not only their inflows.
2.2 Heterogeneity in Turnover Effects among Different Locations and Periods
We extend the benchmark framework (1) to examine in which place the turnover effect is bigger.
For this reason, we construct dummy variables related to each quantile of distributions in population density, share of university graduates, and industrial diversity, and introduce those cross
terms with log(GMrt ) into the regression (1). Thus, the regression model is extended as follows:
CITEDit = α log(GMrt ) +
4
k=2
+
4
θk log(GMrt ) · I(PDQTLk ) +
4
λk log(GMrt ) · I(EDQTLk )
k=2
(3)
φk log(GMrt ) · I(DIQTLk ) + β log(PDrt ) + γEDrt + δDIrt + X irt η + τt + ψk + uit ,
k=2
6
Note that the control variables include the logarithm of municipal area. This coefficient estimate is affected by
population density (Combes and Gobillon, 2015).
6
where PDQTLk , EDQTLk , and DIQTLk are the kth quantiles of distributions in population density,
share of university graduates, and industrial diversity index, respectively; I(·) is an indicator
function; and θk , λk , and φk are parameters of our interests capturing spatial heterogeneity in
turnover effects.7 Dividing the dataset into patents applied in 1980–1990 and patents applied in
1991–2005, we also attempt to look into temporal heterogeneity in turnover effects.
3 Data
Our analysis uses the IIP patent database developed by Goto and Motohashi (2007), which contains
Japanese patent information on application, registration, citation, applicant, inventor, and right
holder from 1964. We focus on patents applied between 1980 and 2005 to match inventors’
addresses with the municipal data of population census.
The IIP patent database has two types of patent citation by the examiners of the Japan Patent
Office (Goto and Motohashi, 2007, p. 1434). The first type relates to examiner rejection. If a patent
application is rejected, prior art references are given by the patent examiners. The second type
of citation is offered in the Patent Gazette when the examiners additionally mention important
prior art references although this is not for the rejection purpose. As discussed by Goto and
Motohashi (2007), citation information in the Patent Gazette is available from 1985 and later as
year of publication. To allow us to cover such a long time span in the database, our analysis
focuses only on patent citations for examiner rejection.8
We construct regional variables from population censuses. Whereas population, area, share
of university graduates, and employed workers by industry for each municipality are publicly
available, interregional migration flows by education level are not available. To make our own
dataset of their interregional migration, we use micro-data of the population census. Japan’s
population census includes questionnaires on residential mobility in five years preceding each
surveys conducted every 10 years, such as those from 1980, 1990, 2000, and 2010. Migration is
defined as the residential difference between where one lives at the time of the census and where
one lived five years before. We counted the number of university graduates who migrated 30 km
or more distance, where the distance is measured between two centroids of municipalities. Each
municipality has in-migration and out-migration, and gross migration is calculated as the sum of
7
Note that these dummy variables for each quantile are constructed from the distributions by year.
We did not use the third case where patents include both types of citations. In addition, we excluded patent
citations if the application years of the cited patents are newer than those of the citing patents.
8
7
in-migration and out-migration. Linear interpolation is implemented between each 10 years.
Municipal population density is calculated by the population divided by inhabitable area
(in km2 ) from 1980, 1985, 1990, 1995, 2000, and 2005 population censuses. Share of university
graduates is calculated as the ratio of university graduates to the total number of graduates.
Industrial diversity index is based on the relative measure suggested by Duranton and Puga
(2000).9 Linear interpolation for these variables is implemented between each 5 years, respectively.
An econometric issue arises since some patents have two or more co-assigning inventors, which
makes the identification of regional factors difficult. We solve this issue by making aggregate
regional variables. Our procedure is as follows. First, we make use of inventors’ addresses to
match patents with regional variables. Second, if a patent has two or more inventors residing in
different cities, we calculate the average regional values as benchmark estimation. For robustness
check, we also use maximum regional values among cities where inventors work.
The team size is defined as the number of inventors in patent i. To account for inventors’
network across regions, we introduce a dummy of invention network that takes the value of 1 if
co-assigning inventors work in different municipalities and 0 otherwise.
The unit of observation is each patent of firms. We might expect that the quality of innovation
will be higher for firms that have bigger investment for R&D activity. In order to control for
such firm heterogeneities, we construct firm dummy variables relying on the firm name database
offered by the National Institute of Science and Technology Policy (2014). This database can be
linked with the IIP patent database. We use the historical id of firms in the firm name database.
This database offered by the National Institute of Science and Technology Policy (NISTEP) covers
firms that have over 100 applications of patents between 1970 and 2010 or listed firms that have
less than 100 applications of patents. In this study, we focus on patents that are related with firms
listed in the firm name database of the NISTEP.
Table 1 presents descriptive statistics for the variable in our models.10 The average number of
citation is approximately 1, but actually as shown in Figure 2, more than half of patents applied for
in both 1980 and 2000 are never cited and that a small number of patents are highly cited. Table 2
shows the correlation matrix of regional variables. The migration flows of university graduates are
highly correlated with share of university graduates in 1980. However, this correlation coefficient
is lower in 2000.
9
Sectors are classified by the major division of the Japan Standard Industrial Classification ver. 10.
We made technology class dummies based on 33 technology classes. See Goto and Motohashi (2007) for more
details on the IIP patent database.
10
8
Figure 3 shows the scatter plots between the number of patent citations and gross migration
flows of university graduates in 1980 and 2000, respectively. We see that there are patents with
a large number of citations in areas where more university graduates immigrate and emigrate.
This is more evidently observed in 2000. The upper limit of patent citations becomes higher in
areas with higher turnover of university graduates. However, note that a lot of patents that are
never cited also exist there. In other words, not all patents in areas with bigger migration flows of
university graduates are shown as high-quality, but very high-quality patents are developed only
there.
[Tables 1–2 and Figures 2–3]
4 Estimation Results
4.1 Active Metabolism Leads to High-Quality Innovation
Table 3 presents the benchmark estimation results of regression (1). Columns (1)–(4) show that,
except for the industrial diversity index, migration flows of university graduates, population
density, and share of university graduates respectively have significant positive effects on the
number of patent citations except industrial diversity index. Column (5) shows that the turnover
effect remains significantly positive after controlling for population density, share of university
graduates, and industrial diversity index.11 In contrast, population density shows negative effects.
These results suggest that the turnover of knowledge workers, which is higher in dense areas,
enhances the quality of innovation. In other words, population density can be viewed as a
mediated factor to enhance the quality of innovation through active interregional migration of
knowledge workers.
Table 4 provides estimation results of regression (2), in which directions of migration flows are
considered. In Columns (1)–(3), we see that migration inflows, rather than the outflows, positively
influence the quality of patents. This finding is almost the same as Faggian and McCann (2009).
However, as a novel point of our study, Column (4) shows that the interaction between migration
inflows and outflows matters, which supports our hypothesis that the turnover of knowledge
workers matters more.
11
In Column (5) of Table 3, the increase of average gross migration flows of university graduates also affects the
average share of university graduates. Therefore, the marginal effect of gross migration flows of university graduates
is predicted to be bigger than that coefficient estimate.
9
[Tables 3–4]
4.2 Heterogeneity in Turnover Effects
Table 5 presents estimation results of regression (3). We see that the turnover effects of university
graduates are highly heterogeneous among different locations and periods. As shown earlier, the
interregional migration of university graduates has positive impact in Columns (1)–(3). Column
(2) shows that the turnover effects are amplified in denser areas with more diversified sectors for
patents applied in 1980–1990. Furthermore, as shown in Column (3), patents applied in 1991–2005
show not only similar results with those of Column (2); they also have bigger turnover effects in
areas with a higher share of university graduates.
Another interesting finding in Table 5 is that the cross term between team size and dummy of
invention network is significant between 1991 and 2005.12 The dummy of invention network has
negative impacts on the number of citations although this estimate is not significant even at the
10% level. This implies that the quality of innovation tends to decrease if co-inventors work in
different cities. As discussed by Criscuolo and Verspagen (2008); Inoue et al. (2013); Murata et al.
(2014), this may be explained by the distance-decay effects of innovation activity. However, the
positive effects of inventors’ diversity exceed negative distance-decay effects as the team size gets
bigger. Therefore, our estimation results suggest that knowledge diversity has been important for
recent innovation activities beyond the physical distance.
Summing up, this extension of our analysis suggests that the turnover of knowledge workers
in agglomerated areas with a higher share of university graduates and more diversified sectors
enhances the quality of patents.
[Table 5]
4.3 Robustness Check
We do two types of robustness check for our previous estimation results. First, we check if the
migration of non-university graduates, such as junior high school and high school graduates, also
might influence innovative activity. To examine which type of people’s migration has impact on
the quality of innovation, we include both migration flows of university graduates and the others
in regressions (1) and (2).
12
In our estimation results, team size shows significant positive impacts on the number of patent citations. This is
also found in Singh (2008).
10
Table 6 presents estimation results of this extension. Columns (1) and (2) show that migration
flows of university graduates, but not of others, positively affect the number of patent citations.
This result supports our emphasis on the role of active interactions between knowledge workers
for a higher quality of innovation.
The second robustness check is related to the aggregation method of regional variables. Thus
far, we have calculated the average values of migration flows, population density, share of university graduates, and industrial diversity index based on inventors’ addresses. However, one may
view this aggregation as a problem. Suppose that two inventors work in a big city and a rural
area, respectively. An inventor working in a rural area can also be affected by a big city through
the co-invention activity, and the use of the average value between two areas might be misleading.
Thus, we run regression (3) using the maximum regional values among inventors’ municipalities.
Table 7 presents estimation results of this regression. Compared to Table 5, the main results
do not change at all. The turnover effect remains significantly positive, and the interregional
migration of university graduates in denser areas with higher share of university graduates and
more diversified sectors has greater impact on the quality of innovation. Therefore, these two
robustness check analyses support our previous findings.
[Tables 6–7]
5 Conclusion
We investigated whether the interregional turnover of knowledge workers enhances the quality of
innovation. Our empirical analysis was based on Japan’s patent database, with which we matched
regional variables, such as migration flows of university graduates, population density, share of
university graduates, and industrial diversity index, with inventors’ addresses. The quality of
innovation was measured as the number of forward patent citations.
We found that the interregional gross migration flows of university graduates positively influence the number of patent citations. A more interesting finding was that this turnover effect shows
spatial and temporal heterogeneity. In other words, the turnover effects of knowledge workers
are bigger in bigger cities. In addition, the quality of patents recently developed in cities with
higher shares of university graduates and more diversified sectors tends to be higher. Furthermore, agglomeration can be viewed as a mediated factor to enhance the quality of innovation—
agglomeration tends to make interregional migration of knowledge workers active.
11
Our empirical findings have important policy implications and suggest that industrial cluster
policy aiming to active innovation does not necessarily work well if interregional migration of
knowledge workers is inactive. Thus, an important view for industrial cluster policy is how to
metabolize cities as a place of innovative activity.
References
[1]
Alcácer, Juan and Michelle Gittelman (2006) “Patent citations as a measure of knowledge
flows: The influence of examiner citations,” Review of Economics and Statistics 88(4), pp.
774–779.
[2]
Alcácer, Juan, Michelle Gittelman, and Bhaven Sampat (2009) “Applicant and examiner
citations in U.S. patents: An overview and analysis,” Research Policy 38(2), pp. 415–427.
[3]
Audretsch, David B. (1998) “Agglomeration and the location of innovative activity,” Oxford
Review of Economic Policy 14(2), pp. 18–29.
[4]
Audretsch, David B. and Maryann P. Feldman (1996) “R&D spillovers and the geography
of innovation and production,” American Economic Review 86(3), pp. 630–640.
[5]
Bacolod, Marigee, Bernardo S. Blum, and William C. Strange (2009) “Skills in the city,”
Journal of Urban Economics 65(2), pp. 136–153.
[6]
Berliant, Marcus and Masahisa Fujita (2012) “Culture and diversity in knowledge creation,”
Regional Science and Urban Economics 42(4), pp. 648–662.
[7]
Bettencourt, Luı́s M. A., José Lobo, Dirk Helbing, Christian Kühnert, and Geoffrey B. West
(2007) “Growth, innovation, scaling, and the pace of life in cities,” PNAS 104(17), pp. 7301–
7306.
[8]
Carlino, Gerald and William R. Kerr (2015) “Agglomeration and Innovation,” in Duranton,
Gilles, J. Vernon Henderson, and William C. Strange eds. Handbook of Regional and Urban
Economics Vol. 5, Amsterdam: Elsevier, Chap. 6, pp. 349–404.
[9]
Carlino, Gerald A., Satyajit Chatterjee, and Robert M. Hunt (2007) “Urban density and the
rate of invention,” Journal of Urban Economics 61(3), pp. 389–419.
[10]
Combes, Pierre-Philippe and Laurent Gobillon (2015) “The empirics of Agglomeration
Economies,” in Duranton, Gilles, J. Vernon Henderson, and William C. Strange eds. Handbook of Regional and Urban Economics Vol. 5, Amsterdam: Elsevier, Chap. 5, pp. 247–348.
[11]
Cotropia, Christopher A., Mark A. Lemley, and Bhaven Sampat (2013) “Do applicant patent
12
citations matter?” Research Policy 42(4), pp. 844–854.
[12]
Criscuolo, Paola and Bart Verspagen (2008) “Does it matter where patent citations come
from? Inventor vs. examiner citations in European patents,” Research Policy 37(10), pp.
1892–1908.
[13]
Duranton, Gilles and Diego Puga (2000) “Diversity and specialisation in cities: Why, where
and when does it matter?” Urban Studies 37(3), pp. 533–555.
[14]
Duranton, Gilles and Diego Puga (2001) “Nursery cities: urban diversity, process innovation,
and the life cycle of products,” American Economic Review 91(5), pp. 1454–1477.
[15]
Faggian, Alessandra and Philip McCann (2009) “Human capital, graduate migration and
innovation in British regions,” Cambridge Journal of Economics 33(2), pp. 317–333.
[16]
Feldman, Maryann P. and David B. Audretsch (1999) “Innovation in cities: Science-based
diversity, specialization and localized competition,” European Economic Review 43(2), pp.
409–429.
[17]
Goto, Akira and Kazuyuki Motohashi (2007) “Construction of a Japanese Patent Database
and a first look at Japanese patenting activities,” Research Policy 36(9), pp. 1431–1442.
[18]
Griliches, Zvi (1990) “Patent statistics as economic indicators: A survey,” Journal of Economic
Literature 28(4), pp. 1661–1707.
[19]
Huber, Franz (2012) “Do clusters really matter for innovation practices in Information
Technology? Questioning the significance of technological knowledge spillovers,” Journal
of Economic Geography 12(1), pp. 107–126.
[20]
Inoue, Hiroyasu, Kentaro Nakajima, and Yukiko Saito-Umeno (2013) “Localization of collaborations in knowledge creation,” RIETI Discussion Paper No. 13-E-070.
[21]
Jaffe, Adam B. and Manuel Trajtenberg (1999) “International knowledge flows: Evidence
from patent citations,” Economics of Innovation and New Technology 8(1–2), pp. 105–136.
[22]
Jaffe, Adam B., Manuel Trajtenberg, and Rebecca Henderson (1993) “Geographic localization
of knowledge spillovers as evidenced by patent citations,” Quarterly Journal of Economics
108(3), pp. 577–598.
[23]
Lucas, Robert E. Jr. (1988) “On the mechanics of economic development,” Journal of Monetary
Economics 22(1), pp. 3–42.
[24]
Murata, Yasusada, Ryo Nakajima, Ryosuke Okamoto, and Ryuichi Tamura (2014) “Localized
knowledge spillovers and patent citations: A distance-based approach,” Review of Economics
and Statistics 96(5), pp. 967–985.
13
[25]
Nagaoka, Sadao, Kazuyuki Motohashi, and Akira Goto (2014) “Patent statistics as an innovation indicator,” in Hall, Bronwyn H. and Nathan Rosenberg eds. Handbook of the Economics
of Innovation Vol. 2, Amsterdam: Elsevier, Chap. 25, pp. 1083–1127.
[26]
National Institute of Science and Technology Policy (2014) “Connection Table Between
NISTEP Firm Name Diccionary and External Databese (Ver. 2014.2),” the Ministry of
Education, Culture, Sports, Science and Technology.
(http://www.nistep.go.jp/research/scisip/data-and-information-infrastructure).
[27]
Singh, Jasjit (2008) “Distributed R&D, cross-regional knowledge integration and quality of
innovative output,” Research Policy 37(1), pp. 77–96.
Appendix A
Constructing the Municipal Panel Data of 1980–2005
The 1980–2005 municipal panel dataset on population density, share of university graduates, and
migration flows of university graduates is constructed by integrating data from the 1980 to 2005
Japanese population censuses. Municipal data on population and share of university graduates
are publicly available. However, migration flows of university graduates are not available online
and therefore calculated from the micro-data of the 1980, 1990, 2000, and 2010 population censuses.
The reference date for geographical information is April 1, 2011, at which date the total number
of municipalities was 1,747 (not counting the Northern Territories). The 23 wards of Tokyo are
counted individually. The cities designated by a government ordinance (Seirei Shitei Toshi) are
counted as cities (shi), rather than subcategories ku. The corresponding cities are Sapporo-shi,
Sendai-shi, Saitama-shi, Chiba-shi, Yokohama-shi, Kawasaki-shi, Sagamihara-shi, Niigata-shi,
Shizuoka-shi, Hamamatsu-shi, Nagoya-shi, Kyoto-shi, Osaka-shi, Sakai-shi, Kobe-shi, Okayamashi, Hiroshima-shi, Kitakyushu-shi, and Fukuoka-shi. In addition, we excluded Miyake-mura
and Ogasawara-mura from the sample. As such, the number of municipalities is 1,745.
Since some municipalities merged between 1980 and 2011, population, the number of university graduates, migration flows of university graduates are re-aggregated based on the information
for these merged municipalities.13
13
Changes in statistical area codes are available on the portal site for Japan’s official statistics, e-Stat:
(URL: http://www.e-stat.go.jp/SG1/hyoujun/initialize.do)
14
Appendix B
Estimation Results by Technology Class
To investigate heterogeneities across technology types, we estimate regression model (1) by dividing our sample according to technology class. We limited our sample to patents applied between
1991 and 2005, which allows us to focus on recent situations of innovation activities.
Table 8 presents the estimation results. The turnover effects of knowledge workers are significantly positive in 10 of 33 technology classes. Especially, the turnover effects are strongly observed
in technology classes related to electronic machinery industry and information technology machinery industry. As discussed earlier, population density is not a direct factor that enhances the
quality of innovation when the interregional migration of knowledge workers is considered. As
a policy implication, we need to pay attention to the fact that the turnover effects of knowledge
workers show heterogeneities across technology classes.
[Table 8]
15
Table 1: Descriptive Statistics of Variables for Regression Analysis
Variables
Mean
S.D.
Min
Max
Number of Patent Citation
Log(Average Gross Migration Flows of Univ. Grads.)
Log(Average Gross Migration Flows of Others)
Log(Average Migration Inflows of Univ. Grads.)
Log(Average Migration Outflows of Univ. Grads.)
Log(Av. Migration Inflows) × Log(Av. Migration Outflows)
Log(Average Migration Inflows of Others)
Log(Average Migration Outflows of Others)
Log(Av. Migration Inflows) × Log(Av. Migration Outflows)
Log(Average Population Density)
Average Share of University Graduates
Average Industrial Diversity Index
Log(Area)
Log(Av. Gross Mig. Flows of Univ. Grad.) × I(PDQTL2 )
Log(Av. Gross Mig. Flows of Univ. Grad.) × I(PDQTL3 )
Log(Av. Gross Mig. Flows of Univ. Grad.) × I(PDQTL4 )
Log(Av. Gross Mig. Flows of Univ. Grad.) × I(EDQTL2 )
Log(Av. Gross Mig. Flows of Univ. Grad.) × I(EDQTL3 )
Log(Av. Gross Mig. Flows of Univ. Grad.) × I(EDQTL4 )
Log(Av. Gross Mig. Flows of Univ. Grad.) × I(DVQTL2 )
Log(Av. Gross Mig. Flows of Univ. Grad.) × I(DVQTL3 )
Log(Av. Gross Mig. Flows of Univ. Grad.) × I(DVQTL4 )
Team Size
D(1: Invention Network)
Team Size × D(1: Invention Network)
0.985
9.498
9.931
8.780
8.804
79.482
9.226
9.231
86.964
8.224
15.405
4.092
4.091
2.329
2.560
2.547
2.363
2.349
2.791
2.267
2.455
2.755
2.271
0.108
0.378
1.808
1.485
1.344
1.470
1.526
26.014
1.348
1.363
24.865
1.213
6.623
1.592
1.182
4.085
4.561
4.344
4.187
4.315
4.565
3.933
4.429
4.577
1.467
0.311
1.202
0.000
2.633
3.875
0.322
2.528
0.814
2.303
3.642
8.386
2.081
1.575
0.838
1.230
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
1.000
0.000
0.000
68.000
12.426
12.450
11.761
11.703
137.639
11.882
11.692
137.992
10.010
35.233
18.446
6.507
12.201
12.426
12.132
11.992
12.325
12.426
12.317
12.426
12.384
38.000
1.000
38.000
Note: The number of observation is 1,903,672. The uppermost 0.001 percentile of the distribution of patent
citations is excluded from the sample as extreme outliers.
1.000
0.964
0.995
0.995
0.964
0.954
0.458
0.500
0.215
0.585
1.000
0.978
0.996
0.996
0.974
0.975
0.523
0.291
0.435
0.625
Application Year: 2000
(1) Log(Av. Migration Gross Flow of Univ. Graduates)
(2) Log(Av. Migration Gross Flow of Others)
(3) Log(Av. Migration Inflow of Univ. Graduates)
(4) Log(Av. Migration Outflow of Univ. Graduates)
(5) Log(Av. Migration Inflow of Others)
(6) Log(Av. Migration Outflow of Others)
(7) Log(Av. Population Density)
(8) Av. Share of Univ. Graduates
(9) Av. Industrial Diversity Index
(10) Log(Area)
(1)
Application Year: 1980
(1) Log(Av. Migration Gross Flow of Univ. Graduates)
(2) Log(Av. Migration Gross Flow of Others)
(3) Log(Av. Migration Inflow of Univ. Graduates)
(4) Log(Av. Migration Outflow of Univ. Graduates)
(5) Log(Av. Migration Inflow of Others)
(6) Log(Av. Migration Outflow of Others)
(7) Log(Av. Population Density)
(8) Av. Share of Univ. Graduates
(9) Av. Industrial Diversity Index
(10) Log(Area)
Variables
1.000
0.973
0.974
0.996
0.997
0.452
0.111
0.463
0.711
1.000
0.957
0.962
0.995
0.994
0.417
0.290
0.231
0.661
(2)
1.000
0.985
0.974
0.966
0.510
0.305
0.426
0.630
1.000
0.981
0.964
0.940
0.406
0.485
0.255
0.622
(3)
1.000
0.966
0.976
0.535
0.279
0.441
0.614
1.000
0.956
0.959
0.503
0.506
0.181
0.545
(4)
1.000
0.986
0.411
0.109
0.461
0.736
1.000
0.979
0.385
0.297
0.252
0.684
(5)
Table 2: Correlation Matrix of Regional Variables
1.000
0.486
0.118
0.464
0.681
1.000
0.447
0.283
0.209
0.629
(6)
1.000
0.491
−0.007
−0.254
1.000
0.498
−0.268
−0.352
(7)
1.000
−0.109
−0.329
1.000
−0.111
−0.171
(8)
1.000
0.468
1.000
0.440
(9)
1.000
1.000
(10)
16
17
Table 3: Turnover Effects by Gross Migration Flows in 1980–2005
Dependent Variable: Number of Patent Citations
Explanatory Variables
Log(Av. Mig. Flows of Univ. Grads.)
(1)
(2)
(3)
Team Size × D(1: Network)
Application Year Dummy
Technology Class Dummy
Firm Fixed Effects
Number of Observations
Adjusted R̄2
Number of Firms
1903672
0.0694
3900
0.0090***
(0.0018)
Av. Industrial Diversity Index
D(1: Invention Network)
0.0486***
(0.0109)
−0.0325***
(0.0106)
0.0052***
(0.0017)
0.0007
(0.0056)
−0.0475***
(0.0134)
0.0786***
(0.0060)
−0.0060
(0.0231)
0.0100
(0.0064)
Yes
Yes
Yes
0.0274***
(0.0084)
Av. Share of University Graduates
Team Size
(5)
0.0388***
(0.0089)
Log(Av. Population Density)
Log(Area)
(4)
−0.0392***
(0.0091)
0.0786***
(0.0060)
−0.0084
(0.0231)
0.0099
(0.0063)
Yes
Yes
Yes
−0.0001
(0.0055)
0.0788***
(0.0060)
−0.0076
(0.0230)
0.0097
(0.0063)
Yes
Yes
Yes
0.0062
(0.0054)
0.0788***
(0.0060)
−0.0074
(0.0230)
0.0098
(0.0064)
Yes
Yes
Yes
0.0023
(0.0056)
−0.0060
(0.0053)
0.0789***
(0.0060)
−0.0009
(0.0235)
0.0098
(0.0063)
Yes
Yes
Yes
1903672
0.0697
3900
1903672
0.0695
3900
1903672
0.0697
3900
1903672
0.0694
3900
Note: Heteroskedasticity-consistent standard errors clustered by technology class are in parentheses. Constant is not reported. * denotes statistical significance at the 10% level, ** at the 5% level, and *** at the 1%
level.
18
Table 4: Turnover Effects by Migration Flows and Directions in 1980–2005
Dependent Variable: Number of Patent Citations
Explanatory Variables
Log(Av. Mig. Inflow of Univ. Grads.)
(1)
(2)
(3)
(4)
0.0320***
(0.0082)
0.0497**
(0.0208)
0.0022
(0.0151)
−0.0325***
(0.0111)
0.0050***
(0.0016)
0.0005
(0.0055)
−0.0516***
(0.0164)
0.0786***
(0.0060)
−0.0061
(0.0231)
0.0100
(0.0063)
Yes
Yes
Yes
−0.0218**
(0.0105)
0.0064***
(0.0018)
0.0014
(0.0057)
−0.0293***
(0.0098)
0.0786***
(0.0060)
−0.0068
(0.0230)
0.0099
(0.0064)
Yes
Yes
Yes
−0.0331***
(0.0108)
0.0049***
(0.0017)
0.0005
(0.0056)
−0.0520***
(0.0151)
0.0786***
(0.0060)
−0.0061
(0.0231)
0.0100
(0.0063)
Yes
Yes
Yes
−0.0367
(0.0288)
−0.0635***
(0.0207)
0.0097***
(0.0024)
−0.0442***
(0.0101)
0.0038**
(0.0017)
0.0036
(0.0060)
−0.0775***
(0.0166)
0.0787***
(0.0060)
0.0017
(0.0224)
0.0101
(0.0063)
Yes
Yes
Yes
1903672
0.0698
3900
1903672
0.0698
3900
1903672
0.0698
3900
1903672
0.0699
3900
0.0514***
(0.0127)
Log(Av. Mig. Outflow of Univ. Grads.)
Log(Av. Mig. Inflow) × Log(Av. Mig. Outflow)
Log(Av. Population Density)
Av. Share of University Graduates
Av. Industrial Diversity Index
Log(Av. Area)
Team Size
D(1: Invention Network)
Team Size × D(1: Invention Network)
Application Year Dummy
Technology Class Dummy
Firm Fixed Effects
Number of Observations
Adjusted R̄2
Number of Firms
Note: Heteroskedasticity-consistent standard errors clustered by technology class are in parentheses. Constant is not reported. * denotes statistical significance at the 10% level, ** at the 5% level, and *** at the 1%
level.
19
Table 5: Spatial and Temporal Heterogeneity in Turnover Effects
Dependent Variable: Number of Patent Citations
Explanatory Variables
Log(Av. Mig. Flows of Univ. Grad.)
Log(Av. Mig. Flows of Univ. Grad.) × I(PDQTL2 )
Log(Av. Mig. Flows of Univ. Grad.) × I(PDQTL3 )
Log(Av. Mig. Flows of Univ. Grad.) × I(PDQTL4 )
Log(Av. Mig. Flows of Univ. Grad.) × I(EDQTL2 )
Log(Av. Mig. Flows of Univ. Grad.) × I(EDQTL3 )
Log(Av. Mig. Flows of Univ. Grad.) × I(EDQTL4 )
Log(Av. Mig. Flows of Univ. Grad.) × I(DIQTL2 )
Log(Av. Mig. Flows of Univ. Grad.) × I(DIQTL3 )
Log(Av. Mig. Flows of Univ. Grad.) × I(DIQTL4 )
Log(Av. Population Density)
Av. Share of University Graduates
Av. Industrial Diversity Index
Log(Av. Area)
Team Size
D(1: Invention Network)
Team Size × D(1: Invention Network)
Application Year Dummy
Technology Class Dummy
Firm Fixed Effects
Number of Observations
Adjusted R̄2
Number of Firms
1980–2005
1980–1990
1991–2005
(1)
(2)
(3)
0.0298***
(0.0107)
0.0016
(0.0020)
0.0086**
(0.0032)
0.0083*
(0.0046)
0.0058***
(0.0015)
0.0058*
(0.0029)
0.0088**
(0.0033)
0.0021
(0.0024)
0.0041*
(0.0021)
0.0062**
(0.0030)
−0.0604***
(0.0110)
0.0025
(0.0023)
−0.0080
(0.0062)
−0.0533***
(0.0141)
0.0788***
(0.0060)
0.0045
(0.0223)
0.0099
(0.0064)
Yes
Yes
Yes
0.0270**
(0.0110)
0.0026
(0.0024)
0.0081**
(0.0036)
0.0086*
(0.0046)
0.0016
(0.0026)
0.0039
(0.0033)
0.0043
(0.0040)
0.0049***
(0.0017)
0.0055**
(0.0023)
0.0069*
(0.0034)
−0.0408**
(0.0160)
−0.0017
(0.0026)
−0.0124
(0.0099)
−0.0443***
(0.0125)
0.0524***
(0.0039)
0.0091
(0.0247)
−0.0062
(0.0072)
Yes
Yes
Yes
0.0355**
(0.0138)
0.0010
(0.0022)
0.0110***
(0.0038)
0.0091*
(0.0048)
0.0110***
(0.0019)
0.0131***
(0.0030)
0.0182***
(0.0043)
0.0016
(0.0031)
0.0066**
(0.0030)
0.0116**
(0.0046)
−0.0816***
(0.0155)
−0.0014
(0.0027)
−0.0151**
(0.0056)
−0.0870***
(0.0209)
0.0907***
(0.0078)
−0.0354
(0.0295)
0.0254***
(0.0073)
Yes
Yes
Yes
1903672
0.0699
3900
725543
0.0521
3011
1178129
0.0839
3779
Note: Heteroskedasticity-consistent standard errors clustered by technology class are in parentheses. Constant is not reported. * denotes statistical significance at the 10% level, ** at the 5% level, and *** at the 1%
level.
20
Table 6: Robustness Check by Migration Flows of University Graduates and the Others
Dependent Variable:
Number of Patent Citations
Explanatory Variables
Log(Av. Mig. Flows of Univ. Grads.)
Log(Av. Mig. Flows of Others)
(1)
0.0847**
(0.0332)
−0.0447
(0.0356)
−0.0283***
(0.0100)
0.0037
(0.0025)
0.0006
(0.0056)
−0.0401***
(0.0131)
0.0786***
(0.0060)
−0.0090
(0.0236)
0.0100
(0.0064)
Yes
Yes
Yes
−0.0481
(0.0420)
−0.0282
(0.0489)
0.0103**
(0.0048)
0.0246
(0.0686)
−0.0546
(0.0758)
−0.0006
(0.0067)
−0.0404***
(0.0094)
0.0023
(0.0021)
0.0034
(0.0059)
−0.0717***
(0.0170)
0.0788***
(0.0060)
−0.0018
(0.0225)
0.0100
(0.0064)
Yes
Yes
Yes
1903672
0.0698
3900
1903672
0.0699
3900
Log(Av. Mig. Inflow of Univ. Grads.)
Log(Av. Mig. Outflow of Univ. Grads.)
Log(Av. Mig. Inflow of Univ.) × Log(Av. Mig. Outflow of Univ.)
Log(Av. Mig. Inflow of Others)
Log(Av. Mig. Outflow of Others)
Log(Av. Mig. Inflow of Others) × Log(Av. Mig. Outflow of Others)
Log(Av. Population Density)
Av. Share of University Graduates
Av. Industrial Diversity
Log(Av. Area)
Team Size
D(1: Invention Network)
Team Size × D(1: Invention Network)
Application Year Dummy
Technology Class Dummy
Firm Fixed Effects
Number of Observations
Adjusted R̄2
Number of Firms
(2)
Note: Heteroskedasticity-consistent standard errors clustered by technology class are in parentheses. Constant is not reported. * denotes statistical significance at the 10% level, ** at the 5% level, and *** at the 1%
level.
21
Table 7: Robustness Check by Maximum Values of Regional Variables
Dependent Variable: Number of Patent Citations
Explanatory Variables
Log(Max Mig. Flows of Univ. Grad.)
Log(Max Mig. Flows of Univ. Grad.) × I(PDQTL2 )
Log(Max Mig. Flows of Univ. Grad.) × I(PDQTL3 )
Log(Max Mig. Flows of Univ. Grad.) × I(PDQTL4 )
Log(Max Mig. Flows of Univ. Grad.) × I(EDQTL2 )
Log(Max Mig. Flows of Univ. Grad.) × I(EDQTL3 )
Log(Max Mig. Flows of Univ. Grad.) × I(EDQTL4 )
Log(Max Mig. Flows of Univ. Grad.) × I(DIQTL2 )
Log(Max Mig. Flows of Univ. Grad.) × I(DIQTL3 )
Log(Max Mig. Flows of Univ. Grad.) × I(DIQTL4 )
Log(Max Population Density)
Max Share of University Graduates
Max Industrial Diversity Index
Log(Max Area)
Team Size
D(1: Invention Network)
Team Size × D(1: Invention Network)
Application Year Dummy
Technology Class Dummy
Firm Fixed Effects
Number of Observations
Adjusted R̄2
Number of Firms
1980–2005
1980–1990
1991–2005
(1)
(2)
(3)
0.0284***
(0.0086)
0.0001
(0.0023)
0.0081**
(0.0035)
0.0058
(0.0049)
0.0060***
(0.0017)
0.0069**
(0.0033)
0.0100***
(0.0035)
0.0010
(0.0026)
0.0034
(0.0023)
0.0047
(0.0030)
−0.0553***
(0.0112)
0.0013
(0.0022)
−0.0049
(0.0050)
−0.0540***
(0.0100)
0.0788***
(0.0060)
−0.0085
(0.0219)
0.0091
(0.0062)
Yes
Yes
Yes
0.0209**
(0.0088)
0.0018
(0.0032)
0.0073
(0.0048)
0.0061
(0.0054)
0.0016
(0.0028)
0.0046
(0.0033)
0.0047
(0.0039)
0.0040**
(0.0018)
0.0054**
(0.0024)
0.0048
(0.0032)
−0.0319*
(0.0158)
−0.0016
(0.0024)
−0.0074
(0.0080)
−0.0384***
(0.0094)
0.0525***
(0.0039)
0.0069
(0.0269)
−0.0055
(0.0068)
Yes
Yes
Yes
0.0221**
(0.0106)
−0.0012
(0.0021)
0.0112***
(0.0037)
0.0083*
(0.0048)
0.0112***
(0.0019)
0.0144***
(0.0037)
0.0193***
(0.0049)
0.0009
(0.0035)
0.0057
(0.0034)
0.0114**
(0.0049)
−0.0734***
(0.0140)
−0.0011
(0.0026)
−0.0128***
(0.0045)
−0.0752***
(0.0145)
0.0908***
(0.0078)
−0.0613**
(0.0243)
0.0252***
(0.0075)
Yes
Yes
Yes
1903672
0.0699
3900
725543
0.0521
3011
1178129
0.0839
3779
Note: Heteroskedasticity-consistent standard errors clustered by technology class are in parentheses. Constant is not reported. * denotes statistical significance at the 10% level, ** at the 5% level, and *** at the 1%
level.
14102
0.0839
569
0.0280
(0.0878)
−0.0002
(0.0636)
−0.0096
(0.0098)
0.0571***
(0.0149)
−0.0353
(0.1066)
0.0255***
(0.0066)
−0.0324
(0.1128)
−0.0115
(0.0352)
Yes
Yes
Yes
8542
0.1034
463
0.1351**
(0.0610)
−0.0518
(0.0633)
−0.0097
(0.0074)
0.0139
(0.0153)
−0.1874**
(0.0730)
0.1467**
(0.0663)
0.2903*
(0.1601)
−0.0946
(0.0645)
Yes
Yes
Yes
ITC=2
Food stuffs
17840
0.0743
805
0.0544
(0.0490)
−0.1192***
(0.0451)
0.0021
(0.0058)
−0.0060
(0.0132)
−0.0338
(0.0549)
0.0281***
(0.0074)
−0.1838
(0.2010)
0.0833
(0.0735)
Yes
Yes
Yes
ITC=3
Personal
and
domestic
articles
8257
0.1295
386
−0.0157
(0.0994)
0.1198
(0.0887)
−0.0208
(0.0138)
0.0137
(0.0243)
−0.1192
(0.1192)
0.0919***
(0.0282)
0.1834
(0.1505)
−0.0832**
(0.0419)
Yes
Yes
Yes
−0.0210
(0.0509)
0.0562
(0.0436)
−0.0060
(0.0069)
−0.0460***
(0.0161)
−0.0121
(0.0571)
0.0745***
(0.0107)
−0.3996**
(0.1988)
0.0986
(0.0644)
Yes
Yes
Yes
44482
0.1079
1015
ITC=5
Drugs
24701
0.0679
1364
0.1021**
(0.0449)
−0.0194
(0.0374)
0.0027
(0.0048)
0.0077
(0.0108)
−0.0661
(0.0522)
0.0801***
(0.0096)
−0.0079
(0.0978)
0.0223
(0.0255)
Yes
Yes
Yes
ITC=6
Separating,
mixing
32038
0.0655
1020
−0.0408
(0.0248)
0.0594***
(0.0201)
0.0046
(0.0033)
−0.0155***
(0.0058)
0.0546*
(0.0315)
0.0640***
(0.0060)
0.0445
(0.0562)
−0.0116
(0.0164)
Yes
Yes
Yes
ITC=7
Machine
tools, metal
working
ITC=8
Casting,
grinding,
layered
product
38756
0.1022
1482
0.1265***
(0.0335)
−0.0874***
(0.0274)
0.0041
(0.0046)
−0.0309***
(0.0091)
−0.1158***
(0.0403)
0.0954***
(0.0085)
−0.1903**
(0.0919)
0.0547*
(0.0300)
Yes
Yes
Yes
Dependent Variable: Number of Patent Citations
ITC=4
Health and
amusement
31544
0.0880
589
0.0958
(0.0591)
−0.1387**
(0.0578)
0.0136**
(0.0069)
0.0153
(0.0139)
−0.1207*
(0.0689)
0.0798***
(0.0098)
0.0536
(0.1182)
−0.0427
(0.0311)
Yes
Yes
Yes
ITC=9
Printing
54609
0.0880
763
0.0485
(0.0341)
−0.0321
(0.0281)
−0.0005
(0.0044)
−0.0063
(0.0069)
−0.0566
(0.0390)
0.0852***
(0.0059)
−0.0384
(0.0695)
−0.0015
(0.0212)
Yes
Yes
Yes
ITC=10
Transporting
41097
0.0737
1491
−0.0062
(0.0278)
−0.0011
(0.0233)
0.0081**
(0.0033)
−0.0073
(0.0061)
−0.0100
(0.0330)
0.0325***
(0.0052)
0.0254
(0.0594)
0.0094
(0.0172)
Yes
Yes
Yes
ITC=11
Packing,
lifting
Note: Heteroskedasticity-consistent standard errors are in parentheses. Constant is not reported. * denotes statistical significance at the 10% level, ** at the 5% level, and *** at the 1% level.
Number of Observations
Adjusted R̄2
Number of Firms
Application Year Dummy
Technology Class Dummy
Firm Fixed Effects
Team Size × D(1: Network)
D(1: Invention Network)
Team Size
Log(Av. Area)
Av. Industrial Diversity Index
Av. Share of University Graduates
Log(Av. Population Density)
Log(Av. Mig. Flows of Univ. Grad.)
Explanatory Variables
ITC=1
Agriculture
Table 8: Robustness Check by Technology Class Using Patents applied in 1991–2005
22
21164
0.0644
873
16209
0.0593
547
0.0073
(0.0384)
−0.0202
(0.0304)
0.0021
(0.0052)
−0.0189*
(0.0107)
−0.0320
(0.0445)
0.0982***
(0.0120)
−0.0802
(0.0744)
0.0123
(0.0206)
Yes
Yes
Yes
−0.0152
(0.0460)
−0.0501
(0.0390)
0.0200***
(0.0052)
−0.0186
(0.0131)
0.0816
(0.0541)
0.0687***
(0.0112)
−0.0241
(0.1160)
0.0149
(0.0328)
Yes
Yes
Yes
32174
0.0604
692
0.1263***
(0.0448)
−0.1016***
(0.0367)
−0.0017
(0.0056)
−0.0427***
(0.0121)
−0.1616***
(0.0500)
0.1554***
(0.0147)
−0.2658**
(0.1289)
0.0731*
(0.0389)
Yes
Yes
Yes
ITC=14
Organic
molecule
compounds
16213
0.0938
762
0.2007**
(0.0794)
−0.1891***
(0.0679)
0.0089
(0.0076)
0.0043
(0.0144)
−0.2543***
(0.0865)
0.0903***
(0.0182)
−0.4821***
(0.1508)
0.1360***
(0.0463)
Yes
Yes
Yes
3746
0.0594
430
−0.0049
(0.0849)
0.0043
(0.0660)
0.0118
(0.0142)
0.0060
(0.0233)
0.0564
(0.1098)
0.1176**
(0.0481)
−0.4190**
(0.1805)
0.0604
(0.0598)
Yes
Yes
Yes
1194
0.1284
183
0.1038
(0.1379)
−0.0652
(0.0939)
−0.0300*
(0.0169)
0.0225
(0.0434)
−0.1632
(0.1613)
0.0399
(0.0411)
−0.2928
(0.2453)
0.0702
(0.0684)
Yes
Yes
Yes
ITC=16
ITC=17
Biotechnology, Genetic
engineering
beer, fermentation
27879
0.0738
729
−0.0151
(0.0331)
0.0222
(0.0255)
0.0045
(0.0037)
0.0073
(0.0076)
−0.0342
(0.0393)
0.0599***
(0.0082)
−0.1162
(0.0883)
0.0305
(0.0238)
Yes
Yes
Yes
ITC=18
Metallurgy,
coating
metals
Dependent Variable: Number of Patent Citations
ITC=15
Dyes,
petroleum
13687
0.0831
501
−0.0564
(0.0639)
0.0037
(0.0452)
0.0016
(0.0089)
−0.0076
(0.0134)
0.0257
(0.0732)
0.0334***
(0.0130)
−0.2226*
(0.1205)
0.0854**
(0.0351)
Yes
Yes
Yes
ITC=19
Textile
2760
0.1037
272
−0.1412
(0.1544)
0.2709**
(0.1359)
−0.0167
(0.0183)
0.0314
(0.0382)
0.0699
(0.1962)
0.1209***
(0.0446)
−0.1338
(0.2381)
−0.0082
(0.0756)
Yes
Yes
Yes
ITC=20
Paper
46593
0.0664
1253
−0.0370*
(0.0208)
0.0242
(0.0196)
−0.0006
(0.0025)
−0.0080
(0.0053)
0.0249
(0.0251)
0.0270***
(0.0047)
0.0001
(0.0396)
0.0128
(0.0106)
Yes
Yes
Yes
ITC=21
Construction
ITC=22
Mining,
drilling
4322
0.0467
229
0.0686
(0.0705)
−0.0220
(0.0639)
−0.0045
(0.0068)
−0.0368***
(0.0133)
−0.0089
(0.0840)
0.0104
(0.0106)
−0.1002
(0.1094)
0.0109
(0.0310)
Yes
Yes
Yes
Note: Heteroskedasticity-consistent standard errors are in parentheses. Constant is not reported. * denotes statistical significance at the 10% level, ** at the 5% level, and *** at the 1% level.
Number of Observations
Adjusted R̄2
Number of Firms
Application Year Dummy
Technology Class Dummy
Firm Fixed Effects
Team Size × D(1: Network)
D(1: Invention Network)
Team Size
Log(Av. Area)
Av. Industrial Diversity Index
Av. Share of University Graduates
Log(Av. Population Density)
Log(Av. Mig. Flows of Univ. Grad.)
Explanatory Variables
ITC=13
Organic
chemistry,
pesticides
ITC=12
Nonorganic
chemistry,
fertilizer
Table 8: Robustness Check by Technology Class Using Patents applied in 1991–2005 (Continued)
23
39844
0.0777
633
30387
0.0916
1167
0.0407
(0.0414)
−0.0482
(0.0359)
0.0081
(0.0052)
0.0120
(0.0087)
−0.0322
(0.0485)
0.0716***
(0.0077)
−0.1547
(0.2086)
0.0953
(0.0686)
Yes
Yes
Yes
−0.0429
(0.0343)
0.0348
(0.0305)
0.0015
(0.0042)
−0.0262***
(0.0082)
0.0278
(0.0462)
0.0607***
(0.0054)
−0.2586**
(0.1148)
0.0775**
(0.0346)
Yes
Yes
Yes
38052
0.0637
915
−0.0729*
(0.0378)
0.0436
(0.0311)
0.0102***
(0.0037)
0.0149*
(0.0083)
0.0894**
(0.0448)
0.0567***
(0.0071)
−0.0150
(0.1090)
0.0461
(0.0293)
Yes
Yes
Yes
ITC=25
Lighting,
steam
generation,
heating
687
0.1281
98
0.2910*
(0.1649)
−0.3593***
(0.1360)
−0.0146
(0.0163)
−0.0115
(0.0381)
−0.4414**
(0.2008)
0.0477
(0.0390)
0.8975**
(0.4082)
−0.1703*
(0.0958)
Yes
Yes
Yes
138392
0.0711
1779
0.0007
(0.0248)
0.0048
(0.0209)
0.0010
(0.0026)
−0.0022
(0.0060)
0.0212
(0.0296)
0.1088***
(0.0045)
−0.0511
(0.1959)
0.0190
(0.0540)
Yes
Yes
Yes
79067
0.0837
1279
0.0926**
(0.0416)
−0.0133
(0.0384)
0.0139***
(0.0040)
0.0255***
(0.0094)
−0.0992**
(0.0471)
0.1024***
(0.0077)
−0.1067
(0.0984)
0.0435
(0.0269)
Yes
Yes
Yes
ITC=27
ITC=28
Measurement, Clock,
optics, pho- controlling,
computer
tography
54911
0.0507
801
0.1980***
(0.0574)
−0.1589***
(0.0513)
−0.0024
(0.0055)
0.0210
(0.0135)
−0.2681***
(0.0633)
0.1122***
(0.0102)
0.3672**
(0.1482)
−0.0444
(0.0330)
Yes
Yes
Yes
ITC=29
Display,
information
storage,
instruments
Dependent Variable: Number of Patent Citations
ITC=26
Weapons,
blasting
2805
0.1235
119
0.0583
(0.1055)
−0.0364
(0.1066)
−0.0097
(0.0107)
−0.0077
(0.0296)
−0.0961
(0.1181)
0.0846***
(0.0193)
0.1615
(0.1679)
−0.0089
(0.0358)
Yes
Yes
Yes
ITC=30
Nuclear
physics
178142
0.0878
1551
0.1022***
(0.0193)
−0.0614***
(0.0175)
0.0070***
(0.0021)
0.0274***
(0.0049)
−0.1150***
(0.0227)
0.1303***
(0.0044)
0.0511
(0.0788)
0.0097
(0.0217)
Yes
Yes
Yes
ITC=31
Electronics
components, semi
conductor
113729
0.0689
830
0.1544***
(0.0431)
−0.0604*
(0.0358)
−0.0001
(0.0034)
0.0157*
(0.0093)
−0.1792***
(0.0489)
0.0927***
(0.0061)
0.1288
(0.1326)
0.0514
(0.0370)
Yes
Yes
Yes
ITC=32
Electronics
circuit, communication
tech.
ITC=33
Others
204
0.3844
56
1.2958
(1.4004)
−0.5356
(1.1176)
0.0278
(0.1213)
−0.4925
(0.2988)
−2.1267
(1.6884)
0.0264
(0.1196)
6.1287*
(3.5550)
−0.8721*
(0.5149)
Yes
Yes
Yes
Note: Heteroskedasticity-consistent standard errors are in parentheses. Constant is not reported. * denotes statistical significance at the 10% level, ** at the 5% level, and *** at the 1% level.
Number of Observations
Adjusted R̄2
Number of Firms
Application Year Dummy
Technology Class Dummy
Firm Fixed Effects
Team Size × D(1: Network)
D(1: Invention Network)
Team Size
Log(Av. Area)
Av. Industrial Diversity Index
Av. Share of University Graduates
Log(Av. Population Density)
Log(Av. Mig. Flows of Univ. Grad.)
Explanatory Variables
ITC=24
Engineering
elements
ITC=23
Engine,
pump
Table 8: Robustness Check by Technology Class Using Patents applied in 1991–2005 (Continued)
24
25
Tokyo
Osaka
1000 or more
1000 or more
100-1000
100-1000
50- 100
Nagoya
Tokyo
10- 50
Osaka
50- 100
Nagoya
10- 50
1- 10
1- 10
less than 1
less than 1
Number of Patents in 1980
(a) 1980 (Application Year)
Number of Patents in 2000
(b) 2000 (Application Year)
Figure 1: Geographical Distribution of Patents
Note: Created by the authors from the IIP Patent Database. Patents that were applied in 1980 and in 2000
and finally registered are 48,397 and 92,950 in our sample, respectively. Some observations with garbled
characters in inventors’ addresses excluded from the data. We assigned location to each patent based on
the inventors’ address. Patents by multiple inventors are divided with equal weights in each inventors’
address.
.6
.6
.5
.5
.4
.4
Fraction
Fraction
26
.3
.3
.2
.2
.1
.1
0
0
5 10 15 20 25 30 35 40 45 50 55 60 65 70
0
Number of Patent Citations
(a) 1980 (Application Year)
0
5 10 15 20 25 30 35 40 45 50 55 60 65 70
Number of Patent Citations
(b) 2000 (Application Year)
Figure 2: Number of Patent Citations
Note: Created by the authors from the IIP Patent Database.
27
70
Number of Patent Citations
Number of Patent Citations
70
60
50
40
30
20
10
0
60
50
40
30
20
10
0
2
4
6
8
10
12
Log(Av. Gross Migration Flows of University Graduates)
(a) 1980 (Application Year)
2
4
6
8
10
12
Log(Av. Gross Migration Flows of University Graduates)
(b) 2000 (Application Year)
Figure 3: Number of Patent Citations and Gross Migration Flows
Note: Created by the authors from the IIP Patent Database. Sample in Table 1.