Professional Documents
Culture Documents
Abstract
Using the 2004 China economic census database, this paper examines the impact of
information and communication technologies (ICT) on the geographical concentration
of manufacturing industries, controlling for a broad set of other determinants of
industrial agglomeration. Contrary to the argument that ICT leads to more dispersion,
it is found that ICT promotes geographical concentration of industries. The results
are robust to different measures of ICT, at different geographical levels and to the
consideration of endogeneity.
1. Introduction
Economists have long recognised that of agglomeration economies is geographical
agglomeration—of both population and proximity. To be more specific, the demand
firms—in cities, yields economic benefits. for geographical proximity stems from the
In Marshall’s view (Marshall, 1920), firms demand for face-to-face communications,
of the same industry concentrate in a city to especially in an innovative business environ-
incorporate specialised labour pooling and ment where tacit, uncodifiable and rapidly
the availability of intermediate inputs, as well changing information is crucial to decision-
as information and knowledge spillovers. In making (Storper and Venables, 2004).
Jacobs’ opinion (Jacobs, 1961), firms of dif- Many recent studies have confirmed the
ferent industries concentrating in a city can existence and significance of agglomera-
facilitate cross-fertilisation of new ideas. In tion economies among firms and among
both arguments, the underlining determinant workers in cities. Further, economists have
Junjie Hong is in the School of International Trade and Economics, University of International Business
and Economics, Beijing, 100029, China. E-mail: hongjunjie@alumni.nus.edu.sg.
Shihe Fu is in the Research Institute of Economics and Management, Southwestern University of
Finance and Economics, Chengdu, China. E-mail: fush@swufe.edu.cn.
tried to disentangle the foundations of firm that agglomeration economies are localised
agglomeration. For example, Rosenthal and and decay with geographical distances. For
Strange (2001) examined the impact of example, Jaffe et al. (1993) found that pat-
knowledge spillovers, labour market pool- ent citations decrease when distance from
ing, input-sharing and natural advantage the company holding the patent increases.
on industrial agglomeration. Audretsch and Previous studies also found that business
Feldman (1996) found that industries that agglomeration economies (Rosenthal and
emphasise research and development (R&D) Strange, 2003) and labour market agglom-
are more likely to concentrate in an area. eration economies (Fu, 2007; Rosenthal and
Lovely et al. (2005) investigated whether the Strange, 2008) attenuate rapidly with distance.
need to acquire information contributes to Although these studies did not explicitly pre-
spatial concentration, while Nakamura (2005) dict the impact of ICT on agglomeration, the
focused on the forward and backward linkage implication is that a decrease in the cost of
externalities.1 communication and transport over distance
Along with the rapid progress and wide will attenuate agglomeration and lead to more
application of information and communica- dispersion.
tion technologies (ICTs) during the past few The second strand of literature predicts
decades, an interesting research question has that ICT will attenuate the demand for face-
been debated: if people can communicate over to-face communications and thus will result
a long distance at decreasing cost via phones, in greater dispersion of economic activities.
fax machines, the Internet and e-mails, is Ota and Fujita (1993) constructed a general
it still necessary for people and firms to be equilibrium model of multiunit firms’ loca-
located close to each other? Or, put in another tion and showed that the development of
way, will the improvement of ICT attenuate, information technology will lead to a greater
or even eliminate the geographical concen- concentration of front-units in the city cen-
tration of economic activities, or even cities? tre and to a dispersion of back-units in the
There is no consensus, either theoretically or far suburbs. Sivitanidou (1997) found that
empirically. between 1989 and 1994 the office-commercial
Theoretically speaking, while ICTs have land value gradients within polycentric Los
led to the de-agglomeration or dispersion Angeles flattened. His interpretation was
of some routine activities, ICTs have also that the recent information revolution had
created a knowledge-based economy result- weakened the attractiveness of large business
ing in more new complex opportunities centres to office-commercial activities, result-
that continue to require face-to-face contact ing in the increasingly dispersed patterns of
(Storper and Venables, 2004). In addition, business locations. Ioannides et al. (2007)
it is argued that information is inevitably developed a formal model showing that the
embedded in social relations since learn- improvement of ICT will increase the disper-
ing (transforming information to people’s sion of economic activities across cities, sug-
knowledge) relies on complex social interac- gesting that city sizes would be more uniform.
tions (Brown and Duguid, 2002). Therefore, They also used cross-country city size data
whether ICT promotes agglomeration or not and found supportive evidence.2
is an empirical question. The third strand, however, has an opposite
Existing empirical studies on the relation- viewpoint. Gaspar and Glaeser (1998) devel-
ship between ICT and the geographical con- oped a theoretical model and demonstrated
centration of industries can be grouped into that ICT and face-to-face interactions could be
roughly three strands. The first strand found complementary, rather than substitute goods.
The reason is that, while some face-to-face out separately, at these three geographical
contacts will be replaced and conducted elec- levels. The empirical results show a positive
tronically, the improved ICT may result in more and strong association between the adoption
face-to-face interactions. If the second force is of ICT and the geographical concentration
dominant, then ICT may strengthen agglomer- of industries. These results are quite robust
ation. They also provided some suggestive evi- to different measures of ICT as well as at
dence, such as the negative correlation between different geographical levels. To deal with
geographical distance and number of phone the concerns on omitted variables and endo-
calls, the complementary relationship between geneity issues, we add two-digit industry
business travel and telecommunications, and dummies and control for government policy
the increase in co-authorship in economics. that might have affected both ICT usage and
Panayides and Kern (2005) extended Gaspar geographical concentration. We also use an
and Glaeser’s model and found that improve- instrumental variable approach based on a
ments in ICT may increase or decrease the predetermined value of ICT. The estimation
demand for face-to-face interactions, depend- results provide further support to the argu-
ing on the cross-price elasticity. If the cross- ment that the adoption of ICT promotes the
elasticity is negative, then city size may increase geographical concentration of industries.
with electronic communications. Kolko (2000) The next section describes the measurement
found that commercial Internet domain density of variables, the econometric model specifica-
(the ratio of commercial Internet domains to tions and identification issues and strategies.
commercial establishments) is higher in larger Section 3 introduces the data and provides
cities, controlling for a broad range of other summary statistics. Section 4 presents the
factors. He interpreted this as evidence that estimation results and section 5 concludes.
face-to-face contact is a complement to elec-
tronic communication.3
2. Variable Definitions and
This study contributes to the literature by
Econometric Model Specification
providing new evidence for unravelling the
dispute between the competing hypotheses This section begins by discussing the mea-
regarding the relationship between ICT and surement of geographical concentration.
the geographical concentration of manufac- The variables that proxy for ICT, and other
turing industries. Specifically, we examine determinants of industrial agglomeration,
the impact of ICT on the spatial concentra- are then defined, followed by the specification
tion of manufacturing industries in China, of econometric models and identification
controlling for other main determinants of strategies.
geographical concentration. The data used in
this research are drawn from the 2004 China 2.1 Measuring Geographical
economic census database, which is believed Concentration
to be the most comprehensive micro-level The two widely used indexes of geographical
database in China thus far. It contains detailed concentration are the Gini coefficient, pro-
information on the entire universe of manu- posed by Krugman (1991), and the Ellison–
facturing firms in China. For each four-digit Glaeser index (Ellison and Glaeser, 1997). The
manufacturing industry, we compute the difficulty with the Gini coefficient is that an
Ellison–Glaeser index (Ellison and Glaeser, industry will be regarded as highly localised if
1997) to measure the level of geographical there are several very large firms concentrat-
concentration at the county, city and province ing in a limited number of locations. The EG
levels. The econometric analyses are carried index, however, can control for differences
in firm size. This paper uses the EG index, own websites where relevant information is
calculated according to the following formula posted, including descriptions of companies
and products, company and industrial news,
G − (1 − ∑ mi2 )H career information and after-sale services.
γ=
i (1) Some company sites also support business
(1 − ∑ mi )(1 − H )
2
transactions and electronic commerce. With
i
the popularity of the Internet, a company’s
where, γ is the EG index (also called the website has become increasingly important.
Gamma index), for a particular industry; One can expect that the companies that have
G = ∑ i (si − mi )2; and s is the ratio of location
i their own website may have more advanced
i’s employment in a particular industry to the information technology. Thus, we use share of
national employment in that industry; mi is companies that have a website in an industry,
the ratio of location i’s total employment to as a proxy of ICT.
the national employment; and H = ∑ j z j
2
E-mails and fax machines have also been
denotes the (employment) Herfindahl index widely used in business and generate new
of the J plants in the industry, with z j repre- options for communication. Some rela-
senting the employment share of the jth plant. tionships that previously would have been
The Gamma index has been widely conducted face-to-face have been replaced
employed in recent studies on industrial by telecommunications. For instance, a com-
agglomeration (for example, Lovely et al., pany manager now can send an e-mail to the
2005; Rosenthal and Strange, 2001). Values clients instead of meeting them in person.
of the Gamma index usually range between Business partners can fax or e-mail draft
-1 and 1. γ takes on a value of zero when contracts to each other instead of delivering
an industry is as concentrated as one would them by hand. However, as noted by Gaspar
expect if the plants in the industry choose and Glaeser (1998), there is an opposing
locations by throwing darts at a map. A posi- effect when telecommunication improves:
tive value of γ indicates excess concentration, advanced ICT makes communication easier
while a negative value of γ implies an excess and hence increases the number of relation-
diffusion of employment. ships. For example, with improved ICT, a
company manager can contact more clients
2.2 Measuring Information and and run more projects, which implies that the
Communication Technologies manager needs more face-to-face contacts.
Geographical distance is a hindrance to face- Since the requirement of face-to-face contacts
to-face contacts and the communication of contributes to agglomeration, the net effect of
ideas. With the improvement of ICT, some telecommunication improvement on indus-
face-to-face contacts are replaced electroni- trial agglomeration is unknown. Therefore,
cally and the costs of communicating ideas we use the share of firms that have a fax number
over distance are reduced, implying that or an e-mail address in an industry, as a proxy
advanced ICT may affect agglomeration for ICT. Since firms that have more computers
forces. The focus of this paper is to test the per worker are believed to have better ICT,
impact of ICT on industrial agglomeration. we also try share of firms with above-average
The rapid development of ICT has been computer share, as an additional proxy.5
characterised by the ever-increasing use of
phones, fax machines, personal comput- 2.3 Controls for Other Variables
ers, the Internet and e-mails, over the past Our model includes a set of control variables
few decades.4 Many companies have their that may affect industrial agglomeration. The
first set of control variables concern labour officials and help to maintain social stability
market pooling. We use share of workers with and serve other social purposes (Bai et al., 2004;
a master’s or college degree6 in an industry Lu and Tao, 2009). Therefore, share of employ-
respectively, as proxies for labour pooling. ment in state-owned enterprises in an industry
We also define average labour intensity in an is used to measure local protectionism.
industry as proportion of number of employ- Our model also considers the impact of
ees over total asset to examine the impact of firm age. Rosenthal and Strange (2001) found
labour markets. The basic idea is that the that, compared with the agglomeration of all
need for human capital and labour intensity establishments, agglomeration of new estab-
may have an impact on a firm’s motivation lishments is not as strongly related to agglom-
to concentrate. erative spillovers and natural advantages. We
The second set of variables controls for include share of young firms in an industry in
natural advantages and knowledge spillovers. the model, where young firms are defined as
Industries concentrate partially due to natural those that are five years old or younger.
advantages, as discussed in Ellison and Glaeser
(1999). We use share of firms in the industry 2.4 Econometric Model Specification
that are located in coastal provinces as a proxy and Identification
for natural advantage, because coastal regions We are interested in how the Gamma index is
in China tend to have flatter terrain, a bet- affected by ICT, controlling for other determi-
ter climate and easier access to seaports and nants of industrial agglomeration, as defined
international markets. Knowledge spillover earlier. The econometric model is specified
is also an important factor in determining as follows
industrial concentration. Because knowledge γ ij = δ X j + eij
(2)
spillover is found to be more significant in
R&D intensive industries, industries that are where, γ ij is the Gamma index for industry j
highly innovative are expected to have higher (four-digit code) at geographical level i; X j is
levels of agglomeration (Lovely et al., 2005). a vector of industry characteristics, including
Similar to previous studies (for example, variables that proxy for ICT, labour pooling,
Lovely et al., 2005), we use number of innova- knowledge spillovers and natural advantages;
tions per worker in an industry, as a proxy for δ is the coefficient vector to be estimated; and
knowledge spillover.7 eij is the error term, assumed to be indepen-
Previous studies have shown that local dently and identically distributed.
protectionism is an important and special The benchmark model is estimated at the
determinant of geographical concentra- county level. To test the robustness of our
tion in China (Bai et al., 2004; Lu and Tao, specification, we also estimate the model at
2009). Local governments have motivation the city and province levels. In China, a city
to protect local firms and industries due to is larger than a county, in terms of land area.
fiscal decentralisation after China’s economic A city normally contains a number of coun-
reform. There was a rise of local protection- ties. There are 31 provinces, 345 cities and
ism in China during the reform era (Young, 2831 counties in mainland China.
2000), creating barriers to trade and imped- The key identification assumption is that,
ing the process of industrial agglomeration. after including the control variables in the
Local governments give more protection to model, the ICT variables are uncorrelated
industries with higher shares of state owner- with the error term. This assumption may
ship, because state-owned enterprises can cre- be violated if there are important omitted
ate much more benefit for local government variables or if firms in highly concentrated
Sedillot, 1999; Rosenthal and Strange, 2001) be very high, but the EG index at the county
used only a proportion of manufacturing level may not be so high. These results are
firms in a country.12 Secondly, all data used also consistent with those of previous studies
in this research are from the same census (for example, Rosenthal and Strange, 2001),
database, therefore the consistency of data suggesting that spillovers go beyond a small
is guaranteed. However, the census database area (Ellison and Glaeser, 1997).
available for research is only at the firm level Table 2 indicates that there are large varia-
and not at the establishment or plant level. tions in spatial concentration across two-
One concern is that, if a firm has several plants digit industries. At the county level, the most
located in different places, and all employees localised industries are cultural, educational
are assigned to the headquarter, a measure- and sports goods, and smelting and pressing
ment error arises when we calculate industrial of non-ferrous metals. At the city and prov-
agglomeration. Fortunately, only a very small ince levels, the most concentrated industries
proportion (about 1.778 per cent) of firms are smelting and pressing of non-ferrous met-
have multiple plants. In addition, according als, and electronics and telecommunication.
to the rule of the State Statistical Bureau of We note that some high-tech industries are
China, all employees of a multiplant firm are highly concentrated (for example, electronics
allocated to the address where the main pro- and telecommunication), while other high-
duction takes place. Thus, we believe that this tech industries are much less concentrated
issue does not lead to serious bias, especially, (for example, medical and pharmaceutical
at the higher geographical levels. products). The pattern of traditional indus-
The data cover 482 four-digit manufac- tries is also mixed. For instance, cultural, edu-
turing industries. Based on the firm-level cational and sports goods are very localised,
data, we calculate the Gamma index at the while the tobacco processing industry is quite
county, city and province levels, for each dispersed.
four-digit industry. Table 1 shows that there An industry is as concentrated as a ran-
is substantially more concentration at the dom allocation when the EG index equals
higher geographical levels: the mean value zero and is excessively concentrated when
of the Gamma index at the province level the index is positive. Using manufacturing
is 0.0657, while at the county level it is only census data for the US, Ellison and Glaeser
0.0155. These results are as expected, since (1997) defined not very localised (state level
provinces are larger than cities and counties. 0 < γ < 0.02), intermediate, and very localised
If employment in an industry is spread over a (state level γ > 0.05) ranges. When we apply
number of counties within one province, the the same classification criteria of Ellison and
Gamma index at the provincial level would Glaeser (1997) to Chinese manufacturing
Table 1. Summary of the EG Gamma index at the four-digit industry level
Correlation with γ at
the level of
Table 3. Distribution of the Gamma index at the four-digit industry level
Gamma γ at the county level γ at the city level γ at the province level
γ≤0 286 144 45
0 < γ ≤ 0.02 135 197 107
0.02 < γ ≤ 0.05 44 92 147
0.05 < γ ≤ 0.10 9 33 113
0.10 < γ ≤ 0.20 8 9 47
γ > 0.20 0 7 23
industries, 107 out of 482 four-digit industries Compared with the spatial concentration
are not very localised at the province level, of US manufacturing industries (for example,
while 183 of them are very localised, as shown Rosenthal and Strange, 2001), Chinese indus-
in Table 3. tries are more concentrated on average. For
example, the average EG index equals 0.0657 websites, e-mail addresses and fax numbers
at the province level in China, while that for are, on average, 6.58 per cent, 8.16 per cent
the US is 0.0485 at the state level (Rosenthal and 42.04 per cent respectively. There are 33.4
and Strange, 2001). However, we need to be per cent of firms with a computer share above
cautious when making such comparisons, the average level. The correlation coefficients
since the Chinese provinces are normally among these ICT-related variables are quite
larger than American states in terms of land high. To avoid multicollinearity, we will not
area. In addition, the official classification of include them in the same equation.
industries is different. Table 4 also reports descriptive statistics for
Table 4 reports descriptive statistics of other control variables. The variable of num-
the explanatory variables. In the census, all ber of innovations per worker is at the three-
firms were required to report their websites, digit industry level and is drawn from the
fax numbers, e-mail addresses and number China Economic Census Yearbook 2004 (State
of computers, if any. There was an auditor, Statistical Bureau, 2006). All other variables
who has been trained formally by the State are computed directly from the census data-
Statistical Bureau, to recheck the data and base and are at the four-digit industry level.
information reported after a firm filled in
the census form, which guarantees the qual- 4. Estimation Results
ity of the database. The value takes one if a
firm has a website, fax or e-mail, and zero 4.1 Benchmark model results
otherwise, based on which we compute ICT- Table 5 presents the estimation results for
related variables at the four-digit industry the benchmark model, with the EG index
level. The shares of firms that have their own at the county level as the explained variable.
Table 5. Benchmark model: regression results at the county level (N = 482)
and Glaeser, 1998; Panayides and Kern, levels. One interesting finding is that the
2005), which motivates firms to agglomerate magnitude of all four ICT variables increases
geographically. at higher geographical levels. For instance,
the coefficient for website is 0.088 at the
4.2 Robustness Tests county level; it increases to 0.113 at the city
In the benchmark model, we regress the level and to 0.280 at the province level. We
Gamma index on possible determinants also note that model fitness is improved sig-
at the county level. Readers might wonder nificantly at higher geographical levels. This
whether our results are sensitive to different pattern is consistent with that in Rosenthal
levels of geography. To address this concern, and Strange (2001).
we estimate the model at the city and prov- Another concern is that our results may be
ince levels. The results are reported in Table 6, subject to outlier bias. Compared with other
where each estimate is taken from a separate industries, the electronics and telecommu-
regression. It shows that all ICT variables are nication industry has a significantly higher
significant at both the city and the province Gamma index, as well as ICT adoption.
levels. This provides supportive evidence that A natural concern is that the estimation
our results are robust at different geographical results may be biased by the inclusion of
Table 6. Robustness test: regression results Table 7. Estimation results: inclusion of
at the city and province levels two-digit industry dummies and policy effects
Table 8. Correlation between ICT variables instrument. With regard to the exclusion
and the instrumental variable restriction (i.e. the instrumental variable does
not affect geographical concentration through
Correlation
between ICT with channels other than the ICT variables), we
instrumental believe that the causality of lagged e-mail is
variablea Coefficientb irreversible: new firm location can respond
to past industry ICT adoption, but not vice
Website 0.557 0.507*** versa. We do a test related to the exclusion
(5.58)
Fax 0.536 1.388*** restriction by regressing the residuals from
(6.39) the second-stage estimations on the instru-
E-mail 0.606 0.593*** mental variable. If the instrumental variable
(6.42) affects the industrial agglomeration through
Computer 0.408 0.678***
(5.19) other channels, the residuals from the second-
stage estimations should be correlated with
a
The instrumental variable is the share of the instrumental variable (Lu and Tao, 2009).
state-owned or above-designated-size firms that
The regression results consistently show that the
have e-mail in an industry in 2003.
b
Coefficient is on the instrumental variable correlation between the two is close to zero in
(lagged e-mail) in a regression of ICT variables magnitude and is statistically insignificant.14
(website, fax, e-mail, computer respectively) on In summary, these tests show that lagged
lagged e-mail. e-mail is a good instrument.
Note: *** indicates the significance at the Table 9 shows that all ICT coefficient esti-
1 per cent level. mates are positive and significant, although
the coefficient estimates lose some signifi-
at the 1 per cent level (column 2 of Table 8), cance compared with the benchmark models.
which indicates the instrument strength. The As in previous studies (for example, Alfaro
IV regression results are reported in Table 9. et al., 2004), the coefficients increase consid-
F-values greater than 10 in the first-stage erably in values compared with the earlier
regressions confirm again the validity of the OLS results in Tables 5 and 6. One possible
First-stage partial
County level γ City level γ Province level γ R2 and F-value
interpretation is that instrumental variable exact mechanisms through which ICT affects
estimation here corrects for measurement industrial agglomeration are left for future
error, which biases the OLS coefficients to research agenda.
zero. Taken together, the results in this sec-
tion lend additional support to the view that
modern information and communication Notes
technologies have resulted in a higher degree 1. For more related studies, see the literature
of industrial agglomeration. review by Duranton and Puga (2004) and
Rosenthal and Strange (2004).
2. There are a few futurists predicting the
5. Conclusions disappearance of cities, see Toffler (1980) and
It has long been speculated that adoption of Naisbitt (1995).
ICT may decrease industrial agglomeration 3. Using individual on-line and off-line
shopping behaviour data, Sinai and Waldfogel
because developments in telecommunications
(2004) found that the Internet can be both
have generated new options for communica- a complement to cities and a substitute for
tions and have replaced face-to-face contacts. cities.
Some people even predict that industrial 4. We do not test the impact of phone lines,
clusters will decline or disappear because because almost all firms use telephones.
firms have no need to locate close to each 5. We do not use average computer share of an
other with improvements in telecommuni- industry as a proxy, because a small proportion
cations technologies. However, others argue of companies have a very large number of
that face-to-face communication and tele- computers. We use share of firms with above-
communication can be complements, since average computer share to avoid outlier bias.
6. The variable share of workers with bachelor’s
face-to-face contact is necessary for learning
degree is not included, since it is highly
and creative activities, and adoption of ICT correlated with other variables. In China,
can increase the number of business relation- college-degree holders normally receive a
ships. Our research examines the impact of three-year education, while bachelor-degree
advanced information and communication holders require a four-year education.
technologies on the geographical concentra- 7. Following Rosenthal and Strange (2001), we
tion of manufacturing industries in China. have also tried to use energy consumption per
We use the 2004 China economic census worker and technological funds per worker
data and compute the Ellison–Glaeser index to measure natural advantage and knowledge
to measure the geographical concentration spillover respectively, and find quite consistent
estimation results.
of four-digit industries. After controlling for
8. Lovely et al. (2005) found that exporter
the main industrial characteristics that may headquarters are more agglomerated when
influence geographical concentration, such foreign market information is difficult to
as labour pooling and natural advantage, we obtain.
find that adoption of ICT actually increases 9. We do not use average export share of an
geographical concentration. These results industry as a proxy, because a small proportion
are quite robust to various measures of ICT, of companies have a very high export share.
at different geographical levels, the inclusion In order to avoid outlier bias, we use share of
of other determinants of agglomeration and firms with above-average export share instead.
10. The causality direction between some
consideration of endogeneity. Our findings
industrial characteristics (such as labour
suggest that knowledge spillovers through quality) and agglomeration can run both
face-to-face contact might still be important ways. For these variables, the coefficients may
for manufacturing industries. However, the reflect the equilibrium relationship rather than
causal effects. Since our focus is the impact of Vol. IV, pp. 2063–2117. Amsterdam: Elsevier
ICT, we will conduct causality analysis only North-Holland.
on ICT effects. Ellison, G. and Glaeser, E. (1997) Geographic
11. Firms that are above designated size are defined concentration in U.S. manufacturing indus-
as those with annual sales of over 5 million tries: a dartboard approach, Journal of Political
Chinese yuan. We did not use predetermined Economy, 105, pp. 879–927.
values of ICT in earlier years as instruments, Ellison, G. and Glaeser, E. (1999) The geographic
because the industry category and code by concentration of an industry: does natural
State Statistical Bureau changed significantly advantage explain agglomeration?, American
in 2003. Economic Review Papers and Proceedings, 89,
12. An exception is a recent paper by Lu (2010) pp. 311–316.
that used the first and second Chinese national Fu, S. (2007) Smart café cities: testing human
establishment censuses in 1996 and 2001, capital externalities in the Boston metropolitan
which cover all manufacturing establishments area, Journal of Urban Economics, 61, pp. 86–111.
in China. Gaspar, J. and Glaeser, E. (1998) Information
13. To save space, the estimation results are not technology and the future of cities, Journal of
reported here, but are available from the Urban Economics, 43, pp. 136–156.
authors upon request. Henderson, J. (2003) Marshall’s scale economies,
14. To save space, the estimation results are not Journal of Urban Economics, 53, pp. 1–28.
reported here, but are available from the Hong, J. (2009) Firm heterogeneity and location
authors upon request. choices: evidence from foreign manufacturing
investments in China, Urban Studies, 46(10),
pp. 2143–2157.
Acknowledgement Ioannides, Y., Overman, H., Rossi-Hansberg, E.
and Schmidheiny, K. (2007) The effect of
Shihe Fu gratefully acknowledges financial support
information and communication technologies
from Project 211 (Phase III) of the Southwestern
on urban structure. Discussion Paper No. 812,
University of Finance and Economics, Chengdu,
Centre for Economic Performance, London
China.
School of Economics.
Jacobs, J. (1961) The Death and Life of Great
References American Cities. New York: Vintage Books.
Jaffe, A. B., Trajtenberg, M. and Henderson, R.
Alfaro, L., Chanda, A., Kalemli-Ozcan, S. and (1993) Geographic localization of knowl-
Sayek, S. (2004) FDI and economic growth: edge spillovers as evidenced by patent cita-
the role of local financial markets, Journal of tions, Quarterly Journal of Economics, 108,
International Economics, 64, pp. 89–112. pp. 577–598.
Audretsch, D. B. and Feldman, M. (1996) R&D Kolko, J. (2000) The death of cities? The death
spillovers and the geography of innovation of distance? Evidence from the geography of
and production, American Economic Review, commercial internet usage, in: I. Vogelsang and
86, pp. 630–640. B. Compaine (Eds) The Internet Upheaval: Raising
Bai, C., Du, Y., Tao, Z. and Tong, S. (2004) Local Questions, Seeking Answers in Communications
protectionism and regional specialization: Policy, pp. 73–97. Cambridge, MA: MIT Press.
evidence from China’s industries, Journal of Krugman, P. (1991) Geography and Trade.
International Economics, 63, pp. 397–417. Cambridge, MA: MIT Press.
Brown, J. S. and Duguid, P. (2002) The Social Lovely, M., Rosenthal, S. and Sharma, S. (2005)
Life of Information. Cambridge, MA: Harvard Information, agglomeration, and the head-
Business School Press. quarters of U.S. exporters, Regional Science and
D u r a n t o n , G . a n d P u g a , D. ( 2 0 0 4 ) Urban Economics, 35, pp. 167–191.
Microfoundations of urban agglomeration Lu, J. (2010) Agglomeration of economic activi-
economies, in: V. Henderson and J. Thisse (Eds) ties in China: evidence from establishment
Handbook of Regional and Urban Economics, censuses, Regional Studies, 44(3), pp. 281–297.
Lu, J. and Tao, Z. (2009) Trends and determinants Rosenthal, S. and Strange, W. (2004) Evidence on the
of China’s industrial agglomeration, Journal of nature and sources of agglomeration economies,
Urban Economics, 65(2), pp. 167–180. in: V. Henderson and J. Thisse (Eds) Handbook
Marshall, A. (1920) Principles of Economics. of Regional and Urban Economics, Vol. IV,
London: Macmillan. pp. 2119–2171. Amsterdam: Elsevier North-
Maurel, F. and Sedillot, B. (1999) A measure of the Holland.
geographic concentration in French manufac- Rosenthal, S. and Strange, W. (2008) The attenu-
turing industries, Regional Science and Urban ation of human capital spillovers, Journal of
Economics, 29, pp. 575–604. Urban Economics, 64(2), pp. 373–389.
Naisbitt, J. (1995) The Global Paradox. New York: Shaver, J. and Flyer, E. (2000) Agglomeration
Avon Books. economies, firm heterogeneity, and foreign
Nakamura, R. (2005) Agglomeration economies direct investment in the United States, Strategic
and linkage externalities in urban manufacturing Management Journal, 21, pp. 1175–1193.
industries: a case study of Japanese cities. Paper Sinai, T. and Waldfogel, J. (2004) Geography
presented at the 45th Congress of the European and the Internet: is the Internet a substitute
Regional Science Association, Amsterdam, or a complement for cities?, Journal of Urban
August. Economics, 56, pp. 1–24.
Ota, M. and Fujita, M. (1993) Communication Sivitanidou, R. (1997) Are center access advantages
technologies and spatial organization of multi- weakening? The case of office-commercial mar-
unit plants in metropolitan areas, Regional kets, Journal of Urban Economics, 42, pp. 79–97.
Science and Urban Economics, 23, pp. 695–729. State Statistical Bureau (2006) China Economic Census
Panayides, A. and Kern, C. (2005) Information Yearbook 2004. Beijing: China Statistics Press.
technology and the future of cities: an alterna- Storper, M. and Venables, A. J. (2004) Buzz: face-
tive analysis, Urban Studies, 42, pp. 263–267. to-face contact and the urban economy, Journal
Rosenthal, S. and Strange, W. (2001) The deter- of Economic Geography, 4, pp. 351–370.
minants of agglomeration, Journal of Urban Toffler, A. (1980) The Third Wave. New York:
Economics, 50, pp. 191–229. Bantam Books.
Rosenthal, S. and Strange, W. (2003) Geography, Young, A. (2000) The razor’s edge: distortions and
industrial organization, and agglomera- incremental reform in the People’s Republic of
tion, Review of Economics and Statistics, 85, China, Quarterly Journal of Economics, 115(4),
pp. 377–393. pp. 1091–1135.