INTRODUCTION

In the economic geography literature, the issue of balanced regional development has revolved around the degree to which agglomeration economies increase the concentration of economic activities in advanced regions, and hence perpetuate regional economic disparities (Ottaviano & Thisse, 2004; Scott & Storper, 2003). Agglomeration economies refer to the beneficial effects for firms stemming from the concentration of economic activities in a region (Caniëls & Verspagen, 2001). They arise, in particular, from the vast opportunities for the exchange and spillovers of information and expertise among organizations in a region. Geographical proximity facilitates spillovers of knowledge because, as knowledge is intrinsically sticky (Winter, 1987), it permits close and repeated interactions between actors, allowing them to understand, identify and integrate relevant knowledge (Antonelli, 2001; Arundel & Guena, 2004; Boschma, 2005; Breschi & Lissoni, 2009; Jaffe, 1989; Ponds, Van Oort, & Frenken, 2010; Verspagen & Schoenmakers, 2004). Confirming this, a vast body of empirical research finds that geographical distance is a barrier to knowledge spillovers (Breschi & Lissoni, 2009; Verspagen & Schoenmakers, 2004; Wang, Meijers, & Szirmai, 2017). From the point of view of peripheral regions, particularly those located very far from centres of excellence, a greater degree of agglomeration economies implies lower prospects of benefitting from the technological advancements of core regions, causing them potentially to fall behind on the technological ladder.

In spite of the advantages of geographical proximity, however, innovation process is displaying a growing trend of cooperation between innovators in different geographical locations (Autant-Bernard, Billand, Frachisse, & Massard, 2007; Waltman, Tijssen, & Eck, 2011). The propensity to collaborate even grows with greater distance (Letaifa & Rabeau, 2013). Accessing expertise from different geographical settings with non-overlapping knowledge bases can facilitate greater possibilities for knowledge combination and create more radical and valuable innovations (Cockburn & Henderson, 1997; Petruzzelli, 2011). More distant collaborations are also believed to lead to higher social impact (Frenken, Ponds, & Van Oort, 2010; Nomaler, Frenken, & Heimeriks, 2013). In the specific context of regional technological development, long-distance cooperation among innovators assumes a singular importance as it can contribute to the technological catching up of regions lagging behind in technological capabilities.

Collaboration network of scientists reflects the cognitive proximity and social connectedness between them (Autant-Bernard et al., 2007; Boschma, 2005; Breschi & Lissoni, 2003). This is because joint research is often developed by co-authors who share the same or related knowledge, but it may also represent certain underlying social connections in that many collaborations stem from familiarity between scientists. Autant-Bernard et al. (2007) suggest that the probability of collaboration is more influenced by social distance than by geographical distance. A collaboration network can, therefore, be a substitute for geographical proximity in facilitating interpersonal interactions and therefore knowledge spillovers. In this vein, Breschi and Lissoni (2003, p. 6) point out that a ‘community of scientists or practitioners exchange “tacit” knowledge even from a long spatial distance’. That is, knowledge spillovers may not necessarily be bound in space and, therefore, need not be a consequence of agglomeration. Waltman et al. (2011) show that the average collaboration distance per joint publication has increased dramatically over the years, partly because technological advancements made it possible to ‘codify and share knowledge across large distances’ (Morescalchi, Pammolli, Penner, Petersen, & Riccaboni, 2015, p. 652).

While recent research has shown that long-distance collaborations improve innovation outcomes (Frenken et al., 2010; Nomaler et al., 2013; Petruzzelli, 2011), there is still only a limited understanding of the extent to which they facilitate technological progress in peripheral regions. This paper seeks to fill this gap in the context of nanotechnology development in China – a country that has been striving to transform itself from an investment-driven to an innovation-driven economy. We focus on the significance of formal collaboration ties in nanotechnology research among scientists based in different Chinese regions for generating spillovers in, in particular, technologically lagging Chinese regions. China presents an interesting context in which to explore the questions addressed in this paper because of its vast regional imbalances, with knowledge creation and innovation activities highly concentrated in the coastal regions.¹ As for the focus on nanotechnology, which represents a set of science-based-enabling technologies that are still in the early stages of their technological life cycles, it is widely seen to create significant long-term pay offs to countries engaging in its development and commercialization and one in which China has made dramatic progress (Niosi & Reid, 2007). Although state support has been an essential feature of nanotechnology development in China, with the bulk of state funding targeting advanced regions, the present data reveal that peripheral regions of China have been making rapid progress in nanotechnology innovations in recent years. We further observe from the data that scientific publications from the latter regions increasingly involve collaborators from other regions, lending weight to the prediction that such collaborations might be a major potential contributor to the dynamism of these regions.

The empirical analysis is carried out on a panel data set of Chinese regions over a 10-year period. The results indicate that collaboration-driven spillovers were a significant contributor to nanotechnology patenting in peripheral regions. The effect of geographical proximity, as well as funding, was limited mainly to core regions that lie along the eastern cost of China.

The paper is structured as follows. The following section provides the theoretical and empirical background of the study and raises the specific questions for empirical scrutiny. The third section presents the data and explains the methods. The fourth section discusses the trends in and patterns of nanotechnology development in China with a regional focus. The results of the econometric analysis are discussed in the fifth section. The final section summarizes the paper’s contribution to the economic geography and catch-up literatures, discusses some policy implications and identifies some avenues for future research.

BACKGROUND AND RESEARCH QUESTIONS

DATA AND VARIABLES

For the econometric analysis, this study uses a panel data set of 30 Chinese regions spanning 11 years (2000–10).² The dependent variable is measured as the count of patent applications (e.g., Basberg, 1987; Griliches, 1990) filed by inventors from a Chinese region at China’s State Intellectual Property Office (SIPO); it is meant to capture a region’s nanotechnology output. The SIPO has three types of patent categories: utility models, industrial design and invention patents. As noted by Li and Pai (2010), the technical threshold for utility models and industrial design is very low and patent filings in these two types do not necessarily represent innovation capacity. Hence, we exclude these two patent categories and consider only invention patent in calculating regional patent output in nanotechnology.

We gathered over 30,000 nano-patent applications from the China Patents Full-text Database.³ In deriving the number of patent applications by a region, we adopted the full counting method. Each patent is counted as 1 for a region if there is any organization from this region appearing in the inventor address of this patent. Accordingly, for example, a patent involving co-inventors from two different regions will be counted as one patent in each region. Data on nano-funding at regional level are collected from the National Natural Science Foundation of China (Statistical Report, various years); expenditure on general R&D is derived from China Statistical Yearbook on Science and Technology (various issues); and data on gross domestic product (GDP) and GDP per capita are collected from China Statistical Yearbook (various issues).

In our framework, interregional spillovers can stem from two sources: nanotechnology-related knowledge of regions measured by patents in this technological area; and the funding for nanotechnology research received by regions. Prior research has shown that R&D expenditure, used as a proxy for innovation activities, carried out in one location can generate spillovers in others through a variety of channels, such as proximity (Funke & Niebuhr, 2005; Wang, Meijers, et al., 2017) and collaboration networks (Fritsch & Franke, 2004; Jaffe, 1989). From this perspective, we suggest that funding, which represents subsidies provided for augmenting R&D expenditure on nanotechnology, can generate spillovers of knowledge to other locations.

We focus on spillovers via both geographical proximity and formal collaboration networks. To capture the former, this study uses the geographical proximity between regions to construct a set of spillover variables. The spatial proximity weight to capture spillovers from region j to region i can be expressed in three different ways depending on the different underlying assumptions (see also Ertur, Le Gallo, & Baumont, 2006; Ponds et al., 2010; Vinciguerra, Frenken, Hoekman, & Van Oort, 2011; Wang, Meijers, et al., 2017). Compared with the two other types of proximity weights (see Note A1 in Appendix A in the supplemental data online), the column standardization of the weight matrixes is considered more appropriate (Ponds et al., 2010). Following the existing literature (e.g., Ertur et al., 2006; Vinciguerra et al., 2011; Wang, Meijers, et al., 2017), this study adopts spatial proximity weights constructed through column standardization:(1) where is the geographical distance between regions i and j. Equation (1) captures knowledge flows from region j to region i with the assumption that spillovers from j to i may be affected by spillovers from j to regions other than i. In particular, if region j is geographically closer to the average region than to region i, there will be fewer spillovers flowing to region i compared with that to the average region.

The knowledge spillovers and funding spillovers that region i receives from other regions through the effect of geographical proximity can be defined respectively as:(2) (3)

where is the number of nanotechnology-related patents in region j in year t; and is the nano-funding received by region j in year t.

In addition to the distance weight above, we also construct weights based on a dynamic interregional collaboration matrix as follows:⁴where an element is the number of co-authored nano-publications involving regions j and i in year t. The knowledge spillovers (TECHSPILL_it) and nano-funding spillovers (FUNDSPILL_it) that region i receives from all other regions are defined respectively as:(4) (5) where is an element of the collaboration matrix defined above and is the number of publications involving scientists from both regions i and j in year t; and is the total number of publications of region j in year t (the column sum in the matrix above).⁵ The scientific collaboration weight is meant to capture the extent to which scientists of a region are engaged with those from other regions. It suggests that when a certain region accounts for a higher proportion of co-authors in the publications of another region, its scientists are engaged in more intense interactions with those from the latter compared with scientists from other regions. A higher degree of interaction with a region in turn implies access to a larger pool of knowledge elements, as well as increased familiarity between scientists that smoothes the process of knowledge exchange (Cohen & Levinthal, 1989, 1990), thereby raising the chances of benefitting more from knowledge spillovers. A collaboration network is particularly salient as a mechanism of knowledge flows in the context of nanotechnology as it is a science-based technology whose diffusion is known to be less constrained by geographical distance (Herstad, Aslesen, & Ebersberger, 2014).

Towards constructing the publication weights (, ) in the above two equations, over 164,000 nanotechnology-related publications from Thomson Reuters’ Web of Science (WoS) were collected. The database is constructed based on an evolutionary lexical query searching and defining strategy developed by the Georgia Institute of Technology (for more details, see Porter, Youtie, Shapira, & Schoeneck, 2008; Wang & Notten, 2010). Next, an international collaboration intensity variable is constructed. This variable captures the effect of potential knowledge spillovers resulting from collaboration with foreign countries, which the literature highlights as an important channel of spillovers (e.g., Lee & Lim, 2001; Wang & Wang, 2017; Wang, Wang, & Philipsen, 2017):(6) where represents the international collaboration intensity in nanotechnology-related publications of region i in year t; is the number of co-authored nanotechnology-related publications involving region i and the foreign country k in year t; and is the total number of nanotechnology-related publications stemming from region i. Each of the 27 foreign countries has had at least 10 papers co-authored with an author based in China during the period of analysis.

As control variables we include regional high-tech R&D personnel (RDH, the number of R&D personnel in a high-tech sector in a region) and graduate share (GRAD) in all models. We use high-tech R&D personnel (Guan & Yam, 2015) as a proxy for the innovation input of a region, given that no precise measure of R&D activities specific to nanotechnology is available. As nanotechnology is a cross-cutting technology that spans a variety of high-tech industries, this measure can be considered as a good proxy for R&D capabilities of a region in nanotechnology research. Nevertheless, the fact that this variable is an imperfect measure, as it may encompass R&D inputs of other sectors, suggests the need for caution when interpreting the results. Graduate share refers to the share of postgraduates (with a master’s degree or a doctorate) in the population in a region, and has often been employed in prior research to capture the endowment of a region’s human capital (Frenz & Ietto-Gillies, 2009; Lau & Lo, 2015; Wang & Yao, 2003). This variable is highly correlated with regional per capita income, pointing to its strong association with regional economic prosperity.

The empirical model is defined as follows:(7) where T_it is the patent applications filed by inventors of region i at the SIPO in time t. In the regression analysis, we sequentially include the key variables and avoid including correlated variables in the same model.

The extent of government funding and interregional collaborations are likely to be influenced by the pattern of innovative capabilities across regions, so the spillover and funding variables may display a certain degree of endogeneity. To tackle this, we lag these variables by one year in the estimation models. In the robustness section, we discuss the results based on more elaborate models that further address concerns of potential endogeneity problems in the model.

TRENDS IN AND PATTERNS OF NANOTECHNOLOGY DEVELOPMENT IN CHINA

This section sets the stage for the econometric analysis conducted below in the fifth section. It discusses the growth in nanotechnology and nanoscience in China, highlighting the major differences between core and peripheral regions with respect to capabilities, state funding for nanotechnology research, publication and patent output, and the extent of collaboration with scientists in other regions and other countries.

In identifying core and peripheral regions, one key consideration is local capabilities that are regarded as an important factor in absorbing knowledge spillovers. As absorptive capacity is ‘largely a function of the level of prior related knowledge’ (Cohen & Levinthal, 1990, p. 128) and as nanotechnology is well recognized as a science-based technology, this study uses the nanotechnology-related scientific knowledge to capture local absorptive capacity. In particular, it conceptualizes the number of nano-publications of a region as representing the basic nanotechnology-related knowledge of that region. Owing to China’s well-recognized interregional economic and technological disparities (Lin et al., 2003; Wang & Szirmai, 2013), Chinese regions are often classified into different groups. Most earlier studies tend to have taken a dichotomous approach, distinguishing between high- and low-capability regions, as respectively those located along the East coast and the rest. While it is well accepted that the high-capability regions indeed tend to form a strong core group along the eastern coast (e.g., Crescenzi et al., 2012), to categorize the rest of the regions as one group is arguable, given that the far-western regions possess very different economic characteristics than those located in the middle. Therefore, to capture better the heterogeneity among Chinese regions, we classify the 30 Chinese regions into three groups: high-, medium- and low-capability regions (see Note A2 in Appendix A in the supplemental data online), based on the level of their nano-related scientific knowledge measured in terms of the total number of scientific publications during the period of study.⁷

RESULTS OF THE ECONOMETRIC ANALYSIS

Summary statistics and correlation matrices for the total sample and the samples for the high-, medium- and low-capability regions are reported in Table A2 in Appendix A in the supplemental data online. The averages of key variables display substantial differences across regions. The average nano-funding in high-capability regions was three times as high as in medium-capability regions, and almost 28 times as high as in low-capability regions (13.08 million yuan for high-capability regions versus 3.68 million yuan for medium-capability regions and 0.48 million yuan for low-capability regions).⁸ Similarly, average R&D personnel in high-capability regions was close to 17 and four times that in low- and medium-capability regions respectively. Differences in the share of graduates in the total population between the three sets of regions also display a similar pattern; however, the extent of regional differences is slightly lower with high-capability regions having about 10 and three times as high the share of graduates compared with low- and medium-capability regions respectively.

As the dependent variable is the number of nanotechnology patents, a count data model such as negative binomial or Poisson is more appropriate than ordinary least squares (OLS). As to the choice between these two models, as Chinese regions exhibit wide variations in patenting, the critical assumption of the equality of mean and variance of the Poisson model does not hold, leading to a preference for the negative binomial model. Given the somewhat short-panel nature of the data set, fixed effects may not produce consistent estimates, so we adopt the population-averaged (PA) model with semi-robust standard errors (Cameron & Trivedi, 2010). As a robustness check, the random effect (RE) model is also employed; the results are similar to those from the PA model and are available in Tables B1 and B2 in Appendix B in the supplemental data online.

Regression results are documented in Tables 2 and 3. All models include the full set of control variables, as well as year dummies to account for unobserved annual events that may affect patenting in all regions. The correlation coefficient between Nanotech spillovers and Funding spillovers (via both collaboration and proximity) is rather high, so these two variables are employed in separate regression models. Furthermore, to illustrate the consistency of the results on spillovers to the exclusion of funding, and vice versa, separate equations – that include only spillovers, only funding and both – are estimated.

Two sets of analysis are carried out: for the full sample and for the three regional samples. Table 2 presents results based on the full sample with different combinations of the key explanatory variables. Table 3 provides separate results for each of the three regional categories (high-, medium- and low-capability regions).

In order to examine the statistical fit of models with the full set of variables compared with the more parsimonious models, we carried out likelihood ratio (LR) tests. These tests show that combining funding and nanotechnology collaboration spillover variables (model 4, Table 2) improves model fit compared with models with only one of the two variables included (models 1 and 2, Table 2) at a significance level of 1%. Similarly, the model that combines funding with collaboration spillovers resulting from funding (model 8, Table 2) also improves the model fit at a 1% level compared with models with only one of these variables included (models 1 and 6, Table 2). However, adding spatial spillover variables to funding does not statistically improve model fit at the 5% level (model 9 compared with model 1, Table 2), in partial agreement with the suggestion that physical proximity between regions is perhaps of lower importance in effecting interregional knowledge flows in a large country such as China. We provide a more nuanced interpretation of the effect of spatial proximity below in the discussion of the results.

DISCUSSION AND CONCLUSIONS

Employing nanotechnology development in China as the research setting, this paper examined the importance of knowledge spillovers for innovation, particularly for peripheral regions. While prior research extensively examined interregional spillovers resulting from geographical proximity, we focused also on spillovers emanating from collaboration network of scientists spanning regions.

The evidence presented in this paper suggests that China’s success in nanotechnology development in general owes in large part the fostering of indigenous scientific capabilities through strong financial support from the state. As to be expected, the bulk of the governmental support targeted the advanced regions of China and the effect of funding had a greater impact in these regions given their ability to ‘generate greater bang for the buck’.

Geographical proximity with other similarly advanced regions tended to display some significant positive spillover effects in high-capability regions which are concentrated along China’s eastern coast, and to a lower extent in medium-capability regions that occupy the East–Central part of China. This confirms the general understanding in the economic geography literature (e.g., Caniëls & Verspagen, 2001) that proximity among advanced regions might generate agglomeration economies in these regions, which, while they help perpetuate their success, may leave peripheral regions behind. Specific to the Chinese context, the limited proximity spillovers received by medium- or low-capability regions may also be seen in the light of the fact that these regions are mostly inland, western regions far from the advanced coastal regions. This finding may have implications beyond the development process associated with a science-based enabling technology such as nanotechnology. Technological context is a critical factor in generating proximity-driven spillovers (Frenken et al., 2010), with studies suggesting that spillovers tend to occur more easily in high-tech sectors (Autant-Bernard, Fadairo, & Massard, 2013). This may suggest that proximity might induce even more limited spillovers of other, less cutting-edge technologies associated, for example, with the low- and medium-technology industries that dominate Chinese manufacturing.

While the above finding may point to a less-than-optimistic prospect for the catching up of lagging regions, a more reassuring picture emerges from the findings on the role of spillovers via the collaboration networks of scientists. Spillovers transmitted through such networks turn out to be the key source of nanotechnology innovation in peripheral regions. Combined with the evidence that direct R&D funding from the state contributed very little in these regions of China, we can conclude that spillovers via collaborations may have been the single most important factor in the rapid progress made in nanotechnology development by peripheral regions in recent years.

The findings of the study have several theoretical implications. First, the economic geography literature recognizes that owing to the presence of formal long-distance channels of knowledge such as collaborations, geographical proximity is not a necessary condition for knowledge spillovers (e.g., Boschma, 2005); the present study contributes to this literature by demonstrating that such formal collaboration linkages can be more salient for lagging regions with limited own capabilities. Second, this study, with its focus on an emerging-economy setting, also contributes to the catch-up literature: it highlights that, on the one hand, targeted technology development programmes may indeed help spur the development of a new technology system in leading regions; and, on the other hand, scientific linkages with these regions can help peripheral regions make rapid technological progress, compensating for their limited scientific capabilities.

This study raises some policy implications as well, particularly in the context of technology development in China and other similar emerging economies. Unlike in Europe, where the European Union’s research funding is geared to promote collaboration between European countries (Hoekman, Scherngell, Frenken, & Tijssen, 2013), the Chinese government’s funding strategy is devoid of any serious measures to stimulate interregional collaborations. So far, the emphasis in the government’s funding guidelines has been limited to either international collaborations or industry–university collaborations. It may very well be the case that funding also continues to flow in the near future more into leading regions where scientific capabilities are stronger. While this is unavoidable on efficiency grounds, it appears imperative that peripheral regions can leverage and expand their scientists’ ties in the broader scientific network within China and benefit from knowledge spillovers. Therefore, the Chinese government’s funding strategy may need to incorporate measures to ensure scientific cooperation that spans regions. Such a strategy would be a natural extension to the current emphasis on university–industry collaboration, which, although it has not been explicitly examined in this paper, is well known to generate knowledge spillovers. As evidenced by this study, cooperation among scientists and technologists can potentially substitute for the relatively weak capabilities of some regions. Nurturing and expanding such networks through the right policies and incentives can play a vital role in helping today’s technologically lagging regions catch up in the production and use of new technologies, thereby ensuring a more even pattern of regional development.

Given the heterogeneity between regions and countries in their propensity to collaborate, future research may explore these questions in other geographical contexts, such as Europe where the geographical, institutional and social conditions are more variegated than in China. This can help answer, in particular, whether the limited impact of geographical proximity in generating spillovers in peripheral regions is due to the peculiar geography of China. In addition, it would be important to understand the extent to which different types of networks act as a mechanism of long-distance knowledge spillovers to peripheral regions in the context of other technologies. This is because for a science-based technology such as nanotechnology, the fairly developed scientific collaboration networks present in China, as well as in other similar emerging economies, are particularly suitable for effecting knowledge flows. However, in the context of other technologies, such as biotechnology, that are characterized by a more complex innovation process with a varied set of actors, such as start-ups, university research laboratories and established pharmaceutical companies, the forces of agglomeration (Zucker et al., 1998) are likely to be harder to overcome. This presents policy-makers in peripheral regions with potentially daunting challenges, such as helping heterogeneous sets of actors build and strengthen linkages with actors in more advanced regions. Understanding this process, particularly in emerging economy settings, presents an important opportunity for future research in economic geography.