Regional innovation system and regional innovation capacity:
an empirical study in the case of China in 1998-2007
Qun, Wang s2179121
Master Thesis for International Economics & Business (2012/2013) Final version, July 2013
Abstract
This paper employs panel regression analysis to explore the determinants of innovation performance between Chinese regions by drawing on regional innovation system (RIS) literature. The results indicate that enterprise R&D, research institutions R&D and the interaction between various R&D performers in RIS are significant determinants of innovation capacity (measured in invention patent applications per million people). Further analysis is also conducted in coastal and inland regions, which gives policy implications in terms of improving regional innovation capacity.
Key words: regional innovation system (RIS), invention patent application, R&D performers, interaction framework of RIS
Supervisor: Bart Los Co-assessor: Ger Lanjcuw Phone: +31645825522
1
Table of Context
1. Introduction ... 2
2. Theoretical background and Hypotheses ... 5
2.1 Regions as innovation systems ... 5
2.2 The concept and framework of RIS ... 6
3. Model and data analysis ... 10
3.1 Sample ... 10
3.2 Dependent Variable ... 11
3.3 Independent variables ... 12
3.4 Control Variables ... 14
3.5 The econometric model ... 17
3.6 Descriptive statistics ... 17
4. Diagnostic checks ... 19
4.1 Fixed and random effect estimation ... 19
2
1. Introduction
Since economic reform and open door policy started in 1978, China has witnessed spectacular economic growth for three decades. The growth, however, is mainly driven by export-oriented, labor intensive manufacturing activities (Young, 1995). In 2008, the total value of exports accounted for approximately 32% of the GDP and tens of millions of workers were engaged in the export sector (Source: China Statistical Yearbook, 2009). From a long run sustainability perspective, the policymakers in China have increasingly realized that the driving force of growth should be transformed into a knowledge-intensive way by emphasizing on the capacity of innovation. In this quest technological innovation is particularly important. Thus far China has primarily served as a manufacturer for western designs. If they can shift from western to indigenous Chinese designs, China can slowly move toward a modern, higher-tech, consumer-driven economy. The year 2006 witnessed an announcement by the State Council, People’s Republic of China as “Medium-to-long Term Plan of National Science and Technology (S&T) Development (2006-2020)”.
3 estimation results.
Figure 1: Number of invention patents application in regions/millions of people
Note: The figures in the table represent the number of patents per millions of inhabitants applied by the
first 10 and last 5 regions ranked (30 regions in total in dataset) based on the years 1998 and 2007.
Source: State Intellectual Property Office (SIPO), 2008
The variety among provinces results from many aspects. The huge geographical area of China is the root of regional dissimilarities. Individual regions have their own characteristics in social, local institutional and economic environment, directing each province to develop their own capabilities to undertake R&D activities and channel it into innovation process (Sun, 2000; Yueh, 2006). On the other hand, the uneven regional development preferred coastal provinces and created a prosperous climate for innovation activities under the support of establishing a robust regional innovation system (RIS) in S&T policies since the open door policy (Fabre and Grumach, 2012).
4
specific characteristics, finding the most appropriate components of RIS (Liu and White, 2001b; Gu and Lundvall, 2006; Li, 2009), or analyzing the diversity of RIS (Chung, 2002). The research conducted by SIPO shows that there is a clear gap between regions when it comes to their regional innovation capacity. Innovation capacity refers to the ability of a country or a region to produce or commercialize a flow of innovation technology over the long term (Stern, et al, 2000). Moreover, Liu and White (2001b) also point out that regional innovation capacity is related to RIS that consists of sufficient R&D expenditure, strong interaction within the organizations in the system, as well as foreign direct investment and other attractive technological infrastructure and human resources in the region. In particular, the crucial roles of different organizations (enterprises, universities and research institutions) which mainly contribute R&D expenditures are highlighted. They support regional innovation activities by providing an innovation platform for new technology creation, fresh knowledge generation, usage, and thus increase the number of patents in RIS (Etzkowitz and Leydesdorff, 1997; Dorloreux and Dionne, 2008).
Given that innovation is critical to sustainable growth in regions, the analysis of regional innovation system becomes prevailed in less developed regions within these years (Doloreux and Dionne, 2008). In China, the underlying determinants of innovation capacity based on the innovation system at the regional level remain relatively rare (Li, 2009). This leads to the research aim of our thesis that explores the framework of RIS on innovation performance through emphasizing the main R&D organizations (enterprises, universities and research institutions) and their network inside it. For this purpose, the main research question is as follows:
“What are the underlying determinants of regional innovation capacity (measured by
patents) in Chinese regional innovation system (RIS)?”
5
Policy under way aims at reducing the gap between regions and establishing a well- functioning regional innovation system by primarily giving more support to innovation of enterprises (State Council, 2006). Thus it inspires me to conduct further analysis to explore whether or not boosting enterprise R&D is the best mechanism to foster regional innovation capacity both in inland and coastal regions.
In the following section, we discuss the theoretical background and raise hypotheses. Section 3 presents the empirical model and data analysis. The Diagnostic checks to deal with dataset are considered in section 4. The estimation results are reported in section 5. In section 6 and 7, I assess the further analysis in the case of coastal and inland regions and present the limitations in the thesis, respectively, followed by the conclusion in the last section.
2. Theoretical background and Hypotheses
In order to compare innovation performance across regions, the unit of analysis should be discussed firstly. Thus, this section will provide the concept and theoretical explanations of regional innovation system to use as a guide for my empirical analysis.
2.1 Regions as innovation systems
6
China, as one of the largest developing economies, has experienced a dramatic change in its innovation system under the transitional context (Chen and Guan, 2011, 2012). From the work of Liu and Lundin (2008), the government functioned as the key coordinator in the innovation system and government investment dominated in performing innovation activities before the economic reform in the early 1990s. However, after the open door policy, the innovation system had been transformed from centrally-planned to market-oriented; the provincial government has been the key executor of facilitating innovation activities and coordinating system elements within a region. In other words, the local government was authorized to take responsibility to create a better regional system of innovation (Chen and Guan, 2011).
So far, we can conclude that the regional dimension should not be overlooked in China since China’s NIS is too large and complicated to be summarized in a single model (OECD, 2008). Furthermore, there are 30 administratively and economically independent provincial-level regions and the innovation process happens more often within, rather than between provincial-level regions (Li, 2009; Chen and Guan, 2011). In this sense, it is more advisable and appropriate for us to take provinces as the unit of analysis and to formulate hypotheses and evaluate the regional variation in innovation capacity.
2.2 The concept and framework of RIS
7
A number of studies have also demonstrated that components and relations between components are commonly formed together to shape a coherent system (Carlsson et al., 2002; Edquist, 2005). There are two main types of components in an innovation system. One component is formally structured and deliberately created organizations: research institutions, enterprises, universities, government agencies and financial institutions responsible for innovation financing in regions (Li, 2009). They are all major innovation performers or actors in spite of their quite different purposes. The other component is some important institutions, as similarly proposed by Doloreux (2004), that include patent laws, tax laws, industrial policies and environmental and safety regulations that potentially influence innovation activities. In the context of the piece of Sigurdson (2004), he suggests that the institutions belonging to the latter component are normally set at the national level and remain constant between regions. The idea of non-institutional differences in regional comparison is consistent with the works of Liu and White (2001a). They follow Sigurdson’s point and go further to propose a general framework for analyzing innovation system, which is anchored around fundamental activities (e.g R&D, implementation, end-use), and focus on interaction effects of system’s structure and dynamics. Correspondingly, Nelson (1993) concludes that the build-up of innovation capacity is mainly contributed by four main various organizations, namely governments agencies, research institutions, enterprises and universities, with their innovation activities jointly in a systematic way under institutions and policies.
Li (2009) and Fritsch and Slavtchev (2007) separate themselves from the other papers by investigating China’s innovation system through which the cross-regional variation in innovation efficiency performance. In both papers, the innovation performance is closely related to technical efficiency, which reflects a region’s capacity of transforming innovation investment into innovation outputs. By realizing the significant impact of components of RISs on efficiency, they evaluate at the regional level in terms of the determinants of innovation efficiency and the degree to which contribution offered by these determinants (participants) in innovation process leads to innovation efficiency.
8
innovation capacity or efficiency of innovation system. Nonetheless, the separate effects of the participation of determinants (e.g. different R&D performers) in RIS are still blurred. Because earlier analyses only focus on the combined effects of them, e.g. the role of URI (universities and research institutions) network (Chen and Kenny, 2007). Hence, we are interested in exploring the individual role of components in RIS.
In consistent with the previous studies of RIS, Li (2009) formulates a conceptual framework with the following factors considered as the most critical ones in determining the realized level of innovation potential:
(1) The innovation effort contributed by three major innovation performers consisting of research institutions, universities and enterprises.
(2) The cooperation and interaction between industry-university-institutions in the innovation system during the innovation process.
(3) The support from government-owned agencies as well as financial institutions responsible for innovation financing in regions (hard to measure because of the difficulties in capturing government behavior under the regional context and in defining financial institutions activities under-developed financial environment, respectively). (4) The region-specific characteristics and innovation environment.
In the innovation system of most advanced regions, firms are considered as the primary locus of innovation and other types of organizations are treated as peripheral elements (Acs et al, 2002; Fritsch & Slavtchev, 2011). The case is different in China where organizations other than firms also have a critical role in R&D performance (Liu and White, 2001b). By relying on the estimations in previous studies focusing on China’s RIS and the framework above as well, research institutions, firms, and universities are directly involved in the process of knowledge generation, serving as carriers and producers of fresh knowledge and innovation output. As a matter of fact, considerable differences existing in organizational objectives, structures and operating practices, the effect of each type of performer undertakes innovation related activities and engages in technological learning can vary to a large extent (Mowery and Sampat, 2004).
9
policies for stimulating the industrial and technological development of Chinese enterprises. When time passes, the behavior of enterprises engaging in R&D activities in each province has become a popular phenomenon (Liu and Lundin, 2008). As a result, the average regional enterprise R&D to total regional GDP was 0.612 in 2007, which was approximately 3 times higher than that in 1998 with 0.230 (Author’s own calculation. Source: China Statistical Yearbook of Science and Technology, 1999-2008). As predicted by Li (2009), enterprise R&D investment can be expected to take the leading position in Chinese regional innovation system in the future. Also, firms can be expected to become the dominant innovator in the future under the support of recent S&T program (State Council, 2006). So we suspect that a region with higher enterprise R&D intensity is associated with a better performance in innovation capacity.
Hypothesis 1: The higher enterprise R&D intensity inspires a better regional innovation capacity measured by the number of patents per million people.
Research institutions. As indicated by previous section, Liu and Lundin (2008) states that
research institution is the main contributor in the early years in innovation performance. Institutions under the support of government still make a great effort in providing key knowledge and skills and play crucial roles in creation, development, accumulation, transfer and utilization of technologies. For example, the institution clusters locate in Beijing, Guangdong, and Shenzhen (Chen and Kenny, 2007; Zhu and Tann, 2007). Due to the fact that Chinese RISs have undergone the transition from institution-dominated to enterprise-dominated, its contribution to R&D expenditure decreased gradually with the time passed. However, we still expect that it has a significant positive impact on regional innovation capacity.
Hypothesis 2: The higher research institution R&D intensity inspires a better regional innovation capacity measured by the number of patents per million people.
Universities. It is widely known that universities R&D expenditure mainly focuses on
10
2013). In this sense, it is less likely to be application-oriented, not supporting appropriate monopolistic profits by applying for patents. However, some argue that among the actors in innovation system, universities’ contribution to industrial innovation is critical (Mowery & Sampat, 2004). I examine its role in regional innovation capacity and propose it would have a significant role in influencing regional innovation performance.
Hypothesis 3: The universities R&D intensity has a significant influence on regional innovation capacity measured by the number of patents per million people.
Interaction. In innovation system, the interaction among various R&D organizations in the
innovation system is very common and provides a prosperous climate for them to conduct interactive learning, exchange ideas and cooperate with each in the innovation process (Etzkowitz and Leytesdorff, 1997; Lundvall et al., 2002). Cook et al (1998) argue that a closer relation between enterprises and research institutions/universities will shape a systematical network in conducting and boosting innovation process. The viewpoint is also supported by Liu and White (2001b), who suggest that public research institutions act as a linkage with each R&D engagers in China under command era. For instance, university graduates will be assigned in R&D institutions, industries and other state-related organizations under the support of Labor Bureau. To establish the linkages among these three R&D actors in innovation process, I use the ratio of universities and research institutions R&D financed by enterprises as a proxy for their interaction. I expect that its sign is positively associate with innovation performance in the model.
Hypothesis 4: The stronger integration of R&D performers, illustrated as tri-relationship between research institution-university-enterprise, will enhance the performance on regional innovation capacity measured by the number of patents per million people.
3. Model and data analysis
3.1 Sample
11
to 2007. All data are mainly drawn from various issues of China Statistical Yearbook (CSY,
1999-2008) and China Statistical Yearbook of Science and Technology (CSYST, 1999-2008), State Intellectual Property Office (SIPO, 2008) and China Internet Network Information Center (CNNIC, 2008) as well. Tibet is excluded in the analysis because most of the
relevant data for it are not available or missing in the observed time period. One reason we choose the starting year is that the data for Chongqing city was not available until 1998. More importantly, China’s R&D activities began to surge in the late 1990s. Therefore, the study based on these ten years can provide insightful policy implications for the National Science and Technology Development Plan (2006-2020). Generally speaking, the exact 30 provinces are as follows and can be grouped according to their relative locations in China: Coastal
Area
12 provinces: Beijing, Tianjin, Hebei, Liaoning, Shanghai, Jiangsu, Zhejiang,
Fujian, Shandong, Guangdong, Guangxi, Hainan Central
Area
9 provinces: Shanxi, Inner, Mongolia, Jilin, Heilongjiang, Anhui, Jiangxi,
Henan, Hubei, Hunan Western
Area
9 provinces: Chongqing, Sichuan, Guizhou, Yunnan, Shaanxi, Gansu,
Qinghai, Ningxia, Xinjiang,
3.2 Dependent Variable
12
According to China’s patent law issued in 1 April 1985, generally speaking, the technology to be patented must pass four tests, which requires the patents to be novel, useful, non-obvious and man-made. There are three types of patents in China: invention, utility and design patents. Especially, the invention patents are more rigorously scrutinized for the characteristics of novelty and non-obviousness while the other two types are only subject to preliminary examination. What’s more, only invention patents can receive the extent of protection for 20 years, which is consistent with the global standard. For the purpose of analyzing Chinese true regional innovation capacities, we consider to use invention patent application since they can better represent high quality ideas and yield high quality data in the analysis. Hereafter, the term patent refers to invention patent application. The initial data for the number of invention patent application are divided by millions of people per province. By applying the transformation to the observations, the dependent variable can eliminate the bias from provincial differences in size.
Invention Patents (Ipatentit): the number of invention patent application per million
people in province i at time t, computed by dividing the total number of invention patent applications by the provincial population for all provinces. The formula is as follows:
𝐈𝐩𝐚𝐭𝐞𝐧𝐭𝒊𝒕= 𝑷𝒂𝒕𝒆𝒏𝒕𝒊𝒕/𝑷𝒐𝒑𝒊𝒕
Ipatentit, the dependent variable, is the number of invention patent per million people
applied by province i at the observed time t (piece). 𝑷𝒐𝒑𝒊𝒕 represents the total population
in province i at time t, measured in millions.
The complete invention patent application data are available in the official database distributed by the State Intellectual Property Office (SIPO): CNPAT ACCESS, which is in English.
3.3 Independent variables
13
analysis distinguishes between different components of the overall three main performers of R&D (Independent Research Institutions, Large & Medium-sized Enterprises, Institution of Higher Education). The reason is that the impact on innovation in general, and patents in particular, may follow different trends according to the various R&D actors and their interaction in a region, aiming at examining the role of Chinese regional innovation system (RIS) on promoting regional innovation capacity. The relevant panel data are available for 30 provinces and there are no missing values. I use the institution R&D, enterprise R&D and universities R&D as a percentage of GDP (R&D intensity) per province to denote the explanatory variables.
Recent studies have shown that interactions between various R&D organizations (industry-university-institutions) also take a variety of forms as joint R&D projects, technology licensing, internships and other collaborations among them (Kodama and Branscomb, 1999). Due to the difficulty in getting access to many of them, not all forms are captured here. Based on the availability of data in China Statistical Yearbook of Science and Technology, the ratio of universities and research institutions R&D financed by enterprises is used as a proxy for the linkages among them. In this way, it can identify the degree of cooperation between R&D organizations in the regional innovation system.
Enterprise R&D intensity (RIentit) denotes the enterprise R&D expenditure per unit of
gross domestic product in province i at time t. The formula is as follows: RIent𝑖𝑡 =RDentit
GDPit *100%
Research Institutions R&D intensity (RIinsit) denotes the research institutions R&D
expenditure per unit of gross domestic product in province i at time t. The formula is as follows:
RIins𝑖𝑡 =RDins𝑖𝑡
𝐺𝐷𝑃𝑖𝑡 *100%
Higher Education R&D intensity (RIheit) denotes the higher education R&D expenditure
per unit of gross domestic product in province i at time t. The formula is as follows: RIhe𝑖𝑡 =RDhe𝑖𝑡
𝐺𝐷𝑃𝑖𝑡 *100%
14
ratio of universities and research institutions R&D financed by enterprises in province i at time t. The formula is as follows:
URFE𝑖𝑡 =(𝐸𝑅𝐷ℎ𝑒(𝑅𝐷ℎ𝑒𝑖𝑡+𝐸𝑅𝐷𝑖𝑛𝑠𝑖𝑡)
𝑖𝑡+𝑅𝐷𝑖𝑛𝑠𝑖𝑡) *100%
Where 𝐸𝑅𝐷ℎ𝑒𝑖𝑡 and 𝐸𝑅𝐷𝑖𝑛𝑠𝑖𝑡 are the R&D expenditures in total universities R&D and
research institutions R&D financed by enterprises in province i at time t, respectively. The production of knowledge usually takes time; I therefore apply one-year lagged terms of these four R&D variables above as the core input of innovation production.
The data are all collected from China Statistical Yearbook of Science and Technology
(CSYST, 1999-2008). Both numerators and denominators in the formulas above are
expressed in million Yuan at current price.
3.4 Control Variables
In addition, other determinants of innovation performance in a region are also captured of which I make use GDP per capita, technological infrastructure, tertiary education level and foreign direct investment. Adding control variables is an appropriate solution. I can reduce the bias for parameter estimation through their participation in my analysis.
GDP per capita (LnGDPcapita)
15
capita. This problem can be relieved by using one-year lag for GDP per capita.
So in order to capture the differences in Chinese provincial initial states, the model choose the nominal GDP per capita measured in ten thousand Yuan which will be transformed into the natural logarithm (LnGdpcapita) in order to modify extreme values into values closer to normal distribution. Thus, for this reason I choose to use the natural logarithm of the nominal GDP per capita, measured in ten thousand Yuan.
The data are taken from China Statistical Yearbook (CSY, 1999-2008).
Infrastructure and Internet usage (Lninternet)
As indicated by previous studies, e.g. Furman et al (2002), he proposed that the infrastructure in a country or a region can directly affect its innovation capacity. Moreover, he mentions that well-structured regional infrastructure will support the innovation activity within and between different regions in an economy. The infrastructure captures the concept of physical and virtual network of innovation, including main roads and railways and information and communication technology (Torrisi, 2009). In specific, information and communication technology includes indicators which measure communication networks, e.g. Internet usage and telephone lines. This virtual network is treated as an enabler of the information society and belongs to innovation drivers, which illustrates the readiness to utilize technology and adopt new innovation (Cooke et al. 2001). In this analysis, we have adopted the number of Internet network user per million of population (lagged terms) in every region as an indicator of technological infrastructure that can influence a region’s ability to communicate with others and thus innovates.
The data were taken from the 3th-23th Survey Report in China Internet Network
Information Center (CNNIC, 2008) where the net users are defined as the citizens who use
the Internet for at least one hour on average per week.
Human capital and level of skills (LnSKILL)
16
capital is closely associated with the skill levels. As the level of skills is widely adopted as a proxy for human capital and the higher the level of attainment in a society, the greater skills it will achieve, and therefore, the greater probability the region’s capability to perform innovation and potentially affect growth (Pavitt, 1991, 1998). In China, provinces with sufficiently high levels of skill in coastal area have more capability of transferring technology and transferring their own R&D into innovation and economic growth. And the low levels of human capital of provinces in the inland area frustrate their innovation capacity and RIS cannot effectively work when below the threshold-level of human capital (Sorensen, 1999).
In general, those with tertiary background who are better educated will have extensive work experience and high skills, and invest more time, energy, and resources in honing their skills can better contribute to innovation process than those with secondary education. As specified above, I choose to represent the skill level by using the ratio of graduates of tertiary level (measured in millions) in a specific year t to total population (in millions) in each province at year t. One-year lag is used for skill variable.
The data are taken from China Statistical Yearbook (CSY, 1999-2008).
Foreign direct investment (LnFDI) (million $)
17
but also a host region with better innovation capacity (e.g. coastal provinces in China) tends to attract more inward FDI (Fu, 2008).
To capture this aspect on regional innovation capacity, we employ the amount of FDI inflows (measured in million dollars) per million inhabitants. From the dataset, I have some extreme values and thus transform by employing the natural logarithm to make them into values closer to a normal distribution, which is usually appropriate for positive monetary terms. Also, I apply a one-year lag to FDI to relieve the endogeneity problem, which will be specifically discussed in the section 4.5.
The data are taken from China Statistical Yearbook (CSY, 1999-2008).
3.5 The econometric model
LnIpatent𝑖𝑡 = 𝛼 + 𝛾1𝐿𝑛𝑅𝐼𝑖𝑛𝑠𝑡𝑖𝑡−1+ 𝛾2𝐿𝑛𝑅𝐼𝑒𝑛𝑡𝑖𝑡−1+ 𝛾3𝐿𝑛𝑅𝐼ℎ𝑒𝑖𝑡−1+ 𝛾4𝐿𝑛𝑈𝑅𝐹𝐸𝑖𝑡−1
+ 𝛿1𝐿𝑛𝐼𝑛𝑓𝑟𝑎𝑖𝑡−1+ 𝛿2𝐿𝑛𝐺𝐷𝑃𝑐𝑎𝑝𝑖𝑡𝑎𝑖𝑡−1+ 𝛿3𝐿𝑛𝑆𝑘𝑖𝑙𝑙𝑖𝑡−1+ 𝛿4𝐿𝑛𝐹𝐷𝐼𝑖𝑡−1 + 𝜀𝑖𝑡
3.6 Descriptive statistics
Before describing the characteristics of the data, outliers should be firstly addressed. Outliers will be detected based on Cook’s distance. Cook’s distance is a measure that combines the information of residual and leverage of observations. Residual is the difference between the predicted and actual value and leverage is a measure of how far an explanatory deviates from its mean. The accuracy of the regression will be distorted by influential observations with high leverage and large residual (UCLA, 2010). Points with Cook’s distance higher than 1 are to be considered as influential. After running the test, the 300 observations in my dataset have a Cook’s D smaller than 1 so none of them will be outliers. Thus the data keep integrated and give a full image of the variation on regional innovation capacity; I use the 30 provincial data in the estimation result and do not drop any data.
18
Table 2-Descriptive Statistics
Variable Obs. Mean Minimum Maximum Std.Dev Ipatent 300 62.04 3.735 1148.990 141.633 RIent 300 0.414 0.015 1.335 0.265 RIinst 300 0.293 0.026 4.526 0.583 RIhe 300 0.102 0.002 0.704 0.112 URFE 300 13.110 0.187 33.926 0.079 Internet 300 6.643 0.022 45.132 7.627 GDPcapita 300 1.263 0.230 6.560 1.010 FDI 300 0.570 0.001 4.734 0.832 Skill 300 0.362 0.106 1.243 0.242 Source: Author’s own calculation, CSY (1999-2008), CSYST (1999-2008) and SIPO (2008) and CNNIC
(2008).
As can be seen in Table 2, there are obvious variation between the minimum and maximum values for main explanatory variables, indicating a significant variation in the role of RIS and innovation capacity across the regions in China. Beijing (2007) further outperforms other regions with the highest number of invention patent applications (Ipatent), 1148.990 (patents per million person) whereas Anhui (1998) shows the weakest ability in patenting, only 3.735. The average level of Ipatent is only 62.04, underlying the fact that many regions are under the average regional level. In combination with the Figure 1 in the introduction section, I conclude that Beijing has led the innovation capacity and kept increasing during the period of 1998-2007, at the same time, with other regions showing the similar upward tendency.
19
larger effort in enterprise R&D investment than its regional counterparts, 1.355% as a percentage of GDP, acting as the center of finance and business in China. Yet, the lowest R&D intensities for enterprise and university lie in Hainan, 0.015% (2005) and 0.002% (2000) respectively, since Hainan is a typical tourist province and pays little attention in innovation. When examining the interaction between organizations, Zhejiang, the richest province in China, performs best in the network of cooperating and interacting in RIS, with the highest ratio of URFE, 33.926% in 2005. The poorest interaction among R&D actors in the network of RIS locates in Hainan, with a ratio of 0.186% in 2004. All these four main variables indicate the large variation in the R&D activity in RIS across regions.
With respect to control variables, the distinction is remarkable between regions. Beijing further exceeds its counterparts in terms of GDP per capita (6.560 thousand Yuan) and the intensity of Internet users (accounting for 45.132% of population) in the year 2007. At the same year, the Tianjin province is outstanding with its large FDI inflows with 4.734 million dollars per million inhabitants potentially because of its attractiveness in the number of exported companies in manufacturing industry. Beijing demonstrates its virtue in nurturing the largest number of current graduates receiving the tertiary education, with the percentage of 1.243% of its population in 2007. In contrast, the lowest figures are all located in inland regions, either in Qinghai or Guizhou in 1998.
4. Diagnostic checks
In this section, diagnostic checks are conducted in order to check the relationship between variables and the validity of the models. Some important checks are shown as follows.
4.1 Fixed and random effect estimation
20 informative.
With respect to the presence of a panel data in the research setting regression equations, I can estimate it via either fixed or random effects. Both methods can control for the region-specific effect ui that varies across regions but being within a region over time,
otherwise it would be captured by the error term 𝜀𝑖𝑡. However, the fixed effect estimation assumes that the intercept captures all the differences between provinces over time. The random effect estimation assumes the intercept capture the differences within provinces but threat the provincial differences as random. To deal with the unobservable individual effect in a panel model, I treat the individual effect as a random variable and fixed parameter by conducting random effect (RE) and fixed effect (FE) models. The Hausman test is employed to judge which one is appropriate. Through the results of using Stata, I find that the P-value (Prob>Chin2) is equal to 0.000 for all regressions (see Section 5). This means I reject the null hypothesis and conclude that fixed-effects model is preferred over a random-effects model in the analysis.
4.2 Normality
A Jarque-Bera test is usually conducted to test for the normality of distribution for variables by checking values of skewness and kurtosis (Hill et al, 2007). Skewness refers to how symmetric the residuals are around zero whereas kurtosis refers to the “peakness” of the distribution. Another easiest way for testing normality is to create histograms then check whether they are bell-shaped.
21
compared to other regions, thus receive much attention from R&D actors (Chen and Kenny, 2007).
Table 3-The Jarque-Bera test Dependent
Variable
Skewness Kurtosis Jarque-Bera P-value Ipatent 4.350 24.712 683.710 0.000 LnRIinst -0.366 2.749 7.477 0.024 LnRIent -1.850 7.642 44.373 0.000 LnRIhe -0.640 3.362 22.146 0.000 LnURFE -1.411 6.138 222.666 0.000 LnInternet -0.926 3.250 43.618 0.000 LnGDPcapita 0.509 2.856 13.199 0.001 LnFDI -0.090 2.350 5.680 0.058 LnSKILL 0.340 2.020 17.771 0.000
22 Figure3: The histogram of the distribution of LnIpatent
A possible solution is to transform the dependent variable into natural logarithm and thus the more extreme variable will be closer to a normal distribution. When including the histogram of LnIpatent (Figure 3), for example, I conclude that the distribution of this variable is closer to the bell-shape than before (Figure 2).
This natural logarithms of other explanatory variables are also utilized in my model, which will modify values of explanatory variables closer to the bell-shape than before as well.
4.3 Multi-collinearity
Multi-collinearity occurs when two or more explanatory variables have a linear relationship (Hill et al, 2007). The higher the degree of multi-collinearity will cause the higher possibility of the instability of the regression model estimates of coefficients. Another concern is the separate effects of independent variables are blurred and thus the least squares estimates are biased.
Table 4-The correlation matrix
Patent RIinst RIent RIhe URFE FDI Gdpcapita Skill Internet
23 RIinst 0.510 1.000 RIent 0.555 0.230 1.000 RIhe 0.242 0.547 0.483 1.000 URFE 0.198 -0.042 0.375 0.126 1.000 FDI 0.553 0.263 0.291 0.187 0.369 1.000 Gdpcapita 0.764 0.312 0.720 0.510 0.437 0.573 1.000 Skill 0.403 0.370 0.225 0.482 0.204 0.633 0.309 1.000 Internet 0.732 0.376 0.124 0.057 0.209 0.168 0.428 0.530 1.000
Source: CSY (1999-2008), CSYST (1999-2008) and SIPO (2008) and CNNIC (2008).
One possible way to detect the problem is the correlation matrix between each pair of variables would be estimated. If the values are lower than -0.8 or higher than 0.8, multi-collinearity does exist (Hill et al, 2008). The correlation matrix is shown in Table 4. It shows that some pairs of correlation are relatively high among each pair of explanatory variables, for example, the presence of high correlation (0.720) between the RIent and GDP per capita. However, the value of correlation is below 0.8 and conforms to the rule of thumb,therefore not worrying about the problem of multi-collinearity.
Table 5-The VIF table
Variable VIF RIinst 5.07 RIent 2.63 RIhe 6.82 URFE 1.57 FDI 6.47 Gdpcapita 8.89 Skill 4.31 Internet 3.86 Mean VIF 4.95
24
4.4 Heteroskedasticity
One potential problem in the model is the existence of heteroskedasticity occurred by the difference of the variances for all observations. To test the presence of heteroskedasticity both the White test is conducted. The result of the tests for Model 1-4 (specified in Section 5) in Table 6 indicates that the null hypothesis about homoscedasticity is rejected since there is a p-value with 0.006 for both tests, which means my equation is biased by significant heteroskedasticity. Due to the nature of heteroskedasticity, the variances and standard errors of coefficient estimates would be certainly biased. In addition, the significance tests for the explanatory variables can be too high or too low, which will underestimate the impacts of R&D organizations in the regional innovation system on regional innovation performance in my equation and thus underestimate the role of RIS on innovation.
Table 6-Result from the White test of heteroskedasticity Model Result from the White test of
heteroskedasticity (prob>chi2)
Conclusion of the White test (null hypothesis: the data are
homoscedastic)
Model 1 0.0002 Reject the null hypothesis at 5% level
Model 2 0.0001 Reject the null hypothesis at 5% level
Model 3 0.0180 Reject the null hypothesis at 5% level
Model 4 0.0500 Reject the null hypothesis at 5% level
To overcome with heteroskedasticity I will use White robust standard errors to increase the p-values and standard errors of the coefficients in the regression of the thesis (Hill et al, 2007).
4.5 Endogeneity
25
(GDPcapita and FDI) in the equation, which is already discussed in the Section 3.4. Therefore, I can use the Hausman to test the presence of endogeneity and in principle, solve it by recommending instrumental variable (Hill et al, 2007). A good instrumental variable should be highly correlated with the endogenous variables but not with the error term in two equations. However, an ideal instrument is difficult to find and beyond the scope of the research.
Bilbao and Rodríguez (2004) applies lagged values of R&D and the work of Fritsch (2002) indicates that exogenous and lagged endogenous variables (e.g. R&D expenditure) can be used to solve endogeneity. Hence I will include this simple but practical method to ease the problem of endogeneity, applying a one-year lag of all R&D activities performed by actors, as well as other explanatory variables. One advantage in applying the lagged values of R&D, FDI and other explanatory variables is that it removes the possible endogeneity problem. Another advantage is to capture the time lags of production of knowledge (measured in patents). However, the approach by using one-year lag for to all explanatory variables will potentially suffer from some problems. For instance, some types of R&D will potentially take longer to make contribution to patents as an indicator of innovation, e.g. higher education R&D expenditures that focus on basic science.
5. Empirical results
5.1 Main regression
Empirical estimates obtained by using fixed effect model are reported in the models1-4, since the Hausman test rejects the null hypothesis that the random effect is appropriate. As discussed before, I use the White robust standard errors to relieve heteroskedasticity. Firstly, model 1 is run for all the 30 regions covered in China to decide the determinants of RIS on regional innovation capacity.
26 (FE) LnRIent_1 0.229** (0.113) LnRIinst_1 -0.171** (0.084) LnRIhe_1 -0.046 (0.046) LnURFE_1 0.134*** (0.048) LnGDPcapita_1 1.121*** (0.180) LnInternet_1 0.098*** (0.025) LnFDI_1 -0.071 (0.057) LnSkill_1 0.130 (0.100) Constant 3.370 (2.064) N 270 R-square 0.7180 𝛘𝟐(𝐇𝐚𝐮𝐬𝐦𝐚𝐧 𝐭𝐞𝐬𝐭) 73.49***
Notes: Figures in parentheses are standard deviations. ***, **, and * indicate significance at 1%, 5%,
and 10% statistical levels, respectively.
Source: CSY (1999-2008), CSYST (1999-2008) and SIPO (2008) and CNNIC (2008).
It is shown that Model 1 has a good explanatory power from the R-square value of 0.7180, which means 71.80% of variation in regional innovation capacity is explained by the including explanatory variables in the model.
The sign of the coefficient of LnRIent_1 is positive and significant at 5% statistical level, supporting the Hypothesis 1 that suggests the higher investment in enterprise R&D as a percent of GDP experiences a better innovation capacity. In this sense, among the organizations of RIS in China, enterprises are more applied-oriented that inclines to apply for patents.
27
that the investment in institutions R&D did not positively contribute to innovation performance in the observed years. There is a trick here: the ratio between growth rate of institution R&D expenditure and growth rate of regional GDP is different. In this case, this type of R&D grew slower than regional GDP. Thus, even though RIinst has a negative signal, it does not imply that R&D expenditure has come down in institutions. Also, it is quite likely that research institutions R&D focuses on basic science, which takes longer to make contribution to patents as a proxy of innovation. Since one-year lag is applied for all types of R&D expenditures, it might be that it takes longer before R&D projects to invent patents. Even though I don’t have a test for this, I would still expect that a rise of institutions R&D may increase patents per million inhabitants.
The coefficient of LnRIhe_1 is insignificant, concluding that there is no evidence that the universities R&D performers in RIS has impact on innovation. So Hypothesis 3 is not supported in general. This implies that unversities’ R&D in China is apt to basic science research and demonstrates their R&D outputs in academic journals or utility and design patent application instead of invention patent application used as dependent variable. Even though R&D expenditure of enterprise, research institutions and universities have different influences on regional innovation capacity I discussed above, the linkage effect of tri-relationship for enterprise-university-academia plays a key influence on region innovation. The variable LnURFE_1, defined as the ratio of institutions and universities financed by enterprises, is found to have the expected positive sign and is statistically significant at 1% confidence level. It implies that RIS stimulates the actors inside it to interact and cooperate by exchanging ideas and fulfilling R&D cooperation, through which basic and applied research can be integrated to effectively develop fresh ideas and thus new technologies. Therefore, hypothesis 4 is supported.
28
examined years as a measure of skill, while the production of innovation is mainly resulted from skilled labors with several working experience. Finally, the result does not show a positive and significant effect of the FDI coefficient on patents, indicating that it has no significant effect on regional innovation capacity. The possible reason is that the FDI inflows can only benefit innovation capacity in the host region through spillover channels that I didn’t take into account in the model.
So far, one thing should be bear in mind is that enterprise R&D is often linked to the creation of new products and production techniques, as well as to a region’s innovation capacity effort. Only enterprise R&D and the relationship between main organizations are positively associated with regional innovation capacity according to my analysis.
5.2 Robustness check
The next step is to check the robustness of the result above. As the capital of China, Beijing is a special region and the nation’s political, cultural and educational center. It has 83 universities and 383 research institutions, by far the largest concentration in any Chinese city (Beijing Statistical Information Net, 2008). The most prominent examples are Peking University, Tsinghua University, as well as the Chinese Academy of Sciences. Also, a number of enterprises have set up their headquarters in Beijing and high-technology firms have become a primary force for the development of innovation and economic development. Additionally, Beijing has absolutely obtained more attention of policy-makers in various ways, which indicates it has exhibited quite different evolutionary trajectories in its regional innovation system (Chen and Kenny, 2007). Therefore, it is important to test what the effect is of omitting the observations of Beijing. Hence, I conduct the robustness check and re-examine the role of RIS on regional innovation capacity without the special case of Beijing.
Table 8-Estimation for 29 regions (excluding Beijing) between 1998 and 2007 Variables Model 2
29 LnRIent_1 0.232* (0.125) LnRIinst_1 -0.175** (0.085) LnRIhe_1 -0.047 (0.047) LnURFE_1 0.133*** (0.047) LnGDPcapita_1 1.176*** (0.228) LnInternet_1 0.094*** (0.026) LnFDI_1 -0.075 (0.066) LnSkill_1 0.102 (0.118) Constant 3.258 (2.105) N 261 R-square 0.7554 𝛘𝟐(𝐇𝐚𝐮𝐬𝐦𝐚𝐧 𝐭𝐞𝐬𝐭) 66.49***
Notes: Figures in parentheses are standard deviations. ***, **, and * indicate significance at 1%, 5%,
and 10% statistical levels, respectively.
Source: CSY (1999-2008), CSYST (1999-2008) and SIPO (2008) and CNNIC (2008).
The effects can be studied by looking at the robustness result in Table 8. In terms of the result, it changes little when Beijing is excluded. The explanatory variables remain their original signals and significance, except for RIent variable. The coefficient of RIent becomes less significant, with a significance level of 10%. When it comes to other R&D variables, the coefficients of RIinst and URFE remain significant, with the significance respectively at 5% and 1%.
30
Beijing when conducting further analysis in terms of the analyses in coastal and inland regions.
6. Further analysis
Are there any differences between provinces belonging to coastal and inland regions in China? This answer will provide policy implications for authority to assess S&T
Development Plan process so far. Model 3-4 present the result for the coastal and inland regions subset, respectively.
Table 9-Estimation for coastal and inland regions respectively (1998 and 2007) Variables Model 3:Coastal
(FE) Model 4:inland (FE) LnRIent_1 0.138** (0.070) 0.220* (0.116) LnRIinst_1 -0.395 (0.267) 0.038 (0.081) LnRIhe_1 -0.112 (0.068) 0.020* (0.057) LnURFE_1 0.293*** (0.075) 0.022 (0.046) LnGDPcapita_1 2.061*** (0.300) 0.356 (0.228) Lninternet_1 0.114*** (0.042) 0.108*** (0.025) LnFDI_1 0.025** (0.013) -0.020 (0.034) LnSkill_1 -0.230 (0.148) 0.420*** (0.130) Constant 2.120 (1.483) 3.741 (2.254) N 99 162 R-square 0.6562 0.7010 𝛘𝟐(𝐇𝐚𝐮𝐬𝐦𝐚𝐧 𝐭𝐞𝐬𝐭) 35.72*** 65.12***
Notes: Figures in parentheses are standard deviations. ***, **, and * indicate significance at 1%, 5%,
and 10% statistical levels, respectively.
31
Coastal
When only coastal regions are taken into consideration, Model 3 also reports a positive effect of enterprise R&D investment in fostering innovation in coastal regions at 5% significant level and an insignificant role of universities R&D intensity. However, the role of research institutions R&D becomes insignificant. A plausible explanation may be provided by the relatively low importance of institutions R&D investment in RIS in coastal regions. In terms of the innovation network (the connection among R&D actors), the interaction variable LnURFE is significant at 1% level and it implies that coastal regions have a well-structured connection among various R&D in RIS.
As in previous analysis, GDP per capita and technological infrastructure variables remain statistically significant. With respect to FDI, its coefficient becomes significant at 5% level, suggesting that FDI inflow to coastal regions has a significant impact on invention patenting of these regions. In terms of the insignificant of the coefficient the level of skill, a plausible explanation is that the input of human capital in innovation from current graduates in coastal regions still need more time to find a job in engaging innovation-related activity because of fierce competition.
Inland
Is this also the same case for inland regions in China? Model 4 shows that the actors in RIS carrying out research activities a bit differ with respect to those in coastal regions. Researchers conducted at enterprises and universities are significantly related to the invention patent application, both at 10% significance level. From this outcome, it seems that universities together with enterprises play a crucial role in boosting innovation capacity, serving as centers and engines of innovation in inland regions in comparison to coastal regions. These results are in accordance with some previous studies which also conclude the importance of universities in many lagging regions (Audretsch and Vivarelli, 1996; Rodriguez-Pose and Refolo, 2003).
32
on regional innovation capacity, possibly implying that the aim of research institution bases in inland regions (like western provinces) is not to produce invention patents, but some basic techniques that can improve the basic life or public infrastructure in western development.
An obvious change also appears in the coefficient of the variable, LnURFE acted as the linkage between various R&D performers in inland regions. Its coefficient becomes insignificant. The result reveals a possible reason that lies in the ineffectiveness of regional innovation network in relatively lagging regions, thus the R&D actors are lack of channels to cooperate and interaction, reflecting a weak regional innovation capacity in inland regions.
With respect to the coefficients of control variables, technological infrastructure maintains its sign and is statistically significant at 1%. The same significantly positive signal can be found in the level of skill, the coefficient of which is significant at 1%. The coefficients of FDI and GDPcapita are however insignificant, so I conclude FDI as well as GDPcapita do not have a significant impact on invention in inland regions. It suggests that the inland regions lack attractiveness to foreigners to invest FDI into relatively lagging regions in China. In terms of the impact of GDPcapita on patents, it is too low to set a platform for inland regions to demonstrate a well innovation performance indicated by patents.
Finally, it is also worth noting the explanatory power of the Model 3 and 4. Almost 65.62% of variation in the dependent variable, invention patent applications, can be explained by the model when the analysis is conducted for coastal regions and 70.10% for inland regions. This suggests that the model better fits regional innovation capacity in inland regions. In the end, I would like to briefly summarize the results obtained by the explanatory variables above. From the Model 1-4, the most support finding is the significant influence of enterprise R&D intensity on innovation. Throughout the regressions, it maintains its significance and the positive signs as I have expected.
33
with the positive innovation network (LnURFE) in fostering innovation. For inland regions, besides enterprise R&D, the research conducted by higher education institutions will also exhibit a positive return to innovation capacity. In contrast, the interaction among organizations in RIS is weak, which shows the ineffective network in RIS in inlands.
7. Limitation
The thesis is marked by several important limitations.
34
in institutions and universities R&D, leaving other dimensions ignored. Moreover, only the development of innovation is considered and other functions of RIS, e.g. diffusion and use of innovation are left aside. From these aspects, I would undermine the network of R&D actors in RIS.
8. Conclusion
This thesis has sought to shed some light upon the determinants of innovation capacity in China, especially the role played by regional innovation system. More specifically, the paper is to determine whether the R&D activities in RIS undertaken by different organizations-enterprises, universities and research institutions-have had different impacts on regional invention patenting. Further analysis is conduct to study if they varied depending on whether a region belongs to coastal or inland regions or not.
Overall, the results show the presence of a positive link between enterprises R&D activities and the genesis of invention patenting. The research activities performed by enterprises have higher rates than research conducted by the remaining organizations. Simply because enterprises R&D investment tends to be more applied, novelty and more commercially oriented and also because the invention patent application is taken as a proxy of innovation in the thesis. In contrast, universities and research institutions R&D investments incline to basic sciences, making their R&D effort on the number of invention patent applications weaker. Finally, the closer interaction among enterprise-university-academia contributes a significant positive impact on regional innovation capacity, suggesting the crucial role of a well-established RIS on promoting a region’s innovation capacity.
35
However, the network of RIS is still immature in terms of generating innovation in inland regions.
As a whole, this paper has emphasized on the complex relationship between R&D performers and regional innovation capacity in RIS. The thesis has also shown the impact of RIS on regional innovation capacity is contingent on some factors. Factors like GDP per capita, the amount of FDI inflows, the intensity of Internet usage, or the availability of skills play an important role in the capacity of a region to generate innovation.
In particular, it confirms the fact that the innovation policy so far pays off in terms of innovation, addressing the effective mechanism in boosting enterprise R&D to foster regional innovation capacity in coastal and inland regions in China. Especially, the enterprise R&D investment in inland regions has a higher return rate than coastal regions on innovation. This finding in the thesis indicates the government can plan and finance more enterprise R&D activities or strengthen the network of RIS in inland regions to reduce the gap between regions and boost sustainable development in China.
Reference
Acs Zoltan J, Anselin L, Varga A. (2002). Patents and innovation counts as measures of regional production of new knowledge. Research policy, 31: 1069-1085.
Beijing Statistical Information Net. (2008). http://www.bjstats.gov.cn/ Retrieved 25.03. 2008.
Bilbao‐Osorio B, Rodríguez‐Pose A. (2004). From R&D to innovation and economic growth in the EU. Growth and Change, 35: 434-455.
Cai, Y., Liu, C. (2013). The roles of universities in Chinese regional innovation systems-a re-examination of the triple helix model. Regional Studies Association European Conference. University of Tampere, Finland.
36
and methodological issues. Research Policy 31: 233-245.
Chen K., M. Kenny (2007). Universities/research institutes and regional innovation systems: The case of Beijing and Shenzhen. World Development 35: 1056-1074.
Chen, K., Guan, J. (2011). Mapping the functionality of China’s regional innovation system: A structural approach. China Economic Review, 22: 11-27.
Chen, K., Guan, J. (2012). Measuring the efficiency of China’s regional innovation system: Application of network data envelopment analysis (DEA). Regional Studies. 46: 355-377.
Chung, S. (2002). Building a national innovation system through regional innovation systems. Technovation 22: 485-491.
Cooke, P., Uranga, M.G. Etexbarria, G. (1997). Regional innovation systems: institutional and organizational R&D. Management Science 48: 1-23.
Cooke, P. (2001). Regional innovation system, clusters, and the knowledge economy.
Industrial and Corporate Change, 10:945-974.
Doloreux D, Parto S. (2004). Regional innovation systems: a critical review. Maastricht,
MERIT, 190: 1-26.
Doloreux, D., S. Dionne. (2008). Is regional innovation system development possible in periphery regions? Some evidence from the case of La Pocatie’ re, Canada.
Entrepreneurship and Regional Development 20: 259-283.
37 10: 563-581.
Edquist, C. (2005). Systems of innovation: perspectives and challenges. In Fagerberg, Jan., et al. (Eds.), The Oxford Handbook of Innovation: 181-208. Oxford University Press.
Etzkowitz, H., Leydesdorff, L. (1997). Universities and the global knowledge economy: A triple helix of university-industry-government relations. London: Printer.
Fabre G., Grumbach, S. (2012). The world upside down, China’s R&D and innovation strategy. Working Papers Series. Fondation Maison Des Science De I’homme, Paries, France.
Freeman, C. (1987). Technology policy and economic performance. London: Printer.
Freeman, C. (2002). Continental, national and sub-national innovation
systems-complementarity and economic growth. Research Policy 31: 192-211.
Fritsch M. (2002). Measuring the quality of regional innovation systems: A knowledge production function approach. International Regional Science Review, 25: 86-101.
Fritsch M, Slavtchev V. (2011). Determinants of the efficiency of regional innovation systems. Regional Studies, 45: 905-918.
Furman, J.L., Porter, M., Stern, S. (2002). The determinants of national innovation capacity.
Research Policy 31: 899-933.
38
Griliches, Z. (1990). Patent Statistics as Economic Indicators: A survey. Journal of
Economic Literature, 1691-1707
Gu, S., Lundvall, B.A. (2006). China’s innovation system and the move toward harmonious growth and endogenous innovation. Innovation: Management, Policy&Practice 8:1-26.
Hill. R.C., Griffiths, W.E. and Lim G.C (2007). Principles of Econometrics, 3rd Edition, New York: John Wiley & Sons, Inc
Hu, A.G., G.H. Jefferson. (2009). A great wall of patents: what is behind China’s recent patent explosion? Journal of Development Economics, 90: 57-68.
Hutschenreiter, G. and G. Zhang (2007). China’s quest for innovation-driven growth-the policy dimension. Journal of Industrial Competition and Trade, 7: 245-254.
Krugman, P. (1991). Geography and trade. Cambridge, MA: MIT Press
Leichenko, R., J. Silva. (2004). International trade, employment and earnings: Evidence from US rural countries. Regional Studies 38: 355-374.
Li, X. (2009). China’s regional innovation capacity in transition: an empirical approach.
Research Policy 38: 338-357.
Liu, X., White, S. (2001b). Comparing innovation systems: a framework and application to China’s transitional context. Research Policy 30: 1091-1114.
Liu X, Lundin N (2008). Toward a market-based open innovation system of China. Retrieved October, 3: 2008.
39
production, innovation and competence building. Research Policy 31: 213-231.
Mowery, D.C., Sampat, B.N. (2004). Universities in National Innovation Systems. The Oxford Handbook of Innovation: 209-239. Oxford and New York: Oxford University Press.
Nelson, R. R. (1993). National innovation systems: A comparative analysis, New York: Oxford University Press.
Pavitt, K. (1991). What makes basic research economically useful? Research Policy 20: 109-119.
Pavitt, K. (1998). The social shaping of the national science base. Research Policy 27: 793-805.
Porter, M., Stern, S. (2001). National innovation capacity. The Global Competitiveness Report 2001-2002: 102-119.
Rodriguez-Pose, A., M.C. Refolo. (2003). The link between local production system and public and university research in Italy. Economic and Planning A, 35: 1477-1492.
Torrisi, G. 2009. Public infrastructure: definition, classification and measurement issues. University of Catania, DEMQ. Mimeo.
Toner P. (2011). Workforce skills and innovation: An Overview of Major Themes in the Literature. OECD Directorate for Science Technology and Industry (STI), OECD 2011.
40
Sigurdson, J. (2004). Regional innovation system (RIS) in China. EIJS. Working paper, No 195.
Sorensen, A. (1999). R&D, learning and phases of economic growth. Journal of Economic
Growth 4: 429-445.
Stern, S., M. Porter, & J. Furman (2000). The determinants of national innovative capacity, NBER Working Paper 7876.
Sun, Y. (2000). Spatial distribution of patents in China. Regional Studies, 34: 441-454.
UCLA (2010). Stata web books, regression with Stata. Chapter 2: Regression diagnostics.
http://www.ats.ucla.edu/stat/stata/webbooks/reg/chapter2/statareg2.htm,
http://www.ats.ucla.edu/stat//stata/dae/rreg.htm
Young, A. (1995). Lessons from the East Asian NICS: A contrarian View. European
Economic Review 38: 946-973.
Yueh, L.Y. (2006). The determinants of innovation: patent laws, foreign direct investment and economic growth in China. University of Oxford, Centre for Economic Performance, LSE.
