This article was downloaded by: [National Science Library] On: 20 March 2012, At: 21:44 Publisher: Routledge Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK Regional Studies Publication details, including instructions for authors and subscription information: http://rsa.tandfonline.com/loi/cres20 Measuring the Efficiency of China's Regional Innovation Systems: Application of Network Data Envelopment Analysis (DEA) Kaihua Chen a & Jiancheng Guan b a School of Management, Beijing University of Aeronautics & Astronautics, 100191, Beijing, China b School of Management, Fudan University, 200433, Shanghai, China E-mail: Available online: 26 Aug 2011 To cite this article: Kaihua Chen & Jiancheng Guan (2012): Measuring the Efficiency of China's Regional Innovation Systems: Application of Network Data Envelopment Analysis (DEA), Regional Studies, 46:3, 355-377 To link to this article: http://dx.doi.org/10.1080/00343404.2010.497479 PLEASE SCROLL DOWN FOR ARTICLE Full terms and conditions of use: http://rsa.tandfonline.com/page/terms-and-conditions This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses should be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material.
24
Embed
Measuring the Efficiency of China's Regional Innovation …or.nsfc.gov.cn/bitstream/00001903-5/111472/1/1000003441248.pdf · Measuring the Efficiency of China’s Regional Innovation
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
This article was downloaded by: [National Science Library]On: 20 March 2012, At: 21:44Publisher: RoutledgeInforma Ltd Registered in England and Wales Registered Number: 1072954 Registered office: MortimerHouse, 37-41 Mortimer Street, London W1T 3JH, UK
Regional StudiesPublication details, including instructions for authors and subscription information:http://rsa.tandfonline.com/loi/cres20
Measuring the Efficiency of China's RegionalInnovation Systems: Application of Network DataEnvelopment Analysis (DEA)Kaihua Chen a & Jiancheng Guan ba School of Management, Beijing University of Aeronautics & Astronautics, 100191,Beijing, Chinab School of Management, Fudan University, 200433, Shanghai, China E-mail:
Available online: 26 Aug 2011
To cite this article: Kaihua Chen & Jiancheng Guan (2012): Measuring the Efficiency of China's Regional InnovationSystems: Application of Network Data Envelopment Analysis (DEA), Regional Studies, 46:3, 355-377
To link to this article: http://dx.doi.org/10.1080/00343404.2010.497479
PLEASE SCROLL DOWN FOR ARTICLE
Full terms and conditions of use: http://rsa.tandfonline.com/page/terms-and-conditions
This article may be used for research, teaching, and private study purposes. Any substantial or systematicreproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form toanyone is expressly forbidden.
The publisher does not give any warranty express or implied or make any representation that the contentswill be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug dosesshould be independently verified with primary sources. The publisher shall not be liable for any loss, actions,claims, proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly orindirectly in connection with or arising out of the use of this material.
CHEN K. y GUAN J. Medicion de la eficiencia de los sistemas de innovacion regional de China: aplicacion de un analisis envolvente
de datos de redes, Regional Studies. En este estudio aplicamos un analisis envolvente de datos relacionales de redes para la evaluacion
sistematica de la eficiencia de la innovacion de los sistemas de innovacion regional de China descomponiendo el proceso de inno-
vacion en los dos subprocesos conectados: el desarrollo tecnologico y la posterior comercializacion tecnologica. Los resultados
muestran que solamente una quinta parte de los sistemas de innovacion regional operan en la frontera empırica de las mejores
practicas durante todo el proceso desde el desarrollo tecnologico hasta la comercializacion. Asimismo se observa que existen con-
siderables incoherencias entre la capacidad de desarrollo tecnologico y la capacidad de comercializacion en la mayorıa de sistemas
de innovacion regional y que la capacidad de comercializacion desempena un papel mas importante en el rendimiento de innova-
cion de los sistemas de innovacion regional.
Sistemas de innovacion regional Proceso de innovacion Eficacia tecnica Analisis envolvente de datos de redes (DEA)
Enfoque Bootstrap China
JEL classifications: O11, O18, O32, O47
INTRODUCTION
Science and technology (S&T) production withinregional innovation systems (RISs) is based on a multi-input, multi-output transformation relation, so it canbe assumed that the aggregate performance of the RIScan be characterized by the efficiency of the input–output relation based on a consideration of all relevantinputs and outputs (ZABALA-ITURRIAGAGOITIA et al.,2007; FRITSCH and SLAVTCHEV, 2006, 2007;BROEKEL and BRENNER, 2007; BROEKEL, 2008;GUAN and CHEN, 2010a). It is an important issue toinvestigate the operational quality related to the trans-formation process of limited innovation resources forimproving regional innovative outputs. Furthermore,China’s national innovation system is too large andcomplex to be summarized with a single model, so theregional dimension should not be overlooked (ORGAN-
ISATION FOR ECONOMIC CO-OPERATION AND
DEVELOPMENT (OECD) and MINISTRY OF SCIENCE
AND TECHNOLOGY (MOST), 2007).China’s regions have played and will continue to play a
key role in the advancement of S&T in China (OECDand MOST, 2007). However, for historic reasons thereare significant interregional inequalities in the develop-ment of S&T and the economy among the regions inChina (LEE, 2000; GUAN and LIU, 2005). These inequal-ities have – to some extent – impeded the harmonicdevelopment of the total economy of China (GUAN
and LIU, 2005). Among the critical issues facing policy-makers of regional governments, as well as of thecentral government, are how to construct and improvethe lagging regional innovation systems, enhance the har-monic development of regional innovative capacities andfurther impel sustainable economic development.
With the transformation of a planned economy to amarket one, China has experienced significant changes
in its economic and ownership structure. This trans-formation has also had a large impact on S&Tand econ-omic management (BROCKHOFF and GUAN, 1996;YAM et al., 2004). Each province is gradually becomingrelatively independent from the central authorities.Regional governments all around the country aremaking great efforts to improve the economic andsocial development as harmoniously as possible. There-fore, a province-level empirical analysis seems to be anappropriate approach for investigating China’s nationalinnovation systems (for example, LI, 2009; and GUAN
and CHEN, 2010a), through which the cross-regionalvariation in innovation efficiency performance can beinvestigated and analysed. Besides, both the mobiliz-ation of labour forces and the operation of the wholeinnovation process happen more often from within,rather than between, provinces. This also provides afeasible basis (for more advantages, see LI, 2009) forthe cross-province empirical analysis.
The technical efficiency of a region to a largedegree reflects its capacity of transforming innovativeinvestment into innovative outputs and economicprofits. Therefore, the innovation performancerelated to the technical efficiency of a region is thekey for this region in order to acquire competenceadvantage. Moreover, measuring and comparing tech-nical efficiencies of peer RISs from the S&T input–output transformation perspective, and then identify-ing the interregional disparities that are of great impor-tance to build a sound national innovation system.However, the innovation processes related to theRISs have not been sufficiently understood in virtueof empirical investigations, except for limited studiesrelated to European regions (for example, FRITSCH,2002; FRITSCH and SLAVTCHEV, 2006, 2007;BROEKEL and BRENNER, 2007; BROEKEL, 2008;
356 Kaihua Chen and Jiancheng Guan
Dow
nloa
ded
by [
Nat
iona
l Sci
ence
Lib
rary
] at
21:
44 2
0 M
arch
201
2
and ZABALA-ITURRIAGAGOITIA et al., 2007). Withrespect to China’s case, some work has beenimplemented (for example, GUAN and LIU, 2005;GUAN and CHEN, 2010a; and LI, 2009). However,these studies only partially characterize the regionalinnovation process.
A typical technological innovation process in theeconomic sense is accomplished only with the firstcommercial transaction (FREEMAN and SOETE, 1997),which is focused on two consecutive but interactivecomponent processes: a technological developmentprocess and a technological commercialization process(PAKES and GRILECHES, 1980; FURMAN et al., 2002;MOON and LEE, 2005). From the perspective ofregional innovation policy-makers, the upstreamdevelopment process is concerned with developmentefficiency; the downstream commercialization processis concerned with commercialization efficiency; andthe whole technological innovation process is con-cerned with overall innovation efficiency. The existingliterature related to the technical efficiency of theRISs mainly focused on the upstream technologicaldevelopment efficiency, and neglected the downstreamcommercialization efficiency and the overall innovationefficiency. Some studies attempted to measure theRISs’ overall innovation efficiency, but usually cameacross the difficulty: are the patents innovative outputsor economic inputs? (for example, ZABALA-ITURRIA-
GAGOITIA et al., 2007).The purpose of the present study is to construct a
complete measurement framework characterizing theRISs’ production framework from original S&T invest-ment to final commercial outputs, and measure theRISs’ process-oriented technical efficiency, which isimplemented in China’s context. It is hoped thatthis study will benefit China’s regional innovation
policy-making, and the conceptual and methodologicalframework can be extended to other regional, nationalor sectoral innovation systems.
The paper is organized as follows. The secondsection introduces a network two-stage conceptual fra-mework for analysing the transformation procedurefrom regional original S&T inputs to intermediate tech-nological outputs and then final economic profits of theinnovation process. The third section describes themethods used and explains how both network DEAand super-efficiency DEA are applied to the proposedtwo-stage network framework for the overall efficiencyof the whole RIS as well as the component efficiency ofthe two internal sub-stages. Moreover, a bootstrapapproach is employed to investigate the influencesfrom outliers (or extremes) and random elements onthe RISs’ efficiency measures. The fourth section pro-vides important empirical findings for evaluatingChinese RISs’ technological development and com-mercialization efficiency. A concluding section providespolicy suggestions for improving the whole regionalinnovation performance in China.
CONCEPTUAL FRAMEWORK AND
MEASURING INDICATORS
An innovation process is essentially a knowledge-generation and -implementation process in which aninnovation production unit utilizes S&T resources fromvarious channels, creates new knowledge and increaseseconomic wealth. Just as DVIR and PASHER (2004)defined, an innovation is the process of convertingknowledge and ideas into a benefit value. Fig. 1 describesthe consecutive innovation process composed of techno-logical development and commercialization processes
Fig. 1. Two-stage conceptual transformation framework of an innovation processNote: For acronyms, see Table 1
Efficiency of China’s Regional Innovation Systems: Application of Network DEA 357
Dow
nloa
ded
by [
Nat
iona
l Sci
ence
Lib
rary
] at
21:
44 2
0 M
arch
201
2
through which the original innovative investments aretransformed into the final market profits. As depicted,the conceptual framework is focused on thetransformation process of innovative elements, that is, atypical innovation process in terms of a productionpoint of view. Specifically, the upstream technologicaldevelopment process from the original S&T investmentsto the incremental S&T knowledge creation (that is,intermediate technological innovation product) is thefirst stage. It is linked with the second stage, that is, thedownstream technological commercialization processfrom the incremental technological knowledge to themarket profits depending on the non-innovativeintermediate inputs, by the incremental technologicalknowledge. Note that the two sub-processes are rela-tional and not independent since they are connected bythe technological innovation product. This means thatthe intermediate technological innovation product hasa ‘double identity’, which is the output in the first sub-process and the input in the second sub-process interms of its role. Moreover, the implementation of thedownstream commercialization process needs the inter-mediate inputs of non-research and development (R&D)labour force and capital. In this sense, the technologicalinnovation process embedded in an RIS is a networktwo-stage production framework. The integrated con-ceptual framework not only leads one to assign primaryimportance to the upstream technological developmentperformance, but also reminds one of attachingparticular importance to the downstream technologicalcommercialization performance in the economic sense.
Note that the conceptual framework does not takeinto account the moderations from the region-specificstructural or contextual factors1 as defined inFURMAN et al. (2002), BROEKEL and BRENNER
(2007), and BRENNER and BROEKEL (2009). As afore-mentioned, the purpose of the present study is focusedon characterizing the transformation process of physicalS&T innovation resources in each regional unit. Thetwo-stage analytical and measurement perspective ofphysical innovative resources transformation in theinnovation process has been frequently embodied inthe extant study at various levels, such innovationprocess analytical frameworks as the ‘R&D projectdiagram’ (GEISLER, 1995), the ‘R&D processdiagram’ based on a typical R&D laboratory (BROWN
and SVENSON, 1998), the ‘National S&Tactivity frame-work’ (MOON and LEE, 2005), the ‘Innovation pro-duction process framework’ based on the China’shigh-technology industries’ innovation activities(GUAN and CHEN, 2010b), as well as the ‘Knowledgeproduction function diagram’ based on the technologi-cal R&D activities (GRILICHES, 1990). Of course, onehas to recognize that the transformation quality deter-mining the output level for the two individual sub-pro-cesses within a region is affected by some non-physicalregion-specific structural or contextual factors(FURMAN et al., 2002; BROEKEL and BRENNER,2007; BRENNER and BROEKEL, 2009) by interactionswith innovation processes. Among them, the industrialstructure related to the regional economic environmentshould receive special attention.2
Even so, to characterize the network two-stage pro-duction framework, a multiple input–output indicatorssystem is necessary (BROWN and SVENSON, 1998;FURMAN et al., 2002; MOON and LEE, 2005; GUAN
and LIU, 2005; ZABALA-ITURRIAGAGOITIA et al.,2007; GUAN and CHEN, 2010a, 2010b). It is importantto measure system performance as a whole ratherthan quantifying particular measures or key indicators(ZABALA-ITURRIAGAGOITIA et al., 2007). Therefore,the overall hypothesis in this paper advocates that theefficiency of the RIS is jointly and interactively affectedby multiple input–output factors, rather than by a single(combined) factor expressed as an index. In the existingstudy related to the efficiency measures of China’s RISs,the Research Group on Development and Strategy ofScience and Technology of China (RGDSST) and theMinistry of Science and Technology (MOST) ofChina have implemented a basic study by employingthirteen input–output indictors related respectively toregional innovation activities. Since the year 2000, thefindings based on each single performance measurehave been reported in the Annual Report of RegionalInnovation Capability of China 2005–2006 (RGDSST,2006). However, the information on a single measureonly presents a partial comparison across China’s RISsin terms of input or output orientation. One possiblereason for this approach is the fact that the role
Table 1. Indicators in the two-stage concept model
Acronym Full name Unit
Original S&T inputs
E(S&T) Expenditure on science and
technology
RMB10 000
P(S&T) Number of science and technol-
ogy personnel
Persons
FDI Foreign direct investment US$10 000
EIT Expenditure on the import of
technology
RMB10 000
EPDT Expenditure on the purchase of
domestic technology
RMB10 000
VCIDTM Value of contractual inflows in
domestic technical markets
RMB10 000
Intermediate technology outputs
INV Invention Item
UM Utility model Item
DES External design Item
Intermediate physical inputs
CAP Capital stock RMB100 million
LAB Labour 10 000 persons
Final commercialization outputs
GDP Gross domestic products RMB100 million
SNP Sale of new products RMB10 000
VOE Value of export US$10 000
AIUR Annual income in urban residents
per capita
RMB per person
358 Kaihua Chen and Jiancheng Guan
Dow
nloa
ded
by [
Nat
iona
l Sci
ence
Lib
rary
] at
21:
44 2
0 M
arch
201
2
differences in heterogeneous input–output measuresrelated to both S&T and economic activities duringinnovation processes as well as complex interactionsamong them are still insufficiently understood. There-fore, it is one of the authors’ motivations to balanceand integrate the incommensurable information ofthese heterogeneous factors for systematically and com-prehensively measuring and comparing China’s RISs.Table 1 gives an explanation of the indicators used inthe model. The two-stage transformation process isfirstly defined/based on the thirteen measures used bythe Annual Report of Regional Innovation Capability ofChina in the evaluation of Chinese RISs’ efficiency. Indoing so, the aggregate performance of an RIS simul-taneously determined by all chosen indicators as wellas the interaction between S&Tand economic activitiesduring the innovation process is emphasized.
The first stage of Fig. 1 concentrates on the techno-logical development efficiency of the upstream activitiesin the regional technological innovation process. Thisstage deals with the efficiency issue of original S&Tinputs from various channels and intermediate techno-logical outputs for an RIS. The input factors at theregional level include:
. Expenditure on science and technology (E(S&T)).
. Number of science and technology personnel(P(S&T)).3
. Foreign direct investment (FDI) in monetary terms.
. Expenditure on the import of technology (EIT).
. Expenditure on the purchase of domestic technology(EPDT).4
. Value of contractual inflows in domestic technicalmarkets in monetary terms (VCIDTM).
With respect to the output of stage 1, that is, theintermediate technological products of an innovationprocess, the patents (applied or granted counts) maybe the most appropriate proxy (for example,FRITSCH, 2002; ACS et al., 2002; GUAN and LIU,2005; FRITSCH and SLAVTCHEV, 2006, 2007;BROEKEL and BRENNER, 2005, 2007; BROEKEL,2008; ZABALA-ITURRIAGAGOITIA et al., 2007; GUAN
and CHEN, 2010a, 2010b; and LI, 2009), althoughGRILICHES (1990) pointed out:
not all inventions are patentable, not all inventions are
patented, and the inventions that are patented differ
greatly in quality.
(p. 1669)
The present study employed the counts of appliedpatents as an indicator for the intermediate innovativeproducts of China’s regions. In China, patents are classi-fied into three categories by the National Patent Officeof China: inventions, utility models and external designs(SUN, 2000). Inventions (INV) are ‘new technical pro-posals for products, methods, or both’.5 Compared withinventions, utility models (UM) and external designs(DES) are more incremental. They can be described
as minor innovations which are industry oriented.UM are described as ‘new technical proposals onshape, structure of a product or the combination ofboth’, while DES refer to ‘aesthetics and industry-appli-cable new designs for shape, design or color of aproduct, or their combination’ (SUN, 2003). All threetypes of patents recognized by the Chinese PatentOffice are employed to describe outputs of the S&Tactivities in stage 1 for an RIS. In this sense, the per-formance in the first stage should be viewed as techno-logical development efficiency.
According to MOON and LEE (2005), patent numbersare not only an R&D output indicator, but also play animportant role in economic activities. However,patents cannot describe the performance of economicactivities of a region, and can only be used as a proxyfor the use of new technology (ACS and AUDRETSCH,1993; ACS et al., 2002). The economic indicators thatare available in the China Statistical Yearbooks and usedby the Annual Report of Regional Innovation Capability ofChina are employed as the economic outputs from thesecond stage. These economic indicators include grossdomestic products (GDP), the sale of new products(SNP), the value of export in monetary terms (VOE),as well as the annual income of urban residents(AIUR). Amongst the outputs, GDP, SNP and VOEindices are measured at the level of a complete region,whereas the AIUR indicator is the annual income perurban resident. Therefore, the performance in thesecond stage may be viewed as technological commer-cialization efficiency.
Besides, the implementation of downstream com-mercialization stage is inevitably involved by the partici-pation of labour force (LAB) and capital (CAP) engagedin non-innovative activities (as shown in Fig. 1). There-fore, the two indicators LAB and CAP were additionallyincluded to incorporate those required non-innovativeelements participating downstream economic activities.To account fully for the accumulative effects of the CAPinvestment series on innovation commercialization,regional capital stocks were constructed from theregional investment series by using the perpetual inven-tory method (PIM) (KUO and YANG, 2008; ENFLO andHJERTSTRAND, 2009). The basic idea with PIM is thatthe capital stock K(t+1) in year t + 1 can be estimated as:
K(t+1) = I(t) + (1 − d)K(t)
where I(t) is the gross investment; and d is the averagedepreciation rate in the current period. The initialstock (K(0)) is calculated as:
K(0) =I(0)
g + d
where g is the growth rate for the capital investmentseries. To simplify the calculation, it was assumed that
Efficiency of China’s Regional Innovation Systems: Application of Network DEA 359
Dow
nloa
ded
by [
Nat
iona
l Sci
ence
Lib
rary
] at
21:
44 2
0 M
arch
201
2
investment grows at a constant annual growth rate. Thegrowth rate was measured by the average growth rate ofthe capital investment series for all provinces in theevaluation during the sample period (KUO and YANG,2008). Furthermore, following LI (2003), the averagedepreciation rate was assumed to be 6% (also YOUNG,2003).
Note that when applying DEA, the input and outputvariables in empirical models are thought of as beingdiscretionary and physical indicators, which are usuallymeasured by statistical and observable aggregatevalues. In the essential view, they are those factors thatcan be considered as part of a production process and,therefore, should be considered in the definition ofthe production possibilities set (PPS). Therefore, allindicators in Table 1 are measured by statistical andobservable aggregate values.
METHODS
As mentioned above, the RISs’ S&T production isbased on a multi-input and multi-output relation, inwhich, differently from standard production activity,both inputs and outputs are not only quantitatively het-erogeneous, but also sometimes truly incommensurablein terms of units of measurement (Table 1). Therefore,the model development of reconciling diverse measuressimultaneously oriented to multiple inputs and multipleoutputs is considered above all. DEA, as discussed in thefollowing, is a method particularly suitable for organiz-ing and analysing these complex data measuringinnovation production systems (GUAN et al., 2006a;GUAN and CHEN, 2010a, 2010b).
DEA is non-parametric and deterministic in thesense that all deviations from the frontier are assumedto be the result of technical inefficiency. It has certainadvantage in the analysis of public-sector activities andsemi-public activities such as the RIS (ZABALA-ITURRIAGAGOITIA et al., 2007). It is an effective wayby which one can simultaneously utilize multipleoutputs and multiple inputs with each being stated indifferent units of measurement (CHARNES et al.,1994). As alternative approaches, FRITSCH (2002) andFRITSCH and SLAVTCHEV (2006, 2007) have appliedparametric approaches for estimating regional inno-vation efficiency. Unfortunately, the parametricapproaches are subject to the non-linear form of pro-duction function as well as the multiple outputs,although they can take into account statistical noiseand random elements. The inherent problems withparametric models have led many researchers to applynon-parametric methods. Moreover, the simulationresults given by BANKER and NATARAJAN (2008) indi-cate that traditional DEA-based procedures performbetter than parametric methods in the efficiency esti-mation of individual decision-making units (DMUs).Therefore, it is not surprising that the existing literature
has preferred to employ the DEA-based approaches toinvestigate regional or national innovation systems (forexample, ROUSSEAU and ROUSSEAU, 1997, 1998;GUAN and LIU, 2005; ZABALA-ITURRIAGAGOITIA
et al., 2007; BROEKEL and BRENNER, 2007;BROEKEL, 2008; SHARMA and THOMAS, 2008; andGUAN and CHEN, 2010a).
With particular respect to modelling the conceptualframework shown in Fig. 1, where the whole RIS trans-formation process is composed of two interactive com-ponent processes, first of all it is required that anintegrated methodological framework simultaneouslyincorporates the two sub-technologies defined by tech-nological development and commercialization activitiessets. In the extant study, the overall efficiency measure-ment of the whole RIS treats reference technologies asblack box (GUAN et al., 2006a; ZABALA-ITURRIAGA-
GOITIA et al., 2007; LI, 2009; GUAN and CHEN,2010a), which neglects the operation of the interactingsub-processes within it. The relational network pathdiagram of the RIS given in Fig. 1 shows that the tech-nological development process is not independent ofthe technological commercialization process. Thismeans that the two sub-technologies are connected byintermediate products into a hybrid network pro-duction technology (FARE and GROSSKOPF, 1996a,1996b). Clearly, traditional standard DEA models arenot appropriate for the overall efficiency measure ofthe network two-stage production system since theyneglect the operation of internal interacting sub-processes (KAO, 2009).
The activity analysis model developed by FARE andGROSSKOPF (1996a, 1996b), which explicitly takesinto account intermediate inputs or products, is suitedto the network two-stage framework. Unfortunately,the interdependent relationship between componentprocesses was not incorporated into the envelopmentmeasurement framework for overall efficiency (KAO,2009). So did COOK et al.’s (2000) multi-componentratio model. In fact, there is a condition for embodyingthe interdependent relationship that the intensity vari-ables (for envelopment form) or the multipliers (forratio form) on the intermediate measures participatingin the two sub-processes must be equal. Moreover,these authors are not explicitly interested in obtainingmeasures of efficiency at each stage. To obtain theperfect modelling for overall efficiency and the twocomponent efficiencies with the consideration of aninterdependent relationship between the two sub-pro-cesses, the present study employs KAO’s (2009) relationalnetwork DEA model, which is an extended model ofFARE and GROSSKOPF’s (1996a, 1996b). This modelcan help to measure the technical efficiency of thewhole RIS and the two sub-processes in an integratedanalytical framework in the sense that the componentand overall efficiencies are estimated simultaneously.
More importantly, KAO’s (2009) relational networkDEA model presents discriminative efficiencies
360 Kaihua Chen and Jiancheng Guan
Dow
nloa
ded
by [
Nat
iona
l Sci
ence
Lib
rary
] at
21:
44 2
0 M
arch
201
2
measures due to including more constraints on theinternal operations in contrast to the traditional one-stage Charnes, Cooper & Rhoades (CCR) DEAmodel (CHARNES et al., 1978). This is an attractiveproperty in the special case of a small sample size. Inempirical studies, a difficulty frequently encounteredin real-world applications of DEA is that the numberof DMUs that are available is usually limited. A smallsample size often produces results that are misleading.Moreover, it produces a relatively large proportion ofefficient DMUs, which makes the subsequent rankingsdifficult. This is also possible even when satisfying therules of thumb that are used to constrain the numericalrelationship between the number of observations andthe sum or product of numbers of inputs and outputs(COOPER et al., 2004; DYSON et al., 2001). In theemployed network DEA, the calculated efficiencyresults are more discriminative and robust (KAO,2009). This will be further confirmed in the empiricalstudy in the fourth section.
It is desirable that the employed network DEA modelmakes the component efficiency measures connected invirtue of constructing a network production frontierdetermined by all heterogeneous component sub-technologies, and it presents cooperative and interde-pendent efficiencies measures. However, the traditionalindependent efficiency measures of individual sub-stages are attractive due to them depending on theirown homogeneous technologies (SEIFORD and ZHU,1999). In this paper, the super-efficiency formulations(ANDERSEN and PETERSEN, 1993) of CCR DEAmodel are further employed to calculate super-efficien-cies, which can produce completely comparableranking results for individual efficiency scores sets.
Despite its popularity, deterministic DEA estimatorshave some obvious drawbacks.6 First, it has long beenrecognized that DEA estimates of inefficiency aresensitive to outliers in the data. Second, DEA estimatorsalso suffer from inherent bias and stochastic elements(SIMAR and WILSON, 1998; ENFLO and HJERT-
STRAND, 2009). Moreover, in many applications,including that used in this paper, there are simply toofew observations available to obtain meaningfulestimates of inefficiency using deterministic DEA esti-mators. This study followed SIMAR and WILSON’s(1998) bootstrap methods to estimate the confidenceefficiency intervals in the context of the out-orientedCCR DEA model, which is used to investigate theinfluences from outliers and random elements.
EMPIRICAL ANALYSIS
Sample data
All the data used in this paper were derived from theChina Statistical Yearbook and China Statistical Yearbookon Science and Technology from 1995 to 2007 publishedby National Bureau of Statistics and Ministry of
Science and Technology between 1996 and 2008.Thirty7 province-level regions were chosen to be usedas the DMUs. Moreover, all the economic data weredeflated by the Chinese retail price index by taking1995 as the base year.
The output in S&T production is lagged, but with alag structure which is not fixed (BONACCORSI andDARAIO, 2003). In the extant work, a two-year lagfrom innovative investment to patent application inChina’s RISs was considered (for example, GUAN andLIU, 2005; and GUAN and CHEN, 2010a). In thisstudy, following their methods, two years’ delaybetween inputs and outputs in each sub-processduring the whole innovation process was taken withinthe RISs. This means that the delay time in the wholeinnovation process, that is, the production period,amounts to four years. In order not to overload thispaper, but providing dynamic trends of each RIS’s inno-vation performance, only three consecutive productionperiods – 1995–1999, 1999–2003 and 2003–2007 –were chosen. In this sense, the original input years are1995, 1999 and 2003, and the corresponding inter-mediate output years are 1997, 2001 and 2005 duethe two years’ lag which are accompanied by the inter-mediate input years. Accordingly, the final output yearsare 1999, 2003 and 2007, respectively, due to anothertwo years’ lag.
Note that the deepening of China’s S&T reform thatstarted in 1995 went through the whole period of years1995–2007 (OECD and MOST, 2007). Therefore, theempirical results at the average level over the threeobserved periods can shed light on the dynamic per-formance of China’s innovation policies. Moreover,the year 2001 when China joined the World TradeOrganization (WTO) belongs to the second period ofyears 1999–2003. In other words, the empiricalresults can additionally provide some evidence of com-paring the change of innovation performance beforeand after China joined the WTO.
Interdependent efficiency using network DEA
It is certain that the whole RIS is technically efficientonly when the two component stages (or processes)are technically efficient since the whole network pro-duction technology is composed of the two connectedproduction sub-technologies controlling two stages.Theoretically, most RISs could be expected to beclose to the network technology frontier and behaveas efficient units, but the results reported in Table 2show that there is only a small percentage being techni-cally efficient. Specifically, six regions (20%) – Fujian,Guangdong, Hainan, Qinghai, Ningxia and Xinjiang– were technically efficient for the whole RIS duringthe first production period across years 1995–1999(column 2); nine regions (30%) – Shanxi, Neimenggu,Shanghai, Fujian, Guangdong, Hainan, Gansu, Qinghaiand Xinjiang – were technically efficient during the
Efficiency of China’s Regional Innovation Systems: Application of Network DEA 361
Dow
nloa
ded
by [
Nat
iona
l Sci
ence
Lib
rary
] at
21:
44 2
0 M
arch
201
2
second production period across years 1999–2003(column 5); and six regions (20%) – Tianjin, Shanghai,Guangdong, Hainan, Qinghai and Xinjiang – per-formed efficiently in the latest production periodacross years 2003–2007 (column 8). Among them,four regions (13%) – Guangdong, Hainan, Qinghaiand Xinjiang – were stably efficient during the threeperiods across years 1995–2007. Beijing, as thebiggest receipt of S&T inputs and with most univer-sity/public research institutes, was on the best-practicefrontier for technological development performance;however, its overall system performs inefficiently dueto its inefficiency in technological commercializationperformance. This is inseparable from its specific inno-vative environment where there is the lion’s share ofbasic research in public institutes but not an industrialbase able to commercialize the innovative results(OECD and MOST, 2007). Therefore, the primarytask for it is to promote the transformation of techno-logical products. Besides, three regions – Hebei, Zhe-jiang and Guizhou – displayed the same performance.The reverse is true for Shanghai, where an activebusiness sector is to some extent deprived of a strong,
application-oriented basic research infrastructure(OECD and MOST, 2007). Therefore, its innovationactivities perform efficiently at the various stagesduring the two later observed periods.
Fig. 2 displays the trend of the average (that is,nation-level) innovation performance of individual effi-ciency scores sets over the three production periods.The change trend of the average efficiency score forthe overall stage as well as stage 2 focused on techno-logical commercialization first decreased during thetransition period 1999–2003 and later increasedduring the nearer period 2003–2007, while thechange trend of the average efficiency score for stage1 focused on technological development behavedoppositely, that is, first increased and later decreased.There are many complex environmental determinantsof the trend changes, such as economic environmentvariations, which are not topic issues in this study. It isargue here that the same trend of the overall stage asstage 2 in terms of average efficiency score indicatesthat stage 2 to a great extent determines the perform-ance change of the overall stage, that is, it plays amore important role in the regional overall innovation
Table 2. Each regional innovation system’s network efficiency in each stage during the three individual production periods
Region
Period 1: 1995–1999 Period 2: 1999–2003 Period 3: 2003–2007
performance in contrast to stage 1. The correlation ana-lyses among individual stages shown in Table 3 furtherconfirm this judgement.
Since the three efficiencies are estimated in an inte-grated framework constrained by the single networkproduction technology, it is appropriate to comparetheir efficiency scores directly. Fig. 2 shows that thecurve of efficiency scores for stage 2 is above thecurve for stage 1. This means that Chinese RISs’ capa-bility of transforming technology into economicalprofits is relative better, which is related to China’sreforming performance in enhancing firm’s innovationcapabilities and the commercialization of pubicresearch. Of course, the superior performance of stage2 over stage 1 is also connected with accelerated learn-ing from international good practices fostered by WTOmembership and observership in the OECD’s Commit-tee on Science and Technological Policy (CSTP) underfast economic growth as well as pressure from technol-ogy-based competition in domestic and internationalmarkets (OECD and MOST, 2007). With respect tooverall performance, it is natural that the average per-formance curve is between them since it is determinedby them simultaneously.
Since the overall efficiency scores are simultaneouslydetermined by the sub-technologies of both stages 1 and
2, it is judged that there is a relational relationshipbetween rank results based on the overall efficiencyscores and the two component efficiency scores. Spear-man’s rank and Kendall’s tau_b correlation coefficientsshown in Table 3 confirm the judgement. However,the rank correlation coefficients between the overallstage and stage 2 have more significant values whetherfor Spearman’s rank or Kendall’s tau_b method, whichfurther confirms that technological commercializationsignificantly influences the innovation performance ofan RIS. Besides, a series of non-significant rank coeffi-cients comparing stages 1 and 2 shows that there is noinevitable relational linkage between component per-formances, although they are physically connectedinto the context of the RISs’ integrated productionframework. The result also reflects the unmatchedrelationship between upstream and downstream com-ponent performances, which is related to the disparitybetween research and economic environments in mostregions.
Fig. 3 illustrates the distribution of collected effi-ciency scores sets (n ¼ 90) obtained from the networkfrontier technology for the three individual stages,where Fig. 3(a)–(c) are, respectively, based on thenetwork efficiency scores of the overall stage, com-ponent stage 1 and component stage 2 during thethree different production periods across years 1995–2007. Some important findings are as follows.
Firstly, the distribution of the three efficiency scoressets is not normal during the whole period, but exhibitsa desirable asymptotic property which is identical to thetrue distribution of efficiency. Therefore, it is reasonableto expect that the problems associated with using effi-ciency estimators will be less acute using a DEA ratherthan a parametric approach (BANKER and NATARAJAN,2008). As hypothesized by ZABALA-ITURRIAGAGOITIA
et al. (2007), since the non-parametric frontier-basedmethodology looks for best-practice observations andtakes them as a benchmark, one can expect a relativelyhigh number of observations to be 100% efficient.Theoretically, most observations could be expected tobe close to the frontier and to behave as efficientunits, but the histograms in Fig. 3 show there is widevariance across China’s RISs in the efficiency perform-ance of each stage. In contrast to the system and
Fig. 2. Comparison between average network efficiency scoresin the three individual production periods
Source: Authors’ calculations
Table 3. Correlation coefficients between three stages in terms of the calculated network efficiency scores during the three individualproduction periods
Method Stage
Period 1: 1995–1999 Period 2: 1999–2003 Period 3: 2003–2007
Note: ∗∗Correlation is significant at the 0.01 level (two-tailed); and ∗correlation is significant at the 0.05 level (two-tailed).
Source: Authors’ calculations.
Efficiency of China’s Regional Innovation Systems: Application of Network DEA 363
Dow
nloa
ded
by [
Nat
iona
l Sci
ence
Lib
rary
] at
21:
44 2
0 M
arch
201
2
commercialization efficiency scores, the histogramrelated to stage 1 in Fig. 3(b) shows there is wider var-iance in the technological development efficiencyscores. In terms of efficiency score ranges, the systemand commercialization efficiency scores are between0.6 and 1.0, while the development efficiency scoresare between 0.2 and 1.0. Moreover, the distributionof the efficiency scores of the technological develop-ment between 0.20 and 0.95 did not display anobvious fluctuation. Both the bigger standard deviation(SD) of 0.2418 and the smaller mean of 0.7916 confirmthe visual judgement.
As demonstrated by ZABALA-ITURRIAGAGOITIA
et al. (2007), who are concerned about comparingEuropean regional innovation performance, underlyingthe research related to regional efficiency-based inno-vation performance is the fact that although theamounts of resources within an RIS are important, itis not evident that those regions with larger amountsof resources are the most efficient ones. The presentstudy, which is based on China’s regional data, alsomade the same conclusion. In the case of China, onemajor reason is that the stock of intellectual capitalfalls behind the capital-intensive investment devotedto building the ‘hardware’ of the innovation system inChina, which creates the over-capacity in the use ofsome research infrastructures (OECD and MOST,2007). A series of negative correlation coefficientsbetween the overall efficiency and the share of S&Tresources (E(S&T) and P(S&T)) at the regional levelreported in Table 4 confirm the conclusion. Besides,the diminishing trend of coefficients (from statisticallysignificant to insignificant) related to both E(S&T)and P(S&T) from 1995 to 2007 shows the gradualimprovement of China’s innovation policy in the tran-sition from a centrally planned innovation model to amarket-based innovation model.
It is noted that in contrast to the calculated efficiencyresults from the one-stage CCR DEA model(CHARNES et al., 1978), the present network efficiencyresults are more discriminative, which can improve thelow discrimination power caused by the limited numberof observations in the one-stage CCR measure. Com-paring the calculated efficiency results respectivelyfrom two models (see Table A1 in the Appendix andTable 2), it is easy to find that the efficiency scoresfrom the network DEA model in this study are nothigher than those from the CCR model. Furthermore,in the efficiency classification, the proportion of effi-cient DMUs in the network DEA measure is relativelysmaller, which is especially true for the overall systemefficiency measures. The summary statistics show thatthe number of efficient DMUs is less than one-thirdin the network DEA measure, while that is more thantwo-thirds in the one-stage CCR measure. Moreimportantly, the employed network DEA measurepresents a desirable logic. In this context, the wholeregional innovation production is efficient only when
Fig. 3. Distribution of united regional innovation systems’(RISs) network efficiency scores during three production periods
for the three individual stagesSource: Authors’ calculations
364 Kaihua Chen and Jiancheng Guan
Dow
nloa
ded
by [
Nat
iona
l Sci
ence
Lib
rary
] at
21:
44 2
0 M
arch
201
2
the two component processes are efficient (Table 2).The efficiency results from the CCR measure do notobey the logic (see Table A1 in the Appendix). Forexample, Tianjin is efficient in the whole regional inno-vation process during period 1 (years 1995–1999),while it is inefficient in stage 1 (the technologicaldevelopment process) during the same period. Clearly,this is not reasonable intuitively when overall systemefficiency is considered.
Independent component super-efficiency
DEA efficiency scores display relative performancesamong peer DMUs, so it is more appropriate tocompare the ranks rather than the numerical scoresunder different samples, that is, in the context of differ-ent technology production frontiers (BROCKETT andGOLANY, 1996). To obtain a complete ranking of allDMUs, there has to be a full degree of freedom forDEA models. The number of degrees of freedom willincrease with the number of DMUs and decrease withthe number of inputs and outputs (COOPER et al.,2004). In the present case, most regions are CCR effi-cient (see Table A1 in the Appendix) for two individualcomponent stages in each period, although it can satisfythe popular rule of thumb, that is:
n ≥ max{m × s, 3(m + s)}
where n is the number of DMUs; m is the number ofinputs; and s is the number of outputs (COOPER et al.,2004).
In order to discriminate efficient regions fully, thesuper-efficiency formulation of CCR model isemployed here to calculate the super-technical effi-ciency scores of the two sub-processes. The rankingsof the super-efficiency for stages 1 and 2 are related tox- and y-axes, respectively, that is, the x-axis refers toa region’s position for technological development effi-ciency performance among thirty regions, while they-axis refers to a region’s position for technologicalcommercialization efficiency performance amongthem. Three rectangular coordinate systems are pre-sented (Fig. 4(a)–(c)) to depict the respective relativeperformance of each region during the three
consecutive periods: years 1995–1999, years 1999–2003 and years 2003–2007.
Five key innovative regions – Beijing, Guangdong,Shanghai, Jiangsu and Zhejiang – showed different per-formance in stages 1 and 2 during the first period acrossyears 1995–1999. As shown in Fig. 4(a), Zhejiang andShanghai displayed opposite status in stages 1 and2. Zhejiang (number (6, 28)) had a relatively bettertechnological development efficiency and a relativelyworse technological commercialization efficiency,while Shanghai (number (25, 1)) shows the opposite.Guangdong (number (7, 2)) performed better in bothcomponent efficiencies. Beijing (number (3, 13)) hasobtained a better technological development efficiencyperformance, while its commercialization efficiencywas in the middle level. Jiangsu (number (26, 9)) paidincreasing attention to business R&D (OECD andMOST, 2007), so it has obtained a relatively better com-mercialization efficiency in contrast to its developmentefficiency. In the case of Guangdong and Shanghai, it isenough to remember that regional industries are crucialto commercial transformation in stage 2. During thesubsequent production period across years 1999–2003, Zhejiang’s commercialization efficiency increased(number (6, 15)), and decreased during the final pro-duction period: years 2003–2007 (number (8, 21)). Inupstream development efficiency, Shanghai hasbecome better during the middle production periodacross years 1999–2003 (number (14, 6)), and hasobtained additional progress in ranking during thefinal production period (number (15, 4)).
With respect to the largest difference, it is Gansuduring the first production period across years 1995–1999 with a ranking difference of 25. The secondlargest difference occurs at Shanghai with a rankingdifference of 24. During both the later two productionperiods across years 1999–2007, it is Guizhou that hasobtained a largest ranking difference of 26. The secondlargest difference occurs for Jiangsu with a ranking differ-ence of 23 during the middle period, and Neimengguwith a ranking difference of 22 during the latestperiod. These considerable gaps between stages 1 and2 in terms of technical efficiency show that there is anunmatched relationship between component efficien-cies that does harm to the whole RISs’ technicalefficiency. Generally speaking in terms of all regions, if
Table 4. Correlation coefficients between the overall network efficiency scores and the share of science and technology (S&T) resources(E(S&T) and P(S&T)) during the three individual production periods
Method
Period 1: 1995–1999 Period 2: 1999–2003 Period 3: 2003–2007
Note: ∗∗Correlation is significant at the 0.01 level (two-tailed); and ∗correlation is significant at the 0.05 level (two-tailed).
Source: Authors’ calculations.
Efficiency of China’s Regional Innovation Systems: Application of Network DEA 365
Dow
nloa
ded
by [
Nat
iona
l Sci
ence
Lib
rary
] at
21:
44 2
0 M
arch
201
2
Fig. 4. Ranking comparisons of regional innovation systems’ (RISs) performance according to the calculated super-efficiency scores ofstage 1 (technological production process) and stage 2 (technological commercialization process) in the three individual production
periodsNote: The first and second numbers in parentheses are the ranking numbers of the super-efficiency scores for stages 1
and 2, respectively. Source: Authors’ calculations
366 Kaihua Chen and Jiancheng Guan
Dow
nloa
ded
by [
Nat
iona
l Sci
ence
Lib
rary
] at
21:
44 2
0 M
arch
201
2
the two component indicators coincided, one wouldexpect the majority of points to be along a 45% line.But this is not the case for three samples. The R2-values are below 0.1. Besides, rank correlation coeffi-cients were calculated in order to check this evidenceempirically. Spearman’s rank correlation coefficientsfor three scenarios through years 1995–1999, years1999–2003 and years 2003–2007 are 0.006, 0.238 and0.115, respectively, which are not statistically significanteven at the 10% level. This indicates that most of China’sregions are in an uncoordinated state from the upstreamtechnological development efficiency to the down-stream technological commercialization efficiency.This result further confirms the above finding.
Influences analysis from outliers
Note that the above empirical analyses are based on thedeterministic DEA models. They do not take intoaccount the outliers,8 inherent bias and noise in the stat-istical sense (SIMAR and WILSON, 2008). As highlightedby BROEKEL and BRENNER (2007) and BROEKEL
(2008), it is key for regional innovation efficiencymeasures since innovation processes, even at a systemlevel, are to some extent non-deterministic in nature.If not considered, there is usually a bias in efficiencyestimates. To provide more information to potentialusers or policy-makers, the influences from outliers inthe sample data are estimated herein.
To shed light on the influences from outliers and stat-istical noise in the regional estimates, SIMAR and
WILSON’s (1998) bootstrap method was followed to esti-mate bias-corrected efficiency scores aswell as confidenceintervals of out-oriented constant returns to scale (CRS)efficiency scores during three periods for each region (seeTables A2(a) and (b) in the Appendix), which can be usedto base policy activities. The results indicate substantialbias. Since the bias estimates are large relative to the stan-dard error estimates, the bias-corrected efficiency esti-mates are preferred to the original estimates. With thebias correction, none of the resulting efficiency estimatesequals 1. Despite the fact that the sample size is rathersmall, the confidence intervals are of moderate length.In this sense, the innovation performance measure forChina’s province-level regions in this study is only an esti-mate, and it therefore has some limitations.
CONCLUSIONS AND POLICY
IMPLICATIONS
The purpose of this study is to measure China’s regionalinnovation performance associated with non-parametricproduction frontier-based measurement modelsoriented to the two interdependent component pro-cesses (stages) as well as multiple heterogeneous andincommensurable inputs/outputs. This motivation notonly makes it possible to integrate the various indicatorscharacterizing component processes for the aggregateperformance comparisons in the whole innovationprocess as well as the individual component processesof peer regions, but also can help in revealing internal
Fig. 4. Continued
Efficiency of China’s Regional Innovation Systems: Application of Network DEA 367
Dow
nloa
ded
by [
Nat
iona
l Sci
ence
Lib
rary
] at
21:
44 2
0 M
arch
201
2
inefficiencies embedded in internal processes as well asthe physical relationships between the whole processand the individual two sub-processes, and between thetwo sub-processes.
For this purpose, this study employed a relationalnetwork data envelopment analysis (DEA) model aswell as the super-efficiency formulation of theCharnes, Cooper & Rhoades (CCR) model to estimatethe innovation performance of thirty of China’s pro-vince-level regions based on three consecutive scenarios(that is, complete production periods) starting in theyear 1995 and ending in the year 2007 with fixedtime lags. In addition, statistical inference was providedfor the out-oriented CCR efficiencies in order to diag-nose the influences from outliers or random elementsfor additional information.
Findings and conclusions
The evidence shows that most of China’s regionsperform inefficiently in the innovation performance.Only one-fifth of the studied regions perform effi-ciently in the whole innovation process and the down-stream technological commercialization process. Thepercentage of the regions performing efficiently in theupstream technological development process becomeslarger, but does not exceed one-third. As OECD andMOST (2007) argued, the inefficiencies in China’sinnovation system are inherent in the position ofChinese actors on the learning curve of best practices.
The authors did not find a significant cooperativerelationship between component processes whichtogether determine the innovation performance of thewhole regional innovation system (RIS). This findingreveals that there are some serious inconsistenciesbetween technology capability and economic perform-ance in most Chinese RISs. This means that theircapability to transform technological development effi-ciency to commercialization efficiency is relativelypoor. The empirical result is consistent with China’sactual situation (GUAN et al., 2006b). One key reasonis that there is a contradiction between upstreampublic research organization-centred research processesand downstream firm-centred commercializationprocesses in China. Moreover, it is revealed that thedownstream commercialization process plays a moreimportant role in the whole regional innovationprocesses in contrast to the upstream developmentprocesses. This fully supports the decision of promotinga firm-centred innovation system by China’s govern-ment in the 2006 national science and technology(S&T) Strategic Plan. Finally, by correlation analysisbetween the regional distribution of S&T-related invest-ments in expenditure and personnel and the overalltechnical efficiency scores, the authors have found thatheavy investment in S&T, although necessary for catch-ing up with highly developed countries, does notnecessarily bring high efficiency for the RIS and
cannot guarantee success in innovation. This confirmsthe OECD and MOST’s (2007) conclusion: theimpressive investment in resources has contributed sig-nificantly to the rapid socio-economic progress regis-tered in China in the last decade, but it has not yettranslated into a proportionate increase in innovationperformance. However, the correlations have obtainedgradual improvement from 1995 to 2007.
Policy implications and recommendations
The findings have important policy implications. First,governments at all levels should change their emphasisfrom current sharply increasing S&T investment to har-monizing tradeoffs between multidimensional innova-tive resources. Only in this way will the investment bereturned in an efficient manner in the economic sense.Second, the Chinese central government allocatingS&T resources should take into account marketdemand and shake off dependence on the planned inno-vation model. The confirmed crucial role of the techno-logical commercialization in the regional innovationperformance provides sufficient evidence. Of course, itis also necessary that government policy on pubicresearch should seek to strike a better balance betweenmission-oriented research and research driven bymarket demand (OECD and MOST, 2007). Moreover,the unmatched relationship between upstream techno-logical development and downstream commercializationfurther promotes the transformation of China’s inno-vation system into a more market-based system fromthe planned economic system. The unmatched result isalso inseparable from the top-down government policyefforts, where government policies have focused on thesupply of innovation inputs and support instruments,and on the development of formal public institutions(for example, publicly funded R&D centres, technologytransfers, education and training, and marketing inforeign markets). However, they usually neglect firminnovation orientations and strategies, absorptive capa-bilities, and specific demands in the region (GUAN
et al., 2009). It is especially true for China in the transitionfrom a reliance on technological import to one of indi-genous innovation.
From the funding instruments, it is advisable toimplement an improved mix of instruments to supportmore efficiently both market-led and mission-orientedS&T development and innovation. However, some fra-mework conditions are insufficiently conducive tomarket-led innovation in China, especially those relat-ing to corporate governance, financing of R&D andtechnology-based entrepreneurship, and enforcementof intellectual property rights (OECD and MOST,2007). Their improvement could create the necessaryconditions for the operation of an open innovationmodel in which indigenous innovation capabilitiesand R&D-intensive foreign investment could bemutually reinforcing.
368 Kaihua Chen and Jiancheng Guan
Dow
nloa
ded
by [
Nat
iona
l Sci
ence
Lib
rary
] at
21:
44 2
0 M
arch
201
2
Limitations and future studies
It is wise to point out several caveats to this empiricalstudy, which motivates future studies. One limitation,which is common for the studies of cross-country orcross-region S&T performance comparisons, is thatone cannot obtain completely independent obser-vations (that is, regions in this study) in terms of S&Tinput–output correspondence. One must recognizethat there would be collaborations and communicationsin S&T activities among those chosen regions. In thissituation, the efficiency measurements would beimproved by collecting more detailed measures infuture work. The second limitation is that this studyonly included various physical and discretionary(controllable) resource-based innovation inputs, andit neglected the exogenous effects of region-specificnon-discretionary (uncontrollable) structural or contex-tual factors related to innovation infrastructure, policysetting as well as economic environment (for example,industrial structure) in efficiency estimations. Onekind of potential study in this context is to incorporateimportant non-discretionary structural or contextualfactors (for example, BROEKEL and BRENNER, 2007;and BRENNER and BROEKEL, 2009) into the presentnetwork DEA model from the systems perspective.However, uncontrollable factors influencing the per-formance of a production unit are out of the controlof the management. The direct method includinguncontrollable environmental variables in the DEAmodels together with physical innovation inputs andoutputs requires a prior understanding of the influencedirection of an uncontrollable variable (YANG andPOLLITT, 2009). Furthermore, in this situation, someinefficient decision-making units (DMUs) mightbecome efficient as the number of the uncontrollablevariables included increases (YANG and POLLITT,2009). Therefore, the direct method is not advisablein the context of DEA-based measurement. Based onthe extant study, there are two motivations worthy ofinvestigation. One is to investigate the effects of struc-tural or contextual factors on the efficiencies of the
whole regional innovation process and the two individ-ual component process variations by a second-stepregression (FRITSCH and SLAVTCHEV, 2010). There isa valuable subsequent study for the regional innovationpolicy development to determine cross-regional effi-ciency variation for the three individual processes.The other motivation is to follow FRIED et al.’s (1999,2002) measurement technique for pure regional inno-vation efficiency removing the exogenous effects ofthe operating environment determined by structuralor contextual factors (WANG and HUANG, 2007).Although the latter will face a challenge in thecontext of the network DEA measurement, it wouldprovide an unbiased efficiency estimation of regionalinnovation processes. The third limitation is attributableto the deterministic property of the network DEAmodel. As indicated by the diagnostic analysis for theinfluences of outliers and statistical noise on innovationefficiency estimations, the efficiency measurement inthe context of deterministic DEA models is biased.Thus, it is necessary in a future study to employ somerobust non-parametric approaches (for example,SIMAR and WILSON’s, 2008, order-m frontier analysis)as the estimation techniques for the current networkDEA model, although such a study would also facemore challenges in the context of the network DEAmeasurement.
Acknowledgements – This research was funded by the
National Social Science Foundation of China (Project
Number 08BJY031), the National Natural Science Foun-
dation of China (Project Number 70773006), and the Shang-
hai Leading Academic Discipline Project (Project Number
B210). The authors are very grateful for the valuable com-
ments and suggestions from the anonymous reviewers and
the Editors, which significantly improved the article. The
authors also thank Professor Ronald Rousseau, KHBO,
Industrial Sciences and Technology, Belgium, for his
English corrections to this paper. The authors’ names are
ordered alphabetically and they contributed equally to the
writing of this paper.
Efficiency of China’s Regional Innovation Systems: Application of Network DEA 369
Dow
nloa
ded
by [
Nat
iona
l Sci
ence
Lib
rary
] at
21:
44 2
0 M
arch
201
2
user
高亮
APPENDIX
Table A1. Each regional innovation system’s Charnes, Cooper & Rhoades (CCR) efficiency in each stage during the three indi-vidual production periods
Region
Period 1: 1995–1999 Period 2: 1999–2003 Period 3: 2003–2007
Mean 0.9836 0.8423 0.9734 0.9679 0.9459 0.9423 0.9840 0.8659 0.9628
Source: Authors’ calculations.
370 Kaihua Chen and Jiancheng Guan
Dow
nloa
ded
by [
Nat
iona
l Sci
ence
Lib
rary
] at
21:
44 2
0 M
arch
201
2
Table A2(a). Out-oriented constant returns to scale (CRS) efficiencies and confidence intervals for the stage 1 (technological development process) by using the SIMAR and WILSON (1998)bootstrap method
Table A2(b). Out-oriented constant returns to scale (CRS) efficiencies and confidence intervals for the stage 2 (technological commercialization process) by using the SIMAR and WILSON