China Economic Review 67 (2021) 101605

Contents lists available at ScienceDirect

China Economic Review

journal homepage: www.elsevier.com/locate/chieco

Establishing science parks everywhere? Misallocation in R&D and

its determinants of science parks in China
Chih-Hai Yang a, *, Wen-Chieh Lee b
a
National Central University, Taiwan
b
National Chengchi University, Taiwan

A R T I C L E I N F O A B S T R A C T

JEL classification: The establishment of science parks is a vital strategy to develop high-tech industries and facilitate
H76 innovations in China. The success of a science park depends heavily on its supportive environ­
O33 ment, suggesting that it is hard to replicate everywhere, while China has established many science
O38
parks across regions in the past decade. This study evaluates the degree of misallocation in
R11
R58
research and development (R&D) and its determinants across science parks in China. Based on an
unbalanced panel data of 145 science parks for the period 2007–2014, we find that the overall
Keywords:
R&D efficiency has decreased sharply since 2011 when China began to initiate many new science
Science Park
R&D parks. The newly constructed science parks exhibit a lower R&D efficiency than their incumbent
Misallocation parks, suggesting a considerable misallocation in R&D resource caused by expanding science
parks everywhere. We further investigate the determinants of R&D misallocation and find that
park characteristics and environmental characteristics matter. Parks which are larger, older, and
having a higher human quality experience a lower R&D misallocation. Parks with closer R&D
collaboration with universities or research institutes, particularly with universities, exhibit a
lower R&D misallocation.

1. Introduction

Inspired by the success of spontaneous industrial clusters in the United States, such as Silicon Valley and Route 128, the idea of
science parks is at the center of establishing high-tech industries, upgrading the technological ladder, and promoting regional eco­
nomic growth in many countries. Consequently, an increased number of science parks (about 1500) were established worldwide up to
the mid-2000s (Wainova, 2009), and the government developed most of them.
There are extensive taxonomies of science parks based on their structure and missions over the past four decades (Bigliardi, Dormio,
Nosella, & Petroni, 2006, Fig. 1). One consensus, reached after the 1990s, is that the establishment of science parks favors locations
near universities and public research institutes. The frequent interaction among the innovation triple helix of industry-university-
government relations generates technological externality and agglomeration economies (Fan & Scott, 2003; Westhead & Batstone,
1998), which promotes innovation and regional economic growth. However, Massey, Quintas, and Wield (1992) and Quintas, Wield,
and Massey (1992) question the effectiveness of the science park model on facilitating innovation and describe science parks as high-
tech fantasies. That is, science parks might have no quantitatively significant impact on their tenants’ performance because science
parks do not encourage on-park firms to create synergies (Macdonald, 1987), whereas more innovative firms self-select to enter science

* Corresponding author at: Department of Economics, National Central University, 300, Jhongda Road, Jhongli District, Taoyuan 320, Taiwan.
E-mail addresses: [email protected] (C.-H. Yang), [email protected] (W.-C. Lee).

https://doi.org/10.1016/j.chieco.2021.101605
Received 6 February 2020; Received in revised form 17 June 2020; Accepted 14 February 2021
Available online 20 February 2021
1043-951X/© 2021 Elsevier Inc. All rights reserved.
C.-H. Yang and W.-C. Lee China Economic Review 67 (2021) 101605

Fig. 1. The Number of STIPs in China, 1988–2015.

parks.
Extant empirical studies also show mixed findings on innovation performance differentials between in-park firms and external
counterparts, e.g., Colombo and Delmastro (2002), Diez-Vial and Fernández-Olmos (2017), Fukugawa (2006), Liberati, Marinucci, and
Tanzi (2015), Löfsten and Lindelöf (2002), Vásquez-Urriago, Barge-Gil, Rico, and Paraskevopoulou (2014), and Yang, Motohashi, and
Chen (2009). Albahari, Barge-Gil, Pérez-Canto, and Modrego-Rico (2018) illustrate science parks’ heterogeneity being relevant to their
tenants’ performance, suggesting that, through services provision and location, science park characteristics can mediate and facilitate
the advantages of cluster dynamics; thereby affecting science park performance. The implication is that the effectiveness of science
parks depends on their heterogeneity. However, comparative analyses on various aspects of performance across science parks are
almost absent: there are only a few science parks in most countries, which prevents rigorous econometric analyses.
Although the evidence of success is not robust, science parks remain a widely-adopted policy tool in emerging economies. From an
economic-geographical perspective (where a science park and its surrounding region form an entity), those parks with high-tech firms
have an evolving structure of inter-firm linkages and agglomerative effects that help promote R&D productivity and output. Devel­
oping technological capability through the cluster of high-tech startups in a superior innovation environment can be an authority’s
primary policy. Based on this assumption, China has established more than 100 science parks in the past two decades.
If the science park policy is successful, can government transplant this success to other regions? Indeed, China’s local governments
generally consider science parks as “development catalysts” rather than “innovation catalysts.” Both employment and income effects
brought about by the establishment of science parks are the key goals of local government, as regional economic growth is the primary
concern (Hu, 2007). The widespread setups of science parks thus comprise many seemingly science parks, which lower the functions
science parks are expected to play in facilitating innovation.1 The concept of science parks has become a popular policy tool that
enhances innovation and regional development; evaluating the performance of science parks and identifying their determinants are
the essential issues of the science parks policy.
However, R&D resource misallocation, an essential aspect of understanding how well a science park operates, is yet to be inves­
tigated in the literature. Restuccia and Rogerson (2008) indicated that the allocation of resources across firms may be a significant
factor in affecting a country’s total factor productivity (TFP). Resource misallocation thus has been identified as a significant source of
variation in productivity in emerging economies (Syverson, 2011). In developing countries, R&D is one of the scarce resources for
innovation, promoting productivity, and raising the value added of production. Better use of scarce resources can efficiently boost
productivity and sustain long-term growth. Directing R&D resource toward government initiated science parks that are located in
cities without favorable innovation environments, and well-functioning regional innovation can potentially result in R&D resource
misallocation; raising the need for research on the evaluation of different efficiencies in the allocation of resources across science parks.
Together, with an understanding of the determinants of R&D misallocation, such research can provide insight into the implications of
technology and regional policies.
This study aims to evaluate R&D misallocation and its determinants across science parks in China. It adds three novelties to this line
of literature. First, as some countries have established dozens of science parks in the past few decades,2 studies comparing performance
across science parks have emerged recently, but have remained limited. For example, Hu, Han, Yeh, and Lu (2010) and Izadikhah and
Saen (2015) adopt the data envelopment analysis (DEA) approach to evaluate the efficiency of science parks in China and Iran,
respectively. Unlike previous studies focusing on evaluating performance differentials between in-park and off-park firms, this park-
level study helps understand the degree and disparity in R&D misallocation across science parks. A counter-factual welfare analysis is

1
Luger and Goldstein (1991) argue that there is no pure science or research park, because STPs shoulder multiple missions.
2
Please see statistics of the United Nations Educational, Scientific, and Cultural Organization (UNESCO) for science parks around the world.
http://www.unesco.org/new/en/natural-sciences/science-technology /university-industry-partnerships/science-parks-around-the-world/

2
C.-H. Yang and W.-C. Lee China Economic Review 67 (2021) 101605

also conducted to simulate the loss of R&D outputs due to research input misallocation on the new established parks.
Second, evaluating the performance of science parks is a rather complex undertaking. Instead of using the DEA approach to assess
the technical efficiency of science parks in existing studies, this study evaluates the R&D resource misallocations of science parks.
Analyzing resource misallocation is an emerging field of literature, while it neglects the misallocation of innovation resource, except
for Li, Lee, and Ko (2017). Establishing many science parks in unfavorable innovation environments (regions) might lead to a
distortion of the R&D resource usage and an association with a larger R&D misallocation. Building on Hsiech and Klenow’s (2009)
methodology into misallocation, we calculate the R&D efficiency of science parks. This efficiency measure assumes a larger value if the
dispersion of revenue productivity, which is a function of a science park’s input and output distortions, is smaller across science parks.
This misallocation measure on “revenue productivity” is called TFPR in Foster, Haltiwanger, and Syverson (2008) and Hsiech and
Klenow’s (2009).
Third, on computing the R&D efficiency measure, we have also obtained individual science park’s TFPR, which can be viewed as an
inverse measure of distortion (misallocation). We further investigate the determinants of misallocation across science parks. As the
science park theory is underpinned by the advantage of agglomeration economies and the network externality of the innovation triple
helix, existing studies focus on factors, like university proximity, interaction within the triple helix, and others, for example, Link and
Scott (2003), Fritsch and Franke (2004), Yang et al. (2009), and Jongwanich, Kohpaiboon, and Yang (2014). Even though science park
characteristics are a crucial dimension of the influential factors affecting performance, they are not well examined. Albahari et al.
(2018) indicate that heterogeneity of science parks is relevant to their tenants’ performance and affects park performance. However,
how science parks’ characteristics relate to their performance is less investigated. This study explores how the park’s characteristics
relate to science parks performance in terms of R&D resource misallocation. Using returns to scale as a conceptual foundation, we also
explore whether and how the growth R&D inputs relate to R&D misallocation.
The remainder of this paper is organized as follows. Section 2 briefly introduces the development and importance of science parks
in China. Section 3 presents the methodology and demonstrates the data source. Section 4 presents and discusses the R&D misallo­
cation across science parks in China and conduct a counter-factual welfare analysis to simulate the R&D misallocation if considering
non-park regions. Section 5 illustrates the empirical results of the determinants of R&D misallocation. Robustness checks are also
conducted. The final section summarizes the concluding remarks and policy implications.

2. The development and importance of science parks in China

2.1. Development of science & technology industrial parks in China

China’s science park policy was initiated in 1988 when the State Council implemented the Torch Program that aims to promote
innovations and develops high-tech industries. One of the foremost strategies is establishing National High-tech Industrial Zones
(hereafter, NHIZs or science parks). Like science parks in advanced countries, the NHIZs were established in suitable locations,
neighboring renowned universities and research institutes, to provide a catalytic-incubator environment. Through stringent entry
criteria, together with liberal policy measures such as tax incentives, R&D grants, and financial assistance to select and support R&D-
intensive high-tech startups, NHIZs are tasked with transforming industrial structure by supporting high-tech firm startups and
fostering technological diffusion (Hu, 2007).
Zhongguancun Science Park was the first national NHIZ that was launched in Beijing in 1988. At that time, Beijing was not an
economic heavyweight of high-tech firms: its selection was attributed to the cluster of research universities and institutions and,
probably also the capital bias from a political correctness point. There followed two waves in the establishment of NHIZs in the 1990s
and 2010s.
In 1991, the State Council approved the establishment of the first group of 24 NHIZs and then approved another 27 NHIZs in 1993,
comprising a total of 52.3 As depicted in Fig. 1, the number of NHIZs increased slowly to 53 in 1997, and 54 in 2007. The first-wave of
the rapid upsurge in NHIZs was in response to Deng Xiaoping’s 1992 “South Trip” talk that declared the intent to speed up the pace of
the open-door policy. NHIZs are utilized as a policy measure to promote domestic high-tech startups as well as attract the inflow of
foreign high-tech firms. The sites of these 54 NHIZs are located either in the largest cities or in business cities to shape a favorable
environment.
The period from 1997 to 2007 was a golden decade for China’s economic development. The achievement of persistent high
economic growth was the most urgent and immediate priority for individual provinces. The building of provincial science and
technology industrial parks and specialized industrial bases to accommodate domestic and foreign firms helps provinces to create jobs
and increase output, thereby helping to reach provincial GDP growth targets. The State Council thus stopped approving the estab­
lishment of new NHIZs and instead focused on reorganizing existing NHIZs, resulting in a stable number of NHIZs.
At the center of China’s 11th Five-Year Plan (2006–2010) initialized in 2006, one of the targets was scientific development, aimed
at achieving technological self-reliance and becoming an innovation-oriented nation. However, China’s economy was also severely
affected by the 2008 financial crisis. The government rethought the NHIZs’ key role in facilitating high-tech industry development and
sustaining economic growth. It upgraded two provincial industrial parks to the national-class science parks in 2009 and then approved
the second-wave of NHIZs by improving existing provincial industrial parks or specialized industrial bases. Fig. 1 illustrates that the

3
The Tiananmen Square protests of 1989 delayed the progress of STIPs project by a few years.

3
C.-H. Yang and W.-C. Lee China Economic Review 67 (2021) 101605

number of NHIZs increased sharply from 56 in 2009 to 83 in 2010, 88 in 2011, 105 in 2012, and 114 in 2013, and finally stabilized at
115 in 2014 and later.
After the second-wave increase, NHIZs are now widespread in every province, except for Tibet. However, their geographical
distribution is unequal and is concentrated mainly in coastal regions, followed by central regions. As demonstrated in Table 1, Jiangsu,
Shandong, and Guangdong rank the top three provinces by owning 15, 12, and 11 NHIZs respectively, whereas Beijing, Tianjin,
Hainan, Guizhou, and Qinghai each have only one NHIZ. Among them, Beijing and Tianjin are municipalities; Guizhou and Qinghai
are less developed provinces, and Hainan is an off-shore island.
As indicated in Columns (1) and (2) combined, the number of new NHIZs in the second-wave (91) is much more than the con­
structed NHIZs in the first-wave (54). Crucially, without a well-developed innovation system, local governments may prioritize science
park initiatives to promote economic growth, meaning that the second-wave established or upgraded NHIZs might locate in an
innovation environment without enough triple helix interactions that the first-wave NHIZs have due to their location. For example,
both Beijing and Qinghai each established only one NHIZ, while these two NHIZs might have diverse performance due to location, park
characteristics, and others. It casts the doubt on whether it is necessary to initiate NHIZs everywhere, and to what extent there is R&D
resource misallocation.
Along with the increase in the number of NHIZs and the number of firms in each park, NHIZs have played an emerging and critical
role in the Chinese economy. Table 2 shows that the output increased approximately 60 times from RMB 310.95 billion in 1997 to RMB
18,601.83 billion in 2015. The corresponding share of GDP increased from 3.90% to 27.14%. Commodities exported from NHIZs
accounted for US$ 6.48 billion in 1997, and 4.08% of total exports. By integrating itself into the world trading system after entering to
the World Trade Organization (WTO) in 2001, China acts as the primary manufacturing center for the East Asian production network
and has gradually become the leading exporter of a variety of products. It has seen a sharp increase in export value to US$ 473.27
billion in 2015, accounting for 20.82% of total exports.
Because the design of NHIZs accommodates R&D-intensive high-tech firms, the tenants naturally devote more efforts to R&D than
their outside counterparts. The outlay of R&D rocketed by more than 29 times from an initial RMB 15.52 billion in 2000 to RMB 452.16
billion, in 2015, experiencing an average annual growth rate of 26.29%. The far right column of Table 2 illustrates that NHIZs play a
critical role in China’s private R&D activity. Their R&D expenditure accounted for 26.30% of total business R&D expenditure in 2000
and, then, increased to an extraordinarily high ratio of 51.66%, in 2007. One surprising phenomenon observed is that although the
number of NHIZs has increased since 2009, the corresponding R&D ratio dropped considerably to 32.26% in 2009 and then steadily re-
climbed to 42.70% in 2015. The possible causes are three-fold. First, the 11th Five-Year Plan (2006–2010) executed other projects for
establishing university technology zones, software zones, and zones for returned talented people to start businesses. The springing up
of new parks also accommodated R&D-intensive firms. Second, in the context of becoming a knowledge economy, firms located
outside NHIZs began to focus more on R&D activity. Finally, and most interestingly, from an R&D perspective, is resource misallo­
cation. The second-wave upgraded NHIZs might spend less on R&D expenditure because their locations are not so favorable for the
innovation network; tenants thus concentrate more on production.

3. Methodology and data

3.1. Measure of R&D misallocation across science parks

The seminal work in Hsieh and Klenow (2009) develops the concept of input distortion in producing outputs to calculate resource
misallocation. By adopting the core ideas of Hsieh and Klenow’s (2009) methodology, Li et al. (2017) estimate a knowledge production
function to evaluate the so-called innovation resource misallocation. Thus, this paper measures the R&D misallocation by estimating a
knowledge production function to compute R&D efficiency across science parks in China.
Assuming that there is a competitive innovation system in China where R&D resources are employed to produce a homogeneous
product in terms of value-added.4 Y is the aggregate value-added across M science parks in China and can be specified as follows:
∑M
Y= i=1
Yi (1)

where Yi is the value-added of science park i. We assume the NHIZ’s innovation production technology is determined by the following
decreasing return-to-scale technology5:
( )γ
Yi = Ai Lαi Ki1− α , γ ∈ (0, 1) (2)

where γ governs a park’s “operative returns to scale” in the innovation system.

Science parks within the country could be heterogeneous in both their innovative technology Ai and in the distortions associated

4
In innovation literature, the number of patents is widely adopted as a measure of innovation output, for example, Griliches (1990) and Nagaoka,
Motohashi, and Goto (2010), where it is obvious that patents are heterogeneous. Either the R&D outputs are processes or products, it aims to raise
the value-added of outputs. Hence, value-added is an adequate measure, which is in concordance with Hsieh and Klenow’s (2009) concept.
5
The innovation production function is assumed to be decreasing returns to scale, as characterized in Jones and Williams (2000) and Weil (2013).
This operative return-to-scale is also referred to as the “span-of-control” parameter described in Lucas (1978). Hsieh and Klenow (2009) have proved
the isomorphic property between a constant returns to scale model with differentiated goods and a Lucas span of control model formation.

4
C.-H. Yang and W.-C. Lee China Economic Review 67 (2021) 101605

Table 1
Geographical Distribution of STIPs in China, 2015.
Province -1 -2 Region

Total Number of NHIZs Number of NHIZs in 2007

Jiangsu 15 4 East Coast

Shandong 12 5 East Coast
Guangdong 11 6 East Coast
Zhejiang 8 2 East Coast
Liaoning 8 3 East Coast
Fujian 7 2 East Coast
Hubei 7 2 Center
Henan 7 2 Center
Sichuan 7 2 West
Jiangxi 7 1 Center
Shaanxi 7 3 Center
Hunan 6 2 Center
Hebei 5 2 East Coast
Jilin 5 2 North
Anhui 4 1 Center
Guangxi 4 2 West
Heilongjiang 3 2 North
Xinjiang 3 1 West
Shanghai 2 1 East Coast
Shanxi 2 1 Center
Neimenggu 2 1 North
Gansu 2 1 West
Ningxia 2 0 West
Chongqing 2 1 West
Yunnan 2 1 West
Beijing 1 1 East coast
Tianjin 1 1 East Coast
Hainan 1 1 East Coast
Guizhou 1 1 West
Qinghai 1 0 West
Tibet 0 0 West
Total 145 54

Source: calculated by the authors.

Table 2
Basic Statistics of STIPs.
Year Output (RMB Export (US$ R&D (RMB Output to GDP ratio Export to total exports ratio R&D to business R&D ratio
billion) billion) billion) (%) (%) (%)

1997 310.92 6.48 n.a. 3.9 4.08 n.a.

1998 433.36 8.53 n.a. 5.09 5.22 n.a.
1999 594.4 11.9 n.a. 6.56 6.81 n.a.
2000 794.2 18.58 15.54 7.92 8.31 26.3
2001 1,011.68 22.66 22.18 9.13 9.45 35.22
2002 1,293.71 32.92 31.45 10.63 11.08 39.91
2003 1,725.74 51.02 41.95 12.56 12.65 43.67
2004 2,263.89 82.38 61.38 13.99 14.9 47.53
2005 2,895.76 111.65 80.62 15.46 15.66 49.08
2006 3,589.90 136.1 106.42 16.36 14.86 51.32
2007 4,437.69 172.81 134.88 16.42 14.95 51.66
2008 5,268.47 201.52 166.82 16.49 14.9 50.38
2009 6,115.14 200.72 134.27 17.52 17.63 32.26
2010 8,431.82 264.8 181.25 20.41 23.14 35.8
2011 10,567.96 318.06 226.9 21.6 16.8 35.34
2012 12,860.39 376.04 274.91 23.8 18.4 36.05
2013 15,136.76 413.33 348.88 25.43 18.7 39.48
2014 16,993.69 435.14 399.57 26.39 18.5 40.7
2015 18,601.83 473.27 452.16 27.14 20.82 42.7

Note: n.a. denotes data being unavailable.

with the use of R&D capital and R&D labor. Following the assumptions in Li et al. (2017), NHIZs might experience two types of
distortions: output distortion dYi, which simultaneously affects productivity of R&D capital and R&D labor; R&D capital distortion dKi,
which drives up the productivity of R&D capital relative to that of R&D labor. A science park’s innovation payoff is given by

5
C.-H. Yang and W.-C. Lee China Economic Review 67 (2021) 101605

πi = (1 − dYi )Pi Yi − wLi − (1 + dKi )RK i (3)

Under the assumptions of a competitive R&D input market and a homogeneous output market, the price of a product is as follows,
assuming the standard first-order condition on demand holds:
Pi = P, for every science parki (4)
To solve the R&D input demand in perfectly competitive factor markets: the derived demand from (4) can be substituted into (3) to
solve for both the input demand and the output supply, whose values are determined by the innovation technology Ai and the
distortion measures dYi and dKi:
Li = Li (Ai , dYi , dKi ) (5)

Ki = Ki (Ai , dYi , dKi ) (6)

Science parks experiencing greater output distortions (dYi) and capital distortions (dKi) will respectively exhibit higher marginal
revenue products of R&D labor and R&D capital, because their production pursues profit maximization. We denote the marginal
revenue products as follows:
1
MRPLi = MRPLi (Ai , Li , Ki ) ≈ w (7)
1 − dYi

1 + dKi
MRPK i = MRPK i (Ai , Li , Ki ) ≈ R (8)
1 − dYi
Given the assumption of decreasing returns, highly distorted NHIZs will have an equilibrium scale of production that is smaller than
the optimal scale.
Following the concept introduced in Foster et al. (2008) and Hsieh and Klenow (2009), Li et al. (2017) differentiate “physical
productivity” from “revenue productivity”. The former is denoted as TFPQ and it is NHIZ-specific, whereas the latter is TFPR and it is
country-specific if there is no difference in the extent of the distortions across science parks. The reduced forms of TFPQ and TFPR for
the science park i can be solved as follows:
Yi
TFPQi ≈ w ( )γ (9)
Lαi Ki1− α

PY i
TFPRi ≈ w (10)
Lαi Ki1− α

If NHIZs can be initiated anywhere, that is, establishing science parks in many cities, TFPR will be country-specific, and there is no
difference across science parks. However, TFPR could vary across science parks if they have different levels of output and capital
distortions. Without the NHIZ-specific output and capital distortions (TFPQi), TFPRi can be represented by the geometric average of a
science park’s marginal revenue products of labor and capital, as indicated in Hsieh and Klenow (2009) and Li et al. (2017). Spe­
cifically, we use Eqs. (7), (8), and (10) to show that a park’s TFPR is, in effect, an indicator of the endured distortions:
[( )α ( )1− α ]γ
MRPLi MRPK i
TFPRi = TFPRi (dYi , dKi )∝
w R
[ ( )1− α ]− γ (11)
1 − dYi
∝ (1 − dYi )α
1 + dKi

Because higher R&D outputs and larger R&D capital distortions raise the marginal products of R&D capital and R&D labor, the
science park i will exhibit a smaller scale of output than the efficient scale if it experiences a larger distortion.
Aggregate R&D output (value-added) can be derived by simply aggregating the individual science park’s R&D output production,
as in eq. (1). Suppose we implicitly define the innovation production efficiency TFP of the country as a whole by:

Y = TFP × Lα × K 1− α
(12)
∑ ∑M
where L = M i=1Liand K = i=1Ki represent the aggregate values of the R&D labor and R&D capital devoted to innovation activities,
respectively. By simplifying the linear aggregate of the production function in the innovation system, the countrywide innovation
production efficiency TFP can be represented by
[ ( )1−1 γ ]1− γ
∑M TFPR
i=1 TFPQi TFPRi
Y
TFP = = ( ) (13)
Lα K 1− α α 1− γ
Lαi Ki1−

where TFPR is a harmonic average of the average marginal revenue product of R&D capital and R&D labor across science parks in

6
C.-H. Yang and W.-C. Lee China Economic Review 67 (2021) 101605

China.
Eq. (13) shows that the countrywide R&D TFP will be that of a CES function aggregated across all of the TFPQi if the revenue
productivity (TFPRi) is equalized across all science parks in China. In this special case, TFP will be
[∑M 1−1 γ ]1− γ
A
TFP = A = ( i=1 i )1− γ (14)
α 1− α
Li Ki

3.2. Computation of misallocation

To compute R&D efficiency, we adopt the following exogenous parameters: First, we assume a rental rate of R&D capital being R =
0.23, following the assumption in Li et al. (2017). The rental rate is a combination of an interest rate (i) of 3% and a depreciation rate
(δ) of 20%.6
Next, we turn to the choice of Lucas span-of-control parameter γ, and the labor share parameter α.7 Previous studies on estimating
the knowledge production functions, such as, Crépon and Duguet (1997), Ramani, EI-Aroui, and Carrere (2008), and Hu and Jefferson
(2009), have shown that innovation production is governed by decreasing returns to scale. Accordingly, in studies estimating firm-
level innovation production function, for example, Jefferson, Bai, Guan, and Yu (2006) and Yang (2018), finding the mean innova­
tion elasticity of R&D inputs being approximately 0.8, we thus choose γ = 0.8.8 The labor share parameter is assumed to be 0.6 based
on the province-level evidence in Li (2009) and Bai (2013) that finds labor share in innovation production is approximately 0.6.
Table 3 summarizes our parameter configuration.
Following Li et al. (2017), the idiosyncratic distortions in R&D labor and R&D capital adoption costs and TFPQs across science
parks are denoted as follows:
1− α wLi
Capital distortion : dKi = × − 1 (15)
α RK i

1 wLi
Output distortion : 1 − dKi = × (16)
αγ RK i
In Eq. (15), if the ratio of the labor share to the capital share is greater than α/(1 − α), we can infer that R&D capital distortion
exists; Eq. (16) illustrates that if the R&D labor share relative to the total output is smaller than αγ, we have output (value-added)
distortion. The TFPQi measurement in Eq. (17) is conceptually similar to the TFP in a neoclassical production function. Indeed, these
measurements of distortion and science parks productivities are the bases for us to gauge the efficiency loss of the innovation system
for science parks.
Li et al. (2017) define the “efficient production” as the output level obtained when there are no idiosyncratic distortions across
science parks. Under the optimal scenario, the marginal revenue products of the innovation inputs are equalized across science parks
within the innovation system in China; thus,

TFPRi = TFPR (17)

We can rewrite TFP in Eq. (14) as
[∑M 1−1 γ ]1− γ
A
A = ( i=1 i )1− γ (18)
Liα Ki1− α

The ratio of the actual and efficient production levels of innovation output is thus denoted as9:
[ ( )1−1 γ ]1− γ
Y ∑M Ai TFPR
YR = = (19)
Yefficient i=1 A TFPRi

Eq. (19) demonstrates that the ratio YR increases as the dispersion of a science park’s TFPR decreases. It reaches the maximum value
(=1) when all science parks’ marginal payoffs of R&D inputs are equalized.10 Thus, YR can be viewed as a measure of innovation

6
The interest rate is about 3% within the sample period in China. Moreover, the R&D depreciation is estimated to hover between 11% and 36% in
various studies (Nadiri & Prucha, 1996).
7
The span-of-control parameter (γ) records the operative returns to scale by labor (L) and fixed capital (K). This operative return-to-scale
parameter is often referred to as the “span-of-control” parameter, as in Lucas (1978), Atkeson and Kehoe (2005) and many other studies. In the
current context, our selection of γ can be viewed as replacing the elasticity of the substitution measure in Hsieh and Klenow (2009), and the gains
from fewer distortions are increasing in γ.
8
For robustness, we also consider alternative values of γ: 0.5 and 0.9. Different choices of γ will affect only the numerical values of the measured
R&D productivity and not the relative ordering or the trend in productivity.
9
The subscript “efficient” means the removal of all idiosyncratic barriers or frictions that cause disparities in the marginal products of labor and
capital.
10
This result has been shown in Hsieh and Klenow (2009).

7
C.-H. Yang and W.-C. Lee China Economic Review 67 (2021) 101605

Table 3
Parameters Used in Calibrations.
α R=δ+i γ

δ i

Parameter values 0.60 0.20 0.03 0.8

Note: The parameters are the same as assumed in Li et al. (2017).

efficiency.

3.3. Determinants of misallocation

After calculating the degree of R&D misallocation, this study next investigates the determinants of misallocation in science parks.
Considering the cross-sectional feature of misallocation comparison, we specify the empirical models as follows
RD misit = α + Xit β + Zit γ + PROV ρ + YEARδ + εit (20)
( )
Here, the dependent variable RD_mis is the degree of misallocation, which is measured byln TFPRit /TFPRt , denoting the deviation
of park i’s TFPR from the mean TFPR in year t. A higher value represents a larger distortion (misallocation). Covariates include two
vectors of factors, mainly science park characteristics (X) and environmental factors (Z). PROV and YEAR are a series of province
dummies and year dummies which are used to control for province fixed effect and macroeconomic shocks on R&D efficiency. The
term ε denotes the white noise error term.
As indicated in Albahari et al. (2018), heterogeneity in science parks affects tenants’ performance: it relates to the park’s R&D
efficiency. Science park characteristics we consider are discussed next. SIZE denotes park size, which is measured by the number of
employees in a science park. Larger parks have potential R&D synergy on R&D output and are more efficient in R&D resource usage.
Moreover, the cluster of more high-tech firms may generate an R&D spillover effect that has a positive effect on the tenants’ R&D
efficiency. Therefore, we expect a negative sign associated with the SIZE variable. New_Park is a dummy variable equaling one if a
science park was established after 2010. From a managerial viewpoint (e.g., Mccann & Folta, 2011), the services provision and
management experience could be relevant to technical efficiency. New_Park is included to capture the managerial experience: more
experienced management of park offices can efficiently coordinate tenants and foster R&D activity, leading to a lower misallocation.
Exports also matter to innovations. Exports might facilitate innovation through learning, competition, and customer feedback.
Yang (2018) has witnessed a positive relationship between exports and R&D for Chinese firms, implying that exports could stimulate
the use of R&D resource more efficiently thereby lower the degree of misallocation. We thus include export intensity (EXP_ratio), which
is measured by the ratio of exports to commodity sale and expect it is expected to have a negative relationship with R&D misallocation.
While Yang (2018) argues that the R&D-enhancing effect of exports depends on the heterogeneity of exports: process exports are found
to have a negative association with R&D. The human capital measured by the ratio of university-educated labors to total employees in
a science park is UNIV_ratio. The success of R&D projects not only depends on the qualified R&D personnel, but also on the coordi­
nation between R&D personnel and higher-skilled employees to apply the R&D output. Therefore, human capital is an essential
prerequisite for efficiently using the R&D resource and is expected to help lower the R&D misallocation.
An essential issue for provincial market development is regional decentralization that empowers the provincial government to have
fiscal autonomy and aligns the interests of local governments with market development (Jin, Qian, & Weingast, 2005). A provincial
government facilitates the innovation performance of science parks mainly by establishing a well-functioning regional innovation
system because the triple helix of university–industry–government relations is widely known for playing a critical role in regional
innovation systems (Etzkowitz & Leydesdorff, 2000). Considering the central role of science parks (industry), we include two variables
of the triple helix. The RES_IND is the degree of government-industry R&D collaboration in a province that is measured as the ratio of
research institutes’ expenditure on science and technology activity financed by industry. The other is UNIV_IND, the degree of uni­
versity–industry R&D collaboration. Correspondingly, it is measured by the percentage of university expenditure on science and
technology activity financed by industry in a province. Jongwanich et al. (2014) argue that science parks play a crucial role in
coordinating R&D collaboration across various R&D performers within the region, and they indirectly contribute to upgrading the
regional technological ladder. It suggests that the network externality of innovation should be helpful to lower R&D misallocation of
science parks.
As R&D activities out of science parks are probably not neglectable in the spatial sense, it suggests that the difference in regional
price (distortion) probably counts for a major part of misallocation calculated for each park. To consider this influence, we adopt two
strategies: the first one is a set of province-year dummies and the other is the R&D misallocation of the province (Prov_distortion)
calculated by Li et al. (2017).11
Accordingly, Eq. (20) is rewritten as follows:

11
Thanks the authors for kindly providing the information regarding R&D misallocation of province they calculated. As the time periods in these
two studies are not same, the Prov_distortion enters the equation in the one-year lagged form and they are the same in the final two years.

8
C.-H. Yang and W.-C. Lee China Economic Review 67 (2021) 101605

RD misit = α + β1 lnSIZEit + β2 New Parkit + β3 EXP ratioit

+β4 UNIV ratioit + γ1 RES INDit + γ2 UNIV INDit (21)
+β5 Prov distortion + PROV ρ + YEARδ + εit

or
RD misit = α + β1 lnSIZEit + β2 New Parkit + β3 EXP ratioit
+β4 UNIV ratioit + γ1 RES INDit + γ2 UNIV INDit (22)
+(PROV*YEAR)γ + εit

3.4. Data source

In this study, the panel-structure dataset used is drawn from two Chinese databases. The first dataset is China Torch Statistical
Yearbook, published by the Ministry of Science and Technology of China. It covers related data of NHIZs, mostly production infor­
mation, financial information, and innovation statistics, which enabled us to calculate the R&D misallocation across science parks.
Moreover, it contains park characteristics that are related to R&D misallocation. As illustrated in Fig. 1, the number of NHIZs shows the
second-wave of substantial increases in the 2010s. We adopted an unbalanced panel data of 145 NHIZs for the 2007–2014 period to
compare the degree of misallocation between old NHIZs and upgraded new NHIZs.
To further examine the determinants of misallocation, we considered both park characteristics and environmental variables for
individual NHIZs. The information of the R&D cooperation between industry and research institute, as well as between industry and
university are drawn from various issues of the China Statistical Yearbook on Science and Technology. Moreover, we obtained the in­
formation of Table 4 to summarize definitions and basic statistics of variables.

4. Empirical results and discussions

Our empirical results will be presented in three parts. First, we compute the annual R&D efficiency of the science park system in
China during our sample period. Next, we look at the difference in the annual R&D efficiencies to see if there are any recognizable
trends. Finally, we discuss the determinants of R&D misallocation of science parks and try to identify the factors that co-move with the
input distortions.

4.1. R&D misallocation across science parks in China

Based on eq. (19), the efficiency measure is the ratio of the actual level of value-added output relative to the “efficient” value-added
production level in the denominator. The measure will take on a greater value if the dispersion of TFPRs is smaller across science parks
functioning in that year, or equivalently, the extent of distortions is similar across difference science parks. This efficiency measure is
computed annually between 2007 and 2014, and the result is presented in Fig. 2.
As depicted in Fig. 2, the efficiency measure increases remarkably during the sample period, starting from 0.42 in 2007 to 0.49 in
2008 and increase even higher at 0.62 in 2009. This increasing pattern of R&D efficiency reveals that although in 2007 about 42% of
the efficiency level of output (value-added) of science parks was realized, and for about three years the trend increasing from 42% to
62% of the efficiency level of output. These findings point to a roughly 20% oscillation in the range of input distortions across science
parks during those three years.
From Fig. 2, we can clearly observe a significant positive change in the measured R&D efficiency before 2010 (the efficiency
measure was highest 0.65), but since 2011 there seems to be a deterioration in efficiency.

4.2. Difference in innovation performance across areas

In Fig. 2, we observe overall large oscillations in the R&D efficiency between 2007 and 2014. Fundamental questions that arise are:
what are the differences in the efficiency of science parks and how do these differences evolve?
We use Fig. 3 and Table 5 to tackle this issue. In Fig. 3, we plot the annual “demeaned” TFPRs for all science parks, where a
demeaned value is calculated by subtracting a science park’s TFPR by the simple average of all TFPRs of science parks in that year. In
Fig. 3, we draw three horizontal lines. The middle line represents zero, so if a science park’s demeaned TFPR occurred on the line, it
implies that the science park’s TFPR equals the simple average of all TFPRs in that particular year. The upper and lower lines represent
the lines above and below one standard deviation of the demeaned TFPRs of all science parks “across all years,” respectively.12 So, if a
science park’s demeaned TFPR occurred above the upper line, it implies that the science park’s TFPR in that year is more than one
(across-years) standard deviation higher than the simple average of all TFPRs in that year; likewise similar interpretation could be
made for science parks’ demeaned TFPRs that are located below the lower line. For convenience, we will call the area below the lower

12
To calculate this “across-years” standard deviation of demeaned TFPRs, we first demeaned all science parks TFPRs by the simple average of all
science parks’ TFPRs in the corresponding years, then calculated the standard deviation of these science parks demeaned TFPRs. The main reason we
consider this “across-years” standard deviation instead of the yearly standard deviations is because we hope to highlight the convergence of the
science parks’ TFPRs throughout the sample period.

9
C.-H. Yang and W.-C. Lee China Economic Review 67 (2021) 101605

Table 4
Variable definitions and basic statistics.
Variable Definition Mean (Std. Dev.)
( )
RD_mis -3.9e-10 (0.994)
R&D misallocation of a science park, measured byln TFPRit /TFPRt
SIZE Park size: number of employees in a park. 130,323 (174,258)
New_Park New park dummy: 1 = parks established in 2011 and onward; 0 = others 0.346 (0.476)
EXP Export of a science park (RMB million) 3685 (6031)
EXP_ratio The ratio of exports to commodity sales of a park 0.170 (0.185)
SKILL_ratio Human capital: the ratio of skilled labors to total employees in a park 0.112 (0.062)
UNIV_ratio Human capital: the ratio of university-educated employees to total employees in a park 0.430 (0.158)
RES_IND The ratio of research institutes’ expenditure on science and technology activity financed by industries (%) 5.287 (2.118)
UNI_IND The percentage of universities’ expenditure on science and technology activity financed by industry 31.389 (10.745)
Prov_dsitortion Province price distortion: proxized by the dependent variable in Li et al. (2017) 1.030 (0.244)
Province wage Proxy variable of province price distortion (RMB thousand) 45.358 (13.238)

Note: Mean and standard deviation are calculated using the 2007–2014 data.

Fig. 2. R&D Efficiency (YR), 2007–2014.

Note: The productivity YR is computed based on Eq. (20), and utilizing the parameters listed in Table 3 .

line zone 1; the areas within one standard deviation below and above the middle line (which equals to zero) are zones 2 and 3,
respectively. The area above the upper line is zone 4.
There are several striking patterns in Fig. 3. First, we see that in earlier years, before 2010, that most science parks’ TFPRs lie in
zone 2 and zone 3, suggesting only a slight divergence in TFPRs across regions. Besides, we can observe that most of the science parks’
TFPRs are in the shallow places (less than 4 in the measure) places into zone 4 for those three years, resulting in a declining trend of
input distortion and increasing the trend of production efficiency. Second, we see that after 2010, TFPRs across regions diverge and lie
in deeper areas of zone 4 with measurements higher than 5. This divergent pattern is most evident in 2011 and 2012. However, in both
2011 and 2012, the TFPR of one science park lies in a very deep area of zone 4.13 Third, since 2010, TFPRs across science parks seem to
diverge continuously. For example, in 2012 there exists the TFPR located in very deep areas of zone 4.
Are there specific patterns of TFPRs over the years in China? To answer this question, in Table 5, we divide each column year into
two sub-columns (except for the first year 2007): old incumbent science parks and new entrant science parks. In this way, we can see
the patterns of dispersions of new science parks relative to the old parks for each year. Table 5 shows that three different periods
(2007–2008, 2009–2011, 2012–2014) exhibit different patterns for new entrant science parks, and we can see how the TFPR for each
science park evolves. In the 2009–2010 period in Table 5, which summarizes the distribution of TFPRs for each year for the entire
period, we see that there is a vast difference in the TFPR distribution across these two years.14 When we compare the TFPRs in different
periods, we find fascinating evolution patterns across the science parks. For science parks already existing in the first period
(2007–2008) and the third period (2012–2014), their TFPRs are less dispersed relative to the new entering science parks in middle
years (2009–2011), in which 50% (2009) and 16% (2011) of the TFPRs of the entrant science parks are in zone 4. However, since 2012,
the TFPRs of existing science parks converge toward the mean, and no new science parks are entering outside the middle zones. Similar

13
However, in 2007–2009, TFPRs located in zones 2 and 3 converge as well, so the measured innovation efficiency is better in that year than in
2019.
14
The definition of new is to summarize the dispersion of TFPRs for those science parks entering in that specific year.

10
C.-H. Yang and W.-C. Lee China Economic Review 67 (2021) 101605

Fig. 3. Distribution of Science Parks’ Demeaned TFPRs (2007–2014).

patterns can be observed in the first period (2007–2008). Thus, we can conclude that in 2009–2011, new entrant science parks have
more aggregated input distortions than those already existing science parks at any other time.

4.3. A counter-factual welfare analysis

Before we discuss what factors affect resource misallocation across science parks, some theoretical grounds are called for. Here we
would like to solidify the theoretical results we have reached in the previous sections. As noted in the misallocation computations, the
innovation input distortions have been found to degrade overall R&D efficiency, as shown in Fig. 2. We also observe the establishment
of many parks, starting from 2011. Subsequently-established parks recorded more input distortions (see Fig. 3 and Table 5). We
therefore doubt that the science parks established after 2011 would have inferior efficiency compared with the existing ones.
We conduct a counter-factual analysis that considers a hypothetical situation to see if post-2011 R&D efficiency (YR) maintains the
same YR as that observed previously. To conduct our hypothetically improved new park setup, the processes are as follows. First, we
categorize all the recorded parks into two groups: the old group of parks established before 2011, and the new group of parks
established starting from 2011 onward. Each park’s TFPQ of the old group in each sample year is calculated based on the Eq. (9). The
next step is picking the lowest TFPQ park of the old group for the years post-2011. The TFPQ of the new group of parks is replaced by
the lowest TFPQ park of the old group in the years post-2011. Based on the eq. (2), using the benchmarked new parks’ TFPQ, we can re-
compute each one’s R&D output belonging to the new group from each sample year, since each new park is now endowed with hy­
pothetical TFPQ and the original innovation inputs.
Utilizing hypothetically-derived R&D outputs of new park groups and the original innovation inputs, we can again compute the
R&D efficiency in the same way as we drew the original YR in Fig. 2. Fig. 2H illustrates the counter-factual results of YR. As depicted in
Fig. 2H, the R&D efficiency improved considerably post-2011, suggesting that even controlling the physical productivity of the new
parks, on the lowest level of the old parks, would significantly improve the aggregate research outputs. The key factor to improved
R&D outputs embedded within this hypothetical setup lies in eliminating the original innovation input distortions that exist ubiq­
uitously in the new parks. Therefore, we continue, so as to observe the distribution of innovation input distortions following this

11
 C.-H. Yang and W.-C. Lee
Table 5
Distribution of demeaned TFPRs across Science Parks.
Distribution of demeaned TFPRs across different science parks (number)

Zone 2007 2008 2009 2010 2011 2012 2013 2014

Old New Old New Old New Old New Old New Old New Old New Old New

Zone1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
Zone2 33 30 0 31 2 34 0 31 27 56 13 63 8 68 0
Zone3 15 18 0 16 0 15 0 14 2 27 1 28 0 27 0
Zone4 6 6 0 7 0 7 0 11 0 2 2 10 0 14 0
Total 54 54 0 54 2 56 0 56 29 85 16 101 8 109 0
12

Distribution of demeaned TFPRs across different science parks (ratio %)

Zone 2007 2008 2009 2010 2011 2012 2013 2014
Old New Old New Old New Old New Old New Old New Old New Old New
Zone1 2% 4% 13% 50% 14% 4% 0% 2% 3% 5%
Zone2 57% 57% 43% 0% 34% 86% 47% 66% 65% 60%
Zone3 33% 31% 31% 0% 34% 5% 38% 24% 20% 26%
Zone4 7% 7% 13% 50% 18% 5% 16% 8% 11% 9%
Total 100% 100% 100% 100% 100% 100% 100% 100% 100% 100%

Note: The values within each cell in the top panel show the frequency. The numbers in the lower panel show the share of the science parks’ demeaned TFPRs located in a particular zone during a specific
year.

China Economic Review 67 (2021) 101605

C.-H. Yang and W.-C. Lee China Economic Review 67 (2021) 101605

Fig. 2H. R&D efficiency (YR), Counter-factual Analysis.

counter-factual analysis.
Correspondingly, we also show the results of innovation input distortions in Fig. 3H and Table 5H, in order to compare with the
original results illustrated in Fig. 3 and Table 5.
Fig. 3H and Table 5H also show that the innovation input distortions improved analogously, leading to improved R&D efficiency, as
recorded in Fig. 2H. Besides, possessing better environmental factors endowed with new parks should generate greatly optimized
innovation input usage.
This section concludes that the worst resource-allocation situation occurs when the least efficient new parks (parks with lower
TFPQ compared with old parks) utilize a big chunk of available innovation inputs. That situation results in lower R&D output in the
equilibrium. Alternatively, remarkable improvement occurs whenever all of the park-level input distortions have been wiped clean to
reflect the optimal allocation based on firm-level TFPQ. Therefore, we determine that better control of new park quality promises
higher quality existing parks, leading to improved R&D productivity. Although this conclusion is not limited to the context of science
parks, it is that R&D discussion for which we have a ready illustration.

5. Determinants of R&D misallocation and R&D scale

5.1. Determinants of misallocation

Table 6 displays the estimates of various specifications to examine the determinants of R&D misallocation across science parks in
China. To begin with, Columns (1) and (2) are a baseline model that includes only park-specific characteristics, as emphasized in
Albahari et al. (2018). A series of province-year dummies are included in Column (2) to capture the price distortion across provinces in
various years.
How does heterogeneity in science parks affect their R&D efficiency (misallocation)? Results obtained using the RE or FE models
are similar.15 The coefficient of size is significantly negative, indicating that larger parks, in terms of the number of employees, have
lower R&D misallocation. As discussed previously, larger parks have potential R&D externality and a spillover effect; a cluster of more
high-tech firms results in parks experiencing a lower R&D misallocation than their smaller counterparts. Albahari et al. (2018) argue
that firms in larger parks have better innovation outputs; this implies that their R&D misallocation could be lower.
We find a significantly positive relation between New_Park and R&D misallocation in Columns (1) and (2), echoing the finding
demonstrated in the previous section that there is a lower R&D efficiency for newly established parks. This result is also consistent with
findings in Albahari et al. (2018) that older parks better perform on innovation outputs, because managerial experience may help
tenants promote R&D efficiency through administrative efficiency.
Exports, in terms of export value or export intensity, exhibit an insignificant relation with R&D misallocation. In other words,
export behaviors have no significant influence on raising or lowering R&D efficiency. Yang (2018) claims that serving the international
market can facilitate innovation through learning, competition, and customer feedback; while ordinary and process has a converse
influence on R&D in China, which are positive and negative, respectively. China’s exports concentrate on processing trade that exploits
the advantages of low-cost labor and tariff-free imported intermediate goods. Although the ratio of processing exports to total exports

15
Through Monte Carlo simulations, Bell and Jones (2015) show that the RE model is more adequate than FE model when the time-invariants
(rarely changed variables) are incorporated, such as the province dummy in Columns (1) and (3).

13
C.-H. Yang and W.-C. Lee China Economic Review 67 (2021) 101605

Fig. 3H. Distribution of science parks’ demeaned TFPRs: Counter-Factual Analysis.

Table 5H
Distribution of demeaned TFPRs: Counter-factual Analysis.
Distribution of demeaned TFPRs across different science parks (number)

Zone 2007 2008 2009 2010 2011 2012 2013 2014

Zone1 1 2 9 8 0 7 7 7
Zone2 31 31 23 20 48 45 47 49
Zone3 18 17 17 18 36 23 23 19
Zone4 4 4 7 10 4 12 11 13
Total 54 54 56 56 88 87 88 88

Distribution of demeaned TFPRs across different science parks (ratio %)

Zone 2007 2008 2009 2010 2011 2012 2013 2014
Zone1 2% 4% 16% 14% 0% 8% 8% 8%
Zone2 57% 57% 41% 36% 55% 52% 53% 56%
Zone3 33% 31% 30% 32% 41% 26% 26% 22%
Zone4 7% 7% 13% 18% 5% 14% 13% 15%
Total 100% 100% 100% 100% 100% 100% 100% 100%

demonstrated a decreasing trend in 2000s, it was still as high as 48.1% in 2008 (Koopman, Wang, & Wei, 2008), and this proportion
might be higher in high-tech industries. Therefore, export activity has no specific relation with R&D misallocation.
Human capital is relevant to R&D efficiency. Human capital not only has a positive impact on technological innovation (Gallie &
Legros, 2012), but also be complementary to firm R&D through absorption capability embedded in human capital. Moreover, higher
educated employees also help coordinate with R&D personnel. Parks with a higher ratio of higher educated employees use R&D
resource more efficiently, resulting in a lower R&D misallocation.
Regarding the linkage effect among triple helix constituent parties, Columns (3)–(6) illustrate results of various specifications. The

14
C.-H. Yang and W.-C. Lee China Economic Review 67 (2021) 101605

Table 6
Determinants of R&D Misallocation across Science Parks.
(1) (2) (3) (4) (5) (6)

RE FE RE FE FE FE

lnSIZE (EMP) − 0.231* (0.120) − 0.380* (0.197) − 0.233* (0.120) − 0.382* (0.197) − 0.575*** (0.133) − 0.581*** (0.134)
New_Park 1.108*** (0.215) 3.235*** (0.402) 1.109*** (0.216) 3.235*** (0.402) 2.952*** (0.332) 2.933*** (0.332)
lnEXP 0.054 (0.050) 0.054 (0.050)
EXP_ratio 0.241 (0.507) 0.241 (0.508) 0.161 (0.368) 0.171 (0.368)
UNIV_ratio − 0.764*** (0.276) − 1.684*** (0.577) − 0.840*** (0.282) − 1.665*** (0.579) − 1.051*** (0.288) − 1.138*** (0.294)
RES_IND 0.003 (0.007) − 0.005 (0.200) 0.001 (0.007)
UNI_IND − 0.009* (0.005) 0.013 (0.047) − 0.009* (0.005)
Prov_distortion − 0.292* (0.163) − 0.298* (0.164)
Constant 1.364 (1.830) 4.298* (2.409) − 0.130 (1.077) 3.972*** (2.533) 6.268*** (1.530) 6.719*** (1.561)
Year FE Yes No Yes No Yes Yes
Province FE Yes No Yes No No No
Province-year FE No Yes No Yes No No
R-square 0.298 0.539 0.299 0.539 0.227 0.231
Observations 571 571 571 571 571 571

Note: Figures in parentheses are heteroscedastic-consistent standard deviations. ***p < 0.01, **p < 0.05, *p < 0.1.

estimated coefficient of RES_IND is insignificant, whereas UNI_IND is, as expected, associated with a significantly negative coefficient
in most estimations. These results are similar to findings in Jongwanich et al. (2014) that there is a positive innovation-enhancing
effect brought about by R&D cooperation between industries and universities, rather than research institutes. Frequent interaction
through R&D collaborations enables high-tech firms in science parks to exchange ideas and research experiences with their R&D
partners. However, this positive network effect on R&D efficiency is not seen in R&D collaboration with research institutes. Con­
trastingly, the university–industry collaborations (UIC) seem to facilitate the use of R&D resources more efficiently, lowering R&D
misallocation.
Firms may prefer R&D collaboration with universities in China for two principal reasons. First, new knowledge generated by
universities can be treated as a public goods, and universities are recognized as a source of knowledge for firms. In particular, the
information generally complements a firms’ R&D (Baba, Shichijo, & Sedita, 2009). This cross-function non-competitive alliance helps
firms to promote R&D efficiency. Second, UIC gives firms inexpensive and low-risk access to universities’ specialist knowledge
(Azagra-Caro, Pardo, & Rama, 2014). The lower transaction cost in R&D helps firms raise their R&D efficiency.
As outlined above, the cooperation between a company’s R&D department and a research institute is advantageous. However, it is
not clear why there is no significant effect on lowering R&D misallocation. One major issue of the UIC linkage or research institu­
te–industry collaboration is that it is fraught with risks due to the uncertainty of innovation commercialization, which is the main aim
of profit-maximizing firms. In China, universities may conduct profit-oriented R&D projects, because university-operated or spin-off
companies are allowed to create profit for universities. Consequently, universities are profit motivated. Alternatively, research in­
stitutes may focus on more generic and basic knowledge and lack strong profit incentives; preventing high-tech firms from
commercializing R&D output in the short-term with no significant stimulating effect on R&D efficiency.

5.2. Robustness checks

We have thus far illustrated some influential factors of R&D misallocation in science parks. In this section, robustness checks are
implemented to deal with several potential problems. First, a park’s characteristics and its R&D activity could have a causal rela­
tionship, or the R&D misallocation might be persistent. To address this potential endogeneity issue in conducting the robustness check,
we adopt the technique of dynamic generalized method of moment (GMM). Second, we use province-specific wages to capture regional
price distortion, which could act as a robustness check. Third, instead of using the standardized TFPR (RD_mis) as the dependent
variable, we run direct regressions on distortion parameters (TFPRi). Although it is not apparent in the literature, this robustness check
helps facilitate understanding of how those distortions emerge. Table 7 summarizes the results of the robustness checks.
Most estimates in Table 7 are similar to those in Table 6, although interesting results emerge and are worth noting, particularly the
triple helix linkages. To begin with, results obtained using the dynamic GMM in Columns (1) and (2) show that park R&D misallocation
seems to be persistent. This implies that the correction of R&D misallocation might be difficult to finalize in the short run. Crucially, we
find that the estimated coefficient of RES_IND becomes significantly negative in Column (1), suggesting that R&D cooperation with
research institutes may also help lower science park R&D misallocation. Next, using the distortions measure (TFPRi) as the dependent
variable, the influences of park characteristics remain the same, while RES_IND and UNI_IND turn out to have significantly negative
and insignificant relationships with R&D misallocation, respectively.
Drawn from estimates of RES_IND and UNI_IND in Tables 6 and 7, both R&D collaboration with research institutes or universities
can help raise the R&D efficiency of science parks, thereby lowering R&D misallocation. It is likely due to weak significance, resulting
in the varied significance of RES_IND and UNI_IND in different empirical specifications. Overall, cross-function, non-competitive
alliance with research institutes and/or universities help firms to promote R&D efficiency, highlighting the importance of national
and/or regional innovation systems.

15
C.-H. Yang and W.-C. Lee China Economic Review 67 (2021) 101605

Table 7
Robustness Checks.
(1) (2) (3) (4) (5)

Dynamic GMM Dynamic GMM Distortion: wage Y = TFPRi Y = TFPRi

lag. RD_miss 0.578*** (0.019) 0.585*** (0.019)

lnSIZE − 0.067*** (0.025) − 0.065*** (0.026) − 0.554*** (0.134) − 0.320*** (0.063) − 0.325*** (0.061)
New_Park 0.332*** (0.081) 0.195*** (0.072) 3.026*** (0.329) 0.650*** (0.154) 0.621*** (0.151)
lnEXP − 0.046* (0.024) − 0.008 (0.023)
EXP_ratio 0.194 (0.198) 0.140 (0.370) 0.133 (0.168)
UNIV_ratio − 0.778*** (0.129) − 0.708*** (0.135) − 1.111*** (0.296) − 0.338*** (0.130) − 0.314** (0.134)
RES_IND − 0.007* (0.0036) − 0.006 (0.0038) 0.002 (0.007) − 0.006* (0.003) − 0.006* (0.003)
UNI_IND − 0.015*** (0.003) − 0.015*** (0.002) − 0.009* (0.005) − 0.002 (0.002) − 0.002 (0.002)
Prov_distortion − 0.459*** (0.171) − 0.508*** (0.168) − 0.045 (0.074) − 0.052 (0.075)
Province wage − 0.421 (0.663)
Constant 2.322*** (0.553) 1.661*** (0.309) 10.358 (6.809) 7.315*** (0.723) 7.248*** (0.712)
Year FE Yes Yes Yes Yes Yes
Province FE No No No No No
R-square 0.266 0.270 0.271
Arellano-Bond test 0.001 0.001
Hansen Sargan 0.014 0.025
Observations 482 482 571 571 571

Note: Figures in parentheses are heteroscedastic-consistent standard deviations. ***p < 0.01, **p < 0.05, *p < 0.1.

Table 8
R&D Growth and R&D Misallocation.
Dep. var. (1) (2) (3) (4)

RD_mis RD_mis TFPRi TFPRi

Δ RD expenditure − 0.091*** (0.030) − 0.047*** (0.013)

Δ RD personnel − 0.119*** (0.041) − 0.065*** (0.017)
Constant − 0.068 (0.055) − 0.070 (0.055) 3.754*** (0.023) 3.754*** (0.023)
Province-Year FE Yes Yes Yes Yes
R-square 0.430 0.429 0.541 0.541
Observations 568 570 568 570

Note: Figures in parentheses are heteroscedastic-consistent standard deviations. ***p < 0.01.

5.3. R&D scale and R&D misallocation

The above analysis identifies how park characteristics relate to R&D misallocation. It is clear that R&D misallocation is affected by
R&D scale. Jones and Williams (1998) indicate that R&D can be increasing or decreasing return to scale, depending on the way past
ideas have affected current R&D productivity. R&D investment risk also declines with scale (Ciftci & Cready, 2011), implying that R&D
productivity increases with scale. In other words, R&D misallocation could fall as R&D inputs grow. Using return-to-scale as a con­
ceptual foundation, we explore how R&D growth affects parks’ R&D misallocation. Table 8 shows the estimation results.
Results using R&D misallocation and TFPR as the dependent variable are shown in Columns (1)–(2) and Columns (3)–(4),
respectively.16 We find a significantly negative relationship between the growth of R&D expenditure (growth of R&D personnel) and
R&D misallocation. From the view of innovation production function, the increases in R&D inputs seem to generate a non-
proportional, larger increase in R&D output, suggesting an increasing return-to-scale with NHIZs. The agglomeration and network
effects, as well as institute-university-park linkages, can also facilitate R&D efficiency. Thus, increases in R&D help lower R&D
misallocation, particularly in the old group of NHIZs.

6. Concluding remarks and policy implication

Establishing science parks is widely adopted as a key strategy for developing high-tech industries and creating jobs to facilitate
regional development across countries. The success of science parks, in terms of R&D and innovation performance, relies heavily on the
park characteristics and the agglomeration effect of various R&D actors: mainly research institutes, universities, and industries.
Adequate locations and the number of science parks are particularly relevant to developing countries where R&D resources are
relatively limited.
Along with economic development, China has established more than 100 national-class science parks in all provinces except for

16
Inasmuch as growth of R&D expenditure and R&D personnel are highly correlated, they are included separately here as covariates. We have also
separated growth of outputs and exports and found no significant influence on R&D misallocation.

16
C.-H. Yang and W.-C. Lee China Economic Review 67 (2021) 101605

Tibet, over the past two decades. There are considerable differences in the degree of economic development across provinces, which
lead to the emergence of crucial issues. Does allocating R&D resources to numerous science parks result in serious R&D misallocations?
How do park characteristics and the triple helix of university–industry–government linkages affect the efficiency of R&D in the science
parks? This study aimed to evaluate R&D misallocation and its determinants across science parks in China. Based on an unbalanced
panel data of 145 science parks from 2007 to 2014, we applied Hsiech and Klenow’s (2009) methodology to calculate R&D efficiency
(misallocation) of science parks and then examined the determinants of R&D efficiency across science parks.
Crucial findings are summarized as follow. First, we found a decreasing trend for R&D efficiency across science parks in China.
Particularly, the average R&D efficiency had decreased sharply since 2011 when China began to start up many new science parks,
suggesting a considerable R&D misallocation caused by establishing science parks in all areas. Second, as argued in Albahari et al.
(2018), the heterogeneity of science parks is relevant to their R&D performance. We found science parks which are larger, older, or
with higher quality of human capital experience lower R&D misallocation. Export activity seems to less relevant to the degree of R&D
misallocation efficiency across parks. Third, the triple helix of university–industry–government linkages matter to science parks’ R&D
efficiency, and the UIC seems to more relevant than the collaboration between the research institutes and industry.
Two crucial policy implications arise from our analyses. First, most of the less R&D efficient science parks are newly established or
are upgraded from the provincial level industrial parks. They are generally located in cities without nationally renowned universities
and research institutes, whereas the first science parks are located in big cities, such as Beijing, Shanghai, and Nanjing, among others.
Although science parks at a national level are nominal, they are indeed managed by the provincial government. Therefore, for some
R&D efficiency parks, the authority may have to coordinate with the central government to improve their R&D efficiency, lowering the
problem of R&D misallocation.
Second, although the importance of the triple helix linkages in the regional innovation system is widely recognized, the research
institute–industry R&D collaboration seems to have less significant R&D efficiency-enhancing effect for science parks. The resources
for R&D that were allocated to public research institutes is higher than that of universities in China. For example, research institutes
and universities accounted for 15.04% and 10.47% of total R&D expenditure in 2011, respectively. By contrast, the collaborative R&D
accounts for a much smaller ratio of total R&D for research institutes than that of universities, as shown in Table 4. The authorities may
consider how to facilitate the technology transfer from public research institutes and how to enhance R&D collaboration with in­
dustries by aiming to complement the industries’ R&D. It would help promote the science parks’ R&D efficiency.

Acknowledgement

Chih-Hai Yang gratefully acknowledges financial support from the Ministry of Science and Technology of Taiwan (NSC107-2410-
H-008-004). Helpful comments and suggestions from participants at the 2019 Asia Pacific Innovation Conference (APIC) are highly
appreciated. Miss. Yi-Fei Li’s research assistant on collecting the science parks information is also appreciated.

References

Albahari, A., Barge-Gil, A., Pérez-Canto, S., & Modrego-Rico, A. (2018). The influence of science and technology park characteristics on firms’ innovation results.
Papers in Regional Science, 97(2), 253–279.
Atkeson, A., & Kehoe, P. (2005). Modeling and measuring organization capital. Journal of Political Economy, 113, 1026–1053.
Azagra-Caro, J., Pardo, R., & Rama, R. (2014). Not searching, but finding: How innovation shapes perceptions about universities and public research organisations.
Journal of Technology Transfer, 39(3), 454–471.
Baba, Y., Shichijo, N., & Sedita, S. (2009). How do collaborations with universities affect firm’s innovative performance? The role of “Pasteur scientists” in the
advanced materials field. Research Policy, 38, 756–764.
Bai, J. (2013). On regional innovation efficiency: Evidence from panel data of China’s different provinces. Regional Studies, 47, 773–788.
Bell, A., & Jones, K. (2015). Explaining fixed effects: Random effects Modeling of time-series cross-sectional and panel data. Political Science Research and Methods, 3
(1), 133–153.
Bigliardi, B., Dormio, A. I., Nosella, A., & Petroni, G. (2006). Assessing science Parks’ performances: Directions from selected Italian case studies. Technovation, 26(4),
489–505.
Ciftci, M., & Cready, W. M. (2011). Scale effects of R&D as reflected in earnings and returns. Journal of Accounting and Economics, 52(1), 6–80.
Colombo, M. G., & Delmastro, M. (2002). How effective are technology incubators? Evidence from Italy. Research Policy, 31, 1103–1122.
Crépon, B., & Duguet, E. (1997). Estimating the innovation function from patent number: GMM on count panel data. Journal of Applied Econometrics, 12, 243–263.
Diez-Vial, I., & Fernández-Olmos, M. (2017). The effect of science and technology parks on a Firm’s performance: A dynamic approach over time. Journal of
Evolutionary Economics, 27(3), 413–434.
Etzkowitz, H., & Leydesdorff, L. (2000). The dynamics of innovation: From National Systems and “mode 2” to a triple Helix of university–industry–government
relations. Research Policy, 29(2), 109–123.
Fan, C. C., & Scott, A. J. (2003). Industrial agglomeration and development: A survey of spatial economic issues in East Asia and a statistical analysis of Chinese
regions. Economic Geography, 79(3), 295–319.
Foster, L., Haltiwanger, J., & Syverson, C. (2008). Reallocation, firm turnover, and efficiency: Selection on productivity or profitability. American Economic Review, 98,
394–425.
Fritsch, M., & Franke, G. (2004). Innovation, regional knowledge Spillovers and R&D cooperation. Research Policy, 33(2), 245–255.
Fukugawa, N. (2006). Science parks in Japan and their value-added contributions to new technology-based firms. International Journal of Industrial Organization, 24,
381–400.
Gallie, E. P., & Legros, D. (2012). Firms’ human capital, R&D and innovation: A study on French firms. Empirical Economics, 43, 581–596.
Griliches, Z. (1990). Patent statistics as economic indicators. Journal of Economic Literature, 28, 1661–1707.
Hsieh, C. T., & Klenow, P. J. (2009). Misallocation and manufacturing TFP in China and India. Quarterly Journal of Economics, 124(4), 1403–1448.
Hu, A. G. (2007). Technology parks and regional economic growth in China. Research Policy, 36(1), 76–87.
Hu, J. L., Han, T. F., Yeh, Y. F., & Lu, C. L. (2010). Efficiency of science and technology industrial parks in China. Journal of Management Research, 10(3), 151–166.
Hu, A. G., & Jefferson, G. H. (2009). A Great Wall of patents: What is behind China’s recent patent explosion? Journal of Development Economics, 90(1), 57–68.
Izadikhah, M., & Saen, R. F. (2015). A new data envelopment analysis method for ranking decision making units: An application in industrial parks. Expert Systems, 32
(5), 596–608.

17
C.-H. Yang and W.-C. Lee China Economic Review 67 (2021) 101605

Jefferson, G. H., Bai, H., Guan, X., & Yu, X. (2006). R&D performance in Chinese industry. Economics of Innovation and New Technology, 15(4/5), 345–366.
Jin, H., Qian, Y., & Weingast, B. R. (2005). Regional decentralization and fiscal incentives: Federalism, Chinese Style. Journal of Public Economics, 89, 1719–1742.
Jones, C. I., & Williams, J. C. (1998). Measuring the social return to R&D. Quarterly Journal of Economics, 113(4), 1119–1135.
Jones, C. I., & Williams, J. C. (2000). Too much of a good thing? The economics of Investment in R&D. Journal of Economic Growth, 5(1), 65–85.
Jongwanich, J., Kohpaiboon, A., & Yang, C. H. (2014). Science Park, triple Helix and regional innovative capacity: Province-level evidence from China. Journal of
Asian-Pacific Economy, 19(2), 333–352.
Koopman, R., Wang, Z., & Wei, S. J. (2008). How much of Chinese exports is really made in China?. In Assessing domestic value-added when processing trade is pervasive,
NBER Working Paper No. 14109.
Li, X. (2009). China’s regional innovation capacity in transition: An empirical approach. Research Policy, 38, 338–357.
Li, H. C., Lee, W. C., & Ko, B. T. (2017). What determines misallocation in innovation? A study of regional innovation in China. Journal of Macroeconomics, 52,
221–237.
Liberati, D., Marinucci, M., & Tanzi, G. M. (2015). Science and technology parks in Italy: Main features and analysis of their effects on the firms hosted. Journal of
Technology Transfer, 41(4), 694–729.
Link, A., & Scott, J. (2003). U.S. science parks: The diffusion of an innovation and its effects on the academic missions of universities. International Journal of Industrial
Organization, 21, 1323–1356.
Löfsten, H., & Lindelöf, P. (2002). Science parks and the growth of new technology-based firms: Academic-industry links, Innovation and Markets. Research Policy, 31,
859–876.
Luger, M. I., & Goldstein, H. A. (1991). Technology in the Garden: Research Parks and Regional Economic Development. Chapel Hill, NC, U.S.A.: The University of North
Carolina Press.
Macdonald, S. (1987). British science parks: Reflections on the politics of high technology. R&D Management, 17(1), 25–37.
Massey, D., Quintas, P., & Wield, D. (1992). High tech fantasies: Science parks in society, science and space. London: Routhledge.
Mccann, B. T., & Folta, T. B. (2011). Performance differentials within geographic clusters. Journal of Business Venturing, 26, 104–123.
Nadiri, M. I., & Prucha, I. R. (1996). Estimation of the depreciation rate of physical and R&D Capital in the US. Total Manufacturing Sector. Economic Inquiry, 34(1),
43–56.
Nagaoka, S., Motohashi, K., & Goto, A. (2010). Patent statistics as an innovation indicator. In B. H. Hall, & N. Rosenberg (Eds.), Vol. 2. Handbook of the Economics of
Innovation (pp. 1083–1127). Amsterdam, Netherland: Elsevier.
Quintas, P., Wield, D., & Massey, D. (1992). Academic-industry links and innovation - questioning the Science Park model. Technovation, 12, 161–175.
Ramani, S. V., EI-Aroui, M. A., & Carrere, M. (2008). On estimating a knowledge production function at the firm and sector level using patent statistics. Research
Policy, 37(9), 1568–1758.
Restuccia, D., & Rogerson, R. (2008). Policy distortions and aggregate productivity with heterogeneous plants. Review of Economic Dynamics, 11, 707–720.
Syverson, C. (2011). What determines productivity? Journal of Economic Literature, 49(2), 326–365.
Vásquez-Urriago, A. R., Barge-Gil, A., Rico, A. M., & Paraskevopoulou, E. (2014). The impact of science and technology parks on Firms’ product innovation: Empirical
evidence from Spain. Journal of Evolutionary Economics, 24(4), 825–873.
Wainova. (2009). Wainova atlas of innovation: Science/technology/research parks and business incubators in the world. Cheshire: Ten Alps Publishing.
Weil, D. N. (2013). Economic growth (3rd ed.). Boston: Pearson.
Westhead, P., & Batstone, S. (1998). Independent technology-based firms: The perceived benefits of a Science Park location. Urban Studies, 35(12), 2197–2219.
Yang, C. H. (2018). Exports and innovation: The role of heterogeneity in exports. Empirical Economics, 55(3), 1065–1087.
Yang, C. H., Motohashi, K., & Chen, J. R. (2009). Are new technology-based firms located on science parks really more innovative? Evidence from Taiwan. Research
Policy, 38(1), 77–85.

18