Do Science Parks Promote Companies’ Innovative Performance? Micro Evidence from Shanghai Zhangjiang National Innovation Independent Demonstration Zone

Abstract

Science parks are considered to be key drivers of innovative economic activities and are important tools for countries and regions to achieve sustainable development. However, there still exists controversy about the positive effect of the science parks on the companies’ innovative performance. In this study, we constructed six hypotheses according to previous studies and tested them in the Shanghai Zhangjiang National Innovation Independent Demonstration Zone to answer two major research questions, i.e., (1) “Do science parks promote companies’ innovative performance?” and (2) “What factors in science parks affect the likelihood and intensity of companies’ innovative performance?”. Specifically, we selected 911 companies within the park and 861 companies outside the park using the coarsened exact matching method and applied the zero-inflated negative binomial model to identify the relationship between the company’s presence within the science park and the company’s innovative performance. Then, we applied the Heckman two-step model to further explore the key impact factors affecting the intensity of the innovation activities of the companies in the park. The results confirmed our first hypothesis that science parks can promote companies’ innovative performance. Moreover, we obtained two other findings. First, if a company is located within a science park, it can greatly improve the probability of innovation of that company, but it does not have any significant impact on the intensity. In other words, science parks promote more innovation among companies lacking innovation experience than those with innovation experience. Secondly, the reason why science parks can promote innovation is the knowledge spillover of innovation supporting institutions and high-tech enterprises in the parks, which is limited to a small geographical range. Our study provided new evidence on the positive role of science parks on companies’ innovative performance and offered a valuable sample for the research of science parks in developing countries. In addition, the policy suggestions we raised have reference value for developing countries to take full advantage of science parks to achieve sustainable development of their innovative economy.

1. Introduction

In recent decades, science parks have emerged as key tools for promoting economic development and competitiveness in both developed countries and developing countries [1]. Equipped with state-of-the-art facilities and services, these designated zones aim to stimulate companies’ innovative performance at the micro level, promote exchanges and cooperation between enterprises and research institutions at the meso level, and ultimately boost national and regional economic growth at the macro level. In addition, the importance of science parks in national and regional sustainability is drawing increasing attention, as they are considered to play a critical role in promoting innovation and sustainable ecosystem development [2]. Science park is an important factor in the innovation ecosystem and a strong promoter of the innovation industry cluster. By bringing together entities such as leading companies, startups, research and development centers, universities, and research institutes in geographical space, science parks help stimulate the flow of knowledge and technology within the region, facilitate the generation of innovation networks, promote the flourishing of high-tech industries, and ultimately contribute to the sustainable development of the region and the country [3].

As a policy tool, science parks play a more prominent role in promoting innovation and sustainable economic development in developing countries where governments intervene in the economy [4]. Many developing countries have achieved great success through the establishment of science parks, among which China is a typical case. Since 1988, China has been laying out science parks, which have helped fuel many of the country’s economic achievements over the decades. Among the many science parks in China, Zhangjiang National Independent Innovation Demonstration Zone (later referred to as “Zhangjiang Demonstration Zone”), one of the first national independent innovation demonstration zones, is the bridgehead for the implementation of China’s innovation-driven development policy. With the recent in-depth development of the technological transformation, China’s economy is shifting to the stage of high-quality development. Therefore, in the recently announced 14th Five-Year Plan, the Chinese government proposed that innovation should play a core role in the modernization drive, and further emphasized the importance of the construction of science parks as a strategy to support national innovation in the transformation of innovative industries.

Despite their popularity and widespread adoption, there is still an ongoing debate about the effectiveness of science parks in fostering companies’ innovative performance and achieving sustainable development. Do science parks really promote the companies’ innovative performance? This issue has garnered widespread attention of scholars. Many scholars believe that science parks have a positive impact on the innovation and growth of companies [5,6], and explored the reasons behind this. They pointed out that science parks can provide key innovation resources, such as financing, expertise, and cooperation networks [1], for companies within the parks. In addition, the positive agglomeration effect and knowledge spillover generated by science parks can also considerably benefit companies [7,8]. However, other scholars are skeptical about the positive effects of science parks. These scholars expressed concern about the potential risks and limitations of science parks and raised some questions, such as whether science parks rely too much on government support [9] and crowd out non-high-tech industries. In addition, some scholars have found that the influence of science parks on companies’ innovative performance is not homogeneous, that is, the benefits gained by all companies located within the same science park are not completely equal [3]. This may be due to the complex heterogeneity of different science parks and different companies. This leads to another important issue, i.e., what factors in science parks affect the likelihood and intensity of companies’ innovative performance? Scholars have explored the heterogeneity factors of science parks, such as the cooperation between science parks and research institutions and the number of companies in the parks.

This paper selected 22 sub-parks of the Zhangjiang Demonstration Zone in Shanghai, China, as research objects, aiming to explore the following two research questions: First, do science parks promote companies’ innovative performance? Second, what factors in science parks affect the likelihood and intensity of companies’ innovative performance? Specifically, we first downloaded data and pre-processed them using the coarsened exact matching (CEM) method to reduce the variability in the sample selection of companies. Second, we used random effects negative binomial and zero expansion negative binomial regression strategies to analyze the influence of science park location on companies’ innovative performance. Finally, we used the Heckman two-stage model to investigate the heterogeneity factors in science parks affecting the companies’ innovative performance. This paper aims to provide new evidence on whether science parks affect companies’ innovation, complement the gap in case studies from developing countries, and provide insights and recommendations to policymakers, practitioners, and academics interested in promoting innovation and sustainable development.

2. Literature Review

2.1. The Controversy about the Positive Effect of Science Parks on Companies

Whether science parks promote companies’ innovative performance is one of the most important research themes in this field. Earlier studies, including Felsenstein [10], Lindelof [11], and Westhead [12], believed that there is no positive correlation between the company’s location inside the park and the company’s innovative performance. Among them, Felsenstein [10] raised the new conception of “innovation enclaves” in which companies rarely interact with research institutions after entering the park. Lindelof [11] believed that even in-park firms are unable to transform the resources they have invested into more innovations. It may be because some firms are more closely connected to subjects outside the park than those inside and fail to benefit from knowledge spillovers from the park. Westhead [12] found that there was no sufficient evidence that can support that in-park companies benefit from science parks at innovative performance than out-of-park companies in terms of innovative efficiency. Some scholars arrived at the opposite conclusion, among which Siegel’s is the most representative [6]. His study integrated five indicators, including expanded R&D inputs and outputs, to form a new indicator of R&D efficiency and found that in-park firms had higher R&D efficiency than out-of-park firms. Other scholars who supported this view have proposed [13] evidence by more patent applications [14], and more new products [15] of the companies within the parks. They attributed this to the knowledge spillovers between firms [10], or cooperation with R&D institutions within the parks [6].

Two reasons make the conclusions of the above studies different from each other. First, different studies have different indicators for measuring the innovation output of firms, such as the number of patents held, the number of new products or sales, etc. From the perspective of patents, Corrocher et al. [1] defined innovation output as the number of patent applications in a single year, Huang [5] defined it as the stock of patents in five years, and Lamperti et al. [13] defined it as the cumulative number of patents in eight years. From the perspective of new products, Ubeda et al. [15] defined innovation output by the number of new products, Ramirez-Aleson et al. [16] by the sales volume of new products, and Vasquez-Urriago et al. [17] by the proportion of sales volume of new products. In addition, Claver-Cortes et al. [18] also defined firms’ innovation as the sum of different types of innovation, including output innovation, process innovation, and commercial innovation. Chan et al. [19] integrated multiple indicators, such as the number of patent applications and the number of new products, and redefined the firm’s innovation output from the perspective of comprehensive indicators.

Second, heterogeneity exists among different firms in different science parks [3,20], but many previous studies only focused on the homogeneity of science parks and firms within them. In other words, the role of science parks in firms’ innovation is not invariable; it will be affected by the different characteristics of the park and the different characteristics of the firms within the park. However, some scholars have already had a preconceived assumption when evaluating the role of parks, that is, all parks will have the same impact on the firms within the parks, and all firms within the parks will benefit in the same way. In recent years, some scholars have begun to question the original homogeneity hypothesis and tried to take into account the heterogeneity of science parks and firms in parks as a regulatory factor affecting the role of parks in research.

Thus, it is necessary to explore the impact of science parks on firms based on a clear definition of innovation intensity with full consideration of firm and science park heterogeneity. We proposed the following hypothesis:

Hypothesis 1:

Science parks can promote the companies’ innovative performance.

2.2. The Potential Impact Factors of Science Parks on Companies

The heterogeneity of science parks and companies is mainly reflected in numerous influencing factors. A large number of studies have been conducted to examine these characteristics. In terms of science parks, existing studies have explored a series of moderating factors related to parks, including age, scale [21], degree of specialization [13], management characteristics [6], regional characteristics [20], and the relationship with local knowledge institutions [13]. Taking the age of the park as an example, scholars found that science parks established earlier had a positive impact on the innovation of firms in the parks [22]. This may be because these parks have more accumulated knowledge and a deeper understanding of the needs of the firms in the park. Moreover, it takes time for the park to have an impact on the firms within it. Therefore, Castell et al. [8] proposed that a scale of 15 to 25 years should be used to evaluate the role of science parks. In addition to the above factors, the heterogeneity of science parks is also closely related to the national conditions and policy factors of the country, such as financial subsidies [23], which is particularly typical in China.

In terms of the firms within the park, previous studies have explored moderating factors such as firm size, age [17], R&D level [24], and managers’ attitudes toward the park [25]. Taking firm size as an example, Liberati et al. [20], Vásquez-Urriago et al. [14], and other scholars found that smaller firms benefit more from the park compared to larger firms, which may be attributed to the fact that smaller firms may obtain more of the positive effects of knowledge spillovers in the cluster. However, Squicciarini [7] found that the increase in the number of employees slightly improves the likelihood of applying for patents. In addition, as a policy tool that mainly serves national economic development, the actual implementation effect of science parks is inevitably affected by different national conditions. Thus, some scholars studied policy-related and country-related features. Based on the influencing factors mentioned above, we selected five of the most representative ones, i.e., industrial output, the area of the science park, the number of high-tech companies in the park, the number of innovation support institutions, and municipal government funding. We hypothesized the following:

Hypothesis 2a:

The industrial output of science parks is positively correlated with companies’ innovative performance.

Hypothesis 2b:

The area of science parks is positively correlated with companies’ innovative performance.

Hypothesis 2c:

The number of high-tech companies in the parks is positively correlated with companies’ innovative performance.

Hypothesis 2d:

The number of innovation support institutions in the science parks is positively correlated with companies’ innovative performance.

Hypothesis 2e:

Government funding received by science parks is positively correlated with companies’ innovative performance.

In addition to whether science parks affect firms’ innovation and the moderating factors mentioned above, the influence mechanism of science parks on firms’ innovation is another topic that has attracted widespread attention. Researchers have explored multiple possible roles of parks when influencing the firms’ innovation, which can be summarized into five categories: bridging intermediary role [13], supporting incubation role [26], preferential policy support role [27], infrastructure protection role [24], and innovation consulting role [28]. The bridging intermediary role means science parks connect companies, universities, and research institutions, stimulating the flow and spillover of knowledge among companies in the parks. At the same time, the integration of industry, university, and research has enriched the sources of innovative knowledge and technology. Supporting incubation role indicates parks actively cultivate potential start-ups to enter the innovation incubator. The preferential policy support role represents parks providing special preferential policies and financial subsidies to increase firms’ R&D investment. Infrastructure protection role means parks offer various infrastructure and shared services. Innovation consulting role means parks provide firms with specialized business and innovation consulting services. By playing the above roles, science parks ultimately help firms reduce innovation costs, increase cooperation frequency and stimulate innovation output.

2.3. Research on China’s Science Parks

Research on China’s science parks has been increasing, as China has made an economic miracle in the past decades as the largest developing country. Most of the studies analyzed the impact of China’s science parks on the regional economy from a macro perspective. Wang et al. [29] examined the effects of science parks on economic development and environmental pollution in China. Yu et al. [30] studied whether science parks contribute to Chinese local economies. Other scholars explored the role of parks on firms’ innovation from a micro perspective. Xie et al. [31] constructed an effect model of technological entrepreneurship of a science park. Torres et al. [4] analyzed the barriers to innovation experienced by tenants in science parks. Up to now, there is a relative lack of research from the perspective of science parks on the promotion of companies’ innovative performance in China.

Specific to the selection of covariates, researchers rarely considered heterogeneity among the research subjects in China. In terms of the studies on impact factors, some economists have found that incubators [32], government subsidies [23], university cooperation [33], and other elements may affect firms’ innovation. However, most of them took an isolated perspective, failing to take the holistic nature of science parks as innovation systems into account. Existing studies assessing the impact of science parks were mostly based on cases from developed countries, with relatively little evidence from developing countries. The impact mechanism of science parks on innovation is closely related to national context and policy factors, thus studies for developing and developed countries may reach different conclusions. There is an urgent need to explore whether science parks in developing countries affect firms’ innovation and the mechanisms of the impact by selecting variables in a comprehensive and accurate manner, taking the heterogeneity of parks and firms as well as the differences in national conditions into account. China, the largest developing country, can be a perfect study choice.

3. Data and Methods

3.1. Data Collection

Zhangjiang Demonstration Zone is located in Shanghai, the leading city of China’s rapid economic development. It is one of the first national high-tech industrial development zones in China. As shown in Figure 1a,b, there are 22 sub-parks in the Zhangjiang Demonstration Zone, covering all districts of Shanghai with a total area of 531.32 square kilometers. These sub-parks are under the management of the Shanghai Promotion Science and Technology Innovation Center Construction Office, but they also retain a certain degree of management autonomy. Important measure indicators, such as scale, age, number of high-tech companies, incubators, national laboratories, and financial subsidies, vary among the sub-parks. In 2020, the Zhangjiang Demonstration Zone hosted nearly 100,000 innovative and technology-oriented enterprises, with the total revenue of enterprises above designated size reaching 905.35 billion US dollars, making it a very representative research zone of science parks in China.

In this study, the company samples were collected from the Orianan Asia Pacific Enterprise Database and Enterprise Search, which is the largest Chinese enterprise search engine. The selection criteria were set as companies established before 2016 and not closed before 2021, with registered addresses in Shanghai, and adopting two-digit code classification of national economy industries (GB/T 4754-2011). The industry categories were selected with reference to the industrial structure of the Zhangjiang Demonstration Zone in the annual report. We collected multiple company characteristic variables including the number of company employees, turnover, capital status, age, R&D investment, ownership, and holding from 2016 to 2020 according to the existing literature [3,17]. In addition, referring to “The Science, Technology and Industry Scoreboard of World Organization for Economic Cooperation and Development 2007” [34], corresponding dummy variables for the technological level of eight different industries were created as the technological level of the industry in which the company is located. Finally, the number of patents previously applied for and patents granted in the current year were collected from the incoPat global patent database to represent the company’s knowledge accumulation. Finally, the raw data of 957 companies within the park and 1012 companies outside the park were obtained, and the spatial distribution of the sample companies is shown in Figure 1c. Whether the park affects company innovation can be influenced by the heterogeneous characteristics of companies, thus this study was conducted under the premise of controlling these characteristics. The company characteristic variables described above were all used as control variables in this study.

3.2. Variable Definition

This paper first explores whether the park influences firm innovation, and the corresponding explanatory variable is the firm’s location variable. Whether a specific firm is located in the park and the name of the sub-park it is located in are obtained from the sub-park map offered by the website of Shanghai Office to Promote the Construction of Science and Technology Innovation Center (https://kcb.sh.gov.cn/) (accessed on 1 December 2021). Then, we studied the impact of the heterogeneous characteristics of the parks on company innovation, and the corresponding explanatory variables are the park characteristics including industrial output, number of employees, area, age, innovation climate, innovation support institutions and financial subsidies refer to the existing literature [13,18]. The above variables of the 22 sub-parks for each year were obtained from the Annual Development Report of Zhangjiang Demonstration Zone from 2016 to 2020 [35] and the website of Shanghai Municipal People’s Government.

The explained variable in this study was the firm’s innovation performance, which can be further divided into the likelihood and intensity of innovation. In a given year, if the company has one patent, it reflects the likelihood of its innovation, while the number of patents reflects the intensity of its innovation. This variable is defined as the number of patents applied by the firm in a given year and the number of patents granted in a lag year, the latter being used only to test the robustness of the model. Like most studies [13,28], we used patent data because it can effectively reflect the companies’ innovation investment and results. In addition, the number of patent applications has high data quality and greater immediacy compared to the number of patents granted. The data comes from the incoPat global patent database, covering the period from 2016 to 2021. The names and definitions of all variables are listed in detail in

Table 1. The name and definition of variables.

Variable	Variable Label	Variable Definition	Unit of Measure
Explained variables
Patent Applications	Inpatapl	The number of patents applied for by the enterprise in the current year *	pcs
Patent Granted	Inpatnext-imp	The number of patents granted to the enterprise in the following year *	pcs
Explanatory variables (firm-related)
Location	park	Within park or not	1: In the park
Explanatory variables (park-related)
Industrial output	pa_Inoutput	Total industrial output value of the sub-park for the year *	billion yuan
Number of employees	pa_Instaff	Number of employees at the end of the year *	million people
Area	pa_Inarea	Campus area for the year *	km2
Age	pa_Inage	The number of years since the establishment of the sub-park to this year *	year
Innovation atmosphere	pa_Inhtcomp	Number of high-tech enterprises in the park for the year *	pcs
Innovation support institutions	pa_Ininsti	The sum of the number of national incubators and national key laboratories in the park for the year *	pcs
Financial subsidies	pa_Infond	Number of municipal special funds received by the park this year *	million
Control variables (company-related)
Number of employees	Instaff	Number of employees for the year *	people
Turnover	Inincome	Turnover for the year *	million yuan
Capital status	Infixa	Number of fixed assets for the year *	ten thousand yuan
Age	age	Number of years since establishment to the current year	year
Knowledge accumulation	Inpatbeapl	Cumulative number of patent applications before the current year *	individual
	Inpatbe-imp	Cumulative number of patents granted before the current year *	individual
R&D investment	Inresa	Capital investment in R&D for the current year *	million yuan
Ownership	stateowned	State-owned or not	1: State-owned
Holding	group	Whether it is a subsidiary held by a large company	1: Held by
Industry Technology Level	techlevelmanu1	Whether it belongs to high-technology level manufacturing industry	1: Belong to
	techlevelmanu2	Whether it is a medium-high technology manufacturing industry
	techlevelmanu3	Whether it belongs to medium-technology manufacturing industry
	techlevelmanu4	Whether it belongs to low-technology manufacturing industry
	techlevelnonmanu1	Whether it is a high-technology level non-manufacturing industry
	techlevelnonmanu2	Whether it belongs to medium-high technology non-manufacturing industry
	techlevelnonmanu3	Whether it belongs to the low-technology non-manufacturing industry
	techlevelnonmanu4	Whether it is a low-tech level non-manufacturing industry

Note: * means that this data is logarithmized and its natural logarithm is returned after adding 1.

, and some data have been logarithmized to reduce the estimation bias caused by the data magnitude.

3.3. Data Matching and Robustness Testing

This study aims to compare whether there exist differences in innovation outcomes between companies within and outside the science park. In order to remove the influence of irrelevant confounders on the comparison results, and to make the control sample of companies consistent as much as possible on characteristics, the CEM method [8] was selected to match the sample companies inside and outside the park based on specified covariates.

The CEM method was chosen as the match strategy in this study for two reasons. First, some of the covariates selected in this study do not conform to normal distribution (e.g., industry), and are not applicable to the distance function of traditional propensity score matching method, however, CEM can handle these categorical variables manually. Second, the CEM method does not require to be limited to the common region of data between the control and experimental group, which can maximize the near-consistency of covariates between experimental and control group data. In this way, the comparability can be increased, and the degree of model dependence and study bias can be reduced.

The process of the CEM method is as follows: All company samples were coarsely classified according to the different values of each covariate. Then, an exact match was conducted. Only the classes containing both experimental and control groups were retained after the result of matching, and the remaining classes and data within them were deleted. The validity of matching was measured by the imbalance of the two groups (L1 statistical value), and the value of L1 was taken in the range of [0–1]. The closer the value of L1 is to 0, the better the balance of the two data sets and the more effective the matching. If the sample sizes of the two groups are not equal after matching, the weight variable is applied to balance the groups.

We referred to the existing literature [13] and selected the age, number of employees, turnover, ownership and industry classification of the company as matching covariates. The progressive coarsening process of the CEM method was conducted using R language [33]. All variables were automatically stratified, except for the number of employees, which was manually stratified. The final matching results were obtained with 911 companies within the park and 861 companies outside the park. The value of L1 decreased from 0.635 before matching to 0.324 after matching, indicating a great matching result. We also conducted a robustness test on the matching results. As shown in

Table 2. The comparison of covariables of companies within and outside the park after matching.

	Stateowned		Instaff		Inincome		Age
	Within Park	Outside Park	Within Park	Outside Park	Within Park	Outside Park	Within Park	Outside Park
N	4555	4305	4555	4305	4555	4305	4555	4305
Mean	0.008	0.010	5.468	5.245	10.260	9.924	14.430	14.420
Median	0	0	5.407	5.303	10.330	9.993	14	14
Std. Dev.	0.090	0.098	1.424	1.491	1.917	1.902	6.707	6.706
Min.	0	0	0.511	0.405	0	0	0	0
Max.	1	1	9.918	9.711	16.230	16.270	39	39
Tests
Kolmogorov–Smirnov test	p = 1.000		p = 0.598		p = 0.497		p = 0.856
Kruskal–Wallis test	p = 0.760		p = 0.354		p = 0.181		p = 0.449
t-test	p = 0.791		p = 0.460		p = 0.266		p = 0.549

, the differences in ownership, number of employees, business income and age between the matched companies within park and outside park are small. The original hypothesis of the tests in the table is that the two sets of data before and after matching are consistent in terms of distribution (Kolmogorov–Smirnov test), median (Kruskal–Wallis test) and mean (t-test), and all p values greater than 0.05 indicated that the original hypothesis is not rejected at the 5% significance level. Figure 2 shows an example of the industry distribution of the matched companies within the park and outside the park. The difference between the two groups was only nominal.

3.4. Model Building

The first type of model was designed to reveal whether science parks will affect companies’ innovative performance. When selecting the benchmark specification model, because the explained variable (number of patent applications) is the count data, the count model is the best choice, such as the negative binomial model or Poisson model. In addition, it can be seen from

Table 3. The descriptive statistics of company’s characteristic variables.

Variable Label	Mean	Standard Deviation	Min	Max	VIF Value
Inpatapl	1.085	1.416	0	7.610	-
park	0.549	0.549	0	1	1.100
Instaff	5.368	1.459	0.405	9.918	7.620
Inincome	10.110	1.917	0	16.27	6.760
Infixa	8.215	2.728	0	15.580	5.490
age	14.430	6.707	5	39	1.370
Inpatbeapl	2.151	2.050	0	9.946	1.580
Inresa	6.138	2.973	0	12.67	1.510
stateowned	0.009	0.094	0	1	1.090
group	0.748	0.434	0	1	1.250
techlevelmanu1	0.152	0.359	0	1	7.970
techlevelmanu2	0.433	0.496	0	1	7.860
techlevelmanu3	0.078	0.269	0	1	4.550
techlevelmanu4	0.027	0.164	0	1	6.390
techlevelnonmanu1	0.007	0.083	0	1	1.410
techlevelnonmanu2	0.181	0.385	0	1	5.530
techlevelnonmanu3	0.052	0.218	0	1	3.810
techlevelnonmanu4	0.069	0.254	0	1	4

that the mean value (1.085) of the explained variable is smaller than its variance (2.005), indicating that the Poisson model is not applicable. This was further confirmed by the results of the likelihood ratio test (LR test), whose output (Prob > chibar2 = 0.000) showed that the original hypothesis was rejected at the 1% significance level. The original hypothesis of the LR test is that there is no excessive dispersion and the Poisson model is applicable. Furthermore, 55.27% of the explained variables are zero, which further indicates that the zero-inflated negative binomial model is the most suitable one.

In addition to selecting the model from the perspective of data characteristics of the explained variables, this study also considered another perspective of quantitative indicators for model evaluation, such as Akaike information criteria (AIC) and Bayesian information criteria (BIC). Both of them are evaluation criteria to measure the fit of the model. AIC calculation formula is based on likelihood function and number of parameters, and BIC adds the number of samples on this basis. When multiple models are compared horizontally, the model with a smaller value is superior. After calculation, the zero-inflated negative binomial model and random effects negative binomial model with smaller AIC and BIC values are selected as the benchmark specification, while mixed regression, random effects regression and negative binomial regression are used as supplementary estimation strategies. The formula constructed is as follows:

$$Intensit{y}_{it}=\alpha +\beta Treatmen{t}_i+\delta X_{it}+{\epsilon }_{it}$$

(1)

where, $Intensit{y}_{it}$ denotes the number of patents filed by company i in year t, $Treatmen{t}_i$ represents the location variable of whether company i is within the park, and $X_{it}$ are the set of control variables of company characteristics. It worth mention that the zero-inflated negative binomial model is divided into two parts: the logit model, which was used for the subsample with zero number of patent applications to investigate whether a company’s presence in the park affects the likelihood of its patent applications; and the negative binomial model, which was used for the subsamples with non-zero number of patent applications to find out the effect of a company’s presence in the park on the number of its patent applications.

The second type of model was used to explore which factors in the science park affect the innovation performance of the company, including the likelihood and intensity of innovation. From 2016 to 2020, 911 companies in the science park were taken as samples to carry out ordinary least square regression and construct Equation (2).

$$Y_{it}=\rho X_{it}+\kappa Z_{it}+{\omega }_{it}$$

(2)

where, $X_{it}$ and $Z_{it}$ are the set of control variables for firm characteristics and the set of explanatory variables for park characteristics, respectively, $Y_{it}$ is the number of patent applications for firm i in year t, ρ and κ are parameters to be estimated, and ${\omega }_{it}$ is the residual term.

Since there exists a subsample of firms in the park without patent applications in the original data, the Heckman two-stage model was selected to avoid potential selection bias. The model constructed two equations: the probability equation of companies within the park having patent applications (Equation (3)), and the regression equation of the effect of park characteristics on company patent applications (Equation (4)). The total samples of firms in the park were first substituted into Equation (3) for Probit estimation to predict the probability of each firm in the park having a patent application, and then the estimated values $\alpha$ and $\sigma$ were brought into Equation (5) to calculate the inverse Mills ratio λ. Then, λ was added to Equation (4) as a new control variable to more accurately estimate the impact of the park on the number of patent applications. In this case, Equation (3) in Heckman’s first stage, used data from a sample of all of the in-park firms (911) between 2016 and 2020, regardless of the number of patent applications in the current year, while Equation (4) in the second stage, only used data from a subsample of in-park firms that have patent applications in the same time period.

$$P_{it}=E\left(Y_{it}>0\right)=\alpha X_{it}{}^{\prime }+{\nu }_i$$

(3)

$$Y_{it}=\beta X_{it}+\eta {\lambda }_i+{\mu }_i$$

(4)

$${\lambda }_i=\frac{\phi \left(-\alpha {X^{\prime }}_{it}/\sigma \right)}{\varphi \left(-\alpha {X^{\prime }}_{it}/\sigma \right)}$$

(5)

where, $P_{it}$ is the probability of a company having patent applications in the current year, which was set to 1 if there is at least one patent application or set to 0 if there is none. $Y_{it}$ is the number of company patent applications in the current year, ${\lambda }_i$ is the inverse Mills ratio, $\alpha$ and $\beta$ are parameters to be estimated, ${\nu }_i$ and ${\mu }_i$ are error terms obeying normal distribution, i is the ith company in the park, and t is the tth year. ${X^{\prime }}_{it}$ is the set of firm-related variables, including control variables and explanatory variables, affecting the presence or absence of patent applications. $X_{it}$ is the set of park-related explanatory variables affecting the number of patent applications. $\phi \left(.\right)$ is the density function of the standard normal, $\varphi \left(.\right)$ is the cumulative distribution function of the standard normal, and $\sigma$ is the standard deviation of ${\nu }_i$. It worth noting that the inverse Mills ratio λ is an important criterion to test the existence of selection bias. If the Mills inverse ratio λ is significantly non-zero, then selection bias exists in the sample, that is, the ordinary regression model is not applicable and the Heckman model should be chosen.

4. Results

In order to identify the impact of science parks on company innovation, this section first answers the question, “Do science parks promote companies’ innovative performance?” using the random effects negative binomial model and the zero-inflated negative binomial model. Then, the park characteristics variables were added and the key factors affecting the intensity of innovation in the park was further investigated using the Heckman two-stage model to answer the second research question, “What factors in science parks affect the likelihood and intensity of companies’ innovative performance?”.

4.1. Evidences for Whether Science Parks Promote Companies’ Innovative Performance

Table 3 presents the results of descriptive statistics for each company characteristic variable. The variance inflation factor is less than 10, which indicates that there is no multi-collinearity among the variables.

Table 4. The regression results of five models.

Model	(1)	(2)	(3)	(4)	(5)
	Mixed Regression Model	Random Effects Model	Negative Binomial Model	Random Effects Negative Binomial Model	Zero-Inflated Negative Binomial Model
					Zero-Count		Non-Zero-Count
					Coefficient	Percentage	Coefficient	Percentage
park	0.523 **	0.68 **	1.221 **	1.313 **	−2.173 **	−88.6%	0.385 ***	47.0%
	(0.021)	(0.019)	(0.02)	(0.033)	(0.051)		(0.017)
stateowned	0.630 **	0.723 *	0.121	0.287 **	−0.248 *	−22.0%	0.166 ***	18.1%
	(0.184)	(0.196)	(0.133)	(0.057)	(1.156)		(0.044)
group	−0.140 ***	−0.187 **	−0.173 ***	−0.246 ***	0.407 ***	60.3%	−0.088 ***	−8.4%
	(0.036)	(0.035)	(0.39)	(0.044)	(0.128)		(0.023)
Instaff	0.060 **	0.084 ***	0.052 **	0.093 ***	−0.116 **	−11.0%	0.033 *	3.4%
	(0.017)	(0.016)	(0.021)	(0.025)	(0.056)		(0.013)
Inincome	0.020	0.024 **	−0.023	−0.031	0.084 *	8.8%	0.023	2.3%
	(0.013)	(0.128)	(0.016)	(0.017)	(0.051)		(0.015)
Infixa	0.019 **	0.037 *	0.014	0.030	−0.018	−1.8%	0.024 *	2.4%
	(0.009)	(0.008)	(0.012)	(0.019)	(0.034)		(0.009)
age	−0.020 ***	0.015 ***	−0.015 ***	−0.021 ***	0.069 ***	7.1%	−0.008 ***	−0.8%
	(0.002)	(0.002)	(0.002)	(0.002)	(0.008)		(0.001)
Inpatbeapl	0.494 ***	0.403 ***	0.431 ***	0.460 ***	−1.156 ***	−68.5%	0.212 ***	23.6%
	(0.01)	(0.012)	(0.013)	(0.024)	(0.041)		(0.009)
Inresa	0.023 **	0.028 ***	0.068 ***	0.076 **	−0.157 **	−14.5%	0.017	1.7%
	(0.013)	(0.01)	(0.021)	(0.019)	(0.067)		(0.014)
techlevelmanu1	0.006	0.009	0.038	0.236 ***	−0.571 *	−43.5%	0.200 ***	22.1%
	(0.107)	(0.069)	(0.094)	(0.067)	(0.312)		(0.05)
techlevelmanu2	0.072	0.038	0.129	0.301 **	−0.395	−32.6%	0.107 ***	11.3%
	(0.1)	(0.055)	(0.086)	(0.104)	(0.298)		(0.048)
techlevelmanu3	0.003	0.139	0.023	0.171 **	−0.321	−27.5%	0.165 **	17.9%
	(0.108)	(0.07)	(0.098)	(0.08)	(0.321)		(0.053)
techlevelmanu4	0.002	0.084	0.016	0.226 **	−0.380	−31.6%	0.181 **	19.8%
	(0.126)	(0.096)	(0.121)	(0.112)	(0.47)		(0.077)
techlevelnonmanu1	0.444 **	0.392 *	0.561 **	0.807 ***	−1.208 **	−70.1%	0.140	15.0%
	(0.185)	(0.2)	(0.136)	(0.099)	(0.5)		(0.086)
techlevelnonmanu2	0.158	0.041	0.045	0.218 *	−0.333	−28.3%	0.072	7.5%
	(0.103)	(0.055)	(0.097)	(0.13)	(0.283)		(0.056)
techlevelnonmanu3	0.126	0.042	0.041	0.192	−0.366	−30.6%	0.124 **	13.2%
	(0.12)	(0.087)	(0.115)	(0.144)	(0.363)		(0.061)
techlevelnonmanu4	0.132	0.038	0.042	0.201	−0.321	−27.5%	0.081	8.4%
	(0.105)	(0.093)	(0.103)	(0.121)	(0.198)		(0.094)
Number of samples	8860	8860	8860	8860	8860
R2	0.568	0.583	0.425	-	-
F	205.89 ***	-	-	-	-
Wald chi2	-	1637.20 ***	2089.03 ***	1762.31 ***	2191.43 ***

Note: *, **, *** denote regression model coefficients significant at the 10%, 5%, and 1% levels, respectively. Standard errors are in parentheses.

shows the regression results for the five models, where the F-test or Wald test for each model is significant at the 1% level, indicating that the models all fit well.

As can be seen in Table 4, the coefficients of the firm location variable in all the models are significantly positive after other firm characteristic variables that may affect innovation were controlled, indicating that companies located in science parks have more patent applications than companies outside the parks. There is a positive influence of science parks on company innovation, which contradicts the findings of some existing studies [15,20]. However, this positive impact of parks may be more biased towards promoting the firm’s patents from zero to a certain number than promoting the growth of the number of patents. In other words, the role of the park is stronger in increasing the likelihood of innovation occurring in firms, and relatively weaker in promoting the intensity of company innovation, which is consistent with the findings of Vásquez-Urriago [17] and Huang [5].

This is also confirmed by model (5). For the zero-count part, the probability of having at least one patent application of the companies within the park is 88.6% higher than companies outside the park in the current year. For the non-zero-count part, the probability of having more patent applications of the companies within the park is 47% higher than companies outside the park in the current year. In other words, firms that lack innovation experience gain more by entering the park than firms that already have innovation experience. This may be due to the positive and negative effects of knowledge spillover in the cluster. Compared to companies lacking R&D experience, companies that already have some R&D experience in the cluster may face the negative effect of technological resource leakage [5]. Companies that lack innovation experience are more likely to suffer from positive effects of the cluster, such as technological talent sharing and invisible knowledge exchange, while the negative effects are minimal.

The coefficients of the control variables in models (1)–(5) are similar to the findings of existing studies [7,16,18], where the number of employees, knowledge accumulation, and R&D investment have a positive effect on firm innovation output, while age has the opposite effect. The innovation performance of state-owned companies is better than that of other ownership companies, while that of group subsidiaries is worse than that of independent companies.

4.2. Impact Factors of Science Parks That Affect the Companies’ Innovative Performance

The above section confirms that science parks have a significant positive effect on company innovation. Next, we tried to reveal the mechanism of park influence on firm innovation by introducing park characteristic variables.

Table 5. The descriptive statistics of park’s characteristic variables.

Variable Label	Mean	Standard Deviation	Min	Max	VIF Value
pa_Inoutput	6.625	1.921	0	8.358	7.95
pa_Instaff	2.7	0.741	0.451	3.735	6.68
pa_Inarea	3.461	0.816	1.141	4.391	6.86
pa_Inage	2.384	0.693	0	3.401	4.66
pa_Inhtcomp	5.688	0.941	2.565	7.451	5.44
pa_Ininsti	2.968	1.391	0	5.347	3.72
pa_Infond	8.107	0.775	5.369	9.609	1.85

shows the results of descriptive statistics for the park characteristic variables, where the variance inflation factors are all less than 10, indicating the absence of multicollinearity. The value of inverse Mills ratio was 1.1827 and was significantly non-zero at the 1% statistical level, indicating the existence of sample selection bias in the original data. Thus, the Heckman model was applied. The results of the Heckman two-stage model are shown in

Table 6. The results of Heckman two-step model.

Regression Equation: ${\mathit{Y}}_{\mathit{i}\mathit{t}}$		Probability Equation: ${\mathit{P}}_{\mathit{i}\mathit{t}}$
Explanatory Variables (Park-Related)		Explanatory Variables (Firm-Related)
pa_Inoutput	0.176 ***	park	2.454 ***
	(0.058)		(0.373)
pa_Instaff	0.018
	(0.097)
pa_Inhtcomp	0.082 **
	(0.042)
pa_Inarea	−0.239 **
	(0.148)
pa_Inage	0.055
	(0.090)
pa_Ininsti	0.024 *
	(0.043)
pa_Infond	0.065
	(0.050)
Control variables (park-related)		Control variables (firm-related)
stateowned	0.839 ***	stateowned	0.953 **
	(0.169)		(0.333)
group	−0.171 **	group	−0.023
	(0.183)		(0.099)
Instaff	0.123 ***	Instaff	0.156 **
	(0.031)		(0.643)
Inincome	0.029	Inincome	0.035
	(0.469)		(0.417)
Infixa	0.057 **	Infixa	0.069 **
	(0.021)		(0.319)
age	−0.041 ***	age	−0.017 **
	(0.005)		(0.009)
Inpatbeapl	0.569 ***	Inpatbeapl	0.523 ***
	(0.024)		(0.037)
Inresa	0.052 **	Inresa	0.058 **
	(0.212)		(0.020)
techlevelmanu1	0.184	techlevelmanu1	0.172 **
	(0.172)		(0.163)
techlevelmanu2	0.051	techlevelmanu2	0.080 *
	(0.167)		(0.157)
techlevelmanu3	0.022	techlevelmanu3	0.076 *
	(0.181)		(0.183)
techlevelmanu4	0.221	techlevelmanu4	0.261
	(0.473)		(0.489)
techlevelnonmanu1	0.260	techlevelnonmanu1	0.600 **
	(0.217)		(0.316)
techlevelnonmanu2	0.154	techlevelnonmanu2	0.211
	(0.156)		(0.168)
techlevelnonmanu3	0.083	techlevelnonmanu3	0.125 *
	(0.196)		(0.189)
techlevelnonmanu4	0.108	techlevelnonmanu4	0.107
	(0.134)		(0.159)

Note: *, **, *** denote regression model coefficients significant at the 10%, 5%, and 1% levels, respectively. Standard errors are in parentheses.

, and the Wald Chi-square value was significant at the 1% level, indicating a good fit of the model.

The probability equation column in Table 6 shows that the coefficient of the location variable deciding whether the company is within the park or not is the largest, which again proves that the likelihood of company innovation is not only influenced by its own characteristics, but also obviously promoted by the science park.

From the regression equation column in Table 6 we obtained many valuable conclusions: There is a significant positive correlation between the number of patent applications and the number of innovation support institutions, which indicates that firms tend to cooperate with research institutions in the park to develop new patents. The number of company patent applications is positively influenced by the industrial output value and the number of high-tech enterprises of the park, which indicates that the knowledge spillover in the form of sharing technological resources and technical talents does promote the frequent occurrence of innovation. The number of firm patent applications is significantly and negatively correlated with the size of the sub-park, implying that an intensive campus size is more conducive to the dissemination of tacit knowledge and possesses stronger positive externalities. In addition, the effect of municipal government funding on the number of patent applications is not significant.

In terms of the control variables, the regression equation gives similar results as the probability equation. The only difference is that the industry technology level of the firm is positively correlated in the probability equation, but becomes insignificant in the regression equation. It indicates that the level of industry technology affects the likelihood of company innovation and has insignificant effect on the intensity of innovation, which is similar to the findings of Albahari [36]. We also conducted corresponding robustness tests with the number of patents granted as the dependent variable, and the conclusions are broadly similar, thus are omitted from the presentation.

5. Discussion

5.1. The Positive Effect of Science Parks on the Companies

In this study, we found that there existed a significantly positive relationship between the coefficients of the firm location variable and the explained variable which validated our first hypothesis, i.e., science parks can promote the companies’ innovative performance. This finding is consistent with numerous previous studies [1,5,6]. Corrocher et al. [1] unfolded the positive effect of Italian science park on tenants after removing the influence of science park and firm heterogeneity. Lamperti et al. [13] found that science parks play an important role in stimulating company innovation in terms of patent applications. Considering that we also applied the number of patents as the definition of innovation intensity as well as removed the heterogeneity, our study can support the views of these scholars. In addition, we can assume that the positive role of science parks may be universal for both developing and developed countries.

Although all the companies benefited from the science park, the benefits were uneven. Our result showed another finding that companies that lack innovation experience gain more by entering the park than those that already have innovation experience, which can be explained by the influence of resource dependence theory. It was mentioned in Huang’s study [5] that companies with different degree of resource scarcity may get different innovation promotion in the science park, thus leading to different innovation performance. In addition, this finding also confirmed part of Squicciarini’s views [7] who believed that the contribution of the science park to the companies is conditional, depending mainly on the match degree of the characteristics of the companies and science parks.

5.2. The Impact Factors of Science Park on the Companies

We put forward five sub-hypotheses based on the impact that may be exerted on the companies by the factors of the science parks. Among them, the positive effect of the number of innovation support institutions, the industrial output value and the number of high-tech enterprises were confirmed.

The results of this study highlighted the importance of innovation institutions in science parks, which are most likely the media of action of science parks in facilitating the innovation performance of companies. Contrary to the “innovation enclaves” view first raised by Felsenstein’s [10], in which companies rarely interact with research institutions after entering the park, we suggested that science parks do act as a “breeding ground” for new collaborations between companies and research institutions in the park. “Lamperti [13] and Squicciarini [7] also suggested that the most important factor that promotes innovation in science parks is the exchange and cooperation between research institutions and companies, such as university laboratories and technology R&D centers located in the parks.

The industrial output and the number of high-tech companies of science parks played a positive role as we expected. Moreover, the continuous expansion of the industrial scale in the park further increases the knowledge stock and expands the cooperation platform, all of which confirm the existence of agglomeration externalities in the park. It is generally accepted that there is a positive correlation between the agglomeration scale of a park and the intensity of knowledge spillover. For example, Yang [33] believed that the potential R&D externalities and spillover effects are stronger in parks with a higher number of firms, and Montoro-Oloms [37] further suggested that the knowledge spillover effects due to innovation within a science park are the strongest. Since several sub-parks in this study were not established for a long time, it can be expected that the positive externalities from this clustering will continue to grow with the number of firms in the coming period.

However, hypothesis 2b (the area of science parks) was not supported by our results. This may be because localization externalities operate only at a smaller geographical scale [38] and decay very fast spatially with distance [39]. Correspondingly, the impact of innovation dynamics received by companies in the park from other companies and research institutions is mainly confined to a 250 m range [40]. Thus, a larger campus size and a sparser distribution of research institutions may lead to companies receiving less knowledge spillover, which are detriment to stimulating innovation. This is similar to Rammer’s [40] finding that innovative firms are generally concentrated within a very limited distance from research institutions, with a rapid decrease in firm concentration after a distance of more than 50 m and hardly distributed after a distance of more than 1 km.

The commonly perceived positive effect of municipal government funding received by science parks was not significant, probably due to the limitations of the data in this study. On one hand, this demonstration zone’s special funding policy excludes the core park and Lingang park, while many companies with high innovation intensity are located in the core park; on the other hand, each park also has independent support incentives, and some major projects have government subsidies across multiple sub-parks, but the relevant data are difficult to obtain.

5.3. Potential Impact Mechanisms of Science Parks on Firm Innovation

The results of this study have demonstrated that science parks have an impact on the firm innovation, but there are many possible mechanisms. Based on the model results and the actual situation in the Zhangjiang Demonstration Zone, we proposed the following assumptions.

First, science parks played a bridging intermediary role in the cooperation between companies in the park and universities or research institutions, as well as stimulated the knowledge spillover. The model results showed that the number of innovation support institutions in the park was positively correlated with firm innovation intensity. In fact, Zhangjiang Demonstration Zone has accelerated the construction of top research institutions led by national laboratories in recent years. In 2020 alone, 330 new national R&D institutions, such as the Li Zhengdao Laboratory, and 14 national major science and technology infrastructure clusters were constructed. At the same time, the park management committee actively organized industry–university research innovation activities, such as innovation summits to match companies with research institutions. These initiatives successfully promoted firm innovation. For example, the stem cell technology of the Zhangjiang Superorgan R&D Center in the core park has directly contributed to the rapid rise of Microchiro Bio, and the knowledge spillover from Tongji University in Yangpu Park has led to a cluster of more than 3000 knowledge-based innovative companies.

Second, the science park played a supportive role in incubating start-up companies in the park and created an innovative atmosphere. The model results showed that the number of high-tech enterprises in the park was positively correlated with the innovation intensity of the firms. In fact, Zhangjiang Demonstration Zone has built a number of incubators based on upstream and downstream industries, open innovation bases, national-level technology business incubators, branded crowdsourcing spaces, etc., which guided a number of startups full of potential to enter, stay in and graduate from the park. They became the main force of high-tech companies in the park. The development and upgrading of incubation services, the strong innovation tendency of the startups, and the cooperation and competition among similar companies in the park have all contributed to the strong innovation atmosphere in the Zhangjiang Demonstration Zone.

5.4. Policy Suggestions

Based on the above findings, we proposed the following policy recommendations:

First, since science parks are more adept at stimulating companies without patent experience to start innovation, the government can appropriately lower the entry threshold of science parks and encourage more small enterprises lacking innovation capability to enter the parks through policy preferences, so that the use of knowledge spillover and innovation incentives in the parks can maximize.

Second, since the number of innovation support institutions in a science park was positively correlated with the innovation intensity of the company, the government can make an effort to attract more research institutions such as university laboratories and technology R&D centers when planning the park. In addition, building a platform for cooperation between industry, academia and research institutes, and organizing regular exchanges can help to maximize the “hotbed” effect of the science park.

Third, since the area of science parks negatively correlated with the innovation intensity of companies, the government can increase the density of companies in the limited area of parks and evenly distribute research institutions to expand the exchange scope of industry–university research so as to improve the efficiency of knowledge spillover. In such way, small but vibrant science parks can be created.

Fourth, considering that the age of a company negatively correlated with its innovation intensity, the government should pay more attention to companies that have moved in early in the science park. The park should actively help these companies escape the trap of innovation stagnation, and stimulate new vitality of innovation development.

6. Conclusions

This study aimed to explore whether science parks promote company innovation performance and aimed to discern the factors impacting the likelihood of such performance. We formulated six hypotheses according to previous studies and carefully designed the test methods. Based on the 2016–2020 incoPat patent database, Orianan Asia Pacific Enterprise Database and Zhangjiang Demonstration Zone annual report data, we conducted an experiment in the Zhangjiang Demonstration Zone, China, using the generalized exact matching, zero-inflated negative binomial model, and Heckman two-stage model.

The results show: First, if a company is located in a science park, it substantially increases their likelihood of innovation and marginally increases the intensity of innovation, which suggests that science parks do promote the companies’ innovation performance. However, they may be better at stimulating firms with no patent experience to start innovating than the firms with existing patent experience. Second, the impact of science parks on company innovation is mainly attributed to knowledge spillovers from innovation support institutions and high-technology firms in the park, but this positive externality is only limited to a small geographical area.

The findings confirmed our first hypothesis, i.e., science parks can promote the companies’ innovative performance. This is consistent with numerous previous studies including Corrocher et al. [1] and Lamperti et al. [13], but contrary to the findings of Felsenstein [10] who believed the science park to act more as seedbeds of innovation rather enclaves. In addition, our findings proved Squicciarini’s [14] theory that the positive effect of science parks on firms is heterogeneous. In terms of impact factors, three hypotheses were verified including the positive effect of science parks’ industrial output, the number of high-tech companies and innovation support institutions while the two other assumptions were overturned.

Our contributions were threefold. First, we provided new evidences on whether science parks can promote companies’ innovative performance based on full consideration of the characteristics of science parks and companies and further verified the important views, such as that companies benefit unevenly from science parks. Second, we examined the effects of several important impact factors of the science parks on the companies, which offered perspectives for thoroughly understanding the influence mechanism of the science parks in the future. Third, this study investigated the role of science parks in the background of developing countries and examined the extent to which national differences affect the contributions of science parks. It also provided a reference for the research of science parks in other developing countries. In addition, the study results offered valuable policy suggestions to the developing countries suggesting them to take full advantage of the science parks and achieve sustainable development of their innovative economy.

Despite the merits mentioned above, there still exist some shortcomings due to data limitations, such as the limited study scope of only sub-parks in the Zhangjiang Demonstration Zone and the lack of data on the characteristics of companies before they enter the park. In-depth research can be conducted in the future based on the data of collaborative innovation, such as patents and papers of companies in the park. The specific direction and pattern of knowledge spillover in the science park, and its detailed impact on companies with different attributes can also be further revealed.