3.1. Empirical Data and Sample Selection
This study aims to determine the factors affecting the innovation performance of universities in China. There are currently 64 universities under the direct management of the Ministry of Education in China. As representatives of Chinese universities, they have played an exemplary role in teaching, scientific research, and social services. In this study, we used secondary data based on a deductive methodology from 2009 to 2019 from 64 Chinese universities administrated by the Ministry of Education of China in STATA 15. The sample data were collected from the “Compilation of Science and Technology Statistics of Higher Education Institutions” published by the Ministry of Education of China from 2010 to 2020, which provides detailed information on the scientific research human resource, funding, institutions, projects, international communication, achievements, and technology transfer of universities in China from 2009 to 2019. According to the research needs and the availability of data, 704 panel data from 64 colleges and universities directly under the Ministry of Education in China in the past 11 years were selected as observation samples.
The dependent variable is the innovation performance of a university (IPU). Guler (2010) used the number of patents and papers as indicators to measure the country’s innovation performance [
21], and Bozeman and Lee (2005) used the total number of academic papers published annually to measure the innovation performance of universities [
22]. The indicators that the Ministry of Education of China considers for the statistics of scientific and technological achievements of its universities each year include the publication of scientific and technological works, academic papers, national-level research projects, patent status, and other intellectual property rights [
23].
In this paper, we selected the number of scientific and technological works, published academic papers, and national-level research projects per year of a university, and integrated them into a comprehensive indicator by principal component analysis (PCA) to observe the innovation performance of the university.
The process of international scientific research collaboration is a process of sharing resources, which includes personnel dispatch and exchanges, international academic conferences, collaborative construction of research platforms, technology introduction, and collaborative publications. In essence, collaborative publications and the collaborative construction of research centers are often the results of the work of university researchers under certain circumstances (such as academic visits and international conferences). In some studies, indicators such as the number of international collaborative papers are considered the result of collaboration to measure collaboration performance [
24].
In this paper, we consider the intensity of international research collaboration of universities (IIC) from two dimensions: personnel exchanges (PE) and international academic conferences (IAC). According to the statistics of the Ministry of Education of China, the annual number of personnel dispatched abroad, personnel receiving visits from abroad, and personnel participating in international academic conferences are selected to reflect the independent variables of PE. Meanwhile, the number of international conferences sponsored by universities each year was selected as the variable of IAC. Therefore, based on hypothesis H1, we propose hypothesis H1a, which is that international personnel exchange has an inverted U-shaped impact on the innovation performance of universities, and H1b, which is that university international academic conferences have an inverted U-shaped impact on the innovation performance of universities.
The moderating variable is the type of university (UT), which is divided into two categories according to whether it is supported by special financial funds for the construction of global first-class universities. The construction plan started in 2017, and the panel data analyzed in this study is from 2009. In fact, the Chinese government’s special construction plans for these universities have been in existence since the 1990s. Among them, the “985 Project” started in 1998, and the current construction plan “Double First-Class Project” is considered a continuation of the “985 Project”: both of which are national strategies proposed by the central government to enhance comprehensive strength; the universities supported by these two plans are almost the same and the 39 universities listed in the “985 Project” have also been supported by the construction of the “Double First-Class Project”.
To discuss the impact of international cooperation on the innovation performance of universities under different university types, we divided the 64 universities directly under the Ministry of Education into two categories according to whether they are currently supported by the “Double First-Class Project” construction plan. Therefore, based on hypothesis H2, we propose the alternative hypothesis H2a, which is that university type has a significant moderating effect between international personnel exchange and the innovation performance of universities, and H2b, which is that university type has a significant moderating effect between international academic conferences and the innovation performance of universities. The classification of 64 universities directly under the Ministry of Education of China is shown in
Table 1.
Controlling factors such as research funding (RF), number of scientific researchers (SR), and time also have an impact on the innovation performance of universities. Therefore, the above three variables are included in the model as control variables. Research funding for a Chinese university mainly includes government funds, enterprises and institutions entrusted funds, and other funds. “Government funds” refers to the scientific research funds allocated by the government to all kinds of universities at all levels. The entrusted funds of enterprises and institutions are obtained by universities from foreign enterprises and institutions, which include the funds allocated by the research institutes of the Chinese Academy of Sciences. Other funds are obtained through other channels for research, development, and scientific and technological services in the current year.
Because the amount of research funding has a certain impact on the output of innovation performance, it is included as a control variable in the model. However, due to the large number of scientific research funds, we used Zhang’s (2018) method to process the funds’ data and changed the unit of scientific research funds from yuan to 10,000 yuan in order to avoid the large difference between the control variable and other variables, which was likely to result in the change of the stability of the data [
25].
The indicator of the number of scientific researchers was selected from the statistics of the Ministry of Education for the number of full-time personnel in R&D. Meanwhile, during the 2009 to 2019 period under study, China’s economic, social, and technological development underwent tremendous changes, and academic research and innovation output has also been affected by changes over time.
It is necessary to control the impact of time on the model. We use years as the unit of time and calculate it in the form of dummy variables. The definition and basis of each variable are shown in
Table 2.
3.2. Model Specification
Panel data are a combination of cross-sectional series data and time series data. Building a dynamic panel model based on panel data can solve the endogenous problems caused by missing variables, measurement errors, or the model’s own causes. It can effectively avoid biased and non-uniform problems caused by random effects or OLS fixed-effect methods [
23].
The panel data used in this paper came from 64 universities in China from 2009 to 2019. Since the level of innovation capacity is a dynamic evolution process, it is suitable to use a dynamic panel estimation model for quantitative dynamic analysis. The GMM model is often used in the dynamic panel estimation model. The GMM model includes the differential generalized method of moments (DIF-GMM) model and the system generalized method of moments (Sys-GMM) model. DIF-GMM removes the effects of individual effects by making first-order differences in the equations. It eliminates the problem of incomplete estimation caused by variables that do not change over time [
26]. Sys-GMM has high estimation efficiency and retains variable coefficients that do not change over time. In addition, it can flexibly select instrumental variables so that the estimation results have less bias [
27]. A basis criterion for Sys-GMM application is N (the number of cross-sections) > T (period), which was fulfilled by the data of this study, where T = 11 and N = 64 (N > T).
To ensure the reliability and accuracy of the regression results, we need to test the quality of the selected data for multicollinearity and stationarity when using panel data for econometric demonstration. Specifically, we mainly tested the stability of panel data and the collinearity before variables. The multicollinearity test is used to test whether the autocorrelation problem exists in the disturbance term, and the stationarity test is used to test whether the data samples tend to be stationary [
28].
Second, the over-identification constraint test and the autocorrelation of residual sequence are carried out, including the over-identification constraint test to test the effectiveness of the instrumental variables selected in the model and the residual sequence autocorrelation test to test the overall robustness of the model.
The empirical analysis part of this paper first tests the overall data and then carries out descriptive statistics and regression analysis. It verifies whether the international research collaboration activities of universities directly under the Ministry of Education of China have a direct impact on the innovation performance of universities and whether the types of universities have a moderating effect.
The general form of the dynamic panel model is as follows:
where
i denotes the cross-sectional units, of which there are 64 in our sample, and
t expresses time, which is 11 years in our sample;
yit−1 is the first-order lag variable of the explained variable, reflecting the influence of historical behavior on current behavior.
Xit is the explanatory variable,
ui is the misspecific variable, and
εit is the random error term.
In order to test the non-linear hypotheses H1a and H1b, we constructed non-linear dynamic panel models of the intensity of university international research cooperation innovation performance of universities from two aspects: personnel exchange and international academic conferences. The non-linear dynamic models of this study can be written as follows:
where
IPU represents the innovation performance of universities,
PE represents the personnel exchange,
IAC represents the international academic conferences,
RF,
SR, and
year are the control variables representing research funding, scientific researcher, and time respectively. If α
2 is positive and α
3 is negative, it indicates that
PE and
IAC have an inverted U-shaped relationship with
IPU, respectively, and hypotheses H1a and H1b can be verified.
Moreover, in order to discuss the moderating effect of the type of university on the relationship between the international research collaboration intensity and the innovation performance of the university, and to test hypotheses H2a and H2b, the interactions between moderator and independent variables are introduced into the models. The moderating impact of university-type interaction terms can be written as follows:
where
UT represents the types of universities,
PE*UT and
PE2*UT are the interaction terms between personnel exchange and the university types, whereas
IAC*UT and
IAC2*UT refer to the interaction terms between international academic conferences and university types. If the coefficients of interaction terms
α4 and
α5 are significant, it indicates that the moderating effect of university types is significant, and hypotheses H2a and H2b can be verified.