Abstract: Jointly building the "Belt and Road" is a major measure to conform to the trend of world multipolarization, economic globalization, cultural diversification and social informatization. Using some public data from the World Bank, the Global Innovation Index and the World Competitiveness Yearbook, this paper evaluates the level of scientific and technological innovation in countries along the "Belt and Road" through the existing four least squares estimation algorithms based on the second-order factor model. Studies have shown that scientific research inputs and outputs have shown relatively large contributions to the improvement of a country's scientific and technological innovation level. With the introduction of the quantile level, the role of the economic and social environment, scientific and technological human resources and scientific research input and output have all shown a changing trend that increases with the rise of the quantile level. The impact of total population and GDP on the socio-economic environment is important, and the differences in their impacts are most pronounced at low quantile levels. The impact of R&D researchers and college enrolment on human resources in science and technology, as well as the impact of the number of papers and patent applications in scientific and technological journals on the input and output of science and technology, also showed similar patterns. At high quantile levels, R&D expenditure shows a relatively significant effect on science and technology input and output.
0 Introduction
In recent years, research on belt and road-related issues can be grouped into the following categories:
First, research on the economic and trade development and influencing factors of countries along the "Belt and Road"
The Belt and Road Initiative has significantly promoted the economic growth of countries along the route, and its role has gradually increased over time[1]. The scale of the added value trade networks of countries along the "Belt and Road" has been expanding, and the geographical proximity, common language, direct investment relations and industrial structure differences of countries have staged differences in the positive effect of promoting the evolution of trade networks[2]. From an international perspective, improving the investment environment, optimizing the investment structure, and cultivating innovation-driven core high-quality elements will help enhance China's position in the global value chain division of labor[3].
Second, research on the development status of various industries in countries along the "Belt and Road"
Mainland retail-related industries have certain advantages in countries along the "Belt and Road", and countries and regions along the "Belt and Road" are important target markets for mainland retail industry to "go global" [4,5]. The virtual land imported by the mainland from some countries along the "Belt and Road" mainly comes from cotton, cereals, fruits and oilseeds, and in the future, the area of virtual land imports based on agricultural products should be further expanded and agricultural products trading partners should be developed[6]. The impact of the "Belt and Road" on the financial performance of enterprises in the construction industry is positive, indicating that the participation of enterprises in the "Belt and Road" will promote the performance of enterprises in the construction industry and benefit the development of enterprises [7].
Third, scientific and technological innovation research in countries along the "Belt and Road"
At present, The cooperation mechanism between China and CEECs in scientific and technological innovation has the characteristics of government leadership, multi-party cooperation in cooperation subjects, and uneven cooperation layout in cooperation [8]. Scientific and technological cooperation can be based on the use of the Innovation Composite Index and the Global Innovation Index to compare the level of scientific and technological development between China and CEECs [9]. Focusing on scientific and technological cooperation and collaborative innovation among countries along the "Belt and Road", research on mechanisms such as ally selection, resource input, division of labor coordination, and information communication has been carried out [10]. In addition, many scholars have carried out in-depth research based on the data related to the "Belt and Road" and achieved high-quality scientific research results.
In recent years, China has also introduced corresponding policy plans to vigorously support the "Belt and Road" scientific and technological innovation construction. Jointly building the "Belt and Road" conforms to the interests of most countries and peoples, helps to promote regional cooperation, maintain the global free trade system and an open world economy, and promote the orderly and free flow of economic factors, efficient allocation of resources and deep market integration. In summary, this paper takes the countries along the "Belt and Road" as the research object, and on the basis of comprehensively considering the relevant data of the World Bank, the Global Innovation Index and the World Competitiveness Yearbook, selects seven indicators including total population, total GDP, R&D researchers, enrollment rate of colleges and universities, R&D expenditure, scientific and technological journal papers and patent applications, and conducts quantitative evaluation of the scientific and technological innovation level of countries along the "Belt and Road" with the help of second-order factor models and partial least squares estimation algorithms. It has certain reference value and reference significance.
1 Study design
1.1 Construction of indicator system
At present, many authoritative institutions will regularly release reports, indicators and data on scientific and technological innovation and competitiveness in various countries in the world. Since 1989, the International Institute of Management Development in Lausanne, Switzerland, has published the World Competitiveness Yearbook; Since 2000, the EU Centre for Innovation Policy Research has issued the European Innovation Union Scoreboard; Since 2006, the European Union has issued the Global Innovation Scoreboard; Since 2007, INSEAD has published the Global Innovation Index; Since 2010, the World Economic Forum has published the World Economic Forum Innovation Capacity Index. In addition, the Development Research Center of the China Association for Science and Technology released the National Innovation Capacity Report in 2009; The China Academy of Science and Technology Development Strategy released the National Innovation Index Report in 2013. Many scholars have also published relevant papers and conducted relevant academic discussions. For example, Wang Zhizhi and Liu Li (2015)[11] pointed out that the European Innovation Union scoreboard does not distinguish between the driving factors and the characterization elements well, and there may be a certain degree of double calculation between the input indicators and the output indicators, and the indicators in the European Innovation Alliance scoreboard are mostly average indicators, mainly considering the proportion, and lack of absolute comparison; The World Bank's Knowledge Economy Index does not include R&D funding in the measurements; Most of the innovation environment-related indicators in the National Innovation Index Report need to be measured through subjective indicators, which creates some subjective bias and makes data collection more difficult.
On the basis of synthesizing the existing research reports and related achievements, considering the principles of comprehensive objectivity, conciseness and feasibility of the index system and applicability, the evaluation of the level of scientific and technological innovation in this paper is mainly carried out from three aspects: economic and social environment, scientific and technological human resources and scientific research input and output, including the total population, total GDP, R&D researchers, enrollment rate of colleges and universities, R&D expenditure, scientific and technological journal papers and patent applications, as shown in Table 1.
Table 1 Evaluation index system of scientific and technological innovation level
1.2 Evaluation Methods
From the perspective of weight assignment methods, the common comprehensive evaluation methods of indicators include two types, one is the subjective empowerment method such as simple addition method, analytic hierarchy method, and comprehensive index method; The other category is the entropy weight method, the second-order factor model and other objective empowerment methods. Subjective empowerment methods generally have the characteristics of simple and easy to understand, clear and easy to operate; The objective empowerment law effectively avoids the shortcomings of the subjective empowerment method in terms of insufficient objectivity in determining the weight, and uses data information to construct the hierarchical relationship and inter-correlation between the indicator variables. In addition, different comprehensive evaluation methods have different limitations. For example, simple linear addition has three main limitations:
(1) The assumptions that the variables of each indicator are independent of each other are difficult to meet in actual conditions.
(2) Due to the importance and differentiation of different index variables in evaluation, there is a certain degree of irrationality in the linear accumulation of equal weights of indicator variables.
(3) The result of linear accumulation depends on the combination and number of indicator variables, and the higher the value of the indicator contained in the combination, the higher the number, the higher the score of the combination.
The evaluation results of the analytic hierarchy method vary depending on the judgment matrix, and the comparison of the indicators by using the nine-level sub-system is prone to contradictions; The composite index method is difficult to determine the comparative criteria, and the evaluation results are too dependent on the comparative criteria [12]; The entropy weight method determines the objective weight according to the size of the index variability, but it is only used in the weight determination process, and the scope of use has certain limitations.
The second-order factor model fully considers the correlation between variables, establishes the relationship between the latent variables and their relationship with the measurable variables, can reflect the actual data more objectively, and has good explanatory performance, which has been widely used [13]. As an estimation method, the partial least squares algorithm does not have particularly strict requirements for the distribution of the data, that is, the measured index does not have to meet the harsh conditions of obeying the multivariate normal distribution and is independent of each other, and the factor score can be calculated while completing the parameter estimation, which is convenient for comprehensive evaluation and ranking [14,15]. Existing partial least squares estimation algorithms mainly include the Repeated Indicators Approach (RI), the Two-step Approach (TS) and the Hybrid Approach (H). Among them, the repeating indicator method is to assign all the measurable variables assigned to the low-order potential variables and then reassign them all to the higher-order potential variables. The two-stage method consists of two steps: (1) calculate the principal components of the measurable variables contained in each low-order potential variable as the value of the low-order potential variables. (2) Based on the structural model and the values of low-order potential variables, the estimation is completed. The hybrid method refers to randomly assigning half of the measurable variables to the low-order potential variables they reflect, and the rest to the higher-order potential variables. In addition, there is a class of Quantile Regression Estimation (QR) based on quantile regression, which is used to show the level of technological innovation at different quantile levels when the data exhibits the characteristics of the bias distribution [16]. According to the characteristics of the national science and technology innovation evaluation index system, four existing least squares estimation algorithms are used to complete the estimation of model parameters and factor scores.
In summary, the expression form of the evaluation model of the scientific and technological innovation level of the countries along the "Belt and Road" includes two parts: the measurement model and the structural model. Unlike traditional structural equation models that partial least squares estimation algorithms, τ in the model represents quantiles and refers to the estimation results of partial least squares estimation algorithms based on quantile regression. For traditional structural equation models such as the repeating index method, the two-stage method, and the mixed method, the least squares estimation algorithm should be ignored.
Equation (1) reflects the multi-level relationship between the measurable variable MVih and the first-order factor FOLVi. λih,τ(i=1,2,3;h=1,2,3;τ=0.1,0.25,0.50,0.75,0.90) represents the load factor of the relationship between first-order factors and measurable variables. εih, τ is the measurement error of the hth measurable variable MVih in the first-order factor FOLVi, the mean is 0, the variance is δ2ih, τ, and is not associated with the first-order factor FOLVi. Equation (2) reflects the multi-level relationship between the first-order factor FOLVi and the second-order factor SOLID. βi, τ (i=1,2,3; τ=0.1,0.25,0.50,0.75,0.90) represents the path coefficient of the relationship between second-order and first-order factors. γi, τ is the measurement error of the first-order factor FOLVi, the mean is 0, the variance is δ2i, τ.
The measurement model (1) and the structural model (2) together constitute the evaluation model of the scientific and technological innovation level of the countries along the "Belt and Road", as shown in Figure 1.
Figure 1 Evaluation model of scientific and technological innovation level of countries along the "Belt and Road"
1.3 Data Sources
The data in this article is from publicly available data such as the World Bank, the Global Innovation Index and the World Competitiveness Yearbook from 2015 to 2017, including 64 countries along the Belt and Road ("China Belt and Road Network": https://www.yidaiyilu.gov.cn/jcsjpc.htm). A table of country names and symbols is shown in Table 2.
Table 2 Country names and symbols
In order to understand the scientific and technological innovation of countries along the "Belt and Road", this paper statistically describes the indicator data by minimum, first quantile, median, mean, third quantile and maximum value, and the results are shown in Table 3.
Table 3 Statistical description of scientific and technological innovation indicators of countries along the "Belt and Road"
From Table 3, it can be found that the values of the 64 countries along the "Belt and Road" in terms of 7 scientific and technological innovation indicators are quite different (the maximum value is subtracted from the minimum value), and it is more appropriate to consider it at different quantile levels when measuring the structural relationship between the indicators. In addition, in the 7 indicator data of 64 countries along the "Belt and Road", 6 indicators except for the total population (MV11) have different degrees of missing conditions. This article uses the median to replace missing data before estimating the model parameters.
2 Evaluation results and analysis
2.1 Parameter estimation results
The above partial least squares estimation algorithm allows you to estimate the path coefficients and load factors in the model. The path coefficient shows that the importance of RI, TS and H to obtain first-order factors is ranked from high to low: scientific research input-output (FOLV3), economic and social environment (FOLV1), and scientific and technological human resources (FOLV2). The importance ranking of the first-order factors obtained by QR is from high to low (positive to negative) as follows: scientific and technological human resources (FOLV2), scientific research input-output (FOLV3), and economic and social environment (FOLV1). Moreover, when the quantile levels are 0.10, 0.25 and 0.50, the path coefficient of the economic and social environment (FOLV1) is negative, indicating that for the medium and low level situations, the impact of the economic and social environment (FOLV1) on scientific and technological innovation (SOLV) in the countries along the route is negative.
The load factor shows that the coefficients between the first-order factors estimated by RI and TS are not much different between the economic and social environment (FOLV1) and the scientific and technological human resources (FOLV2) and their respective corresponding two measurable variables, and the coefficients between scientific research input and output (FOLV3) and the last two measurable variables of scientific and technological journal papers (MV32) and patent applications (MV33) are closer, which is greater than that of scientific research input-output (FOLV3) and measurable variable research and development expenditure (MV31). For example, in RI estimates, the coefficient for the total population (MV11) → economic environment (FOLV1) is 0.948, the coefficient for the economic and social environment (FOLV1) → the total GDP (MV12) is 0.978, the coefficient for science and technology human resources (FOLV2) → R&D researchers (MV21) is 0.890, and the coefficient for science and technology human resources (FOLV2) → the enrolment rate of colleges and universities (MV22) is 0.835. The H method randomly assigns all measurable variables to second- and first-order factors, so the load factors do not all express the relationship between the measurable variables and the first-order factors, and some may be the relationship between the measurable variables and the second-order factors.
The QR method shows that when the quantile level τ = 0.50, the load factor of the total population of the first measurable variable (MV11) is greater than that of the total GDP of the second measurable variable (MV12), and the load coefficient of the total GDP (MV12) of the total GDP (MV12) of the remaining quantile levels is greater than that of the total population (MV11). Human Resources for Science and Technology (FOLV2) reflects the same law in the load factors of the first measurable variable R&D researcher (MV21) and the second measurable variable, college enrolment (MV22). For research input-output (FOLV3), When the quantile level τ is less than or equal to 0.50, the scientific journal paper (MV32) is always the measurable variable that contributes the most to the input and output of scientific research (FOLV3), and its load factor is much greater than that of the other two measurable variables; when the quantile level is greater than 0.50, the R&D expenditure (MV31) becomes the measurable variable that contributes the most to the input and output of scientific research (FOLV3), and the load coefficient of the scientific journal paper (MV32) and the amount of patent application (MV33) is not much different. This is shown in Table 4.
Table 4 Parameter estimation results
2.2 Factor score analysis
For the 64 countries along the Belt and Road, the ranking results of national technological innovation using the repetitive index method (RI), two-stage method (TS), mixed method (H) and partial least squares (QR) based on quantile regression were slightly different. According to the factor score of the state of scientific and technological innovation (SOLV) along the "Belt and Road", this paper ranks the scientific and technological innovation level of all countries and divides them into three segments: high, medium and low. Table 5 shows countries that are ranked in the same rank regardless of the methodology.
Table 5 List and ranking of countries with consistent rankings under all methodologies
Table 5 shows that countries in the highest ranking in terms of the status of scientific and technological innovation include Russia, Turkey, Poland, Slovenia, Lithuania, Singapore, bulgaria; Countries in the middle of the ranking in terms of the status of scientific and technological innovation include Romania, Georgia, Albania, Montenegro, Yemen, Bahrain, Kyrgyzstan, the United Arab Emirates, Thailand, the Philippines, Myanmar, Afghanistan, Armenia, Turkmenistan, Bhutan, Timor-Leste, Maldives, North Macedonia and Kazakhstan; Countries that rank low in the status of scientific and technological innovation include Tajikistan, Azerbaijan, Syria, Laos, Nepal, Jordan, Qatar, Sri Lanka and Uzbekistan.
For countries with different rankings under different methods, this paper focuses on the estimation results of partial least squares estimation (QR) based on quantile regression estimation to study the ranking of countries at different quantile levels. According to statistics (except for the countries shown in Table 5), countries in the highest rank at different quantile levels include Estonia, the Czech Republic, Latvia and Serbia, the countries in the middle rank include Ukraine, Bosnia and Herzegovina, Palestine, Moldova and Indonesia, and the countries in the lower bracket include Vietnam, Bangladesh and Pakistan. Table 6 shows countries that are ranked at different levels of quantile.
Table 6 List and ranking of countries with inconsistent rankings under the QR method of different quantiles levels
Note: QR10, QR25, QR50, QR75, and QR90 indicate partial least squares estimation based on quantile regression estimation when the quantile levels τ are 0.10, 0.25, 0.50, 0.75, and 0.90, respectively.
As can be seen from Table 6, Belarus (BLR) is in the high band at the quantile level 0.10 to 0.75, but belongs to the middle stage at the quantile level 0.90; Croatia (HRV) belongs to the middle stage at the quantile level 0.75 and belongs to the high stage at the other quantile level; Iran (IRN) belongs to the middle stage at the quantile level 0.50 and the high stage at the other quantile level. This shows that under different quantile levels, some countries will have different performances in the ranking of scientific and technological innovation status.
3 Conclusion
In order to study the level of scientific and technological innovation of countries along the "Belt and Road", this paper selects some public indicator data of the World Bank, the Global Innovation Index and the "World Competitiveness Yearbook", and uses the second-order factor model and the partial least squares estimation algorithm as the analysis method to evaluate the scientific and technological innovation level of 64 countries along the "Belt and Road" from three aspects and 7 indicators: economic and social environment, scientific and technological human resources and scientific research input and output.
The results show that no matter which partial least squares estimation algorithm is used, the input and output of scientific research show a relatively large contribution to the improvement of a country's scientific and technological innovation level. When considering different quantile levels, the roles of the economic and social environment, scientific and technological human resources and scientific research inputs and outputs all show a trend of increasing with the rise of quantile levels. The impact of total population and GDP on the economic and social environment is important, and the difference in impact is most pronounced at low quantile levels. The impact of R&D researchers and college enrolment on human resources in science and technology, as well as the impact of the number of articles and patent applications in science and technology journals on the input and output of science and technology, also showed similar patterns. At high quantile levels, R&D expenditure shows a relatively significant effect on science and technology input and output. Therefore, it shows that the partial least squares algorithm based on quantile regression estimation can more carefully and comprehensively portray all aspects affecting the level of scientific and technological innovation and the different performance of each index at different quantile levels, and it is more convenient to find out the key factors and problems in a targeted manner. The least squares algorithm used in this paper also provides method support for the re-ranking of the scientific and technological innovation level of countries along the Belt and Road. In addition, the level of scientific and technological innovation in countries along the "Belt and Road" is not static, but may appear different development and change laws with time, events and other factors. Therefore, on the basis of the research in this paper, considering the dynamic change trend based on the relationship between data structures at different time points, predicting the future development trend of national scientific and technological innovation, and the important factors that may promote or restrict the development of scientific and technological innovation, it is quite promising.
Bibliography:
[1] Cao Xiang, Li Shenting." The impact of the Belt and Road Initiative on the economic growth of countries along the Belt and Road and the role of China[J].World Economic Research,2021,(10).
[2] Ge Chunbao, Yu Jinping, Liu Yapan." The Evolution of the Belt and Road "Added Value Trade Network and Its Influencing Factors[J].Finance and Economics Series,2021,(10).
[3] Cao Jianping, Zhang Miao." The Empirical Evidence of the Belt and Road Direct Investment Network and the Position of Global Value Chains[J].Statistics and Decision-making, 2020,(11).
Chang Jiancong. Analysis of "Global Localization" Business Factors of Mainland Retail Industry Docking the "Belt and Road" Market: An Empirical Study Based on Questionnaire[J].Trade Show Economy,2021,(19).
Zhu Ruiting. Analysis of the Way for China's Retail Industry to "Go Global" and Connect with the "Belt and Road"——Based on the Perspective of Target Market Selection[J].Economic Research Reference,2017,(64)."
Zhao Yuqi,Wang Yunfeng. Analysis of virtual land import benefits of agricultural trade between China and major countries along the "Belt and Road"[J].Agriculture and Technology,2021,41(18).
[7] Qin Ying, Sun Limei, Feng Xiaoyang." Research on the Impact of the Belt and Road Initiative on the Financial Performance of Construction Enterprises in Mainland China: Based on PSM-DID[J].Journal of Beijing University of Civil Engineering and Architecture,2021,37(3).
Zhang Haiyan,Xu Lei. Analysis on the potential and key areas of scientific and technological innovation cooperation between China and Central and Eastern European countries[J].Regional Economic Review,2021,(6).
[9] Xin Hong,Li Zhenqi. Research on Scientific and Technological Innovation Capability and Scientific and Technological Cooperation between China and Central and Eastern European Countries[J].Science and Technology Management Research,2021,41(9).
[10] Fang Weiwei." Research on the Mechanism of Scientific and Technological Cooperation and Collaborative Innovation in The Belt and Road Countries[J].Chongqing Social Sciences,2020,(12).
Wang Zhizhi,Liu Li. Comparative analysis of national innovation ability evaluation indicators[J].Scientific Research Management,2015,36(S1).
Jia Pin,Li Xiaobin,Wang Jinxiu. Comparison of several typical comprehensive evaluation methods[J].Chinese Hospital Statistics,2008,15(4).
Yi Danhui. Structural Equation Model:Methods and Applications[M].Beijing:Chinese Min University Press,2008.]
[14] Lohmoller J B. Latent Variable Path Modeling With Partial Least Squares [M]. Heidelberg：Physica-Verlag,1989.
[15] Robert W G, Bruce R K, Herman O A W. Partial Least Squares Path Modeling With Latent Variables [J]. Analytica Chimica Acta,1979,112(4).
[16] Koenker R. Quantile Regression [M]. Cambridge: Cambridge University Press,2005.
Author: Cheng Hao1, Glory Hua 2
1 Institute of Innovation Strategy, China Association for Science and Technology
2 Faculty of Science, Beijing University of Technology
Project Source: Funded by the National Natural Science Foundation of China (72001197; 11701021）； National Statistical Science Research Project (2021LY052)
This article was originally published in Statistics and Decision-making, No. 7, 2022
Thank you for your support and attention, and welcome to exchange ideas
Submission email: [email protected]