An Empirical Analysis of the Impact of the China Free Trade Zone on the Level of Human Capital Based on the Asymptotic PSM-DID Model

Te article selects relevant data from 283 cities in China from 2000 to 2020 and verifes the impact of the establishment of free trade zones on human capital levels based on the asymptotic PSM-DID model. It is found that the establishment of FTZs can signifcantly contribute to the improvement of regional human capital levels, but the promotion efect has a certain lag. Finally, the article gives constructive suggestions based on the empirical results.


Introduction
As the Chinese economy continues to rapidly develop, in order to better stimulate market vitality and social creativity, China has undertaken a series of domestic deepening reform measures and external globalization strategies, seeking new economic growth points and improving the quality of economic growth through high-level opening-up. Te emergence of free trade pilot zones is an important policy for deepening internal and external reforms and plays a decisive role in China's economic resilience and sustainable development. Innovation of the human resources system is one of the important tasks of free trade zone construction, is a key means to enhance the international competitiveness of the free trade pilot zones, and is the fundamental guarantee for the high-quality development of the free trade pilot zones in China.
In recent years, more and more scholars have begun to pay attention to the research on free trade zone policies. From the perspective of innovation-driven development, the economic growth, international trade, competition, and spillover efects brought about by the establishment of free trade zones will strongly promote the improvement of regional innovation and development levels (Gao and Li [1]). Free trade zones implement the negative list access principle, while canceling the vast majority of tarif and nontarif barriers for goods, promoting the free fow of factors such as goods, capital, technology, and labor. Te new classical growth model emphasizes that capital fows are the source of economic growth. Zhang and Yu [2] combined the construction of free trade zones and believed that the establishment of free trade zones could promote urban capital fows, drive technological innovation, and force domestic market-oriented reforms to improve the level of marketization, thereby afecting urban regional economic growth. Fang [3] believed that the construction of free trade zones could increase the driving force for optimizing and adjusting regional industrial structures through channels such as import expansion efects and fnancial agglomeration efects.
Te research of the above scholars mainly focuses on the relationship between relevant policies, regional innovation, and economic growth. Te establishment of a free trade zone will strongly promote the improvement of regional innovation and development levels. Te implementation of a negative list model and the elimination of tarif barriers will promote factor mobility. However, the efect of human resource system innovation, as an important task of free trade zone policy, lacks empirical evidence. Most scholars have studied the diferences in human capital levels from the perspective of trade openness. However, as a high-level policy of opening to the outside world, there is little research on the free trade zone. Tis paper will conduct an empirical study on the impact of the establishment of a free trade zone on human capital.
Te diference-in-diferences (DID) method is one of the important methods for evaluating policy efects. Zhou et al. [4] used this method to explore the impact of the clean energy demonstration province policy on regional carbon emissions and economic development. Yang et al. [5] used a combination of synthetic DID and generalized synthetic control methods to evaluate the impact of the establishment of a free trade zone on urban green total factor productivity. Shao et al. [6] used the DID method to investigate the efect of the free trade zone policy on regional carbon emissions against the background of the dual carbon targets, and the results showed that the pilot policy of the free trade zone can signifcantly reduce regional carbon emissions. Li and Wang [7] used this method to investigate the impact of the new energy demonstration city policy on urban green vitality based on a quasi-natural experiment. Yang et al. [8] analyzed the impact of the host country's tax environment on investment efciency based on direct investment data from countries along the belt and road.
Based on the above, this paper selects relevant data from 283 prefecture-level cities in China from 2000 to 2020 and uses propensity score matching double diference, a commonly used method for policy evaluation, to verify the causal relationship between free trade zones and human capital level in order to propose relevant suggestions to optimize the human capital resource innovation system in the construction of free trade zones, promote regional human capital investment and accumulation, drive regional innovation and development, and promote regional economic growth, thereby enhancing China's international competitiveness in the world.
Te innovations in this paper mainly include the following aspects: (1) Tis paper verifes the causal relationship between free trade zones (FTZs) and human capital levels, expands the research on FTZs, and enriches the perspective of FTZ research. (2) Te paper corrects the model for evaluating the efect of FTZs on human capital level and uses the propensity score matching double difference method to verify the impact of FTZs on human capital level. Te combination of diference-in-diferences with propensity score matching solves the problems of endogeneity and sample selection bias. In addition, the paper adopts the method of annual matching to avoid the problems of time mismatch and self-matching caused by mixed matching. (3) Most of the existing literature uses provincial panel data, but the establishment of FTZs is in a small area of a prefecture-level city, and the scope is relatively small. Using provincial panel data may exaggerate the efect of FTZ policies and lead to unconvincing conclusions. Terefore, this paper selects data from 283 prefecture-level cities from 2000 to 2020 to make the relationship between the two more reliable.

Theoretical Foundation
From a theoretical perspective, free trade can achieve efcient allocation of resources on a global scale, especially from the perspective of regional comparative advantage, by using the comparative advantages of diferent regions to allocate resources reasonably, achieve complementary advantages, and achieve sustainable development. Te development of free trade has a certain promoting efect on the rapid and free exchange of human capital. Classical political economy emphasizes the role of material capital. American economists Schultz and Becker analyzed the promotion of human capital investment on social and economic development from macro-and microperspectives, respectively. Since 1980s, the new theory of economic growth has been concretized, and mathematical methods have been used to further explain the impact of endogenous variables such as human capital and education level on economic growth.
Currently, there is little research on the impact of the establishment of free trade zones on the level of human capital. Lv and Liu [9] verifed the causal relationship between free trade zones and human capital investment and analyzed the heterogeneity of human capital investment between rural and urban areas. From the above, we can know that the establishment of free trade zones promotes institutional innovation and then promotes free trade. Although there is little research on the relationship between free trade zones and human capital, we can shift our focus to research on the relationship between trade openness and human capital level. For example, Findlay and Kierzkowski [10] introduced human capital into the H-O model, endogenizing both worker wages and education costs, and then verifed that trade openness can afect workers' ability to obtain lower wages from nonskilled labor or higher wages from education investment. Since then, the literature on the relationship between trade openness and human capital has continued to increase. According to existing literature [11], the impact mechanism of trade openness on human capital level can be summarized into four aspects: mainly wage price mechanism, technological progress mechanism, capital input mechanism, and credit constraint mechanism. For the wage price mechanism, scholars believe that trade openness promotes the increase in the ratio of high-skilled labor wages to low-skilled labor wages, followed by the promotion of human capital investment and the accumulation of human capital. For the technological progress mechanism, Acemoglu [12] believes that foreign trade will cause the skill required for a certain job to increase, and correspondingly, talent will continue to improve its skills. Chen and Zhao [13] found that under the background of dual labor market, the efciency and complexity of export technology not only beneft the investment in human capital in urban and rural areas but also help to increase investment in children's education and the long-term human capital investment of workers. Te capital input mechanism is mainly that trade openness promotes cultural exchange, further afecting individual education investment in skills and knowledge, and at the same time, empirical results show that due to the opening of trade, promoting economic growth, residents' income will also increase, and will not be limited by credit constraints, allowing residents to increase their education investment.

Research Hypotheses.
Te formation of human capital stems from investments in human capital. Free trade pilot zones have implemented many measures in the areas of talent introduction and talent service guarantee, such as direct fnancial support and various forms of service guarantee including housing, medical care, and education (Tang and Wu [14]). Currently, Liu and Wang [15] have verifed that the establishment of free trade zones promotes the improvement of regional innovation levels. Under the guidance of the consumption structure, new products, new services, and the diversity of foreign products and services brought about by trade liberalization have promoted the growth of efective demand. According to Keynes' theory of national income determination, the multiplier efect generated by the increase in efective demand will lead to an increase in income, which will have a positive impact on investment in human capital, ultimately afecting the level of human capital. Te author also points out that the radiationdriven efect of free trade zones is conducive to improving the level of regional integration, reducing income disparities, and increasing investment willingness and level. Based on relevant research, this article puts forward the following hypotheses: (1) Te role of industrial structure upgrading in the pilot policies of free trade zones and the level of regional human capital. Free trade zone pilot policies have an obvious promoting efect on industrial structure upgrading (Fang [3]; Li and Li [16]). On the one hand, as a highlevel open policy, the free trade zone simplifes customs clearance procedures and implements a negative list management system, that is, through "rough" trade facilitation measures, breaking down trade barriers and realizing the refow and aggregation of various factors, reducing the fnancing threshold for industrial development, thus promoting regional industrial division of labor and location selection (Hamada [17]; Krugman [18]; Cai and Xu [19]). On the other hand, free trade zone pilot policies are conducive to industrial agglomeration, improving labor productivity, and the cities where free trade zone pilot policies are established have radiation efects, driving the formation of industrial upgrading in surrounding cities. Industrial structure upgrading has a certain promoting efect on the level of regional human capital. Industrial structure upgrading will provide space for the development of technology-intensive industries, and the application of advanced technology will increase the demand for labor skills, driving the improvement of the regional level of human capital. At the same time, areas with advanced production technology and generous wages are more likely to attract the gathering of human capital from surrounding areas (Ni and Ding [20]). (2) Increasing investment in education can signifcantly promote the improvement of human capital. Cities that implement policies to establish free trade zones have more complete transportation, communication, and transportation facilities, which reduce the mobility costs of individuals and make communication between regions easier, thereby having a greater impact on people's ideas and being more conducive to strengthening education investment in various regions (Lv and Liu [9]). At the same time, the free trade zone pilot policy signifcantly promotes regional economic growth (Heand Tang [21]), increases the income level of local residents, and enhances education investment. Education is the main way to form human capital, and government education investment is positively correlated with the accumulation of human capital growth, and education investment has a long-term impact. Te more fnancial investment there is in education, the more educated the population will be, and the length of time that workers receive education will increase, which is benefcial to the improvement of professional human capital. Hou and Liao [22] used the VECM model to verify that government education investment is positively correlated with the accumulation of human capital growth and that there is a stable equilibrium relationship between fnancial education investment and human capital. Te short-term impact of fnancial investment in education on human capital level is less than the long-term impact.

Progressive PSM-DID Model Construction.
Te diference-in-diferences (DID) method is commonly used to evaluate the efects of randomized or natural experiments. It evaluates the changes in the dependent variable based on the counterfactual framework of a policy occurring or not occurring. To make the model more efective, the parallel trend assumption needs to be met, which means that the average level of human capital in the control group and treatment group should have parallel trends over time in the absence of the policy trial. Terefore, the key to implementing the DID model is to fnd an appropriate control group that is not afected by the policy and is extremely similar to the treatment group.
To improve the reliability of the DID model, this paper uses the propensity score matching (PSM) method proposed by Heckman [23]. Teoretically, the PSM method is suitable for nonrandom data, and its theoretical framework is the counterfactual reasoning model. Te selection of trial cities for the free trade experimental zone in this study is not random but related to their economic development, transportation conditions, geographical location, and other factors, indicating that the sample in this study is Discrete Dynamics in Nature and Society nonrandom. In addition, there are signifcant diferences between cities, making it difcult to fnd a control group that is very similar to the treatment group in all aspects. Terefore, the propensity score matching method can solve the problem of sample selection bias, control confounding factors, and reduce interference.
PSM can solve the problem of sample selection bias, but cannot avoid endogeneity caused by omitted variables. Te DID model can solve endogeneity, but there is a problem of sample selection bias. Terefore, this study combines the PSM and DID models to verify the efects of the free trade experimental zone policy on regional human capital levels.
Te PSM-DID model frst needs to perform propensity score matching and then conduct the DID based on the data after matching. Since the free trade experimental zones in this study were established at diferent times, the model used is as follows: where Control it denotes the control variables, and the coefcient of the interaction term between the two β 1 denotes the net policy efect of the establishment of the FTZ which is the focus of the article.
In addition, to further validate the dynamic marginal impact of the pilot FTZ policy on regional human capital levels, the article introduces a time dummy variable based on the model, as follows: where Time2015 and Time2020 correspond to the time dummy variables for 2015 and 2020, respectively. In the meantime, the steps for operating the multitemporal PSM-DID model are shown in Figure 1 , and 2020 as the control group due to the relatively short establishment period, which makes it difcult to refect the efect of the pilot policies of FTZs. Te establishment of the pilot free trade zone policy is only a slice of the world, and the use of provincial panel data would exaggerate the pilot free trade zone policy efect, making the fndings after the PSM-DID test inaccurate and unconvincing. Terefore, the article narrows the scope of the study to prefecture-level cities. Excluding cities with more missing data (Tibetan region, Hainan region, Zhongwei city, Haidong city, Turpan city, Hami city, etc., while the article does not consider Hong Kong, Macao, and Taiwan regions at this time given the availability of data), the fnal sample examination period is 28 prefecture-level cities from 2000 to 2020. Te study targets the frst three batches of established free trade pilot zone; Pingtan city data which is seriously missing will be excluded, so the treatment group is 21 prefecture-level cities, treatment group cities, and the year of establishment as shown in Table 1. For the missing data of some prefecture-level cities, multiple interpolation methods were used to complete the data by referring to relevant literature. In addition, the GDP defator was used to exclude the infuence of price factors and make the results more convincing. Te data in the article were obtained from the China Urban Yearbook, the China Regional Economic Yearbook, and provincial and municipal statistical yearbooks.

Explanatory Variables.
Education is the most important component of human capital, and a higher level of education represents a higher level of education investment. Educational investment is the main way to accumulate human capital, and therefore, the level of education can naturally measure the stock of human capital for individuals or economies. Domestic scholars generally use the average years of education to measure it, but considering the large amount of missing data, this article refers to Lan et al.'s [24] indicator for measuring the level of human capital, selecting the number of students in ordinary high schools as a proxy variable for the level of human capital.

Core Explanatory Variables.
We refer to the cities that established free trade zones during the sample period as the treatment group, and we set "treat" to 1 for this group, while "treat" is set to 0 for the nontreatment group. Tis is followed by the time variable, which is a dummy variable that looks at whether the city is in the year of policy implementation and beyond. If the city has an FTZ in the period under consideration, then the FTZ is assigned a value of 1 for the year in which it was established and for the years that follow; for example, if the Shanghai FTZ was established in 2013, then the Time variable should be assigned a value of 1 for the year in which it was established and 0 for the year before 2013.
Measuring the year in which a city was subject to policy implementation and subsequent years requires the use of the interaction between time variables and group variables Treat × Time, which is also the core explanatory variable of the article.

Matching Variables.
Based on propensity score matching, the study applies the asymptotic double diference method to reduce the initial diferences between the treatment group and the control group, and selects factors that may afect the establishment of pilot free trade zones as much as possible. Te establishment of free trade zones is mainly infuenced by national policies, geographical location, labor Whether it satisfies the common support hypothesis and the balance hypothesis test The matching method was selected as the experimental group matching the The propensity score was calculated using the Logist function    [25], this study selects seven variables: employment level, regional economic development, service industry development level, government size, fnancial development level, real estate investment, and degree of opening to the outside world. Among them, the employment level is represented by the number of employed people, the regional economic development level is measured by the gross regional product, the service industry development level is measured by the proportion of the tertiary industry to the regional GDP, the fnancial development level is measured by the balance of deposits of fnancial institutions, and the degree of opening to the outside world is measured by the ratio of total import and export trade to regional GDP in each city.

Control Variables.
Tis article considers indicators that afect human capital level as control variables. Human capital is formed through human capital investment, with investment sources mainly from government and individual inputs, as well as externalities of human capital. Tis article selects these control variables for the following reasons: Education expenditure: education is the main way to form human capital, and fscal education expenditure provides fnancial support for educational activities. Hou and Liao [22] used the VECM model to verify that government education investment is positively correlated with human capital accumulation and that there is a stable equilibrium relationship between fscal education investment and human capital. Te short-term impact of education fscal investment on human capital levels is less than the long-term impact. Development level of fnancial institutions: fnance is also an important source of education investment, representing the market force, providing important funding sources for various market entities to invest in education, and allowing various entities to rationally fnance to meet their own education investment needs. Yang et al. [26] verify that fnancial development can promote human capital accumulation from the perspective of modern human capital investment theory and based on the dual role of "government-market" in China.
Degree of openness to the outside world: Wen and Dai [11] verify that the impact mechanism of trade openness on human capital level mainly comes from four aspects: wage price mechanism, technological progress mechanism, capital investment mechanism, and credit constraint mechanism. Te greater the degree of openness to the outside world, the more conducive it is to attracting talent. Actual average employee salary: income level determines the resource capacity available to society and families for education investment and is an important determinant of human capital level. Real estate investment amount: Li [27] has proposed that the high-speed growth of real estate investment in the labor cost-efectiveness mechanism signifcantly promotes regional labor wages, and the increase in wages will promote the growth of human capital investment.
Level of economic development: in general, the level of economic development is directly proportional to the degree of talent agglomeration. Employment scale: the larger the number of employment positions, the easier it is to attract out-of-town employees, gradually forming a talent agglomeration.
Among them, the employment scale uses the number of employed persons as a proxy variable, and the development level of the service industry uses the proportion of the tertiary industry to the regional GDP as its proxy variable. For the missing data, multiple imputations are used to fll in the missing values, and logarithmic transformation is applied to real estate investment, regional GDP, and employment scale to eliminate price factors. Table 2 is the variable index required in this paper.

Descriptive Analysis.
Te selected indicators were frst analyzed descriptively. According to Table 3, there are 5,502 samples in the control group and 441 samples in the experimental group in cities without free trade zones. Te average level of human capital in the control group is 2.956, while in cities with free trade zones, i.e., the experimental group, the average human capital level is 3.419. Compared to the control group, the experimental group has a signifcantly higher level of human capital. Based on this observation, this paper puts forth the preliminary hypothesis that the policy of setting up a free trade zone can improve the level of human capital, and the next step is to test this hypothesis.

Propensity Score Matching
Results. Te paper draws on the results of propensity score matching from Becker and Ichino [28], Blundell and Costa Dias [29], and Heyman et al. [30], using kernel matching and a year-by-year matching approach. In addition, to reduce the interference of subjective selection factors on the evaluation of the FTZ policy, this paper refers to Zhang and Yu's [2] research on the factors afecting the pilot FTZs and selects real estate investment, level of service industry, regional economic development level, government size, level of fnancial development, basic medical facilities, and employment level as matching variables. Taking 2013 as an example, the matching method uses kernel matching and combines observable matching variables to frst calculate the predicted probability values of the FTZ policy for each city using a logistic model and then fnds a unique control city without an FTZ for each city with an FTZ policy (the treatment group). After matching, samples that were not successfully matched were removed, and fnally, the corresponding control cities were found for the treatment group cities in 2013. Using the same method, for cities with FTZs in other years, the corresponding control cities were found for the treatment group cities in the respective years. Subsequently, the balance test graph based on propensity score matching also shows that, as can be observed from Figure 2, before PSM matching, the standardized deviations Te proportion of import and export trade volume to gross regional product (GRP)

Control variables
Government spending on education Logarithmic form of education expenditure Level of fnancial development Logarithmic form of deposits balance of fnancial institutions Te degree of openness to the outside world Te proportion of import and export trade volume to gross regional product (GRP) Average salary of employees Logarithmic form of average employee salary Real estate investment Logarithmic form of real estate investment amount Gross regional product (GRP) Logarithmic form of gross regional product (GRP) Employment scale Logarithmic form of the number of employed population Level of service sector development Te proportion of the tertiary industry in gross regional product (GRP)

Parallel Trend Test.
Te overall steps of the parallel trend test are as follows: For the treatment group, according to the time when the free trade zone was established in each city, the treatment variable for policy efectiveness, preintervention, and postintervention is generated. Taking the Shanghai free trade zone as an example, the year of its establishment was 2013, and the treatment variable for policy efectiveness (current � 0) is set as 1. Te variables for preintervention and postintervention are named pre_1, pre_2, post_1, and post_2, respectively, according to the principle mentioned above. Based on the policy intervention time of the treatment group, the core explanatory variables selected are preintervention variables for 10 periods, postintervention variables for 5 periods, and the year of the policy intervention. In addition, to avoid the issue of multicollinearity, the variable for preintervention period 1 is not considered, and the city and time efects are fxed when conducting the parallel trend test. Figure 3 shows the parallel trend test diagram. It shows that "current � 0" represents the starting point of the policy intervention, which is on the left side of the zero axis. Te coefcients for all variables are signifcantly diferent from zero, indicating that there is no signifcant trend diference between the treatment and control groups before the establishment of the free trade zone. On the right side of the zero axis, the coefcients for the fourth and ffth years after the policy intervention are signifcantly diferent from zero. Terefore, the article concludes that the parallel trend hypothesis is satisfed and that the efect of the free trade zone policy on the level of human capital has a signifcant positive impact, which can be further verifed using the diference-indiferences method.

Average Impact Efects.
Te following analysis will use the asymptotic double diference fxed efect model to verify the efects of various pilot policies in each free trade zone, using a multiperiod DID method due to the diferent times of policy implementation. In the asymptotic double diference, the logarithmic form of the number of high school students is used as a proxy variable for human capital, which is the dependent variable in the article. Te interaction term between the treatment group and policy efectiveness is used as the core explanatory variable. Factors that afect the level of human capital are used as control variables. Two experiments are conducted, one with the inclusion of control variables and one without. Te results are shown in Table 5.  Te asterisks in Table 5 represent the signifcance level, and the values in parentheses represent the t-values of the variables. Table 5 shows the results of the baseline regression model. In the frst type of model in the table, which is the asymptotic double diference without the inclusion of control variables, the interaction term between the treatment group and the policy efectiveness is signifcantly diferent from zero at the 5% signifcance level, and its coefcient is positive. In the second type of model, which is the asymptotic double diference with the inclusion of control variables, the interaction term between the treatment group and the policy efectiveness is signifcantly diferent from zero at the 10% signifcance level, and its sign is also positive. Although the coefcient of the interaction term in the model with control variables is smaller than that in the model without control variables, the results suggest that regardless of the inclusion of control variables, there is a clear causal relationship between the establishment of free trade zones and regional human capital levels. In general, free trade zone policies have a signifcant positive efect on regional human capital levels.
In addition, the regression results in Table 5 indicate that the control variables have diferent efects on regional human capital. First, the coefcient of education expenditure is signifcantly positive at the 1% level, indicating that a higher proportion of education expenditure is favorable for improving the level of regional human capital. Human capital is formed through investment in human capital, which is mainly measured by cultural and educational expenditures. Education expenditure is an important way of investing in human capital, and an increase in education expenditure means an increase in investment in human capital, which inevitably leads to an increase in the level of human capital (Lv and Liu [9]). Te coefcient of the proportion of the tertiary industry is signifcantly positive, indicating that the rapid development of the service industry promotes the improvement of regional human capital. Te coefcient of population density is signifcantly positive at the 1% level, indicating that the growth of the population in a certain area will also bring about an increase in the level of human capital, known as the demographic dividend. Second, trade openness can promote the shift  Discrete Dynamics in Nature and Society of production and exports to technology-intensive products, thus increasing the proportion of skilled labor in the total population (Flug and Galor [31]). If the coefcient of trade openness is negative at the 1% level while controlling for the impact of urban employment size, the impact of trade openness on human capital is negative, indicating that trade openness has not brought about an increase in the level of human capital in various cities. When controlling for urban employment size, the coefcients of trade openness are negative, while the coefcient of employment size is signifcantly positive at the 1% level. Te coefcient of the average salary level of employees is signifcantly negative at the 1% level, indicating that the increase in average employee wages has not led to an increase in the level of human capital. Tis may be because the average employee wage cannot truly refect income inequality, which seriously hinders residents with lower initial wealth levels from investing in human capital (Chao and Shen [32]). Moreover, given the availability of data, the employee average wage indicator in the article only measures the overall level of a region, and there may be a phenomenon of excessive income inequality among employees within the region.
On the other hand, the establishment of the free trade zone will inevitably lead to an increase in the degree of trade openness, and the increase in income from exports will raise the opportunity cost of school education, leading to more young people dropping out of school (Atkin [33]).

Dynamic Analysis.
Based on the results of the above tests, it can be seen that the pilot FTZ policy have a positive contribution to the level of human capital. Te following test examines the dynamic policy efects of the pilot free trade zone to analyze how the establishment of the free trade zone policy has evolved to afect the level of human capital. Te article draws on the methodology of Dai and Cao [34] (2015) and sets up to denote the efect of policy implementation in the frst year of the point in time; post_1 denotes the efect of policy implementation in the second year of the pilot and so on, with the article examining the efect of policy implementation pilots up to the ffth year. Table 6 presents the regression results, and the article uses the model without the inclusion of control variables as a control for both models controlling for time and individual fxed efects. As can be seen in Table 6, in the model without the inclusion of control variables, the coefcients are not signifcant from the frst to the third year of the FTZ policy pilot, but in the fourth year of the FTZ policy implementation, the coefcient is positive at the 5% level of signifcance, and in the ffth year of the FTZ policy pilot, its coefcient is positive at the 1% level of signifcance, indicating that the FTZ policy pilot can Tis indicates that the pilot FTZ policy can improve the level of regional human capital, and the efect becomes more obvious over time, indicating that the impact of the pilot FTZ policy on the level of regional human capital has a lagging efect. However, after adding the control variables, the results in Table 6 show that the coefcient of the interaction term Time it × Treat it is not signifcant within 1-4 years after the establishment of the pilot FTZ policy but fve years after the establishment of the pilot FTZ policy, the efect of the pilot FTZ policy on the regional human capital level shows a signifcant boosting efect, i.e., the boosting efect of the pilot FTZ on the human capital level has a signifcant lag. Terefore, as the pilot FTZ is set up as a slice of the city, it is not able to quickly boost the human capital of the city through the radiation efect in a short period. After the inclusion of the control variables, the efects of the pilot policy are largely the same, both showing a signifcant boost to human capital levels in the fourth and ffth years of the pilot free trade zone policy, but this boost has a signifcant lag. Te reason for this is that the FTZ has a relatively small jurisdiction compared to a city, transport facilities are not perfect, and the import and export demand of the FTZ is not stable, the radiation efect of the FTZ is not obvious in the early stages of establishment, and the measures to introduce talents are relatively imperfect, so the ability to attract talents is not very strong.

Stochastic Pseudo-Processing Variables and Stochastic
Pseudopolicy Dummy Variables. To verify the robustness of the PSM-DID results, the method of placebo tests proposed by Huang et al. [35] was used in this study. Specifcally, the policy implementation time points of the free trade zone experiment studied in the article were in 2013, 2015, and 2017, and the policy timing was not a traditional "one-sizefts-all" approach. Terefore, the article frst randomly generated pseudotreatment groups, then generated pseudopolicy dummy variables, and fnally stored the estimated coefcients, standard errors, and P values from 500 regressions. Since the data used in the article is panel data, only one period of data is retained at frst. In this case, we selected the data from the year 2000. Ten, 21 samples are randomly drawn from it (there were 21 treatment groups in the article). Te city names of the selected samples are retained, and then one-to-many matching is performed based on the city names of the data after propensity score matching. Te matched samples are the pseudotreatment group" samples in the article, and the unmatched ones are the control group samples. In the real experiment, there were a total of 283 cities in the control and treatment groups, with 21 cities in the treatment group and 262 cities in the control group. Te sample data after propensity score matching is then randomly shufed using random numbers in Stata software, and a pseudopolicy dummy variable sample is generated by randomly selecting one year from the 283 prefecture-level cities. Te pseudotreatment group and pseudopolicy dummy variable samples are then merged, and the control variables are selected for regression. Te coefcient matrix, standard error matrix, and P value matrix are set up, and the results of 500 regressions are assigned to the corresponding positions in the matrices. Te P values are then calculated and assigned to the matrix, and the results are presented in the form of images. Figure 4 displays the distribution of the estimated coefcients and corresponding P values for the 500 placebo virtual policy variables. Te horizontal dashed line indicates the signifcance level of 0.1. As can be seen from the fgure, most of the estimated coefcients are centered around zero, and the majority of the P values are greater than 0.1 (i.e., not signifcant at the 10% level), indicating that the estimated results are unlikely to be due to chance and that the positive efect of the free trade zone on human capital is not driven by other policies or random factors. Te 500 regression coefcients at diferent (policy implementation) times are also plotted as a kernel density graph. According to Table 5, the coefcient of the core explanatory variable is 0.035, with a standard deviation of 1.69, indicating that the increase in human capital is due to the establishment of the free trade zone pilot.

When the Replacement Treatment Group Policy
Occurred. To eliminate the interference of other factors on human capital, this article changes the time of establishment of the FTZ policy. For the case of multiple time points in the article, it is assumed that the actual time for all prefecture-level cities to establish the FTZ is advanced by 3 periods, 4 periods, and 5 periods, and the establishment time of the FTZ pilot zone is uniformly set as 2010. Individual fxed efects and time fxed efects are simultaneously controlled to examine whether the coefcients of the interaction term between the treatment group and the policy time variable are signifcant in diferent situations. As shown in Table 7, regardless of whether the FTZ policy is uniformly established in 2010 or advanced by 3 periods, 4 periods, or 5 periods, the P values of the interaction term between the treatment group and the policy variable are all greater than 0.05, indicating that the null hypothesis of the interaction term coefcient being 0 cannot be rejected. Tis indicates that the results of this study are robust and that the establishment of the FTZ pilot zone has had an impact on human capital levels.

Changing the Matching Method.
In addition, to ensure the reliability of the results, the paper used a diferent matching method. First, the nearest neighbor one-to-ten matching method was used. Table 8 shows the results of using seven matching variables, including employment size, regional GDP, government fnance scale, service industry development level, real estate investment, fnancial institution development level, and degree of openness to foreign trade, to fnd a new control group for analysis while controlling for individual fxed efects and time fxed efects. Te coefcients of the core explanatory variables and control variables in the model are positive and signifcant at the 5% level, indicating that the experiment with the changed propensity score matching method (nearest neighbor oneto-ten matching) still shows that the establishment of free trade zones can promote the improvement of regional human capital levels. Te paper also validated the original data using the nearest neighbor caliper matching method, and the coefcient of the interaction term between the treatment group and the policy timing is positive at the 10% level of signifcance, indicating that the conclusion drawn using the radius matching method is still valid. Furthermore, the paper validated the original data using the radius matching method, and the coefcient of the interaction term between the treatment group and the policy timing is signifcant at the 1% level, indicating that the conclusion drawn using the radius matching method is still valid.

Replacement of Explanatory Variables.
Robustness tests can also be conducted by replacing the dependent variable to test the sensitivity of the results. Following the method of Liang and Ji [36], some studies have used the number of regular high school teachers as a measure of human capital. Terefore, in the robustness test, the dependent variable will be replaced with the number of regular high school teachers, and it will be processed logarithmically. Te same kernel matching method will be used, and the matching variables of employment scale, regional GDP, government fscal scale, service industry development level,  Discrete Dynamics in Nature and Society real estate investment, fnancial institution development level, and degree of openness to foreign trade will be used as control variables. Individual fxed efects and time fxed efects will also be controlled, and the same control variables will be used to perform a diference-in-diferences analysis.
In addition, a model without control variables will be included for comparison. If the results are consistent with the previous fndings, it indicates that the results are robust, i.e., the pilot policy of the free trade experimental zone helps to promote the improvement of human capital. Table 9 presents the results of the diference-indiferences regression analysis using logarithmically transformed regular high school teacher numbers as the dependent variable after PSM matching. From the table, we can clearly conclude the following: When no control variables are added, the coefcient of the interaction term between the treatment group and policy time is 0.061 and is signifcantly positive at the 1% level, indicating that the pilot policy of the free trade experimental zone has a positive efect on the improvement of regional human capital. In the model with control variables, the coefcient of the interaction term between the treatment group and policy time decreases slightly to 0.036, but its value remains positive at the 5% signifcance level. Overall, the regression results using the alternative dependent variable are consistent with the previous fndings, indicating that the results are robust.

Heterogeneity Analysis.
In this study, geographical division rules were used to classify regions into the eastern, central, and western regions, and further PSM-DID tests were conducted for each region to perform heterogeneity analysis. Table 10 shows that in the western region, the policy efect of the establishment of free trade zones on human capital level is the strongest, followed by the central region, and the eastern region has the weakest efect. Te reason for this may be that the eastern region is generally more developed, and its natural resources, economic development, and talent introduction policies have already played a signifcant role in attracting talent. On the other hand, the western and central regions are relatively less developed, and the establishment of free trade zones promoting free trade openness could have signifcant effects to these regions.

Conclusions and Recommendations
When analyzing the impact of a free trade zone on human capital levels, descriptive analysis and PSM-DID models were frst used to study the human capital levels of cities that have established free trade zones. Te results of the descriptive statistics showed that the average human capital level in the control group sample was 2.956, while the average human capital level in the treatment group sample was 3.419, indicating that the human capital level in the treatment group was signifcantly higher than that in the control group. Next, seven matching variables, including employment scale, fnancial institution development level, regional GDP, real estate investment, government size, service industry development level, and degree of openness to the outside world, were used for propensity score matching, and the results showed that the balance test was met. Parallel trend tests were then conducted using graphical methods and regression methods to eliminate interference from other factors on human capital levels. Te results showed that the coefcients of "current � 0," which represents the starting point of policy establishment on the zero coordinate axis, were not signifcantly diferent from 0, indicating that there was no signifcant diference in the trend of change between the experimental and control groups before policy implementation.
Te average policy efect was then estimated, and the results showed that regardless of whether the control variables were added, the core explanatory variable coefcient was signifcantly positive at the 10% level, indicating that there was a signifcant positive promoting efect of free trade zone policy on regional human capital levels overall. Te dynamic efect analysis demonstrated that the free trade zone had a signifcant lag efect on human capital levels. In addition, the impact of free trade zone policies on human capital levels varies in diferent regions, and there are different policy efects. Among them, the promotion efect of free trade zones on human capital levels is most signifcant in Western regions, followed by central regions, and fnally in eastern regions. Terefore, in the future development path of free trade zones, it is necessary to continuously improve regional infrastructure, promote mutual exchange between regions, and place emphasis on education to increase personal and national investment in education. Secondly, innovative enterprises should be continuously introduced to increase employment demand, and relevant policies should be used to attract talent. Finally, development plans should be formulated based on the geographical location and endowments of each region. Te policy efects of free trade zones in economically developed regions are not very signifcant compared to other regions. Tis may be due to the fact that these regions have inherent advantages that are attractive enough to attract talent, resulting in saturation. In the future, free trade zones in economically developed regions will need to explore other systems to better drive surrounding areas.
Te construction of free trade zones needs to continuously improve the institutional mechanisms and attempt to integrate with other systems into a management approach, form a service model, and create an external image. Tey should focus on establishing a sound joint conference system at a high level and frequency, building a scientifc and effcient coordination mechanism and a unifed planning and development mechanism, and piloting in some areas within the zone to delegate fscal and property rights to the zone for management, clarifying the relationship between public service and market supervision responsibilities, and streamlining management, investment, and distribution mechanisms. Te successful experience of pilot projects should be replicated and promoted within the zone while also being tailored to local conditions and complemented by improved management mechanisms to stimulate reform and innovation.
Innovation in free trade zone regional alliance management mechanisms should focus on encouraging free trade zones to strengthen cooperation based on economic regions, establishing a long-term assessment and evaluation mechanism for free trade zones and an important landmark achievement award system, implementing a diferentiated industrial development strategy, establishing a comprehensive human resource training and exchange mechanism, actively participating in cadres' two-way job rotations with free trade zones in more open areas to promote alignment of ideology and concepts with those in more open areas, establishing a training base for serving enterprises, allowing cadres to follow and learn from the staf of enterprises, deepening their understanding of enterprise needs as ordinary employees, and establishing a personnel exchange mechanism with city-level units to truly achieve the attraction, retention, and efective utilization of innovative talents. In addition, in order to better develop each city, the cities where the free trade zones are located should increase their measures for attracting talents, expand the radiation efect of the free trade zones, promote the human capital levels of surrounding areas, improve infrastructure, and strengthen communication and cooperation between regions.

Data Availability
Te data are available from the China Urban Yearbook, the China Regional Economic Yearbook, and provincial and municipal statistical yearbooks.

Conflicts of Interest
Te authors declare that they have no conficts of interest.