Effects of the Talent War on Urban Innovation in China: A Difference-in-Differences Analysis

Abstract

Talent is an important strategic resource for regional economic development. Based on the background of “the talent war” that has broken out between various cities in recent years, this study empirically verified the influence of the talent policy on urban innovation in 277 prefecture-level cities in China from 2010 to 2019 using the multi-period difference-in-differences model. The results indicated that “the talent war” caused by the talent policy has positively influenced urban innovation, causing, for instance, a dramatic increase in the number of patents for inventions. Among the subsidy methods of “the talent war” policy, the employment and entrepreneurship subsidy had the greatest incentive effect on urban innovation, followed by the talent housing subsidy. Moreover, the “the talent war” policy exerted a positive impact on urban innovation by improving the innovation willingness of cities and the level of talent gathering. At present, “the talent war” cities have, to a certain extent, restrained the improvement of urban innovation in neighboring cities because of the siphon effect, resulting in the division of the regional labor market. A heterogeneity analysis showed that “the talent war” has significantly promoted substantive innovation and the development of coastal cities with a better business environment and a higher degree of intellectual property protection.

1. Introduction

In 2021, the Chinese Talent Work Conference stated that “The foundation of innovation lies in talents, and it is necessary to stimulate the innovation vitality of all kinds of talents” [1]. China’s economic and social development has entered a new era of innovation-driven high-quality development, which requires strong support of all kinds of high-quality laborers [2]. There is no universally uniform definition in academia for the definition of “ talents “. The general view is that talents are those individuals who have certain knowledge or skills, can match social needs for creative labor activities, and have a higher competence and quality labor force in terms of human resources. Under the background of implementing an innovation-driven development strategy and strengthening the country, talents have become the driving force of efficient innovation development for the whole country, as well as its regions and cities. They are also a key factor for transforming and upgrading the country’s economic structure and enhancing the functioning of cities [3,4]. The inflow and agglomeration of a highly skilled workforce usually promotes regional economic growth, bringing higher productivity and externalities [5]. Human capital is embodied in the workforce, including the health status, knowledge skills, and technology level of the workforce [6]. Talent can be understood as a highly skilled human capital [7]. Policymakers generally believe that the agglomeration of human capital plays a role in economic growth through innovation [8,9]. In order to maximize their own benefits, talents tend to “vote with their feet” when choosing cities for employment [10]. Therefore, cities with more employment opportunities, institutional guarantees, and supporting services are more attractive and take their place in “the talent war” more easily.

The talent introduction policy is a means for governments to financially subsidize talents directly for innovative or scientific research activities, which are complemented by salary. In addition to direct monetary subsidies, different degrees of convenience and support are provided, including household registration, housing, children’s education, etc. The policy of talent introduction has certain standards for its subsidized subjects that are usually set as limitations in areas such as academic background, technical level, and scientific research achievements. In the talent introduction policy, unlike traditional government innovation subsidies, the scope of subsidy objects is wider and not only oriented to domestic talents but also eligible subsidized foreign innovative talents. In order to attract top talents, most provincial governments in China, even prefecture-level cities and county-level governments, have introduced various forms of talent recruitment policies, such as preferential policies and increased subsidies, from which “the talent war” developed. “The talent war” began in 2017 and reached its climax in 2018, addressing the urgent need to develop the service economy, innovation economy, and consumption economy in the process of urban economic transformation and upgrading [11]. Cities that have implemented “the talent war” policy include second- and third-tier cities, such as Wuhan, Chengdu, and Xi ‘an, as well as first-tier cities, such as Beijing and Shanghai. Talents have become the primary resource to improve urban innovation ability and competitiveness between cities [12]. Under stiff competition, some local governments have launched more talent attraction initiatives, such as lowering the threshold for settlement, relaxing academic requirements, and providing subsidies for house rental and house purchases. Nevertheless, there is a large gap between the innovation ability of different cities across China, and the phenomenon of uneven economic development still exists.

“The talent war” has important strategic significance for urban development, and the issues worthy of concern are as follows: Firstly, have “the talent war” policies introduced by cities successfully attracted talents and promoted urban innovation? Secondly, which type of talent subsidies meant to attract talents have best improved the level of urban innovation? Thirdly, what makes these policies so efficient? Finally, considering that China is entering a stage of high-quality development, and most cities have successively implemented policies to compete for talents, what impact do these policies have on neighboring cities that are still not involved in “the talent war”? The discussion of the above issues will have significance for the formulation of urban talent policies in the future.

At present, there is little direct research on how the talent policy influences urban innovation. Therefore, the marginal contribution of this paper lies in the following points: Firstly, focusing on the characteristics of prefecture-level cities, this paper empirically analyzes how urban innovation has been impacted by the talent policy. The accuracy of this study’s conclusions is based on empirically obtained original data. Secondly, compared to the existing literature, this paper pays more attention to the different types of talent subsidies and compares their effects on stimulating urban innovation. Thirdly, this paper further analyzes the influence of “the talent war” cities on their neighboring cities. The conclusions of this study may provide some theoretical basis and practical guidance for cities to further implement “the talent war” policy.

2. Policy Background and Theoretical Analysis

2.1. Policy Background

Talent resources have long been highly valued in China. As early as 2000, the Central Economic Work Conference proposed the formulation of a talent strategy. The formation of a “talent-oriented” scientific concept was followed by the development of a strategy to strengthen the country using talents. As a result, governments at all levels paid more attention to talent introduction and carried out a series of talent-related work [13]. In 2014, the reform of the household registration system relaxed the conditions for settlement, which later provided opportunities for the cities to launch “the talent war” policy. In 2017, the 19th National Congress of the Communist Party of China (CPC) once again pointed out that national development and rejuvenation depended on talents. Under the background of the state’s policies and cities’ demands, local governments aimed at “grabbing talents” more directly and fiercely. Nearly 70% of talent policies for cities were released in 2017 and 2018. By 2021, the central and local governments at all levels had issued more than 15,000 policies related to talents, as shown in

Table 1. Release of China’s talent policy.

Local Regulations			Central Regulations
Level of Effectiveness	Issued Quantity	Proportion	Level of Effectiveness	Issued Quantity	Proportion
Local laws and regulations	107	0.77%	Administrative laws and regulations	8	0.52%
Local government regulations	95	0.68%	Judicial interpretation	9	0.59%
Local normative documents	3806	27.30%	Departmental regulations	1295	84.70%
Local judicial documents	2	0.01%	Inner-party laws and regulations	68	4.45%
Local operating documents	9172	65.79%	Group regulation	58	3.79%
Approval of administrative license	273	1.96%	Industry regulation	90	5.89%
Total amount	13,942	100.00%	Total amount	1529	100.00%

Source: The data of local regulations and central regulations were compiled by the author according to Peking University Law Database.

. To ensure effective talent construction, the central government promulgated more than 1500 laws and regulations, including administrative regulations, judicial interpretations, departmental rules, etc. Under the guidance of the central government’s policies, local governments successively issued corresponding talent policies. Formally, the talent policies formulated by the central government were mainly guiding documents and general outlines, while the corresponding implementation plans, talent policies, and operating rules were issued by local governments.

2.2. Literature Review and Theoretical Analysis

2.2.1. Talent Policy and Innovation

Since Schumpeter published the Theory of Economic Development in 1912 and established the theory of “innovation”, scholars have systematically studied the factors influencing innovation and have suggested that talents are the most important factor to promote innovation. With the development of the economy, the competition for talents among regions and even countries is becoming more and more fierce. To a certain extent, talent introduction policy has broken the institutional barriers of talent migration in China and changed its direction and speed, while at the same time making the allocation of human capital among regions more efficient, which has a positive impact on urban innovation. There are not many studies in the existing literature that directly investigate the influence of talent policy on urban innovation, and whether the talent introduction policy can really promote urban innovation is still a controversial point [14]. Some scholars have pointed out that the current talent introduction policy does not pay enough attention to existing talents and focuses on the introduction of new talents instead, while neglecting follow-up talent training [15]. There are also some regions that distort talent introduction policies in order to maintain the urban demographic dividend and achieve performance goals [16]. Zhong and Luo [17] argued that the influence of the talent introduction policy on innovation efficiency was not significant. In contrast, some scholars have concluded that the talent policy has a positive effect on innovation. By comparing talent introduction policies across China, Chen and Huang [18] found that widespread talent introduction policies raised the mobility of human capital and had a positive impact on innovation efficiency. Taking 270 prefecture-level cities in China as research objects, Lu and Zhang [19] showed that the talent introduction policy was highly beneficial for urban innovation. Chen and Xiao [20] used the event analysis model to demonstrate that local talent policies dramatically improve innovation diffusion. Therefore, based on the prior research and the background of the talent policy, this paper proposes the following hypothesis:

Hypothesis 1.

“The talent war” policy can effectively promote urban innovation.

2.2.2. Talent Agglomeration and Innovation

In “Principles of Economics”, Marshall pointed out that production factor convergence in cities helps to reduce costs, including the costs of transporting goods, manpower, and ideas. Innovation factor agglomeration, which refers to the gathering of human, financial, and material resources in a region through innovative activities, can promote the optimization and improvement of its industrial structure. The agglomeration of high-quality talents is one of the main features in the agglomeration of innovation factors and is also the main target of talent policies in cities. The talent policy reduces the cost of human capital flow, attracts talents to quickly gather within the domain, and causes agglomeration and innovation effects [21]. The influence of the talent agglomeration effect on regional innovation mainly includes a knowledge spillover effect and the development of the innovation environment. The generation of new knowledge is a process of long-term accumulation and qualitative change [22], and the generated knowledge can be learned, imitated, and surpassed by others, resulting in a knowledge spillover effect [23,24]. Accumulating many heterogeneous talents with different knowledge reserves in a gathering place improves the opportunities for communication, forms an innovative learning atmosphere, and inspires innovative consciousness. Kurt Lewin, an American psychologist, pointed out in his field dynamic theory that the innovative performance of talents is largely influenced by their environment [25]. In a favorable innovation environment, talents are more willing to actively explore external knowledge.

Most scholars believe that talent agglomeration has a positive impact on regional innovation. For example, Chen and Liu [26] argued that in a short time, the talent policy has changed the urban talent cultivation system and promoted the growth and spatial agglomeration of high-quality human capital, thus stimulating the development of regional innovation. By exploring how talent agglomeration affects regional innovation, Sun and Zhang [27] found that talent agglomeration had a significant spillover effect on regional innovation. Based on this, we put forward hypothesis 2 as follows:

Hypothesis 2.

The talent policy improves urban innovation through talent agglomeration.

2.2.3. Strength of Talent Policy Subsidy and Innovation

The purpose of “the talent war” introduced across the country is to localize the introduced talents [28] and provide them with overall assistance through various talent policies, such as settlement support, living allowances, house purchase allowances, and supporting guarantees [29,30,31]. Based on the method of Wu and Li [32], we classified the preferential policies and subsidies for talents. According to the specific provisions of the talent policies issued by cities, talent policy subsidies are divided into four groups: settlement support, employment and entrepreneurship subsidies, talent housing subsidies, and living service subsidies.

In terms of settlement support, due to the steady implementation of the household registration system reform, all cities that implemented “the talent war” policy during the research period relaxed restrictions on settlement for talents. However, the threshold for settling was different between cities. For example, the minimum education requirements varied across the cities. The education threshold of Xi’an and Zhengzhou was a college degree, while the minimum requirement of Tianjin and Chengdu was an undergraduate degree. In addition, some cities provided the introduced talents with guarantees, such as children’s admission to schools nearby and help for their spouses to settle down. In terms of employment and entrepreneurship subsidies, different amounts of financial subsidies, operating site rental subsidies, and social insurance, as well as loan offers, investments, and other subsidies, were provided for employment and entrepreneurship projects. Nevertheless, 16 of “the talent war” cities did not offer subsidies for employment and entrepreneurship in 2017–2018. In terms of talent housing policies, some cities also provided subsidies, mainly for renting houses, buying houses, allocating rent, etc. For example, in Nanjing, Changsha, and other cities, the financial support and rental subsidies were proportional to the academic qualifications of introduced talents. During the period covered by this study, only 10 cities did not have a clear description of housing subsidies. In terms of living service subsidies, 15 cities provided living subsidies for qualified introduced talents for a certain number of years. Identifying the attractiveness of different subsidy types would help to maximize the effect of the talent policy in cities, which aims to attract more talents and promote urban innovation.

Based on the two-stage model of rural–urban labor migration in underdeveloped countries proposed by Todaro [33,34], we suggest the following model for the theoretical analysis of the strength of the talent policy and talent agglomeration in cities. Theoretically, a city introduces relevant policies to encourage talents to “vote with their feet” and move there in order to gradually become a talent agglomeration region. The city where talents move to is represented by a, and the city that talents move from is indicated by b. The talents transferred due to the talent introduction policy are generally professional talents required for the development of city a. It is highly probable that they will get jobs in a short period of time when they move to the new city. The suggested model regards the opportunity cost caused by frictional unemployment as a part of the one-time migration cost, focusing on the benefits caused by the migration of talents from city b to city a. The consumption expenditure of talents in cities is represented by c, the salary income is $w$, the time before the implementation of the policy is 0, the time after the implementation of the policy is 1, and the number of migrating talents is $N$. Then, the number of talents in city a before the implementation of the policy is $N_{a0}$, the number of talents in city b before the implementation of the policy is $N_{b0}$, and the migration of talents from city b to city a is $\Delta N$. Thus, the following Equation is formed:

$$N_{a1}=N_{a0}+\Delta N,N_{b1}=N_{b0}-\Delta N$$

(1)

The consumption expenditure and the salary income of talents in city a and city b at time t are as follows:

$$c_{at}=c\left(N_{at}\right),c_{bt}=c\left(N_{bt}\right),w_{at}=w\left(N_{at}\right),w_{bt}=w\left(N_{bt}\right)$$

(2)

The real estate industry is a pillar of China’s national economy, and the housing consumption expenditure of urban residents accounts for a relatively high proportion of the consumption expenditure of all Chinese residents. After the introduction of talents, there will be an increased demand for housing, which will drive up housing prices [35]. The rise in housing prices will also affect the cost of living to a certain extent. It is generally believed that this impact is positive, and a city with a larger pool of talents tends to have higher consumer spending. Therefore, $c\prime \left(N\right)>0$; that is, cities with a larger talent scale have a higher consumption expenditure. According to Lu [36], the inflow of talents has little impact on the local wage rate. Thus, we obtain:

$$c\prime \left(N\right)-w\prime \left(N\right)>0$$

(3)

Talents’ “voting with feet” is defined by the wage rate of the target city. Therefore, it is assumed that the percentage change in the number of talents moving into a city is determined by the proportional difference $\alpha$ between the salary income and consumption expenditure of talents before the implementation of the policy; the corresponding relationship is represented by function $F$, which can be written as follows:

$${\alpha }_0=\frac{\left(w_{a0}-c_{a0}\right)-\left(w_{b0}-c_{b0}\right)}{w_{b0}-c_{b0}},\frac{\Delta N}{N_{a0}}=F\left({\alpha }_0\right)$$

(4)

Because the talent stock of city a and city b before and after the implementation of the policy is in a steady state, $F\left({\alpha }_0\right)=F\left({\alpha }_1\right)$, and because the marginal tendency of talent inflow is positive, $F\prime >0$, ${\alpha }_0={\alpha }_1$. Thus, we obtain:

$$(N_{a1}-N_{a0})=-\left(N_{b1}-N_{b0}\right)=\Delta N$$

(5)

There are four types of subsidies associated with the talent introduction policy, the quantifiable amount of which is $m$. Thus, we obtain:

$$\frac{\left(w_{a0}-c_{a0}\right)}{w_{b0}-c_{b0}}=\frac{\left(w_{a1}-c_{a1}\right)+m}{w_{b1}-c_{b1}}$$

(6)

By performing the Taylor expansion of $N_{a0}$ and $N_{b0}$ at $w_{a1}$ and $w_{b1}$, respectively, and ignoring the quadratic term, we obtain:

$$w_{a1}\approx w_{a0}+w\prime \left(N_{a0}\right)\Delta N,w_{b1}\approx w_{b0}+w\prime \left(N_{b0}\right)\Delta N$$

(7)

$$c_{a1}\approx c_{a0}+c\prime \left(N_{a0}\right)\Delta N,c_{b1}\approx c_{b0}+c\prime \left(N_{b0}\right)\Delta N$$

(8)

By substituting the above equation to the right of Equation (6),

$$\frac{\left(w_{a0}-c_{a0}\right)}{w_{b0}-c_{b0}}\approx \frac{\left(w_{a0}-c_{a0}\right)-\left[c\prime \left(N_{a0}\right)-w\prime \left(N_{a0}\right)\right]\Delta N+m}{\left(w_{b0}-c_{b0}\right)+\left[c\prime \left(N_{b0}\right)-w\prime \left(N_{b0}\right)\right]\Delta N}$$

(9)

The following equation is then obtained:

$$\Delta N\approx \frac{m}{\left[c\prime \left(N_{a0}\right)-w\prime \left(N_{a0}\right)\right]+\left(1+{\alpha }_0\right)\left[c\prime \left(N_{b0}\right)-w\prime \left(N_{b0}\right)\right]}$$

(10)

Since the denominator in formula (10) is greater than zero, $\frac{\partial \Delta N}{\partial m}>0$, which shows a positive linear relationship between the flow of talent from city b to city a ($\Delta N$) and the subsidy intensity of the talent policy ($m$). That is, the talent policy can attract talents, and the greater the subsidies of the talent policy, the stronger the capacity to attract talents or to promote talent agglomeration.

Housing subsidies are one of the four types of subsidies associated with the talent introduction policy, and most cities under review liberalized settlement by lowering the requirements. Meanwhile, employment and entrepreneurship subsidies place emphasis on projects that help talents find employment and entrepreneurship in new cities. This is also the most direct way to subsidize the employment of talents. Stevenson [37] considered that the essence of entrepreneurship subsidies is to support the development of entrepreneurship, and the government’s assistance in this field can further stimulate talents [38]. Chen and Xiao [39] used the Delphi method to measure the effectiveness and intensity of the urban talent policy and found that the average effectiveness of the financial subsidy policy, including the project subsidy, was the highest, even higher than that of the living support subsidy. Therefore, it can be predicted that the employment and entrepreneurship subsidy will have the greatest effect on attracting talents, and its impact on urban innovation will be the most significant.

As for talent housing subsidies, a study by Liu and Jin [40] found that the newly entered talents had little demand for house purchasing in the short term, and the rental subsidy could quickly relieve the housing pressures on the talents. The rental subsidy provided by some cities, such as Changsha, can basically meet rental needs, while the rental subsidy provided by other cities, such as Nanjing, is hardly enough. Living service subsidies generally involve subsidizing talents with certain amounts of money for a particular period of time. Sun and Luo [41] found that although the proportion of such monetary subsidies is still low, it is rapidly growing as competition for talents becomes fiercer. Increasing talent agglomeration is directly correlated with growth in the amounts of subsidies. Under the condition that hypothesis 2 is established, the following hypotheses can be put forward:

Hypothesis 3.

While there is a positive relationship between all subsidies of the talent policy and urban innovation, the employment and entrepreneurship subsidy has the greatest effect on attracting talents.

3. Materials and Methods

3.1. Research Methods and Model Building

Here, we selected a difference-in-differences model to estimate the causal effect of the talent policy, effectively avoiding missing variables and sample selection errors and reducing endogenous problems [42]. Talents are a high-quality human resource with greater abilities. If a city has a policy of introducing talents, liberating restrictions for those who meet certain criteria at the academic or technical level, and providing corresponding employment entrepreneurship subsidies, house purchase subsidies, and life services subsidies, etc., then the city is considered to participate in “the talent war”. In the sample, the cities that implemented “the talent war” policy were set as the treatment group—that is, $Trea{t}_i=1$—and the cities that did not implement “the talent war” policy were set as the control group—that is, $Trea{t}_i=0$. The value of the year before the implementation of the policy was 0; that is, $Pos{t}_{it}=0$. The value of the year after the implementation of the policy is 1; that is, $Pos{t}_{it}=1$. The net effect of the implementation of the talent policy is expressed by the coefficient of the interaction term $Polic{y}_{it}=Trea{t}_i\times Pos{t}_{it}$ of the dummy variables $Trea{t}_i$ and $Pos{t}_{it}$. As the implementation time of “the talent war” in different cities was not consistent, and the traditional difference-in-differences model requires that the implementation time of policies be at the same node, we adopted the multi-period difference-in-differences model. The regression model is as follows:

$$Innovatio{n}_{it}={\beta }_{1}Polic{y}_{it}+{\beta }_2{X}_{it}+{\gamma }_t+{\mu }_i+{\epsilon }_{it}$$

(11)

where subscript $i$ represents a city, $t$ represents time, and $Innovatio{n}_{it}$ represents the logarithmic level of urban innovation. If the city $i$ introduced “the talent war” policy in year $t$, the value of $Polic{y}_{it}$ before year $t$ is 0; otherwise, it is 1. The characteristic variables $X_{it}$ control the characteristics of cities and the changes in cities with time. ${\gamma }_t$ and ${\mu }_i$ represent the time (year) fixed effect and individual (city) fixed effect, respectively, which are used to control the impact, which changes with the year, and other unobserved urban features that do not change with time but may affect urban innovation. ${\epsilon }_{it}$ is a random error term and ${\beta }_0$ is a constant term. The coefficient ${\beta }_1$ is used to estimate the impact of “the talent war” policy on urban innovation.

In order to further verify whether the improvement of innovation in each city was brought about by subsidizing a specific policy of “the talent war”, the intersection of the policy subsidies and dummy variables of talent policy was added into the model. The model is constructed as follows:

$$Innovatio{n}_{it}={\alpha }_{1}Polic{y}_{it}\times Detai{l}_j+{\alpha }_{2}Polic{y}_{it}+{\alpha }_{3}Detai{l}_j+{\alpha }_4{X}_{it}+{\gamma }_t+{\mu }_i+{\epsilon }_{it}$$

(12)

where $Detai{l}_j(j=1,2,3,4)$ is a variable that indicates the specific policy type of “the talent war”: $Detai{l}_1$ indicates settlement support, $Detai{l}_2$ represents the employment and entrepreneurship subsidy, $Detai{l}_3$ is the talent living subsidy, and $Detai{l}_4$ designates the living service subsidy. In this equation, if the content of “the talent war” policy covers the jth type of subsidy, then $Detai{l}_j=1$; otherwise, it is 0. Specifically, if settlement support is provided, it will be assigned as 1; otherwise, it will be 0. The intensity of subsidies is described by the value of subsidies. If there are subsidies, the value is 1; otherwise, it is 0. Equation (12) mainly demonstrates whether there is a relationship between implementing different subsidy methods and the strength of urban innovation; the coefficient ${\alpha }_1$ of the interaction terms $Polic{y}_{it}\times Detai{l}_j$ reflects the influence of the different subsidy methods and strengths of “the talent war” policy on urban innovation.

In addition, in order to verify whether the impact of “the talent war” policy on neighboring cities in the treatment group produced the spillover effect or siphon effect, we used dummy variables ($Nea{r}_i$). If a city neighbored a “the talent war” city, then $Nea{r}_i=1$. If a city neither implemented “the talent war” policy itself nor was located around a city that implemented it, then $Nea{r}_i=0$. The following regression model can then be constructed:

$$Innovatio{n}_{it}={\delta }_{1}Nea{r}_i\times Pos{t}_{it}+{\delta }_2{X}_{it}+{\gamma }_t+{\mu }_i+{\epsilon }_{it}$$

(13)

where the treatment group is the neighboring city and the new control group subsumes “the talent war” cities without their neighboring cities. If ${\delta }_1$ is significantly negative, it means that “the talent war” policy has a siphon effect on neighboring cities; if ${\delta }_1$ is significantly positive, it means that “the talent war” policy has a spillover effect on neighboring cities.

Figure 1, created using ArcGIS 10.7, shows the cities that implemented “the talent war” policy in 2017 and 2018 and their neighboring cities. In Figure 1, “reformcity” indicates the city that implemented “the talent war” policy and “nearcity” is the neighboring city of “the talent war” city. It can be seen that the cities that implemented “the talent war” policy and their neighboring cities are mainly located in the eastern and central regions.

3.2. Variable Definition

3.2.1. Explained Variable

The explained variable in this study was urban innovation, and there was no clear standard for the selection of this index. In line with the existing literature [17,43], this study used the number of patents obtained by a city in a particular year as a measure of urban innovation. In this paper, the explained variable was treated by adding 1 to take the logarithm. The total number of patents ($\mathrm{Sumaa}$) obtained could be specifically divided into patents for inventions ($\mathrm{Invaa}$), patents for utility models ($\mathrm{Umaa}$), and patents for appearance design ($\mathrm{Desaa}$). Invention patents are the crystallization of new high-level technological achievements, while utility model patents and design patents have relatively low technical content and are generally based on the original technology. In this paper, in addition to the regression of the number of invention patents, the regression of the number of the other two patent types, as well as of the total number of patents obtained by cities in the year under review, were calculated. This allowed us to evaluate the specific innovation effect of “the talent war”.

3.2.2. Explanatory Variable

The explanatory variable in this study was the dummy variable of “the talent war” policy. If a city implemented the policy during a certain period, the value of $Polic{y}_{it}$ was 1, and it was placed in the treatment group. If a city did not implement the policy within the sample time range, the value was 0 and it was placed in the control group. For cities that did not introduce the policy during 2010–2019, the value was also 0.

3.2.3. Control Variables

In order to improve the accuracy of the model estimation results, we referred to the existing research [44] and selected seven key factors that could affect urban innovation as control variables. The specific control variables are as follows:

• The level of urban industrial structure. The level of industrial structure in a region is directly proportional to its industrial economic benefits. The higher the level of industrial structure, the higher the corresponding labor productivity and economic benefit, which leads to more demand for innovation and thus plays a certain driving role for urban innovation. In this study, the proportion of secondary industry and tertiary industry in the GDP of each prefecture-level city was selected to measure the level of industrial structure in each city.
• The level of urban human capital. Human capital is one of the inputs of innovation activities, and the level of human capital is critical to how innovation activities are conducted. A higher level of human capital represents a good environment for innovation and thus has an important impact on urban innovation. In this study, we selected the number of college students in each city to measure the level of urban human capital.
• The degree of city openness to the outside world. The degree of city openness reflects the liquidity of its production factors. Foreign direct investment brings advanced technology, management, capital, and talent support to cities and is an important driver of the development of urban innovation; thus, the degree of openness to the outside world becomes an important factor affecting urban innovation. In this study, we selected the logarithmic amount of foreign capital used in a given year to measure the openness level of each city.
• Investment in urban education. Education is a direct way to develop talents, and investment in urban education indicates a city’s commitment to developing human capital with a higher education level, which delivers innovative talents directly to the city. The investment in education will have an impact on the innovation potential of the whole city. This study measured the investment in urban education as the proportion of the local budget expenditure that was dedicated annually to urban education.
• The level of science and technology communication. The development of the modern internet has reduced the cost of the communication of information, such that individuals in cities have more convenient and efficient access to scientific and technological information [45]. The increased level of science and technology communication has accelerated the spread of innovative knowledge and knowledge spillover and has become an important factor in promoting urban innovation. The number of mobile phone users represents the speed of information propagation to some extent; thus, in this study, the level of science and technology communication was represented by the number of mobile phone users in each city.
• The level of medicine and healthcare. The level of medicine and healthcare in a city reflects the comprehensive service ability for people’s livelihood and is relevant to the health life of talents. The attractiveness of a city will be affected by the local level of medical resources, and a good level of healthcare provides health guarantees and living environments that allow talents to better settle in the city [46,47]. Cities with a lower medicine and healthcare level are less likely to attract talent settlement, having a negative effect on the generation of innovation. Thus, we chose the medical health level as one of the factors influencing urban innovation. Specifically, we selected the proportion of the number of doctors in the city to the total population at the end of the period under review to represent this variable.
• The level of city environmental construction. Talents’ choice of a city will be affected by the livability level of the urban environment. The more suitable the living environment of a city is, the greater the attraction to talents and the higher the probability that talents will choose to settle there, which in turn will generate a positive effect that drives urban innovation [48]. Therefore, this indicator was chosen as one of the control variables in this study. Greening cover rate can reflect the ecological status and living conditions of a city, as well as the level of urban environment construction; thus, the level of city environmental construction was represented by the green coverage rate of built-up area in this study.

3.3. Data Source and Variable Description

This study used the panel data of China’s prefecture-level cities from 2010 to 2019 to evaluate the effect of “the talent war” policy on urban innovation. We excluded all the samples of cities that underwent an adjustment in administrative areas during the reviewed period, as well as the samples of cities that showed serious data loss. As a result, 277 prefecture-level cities were selected as research objects, where the relevant data represented the statistical standard of the whole city. The data for the explained variables used to measure urban innovation were obtained from the Chinese Research Data Services (CNRDS) database. The data on the specific time when each city introduced “the talent war” policy in 2017–2018 were obtained from the local government websites, while the data for the controlled variables were obtained from the China Urban Statistical Yearbook (2011–2020), the Economy Prediction System (EPS) data platform, and the statistical database of China Economic Net. A few missing data points were filled in using the smoothing index method. The map used in this paper was based on a standard map downloaded from the Standard Map Service Network of the State Bureau of Surveying, Mapping and Geographic Information (approval number: GS(2016)1554). The base map was not modified.

Table 2. Descriptive statistical results of variables.

Variable	Unit	Obs	Mean	S.D.	Min	Max
“The talent war” policy		2770	0.012	0.110	0.000	1.000
$\mathrm{Number} \mathrm{of} \mathrm{inventions} \mathrm{obtained} \left(\mathrm{Invaa}\right)$	piece	2770	10.070	1.859	1.099	14.940
$\mathrm{Number} \mathrm{of} \mathrm{utility} \mathrm{models} \mathrm{acquired} \left(\mathrm{Umaa}\right)$	piece	2770	4.826	1.789	0.000	10.880
$\mathrm{Number} \mathrm{of} \mathrm{designs} \mathrm{obtained} \left(\mathrm{Desaa}\right)$	piece	2770	6.669	1.607	1.792	11.380
$\mathrm{Total} \mathrm{number} \mathrm{of} \mathrm{patents} \mathrm{obtained} \left(\mathrm{Sumaa}\right)$	piece	2770	5.699	1.735	0.000	11.080
Proportion of secondary industry in GDP	%	2770	7.197	1.605	2.565	12.020
Proportion of tertiary industry in GDP	%	2770	3.826	0.253	2.433	4.496
Actual amount of foreign capital used in that year	ten thousand	2770	3.672	0.247	2.282	4.425
Number of college students in higher education	person	2770	5.802	0.767	3.620	8.313
Number of mobile phone users at the end of the year	ten thousand	2770	0.178	0.040	0.018	0.356
Proportion of education expenditure in local fiscal budget expenditure	%	2770	3.032	0.437	1.763	4.588
The proportion of doctors in the total population at the end of the year	%	2770	3.668	0.226	0.020	5.957
Greening coverage rate of built-up area	%	2770	10.540	1.313	5.442	13.900

Source: The data was calculated by the author with STATA.

shows the logarithmic descriptive statistical results for each variable. There were 2770 samples used in this study. As for the total number of patents obtained and the number of patents of each kind, there was a considerable difference between the maximum and the minimum value, which indicated that the innovation level among different cities was obviously unbalanced. The main research purpose of this study was to explore if this effect was caused by the talent policy. The average value of “the talent war” policy was 0.012, and the average GDP was 2.190. The average number of invention patents obtained ($\mathrm{Invaa}$) was 10.070, the average number of utility model patents ($\mathrm{Umaa}$) was 4.826, and the average number of design patents ($\mathrm{Desaa}$) was 6.669. The average value of total patents ($\mathrm{Sumaa}$) was 5.699. The standard deviation of the proportion of the secondary industry in GDP and the green coverage rate of built-up areas was large, which implied that there was a considerable difference between secondary industry and green coverage in the selected cities.

4. Results

The consistency of the changing trend in the treatment group and control group before the implementation of “the talent war” policy is an important prerequisite for estimating its effect when using the difference-in-differences model. In this work, the event study method was used to test whether this premise was satisfied. Firstly, the interactive terms between the year and the policy implementation time of the treatment group were generated and further used as explanatory variables for regression. The difference between the treatment group and the control group in a specific year was reflected by the coefficient of the interaction item; if the coefficient of the interaction item was not significant, then the precondition was considered to be met. Because there were differences in the start times of “the talent war” among cities, we eliminated the data from the -1 period to avoid multicollinearity.

As can be seen from Figure 2, before the implementation of the policy, the 90% confidence intervals covered 0, i.e., there was no obvious difference between the treatment group and the control group before the implementation of “the talent war”. However, this gap grew significantly one year after the implementation of the policy. Thus, the parallel trend test was satisfied, and the difference-in-differences model could be used to estimate the relationship between talent policy and urban innovation.

4.1. The Impact of Talent Policy on Urban Innovation

We used benchmark regression to estimate the effect of talent policy on urban innovation to test Hypothesis 1. The regression results are shown in

Table 3. Benchmark regression results.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
	$\mathrm{S}\mathrm{u}\mathrm{m}\mathrm{a}\mathrm{a}$	$\mathrm{I}\mathrm{n}\mathrm{v}\mathrm{a}\mathrm{a}$	$\mathrm{U}\mathrm{m}\mathrm{a}\mathrm{a}$	$\mathrm{D}\mathrm{e}\mathrm{s}\mathrm{a}\mathrm{a}$	$\mathrm{S}\mathrm{u}\mathrm{m}\mathrm{a}\mathrm{a}$	$\mathrm{I}\mathrm{n}\mathrm{v}\mathrm{a}\mathrm{a}$	$\mathrm{U}\mathrm{m}\mathrm{a}\mathrm{a}$	$\mathrm{D}\mathrm{e}\mathrm{s}\mathrm{a}\mathrm{a}$
$\mathrm{Treat}\times \mathrm{Post}$	0.1835 ***	0.3224 ***	0.1984 ***	0.1300 *	0.1649 ***	0.3128 ***	0.1677 ***	0.1756 **
	(0.0393)	(0.1117)	(0.0394)	(0.0739)	(0.0420)	(0.1077)	(0.0396)	(0.0809)
Control variable	N	N	N	N	Y	Y	Y	Y
City fixed effect	Y	Y	Y	Y	Y	Y	Y	Y
Year fixed effect	Y	Y	Y	Y	Y	Y	Y	Y
Observations	2770	2770	2770	2770	2770	2770	2770	2770
adj. R2	0.961	0.957	0.957	0.904	0.963	0.959	0.959	0.906

Note: *, ** and *** are significant at 0.1, 0.05 and 0.01 respectively, and the values in brackets are standard errors.

. In Table 3, the columns (1)–(4) show the city fixed effect and the year fixed effect, without the addition of control variables, and the coefficients in columns (1)–(3) were significant at the level of 1%, indicating that there was a significant positive causal relationship between the introduction of the talent policy and urban innovation, i.e., the talent policy significantly promoted urban innovation. On this basis, columns (5)–(8) suggest a series of control variables to alleviate possible endogenous bias. It can be seen that after the control variables were added, the coefficients and significance changed, and the values and significance of the coefficients in columns (5)–(7) decreased. Nevertheless, they were still significant at a statistical level of 1% or above and were basically consistent with the empirical results of a series of robustness tests presented below, thus proving the reliability of the estimated results. Judging from the estimated coefficients in columns (5)–(8), “the talent war” policy significantly improved urban innovation, which was reflected not only in the total number of patents obtained, but also in the number of inventions, utility models, and design patents. Judging from the estimated coefficients, “the talent war” policy increased the total number of patents by 16.49%, the number of invention patents by 31.28%, the number of utility model patents by 16.77%, and the number of design patents by 17.56%. It is worth noting that a greater number of invention patents was obviously beneficial for urban innovation, both before or after the control variables were added, due to their high technical content; they make a more valuable contribution to urban innovation compared to utility model patents and design patents. Generally speaking, the benchmark regression results showed that the introduction of talent policy played a significant role in promoting urban innovation, and the estimated coefficient values were highly consistent with the findings.

4.2. The Impact of Talent Policy Subsidy Mode on Urban Innovation

Table 4. Regression results of the influence of talent policy subsidy mode on urban innovation.

	$\mathrm{S}\mathrm{u}\mathrm{m}\mathrm{a}\mathrm{a}$	$\mathrm{I}\mathrm{n}\mathrm{v}\mathrm{a}\mathrm{a}$	$\mathrm{U}\mathrm{m}\mathrm{a}\mathrm{a}$	$\mathrm{D}\mathrm{e}\mathrm{s}\mathrm{a}\mathrm{a}$
Settlement support	0.2126 *	0.2290	0.1651	0.2595
	(0.1108)	(0.2153)	(0.1003)	(0.2471)
Employment subsidy	0.2779 **	0.3938 *	0.2634 ***	0.1374
	(0.1101)	(0.2049)	(0.1006)	(0.2189)
Talent housing subsidy	0.2543 ***	−0.0189	0.2586 ***	0.1756
	(0.0926)	(0.2187)	(0.0865)	(0.1861)
Living service subsidy	0.2082 *	0.2536	0.1931 **	0.1704
	(0.1071)	(0.2279)	(0.0862)	(0.2544)
Control variable	Y	Y	Y	Y
City fixed effect	Y	Y	Y	Y
Year fixed effect	Y	Y	Y	Y

Note: *, ** and *** are significant at 0.1, 0.05 and 0.01 respectively, and the values in brackets are standard errors.

shows the coefficients (${\alpha }_1$) and significance of the interactive items from the regression results. Judging from the total number of patents obtained, most of the interactive item coefficients for policy subsidies were significantly positive. Compared with the control group, the interactive item coefficient of the employment and entrepreneurship subsidy was significantly positive at a level of 5%, with a policy effect of 27.79%. After introducing settlement support, the total number of obtained patents increased by 21.26%; the policy effect of the talent housing subsidy was 25.43%, followed by the living service subsidy at 20.82%.

The coefficient of the employment and entrepreneurship subsidy was the largest. It can be seen that the employment and entrepreneurship subsidy contributed the most to the overall policy efficiency, followed by the talent housing subsidy. Providing settlement support and living service subsidies only passed the significance test at a level of 10%, which was consistent with the expectation of hypothesis 3. Judging from the number of invention patents obtained, the multiplication coefficient of the employment and entrepreneurship subsidy was significantly positive, indicating that the greater the employment and entrepreneurship subsidy, the greater the impact on the number of invention patents. The employment and entrepreneurship subsidy alone was able to increase the number of invention patents by 39.38%, while the other subsidy policies showed no significant increase. Comparing the coefficients of the employment and entrepreneurship subsidy for the three types of patents, it is evident that this subsidy type most successfully stimulated growth in the number of invention patents, followed by the number of utility model patents. However, its promotion effect on the number of design patents was not significant. The policy of subsidizing settlement was not notably effective in increasing the number of each of the three patent types. However, it had a certain promotional effect on the growth of the total number of patents. Subsidies for housing and living services were shown to raise the total number of patents by increasing the number of utility model patents. Generally speaking, the greater the subsidies of the talent policy, the greater the total number of patents obtained in the region in that year, and the more positive the impact on urban innovation. The employment and entrepreneurship subsidies proved to be the most efficient at increasing the total number of patents, where the number of invention patents and utility model patents grew the most. Therefore, the best way to increase the number of invention patents, which represent substantive innovation, is to increase employment and entrepreneurship subsidies.

4.3. The Impact of Innovation Willingness and Talent Agglomeration on Urban Innovation

The investment of cities in innovation activities shows to what extent they are willing to be leaders in innovation [49], greatly affects the actual innovation level, and determines the implementation effect of the talent policy [50]. Referring to the existing literature [51,52], in this study, the proportion of R&D investment and local financial expenditure in each city was used to measure the willingness of cities to be involved in innovation. At the same time, the number of talents per million people was used to represent the talent agglomeration level of cities, implying that “the sum of the number of employees in scientific research, technical services and geological exploration industries and information transmission, computer services and software industries in each city accounts for the proportion of the total population of cities at the end of the year”. The innovation willingness of cities and talent agglomeration level were used to replace urban innovation, while urban fixed effect and time fixed effect were also included in the model. The regression results are shown in

Table 5. Regression results of innovation willingness and talent agglomeration.

	(1)	(2)	(3)	(4)
	Innovation Will	Innovation Will	Talent Agglomeration	Talent Agglomeration
$\mathrm{Treat}\times \mathrm{Post}$	0.3332 ***	0.2081 **	0.5878 ***	0.4020 **
	(0.0772)	(0.0809)	(0.1926)	(0.1991)
Control variable	N	Y	N	Y
City fixed effect	Y	Y	Y	Y
Year fixed effect	Y	Y	Y	Y
Observations	2770	2770	2770	2770
adj. R2	0.789	0.815	0.722	0.727

Note: ** and *** are significant at 0.05 and 0.01 respectively, and the values in brackets are standard errors.

. Column (1) and column (3) show the city fixed effect and year fixed effect, respectively, without the addition of control variables. The regression result was significant at the 1% level, which indicates that the introduction of talent policy had a significant positive causal relationship with the innovation willingness of cities and their level of talent agglomeration. Columns (2) and (4) show the regression results with the addition of control variables. The coefficient of the dummy variable, “the talent war”, was significantly positive at the level of 5% its effect on the innovation willingness of cities was 20.81%, and its effect on the talent agglomeration level was 40.20%. These results show that when compared with cities that did not implement “the talent war” policy, the talent policy had a significant effect on the innovation willingness of cities and their talent agglomeration level. Thus, Hypothesis 2 was verified.

Compared with utility model patents and design patents, invention patents represent substantive innovation and make a great contribution to urban innovation. Therefore, this study took the number of invention patents as an example to verify the influence mechanism of the four kinds of talent policy subsidies on urban innovation. The results are shown in

Table 6. Regression results of influence mechanism of talent policy subsidy mode.

	(1)	(2)
	$\mathrm{I}\mathrm{n}\mathrm{v}\mathrm{a}\mathrm{a}$	$\mathrm{I}\mathrm{n}\mathrm{v}\mathrm{a}\mathrm{a}$
Settlement support	0.3444 **	0.0269
	(0.1421)	(0.0801)
Employment subsidy	0.3931 ***	0.1716 **
	(0.1207)	(0.0851)
Talent housing subsidy	0.2509 *	0.0150
	(0.1411)	(0.0488)
Living service subsidy	0.3461 **	0.0248
	(0.1705)	(0.0640)
Control variable	Y	Y
City fixed effect	Y	Y
Year fixed effect	Y	Y

Note: *, ** and *** are significant at 0.1, 0.05 and 0.01 respectively, and the values in brackets are standard errors.

. Column (1) shows the regression results of the interaction between the four subsidy modes and the innovation willingness of cities. It can be seen that the four subsidy modes, including settlement support, the employment and entrepreneurship subsidy, the housing subsidy, and the living service subsidy, could all improve the innovation willingness of cities, thus promoting urban innovation. Column (2) shows the regression results of the interaction among the four types of talent policy subsidies and talent agglomeration. It can be seen that among the four types of talent policy subsidies, only the employment and entrepreneurship subsidies promoted urban innovation by improving the level of talent agglomeration, while the other talent policy subsidies had not yet entered a benign interactive state of promoting urban innovation through talent agglomeration.

4.4. Robustness Test

4.4.1. Elimination of Interference from Other Policy Effects

The implementation of other policies or institutions in the research period could interfere with the effect of talent policy implementation [53]. This study selected other representative policies in the sample period to control for them accordingly. According to the appraisal of “China’s Top Ten Innovative Cities” at the 2011 Annual Independent Innovation Conference, cities that have won the national innovation title have the opportunity to obtain more R&D funds or other innovation support, and the announcement itself will also attract more innovative talents to settle in these cities, thus injecting more vitality into innovation. In order to control for the influence of these cities, the model used in this paper included interactive terms of the binary variable of the top ten innovative cities and the binary variable of the evaluation time of innovative cities. It can be seen from

Table 7. Regression results of eliminating interference.

	$\mathrm{S}\mathrm{u}\mathrm{m}\mathrm{a}\mathrm{a}$	$\mathrm{I}\mathrm{n}\mathrm{v}\mathrm{a}\mathrm{a}$	$\mathrm{U}\mathrm{m}\mathrm{a}\mathrm{a}$	$\mathrm{D}\mathrm{e}\mathrm{s}\mathrm{a}\mathrm{a}$
$\mathrm{Treat}\times \mathrm{Post}$	0.1651 ***	0.3138 ***	0.1675 ***	0.1764 **
	(0.0419)	(0.1067)	(0.0397)	(0.0813)
Control variable	Y	Y	Y	Y
City fixed effect	Y	Y	Y	Y
Year fixed effect	Y	Y	Y	Y
Observations	2770	2770	2770	2770
adj. R2	0.963	0.959	0.959	0.906

Note: ** and *** are significant at 0.05 and 0.01 respectively, and the values in brackets are standard errors.

that the interaction coefficient was still significant after controlling for this influence.

4.4.2. Differences in Administrative Level

The administrative level of municipalities directly under the central government is different from those of other prefecture-level cities. On the one hand, municipalities under the central government are directly controlled by the jurisdiction of the central government, while there is a certain subordinate relationship between other prefecture-level cities and their provinces. Although the government decision-making departments of the two administrative units are relatively independent, there is still a certain pressure transmission mechanism that restricts decisions in prefecture-level cities to provincial governments [54]. On the other hand, municipalities directly under the central government are generally the center of China’s economy and culture, and they have a relatively complete industrial system. Therefore, municipalities directly under the central government are generally superior to other prefecture-level cities in terms of factor endowment and economic foundation, and they are bound to take the vanguard position in policy reforms. The same transfer of payment will bring economies of greater scale, which may lead to a policy tilt effect caused by the city’s strength, environmental factors, and other conditions [55]. Due to the particularity of the administrative level of municipalities directly under the central government, this study excluded the corresponding samples and re-estimated the results. The results are shown in

Table 8. Regression results of excluding samples from municipalities directly under the central government.

	$\mathrm{S}\mathrm{u}\mathrm{m}\mathrm{a}\mathrm{a}$	$\mathrm{I}\mathrm{n}\mathrm{v}\mathrm{a}\mathrm{a}$	$\mathrm{U}\mathrm{m}\mathrm{a}\mathrm{a}$	$\mathrm{D}\mathrm{e}\mathrm{s}\mathrm{a}\mathrm{a}$
$\mathrm{Treat}\times \mathrm{Post}$	0.1622 ***	0.3100 ***	0.1663 ***	0.1669 **
	(0.0425)	(0.1084)	(0.0404)	(0.0813)
Control variable	Y	Y	Y	Y
City fixed effect	Y	Y	Y	Y
Year fixed effect	Y	Y	Y	Y
Observations	2730	2730	2730	2730
adj. R2	0.961	0.956	0.956	0.902

Note: ** and *** are significant at 0.05 and 0.01 respectively, and the values in brackets are standard errors.

. It can be seen that after excluding the samples of municipalities directly under the central government, the regression results were basically consistent with the previous results, and the number of patents was still significantly positive at the level of 0.01, verifying the positive effect of “the talent war” policy on urban innovation.

4.5. Heterogeneity Test

4.5.1. The Perspective of Geographical Location

China’s coastal cities have relative superiority in geographical position and infrastructure and are characterized by convenient transportation, prosperous business opportunities, significant logistics advantages, a strong population agglomeration capacity, and rapid economic development [56]. However, cities in non-coastal areas have a relatively weak ability to attract migrants and a low level of population agglomeration due to their traffic and livability conditions. Based on this, we examined whether geographical location affected urban innovation. If the city under consideration was located in a coastal province, it was classified as a coastal city; otherwise, it was classified as an inland area city. In

Table 9. Regression results of urban geographical location.

	Coastal Areas				Inland Area
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
	$\mathrm{S}\mathrm{u}\mathrm{m}\mathrm{a}\mathrm{a}$	$\mathrm{I}\mathrm{n}\mathrm{v}\mathrm{a}\mathrm{a}$	$\mathrm{U}\mathrm{m}\mathrm{a}\mathrm{a}$	$\mathrm{D}\mathrm{e}\mathrm{s}\mathrm{a}\mathrm{a}$	$\mathrm{S}\mathrm{u}\mathrm{m}\mathrm{a}\mathrm{a}$	$\mathrm{I}\mathrm{n}\mathrm{v}\mathrm{a}\mathrm{a}$	$\mathrm{U}\mathrm{m}\mathrm{a}\mathrm{a}$	$\mathrm{D}\mathrm{e}\mathrm{s}\mathrm{a}\mathrm{a}$
$\mathrm{Treat}\times \mathrm{Post}$	0.1465 ***	0.4605 ***	0.1478 ***	0.2113 **	0.1714 ***	0.1742	0.1733 ***	0.1420
	(0.0437)	(0.1065)	(0.0432)	(0.0888)	(0.0625)	(0.1642)	(0.0576)	(0.1221)
Control variable	Y	Y	Y	Y	Y	Y	Y	Y
City fixed effect	Y	Y	Y	Y	Y	Y	Y	Y
Year fixed effect	Y	Y	Y	Y	Y	Y	Y	Y
Observations	2540	2540	2540	2540	2620	2620	2620	2620
adj. R2	0.958	0.950	0.952	0.897	0.959	0.954	0.953	0.892

Note: ** and *** are significant at 0.05 and 0.01 respectively, and the values in brackets are standard errors.

, columns (1)–(4) list the regression results of the coastal areas, and columns (5)–(8) list the regression results of the inland areas. As can be seen from Table 9, the correlation coefficient of columns (1)–(3) passed the significance test at 1%, and column (4) passed the significance test at 5%. The regression coefficients for coastal areas were all positive, while the correlation coefficient of columns (5)–(8) was only significant at the level of 1% in columns (5) and (7). In terms of the total number of patents, although the level of urban innovation in inland areas was higher than that in coastal areas, urban innovation in inland areas was contributed to by the number of utility model patents, rather than substantive innovation. In coastal areas, the three types of patents simultaneously played a significant role in promoting urban innovation, where substantive innovation made the greatest contribution with a significantly positive coefficient at the level of 1% and a policy effect of 46.05%. A possible reason for this is that the implementation of the talent policy played a timely role in improving urban innovation in inland areas, significantly promoting the growth of total inventions there. However, due to the weak accumulation of knowledge in inland areas, talents cannot give full play to the leading role of innovation. Thus, only non-substantive innovations, rather than invention patents representing substantive innovations, increased. Compared with inland cities, the talent introduction policy had a more significant effect on substantive urban innovation in coastal cities, which proves to some extent that the geographical position of cities plays an important role in how the talent policy affects urban innovation.

4.5.2. The Perspective of Urban Business Environment

In fact, the selective implementation of talent policies is influenced by the government’s intervention in the market. A good urban business environment means that the government has simplified administration and decentralized power, reduced their intervention in the market and the cost of innovation, and made preferential policies more predictable [57,58]. Therefore, the effects of talent policy implementation may be different for different urban business environments. Innovation activities need not only capital investment, but also a business environment, e.g., external legal and governmental support [59].

In the general free market environment, the efficiency of factor allocation and innovation is higher. Fan [60] compiled a market-oriented index of each region, among which the index of “reducing intervention in enterprises” reflected the degree of government intervention in local enterprises. This study used this index to measure the urban business environment. The smaller the index, the stronger the degree of government intervention. The market index used in this analysis and the following heterogeneity test was obtained from the Wind database. In this study, the samples were divided into high and low groups. If the degree of government intervention in the area where the city is located was higher than the average value of this index in the same year, the city was placed in the group with a high degree of government intervention; otherwise, it was placed in the low group. As can be seen from

Table 10. Regression results of urban business environment.

	Low Degree				High Degree
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
	$\mathrm{S}\mathrm{u}\mathrm{m}\mathrm{a}\mathrm{a}$	$\mathrm{I}\mathrm{n}\mathrm{v}\mathrm{a}\mathrm{a}$	$\mathrm{U}\mathrm{m}\mathrm{a}\mathrm{a}$	$\mathrm{D}\mathrm{e}\mathrm{s}\mathrm{a}\mathrm{a}$	$\mathrm{S}\mathrm{u}\mathrm{m}\mathrm{a}\mathrm{a}$	$\mathrm{I}\mathrm{n}\mathrm{v}\mathrm{a}\mathrm{a}$	$\mathrm{U}\mathrm{m}\mathrm{a}\mathrm{a}$	$\mathrm{D}\mathrm{e}\mathrm{s}\mathrm{a}\mathrm{a}$
$\mathrm{Treat}\times \mathrm{Post}$	0.1537 ***	0.3722 ***	0.1550 ***	0.1580 *	0.1637 *	0.0435	0.1531	0.2848 *
	(0.0439)	(0.1210)	(0.0423)	(0.0842)	(0.0862)	(0.1193)	(0.0948)	(0.1583)
Control variable	Y	Y	Y	Y	Y	Y	Y	Y
City fixed effect	Y	Y	Y	Y	Y	Y	Y	Y
Year fixed effect	Y	Y	Y	Y	Y	Y	Y	Y
Observations	2656	2656	2656	2656	2503	2503	2503	2503
adj. R2	0.962	0.957	0.957	0.904	0.953	0.946	0.947	0.881

Note: * and *** are significant at 0.1 and 0.01 respectively, and the values in brackets are standard errors.

, columns (1)–(4) show the regression results of cities with a low degree of intervention. It can be seen that the coefficients in columns (1)–(3) were not only positive, but also passed the significance test at a level of 1%. This indicated that in cities located in areas with good business environments, the implementation of policies significantly increased the total number of patents obtained, including the number of invention patents and utility model patents. In such cities, it was substantive innovation that contributed the most to urban innovation, with a policy effect of 37.22%. Columns (5)–(8) list the regression results of areas with poor business environments. It can be seen that the coefficient corresponding to the number of invention patents and the number of utility model patents was not significant, indicating that in cities with strong government intervention, the role of the talent policy in substantive innovation was not obvious. However, judging from the total amount of patents, the talent policy had a significant promotional effect on urban innovation in cities with a low or high degree of government intervention. On the other hand, judging from the level of substantive innovation, the incentive’s effect on urban innovation in cities with a low degree of government intervention was significant at a level of 1%, while that of cities with a high degree of government intervention was not significant. This shows that when compared with the cities with high government intervention in the market, the impact of the talent policy on substantive innovation was more obvious in the cities with low government intervention.

4.5.3. The Perspective of the Degree of Intellectual Property Protection

Usually, innovation has strong externalities and is easy to imitate, which results in free-riding behavior and damages the interests of the innovation subject. Therefore, the degree of intellectual property protection in cities directly affects the innovation ability of the city. With the optimization of intellectual property rights and the improvement in the efficiency of property rights enforcement, the phenomenon of free riding in the process of innovation can be effectively reduced, and the effect on innovation can be enhanced [61].

This study used the index “Maintaining the Legal Environment of the Market” compiled by Fan [60] to measure the degree of urban intellectual property protection. The research samples were divided into two groups according to the degree of intellectual property protection, so as to explore the heterogeneous effects of the talent policy on urban innovation under different degrees of intellectual property protection. The regression results are shown in

Table 11. Return results of intellectual property protection.

	High Degree				Low Degree
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
	$\mathrm{S}\mathrm{u}\mathrm{m}\mathrm{a}\mathrm{a}$	$\mathrm{I}\mathrm{n}\mathrm{v}\mathrm{a}\mathrm{a}$	$\mathrm{U}\mathrm{m}\mathrm{a}\mathrm{a}$	$\mathrm{D}\mathrm{e}\mathrm{s}\mathrm{a}\mathrm{a}$	$\mathrm{S}\mathrm{u}\mathrm{m}\mathrm{a}\mathrm{a}$	$\mathrm{I}\mathrm{n}\mathrm{v}\mathrm{a}\mathrm{a}$	$\mathrm{U}\mathrm{m}\mathrm{a}\mathrm{a}$	$\mathrm{D}\mathrm{e}\mathrm{s}\mathrm{a}\mathrm{a}$
$\mathrm{Treat}\times \mathrm{Post}$	0.1819 ***	0.3754 ***	0.1922 ***	0.1713 *	0.1320 **	0.0915	0.1133 *	0.2098 *
	(0.0497)	(0.0951)	(0.0418)	(0.0980)	(0.0526)	(0.0833)	(0.0592)	(0.1175)
Control variable	Y	Y	Y	Y	Y	Y	Y	Y
City fixed effect	Y	Y	Y	Y	Y	Y	Y	Y
Year fixed effect	Y	Y	Y	Y	Y	Y	Y	Y
Observations	2618	2618	2618	2618	2530	2530	2530	2530
adj. R2	0.961	0.956	0.956	0.902	0.955	0.948	0.949	0.885

Note: *, ** and *** are significant at 0.1, 0.05 and 0.01 respectively, and the values in brackets are standard errors.

. Columns (1)–(4) show the cities with a high degree of intellectual property protection, and the regression coefficients in columns (1)–(4) indicate that in these cities, the talent policy played a positive role in promoting urban innovation. Specifically, for each additional unit of the talent introduction policy, the total number of patents increased by 18.19% and the number of invention patents increased by 37.54%. Columns (5)–(8) list cities with a low degree of intellectual property protection; their coefficients and significance were lower than those of cities with a high degree of intellectual property protection. In particular, there was a lack of invention patents representing substantive innovation, which showed that in cities with a low degree of intellectual property protection, the impact of the talent policy on urban innovation was lower.

5. Further Analysis

Each city participates in the competition for talents, and “the talent war” policy was introduced to urge talents to “vote with their feet”, a process by which all kinds of high-quality talents flow across regions to specific cities in order to maximize their own benefits. “The talent war” cities make use of their own advantages and policy bias to stimulate the agglomeration of talents in the region and enhance their own innovation level. Although the talent policy has had a positive impact on urban innovation by improving the innovation willingness of cities and promoting the agglomeration of talents, the talent policy may not improve the level of urban innovation in neighboring cities; on the contrary, it may inhibit urban innovation in neighboring cities to some extent, i.e., there exists a “siphon effect”. Cities that implement “the talent war” policy will provide more employment opportunities and relatively high capital rewards, not only absorbing local talents and capital, but also attracting talents and capital from neighboring cities [62]. Therefore, although talents flowing into “the talent war” cities benefit from the agglomeration economy—and although these cities gradually improve their urban innovation ability by constantly obtaining the necessary labor force, capital, and other production factors from neighboring cities—the innovation ability of cities with talent outflow is restrained.

Whether the surrounding cities that have not yet joined the competition are affected by “the talent war” cities and whether “the talent war” cities have a spillover or siphon effect on the urban innovation of other cities are questions worthy of discussion. In this paper, the econometric model in formula (13) was constructed to test the impact of implementing “the talent war” policy on neighboring cities. The results are shown in

Table 12. Policy effects to neighboring cities.

	$\mathrm{S}\mathrm{u}\mathrm{m}\mathrm{a}\mathrm{a}$	$\mathrm{I}\mathrm{n}\mathrm{v}\mathrm{a}\mathrm{a}$	$\mathrm{U}\mathrm{m}\mathrm{a}\mathrm{a}$	$\mathrm{D}\mathrm{e}\mathrm{s}\mathrm{a}\mathrm{a}$
$\mathrm{Near}\times \mathrm{Post}$	−0.0870 *	0.0578	−0.0829	−0.1354 *
	(0.0471)	(0.0510)	(0.0523)	(0.0775)
Control variable	Y	Y	Y	Y
City fixed effect	Y	Y	Y	Y
Year fixed effect	Y	Y	Y	Y
Observations	2400	2400	2400	2400
adj. R2	0.951	0.940	0.946	0.876

Note: * is significant at 0.1, and the values in brackets are standard errors.

. As can be seen from Table 12, the interaction coefficient of total patents was −0.087 and the interaction coefficient of design patents was −0.1354, both of which were significantly negative at the level of 0.1, indicating that the impact of “the talent war” on neighboring cities at present does not show positive externalities, but instead has a certain restraining effect; cities that implement “the talent war” policy draw talents and capital from neighboring cities through the siphon effect [63]. The possible reason for this is that “the talent war” cities often give higher wages to introduced talents, which interferes with the free flow of the labor force to a certain extent, resulting in labor market segmentation. The reward for the same labor can vary significantly in different cities. In addition, the policy was not shown to have a notable impact on the number of invention patents in neighboring cities, which indicates that “competing for people” does not improve local substantive innovation.

6. Conclusions

This paper took 277 prefecture-level cities in China in the time period of 2010–2019 as research objects and empirically tested the effect of the talent policy on urban innovation by using the difference-in-differences model. Firstly, we found that the cities participating in “the talent war” significantly improved the development of urban innovation, which indicates a positive role of “the talent war” policy in promoting local innovation activities; specifically, the increase in the number of invention patents was the most beneficial for urban innovation.

Secondly, the greater the subsidy of the talent policy, the better the impact on urban innovation; the employment and entrepreneurship subsidy had the strongest incentive effect on urban innovation. Thus, the best way to increase the number of invention patents is to increase employment and entrepreneurship subsidies.

Thirdly, to successfully promote urban innovation through the implementation of “the talent war” policy, innovation willingness and talent agglomeration mechanisms must be improved. Willingness to innovate was improved to a different extent by the four types of subsidies for talents. Specifically, the employment and entrepreneurship subsidy was beneficial for talent agglomeration, leading to the promotion of substantive innovation. However, the other types of talent subsidies have not yet entered a benign interactive state to promote urban innovation.

Fourthly, at present, “the talent war” cities have had a siphon effect on the innovation of neighboring cities, which to some extent inhibits innovation there.

Fifthly, whether a city is located in a coastal area or an inland area, whether the business environment is good or bad, or whether the degree of intellectual property protection is high or low, “the talent war” policy has a certain effect on urban innovation in terms of the total number of patents. However, judging from the number of invention patents that represent substantive innovation, “the talent war” policy more significantly promoted substantive innovation and development in coastal cities and cities with good business environments and a high degree of intellectual property protection.

Based on the results of this study, the following recommendations for the implementation of the talent policy can be formulated. Firstly, we should encourage more cities to cultivate an innovative atmosphere by introducing the talent policy. The conclusion of this research shows that the talent policy can significantly improve the innovation level of a city. Thus, local governments should be encouraged to intensify the propaganda associated with the talent policy and rationally guide regional talent flow to increase the efficiency of labor allocation. Secondly, the employment and entrepreneurship subsidy was shown to have the best incentive effect on urban innovation. Therefore, in the process of formulating and implementing the talent policy, attention should be paid to subsidizing employment and entrepreneurship, as well as talents’ housing; this would give full play to the advantages of the talent policy in policy design, and further enhance the attractiveness of cities for talents as agglomeration regions. Thirdly, local governments should not only focus on competing for new talents, but also consider the cultivation of existing talents and develop management mechanisms to improve their retention.

Fundamentally, governments should establish and improve a high-quality development model that promotes urban innovation through market competition and gives play to the decisive role of the market in the allocation of innovative resources. In addition, a good business environment and a perfect intellectual property protection system will enhance the incentive effect of the talent policy on substantive urban innovation. Therefore, the construction of urban business environments should be strengthened such that the talent introduction policy can better promote the development of urban innovation.