Market Access,Population Mobility and Labor Force Structure: Evidence From China’s High-Speed Railway

Abstract

Population mobility is becoming a common phenomenon worldwide which can contribute to the development of individuals and societies. This paper offers that the launch of HSR would facilitate population mobility and spatial disparity. We also find that the HSR connection mainly affects migration of people under 14 and 15 to 64 years old, with no effect on the old over 65. This implies that labor-led family migration has become the main form of movement under the effect of HSR. Furthermore, HSR connection has promoted the concentration of high-skilled labor and the outflow of low-skilled labor in large cities. This study offers useful insights for policy makers that concerns transport infrastructure and the issue of household and workforce structure it raises.

JEL: O30, O18, D60, L31.

Introduction

The power of population mobility has long been recognized, as is a key driver of economic growth (Chen & Fang, 2013; Greenwood, 1973), labor market flexibility (X. Wang et al., 2011; Zhao, 2005), skill enhancement (Aminuzamman, 2007; Clemens, 2013), and regional development (Fu & Gabriel, 2012; Gennaioli et al., 2013). While population movement becomes more and more powerful, the importance of exploring the mechanisms behind it and the various implications it may have also increases.

This paper concerns whether the population size and distribution across cities will change as a result of improvements in transportation infrastructure. According to Harris (1954), areas closer to surrounding markets have the ability to deliver quickly to the consumers and furnish suitable conditions for the development of economy. Therefore, traditional wisdom would argue “yes” because the reduction of transportation costs apparently shortens the distance from a city to the surrounding markets, which in turn creates a favorable environment attracting population and labor force concentration. However, this view may be challenged in an era of HSR, which mainly caters to passenger transport. Does the reduction in travel cost can still yield similar results?

We investigate this question using China (2007–2014) as a quasi-natural experiment. The rapid development of HSR in China during this period provides sufficient variation in the space and time distance between cities. The results show that the launch of HSR would facilitate population mobility and spatial disparity, regardless of the model setting. We further focus on the differences in the population by age group, finding that the HSR connection mainly affects migration of people under 14 and 15 to 64 years old, with no effect on the old over 65. This implies that labor-led family migration has become the main form of movement under the effect of HSR. We then use the employment data across sectors to check the changes in the structure of labor force and find that, HSR connection has promoted the concentration of high-skilled labor and the outflow of low-skilled labor in large cities. Finally, we provide an explanation of population mobility in terms of market access.

This paper makes several contributions to the existing literature. First, it enriches the literature on the role of HSR in population mobility and its mechanisms. Previous studies focused on the positive or negative effects of transportation on socio-economic outcomes, such as economic growth (Donaldson & Hornbeck, 2016; Qin, 2016), inequality (Bittencourt & Giannotti, 2021; Pereira et al., 2017), environmental quality (Allen et al., 2021; Saidi & Hammami, 2017), among other outcomes. However, research on the effects of transportation infrastructure, particularly HSR, on population mobility has not received extensive attention. The results have gone in multiple directions, ultimately resulting in a lack of consensus. According to Xu and Sun (2021), HSR has promoted intercity migration, with large cities exerting a significant siphon effect on small cities. Other researchers also share the same arguments (Guirao et al., 2018; Monzon et al., 2019; Yu, 2021). Conversely, some studies suggest that the impact of HSR on population mobility is not entirely positive (Deng et al., 2019; F. Wang et al., 2019, 2022). For example, F. Wang et al. (2022) find that HSR connection can lead to economic growth in large cities, but it would have negative effects on small and medium-sized cities, such as population loss and economic decline. Regarding the channels behind, researches show that HSR can facilitate migration by increasing connectivity and accessibility (L. Wang, 2018), reducing travel time (F. Wang et al., 2019), or increasing business and job opportunity (Dong, 2018). Our study sheds light on the causes of population mobility in terms of market access and reconciles the contradictions between population growth and decline.

Second, on the issue of population mobility, a large body of research has focused on motivation analysis, for instance, investigating the determinants of migration such as high wages (Dustmann & Weiss, 2007; Hanson & Spilimbergo, 1999; Knapp et al., 2013), good health care (de Haas et al., 2019; Kureková, 2013), strong educational systems (Crivello, 2011; OECD, 2014), or political and institutional issue (Adamson & Tsourapas, 2020; Czaika & Hobolth, 2016). Our study suggests that better intercity passenger transportation could also affect population movements, particularly family migration. Family-based migration has received widespread attention because it might facilitate integration into new environment and socio-economic well-being (Gubernskaya & Dreby, 2017). This paper adds to the literature with a study on how HSR could affect family migration by examining different age groups.

Third, it contributes to the discussion on the distribution effect of HSR on labor force. According to Lin (2017), HSR connection could promote the delivery of services across cities, facilitate cross-city labor sourcing and knowledge exchanges, and ultimately shift the specialization pattern of connected cities toward skilled and communication intensive sectors. Also, HSR connection may facilitate the movement of low-skilled labors to connected cities, thus reshaping employment structures in the cities (Kong et al., 2021). A study that is most related to ours is Feng et al. (2023) which investigated the impact and mechanism of the launch of HSR on the mobility of labor force in China. Using the data of prefecture-level cities in China, they find HSR can significantly promote the agglomeration of labor force with higher education, especially for the large cities. To the best of our knowledge, this is the first study to examine the heterogeneity effect of HSR on high-skilled labor mobility. We extend this research by investigating the differential impact of HSR on employment across sectors.

The remainder of this paper’s structure is as follows. Section 2 describes the background of this study. Section 3 presents the theoretical framework and empirical strategy. Section 4 presents the estimation results and robustness tests. Section 5 presents the heterogeneity analysis. Section 6 presents additional estimated results to further investigate the channels at work. Section 7 serves as the conclusion part.

Research Background

Since the first launch of the HSR line between Beijing and Tianjin in 2008, China has constructed an extensive modern HSR network. Due to the convenience of transportation, HSR has enhanced the regional location advantage of connected areas and formed the spatial characteristics of HSR hinterland as the agglomeration area. In terms of spatial distribution, the area within 50 km of routes accounts for 13% of the national land area, 66% of population and 51% of GDP. The area within 100 km of routes accounts for 20% of the national land area, 77% of population and 67% of GDP (Table 1). In terms of spatial-temporal distribution, the current level of HSR service tends to decrease from east to west, with the east enjoying the densest HSR network and the central and western regions being relatively weak. Figure 1 shows that the 1-hr commuting buffer of the HSR network already covers three major metropolitan areas, that is, Pearl River Delta, Shanghai-Hangzhou-Ningbo, and Beijing-Tianjin-Hebei. The 2-hr commuting buffer has formed continuous urban clusters in the eastern region, making the cities along the route highly integrated.

Table 1.

Properties of HSR Coverage Area.

Distance buffers (km)	GDP share (%)	Population share (%)	Area share (%)
50	51	66	13
100	67	77	20

Figure 1.

Spatial-temporal buffer around the HSR network.

The above analysis shows that the HSR network can basically respond to the needs of spatial economic development and play a positive role in promoting factor flow and agglomeration. In this paper, we will use this context to study the impact of HSR on population mobility and spatial disparity.

Theoretical Framework and Empirical Strategy

Theoretical Basis

Intuitively, when a city is in a more accessible area, it will attract businesses and entrepreneurs to establish themselves in that city, as they can capitalize on the demand for their product or service. This can lead to an increase in job opportunities and economic growth, which in turn can attract more people to move to the city in search of employment. To formalize the intuition on the general equilibrium, we borrow a reduced-form model from Donaldson and Hornbeck’s (2016) to illustrate how transport improvements affect urban population. Assuming that consumers face a continuous set of differentiated goods and satisfy the CES consumer utility function $U={\left[{\int }_0^{1}Q{(j)}^{(\sigma -1)/\sigma }\mathrm{dj}\right]}^{\left({}^{(\sigma -1)/\sigma }\right)}$ . The productivity z_o(j) is independently and identically distributed for all cities and receives a Fréchet distribution $F_o\left(z\right)=\mathrm{Pr}[Z_i\le z]=e^{-T_o{z}^{-\theta }}$ , in which the shift parameter T_o > 0 is location specific reflecting the technological level of the city whereas the parameter θ > 1 is the degree of variation in technology. Bilateral trade is possible between all cities, subject to iceberg transportation costs, ${\tau }_{\mathrm{od}}$  > 1. We denote the domestic city by the subscript o (origin) and the target city by the subscript d (destination). All cities face the Cobb-Douglas production function $c_o=w_o^{\alpha }{r}_o^{\beta }{q}_o^{1-\alpha -\beta }$ , where $w_o$ is the wage, $r_o$ is the rental rate of capital, and $q_o$ is the rental rate of land. Under the assumption of perfect competition and free flow of capital, we can derive following Eaton and Kortum (2002) the equilibrium value of trade flows between each pair of domestic origin and destination locations:

$$\begin{array}{c}\multicolumn{1}{c}{X_{\mathrm{do}}=\frac{T_o{\left(w_o^{\alpha }{r}^{\beta }{q}_o^{1-\alpha -\beta }{\tau }_{\mathrm{od}}\right)}^{-\theta }}{\sum _{o=1}^N{T}_o{\left(w_o^{\alpha }{r}^{\beta }{q}_o^{1-\alpha -\beta }{\tau }_{\mathrm{od}}\right)}^{-\theta }}}\\ \multicolumn{1}{c}{Y_d=\frac{{\left[\mathrm{\Gamma }\left(\frac{\theta +1-\sigma }{\theta }\right)\right]}^{\frac{-\theta }{1-\sigma }}{r}^{-\theta \beta }}{{\left(P_d\right)}^{-\theta }}{T}_o{\left(w_o^{\alpha }{q}_o^{1-\alpha -\beta }{\tau }_{\mathrm{od}}\right)}^{-\theta }{Y}_d}\end{array}$$ (1)

where $P_d$ is the consumer price and $Y_d$ is the total income of the target city. Summing over the value of all trade flows of target cities, we can derive the total income or GDP at the domestic city. Then, again based on the Cobb-Douglas production function, we can obtain the logarithmic form of the population:

$$\begin{array}{c}\multicolumn{1}{c}{\ln N_o=\kappa +\frac{1}{1+(1-\alpha -\beta )\theta }\ln }\\ \multicolumn{1}{c}{T_o+\frac{(1-\alpha -\beta )\theta }{1+(1-\alpha -\beta )\theta }}\\ \multicolumn{1}{c}{\ln L_o+[\frac{1+\mathrm{\alpha }}{1+(1-\alpha -\beta )\theta }+\frac{1}{\mathrm{\theta }}]\mathrm{lnM}{A}_o}\end{array}$$ (2)

where $\kappa$ is a polynomial over the parameters, $L_o$ is the fixed quantity of land, and $M{A}_o=\sum _{d=1}^N{\tau }_{\mathrm{od}}^{-\theta }{\left(P_d\right)}^{\theta }{Y}_d$ is the market access of the domestic city. From Equation 2, we can see that as the cost of transport between the local city and other cities decreases, its market access increases and further increases its population. This leads us to the first hypothesis.

• H1. The launch of HSR will increase the population of the connected cities.

The recent China mobile population report released by the National Health and Family Planning Commission highlights the trend of family-led migration, which is now an important feature of population movements worldwide (OECD, 2022), with a positive effect on mobile populations (Gubernskaya & Dreby, 2017). Combined with the role of HSR, we propose our second hypothesis.

• H2. The launch of HSR can potentially facilitate family-related migration.

Furthermore, the 2020 China Census shows that, the mobile population, mainly the labor force, is gradually moving to developed areas and metropolitan regions centered on large cities. There is already evidence that the HSR has the potential to transform the structure of labor force (Lin, 2017), leading to growth in high-skilled workforce in large cities (Feng et al., 2023). Therefore, we propose the following hypothesis.

• H3. HSR connection may change the workforce structure in cities, especially large cities.

Data

The data applied in this paper include socioeconomic data collected from statistical yearbooks, population census and GIS. Among these data, the socioeconomic statistics are from the county-and prefecture-level statistical yearbooks (2003–2014) and the census data. China’s railway and highway maps are from the Institute of Geography of the Chinese Academy of Sciences. The HSR maps are digitally generated by ArcGIS, and their information from 2008 to 2014 is collected from the National Railway Administration and the 12306 online ticketing system. In addition, the Baidu API was employed to generate the digital map by acquiring the longitude and latitude. The details are as follows.

1. Resident population, registered population, number of elementary school students, GDP, savings balance, fixed asset investment, and administrative area data were collected from the county-and prefecture-level statistical yearbooks.

2. Non-agricultural population, number of employed people, number of people employed in the three industries, and number of people employed in the service industry of 2000 came from China Population Census 2000.

3. The county-level administrative map, the geographic digital map, the railroads map and highways map were from the National Earth System Scientific Data Sharing Platform and OpenStreetMap.

4. China 1 km resolution digital elevation model dataset (DEM) and IGBP China 2000 land cover digital map were collected from the Cold and Arid Regions Science Data Center.

5. HSR lines and stations (2008–2014) were extracted from the National Railway Administration and 12,306 online ticketing system. We use Baidu API to obtain latitude and longitude and generate digital maps manually.

We define each county with administrative boundaries to the nearest HSR within a distance less than 10 km as a connected county (launch of HSR). The two boundary dummy variables used in the paper-whether they lie on the provincial boundary and whether they lie on the prefectural boundary-are extracted by the digital map. The national land slope data used to construct the minimum spanning tree instrumental variables is generated by 1 km DEM map. According to the “Mid-to long-term Railway Network Plan,” the HSR are directly connected to each station city, and the station cities are obtained from 12,306 website with latitude and longitude being obtained from Baidu API. China Statistical Yearbook divides China into eastern, central, western and northeastern regions, and the dummy variables are set accordingly (east_dum, cent_dum, west_dum). Whether it belongs to coastal area (sea_dum) is based on the division of China Marine Statistical Yearbook. The missing data are supplemented by the city statistical yearbook or provincial statistical yearbook. We exclude those counties having experiencing administrative division adjustment, and keep counties that have changed their names after revision. In this paper, Hong Kong, Macau, Taiwan, and Tibet are excluded from the study, and the descriptive statistics of the main variables are shown in Table 2.

Table 2.

Summary Statistics.

Variables	Definition	Mean	S.D.	Data coverage
CRH_pass	Launch of HSR (0/1)	0.098	0.298	2007–2014
Prefect	=1 if a county is in prefect; =0 otherwise	0.12	0.325	2004
City_statu	=1 if a county is a county-level city; =0 otherwise	0.171	0.377	2004
Prov_b	=1 if a county lies on the province boundary	0.476	0.499	2004
Prefect_b	=1 if a county lies on the prefectural boundary	0.924	0.264	2004
Area	Administrative area (km2)	2724.161	5214.509	2007–2014
GDP03	GDP in 2003 (100 million ¥)	62.369	209.588	2003
Population	Resident population (10,000)	65.954	93.974	2007–2014
Pop_reg	Registered population (10,000)	62.041	64.855	2007–2014
Mobility_rate	Resident population/Registered population	0.967	0.19	2007–2014
Primary_student	Number of primary school students (10,000)	4.586	5.268	2007–2014
east_dum	=1 if a county is in the eastern region; =0 otherwise	0.282	0.45	2004
cent_dum	=1 if a county is in the central region; =0 otherwise	0.287	0.452	2004
west_dum	=1 if a county is in the western region; =0 otherwise	0.343	0.475	2004
hori_dum	=1 if a county lies on the horizontal line; =0 otherwise	0.083	0.276	2007–2014
verti_dum	=1 if a county lies on the vertical line; =0 otherwise	0.145	0.352	2007–2014
coastal_dum	=1 if a county is in the coastal area; =0 otherwise	0.391	0.488	2004

Model Setting

To identify the impact of HSR on population mobility and regional spatial structure, we use DID method for econometric analysis to compare the difference between the counties along the route and non-route counties. Figure 2 illustrates the validity of our identification strategy. It shows the time trends of the logarithm of the population in the treatment group (counties along the HSR) and the control group (counties not along the HSR) during 2004 to 2014 (Treatment group refers to all counties along the route as of 2014.). The treatment group and the control group show similar trends before 2007, a year before the launch of HSR. But after 2008, with the successive launch of HSR, the population growth of the two groups diverges significantly: the population of routine counties grew significantly faster than that of non-routine counties. The baseline model has the following specification:

$$\log \left(\mathrm{populatio}{n}_{\mathrm{it}}\right)={\alpha }_i+\beta \mathrm{CRH}\_\mathrm{pas}{s}_{\mathrm{it}}+{\gamma }_t+{ϵ}_{\mathrm{it}}$$ (3)

Figure 2.

Population growth trends comparison.

where i and t denote county i and year t, respectively; $\log \left(\mathrm{populatio}{n}_{\mathrm{it}}\right)$ represents the logarithm of resident population of city i in year t; ${\alpha }_i$ is the county fixed effect, capturing all the time-invariant characteristics of the counties, which might influence the outcome of interest; $\mathrm{CRH}\_\mathrm{pas}{s}_{\mathrm{it}}$ is the regressor of interest in this paper, indicating whether county i launches HSR in year t; ${\gamma }_t$ is the year fixed effect; and ${ϵ}_{\mathrm{it}}$ is the random error term. To address the potential serial correlation and heteroskedasticity, we cluster the standard errors at the county level.

Identifying Assumption and Checks

The validity of the DID estimation requires that the treatment group and the control group would have followed the same time trends under the condition of no policy imposed. The planned routes of the HSR may not be randomly selected, which would lead to the fact that the difference in population growth after 2008 shown in Figure 2 may be caused by some pre-existing differences in the characteristics between the two groups. Therefore, we need to add several key control variables to Equation 3 so that these factors can affect the difference in growth trends can be effectively controlled. For example, according to the “Mid-to long-term Railway Network Plan” and the National Railway Administration, HSR should pass through economically developed areas, connect various urban clusters, coordinate point-line capacity, and preserve intensive land use. Generally speaking, prefectural districts and county-level cities have better location advantages compared to counties, and the centers of urban clusters are more likely to be located in provincial city centers. Six key control variables were selected: prefectural district dummy (prefect), county-level city dummy (city_statu), provincial boundary dummy (prov_b), prefectural boundary dummy (prefect_b), administrative area (area), and total economy in 2003 (GDP03).

Table 3 shows the balancing checks with the six control variables conducted to verify whether controlling for six key determinants of HSR counties selection can lead to better balance between the treatment and control groups. As seen from the statistics of control variables in panel A, 23% of the HSR counties are prefectural districts, and only 9% of the non-HSR counties are prefectural districts; 26% of HSR counties are county-level cities, versus 15% of the non- HSR counties; 43% of the HSR counties are located on provincial boundaries, and 49% of the non-HSR counties are located on provincial boundaries; 89% of the HSR counties are located on prefecture boundaries versus 93% of the non-HSR counties; and the area of the HSR counties are smaller compared to the area of non-HSR counties, while the total economy is larger. Overall, the data illustrate that most of these criteria play an important role in determining the treatment status. Panel B compares the treatment and control groups on various economic and social development variables in the initial year, including the non-agricultural population, total employment, primary industry employment, secondary industry employment, tertiary industry employment, total social savings balance, and fixed asset investment. Column 3 shows that there are significant differences between the HSR counties and non-HSR counties. On average, HSR counties have a larger population, higher employment, larger savings, and higher fixed asset investment. But their primary industry employment rate is lower. However, as shown in column 4, after controlling for the six key factors mentioned above, none of these characteristics exhibit any statistically or economically significant difference between the treatment and control groups. The treatment and control samples are balanced, which is crucial for the identification.

Table 3.

Balance Checks.

Variables	Treatment group	Control group	Unconditional diff.	Conditional diff.
Panel A: Selection criteria
prefect (%)	0.229508 (0.420578)	0.090363 (0.286712)	0.139*** (0.00764)
citystatu (%)	0.262295 (0.439946)	0.146603 (0.353724)	0.256*** (0.00935)
prov_b (%)	0.428571 (0.494944)	0.489415 (0.499907)	−0.0608*** (0.00956)
prefect_b (%)	0.892271 (0.310082)	0.933017 (0.250001)	−0.0407*** (0.00575)
area (10,000 km2)	1559.932 (1047.243)	3038.3 (5811.046)	−1,478*** (54.67)
GDP03 (10,000 ¥)	163.2869 (427.295)	35.13934 (55.10142)	128.1*** (7.327)
Panel B: Other characteristics
Population in 2000	174,947.9 (251,087.2)	90,942.51 (83,972.34)	84,005*** (4.972)	−12,091 (11,561)
Employment in 2000	37,389.57 (30,055.34)	23,810.99 17,149.51	13,579*** (609.0)	1,115 (1,753)
Primary industry employment in 2000	71.24942 (18.37023)	77.48089 (13.45174)	−6.231*** (0.380)	−0.221 (1.617)
(10,000)
Secondary industry employment in 2000	14.34251 (13.74722)	8.969927 (7.930509)	5.373*** (0.279)	0.205 (1.135)
(10,000)
Tertiary industry employment in 2000	14.40838 (5.97825)	13.5491 (7.157803)	0.859*** (0.135)	0.0161 (0.634)
(10,000)
Service sector employment in 2000	2934.569 (4502.53)	1290.604 (1185.929)	1,644*** (88.72)	−28.79 (193.1)
(10,000)
Saving balance in 2003	34.19128 (30.74189)	16.87802 (19.9765)	17.31*** (0.629)	−1.853 (1.170)
(100 million ¥)
Fixed asset investment in 2003	96.46851 (222.4681)	18.46299 (30.22142)	78.01*** (5.010)	2.220 (2.463)
(100 million ¥)
Observations	1,579	14,497

Note.***p < .01. **p < .05. *p < .1.

Another obstacle to identification is the possibility of simultaneous investments in other areas that local governments may take HSR connection as a new engine for local economic growth and invest heavily in related projects (Lin, 2017). Furthermore, other transportation also affects population movements, especially in the past 20 years. For example, China has made great achievements in the construction of highways which “shorten” the distance between two cities/counties and facilitate people’s travel. To deal with these concerns, we control the GDP, the total area of new urban roads built, the distance from each city to the highways and rivers, and the rail passenger traffic. Therefore, throughout our analysis, our preferred specification is

$$\begin{array}{c}\multicolumn{1}{c}{\log \left(\mathrm{populatio}{n}_{\mathrm{it}}\right)}\\ \multicolumn{1}{c}{={\alpha }_i+\beta \mathrm{CRH}\_\mathrm{pas}{s}_{\mathrm{it}}+Z{\mathrm{\text{'}}}_{\mathrm{it}}\sigma +(X\times f\left(t\right)}\end{array})\mathrm{\text{'}}\delta +{\gamma }_t+{ϵ}_{\mathrm{it}}$$ (4)

where vector $Z_{\mathrm{it}}$ includes the indicators of city i in year t such as the GDP, the total area of new urban roads built, transportation conditions, etc., and X are the six key variables mentioned above and f(t) is a polynomial function of time. We interact the six ex-ante variables with time, which can control for the time effects of X.

Empirical Results

Main Results

The main results are reported in Table 4. In Column 1, we do not control for the city-level characteristics; in Columns 2 to 5, we do. Column 1 shows a positive and significant relationship between CRH_pass and resident population. This finding implies that the HSR may have promoted population agglomeration in the affected counties. Columns 2 and 3 report the results allowing interactions between a flexible function of time and all of the major determinants of the HSR introduction, as elaborated in the previous section together with the other control variables. Specifically, interactions of the six key selection variables with a third-order polynomial function of time are included in column 2. Interactions of the six key selection variables with year dummies are included in the estimation reported in column 3. We consistently find a positive and statistically significant effect of the launch of HSR, despite of a significant drop in its estimated magnitude which indicates that the initial factors of these cities influence their future trends. Column 4 replaces the dependent variable with the population mobility rate, which is intended to examine the effect of HSR on population flow. The result shows that the launch of HSR increases population inflow by an average of 2.5%, indicating the population of non-HSR counties converges to HSR counties. Combined, these findings are due to the fact that the reduced travel cost enhances location advantage, and enterprises and labor force are willing to choose connected cities that can obtain more agglomeration benefits. This conclusion is also a validation of our theoretical model above. The location advantage can be explained by the market access, which we will further test later on. Besides, the positive coefficients of Log(GDP) and Log(road area)in columns 2 to 5 indicate that higher economic level and better urban development will promote population inflow. We don’t find a positive effect of rail development on population growth at the county level, at least in our sample. An interesting finding is that the coefficient is not statistically significant if we substitute registered population for the dependent variable, indicating that the population cannot move freely under the restrictive regime of the household registration system (hu-kou system), and the separation of residence and household registration is still very evident.

Table 4.

Main Results.

	Log(population)			Mobility_rate	Log(pop_reg)	Log(primary_student)
Dependent variable	(1)	(2)	(3)	(4)	(5)	(6)
CRH_pass	0.046*** (0.010)	0.029*** (0.010)	0.025*** (0.009)	0.025*** (0.008)	0.000 (0.004)	0.019* (0.010)
Log (GDP)		0.020*** (0.005)	0.215*** (0.035)	0.087*** (0.022)	0.124*** (0.029)	0.049* (0.027)
Log (distance to highway)		0.000 (0.004)	0.006* (0.003)	0.003 (0.003)	−0.003* (0.002)	−0.002 (0.006)
Log (rail passenger traffic)		0.004 (0.004)	0.004 (0.004)	0.003 (0.003)	0.003 (0.002)	−0.001 (0.007)
Log (road area)		0.058*** (0.012)	0.043*** (0.012)	0.020** (0.008)	0.013** (0.006)	0.008 (0.014)
Constant	3.802*** (0.002)	3.510*** (0.081)	3.669*** (0.077)	0.836*** (0.056)	3.866*** (0.035)	1.304*** (0.097)
City fixed effect	YES	YES	YES	YES	YES	YES
Year fixed effect	YES	YES	YES	YES	YES	YES
X × f(t)		YES
X × yeardummy			YES	YES	YES	YES
Observations	11,297	11,297	11,297	11,297	16,067	16,061
R-squared	.991	.989	0.990	.753	.994	.977

Note. X are the six key variables and f(t) is a polynomial function of time. The robust standard errors in the brackets are clustered at the county level.

***

p < .01. **p < .05. *p < .1.

Robustness Tests

Alternative Measure of Population Mobility

China’s mobile population has gradually expanded in recent years. The China Mobile Population Development Report 2016 points out that the scale of mobile population reached 247 million in 2015, accounting for 18% of the total population, and the trend of family-oriented mobility becomes more pronounced. Data from the sixth national census shows that compulsory education for children of the mobile population is now basically guaranteed, with only 2.94% of school-age children failing to receive compulsory education. Considering the trend of family-oriented mobility, an increase in the inflow of people to a county must be accompanied by a corresponding increase in the number of primary school students, as can be used to approximate the trend of the change in the resident population. Hence, we replace the resident population with the number of primary school students in Equation 2. As show in column 6 of Table 4, with the launch of HSR, the number of elementary school students increases by 1.93% and is significant at 10% level, consistent with the previous conclusion of the effect of HSR on population agglomeration.

Placebo Test

To check the extent to which the results are influenced by any omitted variables, a placebo test is conducted by randomly assigning the launch of HSR to counties (Chetty et al., 2009; Li et al., 2016). In order to preserve consistency with the original HSR (i.e., 8 years with HSR launch {16, 25, 89, 58, 8, 1, 97, 37}), we first randomly select 16 treatment group counties in 2008. Then, for year 2009, we randomly select 25 counties from the remaining non-HSR counties to become HSR counties, and so on. Using this false CRH_pass variable, a placebo DID estimation is conducted using the specification in column 5 of Table 4. Given the random data generation process, the false CRH_pass variable should have produced no significant estimate with a magnitude close to zero; otherwise, it would indicate a mis-specification of the DID estimation. To increase the identification power of this placebo test, it is repeated 500 times. Figure 3 shows the distribution of the estimates from the 500 runs along with the benchmark estimate, 0.025, from column 5 of Table 4. The distribution of estimates from random assignments is clearly centered around zero with a standard deviation of 0.0086, indicating that there is no effect with the randomly constructed HSR network. Meanwhile, the vertical line of the benchmark estimate is located at the tail of the distribution. Combined, these observations suggest that the positive and significant effect of the HSR on population growth is not driven by unobserved factors.

Figure 3.

Placebo test.

Event Study

To test the hypothesis that the treatment and control groups have the same time trend before the launch of HSR, we add the time indicator to Equation 2.

$$\begin{array}{c}\multicolumn{1}{c}{\log \left(\mathrm{populatio}{n}_{\mathrm{it}}\right)={\alpha }_i+\beta \sum _{k\ge -3}^{3+}\mathrm{CRH}\_\mathrm{pass}\times }\\ \multicolumn{1}{c}{1\{\mathrm{yea}{r}_{t-t0}=k\}+Z{\mathrm{\text{'}}}_{\mathrm{it}}\sigma +X\times f\left(t\right)\mathrm{\text{'}}\delta +{\gamma }_t+{ϵ}_{\mathrm{it}}}\end{array}$$ (5)

where $1\{\mathrm{yea}{r}_{t-t0}=k\}$ is the event time indicator and t₀ denotes the year when a county launches HSR. For example, if a county launches HSR in 2010, k = 0, while in 2009, k = −1. This method allows us to normalize the years of the sample counties to −3, −2, −1, 0, 1, 2, and 3 based on the launch time. Figure 4 plots the event study coefficients and 95% confidence intervals within a 6-year event window. The point estimates represent the time path of outcome variables, that is, population and mobility rate affected by HSR relative to non-HSR counties. Both graphs support the validity of the design because none of the coefficients are significantly different from zero prior to the launch of HSR in year 0, suggesting that HSR and non-HSR counties followed similar time trends at least 3 years before the launch of HSR. Meanwhile, right after the launch of HSR, the coefficients become statistically significant and gradually increase in magnitude. Taken together, these point estimates show that the difference in population growth between HSR and non-HSR counties begins to diverge with the launch of HSR.

Figure 4.

Event study.

IV Estimation

The main challenge for the empirical work is that infrastructure measures may be partly determined by some of the same unobservables that predict outcomes of interest. The balance test in section 3 shows that there are significant differences between HSR counties and non-HSR counties. In fact, the “Mid-to long-term Railway Network Plan” has explicitly mentioned 21 regional central cities in the setting of “four vertical and four horizontal” network. Faber (2014) constructs an instrumental variable by the least-cost path. We borrow this approach to conduct a robustness check on the previous results. Considering the 21 regional central cities identified in the “Mid-to long-term Railway Network Plan”, we construct a HSR network subject to global construction cost minimization that connects these cities. In the first step, we use the IGBP China land cover digital map and the 1 km DEM model dataset to build the cost raster map with following cost function.

$$\begin{array}{c}\multicolumn{1}{c}{\cos t_i=\mathrm{wate}{r}_i*25+\mathrm{buildin}{g}_i*25+\mathrm{slop}{e}_i\times 0.6}\\ \multicolumn{1}{c}{+\mathrm{Undulatio}{n}_i\times 0.4}\end{array}$$ (6)

where $\cos t_i$ denotes the cost of crossing a pixel of land i, $\mathrm{wate}{\mathrm{r}}_{\mathrm{i}}$ is a dummy indicator whether i is covered by water, $\mathrm{buildin}{g}_i$ denotes whether i is a built-up area, $\mathrm{slop}{e}_i$ is i’s average slope gradient and $\mathrm{Undulatio}{n}_i$ denotes i’s undulation degree. In the second step, we construct 210 pairs ((21 × 20)/2) of weighted minimum cost paths between all the 21 regional central cities, using Dijkstra’s algorithm and the cost map generated in the first step. In the third step, the minimum spanning tree network connecting the 21 cities is constructed according to Kruskal’s algorithm. We additionally construct a minimum spanning tree connecting each city with a straight line (Figure 5).

Figure 5.

Minimum spanning tree network.

The two hypothetical minimum spanning tree HSR network as instruments for actual route placements are only related to the geographical cost, so that the instrumental variables are generated without being influenced by the subjective intentions of the planners. Also, this hypothetical network does not exist in reality and therefore does not affect people migration choices. Table 5 presents the first stage results for the minimum spanning tree network instruments as well as their combined first stage results. If a city is connected by the hypothetical network, we assign 1 to the binary connection indicator. Both networks are strongly significant predictors of actual HSR placements conditional on the full set of socio-economic county characteristics. The second stage results are reported in Table 6. The Cragg-Donald Wald F statistics and Hansen J statistics show no weak IV and no over-identification problems. As shown in columns 1 to 3, the launch of HSR increases the population by about 2% to 3%. The results in columns 4 to 6 show that the estimated results are positive at 10% level when we use the mobility rate as the dependent variable.

Table 5.

First Stage Regressions.

	CRH_pass
Dependent variable	(1)	(2)	(3)
Straight line	0.668*** (0.022)		0.346*** (0.040)
Minimum spanning tree		0.686*** (0.021)	0.447*** (0.039)
Log(GDP)	−0.010 (0.021)	−0.001 (0.019)	−0.006 (0.019)
Log(distance to highway)	0.026*** (0.005)	0.029*** (0.005)	0.025*** (0.00)
Log(rail passenger traffic)	0.025** (0.010)	0.040*** (0.010)	0.031*** (0.010)
Log(road area)	−0.002 (0.011)	0.011 (0.012)	0.001 (0.011)
Other control variables	YES	YES	YES
Constant	YES	YES	YES
Observations	16,076	16,076	16,076
R-squared	.778	.765	.794

Note. The variable “Straight line” is the binary connection indicator generated by the linear minimum spanning tree. The variable “Minimum spanning tree” is the binary connection indicator generated by the minimum spanning tree. The robust standard errors in the brackets are clustered at the county level.

***

p < .01. **p < .05. *p < .1.

Table 6.

Second Stage Regressions.

	Log(population)			Mobility_rate
	MST IV	Line IV	Both IV	MST IV	Line IV	Both IV
Dependent variable	(1)	(2)	(3)	(4)	(5)	(6)
CRH_pass	0.030** (0.014)	0.026*** (0.010)	0.025* (0.014)	0.022* (0.013)	0.022* (0.012)	0.020* (0.011)
Control variables	YES	YES	YES	YES	YES	YES
Constant	YES	YES	YES	YES	YES	YES
Cragg-Donald Wald F.	97.222	69.497	53.297	97.222	69.497	53.297
Hansen J. P value			0.692			0.608
Observations	11,297	11,297	11,297	11,297	11,297	11,297
R-squared	.990	.990	.990	.747	.748	.747

Note. The robust standard errors in the brackets are clustered at the county level.

***

p < .01. **p < .05. *p < .1.

Heterogeneity Analysis

The launch of HSR shortens the spatial-temporal distances between counties, and as shown in above results, the population in general clusters toward counties along the HSR lines. With the policy of regional balance in recent years, does the decrease in travel cost lead to differences between regions? To test the heterogeneity in different regions, this paper divides the sample into three parts (the eastern, the western and the rest regions). Columns 1 to 3 of Table 7 show that counties in the eastern region has a greater population inflow than that of western region influenced by the launch of HSR. In particular, column 1 shows that the launch of HSR will lead to a 6.5% increase in the resident population relative to the reference group, while it is not significant in the western region. The results in column 3 shows that the launch of HSR leads to a relative decrease of 10.8% in the scale of primary school students in the western counties. Column 7 reports the IV results which is consistent with the previous regressions. Combined, these observations suggest that the launch of HSR facilitates cross-regional population movement, clustering from the western to the eastern region.

Table 7.

Heterogeneity Analysis by Region.

	Log(population)	Mobility_rate	Log (primary student)		Log (population)		Log (population) MST IV
Dependent variable	(1)	(2)	(3)	(4)	(5)	(6)	(7)
CRH_pass	−0.004 (0.012)	0.010 (0.010)	−0.025 (0.016)	0.001 (0.012)	−0.004 (0.014)	−0.001 (0.013)	0.049*** (0.018)
East_dum*CRH_pass	0.065*** (0.021)	0.034* (0.017)	0.108*** (0.022)				0.061*** (0.019)
West_dum*CRH_pass	−0.003 (0.030)	−0.004 (0.029)	−0.066** (0.029)				−0.006 (0.028)
Northeast_dum*CRH_pass	0.026 (0.029)	0.030 (0.038)	−0.004 (0.025)
Verti_dum*CRH_pass				0.043** (0.017)
South_dum × CRH_pass					0.029* (0.018)
Coastal_dum × CRH_pass						0.048*** (0.019)
Control variables	YES	YES	YES	YES	YES	YES	YES
Constant	YES	YES	YES	YES	YES	YES	YES
Observations	11,297	11,297	16,061	11,297	11,297	11,297	11,297
R-squared	0.990	0.748	0.977	0.990	0.990	0.990	0.990

Note. The robust standard errors in the brackets are clustered at the county level.

***

p < .01. **p < .05. *p < .1.

A large number of studies focus on east-west differences in China, and less attention has been paid to the north-south differences. The “four vertical and four horizontal” HSR network connects the less developed western region with the developed eastern region, as well as the north and the south. For completeness, we compare the impact of the “four vertical” routes with the “four horizontal” routes (Figure 6). The results in Table 7 indicate that the vertical routes have a greater effect on population agglomeration than the horizontal ones. The results in Column 4 shows that the population of HSR counties along the vertical routes increases by 4.3% compared to the horizontal lines at 5% level. The result in column 5 of Table 7 shows that the population of HSR counties in the southern region is 2.9% greater than that of the northern region, indicating the movement of population from the north to the south influenced by HSR. In addition, the result in column 6 indicates that the launch of HSR promotes population migration from hinterland to coastal areas.

Figure 6.

The “four vertical” routes and “four horizontal” routes (2014).

Past studies suggest that large cities are more attractive than small and medium-sized cities. Can HSR widen this difference in each region? As shown in Table 8, the counties in the eastern central cities (prefecture level) influenced by HSR do not attract as much population as those small and medium-sized periphery counties (in non-central cities), indicating that HSR promotes the inflow of population to counties in small and medium-sized cities and weakens the attractiveness of counties in large cities. Column 1 of Table 8 shows that the population of counties in small and medium-sized cities in the eastern region increases by 6.7% with the launch of HSR relative to other counties, while counties in the central cities do not show a significant increase. The IV estimation shows the same result (column 5). As for the central region, we do not see the same trend. On the contrary, in the relatively less developed central region, the launch of HSR promotes population agglomeration in regional central cities. Column 3 of Table 8 shows that the launch of HSR increases the population of counties in central cities by 6.9%, while the population of counties in small and medium-sized cities has decreased by 5.8%. The result in column 4 shows that the launch of HSR increases population mobility rate of counties in central cities by 6.3%, while the inflow of population of counties in small and medium-sized cities also shows an opposite trend. Combined, the above results indicate that the improvement of transportation in developed areas can benefit small and medium-sized cities, whereas it may weaken the agglomeration power of small and medium-sized cities in underdeveloped area.

Table 8.

Heterogeneity Analysis by Hierarchy.

	Log(population)	Mobility_rate	Log(population)	Mobility_rate	Log(population) MST IV
Dependent variable	(1)	(2)	(3)	(4)	(5)
CRH_pass	−0.002 (0.012)	0.011 (0.011)	0.045*** (0.015)	0.041** (0.016)	0.003 (0.018)
Central city × east_dum × CRH_pass	0.014 (0.063)	−0.092 (0.010)			0.011 (0.062)
Non-central city × east_dum × CRH_pass	0.067*** (0.023)	0.043** (0.019)			0.064*** (0.022)
Central city × cent_dum × CRH_pass			0.069 (0.055)	0.063* (0.034)
Non-central city × cent_dum × CRH_pass			−0.058*** (0.020)	−0.037* (0.021)
Control variables	YES	YES	YES	YES	YES
Constant	YES	YES	YES	YES	YES
Observations	11,297	11,297	11,297	11,297	11,297
R-squared	.990	.754	.990	.740	.990

Note. The robust standard errors in the brackets are clustered at the county level.

***

p < .01. **p < .05. *p < .1.

The pattern of people who chose to migrate is also an issue of concern to us. As mentioned above, the trend of family-oriented mobility becomes more pronounced. Does this mean that the launch of HSR facilitate for the labor force population to move together with their families, or just the labor force population itself? To test this, we run regressions using data from census 2000 and 2010 grouped by age, that is, 0 to 14 years for the children, 15 to 64 years for the labor force and 65+ years for the old. The results in Table 9 partially verify our conjecture that the HSR may have promoted labor force movements together with their children (columns 1 and 2). However, we do not find it works for the old as shown in column 3. This is likely due to the fact that older people may have established deep roots in their communities and have health concerns that make it difficult to travel or adapt to new environments and new culture.

Table 9.

Heterogeneity Analysis by Age.

	Log(population)
	Age 0-14	Age 15-64	Age 65+
Dependent variable	(1)	(2)	(3)
CRH_pass	0.081* (0.046)	0.075** (0.030)	−0.035 (0.029)
Control variables	YES	YES	YES
Constant	YES	YES	YES
Observations	2,720	2,720	2,660
R-squared	.923	.973	.982

Note. The robust standard errors in the brackets are clustered at the county level.

***

p < .01. **p < .05. *p < .1.

The above findings suggest that the population movement in China under the influence of HSR is a family-based movement driven by the labor force population. With the labor force accounting for a large share of the population, this mobility pattern also has important implications for social development. On the one hand, the 2020 China Census shows that the mobile population, mainly the labor force, is gradually moving to developed areas and metropolitan regions centered on large cities. On the other hand, according to Lin (2017), improvements in transport infrastructure can change the employment structure of a city. The changes in the spatial pattern brought about by HSR, and in particular the structural impact on the workforce in large cities, should therefore be of concern to us. For this purpose, we integrate the county-level data and combine it with employment data from prefecture-level cities. Table 10 shows the results on employment across sectors in large cities relative to other cities. It is evident from the interaction terms of column 2 and 3 that the employment in high value-added sectors such as finance and business services has grown significantly in large cities due to the launch of HSR, while manufacturing and residential services have seen a relative decline in employment (column 1 and 4). These findings have also echoed Feng et al. (2023) that HSR connection may promote the agglomeration of high-skilled labor in large cities. Besides, column 5 shows that there is no evidence that HSR has an impact on technical workers in large cities.

Table 10.

Heterogeneity Analysis by Employment.

	Log(employment)
	Manufacture	Finance	Commerce	Residence	Technology
Dependent variable	(1)	(2)	(3)	(4)	(5)
CRH_pass	0.080** (0.033)	−0.009 (0.070)	−0.063 (0.120)	0.263** (0.102)	−0.053 (0.043)
Central city × CRH_pass	−0.0886* (0.046)	0.313** (0.156)	0.536*** (0.196)	−0.424*** (0.147)	−0.028 (0.055)
Control variables	YES	YES	YES	YES	YES
Constant	YES	YES	YES	YES	YES
Observations	1,864	1,294	1,101	1,613	1,420
R-squared	.982	.989	.974	.977	.994

Note. Employment data from the upper table are a panel of Chinese prefecture cities. Control variables at prefecture level are included in all regressions. The robust standard errors in the brackets are clustered at the prefecture level.

***

p < .01. **p < .05. *p < .1.

Interpretation From Market Access

The launch of HSR promotes population movement and spatial agglomeration. Figure 7 shows that the 3-hr commuting buffer of Wuhan in 2007 is just limited to Hubei Province, while it reaches the Yangtze River Delta region in 2014 with the construction of HSR. The reduction in spatial-temporal distance has led to new options for enterprise location and labor employment. On the one hand, the reduced travel cost enhances location advantage, and enterprises and labor force are willing to choose large cities that can obtain more agglomeration benefits. On the other hand, the higher cost of large cities makes enterprises and households prefer to locate in the surrounding second-tier cities.

Figure 7.

Commuting buffer of Wuhan (2007, 2014).

Hence, HSR affects resource allocation by changing market access. Donaldson and Hornbeck (2016) quantify the global impact brought about by rail using the market access indicator, which gives us a good reference for analyzing the change in accessibility and spatial disparity caused by HSR in China. We set the model as follows.

$$\begin{array}{c}\multicolumn{1}{c}{\log \left(\mathrm{populatio}{n}_{\mathrm{it}}\right)={\alpha }_i+\beta \log \left(M{A}_{\mathrm{it}}\right)}\\ \multicolumn{1}{c}{+Z{\mathrm{\text{'}}}_{\mathrm{it}}\sigma +X\times f\left(t\right)\mathrm{\text{'}}\delta +{\gamma }_t+{ϵ}_{\mathrm{it}}}\end{array}$$ (7)

where $M{A}_{\mathrm{it}}$ denotes the market access of county i in year t which affects its population. Meanwhile, HSR affects market access of a county.

$$\log \left(M{A}_{\mathrm{it}}\right)=\beta \mathrm{CRH}\_\mathrm{pas}{s}_{\mathrm{it}}+Z{\mathrm{\text{'}}}_{\mathrm{it}}\sigma +(X\times f\left(t\right))\mathrm{\text{'}}\delta +{\gamma }_t+{ϵ}_{\mathrm{it}}$$ (8)

where the market access is as follows.

$$\underset{d=1}{\overset{n}{M{A}_o=\sum }}{c}_{\mathrm{od}}^{-\theta }{P}_d$$ (9)

where $c_{\mathrm{od}}^{-\theta }$ denotes the minimum travel cost from county o to county d, θ is the trade elasticity, $P_d$ is the market size of county d, and n is the number of counties.

The estimation results are reported in Table 11. We first use 3.8 as the trade elasticity θ referring to Donaldson and Hornbeck (2016) with robustness tests using 1 and 2.5. The market size indicator is measured by the population of the city. Columns 1 to 3 show that the launch of HSR has a significant impact on market access with different settings of θ. Columns 4 to 6 replace the binary CRH_pass indicators with the log distance to the nearest HSR segment. Specifically, column 4 shows that a 1% decrease in the distance to HSR will lead to a 5.5% increase in market access at 1% level. Columns 5 and 6 show the same results. The result in column 7 indicates that a 1% increase in market access will lead to a 2.6% increase in the resident population. The above results suggest that the launch of HSR improves the market access of a county, and further causes population concentration. To further verify the robustness of the estimation results, we replace the market size parameter with GDP and the results in Table 12 are in line with above findings.

Table 11.

HSR and Market Access (Market Size-Population).

	Log (market access)						Log (population)
	θ = 3.8	θ = 1	θ = 2.5	θ = 3.8	θ = 1	θ = 2.5	θ = 3.8
Dependent variable	(1)	(2)	(3)	(4)	(5)	(6)	(7)
CRH_pass	0.583*** (0.040)	0.148*** (0.006)	0.437*** (0.022)
Log (dist_to_HSR)				−0.055*** (0.005)	−0.032*** (0.002)	−0.065*** (0.004)
Log (market access)							0.026*** (0.010)
Control variables	YES	YES	YES	YES	YES	YES	YES
Constant	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	16,064	16,064	16,064	12,876	12,876	12,876	11,294
R-squared	.986	.973	.984	.996	.978	.991	.753

Note. The robust standard errors in the brackets are clustered at the county level.

***

p < .01. **p < .05. *p < .1.

Table 12.

HSR and Market Access (Market Size-GDP).

	Log (market access)
	θ = 3.8	θ = 1	θ = 2.5	θ = 3.8	θ = 1	θ = 2.5
Dependent variable	(1)	(2)	(3)	(4)	(5)	(6)
CRH_pass	0.698*** (0.048)	0.499*** (0.028)	0.170*** (0.007)
Log (dist_to_HSR)				−0.074*** (0.007)	−0.038*** (0.002)	−0.081*** (0.006)
Control variables	YES	YES	YES	YES	YES	YES
Constant	Yes	Yes	Yes	Yes	Yes	Yes
Observations	16,064	16,064	16,064	12,876	12,876	12,876
R-squared	.985	.983	.988	.995	.989	.991

Note. The robust standard errors in the brackets are clustered at the county level.

***

p < .01. **p < .05. *p < .1.

Finally, we explore the reason for regional disparity influenced by HSR. Columns 1 to 3 of Table 13 show that HSR exerts opposite effects on counties in the east and west. In particular, column 2 shows that the launch of HSR increases market access of the eastern counties by 5.3% relative to the others at the 1% level, while it decreases market access of the western counties by 7.9% relatively, which is in line with the cross-regional movement of population from west to east. Columns 4 to 6 show that the launch of HSR increases market access of southern counties more than that of northern counties, which is the reason for the population movement from north to south as previously proposed. Column 8 shows that the launch of HSR can increase market access of counties in eastern small and medium-sized cities by 5.5% at 1% level, while the coefficient of eastern central cities is not statistically significant. This finding explains why the launch of HSR can promote population concentration of counties in eastern small and medium-sized cities, while the population inflow in central cities tends to slow down. Zheng and Kahn (2013) find that HSR improves market access, provide more possibilities for employment and investment, and can significantly increase the value of second- and third-tier cities through industrial transfer and intra-regional market integration. Hence, with the improvement of transportation infrastructure in developed region, the spatial-temporal distance between small cities and large cities becomes short, significantly enhancing their market access. At the same time, as large cities face the problem about growing too big and household register constraint, they bring more opportunities to periphery small and medium-sized cities, leading to population flows to small and medium-sized cities in developed regions. In addition, as shown in Figure 1, the 2-hr commuting buffer of HSR network basically covers periphery counties in the eastern region, but this is not the case in the central region, which is an important reason for the regional disparity between the two regions.

Table 13.

Market Access and Regional Disparity.

	Log (market access)
	θ = 3.8	θ = 1	θ = 2.5	θ = 3.8	θ = 1	θ = 2.5	θ = 3.8	θ = 1	θ = 2.5
Dependent variable	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
CRH_pass	0.534*** (0.059)	0.124*** (0.001)	0.405*** (0.033)	0.441*** (0.045)	0.135*** (0.006)	0.352*** (0.024)	0.546*** (0.050)	0.124*** (0.008)	0.398*** (0.028)
East_dum × CRH_pass	0.096 (0.085)	0.053*** (0.012)	0.080* (0.047)
West_dum × CRH_pass	−0.082 (0.150)	−0.079*** (0.015)	−0.130 (0.080)
Northeast_dum × CRH_pass	0.117 (0.132)	0.067*** (0.015)	0.061 (0.068)
South_dum × CRH_pass				0.288*** (0.077)	0.027** (0.011)	0.173*** (0.042)
Central city × east_dum × CRH_pass							0.444 (0.366)	−0.000 (0.031)	0.212 (0.153)
Non-central city × east_dum × CRH_pass							0.071 (0.079)	0.055*** (0.011)	0.083* (0.043)
Control variables	YES	YES	YES	YES	YES	YES	YES	YES	YES
Constant	YES	YES	YES	YES	YES	YES	YES	YES	YES
Observations	16,064	16,064	16,064	16,064	16,064	16,064	16,064	16,064	16,064
R-squared	.987	.974	.984	.987	.974	.984	.987	.974	.984

Note. The robust standard errors in the brackets are clustered at the county level.

***

p < .01. **p < .05. *p < .1.

Conclusion

Transportation infrastructure is often used as an important tool for a country to promote growth and balance regional development, yet it is often difficult to estimate their effects due to the lack of data. This paper exploits a quasi-natural experiment in the context of the largest developing country, China, to examine the effect of high-speed railway on population mobility and its related impacts. The results suggest that the launch of HSR has facilitated population mobility and spatial disparity by improving market access. Labor-led family migration has become the main form of movement under the effect of HSR, but this has also given rise to the social problem of the elderly people left behind. By analyzing the heterogeneity of cities, we have also found that the development of HSR has promoted the concentration of high-skilled labor and the outflow of low-skilled labor in large cities.

The above findings can give us some insights. First, the development of HSR may increase regional imbalances and we need to focus on the less developed regions that are affected. Second, the government needs to pay attention to the elderly left behind and the children of the migrant population who have moved with them, such as increasing funding for social programs to elderly individuals and providing access to education to help the children overcome language barriers and cultural differences. Finally, regarding the outflow of low-skilled labor, the government can take some measures such as investing in job training and education, increasing access to affordable housing, and developing regional economic development strategies to distribute job opportunities more evenly across the region. However, it is not clear whether this change in the structure of the workforce facilitated by HSR will be beneficial or detrimental to urban development, and this will be an important concern for the future.