5.1 Baseline results
In Table 3, we present the findings of our locally linear estimation of equation (1) using various bandwidths (50 km, 75 km, 100 km). Our results demonstrate that the TF has enduring effects on the TF regions, hindering the development of public services in these regions. These outcomes are all statistically significant at either the 1% or 5% level. Specifically, our estimates in Table 3, column 1, using a 50 km bandwidth, indicate that the number of public services (middle schools, primary schools, hospitals, health centres) in the TF regions declined by 6.1%, 10.3%, 9.8%, and 7.7%, respectively. This suggests that the TF investments made in the past have had a negative impact on the long-term provision of public services in these regions. Such impacts could be attributed to government-led policies that distorted resource allocation, resulting in a lack of incentives and inefficiencies. Table 3, columns 2 and column 3 demonstrate the results of our bandwidth-adjusted estimates, where we expand the bandwidth by a factor of 1.5 and 2, respectively. We observe that the estimates remain stable and significant.
Table3 Baseline RD results
Sample within
|
<50km
(1)
|
<75km
(2)
|
<100km
(3)
|
Dependent variable
|
Middle
School
|
Primary
\School
|
Hospital
|
Health centre
|
Middle
School
|
Primary
School
|
Hospital
|
Health centre
|
Middle
School
|
Primary
School
|
Hospital
|
Health centre
|
West
|
-0.061**
(0.027)
|
-0.103***
(0.038)
|
-0.098**
(0.049)
|
-0.077*
(0.041)
|
-0.061**
(0.023)
|
-0.111***
(0.034)
|
-0.104**
(0.045)
|
-0.080**
(0.0)
|
-0.056**
(0.023)
|
-0.117***
(0.033)
|
-0.099**
(0.042)
|
-0.082***
(0.028)
|
Observations
|
2886
|
2886
|
2886
|
2886
|
4019
|
4019
|
4019
|
4019
|
4918
|
4918
|
4918
|
4918
|
R-squared
|
0.007
|
0.080
|
0.077
|
0.057
|
0.006
|
0.071
|
0.053
|
0.048
|
0.003
|
0.007
|
0.078
|
0.044
|
latitude and longitude control
|
YES
|
YES
|
YES
|
YES
|
YES
|
YES
|
YES
|
YES
|
YES
|
YES
|
YES
|
YES
|
Distance to boundary control
|
YES
|
YES
|
YES
|
YES
|
YES
|
YES
|
YES
|
YES
|
YES
|
YES
|
YES
|
YES
|
Distance to coastline control
|
YES
|
YES
|
YES
|
YES
|
YES
|
YES
|
YES
|
YES
|
YES
|
YES
|
YES
|
YES
|
Notes: *, **, and *** indicate statistical significance at 10%, 5%, and 1% levels, respectively; Robust standard errors are clustered at the county level.
Table 4 presents the estimated coefficients controlling for latitude, longitude, distance to the coastline, distance to the Beijing-Guangzhou Railway, and their quadratic polynomials. The results show a consistent negative relationship between the TF and public services over time, which is in line with our previous findings. Moreover, our bandwidth-adjusted estimates demonstrate this relationship with stable and significant estimates. Figure 3 illustrates the relationship between distance from the TF boundary and the number of public services, indicating a significant jump in the level of public services between the eastern and western regions near the boundary line. This observation highlights that the TF has enduring consequences on public service provision and may be caused by inefficient resource allocation policies and a lack of incentives.
Table 4 Baseline RD result
Sample within
|
<50km
(1)
|
<75km
(2)
|
<100km
(3)
|
Dependent variable
|
Middle
School
|
Primary
School
|
Hospital
|
Health centre
|
Middle
School
|
Primary
School
|
Hospital
|
Health centre
|
Middle
School
|
Primary
School
|
Hospital
|
Health centre
|
Control latitude and longitude
|
West
|
-0.058**
(0.024)
|
-0.111***
(0.028)
|
-0.087**
(0.039)
|
-0.084***
(0.031)
|
-0.059***
(0.021)
|
-0.113***
(0.025)
|
-0.094***
(0.034)
|
-0.085***
(0.027)
|
-0.055***
(0.020)
|
-0.118***
(0.024)
|
-0.097***
(0.032)
|
-0.083***
(0.026)
|
Observations
|
2886
|
2886
|
2886
|
2886
|
4019
|
4019
|
4019
|
4019
|
4918
|
4918
|
4918
|
4918
|
R-squared
|
0.005
|
0.063
|
0.019
|
0.045
|
0.004
|
0.062
|
0.022
|
0.039
|
0.005
|
0.071
|
0.027
|
0.039
|
Control the distance to the Coastline
|
West
|
-0.051**
(0.023)
|
-0.107***
(0.027)
|
-0.063***
(0.038)
|
-0.091***
(0.030)
|
-0.058***
(0.019)
|
-0.125***
(0.024)
|
-0.088***
(0.032)
|
-0.098***
(0.026)
|
-0.062***
(0.018)
|
-0.150***
(0.022)
|
-0.100***
(0.030)
|
-0.108***
(0.024)
|
Observations
|
2886
|
2886
|
2886
|
2886
|
4019
|
4019
|
4019
|
4019
|
4918
|
4918
|
4918
|
4918
|
R-squared
|
0.002
|
0.051
|
0.012
|
0.013
|
0.003
|
0.042
|
0.006
|
0.010
|
0.003
|
0.043
|
0.006
|
0.009
|
Control the distance to the boundary
|
West
|
-0.053**
(0.022)
|
-0.160***
(0.027)
|
-0.119***
(0.036)
|
-0.110***
(0.030)
|
-0.050***
(0.019)
|
-0.183***
(0.023)
|
-0.134***
(0.031)
|
-0.114***
(0.025)
|
-0.046***
(0.017)
|
-0.218***
(0.021)
|
-0.136***
(0.028)
|
-0.128***
(0.023)
|
Observations
|
2886
|
2886
|
2886
|
2886
|
4019
|
4019
|
4019
|
4019
|
4918
|
4918
|
4918
|
4918
|
R-squared
|
0.004
|
0.013
|
0.054
|
0.017
|
0.002
|
0.015
|
0.035
|
0.013
|
0.001
|
0.021
|
0.028
|
0.011
|
Control square polynomial of latitude and longitude
|
West
|
-0.056**
(0.024)
|
-0.089***
(0.028)
|
-0.089**
(0.039)
|
-0.088***
(0.031)
|
-0.054***
(0.021)
|
-0.092***
(0.025)
|
-0.094***
(0.035)
|
-0.092***
(0.028)
|
-0.045**
(0.020)
|
-0.090***
(0.024)
|
-0.085***
(0.032)
|
-0.075***
(0.026)
|
Observations
|
2886
|
2886
|
2886
|
2886
|
4019
|
4019
|
4019
|
4019
|
4019
|
4019
|
4019
|
4019
|
R-squared
|
0.010
|
0.078
|
0.028
|
0.049
|
0.016
|
0.074
|
0.026
|
0.049
|
0.020
|
0.084
|
0.030
|
0.046
|
Control square polynomial of the boundary coastline
|
West
|
-0.053**
(0.022)
|
-0.162***
(0.027)
|
-0.112***
(0.036)
|
-0.114***
(0.030)
|
-0.050***
(0.019)
|
-0.183***
(0.023)
|
-0.134***
(0.030)
|
-0.115***
(0.025)
|
-0.047***
(0.017)
|
-0.217***
(0.021)
|
-0.144***
(0.027)
|
-0.125***
(0.023)
|
Observations
|
2886
|
2886
|
2886
|
2886
|
4019
|
4019
|
4019
|
4019
|
4019
|
4019
|
4019
|
4019
|
R-squared
|
0.004
|
0.016
|
0.067
|
0.021
|
0.003
|
0.016
|
0.0051
|
0.017
|
0.002
|
0.044
|
0.015
|
0.003
|
Control square polynomial of the coastline
|
West
|
-0.051**
(0.023)
|
-0.103***
(0.027)
|
-0.063*
(0.038)
|
-0.088***
(0.030)
|
-0.058***
(0.019)
|
-0.118***
(0.024)
|
-0.089***
(0.032)
|
-0.094***
(0.026)
|
-0.061***
(0.018)
|
-0.143***
(0.022)
|
-0.100***
(0.030)
|
-0.104***
(0.024)
|
Observations
|
2886
|
2886
|
2886
|
2886
|
4019
|
4019
|
4019
|
4019
|
4019
|
4019
|
4019
|
4019
|
R-squared
|
0.002
|
0.061
|
0.012
|
0.018
|
0.003
|
0.049
|
0.006
|
0.011
|
0.003
|
0.049
|
0.006
|
0.011
|
Notes: *, **, and *** indicate statistical significance at 10%, 5%, and 1% levels, respectively; Robust standard errors are clustered at the county level.
Our study examined the dynamic impact of the TF on local public service provision; we extended our analysis further to investigate the dynamic impacts of the TF on regional public services. To measure the growth of local public services, we calculated the difference between the regional public service volumes in 2021 and 2012, and the results are presented in Table 5. We found that the TF leads to lower public service growth over time. This observation further underscores that the TF impact on regional public service supply behaviour persists and that it continues to shape the public service demand-supply dynamics of regional governments. Consequently, the gap between public service provision levels in the TF and non-TF regions continues to widen.
Table 5 The effect of the TF on public service growth
Sample within
|
<50km
|
Dependent variable
|
Middle School
|
Primary School
|
Hospital
|
Health centre
|
West
|
-0.128**
(0.051)
|
-0.200**
(0.093)
|
-0.473*
(0.279)
|
-0.666**
(0.030)
|
Observations
|
2886
|
2886
|
2886
|
2886
|
Notes: *, **, and *** indicate statistical significance at 10%, 5%, and 1% levels, respectively; Robust standard errors are clustered at the county level.
We conducted a placebo test by manually creating the boundary to validate the rationality of using the Beijing-Guangzhou Railway as a breakpoint for the TF, and we shifted the boundary line 30km to the east and west respectively. Table 6 reports the results of our regression using equation (1) on the sample near the leveled boundary. We found that the difference in the level of public services in 2021 between the areas on both sides of the boundary is not significant, with only the number of primary schools being statistically significant at the 10% level for each of the boundary shifts to the west, while the estimates of the other variables are not significant. The results of the placebo test also demonstrate that the Beijing-Guangzhou Railway is indeed an effective breakpoint in delineating the TF regions and non- TF regions.
Table 6 Placebo test
Dependent variable
|
Middle School
|
Primary School
|
Hospital
|
Health centre
|
Move Eastward
(1)
|
0.020
(0.027)
|
-0.053
(0.036)
|
-0.61
(0.048)
|
-0.034
(0.043)
|
Observations
|
2962
|
2962
|
2962
|
2962
|
R-squared
|
0.009
|
0.099
|
0.055
|
0.085
|
Move westward
(2)
|
-0.035
(0.026)
|
-0.041
(0.035)
|
-0.050
(0.049)
|
-0.080**
(0.041)
|
Observations
|
2782
|
2782
|
2782
|
2782
|
R-squared
|
0.008
|
0.050
|
0.040
|
0.035
|
Notes: *, **, and *** indicate statistical significance at 10%, 5%, and 1% levels, respectively; Robust standard errors are clustered at the county level.
A potentially serious problem when using geographical breakpoint regression to study problems is the spillover effect between regions. On the one hand, large industrial migration from policy-influenced areas in the TF provides employment opportunities that can attract areas near the border that are not affected by the policy to move to them, increasing the demand for public services. On the other hand, public services can spill over to neighbouring areas close to the border itself (Ferraresi et al., 2018). An increase in public services on one side of the border may benefit the population on the other side, thereby reducing the demand for public services. We use a spatial exclusion approach (Ehrlich and Seidel ,2018), which assumes that the opportunity for such spatial spillover effects decreases as the cost of transfer increases with distance. We present towns near the border at 5 km, 10 km, and 15 km, respectively, where it is generally assumed that the closer the town is to the border, the more likely it is to be affected by migration. Table 7 reports the results of the estimation excluding the codified towns, where the policy effects are enhanced compared to the baseline regression results and are all statistically significant at the 1% or 5% level. In addition, China's hukou system reduces the likelihood that people living near the border will cross it in order to access public services. Since its establishment in 1958, the hukou system has significantly curtailed spontaneous population migration, and planned state-organized migration has become the norm. Despite the relaxation of the hukou system in the 1990s, the probability of people relocating across the border remains low.
Table 7 Spillover effect
Dependent variable
|
Middle School
|
Primary School
|
Hospital
|
Health centre
|
5-50km
|
-0.053**
(0.025)
|
-0.138***
(0.030)
|
-0.085**
(0.040)
|
-0.089***
(0.033)
|
Observations
|
2511
|
2511
|
2511
|
2511
|
00R-squared
|
0.006
|
0.092
|
0.055
|
0.046
|
10-50km
|
-0.063**
(0.027)
|
-0.145***
(0.033)
|
-0.075*
(0.042)
|
-0.104***
(0.036)
|
Observations
|
2137
|
2137
|
2137
|
2137
|
R-squared
|
0.004
|
0.096
|
0.042
|
0.040
|
15-50km
|
-0.100***
(0.031)
|
-0.159***
(0.037)
|
-0.119**
(0.048)
|
-0.127***
(0.040)
|
Observations
|
1811
|
1811
|
1811
|
1811
|
R-squared
|
0.007
|
0.090
|
0.032
|
0.037
|
Notes: *, **, and *** indicate statistical significance at 10%, 5%, and 1% levels, respectively; Robust standard errors are clustered at the county level.
Table 8 presents the results of our robustness tests using various methods to explore the consistency of our estimates. Panel A shows the estimation results after incorporating the boundary fixed effects, where we divided the boundary into 5 segments of the same length, which produced indicator functions of 1 if grid i is closest to segment n and 0 if otherwise. This allowed us to compare samples from the same segment along the boundary. The results find that the estimates remain consistent and statistically significant with the baseline regression results. In Panel B, we exclude municipalities directly under the central government (Beijing and Tianjin) and find that the negative impact of the TF on public service provision over the long term remains significant. Furthermore, we exclude towns governed by provincial capitals (Shijiazhuang, Zhengzhou, Wuhan, Changsha, and Guangzhou) from the sample, which are often situated further from TF projects due to the policy's aim of accommodating war preparations. The results reported in Panel C demonstrate that the sample after removing these large cities exhibits a greater effect size on the number of primary and secondary schools compared to the baseline regression findings. In Panel D, we estimate our results by using a 10km*10km grid near the border that is not affected by administrative districts (Dell,2010). Our estimates are consistent with the baseline regression, highlighting the robustness of our findings to administrative boundary definitions. The next step was to consider the administrative districts; we used administrative units at the county levels in China. We then counted the number of public services in each county and district as the sample for the regression, with the outcomes reported in Panel E. We added the covariates (elevation,sun hour,pm2.5,population) from the balance test to Equation (1) on a county-level administrative district basis, and the results are reported in Panel F. The results obtained by the robustness test indicate that the level of provision of public services in the TF region is lower than that in the eastern region, which is consistent with the findings of previous studies.
Table 8 Robustness tests
Dependent variable
|
Middle School
|
Primary School
|
Hospital
|
Health centre
|
Panel A Control of boundary fixed effect
|
West
|
-0.051**
(0.023)
|
-0.089***
(0.028)
|
-0.082**
(0.037)
|
-0.066**
(0.030)
|
Observations
|
2886
|
2886
|
2886
|
2886
|
R-squared
|
0.023
|
0.096%
|
0.090
|
0.066
|
Panel B Exclusion of municipalities from the sample
|
West
|
-0.060**
(0.024)
|
-0.100***
(0.028)
|
-0.073***
(0.038)
|
-0.071**
(0.031)
|
Observations
|
2824
|
2824
|
2824
|
2824
|
R-squared
|
0.008
|
0.086
|
0.072
|
0.060
|
Panel C Exclusion of provincial capital from the sample
|
West
|
-0.070***
(0.026)
|
-0.097***
(0.033)
|
-0.048**
(0.042)
|
-0.081**
(0.035)
|
Observations
|
2121
|
2121
|
2121
|
2121
|
R-squared
|
0.008
|
0.148
|
0.079
|
0.102
|
Panel D Using 10km*10km grids sample
|
West
|
-0.224***
(0.040)
|
-0.344***
(0.046)
|
-0.321***
(0.062)
|
-0.122***
(0.042)
|
Observations
|
2151
|
2151
|
2151
|
2151
|
R-squared
|
0.153
|
0.231
|
0.184
|
0.1045
|
Panel E Using county sample
|
West
|
-0.165*
(0.087)
|
-0.223**
(0.102)
|
-0.312***
(0.102)
|
-0.123
(0.100)
|
Observations
|
233
|
233
|
233
|
233
|
R-squared
|
0.028
|
0.102
|
0.120
|
0.042
|
Panel F Controlling for covariates
|
West
|
-0.122*
(0.070)
|
-0.172**
(0.091)
|
-0.280***
(0.091)
|
-0.061
(0.087)
|
Observations
|
229
|
229
|
229
|
229
|
R-squared
|
0.401
|
0.312
|
0.330
|
0.315
|
Notes: *, **, and *** indicate statistical significance at 10%, 5%, and 1% levels, respectively; Robust standard errors are clustered at the county level.
Numerous studies indicate that transportation infrastructure construction can impact local economic activities (Redding and Turner, 2015; Berg et al., 2017). Given that our selected boundary shares the Beijing-Guangzhou Railway, an essential Chinese railway trunk line, we recognized that this may influence public service provision on both sides of the boundary and potentially skew estimation results. To mitigate the railway facility's impact on public services, we identified the Beijing-Kowloon Railway, an adjacent railway trunk line, as a proxy policy boundary. Utilizing equation (1), we estimated the level of public service provision within a 50km bandwidth on both sides of the boundary and presented our findings in Table 9. Remarkably, we discovered no significant difference in the level of public services between the two sides of the Beijing-Kowloon Railway, suggesting that railway facilities are not liable for the substantial discrepancy in public service provision between TF regions and non-TF regions.
Table 9 Robustness tests
Dependent variable
|
Middle School
|
Primary School
|
Hospital
|
Health centre
|
West
|
-7.13
(0.021)
|
-7.81
(0.026)
|
1.81
(0.025)
|
-3.59
(0.028)
|
Observations
|
3048
|
3048
|
3048
|
3048
|
R-squared
|
0.003
|
0.055
|
0.013
|
0.072
|
Notes: *, **, and *** indicate statistical significance at 10%, 5%, and 1% levels, respectively; Robust standard errors are clustered at the county level.
Last, it is important to acknowledge that our investigation into the long-term effects of the TF may be complicated by the Western Development Policy, a significant scheme implemented by the Chinese government to achieve regional balance in the 21st century. This initiative, which has been in operation for over two decades, covers 12 provinces and three single prefecture-level administrative regions in western China, including Enshi Prefecture in Hubei Province, Xiangxi Prefecture in Hunan Province, and Yanbian Prefecture in Jilin Province—areas that partially overlap with those involved in the TF. As a result, the accuracy of our estimation results may be affected. However, most of the provinces near the TF's eastern border that we selected for our sample were not influenced by the Western Development Policy. Only Enshi Prefecture in Hubei Province and Xiangxi Prefecture in Hunan Province were impacted, and we utilized the boundary for estimation in our selected sample. Our results show that the number of primary schools, middle schools, hospitals and health centres were not statistically significant. This indicates that how TF hindered the level of public service provision, which was reflected in our results, was not influenced by the Western Development Policy.
Table 10 The effect of WDP on public service
Dependent variable
|
Middle School
|
Primary School
|
Hospital
|
Health centre
|
West
|
0.045
(0.062)
|
0.024
(0.068)
|
0.150
(0.098)
|
0.128/
(0.078)
|
Observations
|
459
|
459
|
459
|
459
|
R-squared
|
0.049
|
0.060
|
0.042
|
0.050
|
Notes: *, **, and *** indicate statistical significance at 10%, 5%, and 1% levels, respectively; Robust standard errors are clustered at the county level.