RETRACTED: The Effects of Environmental Regulations on Medical Expenses: Evidence from China

Abstract

In this study, we examine the effects of a special period regulation (SPR), implemented in the Chang-Zhu-Tan (Changsha City, Zhuzhou City, and Xiangtan City; CZT) region, regarding medical expenses paid by the Urban and Rural Resident Basic Medical Insurance (URRBMI) and Urban Employee Basic Medical Insurance (UEBMI) programs, using a difference-in-differences (DID) design. We find that the SPR significantly reduces medical expenses in the CZT region, which appears to be driven by improved air quality. Furthermore, this regulation has a significantly negative and positive impact on medical expenses paid by the UURBMI and UEBMI, respectively. In summary, our results provide empirical evidence for the orderly implementation of command-and-control environmental regulation policies from the perspective of health benefits.

1. Introduction

Air pollution is gradually becoming a serious problem that threatens human physical and mental health [1,2,3,4,5,6]. Strengthening environmental regulations is the main strategy to deal with environmental degradation worldwide, especially in developing countries. It is crucial to comprehensively evaluate the costs and benefits of environmental regulations for policymakers. The benefits of environmental regulations arise mainly from the improvement in air quality, which will reduce medical expenses and defensive expenditures [7,8,9]. Because there is no direct evidence of the relationship between environmental regulations and health benefits, accurately evaluating the cost–benefit of environmental regulations is a key challenge. Fortunately, Hunan Province, located in southern China, provides an exogenous shock via which to study the impact of environmental regulations on medical expenses, which facilitates the direct assessment of the health benefits of environmental regulations.

The Hunan provincial government designed the SPR to prevent and control air pollution in the CZT region in 2015. With the rapid development of the economy, it is well known that air pollution has become an increasingly serious problem in the CZT region, especially in autumn and winter. Therefore, the special period was determined from October to February of the next year (5 months). During this period, it is necessary to strengthen law enforcement inspections and enable the joint prevention and control of air pollution to improve air quality. The SPR, a typical regional command-and-control environmental regulation, can significantly improve air quality [10]. However, whether it can bring health benefits and reduce medical costs remains to be determined.

The purpose of this study is to evaluate the impact of environmental regulations pertaining to air pollution on medical expenses in Hunan Province, China. Using the SPR as a quasi-natural experiment, we compare the changes in medical expenses before and after the regulations between counties in the CZT region and other regions with a DID approach. To implement the analysis, we obtained data from the Hunan Medical Insurance Bureau, which provides data on county-level medical expenses paid by the UURBMI and UEBMI programs from 2013 to 2018. We found that the SPR has a significantly negative impact on medical expenses. The DID estimates suggest that the regulation leads to 11.1% less medical expenses in CZT counties than in other counties. Notably, heterogeneity analyses show that the SPR has a significantly negative impact on medical expenses paid by the UURBMI and a significantly positive impact on medical expenses paid by the UEBMI. Various robustness and falsification checks were conducted, and they did not alter our conclusions. This paper further provides suggestive evidence that reducing pollution is an important driver behind declines in medical expenses.

This study makes three significant contributions to the literature. First, to the best of our knowledge, this paper is among the first to establish a comprehensive framework for evaluating the health benefits of environmental regulations from the perspective of medical expenses. Using the SPR as a quasi-natural experiment, we directly evaluate the financial health benefits of environmental regulations with a DID approach. Our results show that the implementation of the SPR significantly reduces medical expenses.

Second, our DID estimations suggest that regional environmental regulations significantly negatively affect medical expenses, complementing existing research mainly focusing on health outcomes (e.g., infant mortality, fetal health). In addition, the effectiveness of environmental regulations in mitigating these health effects has not been consistently confirmed in different empirical settings. This paper fills the gap and provides evidence from China, the largest developing country in the world.

Third, mechanism analysis provides suggestive empirical evidence that improving air quality is an essential driver behind reductions in medical expenses observed after the implementation of the SPR. Our results suggest that the SPR can improve air quality and further reduce medical expenses. These findings suggest that the benefits of the implementation of the SPR are enormous.

The remainder of this paper is organized as follows. The following section discusses related literature on the effects of environmental regulations and provides institutional background information for the SPR in Hunan Province. Section 3 describes the empirical specifications. Section 4 introduces major data sources and summary statistics. In Section 5, we present the study results, main empirical results, robustness checks, and analyses of the mechanisms. Finally, Section 6 provides conclusions.

2. Literature Review and Institutional Background

2.1. The Effects of Environmental Regulations

Over the past four decades, China’s unprecedented economic growth has lifted hundreds of millions of people out of poverty. However, because of the heavy reliance on fossil fuels and lax environmental regulations, pollution problems have intensified [11,12]. China’s air pollution level, which is 6–8 times that of the United States, has consistently exceeded the standards recommended by the World Health Organization (WHO) [13]. In response to environmental challenges, China has developed a multifaceted regulatory system that includes environmental laws, standards, and regulations. China’s environmental regulations mainly include three different types: command-and-control, market-based, and public participation and information provision regulations. These are composed of three stages: regulation legislation, planning process, and implementation mechanisms.

With the development of China’s environmental regulation system, a handful of studies have empirically estimated the causal effects of different types of environmental regulations on environmental outcomes by employing a DID approach. They found that strengthening environmental regulations is associated with improvements in environment quality [14,15,16,17,18,19]. The improvement of corporate environmental performance may be an important channel behind declines in air pollution [20].

Although effective environmental regulations can improve air quality, reducing pollution is accompanied by direct or indirect economic costs. To meet the target of environmental regulations, firms need to take measures to adjust production activities, install emission abatement facilities, and increase environmental protection investment [21,22]. There is a growing body of evidence indicating that stricter environmental regulations reduce exports [23,24], foreign direct investment [25], labor demand [26,27], and productivity [28]. In addition, environmental regulation may be a blessing, enhancing the positive effect of clean energy consumption on economic growth [29].

Furthermore, many studies focus on environmental policy and its impacts on health outcomes, including infant mortality [13,30,31,32], and fetal health [33]. Different types of environmental regulations lead to a reduction in infant mortality and the likelihoods of having negative fetal health in both developed and developing countries, which is mainly driven by reducing air and water pollution.

In summary, despite mounting research on the economic, environmental, and health effects of air and water pollution environmental regulations, knowledge is scarce on the influence of environmental regulations on medical expenses in developing counties, where medical expenditure increases year by year. This paper fills this gap and provides empirical evidence of the health benefits of regional air pollution regulation in China.

2.2. The SPR

The CZT region is the economic growth core of Hunan Province, which accounted for 60% of the provincial GDP in 2017 [34]. In 1997, the provincial party committee and provincial government put forward the Changsha-Zhuzhou-Xiangtan economic integration development strategy. The CZT city group has been approved as a “National Resource-economical and Environment-friendly Society Integrated Support Reform Experimental Zone” in 2007 [35]. With the rapid development of the economy, the ecological environment has been continuously deteriorating in the CZT region, which has become one of the key areas for regional air pollution joint prevention and control in China [36]. In order to solve the environmental problems, the Hunan provincial government approved a work plan for environmental cooperative governance in the CZT region from 2010 to 2020 and issued guidelines for promoting the joint prevention and control of air pollution in the CZT region in 2012. Since then, the joint prevention and control of air pollution has become an important measure to improve air quality in the CZT region.

The Hunan provincial government issued a work plan for the prevention and control of air pollution in the CZT region to reduce the frequency of heavily polluted weather in autumn and winter in 2015. The period from October to February of the following year (five months in total) was determined as the special period. During the special period, coal-fired, exhaust gas, and dust pollution are the key sources of air pollution. The CZT region must unify time, measures, and actions to form an overall synergy of regional air pollution management and control. As of 2019, the SPR has been implemented four times, from 1 October 2015 to 29 February 2016; from 1 October 2016 to 28 February 2017; from 16 October 2017 to 15 March 2018; and from 16 October 2018 to 15 March 2019. The core of the SPR is to improve air quality and the main goal of the SPR is to reduce pollutant emissions. The implementation of the SPR has been proved to be effective in reducing air pollution [10].

3. Empirical Strategy

3.1. DID Estimation

The main objective of this study is to assess the effect of air pollution regulations on medical expenses. Equation (1) is the baseline DID model to estimate the causal effect of the SPR on medical expenses:

$$Y_{ct}={\beta }_0+{\beta }_{1}Trea{t}_c\times Pos{t}_t+\theta C_{ct}+\gamma W_{ct}+{\alpha }_c+{\pi }_t+{\epsilon }_{ct}$$

(1)

where $Y_{ct}$ is the medical expenses paid by the UURBMI and UEBMI programs; $Trea{t}_c$ is an indicator variable that takes on the value 1 if county c is located in the CZT region; $Pos{t}_t$ is an indicator variable that takes on the value 1 for years after 2015; $C_{ct}$ denotes a vector set of county-level control variables; $W_{ct}$ are weather control variables, including the number of days with daily mean temperature falling into the kth bin of {$\ge$30 °C, 25–30 °C, 15–20 °C, 10–15 °C, 5–10 °C, 0–5 °C, $<$0 °C} and other weather characteristics (total precipitation amount, average relative humidity, average wind speed, sunshine duration, and atmospheric pressure); ${\alpha }_c$ are county fixed effects, which control the time-invariant city characteristics (e.g., geographical location); ${\pi }_t$ are year fixed effects, which control the year-specific shocks that are common to both CZT and non-CZT regions. All standard errors are clustered at the county level, allowing for an arbitrary correlation within the city over time. We also perform two-way clustering of the standard errors at the county-year and city-year levels as robustness checks. All dependent and control variables are in logarithmic form.

${\beta }_1$ is the central parameter; it captures the difference in the changes in medical expenses before and after the SPR in 2015, between the counties located in the CZT and non-CZT regions. If the environmental regulations contributed to significant reductions in medical expenses in the CZT region relative to the non-CZT regions, ${\beta }_1$ is expected to be negative.

3.2. Event Study

The key identification assumption for Equation (1) to provide causal effects is that the county-level districts and counties located in non-CZT regions provide valid counterfactual changes in medical expenses for the treatment group. We constructed the following event study model to test the parallel trend assumption:

$$Y_{ct}={{\displaystyle \sum }}_{k=-2}^{3}Trea{t}_c\times Yea{r}_k{\beta }_k+\theta C_{ct}+\gamma W_{ct}+{\alpha }_c+{\pi }_t+{\epsilon }_{ct}$$

(2)

In this equation, we replace the $Pos{t}_t$ in Equation (1) with one of six dummy variables that equals 1 depending on the year in which the county established the SPR. These variables are as follows: $Yea{r}_{-2}$ if established 2 years earlier, $Yea{r}_{-1}$ if established the previous year, $Yea{r}_0$ if established that year, $Yea{r}_1$ if established 1 year later, $Yea{r}_2$ if established 2 years later, $Yea{r}_3$ if established 3 years later. The $Yea{r}_0$ was omitted. The coefficients of the interaction terms should be insignificant before the implementation of the SPR, suggesting that there is no systematic difference in medical expenses between the CZT region and other regions.

4. Data Sources and Summary Statistics

4.1. Data Sources

We obtained county-level medical expense data from the Hunan Provincial Medical Insurance Bureau, which provides medical expenditures of all the individuals who joined the UURBMI and UEBMI in Hunan Province from 2013 to 2018. The UURBMI is a public health insurance scheme for urban and rural non-working residents, such as young children, adolescents, university students, seniors, and the disabled [37]. The UEBMI covers all current urban employees and retirees of past urban employees, including those in the public and private sectors [38]. The UURBMI and UEBMI are the largest public insurance schemes in China, and their total number of enrollees reached 1.345 billion in 2018, accounting for approximately 95% of China’s urban population.

The data for the weather control variables comes from the China Meteorological Data Sharing Service System, including daily average, maximum and minimum temperatures, total precipitation amount, average relative humidity, average wind speed, sunshine duration, and atmospheric pressure. To merge the weather data with county-level variables, we first calculate the daily weather variables for each county from 2013 to 2018, based on station-level weather records, following the inverse-distance weighting method. We then calculate the average value of the other meteorological characteristics for the past year. Finally, we calculate the number of days when the daily mean temperature fell into each of the eight 5 °C bins. Bin 1 comprises temperatures higher than 30 °C, Bin 2 comprises temperatures from 25 °C to 30 °C, …, and Bin 8 comprises temperatures less than 0 °C. Bin 3 (20–25 °C) was omitted.

Air pollution data is obtained from the Chinese air quality real-time publishing platform, published by the China Environmental Monitoring General Station. We aggregate this data by year and city to form a yearly-county panel consisting of 115 counties.

County-level economic data comes from the China County Statistical Yearbook from 2014 to 2019. Other control variables include the per capita regional gross domestic product; proportion of tertiary industry output value in total gross domestic product; population size; and numbers of health institutions, professional doctors, and registered nurses.

4.2. Summary Statistics

Table 1. Summary statistics.

Variable	Obs.	Mean	Std. Dev.	Min.	Max.
Panel A: core variables
Total medical expenses	690	27,460.840	19,081.130	1733.760	112,050.900
Medical expenses (UURBMI)	690	23,001.510	16,556.170	991.150	89,080.450
Medical expenses (UEBMI)	690	4459.332	3406.938	261.832	24,481.330
Treat	690	0.157	0.364	0.000	1.000
Post	690	0.500	0.500	0.000	1.000
Panel B: economic control variables
Per GDP	690	4.522	3.714	0.940	22.390
Tertiary industry	690	0.441	0.137	0.109	0.957
Population	690	56.970	28.299	6.000	135.130
Institutions	690	508.562	275.163	45.000	1339.000
Doctors	690	941.022	821.798	53.000	5661
Nurses	690	1253.874	1156.323	53.000	7901.000
Panel C: weather control variables
Number of days (AT ≥ 30 °C)	690	21.933	16.946	0.000	68.000
Number of days (AT 25–30 °C)	690	68.975	16.327	3.000	108.000
Number of days (AT 20–25 °C)	690	76.172	15.177	50.000	130.000
Number of days (AT 15–20 °C)	690	58.807	11.297	27.000	97.000
Number of days (AT 10–15 °C)	690	56.145	6.681	38.000	83.000
Number of days (AT 5–10 °C)	690	60.370	11.825	27.000	91.000
Number of days (AT 0–5 °C)	690	19.655	7.656	0.000	53.000
Number of days (AT < 0 °C)	690	3.080	4.470	0.000	34.000
Precipitation	690	4.046	0.680	2.444	6.739
Relative humidity	690	78.617	3.931	64.707	89.295
Wind speed	690	1.826	0.636	0.859	4.985
Sunshine duration	690	3.798	0.646	2.468	5.780
Air pressure	690	9887.207	217.101	8953.592	10,120.600
Panel D: air pollution variables
PM2.5	672	50.004	13.068	26.283	86.641
PM10	672	77.800	16.713	41.897	141.593
SO2	672	20.709	8.654	5.043	43.481

Note: This table presents the summary statistics for the sample used for our main regression analysis. In Panel A, “Total Medical Expenses” is the sum of “Medical expenses (UURBMI)” and “Medical expenses (UEBMI)”. In Panel B, “Per GDP” is calculated as county GDP divided by county population. “Tertiary industry” is the proportion of tertiary industry, calculated as the output value of tertiary industry divided by GDP. In Panel C, “AT” is an abbreviation for “average temperature”. In Panel D, the unit of PM_2.5 and PM₁₀ is μg/m³. The unit of SO₂ is thousand tons.

presents the summary statistics of the variables. The sample data covers 115 county-level districts and counties belonging to 14 prefecture-level cities in Hunan Province, accounting for 94% of all districts and counties. The treatment group includes 18 county-level districts and counties in the CZT region, accounting for 15.7% of all samples. The control group includes 97 county-level districts and counties in non-CZT regions.

The mean value of the total medical expenses paid by UURBMI and UEBMI is RMB 274.608 million. The mean value of medical expenses paid by the UURBMI is RMB 230.015 million, which is significantly five times higher than RMB 44.593 million paid by the UEBMI. The coverage of the UURBMI is wider. The elderly and infants have more precarious health statuses and have greater demand for medical services.

We then plot the variation trend of county-level medical expenses in the CZT region (treatment group) and other regions (control group) during the sample period. This is shown in Figure 1. The solid vertical line indicates the timing of the SPR’s implementation in 2015. The figure shows that the medical expense trends were similar before 2015. The medical expenses of the control group showed an overall upward trend, whereas those of the treatment group showed an upward trend before 2015. From 2013 to 2015, the medical expenses of the treatment group were higher than those of the control group and the gap continued to widen. However, a trend break appeared around 2015 when medical expenses began to drop only among the treatment group, whereas medical expenses remained somewhat constant within the control group. Although the medical expenses of the treatment group increased in 2017, the gap gradually narrowed. The control and treatment groups showed statistically similar pre-regulation trends in medical expenses. A similar pre-trend between the two sets of counties suggests that the post-trend would have been similar in the absence of the SPR. It should be noted that the descriptive conclusions in Figure 1 represent only correlations rather than causality.

The maximum annual mean of PM_2.5 is 10 μg/m³ and the maximum mean of PM_2.5 within a 24-h period is 20 μg/m³. These are the values recommended by the WHO. In the sample, the mean annual PM_2.5 level is 50.004 with a high of 86.641, which is much higher than the safe standard for human health.

5. Empirical Results

In this section, we first present the results of event study to test the parallel trend assumption. Then, we report the main results and PSM-DID estimations. Thereafter, we show that our results are consistent across a battery of robustness checks. Finally, we discuss the potential mechanisms.

5.1. Parallel Trend Assumption

The results of Equation (2) are reported in Figure 2, the dummy variables $Yea{r}_{-2}$ and $Yea{r}_{-1}$ allow us to assess whether any effect on medical expenses can be found before the implementation of the SPR. However, the coefficients of $Yea{r}_{-2}$ and $Yea{r}_{-1}$ are not significant. The estimated coefficients of the $Yea{r}_1$ and $Yea{r}_3$ dummies are negative and statistically significant, indicating that the SPR helps reducing the scale of medical expenses over the long term.

5.2. Main Results

We present the DID estimates of the regulation’s effect on medical expenses in

Table 2. Impact of the SPR on total medical expenses borne by the UURBMI and UEBMI.

	(1)	(2)	(3)	(4)	(5)
Treat × Post	−0.085 *	−0.099 **	−0.111 **	−0.111 **	−0.111 ***
	(0.051)	(0.047)	(0.055)	(0.056)	(0.040)
Per GDP		0.087	0.125	0.125	0.125
		(0.240)	(0.234)	(0.208)	(0.222)
Tertiary industry		−0.208	−0.196	−0.196	−0.196
		(0.172)	(0.179)	(0.152)	(0.173)
Population		0.617 *	0.669 **	0.669 **	0.669 *
		(0.362)	(0.328)	(0.315)	(0.343)
Institutions		−0.113 **	−0.122 **	−0.122 *	−0.122 *
		(0.051)	(0.052)	(0.071)	(0.067)
Doctors		0.063	0.098	0.098 *	0.098
		(0.069)	(0.071)	(0.054)	(0.075)
Nurses		−0.038	−0.029	−0.029	−0.029
		(0.049)	(0.049)	(0.054)	(0.077)
Number of days (AT ≥ 30 °C)			−0.002	−0.002	−0.002
			(0.002)	(0.002)	(0.002)
Number of days (AT 25–30 °C)			−0.000	−0.000	−0.000
			(0.002)	(0.002)	(0.002)
Number of days (AT 15–20 °C)			0.000	0.000	0.000
			(0.002)	(0.002)	(0.002)
Number of days (AT 10–15 °C)			−0.004	−0.004	−0.004
			(0.003)	(0.003)	(0.003)
Number of days (AT 5–10 °C)			0.000	0.000	0.000
			(0.003)	(0.003)	(0.003)
Number of days (AT 0–5 °C)			0.002	0.002	0.002
			(0.004)	(0.004)	(0.004)
Number of days (AT < 0 °C)			0.006	0.006	0.006
			(0.007)	(0.007)	(0.006)
Precipitation			0.002	0.002	0.002
			(0.013)	(0.013)	(0.016)
Relative humidity			0.001	0.001	0.001
			(0.004)	(0.004)	(0.005)
Wind speed			−0.021	−0.021	−0.021
			(0.041)	(0.039)	(0.034)
Sunshine duration			0.013	0.013	0.013
			(0.023)	(0.027)	(0.037)
Air pressure			0.000	0.000	0.000
			(0.000)	(0.000)	(0.000)
County FE	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes
Cluster	County	County	County	County-Year	City-Year
Observations	690	690	690	690	690
R2	0.664	0.671	0.677	0.962	0.962

Note: The sample period ranged from 2013 to 2018. Treat × Post is a dummy variable that = 1 for county-level districts and counties in the CZT region in the special regulation period (2015–2018). Robust standard errors clustered at the county level are in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01. The result is obtained by regression analysis using stata14.0 software.

. The dependent variable is the logarithm of medical expenses. Column 1 shows the result from Equation (1) with city and year fixed effects, indicating that the SPR is associated with reductions of 8.5% in medical expenses. Additionally, Column 2 controls for time-varying county-level economic variables. In Column 3, we further control for additional meteorological characteristics. Relative to non-CZT regions, the SPR leads to a decrease of 11.1% in medical expenses in the CZT region. The estimated effects are robust to the inclusion of these covariates and remain significant at the 10% level. In Columns 4 and 5, standard errors are clustered at the county-year level and city-year level, allowing for spatial and serial correlations within each county year and city year. Overall, the estimated impacts of the SPR are similar across the five specifications, although more stringent controls in the model lead to larger estimated coefficients for medical expenses. Because Column 3 includes the strictest set of controls, we use this specification to interpret the remaining results.

Economic development and medical facilities may also affect medical expenditure. Table 2 shows that population size has a significantly positive effect on the total medical expenses. However, the number of health institutions has a significantly negative effect on the total medical expenses. Per GDP, the industry structure and the number of professional doctors and registered nurses have no significant impacts on medical expenses. Due to endogeneity issues, the estimates of these control variables are biased.

Other meteorological variables may also affect physical and mental health, thereby leading to changes in medical expenses. Our results show that the coefficients of weather control variables are insignificant. There are two reasons: first, this may be due to insufficient annual variation in weather control variables. Second, our research direction hasn’t penetrated these compensation intervals (We thank an anonymous referee for this point.).

There are differences in the income level and reimbursement ratio of the residents insured under the UURBMI and UEBMI. They also have different willingness to pay and demand elasticity for medical services. Therefore, there is a large gap in medical expenses between the patients insured with the two different insurance schemes. The impacts of the SPR on medical expenses paid by UURBMI and UEBMI may have systematic differences. Columns 1 and 4 in

Table 3. Impact of the SPR on medical expenses: by insurance type.

	UURBMI			UEBMI
	(1)	(2)	(3)	(4)	(5)	(6)
Treat × Post	−0.217 ***	−0.217 **	−0.217 ***	0.126 ***	0.126 ***	0.126 ***
	(0.081)	(0.088)	(0.054)	(0.039)	(0.034)	(0.036)
Economic controls	Yes	Yes	Yes	Yes	Yes	Yes
Weather controls	Yes	Yes	Yes	Yes	Yes	Yes
County FE	Yes	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes	Yes
Cluster	County	County-Year	City-Year	County	County-Year	City-Year
Observations	690	690	690	690	690	690
R2	0.536	0.938	0.938	0.714	0.960	0.960

Note: The sample period ranged from 2013 to 2018. Treat × Post is a dummy variable that = 1 for county-level districts and counties in the CZT region in the special regulation period (2015–2018). Economic controls include per capita regional gross domestic product, the proportion of tertiary industry output value in total gross domestic product, the number of population, the number of health institutions, the number of professional doctors, and the number of registered nurses. Weather controls include the number of days with daily mean temperature falling into the kth bin of {≥30 °C, 25–30 °C, 15–20 °C, 10–15 °C, 5–10 °C, 0–5 °C, <0 °C}, total precipitation amount, average relative humidity, average wind speed, sunshine duration, and atmospheric pressure. Robust standard errors clustered at the county level are in parentheses. ** p < 0.05, *** p < 0.01. The result is obtained by regression analysis using stata14.0 software.

present these results. The results indicate that the core explanatory variable $\mathrm{Treat}\times \mathrm{Post}$ has significantly negative and positive effects on medical expenses covered under the UURBMI and UEBMI, respectively. After the implementation of the SPR, the medical costs borne by the UURBMI system decreased by 21.7% and the medical costs borne by the UEBMI system increased by 12.6%. We allow for two-way clustering of errors by county year in Columns 2 and 5. In Columns 3 and 6, standard errors are clustered at city-year level. All results are similar to baseline results in Columns 1 and 4.

Children and the elderly in rural and poor areas, who are covered by the UURBMI system, are more vulnerable to the effects of air pollution [39,40]. Effective environmental regulations improve air quality, thus, the morbidity and mortality from air pollution-related diseases will decrease significantly [30]. The health status of the elderly and children will also improve and the demand for medical services will decline. Hence, the medical costs borne by the UURBMI system will be reduced.

The UEBMI includes the urban working population, whose income, education level, and awareness of air pollution and its hazards are relatively higher. Consequently, they are more responsive to pollution information and more capable of taking preventive measures to avoid exposure to air pollution [41]. In addition, urban workers usually work indoors and have shorter air pollution exposure times. Therefore, improving air quality through environmental regulations might have an insignificant effect on their health and medical expenses. Nevertheless, the strengthening of environmental regulations may reduce the production activities, increase environmental protection investments, and reduce profits and labor demand [21,26,27,42], which are directly related to the economic and social welfare of employees. Fewer job opportunities and lower income have been widely proven to result in workers gaining weight and impairing their mental health [43,44]. Therefore, the demand for medical services for employees has increased and the medical expenses paid by the UEBMI have risen significantly.

Figure 3 plots the estimated results of Table 2 and Table 3.

5.3. PSM-DID Strategy

As described above, 18 county-level districts and counties in the CZT region are affected by the SPR policy, which allows us to estimate the causal effects of environmental regulations on medical expenses using a DID strategy. When the treatment is randomly set, or at least when the observable characteristics can be used to control for the treatment, then the DID estimation becomes most appropriate for capturing the effects of environmental regulation. However, the SPR does not easily satisfy the requirements of random experiments. The cities of the CZT region are geographically close to each other and this region comprises the core growth pole of economic development in Hunan Province. The assignment of the regulated region in our case is not random, relying on economic and political conditions. Therefore, we first use the Propensity Score Matching (PSM) approach to construct a comparable control group that is statistically similar to that of untreated counties. PSM uses logistic regression based on a set of initial variables in 2013 to obtain the propensity score of each city. Based on their propensity scores, cities are then matched by Nearest-Neighbor, Radius, and Kernal matching techniques (We thank an anonymous referee for this point.). Based on the matched control group, we then use a DID estimation to investigate how the SPR policy affected medical expenses in treatment group.

Table 4. Robustness checks: PSM-DID.

	Nearest-Neighbor			Rdius			Kernal
	(1) Total	(2) UURBMI	(3) UEBMI	(1) Total	(2) UURBMI	(3) UEBMI	(1) Total	(2) UURBMI	(3) UEBMI
Treat × Post	−0.125 **	−0.232 **	0.116 ***	−0.111 **	−0.217 ***	0.126 ***	−0.117 **	−0.223 ***	0.119 ***
	(0.060)	(0.089)	(0.040)	(0.055)	(0.081)	(0.039)	(0.057)	(0.084)	(0.039)
Economic controls	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Weather controls	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
County FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	680	680	680	690	690	690	685	685	685
R2	0.673	0.533	0.708	0.677	0.536	0.714	0.675	0.534	0.712

Note: The sample period ranged from 2013 to 2018. Treat × Post is a dummy variable that = 1 for county-level districts and counties in the CZT region in the special regulation period (2015–2018). Economic controls include: per capita regional gross domestic product, the proportion of tertiary industry output value in total gross domestic product, the number of population, the number of health institutions, the number of professional doctors, and the number of registered nurses. Weather controls include: the number of days with daily mean temperature falling into the kth bin of {≥30 °C, 25–30 °C, 15–20 °C, 10–15 °C, 5–10 °C, 0–5 °C, <0 °C}, total precipitation amount, average relative humidity, average wind speed, sunshine duration, and atmospheric pressure. Robust standard errors clustered at the county level are in parentheses. ** p < 0.05, *** p < 0.01.

provides the results of the PSM-DID approach. Compared to the baseline results in Table 2, the impacts of environmental regulations on medical expenses remain robust and the values of the coefficients are similar.

5.4. Robustness Checks

In this subsection, we show that our results are consistent across a battery of robustness checks: including alternative samples and placebo tests.

5.4.1. Alternative Samples

After the implementation of the SPR, the Hunan provincial government promulgated a work plan for the prevention and control of air pollution in the CZT region during the special period of 2018 to 2020 in 2018. In this work plan, it was noted that based on the CZT city cluster, Yueyang, Changde, and Yiyang would be added to the key areas for controlling air pollution.

On one hand, the study period in our paper is from 2013 to 2018. To avoid the possibility that the promulgation of the work plan in 2018 may interfere with the main results, we exclude the sample in 2018 and re-estimate Equation (1).

Table 5. Robustness checks: exclude other samples in 2018.

	(1) Total	(2) UURBMI	(3) UEBMI
Treat × Post	−0.134 *	−0.252 **	0.102 ***
	(0.071)	(0.111)	(0.034)
Economic controls	Yes	Yes	Yes
Weather controls	Yes	Yes	Yes
County FE	Yes	Yes	Yes
Year FE	Yes	Yes	Yes
Observations	575	575	575
R2	0.559	0.397	0.678

Note: The sample period ranged from 2013 to 2018. Treat × Post is a dummy variable that = 1 for county-level districts and counties in the CZT region in the special regulation period (2015–2018). Economic controls include: per capita regional gross domestic product, the proportion of tertiary industry output value in total gross domestic product, the number of population, the number of health institutions, the number of professional doctors, and the number of registered nurses. Weather controls include: the number of days with daily mean temperature falling into the kth bin of {≥30 °C, 25–30 °C, 15–20 °C, 10–15 °C, 5–10 °C, 0–5 °C, <0 °C}, total precipitation amount, average relative humidity, average wind speed, sunshine duration, and atmospheric pressure. Robust standard errors clustered at the county level are in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01.

presents the results of the study. Compared with the baseline results in Table 2 and Table 3, the values of coefficients and significance remain similar.

On the other hand, Yueyang, Changde, and Yiyang, three main transmission channel cities in Hunan Province, are located upwind of the CZT region. The pollutant emissions of these three cities are closely related to those of the CZT region, and their pollution prevention and control activities are important for coordinated air pollution control in the CZT region. Environmental protection has become a key factor in the promotion of local officials. After the implementation of the SPR policy in the CZT region, to respond to the call of the provincial government, Yueyang, Changde, and Yiyang, the second echelon of economic development in Hunan Province, could also actively take measures to strengthen pollution prevention and control. Hence, using Yueyang, Changde, and Yiyang as the control groups may underestimate the impact of the SPR policy. In

Table 6. Robustness checks: exclude other samples in Yueyang, Changde and Yiyang.

	(1) Total	(2) UURBMI	(3) UEBMI
Treat × Post	−0.131 **	−0.252 ***	0.145 ***
	(0.058)	(0.087)	(0.041)
Economic controls	Yes	Yes	Yes
Weather controls	Yes	Yes	Yes
County FE	Yes	Yes	Yes
Year FE	Yes	Yes	Yes
Observations	546	546	546
R2	0.736	0.593	0.700

Note: The sample period ranged from 2013 to 2018. Treat × Post is a dummy variable that = 1 for county-level districts and counties in the CZT region in the special regulation period (2015–2018). Economic controls include: per capita regional gross domestic product, the proportion of tertiary industry output value in total gross domestic product, the number of population, the number of health institutions, the number of professional doctors, and the number of registered nurses. Weather controls include: the number of days with daily mean temperature falling into the kth bin of {≥30 °C, 25–30 °C, 15–20 °C, 10–15 °C, 5–10 °C, 0–5 °C, <0 °C}, total precipitation amount, average relative humidity, average wind speed, sunshine duration, and atmospheric pressure. Robust standard errors clustered at the county level are in parentheses. ** p < 0.05, *** p < 0.01.

, we exclude the county-level sample under the jurisdiction of Hunan Province and re-estimate Equation (1). The absolute values of the coefficients are larger than the baseline results, suggesting that the benchmark results are underestimated. However, these results are generally consistent with our main conclusions.

5.4.2. Placebo Tests

Although the DID method can control for time-invariant unobserved variables, other confounding factors may have affected the causal effects. If there were some unobservable factors and neglected policies before 2015 that reduced medical expenses, the effect of the SPR on medical expenses may have been overestimated. Several placebo tests are conducted to check the robustness of the estimated results. We first assume that the SPR was enforced in 2014. Based on the above analysis, the SPR should not affect medical expenses during the fictitious treatment period. The results in

Table 7. Robustness checks: alternative treatment period.

	(1) Total	(2) UURBMI	(3) UEBMI
Treat × Post	0.006	−0.068	−0.046
	(0.047)	(0.066)	(0.154)
Economic controls	Yes	Yes	Yes
Weather controls	Yes	Yes	Yes
County FE	Yes	Yes	Yes
Year FE	Yes	Yes	Yes
Observations	690	690	690
R2	0.673	0.523	0.774

Note: The sample period ranged from 2013 to 2018. Treat × Post is a dummy variable that = 1 for county-level districts and counties in the CZT region in the special regulation period (2015–2018). Economic controls include: per capita regional gross domestic product, the proportion of tertiary industry output value in total gross domestic product, the number of population, the number of health institutions, the number of professional doctors, and the number of registered nurses. Weather controls include: the number of days with daily mean temperature falling into the kth bin of {≥30 °C, 25–30 °C, 15–20 °C, 10–15 °C, 5–10 °C, 0–5 °C, <0 °C}, total precipitation amount, average relative humidity, average wind speed, sunshine duration, and atmospheric pressure. Robust standard errors clustered at the county level are in parentheses.

suggest that the fictitious SPR has no significant effect on medical expenses covered under the UURBMI and UEBMI systems.

In addition, if there were some unknown confounding factors affecting the medical expenses, our estimated causal effect of the SPR on medical expenses happened to be a coincidence. Thus, we randomly select 500 fictitious treatment groups from our sample and design placebo tests to prove that our results are not caused by other factors. We use the fictitious treatment groups and re-estimate Equation (1) to study the effects of the SPR. As shown in Figure 4, the mean of the fictitious treatments is zero. The actual treatment effects of the SPR on total medical expenses and medical expenses covered under the UURBMI are located in the left tail of the distributions. The actual treatment effects of SPR on the medical expenses covered under the UEBMI are located in the right tail of the distribution. The results show that the fictitious treatment effects are much smaller than the real treatment effects. Therefore, our main conclusions are not chance findings.

5.5. Mechanisms

Existing literature has suggested that air pollution impairs physical and mental health [2,5,6,39], which can lead to an increased demand for medical care. As a result, medical expenses have increased. A large literature in economics documents the negative effects of effective environmental regulations on air pollution [17,45]. In this section, we estimate the effects of the SPR on air pollution using a DID approach based on Equation (1). We chose PM_2.5, PM₁₀, and SO₂ as dependent variables, and the results are presented in

Table 8. Mechanism test: air pollution.

	(1) PM2.5	(2) PM10	(3) SO2
Treat × Post	−0.589 ***	−0.433 ***	−0.945 ***
	(0.034)	(0.037)	(0.103)
Economic controls	Yes	Yes	Yes
Weather controls	Yes	Yes	Yes
County FE	Yes	Yes	Yes
City-by-year FE	Yes	Yes	Yes
Observations	672	672	672
R2	0.988	0.981	0.978

Note: The sample period ranged from 2013 to 2018. Treat × Post is a dummy variable that =1 for county-level districts and counties in the CZT region in the special regulation period (2015–2018). Economic controls include per capita regional gross domestic product, the proportion of tertiary industry output value in total gross domestic product, the number of population. Weather controls include mean temperature, total precipitation amount, average relative humidity, average wind speed, sunshine duration, and atmospheric pressure. Robust standard errors clustered at the county level are in parentheses. *** p < 0.01.

. We find that the SPR has significantly negative effects on different pollutants at the 1% level. The PM_2.5, PM₁₀, and SO₂ concentrations of the treatment group are 58.9%, 43.3%, and 94.5% lower than those of the control group, respectively.

We then compare our estimates to evaluate the effect of PM_2.5 on economic costs. To do so, we combine the effects of the SPR on medical expenses and assume that the decline in medical expenses in the CZT region is totally caused by the reduction of PM_2.5 emissions. As shown above, the implementation of the SPR decreases medical expenses by 11.1%, to about RMB 30.482 million (USD $4.780 million). Therefore, a 1% decrease in PM_2.5 decreases medical expenses by RMB 0.518 million (USD $0.081 million) in our case. A 1% decrease in PM_2.5 concentration reduces the expenditure on mental illness by USD $600 million [6]. An exogenous 1% decrease in PM_2.5 nationwide increases the output across all firms by USD $1200 million annually [46]. These estimates are much larger than those of the present study. Because our estimates are only for Hunan Province, whereas their estimates are for nationwide provinces. Furthermore, our estimates should be interpreted with caution because our study does not directly estimate the causal effect of air pollution on medical expenses. In addition, the medical expenses only include the part borne by the UURBMI and UEBMI, excluding those borne by the insured and other insurance schemes.

6. Conclusions

This study estimates the effects of the SPR, a command-and-control environmental regulation implemented in the CZT region, on medical expenses in Hunan Province, using a DID method. The main results show that the implementation of the SPR significantly reduces medical expense paid by the UURBMI and UEBMI programs in the CZT region. Compared with the control group, the medical expenses significantly decrease by 11.1% in the treatment group. Our results are robust after conducting a battery of robustness checks and placebo tests. Furthermore, heterogeneous analyses show that this regulation has a significantly negative impact on the medical expenses paid by the UURBMI and a positive impact on those paid by the UEBMI. Finally, regarding the influencing channels, the SPR significantly decreases air pollution in the CZT region.

Due to data availability, we could not comprehensively evaluate the effects of the SPR on medical expenses borne by the insured and other insurances, nor could we evaluate defensive expenditures. Hence, our results partly underestimate the health benefits of the policies. We are also unable to directly examine the health mechanism channels and study the heterogeneous effects of SPR on medical expenses for different individuals and diseases.

The costs and benefits of China’s environmental regulations are fruitful areas. We leave these limitations and shortcomings for future research. This paper provides an empirical analysis framework for this.