Residents’ Health Effect of Environmental Regulations in Coal-Dependent Industries: Empirical Evidence from China’s Cement Industry

Abstract

Coal-dependent industries are the economic pillar of many countries; however, their contribution to air pollution also restricts long-term economic development. While the negative effects of environmental regulations on coal-dependent industries has attracted much attention, the health effects of such regulations remains barely quantitatively studied. Our study is based on a quasi-natural experiment created by a command-and-control policy in China’s cement industry, whereby cement enterprises are forced to stop production for a specific period of time every year. This paper adopted DID methods and found that direct pollution control measures for coal-dependent industries could significantly improve residents’ health levels and that the resultant reduction in medical expenditures could save 456.8 RMB yuan per capita per year. Additionally, our mechanism analysis found that the COPP is beneficial to residents’ health in that it reduces air pollution and sewage treatment rates. This means that pollution from coal-dependent industries has a large and underestimated impact on residents’ health. Environmental regulation of the coal-dependent industry could greatly improve the health of residents.

1. Introduction

The existing literature regards air pollution as a primary restraining factor for long-term economic development. The main mechanisms include damage to physical and mental health [1,2], a decrease in labor productivity [3,4,5], a decline in family income [6], an increase in social medical expenditures [7], and annual direct GDP losses [8,9]. Among all pollution sources, coal-dependent industries are the main source of anthropogenic emissions, which affect residents’ health by increasing cardiovascular and respiratory morbidity [10,11,12]. To control the pollution emissions of coal-dependent industries, governments have formulated a large number of regulatory policies. However, to determine whether these policies can achieve the expected results, the current analysis is still insufficient.

First, although there are many studies on the causal effects of pollution and residents’ health in the current literature, most of them are aimed at developed countries [13,14]. However, pollution problems are often more severe in developing countries, where the vast majority of residents live in an environment with excessive pollution [15]. Second, industrial pollution mainly comes from coal-dependent industries [16,17], such as thermal power [18,19]. However, some other heavy industries, such as the cement industry, have not been studied in the literature. Third, coal-burning pollution has a great negative impact on residents’ health [20]. However, the current literature mainly focuses on respiratory and cardiovascular diseases [21,22], with particular studies focusing on the impact of pollution on residents’ health as shown through hospital visits [23]. These studies ignore non-respiratory and non-cardiovascular diseases and non-hospital visits and may thus underestimate the impact of coal-burning pollution on residents’ health. Considering the above gaps, in view of the Cement Off-Peak Production Policy (the COPP), we take China’s cement industry as our research object to study the impact of environmental regulations in the cement industry on residents’ health.

The COPP was implemented in 2014 in China. Based on the COPP, cement enterprises are forced to stop production for a specific period every year, which greatly reduces air pollution emissions. We used the difference-in-differences method to study the impact of the COPP on residents’ health. The results indicate that the COPP could greatly improve residents’ health and save medical expenditures. A series of robustness tests and placebo tests supported our conclusion. Furthermore, a mechanism study found that this was mainly due to the improved air and water quality caused by the COPP, which improved residents’ health and reduced medical expenditures.

This paper makes three contributions. First, the current literature mainly studies the impact of pollution on the health of residents of developed countries. We take China’s cement industry as the research object and focus on pollution and the health of residents of developing countries. Second, while there is some research on the health effects of pollution in coal-dependent industries, these studies mainly focus on the thermal power sector. With the cement industry as our background, we are able to extend the current research to other coal-dependent industries. Third, we use the residents’ self-rated health index to study the causal effect of pollution on residents’ health, which is better for estimating the impact of pollution. This method could better compensate for the deficiencies of the current literature, which only uses the index of respiratory and cardiovascular diseases or the index of hospital visits.

This paper is arranged as follows. The Section 2 reviews the relevant literature and presents the policy background. The Section 3 introduces the model construction, variables, and data. The Section 4 outlines the empirical results and corresponding analysis. The Section 5 provides the conclusion and policy implications.

2. Literature Review and Background

2.1. Literature

Our study closely relates to two strands of the current literature. The first is the impact of total air pollution levels on residents’ health [24,25,26,27,28,29,30,31]. However, the current literature mainly focuses on respiratory and cardiovascular diseases [32,33], with various studies focusing on the impact of pollution on residents’ health as indicated through hospital visits [23]. Whether through respiratory and cardiovascular diseases or hospital visits, these studies found that pollution significantly impacted residents’ health [34,35]. However, it is insufficient to use the above two methods to study the causal effects of pollution and health. Using respiratory and cardiovascular diseases may lead researchers to disregard non-respiratory and non-cardiovascular diseases, and using hospital visits may lead them to disregard non-hospital visits. Consequently, these studies may underestimate the impact of pollution on residents’ health. The existing literature also mainly focuses on developed countries. Only a few studies have focused on China, such as He et al. (2020) [36] and Tanaka (2015) [37], who studied the impact of pollution emissions and environmental regulation on residents’ health in China.

The second branch of the literature focuses on the health impact of pollution emissions from coal-dependent industries. Yang and Chou (2018) [18], Luechinger (2014) [19], and Barreca et al. (2021) [20] used the closure of coal-fired power plants as an exogenous shock to quantify the impact of pollution emissions from coal-dependent industries on residents’ health. Currie et al. (2015) [38] used quasi-natural experiments generated by the opening and closing of 1600 chemical plants to find that air pollution emissions could reduce newborn body weight by 3%. However, these studies mainly focus on developed countries. Some studies have focused on China; for example, Chen et al. (2021) [39] used county-level data from disease surveillance points in China between 2004 and 2008 to examine the spillover effects of air pollution from neighboring coal-fired power plants on local cardiovascular and respiratory disease-related deaths. However, the cement industry is also a heavy pollution emission industry, and there are few cement-focused studies on the harm to residents’ health caused by pollution emissions in the current literature.

To conclude, there are gaps in the current literature. First, most previous literature focuses on the health effects of environmental pollution in developed countries. Relevant research with respect to developing countries remains inadequate, especially research on coal-dependent industry pollution. Second, few studies have focused on the health effects of coal-dependent industries’ pollution. This is one of major sources of pollution emissions in the economy. Third, previous research has mainly employed the index of respiratory and cardiovascular diseases or the index of hospital visits. This study exploited the self-evaluated health index of participants randomly sampled from the whole residential group, which makes up for the potential underestimates of health impacts. To fill these gaps, taking China’s cement industry as our research object, based on the Cement Off-Peak Production Policy, we study the impact of environmental regulations in the cement industry on residents’ health.

2.2. Policy Background

The cement industry is a coal-dependent industry, mainly using coal as input. Coal burning is the most significant contributor to pollution in developing countries, especially in China. The cement industry produces severe pollution every year. The primary material used for cement production is clinker. The production of clinker requires a large amount of coal, generating severe air pollutants such as SO₂ and NO₂. This means that the cement industry, like other coal-dependent industries, is responsible for significant health effects due to its industrial pollution emissions.

To control pollution, the Chinese government decided to uniformly stop the production of cement clinker, which causes pollution, in the heating season, in order to reduce pollution and alleviate the effects of excess cement production. Subsequently, the policy was extended to the whole country.

Table 1. Provinces implementing the COPP.

Year	New Provinces Implementing the COPP
2014	Heilongjiang, Jilin, Liaoning
2015	Hebei, Henan, Shandong, Shanxi, Shaanxi, Qinghai, Inner Mongolia Autonomous Region, Ningxia Hui Autonomous Region, Xinjiang Uygur Autonomous Region, Gansu, Beijing, Tianjin
2016	Sichuan
2017	Zhejiang, Fujian, Guangdong, Guangxi Zhuang Autonomous Region, Hubei, Hunan, Jiangsu, Jiangxi, Guizhou, Chongqing

Note: The years in this table refer to the years when the COPP was implemented in each province.

shows the provinces that participated in the policy and when it started. In the north, the COPP is mainly implemented in winter, while, in the south, the COPP is mainly implemented in winter and the rainy season. Since the number of enterprises in the cement industry is relatively fixed in the short term, China prohibits new production capacity, so this policy is a typical multiperiod DID quasi-natural experiment.

Since the implementation of the COPP in 2014, the effects have been remarkable. In 2015 and 2016, the clinker output in the northern provinces decreased by 85.56 and 237.92 million tons, respectively, according to the reports on China Cement Association Web, which indicates that the COPP effectively reduced clinker production. This trend is displayed in Figure 1, where the monthly clinker outputs of six clinker firms abruptly dropped to zero from January to March and from November to December in 2015.

The government’s report on the COPP indicated that, in 2016 and 2017, there were sulfur dioxide reductions of 11,500 and 12,800 tons, respectively; nitrogen oxide reductions of 428,000 and 475,000 tons, respectively; and particulate emission reductions of 5800 and 6400 tons, respectively; thus, the report indicated that the policy effectively reduced pollutant emissions in the cement industry.

The cement industry causes serious pollution emissions and social health damage [17]. There have been studies on the impact of the COPP on the improvement of air pollution [40], and there are also studies on the impact of the COPP on the economic performance of enterprises from an economic perspective [41]. However, there is a lack of research on the impact of the pollution from the cement industry on residents’ health. The purpose of this paper is to study the impact of coal-dependent industries on residents’ health based on the COPP, using the cement industry as our object.

3. Empirical Strategy

3.1. Model Specification

To estimate the causal effect of the COPP on residents’ health status, we employ a multiperiod DID method as follows, based on Beck et al. (2010) [42]:

$$Y_{ijt}=\alpha +\beta Trea{t}_{jt}+Individua{l}_{ijt}\delta +Provinc{e}_{jt}\rho +{\mu }_j+{\phi }_t+{\epsilon }_{ijt}$$

(1)

where $i$ is the $ith$ individual sample, $j$ is province $j$, and $t$ is year $t$. $Y_{ijt}$ is the explained variable and is proxied by health status and medical expenditure. The residents’ self-rated health index is a discrete variable, and medical expenditure is a continuous variable. Therefore, the linear probability model is used to analyze the impact of the policy on residents’ health, and the ordinary linear regression model is used to analyze the impact of the policy on medical expenditure. $\alpha$ is the constant term. $Trea{t}_{jt}$ is the explanatory variable. $Trea{t}_{jt}=1$ indicates that the COPP is implemented in province $j$ and year $t$; thus, the sample is treated by the COPP, while $Trea{t}_{jt}=0$ indicates the opposite. $Individua{l}_{ijt}$ is a matrix of individual-level characteristic variables for the individual $i$ in the year $t$ (including education level, age, net family income, and marriage status). Residents with different education levels, marital statuses, or ages may have heterogeneous reactions to pollution exposure [43]. Additionally, there is a bidirectional causality between income and health [44]. Failure to control for these factors can lead to bias. To adjust the influence of different provinces’ economic development and climatic conditions on health [30], we add the $Provinc{e}_{jt }$ variable. $Provinc{e}_{jt }$ is a matrix of province-level control variables, including economic variables (GDP) and climate variables (temperature, relative humidity, wind speed, and direction). ${\mu }_j$ controls for the provincial fixed effects, and ${\phi }_t$ controls for the year fixed effect. ${\epsilon }_{ijt}$ is the error term that includes unobservable variables not controlled for in the model. $\beta$ is the key coefficient that indicates whether the COPP shows a significant effect on the residents’ health.

To examine whether Model (1) satisfies the common trend assumption of the DID method, and to identify the dynamic effect of the COPP, we employ the following model:

$$Y_{ijt}={\displaystyle \sum }_{\tau \ne 0}{\alpha }_{\tau }Trea{t}_{j\tau }+Individua{l}_{ijt}\delta +Provinc{e}_{jt}\rho +{\mu }_j+{\phi }_t+{\epsilon }_{ijt}$$

(2)

where $\tau =t-policy year$ + 1, which is the standardized year, and $\tau$ = 1 indicates the first year of COPP implementation. Within the same standard year, if province $j$ implemented the COPP, $Trea{t}_{j\tau }$ = 1; otherwise, $Trea{t}_{j\tau }$ = 0. The setting of the control variables here is the same as in Model (1).

Furthermore, we analyze the impact mechanism of the COPP on residents’ health, and the model is set as follows:

$$Y_{kt}=\alpha +\beta Trea{t}_{kt}+X_{kt}^{\text{'}}\delta +{\mu }_k+{\phi }_t+{\epsilon }_{kt}$$

(3)

where k is city $k$, and $t$ is year $t$. $Y$ is the explained variable, which is the logarithm of the SO₂ and NO₂ concentrations. The advantage of using a logarithmic form is that the estimated coefficient represents the percentage of the impact of the COPP on the SO₂ and NO₂ concentrations, and the logarithmic form can also alleviate the possible heteroscedasticity and non-stationary series problems of panel data. $\alpha$ is the constant term. $Treat$ is a dummy variable. $Treat$ = 1 indicates that the COPP is implemented, while $Treat=0$ indicates the opposite. $\beta$ is the key coefficient. $X^{\prime }$ is the vector of a series of control variables. To control for the impact of economic development in different regions on air pollution, we control for the logarithm of the GDP growth rate of each city and the logarithm of the proportion of the secondary industry [24]; to control for the influence of the heating policy in the north on the air quality in the south and north, we add the logarithm of the heating volume [31]; to control for the impact of climate factors on pollution emissions, we have controlled for the logarithm of air humidity, temperature, wind speed, and wind direction [30]. $\mu$ controls for the city fixed effects, and $\phi$ controls for the year fixed effect. $\epsilon$ is the error term that includes unobservable variables not controlled for in the model.

For Model (1), the CFPS database discloses the province where the observations are located, not the individual locations. The explanatory variable represents the implementation time of the COPP in each province. Therefore, Model (1) is a provincial-level analysis. However, the mechanism analysis of Model (3) uses city-level air pollution concentration data, and the explanatory variable is also the COPP implementation time at the city level. Therefore, Model (3) is a city-level analysis, which is different from Model (1)

3.2. Variable Description and Data Sources

3.2.1. Explained Variable

We derive indices that provide detailed descriptions of individual health status from the CFPS dataset, a survey database published by the Institute of Social Science Survey, Peking University. The CFPS database mainly focuses on economic activities, education, and physical health at the individual, family, and community levels. The survey sample covers 25 provinces and municipalities, with a scale of 16,000 families, including 33,000 adults and 9000 underage children per year. These biennial data were initiated in 2010 and provide five sample periods during our observation period of 2010–2018.

We employ two indices, self-evaluated health status and medical expenditure, as the explained variables, sequentially, both from the CFPS database.

Self-evaluated health status: The main examination employs citizen-level self-evaluated health status as the explained variable. This index details the subjective health status of the individual when interviewed by dividing the health status into five levels: very unhealthy, unhealthy, relatively unhealthy, average, and healthy. To quantify the description, we assign a value of 1 to 5 to each status, where very unhealthy = 1 and healthy = 5. The CFPS database also provides relative health indices such as whether an individual feels ill or experiences a chronic disease. However, the effect of air pollution on health is a continuously distributed variable that varies across physical differences. In most cases, air pollution will not cause noticeable symptoms. The feeling of illness and chronic disease experience indices can underestimate the air pollution effect.

To make the sample comparable before and after treatment, we made the following modification. First, the definition of this index in 2010 is somewhat different from the following survey periods and thus is standardized to remain comparable. Second, observations without self-evaluated health information, or those evaluated to be uncertain or inapplicable, are eliminated. Third, we eliminate the individual observations that are not available before and after the COPP. After this, the sample consists of 24,685 individuals who contribute up to 99,832 observations.

Two dummy variables that describe whether the interviewee felt uncomfortable within the last two weeks (yes = 1, no = 0) and whether the interviewee had a chronic disease within the last half year (yes = 1, no = 0) are employed as the explained variables for the robustness check. These data are also from the CFPS database.

Medical expenditure: To further monetize the health effect, this paper employed personal health expenditure data as the explained variable. One important mechanism by which air pollution causes economic losses is the increase in individual health expenditures. Therefore, quantifying the savings in personal medical expenditures caused by the COPP can well evaluate the economic benefits that the COPP creates by improving residents’ health. These data are also obtained from the CFPS database.

3.2.2. Explanatory Variable

The explanatory variable is a dummy variable that indicates whether the observation is made during the COPP treatment. If province j in period t was regulated during the shutdown period by the COPP, then $Trea{t}_{jt}$ = 1; otherwise, $Trea{t}_{jt}$ = 0. This variable is manually sorted according to the formal policy papers that are published annually by the local and central governments. The methods to generate this dummy followed Wang et al. (2021) [41].

3.2.3. Control Variables

We control for individual-level characteristics, provincial GDP, and climate variables while isolating the effect of the COPP on health status and medical expenditure. The individual characteristics include education level, age, net family income, and marital status, which are attained simultaneously with health and medical indices by the CFPS. Provincial GDP is used to control for the differences in economic development across provinces. The GDP data are from the National Statistics Bureau. Climate variables are an assignable factor that covaries with the health status and the geological distribution of COPP implementation. Therefore, a matrix of climate variables (temperature, relative humidity, wind speed, and wind direction) is controlled for. The climate data are derived from the Goddard Data and Information Service Center M2T1NXFLX_V5.12.4, a level-2 satellite remote sensing dataset with a resolution of 50 km × 62.5 km and an hourly frequency. We employ the data in the form of the annual average.

3.2.4. Other Variables

To explain the impact mechanism of the COPP on the health level of residents, this paper also uses other variables:

(1)

The explained variables used in the mechanism analysis are the SO₂ concentration and NO₂ concentration, both of which are from satellite data [45]. The SO₂ concentration data are derived from the Goddard Data and Information Service Center M2T1NXFLX_V5.12.4, a level-2 satellite remote sensing dataset with a resolution of 50 km × 62.5 km and an hourly frequency. The NO₂ concentration data are derived from the POMINO satellite data provided by Peking University, a level-3 satellite remote sensing dataset with a resolution of 50 km × 66.7 km and six hourly frequencies. Variables such as the centralized treatment rate of sewage treatment plants, the comprehensive treatment rate of general industrial solid waste, and the harmless treatment rate of domestic waste are used to replace the explained variables for further verification. They are all from the Statistical Yearbook of Chinese Cities.

(2)

The explanatory variable $Trea{t}_{kt}$ used in the mechanism is a dummy variable that indicates whether the city stops production or not. If the sample city is in the process of stopping production due to the COPP, then $Trea{t}_{kt}$ = 1; otherwise, $Trea{t}_{kt}$ = 0. This variable is manually sorted according to the formal policy papers that are annually published by the local and central governments. The methods to generate this dummy followed Wang et al. (2021) [41].

(3)

The control variables used in the mechanism include the city’s GDP growth rate, the proportion of the secondary industry of each city, the heating capacity of each city, and the air humidity, air temperature, wind speed, and wind direction. The first two variables are from the Statistical Yearbook of Chinese Cities, and the missing data are supplemented according to the corresponding provincial or city statistical yearbook; the variables for the city heating volume are from the Statistical Yearbook of City Construction in China; the climate data are derived from the Goddard Data and Information Service Center M2T1NXFLX_V5.12.4, a level-2 satellite remote sensing dataset with a resolution of 50 km × 62.5 km and an hourly frequency. The definitions, abbreviations, units, and data sources of all variables are shown in

Table 2. Variable definition and data sources.

Variables	Abb	Unit	Data Source
Main variables
Residents’ health levels	Health	-	CFPS database
Medical expenditure	Expenditure	yuan	CFPS database
Education	Education	-	CFPS database
Age	Age	year	CFPS database
Log (1 + net household income)	Income	yuan	CFPS database
Marriage	Marriage	-	CFPS database
Province GDP	PG	100 million yuan	National Statistics Bureau
Air temperature	Temperature	K	M2T1NXFLX_V5.12.4
Relative humidity	Humidity	kg/m2	M2T1NXFLX_V5.12.4
Wind speed	Speed	m/s	M2T1NXFLX_V5.12.4
East wind	East	-	M2T1NXFLX_V5.12.4
North wind	North	-	M2T1NXFLX_V5.12.4
Other variables
SO2 concentration	SO2	μg/m2	M2T1NXFLX_V5.12.4
NO2 concentration	NO2	μg/m2	POMINO
City GDP growth rate	CGGR	%	China City Statistical Yearbook
City secondary industry proportion	CSIP	%	China City Statistical Yearbook
City heating capacity	CHC	10,000 GJ	China City Construction Statistical Yearbook
Air humidity	AH	kg/m2s	M2T1NXFLX_V5.12.4
Wind speed	WS	m/s	M2T1NXFLX_V5.12.4
Air temperature	AT	K	M2T1NXFLX_V5.12.4
Wind direction	WD	-	M2T1NXFLX_V5.12.4
Centralized treatment rate of sewage treatment plants	The sewage treatment rate	%	China City Statistical Yearbook
Comprehensive treatment rate of general industrial solid waste	The solid waste treatment rate	%	China City Statistical Yearbook
Harmless treatment rate of domestic waste	The domestic waste treatment rate	%	China City Statistical Yearbook

.

4. Results and Discussion

4.1. Summary Statistics of the Variables

The summary statistics for the variables described above are shown in

Table 3. Summary statistics of the variables.

Abb	Observations	Mean	S.D.	Min	Max
Main variables
Health	99,832	3.28	1.29	1	5
Expenditure	34,350	3003.82	12,029.51	0	550,000
Education	62,534	2.73	1.57	1	10
Age	42,837	46.46	15.09	12	98
Income	40,217	10.07	1.38	0	16.15
Marriage	40,212	2.10	0.81	1	5
PG	243	28,290.73	20,648.48	3943.7	107,986.9
Temperature	243	286.23	5.72	273.64	298.29
Humidity	243	4 × 10−5	2 × 10−5	8.04 × 10−6	1.02 × 10−4
Speed	243	5.13	0.75	3.21	6.98
East	243	−0.12	1.41	−2.99	2.81
North	243	−0.13	0.76	−1.58	2.52
Other variables
NO2	2280	7048.5	6017.41	398.58	33,844.77
SO2	2280	0.69	0.31	0.35	2.91
CGGR	2280	10.3	7.81	−19.8	38.6
CSIP	2280	47.86	11.33	5.6	89.75
CHC	2280	164.03	559.26	0	6034
AH	2280	0.01	0	0	0.02
WS	2280	5.13	0.91	2.61	7.59
AT	2280	0.08	0.05	−0.01	0.3
WD	2280	1.57	0.85	0	4.78
Sewage treatment rate	1947	82.09	15.42	9.12	100
Solid waste treatment rate	1943	81.44	21.88	0.49	100
Domestic waste treatment rate	1924	88.98	17.44	5.49	100

. We use the main variables in Table 3 to study the impact of the COPP on residents’ health. The mean of the health variable is 3.28. The number of observations is 99,832 without controlling for any variables except for the individual-fixed and the time-fixed effects in the baseline regression. The number of observations shrinks when control variables are added. The number of observations is 39,876 with all controls added. We use other variables for mechanism research in Table 3. This is a city-level regression analysis with 285 cities. The number of observations will drop in further analyses. Additional variable information is shown in Table 3.

4.2. Baseline Results

Table 4. Impact of the COPP on residents’ health levels.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)
	Health	Health	Health	Health	Health	Health	Health
Treat	0.100 ***	0.112 ***	0.171 ***	0.176 ***	0.175 ***	0.174 ***	0.174 ***
	(0.0234)	(0.0260)	(0.0279)	(0.0288)	(0.0288)	(0.0340)	(0.0408)
Education		0.0215 ***	0.0210 ***	0.0245 ***	0.0247 ***	0.0260 ***	0.0261 ***
		(0.00459)	(0.00427)	(0.00460)	(0.00456)	(0.00477)	(0.00445)
Age			−1.28 × 10−5	0.00415	0.0240	0.0230	0.0224
			(0.0146)	(0.0176)	(0.0157)	(0.0163)	(0.0162)
Income				0.0107 **	0.0107 **	0.0102 *	0.0108 **
				(0.00500)	(0.00499)	(0.00528)	(0.00514)
Marriage					−0.0139	−0.0129	−0.0135
					(0.0105)	(0.0105)	(0.0105)
PG						−4.63 × 10−7	4.41 × 10−7
						(1.57 × 10−6)	(1.01 × 10−6)
Temperature							0.0767 **
							(0.0364)
Humidity							3240
							(2009)
Speed							0.103
							(0.109)
East							−0.158 **
							(0.0592)
North							0.0691
							(0.0611)
Individual FE	Y	Y	Y	Y	Y	Y	Y
Time FE	Y	Y	Y	Y	Y	Y	Y
observations	99,832	62,534	42,837	40,217	40,212	39,876	39,876
R-squared	0.570	0.706	0.746	0.747	0.747	0.747	0.748

Note: (1) Numbers in brackets are standard errors; (2) ***, ** and * represent a significance level of 1%, 5%, and 10% respectively; (3) FE represents fixed effects, as follows; (4) Y, which means YES, controls the corresponding fixed effects.

represents the linear probability regression results for the effect of the COPP dummy on the residents’ self-evaluated health indicators via Model (1). Column (1) is the benchmark result without control variables, except for the individual-fixed and the time-fixed effects. The benchmark result indicates that the COPP significantly (p < 0.01) increases residents’ health level by 10%. Residents with different education levels, marital statuses, or ages may have heterogeneous reactions to pollution exposure [43]. Additionally, there is bidirectional causality between income and health [44]. Failure to control for these factors can lead to bias. Columns (2) to (5) control for these variables in turn. The core explanatory variable coefficients (treat) are all significant at the 1% level, and the coefficients are 11.2%, 17.1%, 17.6%, and 17.5%, respectively, which indicates that the COPP significantly improves the health of residents. To adjust the influence of different provinces’ economic development and climatic conditions on health [30], the GDP and climate variables for each province are controlled for in Columns (6) and (7), respectively. GDP variables for each province are added to Column (6) based on controlling for all of the personal characteristics variables in Column (5). Column (7) adds climate variables based on the Column (6) control variables. The core explanatory variable coefficients (treat) for Columns (6) and (7) are significant at the 1% level, with coefficients of 17.4%. The article takes Column (7) as the benchmark result; on average, the policy increases the individual’s health probability by 17.4%.

4.3. Robustness Test

This section examines the robustness of the benchmark regressions by substituting the regression technique and the explained variable. For discrete explained variables, another commonly used regression technique is the logit model. Therefore, logit regression is conducted via Model (1) to examine the reliability of the linear probability regression in Section 4.2. Among the five degrees of the health status, the Very Unhealthy group is chosen as the benchmark group.

Table 5. Robustness test results using logit model.

		(1)	(2)	(3)	(4)
		Unhealthy	Less Healthy	Normal	Healthy
Treat	-	−0.320 **	0.227	0.380 *	0.578 ***
		(0.152)	(0.142)	(0.216)	(0.134)
RRR	-	0.726	1.254	1.461	1.78
Marginal effects	−0.012 *	−0.089 ***	−0.011	0.032	0.080 ***
Delta-method SE	(0.006)	(0.011)	(0.015)	(0.033)	(0.028)
Control variables	-	Y	Y	Y	Y
Individual FE	-	Y	Y	Y	Y
Year FE	-	Y	Y	Y	Y

Note: (1) Numbers in brackets are standard errors; (2) ***, ** and * represent a significance level of 1%, 5%, and 10% respectively; (3) FE represents fixed effects, as follows; (4) Y, which means YES, controls the corresponding control variables and fixed effects.

shows the results. The first row shows the log-odds of each group relative to the benchmark group. To better represent the change rate for the log-odds of each group compared with the benchmark group, the relative-risk ratios (RRR) are also reported in Table 5. The results show that the COPP significantly increases the log-odds of the Healthy (p < 0.01) and Average groups (p < 0.1), which indicates that residents’ health status is improved by the COPP. Meanwhile, the log-odds of the Unhealthy group significantly decreased by 32% (p < 0.05). The coefficient means that the probability of being rated as unhealthy decreases. Furthermore, the marginal effects of the COPP show that the probability of being unhealthy and relatively unhealthy decreases, while the probability of being healthy obviously increases. The results indicate that more residents’ self-evaluations change from very unhealthy to unhealthy to healthy. This proves the robustness of the benchmark result and the reliability of the linear probability regression. Therefore, the following examinations are conducted via linear probability regression.

The self-evaluated health indicator outweighs the better measurement of the overall physical health status, but there may be self-report bias, which threatens the reliability of the results. Therefore, this section examines the effect of the COPP treatment on the probability of having physical discomfort or chronic symptoms as a robustness test.

Table 6. Robustness test results with the substitution of the health indices.

	CFPS Data
	(1)	(2)
Variables	Physical Discomforts	Chronic Symptoms
Treat	−0.103 *	−0.0311 **
	(0.0569)	(0.0127)
Control variables	Y	Y
Individual FE	Y	Y
Year FE	Y	Y
Observations	34,350	34,350
R-squared	0.749	0.593

Note: (1) Numbers in brackets are standard errors; (2) ***, ** and * represent a significance level of 1%, 5%, and 10% respectively; (3) FE represents fixed effects, as follows; (4) Y, which means YES, controls the corresponding control variables and fixed effects.

shows the results, where the probability of reporting physical discomfort significantly decreases by 10.3% (p < 0.1) (seen in Column (1)) and chronic symptoms decrease by 3.11% (p < 0.05) (seen in Column (2)).

The robustness checks above indicate that the benchmark regression results are reliable and that the self-evaluated health indicators ideally reflect the changes in individual health levels. In terms of the expression of the health effect, the COPP can not only significantly reduce the degree of physical discomfort but can also significantly reduce the incidence of chronic diseases, perhaps because the alleviation of air pollution significantly reduces the complications of chronic diseases in the population. More importantly, there are also some unperceived or slightly perceived health effects of air pollution on health that cannot be reflected by the routine health indicators employed in the literature. Self-evaluated health indicators can reflect the health benefits of environmental regulation in a more comprehensive way.

4.4. Dynamic Effect Analysis

The premise for employing a DID model is that the common trend assumption should be satisfied. Here, we examine the dynamic effect of the COPP on residents’ health statuses via Model (2). Figure 2 shows the results. Figure 2 visualizes the dynamic effects of the COPP on health level, using the standard year zero as the reference, where the black line connects the estimates and the blue line indicates the 95% confidence intervals. The corresponding coefficients are estimated using Model (2). The coefficient of the COPP remains negative before the standard year and insignificant in the first two periods, which indicates that there is no significant difference in pre-trends between the treatment group and the control group, and the common trend assumption is satisfied. Meanwhile, the coefficient after the stand year becomes positive and stays significant throughout the year. The common trend assumption test supports the main conclusion of this paper.

4.5. Placebo Test

A placebo test is conducted to further validate the main conclusion. To examine whether the effect is caused by some unobserved factor, we conducted 200 random placebo experiments.

Table 7. Placebo test results.

Panel A: One Result from 200 Placebo Tests.
	(1)
	Health
Placebo effect	−0.03
	(0.0187)
Control variables	Y
FE	Y
Panel B: Results of 200 iterations of placebo sampling, number of estimates landing above, below, and within 95 percent confidence interval around 0
	Significant		Insignificant
	Above 0	Below 0
Health	9	20	171
Panel C: Cumulative probability distribution of 200 Placebo tests

Note: (1) Numbers in brackets are standard errors; (2) ***, ** and * represent a significance level of 1%, 5%, and 10% respectively; (3) FE represents fixed effects, as follows; (4) Y, which means YES, controls the corresponding control variables and fixed effects.

Panel A shows 1 result among the 200 placebo tests and indicates that the placebo effect is insignificant. Panel B shows the number of runs in which each outcome of interest is above or below zero and is significant at the 5% level versus not significant at the 5% level. Panel B shows the symbols of the 200 coefficients, where 171 are not significant, 20 are significantly negative (with a ratio of 10%), and 9 are significantly positive (4.5%). Panel C shows that the accumulative distribution probability of the coefficients satisfies the characteristics of the accumulative normal distribution. The vertical lines show the mean coefficients of Treat. The standard errors are clustered at the province level. The average value of the coefficients is close to zero, while the true causal coefficient is far above zero. The placebo test indicates that there is no evidence showing that the effect of the COPP on physical health is due to some unobserved factor.

4.6. Mechanism Analysis

Why does the COPP positively affect residents’ health level? Based on this policy, cement enterprises are forced to stop production at a specific time, which leads to a decline in cement production, thus reducing the concentration of air pollution. This may have a positive impact on residents’ health. Wang et al. (2021) [41] found that the COPP has led to a significant decline in the production of clinker and cement, which has led to an increase in the price of clinker and cement in the market. This means that the COPP may lead to a decline in the concentration of air pollution. In this paper, Model (3) is used to test whether the concentration of air pollution decreases significantly. The regression results are shown in

Table 8. Mechanism analysis results.

	(1)	(2)	(3)	(4)	(5)
	Log (SO2)	Log (NO2)	The Sewage Treatment Rate	The Solid Waste Treatment Rate	The Domestic Waste Treatment Rate
Treat	−0.0611 *** (0.0111)	−0.0766 *** (0.0275)	−0.0607 ** (0.0254)	−0.00643 (0.0342)	−0.0392 (0.0351)
Control variables	Y	Y	Y	Y	Y
City FE	Y	Y	Y	Y	Y
Year FE	Y	Y	Y	Y	Y
Observations	2280	2280	1947	1943	1924
R-squared	0.474	0.427	0.631	0.678	0.454

Note: (1) Numbers in brackets are standard errors; (2) ***, ** and * represent a significance level of 1%, 5%, and 10% respectively; (3) FE represents fixed effects, as follows; (4) Y, which means YES, controls the corresponding control variables and fixed effects.

, and they control for the city’s GDP growth rate, the proportion of the secondary industry of each city, the heating capacity of each city, air humidity, air temperature, wind speed and wind direction, and city and year fixed effects. Column (1) is the regression result for SO₂, and Column (2) is the regression result for NO₂. The COPP reduced the SO₂ concentration by 6.11% and the NO₂ concentration by 7.66%. In general, the COPP significantly reduces the concentrations of SO₂ and NO₂, which is consistent with the research results of Xu et al. (2021) [40].

To examine how the COPP affects residents’ health, we also conduct the following analysis. The COPP mainly targets the cement production end and restricts the operation of the production line. As the cement industry’s pollution emissions mainly come from the clinker production process, shutting down clinker production leads to a reduction in pollutant emissions. At the same time, the production of clinker also produces a large amount of sewage, and shutting down clinker production also reduces the discharge of sewage. This could reduce the sewage treatment rate. We use the pollution emission data from the China City Statistical Yearbook for further testing, and the regression results are shown in Table 8. The results in Column (3) show that the COPP reduced the sewage treatment rate by 6.07%. The regression results in Column (4) and Column (5) show that the coefficients for the solid waste treatment rate and the domestic waste solid waste treatment rate are negative and insignificant. In summary, our empirical analysis found that the COPP significantly reduces air pollution emissions and the sewage treatment rate, which may have a positive impact on residents’ health.

4.7. Medical Expenditure

The above analysis validated that the COPP significantly increased the residents’ health level, from which naturally follows the question of the value of this health effect. Since the literature has already found that air pollution damages economic development by significantly increasing social medical expenditures [46], this section further analyzes the effect of this environmental regulation on medical expenditures and gives the monetized health gains. The analysis was conducted via Model (1) using the medical expenditure data from CFPS.

Table 9. The impact of the COPP on medical expenditure.

	(1)	(2)	(3)
	Expenditure	Physical Discomforts Expenditure	Chronic Symptoms Expenditure
Treat	−456.8 *	−876.9 **	−1050 *
	(257.0)	(393.5)	(513.8)
Control variables	Y	Y	Y
Region FE	Y	Y	Y
Year FE	Y	Y	Y
Observations	34,350	29,578	20,106
R-squared	0.593	0.501	0.501

Note: (1) Numbers in brackets are standard errors; (2) ***, ** and * represent a significance level of 1%, 5%, and 10% respectively; (3) FE represents fixed effects, as follows; (4) Y, which means YES, controls the corresponding control variables and fixed effects.

Column (1) shows the result that the COPP can significantly reduce annual personal medical expenditures by 456.8 RMB yuan (p < 0.1). Additionally, the examinations were repeatedly carried out within the group with physical discomfort only and the group with chronic symptoms only. As seen in Columns (2) and (3), the COPP can significantly save the group with physical discomforts and the group with chronic symptoms 877 RMB yuan (p < 0.05) and 1050 RMB yuan (p < 0.1) per capita in medical expenditures.

The above results show that the COPP can significantly reduce personal medical expenditures by improving the health levels of the residents. Furthermore, the magnitude of the reduction in medical expenditure is even larger for people with physical discomfort and chronic symptoms. Meanwhile, these effects on medical expenditures imply that the unperceived or slightly perceived symptoms caused by air pollution can also effectively affect the cost of health care and, thus, the overall social welfare instead of being a negligible illusion. Environmental regulations in energy-intensive industries can effectively reduce such costs.

4.8. Discussion

The COPP aims to reduce air pollution emissions in the cement industry. We found that, after the implementation of the COPP, the health level of residents was improved, and medical expenditure was reduced. A series of robustness tests also proved this conclusion. Furthermore, our mechanism analysis found that the COPP will reduce air pollution emissions and the sewage treatment rate, which may have a positive impact on residents’ health. Although we have not found any significant changes in the solid waste treatment rate, this does not mean that the COPP has no impact on the solid waste treatment rate.

However, according to Wang et al. (2021) [41], the COPP has reduced the incomes and profits of enterprises, which may lead to a decline in the income of employees in the cement industry or other residents in related areas. This means that there may be a potential economic channel that has a negative impact on residents’ health. If this is indeed the case, then the baseline results estimate the net effect of the channel we found above and the potential economic channel. Therefore, the baseline results give the lower limit of the policy impact on residents’ health. The baseline conclusion is unchanged.

5. Conclusions and Policy Implications

The main conclusions of this paper are as follows. First, based on the COPP policy, this paper uses the self-rated health index to study the impact of pollution emissions from coal-dependent industries on residents’ health. The results show that controlling the production of the cement industry during the period of serious air pollution can improve the overall health of residents and reduce the possibility of physical discomfort and chronic symptoms. Second, the improvement of residents’ health by the COPP is reflected in their medical expenditures, and its magnitude is large. Third, our mechanism analysis found that the COPP policy reduced air pollution emissions and sewage treatment rates, which may improve residents’ health and reduce medical expenditures.

The policy implications of the conclusions in this paper are as follows. From the perspective of policy design, coal-dependent industry pollution has a huge negative impact on social health. Direct production regulation for specific coal-dependent industries will have significant health effects. Theoretically, direct production control measures can also produce the same significant health benefits in other coal-dependent industries. From the perspective of social welfare, pollution has a hidden impact on residents’ health, such as non-respiratory and non-cardiovascular diseases or non-hospital visits. Therefore, environmental pollution may have a great impact on social welfare, and environmental regulation may have great benefits for social health. From the perspective of industrial pollution emissions, most coal-dependent industries, such as the firepower industry and the cement industry, have environmental pollution problems. The factories of coal-dependent industries are built in the same industrial park, which is beneficial for the exchange and use of waste between coal-dependent industries and the reduction of pollution emissions. Moreover, coal-dependent industries are concentrated in the same industrial park and use the same set of complete pollution treatment facilities, which can reduce the cost of industrial pollution treatment.