The Impact of Sustained Exposure to Air Pollutant on the Mental Health: Evidence from China

Abstract

Emerging evidence suggests that poor mental health is particularly pronounced among Chinese residents, who are exposed to nearly the worst air pollution worldwide. However, the correlations between air pollutant concentration and mental health have not been consistently reported in previous studies. Methodologically speaking, a sufficiently rigorous design is required to demonstrate the causal relationship between the two factors. In this study, we aimed to infer the causal relationship between air pollutant concentration and mental health. In this panel research, the data were compiled through a combination of statistics from the China Family Panel Study, China Environmental Statistics Yearbook, World Meteorological Association, and China National Bureau of Statistics. Ultimately, this study enrolled 65,326 individuals whose mental health, air pollutant concentration, and other demographic information was available and robust. The RD design of this study utilizes the discontinuous variation in air pollutant concentrations and mental health as one crosses the Huai River boundary, which is an arbitrary heating policy that caused the difference in air pollutant concentrations between the north and south of China. In this study, we found that a 10 μg/m³ increase in air pollutant concentrations (air particulate matter smaller than 10 μm (PM₁₀)) leads to a 4.9-unit decrease in the mental health of the Chinese residents(coeff = 0.49, SD = 0.07, p < 0.05), equivalent to 36% of the average of Chinese residents. In the heterogeneity model, the impairment of mental health by air pollutant concentrations was more pronounced in male residents (coeff = 1.37, SD = 0.10, p < 0.05) compared to female residents (coeff = 0.42, SD = 0.04, p < 0.05) and smokers (compared to non-smokers). The robustness of the results is ensured by changing the RD bandwidth and polynomial order, and by two unique sensitivity analyses. The results indicate that air pollutant concentrations significantly impair the mental health of Chinese residents, which provides empirical evidence supporting the Chinese government’s decision to invest more in combating air pollution and ensuring the mental health of Chinese residents.

1. Introduction

Mental health is defined as a healthy state of managing stress, working productively, and contributing to the community [1]. Mental disorders are the second leading cause of the global burden of disease, posing a widespread threat to human health. The mental health crisis is particularly pronounced among Chinese residents. More specifically, the cumulative number of people with mental disorders in China is 95 million, equivalent to one-third of the United States’ population. This results in an indirect economic loss of 4.85 trillion dollars annually, which is equivalent to the total losses from the 2008 sub-prime mortgage crisis. Therefore, to safeguard the well-being of all people and reduce health burdens, it is necessary to identity the potential undesired effects associated with poor mental health. Scholars have extensively explored the correlation between mental health and conventional factors, such as drug abuse, maternal infection, physical activity, hormonal changes, lifestyle, urbanization, etc. [2,3,4,5,6,7,8]. However, despite the severity of the worldwide air pollution problem, no conclusive evidence exists describing the connection between air pollutants and mental health.

The previous literature examining the relationship between air pollutants and mental health suffers from the following three major methodological flaws: (a) air pollutants are highly correlated with the urban economy, industrial structure, and other confounding factors, all of which would influence mental health. Previous studies, despite accounting for personal income, have omitted these confounders, which could underestimate air pollutants’ influence on mental health; (b) there may be a reverse causality between air pollutants and mental health. Deterioration in mental health may lead to lower labor supply productivity and affect the pollutant emissions associated with economic activities. Some studies tried to use the inverse temperature phenomenon as an instrumental variable to reduce the reverse causality. However, they were unable to significantly improve the accuracy of the model; and (c) The irrational behavior of people in choosing where to live (exposure to air pollutants) invokes the problem of selective sorting in previous studies. More specifically, people with better mental health may prefer cleaner air and choose to live in cities with lower air pollution. This irrational behavior would result in a non-random distribution of air pollutants in the population and skew the estimated relationship between air pollution and mental health.

Fortunately, China’s free heating policy for the Huai River boundary could be a possible quasi-experimental design to study the relationship between air pollutants and mental health, overcoming the three limitations described above. Established during Mao’s period, the free heating policy aimed to provide free or highly subsidized coal for indoor heating for cities north of the Huai River boundary. This policy ensured that low temperatures would not seriously affect the daily lives of northern Chinese residents. However, burning a vast quantity of fossil fuels released numerous air pollutants and damaged the mental health of northern residents, which is a problem still existing now. This research aims to illustrate, through an empirical model, that there is no discontinuous change in the confounders (e.g., income, physical activity, or smoking) which affect mental health among urban residents living north and south of the Huai River boundary. Accordingly, this research was able to isolate the net effect of air pollutants on mental health.

Using an RD (Regression Discontinuity) design necessitates ensuring that the dependent variable (mental health) and independent variable (air pollution) are completely authentic, simultaneously. Mental health data used in this research are the resident self-reported mental health index (measured by the Center for Epidemiological Survey Depression Scale, CES-D) from the Chinese Family Panel Studies (CFPS), which objectively reflect the mental health of the Chinese residents. The panel data of air pollutants are from the China Environmental Statistics Yearbook. Though criticized due to the possible political manipulation, the data are nevertheless acceptable, as politicians do not have sufficient incentive to falsify environmental statistics when their performance is evaluated primarily based on economic rather than environmental performance [9,10].

The results of the RD regression indicate that a 10 μg/m³ rise in air pollutants would lead to a significant decrease of 4.9 units in mental health (SD = 0.07, p < 0.05). More specifically, with the arbitrary heating policy, China has 3.5 million more mentally unhealthy residents, which is similar to the population of Los Angeles. Meanwhile, different tests were applied to prove the robustness of the above empirical results.

2. Methods

2.1. Population and Study Design

The study is based on the China Family Panel Studies (CFPS), which are large-scale (contain 25 provinces and over 100,000 respondents), multi-ethnic (contain 14 ethnic groups), interdisciplinary social surveys executed by the Chinese Social Science Survey Center (ISSS) of Peking University. The CFPS cover hundreds of thousands of respondents and track 30 percent of them, which provides a detailed description of the socio-economic characteristics of the Chinese residents. More importantly, the CFPS use multilevel random sampling to maintain geographic balance. In other words, we were able to obtain a sample with a relatively balanced north–south distribution. This feature is important for our analysis of the causal relationship between air pollutant concentrations and mental health using the RD design. To enhance the comparability and authenticity of our data, we chose the CFPS 2016–2020 to construct the panel data, and the selected data use a consistent, standardized questionnaire for the measurement of mental health. After screening out invalid data, a final sample of 65,326 was included in this study.

2.2. Huai River Policy and Discontinuous APCs in Northern and Southern China

It is well-known that China is a developing country with a vast territory. Geographically speaking, China is divided by the Huai River into two significantly different climates, namely a Monsoon climate of medium latitudes (geographically located in Northern China) and a Subtropical monsoon climate (geographically located in Southern China). More specifically, northern cities experience extremely low temperatures in winter than their southern counterparts. This phenomenon makes it difficult for the residents of northern China to engage in outdoor production during the winter, which negatively affects the development of the Chinese economy.

Mao proposed an arbitrary heating policy (namely the Huai River policy) in the 1960s to enable northern Chinese residents to participate in outdoor productions during the winter. The Huai River is used as a border dividing China into two heterogeneous parts, with the north of the Huai River being the northern part of China, which receives free heating in winter. South of the Huai River is without heating.

However, due to technological limitations, Chinese heating is mainly accomplished through the burning of fossil fuels, which results in significantly higher concentrations of air pollutants (PM₁₀ (Particulate Matter 10), NO₂ (Nitrogen Dioxide), SO₂ (Sulfur Dioxide), CO (Carbon monoxide), O₃ (Ozone), etc.) in northern China. A representative article once found that air pollutant concentrations in northern cities were twice as high as those in southern cities for the arbitrary Huai River policy. The influence of the Huai River policy on China is depicted graphically in Figure 1.

2.3. Air Pollutant Exposure Assessment

Air pollutant data (including the average concentrations of PM₁₀, NO₂, SO₂, CO, and O₃) were collected from the China Environment Yearbook, which contains nearly-comprehensive statistics for 300 cities from 2016 to 2020. To construct a nested database, we solely selected air pollution data from 125 cities where city Id was available in the CFPS.

We chose PM₁₀ concentration as a proxy for air pollutants for the following reasons. In a pathological context, PM₁₀ is more likely to affect mental health than other pollutants. A large body of literature has confirmed that PM₁₀ inhalation exacerbates the development of systemic or brain-based oxidative stress and inflammation, as well as causing severe sadness and anxiety [11,12]. Additionally, PM₁₀ has a large diameter and is more likely to affect air visibility compared to other pollutants. Accordingly, it is more likely to affect residents’ mental health psychologically [13].

Figure 2 illustrates the graphical relationship between city location and air pollutant concentrations using ARCGIS 17.2 (where the darker the color, the more serious the air pollution). The blue line denotes the Huai River boundary, and each color block shows the location of the city and the concentration of air pollutants. Significant differences in air pollutant concentrations can be clearly identified between the cities on the north and south sides of the Huai River boundary. We detail in

Table 1. North–South differences in air pollutant concentrations.

Variables	Mean in South Group	Mean in North Group	Mean Difference
PM10	45.317 (0.179)	66.564 (0.116)	−21.247 ***
SO2	17.658 (0.221)	49.349 (0.076)	−31.690 ***
CO	1.318 (0.006)	2.318 (0.003)	−1.000 ***
NO2	34.946 (0.098)	47.201 (0.091)	−12.255 ***
O3	117.511 (0.312)	104.074 (0.281)	13.437 ***

Note: *** significant at 1%.

the north–south disparities in PM₁₀, NO₂, SO₂, CO, and O₃ concentrations. It can be found that the concentrations of five major air pollutants in northern cities are significantly higher than those in southern cities. Furthermore, the discontinuous change of air pollutants partly justifies the RD design.

2.4. Mental Health Assessment

The data for mental health were from the mental health module of the CFPS database, which was divided into 6 questions, including the perceptions of depression, low self-efficacy, dyssomnia, upset, loneliness, and hopelessness, respectively. All items were taken from the standard CES-D questionnaire, and we summed them up to obtain an index indicating the mental health of individual Chinese residents, referring to the strategies of other psychology researchers [14].

Similarly, the difference in mental health between northern and southern residents on the Huai boundary is graphically represented in Figure 3, where the depth of the color indicates the severity of the resident’s mental problems. The Huai boundary results in a statistically significant decline in the mental health of the northern residents compared to that of their southern counterparts. Similarly, we systematically report North–South differences in mental problems in

Table 2. North–South differences in mental problems.

Variables	Mean in South Group	Mean in North Group	Mean Difference
Mental Problem total	13.285 (0.022)	13.383 (0.024)	−0.098 ***
Mental Problem: depression	1.663 (0.004)	1.737 (0.004)	−0.074 ***
Mental Problem: low self-efficacy	1.717 (0.004)	1.73 (0.005)	−0.013 *
Mental Problem: dyssomnia	1.792 (0.005)	1.771 (0.005)	0.020 ***
Mental Problem: upset	2.062 (0.005)	2.036 (0.005)	0.027 ***
Mental Problem: loneliness	1.427 (0.004)	1.474 (0.004)	−0.047 ***
Mental Problem: hopelessness	1.934 (0.005)	1.906 (0.006)	0.028 ***

Note: * Significant at 10%, *** significant at 1%.

. The results indicate that residents in the North generally suffer severer mental problems compared to their southern counterparts, and were statistically significant. Taken together, Figure 1 and Figure 2 demonstrate the discontinuous change in air pollution and mental health near the Huai boundary, which fully justifies the RD design.

2.5. Covariates

Our study includes two levels of covariates: (a) level 1 includes conventional individual demographic characteristics such as age, sex, residence, wage, education, employment, BMI (Body Mass Index), physical activity, and smoking habits (as shown in Panel 3 of Table 1) and (b) level 2 includes provincial variables that have rarely been considered in previous studies, such as GDP, per capita GDP, proportion of secondary industry, temperature, and wind (as shown in Panel 2 of Table 1). Relevant studies have revealed that the characteristics listed above have statistically significant effects on mental health [15,16,17]. Level 1 covariates are from the CFPS individual self-reported questionnaire, and level 2 covariates were compiled through a combination of the statistic from the World Meteorological Association and China National Bureau of Statistics.

2.6. Statistical Analysis

We used three approaches to assess the relationship between air pollutant concentrations and mental health. The first approach was to use the traditional ordinary least squares estimation as Equation (1):

$$M_i={\alpha }_0+{\gamma }_1{X}_{i,j}+{{\displaystyle \sum }}_{i=1}^n{\gamma }_i{Z}_i+{\epsilon }_j$$

(1)

where X_i,j denotes the air pollutant concentration in city j where individual i is located, Zi is a vector indicating the covariates affecting mental health, M_i represents the mental health of individual i, and the coefficient γ₁ indicates the impacts of air pollutant concentration on mental health. Consistent estimates of γ₁ require that unobserved mental health influences do not covary with X_i,j after adjusting for Z_i. However, this condition is hard to satisfy in practice because of omitted variables and sample self-selection problems. More importantly, we cannot use traditional OLS (Ordinary Least Squares) regression because our dependent, independent, and covariate variables are clearly nested data.

Therefore, we used the Hierarchical Linear Models (HLM) to further estimate the relationship between air pollution and mental health as in Equation (2):

$$M_{i,j}={\gamma }_{0,0}+{\gamma }_{0,1}{X}_{i,j}+{\pi }_{i,j}Individua{l}_{i,j}+{\theta }_{i,j}provincia{l}_{i,j}+{\epsilon }_{i,j}+{\mu }_{0,j}$$

(2)

where π_i,j is the slope of individual-level variables for individual i in province j and θ_i,j is the slope of provincial variables for province j. γ_0,1 is a vector that denotes the coefficient between air pollutant concentrations and mental health based on HLM. However, although HLM can analyze nested data, it cannot cope with the endogeneity problems caused by omitted variables, reverse causality, and sample self-selection, which are not negligible in estimating the relationship between air pollution and mental health.

The RD design is ideal for eliminating endogenous problems and inferring a causal relationship between air pollutant and mental health. This study’s RD design utilizes the discontinuous variation in air pollutant concentrations and mental health as one crosses the Huai River boundary. The necessary precondition is that any covariates affecting mental health are smooth around the boundary. Adjustment for a sufficiently flexible polynomial in distance from the river will eliminate all potential causes of bias and provide the following causal inferences if the precondition is met:

$$X_{i,j,t}={\alpha }_0+{\alpha }_1{N}_{i,j,t}+{\alpha }_{2}f(L_{i,j,t})+C_{i,j,t}k+{\vartheta }_j$$

(3a)

$$Y_{i,j,t}={\delta }_0+{\delta }_1{N}_{i,j,t}+{\delta }_{2}f(L_{i,j,t})+C_{i,j,t}k+{\vartheta }_j$$

(3b)

Equation (3a,b) represent the RD estimates of air pollutant concentrations and mental health for year t in city j of individual i, respectively. X_i,j,t and Y_i,j,t denote air pollutant concentrations and mental health in city j of individual i in year t. N_j is the driver variable, taking values of 1 and 0 to indicate that city j is on the northern or southern side of the Huai River boundary, respectively. f(L_j) is a polynomial used to assess the distance of city j from the boundary. C_j,i,t refers to all province-level, individual-level covariates affecting mental health in year t. X_i,j,t, Y_i,j,t, and C_j,i,t are panel data, simultaneously.

Furthermore, we used the RD design to infer the causal relationship between air pollutant concentrations and mental health. More Specifically, we choose an appropriate bandwidth so that the Huai River boundary only affects mental health through air pollutant concentrations. Then, it was valid to treat Equation (3a) as the first stage in a two-stage least-squares (2SLS) system of equations. In this way, 2SLS estimation could assess the causal relationship between air pollutant concentrations and mental health as in Equation (3c):

$$Y_{i,j,t}={\rho }_0+{\rho }_1\widehat{X_{i,j,t}}+{\rho }_{2}f(L_{i,j,t})+C_{i,j,t}+{\epsilon }_j$$

(3c)

By combining Equations (1)–(3), we can obtain an unbiased causal effect of air pollutants on mental health, excluding the endogeneity problem of omitted variables, confounders, reverse causation, and sample self-selection.

3. Results

3.1. Study Population Description

Table 3. Characteristics of the study population (n = 65,326).

Characteristics	Overall (n = 65,326) (1)	North (n = 29,039) (2)	South (n = 36,285) (3)	Mean Difference (4)	Adjusted Difference (5)	p-Value (6)
Panel 1: air pollutant exposure and mental health difference between China’s Huai River.
Mental Problem	13.34 (8, 32)	13.38 (8, 32)	13.28 (8, 32)	0.098 **	0.426 ***	0.003/<0.001
PM10	57.21 (11, 164)	66.56 (18, 164)	45.32 (11, 105)	21.247 ***	16.08 ***	<0.001/<0.001
NO2	41.85 (7, 160)	47.20 (14, 160)	34.95 (7, 114)	12.254 ***	12.62 ***	<0.001/<0.001
SO2	35.51 (2, 272)	49.35 (18, 164)	17.66 (11, 105)	31.69 ***	34.30 ***	<0.001/<0.001
CO	35.51 (0.3, 9.5)	2.32 (0.5, 9.5)	1.31 (0.3, 3)	1.00 ***	1.08 ***	<0.001/<0.001
O3	109.94 (7, 224)	104.07 (7, 224)	117.51 (8, 186)	−13.436 ***	−15.37 ***	<0.001/<0.001
Panel 2: Economic and climate difference between China’s Huai River.
Ln (GDP)	17.14 (15.01, 19.76)	16.83 (15.01, 19.68)	17.54 (15.23, 19.76)	17.54 ***	17.92 ***	<0.001/<0.001
Ln (PGDP)	10.83 (9.45, 12.10)	10.66 (9.45, 12.01)	10.05 (10.10, 12.10)	11.05 ***	11.23 ***	<0.001/<0.001
Proportion of secondary industry	39.43 (12.08, 65.05)	38.71 (12.08, 65.05)	40.31 (21.7, 56.54)	40.30 ***	40.75 ***	<0.001/<0.001
Temperature	2.42 (−23.5, 21)	−2.73 (−23.5, 11.5)	8.91 (−2.5, 21)	8.91 ***	6.18 ***	<0.001/<0.001
Wind	2.72 (1, 5)	2.66 (1, 5)	2.79 (1, 5)	2.79 ***	2.44 ***	<0.001/<0.001
Panel 3: Demographic features difference between China’s Huai River
Age	46.35 (9, 104)	46.32 (9, 104)	46.40 (9, 104)	−0.08	-0.08	0.601/0.600
Sex (Female)	32,008 (49.00%)	14,347 (49.41%)	17,661 (48.67%)	0.01 *	<0.01	0.063/0.582
Residence (city)	40,178 (71.67%)	23,071 (73.37%)	17,107 (69.51%)	−0.06 ***	0.02	<0.001/0.124
Ln (wage)	10.14 (4.03, 13.82)	10.05 (4.03, 13.82)	10.24 (4.79, 13.12)	−0.19 ***	−0.023	<0.001/0.692
EduYear	7.69 (0, 22)	7.81 (0, 22)	7.53 (0, 22)	0.28 ***	0.18	<0.001/0.279
Employment (employed)	40,433 (74.58%)	22,653 (74.62%)	17,780 (74.53%)	−0.01	−0.12	<0.001/0.818
BMI	22.66 (8.16, 130)	23.01 (8.86, 130)	22.19 (8.16, 90.91)	0.82 ***	−0.16	<0.001/0.307
Physical Activity	95.33 (0, 2500)	97.96 (0.1, 2500)	92.07 (0, 2000)	5.89 ***	−2.614	<0.001/0.152
Smoke (smoked)	15,791 (26.45%)	9046 (27.03%)	6745 (25.71%)	0.01 ***	<0.01	0.01/0.180

Note: The values in brackets indicate the range (for continuous variables) or the percentage (for categorical variables) of each variable. * Significant at 10%, ** significant at 5%, *** significant at 1%. Sources: China Environmental Statistics Yearbook (2016–2020), World Meteorological Association (2016–2020), China National Bureau of Statistics (2016–2020), and Chinese Family Panel Studies (2016–2020).

presents a summary of numerous associated factors and evidence for the validity of the RD design. The means of key variables for the full sample, north, and south residents are shown in Columns (1)–(3), respectively. Column (4) shows the average difference of key indicators among the North and the South. Similar results are reported in Column (5), adjusted for a cubic polynomial in degrees north of the Huai River. It is a test for a discontinuous change at the Huai River boundary. An essential prerequisite of the RD design is that the covariates change continuously around the boundary, and Column (5) indicates that, after adjusting for a cubic polynomial in degrees north of the Huai River, there is no significant difference in the covariates (Panel 3 in Column (5)). However, it is also a desirable result if the covariates exhibit a smooth change around the boundary. Column (6) reports the p values associated with the tests, and that the heterogeneity in Columns (4) and (5) are equal to zero. Taken together, the results of Table 1 prove that the covariates change smoothly around the boundary, and the air pollutant concentration and mental health exhibited a discontinuous variation near the Huai River boundary.

Two preliminary results are clearly presented in Table 1. First, significant heterogeneity of air pollutant and mental health exists among southern and northern residents, which may indicate a potential correlation between these two factors. Second, there are significant differences in several demographic characteristics (e.g., wage, education, BMI) between the residents north and south of the boundary after accounting for the cubic polynomial in latitude. The above findings systematically fulfill the essential prerequisite of the RD design and make it possible to infer the causal relationship between air pollution and mental health.

3.2. Graphical Analysis

Figure 4 methodically displays the graphical results used to examine the RD design’s validity. The left half of Figure 4 represents the average air pollutant of each city in relation to their latitude north of the Huai River boundary. The horizontal coordinate takes the value of 0 to indicate the Huai River boundary, and each point represents the means of air pollutant (in Figure 4) across locations within a standard bin from the boundary. The discontinuous change in air pollutant on the boundary of the Huai River indicates that the arbitrary heating policy has led to a 11.4-(μg/m³) increase in PM₁₀ concentrations in the North relative to the South.

The difference in mental health between northern and southern residents on the Huai boundary is graphically represented in the right half of Figure 4. The Huai boundary causes a 0.47-unit decrease in the mental health of the northern residents compared with their southern counterparts. Taken together, Figure 4 demonstrates the discrete variations in air pollutants and mental health near the Huai boundary.

In addition, Figure 5 shows graphically the validity of the RD design by charting expected mental health differences (calculated as the fitted value from a hierarchical linear model (HLM) regression of mental health on all covariates except PM₁₀) at the Huai boundary. Notably, all covariates cumulatively explained 33.05% of the change in mental health. After removing the impacts of air pollution, it is evident that mental health is practically equivalent north and south of the border (p value of predicted mental health among the northern and southern residents of the border is 0.32, and statistically insignificant). Taken together, the above findings evidently prove that the potential covariates can’t explain the discontinuous decline in mental health north of the boundary, and uncontrolled air pollutants may be the cause of mental health differences between the North and South.

3.3. Regression Results of Air Pollutant Concentrations and Mental Health

Table 4. Association between air pollutant concentrations and mental health.

Dependent Variable	(1) OLS	(2) OLS	(3) HLM
Mental Problem	−0.006 *** (0.001)	−0.008 *** (0.003)	0.004 *** (0.001)
Intercept	13.706 *** (0.035)	15.147 *** (0.450)	13.411 *** (0.352)
Level 1 covariates (panel 2 in Table 1)	No	No	Yes
Level 2 covariates (panel 3 in Table 1)	No	Yes	Yes

Note: Each cell in the table represents the coefficient from a separate regression, and heteroskedastic-consistent SEs are reported in parentheses. Column 1 uses OLS without any covariates, column 2 uses OLS with individual-level covariates, and column 3 uses HLM with individual-level and provincial-level covariates, simultaneously. *** significant at 1%.

reveals the correlation between PM₁₀ and mental health using OLS and HLM. The set of covariates is shown in the last two rows of the table. A biased conclusion that PM₁₀ concentrations positively predict mental health will be drawn if we only take the first two columns into consideration, as the mental health of the population of a province is affected by its economic and medical situations. The two-level HLM model in Column 3 clearly reveals that a 100 μg/m³ increase in PM₁₀ concentration results in a 0.4-unit decrease in Chinese individuals’ mental health, and is statistically significant (95% CI: 0.001, 0.007).

Table 5. Association between air pollutant concentrations and mental health through RD.

Dependent Variable	(1)	(2)	(3)	(4)	(6)	(7)
Panel 1: Impact of “North” on the listed variable, RD robust with OLS
PM10	24.49 *** (0.67)	27.28 *** (0.70)	12.48 *** (0.47)	13.55 *** (0.55)	12.53 *** (0.41)	11.60 *** (0.47)
Mental Problem	0.21 *** (0.06)	0.11 *** (0.03)	0.65 *** (0.15)	0.59 *** (0.18)	0.60 *** (0.13)	0.74 *** (0.14)
Panel 2: Impact of mental health on the listed variable, RD robust with two-stage least squares
Mental Problem	0.49 *** (0.07)	0.35 *** (0.10)	0.81 *** (0.19)	0.82 *** (0.26)	0.68 *** (0.14)	0.88 *** (0.17)
Level 1 covariates (panel 2 in Table 1)	Yes	Yes	Yes	Yes	Yes	Yes
Level 2 covariates (panel 2 in Table 1)	Yes	Yes	Yes	Yes	Yes	Yes
RD type	Polynomial	Polynomial	Polynomial	Polynomial	Polynomial	Polynomial
Polynomial function	Linear	Second	Linear	Second	Linear	Second
Sample	2°	2°	5°	5°	8°	8°

Note: Each cell in the table represents the coefficient from a separate regression, and heteroskedastic-consistent SEs are reported in parentheses. Column 1 and 2 both restrict the sample to locations within 2° latitude of the boundary. Column 3 and 4 both restrict the sample to locations within 5° latitude of the boundary. Column 5 and 6 both restrict the sample to locations within 8° latitude of the boundary. *** significant at 1%.

displays the RD regression from the estimation of Equation (3a,3b) in panel 1 and the 2SLS results from Equation (3c) in panel 2. Column 1 introduces the cubic in distance between the location of the residents and the Huai boundary to account for possible deviations, and it limits the sample to locations within 2° latitude of the boundary to improve the accuracy of the RD estimation. Latitude is estimated by the first-order polynomial. Column 2 is identical to column 1, with the exception of the usage of a second-order polynomial to estimate the latitude. For robustness, the findings for 5° and 8° latitude are modeled with the first and second order polynomials, respectively, as shown in Columns 3 through 7.

Panel 1 of Table 5 shows a statistically significant difference in air quality and mental health between the southern and northern residents near the Huai boundary after adjusting for the covariates. The Huai River boundary resulted in a discontinuous increase in PM₁₀ concentrations in northern Chinese cities by 11.60–27.28 μg/m³, which is similar to the annual PM₁₀ concentration in Australia. Meanwhile, a result of the arbitrary heating practices, the mental health of residents to the north of the boundary has decreased by 0.11 to 0.74 units relative to those to the south.

Panel 2 reports the results of 2SLS in RD design. Column 1 presents the optimal RD model selected according to the AIC (Akaike information criterion) and BIC (Bayesian Information Criterion) optimality principle. Columns 2–7 provide evidence of robustness by correcting for polynomial order and latitudinal distance, respectively. Column 1 reveals that a 100-μg/m³ increase in PM₁₀ concentration leads to a 4.9-unit decrease in the mental health of Chinese residents (SD = 0.07, p < 0.05), equivalent to 36% of the average of Chinese residents.

The heterogeneity of the results across the demographic characteristics is presented in

Table 6. Comparison of RD results for samples with different demographic characteristics.

Variables	Male	Female	Smoker	Non-Smoker
Dependent variable	(1)	(2)	(3)	(4)
Panel 1: Impact of “North” on the listed variable, RD robust with OLS
PM10	15.69 *** (2.89)	17.38 *** (1.75)	19.21 *** (1.32)	21.05 *** (0.77)
Mental Problem	0.36 *** (0.04)	0.31 *** (0.04)	1.25 *** (0.33)	0.61 *** (0.22)
Panel 2: Impact of mental health on the listed variable, RD robust with two-stage least squares
Mental Problem	1.37 *** (0.10)	0.42 *** (0.04)	1.73 *** (0.46)	0.83 *** (0.29)
Level 1 covariates (panel 2 in Table 1)	Yes	Yes	Yes	Yes
Level 2 covariates (panel 2 in Table 1)	Yes	Yes	Yes	Yes
RD type	Polynomial	Polynomial	Polynomial	Polynomial
Polynomial function	Second	Second	Second	Second
Sample	5%	5%	5%	5%

Note: To ensure that the results are comparable, we restricted the sample to locations within 5° latitude of the boundary and fit the distance using a second-order polynomial. *** significant at 1%.

. Columns 1 and 2 reveal that the negative impact of air pollutants on mental health is more pronounced for male residents (coeff = 1.37, SD = 0.10, p < 0.05), compared to female residents (coeff = 0.42, SD = 0.04, p < 0.05). This is because more male residents work outdoors and are thus more likely to be exposed to air pollutants. Meanwhile, the damage to mental health from air pollutants is more pronounced in the smoking population (coeff = 1.73, SD = 0.46, p < 0.05) than in the non-smoking population (coeff = 0.83, SD = 0.29, p < 0.05). Thus, smoking positively moderates the negative mental health impacts of air pollutants.

3.4. Sensitivity Analysis

To prove the robustness of the RD design, we further used alternative approaches. We tested the validity of RD in the following two ways. Firstly, Figure 6 depicts the correlation between sample density and distance from the Huai River boundary. No significant differences in sample density can be found within a standard bandwidth (h = 5) north and south of the Huai River boundary. This conclusion indicates that the distribution of the drive variable (the distance north from the boundary) is completely random around the breakpoint. Secondly, RD designers pointed out that the likelihood of sample manipulation increases as the sample approaches the breakpoint. Using the donut hole test, we next determined if the results remained statistically significant after deleting 1%, 2%, and 3% of the samples near the breakpoints. Figure 7 clearly reveals that the coefficients remain statistically significant after removing 1% (95% CI: 0.47, 2.21), 2% (95% CI: 0.16, 3.11), and 3% (95%CI: 0.14, 3.54) of the samples near the breakpoints, respectively, which indicates that our samples were not manipulated near the breakpoint.

4. Discussion

The results of the RD regression indicate that a 100-μg/m³ rise in air pollution leads to a 49-unit decrease in mental health (SD = 0.07, p < 0.05). This estimate is ten times larger than results using the conventional HLM. The bias is due in part to sample self-selection, since people with better mental health may prefer cleaner air and choose to move to areas with better air quality, and this self-selection behavior confounds the estimation. In addition, county fixed effects will absorb permanent variations in economic activities across counties, but cannot absorb time-varying differences within counties, which might still bias the estimates downward. Meanwhile, this is approximately 1.25 times the estimated impact of PM_2.5 on the mental health of the Chinese population from Chen’s representative research [18]. The differences in the results of the two studies stem from three aspects: (a) PM₁₀ is more sensitive to induce the dysregulation in cytokine signaling and result in the occurrence of mental disorders; (b) the data from the study of Chen et al. are from 2014–2015, whereas we used the longitudinal data of 2016–2020, a period when Chinese residents were more sensitive to air pollutants for the unpredictable epidemic; and (c) this research uses an RD method, which can simulate random distribution near breakpoints and assess the causal effects between air pollutants and mental health.

Several European, American, and Australian researchers have also assessed the relationship between air pollutants and mental health using their own national databases, allowing for a cross-country comparison [19,20,21]. It turns out that PM₁₀ has the most devastating impact on the mental health of China’s residents (who are exposed to the most polluted air in the world) compared to residents of other countries. This conclusion implies that the effect of air pollutants on mental health is a concave function, i.e., the negative impact of air pollutants on mental health is more pronounced in severely polluted areas.

A natural concern related to the RD design is that the Huai River heating policy might influence the behavioral patterns of Chinese residents and induce bias in our estimation. This policy may promote mental health because northerners are likely to spend more time indoors and be exposed less to air pollutants. However, mental health could also be hindered if more time spent indoors leads to a stronger addiction to cigarettes and alcohol. Moreover, this free heating policy saves certain living costs and increases disposable income for northerners, which may further be spent on mental-health-benefitting (e.g., access to a psychiatrist, increased health insurance) or -harming (e.g., gambling, sex, drug use) behaviors and activities. The above situations could all lead to biased causal effects estimated by the RD design. Eventually, the evaluated effects of air pollutants on mental health should be interpreted with these constraints in mind, given the lack of data to test for these behavioral responses.

Unfortunately, it is hard to take the COVID-19 epidemic into our consideration, as the CFPS did not address the relevant individual characteristics. However, with China’s nationwide lockdown policy, we think that the impact of the epidemic on the causal relationship between APCs and mental health is negligible, especially with the RD design. More specifically, the impact of the epidemic on the mental health of all Chinese residents is almost homogeneous. That is, the impact of the epidemic on residents’ mental health will be the random error term in our regression (${\mathsf{\epsilon }}_{\mathrm{j}}$ in Equation (3c)).

Additionally, there are several limitations to be interpreted carefully: (a) as with the RD design, the fixed effects are not taken into our consideration. Although consistent with the traditional approach of contemporary empirical researchers dealing with breakpoint regressions on panel data, this may cause some estimation bias; (b) although containing 125 cities distributed on both sides of the Huai River boundary, data used still do not cover the entire potential sample; and (c) research design failed to account for the spatial autocorrelation between cities, which might cause several endogeneity problems. Despite the above limitations, this research is still theoretically and empirically significant, with an accurate and robust assessment of the relationship between air pollutants and mental health.

Results of this research could endorse Chinese ongoing policy and investment in pollution treatment. Because of arbitrary heating policies, China has an additional 3.5 million mentally unhealthy residents, the equivalent of the entire population of Los Angeles, California. Fortunately, since 2020, the Chinese government and environmental NGOs have been strongly appealing to residents to reduce carbon emissions. Empirical evidence indicates that it is vital for China’s government to strengthen its advocacy for a low-carbon lifestyle.

This paper is pioneering in attempting to apply an RD design to infer the causal relationship between air pollutant concentrations and mental health, and accordingly, provides several meaningful insights for future research. Researchers could compare the heterogeneous effects of air pollutant concentrations on the mental health between populations in developing and developed countries, as more cross-national studies based on RD designs are extremely meaningful. Moreover, future research could explore the potential relationship between air pollutant concentrations and mental health with different models such as the structural equation modeling model.

5. Conclusions

The results indicate that air pollutant concentrations significantly impair the mental health of Chinese residents, while the robustness of the results is ensured by changing the RD bandwidth and polynomial order and two unique sensitivity analyses. Meanwhile, the effect of APCs on mental health was more pronounced in male residents (compared to that in female residents), and smokers (compared to non-smokers). Additionally, we compared the results of the study with other literature and found that the effect of air pollutants on mental health is a concave function, i.e., the negative impact of air pollutants on mental health is more pronounced in severely polluted areas (such as China). More specifically, rising air pollutant concentrations due to arbitrary heating policies (shaping the Huai River boundary) have reduced the mental health of Northern Chinese residents by 0.49 units, equivalent to 3.6% of the average Chinese residents. These findings give empirical basis for the Chinese government to invest more in combatting air pollution and ensuring the mental health of its citizens.

Acronyms

RD	Regression Discontinuity
OLS	Ordinary Least Squares
HLM	Hierarchical Linear Model
BMI	Body Mass Index
APCs	Air Pollutant Concentrations
AIC	Akaike Information Criterion
BIC	Bayesian Information Criterion