During the past decades, China has experienced unprecedented economic growth, but it has also brought with it the byproduct of severe air pollution (37). PM2.5, as the main component of air pollution, has caused substantial welfare losses to individuals and society (38). The comparison of PM2.5 concentration maps for China in 2000 and 2020 reveals a substantial improvement in air quality over the two decades in Figure 1. The 2020 map shows a significant reduction in severe and heavy pollution areas, especially along the southeastern coast. This trend indicates the effectiveness of environmental policies and measures in reducing pollution, although some regions still require further attention to achieve better air quality standards. Furthermore, Figure 2 illustrates the difference in the annual average PM2.5 concentration level across cities of China in 2000 and 2020. In general, air pollution concentrations have significantly increased over the period. Moreover, there is a striking spatial disparity in the PM2.5 level.

To address this problem, China has implemented several policies to solve the issue of air pollution. For example, the Two Control Zone (TCZ) policy was launched in 1998, designating certain geographic areas where strict measures are applied to control pollutants, including PM2.5, sulfur dioxide (SO₂), and nitrogen oxides (NOx). China has continuously implemented various air quality action plans, including the National Air Pollution Control Action Plan (2013–2017), which set targets for reducing PM2.5 levels in critical regions, such as Beijing-Tianjin-Hebei, the Yangtze River Delta, and the Pearl River Delta. China has implemented stricter vehicle emission standards and promoted the use of electric vehicles (EVs) to address its significant air quality challenges.

China is experiencing significant demographic aging, with its population structure undergoing a dramatic transformation. Figure 3 displays the proportion of China's older adults population to the total population from 2013 to 2022. There are an increasing number of older adults. The working-age population is expected to decline by 23% from 925 million in 2011 to 800 million in 2050 (39). This demographic shift is primarily driven by two factors. First, the one-child policy, implemented from 1979 to 2015 to control population growth, significantly reduced birth rates. Second, advancements in healthcare have improved life expectancy, contributing to the rise in the older adults. Additionally, declining mortality rates further accelerate this trend. According to the Seventh National Population Census published in 2021, China's total fertility rate (TFR) dropped from 5.6 in 1970 to an estimated 1.3 in recent years. This demographic transition presents both challenges and opportunities. It calls for comprehensive reforms in social care, healthcare systems, and economic strategies to support the aging population. At the same time, advancements in technology and healthcare offer opportunities to enhance the quality of life for older adults.

The prevalence of depression among the middle-aged and older adults in China is a significant concern. The pooled overall prevalence of depressive symptoms among older adults in China was ~20.0%, indicating a considerable burden of depression in this demographic (40). The incidence of depressive symptoms has shown a slow increase since the 1990s. Especially, the evolution of depression among the Chinese older adults from 2011 to 2018 revealed a significant upward trend in the detection rates of depression, reaching 44.5% by 2018 (41). The prevalence of depression in the middle-aged and older adults population is of even greater concern. There is a clear need for policies and programs that consider the specific needs of vulnerable older adults, focusing on improving social participation and access to mental health services to mitigate the risk of depression. We provide new perspectives to explore how air pollution, as a potential causative factor, affects the mental health status of the middle-aged and older adults.

Our empirical objective is to examine the mental health effects of air pollution. It is typically measured using “pollution gradients,” i.e., ∂MentalHealth∂Pollution for pollution's mental health damage (45). We focus on this marginal approach and estimate how air pollution on mental health.

Taking advantage of the regional and temporal variations in air quality, this paper seeks to identify the impacts of air pollution on the mental health of middle-aged and older adults. Our empirical strategy anchors on a fixed-effect model. The model is given as follows:

In this specification, the outcome variable MH_icym denotes the mental health of middle-aged and older adults i from city c in month m year y. pollution_cym represents the monthly mean fine particulate matter (PM2.5). The coefficient, β₁, captures the depression effect of air pollution. The vector X_icmy refers to individual controls. The vector W_cmy refers to weather characteristics, including the average humidity, sunshine hours, wind speed, and precipitation. We further control for individual fixed effects (μ_i) to account for unobservable factors that are constant over time at the individual level, city fixed effects (σ_c) to account for unobservable factors that are constant over time at the city level, and year (π_y) and month fixed effects (ω_m) to control for external factors that are common to all units of observation in specific years and months. T_t refers to the monthly time trend that the interview date falls. ε_icmy is the error term. We cluster standard errors at the city level.

First, we investigate the baseline results of how air pollution affects the mental health of middle-aged and older adults. Table 2 presents the corresponding estimated coefficients for CES-D scores. Column (1) presents a parsimonious model that controls for the PM2.5 variables, individual FE, city FE, month FE, and year FE. Column (2) additionally controls the meteorological characteristics. Column (3) adds individual characteristics. Column (4) further controls for the linear monthly time trends. The increase in PM2.5 concentration leads to the CES-D score rising by 0.0168, 0.0176, 0.0313, and 0.0313 in columns (1) to (4), respectively. In conclusion, the model with tighter controls leads to a higher CES-D score. This implies that PM2.5 increases CES-D scores, which is statistically significant. From a mental health perspective, the magnitude of this effect is substantial. For example, according to the findings of Zhang et al. (46), the proportion of China's population living in areas with high PM2.5 concentrations (annual average PM2.5 concentration exceeding 75 μg/m3, compared to the WHO-recommended annual average of ≤ 5 μg/m3) accounted for ~5% of the total population after 2015. This exposure level suggests that elevated PM2.5 concentrations could significantly exacerbate depressive symptoms. It further verifies that air pollution exposure poses a potential threat to the mental health of the middle-aged and older adults.

The major components of depression include depressed mood, feelings of guilt and hopelessness, feelings of meaninglessness, and loneliness, which were identified from the clinical literature and factor analytic studies (47). The K-10 questions contain an inquiry about the frequency of these depressive symptoms. We next explore the effects of air pollution on the degree of each depression symptom separately. Table 3 displays the results, where the larger coefficients indicate more severe symptoms of depression. We find that air pollution significantly increases the occurrences of depressive symptoms of bothered, lonely, depressed, and meaningless.

We further used Air Quality Index (AQI), which was later substituted in robustness checks to strengthen the baseline results. The AQI data were obtained from the China National Environmental Monitoring Center through the web crawler technology, which can provide a broader measure of air quality, capturing multiple pollutants beyond PM2.5 alone. We try to validate the robustness of the findings, examining whether the effects identified with PM2.5 hold when evaluated against a more comprehensive air quality index. This approach enables an inclusive assessment of pollution-related impacts on mental health outcomes. Specifically, the AQI tracks the levels of various air pollutants, including ground-level ozone, particulate matter (such as PM2.5 and PM10), carbon monoxide, sulfur dioxide, and nitrogen dioxide. We categorize the AQI value into levels of increasing health concern. When the AQI value is from 0 to 50, the air quality is considered satisfactory, and air pollution poses little or no risk. The AQI value is from 51 to 100. The air quality is acceptable; however, there may be some health concerns for a very small number of people who are unusually sensitive to air pollution. The AQI value is from 101 to 150. The air quality is unhealthy for sensitive groups, but the public is unlikely to be affected. The AQI value is from 151 to 200. The air quality is unhealthy, and everyone may experience health effects. The AQI value is from 201 to 300. The air quality is very unhealthy; everyone may experience more serious health effects. The AQI value is from 301 to 500. The air quality is hazardous, which is a health warning of emergency conditions.

We use the following model to examine the effects of different pollution gradients on mental health:

Since the AQI value is from 0 to 50, which is no potential threat to the public, we use the 0–50 (excellent air quality) as the baseline group. For each observation, pollution_cym represents the air quality falling into the kth bin of {301–500 (seriously unhealthy,); 201–300 (unhealthy); 151–200 (moderately unhealthy); 101–150 (slightly unhealthy); 51–100 (acceptable)}. The rest of this model is consistent with Equation 1. Table 4 presents the results indicating that severe air pollution levels were associated with higher CES-D scores and more pronounced depressive symptoms, worsening mental health status. Moreover, the results may suggest that the relationship between air pollution levels and mental health outcomes might not be strictly linear across different pollution bins.

Although we have added adequate control variables at the city level that can affect air pollution, the empirical strategy may still be confounded by other unobservable factors that affect local air pollution and thus lead to estimation bias. Following Ferrara et al. (48), we use the placebo test by randomly assigning PM2.5 values. If no other factors disturb the policy, the coefficient will equal zero, meaning = 0. Next, we repeated the regression of Equation 1 500 times. In each round, the values of PM2.5 are shuffled across the city, year, and month, and the estimated coefficient is recorded. Figure 5 displays the distribution of these estimates, where the distribution of the coefficients follows a normal distribution with zero as the mean, which is far from the vertical dashed line indicating the actual coefficient estimate obtained from Column (4) in Table 2. This suggests that our identification is unbiased, and no unobserved characteristics affect our empirical results. It further verifies that the significant actual estimates rightfully refer to the impacts of air pollution on mental health rather than mere coincidence.

We consider another falsification test. Future air pollution is usually difficult and does not directly affect the current mental health status. Based on this, we follow Burnet and Kogan (49) and use the 2-year forward PM2.5 value as an explanatory variable in Equation 2 to conduct the falsification test. The results in column (1) of Table 5 signify that air pollution does not significantly affect mental health two years forward. This falsification test further demonstrates that our conclusions are not confounded by unobserved factors related to air pollution.

In spring 2020, COVID-19 started to spread in China. The fear of contracting the virus and the uncertainty surrounding the pandemic have contributed to heightened levels of anxiety and stress. Moreover, the social isolation measures of epidemic prevention, including lockdowns, social distancing, and quarantine, have led to increased feelings of loneliness and isolation, contributing to depression. Therefore, we exclude 2020, the worst year of the epidemic. The results shown in column (2) of Table 5 indicate that even if the effects of the epidemic are excluded, air pollution still raises CES-D scores and has a negative impact on the mental health of middle-aged and older adults.

Given the potential heterogeneity in PM2.5 concentration changes across different regions, we further stratify our sample into four major regions in China—Eastern, Central, Western, and Northeastern—to examine the impact of PM2.5 concentration on the mental health of middle-aged and older adults. However, due to the relatively small sample size in the Northeastern region (observations = 2,806), we focus our empirical analysis on the Eastern, Central, and Western regions. As depicted in Figures 1, 2, the Western regions experienced an increase in PM2.5 concentration from 2000 to 2020. In contrast, the Eastern region witnessed a significant decline in PM2.5 concentration over the same period. The results presented in Table 6 indicate that regions with more substantial pollution changes exhibit more pronounced negative impacts on mental health, thereby providing further evidence that air pollution is a significant contributing factor. Specifically, each unit increase in PM2.5 concentration is associated with a 0.0316-point increase in the CES-D score among middle-aged and older adults in the Western regions, respectively. Moreover, this differential impact across regions with varying pollution trends supports the argument that air pollution is a significant contributing factor to mental health outcomes.

Table 7 includes results for four more robustness checks using alternative model specifications. In column (1), we include year-month FE as a robust check. In column (2), we control for the city-month FE so that the estimated coefficients will be more accurate by accounting for potential omitted variable bias that varies across cities and over time. The results show that average air pollution affects mental health. Increases the CES-D score by 0.0347, indicating that a causal effect, large positive deviations from air pollution significantly deteriorate mental health. In column (3), we include the city-year FE as a robustness check. We aim to eliminate excessive variations in meteorological variables due to the data collection in the same city being usually conducted and focus on annual changes within each city. In column (4), we use the significance level of the estimates from the two-way clustering standard by city and interview month.

Some relevant literature pointed out that air pollution exposure may lead to a higher risk of depression later in life (50). We further divided the respondents into three groups: middle-aged (ages 45–60), older adults (ages 61–70), and old seniors (ages 71 and above), and conducted regression analyses for each group. Results in Table 8 indicate the heterogeneous effect of air pollution on mental health among different age groups. We find that middle-aged and older seniors are more susceptible to the adverse effects of air pollution on their mental health. Furthermore, the marginal impact on the mental health of older seniors is slightly higher than that on middle-aged individuals. However, for individuals aged between 61 and 70, referred to as the older adults, the air pollution does not have significant effects on the CES-D scores. For middle-aged individuals, frequent social activities, commuting, and outdoor work increase their exposure to polluted environments, thus rendering them more susceptible to the adverse effects of air pollution. In contrast, older seniors are more likely to rely on family and social care, with lower levels of autonomy in daily living and weaker adaptive capacity to environmental changes, making them more vulnerable to the impacts of air pollution. Consequently, social and family support play more critical roles in the mental health of this age group.

Some studies suggest that the impact of air pollution on mental health has a significant modification by gender (51). Therefore, we proceed to examine the heterogeneous impact of air pollution on mental health between males and females. Results in column (1) of Table 9 show that males are more prone to mental illness induced by air pollution than females. The observed gender disparity may stem from several interrelated factors. Behaviorally, due to occupational differences, men are often more exposed to polluted environments. Furthermore, psychological literature suggests that men are less likely to seek emotional support, potentially exacerbating pollutant-induced stress effects.

The psychological state of widowed older adults can indeed be more fragile compared to their married or non-widowed counterparts (52). This fragility may stem from the significant emotional loss and the subsequent adjustment to life without their spouse, which can profoundly impact their mental health. Results in column (2) of Table 9 show that the widowed older adults are more susceptible to the effects of air pollution on mental health.

Education may equip individuals with more knowledge on how to cope with air pollution (53). The survey population in CHARLS was mostly born in the 1970s, when educational attainment was generally low. Based on this, we next define a dummy variable of low educational attainment based on the 9-year compulsory education in the interviewers. Respondents who have obtained at least 9 years of education are classified into the higher education group, and those with < 9 years of education are classified into the lower education group. Results are displayed in column (3) of Table 9. We find that educational level does not significantly moderate the impact of air pollution on the mental health of middle-aged and older adults. Existing evidence suggests that air pollution impacts mental health primarily through biological mechanisms, including neuroinflammation, oxidative stress, and hypothalamic–pituitary–adrenal (HPA) axis dysregulation, that are independent of conscious awareness or behavioral mediation (54). Furthermore, physiological vulnerabilities associated with aging, such as increased prevalence of chronic diseases, may reduce the differential protective effects of education across exposure groups. Notably, individuals with higher education levels may exhibit greater environmental risk perception and awareness, potentially increasing psychological stress or anxiety related to pollution exposure (55). Consequently, educational attainment may not significantly moderate the relationship between air pollution exposure and mental health outcomes in the older adults population.

Existing literature has suggested that air pollution exposure impairs physical health, which can lead to deteriorating mental health (56). Specifically, air pollution exposure may also change an individual's lifestyle and health behaviors. Studies have shown that exposure to higher concentrations of particulate matter (PM2.5) is associated with poor sleep quality (57). Notably, poor sleep quality is associated with an increased risk of mental health issues. Poor and insufficient sleep increases negative emotional responses to stressors, decreases positive emotions, and may exacerbate psychiatric symptoms (58).

Scientific studies recommend that older adults, typically those aged 65 and older, should aim for 7–8 h of sleep per night (59). However, we take individual variations into account. Specifically, some older adults may feel natural and energized waking up after 6 h of sleep, while others may need up to 9 h of sleep to feel rejuvenated. Moreover, CHARLS collects information on the duration of sleep at night over the past month. Based on this information, we construct a sleep quality indicator (sleep) based on their sleep duration, i.e., if a respondent sleeps more than 6 h and < 9 h during the night, we consider that he/she has good sleep quality, which takes the value of 1. Otherwise, it is equal to zero. Further, we divide the sample into two groups: the middle-aged group (45–65) and the older adult group (>65). We next estimated whether air pollution affects the status of individual health behaviors and sleep quality using a linear probability model. Results are displayed in column (1) and column (2) of Table 10. The air pollution exposure has a significant negative impact on older adults, but not on the middle-aged group (45–65 years old), suggesting that severe air pollution impairs sleep quality and affects the lifestyles of older adults. Poor sleep quality, in turn, can negatively impact mental health, leading to increased anxiety and depression.

Existing literature suggests that air pollution impairs physical health, which can lead to deteriorating mental health (60). We estimate whether air pollution affects the cardiopulmonary health status of individuals. Based on this, CHARLS asks each interviewee whether they feel chest pain when climbing stairs (uphill) or walking quickly, which is usually used to assess cardiopulmonary health conditions. We define a dummy variable of individual cardiopulmonary health status that he/she feels well in the chest and walks quickly, which takes the value of 1. Otherwise, it is equal to zero. We estimate the effects of air pollution on this dummy variable using a linear probability model and differentiate the effects between the middle-aged and older adults. The primary findings are presented in columns (3) and (4) of Table 10. Specifically, a one-unit increase in PM2.5 concentration is associated with a 0.0064 rise in the probability of cardiovascular and cerebrovascular deterioration among the older adult population. This suggests that even minor fluctuations in PM2.5 levels can exert a significant impact on the cardiopulmonary health of this vulnerable group. For older adults who already suffer from cardiovascular and cerebrovascular diseases, the heightened risk may lead to a higher likelihood of disease exacerbation. This, in turn, indicates that air pollution not only directly harms the cardiopulmonary health of older adults but also indirectly exacerbates their mental health status through the worsening of physical health conditions. However, the effect of air pollution on the cardiopulmonary health status of the middle-aged group is not significant.

Air pollution has been found to impact life satisfaction negatively (61). A study explored the effect of ambient air pollution on individual levels of subjective wellbeing. This suggests that even relatively low levels of PM10 air pollution can adversely affect people's subjective assessments of well-being and its effects on physical health.

Based on this, we estimate the implication of air pollution on self-reported subjective wellbeing. The CHARLS asked each respondent about their recent life satisfaction. The self-reported future confidence is a discrete numerical variable that takes integer values from 1 to 5. Scoring values are as follows: completely satisfied, very satisfied, somewhat satisfied, not at all satisfied. The mean reported life satisfaction in the survey is 2.82 out of 5. The results are in columns (1) and (2) of Table 11. A one standard deviation rise in PM2.5 concentration results in a decrease of 0.034 units in life satisfaction, which means a decline in an individual's overall sense of wellbeing and contentment with life. Hence, it is crucial to improve air quality to enhance the quality of life for older adults. High PM2.5 concentrations can render the air turbid and reduce visibility, thereby restricting outdoor activities for older adults. This diminishes their comfort and convenience in daily life.

Cognitive ability predicts health outcomes, academic achievement, and mental health. Therefore, we try to figure out the effect of air pollution exposure on human cognition. This has significant implications for multiple areas of policy. The scope of cognitive evaluation includes date recognition ability, numeracy, and situational memory capacity (62). Specifically, in the assessment of date recognition, respondents were asked to answer the current day of the week, date, month, season, and year, with each correct answer earning 1 point, for a total of five points. In the numeracy assessment, respondents were asked to answer 100 minus 7, with five consecutive subtractions, where 1 point was awarded for each correct answer, totalling 5 points. In the assessment of drawing ability, the respondent draws the picture that they have seen and scores 1 point for a successful drawing. The assessment has a total score of 11, with higher scores representing greater cognitive ability. The results are displayed in columns (1) and (2) of Table 12. A one standard deviation increase in PM2.5 concentration is associated with a decrease in cognitive ability scores by 3.0361 for middle-aged individuals yet exhibits no significant impact on older adults. We found that air pollution impairs cognitive performance in both middle-aged and older age groups, but to a greater extent in older adults than in middle-aged adults.

Further, we construct an indicator of cognitive ability in the following Equation 3. CogIndex is the proportion of the overall number of indicators in which an individual has a health risk for each of the given health indicator ratings (63),

Where CogIndex_i is the cognitive ill-health index. N is the total score of the cognitive health index variable (in our research, N = 11). D_ij is the value corresponding to the jth variable reflecting the cognitive health ability of an individual i. D_ij = 1 means the jth cognitive ability variable is healthy. D_ij = 0 means that the jth cognitive ability variable is in a state of health risk. Therefore, the larger value of CogIndex_i indicates that the individual's cognitive ability is more excellent. The results are presented in columns (3) and (4) of Table 12. A one standard deviation increases in PM2.5 concentration is associated with a 0.3307 reduction in the cognitive index for older adults. Our findings suggest that exposure to air pollution is correlated with cognitive decline, which is further associated with an elevated risk of depression, particularly for older adults. Meanwhile, relevant studies have indicated that individuals exhibiting higher levels of depressive symptoms at the initial stage tend to experience a more pronounced cognitive decline over time, suggesting a bidirectional relationship (64, 65). This finding underscores the complexity of the interplay between air pollution, cognitive function, and mental health. The bidirectional relationship between cognitive decline and depression has important implications for understanding the impact of air pollution on mental health. While our study primarily focused on cognitive decline as a mechanism linking air pollution to depression, the reverse pathway from depression to cognitive decline cannot be ignored.

The high internet and mobile phone penetration in China provides a great opportunity to investigate pollution information using digital sources. The network significantly expands access to pollution information and elevates people's awareness about pollution issues. This information sharing may influence the avoidance behaviors of individuals who can take action. Like Google Trends, the Baidu Index is a data analysis tool provided by Baidu, the most widely used search engine in China. It allows users to see the research frequency of specific keywords in each city over a specified period since 2011. We focus on the search index for “haze pollution,” the internet catchphrase for air pollution, to illustrate personal access to the pollution information.

We estimate the moderating effects of private avoidance behaviors on the impact of air pollution on mental health. The results are shown in columns (1) and (2) of Table 12. The wider personal access to pollution-related information, the more effective they can be in counteracting the damage to mental health caused by air pollution, both for middle-aged and older adults. The results were not statistically significant for older adults. We assume that it may be related to the older population's lower proficiency in Internet use.

Vegetation, especially trees and shrubs, can act as a natural air filter. They capture particulate matter on their leaves and surfaces, which helps to reduce the concentration of these particles in the air. Hence, we estimate the filtering effect of urban vegetation density (we call the urban forest).⁵ We use the NDVI indicator (normalized difference vegetation index), which is a graphical indicator used to analyze remote sensing measures of vegetation density to measure whether a city is an urban forest city and investigate the modifying role of greenness.

The formula for NDVI is:

Where NIR is the near-infrared light reflected by vegetation, and RED is the visible light reflected. NDVI values range from −1 to +1. Healthy vegetation can reflect a lot of near-infrared light (NIR) and absorb most infrared light (RED). Based on this, a high value (closer to +1) indicates high green, healthy vegetation concentrations, while lower values (closer to −1) indicate sparse or unhealthy vegetation or non-vegetative surfaces.

The NDVI data are obtained from the NASA Moderate Resolution Imaging Spectroradiometer (MODIS) product MOD13Q1.061, which provides vegetation indices at a spatial resolution of 1 km. We obtained monthly data based on when and where respondents were interviewed. We further matched the data with the NDVI values of the respondents' locations at the time of the survey. It is generally accepted that an NDVI value above 0.6 indicates that the location is rich in vegetation and appears very green.⁶ Further, we defined a dummy variable for urban forest cities, where a value of 1 indicates that the NDVI value in the area is above 0.6 (i.e., the area has dense vegetation), and a value of 0 indicates otherwise. We examined the interactions of this variable with urban air quality variables. The results are displayed in columns (3) and (4) of Table 13; compared with those interviewees in non-urban forest cities, respondents are less mentally impaired by air pollution. The improvement of mental health for older adults is more pronounced than for the middle-aged. The evidence points out that greenness, as a nature-based intervention, can mitigate the residential air pollution of physiological stress.

Furthermore, we have conducted additional calculations to explore the marginal effect of high vegetation density (i.e., NDVI = 1) on the CES-D scores of middle-aged and older adults under varying concentrations of PM2.5. As illustrated in Figure 6, which presents the marginal effects of the interaction term, the influence of high vegetation density on the marginal effect of PM2.5 on the CES-D scores of middle-aged and older adults persists as PM2.5 concentrations increase. This effect significantly mitigates the CES-D scores, thereby improving psychological health status, although the magnitude of change is not substantial. These findings further verify the notion that a higher degree of greening can effectively ameliorate the impact of air pollution on the mental health of the middle-aged and older adults.