Impact of Natural Disasters on Mental Health: A Cross-Sectional Study Based on the 2014 China Family Panel Survey

Mental health problems are a leading cause of disability in both developed and developing countries, and the consequences of mental health disorders for individuals, families, and society as a whole could be severe and costly. To supplement relevant research and provide insightful policy suggestions to families, government and societies, this study investigates the nexus between natural disasters and mental health for middle-aged and older adults in rural China. Based on data of 8721 observations from the 2014 China Family Panel Studies, we estimate the effects of natural disasters on mental health using ordinary least squares and propensity score matching. Our findings suggest that natural disasters have a significant negative effect on middle-aged and older adults’ mental health in the case of rural China. This effect is heterogeneous depending on individuals’ education level and their agricultural production status. Finally, individuals’ happiness and life satisfaction are shown to be the potential mechanism through which the effect of natural disasters on mental health operates.


Introduction
Natural disasters, which are a possible result of global warming, play a crucial role in the relationship between humans and nature. For this reason, natural disasters have been widely studied, with researchers exploring their impact on society and aspects such as household finance, poverty, family violence, the macroeconomy, and energy consumption [1][2][3][4][5][6]. It is obvious that natural disasters threaten people's lives and physical health; however, little attention has been paid to their impact on mental health. Natural disasters can cause anxiety, sleep disturbances, impaired interpersonal relationships, and depression, among other mental problems [7][8][9].
The importance of mental health is confirmed by the World Health Organization [10], which states that: "Health is a state of complete physical, mental and social well-being and not merely the absence of disease or infirmity". Mental health is closely related to daily life and work, and it affects our attitude to life and work productivity. Research also indicates that mental illness can be costly for individuals and society [11][12][13][14][15][16]. Serious mental illness can even lead to suicide. Globally, around 703,000 people die by suicide every year (World Health Organization, 2021) [10]. Hence, exploring the causes of mental health problems is crucial for the well-being of individuals, their families, and society as a whole.
Given the importance of mental health, the determinants of mental health problems attract scholars' interests. Most of the research focuses on the impact of human activity on mental health. For example, Ettner [17] found that an increase in individuals' income can improve their mental health. In contrast, unemployment has a negative impact on mental health, a relationship that has been investigated by Scutella and Wooden [18]. Furthermore, Hypothesis 1. Natural disasters have a significant negative effect on middle-aged and older adults' mental health.
Belo et al. [44] found that well-educated people tend to have a higher income, a healthy diet, and an optimistic attitude towards life. Natural disasters might destroy immovables, cause massive loss of human life, and destruction of resources. Compared with lesseducated individuals, well-educated people own more social and economic resources. Those well-educated individuals could better cope with the negative impact of natural disasters. The effect of natural disasters on mental health might not be homogeneous for people at different education levels. Hence, Hypothesis 2 arises. Hypothesis 2. The impact of natural disasters for well-educated individuals is less strong than it is for their less-educated counterparts.
Most individuals have sustained heavy financial losses due to natural disasters. People involved in agricultural production suffer more losses from natural disasters [45]. Furthermore, property loss induces anxiety or other mental health problems in these people. Second, compared with the individuals who are not involved in agricultural production, natural disasters can be more devastating for those who are. Since the damage affects not only financial property, but also people's agricultural livelihoods [3], the double loss might result in mental health problems. The effect of natural disasters on mental health might also vary depending on the family's agricultural production status. Hence, we propose Hypothesis 3.

Hypothesis 3.
Individuals show a stronger response to natural disasters if they have a family member involved in agricultural production, compared to those who do not.
Individuals with a higher level of happiness have more positive emotions and attitude to life than the ones with a lower level. Previous studies have recognized the important role of happiness in an individual's mental health [46]. Furthermore, the existing literature indicates that natural disasters have a significant negative impact on individuals' happiness [47,48]. Hence, we propose Hypothesis 4.

Hypothesis 4.
Natural disasters have an impact on mental health through their effects on happiness.
Natural disasters are linked with reduced satisfaction. Effects of natural disasters on life satisfaction fall into two broad categories: psychic costs and financial losses. Luechinger and Raschky [49] found that flood disasters have a negative effect on individuals' life satisfaction. Individuals' life satisfaction scores embody specific information on a subjective assessment of their daily life. Respondents with a higher score of life satisfaction are less likely to experience a psychological problem. Hence, Hypothesis 5 is proposed.

Hypothesis 5.
Natural disasters can harm mental health through their effects on life satisfaction.

Statistical Analysis
Statistical analysis was conducted using econometric software STATA version 15.1 (StataCorp, College Station, TX, USA). We report the mean, standard deviation, minimum, and maximum of variables in Table 1. Given mental health is a continuous variable, OLS was constructed to investigate the causal relationship between natural disasters and middleaged and older adults' mental health. (We used the STATA package "regress" for the OLS regression). In our robustness check, we estimate the effect of natural disasters on mental health using propensity score matching (PSM). (We used the STATA package "psmatch2" to calculate the average treatment effect on the treated (ATT) of the various propensity score matching methods). To investigate the mechanisms, we estimate the impact of natural disasters on individuals' happiness and life satisfaction using OLS and the ordered probit model. (We used the STATA package "oprobit" for the ordered probit model). All reported p-values were two-tail. The level of statistical significance was set at p < 0.1.

Variables and Descriptive Statistics
Outcome variable: middle-aged and older adults' mental health The main outcome variable in this paper is the mental health of middle-aged and older adults in rural China. Following existing studies [43,50], the mental health index is derived from the 6-item short form of the Center for Epidemiologic Studies of Depression (CES-D) in the CFPS. (CES-D questions: 1. How often did you feel depressed that nothing could cheer you up during the past 30 days? 2. How often did you feel nervous during the past days? 3. How often did you feel restless or fidgety during the past 30 days? 4. How often did you feel hopeless during the past 30 days? 5. How often did you feel that everything was an effort during the past 30 days? 6. How often did you feel that life was meaningless during the past 30 days? Individuals were asked to indicate the frequency of their feelings on a five-scale metric-"Almost daily", "Often", "Half of the time", "Sometimes", and "Never". These responses are coded from 1 to 5, respectively). The response for each question is coded from 1 to 5. There are six questions to assess mental state in the survey, and each one is constructed and standardized to have a mean of zero and a standard deviation of one. The final score is calculated by aggregating the multiple measures into indices. The higher the index value, the better the individual's mental health.
Independent variable: natural disaster We consider two measures of natural disaster as the independent variable. The first one is captured by a dummy variable (Disaster_d). It equals 1 if the middle-aged or older adult has experienced at least one type of natural disaster, and otherwise 0. (The types of natural disasters include typhoons, floods, storm surges, forest fires, frost, hail, landslides, debris flow, earthquakes, infectious diseases, agricultural and forestry pests, etc.). The second is constructed as a continuous variable (Disaster_n), which measures the number of types of natural disaster that the middle-aged or older adult has experienced.
Control variables and descriptive statistics We include the following control variables: age, a dummy variable for sex, education level, marital status, cognitive abilities, income, medical insurance, and a dummy variable for agricultural production. In addition, we control for family size, house value, and family expenditure. Descriptive statistics of the variables used in the paper are reported in Table 1, where it can be seen that the sampled middle-aged and older adults were 58.41 years old on average, and 50.4 percent of them were male. The average mental health score is −0.339. About 75 percent of middle-aged and older adults have experienced at least one type of natural disaster. The value of Disaster_n varies from 0 to 5. That is to say, the most types of disasters that have been experienced by a person is 5, and the least is 0 in our sample.

Empirical Methodologies
The effect of natural disasters on middle-aged and older adults' mental health is estimated using ordinary least squares, as follows: where mental i represents the dependent variable (middle-aged and older adults' mental health), disaster_d represents the natural disaster dummy variable (dummy variable equal to 1 if the middle-aged or older adult experienced at least one type of natural disaster, and otherwise 0), disaster_n represents the number of times a natural disaster was experienced, and control i is a vector of observable determinants of middle-aged and older adults' mental health.

The Basic Correlation
The basic relationship between natural disasters and mental health is presented in Figure 1. The graph indicates that a negative correlation exists between natural disasters and middle-aged and older adults' mental health. family expenditure. Descriptive statistics of the variables used in the paper are reported in Table 1, where it can be seen that the sampled middle-aged and older adults were 58.41 years old on average, and 50.4 percent of them were male. The average mental health score is −0.339. About 75 percent of middle-aged and older adults have experienced at least one type of natural disaster. The value of Disaster_n varies from 0 to 5. That is to say, the most types of disasters that have been experienced by a person is 5, and the least is 0 in our sample.

Empirical Methodologies
The effect of natural disasters on middle-aged and older adults' mental health is estimated using ordinary least squares, as follows: where i mental represents the dependent variable (middle-aged and older adults' mental health), _ disaster d represents the natural disaster dummy variable (dummy variable equal to 1 if the middle-aged or older adult experienced at least one type of natural disaster, and otherwise 0), _ disaster n represents the number of times a natural disaster was experienced, and i control is a vector of observable determinants of middle-aged and older adults' mental health.

The Basic Correlation
The basic relationship between natural disasters and mental health is presented in Figure 1. The graph indicates that a negative correlation exists between natural disasters and middle-aged and older adults' mental health.

Baseline Results
When investigating the causal relationship between natural disasters and mental health, an individual's math and language abilities are generally highly correlated. High correlation among variables gives rise to concerns about multicollinearity, which may lead to considerable bias in the estimation. We use the variable inflation factor (VIF) to check for multicollinearity in our model. Table 2 reports the VIF of each variable. In each case, the VIF is less than the rule-of-thumb value of 10, indicating that multicollinearity is not a major issue.  Table 3 reports the baseline results on the effects of natural disasters on middle-aged and older adults' mental health. Columns (1) and (3) include only the dummy of natural disasters and the intensity of natural disasters, respectively. A set of control variables affecting middle-aged and older adults' mental health is included in columns (2) and (4). The effects in columns (1) and (3) suggest a salient negative effect of natural disasters on mental health for middle-aged and older people. When controlling for a set of covariates in columns (2) and (4), results from OLS models indicate that natural disasters are a significant predictor of middle-aged and older adults' mental health, showing a negative correlation. Those results verify Hypothesis 1. In addition, sex shows a positive sign in columns (1) and (3). This indicates that males have better mental health than females. The results for education report positive signs, indicating that education has a positive impact on mental health. The coefficients of marital status are positive and statistically significant. The results indicate that the mental health status of married adults is higher than in their unmarried counterparts. Math abilities, income, insurance, and house value show a salient positive impact on mental health.

Endogeneity
Bearing selection bias in mind, we estimate the causal effect of natural disasters on mental health using the propensity score matching (PSM) technique. In this case, we use a dummy variable equal to 1 if the middle-aged or older adult experienced at least one type of the natural disaster (treatment group), or otherwise 0 (control group).
An important step when applying PSM is to check the covariate balance of the treatment and control group, which is achieved if both groups have similar observable covariates. This paper uses two methods to check the covariate balance of the two groups. The first one is essentially based on comparing the mean (after matching) of observable covariates in the two groups. The second one is based on the standardized bias. Table 4 reports the results of the mean of the observable covariates in the two groups. The results in column (5) indicate that the p-values (after matching) are larger than 0.1 in most of the cases. Additionally, we report the standardized bias in Figure 2. The standardized bias reduction is below 5%, providing evidence that the covariates are balanced in the two groups.   According to Heckman et al. [51], a crucial step when applying PSM is to examine the overlap and region of common support between treatment and control groups. Figures 3  and 4 report the estimation of the density distribution in the two groups, indicating that most samples fall into the region of common support. According to Heckman et al. [51], a crucial step when applying PSM is to examine the overlap and region of common support between treatment and control groups. Figures 3 and 4 report the estimation of the density distribution in the two groups, indicating that most samples fall into the region of common support.   According to Heckman et al. [51], a crucial step when applying PSM is to examine the overlap and region of common support between treatment and control groups. Figures 3 and 4 report the estimation of the density distribution in the two groups, indicating that most samples fall into the region of common support.   Following Rosenbaum and Rubin [52], this paper presents different types of matching estimators, including kernel matching, local linear matching, radius matching, and nearest-neighbor matching (k = 1, k = 4). Table 5 presents the results of the different matching strategies. The results indicate that the average treatment effect on the treatment of the various matching methods is all negative and significant, indicating a negative correlation between natural disaster and mental health in our sample.

Robustness Check
Two lines of the robustness check are conducted to confirm the reliability of the results in the previous section. In the first, we use an alternative measure of mental health. In the second, we employ another methodology to estimate the effect of natural disasters on middle-aged and older adults' mental health.
One concern may be that our findings might be driven by the measurement of mental health. To analyze this, we construct an alternative index to measure mental health in our sample. We adopt a factor analysis approach to assess the individual's mental health. The results show that the KMO-statistics are all larger than 0.8, Cronbach's alpha is 0.86, and the p-values of the Bartlett test of sphericity are all less than 0.01, confirming that exploratory factor analysis fits well as a method to measure mental health. We also use this index to regress our model using OLS, and the results are reported in Table 6. The results indicate that natural disasters have a negative effect on middle-aged and older adults' mental health, consistent with the OLS results of Table 3. In the second robustness check, we apply the probit methodology to estimate the effect of natural disasters on middle-aged and older adults' mental health. To do this, we replace individuals' mental health by using a dummy. This dummy takes 1 if the value of mental health is more than −0.339, and otherwise 0. Columns (3)-(4) of Table 6 show the estimated results, which indicate that natural disasters have a negative effect on mental health. We also report the margin effects in columns (5)-(6) of Table 6. The marginal effects are all negative and statistically significant. All in all, the results in Table 6 are consistent with the results in Table 3.

Heterogeneity
To better understand the relationship between natural disasters and mental health, we examine the heterogeneity of effects by splitting the sample into different education levels and agricultural production status.
In order to check Hypothesis 2, Table 7 presents the results of the heterogeneous effect of natural disasters on mental health for different education levels. The results suggest that less-educated adults show a stronger response to natural disasters than well-educated ones. These results verify Hypothesis 2. To check Hypothesis 3, we include a dummy variable to measure the agricultural production status in a family. The dummy equals 1 if the individual belongs to a family involved in agricultural production, and otherwise 0. Table 8 reports the effect of natural disasters considering the family's agricultural production status. The results indicate that middle-aged and older adults have a stronger response to natural disasters if they have a family member involved in agricultural production, compared to those that do not. The results in Table 8 verify Hypothesis 3.

Mechanisms
To explore the mechanisms through which natural disasters affect middle-aged and older adults' mental health, two channels are studied in this section: happiness and life satisfaction.
To test Hypothesis 4, we estimate the impact of natural disasters on happiness for middle-aged and older adults by means of OLS. (The happiness index ranges from 1-10, where 0 is the least happy and 10 is the most happy). Given that happiness is reported on an ordinal scale, we also employ the ordered probit model to investigate the impact of natural disasters on mental health. Table 9 reports the results for the effect of natural disasters on middle-aged and older adults' happiness, revealing negative and statistically significant coefficients. This indicates that natural disasters have an impact on middle-aged and older adults' mental health through their effects on happiness. The results in Table 9 verify Hypothesis 4. In order to test Hypothesis 5, we investigate whether natural disasters can affect life satisfaction. This indicator is also available in the CFPS survey, with a higher value meaning higher life satisfaction. (The life satisfaction index ranges from [1][2][3][4][5]. The corresponding estimates are presented in Table 10. Columns (1)-(2) of Table 10 report the results of the OLS. Life satisfaction is reported on an ordinal scale, which allows us to estimate the effects of natural disasters on mental health with the ordered probit model. The results show that the coefficients of the natural disasters are negative and statistically significant, indicating that natural disasters can harm mental health through their effects on life satisfaction. These results verify Hypothesis 5.

Discussion
There are a number of studies that are related to what we have examined in this paper, but they are in the spirit of focusing on a specific disaster. For instance, Kovats and Hajat [40] conducted a meta-analysis of previous studies and found that extreme hot weather threatens public health and can also be a cause of mortality. Furthermore, studies by Rataj et al. [41] and Weilnhammer et al. [53] show that extreme weather has a negative impact not only on physical health, but also on mental health. However, the abovementioned research is based on descriptive studies and lacks empirical support. Our research employs ordinary least squares and propensity score matching to investigate the causal impact of natural disasters on middle-aged and older adults' mental health and provides empirical evidence on the effects of natural disasters on mental health. The baseline results are in line with previous studies [54,55]. Moreover, most of the research is derived from studies of flood-exposed regions. Unlike the research of Fernandez et al. [56], our research is derived from large-scale micro population survey data (CFPS). Based on Adult Psychiatric Morbidity Survey data in England, Graham et al. [42] investigated the impact of storms and floods on individuals' mental health, but their paper makes no attempt to provide the mechanism analysis. Our research not only investigates the impact of natural disasters on individuals' mental health, but also strives to ascertain the mechanism between natural disasters and mental health. Furthermore, previous studies found that experiencing an earthquake may influence sleep quality and interpersonal relationships, or even lead to suicide [7,8].
Despite mounting evidence indicating that heat, floods, and hurricanes might cause a negative impact on individuals, little has been said to discuss the impact of all kinds of natural disasters as an external shock on middle-aged and older adults' mental health. Furthermore, our research also indicates that the impact of natural disasters on middle-aged and older adults' mental health is heterogeneous depending on individuals' education level and their agricultural production status. Our study found that well-educated individuals have a weaker response to natural disasters than their less-educated counterparts. Middleaged and older adults show a stronger response to natural disasters if they have a family member involved in agricultural production, compared to those that do not. Last but not least, our findings provide new evidence on the causal mechanism between natural disasters and middle-aged and older adults' mental health.
However, this paper is limited in some facets. First, we estimate the short-run effects of natural disasters on middle-aged and older adults' mental health. Regretfully, due to data constraints, we fail to consider the long-term effects of natural disasters. Second, as well as the data limitations, we measure mental health in a very general way. For instance, posttraumatic stress disorder (PTSD) is highly related to disaster survivors [29,55]. Given the lack of relevant data to PTSD, we do not investigate the impact of natural disasters on PTSD. Third, although we try our best to include the factors that might affect individuals' mental health, the model could not include some further external factors that affect individuals' mental health, which are difficult to measure. An interesting future research avenue could be projected on the long-term and dynamic effects of natural disasters on middle-aged and older adults' mental health. Furthermore, research could also investigate the causal effect between natural disasters and a specific mental problem, such as PTSD.
Several policy implications can be derived from this analysis. First, our study suggests that natural disasters have a notable adverse impact on middle-aged and older adults' mental health. Thus, the government and society as a whole might need to provide aid to the middle-aged and older adults who have suffered from natural disasters. This help should be targeted not only at infrastructure reconstruction and financial subsidies but also at effective mental health care. Particular attention should be paid to those people who have a low level of education and are involved in agricultural production. Finally, the government might also focus more on helping the middle-aged and older adults of disaster-stricken regions by improving their happiness and life satisfaction.

Conclusions
Given the importance of mental health in daily life, there has been a growing amount of research on this topic. In this paper, we investigate the causal relationship between natural disasters and mental health in the case of middle-aged and older adults in rural China by using 8721 observations from 2014 CFPS survey data. One of the most important findings to emerge from this paper is that natural disasters have a negative impact on mental health.
Further analysis on heterogeneous effects is conducted by splitting the sample according to educational level and family agricultural production status. On the one hand, the results show that natural disasters have a slightly stronger impact on less-educated people than their better-educated counterparts. On the other hand, compared with those whose family members are not involved in agricultural production, those who are involved in agriculture show a stronger response to natural disasters. Our study also investigates the mechanisms through which natural disasters can have an impact on mental health, indicating that they influence mental health through their effect on the individual's level of happiness and life satisfaction.