Optimal Residence: Does Air Quality Affect Settlement Decisions of Urban Floating Migrants?

Abstract

There is a strong association between urban prosperity and human resources. Rapid urbanization and industrialization have caused severe air pollution to the urban cities over the past decades, and permanently moving to locations with clean air has been considered to be one of the most effective strategies to avoid air pollution. However, whether this is in fact still unclear. In order to fill the research gaps, this study combines individual-level data derived from a national survey and city-level data obtained from various sources to empirically exam the effect that air quality has on settlement decisions of urban floating migrants in China. We focus on the target floating migrants who have urban hukou in the origin city and housing in the destination city, and investigate the relationship between air quality and settlement decisions from the move-in and move-out city perspective. In addition, we exam not only the willingness of urban floating migrants to settling in the move-in city, but also their actual migration decision in the reality. The bivariate probit analysis and panel data regression are applied. Results show that poor air quality in the move-out cities has a significant positive impact on urban floating migrants’ willingness to settling in the move-in cities, while bad air quality in the move-in cities has a significant negative impact, indicating that poor air quality does act as a push force to drive floating migrants away but good air quality can be a pulling force for floating migrants to settling down. In addition, the robust tests show good agreement with the main regression results, with almost all relationships reported having the same direction as those indicated by the regression analysis. Our findings can help policy makers and urban planners assess and design interventions to improve the urban air quality aimed at attracting talent, with broader implications for urban sustainable development.

1. Introduction

There is a strong association between urban prosperity and human resources [1,2,3]. Rapid rural to urban migration (termed as urbanization) has happened in the world over the past decades and it is estimated that urban residents will account for 67% of the world population in 2050 [4]. For decades, industrial economic activities were prioritized over environmental sustainability in order to satisfy the needs of the increasing urban population, and this has caused severe air pollution. It is estimated that air pollution has caused approximately seven million deaths in the world [5], and this problem is more prominent in developing countries, such as China. Permanently moving to locations with clean air has been considered to be one of the most effective strategies to avoid air pollution [6,7,8,9]. However, whether this is in fact still unclear, especially in a developing country, such as China, where permanent migration is costly as Chinese citizens are facing many restrictions (e.g., hukou (The hukou system is the Chinese household registration system initially established to control the influx of rural migrants to cities. Living in a region with a local resident hukou is associated with many social benefits, such as free compulsory education for children, junior/senior high school entrance examination locations, and social security services. Migrants without a local hukou need to pay additional operating and maintenance fees to the public school to use educational resources that are intended for local residents.) and self-owned housing) before being allowed to access to local public services such as health care and public schools [10,11,12]. This study aims to empirically investigate the association between air quality and settlement decisions of urban floating migrants who have urban hukou in the origin city and self-owned housing in the destination from both move-in and move-out city perspective. The rural-to-urban urbanization is the initial stage of migration and the advanced stage is the intercity migration which is becoming observable as that of rural-to-urban urbanization in developing countries [13], and the intercity migration trend has continued in industrialized nations such as the United States, therefore, our research can be generalized to a wider community. In addition, we exam not only the willingness of floating migrants to settling in the destination city but also their actual migration behavior in practice.

Based on the economic theory, the factors that affect people’s settlement decisions can be classified as economic and non-economic. The economic determinants mainly include income, public goods and services (e.g., school and education, hospital and health services, social welfare and public infrastructure), and people tend to move to a place with high income and sound public goods and services [14,15,16]. The non-economic factors refer to those can satisfy multi-needs valued by the floating migrants, including housing conditions, work environment, comfort, and air quality. A body of literature have acknowledged the impact that economic factors have on settlement decisions of migrants [17,18,19]; however, the current literature investigating the impact of air quality on migration mainly focused on overall population [20] or rural migrant workers [21], and aimed at estimating the net air quality effect (air quality difference between the origin and destination city) [22,23]. Few studies have paid attention to urban floating migrants and examined the association between air quality and both willingness and actual settlement decisions of urban floating migrants, and from the perspective of origin and destination cities [1,24]. This study aims to fill the research gaps and provide significant implications to policies.

We focus on a specific group of urban floating migrants who have urban hukou in the move-out city and self-owned housing in the move-in city. There are a number of reasons for selecting such a group of migrants. First, floating migrants who have urban hukou in the origin city and self-owned housing in the move-in city have good financial status and sufficient leisure time. It is this group of people who are likely to be affected by the non-economic factors (e.g., air quality) when making migration decisions. Previous research has found that it is the destination’s good air quality that has an impact on intercity migration of skilled labor and wealthy residents, rather than job opportunities that continue to be the main reasons for rural residents to flow into cities [25,26]. Second, to own knowledge, few studies have investigated the impact of air quality on settlement decisions of this specific group of floating migrants. Third, it is this group of urban floating migrants that makes great contribution to urban development and prosperity. It is estimated that the private consumption made by China’s middle class is expected to reach US$2.3 trillion by 2022, accounting for 24% of its GDP [27]. Fourth, if the research objects were not correctly defined, the estimates of the impact of air quality on migration decisions may be biased. Therefore, our research draws public attention on this target urban floating migrants and investigates the association between their settlement decisions and air quality.

Based on Lee’s push-pull theory [28], this study investigates the association between air quality and individual’s settlement decisions from both push and pull sides other than merely from net push-pull aspect. That is to say, we investigate whether the bad air quality in initial residential city acts as a push force to drive urban residents away while the good air quality in destination city acts as a pull force to attract people to reside. This message meets the theoretical basis and is vital to understand whether the settlement decisions of urban floating migrants are caused by air quality.

In addition, we exam not only the effect that air quality has on urban floating migrants’ willingness to settling in the destination city where they migrate to but also on their actual migration behavior in the reality. More specifically, we further investigate the associations between air quality and expected long-term settlement and actual migration decisions in the robust tests. This message is very important to fully understand settlement decisions of urban floating migrants. Data is derived from China, the world’s second largest economy with a rapid urbanization and can be a representative of understanding the association between air quality and settlement decisions in the globe.

This study makes four contributions to the literature. First, this is one of the first studies to investigate urban residents’ reaction to air pollution from the perspective of migration. In comparison to rural migrant workers, urban citizens are more likely to suffer from air pollution and those who have housing in the move-in city have a good financial situation to afford the costly migration [22]. Second, we exam not only the impact that air quality has on willingness to settling, but also on the actual behavior of settlement decisions. In the existing literature, studies have investigated either individual’s willingness to settling in the destination city [29], or actual migration behavior [7], but few has considered both of them together. Third, in order to address the endogeneity concerns, we employ ventilation coefficient and precipitation as the instrumental variables of air quality. Finally, the results empirically reveal that floating migrants who have urban hukou at the origin city and housing in the destination do chase clean air and avoid the air pollution in the reality by leaving the origin place with bad air quality and moving to a location with good air quality. This information will assist urban planners and policy makers to make decisions in terms of bringing interventions to improve the urban air quality to attract the high-quality workforce to work and reside in the city.

The rest of the paper is structured as follows: Section 2 demonstrates the method that we developed to empirically investigate the topic, Section 3 presents the results, Section 4 is the discussion, and Section 5 concludes the study.

2. Methods

2.1. Conceptual Framework

We base our research on Lee’s migration model [28] that describes the push and pull factors of migration which are basically reasons for emigration and immigration. A push factor is something that is unfavorable about the area that someone lives in and is a reason for them to leave. A pull factor is a factor that attracts someone into an area. This model does not isolate certain push and pull factors but the factor could be environmental. We propose a conceptual framework (Figure 1) investigating whether the bad air quality acts as a push force in emigration city that forces people to leave, and the good air quality acts as a pull factor in immigration city that attracts people to reside. Inspired by the theory of planned behavior (TPB), we divide settlement decisions into two categories: willingness to settling and actual migration behavior [30]. As the expected long-term settlement has also been applied to studies of migration, we include it as a separate type of settlement decision parallel to willingness to settling and actual migration behavior. We first investigate the effect that air quality has on urban floating migrants’ willingness to settling in the destination city, and followed by the robust tests of the association between air quality and expected long-term settlement and actual migration behavior. In case of double counting, we only consider the urban floating migrants in destination city. It is this group of people that is pushed by the poor air quality in the origin city and pulled by the good air quality in the destination city.

2.2. Econometric Models

To estimate the association between air quality and willingness to settling in the destination city, the following bivariate probit model is introduced:

$$W{S}_{ij}={\beta }_0+{\beta }_1{P}_i+{\beta }_2{P}_j+{\beta }_{3}X+{\gamma }_{ij}+{\epsilon }_{ij}$$

(1)

where $W{S}_{ij}$ refers to the willingness to settling down in destination city $j$, after migrating from origin city $i$. It has two target values, 1 and 0, and 0 refers to ‘no willingness to settling in the destination city’, and 1 means there is a willingness. Since the dependent variable is recorded as a dummy variable, a bivariate probit model of Equation (1) is employed to exam the relationship between air quality and willingness to settling. The explanatory variable $P_i$ is air quality in origin city $i$ and the corresponding coefficient under this setup is ${\beta }_1$. $P_j$ is the air quality in move-in city $j$, and the coefficient of interest is ${\beta }_2$. $X$ is a vector of control variables at both the individual and city levels, and the corresponding coefficient is ${\beta }_3$. ${\gamma }_{ij}$ is the city fixed effects to control for fixed characteristics including city size, city economic difference between city $i$ and $j$. The idiosyncratic error is represented by ${\epsilon }_{ij}$.

Based on the conceptual framework demonstrated in Section 2.1, we expect ${\beta }_1$ to be negative if move-out city $i$ had good air quality, while ${\beta }_2$ to be positive if move-in city $j$ had good air quality. The air pollution in move-out city $i$ acts as a push factor that forces people to leave, while the sound air quality in destination city $j$ can be considered a pull factor that attracts people to move in and settling down.

In order to address the concern over endogeneity, we apply an instrumental variable linear regression model to obtain the unbiased estimates of the association between air quality and settlement decisions. Specifically, the key endogenous independent variable, air quality, is regressed on the instrumental variables, ventilation coefficient ($VC$) and precipitation, and other control variables at the city level. The instrument variable regression model is as follows:

$$P={\delta }_0+{\delta }_{1}VC+{\delta }_{2}precipitation+{\delta }_{3}X+\epsilon$$

(2)

where $P$ is the air quality in either move-out or move-in city. $VC$ and $precipitation$ are the instrumental variables ventilation coefficient and precipitation, and ${\delta }_1$ and ${\delta }_2$ are the corresponding parameters, respectively, to be estimated. $X$ represents a vector of control variables at both the individual- and city- level, and the coefficient is ${\delta }_3$. $\epsilon$ is the error term.

Moreover, we conduct the robust tests using panel data regression model:

$$S{L}_{jt}={\alpha }_0+{\alpha }_1{P}_{jt-1}+{\gamma }_j+{\epsilon }_j$$

(3)

$$A{M}_{jt}={\phi }_0+{\phi }_1{P}_{jt-1}+{\gamma }_j+{\epsilon }_j$$

(4)

where $S{L}_{jt}$ refers to the expected long-term settlement of urban floating migrants in the destination city. $A{M}_{jt}$ refers to actual migration behavior. $P_{jt-1}$ is the air quality in destination city $j$ in the year $t-1$. ${\alpha }_1$ and ${\phi }_1$ are the key parameters to be estimated. ${\gamma }_j$ refers to the city fixed effects. ${\epsilon }_j$ is the idiosyncratic error. We predict ${\alpha }_1$ and ${\phi }_1$ to be positive if move-in city $j$ had good air quality, otherwise negative if destination city $j$ had bad air quality.

2.3. Variables and Data Sources

2.3.1. Air Quality

Air quality refers to the concentration of pollutants in the air and is measured by Air Quality Index (AQI) accepted by China Environmental Protection Agency and commonly used by many literature [23]. Considering the hysteresis relationship between migration and changes in air quality, that is, the settlement decisions of floating migrants depend not only on the current state of air quality, but also upon its past history, we select 2016 as a reference year. That is to say, we assume that the settlement decisions in 2017 were influenced by air quality in the previous year, 2016. The AQI value is calculated by using the method provided by Ambient Air Quality Index Regulations (Trial), and it is the annual average concentration of five pollutants, particulate matter 2.5 (PM2.5), particulate matter 10 (PM10), sulfur dioxide (SO₂), nitrogen dioxide (NO₂), and ozone (O₃), all of which are the main components of haze. A high AQI value implies poor air quality, while a low AQI value indicates good air quality. There are 113 cities above prefecture-level included in the analysis, and the floating migrants migrated across county but within the prefecture-level cities are not included in the sample. The data is obtained from the China Environment Statistical Yearbook (CESY). The calculation formulation is presented below:

$$AQI=max\left\{IAQ{I}_1,IAQ{I}_2,IAQ{I}_3,\dots ,IAQ{I}_n\right\}$$

(5)

$$IAQ{I}_P=\frac{IAQ{I}_{Hi}-IAQ{I}_{Lo}}{B{P}_{Hi}-B{P}_{Lo}}\left(C_P-B{P}_{Lo}\right)+IAQ{I}_{Lo}$$

(6)

where $AQI$ refers to air quality index; $n$ represents the number of pollutants and there are five items in total in this study; $IAQ{I}_P$ refers to the index of air quality of an individual pollutant $P$; $C_P$ defines the concentration value of the pollutant item $P$; $B{P}_{Hi}$ refers to the maximum value of $C_P$; $B{P}_{Lo}$ refers to the minimum value of $C_P$; $IAQ{I}_{Hi}$ refers to the sub-index of air quality in accordance with $B{P}_{Hi}$; $IAQ{I}_{Lo}$ means the sub-index of air quality corresponding to $B{P}_{Lo}$.

2.3.2. Willingness to Settling

We define willingness to settling in the destination city as a psychological tendency to permanently reside in a city. In order to measure the willingness to settling in the move-in city, we use two stepwise survey questions and the first being ‘Do you plan to stay here for some time from now on?’ The answer has three options: ‘yes’, ‘no’ and ‘undecided’. Respondents who choose ‘yes’ are then asked to answer the second question: ‘How long do you plan to stay in this city?’ The answer has six options and the first one is 1–2 years, the second 3–5 years, the third 6–10 years, the fourth more than 10 years, the fifth permanently, and the sixth undecided. We categorize those who select the fifth option as willing to settle in the destination city, while those who choose one of the first four options as unwilling. We drop those who have not decided how long to stay in the move-in city. Because those undecided floating migrants could fall into any of the top five options, or none at all. In addition, we code the permanent settlement intention of floating migrants as ‘1’ and otherwise ‘0’. We do not classify those who select the first option as the least willingness and the fifth the strongest, as we feel that having or not having a willingness to settling in the destination city is in line with the topic of this study.

The data is derived from China Migrants Dynamic Survey (CMDS) provided by the Migrant Population Services Center of the National Health Commission of the People’s Republic of China. In order to assist policies aimed at improving the living conditions of the floating migrants, a nationwide survey covering all of the provinces of mainland China has been conducted by the National Health and Family Planning Commission of People’s Republic of China every year since 2009. Questionnaire is the main form of the survey. The objects of the survey are those who are aged more than 15 and have been living in the destination cities for more than one month, but without local hukou. The questionnaire contains five parts: (1) personal information (e.g., family members and income), (2) employment status, (3) willingness to leave or to stay in the destination city, (4) health and public services, and (5) social network. There are about ten questions in each section, designed to capture the exact information of the floating migrants. The objects of the survey are selected based on the probability proportionate to size sampling techniques. This dataset contains records of 169,989 floating migrants across all cities above prefecture-level in China. The investigators are experienced and have received intensive training before conducting the investigation to ensure that they have an accurate and unified understanding of the purpose and content of the investigation, and master survey skills.

We employ this data source for a number of reasons. First, it provides detailed internal migration information including floating migrants’ individual socioeconomic characteristics (e.g., hukou identity, gender, education, and income), housing situation (e.g., own housing or rented), employment status (e.g., permanent or non-permanent job), which may have an impact on settlement decisions of floating migrants. Second, it includes the very group of floating migrants who have urban hukou in the origin city and housing in the destination city. These people are more likely to be affected by air quality and are of interest of this paper. Third, it is reliable as the survey is conducted by Chinese official authorities and covered mainland China. Fourth, much literature has used this database [17,23,31,32,33,34].

2.3.3. Expected Long-Term Settlement

We define the expected long-term settlement as the willingness to stay in the destination city for more than five years. It is measured by the percentage of urban floating migrants who are willing to stay in the destination city for more than five years that accounts for the total amount of urban floating migrants investigated in the move-in city in a certain year. We use two steps to measure the expected long-term settlement. In the first stage, we aggregate the respondents who report that they are planning to stay in the relocated city for more than five years. We then divide the aggregates by the total amount of urban floating migrants surveyed in the move-in city that year. In the aggregation stage, we use the questionnaire and the question is: ‘Have you planned to stay here for a long time (more than five years)?’, which has three alternatives: ‘yes’, ‘no’, and ‘undecided’. We aggregate those who choose ‘yes’ as the number of urban floating migrants who are willing to settling in the destination city for a long time.

2.3.4. Actual Migration Behavior

Actual migration behavior refers to the migration of floating migrants from one city to another. This variable is measured by the amount of floating migrants who have urban employees’ medical insurance or public health insurance (noted as UEMIPI unless otherwise stated) in a certain city that accounts for the total number of floating migrants with UEMIPI in the 105 (Due to the lack data on ventilation and precipitation in 8 cities, there are 105 above prefecture-level cities included in this context) cities. This percentage is an absolute number and cannot reflect the attractiveness of a city, therefore, we employ the difference of this percentage between two consecutive years as a proxy for the dependent variable of actual migration behavior. The higher the proxy, the more floating migrants migrate to, and the more attractive the destination city. The formulation of measuring the actual migration behavior is as follows:

$$\begin{gathered} \frac{Number of floating migrants that obtain UEMIPI in a city in the Tth year}{Total number of floating migrants that obtain UEMIPI in 105 cities in the Tth year}\times 100\% \\ -\frac{Number of floating migrants that obtain UEMIPI in a city in the (T-1)th year}{Total number of floating migrants that obtain UEMIPI in 105 cities in the (T-1)th year}\times 100\% \end{gathered}$$

(7)

Both the numerator and denominator are a certain number within a calendar year, therefore, the difference of the percentage between two consecutive years can be considered a proxy of a city’s attraction. Since there are only four years’ data available, 2014, 2015, 2016 and 2017, we end up with three-year values for each individual city after conducting subtracting between two consecutive years. The data is derived from CMDS.

2.3.5. Control Variables

Control Variables for Bivariate Probit Model

In order to control for important factors that may affect settlement decisions of urban floating migrants, we include nine control variables at both the individual and city levels in bivariate probit model. At the individual level, they are age, gender, education, income and social network. The data is derived from CMDS. At the city level, they are growth rate of GDP, sales acreage of commercial housing, total passenger flow of transportation and location of a city, and the data is obtained from CESY.

Age is a continuous variable representing the respondents’ age in the survey year, and its impact on settlement decisions is complex. Young people are facing low cost of migration (the minimum influence made by hukou and no mortgage payments), they are less likely to settling in a particular city at a young age, while tend to settling down when they get aged [29]. Therefore, we include both age and age square in the model. Gender refers to either male or female, because the two sexes are the dominant genders in China. Compared with female, men are wealth-oriented and more likely to develop this potential by searching for good job opportunities rather than focusing on the environmental quality, as a result, they are expected to be less willing to settling in the destination city merely for its good air quality [10,29]. Education means the number of years the floating migrants have been to school. It is measured by the survey and the question is “what is your education level?” There are seven options in corresponding to Chinese education system and they are never been to school, primary school, secondary school, high school, college, university, and postgraduates. We translate the seven options into number of years of education, and they are 0, 6, 9, 12, 15, 16, and 19 respectively. The more years of education, the higher the level of education. In general, people who have received high education are more concerned about the environmental conditions and are more likely to stay in a city with good air quality [29,31].

Income refers to the average monthly household income. This is a fill-in question, and the interviewee is asked to write down the most probable number if s/he refused to answer this question. In order to weaken the collinearity and heteroscedasticity of the model and reflect the income elasticity, we employ the logarithm of income as the proxy to conduct econometric analysis. The higher the income, the more sensitive to air pollution and the more likely to stay in the destination city with good air quality [23,29,31].

Social network refers to the relationships established by the floating migrants in the move-in city. It is measured through a survey and the question is “who do you often contact with in your spare time (not including your customers or relatives)?” There are six options for this question. The first option is fellow countryman with hukou in the destination city, the second fellow countryman with hukou in the origin city, the third fellow countryman with hukou in a place other than the origin nor destination city, the fourth natives, the fifth remaining migrants who moved to this city, and the sixth seldom contact with others. In order to simplify the analysis but better reflect the actual social networks in reality, we group those who choose any of the first three options into the same category. This category indicates a broader social network compared with those who selected the sixth option. Those who pick the fourth or fifth option are classified into the same category, demonstrating an even broader range of social connections compared with those who choose any of the first three options. We define those who select the sixth option as having the narrowest social network.

The growth rate of GDP, sales acreage of commercial housing and the total passenger flow of transportation (including highways, waterways and civil aviation) are the indicators of the economic development of a city. The higher these indicators, the more development a city is. The sales acreage of commercial housing is the total area of commercial housing sold in a city in a year. The transportation’s passenger flow is the total number of passenger times carried by the transportation department in a given year. The location of a city refers to the geological position of the city. In accordance with the economic development, Chinese provinces (is consisted of cities) can be divided into four regions and they are the eastern, the central, the western and the northeastern. The eastern region includes ten provinces (and municipalities) and they are Beijing, Shanghai, Tianjin, Shandong, Guangdong, Jiangsu, Hebei, Zhejiang, Hainan and Fujian. The central area is consisted of six provinces and they are Anhui, Shanxi, Jiangxi, Henan, Hubei and Hunan. Twelve provinces are classified as the western region of China and they are Yunnan, Inner Mongolia, Sichuan, Ningxia, Guangxi, Xinjiang, Gansu, Tibet, Guizhou, Chongqing, Shaanxi and Qinghai. The remaining three provinces located in the northeastern of China are classified into the fourth category and they are Jilin, Liaoning and Heilongjiang.

In China, the eastern region contributes more than half of the country’s GDP, the middle contributes the second, the western the third and the northeastern contributes the least [35]. Migrants who are originally from the eastern region of China are facing many job opportunities and sound public good and services; as a result, those people are less likely to settling in the move-in city. In addition, Chinese population is concentrated in the eastern region and it is estimated that this region accounts for 38.6% of the whole nation’s population in 2019 [35]. There is no big difference between the central and the western region regarding the proportion of the national population, with 26.5% and 27.2% respectively in 2019 [35]. The population in the northeastern region only accounts for 7.7% of the country’s population and this is one fifth of that of the eastern region.

Control Variables for Panel Data Regression

When performing panel data regression, we include seven city-level control variables that may interfere with the analysis. They are the proportion of employees, green GDP, total passenger flow of transportation, sales acreage of commercial housing, the growth rate of GDP, rate of floating migrants, and the number of hospitals and colleges and universities. The proportion of employees measures the employment status of a city, and the higher the number, the more attractive the city is and the longer the floating migrants expect to stay. Green GDP refers to the amount of industrial wastewater caused by per unit of GDP. The larger the green GDP, the more serious the pollution, and the shorter the floating migrants want to stay. Total transportation’s passenger flow, the sales acreage of commercial housing and the rate of floating migrants are indicators to measure the openness of a city. The larger the value of these variables, the higher the acceptance of the floating migrants in the city. In addition, the floating migrants are willing to stay long in a location with high openness. The number of hospitals and tertiary institutions represents the infrastructure condition of a city, and the more the better. Floating migrants tend to stay long in a city with high growth rate of GDP and large number of hospitals and tertiary institutions. The data is derived from the CESY and the statistical yearbooks and bulletins of various provinces and cities in China.

Due to the lack of data on the floating migrants who possess urban health insurance or public insurance in several cities, and the missing data on control variables in several cities, the panel data regression models finally include 105 cities.

2.3.6. Instrumental Variables

In order to address the endogenous problem, we introduce the ventilation coefficient and precipitation as two instrumental variables of air quality index. The ventilation coefficient and precipitation are considered to be highly correlated with the AQI, but have no effect on the migration of the floating migrants. Furthermore, they have been used in many literature [36,37]. On the one hand, the larger the ventilation coefficient and precipitation, the less the air pollutants and the better the air quality, which meets the correlation hypothesis of a valid instrumental variable [37]. On the other hand, the ventilation coefficient and precipitation are influenced by the atmospheric boundary layer, which is determined by complex meteorological and geographic conditions. Therefore, it satisfies the exogenous assumption of a valid instrumental variable [36]. The formula for calculating the ventilation coefficient is as follows:

$$V{C}_{it}=W{S}_{it}\times BL{H}_{it}$$

(8)

Among them, the $V{C}_{it}$, $W{S}_{it}$, and $BL{H}_{it}$ represent ventilation coefficient, wind speed, and the height of atmospheric boundary layer, respectively. The raw data of wind speed ($W{S}_{it}$) and the height of atmospheric boundary layer ($BL{H}_{it}$) are derived from the longitude and latitude meteorological data released by European Centre for Medium-Range Weather Forecasts (ECMWF). We employ ArcGIS technology to convert these data into ventilation coefficient in accordance with the Formula (8) mentioned in this section. The precipitation is measured by average daily precipitation. The data of precipitation is derived from “The dataset of daily values of climate data from China National Surface Weather Station” provided by the Meteorological Information Center of the National Meteorological Administration of China.

Table 1. Descriptive statistics of variables in the bivariate probit analysis.

Variables	Move-In City		Move-Out City
	Mean	SD	Mean	SD
WS	0.7402	0.4385	0.4696	0.4991
AQI	0.7505	0.1723	0.7600	0.1749
Age	40.3439	13.2578	38.7100	12.4557
Age2	18.0338	12.6522	16.5345	11.3732
Gender	0.4828	0.4997	0.5124	0.4999
Education	13.7530	2.8761	13.2847	2.9705
Income	8.0634	0.7540	7.9425	0.7566
Social network	2.4428	0.7876	2.3831	0.8032
Growth rate of GDP	7.2808	3.0781	7.4220	3.3126
Sales acreage of commercial housing	9.4560	0.8683	8.9756	1.0967
Passenger flow of transportation	9.3492	1.1531	8.8520	1.4916
Location of the city
Eastern	0.4604	0.4985	0.4040	0.4907
Central	0.1169	0.3213	0.1616	0.3681
Western	0.2912	0.4543	0.3081	0.4617
Northeastern	0.1316	0.3380	0.1263	0.3322
Ventilation coefficient	1.4161	0.5017	1.2295	0.4679
Precipitation	2.6702	1.4857	2.9986	1.6020

Notes: SD means standard deviation. WS refers to willingness to settling in the destination city. Age² refers to age squared.

describes the characteristics associated with all variables included in the bivariate endogenous probit analysis. The descriptive statistics with respect to all variables included in the panel data regression is depicted in

Table 2. Description of variables in the panel data regression.

Variables	Move-In City
	Mean	SD
Long-term settlement	0.6923	0.2098
Actual migration behavior	−0.1491	0.5889
AQI	75.7232	18.8727
Proportion of employees	0.1666	0.0967
Green GDP	2.6975	1.8123
Passenger flow of transportation	0.9795	1.4665
Sales acreage of commercial housing	8.4161	0.993
Growth rate of GDP	7.7971	2.9186
Immigration rate	0.0443	0.1595
Number of hospitals and colleges and universities	273.7847	215.1742

Notes: SD means standard deviation.

.

3. Results

3.1. The Effects of Air Quality on Willingness of Urban Floating Migrants to Settling in the Destination City

The probit regression results are integrated in

Table 3. The probit regression results.

Variables	Move-In City			Move-Out City
	Model 1	Model 2		Model 3	Model 4
	DV: Willingness to Settling	DV: AQI	DV: Willingness to Settling	DV: Willingness to Settling	DV: AQI	DV: Willingness to Settling
AQI	−0.3260 ***		−0.1924 *	0.0982		0.2699 **
	(0.0852)		(0.1144)	(0.0629)		(0.1186)
Age			−0.0166 **			0.0110 **
			(0.0077)			(0.0054)
Age2			0.0159 **			−0.0003
			(0.0080)			(0.0059)
Gender			−0.0671 **			−0.1739 ***
			(0.0309)			(0.0231)
Education			0.0191 ***			0.0516 ***
			(0.0060)			(0.0043)
Income			0.0434 *			0.0484 ***
			(0.0225)			(0.0174)
Social network			0.1032 ***			0.1624 ***
			(0.0194)			(0.0146)
Growth rate of GDP		−0.0147 ***	0.0210 ***		−0.0047 ***	−0.0100 **
		(0.0005)	(0.0060)		(0.0004)	(0.0042)
Sales acreage of commercial housing		0.0129 ***	−0.0738 ***		0.0264 ***	−0.0307 **
		(0.0017)	(0.0247)		(0.0015)	(0.0151)
Passenger flow of transportation		0.0012	−0.0637 ***		−0.0056 ***	−0.0177 *
		(0.0017)	(0.0198)		(0.0013)	(0.0102)
Location of the city
Central		0.0200 ***	−0.3116 ***		0.0116 ***	−0.1789 ***
		(0.0057)	(0.0544)		(0.0035)	(0.0353)
Western		−0.1512 ***	−0.4045 ***		−0.0897 ***	−0.2284 ***
		(0.0041)	(0.0402)		(0.0035)	(0.0292)
Northeastern		−0.1482 ***	−0.1901 ***		−0.1431 ***	−0.1632 ***
		(0.0037)	(0.0575)		(0.004)	(0.0429)
Ventilation coefficient		−0.2466 ***			−0.1340 ***
		(0.0044)			(0.0032)
Precipitation		−0.0773 ***			−0.0628 ***
		(0.0008)			(0.0009)
_cons	0.8899 ***	1.3400 ***	0.2876	−0.1510 ***	0.9974 ***	−1.4431 ***
	(0.0660)	(0.0246)	(0.3134)	(0.0491)	(0.0132)	(0.2292)
Corr (AQI, Willingness to Settling)		−0.0464 **			−0.0626 ***
		(0.0186)			(0.0197)
Number of observations	8301	8149		13,016	12,440

Notes: Standard errors are listed in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01. DV represents dependent variable.

. The ventilation coefficient and precipitation are two instrumental variables of air quality, and the dependent variable is willingness to settling in the destination city. Model 1 and 3 perform probit regression and do not include control variables, while Model 2 and 4 add all of the control variables including individual- and city-level variables, resulting in endogenous probit analysis. The AQI and city-level variables in Model 1 and 2 are derived from the move-in cities, and those variables in Model 3 and 4 are derived from the move-out cities. In addition, the objects in Model 1 and 2 are those floating migrants who have urban hukou in the origin city and housing in the move-in city, while Model 3 and 4 release these restrictions and focus on the total number of floating migrants. In the baseline model with no additional control variables, the association between AQI at the move-in city and willingness to settling is negative, while positive for the AQI at the move-out city. The ventilation coefficient and precipitation are found to be significantly negatively correlated with AQI. The preferred regression results are in line with our expectations, with good air quality acting as a pull factor to attract urban floating migrants and poor air quality acting as a push factor to drive floating migrants away.

In the baseline model without additional controls, a unit increase in the air quality index of the move-in city will decrease the willingness of urban floating migrants who have housing in the destination city to settling in the move-in city by around 0.33%, while a unit increase in the air quality index at the move-out city will increase the willingness of the total number of floating migrants to settling in the move-in city by around 0.1%. Counterintuitively, the higher the air quality index the lower the air quality. In our preferred specification where individual-level and city-level variables are added to the regression model, the marginal effect of the air quality index at the destination city declines to around 0.19%, while the marginal effect of AQI at the origin city increases to around 0.27%. These results tell us that the air quality index at both origin and destination cities are equally important in directing settlement choices. The findings support the hypothesis that the effects we have identified so far are motivated by individuals’ avoidance of air pollution. For instance, if it was found that only the air quality at the move-in city that has an impact on settlement decisions of urban floating population, then the determination of the AQI coefficient may be confounded by other factors [25].

In terms of the control variables, both individual- and city-level variables turn out to have significant impacts on floating migrants’ willingness to settling in the destination city, and the estimates are basically in line with our expectations and findings of previous research [6,10,17,23,25,31].

3.2. The Effects of Air Quality on Expected Long-Term Settlement and Actual Migration Behavior

In order to ensure the reliability of the results, we conduct two robust tests, considering the effects that air quality has on expected long-term settlement and actual migration behavior. The results are shown in

Table 4. The results of robust tests.

Variables	DV: Long-Term Settlement		DV: Actual Migration
	Model 1	Model 2	Model 1	Model 2
AQI	−0.0064	−0.0267 *	−0.0311 ***	−0.0385 **
	(0.0063)	(0.0155)	(0.0074)	(0.0160)
Proportion of employees		0.4792		4.7326 *
		(0.3484)		(2.6915)
Green GDP		0.0397		0.0168
		(0.0287)		(0.0488)
Passenger flow of transportation		0.0788		−0.0186
		(0.0513)		(0.1351)
Sales acreage of commercial housing		−0.2588 *		−0.0418
		(0.1403)		(0.3556)
Growth rate of GDP		0.0300 **		0.0311
		(0.0129)		(0.0289)
Immigration rate		−0.6661		4.0979 *
		(1.0512)		(2.3304)
Number of hospitals and colleges and universities		−0.0030		−0.0008
		(0.0018)		(0.0036)
F	1.03	1.14	17.75	0.0007
Centered R2	−0.0862	−0.5521	0.0586	0.0632
Number of observations	292	281	277	266
Number of groups	99	97	94	92

Notes: Standard errors are listed in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01. DV represents dependent variable.

. Model 1 and 2 use the long-term settlement as the dependent variable, while Model 3 and 4 employ the difference of the percentage of the number of floating migrants who have UEMIPI in a certain city that accounts for that of in the 105 cities in two continuous years. Model 1 and 3 offer the baseline regressions that exclude the control variables, while Model 2 and 4 include the controls. All city-level variables are derived from the move-in cities.

It can be seen from Table 4 that the associations between the air quality and expected long-term settlement and actual migration behavior are significantly negative across all four models, indicating that poor air quality at the destination city reduces the long-term settlement intention of urban floating migrants as well as the attractiveness of the city. The coefficient of the AQI is around −0.027 for the expected long-term settlement, and −0.039 for the actual migration behavior. The findings support the aforementioned results, indicating that people do optimize residence by leaving the current city with poor air quality and moving to a city with clean air. The results also indicate that the long-term settlement intention of urban floating migrants is strongly associated with air quality, and the better air quality of a city the longer the floating migrants want to stay. In addition, urban floating migrants do migrate to the cities with clean air in the reality. These results are very important for urban planners to (re)design the policies aimed at improving the air quality of the city as well as to predict migration both in developed economies and developing countries.

4. Discussion

This study has explicitly examined the presumed association between air quality and settlement decisions of urban floating migrants who have urban hukou in the origin city and housing in the destination city. The results support the hypothesis that people do optimize their residence by leaving a city with poor air and moving to a city with clean air. The willingness of urban floating migrants to settling in the destination city is associated with air quality in both move-in and move-out city, with poor air quality in move-out city acting as a push force to drive floating migrants away while the good air quality in move-in city acting as a pull force to attract migrants to settling. In this study, the higher the air quality index, the lower the air quality. For 1 per cent increase in the air quality index of the move-in city, the willingness of urban floating migrants to stay in the destination city decreases by 0.19 per cent; indicating that the poor air quality in move-in city will reduce the willingness of the floating migrants to settling in the destination city. However, a unit increase in the air quality index of the move-out city would cause 0.27 per cent increase of willingness to settling in the move-in city, suggesting that the poor air quality in move-out city will increase the willingness of floating migrants to settling in the move-in city. This message is important to assist urban planners and policy makers to design political interventions aimed at reducing the air pollution to attract talent to settling.

Other factors accounted for in this study, including age, gender, education, income, social network, growth rate of GDP, sales acreage of commercial housing, passenger flow of transportation and location of the city, are also significantly related to the willingness to settling in the destination city. The robust tests with respect to the associations between air quality and the expected long-term settlement and actual migration behavior of urban floating migrants show good agreement with the main regression results, with almost all relationships reported having the same direction as those indicated by the main regression analyses. The results imply that people do chase clean air and make the migration choices in accordance with the air quality. The better the air quality of a city, the longer the urban floating migrants want to stay, and the more migrants move to in reality.

The main contribution of this study is that we identify the very group of target floating migrants who are more likely to make relocation decisions due to the air quality. They are urban citizens with urban hukou in the origin city and housing in the destination city. Also, we empirically evaluate the effect that air quality has on settlement decisions of urban floating migrants from the perspective of both the move-in and move-out cities. In addition, we investigate not only the willingness to settling in the destination city, but also the expected long-term settlement and the actual migration behavior made by urban floating migrants in reality. This brings the integration of the individual- and city-level data, which is an advance in this topic. The results show that poor air quality leads to out-migration, while good air quality creates in-migration, implying that environmental quality has become an important attribute of migration.

Findings from this research could contribute several implications to future research and policies. First, this study provides a better understanding of the consequences of air pollution in urban cities. The poor air quality can not only affect human health but also drive them away, while the clean air can be a soft power for cities to attract urban floating migrants to migrate to and settling down. The urban floating migrants with urban hukou in the origin city and housing in the destination city has the greatest impact. Policy makers can use this information to make interventions to improve the air quality of the city and attract talent. Second, this research can be seen as a logical extension of the recent studies on natural-based solutions that aimed at bringing nature to the urban cities. Therefore, it is useful for policy makers to make interventions to stop prioritizing industrial economic activities over environmental sustainability, and include the air quality as a performance indicator instead.

Due to the nature of the data set, there are several limitations that need to be addressed in future research. First, the observations in this study are limited to the floating migrants who have purchased house in the destination city but without local hukou, and the data is derived from CMDS. This sample selection would exclude a group of urban floating migrants who have moved to large cities, such as Beijing and Shanghai, as they are not allowed to purchase house if not have local hukou. Moreover, once they obtained local hukou, they become natives which are not the targets of CMDS. Ideally, a complete data set should contain information on migrants who have already obtained local hukou in the move-in city. This information then would allow another empirical analysis that excludes the influence of hukou. Domestic migration in China is costlier compared to most other countries as Chinese citizens need to apply for hukou which can be very difficult, especially in Beijing and Shanghai, and migrants are unable to access to the amenities such as education and health care, if not have local hukou. Second, settlement is a very complicated decision and influenced by many determinants. The effects that we have investigated in this study may be overstated. This problem is, however, likely to be mitigated by controlling for more economic factors in the regression analyses. Third, the survey data used in this study does not include information on migrants prior to their migration, such as their previous occupations and income levels. As a consequence, we are unable to compare the influences of these variables before and after migration. Fourth, it is not a follow-up survey, therefore, we cannot conduct a more convincing panel regression analysis.

Despite these limitations, this study provides valuable evidence with respect to the effect that air quality has on settlement decisions of urban floating migrants.

5. Conclusions

Our study empirically examines the association between air quality and settlement decisions of urban floating migrants by using integrated individual- and city-level data from which we extract the target migrants and empirical approaches to address the potential endogeneity of air quality. We investigate such an association from both move-in and move-out city representing push and pull force respectively. In addition, we exam not only the willingness of urban floating migrants to settling in the move-in city, but also their actual migration decision in the reality. The bivariate probit analysis and panel data regression are applied. The results show that the poor air quality in the move-out city acts as a push force to drive migrants away, while the good air quality in the move-in city acts as a pulling force to attract migrants to stay. This message is very important for urban planners and policy makers to design interventions aimed at improving air quality so as to attract talent and eventually to achieve prosperity. More specifically, this study provides empirical support for the strategy of promoting population redistribution from mega-urban centers to middle- and small-sized cities in the world.