Has Cross-City Commuting Promoted Housing Purchases among the Workforce within Metropolitan Areas?—An Empirical Analysis from Micro Survey Data from China’s Three Major Metropolitan Areas

Abstract

The ability of the cross-city commuting labor force to obtain housing has a profound impact on the development of the housing market and the enhancement of social welfare, but whether cross-city commuting has facilitated housing purchases remains to be verified However, the research on whether cross-city commuting behavior promotes labor force housing purchase in metropolitan areas is still lacking, especially in China, where the culture of buying houses is relatively special. This article used field survey data from the 2023 China Metropolitan Area Occupation and Housing Status Sampling Survey to empirically analyze whether cross-city commuting has facilitated housing purchases within metropolitan areas. The analysis was conducted by constructing a baseline model, a mediation effect model, and a subsample regression model. The results show that the cross-city commuting facilitated housing purchase within metropolitan areas, and the location preference is to purchase a house with a distance of 20–40 km from the workplace, but the contribution of the cross-city commuting to multi-suite purchases is relatively low. Mechanism analysis shows that compared to the workers who work and live in peripheral areas or the workers who work and live in cores, intercity commuters are promoted to purchase housing by relatively higher income and inducement of the housing price gap. The above conclusions still hold after controlling potential endogeneity issues and in robustness tests. The research of this paper can provide a new perspective for alleviating the housing inequality in the metropolitan area.

1. Introduction

With the continued advancement of regional integration in China, a more pronounced phenomenon of cross-city commuting among the labor force has emerged, especially within metropolitan areas centered around megacities. Cross-city commuting refers to the phenomenon of long-duration commuting due to the separation of workers’ residences and workplaces across different administrative units [1]. Essentially, it is a form of job–residence separation, often characterized by workers being employed in central cities while residing in peripheral cities. This commuting behavior results from a comprehensive consideration of factors such as job opportunities, career development, housing costs, commuting costs, and access to public services. It is also commonly used to measure the development level of metropolitan areas [2]. According to the 2022 National Major Cities Commuting Monitoring Report released by the China Academy of Urban Planning and Design, more than 14 million people in 44 major cities endure extreme long-distance commuting of more than one hour, accounting for 13% of the entire commuting population. There were over 8 million people with extreme long-distance commuting in three megacities, namely, Beijing, Shanghai, and Guangzhou, accounting for 63% of the total commuting population [3]. With a further urbanization process, the population will continue to gather in the metropolitan areas with megacities as the core [4], and the peripheral cities within the metropolitan area will gradually be included in the commuting range. Thus, the connection between the residence and workplace of the cross-city commuters will continue to be intensified. It is foreseeable that the size of the cross-city commuters within the metropolitan area, which is an important carrier of urban development and population concentration in China, will also continue to expand [5].

Whether such a large-scale cross-city commuting population can achieve a balance between “livability” and “workability” is a crucial issue that affects the overall welfare of society. The concentration of high-quality resources in core cities within metropolitan areas leads to higher income levels, housing demand, and housing prices in these core cities, while peripheral cities experience relatively lower levels. Consequently, workers weigh these factors and make decisions about intercity commuting and renting or purchasing housing based on their circumstances to maximize household utility [6,7,8]. On one hand, workers who choose cross-city commuting are more likely to earn higher incomes compared to those in peripheral cities [9,10,11], enhancing their ability to purchase housing in peripheral cities [12,13,14]. Additionally, influenced by traditional Chinese beliefs such as “a home is only complete with property” and “settling down and establishing a career”, they are more inclined to buy rather than rent a home, aiming for stable housing [15]. On the other hand, the housing costs for cross-city commuters are significantly reduced due to residing in peripheral cities, which facilitates home purchases. Because the land rents and prices gradually decline from the city center to the suburbs [16], housing prices will also decrease as the distance from the central city increases [17]. And the lowering cost of housing will promote the purchase of housing by cross-city commuters.

However, existing studies primarily focus on issues such as traffic congestion, job–housing imbalance, and the physical and mental strain experienced by commuters due to cross-city commuting [5]. Consequently, there remains a dearth of evidence regarding the correlation between cross-city commuting behavior and housing purchases. Whether cross-city commuting has enhanced housing purchases among the workforce within metropolitan areas, the mechanisms through which this is achieved, and the characteristics of this process are questions that remain to be answered. Based on the above analysis, empirical testing and analysis were conducted by using micro-survey data from the metropolitan areas of Beijing, Shanghai, and Guangzhou in 2023. The base model, intermediary effect model, and sub-sample regression model were constructed to investigate the promotion effect, characteristics, and specific mechanism of cross-city commuting behavior among laborers in metropolitan areas. This paper’s marginal contribution primarily lies in four aspects compared to existing research results. First, this paper systematically examined the impact of cross-city commuting on the housing purchasing behavior of the labor force in metropolitan areas. Second, it thoroughly analyzed whether cross-city commuting affects the housing purchasing behavior of the labor force in metropolitan areas through two mechanisms: increasing income and reducing costs by leveraging housing price gaps. Furthermore, this study delved into the characteristics of location preferences for housing purchases and multi-house purchase decisions among cross-city commuter workers. Additionally, field research was conducted to gather first-hand data, ensuring the reliability of the research findings. From a perspective centered on cross-city commuting behavior, this paper discusses its positive influence on housing markets, providing valuable policy insights for mitigating housing inequality within metropolitan areas, promoting stable and healthy development in real estate markets, achieving high-quality urbanization, and enhancing social welfare effects.

2. Theoretical Analysis and Research Hypothesis

2.1. Cross-City Commuting and Housing Purchase of Labor Forces in Metropolitan Areas

The change of housing market demand is closely related to population migration or household relocation [18,19,20,21]. The dynamic evolution theories of metropolitan areas, such as the urban evolution model and urban spatial cycle theory, suggest that population agglomeration and diffusion directions vary across the different development stages of metropolitan areas [22,23]. The maturation of the metropolitan area promotes the development of the housing market, and the relationship between housing supply and demand in different parts of the metropolitan area will be different, which brings about changes in the number of housing floors and building density in different layers of the metropolitan area [24]. According to the multi-city system equilibrium model, as shown in Figure 1, it is assumed that in a closed system consisting of two linear single-center cities, city A is the core city and city B is the peripheral city; the total population size of the two cities is unchanged; and the employment opportunities are mainly located in the central business districts of the two cities, CBD1 and CBD2. C is the boundary between the two cities, with residents living between the central business district and the city boundary, which will change as the ease of commuting improves. Residents maximize their utility by balancing the relationship between where they live, the city they work in, and the cost of commuting.

As a typical feature of population mobility, cross-city commuting has a profound impact on the purchasing behavior of the labor force in the metropolitan area. The capability theory holds that individual freedom of choice is constrained by their own capabilities. Hence, not everyone can achieve the maximum utility or preference, and welfare disparities arise due to different endowments. High-quality resources in the metropolitan area are mainly concentrated in the core cities, inevitably leading to population agglomeration. However, not all individuals have the capability to purchase housing in the core cities, and they will make suitable choices according to their own conditions after weighing and choosing. This paper argues that the main reason why the labor force in the metropolitan area has cross-city commuting behavior is the lack of capability to purchase or rent a house in the core city. However, in the peripheral cities, their freedom of choice has been improved, and under the influence of traditional concepts such as “only with a house can one have a home” and “settling down with property”, they are more inclined to buy a house rather than rent, aiming to achieve the goal of having a place to live. Therefore, the first research hypothesis of this paper is as follows:

Hypothesis 1. 

Cross-city commuting behavior of labor forces in metropolitan areas has a positive effect on housing purchase.

2.2. Mechanism of Cross-City Commuting Affecting Housing Purchase

The influence of cross-city commuting on the housing purchase behavior of the labor force in a metropolitan area may be realized through two pathways or mechanisms of using the income gap to increase income and the housing price gap to reduce cost:

(1)

Increased income pathway

Cross-city commuting raises the income level of the labor force, which in turn promotes the increase in housing consumption. Cross-city commuting behavior is mostly manifested by the situation of labor forces who live in peripheral cities and work in core cities. Generally speaking, the income level of labor forces in core cities is higher than that in their places of residence. On the one hand, core cities have a knowledge spillover effect and higher human capital level, so the wage level is higher than that of peripheral cities [25]. On the other hand, labor can accumulate more human capital in core cities through free flow, which is more conducive to improving their own income level [9,10,11]. According to Keynes’ absolute income theory, with the increase in residents’ disposable income, their consumption expenditure will also increase [26], which has a very significant promoting effect on housing consumption [12,13]. For example, the research of Zhao and Zou (2012) shows that the group with high income has higher housing consumption demand than the group with low income [14]. Therefore, the research hypothesis of this paper is as follows:

Hypothesis 2. 

Cross-city commuting behavior in metropolitan areas is conducive to promoting housing purchase through the increase in labor income level.

(2)

Reduced costs pathway

Cross-city commuters can also take advantage of the housing price gradient between the places of work and residence to reduce the cost of housing and realize housing purchase. In the 1960s, William Alonso established a rent competition model with landmark significance of modern classical city location theory—the rent competition model of single central city land price. According to this model, competitors that are more sensitive to location and have strong ability to pay land rent (such as commercial service industry) will obtain the land use right in the downtown area, and the land use of other activities will be extrapolated in turn, with the land rent price gradually declining from the downtown area to the suburbs [16]. Later, in the AMM (Alonso–Mears–Mutter) single-center model, Mills and Mutter replaced the land market with the housing market and argued that housing prices would decrease with the increase in distance from the central city [16]. Prices in peripheral cities are lower than those in central cities. Existing research also indicates that young people with lower incomes, especially new urban residents, are often willing to endure longer commutes to save housing costs and achieve home ownership [27,28,29]. Housing affordability of the labor force is relatively high. And housing demand is easier to achieve. Based on this analysis, the research hypothesis made in this paper is as follows:

Hypothesis 3. 

Cross-city commuting behavior in metropolitan areas is conducive to the labor force taking advantage of the housing price gradient between their places of work and residence, which promotes housing purchase.

3. Research Design

3.1. Selection and Geographical Definition of Metropolitan Area

According to the China Metropolitan Area Development Report 2021 released by Tsinghua University’s Institute for China Sustainable Urbanization, there are 34 metropolitan areas in China that are categorized into three groups: mature, developmental, and nurturing. Among them, the Capital Metropolitan Area, Shanghai Metropolitan Area, and Guangzhou Metropolitan Area are recognized as mature metropolitan areas that are characterized by high-quality development patterns and a high degree of freedom in the flow of various resources. As both the political and economic center of China, Beijing has become a megacity with multiple functions. The Capital Metropolitan Area, also known as JingJinJi Metropolitan Area, centering around Beijing, plays a crucial role in decentralizing non-capital functions and constructing a world-class metropolitan area. Shanghai and Guangzhou serve as economic centers in China, having consistently been at the forefront of urbanization levels within their respective metropolitan areas. These regions boast strong internal connections and witness a substantial number of cross-city commuters. Therefore, selecting these three metropolitan areas as research objectives is of great representative significance.

According to the Beijing High-Tech Industry Development Plan during the 14th Five-Year Plan Period and the Beijing Urban Master Plan (2016–2035), the spatial scope of the capital metropolitan area mainly includes all districts and counties of Beijing and Tianjin, as well as some districts and counties of Hebei Province (all districts and counties of Langfang, Baoding, Zhangjiakou, Chengde, Cangzhou, and Tangshan and Xiongan New Area in Hebei Province). The core city is Beijing, and the rest are peripheral cities. According to the Master Plan for the City of Shanghai (2017–2035) and the Outline Plan for the Integrated Development of the Yangtze River Delta Region, the spatial scope of the Shanghai metropolitan area mainly includes all districts and counties in Shanghai and some districts and counties in Jiangsu and Zhejiang provinces (all districts and counties of Wuxi, Changzhou, Suzhou, and Nantong in Jiangsu Province; all districts and counties of Ningbo, Huzhou, Jiaxing, and Zhoushan City in Zhejiang Province). The core city is Shanghai, and the rest are peripheral cities. According to the Draft of the General Urban Plan of Guangzhou City (2017–2035), the 14th Five-Year Plan for National Economic and Social Development of Guangzhou City and the Outline of Long-Term Goals for 2035, and the 14th Five-Year Plan for National Economic and Social Development of Guangdong Province and the Outline of Long-term Goals for 2035, the spatial scope of Guangzhou metropolitan area mainly includes all the districts and counties of Guangzhou City and Foshan City and the urban areas of Zhaoqing City, Qingyuan City, Yunfu City, and Shaoguan City (Duanzhou District, Dinghu District, and Gaoyao District of Zhaoqing; Qingcheng District and Qingxin District of Qingyuan; Yuncheng District and Yun’an District of Yunfu; Wujiang district, Zhenjiang district, and Qujiang district of Shaoguan). The core city is Guangzhou, and the rest are peripheral cities. The spatial scope and layer definition of the three metropolitan areas are shown in

Table 1. Spatial scope and layer definition of the three metropolitan areas.

Functional Partition	Capital Metropolitan Area	Shanghai Metropolitan Area	Guangzhou Metropolitan Area
Core city	Beijing	Shanghai	Guangzhou
Peripheral cities	Langfang, Baoding, Zhangjiakou, Chengde, Cangzhou, Tangshan, Xiongan New Area, and Tianjin	Wuxi, Changzhou, Suzhou, Nantong, Ningbo, Huzhou, Jiaxing, and Zhoushan	Foshan, Zhaoqing, Qingyuan Yunfu, and Shaoguan

Source: The classification of city types is based on the relevant planning documents.

.

3.2. Data Sources and Processing

The data used in this study was obtained from the Sampling Survey of Employment and Housing Status of the Population in China’s Metropolitan Areas, which was conducted in January 2023. This survey received support from the National Natural Science Foundation of China project and primarily focused on examining the employment and housing conditions among individuals aged 18 and above who constitute the permanent labor force in metropolitan areas (the reason why the labor force aged 18 and above was selected as the research object is that this group has certain self-decision-making and behavioral ability, can make decisions on migration or cross-city commuting, and has less interference by external factors such as parents and elders, which ensures the independence of individuals in the research sample to a greater extent). In collaboration with the NetEase Cloud Business online research platform, our research group sent questionnaires for people to fill out according to the set conditions.

A total of 3090 questionnaires were distributed, and 3090 questionnaires were collected, with a recovery rate of 100%, with 1030 questionnaires distributed in each metropolitan area. The number of questionnaires distributed in each district and county within the metropolitan area followed the proportion of the permanent resident population announced in the 7th National Population Census to the total permanent resident population in this metropolitan area (considering the fact that the metropolitan area of the capital is too large and the connection is not high, only the districts and counties within 50 km of Beijing were issued questionnaires). Then, by accessing the big data port of personal information of mobile phone users, questionnaires were sent to all eligible people of different ages, genders, and educational backgrounds in each district and county according to the statistical proportion of seven popular data, and random screening was conducted to determine the final respondents to fill in the questionnaire according to the acceptance of users. Samples that were excluded from analysis include outliers with abnormal data, respondents selecting “don’t know” or “not clear” as answer options, individuals working outside but residing within the metropolitan area, individuals working in peripheral cities while residing in core cities, and individuals living and working across different peripheral cities (considering that the cross-city commuting behavior of the labor force in the metropolitan area is mainly living in the peripheral city but working in the core city (experimental group) and living and working in both the core city or the peripheral city (control group), this paper investigated the samples, while other situations were not considered). The final sample size used for analysis was 3051: 1016 from both the Capital metropolitan area and the Shanghai metropolitan area, and 1019 from the Guangzhou metropolitan area (if any authors are interested in the questionnaire, please contact the first author or corresponding author).

It is important to emphasize that, in order to ensure the scientific validity of the questionnaire data, we implemented a rigorous testing mechanism both during and after the investigation process. Firstly, during the questionnaire design phase, certain questions were equipped with test items to assess the consistency of respondents’ choices. Secondly, a half month after the completion of the survey, a survey with a part of questions from the original questionnaire was reconducted for the same interviewees to compare whether they made the consistent choices.

3.3. Model Setting

Based on the nature of the explained variables and the needs of the research, in order to further explore the relationship between the cross-city commuting behavior (the core explanatory variable) and the housing purchase (the explained variable), this paper constructed a binomial Logistic regression model for empirical analysis of the benchmark regression model. The model settings are as follows:

$$logit{bh}_i=\alpha +{\alpha }_0{ccb}_i+\sum {\alpha }_j{controlvar}_i+{\epsilon }_i$$

(1)

Among them, $logitbh$ is the logistic conversion form of whether the dependent variable buys a house, which mainly uses the measurement of whether micro-individuals own their own houses. $ccb$ is the core explanatory variable, $controlvar$ are the control variables, $\alpha$ on the right side of the model is a constant term, ${\alpha }_0$ and ${\alpha }_j$ are the coefficients to be estimated ($j\ge 1$ and is an integer), ${\alpha }_0$ is the total effect of the influence of cross-city commuting on housing purchase, and ${\epsilon }_i$ is a random error term. At the same time, in order to further explore the intermediary effect of income gap and housing price gap on cross-city commuting affecting the housing purchase behavior of labor force in the metropolitan area, the following test models were established:

$${mediatingvar}_i={\beta }_0+{\beta }_1{ccb}_i+c_i+{\omega }_i$$

(2)

$$logit{bh}_i={\gamma }_0+{\gamma }_1{ccb}_i+{\gamma }_2{mediatingvar}_i+\sum {\gamma }_j{controlvar}_i+{\delta }_i$$

(3)

where $mediatingvar$ is the intermediary variable and $c_i$ is the control variable.

Substitute Equation (2) into Equation (3) to obtain

$$logit{bh}_i={\gamma }_0+{\beta }_0{\gamma }_2+{(\gamma }_1+{\beta }_1{\gamma }_2){ccb}_i+\sum {\gamma }_j{controlvar}_i+{\gamma }_2{c}_i+{{\delta }^{\prime }}_i$$

(4)

Among them, ${\gamma }_1$ is the direct effect of cross-city commuting on housing purchase, and ${\beta }_1{\gamma }_2$ is the indirect effect of cross-city commuting on housing purchase through intermediary variables. The testing mechanism of the intermediary effect is that if ${\alpha }_0$, ${\beta }_1,$ and ${\gamma }_2$ are all significant and ${\gamma }_1$ becomes less significant than ${\alpha }_0$, it is considered that there is an intermediary effect [30].

3.4. Variable Selection

(1)

Dependent variable ($bh$)

The dependent variable in the binomial logistic regression model employed by this paper is the decision to purchase a house, represented as a binary variable. Based on the question “What type of housing do you currently reside in?” in the questionnaire, the dependent variable refers to whether an individual chooses to buy a house. The value of “1” is assigned to respondents who indicate ownership of a house, indicating that they have indeed purchased housing, whereas a value of “0” is assigned to those who report living in rented houses or dormitories, suggesting that they primarily rely on rental accommodations and have not yet made a housing purchase.

(2)

Core explanatory variable ($ccb$)

The core explanatory variable is whether there is a cross-city commute, which is a binary variable. According to the questions “May I ask your current place of residence” and “May I ask your current place of work” in the questionnaire, samples that answered “live in peripheral cities but work in core cities” are regarded as having cross-city commuting behavior and assigned a value of “1”. The samples that answered “both the residence and work place are in the core city or the same city outside” were considered as having no cross-city commuting behavior and assigned a value of “0”.

(3)

Intermediate variables ($mediatingvar$)

The intermediate variables in this paper include increasing income and reducing costs. Specifically, the increase in income is determined by asking participants to compare their current workplace salary level with their current residence. Participants who respond with “higher” are considered to have experienced an increase in income and are assigned a value of “1”, while those who respond differently are assigned a value of “0”. On the other hand, cost reduction is assessed by comparing housing prices near the participants’ current workplace and their current residence. If the ratio between these two prices exceeds “1”, it indicates a reduction in house purchase cost, and such samples are assigned a value of “1”; otherwise, they receive a value of “0”.

(4)

Control variables ($Controlvar$)

In terms of the selection of control variables, drawing on the research results of Wang Zhenpo et al. (2018) [31], six control variables, namely, gender, education level, family size, relocation intention, residence registration location, and annual family income, are mainly selected.

Among the above variables, the binary variable assignment method is employed for gender variables, with the male sample assigned a value of “1” and the female sample assigned a value of “0”. Multiple categorical variables are assigned values based on education level. Respondents’ answers to the questionnaire question “What is your highest education?” are coded as follows: “did not go to school” = 1, “primary school” = 2, “went to middle school” = 3, “went to high school/secondary school/vocational high school” = 4, and “went to junior college/vocational high school” = 5. Respondents answering with “undergraduate” are assigned a value of 6, while those answering with “master’s degree or above” are assigned a value of 7. To determine family size, a method of assigning multiple categorical variables is employed. In response to the questionnaire item “What is the total population of your household?”, a value of “1” is assigned for a “single person”, “2” for “both husband and wife”, and “3” for “husband and wife living with children and elderly individuals”. Regarding intention to move, answers to the question “Do you have any intention or plan to relocate within the next 5 years?” are coded as “1” for affirmative responses (“yes”) and as “0” for negative responses (“nothing”). For household registration location, two categorical variables are assigned based on participants’ answers to the question regarding their place of registration. Those who indicate their registration in either the county/district or other counties/districts within this city (prefecture-level city/municipality directly under central government) are considered as having city registration (coded as “1”), while those indicating registrations in other cities within this province or in other provinces are considered without city registration (coded as “0”). The method employed for assigning multiple categorical variables to the annual family income variable involved categorizing responses to the questionnaire question “What is your annual family income?” as follows: respondents indicating “100,000 yuan and below (including 100,000)” are assigned a value of “1”, those indicating “100,000–200,000 yuan (including 200,000)” are assigned a value of “2”, those indicating “200,000–300,000 yuan (including 300,000)” are assigned a value of “3”, and those indicating “300,000–400,000 yuan (including 400,000)” are assigned a value of “4”. Respondents who indicate an annual family income range of “400,000–500,000 yuan (including 500,000 yuan)” are given a value of “5”, while respondents who reported an annual family income exceeding 500,000 yuan receive a score of “6”.

The descriptive statistics of the variables are shown in

Table 2. Descriptive statistical results of variables.

Variable Name	Variable Assignment	Full Sample		Living in the Peripheral Cities (Control Group: Both Work and Residence in Peripheral Cities)		Working in the Core Cities (Control Group: Both Work and Residence Are in the Core City)
		Mean Value	Standard Deviation	Mean Value	Standard Deviation	Mean Value	Standard Deviation
Whether to buy a house	Buy a house = 1	0.63	0.48	0.63	0.48	0.64	0.48
Whether to commute across town	Commute across town = 1	0.08	0.26	0.16	0.36	0.13	0.33
Raise income	The income of the workplace is higher than that of the residence = 1	0.49	0.50	0.52	0.50	0.49	0.50
Reduce cost	Housing prices are higher where you work than where you live = 1	0.18	0.38	0.30	0.46	0.17	0.37
Gender	Male = 1	0.55	0.50	0.48	0.50	0.59	0.49
Educational level	Not going to school = 1; Primary school = 2; …; Undergraduate = 6; Graduate = 7	4.69	1.35	4.63	1.40	4.75	1.32
Family size	1 single person = 1; Husband and wife = 2; Couples, children, and the elderly = 3	1.82	0.70	1.86	0.62	1.82	0.74
Relocation intention	Intention or plan to relocate = 1	0.27	0.44	0.20	0.40	0.32	0.47
Household registration	Registered in the county/district and other counties/districts of the city (prefecture-level city/municipality directly under the Central Government) = 1	0.81	0.39	0.84	0.37	0.79	0.41
Gross annual household income	100,000 yuan and below = 1; 100,000 to 200,000 yuan = 2; …; More than 500,000 yuan = 6	2.24	1.12	2.19	1.02	2.37	1.22
Sample size		3051		1451		1829

Note: The data used in this paper were from the 2023 Sampling Survey of Employment and Housing Status of Population in China’s Metropolitan Areas.

.

4. Empirical Results and Analyses

In order to elucidate the relationship between cross-city commuting behavior and housing purchase, as well as the extent of influence and specific mechanisms at play, this paper employed quantitative analysis through descriptive analysis, constructing a benchmark model and incorporating an intermediary effect model that accounts for income increase and cost reduction. All empirical results presented below were obtained via regression analysis using Stata16.0 software.

4.1. The Characteristics of Housing Purchase Behavior of Different Types of Labor Force in the Metropolitan Area

According to the statistics on the rental and purchase behaviors of different types of labor force in the metropolitan area, the proportion of cross-city commuter labor force housing purchase was found to be significantly higher than that of non-cross-city commuter labor force (who both live and work in peripheral cities or core cities), as shown in Figure 2. Specifically, the proportion of housing purchase in peripheral cities by cross-commute workers was 75.98%, which surpassed that of non-cross-city commuter workers (61.84%) and was higher than that of workers who live in peripheral cities (61.13%) and core cities (62.38%). It indicates a discernible correlation between cross-city commuting behavior and housing purchase.

4.2. An Empirical Analysis of the Effects and Mechanisms

So, what is the degree of this correlation, and is the mechanism of action outlined above effective? This paper continues to test and analyze the estimations through the empirical model. To address the issue of multicollinearity, this paper first conducted a multicollinearity test on the proposed model and all explanatory variables. The results showed that the mean VIF of all models was less than 10, and the VIF value of each variable was also less than 10. Therefore, it was determined that the model did not have a multicollinearity problem. Regression analysis was performed, and the regression results are shown in

Table 3. Estimates the impact of cross-city commuting behavior on housing purchases.

Variable Name	The Estimated Results of the Benchmark Model			The Estimated Results of the Intermediary Effect Model
	Full Sample	Residential Peripheral City	Work Core City	Full Sample	Residential Peripheral City	Work Core City
Whether to commute across town	1.71 ***	2.33 ***	1.50 **	1.82 ***	1.97 ***	1.05
Raise income				1.76 ***	2.08 ***
Reduce cost				0.73 **		1.66 **
Gender	1.17 *	1.30 **	1.01	1.17 *	1.24 *	1.00
Educational level	1.22 ***	1.18 ***	1.22 ***	1.24 ***	1.18 ***	1.22 ***
Family size	1.98 ***	1.80 ***	2.06 ***	1.96 ***	1.77 ***	2.07 ***
Relocation intention	0.57 ***	0.51 ***	0.51 ***	0.56 ***	0.50 ***	0.51 ***
Household registration	4.20 ***	4.91 ***	4.16 ***	4.19 ***	4.96 ***	4.21 ***
Gross annual household income	1.07 *	0.96	1.09 *	1.09 **	0.96	1.08
Constant	0.06 ***	0.07 ***	0.06 ***	0.04 ***	0.05 ***	0.06 ***
Wald χ2	377.42 ***	178.92 ***	244.65 ***	420.51 ***	208.86 ***	256.43 ***
Pseudo R2	0.1140	0.1098	0.1299	0.1276	0.1293	0.1327
Number of valid samples	3051	1451	1829	3051	1451	1829

Note: *, **, and *** represent the significant levels of 10%, 5%, and 1%, respectively; all variable coefficients are OR values.

. It should be noted that the first three columns in Table 3 are the estimated results of the benchmark model of the whole sample, living in the peripheral city (control group, all work and live in the peripheral city) and working in the core city (control group, all work and live in the core city), and the last three are the estimated results of the intermediary effect model.

4.2.1. The Estimation Results and Analysis of the Benchmark Model

From the effect of each variable in the benchmark regression model on the dependent variable, the results were as follows:

(a)

The cross-city commuting behavior of the labor force in the metropolitan area significantly promotes the increase in housing purchases. Specifically, regardless of the whole sample, the subset residing in the peripheral cities, or the subset employed in the core cities, the probability of the labor with cross-city commuting behavior choosing to buy housing in the peripheral cities of the metropolitan area was significantly higher than that of the labor without cross-city commuting behavior when other variables were controlled. In the whole sample, the probability that the labor force with cross-city commuting behavior chose to buy housing in the outer cities was 1.71 times that of the labor force without commuting behavior; in the sample living in the outer cities, the probability that the labor force with cross-city commuting behavior chose to buy housing in the outer cities was 2.33 times that of the labor force without commuting behavior; in the sample working in the core cities, the probability of purchasing housing in the outer cities was 2.33 times that of the labor force without commuting behavior. The probability that the labor force with cross-city commuting behavior chose to buy housing in the outer cities of the metropolitan area was 1.50 times that of the labor force without commuting behavior, and the above results were highly significant at the significance level of 1% or 5%. It shows that the cross-city commuting behavior of the labor force in the metropolitan area significantly affects the spatial distribution of housing purchases and increases the housing purchase demand in the peripheral cities. Hypothesis 1 was tested.

(b)

From the effect of each control variable on the housing purchase, gender, education level, family size, relocation intention, and household registration all significantly affected the housing purchase choice in the metropolitan area. The main finding suggests that the probability of buying housing is higher for male workers with no intention to move and local household registration than for female workers with intention to move and local household registration. The higher the education level, the larger the family size, and the higher the annual household income, the higher the probability of buying housing.

4.2.2. Estimation Results and Analysis of the Intermediary Effect Model

The last three columns in Table 3 report the estimation results of the test of the influence mechanism of cross-city commuting behavior of the labor force on housing purchase in metropolitan areas.

Table 4. Estimates of intercity commuting behavior on intermediary variables.

Variable Name	Full Sample		Sample of Residents in Peripheral Cities		Sample Working in Core Cities
	Raise Income	Reduce Cost	Raise Income	Reduce Cost	Raise Income	Reduce Cost
Whether to commute across town	3.05 ***	27.50 ***	2.85 ***	/	/	53.18 ***
Control variable	Control	Control	Control	/	/	Control
Wald χ2	68.17 ***	364.87 ***	54.72 ***	/	/	416.73 ***
Pseudo R2	0.0183	0.1765	0.0307	/	/	0.3596
Number of valid samples	3051	3051	1451	/	/	1829

Note: *** represent the significant levels of 1%; “/” indicates that this situation is not applicable.

shows the regression estimation results of cross-city commuting on income improvement and cost reduction of intermediary variables. The following can be seen:

First of all, regardless of the whole sample, the sample living in the peripheral city or the sample working in the core city, when controlling for other variables, cross-city commuting behavior exhibited a positive promotional effect on the increase in income and the reduction of cost of the intermediary variable, and the regression estimation results were highly significant at a 1% significance level. It shows that the labor force with cross-city commuting behavior achieves the purpose of increasing income and reducing housing cost.

Secondly, in the whole sample, after adding intermediary variables, although the effects of increasing income and reducing cost on housing purchase were highly significant at the significance level of 1%, the regression coefficient of cross-city commuting increased (1.71 < 1.82), and the effect of reducing costs on housing purchases was negative (OR value < 1). This indicates that there was a masking effect in the intermediate effect of this sample (a masking effect is a special case of mediating effect, which is mainly manifested by the different symbols of indirect effect and direct effect). The possible reasons are as follows: First, the reference group was mixed in the total sample, which led to the masking of the conduction effect of the intermediary effect. The reference group of the sample (non-cross-city commuter labor force) included both the samples who both work and live in peripheral cities and the samples who both work and live in core cities, with the intermediary mechanism of these two samples being different from that of the experimental group (the labor force with cross-city commuting behavior). For example, when the samples who both work and live in peripheral cities were selected as the reference group, the housing market where all the samples are located can be considered homogeneous, and cross-city commuters may mainly promote housing purchase through the intermediary effect of increasing income, while the path of reducing the cost of housing purchase is not applicable. When the samples who both work and live in core cities were selected as the reference group, the income level of all samples can be considered to be the same. Cross-city commuters may promote housing purchase mainly through the intermediary effect of reducing the house purchase cost, while the path of increasing income is not applicable. If they are not distinguished, the results obtained will be disturbed. Second, in the total sample, there was an overvaluation of the housing price in the place of living than the housing price in the place of work, resulting in a negative intermediary effect of cost reduction. Although housing prices in core cities are generally higher than those in peripheral cities, and there is a significant difference in housing prices, there is no problem that there is an intermediary mechanism to reduce costs between core cities and peripheral cities. However, in core cities or peripheral cities, the degree of integration between workplace and residence is relatively high, and the labor force generally lives in areas with high housing prices but work in areas with low housing prices. In the statistical process, samples that should be identified as having no difference in housing prices are classified into cases where the housing prices of places to live are higher than those of workplaces, resulting in deviations in the estimation results.

Finally, the regression results of the sub-sample intermediary effect model of the reference group were determined, which showed that compared with the labor force who both work and live in the peripheral cities or the labor force who both work and live in the core cities, cross-city commuters promoted the housing purchase by increasing the relative income and the housing price gap to reduce the cost, respectively. Hypothesis 2 and Hypothesis 3 were verified. Hypothesis 2 and Hypothesis 3 were tested.

After sub-sample processing, in regression models with mediating effects, the positive promotion effects of increasing income and reducing cost on housing purchase were highly significant at the significance levels of 1% and 5%, respectively, and the regression coefficients of cross-city commuting variables decreased (2.33 > 1.97; 1.50 > 1.05). Since the coefficient of cross-city commuting was significant in the samples where people live in peripheral cities (the control group both working and living in peripheral cities), while it was not significant in the samples where people work in core cities (the control group both working and living in core cities), it can be preliminarily concluded that there was a partial intermediary effect in increasing income in the former sample. The cost reduction in the latter sample was a fully mediating effect. Further, using the KHB test method, it was found that the Z-values of income improvement and cost reduction of the intermediary variables in the two samples were both greater than 1.65 (since whether the explained variable buys a house is a binary categorical variable, the traditional Sobel statistic was no longer applicable, so the KHB test method was used), indicating again that compared with the labor force working and living in peripheral cities, cross-city commuters mainly promote housing purchase through the path of increasing relative income. Cross-city commuters mainly promote housing purchases through the path of cost reduction through the housing price gap, as shown in

Table 5. Results of the mediation effect test and decomposition.

Variable Name	Full Sample		Sample of Residents in Peripheral Cities		Sample Working in Core Cities
	Coefficient (Z)	Relative Contribution (%)	Coefficient (Z)	Relative Contribution (%)	Coefficient (Z)	Relative Contribution (%)
Direct impact ${\gamma }_1$	1.82 *** (2.93)	111.96	1.97 *** (3.79)	79.36	1.05 (0.20)	11.56
Indirect effect ${\beta }_1{\gamma }_2$	0.94 (−0.70)	−11.96	1.19 *** (4.50)	20.64	1.45 ** (2.36)	88.44
Raise income	0.15 *** (5.36)	27.44	0.18 *** (4.50)	20.64	/	/
Reduce cost	−0.21 *** (2.41)	−39.40	/	/	0.37 *** (2.36)	88.44

Note: **, and *** represent the significant levels of 5%, and 1%, respectively; the value in brackets is the z-statistic (equal to the coefficient/standard error of the intermediate variable); “/” indicates that this situation is not applicable.

. According to the channel decomposition of the intermediary effect, it was found that for the sample whose residence was in the peripheral city (the control group both working and living in peripheral cities), 20.64% of the total impact of cross-city commuting on housing purchase was generated through the indirect channel of increasing income, and the remaining 79.36% was directly affected. In the sample whose workplace is all in core cities (the control group both working and living in core cities), 88.44% of the total impact of cross-city commuting on housing purchase was indirect through cost reduction, and the remaining 11.56% was direct impact.

4.3. Endogeneity Tests

In addition, the model may be subject to the endogeneity issue arising from omitted variables and reverse causality. For instance, the reason why the labor force produces cross-city commuting behavior is mainly due to them having bought houses in peripheral cities before working. To address the endogeneity issue, this paper used commuting efficiency, the ratio of commuting distance to commuting time, as an instrumental variable for the endogeneity test. In principle, the logic of choosing commuting efficiency as an instrumental variable is that individuals with higher commuting efficiency are generally more likely to be cross-city commuters. This is because non-cross-city commuters have shorter commuting distances but less efficient modes of transportation, resulting in lower commuting efficiency. According to the 2020 Annual Report on Transportation Development of Guangzhou published by the Guangzhou Municipal Bureau of Planning and Natural Resources in August 2021, one of the reasons for the labor force to choose cross-city commuting is the longer commute time within the city (source: https://baijiahao.baidu.com/s?id=1720637333094551725&wfr=spider&for=pc (accessed on 6 July 2024)), so it can be considered that the commuting efficiency meets the correlation of instrumental variables, but the commuting efficiency has no direct impact on whether individuals buy or rent houses. It also satisfies the externality of instrumental variables. Learning from Yuan Wei’s (2018) method [30], the instrumental variable IV Probit method was used to conduct the endogeneity test, and the results are shown in

Table 6. Results of the endogeneity test.

Variable Name	Full Sample	Sample of Residents in Peripheral Cities	Sample Working in Core Cities
One-stage coefficient	0.06 ***	0.28 ***	0.04 ***
One-stage F value	139.98 ***	100.54 ***	198.41 ***
Two-stage Wald test results	p = 0.36	p = 0.46	p = 0.75

Note: *** indicate the significance levels of 1%.

. The results show that the p-values of the Wald test for all samples were greater than 0.1, and the null hypothesis that the core explanatory variable is an exogenous variable was not rejected. Moreover, from the regression results of the first step, the influence coefficients of commuting efficiency of instrumental variables on the cross-city commuting of explanatory variables were highly significant at the significance level of 1%, and the instrumental variables had a strong explanatory power on the cross-city commuting of explanatory variables. Therefore, it can be considered that cross-city commuting is an exogenous variable, and there is no endogenous problem in the model, that is, the research conclusion of this paper is reliable.

4.4. Robustness Tests

The above regression results may have some contingencies due to model selection or special sample size, which may lead to the unrobustness and untrustworthiness of the regression results. For example, due to the special historical background of the labor force over the age of 60 in the sample, the proportion of homeownership is very high, which may interfere with the conclusion. In order to ensure the robustness and reliability of the above estimation results, this paper conducted a robust test by transforming the model (using the Probit model) and changing the sample size (deleting samples over 60 years old), respectively (due to space limitations, the detailed results of the robustness test are not shown; if you are interested, please contact the author). From the estimation results, no matter whether the model was transformed or the sample size was changed, the influence direction and significance of each variable coefficient did not change much. Therefore, the estimation results of the binomial Logistic model above can be considered robust and credible.

5. Further Discussion

The cross-city commuting behavior promotes the housing purchase of the labor force and has an important impact on the development of the housing market in the metropolitan area. What are the characteristics of the housing purchasing behavior for the cross-city commuters? Are there certain patterns in the location choices and the number of purchases? In this regard, this paper produces a further discussion from the two aspects of location preference and multi-suite decision making.

5.1. The Geographical Preference of Cross-City Commuter in Housing Purchase

The housing purchase of cross-city commuter workers is a decision made after comprehensive consideration of various factors to pursue the maximization of welfare. Cross-city commuters should not only consider their actual purchasing power, but also take other factors such as commuting and living environment into account. In this paper, sub-sample analysis was carried out on the samples whose work–residence distance was less than 20 km, between 20 and 40 km, and greater than 40 km to clarify the geographical location preference of cross-city commuters in housing purchase. As can be seen from the empirical results in

Table 7. Estimates of the effect of cross-city commuting behavior on home purchase geographic preference.

Variable Name	The Distance between Work and Residence <20 km	The Distance between Work and Residence >20 km, <40 km	The Distance between Work and Residence >40 km
Whether to commute across town	0.47	6.82 ***	1.01
Control variable	Control	Control	Control
Wald χ2	356.49 ***	91.93 ***	16.01 **
Pseudo R2	0.1268	0.2325	0.1167
Number of valid samples	2573	343	135

Note: **, and *** represent the significant levels of 5%, and 1%, respectively.

, compared with the samples whose work–residence distance is less than 20 km and more than 40 km, cross-city commuter workers were found to be more inclined to buy housing within 20–40 km from their workplace. The possible reasons are that, on the one hand, housing prices close to the workplace are also high, and workers’ purchasing power is limited, and on the other hand, although the housing price is lower in areas farther from the workplace, the level of basic public services such as healthcare and education is also not high, resulting in a lower willingness of cross-city commuters to buy houses.

5.2. Cross-City Commuting Behavior and Multi-Suite Purchase Decision

Purchasing multiple dwellings is an important way for residents to invest and meet the needs of higher levels of living. How do cross-city commuters fare when it comes to multi-home purchases? In this paper, regression analysis was carried out on the whole sample, the sample living in peripheral cities, and the sample working in core cities. The results show that in the whole sample and the sample working in core cities, compared with the non-cross-city commuters, the probability of cross-city commuting behavior of the labor force is lower, indicating that cross-city commuting behavior does not promote the multiple dwellings purchase of the labor force. However, among the samples living in peripheral cities, cross-city commuting behavior has no significant effect on whether to buy multiple apartments, indicating that there is little difference between cross-city commuters and workers living in peripheral areas, as shown in

Table 8. Estimates of the impact of cross-city commuting behavior on the purchase of multiple homes.

Variable Name	Full Sample	Sample of Residents in Peripheral Cities	Sample Working in Core Cities
Whether to commute across town	0.59 ***	0.75	0.48 ***
Control variable	Control	Control	Control
Waldχ2	215.02 ***	101.94 ***	129.60 ***
Pseudo R2	0.0643	0.0755	0.0588
Number of valid samples	3051	1451	1829

Note: *** indicate the significance levels of 1%.

. Therefore, cross-city commuting behavior does not promote multiple dwellings purchases by the labor force.

6. Conclusions and Outlook

Applying the data of the 2023 Sampling Survey on the Employment and Housing Status of the Population in China’s Metropolitan Areas, this paper conducted rich and rigorous empirical analysis on the relationship between cross-city commuting behavior and housing purchase in order to reveal the underlying mechanisms. The findings are as follows: (1) The cross-city commuting behavior of the labor force in the metropolitan area has a significant positive promoting effect on housing purchase. (2) Compared with those who work and live in the peripheral cities, the cross-city labor force mainly promotes housing purchases through the means of increasing income. (3) Compared with the labor who works and lives in the core cities, the cross-city labor force mainly promotes housing purchase through cost reduction. (4) Cross-city commuters prefer to buy a house 20–40 km away from their workplace, but fewer are multiple dwellings owners.

With the gradual development of the metropolitan area, the phenomenon of cross-city commuting will become more common. Reasonable and efficient guidance of the cross-city commuting behavior of the labor force and alleviation of the contradiction between working and living will benefit both workers and the construction of the modern metropolitan area. (1) The quality of the metropolitan area should be enhanced to promote the mobility of the elements within the metropolitan area including the labor force, for example, the development of a public transport network system to harmonize the spatial dislocation between working and living. (2) It is necessary to coordinate the sharing of resources in the metropolitan area and narrow the gap in the public goods between the central city and the peripheral cities, thus achieving steady development of the housing market in the metropolitan area. Public goods such as education and healthcare should be decentralized from the central city to the peripheral cities through the establishment of branch campuses, so as to attract more cross-city commuters to settle down in the peripheral cities. (3) More employment should be provided to support cross-city commuters in core cities. Corresponding measures include but are not limited to providing transportation and catering subsidies for cross-city commuters. (4) It is necessary to stabilize housing prices and strengthen the monitoring of housing prices in peripheral cities. It is also necessary to prevent the sharp rise of housing prices, especially in peripheral cities. Maintaining the stability of the housing market and improving the affordability of housing in peripheral cities is of great significance for improving the welfare of cross-city commuters by achieving their homeownership.

Of course, there are still some limitations in this paper. For example, the scope of application of the model has certain Chinese characteristics, which may not be suitable in some countries. This stems from the Chinese residents’ attachment to housing. Perhaps as residents’ attitudes change, the results will favor renting rather than buying, and we will continue to conduct research and share these changes.