How Does Perceived Neighborhood Environment Affect Commuting Mode Choice and Commuting CO2 Emissions? An Empirical Study of Nanjing, China

Exploring the impacts of perceived neighborhood environment on commuting behavior and travel-related CO2 emissions helps policymakers formulate regional low-carbon transport policies. Most studies have examined the impact of the objective measures of built environment on travel behavior and related CO2 emissions, and few studies have focused on perceived neighborhood environment. This study develops a structural equation model and uses data from a self-administered survey of urban full-time employees in Nanjing, China to examine the direct and indirect effects of perceived neighborhood environment on commuting mode choice and commuting CO2 emissions. The study shows that perceived service facilities has a significant direct effect on commuting mode and a significant indirect effect on commuting CO2 through the mediating effect of commuting mode choice. While socio-demographic variables such as gender have a significant direct impact on commuting mode and commuting CO2 emissions, they have an indirect impact on commuting mode and commuting CO2 emissions through the intermediate variables (such as car ownership, perceived neighborhood environment and commuting distance). The conclusions of this study show that the potential of commuting CO2 emissions reduction in China is enormous, and that policy interventions on commuting would help developing countries such as China achieve the goals of low-carbon transport and sustainable development.


Introduction
Cities are not only the centers of economic development and human activities, but also tremendous sources of carbon emissions [1]. In cities, transportation is identified as one of the priority sectors for decarbonization [2]. However, transportation is the most non-renewable energy-dependent sector in the world [3]. Compared to other sectors, transportation has experienced the most difficulty in achieving CO 2 emissions reduction [4]. Clearly, reducing energy consumption and CO 2 emission in the transportation sector contributes significantly to climate change mitigation [5]. Some of the countermeasures to achieve low-carbon cities and green transport should also include reducing CO 2 emissions from the transportation sector [6,7]. As an important travel purpose, commuting accounts for nearly 50% of the total travel [8,9]. Discovering the reasons behind commuting mode choice and related CO 2 emissions is vital for the construction of low-carbon cities and the formulation of sustainable transport policies and schemes in regions. Therefore, encouraging non-motorized commuting and reducing CO 2 emissions while commuting will help developing countries such as China reduce CO 2 emissions in urban transportation.
Unlike western developed countries, China is in the process of rapid urbanization and urban construction. High-speed economic development, urban spatial expansion and increased car ownership have resulted in an increasing proportion of employees who choose cars to commute. This shift has led to rapid growth in urban transportation demand with related energy consumption and CO 2 emissions. In 2019, transport sector CO 2 emissions accounted for about 10% of China's total energy-related CO 2 emissions [10]. Meanwhile, with the acceleration of China's urbanization process, the urban environment has also undergone tremendous changes such as diversification of land use and suburbanization of housing. Changes in the urban built environment affect residents' travel behavior and then generate related CO 2 emissions [5,11]. Furthermore, changes in the objective built environment of cities affect residents' perception of the built environment. Changes in the perception of the built environment may affect residents' travel behavior, which further affects residents' travel CO 2 emissions [12,13]. Therefore, examining the impact of the perceived built environment on travel behavior and related environmental consequences is important for policymakers to promote low-carbon travel behaviors by improving the urban environment.
While scholars have been increasingly interested in combining travel behavior, travel CO 2 emissions with urbanization [14,15], urban form [16][17][18] and land use [19][20][21] to reduce CO 2 emissions in the transportation sector, the current literature is more focused on exploring the correlations among objective measures of the built environment, travel behavior and travel CO 2 emission [22][23][24][25][26][27]. As for the impacts of the perceived built environment on travel behavior and travel-related CO 2 emissions, researchers have paid limited attention [28][29][30], especially regarding the impact of perceived neighborhood environment on travel CO 2 emissions. In fact, the perceived environment not only has the mediating effect in the impact of the built environment on travel behavior [13], but also has a direct impact on active travel behavior [31][32][33]. Currently, we know little about the impact of the perceived built environment on travel modes other than active travel.
The built environment is an important determinant of travel behavior [34]. Urban-level built environment and transportation planning are of great benefit to achieving the regional sustainable planning goals [23,35,36]. With the support of the above views, numerous studies have linked the built environment with travel behavior and travel CO 2 emissions. Some studies have suggested that built environment factors such as land use diversity and density affect travel distance and travel mode choice, and travel distance and travel mode are closely related to transportation CO 2 emissions [37][38][39][40][41][42]. Ding et al. examined the effects of the built environment on travel distance and energy consumption and found significant differences between commuting and non-commuting trips [24]. Ten Dam found that fulltime work is associated with higher energy consumption [43]. These studies suggest that the impact of the built environment may be different for different travel purposes and types of work. Therefore, further research on commuting travel of urban full-time employees is of great significance.
Most of the above studies only have focused on the impact of objective measures of the built environment on travel behavior and related CO 2 emissions. Research on the perceived built environment has been more focused on its association with physical activity (PA) [44][45][46][47]. Other studies have focused on the relationship between perceived environment and active travel, but most of these studies only focused on the relationship between perceived environment and children or adolescents' active commuting to school [48,49] and adults' active travel [31,32,50], and seldom on the impact of the perceived built environment on commuting behavior and related CO 2 emissions of urban full-time employees. Recent studies have found that perceived high land use diversity, the existence of alternative routes, perceived cycling infrastructure, aesthetic characteristics and green space can promote pedestrian traffic and bicycle traffic [31,50]. However, we know little about the impacts of the perceived built environment on travel modes other than active travel modes [29]. A study in rural China found that perceived accessibility and preference have positive impacts on the probability of choosing to walk, and safety and neighborhood harmony have positive impacts on the frequency of motorcycle and private car trips [51]. In addition, perceptions had a mediating effect in the impact of the objective built environment on travel behavior [13]. Hong and Chen found that built environment factors such as traffic convenience and density affect perceived safety from crime, and further affect walking behavior [12].
In summary, without examining the effects of the perceived built environment on travel modes other than active travel mode and on travel-related CO 2 emissions, the abovementioned studies focus primarily on the impact of the objective measures of the urban built environment on travel behavior and related CO 2 emissions. However, the existing studies seldom focus on the specialized group of urban full-time employees for whom the daily commuting behavior has an important impact on transportation CO 2 emissions. Furthermore, most of the current studies only consider the direct effect of the built environment on travel CO 2 emissions while ignoring the mediating effect of travel behavior. Therefore, this paper develops a structural equation model to examine the direct and indirect effects of perceived neighborhood environment on commuting mode choice and commuting CO 2 emissions of employees in Nanjing, and to better understand the mechanism of the connection among perceived environment-travel behavior-environmental consequences. The core questions of this study are: (1) Does perceived neighborhood environment affect commuting mode choice and commuting CO 2 emissions? (2) Does commuting mode choice have a mediating effect in the impact of perceived neighborhood environment on commuting CO 2 emissions?
This study contributes threefold to the literature. First, we examined the impact of perceived neighborhood environment rather than objective measures of built environment on commuting mode choice and commuting CO 2 emissions, which has received less attention in the literature. Second, our research objects are urban full-time employees with relatively fixed daily commuting behavior. Their commuting behavior has an important impact on urban transportation CO 2 emissions as research has found that full-time work is associated with higher energy consumption [43]. Third, we used a structural equation model to examine the mediating effect of commuting mode choice in the impact of perceived neighborhood environment on commuting CO 2 emissions. Our research provides implications for the formulation of urban commuting CO 2 emissions reduction policies.
The remainder of this paper is organized as follows. Section 2 describes the study area, self-administered survey, variables in the model and modeling approaches. Section 3 presents calculation results of commuting CO 2 emissions and results of the structural equation model. Section 4 presents research conclusions and discussion. This study uses Nanjing, a core city in the Yangtze River Delta region in eastern China, as the study area. Nanjing is the capital city of Jiangsu province with its socioeconomic development level at China's forefront. As with the development of most Chinese large cities, Nanjing has experienced rapid urbanization and motorization since the 21st century. From 2000 to 2019, the urban construction area increased from 194 km 2 to 972 km 2 , the urbanization rate increased from 53.41% to 83.20% [52,53] and the number of private cars increased from 27,413 to 2,111,876 [54,55]. The rapid urbanization and motorization of Nanjing has led to corresponding changes in the commuting mode of urban employees. Employees are increasingly dependent on private cars for commuting, and commuting CO 2 emissions have also entered a stage of rapid growth. In addition, Nanjing has continuously strengthened its efforts in the construction of urban environment and transportation infrastructure in recent years. These efforts will inevitably affect the subjective perception of urban employees on the built environment, and this in turn will lead to a new impact on the commuting mode. Therefore, based on the highly representative study area of Nanjing, this is an important case for empirical research on Chinese cities.

Data Collection
Data for the study were obtained from a retrospective questionnaire survey conducted between November 2017 and January 2018. The administrative division of Nanjing includes 11 urban areas. We selected the main urban area (including 6 administrative districts) with relatively concentrated population and employment for research. Our sampling rule was to randomly select 8 administrative streets in the main urban area and randomly select a community in each street. According to the traffic environment, leisure environment and socio-demographic characteristics of the communities, we divided these communities into 4 types (Figure 1). Among the 8 communities, Yunnanlu community and Yujiaxiang community, located in the old town area, have a good traffic environment and a poor leisure environment, and belong to type I community. Zhong'ao community and Fengqiyuan community, located in Hexi new town, have a good traffic environment and a good leisure environment, and belong to type II community. Huilinlvzhou community and Suojincun community, located close to Xuanwu Lake scenic area, have a poor traffic environment and a good leisure environment, and belong to type III community. Jingmingjiayuan community and Xingweicun community, located at the edge of the main city, have a poor traffic environment and a poor leisure environment, and belong to type IV community. Before conducting the questionnaire survey, we established contact with eight community neighborhood committees to explain the purpose and use of our questionnaire and invite them to participate. After receiving affirmation, we entered the communities to issue questionnaires. The members of our research group pre-distributed the questionnaire. Based on the feedback from the research group members, we revised the questionnaires and distributed them formally. to a new impact on the commuting mode. Therefore, based on the highly representative study area of Nanjing, this is an important case for empirical research on Chinese cities.

Data Collection
Data for the study were obtained from a retrospective questionnaire survey conducted between November 2017 and January 2018. The administrative division of Nanjing includes 11 urban areas. We selected the main urban area (including 6 administrative districts) with relatively concentrated population and employment for research. Our sampling rule was to randomly select 8 administrative streets in the main urban area and randomly select a community in each street. According to the traffic environment, leisure environment and socio-demographic characteristics of the communities, we divided these communities into 4 types ( Figure 1). Among the 8 communities, Yunnanlu community and Yujiaxiang community, located in the old town area, have a good traffic environment and a poor leisure environment, and belong to type I community. Zhong'ao community and Fengqiyuan community, located in Hexi new town, have a good traffic environment and a good leisure environment, and belong to type II community. Huilinlvzhou community and Suojincun community, located close to Xuanwu Lake scenic area, have a poor traffic environment and a good leisure environment, and belong to type III community. Jingmingjiayuan community and Xingweicun community, located at the edge of the main city, have a poor traffic environment and a poor leisure environment, and belong to type IV community. Before conducting the questionnaire survey, we established contact with eight community neighborhood committees to explain the purpose and use of our questionnaire and invite them to participate. After receiving affirmation, we entered the communities to issue questionnaires. The members of our research group pre-distributed the questionnaire. Based on the feedback from the research group members, we revised the questionnaires and distributed them formally. The data were collected through face-to-face structured questionnaires filled out by random sampling method, and the respondents were recruited in community public spaces. When distributing the questionnaires, we prepared small gifts (including some The data were collected through face-to-face structured questionnaires filled out by random sampling method, and the respondents were recruited in community public spaces. When distributing the questionnaires, we prepared small gifts (including some daily necessities such as handkerchief papers or wet wipes, worth about USD 2) as an incentive to participants. We recorded the commuting behavior, perceived neighborhood environment and socio-demographic characteristics of full-time employees aged 18 and above. Respondents were asked to recall their commuting behavior from the past week, including the different modes of transportation they chose to travel between home and work and their corresponding time. If there was a transfer behavior, participants could record different transportation modes and their corresponding times. Respondents were also asked to provide their workplace address. Some respondents, concerned about potential privacy leakage, were unwilling to provide the address of their workplace address, and so we asked them to indicate the bus station or metro station closest to their workplace. We used the above information to calculate commuting distance and commuting CO 2 emissions of employees. Respondents also filled in their perceptions of the neighborhood environment. These question settings used a 5-point Likert scale, with 1 representing "completely disagree" and 5 representing "completely agree". Meanwhile, we determined whether the respondents had moved in the past five years according to the number of years they had lived in the current community to eliminate the impact of residential selfselection. Considering that employees are usually away from their home on weekdays, we chose to conduct questionnaire surveys during the weekends when employees were at home. We distributed a total of 1200 questionnaires (150 in each community), recovered 1102 questionnaires and collected 622 questionnaires from employees who had complete commuting information and had not moved in the past five years. The built environment characteristics and sample characteristics of different types of communities are shown in Table 1.  1 China's hukou system refers to a household registration system required by law to officially identify every citizen as a resident of a certain area. Under this system every citizen is categorized according to the type of hukou (agricultural/non-agricultural) and the place of hukou registration (urban/rural areas) [56,57].

Calculation of Commuting CO 2 Emissions
Considering the availability and accuracy of data, this study uses a method commonly used internationally to calculate individual commuting CO 2 emissions. In other words, this study uses the commuting mode and commuting distance of each employee to calculate commuting CO 2 emissions [22,51,58,59]. The commuting distance (CD) of each sample was calculated by multiplying the average speed of each employee's chosen mode of transportation by the corresponding commuting time. Commuting distance was calculated using the following formula: where CD ij represented the commuting distance of the employee i using the transportation mode j, V ij was the average speed of the transportation mode j during the peak commuting period (obtained through field investigation) and T ij was the commuting time of the employee i using the transportation mode j. Once the commuting distance was measured for each employee using different modes of transportation, the CO 2 emissions per commute could be calculated. According to different transportation modes, a CO 2 emission factor was assigned to each distance. The  Table 2. The calculation of CO 2 emissions was as follows: where CE i represented the daily commuting CO 2 emissions of employee i, CD ij was the one-way commuting distance of employee i using transportation mode j and F j was the CO 2 emission factor corresponding to transportation mode j. It is assumed that employees commute to and from work twice a day and use the same means of transportation each time. Due to the limitation of data acquisition, the calculation of commuting CO 2 emissions in this study did not consider the impact of fuel type, vehicle type, vehicle speed and other factors. In this study, commuting mode (CM) was classified into four categories according to the CO 2 emission factors for different transportation modes. The CO 2 emission factors for walking and biking were both 0, and so they were classified into one category of commuting mode (namely walking/biking commuting mode). Electric bicycle, as a more common mode of motorized transportation for short-distance travel in China, had a small CO 2 emission factor, and so it was classified as the electric bicycle commuting mode. The CO 2 emission factors for public transport such as subway, bus and unit shuttle bus were relatively large, and so they were regarded as the public transportation commuting mode. The CO 2 emission factor was the largest when private cars and taxis were chosen, and so we classified cars and taxis into one category (namely car commuting mode).
For the 622 samples collected in this paper, in terms of commuting mode choice, the proportion of employees who chose public transport commuting mode was the largest at 36.66%, followed by the walking/bicycle commuting mode for 30.23% of the employees. The proportion of employees choosing car commuting mode and electric vehicle commuting mode was 22.51% and 10.61%, respectively. Descriptive statistics of commuting modes are shown in Table 3.   (Table 4). This indicates that the validity of the data in the questionnaire was good. Next, we performed exploratory factor analysis on 18 variables of perceived neighborhood environment in the questionnaire. To analyze whether the perceived neighborhood environment variables satisfy the prerequisites of factor analysis, that is, whether there was a strong correlation among the items, KMO and Bartlett tests were conducted on the 18 variables of perceived neighborhood environment. It was verified that the KMO value was 0.777, and the significance of the Bartlett sphericity test value was 0.000. This indicates that the correlation coefficients among the items were both significant and suitable for factor analysis.
The factors were further rotated orthogonally using the maximum variance method to make them more convincing and explanatory. The cumulative variance contribution rate of the five common factors was 58.22%. From the rotation component matrix table (Table 5), combined with the meaning of each item, the five common factors were defined as service facilities perception, environmental quality perception, road condition perception, traffic safety perception and community safety perception, forming the latent variables of perceived neighborhood environment in the structural equation model. The means, standard deviations and standard errors of the latent variables of perceived neighborhood environment are shown in Table 6.

Socio-Demographic Characteristics
Socio-demographic characteristics are crucial to understanding travel behavior and travel CO 2 emissions. The existing relevant literature has proven that gender, age, income, education, occupation, household size and hukou not only affected travel behaviors [62,63], but also affected travel CO 2 emissions [5,11,22,24,60]. In this study, the above socio-demographic characteristics were considered as exogenous variables introduced into the model. The impact of car ownership is more complicated, with some scholars believing that car ownership has a mediating effect between the exogenous variables and other endogenous variables [11,64,65]. Therefore, car ownership was set as an endogenous variable in this research. Table 7 presents the socio-demographic characteristics of the sample.

Structural Equation Model and Conceptual Framework
In recent years, scholars have often used structural equation modeling to analyze the complex relationships among built environment, travel behavior and related CO 2 emissions [11,17,64,66]. Structural equation modeling is a multivariate data analysis tool that analyzes the relationships among variables based on the covariance matrix of variables. It integrates factor analysis and path analysis. Structural equation models can not only solve the endogeneity problem among variables, but also allow the existence of mediating variables [26,66]. Because there were mediating variables such as commuting distance and commuting mode in this study, it would have been difficult to support the analysis using traditional multiple regression methods. Thus, we used structural equation modeling to analyze the direct and indirect impact of perceived neighborhood environment on commuting mode choice and commuting CO 2 emissions through the effect values. Because of the subjective aspect of human behavior and their different life experiences, attitudinal preferences and socio-demographic characteristics, even in the face of the same urban built environment, different people have different subjective perceptions of the built environment, and their commuting mode choices and commuting CO 2 emissions will differ to some extent. Therefore, the mechanism behind commuting mode choice and commuting CO 2 emissions cannot be fully explained from the perspective of the objective measures of the urban built environment. The impact of perceived neighborhood environment must also be considered. In addition, commuting distance has been considered an important factor for commuting mode choice and travel CO 2 emission [11,67]. Thus, commuting distance is included in the model in this paper. Socio-demographic characteristics significantly affect commuting mode choice and commuting CO 2 emissions, so they are also included in the model.
In summary, the conceptual framework of the structural equation model is shown in Figure 2. Through this conceptual framework, we can intuitively see the direct effect of perceived neighborhood environment on commuting mode and commuting CO 2 emis-sions, and how perceived neighborhood environment ultimately affects commuting CO 2 emissions through the mediating effect of commuting mode.

Calculation Results of Commuting CO 2 Emissions
Through calculation, the average one-way commuting distance of urban employees in Nanjing is 12.42 km, the one-way commuting time is 29.59 min, the daily commuting CO 2 emissions per capita is 1.14 kg and the corresponding standard deviations are 13.17 km, 22.19 min and 1.82 kg, respectively. From comparison with other studies in Table 8, we found that the daily commuting CO 2 emissions of our sample are close to the results of other scholars' research on China [5,60], but much lower than the result of Ohnmacht et al. for Switzerland [58].

Goodness of Fit for Structural Equation Model
In this paper, we used AMOS 22 to build the initial model. The model was estimated using the Bollen-Stine Bootstrap because our data were not normally distributed [11,32]. We removed non-statistically significant links (p > 0.1) and re-estimated the model. We then modified the model according to the modification indices (MI) to obtain the final model. The model fit indices and its corresponding reference values [69] are given in Table 9. All indices show that the model fits well and is statistically significant.

Effects among Endogenous Variables
The relationships among endogenous variables are shown in Table 10. In terms of the direct effects among endogenous variables, the service facilities perception in the perceived neighborhood environment variables has a significant direct effect on commuting mode, which indicates that employees with a positive perception of service facilities around the community have a higher probability of choosing walking/bicycle commuting methods. It is understandable that employees have a positive perception of service facilities around the community, meaning they may live closer to the center of the main city rather than the edge of the main city, so they are closer to the workplace and are more likely to choose the walk/bicycle commuting mode. Meanwhile, car ownership and commuting distance have a significant direct effect on commuting mode. This shows that employees with more car ownership and longer commuting distance have a higher probability of choosing car commuting mode. In addition, commuting mode, commuting distance and car ownership have a significant direct effect on commuting CO 2 emissions, which means that employees who choose to commute by car, commute longer distances or own more cars and emit more CO 2 when commuting.
However, perceived neighborhood environment variables have no direct effect on commuting CO 2 emissions. Service facilities perception and commuting distance indirectly affect commuting CO 2 emissions through the mediating effect of commuting mode. That is, employees with a positive perception of service facilities around the community have a higher probability of choosing walking/biking commuting mode, and CO 2 emissions are lower when they choose walking/biking commuting mode; employees with longer commuting distance are more likely to choose the car commuting mode, and CO 2 emissions are higher when choosing the car commuting mode. Car ownership indirectly affects commuting mode choice through the mediating effect of service facilities perception. In addition, car ownership indirectly affects commuting CO 2 emissions through the mediating effect of service facilities perception and commuting mode.

Effects of Socio-Demographic Variables on Endogenous Variables
The relationships between socio-demographic variables and endogenous variables are shown in Table 11.  Among socio-demographic variables, only income directly impacts on car ownership. This indicates that high-income employees own more cars.
Socio-demographic variables directly impact perceived neighborhood environment. Male employees have a positive perception of road conditions and a negative perception of environmental quality, which means that male employees better understood road condi-tions, while female employees better understood environmental quality. Older employees have a positive perception of traffic safety. Freelance employees have a negative perception of traffic safety and a positive perception of road conditions. Meanwhile, employees with local hukou have a positive perception of environmental quality. In addition, sociodemographic variables also have significant indirect effects on perceived neighborhood environment, which comes from the mediating effect of car ownership. Higher-income employees own more cars, while employees with more cars have a positive perception of community safety and a negative perception of service facilities. It is not difficult to understand that most of the high-income employees live in high-end communities, and the safety of such communities is more guaranteed; employees with more cars can easily reach farther distances to obtain services, so they have a negative perception of service facilities around their communities.
Socio-demographic variables directly impact commuting distance, commuting mode and commuting CO 2 emissions. Male employees, employees with larger household size and employees with local hukou commute longer distance, while freelance employees commute relatively shorter distances. Male employees, highly educated employees and employees with local hukou have a higher probability of choosing the car commuting mode, while older employees and freelance employees have a higher probability of choosing the walking/biking commuting mode. Meanwhile, male employees and freelance employees emit more CO 2 when commuting. In addition, socio-demographic variables also indirectly impact commuting mode and commuting CO 2 emissions. This comes from the mediating effects of car ownership and commuting distance. Higher-income employees own more cars, and so they have a higher probability of commuting by car and emit more CO 2 . Male employees, local hukou employees and employees with larger household size have longer commuting distances, and so they are more likely to choose cars to commute and emit more commuting CO 2 emissions. Freelance employees have relatively shorter commuting distances, and so they have a higher probability of choosing the walking/bicycle commuting mode; this results in lower commuting CO 2 emissions. Highly educated employees are more likely to choose cars to commute, and so their commuting CO 2 emissions are relatively higher. Older employees are more likely to choose the walking/biking commuting mode, and so their commuting CO 2 emissions are relatively lower.
In general, the indirect effects of socio-demographic characteristics on commuting CO 2 emissions are more significant than the direct effects, and so we cannot ignore the mediating effects of commuting behavior, including commuting distance and commuting mode.

Conclusions and Discussion
Taking Nanjing as the case and using questionnaire data to estimate the conceptual model of a structural equation model, this study examined the direct and indirect effects of the perceived neighborhood environment on commuting mode choice and commuting CO 2 emissions of urban full-time employees. The following main conclusions and policy implications were determined: (1) Using full-time employees in Nanjing as a sample, the average daily commuting CO 2 emission per employee is 1.14 kg. If Chinese people work 250 days per year, each urban employee would emit 285 kg of CO 2 every year due to commuting. It is evident that commuting is an important part of residents' daily travel, and its CO 2 emission reduction potential is enormous. If we can encourage employees to shift from high-carbon car commuting to green and low-carbon walking/biking and public transportation commuting from the perspective of changing their commuting behavior, it will not only alleviate traffic congestion in Chinese large cities and promote the construction of low-carbon cities in China, but also take part in achieving China's carbon neutrality target by 2060. (2) Among the perceived neighborhood environment variables, the service facilities perception directly affects commuting mode choice, and perceived neighborhood environment ultimately affects the commuting CO 2 emissions of employees indirectly through the mediating effect of commuting mode. Therefore, in climate change mitigation, it is more beneficial to change residents' behavior patterns through urban planning tools. In some developed countries, perceived safety and aesthetic characteristics often promote walking and cycling [30,50], so improving residents' perceptions of safety and aesthetics can promote low-carbon travel. In developing countries such as China, a good service facilities perception may be more important for promoting low-carbon commuting mode choice and reducing related CO 2 emissions. Therefore, improving service facilities around communities should become one of the key dimensions of urban low-carbon transportation construction in China. (3) Commuting distance and commuting mode directly affect commuting CO 2 emissions, and commuting distance indirectly affects commuting CO 2 emissions of employees through the mediating effect of commuting mode choice. Since the impact of perceived neighborhood environment on commuting mode choice is limited (only the impact of service facilities perception is significant), shortening the commuting distance of employees and promoting their choice of walking/bicycle commuting are some of the effective measures to reduce the commuting CO 2 emissions. In the past two decades, China's urban form has continued to expand, and the job-housing imbalance is one reason for the increasing commuting distance of urban employees [70,71]. Thus, it is necessary to develop compact urban forms in the future. Meanwhile, with the advancement of internet communication technologies, more diverse forms of work can be explored. Current studies have also proven that coworking, working from home or teleworking could reduce energy consumption and greenhouse gas emissions [58,72]. In addition, accelerating the construction of pedestrian greenways, bicycle paths and public transportation systems, and advocating low-carbon travel behavior are of great significance for encouraging urban employees to choose a green commuting mode and reducing commuting CO 2 emissions. China's first urban mobility industry report, titled "The 2021 Urban Sustainable Mobility Observation Report", showed that public transportation to car commuting time ratios in eight big cities in China were generally from 1.5 to 2.5 [73], indicating that the priority development strategy of public transportation in China still needs to be improved. (4) Socio-demographic characteristics such as gender, occupation and car ownership directly influence commuting CO 2 emissions and indirectly influence commuting CO 2 emissions through mediating variables of commuting distance and commuting mode. These conclusions are beneficial to formulate commuting CO 2 reduction policies for specific groups. For example, for groups with a high propensity to commute by cars-this paper refers to male employees, higher-income employees and employees with more car ownership and local hukou-the Chinese government should increase the cost of their car use. Through measures such as car purchase restrictions, higher vehicle purchase taxes and parking fees, the Chinese government can control the use of cars by urban employees within a reasonable range and encourage such employees to choose a lower-carbon commuting mode, thereby reducing CO 2 emissions from commuting.
Overall, this study fills a research gap on the effects of perceived neighborhood environment on commuting mode choice and related CO 2 emissions, providing some new evidence for the current construction of sustainable transport and low-carbon cities in China and other developing countries. The main limitation of this study is related to the use of CO 2 emission factors, which are closely related to fuel type, vehicle type, vehicle speed and other factors. Due to the difficulty in obtaining such data, this study did not take these factors into account and could only estimate the model using the average CO 2 emission factor for each mode of transportation.