The Impact of New Subway Construction on the Commuting Methods and Time Utilization of Residents Along the Line: A Comparison Before and After the Completion of the Subway in Kunming City

Abstract

Subway construction changes land use patterns, disrupting the balance between traffic supply and demand and influencing residents’ activity and travel behaviors along the route. Previous studies have often overlooked the multiple attributes of the subway and their varying impacts on daily activities. To understand how subway construction affects travel and activity patterns, this study analyzed travel data from two years before and after the opening of the Kunming Subway. Propensity score matching was used to control for socioeconomic changes unrelated to the subway. The findings show the following: (1) The subway’s introduction leads to the development of commercial and civic amenities around its stations, attracting wealthier and more educated residents, which contributes to gentrification. (2) Overall, subway construction extends urban residential and employment areas, increasing residents’ dependence on cars and promoting a more motor-centric lifestyle. As a result, the subway’s ability to reduce car usage is limited. (3) The subway alters the impact of the built environment on travel behavior, with residents closer to the subway experiencing shorter travel distances and reduced activity spaces, while those further away maintain longer travel distances and greater activity areas.

1. Introduction

For a long time, the study of the impact of the built environment on residents’ travel behavior has been a key topic in the field of transportation research [1]. With the rapid acceleration of urbanization in China, the significant expansion of urban areas, and the large-scale construction of transportation infrastructure, the built environment has undergone dramatic changes. At the same time, accompanied by rapid socioeconomic development, residents’ travel behavior has become more diverse. In this context, the impact of the built environment on residents’ travel behavior is undergoing profound changes. Among them, the metro, as a special type of built environment, exerts dynamic impacts on the surrounding environment during its construction, and these effects indirectly influence residents’ travel behavior. Therefore, accurately assessing the impact of metro projects is crucial [2].

It is worth noting that the subway has multiple attributes simultaneously; it can function as a built environment, like a Transit-Oriented Development (TOD) community. It can also serve as a type of transportation facility, such as rail stations. Furthermore, it provides a means of transportation. Depending on its classification, the subway exerts distinct impacts: as a travel mode, it directly influences residents’ commuting patterns; as transportation infrastructure, it indirectly alters travel expenses by enhancing accessibility, thereby influencing residents’ travel decisions [3]; and as a unique built environment, it transforms land use patterns, alters daily lifestyles, and ultimately affects individual travel behaviors [4]. Nevertheless, the current research on the influence of subway on travel behavior focuses more on the single attribute of subway, neglecting to integrate all three key attributes of subway systems and their interrelationships. Consequently, findings on fundamental inquiries remain inconclusive, particularly regarding whether the subway effectively reduces reliance on car travel. This is because the central debate revolves around whether shifts in residents’ travel patterns induced by subway systems stem from alterations in the surrounding built environment due to subway construction or enhancements in transportation accessibility facilitated by the subway itself. Moreover, it questions whether the subway’s influence on the built environment subsequently affects individual mobility. Clarifying these underlying mechanisms is imperative.

Indeed, as a reliable and efficient mode of transportation, the subway has markedly reduced car ownership and usage rates while mitigating surface-level traffic congestion [5,6]. Li studied the impact of metro accessibility on the travel mode choices of the 1980s generation in Shanghai. The study found that improved metro accessibility contributes to reducing car ownership and increasing the likelihood of metro commuting, but there is no evidence to suggest that it has an impact on car usage [7]. Based on two mobility survey datasets, IBRAEVA used an autoregressive method to study the impact of TOD on the travel share of cars and buses. The results show that TOD has a significant suppressive effect on car usage. When TOD is present at both ends of the trip, the car share in the origin–destination (OD) pair decreases even more [8]. Nevertheless, when considering built environment attributes, the subway not only influences residents’ travel choices but also profoundly impacts their behaviors and lifestyles, thereby extending its influence across all facets of daily life [9]. For instance, Yan discovered that while the construction of subways reshapes the urban spatial structure, it also tends to exacerbate the phenomenon of the separation of residence and work. This leads to a significant increase in the spatial scale of residents’ daily household activities [10], resulting in a greater dependence on private cars for both individual commuting and daily family life, potentially further suppressing the use of public transportation [11]. Chatman’s study also found similar results. After controlling for built environment variables, residents living near train stations tended to make longer motor vehicle trips [12]. Hence, by solely focusing on the travel mode aspect of subways while overlooking their built environment influence, one might mistakenly perceive subway construction as fostering car dependency. In light of this, while investigating the influence of subway systems on residents’ behavior, certain scholars additionally accounted for built environment variables. They designated analogous urban zones as the control counterparts for subway routes, examining the effect on leisure walking frequency. Their findings indicate a notable and distinct correlation between subway presence and increased walking activity [13]. Nonetheless, significant disparities persist between this comparable area and an actual subway corridor concerning spatial built environments and demographic structures. To address these methodological limitations, subsequent researchers confined their study to the same spatial scope—specifically, alongside the subway route. Subsequently, researchers conducted comparative analyses using two temporal phases—pre- and post-subway inauguration—to examine shifts in residents’ travel behaviors from the subway’s inception to its establishment. DAI analyzed cross-sectional survey data before and after the opening of the Singapore Circle Line and found that the opening of the Circle Line increased the proportion of rail transit usage while decreasing the proportion of private car usage. However, there was no significant impact on the proportion of bus usage or travel patterns [14,15]. This approach embodies greater scientific rigor and methodological soundness. The primary challenge lies in acquiring relevant data from both pre- and post-subway phases, while mitigating confounding variables like urban social and economic developments during these periods.

Furthermore, according to individual activity theory, travel is a behavior originating from individuals seeking to satisfy their needs and engage in social interactions. Hence, travel behavior is regarded as subordinate to activity planning [16]. Consequently, the subway system exerts distinct effects on both individual travel patterns and overall activity levels. For example, some affluent and well-educated residents reside in proximity to subway stations due to the augmented amenities brought about by the subway’s inauguration. However, their choice of residence does not stem from the intention to utilize the subway for commuting purposes [17]. And further, the decline in car usage may not solely be attributed to the introduction of the subway. Simultaneously, as a result of the incremental enhancement of amenities along the subway route, the lifestyle and commuting patterns of this demographic are likely to undergo substantial transformations compared to their pre-relocation norms. For instance, following relocation to the subway corridor, the concentrated placement of schools mitigated the initial imbalance between educational access and residential placement. Moreover, the profusion of commercial and corporate amenities transformed the routine of parental school commutes from a linear to a multifaceted network, thereby reshaping the traditional ‘home-to-school’ commuting patterns and lifestyle [18]. On the other hand, as a type of infrastructure, the subway exhibits distinct impacts on its surrounding areas [17,19]. Nevertheless, the majority of extant studies confine their analyses to single geographic areas, thereby failing to capture these differential effects.

Finally, few studies have examined the impact of metro systems on residents’ behavior under different scenarios, with most of these studies being conducted in European and American countries. For example, Cao focused on the influence of the Minneapolis metro corridor on residents’ public transport usage [20]. However, these countries have well-established metro systems and excellent urban designs, which differ significantly from the actual situation in many Chinese cities, most of which are in the early stages of developing rail transit. Therefore, the findings from these studies are not directly applicable to practical guidance in China. For this reason, this paper selects the representative “sparsely distributed metro city” of Kunming as the research object to explore the impact of the process of metro construction—going from no metro system to having one—on urban residents’ travel behavior.

In conclusion, underpinned by extensive big data and open-source datasets, and complemented by micro-behavioral survey data, this study categorizes spatially into two types: continuous corridors and discrete stations along subway lines, guided by the principles of scenario classification, stratification, and sub-scaling. The analysis employs nuclear density analysis and other methodologies to assess how subway openings influence the built environment within 0–500 m and 500–1000 m from stations and along corridors. Secondly, leveraging survey data on residents’ travel patterns pre- and post-subway inauguration, we apply the propensity score matching (PSM) method to mitigate the effects of non-subway variables across distinct time frames. Subsequently, we compare residents’ behavioral adaptations along the transit route in both scenarios—with and without the subway—to underscore its impact on daily routines and travel habits. Finally, we designate car travel choice and distance as travel-level variables, and travel time and activity duration as activity-level variables. Separate models are developed for each of these four variables, followed by an analysis of their paths and influential mechanisms to discern direct impacts attributable to the subway and indirect effects mediated by alterations in the built environment.

The paper is organized as follows: Section 1 introduces the research question, reviews the relevant literature, and identifies the research gaps. Section 2 presents the study area and data. Section 3 outlines the research design and modeling methodology. Section 4 analyzes the changes in the built environment and residents’ behavior characteristics before and after the opening of the subway. Section 5 presents the empirical results. Section 6 summarizes the findings and discusses the policy implications.

2. Study Area and Data

2.1. Study Area

The data for this study is derived from the districts of Panlong, Guandu, Wuhua, Xishan, and Chenggong in Kunming City. Kunming, an important central city in Southwest China, is a typical monocentric city. For many years, the majority of the urban population, government administrative agencies, and public amenities were concentrated within the Second Ring Road of the old city, leading to progressively worsening traffic congestion. In response, government agencies, research institutions, and universities relocated to the southern New City area, prompting a significant influx of businesses and residents to Chenggong New Town. During this period, the urban land structure underwent substantial changes, with land development accelerating, and Chenggong New Town in the southern part of the city gradually expanding. However, large-scale infrastructure construction in the suburbs failed to effectively transform the traditional monocentric urban development model, instead exacerbating the job-housing separation in some areas. As a result, many residents began using private cars to commute between the main city and the new city. Meanwhile, the layout of the subway network and the expansion of transportation infrastructure were also incorporated into the city’s development agenda. Consequently, the transformation of the urban job-housing structure, driven by land development, and the large-scale construction of public transportation facilities have placed Kunming at a critical crossroads, where public transportation and private cars are in direct competition.

This research primarily focuses on Subway Lines 1 and 2, which comprise a total of 35 stations. These lines serve as vital transportation corridors in the city, characterized by high population density and discrepancies in urban development. Subway Line 6 is not included in the study due to its minimal influence on demographic and environmental factors. Refer to Figure 1 for the geographical coverage.

2.2. Study Data

This study employed behavioral data from resident travel surveys conducted in Kunming in 2011 and 2016. The surveys were conducted prior to and following the inauguration of Kunming’s inaugural subway system, encompassing the city’s principal urban districts. In 2011, the survey involved participation from over 38,000 individuals across 20,643 households, resulting in 90,400 pieces of detailed travel data. In 2016, the survey included 32,664 households with over 80,000 residents, resulting in more than 183,000 travel records. The surveys collected socio-economic characteristics and travel patterns of respondents and their family members across the entire day.

In addition, we compared the basic social demographics, built environment, and public transportation information of Kunming City in 2011 and 2016, as shown in

Table 1. Basic information of Kunming city in 2011 and 2016.

Indicator	2011	2016	Rate of Change
Kunming City population (ten thousand)	648.64	672.8	1.83%
Regional Gross Domestic Product (billion)	2510	4300	26.28%
Built-up area (km2)	298	412	16.06%
Population density (people/km2)	300.5	320	3.14%
Population density of built-up areas (people/km2)	217.66	163.3	−14.27%
Average annual disposable income (RMB/year)	21,966	36,739	25.16%
Number of vehicles (ten thousand)	150.8	226.8	20.13%
Number of buses	4292	6397	19.69%
Length of subway line (km)	0	64.7	100.00%
Road length (km)	1642	1997	9.76%
Number of motor vehicle parking lots (ten thousand)	11	32.3	193.64%

. Over the five years, with the expansion of urban space and the widening of the urban framework, the activity space for residents has also increased. However, the development of basic service facilities has lagged behind, leading to an increase in travel distances for urban residents. The growth in per capita income and GDP has driven a rapid increase in the number of motor vehicles. At the same time, the accessibility of public transportation has also greatly improved, especially with the development of the subway (which went from non-existent to established). From 2016 to 2025, Kunming made significant progress in the development of urban rail transit. The metro network gradually improved, with the opening of Lines 4 and 5, forming an efficient transportation system with a “cross-shaped” structure as its backbone.

3. Methods

3.1. Research Design

The subway plays an important role in urban expansion and urban renewal. At the same time, the construction of the subway can influence people’s daily travel choices and their responses to the urban environment, indicating that the impact of the built environment on travel behavior changes with the emergence of the subway. Moreover, the subway has multiple attributes, and each of these attributes can have different levels of impact on the residents’ daily behaviors. Therefore, we hypothesize that the impact of the built environment on travel behavior will change during the process of the subway being established. By comparing modeling results from different periods, we aim to gain a deeper understanding of the long-term dynamic relationship between the subway, the built environment, and activity-based travel.

3.2. Variable Selection

The factors influencing residents’ travel and activities mainly come from three aspects: the built environment, residents’ socioeconomic attributes, and self-selection (attitude). First, existing research has shown that as population/employment density increases, the impact of built environment variables (such as population density) on travel behavior tends to weaken, and in some cases, may even reverse direction. Therefore, the influence of built environment factors (such as population density) on travel behavior may differ in the early stages of subway construction (when population density is low) compared to the later stages (when population density is high). Additionally, to eliminate the interference of confounding factors such as gender, household income, and education level, and to assess the true changes in residents’ travel and activities, we have controlled for the socioeconomic attributes of residents in the same strata across different years before modeling. The issue of self-selection has also been mitigated due to geographic location constraints, and the specific methods will be explained in the next section.

Correlations and collinearities among independent variables are examined after selecting indicators. After excluding variables with significant collinearity (VIF > 10), 9 built environment variables and 7 socioeconomic attributes were kept.

Table 2. Classification and definition of variables.

Variable Category	Variable Name	Variable Type
Built environment	Distance to nearest bus stop	Continuous
	Distance to the nearest subway stop	Continuous
	Mixing degree	Continuous
	Number of bus stops	Continuous
	Number of schools	Continuous
	Number of recreational facilities	Continuous
	Number of shopping facilities	Continuous
	Number of site population	Continuous
	Number of intersections	Continuous
Social economic attributes of individuals	Household size	Continuous
	Number of cars in the family	Continuous
	Number of electric bicycle in households	Continuous
	Gender (a)	Categorical
	Annual household income (b)	Categorical
	Level of education (c)	Categorical
	Occupation (d)	Categorical

Note: (a) male = 0; female = 1. (b) Low income = 1; middle income = 2; high income = 3. (c) Elementary education = 1; secondary education = 2; advanced education = 3. (d) Science, education, culture, and health personnel = 1; workers in production enterprises = 2; unemployed and retired = 3; private owners = 4; students = 5.

shows the relevant variables. The diversity index, which measures land use mixing, is calculated using the formula

$$H=-{\displaystyle \sum _{i=1}^n{P}_i\ln P_i}$$

(1)

H is the mixing degree to be calculated; n is the category of POI;$P_i$ is the percentage of the number of Class i POIs in the total number of POIs.

The study analyzes changes in station infrastructure before and after subway inauguration using information entropy, dominance, and balance indices to assess diversity and equilibrium of station amenities. Information entropy measures diversity and complexity of local facilities, dominance and equilibrium indices evaluate evenness of facility distribution within a locale. Equations (1)–(3) present the formulas for these indices.

$$D={\log }_2(n)+{\displaystyle \sum _{i=1}^n{P}_i\ln P_i}$$

(2)

$$E={\displaystyle \frac{{\displaystyle \sum _{i=1}^n{P}_i\ln P_i}}{{\log }_2(n)}}$$

(3)

3.3. Propensity Score Matching

Before and after the subway opened, the social economy experienced rapid development, and the mobility of urban residents further increased. Whether during the construction or operation of the subway, it attracts groups who favor public transportation or are averse to the disturbances caused by traffic, leading to significant changes in the structural proportions of residents from various social strata within the research area, resulting in selection dilemmas. Therefore, simply dividing and conducting statistics based on spatial location may introduce certain biases in the final conclusions. To address this issue and eliminate the confounding effects of socioeconomic changes such as gender, household income, and educational level, we employed the Propensity Score Matching (PSM) method to control for temporal heterogeneity.

Propensity Score Matching (PSM) is a method used to address issues of data bias and confounding factors, commonly applied in fields such as econometric research. Its principle is based on the idea of matching estimators, using observable variables to match data from different groups. The main steps are as follows:

(1)

ArcGIS (version 10.8.1) is a powerful geographic information system software used for map creation, spatial analysis, data management, and visualization. The buffer tool in ArcGIS is used to delineate two circular buffer zones, strictly based on Euclidean distance, namely 0–500 m and 500–1000 m, around the site. Based on these buffer zones, household, member, and travel information located in different ranges before and after the subway opening are extracted. The relevant information extracted is shown in

Table 3. Summary of household travel information of different locations before and after the opening of metro.

	0–500 m Layer			500–1000 m Layer
Time Node	Number of Households	Population of Residents	Travel Information Number	Number of Households	Population of Residents	Travel Information Number
pre-subway opening	1808	3051	7286	2286	3760	8882
post-subway opening	1762	4232	9646	3459	8320	19,603

.

(2)

The residents’ samples after the subway opened are classified as the treatment group, while the residents’ samples before the subway opened are classified as the control group. The samples after the subway opening are used to match the samples before the subway opening, which makes this matching method more aligned with actual conditions and results in minimal loss of useful information.

(3)

A parallel trend test is conducted on the matching results of the treatment group and the control group, with the test results shown in

Table 4. Sample parallel hypothesis test.

Variable	Status	0–500 m Layer				500–1000 m Layer
		Experimental Group	Control Group	Standardized Deviation (%)	p	Experimental Group	Control Group	Standardized Deviation (%)	p
Gender	Before matching	0.499	0.532	−6.63	0.036	0.509	0.507	0.35	0.903
	After matching	0.507	0.503	0.76	0.862	0.499	0.501	0	0.94
Type of work	Before matching	2.665	2.735	−5.99	0.057	2.606	2.709	−8.45	0.003
	After matching	2.69	2.642	4.2	0.334	2.578	2.536	3.55	0.348
Household income	Before matching	1.724	1.198	112.31	0	1.757	1.231	105.47	0
	After matching	1.572	1.572	0	1	1.615	1.615	0	1
Educational background	Before matching	2.196	2.081	17.56	0	2.24	2.153	13.25	0
	After matching	2.165	2.212	−7.36	0.091	2.267	2.288	−3.32	0.38

. From the table, it can be seen that the absolute value of the standardized bias after matching is within 10%, and the reduction in standardized bias is significant, while the p-value after matching is not statistically significant. This indicates that the distribution of covariates between the treatment group and the control group is relatively balanced, and the matching is quite successful. After matching, the inner and outer circles obtained 1056 and 1405 pairs of resident samples, respectively.

3.4. Modeling Method Identification

The built environment data used in this study is considered aggregate data as it comes from the station area level, while socioeconomic attribute data is disaggregated, sourced from individual residents. To avoid ecological fallacy in modeling, the variability of dependent variables at the site level was evaluated. A two-level null model without explanatory variables was established using Equations (4) and (5), and the inter-group correlation coefficient (ICC) was used for assessment. If the ICC at the station level exceeded a certain threshold, a multi-level linear model (HLM) was used; otherwise, a traditional linear regression model (OLS) was employed.

$$Y_{ij}={\beta }_{oj}+r_{ij}$$

(4)

$${\beta }_{oj}={\gamma }_{oo}+{\mu }_{oj}$$

(5)

Y_ij denotes the travel activity characteristics exhibited by the i-th resident at station j; β_oj signifies the intercept at the individual level; β_oj is considered as the random variable at the individual level; γ_oo represents the intercept derived from regression excluding station-specific considerations; μ_oj signifies the residual variance between the intercept of station j and the overall intercept.

Following the null model variance test, the results suggest that each dependent variable shows distinct site-level concentrations but generally displays weak overall associations. The academic community has yet to establish a clear critical value for ICC. For instance, Cohen proposed an ICC critical value of 0.059 [21], James observed a median ICC value of 0.138 [22], and Mathieu suggested using a cross-layer model when ICC falls between 0.15 and 0.30 [23]. Based on these findings, a multiple linear regression model (OLS) was employed to analyze travel distance, travel time, and activity duration. The binary Logit model is used to study the impact of the built environment on travel behavior for car trips.

Multiple linear regression is a statistical method that combines two or more independent variables linearly to estimate or predict a dependent variable. The expression is as follows:

$$Y_i={\beta }_0+{\beta }_1{X}_{i1}+{\beta }_2{X}_{i2}+\dots +{\beta }_n{X}_{in}+{\epsilon }_i(i=\mathrm{1,2}\dots n)$$

(6)

In the equation, ${\beta }_0$ represents the constant term; ${\beta }_1\dots {\beta }_n$ represents the regression coefficient; $X_{in}$ represents the n-th independent variable of individual i; ${\epsilon }_i$ represents the random error.

The binary Logit model is a type of generalized linear model where the dependent variable is a binary classification variable. It requires that the sum of the residuals is zero and follows a binomial distribution. Unlike linear models, this model uses the maximum likelihood method for estimation. The expression is as follows:

$$\mathrm{L}\mathrm{o}\mathrm{g}\mathrm{i}\mathrm{t}\left(p\right)={\beta }_0+{\beta }_1{X}_1+\dots +{\beta }_n{X}_n$$

(7)

In the formula, ${\beta }_0$ is the constant term; ${\beta }_1\dots {\beta }_n$ is the regression coefficient; ${X_1\dots X}_n$ is the independent variable of the model.

4. Descriptive Statistical Analysis

4.1. Analysis of Environmental Changes

This section investigates the impact of subway openings on the built environment along its route. The study focuses on discrete stations and continuous corridors as its subjects. Research shows that walking remains the main mode of transport before and after subway implementation, with over 90% connectivity within 500 m of stations [24]. Thus, research scales into 0–500 m and 500–1000 m. It also analyzes facility clustering, land use patterns, and spatial distribution along subway corridors over different timeframes. To ensure data collection feasibility, the study uses a 2 km radius around the subway line [25], utilizing a GIS grid of 500 m × 500 m as the research unit.

4.1.1. The Changes in the Built Environment of Subway Stations

Information entropy, dominance, and balance were used to measure the changes in the built environment of stations before and after the opening of the subway. The relevant measurement results were summarized in

Table 5. Summarizes diverse index values across varying scale ranges before and after the opening of subway.

Indicators	In One Ring		First toSecond Ring		Second to Third Rings		Beyond the Third Ring		All Lines
	Subway Opening
	Before	After	Before	After	Before	After	Before	After	Before	After
0–500 m layer
Information entropy	1.86	1.7	1.73	1.64	1.61	1.36	1.28	1.34	1.46	1.42
Dominance Index	0.71	0.88	0.85	0.94	0.96	1.22	1.30	1.24	0.85	0.94
Index of equalization	0.72	0.66	0.67	0.64	0.63	0.53	0.50	0.52	0.56	0.55
500–1000 m layer
Information entropy	2.00	1.65	1.74	1.49	1.68	1.30	1.62	1.42	1.69	1.43
Dominance Index	0.59	0.94	0.84	1.10	0.91	1.29	0.97	1.17	0.89	1.15
Index of equalization	0.77	0.64	0.67	0.58	0.65	0.50	0.63	0.55	0.66	0.56

.

Upon comparing outcomes from both layers, it becomes evident that prior to the subway’s inauguration, the peripheral zone maintained a robust balance of amenities, making it susceptible to potential disruptions. As facilities continue to accumulate, the richness and balance of amenities decline rapidly. Following the subway’s inauguration, the information entropy and equilibrium of facilities at third-ring stations generally decrease, while those beyond exhibit diverse exponential transitions depending on their location. This phenomenon arises because the commencement and operation of the subway system have led to the concentration of commercial and other facilities around stations, thereby disturbing the equilibrium among them.

4.1.2. The Changes in the Built Environment Along the Subway Corridor

Using the ARCGIS kernel density function, a kernel density distribution map of various facilities within the 2 km subway corridor was generated to compare the agglomeration and distribution of these facilities before and after the subway opening.

As illustrated in Figure 2, the changes in core density of industrial and enterprise facilities before and after the subway’s opening both demonstrate a decreasing trend outward, with the change in the Chenggong New Town area being particularly pronounced. Simultaneously, commercial facilities along the subway line have experienced a sharp increase, and their spatial distribution has transitioned from a pattern of higher density in the north and lower density in the south to a more balanced distribution, with core density reaching six times that of the period prior to the subway’s opening. As illustrated in Figure 3, following the subway’s inauguration, the density of public facilities in the central urban area increased by 1.86 times, while the concentration in the urban outskirts also rose. However, there were no significant changes observed in the distant suburbs. As illustrated in Figure 4, following the subway’s opening, the core density of residential facilities decreased by approximately 32%, with a shift occurring towards the two ends of the subway line.

To further investigate the extent of changes in land functions and their spatial distribution along the subway line before and after its construction, a quantitative identification method for urban functional zones was employed, followed by accuracy verification. This process resulted in the creation of spatial distribution maps illustrating various land functions within the corridor before and after the subway’s opening. The results (Figure 5) indicate that after the subway was completed, both the land functions and spatial distribution along the line experienced significant changes. Specifically, the area with no data decreased by 48.3%, releasing 25.25 km² of land.

The findings indicate that building subway systems boosts land development along their routes, speeding up the process (Figure 6). This leads to a greater variety and quantity of facilities, enhancing land functions. Research has additionally demonstrated that subway construction increases land values, offering better long-term returns for residential and commercial properties. Residential and commercial land areas along subway routes have expanded. Conversely, green spaces have limited economic benefits, and industrial and commercial land users relocate to reduce costs. There is a trend towards intensified land development around subway stations, reducing the gap in peripheral facilities around the city center. Furthermore, there are marked spatial variations in alterations to the built environment within the directly affected areas by the subway, affirming the cyclical impact of subway systems as urban infrastructure.

4.2. Analysis of Changes in Travel Behavior

Based on the aforementioned analysis, the inauguration of the subway system has precipitated notable transformations in the architectural landscape encompassing the corridor and station precincts. These alterations are poised to directly impact the diverse behavioral patterns exhibited by residents. Therefore, this section, drawing on Activity Theory, compares the differences in residents’ behavior before and after the subway’s opening across several aspects, including travel modes, travel distance, commuting space, and activity duration, in order to capture the impact of the subway on residents’ travel behavior along the route.

On the other hand, the impact of a station on neighboring residents varies. To analyze this, the station vicinity is divided into two zones: 0–500 m and 500–1000 m. Demographic and travel data before and after the subway opening are collected from these zones. Moreover, to discern and assess genuine shifts in travel behaviors while mitigating confounding social and economic factors the propensity score matching (PSM) [26] method was employed to juxtapose pre- and post-subway opening samples from varied cohorts. Following matching, 1056 and 1405 resident pairs were derived from the inner and outer concentric zones, respectively.

4.2.1. Travel Mode Selection

Figure 6 show that after the subway opening, motorized travel in the inner circle increased while decreasing in the outer circle. Walking remains above the city average, but walking in the inner circle decreased while increasing in the outer circle. Electric bicycle and bicycle travel saw little change, with electric bicycles being more popular. Car travel decreased slightly post-subway opening, suggesting the subway’s role in reducing car use. There is no evidence to show whether the subway has a direct or indirect impact on car use. Additionally, from a spatial location perspective, the inhibitory effect is better in the outer circle farther away from the subway station. This finding is inconsistent with the expected pattern of the subway’s effect on cars.

4.2.2. Matching Sample Space Utilization Characteristics

Regarding travel distance (

Table 6. Matching sample travel distance statistics.

Statistical Category	0–500 m Layer			500–1000 m Layer
	Pre-Subway Opening (m)	Post-Subway Opening (m)	Cohen’s d	Pre-Subway Opening (m)	Post-Subway Opening (m)	Cohen’s d
Distance traveled throughout the day	7986	7913	0.092 **	9128	9324	−0.212 ***
Average travel distance per trip	3389	3301	0.26 ***	3651	3740	−0.24 ***
Average commuting distance	4087	3759	0.835 ***	4274	4194	0.189 ***
Average non-commuter distance	2676	3038	−1.26 ***	3218	3296	−0.24 ***

Note: The paired t-test is used to examine whether there is a significant difference in travel behavior between the groups before and after the subway opening. ** indicates significance at the 5% level; *** indicates significance at the 1% level.

), prior to the subway’s opening, residents in the outer circle traveled up to 9.12 km, while those in the inner circle averaged less than 8 km. After the subway’s inauguration, residents in the outer circle traveled an average of 9.32 km, and the difference in distances within the inner circle decreased, showing improved convenience in residents’ daily lives. Non-commuting travel distances slightly increased, while commuting distances decreased, suggesting a slight improvement in the balance between employment and housing, possibly due to subway access and improved local amenities.

Travel distance provides some insight into residents’ activity paths, but it does not fully capture their entire activity range. An analysis was conducted on residents’ average daily activity space before and after the subway’s inauguration. Results show that after the inauguration, residents in the inner circle of the station had a slight decrease in daily activity space, from 11.47 km² to 11.42 km². In contrast, residents in the outer circle saw a significant expansion in their daily activity space, increasing from 12.40 km² to 14.32 km².

Let us look at the resident commuter OD flow (Figure 7). Prior to the subway’s inauguration, a majority of residents were employed within the central district of the historic city, leading to centralized commuting patterns. Residents in the outer circle worked more widely and deviated from the city center. After the opening, the distribution of workplaces changed significantly. The concentration of employment areas in the main city decreased, small-scale employment areas emerged in the suburbs, and the commuting distance of suburban residents shortened. The east–west commuting link was strengthened, aligning with the urban development strategy.

Based on the distribution and evolution of active points (

Table 7. OD distribution of matched samples.

The Circle of Residence	Residential Location, (Non-) Workplace, and Spatial Correlation	Proportion Pre-Subway Opening (%)	Proportion Post-Subway Opening (%)
0–500 m		24.72	30.95
		26.6	27.3
500–1000 m		17.59	18.78
		24.41	23.62
		22.84	11.99
		31.36	33.79
0–500 m buffer zone	500–1000 m buffer zone	Travel orientation

) before and after the subway inauguration, significant changes are observed in residents’ workplace activities near the inner circle. Non-occupational activities among residents living in the station’s periphery show a noticeable shift away from the station’s core. These findings support the statistical evidence of an increased activity space for residents within this radius.

4.2.3. Matching Sample Activity Time Utilization Characteristics

The duration of activities is an important dimension reflecting the characteristics of residents’ stay. As shown in

Table 8. Matching sample activity duration statistics.

Statistical Category	0–500 m Layer			500–1000 m Layer
	Pre-Subway Opening	Post-Subway Opening	Cohen’s d	Pre-Subway Opening	Post-Subway Opening	Cohen’s d
Total daily activity duration (min)	448	429	0.43 ***	470	454	0.35 ***
Constraint-based activity duration (min)	401	345	1.50 ***	416	372	1.12 ***
Unconstrained activity duration (min)	144	124	1.49 ***	121	147	−1.93 ***

Note: The paired t-test is used to examine whether there is a significant difference in travel behavior between the groups before and after the subway opening. *** indicates significance at the 1% level.

, after the subway was opened, the daily activity duration of residents in both circles decreased, but the reduction was relatively small. After statistics based on the nature of activities, it was found that the duration of restricted activities of residents in both circles decreased significantly, while the variation pattern of the duration of free activities showed differences among circles. The specific manifestation is that the duration of free activities of residents in the 0–500 m circle of the station decreases, while that of residents in the 500–1000 m circle of the station increases.

5. Analysis of Influence on Residents’ Travel Activity Before and After Subway Opening

In the aforementioned feature analysis, the observed decrease in residents’ travel distances contradicts the anticipated expansion of residents’ activity spaces facilitated by subway infrastructure. The more substantial decline in the percentage of car journeys in the outer zones deviates from the anticipated influence of subway systems on automobile usage. Thus, travel distance and car travel will be scrutinized as variables influencing residents’ travel behaviors. Furthermore, residents’ time allocation is influenced by familial or personal factors, which operate distinctively from those impacting travel decisions. Accordingly, the consumption of travel time and the duration of activities, as variables influencing residents’ activities behaviors.

Linear regression models (OLSs) were constructed using the variables listed in Table 2. The residuals of each model exhibited normal distribution, with a Durbin–Watson (DW) index close to 2, and the models passed the F-test. Therefore, we can confidently assert the validity of the model, given its adherence to the fundamental assumptions of linear regression. Subsequent analysis delves into the outcomes of these models.

5.1. Dependent Variable Is the Result of the Model of Car Travel Choice

Table 9. Binomial logit regression results of car travel at different scales.

Variable Name	0–500 m Layer		500–1000 m Layer
	Per-Subway Opening (B)	Post-Subway Opening (B)	Pre-Subway Opening (B)	Post-Subway Opening (B)
Constant term	−0.353	−3.561 **	−5.986 ***
Number of shopping facilities	−0.020	0.519 **	0.113 *	0.001
Number of schools	−0.082	−0.867 ***
Number of recreational facilities			−0.022 **	−0.078
Distance to nearest bus stop			−0.058	−2.051 **
Mixing			3.255 ***	1.911
Site population	−0.042 **	−0.039 *
Household population	−0.472 ***	−0.062
Number of electric bicycle in households	−1.169 ***	−0.540 ***	−1.083 ***	−0.419 ***
Number of family cars	2.772 ***	2.924 ***	2.575 ***	2.966 ***
Occupation (scientific research, the literature and health as reference)
Retired population	−1.600 ***	−1.641 ***	−1.389 ***	−2.108 ***
Private owners	1.192 ***	0.364	0.760 ***	0.800 ***
Students	−1.510 ***	−0.380
Education (primary school and below for reference)
Middle school and high school	−1.078 **	0.214
Bachelor’s degree or above			1.198 ***	1.045 **
Income (low income for reference)
Middle income	0.496 *	0.374	−0.037	0.578 **
High Income			0.942 *	1.609 **
Female (male reference)	−0.804 ***	−2.015 ***	−0.709 ***	−1.882 ***
Pseudo-R2	0.375	0.367	0.381	0.395

Note: * indicates significance at the 10% level; ** indicates significance at the 5% level; *** indicates significance at the 1% level.

illustrates the impact of built environment and socioeconomic attributes on travel behavior in various scenarios and scales. Before the subway open, factors influenced car usage near the station. Limited parking in densely populated historic areas led residents to prefer non-car transport. Additionally, each extra family car increased the likelihood of car travel by 16 times. Conversely, in the outskirts, abundant recreational facilities reduced car dependency. Urban integration was found to increase car use, contrary to previous studies, possibly due to higher incomes and more car ownership in integrated areas, indirectly raising the likelihood of car travel.

Following the inauguration of the subway system, the proliferation of schools within the inner city area has notably fostered a heightened preference among residents for car-independent modes of transportation. This trend can be ascribed to diminished vehicular necessity facilitated by the close proximity and comprehensive accessibility of educational institutions. Concurrently, the seamless operation of the subway system affords residents the option to utilize it for their daily commutes, thereby mitigating reliance on automobiles. Despite car-owning households constituting the predominant demographic for vehicular travel, there has been a discernible rise in car usage subsequent to the subway’s commencement, relative to its pre-commencement period. In outlying areas, the proximity of bus stations significantly augments the accessibility of bus transit options, thereby diminishing residents’ dependency on personal vehicles to a certain degree.

5.2. The Model Results with the Dependent Variable Set as the Daily Travel Distance

Table 10. Shows the regression results of travel distance for dependent variables at different scales.

Variable Name	0–500 m Layer		500–1000 m Layer
	Pre-Subway Opening (B)	Post-Subway Opening (B)	Pre-Subway Opening (B)	Post-subway Opening (B)
Constant term	2.043 ***	1.624 ***	1.130 ***	1.114
Number of shopping facilities			−0.023 *	−0.006 ***
Number of schools	−0.095 **	0.243 ***	−0.039 *	−0.009 ***
Number of recreational facilities	−0.019 *	−0.072	0.002	0.033 *
Distance to the nearest subway stop	−0.173	2.197 **
Distance to nearest bus stop	0.593	−2.01 ***	0.131	−0.646 **
Mixing	−0.268	−0.79 *	0.821 ***	1.252 **
Site population	0.017 **	−0.016 **
Number of electric bicycles in households			0.01	0.107 **
Number of family cars	0.351	0.357 ***	0.112 ***	0.366 ***
Occupation (scientific research, the literature and health as reference)
Retired population	−0.419 ***	−0.167	−0.201 ***	−0.205 *
Students	0.619 ***	−0.307	−0.28 ***	−0.18
Education (primary school and below for reference)
Middle school and high school	0.31 **	0.38 ***	0.172 **	0.551 ***
Bachelor’s degree or above	0.438 **	0.596 ***	0.269 ***	0.777 ***
Income (low income for reference)
Female (male reference)	−0.197 **	−0.204 ***	−0.086 ***	−0.347 ***
Pseudo-R2	0.119	0.116	0.114	0.156

Note: * indicates significance at the 10% level; ** indicates significance at the 5% level; *** indicates significance at the 1% level.

displays the model results, showing that before the subway opening, built environment variables affecting residents’ activity times in the inner circle are mainly related to density, while in the outer circle, they are influenced by both density and diversity. The inner circle model’s variables impacting travel distance changed pre and post subway opening, whereas the outer circle model’s variables showed consistent influence on travel distance.

Before the subway opening, the presence of schools and recreational facilities near the station increased travel distances, while the station population had the opposite effect. The reason is that the dense distribution of schools alleviates the unbalanced student-residence pattern and greatly reduces the travel distance of students and their companions. After the subway opened, the impact of school density on travel distances changed from negative to positive. This shift may be due to improved infrastructure near subway stations, changing residents’ commuting patterns. Before the subway, increased station integration in the outer circle led to longer travel distances due to enhanced infrastructure. Post-subway opening, substantial built environment factors within this region are expected to escalate.

5.3. The Model Result with the Dependent Variable Being the Duration of Activities Throughout the Day

Based on the model results (

Table 11. Displays the regression results of activity duration for dependent variables at various scales.

Variable Name	0–500 m Layer		500–1000 m Layer
	Pre-Subway Opening (B)	Post-Subway Opening (B)	Pre-Subway Opening (B)	Post-Subway Opening (B)
Constant term	14.80 ***	6.041 ***	19.396 ***	16.978 ***
Number of shopping facilities	0.38 **	−0.009
Number of schools	0.10	1.711 ***
Distance to nearest subway stop	−0.13	9.488 ***
Mixing	0.90	−4.406 ***
Site population	−0.06 **	0.038 *	0.028	0.019 *
Household population	0.28	0.501 ***
Number of electric bicycles in households			0.157	0.987 ***
Number of family cars			0.98 **	0.398
Occupation (scientific research, the literature and health as reference)
Retired population	−9.41 ***	−4.851 ***	−14.3 ***	−8.721 ***
Students	1.55 *	1.049 *	4.299 ***	2.095
Education (primary school and below for reference)
Middle school and high school	0.81	0.936 **	3.343 ***	1.167 *
Bachelor’s degree and above	0.48	1.921 ***	3.395 ***	1.582 **
Income (low income as reference)
Middle income	−0.15	−0.807 ***
High earnings	−0.73	1.736 *
Female (male reference)	0.06	−0.443 *
Pseudo-R2	0.382	0.376	0.368	0.29

Note: * indicates significance at the 10% level; ** indicates significance at the 5% level; *** indicates significance at the 1% level.

), before the subway’s opening, the number of shopping facilities in the inner circle correlated positively with residents’ activity duration, while station population density showed a negative correlation. After the subway’s opening, there was a significant increase in factors affecting residents’ activity duration. The number of schools, station population density, and proximity to subway stations positively impacted activity duration, while population density had a negative correlation. Occupation had a significant influence on activity duration before the subway’s launch. After the inauguration, factors such as family size, income, and gender influenced activity duration, indicating changes in residents’ activity patterns due to subway construction. Before the subway’s opening, outer circle residents’ activity duration was influenced only by social and economic factors. After the opening, the built environment started to influence activity duration, with station population density and time savings from electric bicycle use showing positive correlations with activity duration.

6. Discussion and Conclusions

6.1. Discussion

This study explores the effects of the subway on residents’ travel behaviors and the relationship between the subway, built environment, and travel behavior. It examines changes in the built environment, residents’ characteristics, and behaviors before and after the subway’s introduction, guided by scenario analysis, hierarchical structuring, and scalar perspectives. The study has generated valuable research findings.

Firstly, the opening of the subway has a significant impact on the built environment of the surrounding corridors and stations, a fact that has been corroborated by previous research [27]. After the subway opening, there is a noticeable increase in various amenities along its route, balancing out the distribution around the city center. There is a clear trend of increased land development and utilization. Land use at both ends of the line improves, with reduced facility density at stations, showing a clear stratification of the peripheral area.

Subsequently, the inauguration of the subway system has profoundly influenced the daily behaviors of residents living along its route. Notably, there has been an increase in the presence of highly educated, high-income individuals, and those with specialized occupational skills near the subway stations, alongside a significant rise in automobile ownership among local households—signals of the onset of gentrification. LIANG’s research also shows that the influx of higher-income, highly educated individuals and the outflow of lower-income residents are closely linked to the development of subway infrastructure [28]. Moreover, there has been a noticeable increase in the duration, span, and frequency of residents’ trips, particularly in areas near the subway stations. Upon adjusting for residents’ economic characteristics, significant differences in travel behavior across both time and space become apparent. In the inner circle, residents’ travel distances have decreased, their activity spaces have contracted, and their travel intensity has diminished—this stands in stark contrast to the outer circle areas. Furthermore, following the subway’s introduction, there has been a slight reduction in the proportion of car trips, indicating the subway’s role in curbing car usage. From a spatial perspective, the effect on car usage is more pronounced in areas farther from the subway stations. This contradicts the general pattern where the subway’s influence on car usage diminishes with distance from the stations. This phenomenon may be explained by the fact that residents in outer areas often face challenges such as limited coverage and infrequent traditional public transport options, such as buses. The subway’s opening has expanded their travel choices, especially during peak hours when the subway’s advantages over buses or private cars are most evident, contributing to a reduction in car usage.

Finally, the inauguration of the subway system fundamentally alters the mechanisms through which the built environment influences residents’ travel behaviors. In terms of temporal scenarios, there is an escalation in both the quantity and dimensions of pivotal variables, with bus proximity emerging as a notable dimension, thereby altering how the built environment shapes residents’ travel behaviors. Regarding spatial scales, factors such as design and diversity within the built environment begin exerting influence on travel behaviors in the inner sphere, whereas density gains prominence in the outer periphery. In terms of influence mechanisms, the inner built environment undergoes fundamental changes in its impact on residents’ behaviors, whereas in the outer periphery, the magnitude of influence varies. Regarding residents’ travel patterns and activity levels, pre-subway inauguration, high-density facilities shorten travel distances, increased land mixing extends travel distances and encourages automobile usage, while shopping amenities influence residents’ activity durations. Post-subway opening, high-density facilities and expanded station networks diminish residents’ travel distances, enhanced bus services extend travel distances while boosting public transit usage, and land mixing limits the escalation of activity durations.

6.2. Conclusions

This study on the impact of subway openings on residents’ travel activities has yielded several key findings. First, in comparison to the built environment, residents’ socioeconomic attributes have a relatively stable influence on travel behavior. This suggests that the built environment plays a crucial role in shaping residents’ travel behaviors, making it easier to guide and influence these behaviors. As a result, it is important to prioritize a comprehensive workflow for planning, developing, and post-evaluating the built environment in station areas. Second, the impact of the built environment on residents’ travel behavior varies not only by context but also across different spatial scales. Therefore, when planning and developing the built environment in station areas, attention should be given to location and scale factors, and land development strategies should be tailored to the specific characteristics of each station zone. Finally, after the subway opens, proximity to bus services significantly influences nearly all residents’ travel activities. This underscores the growing importance of public transportation as urban areas expand. Consequently, a dual approach is required, focusing on both enhancing physical infrastructure and improving service quality. These research conclusions and policy recommendations offer valuable insights for cities in the early stages of developing their rail transit systems.

Furthermore, this study is not without its constraints. The travel data of residents utilized in this study has been derived from a selected sample segment, and the survey method employed is a door-to-door RP survey, which may introduce bias into the residents’ self-assessment of their travel patterns. Hence, it remains uncertain whether these data can effectively capture the authentic travel behaviors of residents along the route. With the advent of the big data era, encompassing sources like mobile phone signals and GPS data, researchers are poised to provide more precise insights into individual daily travel activities, with future studies well-positioned to exploit this opportunity fully. Moreover, in contrast to panel data, the mixed cross-sectional data employed here offers limited insights into causality, thus warranting future investigations into the longitudinal dynamics of individuals within a narrow scope.